Encyclopaedia of

Mathematical Sciences
Volume 70

Editor-in-Chief: R.V. Gamkrelidze

 Yu.G. Reshetnyak (Ed.)

Geometry IV
Non-regular Riemannian Geometry

With 58 Figures

Springer-Verlag Berlin Heidelberg GmbH

 Consulting Editors of the Series:
A.A. Agrachev, A.A. Gonchar, E.F. Mishchenko,
N.M. Ostianu, V.P. Sakharova, A.B. Zhishchenko

Title of the Russian edition:

Itogi nauki i tekhniki, Sovremennye problemy matematiki,
Fundamental'nye napravleniya, VoI. 70, Geometriya 4,
Publisher VINITI, Moscow 1989

Mathematics Subject Classification (1991):

52BXX, 53AXX, 53BXX, 53C20, 53C21, 53C45, 58G30

ISBN 978-3-642-08125-5

Library of Congress Cataloging-in- Publication Data

GeometriIâ IV. English. Geometry IV: non-regular Riemannian geometry (Yu. G. Reshetnyak (ed.). p. cm.-
(Encyclopedia of mathematieal sciences; v. 70)
Includes bibliographical references and indexes.
ISBN 978-3-642-08125-5 978-3-662-02897-1 (eBook)
DOI 10.1007/978-3-662-02897-1
1. Geometry, Riemannian. 1. Reshetniâk, tUrii Grigor'evich.
Il. Title. III. Title: Geometry 4. IV. Series.
QA649.G46713 1993 516.3'73-dc20 93-13858

This work is subject to copyright. AII rights are reserved, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, reuse of iIIustrations, recitation, broadcasting, reproduction on
microfilms or in any other ways, and storage in data banks. Duplication of this publication or parts thereof is
permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version,
and permission for use must always be obtained from Springer-Verlag. Violations are liable for prosecution under
the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1993
OriginaIly published by Springer-Veriag Berlin Heidelberg New York in 1993
Softcover reprint of the hardcover Ist edition 1993
Typesetting: Aseo Trade Typesetting Ltd., Hong Kong
41(3140-543210 - Printed on acid-frec paper
 List of Editors, Authors and Translators

Editor-in-Chief

R.V. Gamkrelidze, Russian Academy of Sciences, Steklov Mathematical Insti-

tute, ul. Vavilova 42, 117966 Moscow, Institute for Scientific Information
(VINITI), ul. Usievicha 20 a, 125219 Moscow, Russia

Consulting Editor

Yu.G. Reshetnyak, Institute of Mathematics, Siberian Branch of the Russian

Academy of Sciences, Universitetskij pro 4, 630090 Novosibirsk, Russia

Authors

V.N. Berestovskij, Omsk State University, ul. Mira, 55 "A", 644077 Omsk,
Russia
I.G. Nikolaev, Institute of Mathematics, Siberian Branch of the Russian Acad-
emy of Sciences, Universitetskij pro 4, 630090 Novosibirsk, Russia
Yu.G. Reshetnyak, Institute of Mathematics, Siberian Branch of the Russian
Academy of Sciences, Universitetskij pro 4, 630090 Novosibirsk, Russia

Translator

E. Primrose, 12 Ring Road, Leicester LE2 3RR, England

 Contents

I. Two-Dimensional Manifolds of Bounded Curvature

Yu.G. Reshetnyak
3

II. Multidimensional Generalized Riemannian Spaces

V.N. Berestovskij, I.G. Nikolaev
165

Author Index
245

Subject Index
246
 I. Two-Dimensional Manifolds of Bounded Curvature

Preface

The book contains a survey of research on non-regular Riemannian geome-

try, carried out mainly by Soviet authors. The beginning of this direction oc-
curred in the works of A.D. Aleksandrov on the intrinsic geometry of convex
surfaces. For an arbitrary surface F, as is known, all those concepts that can be
defined and facts that can be established by measuring the lengths of curves on
the surface relate to intrinsic geometry. In the case considered in differential
geometry the intrinsic geometry of a surface is defined by specifying its first
fundamental form. If the surface F is non-regular, then instead of this form it
is convenient to use the metric PF' defined as follows. For arbitrary points
X, Y E F, PF(X, Y) is the greatest lower bound of the lengths of curves on the
surface F joining the points X and Y. Specification of the metric PF uniquely
determines the lengths of curves on the surface, and hence its intrinsic geometry.
According to what we have said, the main object of research then appears as a
metric space such that any two points of it can be joined by a curve of finite
length, and the distance between them is equal to the greatest lower bound of
the lengths of such curves. Spaces satisfying this condition are called spaces with
intrinsic metric. Next we introduce metric spaces with intrinsic metric satisfying
in one form or another the condition that the curvature is bounded. This condi-
tion is introduced by comparing triangles in space with triangles on a surface of
constant curvature having the same lengths of sides.
The book contains two articles. The first is devoted to the theory of two-
dimensional manifolds of bounded curvature. This theory at present has a com-
plete character. It is a generalization of two-dimensional Riemannian geometry.
For a manifold of bounded curvature there are defined the concepts of area and
integral curvature of a set, the length and turn (integral curvature) of a curve.
One of the main results of the theory is the closure of the class of two-
dimensional manifolds with respect to the passage to the limit under certain
natural restrictions. In particular, this enables us to define two-dimensional
manifolds of bounded curvature by means of approximation by polyhedra. The
proof of the possibility of such an approximation is one of the main results of
the theory. In the account given here it is essential to use the analytic representa-
tion of two-dimensional manifolds of bounded curvature by means of a line
element of the form ds 2 = A(z)(dx 2 + dy2). The function A(Z) is such that its
logarithm is the difference of two subharmonic functions. In contrast to the case
of Riemannian manifolds the function A here may vanish and have points of
discontinuity. Some results in the theory of manifolds of bounded curvature do
not have a complete analogue in two-dimensional Riemannian geometry. Here
we should refer to some estimates and solutions of extremal problems, the theo-
rem on pasting, and so on.
The second article is devoted to the theory of metric spaces whose curvature
is contained between certain constants Kl and K 2 , where Kl < K 2 • The main
result of this theory is that these spaces are actually Riemannian. In each such
2 Yu.G. Resbetnyak

space we can locally introduce a coordinate system in which its metric is defined
by a line element ds 2 = gil dx i dxl, where the functions gil satisfy almost the
same regularity conditions as in ordinary Riemannian geometry. (We say "al-
most the same" because the functions gil only have second derivatives, general-
ized in the sense of Sobolev, that are summable in any degree p > 0; this implies
that the coefficients gil are continuous.) The theory of curvature in Riemannian
geometry can be transferred to the case of such spaces. Some relations here
are satisfied only almost everywhere (for example, the formula for representing
the sectional curvature of a manifold). In this article the authors also consider
some questions of Riemannian geometry. Applications are given of the theorem
on the Riemann property of spaces of two-sided bounded curvature to global
Riemannian geometry.
In particular, an axiomatic definition of a Riemannian space is obtained here,
based on representations in the spirit of synthetic geometry. A priori it is not
required that the spaces under consideration should be manifolds. This fact
follows from other axioms.

Yu.G. Reshetnyak
 I. Two-Dimensional Manifolds of
Bounded Curvature
Yu.G. Reshetnyak
Translated from the Russian
by E. Primrose

Contents

Chapter 1. Preliminary Information ............................... 6

§ 1. Introduction ............................. '. . . . . . . . . . . . . . . . . . 6

1.1. General Information about the Subject of Research and a Survey
of Results ............................................. 6
1.2. Some Notation and Terminology ......................... 10

§ 2. The Concept of a Space with Intrinsic Metric ............... . . . 11

2.1. The Concept of the Length of a Parametrized Curve ........ 11
2.2. A Space with Intrinsic Metric. The Induced Metric .......... 13
2.3. The Concept of a Shortest Curve ......................... 16
2.4. The Operation of Cutting of a Space with Intrinsic Metric ... 16

§ 3. Two-Dimensional Manifolds with Intrinsic Metric .............. 18

3.1. Definition. Triangulation of a Manifold ................ . . . 18
3.2. Pasting of Two-Dimensional Manifolds with Intrinsic
Metric ................................................ 22
3.3. Cutting of Manifolds ........... . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4. A Side of a Simple Arc in a Two-Dimensional Manifold ..... 31

§4. Two-Dimensional Riemannian Geometry ..................... 36

4.1. Differentiable Two-Dimensional Manifolds ................ 36
4.2. The Concept of a Two-Dimensional Riemannian Manifold ... 37
4.3. The Curvature of a Curve in a Riemannian Manifold.
Integral Curvature. The Gauss-Bonnet Formula ............ 43
4.4. Isothermal Coordinates in Two-Dimensional Riemannian
Manifolds of Bounded Curvature .............. . . . . . . . . . . 48
4 Yu.G. Reshetnyak

§ 5. Manifolds with Polyhedral Metric ............................. 49

5.1. Cone and Angular Domain ............................... 49
5.2. Definition of a Manifold with Polyhedral Metric ............ 53
5.3. Curvature of a Set on a Polyhedron. Turn of the Boundary.
The Gauss-Bonnet Theorem ............................. 58
5.4. A Turn of a Polygonal Line on a Polyhedron ............... 61
5.5. Characterization of the Intrinsic Geometry of Convex
Polyhedra ............................................. 63
5.6. An Extremal Property of a Convex Cone. The Method of
Cutting and Pasting as a Means of Solving Extremal Problems
for Polyhedra .......................................... 64
5.7. The Concept of a K-Polyhedron .......................... 68

Chapter 2. Different Ways of Defining Two-Dimensional Manifolds

of Bounded Curvature .......................................... 71
§ 6. Axioms of a Two-Dimensional Manifold of Bounded Curvature.
Characterization of such Manifolds by Means of Approximation
by Polyhedra ............................................... 71
6.1. Axioms of a Two-Dimensional Manifold of Bounded
Curvature ............................................. 71
6.2. Theorems on the Approximation of Two-Dimensional
Manifolds of Bounded Curvature by Manifolds with
Polyhedral and Riemannian Metric ....................... 76
6.3. Proof of the First Theorem on Approximation .............. 77
6.4. Proof of Lemma 6.3.1 ............ " .... " . . . .. .. . .. ... . . 90
6.5. Proof of the Second Theorem on Approximation ............ 94
§ 7. Analytic Characterization of Two-Dimensional Manifolds of
Bounded Curvature ......................................... 97
7.1. Theorems on Isothermal Coordinates in a Two-Dimensional
Manifold of Bounded Curvature .......................... 97
7.2. Some Information about Curves on a Plane and in a
Riemannian manifold ................................... 105
7.3. Proofs of Theorems 7.1.1, 7.1.2, 7.1.3 ....................... 112
7.4. On the Proof of Theorem 7.3.1 ............................ 119

Chapter 3. Basic Facts of the Theory of Manifolds of Bounded

Curvature ..................................................... 123
§ 8. Basic Results of the Theory of Two-Dimensional Manifolds
of Bounded Curvature ....................................... 123
8.1. A Turn of a Curve and the Integral Curvature of a Set ....... 123
8.2. A Theorem on the Contraction of a Cone. Angle between
Curves. Comparison Theorems ........................... 135
8.3. A Theorem on Pasting Together Two-Dimensional Manifolds
of Bounded Curvature ................................... 141
 I. Two-Dimensional Manifolds of Bounded Curvature 5

8.4. Theorems on Passage to the Limit for Two-Dimensional

Manifolds of Bounded Curvature ......................... 142
8.5. Some Inequalities and Estimates. Extremal Problems for
Two-Dimensional Manifolds of Bounded Curvature ......... 147
§9. Further Results. Some Additional Remarks 151

References 159
6 Yu.G. Reshetnyak

Chapter 1
Preliminary Information

§ 1. Introduction

1.1. General Information about the Subject of Research and a Survey of Results.
The theory of two-dimensional manifolds of bounded curvature is a generaliza-
tion of two-dimensional Riemannian geometry. Formally a two-dimensional
manifold of bounded curvature is a two-dimensional manifold in which there
are defined the concepts of the length of a curve, the angle between curves
starting from one point, the area of a set, and also the integral curvature of a
curve and the integral curvature of a set. For the case when the given manifold
is Riemannian, the integral curvature of a curve is equal to the integral of the
geodesic curvature along the length of the curve, and the integral curvature of a
set is equal to the integral of the Gaussian curvature of the manifold with respect
to the area. The remaining concepts in this case have the meaning that is usual
in Riemannian geometry. For an arbitrary two-dimensional manifold of bounded
curvature the integral curvature is a completely additive set function, which may
not admit representations in the form of an integral with respect to area.
Another particular case of two-dimensional manifolds of bounded curvature
consists of surfaces of polyhedra (not necessarily convex) in three-dimensional
Euclidean space. For them the integral curvature is an additive set function
concentrated on some discrete set, namely the set of vertices of the polyhedron.
If the set consists of a unique point, a vertex of the polyhedron, then its integral
curvature is equal to 2n - e, where e is the total angle of the polyhedron at this
vertex, that is, the sum of the angles of all its faces that meet at this vertex.
Three methods are known for introducing two-dimensional manifolds of
bounded curvature. The first of them is axiomatic. A two-dimensional manifold
of bounded curvature is defined as a metric space satisfying some special
axioms. The second method is based on approximation by two-dimensional
Riemannian manifolds or manifolds with polyhedral metric. It turns out that
under certain natural assumptions the limit of the sequence of two-dimensional
manifolds of bounded curvature is also a manifold of bounded curvature. Exact
formulations are given later; here we just mention that a certain condition of
boundedness of the curvature is fundamental in these assumptions. In particu-
lar, the limit of a sequence of manifolds with polyhedral metric is a manifold of
bounded curvature. This fact can be used for the definition of the class of two-
dimensional manifolds of bounded curvature.
A two-dimensional Riemannian manifold is a smooth manifold such that
for each local coordinate system there is defined in it a differential quadratic
form
 I. Two-Dimensional Manifolds of Bounded Curvature 7

The main concept of two-dimensional Riemannian geometry is the Gaussian

curvature. In order that it can be defined it is necessary to require that the
coefficients gij (i, j = 1, 2) have partial derivatives of the first and second orders.
There naturally arises the idea of considering "generalized" Riemannian geo-
metries obtainable if we weaken the requirements of regularity imposed on
the coefficients of the quadratic form ds 2 • It turns out that two-dimensional
manifolds of bounded curvature can be defined in such a way. We shall show
later how to do this. Here we consider the case when the line element of the
manifold has a certain special structure, namely such that gl1 = g22' g12 == o.
For the case of Riemannian manifolds the line element can always be reduced to
such a form by a transformation of the coordinates. The system of coordinates
for which gl1 == gn, g12 == 0 is called isothermal. Using such a form of the
coordinate system, we obtain a third analytic method of introducing two-
dimensional manifolds of bounded curvature.
The general plan of the theory of two-dimensional manifolds of bounded
curvature is due to A.D. Aleksandrov, who developed the geometrical aspects
of this theory (see Aleksandrov (1948b), (1948c), (1949b), (1950), (1954), (1957a),
(1957b), Aleksandrov and Burago (1965), Aleksandrov and Strel'tsov (1953),
(1965), Aleksandrov, Borisov and Rusieshvili (1975). An account of the theory
constructed by Aleksandrov is given in a monograph of Aleksandrov and
Zalgaller (1962). An analytic approach to the introduction and study of
two-dimensional manifolds of bounded curvature is due to Yu.G. Reshetnyak
(Reshetnyak (1954), (1959), (1960), (1961b), (1962), (1963a), (1963b)). Other au-
thors also took part in the development of individual aspects of the theory (the
corresponding references are given later in the main text).
The concept of a two-dimensional manifold of bounded curvature was in-
troduced as a development of the research of Aleksandrov on the intrinsic
geometry of convex surfaces (Aleksandrov (1944), (1945a), (1945b), (1947),
(1948a», and presented completely in his monograph Aleksandrov (1948a).
Chapter I of this article has an auxiliary character. Two-dimensional mani-
folds of bounded curvature are defined as metric spaces satisfying certain special
conditions. One of these conditions is that the metric of the space must be
intrinsic. In § 2 we give necessary conditions and a summary of the basic facts
relating to the theory of metric spaces with intrinsic metric. In § 3 we consider
two-dimensional manifolds with intrinsic metric. Here we go into details on the
definition of the operations of cutting up and pasting such manifolds. In addi-
tion, the concept of a side of a simple arc in a two-dimensional manifold has
important significance for what follows.
In § 4 we give a summary of the basic results of two-dimensional Riemannian
geometry. The main information concerning two-dimensional manifolds with
polyhedral metric is contained in § 5. In particular, for such manifolds we define
the concepts of integral curvature (or the turn) of a curve and the curvature of a
set, and we study the structure of a shortest curve on a two-dimensional poly-
hedron. Polyhedra play a special role in the theory of manifolds of bounded
curvature. By approximating an arbitrary manifold by polyhedra, in many cases
8 Yu.G. Reshetnyak

it turns out to be possible to reduce the solution of this or that problem to the
case of polyhedra, for which it becomes a problem with respect to the formula-
tion belonging to elementary geometry. This enables us to use for the solution
of such problems arguments based on intuitive geometric representations. For
example, some extremal problems are related to a number of problems for
which such a way of action leads to success.
The definition of two-dimensional manifolds of bounded curvature is given
in § 6. We regard the axiomatic definition as fundamental. The following fact
relating to classical Riemannian geometry is well known. Let T be a triangle in
a two-dimensional Riemannian manifold, that is, a domain homeomorphic to a
disc whose boundary is formed by three geodesics. We denote by (x, Pand y the
angles of this domain at the vertices ofthe triangle T and let w(T) be the integral
over T of the Gaussian curvature with respect to area. We put c5(T) = (X + p +
y - n. The quantity c5(T) is called the excess of the triangle T. As we know,
e>(T) = w(T). (This statement is a special case of the Gauss-Bonnet theorem; see
§ 4.) If the Gaussian curvature is non-negative, then it follows that c5(T) ~ 0 for
any triangle.
Let U be an arbitrary domain in a two-dimensional Riemannian manifold.
For any system of pairwise non-overlapping geodesic triangles 7; c U, i = 1,2,
... , m, we have the inequality

i~ w(7;) ~ Iv [Jf'(x)] +dO'(x),

where Jf' is the Gaussian curvature, dO' is the element of area, a+ = max{a, O}.
This property is taken as the basis for constructing the axiomatics of a two-
dimensional manifold of bounded curvature. A two-dimensional manifold of
bounded curvature is defined as a certain metric space. A geodesic is a curve,
any sufficiently small arc of which is a shortest curve, that is, such that its length
is equal to the distance between the ends. The concept of a shortest curve is
naturally defined for the case of metric spaces. It is also clear what we need to
call a triangle. In order to define the concept of the excess for a triangle in an
arbitrary metric space we need to know what the angle between two curves
starting from one point is, in the given case the angle between the sides of the
triangle. The corresponding definition is given in § 6. A manifold of bounded
curvature can be defined as a metric space that is a two-dimensional manifold
and is such that for any point of it there is a neighbourhood U for which the sum
of the excesses of pairwise non-overlapping geodesic triangles contained in U
does not exceed some constant qU) < 00, however these triangles are chosen.
The exact formulations are given in § 6. The final version of the axiomatics of
two-dimensional manifolds of bounded curvature is defined by the argument
that of the different equivalent forms of the axiomatics we must choose the
weakest.
One of the main results of the theory of two-dimensional manifolds of
bounded curvature is the characterization of such manifolds by means of
approximation by two-dimensional polyhedra, or, which reduces to the same
 I. Two-Dimensional Manifolds of Bounded Curvature 9

thing, by two-dimensional Riemannian manifolds. The difficulties that must be

overcome here are connected with the fact that starting from the axioms of
a manifold of bounded curvature it is required to establish some very deep
properties of it. In § 6 of this article we give an outline of the proof of theorems
on the approximation of a two-dimensional manifold of bounded curvature
by Riemannian manifolds. The reader can find complete proofs in the mono-
graph Aleksandrov and Zalgaller (1962). The proof of the necessary conditions
is based on arguments that are a development of the ideas worked out by
Aleksandrov in the study of the intrinsic geometry of convex surfaces. The proof
of the sufficient conditions outlined in § 6 is based on arguments different from
those given in Aleksandrov and Zalgaller (1962). (For a complete account of this
proof, see Reshetnyak (1962).)
The analytic characteristic of two-dimensional manifolds of bounded curva-
ture is given in § 7. We dwell on it in more detail, bearing in mind the fact that
for specialists thinking in terms of categories of mathematical analysis it is
the shortest path towards determining what is a two-dimensional manifold of
bounded curvature.
We first consider Riemannian manifolds. In a neighbourhood of any point
of such a manifold we can introduce a coordinate system in which the line
element of the manifold is expressed by the formula
ds 2 = ),(x, y)(dx 2 + dy2).
(As we said above, such a coordinate system is called isothermal.) The Gaus-
sian curvature .Yt" of a given manifold in this coordinate system admits the
representation
1
.Yt"(x, y) = - 2),(x, y) Aln ),(x, y).

Using known results of potential theory, we thus obtain

In ),(z) = ~ fG
fin Iz ~ ".Yt"(O)'(O d~ d" + h(z).
Here z = (x, y), , = (~, ,,), G is a domain on the plane, and h(z) is a harmonic
function. We now observe that for an arbitrary set E c G

w(E) = ff
E
.Yt"(O)'(O d~ d"
is the integral curvature of the corresponding set in the Riemannian manifold.
By virtue of this the integral representation for In ),(z) given above can be writ-
ten in the form

In ),(z) = ~ f
G
fin Iz ~ "dw(O + h(z). (1)
10 Yu.G. Reshetnyak

The last relation naturally suggests that if we wish to have generalized

Riemannian manifolds in some sense, for which the integral curvature is an
arbitrary completely additive set function, then it is sufficient in (1) to substitute
such a function for ro, and then to consider the geometry defined by the cor-
responding line element ds 2 = l(z)(dx2 + dy2). Such a path leads to a two-
dimensional manifold of bounded curvature.
In § 7 we give only drafts of the necessary proofs. A complete account of them
can be found in Reshetnyak (1954), (1960), (1961a).
For an arbitrary two-dimensional manifold of bounded curvature there are
defined the concepts of integral curvature and area of a set, and the integral
curvature (or the tum) of a curve. In § 8 we show how all these concepts can be
defined. We rely on the analytic representation of two-dimensional manifolds of
bounded curvature described in § 7.
In § 8 we give a survey of the main results of the theory of two-dimensional
manifolds of bounded curvature. Here we are concerned first of all with a
theorem on pasting of two-dimensional manifolds of bounded curvature and
theorems on passage to the limit. The class of two-dimensional manifolds of bounded
curvature turns out to be closed with respect to passages to the limit under
significantly weaker assumptions than for the class of Riemannian manifolds.
Among the main results of the theory of manifolds of bounded curvature
there are, in particular, those that concern extremal problems for manifolds of
bounded curvature. One of the main instruments for research is the method of
cutting and pasting created by Aleksandrov. This method uses essentially the
specific character of two-dimensional manifolds of bounded curvature. The
totality of all such manifolds is invariant with respect to operations connected
with the method indicated, which we cannot say, for example, about the class of
Riemannian manifolds.
In § 9 of this chapter we give a survey of further research into the theory of
two-dimensional manifolds of bounded curvature. The author has tried to ex-
press everything that is most essential in this topic.
1.2. Some Notation and Terminology. Later we assume that the concepts of
topological and metric spaces are known, like all the basic facts of general
topology. In particular, we assume that the reader knows what is a neighbour-
hood of a point in a topological space, an open or closed set, a connected
component, and so on.
Let us recall some standard notation, used in what follows.
Let A be a set in a topological space 9t Then A denotes the closure of A, A 0
denotes the totality of all interior points of A, and aA = A\A O denotes the
boundary of A.
The symbol ~n denotes the n-dimensional arithmetic Euclidean space of
points x = (Xl' X 2 , ... , x n ), where Xl> X 2 , ... , Xn are arbitrary real numbers. For
X = (Xl' X 2 , ... , Xn) E ~n we put

Ixi = Ji~ xf.

 I. Two-Dimensional Manifolds of Bounded Curvature 11

For arbitrary points x, Y E IR" the distance between x and y is assumed to be

equal to Ix - YI. The function p: (x, y) -+ Ix - yl is the metric. In a well-known
way a given metric defines some topology in IR". Speaking of IR" as a topological
space, we shall always have this topology in mind.
The space 1R2 will be called an arithmetic Euclidean plane. The symbol C
denotes the set of complex numbers. Later we shall often identify 1R2 and C,
regarding the point (x, y) E 1R2 and the complex number z = x + iy as one and
the same object.
The usual Euclidean plane is denoted later by the symbol Ie. As a metric
space 1E2 is isometric to 1R2.
Let B(O, 1) be the open disc {(x, y)lx 2 + y2 < I} in the plane 1R2, and B(O, 1)
the closed disc {(x, y)lx 2 + y2 ~ I}.
Henceforth the statement that some set in a topological space is homeo-
morphic to a disc (a closed disc) always means that this set is homeomorphic to
the disc B(O, 1) (respectively, the disc B(O, 1).

§ 2. The Concept of a Space with Intrinsic Metric

2.1. The Concept of the Length of a Parametrized Curve. We assume that the
concept of a metric space and some of the simplest information relating to it are
known.
Let M be a set in which a metric p is specified. We shall denote the metric
space obtained in this way by the symbol (M, pl. This notation is appropriate in
that later there will often arise the necessity of considering different metrics on
the same set. When no misunderstanding is possible we shall simply talk about
a metric space M.
Let M be a metric space and p its metric. A parametrized curve or path in the
space M is any continuous map x: [a, b] -+ M of the interval [a, b] of the set of
real numbers IR into M. We shall say that the path x: [a, b] -+ M joins the points
X, Y E M if x(a) = X, x(b) = Y.
A metric space M with metric p is called linearly connected if for any two
points X, Y of it there is a path joining these points.
A path x: [a, b] -+ M is called simple if it is a one-to-one map of the interval
[a, b]. A set L in the space M is called a simple arc if there is a simple path
x: [a, b] -+ M such that L = x([a, b]). Any simple path x: [a, b] -+ M satisfying
this condition is called a parametrization of the simple arc L.
A set r in a metric space (M, p) is called a simple closed curve in M if it is a
topological image of the circle S(O, 1) on the plane ~2. If r is a simple closed
curve in the metric space M, then there is a path x: [a, b] -+ M such that x(a) =
x(b), x([a, b]) = r and for any t 1 , t2 E [a, b] such that tl # t2 and at least one
of the points tl and t2 is not an end of the interval [a, b] the points X(tl) and
x(t 2 ) are distinct.
12 Yu.G. Reshetnyak

We shall call a set L eM a simple curve if L is closed and is either a simple

closed curve in M or a topological image of an arbitrary interval of the number
line IR (which, generally speaking, may not be closed).
Suppose we are given a path x: [a, b] -+ M in a metric space M. We specify
arbitrarily a finite sequence Q( = {to, t l' ... , t m } of points of the interval [a, b]
such that to = a ~ tl ~ ... ~ tm = b and put
m
s(x, Q() = L p[X(t
i=l
i - 1 ), x(t i )].

The least upper bound of s(x, Q() on the totality of all sequences Q( satisfying
the conditions mentioned above is called the length of the path x and denoted
by the symbol sp(x; a, b) or simply s(x; a, b) when no misunderstanding is possi-
ble. (The notation six; a, b) is necessary for those cases when we consider differ-
ent metrics in M and compare the lengths of the path x: [a, b] -+ M in these
metrics.)
We mention the following properties of length that follow immediately from
the definition.
I. Any path x: [a, b] -+ M in the space (M, p) satisfies the inequality
p[x(a), x(b)] ~ sp(x; a, b).
II. Suppose we are given a path x: [a, b] -+ M. Then for any c such that
a < c < b we have
six; a, b) = sp(x; a, c) + sp(x; c, b).
III. Suppose we are given a sequence of paths (xv: [a, b] -+ M), v = 1, 2, ...
and a path Xo: [a, b] -+ M. We assume that xo(t) = lim xv(t) for any t E [a, b].
Then 1-+0

sp(x; a, b) ~ lim sp(x v; a, b).

v-+ 00

Let L be a simple arc in the metric space (M, p). We specify arbitrarily a
parametrization x: [a, b] -+ M of the arc L. Then it is easy to establish that
sp(x; a, b) does not depend on the choice of parametrization x of the arc L. In
this case we shall call sp(x; a, b) the length of the simple arc L and denote it by
sp(L) or simply s(L).
Similarly, if r is a simple closed curve and x: [a, b] -+ M is an arbitrary
parametrization of it, then sp(x; a, b) does not depend on the choice of this
parametrization and is denoted henceforth by sp(r) or simply s(r).
Let L be a simple arc in the metric space M, and x: [a, b] -+ L a parametriza-
tion of L. The points A = x(a) and B = x(b) are called the end-points of L. All
the remaining points of L are called interior points of it. Let X = x(t 1 ) and
Y = x(t 2 ), t 1 < t 2 , be two arbitrary points of the simple arc. The set of all points
Z = x(t), where tl ~ t ~ t 2 , is obviously a simple arc. We shall denote it by
[X Y]. From property I of the length of a parametrized curve it follows that for
any simple arc L with end-points A and B we have
p(A, B) ~ s(L).
 I. Two-Dimensional Manifolds of Bounded Curvature 13

From property II it follows that for any interior point C of the simple arc
s([AB]) = s([AC]) + s([CB])
(A and B are the end-points of L).
Let F be a simple closed curve and A an arbitrary point of it. Then there is a
parametrization x: [a, b] -+ M of F such that x(a) = x(b) = A. Any two distinct
points X, Y of the simple closed curve F split it into two simple arcs, which we
denote by Fl and F2 • From Property II of the length of a parametrized curve it
follows that
sp(F) = Sp(Fl) + sp(F2 )·
In the given definitions, in principle, there can be an infinite value of the
length. If the length of the path x: [a, b] -+ M in the metric space (M, p) is finite,
then the given path is called rectifiable. Similarly, a simple arc (simple closed
curve) is called rectifiable if its length is finite.
Let L be a simple arc in the space (M, p). We assume that L is rectifiable.
Then it admits a parametrization x: [0, I] -+ M such that s is equal to the length
of the arc [x(O)x(s)] for each s E [0, I].

2.2. A Space with Intrinsic Metric. The Induced Metric. Suppose we are given
a metric space M and a set A eM. We shall say that the path x: [a, b] -+ M lies
in the set A (or goes into A) if x(t) E A for all t E [a, b].
A set A in a metric space (M, p) is said to be metrically connected if for any
two of its points there is a rectifiable path joining these points and lying in the
set A. In particular, the space (M, p) itself is said to be metrically connected iffor
any two of its points X, Y there is a rectifiable path joining these points.
A metric space (M, p) is called a space with intrinsic metric if it is linearly con-
nected and for any two of its points X, Y the quantity p(X, Y) is equal to the
greatest lower bound of lengths of arcs joining these points.
If (M, p) is a space with intrinsic metric, then M is metrically connected.
Suppose, for example, that M is the usual plane 1E2. For arbitrary points
X, Y E 1E2 suppose that p(X, Y) = 0 if X = Y and that p(X, Y) is equal to the
length of the interval with end-points X and Y if X #- Y. The metric defined in
this way on the plane 1E2 is obviously intrinsic.
Similarly, if M is a sphere 1:K ofradius r = 11ft in the space 1E3, then taking
for p(X, Y) the length of the shortest arc of the great circle passing through the
points X and Y, we obtain an intrinsic metric on the sphere 1:K. At the same
time, the metric Po(X, Y), where Po(X, Y) is the length of the interval in 1E3
joining the points X and Yon the sphere 1:K, is not intrinsic.
The metric spaces known from analysis, namely Hilbert space and, more
generally, any normed vector space, are spaces with intrinsic metric.
Let (M, p) be a metric space and a E M an arbitrary point of M. Let us specify
arbitrarily a number r > o. We denote the set of all points x E M such that
p(x, a) < r by the symbol B(a, r) and call it the open ball with centre a and radius
r. In certain cases considered later, instead of the word "ball" we shall say "disc".
14 Yu.G. Reshetnyak

The totality of all points x E M for which p(x, a) = r is denoted by the symbol
S(a, r) and called the sphere with centre a and radius r. In those cases when
B(a, r) is called a disc we shall call the set S(a, r) a circle. We put B(a, r) =
B(a, r) u S(a, r). The set B(a, r) is called the closed ball with centre a and radius r.
We mention the following properties of spaces with intrinsic metric.
Theorem 2.2.1 (Aleksandrov and Zalgaller (1962». Let (M, p) be a space with
intrinsic metric. If M is locally compact (that is, any point X E M has a neighbour-
hood whose closure is compact), then for any r> 0 the closed ball B(X, r) is a
compact set.
The metric space (M, p) is called complete if any sequence (xv), v = 1,2, ... , of
points of this space for which lim p(xv' xl') = 0 is convergent. According to
v-+oo,I'-+W
a well-known theorem of Hausdorff, for any metric space (M, p) there is a com-
plete metric space (M, p) such that M c M, p(x, y) = p(x, y) fOr any x, y E M,
and the set M is everywhere dense in M. The space (M, p) is unique in the
following sense. If (M', p') is another metric space connected with (M, p) like
(M, p), then there is a map j: M ..... M' such that j(M) = M', j(x) = x for any
x E M, and p'[j(x), j(y)] = p(x, y) for any x, y E M. We shall call (M, p) the
Hausdorff completion of the space (M, pl. Henceforth the metric ofthe Hausdorff
completion will be denoted like the metric of the original space.
Theorem 2.2.2 (Aleksandrov and Zalgaller (1962». The Hausdorff completion
of a metric space with intrinsic metric is also a space with intrinsic metric.
We mention here a general scheme for constructing the metric. Suppose we
are given a metric space (M, p), and let A be a connected set of this space. The
set A with metric p is itself a metric space - a subspace of (M, p). Even if (M, p)
is a space with intrinsic metric, the metric space (A, p) may not be of this kind.
Let us define a metric in the set A, which we denote by p". Namely, for arbitrary
points X, YEA we denote by p,,(X, Y) the greatest lower bound of lengths of
paths in the space (M, p)joining the points X and Y and lying in the set A.
Theorem 2.2.3. If A eM is a metrically connected set of the space (M, p),
then the function (X, Y) ..... p,,(X, Y) of a pair of points of A, defined in the way
indicated above, is a metric on the set A. This metric is intrinsic and for any path
x: [a, b] ..... M lying in the set A we have sp(x; a, b) = spJx; a, b).
The metric p" is called the induced intrinsic metric on the set A of the metric
space (M, pl.
Suppose, for example, that M is the three-dimensional Euclidean space 1E 3 ,
and that the set A is the sphere S(a, R) in this space. It is easy to show that in
the given case the quantity p,,(X, Y) is equal to the length of the shortest arc of
the great circle passing through the points X and Y, that is, the induced intrinsic
metric on the sphere S(a, R) coincides with the metric defined above.
Let L be a rectifiable simple arc in the metric space (M, p). Then for arbitrary
points X, Y E L the quantity PL(X, Y) is equal to the length of the arc [XY] of
the curve L. We assume that r is a rectifiable simple closed curve in the metric
 I. Two-Dimensional Manifolds of Bounded Curvature 15

space (M, p). We take arbitrarily the points X, Y E r. Then Pr<X, Y) = 0 if

X = Y, and if X #- Y, then X and Y split r into two arcs r 1 and r2 and
Pr(X, Y) = min {sp(rd, sp(r2 )}.
In the definition of the induced metric we do not exclude the case when
A = M. If (M, p) is a space with intrinsic metric, then the induced metric PM
coincides identically with the metric P already existing in M.
We note that, generally speaking, the induced intrinsic metric may tum out
to be poorly connected with the set A. For example, the topology determined by
the induced metric may be different from the natural topology of A as a subset
of the metric space. The following example serves to illustrate this statement.
Let H be a plane in the three-dimensional Euclidean space 1E 3 , and r a simple
closed curve in H such that any arc of rhas infinite length. We take arbitrarily
a point 0 not lying in H, and let A be the cone formed by the rays OX, where X
ranges over the curve r. The set A is metrically connected. In fact, if the points
Y, Z do not lie on one of the rays OX, X E r, then the simple arc composed of
the intervals [YO] and [OZ] joins these points. It is obviously rectifiable and is
contained in A.1f Yand Z lie on one ray OX, X E r, then the interval of this ray
with ends Y and Z is also a rectifiable simple arc joining Y and Z in the set A.
It is easy to verify that if Y and Z lie on one ray OX, X E r, then PA(Y, Z) is
equal to the length ofthe interval [YZ].1f Yand Z belong to different rays OX1 ,
OX2, Xl' X 2 E r, then PA(Y, Z) is equal to the sum of the lengths of the intervals
[YO] and [OZ]. The set A of the space 1E3 is homeomorphic to the plane 1E2. At
the same time the metric space (A, PA) does not admit a topological map into
a plane. This follows, for example, from the fact that if the sequence (Xv), v =
1, 2, ... , of points of the curve r converges to a point Xo on the plane H, then
PA(Xv, X o) -. 210Xol > 0 and so the sequence (Xv) does not converge to the
point Xo in the space (A, PA).
We mention an important special case when the topology determined by the
induced intrinsic metric in the set A coincides with the natural topology of A.
Namely, the following proposition is true.
Lemma 2.2.1. Let (M, p) be a metric space with intrinsic metric, and U an
open set of the space M. If the set U is connected, then it is metrically con-
nected. Moreover, if the point X o E U and the number b > 0 are such that the ball
B(Xo, b) c U, then for any points X, Y E B(Xo, b/2) we have

Pu(X, Y) = p(X, Y). (2.1)

The proof of (2.1) is based on the argument that if the path x: [a, b] -. M
joining the points X and Y of the ball B(Xo, b/2) contains points lying outside
the ball B(Xo, b), then its length must be not less than b. At the same time,
p(X, Y) ~ p(X, X o) + p(Xo, Y) < b, and so by the definition of p(X, Y) we can
consider only paths that are contained in the ball B(a, b), and therefore in the
set U. Hence it follows that Pu(X, Y) = p(X, Y).
Corollary. In the conditions of Lemma 2.2.1 the metrics Pu and P in the set U
are topologically equivalent (that is, they determine the same topology in U).
16 Yu.G. Reshetnyak

2.3. The Concept of a Shortest Curve. Suppose we are given a metric space
(M, pl. A simple arc L in the space M is called a shortest curve if its length is
equal to the distance between its end-points A and B,
siLl = p(A, B).
The concept introduced in the theory described here is one of the fundamental
concepts. Let us state some theorems that establish the conditions that guaran-
tee the existence of a shortest curve joining two arbitrary points of the given
metric space (M, pl.
Theorem 2.3.1. Let (M, p) be a locally compact metric space with intrinsic
metric. Then for any point X of the space M we can find a neighbourhood V of X
such that for any two points Y, Z E V there is a shortest curve joining these points.
Theorem 2.3.2. Let (M, p) be a locally compact metric space with intrinsic
metric. If M is complete, then for any two points X, Y E M there is a· shortest
curve joining these points.
It is obvious that a rectilinear interval on the plane 1E2 (and more generally in
the space IE" for any n ~ 1) is a shortest curve. Let M be the sphere S(a, R) in the
space IE" endowed with the metric induced from IE". Then any arc of a great circle
on S(a, R) whose length does not exceed nR is a shortest curve.

2.4. The Operation of Cutting of a Space with Intrinsic Metric. In the

investigation of two-dimensional manifolds of bounded curvature a method
proposed by A.D. Aleksandrov turns out to be useful; this is based on a trans-
formation of the manifolds under consideration by successive application of the
operations of cutting and pasting. The definition of the operation of pasting will
be given below. Here we define the operation of cutting. Formally its definition
is somewhat simpler and can be stated in a very general form.
Let (M, p) be an arbitrary metric space with intrinsic metric, and A a non-
empty closed set in M.
Let U be an arbitrary connected component of the set M\A, and Pu the
induced intrinsic metric in U. We thus obtain a collection of metric spaces
(U, Pu), where U is a connected component of M\A. To each of these spaces we
adjoin new elements that can be regarded as points lying on the bank of the cut.
For this we first construct the completion (D, Pu) of the space (U, Pu). However,
D may contain points that must be regarded as superfluous in the given case.
Suppose that X E D\ U, and that (Xm ), m = 1,2, ... , is an arbitrary sequence of
points of U that converge to the point X. Since p(X, Y) ~ Pu(X, Y) for any
X, Y E U, in the space (M, p) this sequence is fundamental. There is no need to
assume that the space (M, p) is complete, and so the sequence (Xm) in the origi-
nal space (M, p) may not have a limit. If, however, the limit lim Xm in the space
m-+oo
M exists (in this case it belongs to A), then we can call the point X E D\ U an
admissible point of D. We shall also regard any point X E U as admissible. Let
a be the totality of all admissible points of the set D. As a result we obtain a
 I. Two-Dimensional Manifolds of Bounded Curvature 17

metric space (0, Pu). The metric of this space is intrinsic. Carrying out a similar
procedure over each connected component of the set M\A, we obtain a collec-
tion of spaces with intrinsic metric. We shall say that this collection arises as a
result of cutting the space M with respect to the set A, or cutting out the set A
from M. (If A has internal points, then the term "cutting out" seems more
natural in the given context.) Let us take arbitrarily a point X E O. Let (Xm ),
m = 1, 2, ... , be a sequence of points of the set U such that Pu(X, Xm) -+ 0 as
m -+ 00. This sequence will have a limit in the space M. The value of this limit
does not depend on the choice of the sequence (Xm). We denote it by n(X). If
X E U it is obvious that n(X) = X. If X 1/ U, then n(X) E A. We thus obtain a
map n: 0 -+ M. This map is continuous. A point X E A is the image of some
point X E 0 with respect to the map n if and only if there is a sequence (Xm)'
m = 1,2, ... , such that Xm E U for each m and Xm -+ X in the space M, and the
sequence (Xm) is fundamental with respect to the metric Pu in U. If X satisfies
this condition, we shall say that X is an accessible point of the set A from M\A.
In the general case the construction we have described can be reduced to
objects of a rather pathological nature.
Let us give some simple examples. Let M be the plane 1E2, endowed with its
natural metric, that is, p(x, y) is the length of the interval with ends x and y for
x, y E 1E2. Let us assume that A consists of a single point P. We put U = 1E2\ {Pl.
It is not difficult to see that Pu(X, Y) = IXYI for any X, Y E U. In the given case
the set 0\ U = n- 1(p). Let (Xv), (y"), v = 1,2, ... , be two arbitrary sequences of
points of U converging to points P', P" of the set 0\ U. These sequences on the
plane 1E2 converge to the point P. Since pu(Xv , y") = IXv Y"I-+ 0 as v -+ 00, we
have PU(P', PII) = 0 and so we deduce that the set n- 1(p) consists ofthe unique
point P'. For any point X =f: P we have Pu(X, Pi) = IXPI. Consequently we
deduce that the metric space (0. Pu) is isometric to 1E2. It is naturally identified
with the plane 1E2. Thus we deduce that the plane cannot be cut with respect to
a one-point set.
We now consider the case when the set A is an interval [PQ], P =f: Q (Fig. 1).
The line PQ splits 1E2 into two half-planes, which we denote by IE~ and IE:. In the
given case the situation is such that for any internal point X of the interval [PQ]
the set n- 1 (X) consists of exactly two elements X' and X". Here X' is the limit
of a sequence (Xv), v = 1,2, ... , where XII E IE! for all v, and X" is the limit of a
sequence (y"), v = 1, 2, ... , where Y" E IE! for each v. We omit the proof ofthis
simple fact. It is natural to assume that X' belongs to the upper and X" to the
lower bank of the cut of the plane by the interval [PQ]. If X is one of the ends
of [PQ], then n- 1 (X) consists of a unique point. The points that belong to the
upper bank of the cut, if we add to them pi = n- 1(p) and Q' = n-1 (Q), form a

Fig. 1
18 Yu.G. Reshetnyak

Fig. 2

simple are, which is mapped one-to-one onto [PQ] by the map 7t. In the same
way the points of the lower bank, if we again add to them the points P and Q,
form a simple are, which is also mapped one-to-one onto [PQ] by the map 7t.
If A is the union of two intersecting intervals [PQ] and [RS] (Fig. 2) and T
is their point of intersection, then in this case for the point T the set 7t- 1 (T)
consists of four elements. In fact, if the sequence (Xy), v = 1, 2, ... , of points of
the set U = 1E2\A, all of whose points lie in one of the four quadrants into which
PQ and RS divide the plane, converges to T, then it is convergent in (0, Pu), as
we can easily see, and its limit is a point belonging to 7t- 1 (T). Let (Xy), (Y,,),
v = 1, 2, ... , be sequences convergent to T, where all points of each of them
lie in one of the four quadrants mentioned. If Xm , m = 1, 2, ... , and Ym , m =
1,2, ... , lie in the same one of the given angles, then in the space (0, Pu) they
converge to the same element of the set 7t- 1 (T). If these sequences lie in different
angles, then their limits are different points of the set 7t- 1 (T). It remains to
observe that if the sequence (Xm), m = 1,2, ... , converges to the point T, and Xm
skips from one angle to another for arbitrarily large m, then in the space (0, Pu)
this sequence is not convergent.

§ 3. Two-Dimensional Manifolds with Intrinsic Metric

3.1. Definition. Triangulation of a Manifold. We first give a definition of

what such a two-dimensional manifold is.
We recall that a topological space 91 is called a Hausdorff space iffor any two
distinct points X and Y of it we can find a neighbourhood U of X and a
neighbourhood Vof Y whose intersection is empty.
Let 91 be a topological space and ~ a set of open subsets of 91. We say that
~ is a base of 91 if any open set U of 91 can be represented as the union of some
set of sets belonging to ~.
A neighbourhood of a point X of the topological space 91 is any open set of
this space containing X.
The family U of neighbourhoods of a point P of the space 91 is called a basis
of the set of neighbourhoods of P if for any neighbourhood U of P there is a
 I. Two-Dimensional Manifolds of Bounded Curvature 19

V E U such that U ::::> V. The topology of 9t is completely determined if for any

point of it we find a basis of the set of neighbourhoods of this point.
Let us introduce the following auxiliary concept. We shall call a subset G of
the plane 1R2 a two-dimensional standard domain if G is either the disc
B(O, 1) = {(x, y) E 1R21x 2 + y2 < I},
or the half-disc
B+(O, 1) = {(x, y) E 1R21x 2 + y2 < 1, y ~ O}.
The point 0 is called the centre of the two-dimensional standard domain. We
note that in the case when the two-dimensional standard domain G is a half-
disc, all the points of the diameter {(x, y)ly = 0, -1 < x < I} belong to G.
The topological space 9t is called a two-dimensional manifold with boundary
if it is a Hausdorff space, it has a denumerable basis, and for any point X
of it there is a neighbourhood U of X homeomorphic to the standard two-
dimensional domain G and a topological map cp of U onto G under which the
centre of G corresponds to X. We shall call this neighbourhood U of X a special
neighbourhood of X, and the homeomorphism cp a special coordinate system
corresponding to the neighbourhood U and the point X.
Let M be a two-dimensional manifold with boundary. We shall call a point
X E M an interior point if it has a special neighbourhood homeomorphic to a
disc. Otherwise X is called a boundary point of M. By Brouwer's theorem on the
invariance of open sets with respect to topological maps, the same point X E M
cannot be simultaneously an interior point and a boundary point of M. The
totality of all boundary points of M is called its boundary and denoted by aM.
We assume that X is an interior point of a manifold M with boundary, and
U an arbitrary special neighbourhood of X. Then, obviously, any point Y E U
is also an interior point of the manifold. Hence it follows that the totality of all
interior points of M is an open subset of M, and so aM is a closed subset of M.
Let X be a boundary point of M, let U be any special neighbourhood of it,
and let cp: U -+ B+(O, 1) be the corresponding special coordinate system. Obvi-
ously, each point Y E U whose image cp(Y) belongs to the bounding diameter of
the half-disc B+(O, 1) is a boundary point.
The boundary of an arbitrary two-dimensional manifold, if it is not empty,
consists of at most a denumerable set of connected components, each of which
is homeomorphic to either a circle or a straight line. In particular, if the mani-
fold M is compact, then its boundary consists of finitely many (possibly zero)
connected components, each of which is homeomorphic to a circle.
Numerous examples of two-dimensional manifolds are known (Fig. 3). A
sphere in three-dimensional space, the surface of an ellipsoid, hyperboloids of
two sheets and one sheet, a paraboloid, the surface of an infinite circular cy-
linder in three-dimensional space - all these are two-dimensional manifolds.
At the same time a right circular cone, that is, a surface described by a straight
line under rotation around an axis intersecting it, is not a manifold. The ver-
tex of the cone - the point of intersection of its generators with the axis of
20 Yu.O. Reshetnyak

Fig. 3

rotation - does not have a neighbourhood that is homeomorphic to a standard

two-dimensional domain. In exactly the same way, a set that is the union of
two intersecting planes is not a two-dimensional manifold. In this case the
singular points, which do not have a neighbourhood homeomorphic to a two-
dimensional standard domain, lie on the line of intersection of the given
planes.
A hemisphere and the lateral surface of a finite right circular cylinder give
examples of two-dimensional manifolds with boundary.
Let us consider the following example. Let H be the half-plane {(Xl' Xl) E
~llxl > O}, and 10 the interval (-1,1) lying on the line Xl = O. The set H u 10,
endowed with the topology induced from ~2, is a two-dimensional manifold
with boundary. Its boundary is the interval 10 , At the same time, if we adjoin
to H the closed interval [ -1, 1] = 1, then the resulting set H.u 1 is not a two-
dimensional manifold, since the points ( -1, 0) and (1, 0) do not have neighbour-
hoods in it that are homeomorphic to a two-dimensional standard domain.
One of the fundamental operations by means of which from some two-
dimensional manifolds it is possible to obtain others is the pasting operation.
Applications of it in topology are well known. In the theory of two-dimensional
manifolds of bounded curvature this operation forms the ,basis of the method of
cutting and pasting developed by A.D. Aleksandrov, which we shall discuss
later.
If a manifold M with boundary is connected, then it is linearly connected,
that is, for any two points X and Y of it there is a simple arc L c: M joining X
and Y. All the manifolds considered later are assumed to be connected unless we
explicitly say otherwise. In the questions studied here this restriction turns out
to be immaterial.
Henceforth M denotes an arbitrary two-dimensional manifold with boundary.
A set G in M will be called a simple domain if G is homeomorphic to a closed
disc on the plane ~2.
 I. Two-Dimensional Manifolds of Bounded Curvature 21

A topological triangle in M is any simple domain T c M for which there are

three distinct points Xl' X 2 , X3lying on the boundary of T. The points Xl' X 2 ,
X 3 are called the vertices of the triangle T. They split the boundary of T into
simple arcs Ll = [X 2X 3], L2 = [X3X l ], L3 = [X1 X 2 ], the pairs of which do
not have any points in common except the ends. The arcs L l , L 2, L3 are called
the sides of the topological triangle T.
A triangulation of a two-dimensional manifold M with boundary is any
finite or denumerable set K of topological triangles that satisfies the following
conditions.
I. Any point X E M lies in at least one of the triangles T belonging to K.
II. Each point X E M has a neighbourhood that intersects only finitely many
triangles T E K.
III. The intersection of any two distinct topological triangles T, Til E K is
either empty, or consists of a unique point that is the vertex of each of them, or
it is a simple arc that is a side of each of the given triangles.
The topological triangles that form the triangulation K are called the faces of
K, their sides are called the edges of K, and their vertices are called the vertices
ofK.

Theorem 3.1.1. (Rado's theorem, Rado (1925». Any two-dimensional manifold

with boundary admits a triangulation.

The proof of Theorem 3.1.1 is given in Kuratowski (1968) and Rado (1925),
for example. It is rather cumbersome and relies on the following strengthened
form of Jordan's theorem.

Theorem 3.1.2 (Schoenflies's theorem). Let r be a simple closed curve on the

plane 1R2, and let D be the bounded planar domain whose boundary is r. Then the
set D uris homeomorphic to the closed disc B(O, 1) = {(x, y) Ix 2 + y2 ~ I} on the
plane 1R2.

Later we shall study simple arcs in two-dimensional manifolds. In particular,

it is necessary to make precise the concept of a side of a simple arc (see 3.4 later).
For this we require the following assertion.

Theorem 3.1.3. Let M be a two-dimensional manifold with boundary, and L a

simple arc in M not containing boundary points of M. Then in M there is an open
set U containing L and homeomorphic to the disc B(O, 1) on the plane 1R2.

We assume that M is a compact two-dimensional manifold with boundary

(the latter, in particular, may be an empty set). From condition II in the defini-
tion of triangulation, by a theorem of Borel it follows that any triangulation of
M is based on finitely many triangles. Let no be the number of vertices, n l the
number of edges, and n 2 the number of faces of the triangulation K of the
manifold M. The quantity
22 Yu.G. Reshetnyak

does not depend on the choice of the triangulation of M and is called its Euler
characteristic. It is a topological invariant of M. If M is a manifold homeo-
morphic to a two-dimensional sphere S2 in the space 1E 3 , then X(M) = 2. If Mis
homeomorphic to the closed disc B(O, 1) in 1R2, then X(M) = 1.

3.2. Pasting of Two-Dimensional Manifolds with Intrinsic Metric. One of the

main methods used in topology in the study of two-dimensional manifolds
consists in the systematic application of the operations of cutting and pasting.
As Aleksandrov showed, this method can also be applied successfully in the
study of two-dimensional manifolds of bounded curvature. Its effectiveness in
the given case is connected with the fact that the conditions that must be satis-
fied, in order that as a result of applying the operations of cutting and pasting
to manifolds of bounded curvature we again obtain manifolds of bounded cur-
vature, have sufficiently general character. For Riemannian manifolds, for ex-
ample, similar conditions are significantly more restrictive, which makes the
application of the operations to Riemannian manifolds less useful.
The operation of pasting has quite a simple intuitive sense, although its for-
mal definition, which we give later, is rather cumbersome. The awkwardness of
the definition is caused by the desire to describe as far as possible a more general
situation. We first consider some examples.
It is well known that the surface of a cube can be pasted from some cross-
shaped polygon T. Figure 4 shows how this can be done. The perimeter of the
polygon splits into segments. Those that are pasted together in the construction
of the cube are denoted by the same letters. The process of pasting the cube from
the polygon T is represented on the right of Fig. 4. The segments denoted by one
letter are combined so that the arrows marked on each of them are combined.
As a result of the pasting we obtain some surface Q. On it we define the intrinsic
metric induced by the metric of the space. The distance between points in this
metric is equal to the greatest lower bound of the lengths of the curves on the
N

L It

K P
K P
\

M Q
M Q

Fig. 4
 I. Two-Dimensional Manifolds of Bounded Curvature 23

surface Q that join these points. We shall see how we can define this metric, start-
ing from the polygon T, the development of the surface Q. Any curve on Q under
the transition from Q to the polygon T splits into separate pieces and its length
is equal to the sum of the lengths of these pieces. The number of such pieces may
tum out to be infinite. In the definition of the metric in the given case it is possi-
ble, as we can easily show, to restrict ourselves to a consideration of such curves,
which on transition from Q to T split into finitely many arcs lying on T. The
length of any such curve in space is equal to the sum of the lengths of these arcs.
Figure 5 shows a collection of planar domains, by the pasting of which we
can obtain the surface of a right circular cylinder in space. One of these domains
is a rectangle of height h and base length 2nr, and the other two are circles of
radius r. The boundary of each of the given domains is represented as the union
of finitely many simple arcs. Arcs denoted by the same letters have equal lengths
and the cylinder is obtained by pasting together such arcs.
Let H be an infinite strip on the plane, bounded by two parallel lines at a
distance 2k > 0 from each other. Rolling this strip in space so as to combine
points of the boundary lying on lines perpendicular to the boundary lines (Fig.
6), we obtain a surface isometric to the lateral surface of an infinite right circular
cylinder. We note that the boundary of the strip H can be pasted so as to
combine points of the boundary obtained in the intersection of the boundary H
by a family of parallel lines not perpendicular to the boundary lines (see Fig. 7).

,QsrQu u
T U

T
P D3 P P

5
,-- ....

R 5
R
Fig. 5

IIII
---
H

l\ S\ \
--

G~~() () ()--~ H

Fig. 6 Fig. 7
24 Yu.G. Reshetnyak

X A

~H A' X'

Fig. 8

~ ex: ~
Fig. 9

In this case the pasting can also be brought about in practice by winding H as
a band infinite on both sides onto a suitable right circular cylinder. The bound-
ary of the strip H goes over to a spiral curve on the surface of the cylinder.
Figure 8 presents another way of pasting together the edges of a strip H
bounded by two parallel lines on the plane. We choose a point 0 lying in H at
the same distance from the boundary lines 11 and ' 2. To any point X E '1 there
corresponds a point X' symmetrical to X with respect to 0, and these points X
and X' are combined as a result of the pasting. In the given case the pasting
cannot be realized by a bending of the strip in space; in any case, it is not
obvious how this could be done in a sufficiently simple way (if we cut out from
the strip H a narrow band containing the segment AA' perpendicular to 11 and
12 and passing through 0, then the pasting can be carried out in practice, and as
a result we obtain the well-known model of a Mobius band (see Fig. 9), but it is
not clear how to realize the pasting together of the whole strip). In this connec-
tion there arises the necessity of making more precise the procedure of pasting
together in the given case. This can be done as follows. For an arbitrary point
X E H we define an object p(X). If X is an interior point of H, then p(X) = X. If
X belongs to the boundary of H, then p(X) is the pair (X, X'), where X' is the
point symmetrical to X with respect to O. The totality of all p(X), where X E H,
is denoted by M. We have a map X E H 1--+ p(X) E M. In the set M we define a
metric p. This metric is constructed in the same way as the intrinsic metric of the
surface of a cube is defined with respect to its development (that is, a cross-
shaped polygon T, by the pasting of which the surface of the cube is obtained).
Instead of curves it turns out to be convenient to consider parametrized curves.
We shall say that a map x: [a, b] -+ M is an admissible path in M if we can find
a finite collection of paths going into the strip H, ei: [ai-1, a;] -+ H, i = 1, 2, ... ,
m, where ao = a < a1 < ... < alll - 1 < alii = b, such that p[ei(t)] = x(t) for each
t E [ai-1, a;], where i = 1, 2, ... , m. In particular, the equalities p[ei(ai)] =
p[ei+1 (a i )] hold. We shall call the sum ofthe lengths ofthe paths ei' i = 1,2, ... ,
m, the length of the path x. For a given admissible path x: [a, b] -+ M there can
 I. Two-Dimensional Manifolds of Bounded Curvature 25

exist different collections of paths ~i satisfying the conditions mentioned above,

but it is easy to show that the sum of their lengths does not depend on the choice
of such a collection. We shall say that an admissible path x: [a, b] -+ M in the
set M joins the points P, Q E M if x(a) = P, x(b) = Q. We denote the greatest
lower bound of the lengths of admissible paths in M joining the points P, Q E M
by p(P, Q). Thus, a metric is defined on the set M. This metric is intrinsic. Any
point P E M has a neighbourhood that is isometric to a disc on the plane in the
metric p.
Let us prove the last assertion. Let P be an arbitrary point of M. We first
assume that P is an interior point of the strip H. Then there is a ~ > 0 such that
the disc B(P, ~) c H. We show that the disc U = B(P, ~/2) is the desired neigh-
bourhood of the point P. Any path in H joining an arbitrary point X E U to a
point lying outside the disc B(P, ~) has length at least ~/2. Hence it follows that
any admissible path in M joining two arbitrary points X, Y E U and not con-
tained in the disc B(P, ~) has length at least ~. Consequently, in the definition of
the distance between the points X, Y E U it is sufficient to consider only those
paths that are contained in the disc B(P, ~). Hence it is obvious that the distance
between the points X, Y E U is equal to the length of the segment joining the
points X and Y, that is, the metric p in U coincides with the usual Euclidean
metric. In the case when P = (Q, Q'), where Q E 11 and Q' E 12 are points on the
edge of the strip symmetrical with respect to the point 0, the desired neighbour-
hood is obtained as follows. Let 2k > 0 be the distance between the lines 11 and
12 • For t E (0, k) we put B+(t) = B(Q, t) n H, B-(t) = B(Q', t) n H and suppose
that U(t) = p[B+(t) u B-(t)]. The set U(t) is obtained by pasting together the
half-discs B+(t) and B-(t) along their diameters. It is easy to establish that in the
given case U(t/2) is the required neighbourhood ofthe point P.
Thus from the strip H we obtain a metric space M. This space is complete,
and like the surface of a cylinder it is locally Euclidean. We shall call it a Mobius
surface.
The operation of cutting is in some sense the opposite of the operation of
pasting. Thus, for example, cutting a cube (Fig. 4) along its edges, denoted by the
letters K, L, M, N, 0, P, Q, and straightening the surface of the cube into a plane,
we obtain the cross-shaped polygon from whose pasting the cube was obtained.
Cutting the surface of a finite right circular cylinder along the base circles (Fig.
5), we obtain a collection of three surfaces: two discs and the lateral surface of
the cylinder. Cutting the latter along a generator and straightening into a plane,
we obtain the same collection of surfaces by pasting which (Fig. 8) we obtain the
surface of the cylinder, and so on.
We now give the necessary formal definitions.
Suppose we are given a finite or denumerable set (Dk ), k = 1, 2, ... , of two-
dimensional manifolds with boundary, each of which is endowed with the in-
trinsic metric, and for each k the boundary of the manifold Dk is locally rectifi-
able, that is, any simple arc contained in fJDk is rectifiable. Let Pk be the metric,
specified in Dk • From the given collection of manifolds we can obtain a new
manifold with boundary by pasting them together along simple curves, each of
26 Yu.G. Reshetnyak

which lies on the boundary of one of the given manifolds. Let us describe the
pasting rule in detail. We assume that pairs of the sets (D,,), k = 1,2, ... , do not
have elements in common. We introduce the following notation. We put

A= U
11:=1.2....
Db aA = U aD".
"

In the set A we introduce a topology by the stipulation that a set U c A is open

in A if and only if for each k = 1, 2, ... the intersection U n D" is an open set in
Dk • It is not difficult to see that the set A, endowed with such a topology, is a
two-dimensional manifold with boundary and aA is its boundary. Thus, instead
of a collection of manifolds we obtain one manifold; it is true that it consists of
separate pieces, not connected with each other. The pasting rule is defined in the
following way. We first specify a no more than denumerable set R of simple
curves, each of which is contained in the set aA. We assume that any two
different curves of R have no points in common except the ends. The curves
belonging to R are those along which the pasting is carried out. We then specify
the order in which the curves of R are pasted together. For this the set of curves
R is split into pairs (L j , La and for each of them there is specified a topological
map ({Jj: L j -+ L;. The following conditions must be satisfied:
1) the union of the curves constituting R is a closed subset of A;
2) for each i the curves L j and L; are different and any curve K E R belongs
to one and only one of the pairs (Ljo L;);
3) for each i, ({Jj is a map of L j onto L;;
4) any simple arc is transformed by the map ({Jj into an arc of the same length.
In the manifold A we introduce a relation between its elements, denoted by
the symbol "". Let X and Y be two arbitrary points of A. If one of them does
not belong to any of the curves of the set R, we suppose that X ,... Y if and only
if X = Y. If one of the given points X and Y lies on a curve of the set R, we shall
suppose that X ,... Y if and only if either X = Y or Y can be obtained from X in
finitely many steps by applying the maps ({Jj and ({Jj-1. It is easy to verify that the
relation ,... introduced by the method described is an equivalence relation (that
is, it is reflexive, symmetric and transitive). The last condition that the pasting
rule must satisfy is:
5) for any point X E A the set of all YEA such that X ,... Y is finite.
Figure 10 (A, B, C) represents some examples that illustrate the general
scheme described here.
We denote by D the set obtained from A by identifying any two points
X, YEA such that X ,... Y. Formally the elements of the set D are equivalence
classes in the set A with respect to the relation "". Let p be a canonical map of
A onto D. For an arbitrary X E A, p(X) is the element of D generated by this X,
that is, p(X) is the totality of all YEA such that X ,... Y. The topology in D is
defined by the stipulation that a set U c D is open if and only if p-1(U) is an
open set in A.
It is easy to verify that the topological space D constructed as described
above is a two-dimensional manifold. We make the following remarks about
 I. Two-Dimensional Manifolds of Bounded Curvature 27

Fig. 10

Fig. 11

how D is constructed. Let us take an arbitrary point XED. Let x be any point
of the manifold L1 such that p(x) = X. If x does not belong to one of the curves
of the set R, then p -1 (x) consists of a unique element - the given point x, and the
map p maps a neighbourhood U of the point x one-to-one onto a neighbour-
hood of the point X. We suppose that x E p-1(X) is an internal point of some
curve L E R (see Fig. 11). Then p-1(X) consists of two distinct elements x E L
and x' E L', where the curves Land L' are identified according to the given
pasting rule. In the given case the point X has a neighbourhood obtained by
28 Yu.G. Reshetnyak

pasting together some neighbourhood V of the point x and some neighbourhood

V' of the point x'. Here V and V' can be chosen so that each ofthem is homeo-
morphic to a half-disc and the pasting is carried out along the boundaries of
these neighbourhoods. Finally we consider the case when some point x E p-l(X)
is an end of one of the curves L E R. The set p-l(X) is finite. Let Xl' X2' ... , Xm
be all its elements. The points Xi are all ends of curves belonging to R. In the
given case we find neighbourhoods Vi of the points Xi' i = 1, 2, ... , m, such that
each of them is homeomorphic to a half-disc and for the point Xi the boundary
of the neighbourhood Vi is divided into two arcs Li and L;. The numbering of
the points that constitute p-l(X) and the neighbourhoods Vi can be chosen so
that under the pasting by means of which D is obtained from LJ the neighbour-
hoods Vl and V 2 are pasted to each other along the arcs L~ and L 2 , V 2 and V3
along the arcs L~ and L 3 , and so on, ending with Vm- l and Vm along the arcs
L~-l and Lm. Then two cases are possible. The first possibility is that the arc L~
is not contained in one of the curves belonging to R, and then also the arc Ll is
not contained in one of the curves of R. In this case the point X (Fig. 12) is a
boundary point of the manifold D, and the neighbourhoods Vi> pasted together,
form a neighbourhood of the point X homeomorphic to a half-disc (see Fig. 12,
where m = 4). The other possibility is that the neighbourhoods Vl and Vm are

L'D
pasted together along the arcs Ll and L~ (Fig. 13, where m = 3). In this case X

6:....
1

Llf '
"L,
""

Fig. 12

Fig. 13
 I. Two-Dimensional Manifolds of Bounded Curvature 29

is an interior point of the manifold D, and by pasting together the neighbour-

hoods Ui we obtain a neighbourhood of the point X homeomorphic to a disc.
The case m = 1 is allowed by the construction. If this holds, then the arcs Ll and
L'l that constitute the boundary of the neighbourhood Ul are pasted together.
All the constructions described above have a purely topological character.
(We observe that condition 4 of the equality of lengths of pasted arcs has not
been used.) Questions relating to the metric have not yet been considered. We
now show how we can introduce an intrinsic metric in the manifold D. In
principle the manifold D, constructed as we have described, may consist of
separate pieces in no way joined to each other, and in this case an intrinsic
metric in D cannot be defined. We shall assume henceforth that the manifold D
is connected.
Let X and Y be two arbitrary points ofthe manifold D. The sequence Xl' Yl'
X2' Y2, ... , X n , Yn of points of the manifold LJ will be called a chain joining the
points X and Y if P(Xi) = X, P(Yn) = Y, P(Yi) = P(Xi+l) when i = 1, 2, ... , n - 1,
and for each i = 1, 2, ... , n the points Xi and Yi belong to the same manifold D.,.
We shall denote the greatest lower bound of the sums
n
L P•• (Xi' Yi)
i=l

on the totality of all chains in the manifold LJ joining the points X and Y by the
symbol p(X, Y). It is easy to verify that the function p(X, Y) of a pair of points
of D defined in this way is a metric. This metric is intrinsic and has the following
property. Let D~ be the totality of all interior points of the manifold D•. The map
p is one-to-one on the set D~; in this connection we shall identify any point
X E D~ with the point p(X) of the manifold D, and in accordance with this we
shall further identify D~ with p(D~). The metric induced in the domain D~ of the
metric space (D, p) coincides with the metric p•. The metric p in the manifold D,
defined as described above, is uniquely determined by this property and the
requirement that it is an intrinsic metric.
Let us make one more remark concerning condition 4 in the description of
the pasting rules. From the point of view of topology it is superfluous. In our
case the fulfilment of condition 4 is that to each of those arcs along which
pasting takes place, in the manifold obtained as a result of the pasting there
corresponds an arc of the same length. In principle the fulfilment of condition 4
need not be required in the "metrical case", which is of interest to us. But then
the application of the pasting operation will lead to manifolds of a rather patho-
logical character, which are of no interest to us.
As an example we mention a special case of the pasting construction, which
is interesting in that in many cases it enables us to reduce the case of a manifold
with boundary to the case when the boundary of the manifold is empty. Let D
be an arbitrary two-dimensional manifold with intrinsic metric, where the boun-
dary of D is not empty and any simple arc contained in aD is rectifiable. Let Do
and Dl be two different copies of D. Formally we can represent the pair of
manifolds Do and Dl as the direct product of D and a topological space consist-
30 Yu.G. Reshetnyak

ing of two elements - the numbers 0 and 1. The elements of Do are pairs of the
form (x, 0), where xED, and the elements of Dl are pairs of the form (x, 1),
where XED. The boundary of D consists of a no more than denumerable set of
simple curves (Km), m = 1,2, ... Let K~ be the set of all points (x, 0) E Do, where
x E K m, and K~ the set of all points (x, 1) E Dl , where x E Km. Let R be the
totality of all curves K~, K~, m = 1,2, ... In R we naturally select the set of pairs
(K~, K~). For any m let CfJm be the map (x, 0) E K~ --+ (x, 1) E K~. For each m the
map is topological. It is not difficult to see that conditions 1-5, which the
pasting rule must satisfy, are all fulfilled in the given case. Identifying the points
corresponding to each other under the map CfJm' m = 1, 2, ... , we consequently
obtain a two-dimensional manifold D. We shall call it the twice covered manifold
D.1t is not difficult to see that the boundary of D is empty. As a result of pasting
together the curves K~ and K~ we obtain a curve Km c D.
The curves Km , m = 1, 2, ... , split D into two domains D' and D", each of
which is homeomorphic to D. The set U Km is the common boundary of these
domains. If we introduce in D' and D" the metrics induced from D, then each of
the domains D' and D" will also be isometric to the interior of D.

3.3. Cutting of Manifolds. A formal definition, moreover in a very general

situation, was given in 2.4. Here we just make some remarks related to the case
of two-dimensional manifolds.
In the case when the space (M, p) with intrinsic metric is a two-dimensional
manifold, and the set A is constructed rather simply in the topological respect,
we can determine a natural condition for the accessibility of a boundary point
of the set A. Let us introduce an auxiliary concept.
Let U be a connected open subset of the metric space (M, p) with intrinsic
metric. We introduce in U the induced intrinsic metric Pu. (We recall that by
definition Pu(X, Y), where X, Y E U, is the greatest lower bound of lengths of
curves joining the points X and Y and lying in the domain U.) The greatest
lower bound of Pu(X, Y), where X, Y E U, is called the internal diameter of the
set U and denoted henceforth by A(U).
Theorem 3.3.1. Let (M, p) be a two-dimensional manifold with intrinsic metric.
We assume that A c M is a closed set such that A "# M and the boundary of A
can be represented as the union of a finite or denumerable set of simple arcs L k ,
k = 1,2, ... , so that any two of them do not have points in common other than the
ends and any compact subset E c M intersects finitely many arcs L k • Let X be an
arbitrary boundary point of A. If for any e > 0 we can find a neighbourhood Vof
the point X such that the internal diameter of each component of the set V\A is
less than e, then X is an accessible point.
In fact, we assume that the set A satisfies the conditions of the theorem. We
take arbitrarily a boundary point X of the set A. We represent the boundary of
A as the union of a no more than denumerable set of simple arcs satisfying
the conditions of the theorem. Let L l , L 2 , ..• , Lm be those arcs that contain
the point X. Then there is a neighbourhood VI of X and a homeomorphism
 I. Two-Dimensional Manifolds of Bounded Curvature 31

qJ: VI -+ 1R2 such that qJ(V) is the disc B(O, 1), the point X is transformed by qJ to
the centre of B(O, 1), and the intervals L j n VI of the arcs L j contained in VI are
taken by qJ into radii of the disc. The existence of a neighbourhood VI and a
homeomorphism qJ satisfying all these conditions is easily established by means
of Schoenflies's theorem. The arcs VI n L j split VI into sectors that are transformed
under the map qJ into corresponding sectors of B(O, 1). Each of these sectors is
contained either in the set A or in M\A. Let G1 , G2 , ••• , G1 be those of them
contained in M\A. Let (Xy), v = 1,2, ... , be an arbitrary sequence of points lying
in the set Gj that converges to a point X. The set Gj is connected, and so it is

°
contained in some connected component of M\A. Let Uo be this component.
We specify /:: > arbitrarily and find with respect to it a neighbourhood W of the
point X such that any connected component of W\A has an interior diameter

° °
less than /:: (we assume that the point X E aA is such that there is such a neigh-
bourhood for it, whatever /:: > is). We put v., = qJ -1 [B(O, ~)], where < ~ ~ 1.
We fix the value ~ E (0, 1) such that v., c W. The intersection v., n Gj is con-
nected, since it is transferred to some sector of the disc B(O, ~) by the homeo-
morphism qJ. The set v., n Gj is contained in some connected component of W\A.
Let Wo be this component. We find Vo such that Xy E v., when v ~ Vo. For any
°
v ~ vo, J.l ~ Vo we have Pvo(x y , xI') ~ Pwo(x y , xI') < /: and since /:: > is arbitrary,
we have thus proved that the sequence (xv) is fundamental in the space (Uo, PuO>-
Hence it follows that the point X is accessible, which we needed to prove.
We note that, as is easy to see, each of the sectors Gj that figure in the
argument advanced above determines exactly one point in the collection of
metric spaces obtained by cutting the manifold M along the set A, and all these
points are pairwise distinct.
The operation of cutting is in some sense the reverse of the pasting operation.
Namely, we assume that the manifold M with intrinsic metric P is obtained by
pasting together the manifolds D". Let A be the closed set in M consisting of all
points of M that correspond to points lying on these curves along which the
pasting is carried out. Cutting the space (M, p) along the given set A, we obtain
the original collection of two-dimensional manifolds.

3.4. A Side of a Simple Arc in a Two-Dimensional Manifold. Let M be an

arbitrary two-dimensional manifold with boundary, and L a simple arc in M.
We introduce here some concepts that enable us to give an exact meaning to
words "on a given side of L" or "on the same side of L" (see p. 21).
We require the concept of the orientation ofa simple closed curve on a plane.
It has the following intuitive meaning. A simple closed curve r is oriented if we
are given a direction for going round the curve r. According to Jordan's theo-
rem the curve r splits the plane into two domains G and G', one of which
(suppose it is G) is bounded. If on going round r in the given direction G turns
out to lie on the left of r, we shall say that the curve is positively oriented. If the
domain lies to the right of r on going round r in the given direction, we shall
say that r is negatively oriented. For example, the curve r1 in Fig. 14 is posi-
tively oriented, and r2 is negatively oriented.
32 Yu.G. Reshetnyak

Fig. 14

Fig. 15

Let K be a simple arc on the plane 1R2. We shall say that the curve K is
oriented if one of its end-points is called the beginning. Let K and L be two
oriented simple arcs with common beginning at a point A and common end B.
We assume that K and L have no other common points. Then together they
r
form a simple closed curve r. We orient so that under a motion along K the
r
point A precedes the point B. If is positively oriented, we shall say that Lis
r
situated to the left of K. If the orientation of is negative, we shall say that Lis
situated to the right of K. For example, the simple arc Ll in Fig. 15 is situated to
the left, and the simple arc L2 to the right of K. It is not difficult to see that if a
simple arc L lies to the left of K, then in turn K lies to the right of L.
What "right or left of a simple arc on a manifold" means we shall define later.
Without difficulty we can give an exact meaning to the intuitive definitions
given above in the case when the arcs under consideration are sufficiently
smooth. In the general case the reader can perceive in them a kind of vicious
circle.
We now give the exact definitions. We first make some remarks.
Suppose that z = (x, Y) E 1R2, where Izl = r =1= O. Then the polar coordinates
are defined by the conditions
x = r cos (), Y = r sin (). (3.1)
Any number () for which the equalities (3.1) are satisfied is called the polar
angle of the point z.
 I. Two-Dimensional Manifolds of Bounded Curvature 33

Let z(t}, a ~ t ~ b, be an arbitrary closed path on the plane ~2, that is, such
that z(a) = z(b). We denote by L the set of points on the plane sketched out by
the point z(t) when t runs through the interval [a, b], that is, L = z([a, b]). We
take a point c If L arbitrarily. Then we can define a continuous function <p(t),
a ~ t ~ b, such that for each t E [a, b] <p(t) is the polar angle of the vector
z(t) - c. If <P1 (t) and <P2(t) are two functions satisfying this condition, then
<P1 (t) - <P2(t) == 2nm = const, where m is an integer. The difference <p(b) - <p(a)
therefore does not depend on the choice of the function <p(t). The number
v(c, z) = [<p(b) - <p(a)]/2n is an integer. The quantity v(c, z) is called the index of
the point c with respect to the closed path z(t), a ~ t ~ b. The function c H v(c, z)
of the variable c is defined on the set ~2\L and is constant on each connected
component of it.
Let r be a simple closed curve on the plane, z(t), a ~ t ~ b, an arbitrary
parametrization of it, that is, a path in ~2 such that z(a) = z(b), and suppose that
the function z maps the half-open interval (a, b] one-to-one onto the set r. We
shall assume that the function z(t) is extended periodically with period T =
b - a onto the whole set ~. (For any t E ~ there is one and only one integer m
such that t - Tm E (a, b]. For this t we put z(t) = z(t - Tm).) Henceforth we
shall assume that this condition is automatically satisfied.
If the periodic functions Z1 with period T1 and Z2 with period T2 are parame-
trizations of a simple closed curve r, then Z2(t) = z 1 (<p(t», where <p: ~ -+ ~ is a
continuous strictly monotonic function such that if t2 - t1 = T1, then <P(t2)-
<p(td = ± T2. We shall say that the parametrizations Z1 and Z2 are oriented
in the same way if <p is an increasing function, and oriented in the opposite
way if qJ is a decreasing function. The set of all parametrizations of a simple
closed curve is split into two disjoint classes so that two parametrizations of one
class are oriented in the same way, and parametrizations belonging to different
classes are oriented in the opposite way. We shall say that the simple closed curve
r is oriented if the parametrizations of one class are called right, and the para-
metrizations of the other class are called left. Intuitively this is equivalent to the
designation of a definite direction of going round the curve r.
Let rbe an oriented simple closed curve on the plane ~2, z(t) a right parame-
trization of this curve, and T> 0 the period of the function z(t). We have a
closed path z(t), 0 ~ t ~ T, and for any point c If r there is defined a number
v(c, z). This number does not depend on the choice of the parametrization of r.
The curve r splits ~2 into domains G and G', one of which is bounded. We shall
assume that this is the domain G. For any point c E G', v(c, z) = 0, and for c E G,
v(c, z) = ± 1. The curve r is said to be positively oriented if v(c, z) = 1 for all
c E G, and negatively oriented if v(c, z) = -1 for c E G.
The following assertion is true.

Theorem 3.4.1. Let U be a domain in ~2, and <p: U -+ ~2 a topological map.

Then either for any oriented simple closed curve r c U the curve rand <p(r) are
oriented in the same way, or for any such curve the curves rand <p(T) are oriented
in the opposite way.
34 Yu.G. Reshetnyak

If the first of the two possibilities holds, we shall say that the map qJ preserves
the orientation. In the second case we shall say that qJ changes the orientation.
Let K be a simple arc on a plane. We shall say that the arc K is oriented if one
of its end-points (we denote it by A) is called the beginning and the other (we
denote it by B) is called the end ofthe arc. We shall call the parametrization z(t),
a ~ t ~ b, of the oriented simple arc K right if z(a) = A, z(b) = B, and left if
z(a) = B, z(b) = A.
Let K and L be two oriented simple arcs on a plane such that the beginning
of each of them is the point A, and the end is the point B. The curves K and L
together form a simple closed curve F. We construct a parametrization of F.
Namely, we specify a, b, c arbitrarily such that a < c < b, and let ZI(t) be a
parametrization of the arc K such that ZI (a) = A, ZI (c) = B, and Z2(t) a parame-
trization of the arc L such that Z2 (c) = B, z2(b) = A, and put z(t) = Z1 (t) when
a ~ t ~ c, and z(t) = Z2(t) when c ~ t ~ b. It is not difficult to see that all the
parametrizations of F that can be obtained in this way are of the same name.
Calling them right, we obtain an orientation of F. We shall say that Flies to the
left of K if F is positively oriented. If F is negatively oriented, we say that L is
situated to the right of K.
Thus, we have given an exact meaning to the assertion "the simple arc L lies
to the left (right)" of the simple arc K for the case of a plane. As we did above,
we can now define the concept of left and right semineighbourhood of the
oriented simple arc K on the plane.
Let us consider the case of a simple arc on an arbitrary two-dimensional
manifold. Let M be a two-dimensional manifold with boundary, and K an
arbitrary simple arc in M not containing boundary points of M. We shall call
any open set V => K such that the closure of V is homeomorphic to a closed disc
on the plane ~2 a canonical neighbourhood of the arc K (the existence of an open
set V satisfying these conditions follows immediately from Theorem 3.1.3). Let
qJ: V -+ ~2 and 1/1: V -+ ~2 be topological maps of the domain V into ~2. The
sets G = qJ(V) and H = I/I(V) are open, and (1 = 1/1 0 qJ -1 is a topological map of
G onto H. We shall say that qJ and 1/1 are ofthe same name if the map (1 preserves
the orientation, and of different names if (1 changes the orientation. The set of all
homeomorphisms of V into R2 splits into two classes in such a way that two
homeomorphisms belonging to one class have the same name, and homeomor-
phisms belonging to different classes have different names. We shall say that the
domain V is oriented, or in other words that a definite orientation of V is
specified, if the homeomorphisms of one class are called right, and homeomor-
phisms of the other class are called left.
Let VI and V 2 be two canonical neighbourhoods of a simple arc K. We
denote by V the connected component ofthe set VI n V 2 that belongs to K. We
assume that VI and V2 are oriented, and let qJ: VI -+ ~2 and 1/1: V 2 -+ ~2 be right
homeomorphisms ofthese neighbourhoods. Let G = qJ(V), H = I/I(V). Then the
homeomorphism (1 = 1/1 0 qJ -1: G -+ H is defined. The sets G and H are con-
nected. We shall say that VI and V 2 are oriented in the same way, or in other
words that their orientations are consistent, if (1 preserves the orientation. Other-
 I. Two-Dimensional Manifolds of Bounded Curvature 35

wise we shall say that VI and V 2 are oriented in the opposite way. The set of
canonical neighbourhoods of the simple arc K thus splits into two classes so
that two neighbourhoods of one class are oriented in the same way, and neigh-
bourhoods of different classes are oriented in the opposite way. We shall say
that along K there is specified a definite orientation of the manifold M if the
elements of one of these two classes are called right neighbourhoods of K (then
the elements of the other class are called left neighbourhoods of K).
Thus, let K be an oriented simple arc in a two-dimensional manifold M. We
assume that an orientation is specified along K. Let V be a right canonical
neighbourhood of K, and cp a topological map of the closure of U onto the disc
B(O, 1) such that the beginning A ofthe simple arc K is transformed by this map
into the point P = (-t, 0) E 1R2, and the end B of K is mapped into the point
Q = (t, 0), and the simple arc itself is transformed by a given homeomorphism
into the interval [PQ]. The existence of such a homeomorphism is easily estab-
lished by applying Schoenflies's theorem. Obviously we can assume that this
homeomorphism cp is right. Let L be an oriented simple arc contained in U,
where A is its beginning and B its end, and there are no other points common
to K and L. We shall say that L lies to the left (right) of the curve K if cp(L) lies
to the left (right) of the interval [PQ] on the plane. Let G c V be the open
domain bounded by the simple arcs K and L. Then we shall call G a left semi-
neighbourhood of K if L lies to the left of K, and a right semineighbourhood if L
is situated to the right of K. Let (X.), v = 1, 2, ... , be an arbitrary sequence of
points of M that converges to some interior point of K. Then there is a Vo such
that when v ~ Vo the point X. belongs to V. When v ~ Vo there is defined a point
z. = cp(X.), z. = (x., y.). We shall say that X. converges to a point Xo to the
right (left) with respect to K if there is a VI ~ Vo such that y. > 0 (y. < 0) for all
v ~ VI. It is easy to prove that the property that the sequence (X.), v = 1,2, ... ,
converges to the right (left) to an arbitrary interior point Xo does not depend
on the choice of the right neighbourhood V of K nor on the choice of the
homeomorphism cpo
Let us define what is meant by "a sequence of simple arcs converges on the
left (right) to a simple arc K on a manifold M". We first introduce a concept that
refers to plane curves. Suppose we are given simple arcs K and K., v = 1, 2, ... ,
on the plane 1R2. Then we shall say that K. converges to K as v -+ 00 ifthe arcs
K and K. admit the parametrizations X(t), a ~ t ~ b, and X.(t), a ~ t ~ b, v =
1,2, ... , such that IX.(t) - X(t)l-+ 0 as v -+ 00 uniformly on [a, b].
Let K be a simple arc in the manifold M that does not contain boundary
points of M. We assume that K itself is oriented and along it there is specified
an orientation of M. We specify arbitrarily a right neighbourhood U of the arc
K and construct a homeomorphism cp of the set U into the plane that satisfies
all the conditions listed above (that is, cp(U) is the disc B(O, 1), and cp(K) is the
interval [PQ], where P = (-t, 0) = cp(A), Q = (t, 0) = cp(B), A is the beginning
and B the end of the arc K; in addition, we require that cp is a right homeomor-
phism). Let (K.), v = 1, 2, ... , be a sequence of oriented simple arcs in M, each
of which has its beginning at A and its end at B. We shall say that the simple
36 Yu.G. Reshetnyak

arcs K. converge to the simple arc K on the left (right) if there is a number Vo
such that K. c U for all v ~ Vo and K. lies to the left (right) of K, and as v -+ 00
the arc (j)(K.) converges to the interval [PQ].

§ 4. Two-Dimensional Riemannian Geometry

4.1. Differentiable Two-Dimensional Manifolds. The concept of a two-

dimensional manifold with boundary has been defined in 3.1. Here we shall
consider only manifolds whose boundary is empty, and we shall use the term
"two-dimensional manifold" for their designation.
Let M be an arbitrary two-dimensional manifold. A chart or local coordinate
system in M is any topological map (j): U -+ 1R2, where U is an open set of the
space M. If (j): U -+ 1R2 is a chart in a two-dimensional manifold M, then V =
(j)(U) is an open set in 1R2. U is called the domain of definition of the chart (j), and
V is called its range. Suppose that p E U and (j)(p) = (t 1, t 2). The numbers t 1, t2
are called the coordinates of the point p E U with respect to the given chart.
The charts (j)1: U1 -+ 1R2, (j)2: U2 -+ 1R2 of a two-dimensional manifold Mare
said to be overlapping if U1 n U2 is not empty. In this case there are defined open
sets in 1R2: G1 = (j)1(U 1 n U2) and G2 = (j)2(U1 n U2), and topological maps (J =
(j)2 0 (j)1 1: G1 -+ G2,. = (j)1 0 (j)2 1: G2 -+ G1. Obviously (J = .-1. We take a point
p E U1 n U2 arbitrarily. Let (j)1(P) = (t1' t 2), (j)2(P) = (u 1, U2)' The numbers t 1, t2
are the coordinates ofthe point p with respect to the chart (j)1' and u 1, U2 are its
coordinates with respect to the chart CP2' We have U(t1' t 2) = (u 1, U2)' so the
function (J enables us to calculate the coordinates of a point p E U1 n U2 with
respect to the chart CP2 from its coordinates in the chart (j)1' Similarly, (t1' t 2) =
.(u 1, U2)' We shall call (J and. the transition functions for the given charts (j)1
and (j)2'
Let (j)1: U1 -+ 1R2 and (j)2: U2 -+ 1R2 be two overlapping charts in a manifold
M. We shall say that the charts (j)1 and (j)2 are smoothly compatible if the transi-
tion functions (j)1 0 (j)2 1 and (j)2 0 (j)1 1 corresponding to them have all the partial
derivatives of any order, and these derivatives are continuous.
Let m: be a set of charts of a two-dimensional manifold M. Then we shall say
that m: is an atlas of class coo if the domains of definition of the charts belonging
to the atlas m: cover M, and any two overlapping charts of m: are smoothly
compatible.
A differentiable two-dimensional manifold is a two-dimensional manifold on
which an atlas m: of class Coo is specified. We shall call the charts belonging to
the atlas m: basic charts. A chart (j): U -+ 1R2 in a manifold M, if it is smoothly
compatible with any basic chart 1/1: G -+ 1R2 such that Un G is non-empty, is
said to be admissible. The set of all admissible charts in a two-dimensional
manifold forms an atlas of class Coo.
Let us define the concept of an orientable two-dimensional manifold and the
concept of an orientation in it.
 I. Two-Dimensional Manifolds of Bounded Curvature 37

Let M be an arbitrary differentiable two-dimensional manifold and let C(Jl:

Ul -+ 1R2 and C(J2: U2 -+ 1R2 be two overlapping admissible charts in M. Let (I =
C(J2 0 C(Jll and 1: = C(Jl 0 C(Ji l = (1-1 be the transition functions corresponding to
them. If the Jacobian of the map (I is positive everywhere (obviously in this case
the Jacobian of 1: is also positive everywhere), we shall say that the given charts
are oriented in the same way. If the Jacobian of (I is negative everywhere (then the
Jacobian of 1: is negative at every point of the domain of definition), then the
charts C(Jl and C(J2 are said to be oriented in the opposite way.
A differentiable two-dimensional manifold M is said to be orientable if the set
of all its admissible charts can be split into two classes in such a way that any
two overlapping charts belonging to one class are oriented in the same way, and
charts taken from different classes are oriented in the opposite way. We say that
a definite orientation of M is specified, or in other words that M is oriented, if
all the charts belonging to one of these classes are called right, and all the charts
of the other class are called left.
Let M be a differentiable two-dimensional manifold, and GeM an open
subset of M. Then G as a topological space with the topology induced from M
is obviously a two-dimensional manifold. We assume that C(J: U -+ 1R2 is an ad-
missible chart in M such that U n G is non-empty. Then the restriction of C(J to
the set U n G is a chart in the two-dimensional manifold G. The totality of all
charts in G that can be obtained in such a way is an atlas of class Coo. We shall
say that this atlas determines in G the structure of a differentiable manifold
induced from the manifold M.
Any connected component of the two-dimensional manifold M is a two-
dimensional manifold. If M is a differentiable manifold, then by the construction
indicated above any connected component of M is turned into a differentiable
two-dimensional manifold.

4.2. The Concept of a Two-Dimensional Riemannian Manifold. From an ele-

mentary course of differential geometry the reader will be familiar with such
concepts as the first and second fundamental forms of a surface. The first funda-
mental form gives the possibility of calculating the lengths of curves lying on
a surface, and the second fundamental form characterizes the bending of the
surface in space. All the concepts relating to a surface that can be defined by
means of the first fundamental form, and those properties of a surface that can
be established by means of this fundamental form, determine the topic of the
intrinsic geometry of a surface.

The concepts of a smooth and piecewise smooth path in a differentiable mani-

fold. A two-dimensional Riemannian manifold is a differentiable manifold in
which the length of a curve is defined. If the curve is sufficiently smooth, then its
length can be calculated, starting from some differential quadratic form, as in
the case of surfaces in three-dimensional Euclidean space.
Let x(t) = (Xl (t), X2(t)), a ~ t ~ b, be a path on the plane 1R2. Then we shall
say that X is a smooth path of class C, where r ~ 1 is an integer, if the functions
38 Yu.G. Reshetnyak

Xl (t) and X2(t) have continuous derivatives d~;'l (t) and d~;'2 (t) in [a, b], and the
vector x'(t) = (x~(t), x;(t)) is non-zero for all t E [a, b]. We shall call the path
x(t), a ~ t ~ b, piecewise smooth of class cr if we can find a finite sequence of
points a = to < t I < ... < tm = b such that the restriction of X to each of the
intervals [ti-l' tais a smooth path of class C. This means that the function x(t)
in each of the intervals (t l- 1 , tl) has a continuous derivative x(r)(t) and at the
points tl the function x(r)(t) has finite limits when t tends to tl from the left or
right. Hence it follows, in particular, that the first derivative x'(t) also has limits
to the left and right at these points. For the function x'(t) these limits must be
non-zero.
Let M be a differentiable two-dimensional manifold, and e:
[a, b] -+ M a
parametrized curve in M. We assume that there is an admissible chart qJ: U -+
e
~2 such that W) E U for all t E [a, b]. We shall say that is a smooth (piecewise
smooth) path of class C if the path x(t) = qJ [e(t)], a ~ t ~ b, on the plane ~2 is
smooth (piecewise smooth) of class C. Let 1/1: V -+ ~2 be any other admissible
chart such that the point e(t) E V for all t E [a, b]. Then the function y(t) =
I/I[e(t)] is defined. We obviously have y(t) = q[x(t)], where q = qJ 0 1/1-1 is the
transition function for the given charts qJ and 1/1. The function q has all the
partial derivatives of any order, and these derivatives are continuous. Hence it
follows that y(t) is a piecewise smooth path of class C. We should observe that
the first derivatives of the components Yl (t), Y2(t) of the function yare expressed
as follows:
0Yl, (t ) +~X2
Yl, (t ) =~Xl 0Yl , (t,) iJY2, (t ) +~X2
Y2, (t ) =~Xl iJY2, (t.)
UX I UX 2 uXI uX2
Here the symbols Yl and Y2 denote the components of the function q.1f x'(t)-#
0, then since the Jacobian of the function q is non-zero it follows from the
equalities indicated that y'(t) -# o.
Definition of a two-dimensional Riemannian manifold. Let M be an arbitrary
differentiable two-dimensional manifold. Then we shall say that a Riemannian
geometry is specified in M, or briefly that M is a Riemannian manifold, if with
e:
any piecewise smooth path [a, b] -+ M of class C 1 there is associated a num-
ber l(e; a, b) such that the following conditions are satisfied.
Rl. For any c E (a, b) the following equality is satisfied:
l(e; a, b) = l(e; a, c) + l(~; c, b). (4.1)
R2. For any admissible coordinate system qJ: U -+ ~2 continuous functions
X2)' i, j = 1,2, are defined such that g12(X 1, x 2) = g21(Xl' x 2) and for any
glj(X 1 ,
X E G = qJ(U) the quadratic form
2 2
g(x; z) == L L glj(x 1 , X2)Z IZj
1=1 j=l
(4.2)
in the variables ZI' Z2 is positive definite and for any piecewise smooth path
~: [a, b] -+ M of class C 1 passing into the domain of definition of the given chart
 I. Two-Dimensional Manifolds of Bounded Curvature 39

r
the follow~ng equality is satisfied:
2 2
l(e; a, b) = L L gjj[x(t)]xat)xj(t) dt.
j=l j=l
(4.3)

e:
The quantity l(e; a, b) is called the length of the path [a, b] -+ M in the
Riemannian manifold M. The value of l(e; a, b) is completely determined by the
e
conditions Rl and R2 for any piecewise smooth path in the manifold M.
We assume that the manifold M is connected. Then for any two points
p, q E M there is a piecewise smooth path in M that joins these points. The greatest
lower bound of the lengths of such paths is denoted by PM(P, q) and is called the
distance between the points p and q in the Riemannian space M. The function
PM of a pair of points of the two-dimensional manifold M defined in this way is
a metric. We shall call it the natural metric of the given Riemannian manifold.
It is convenient to write the quadratic form (4.2) in the form

(4.4)

This notation goes back to the infinitesimal representations of the time of the
creation of mathematical analysis. We can give it a completely modern meaning
by interpreting dXj as a linear function such that for the vector Z = (Zl' Z2) we
have dxj(z) = Zj. We shall call the quadratic form (4.4) a representation of the
line element of the manifold M by means of the chart cpo We shall call the
quantity ds, that is, the square root of the quadratic form on the right-hand side
of (4.4), the line element of the Riemannian geometry in the manifold M. We
shall also call the quadratic form (4.4) the metric tensor of the Riemannian
manifold M.
The ordinary Euclidean plane is a Riemannian manifold. In the orthogonal
Cartesian coordinate system its line element is defined by ds 2 = dxf + dx~. It
is easy to construct other examples of two-dimensional Riemannian manifolds.
For example, a sphere in ~3, the surface of an infinite circular cylinder, and
the Lobachevskij plane are two-dimensional Riemannian manifolds. Any two-
dimensional surface in ~3 that satisfies the regularity conditions adopted in
differential geometry is turned into a Riemannian manifold if we introduce on it
the metric induced from 1R3.
Properties of the natural metric of a Riemannian manifold. We show that the
metric PM in a Riemannian manifold M is compatible with the topology of this
space. For this we first prove the following proposition.
Lemma 4.2.1. Let cp: U -+ 1R2 be an arbitrary admissible chart in the
Riemannian manifold M, Po E U, X o = cp(Po), V = cp(U). Let B, denote the disc
B(xo, r) on the plane ~2, and let G, = cp-l(B,). Then there are numbers b > 0 and
L < 00, L ~ 1 such that if 0 < r ,.;; ~, then
BM(po, rlL) c G, c BM(po, rL)
(the symbol BM(po, r) denotes a ball in the sense of the metric PM in the
manifold M).
40 Yu.G. Reshetnyak

Proof. Let b > 0 be such that the disc B(xo, 2b) c V. Then the closed disc
B(xo, b) is contained in V. Since the disc B(xo, b) is compact and the coefficients
of the quadratic form g(x, h) by means of which the functional l(p; a, b) is de-
fined are continuous, there is a number L < 00 such that for any vector h E 1R2
and any point x E B(xo, b) we have
1 2 2 2
L21hl ::;; g(x, h) ::;; L Ihl .

We show that the given Land b are the required quantities.

Assume that 0 < r ::;; b. We take a point p E Gr arbitrarily. Let x = cp(p),
x(t) = (1 - t)xo + tx, 0 ::;; t ::;; 1, p(t) = cp -1 [x(t)]. Suppose that x(t) is a para-
metrized interval in 1R2 with ends Xo and x. We have

PM(PO' p) ::;; l(p; 0, 1) = II J g[x(t), X'(t)] dt

::;; L II Ix'(t)1 dt = Llx - xol < Lr.

Since p E Gr is taken arbitrarily, we have thus proved that Gr c BM(po, Lr).

Let ~(t), a ::;; t ::;; b, be an arbitrary piecewise smooth path such that p(a) =

r
Po, q = p(b) ¢ G. There is a c ::;; b such that when a ::;; t ::;; c the point ~(t) E Gr
and ~(c) ¢ Gr. Then, putting x(t) = cp(~(t)), we obtain

l(~; a, b) ~ l(p; a, c) = J g[x(t), x'(t)] dt

~ ~L Iea
Ix'(t)1 dt ~ ~ Ix(c) - x(a)1 = ~.
LL
Hence it follows that PM(P, q) ~ rlL for any point q ¢ Gr. Hence, if PM(P, q) <
rlL, we have q E Gr , that is,

This proves the lemma.

Corollary. The natural metric PM of the two-dimensional Riemannian manifold
M is compatible with the natural topology of M.
In order to prove the given assertion we first need to prove that any neigh-
bourhood of an arbitrary point Po of the manifold M contains some disc
BM(po, 8). The truth of this follows from the first inclusion of the lemma. Sec-
ondly, we need to prove that any disc, in the sense of the metric PM' with centre
Po contains some neighbourhood of the point Po. This follows from the second
inclusion, and thus the corollary is proved.
The following assertion is true.
Theorem 4.2.1. The metric PM on the two-dimensional Riemannian manifold M
is the intrinsic metric. For any piecewise smooth path~: [a, b] ..... M of class C 1 in
the manifold M, l(~; a, b) is its length with respect to the metric PM'
 I. Two-Dimensional Manifolds of Bounded Curvature 41

We shall omit the proof of the theorem, as it is rather cumbersome (see

Kobayashi and Nomizu (1963), Ch. IV).
The rule for transforming the coefficients of the metric tensor on going over to
another coordinate system. In the classical handbooks on Riemannian geometry
the rule for transforming the coefficients of the metric tensor is included in the
definition of a Riemannian manifold. We show how to derive this rule, starting
from the definition given here.
Let cp: U -+ 1R2 and 1jJ: U -+ 1R2 be two admissible coordinate systems with a
common domain of definition U. Suppose that the differential quadratic forms
2 2 2 2
L L giix1,X2) dX i dx
i=l j=l
j, L L hdYl' Y2) dYk dYI'
k=l 1=1

are representations of the line element of the Riemannian manifold by means

of the charts cp and IjJ respectively. Let us explain how the functions gij and hkl
are connected. Let G1 = cp(U), G2 = IjJ(U). We take arbitrarily a point Y E G2
and a vector z #- 0 on the plane 1R2. We put y(t) = y + tz, 0 ~ t ~ u. We shall
assume that u is chosen so that x(t) E G1 for all t E [0, u]. Suppose that ~(t) =
IjJ -1 [y(t)], y(t) = cp [~(t)] = u [y(t)], where u = cp 0 IjJ -1 is the transition function
for the given charts cp and 1jJ. Obviously ~(t), 0 ~ t ~ u, is a piecewise smooth
path of class C 1 lying in the domain U. We have

l(p; 0, u) = J: J g[x(t), x'(t)] dt,

on the other hand,

l(p; 0, u) = J: Jh[y(t), y'(t)] dt.

Differentiating the given equalities with respect to u and putting u = 0, we

obtain
g[x(O), x'(O)] = h(y, z).
We have x = x(O) = r(y). The vector, = x'(O) is expressed in terms of the vector
z by the formula, = dr(y, z). Here dr(y) is a linear map - the differential of the
function r at the point y. Written in coordinates, the last equality takes the form

As a result we obtain
h(y, z) = g(x, dr(y, z)). (4.5)
In expanded form this equality is
42 Yu.G. Reshetnyak

Since the vector z is arbitrary, we conclude that

(4.6)

where x = ,(y). This is the required representation of the coefficients of the

quadratic form hkl •
From (4.5) and (4.6), we obtain in turn

g(x, z) = h(y, d,-l(y, z)), (4.7)

where ,-1 = '" 0 cp-1, Y = ,-l(X) and furthermore

(4.8)

Properties of being approximately Euclidean in the small. Let Po be an arbi-

trary point of a two-dimensional Riemannian manifold. Then in a neighbour-
hood of this point we can define an admissible chart cp: U ~ jR2 such that the
coefficients of the differential quadratic form that determines the line element
of the manifold at the point Xo = cp(Po) take the values gll (xo) = 1, g12(XO) =
g21(X O) = 0, g22(XO) = 1 so at the point Xo this quadratic form is ds 2 = dx~ +
dx~.
The coefficients gij of the metric tensor of the manifold are continuous, and
as x ~ Xo, gll (x) ~ 1, g22(X) ~ 1, g12(X) = g21 (x) ~ O. The Riemannian geome-
try determined by the line element dxf + dx~ on the plane ~2 is the usual
Euclidean geometry. Therefore the result established here can be interpreted in
the following way. In an infinitely small neighbourhood the Riemannian metric
is Euclidean.

Area as a function of sets in a Riemannian manifold. On any differentiable

two-dimensional Riemannian manifold we can define a totally additive set func-
tion S, called the area.
A set A c M will be called a Borel set if for any admissible chart cp: U ~ jR2
in the manifold M the set cp(A n U) is a Borel set.
A set A in a Riemannian manifold M is said to be bounded if the closure of A
is compact.
Let ~(M) denote the totality of all Borel sets of the two-dimensional mani-
fold M. The symbol ~o(M) will denote the totality of all bounded Borel sets
inM.
Let A be an arbitrary Borel set in the two-dimensional Riemannian manifold
M. We assume that A is contained in the domain of definition of some admissi-
ble chart cp: U ~ jR2. Let gll' g12 = g21' g22 be the coefficients ofthe differential
quadratic form ds 2 with respect to this chart. Since the quadratic form g(x, z) =
gll(X)'Z~ + 2g12(x)Zl'Z2 + g22(X)'Z~ is positive definite, the quantity g(x) =
gll(X)'g22(X) - [gdX)]2 is positive for all x = (Xl' X2) for which it is defined.
 I. Two-Dimensional Manifolds of Bounded Curvature 43

We put

S(A) = f f JgW dX 1 dX2·

'P(A)

From the classical formula for changing the variables in a mUltiple integral and
the rule for transforming the coefficients of a quadratic form on transition from
one coordinate system to another it follows that the quantity S(A) does not
depend on the choice of local coordinate system in whose domain of definition
A is contained.
If A is an arbitrary Borel set in M, then since M is a space with denumerable
base A can be represented in the form
ao
A= U Am'
m=l
(4.9)

where each of the sets Am is a Borel set and is contained in the domain of
definition of some admissible chart, and the sets Am are pairwise disjoint. We
put
ao
S(A) = L S(Am)·
m=l

It is easy to establish that the sum on the right-hand side does not depend on
the choice of the representation (4.9) of the set A, so the given definition is
reasonable.
We shall call the quantity S(A), where A is a Borel set in a Riemannian
manifold M, the area of the set A. A function of the set S defined here is com-
pletely additive, that is, for any sequence (Am), m = 1, 2, ... , of pairwise disjoint
Borel sets in M we have

The purely formal definition of the area of a set given here is justified by the
fact that in the case when the Riemannian manifold is a smooth surface in the
space 1E3 with the metric induced from 1E 3, S(A) exactly coincides with what we
call the area of a set on a surface in differential geometry. We note that other
more natural geometrical definitions of area in a Riemannian manifold lead to
the same quantity S(A).

4.3. The Curvature of a Curve in a Riemannian Manifold. Integral Curvature.

The Gauss-Bonnet Formula. In the definition of a Riemannian manifold given
in 4.2, for the coefficients gl) in the representation of the line element we assume
only that the coefficients are continuous. The main content of Riemannian
geometry is the theory of curvature of Riemannian manifolds. The latter can
be constructed only on the assumption that the functions gij have all the
partial derivatives of not less than the second order, and all these derivatives are
44 Yu.G. Reshetnyak

continuous. For simplicity we shall assume henceforth that the functions gij
belong to the class Coo, that is, each of them has all the partial derivatives of any
order and these derivatives are continuous. If this condition is satisfied, we shall
say that M is a Riemannian manifold of class Coo.
Let M be a two-dimensional Riemannian manifold of class Coo. We assume
that M is connected. We shall denote the natural metric of M by p.
Theorem 4.3.1 (Kobayashi and Nomizu (1963), Ch. IV). Any point X of a
Riemannian manifold of class Coo has a neighbourhood U such that for any two
points Y, Z E U there is a unique shortest curve of the manifold M that joins
these points. Any shortest curve in a Riemannian manifold of class Coo is a curve
of class COO.
Curvature and turn of a simple arc. Let K be a smooth simple arc of class C 2
in a two-dimensional Riemannian manifold M. We assume that the arc K is
oriented and along it there is specified a definite orientation of M. Then for each
point X of the curve K we can define the number k,(X), which we shall call the
left curvature of Kat X. We put kr(X) = -k,(X). The number kr(X) is called the
right curvature of K at X. We do not give a formal definition of k,(X). From
those properties of the curvature at a point of a curve that we shall give later
there actually follows a way of calculating the curvature (the main property is
the one contained in the Gauss-Bonnet theorem stated later). We note that if we
change the orientation of the simple arc K, then k,(X) changes sign. In the same
way, if under a fixed orientation of the simple arc we change the orientation of
the manifold M along this arc, then k,(X) is also multiplied by -1.
If K is a shortest curve, then its curvature at each point is zero.
We assume that K is an oriented simple arc of class C 2 in a Riemannian
manifold and that along K there is specified an orientation of M. Let x (s),
o ~ s ~ I, be a parametrization of K, where the parameter s is arc length. We
put

K,(K) = I k,[x(s)] ds, Kr(K) = -K,(K).

We shall call K,(K) the left turn of K, and Kr(K) the right turn of K. All that we
have said above about the behaviour of the curvature under a change of orienta-
tion of the curve K and a change of orientation of the manifold along K remains
valid in connection with a turn.
The introduction of the two quantities K,(K) and Kr(K), which differ only in
sign, may seem odd, and it is usually not done in the standard textbooks on
differential geometry. We shall define the left and right turns later for simple arcs
in an arbitrary two-dimensional manifold of bounded curvature, and there the
relation Kr(K) = - K,(K) may not be satisfied, generally speaking.
The Gaussian Curvature at a Point of a Manifold. For any point X of a
two-dimensional Riemannian manifold M we can define a number f(X), the
Gaussian curvature of the manifold M at the point X. Later we shall give
 I. Two-Dimensional Manifolds of Bounded Curvature 45

formulae for calculating %(X) in some special local coordinate systems on the
manifold. We note, however, that those properties of the curvature that we give
later can be used, in principle, for its definition. (In the given case the main
property is the Gauss-Bonnet formula.) With the help of %(X) we can introduce
some additive set functions. Namely, let E be an arbitrary Borel set in M.
Then if E is bounded (that is, the closure of E is compact), there is defined the
integral

f f %(p) dS(p) = w(E).

E

We shall call w(E) the integral curvature of the set E. In addition, we introduce
the following quantities:

Iwl(E) = f f'%(p), dS(p),

E

w+(E) = f f %+(p) dS(p), w-(E) = f f %-(p) dS(p).

E E
Here %+(p) = max {%(p), o} =(I%(p)1 + %(p))j2, %-(p)=max{ -%(p), o}=
(1%(p),I- %(p))j2. The quantities Iwl(E), w+(E) and w-(E) are defined for any
Borel set E in the manifold M (the boundedness of E is not required). Infinite
values of the integrals are allowed. The quantity Iwl (E) is called the absolute
integral curvature of the set E, and w + (E) and w - (E) respectively the positive
and negative parts of the curvature of E. For any E for which w(E) is defined we
have
w(E) = w+(E) - w-(E),
Iwl(E) = w+(E) + w-(E).
The Gauss-Bonnet formula (Alekseevskij, Vinogradov and Lychagin (1988)).
Here we give a relation connecting a turn of a curve and the integral curvature
of a set.
We shall call a set G in a two-dimensional manifold M a simple domain if G
is homeomorphic to the closed disc {(x, Y)lx 2 + y2 ~ 1} on the plane ~2. If Gis
a simple domain, then its boundary r is a simple closed curve.
Let G be a simple domain in a Riemannian manifold M, and r the boundary
of G. We assume that the curve ris piecewise smooth of class C 2 , that is, that r
can be split into finitely many simple arcs that are smooth of class C 2 • Let Xl'
X 2 , ••• , Xm , Xm +1 = Xl be the points of rthat bring about this splitting. Let Ij
be the arc [Xj' X j + 1 ] of this splitting, where j = 1, 2, ... , m. We specify arbi-
trarily a topological map cp of the domain G onto the disc B(O, 1) in ~2. We
orient r so that the following condition is satisfied. If p(t), a ~ t ~ b, is a right
parametrization of r, then the function x(t) = cp[p(t)], a ~ t ~ b, defines in ~2
the positive orientation of the boundary of the disc B(O, 1). In accordance with
46 Yu.G. Reshetnyak

this, each of the arcs Ij is oriented. Along each of the arcs Ij we specify an
orientation of M by means of the following agreement. Let U be a canonical
neighbourhood of Ij and 1/1: U -+ /R 2 a right homeomorphism of U into /R 2 • Let
H be the connected component of the set GnU containing Ij. Then 1/1 0 q> -1 is
a topological map of the set q>(M) that preserves the orientation. The orienta-
tion of M along Ij is completely determined by this condition. We shall say that
it is an orientation of the arc Ij compatible with the specified homeomorphism
q>: G -+ /R 2 • We note that the construction described here will be necessary when
we consider general two-dimensional manifolds of bounded curvature.
For each j there is defined the left turn of the arc Ij. We shall also denote it
by "G(Ij) and call it the turn ofthe arc Ij on the side ofthe domain G. For each
of the points Xj we define a number (J.j' the angle of the domain Gj at this point.
To this end, for each j = 1,2•... , m we define a local coordinate system q>j: Uj -+
/R 2 such that q>j(Xj) = 0 and the metric differential form of the manifold at the
point (0,0) takes the form dx~ + dx~. We shall assume that Uj is sufficiently
small and that q>j(Uj ) is the disc B(O, (5). Then the intersection Uj n r consists of
two arcs contained in Ij-l and Ij. and the images of these arcs are smooth
simple arcs starting from the point 0 and splitting the disc B(O, (5) into two
domains Bl and B2 • One of them (suppose it is B l ) is the image of ~ n Gunder
the map q>j. We put (J.j equal to the angle of the domain Bl at the point 0 (see
Fig. 16). We have 0 ::s;; (J.j ::s;; 2n. The quantity (J.j thus defined does not depend on
the choice of special coordinate system.
Theorem 4.3.2 (Alekseevskij, Vinogradov and Lychagin (1988». Let G be a
simple domain in a two-dimensional Riemannian manifold M of class coo. We
assume that the boundary r of G is split by points Xl. X 2 , ••• , Xm into smooth arcs
of class C 2 • Let (J.j be the angle of G (defined as mentioned above) at the point X j•
and "G(Ij) the turn on the side of G of the simple arc Ij = [Xj - l Xj]. Then we have
m m
L "G(Ij) + L (n -
j=l j=l
(J.j) = 2n - w(G)
(the Gauss-Bonnet formula)l.

Fig. 16

1 Encyclopaedia of Mathematical Sciences, vol. 28 (1991), p. 36

 I. Two-Dimensional Manifolds of Bounded Curvature 47

A geodesic triangle in the Riemannian manifold M is any domain T c: U

whose boundary consists of three shortest curves successively joining the points
A, Band C on the boundary of T and having no common points except the
ends. These shortest curves form a simple closed curve that is piecewise smooth
of class C 2 • We denote by oc, Pand y the angles ofthe boundary of the triangle
T at the points A, Band C respectively measured from the side of T.
The quantity oc + P+ y - n = <5(T) is called the excess of the geodesic tri-
angle T.
Corollary 1. For any geodesic triangle T in the Riemannian manifold M we
have
<5(T) = ro(A).
This result, in particular, enables us to find a way of calculating the Gaussian
curvature at an arbitrary point of a two-dimensional Riemannian manifold M.
Corollary 2. Let X be an arbitrary point in a Riemannian manifold M, and X m,
Ym, Zm' m = 1, 2, ... , sequences of points in M such that Xm -+ X, Ym-+ X and
Zm -+ X as m -+ 00 and for each m the points X m, Ym and Zm do not lie on one
shortest curve. Let Tm be the geodesic triangle bounded by the shortest curves
joining the sequences Xm and Ym, Ym and Zm, and finally Zm and Xm. Then
we have

m-+oo

Let us show how, by using the Gauss-Bonnet formula, we can define a tum
of a simple arc in a Riemannian manifold. Let K be a simple arc in the manifold
M. We assume that K is oriented and that along K there is specified an orienta-
tion of M. Let A be the beginning and B the end of K. We assume that we can
find a finite sequence Xo = A, Xl' X 2 , ••• , Xm = B of points of the curve K,
numbered in the order of disposition on K, such that each ofthe arcs [Xi - l , Xa,
i = 1, 2, ... , m, of K is a shortest curve. In this case we shall call K a geodesic
polygonal line in the manifold M. The points Xi' i = 0, 1, ... , m, are called its
vertices. For the polygonal line K we define two numbers KI(K) and K,(K), which
we shall call the left and right turns of K respectively. Namely, G and Hare
arbitrary left and right semineighbourhoods of M. We denote by oci , i = 1,2, ... ,
m - 1, the angle at the point Xi of the domain G, and by Pi the angle of the
domain H at the same point (see Fig. 17). Obviously OCi + Pi = 2n. We put
m-l m-l
KI(K) = L (n -
i=l
oci ), K,(K) = L
i=l
(n - Pi)'
We have

K1(K) + K,(K) = O.
The next assertion gives the answer to the question of how to define a tum of
an arbitrary simple arc of class C2 •
48 Yu.G. Reshetnyak

Fig. 17

Corollary 3. Let K be a simple arc of class C 2 in a two-dimensional Rieman-

nian manifold M. We assume that the arc K is oriented, A and B are the beginning
and end of K, and along K there is specified an orientation of M. Let (L m), m = 1,
2, ... , be a sequence of geodesic polygonal lines joining the end-points of K,
converging to K on the left (right). Let IXm and Pm be the angles at A and B of the
domain contained between Lm and K. If IXm ~ 0 and Pm ~ 0 as m ~ 00, then
K,(L m) ~ K,(K)
(respectively, Kr(Lm) ~ Kr(K) as m ~ 00).

Proof We confine ourselves to the case when the polygonal lines Lm con-
verge to K on the left. Let Gm be the domain included between K and Lm. We
denote by Yl' Y2' ... , Yn the angles of Gm at the vertices of Lm. The turn of each of
the arcs into which Lm is split by its vertices is equal to zero, and applying the
Gauss-Bonnet formula we obtain
n
K,(K) + (n - IXm) + (n - Pm) +
i=l
L (n - Yi) = 2n - w(Gm)·

Since Gm lies to the right of L m ,

n
L (n -
i=l
Yi) = Kr(Lm) = - K,(L m)

and as a result we obtain

By hypothesis IXm ~ 0, Pm ~ 0 as m ~ 00. Since the Lm converge to K as m ~ 00,

it follows that w(Gm) ~ 0 as m ~ 00, and so K,(K) = lim K,(L m), as required.
m"'OO

4.4. Isothermal Coordinates in Two-Dimensional Riemannian Manifolds of

Bounded Curvature. Let M be a two-dimensional Riemannian manifold, and
cp: U ~ ~2 an admissible chart in M. We say that cp is an isothermal coordinate
system in M if the line element of the manifold in this coordinate system has
the form
 I. Two-Dimensional Manifolds of Bounded Curvature 49

We give an expression for some quantities connected with a two-dimensional

Riemannian manifold in the isothermal coordinate system.
We have the following expression for the Gaussian curvature:
1 1 (8 2 In 82 In A A)
$'= -2A' AlnA = -2A' ~+~ .
Expressing In A in terms of K by the formula for solving the Poisson equation,
we obtain

In A(Z) f
= ~ fin Iz ~ ,,' $'(0' Am d~ drt + h(z),
G

where G = q>(U), and h(z) is a harmonic function of the variable z = x + iy. The
last equality can be rewritten as follows:

In A(Z) =~. ffin Iz ~ "dwm + h(z), (4.10)

G

where w(E) is a totally additive function, the integral curvature of a set.

The equality (4.10) is interesting in that it suggests an analytic approach
to the solution of the problem of introducing two-dimensional manifolds of
bounded curvature. For these the integral curvature is an arbitrary totally addi-
tive set function. It is natural to consider two-dimensional Riemannian mani-
folds with a line element which in a neighbourhood of each point admits the
representation
ds 2 = A(X, y)(dx 2 + dy2),
where the function A(Z), z = x + iy, admits a representation of the form (4.10) in
which w(E) is an arbitrary set function. It turns out that the general concept of
a two-dimensional manifold of bounded curvature can actually be defined in
this way. More details on this will be given below.

§ 5. Manifolds with Polyhedral Metric

5.1. Cone and Angular Domain. We first describe a general construction re-
lating to arbitrary metric spaces. Suppose we are given a metric space (M, pl.
We denote the ray {x E IRlx > O} of the number line IR by IRri. We construct the
direct product M x IRri. Formally M x IRri is the set of all pairs (x, t), where
x E M, t > O. For the element z = (x, t) ofthe set M x IRri we put r(z) = t. To the
set M x IRri we adjoin an element 0, for which we shall assume that r(O) is equal
to zero. The set M x IRri, augmented by the element 0, will be denoted by QM'
It is convenient to identify 0 with the set of all pairs of the form (x, 0). For
z E QM the number r(z) is called the polar radius of the point z.
50 Yu.G. Reshetnyak

We now introduce in QM a metric p*. Let Zl and Z2 be two arbitrary elements

of QM' If Zl = Z2' we put p*(z, z) = 0. If Zl = 0, Z2 =I< 0, we put P*(ZI' Z2) =
P*(Z2' zd = r(z2)' Suppose that Zl = (Xl' r l ), Z2 = (X2' r2)' where r l > 0, r 2 > 0.
If P(XI' x 2) ~ 11:, we put P*(ZI' Z2) = r l + r2' If P(XI' X2) < 11:, then P*(Zlo Z2) is
defined by the formula
P*(ZI' Z2) = Jr~ + r~ - 2rl r2 cos P(XI' X2)'
Obviously P*(ZI' Z2) in this case is equal to the side BC of the planar triangle
ABC, for which IABI = r l , lAC! = r2' LBAC = P(XI' X2)' It is easy to verify
that P*(YI' Y2) is the metric in QM' The metric space QM defined in this way, and

°
also any space isometric to it, will be called a cone over the metric space (M, pl.
The point is called the vertex ofthe cone QM'
Let a be an arbitrary point of the metric space (M, pl. The totality of all
points of the cone QM of the form Z = (a, r), where r ~ 0, will be called its
generator corresponding to the point a E M and denoted by y(a). Let (a, rd = Zl
and (a, r2) = Z2 be two arbitrary points ofthe generator y(a) of the cone QM' The
set of all points Z E QM of the form (a, r), where r lies between r l and r2' is called
the interval of the generator y(a) with ends Zl and Z2' For any two points Zl =
(a, r l ) and Z2 = (a, r2) we have P*(ZI' Z2) = Irl - r21. Hence it follows that the
length of the interval of the generator of QM with ends at the points ZI = (a, r I)

°
and Z2 = (a, r2) is equal to Irl - r21, and so any such interval is a shortest curve
in the space QM' Let be the vertex of the cone QM' For any point Z = (x, r) E
QM we have, by definition, p*(O, z) = r. Let us specify arbitrarily a number
h > O. The set QM(h) of all points z = (x, r) E QM for which r = p*(O, z) < h will
be called a finite cone over the space (M, p) with length of generator equal to h.
The set r,. of all points z = (x, r) for which r = p*(O, z) = h is later called the
cross-section of the cone QM at a distance h from its vertex.
If the metric of the space (M, p) is intrinsic, then (QM' p*) is also a space with
intrinsic metric.
If the metric spaces(MI' pd and (M 2, P2) are isometric, then the cones QMI
and QM2 are isometric spaces.
Later we shall require the following two special cases of the general construc-
tion described here.
We obtain the first case by choosing for M the interval [a, b], a < b, of the
number line lit In this case for x, Y E [a, b] we put p(x, y) = Ix - Yi. The cone
QM corresponding to the case when M is the interval [a, b] is called an angular
domain, and the number 0 = b - a is called the magnitude of this angular do-
main. If the intervals MI = [aI' bi ] and M2 = [a2' b2] have equal lengths, then
the cones QMI and QM2 are isometric. The angular domain whose magnitude is
equal to 0 will be denoted by A(O). Obviously for any 0 > 0 the space A(O) is
homeomorphic to a half-plane. Let us define a map j of the cone A(O) = Q[II,b]
into the plane ~2 by putting for (x, r) E Q[II,b]
j(x, r) = (r cos(x - a), r sin(x - a».
In a neighbourhood of any point Y = (x, r) of the cone A(O) such that r > ° the
 I. Two-Dimensional Manifolds of Bounded Curvature 51

mapj is topological. It is easy to verify that each curve in A(O) is transformed by

j into a curve of the same length, so it follows, in particular, that j is isometric in
a neighbourhood of any point y E A(O) other than the vertex of the cone A(O).
In the case when 0 < 2n the map j is one-to-one, and j(A(O)) is a planar angular
domain, which in polar coordinates (r, <p), where r is the distance from the origin
and <p is the polar angle, is defined by 0:::; <p :::; O. The generators y(a) and y(b) of
A(O) are called its boundary rays. If 0 = n, the cone A(O) is isometric to a closed
half-plane.
We obtain the second (special) case by choosing for M a circle in the plane
1R2. Let us specify a number 0 > 0 arbitrarily, and let Co be a circle in the plane
£2 with centre at the point (0,0) and such that its length is equal to O. For
arbitrary points x, y E Co we put p(x, y) = 0 if x = y, and in the case when x i= y
let p(x, y) be the length of the shortest arc into which Co is split by the points x
and y. Obviously p(x, y) is the metric on Co. This metric is intrinsic. We shall call
the cone over the metric space (Co, p) a circular cone and denote it by Q(O). (In
the case when 0:::; 2n the cone Q(O) is isometric to the lateral surface of an
infinite right-circular cone in the space Ie.) The number 0 is called the total
angle of the circular cone Q(O), and the difference 2n - () = w[Q«())] is called its
curvature. Any point of the cone Q«()) other than its vertex has a neighbourhood
isometric to a piece of the plane. If w = 0, then some neighbourhood of the
vertex of the cone Q«()) is also isometric to a flat domain and the space Q«()) is
isometric to the plane 1E2. If w is non-zero, then no neighbourhood of the vertex
of Q«()) is isometric to a flat domain. To verify this let us consider the cross-
section I;. of the cone Q«()) at a distance h from its vertex. The set I;. is a simple
closed curve whose length is equal to Oh. If some neighbourhood of the point Q,
the vertex of the cone Q«()), were isometric to a flat domain, then for sufficiently
small h the length of I;. would be equal to 2nh i= ()h.
Suppose we are given a number h > O. The set of all points x of the cone A«())
at a distance less than h from its vertex will be called a circular sector of radius
h with angle () and denoted by A«(), h). The set of all points of the circular cone
Q«()) at a distance less than h from its vertex will be called a finite circular cone
with total angle () and length of generator h and denoted by Q«(), h).
Any metric space isometric to a circular (finite circular) cone will also be
called a circular cone (finite circular cone respectively).

Shortest curves in the cones A«()) and Q«()). Let M be an arbitrary metric
space with intrinsic metric. We construct a cone QM and take an arbitrary
generator y(a) of it, where a E M. Let X = (a, rd and Y = (a, r2) be two arbi-
trary points of y(a), where r1 < r2. The set of all points Z = (a, r), where r 1 :::;
r :::; r2, is a simple arc L = [X, Y], which we shall call an interval of the given
generator. It is easy to verify that if Z ¢ [X, Y], then p*(X, Y) < p*(X, Z) +
p*(Z, Y) (a strict inequality). Hence it follows that if the path X(t), p :::; t :::; q, in
the space QM, joining the points X and Y, is not a parametrization of the
interval [X, y], then its length I(X; p, q) > p*(X, Y) = r2 - r 1 • This enables us
to conclude that if the points X and Y in the cone QM lie on one generator
52 Yu.G. Reshetnyak

of QM' then they are joined in QM by a unique shortest curve - the interval
[X,Y].
We now consider a special case of the cone Q(O). Let X and Y be two arbi-
trary points of it. If one of them is the vertex of Q(O), then X and Y lie on one
generator, and so in the given case they are joined by a unique shortest curve -
an interval of this generator. Let X = (a, r 1), Y = (b, r2), where a and bare
points of the circle CII • We shall assume that a :F b, since otherwise X and Y lie
on one generator of Q(O). The points a and b split CII into simple arcs F1 and F2 •
Let 01 and O2 be the lengths of these arcs. We denote by Q1 the set of all points
(x, r) of Q(O) such that x E F1, r ~ 0, and by Q2 the totality of all (x, r) E Q(O) for
which x E F2 • In the induced intrinsic metric the domain Q1 is isometric to the
angular domain A(Od, and the domain Q2 is isometric to A(02)' Under the
isometry, to the vertex of Q(O) there correspond the vertices of A(Od and A(02)'
The distance between the points a and b on CII is equal to min{01' 02}' and if
01 ~ n, O2 ~ n, then p*(X, Y) = r 1 + r2' The simple arc composed of the inter-
vals [X, 0] and [0, Y] of generators of Q(O) has length equal to r1 + r 2 =
p*(X, Y), and so it is the shortest curve joining X and Y. It is easy to verify that
for any point Z :F 0 we have p*(X, Z) + p*(Z, Y) > r1 + r2 (a strict inequality).
Hence it follows that the shortest curve joining X and Y is unique. In the case
when min{01' 02} < n we have p*(X, Y) < r1 + r 2, so in this case the simple arc
composed of the intervals [X, 0] and [0, Y] is not a shortest curve. Suppose,
for definiteness, that min{01' 02} = 01, The set Q1 is isometric to an angular
domain on the plane equal to 01 , Let us map Q1 isometrically onto this angular
domain. To the points X and Y there correspond points X' and Y' on the sides
of the angular domain. It is obvious that the simple arc that goes over to the
interval joining X' and Y' under this map is a shortest curve in the cone Q(O),
joining X and Y. If O2 = 01 < n, then applying our arguments to the domain Q2
we deduce that in Q(O) there are two distinct shortest curves joining X and Y.
One of them goes into Q1 and the other into Q2 and the vertex of the cone lies
inside the domain bounded by these shortest curves.
From what we have said here about shortest curves in Q(O) there follows a
circumstance in relation to which the case 0 < 2n is qualitatively different from
the case 0 ~ 2n. Namely, if 0 < 2n, then a shortest curve in the cone Q(O) cannot
pass through its vertex (that is, the vertex of the cone cannot be an interior point
of the shortest curve). If 0 ~ 27t, then in the cone there are shortest curves
passing through its vertex.
Let us investigate the arbitrariness with which a shortest curve in Q(O), where
o ~ 27t, that ends at the vertex of the cone, can be extended beyond this vertex
so that the resulting curve remains a shortest curve. In the case 0 = 27t the cone
Q(O) is isometric to a plane and here everything is clear. We shall assume that
o> 2n. Let X = (a, r), where r > 0, be an arbitrary point of Q(O). On the circle
Ce we layoff from the point a on different sides of it the arcs [a, b] and [a, c]
such that the length of each of them is equal to 7t, and let F be the arc [b, c] of
Ce, a ¢ [b, c]. If Y = (p, s), where p E F, s > 0, then the simple arc composed of
intervals of the generators [X, 0] and [0, Y] is a shortest curve. If p ¢ F, then
 I. Two-Dimensional Manifolds of Bounded Curvature 53

this simple arc is not a shortest curve. Thus, moving from X to 0 along the
generator and then from 0 along any generator l(p), where pEr, we obtain a
shortest curve in Q(O) passing through 0, and all the shortest curves that are
extensions of the shortest curve [X, 0] beyond 0 can be constructed in this way.
The cone Q(O) can be obtained from A(O) by pasting together the extreme
rays of the cone A(O), identifying points of these rays that are equidistant from
the vertex of the cone.
We assume that there is specified a finite collection of cones A(Oj), A(Oj) =
Q[lIj.bd' i = 1, 2, ... , m, where the intervals [aj, bl] are pairwise disjoint. Pasting
together the angular domains A(Oi) along extreme rays, we obtain from them a
metric space that is isometric to A(O), where 0 = 01 + O2 + ... + Om. Formally in
a given case the pasting is carried out as follows. Let Li = l(ai) and L; = l(bi) be
boundary rays ofthe angular domain A(OJ For each i = 1,2, ... , m we paste the
boundary ray L; ofthe cone A(Oi) to the ray Li+1 ofthe cone A(0i+1), identifying
points of these rays that are equidistant from the origin.
From what we have said it follows, in particular, that in the general case both
A(O) and Q(O) can be obtained by pasting together finitely many flat convex
angular domains.

5.2. Definition of a Manifold with Polyhedral Metric. Let M be an arbi-

trary two-dimensional manifold with boundary (see Alekseevskij, Vinogradov
and Lychagin (1988». We assume that M is connected and an intrinsic metric
p(X, Y) is defined in it. Then we shall say that the metric p(X, Y) is polyhedral
and the metric space (M, p) is a two-dimensional polyhedron ifthe conditions A
and B formulated below are satisfied. The first of these conditions concerns the
structure of the metric space (M, p) close to its interior points, and the second
close to the boundary points of M.
A. For any interior point X of the manifold M we can find a neighbourhood
that admits an isometric map qJ onto some circular cone Q(O, h), where 0 > 0,
h > 0, such that qJ(X) is the vertex of this cone (0 and h depend on the point X,
of course).
Let X be an interior point of M, and V a neighbourhood of this point such
that condition A is satisfied. The quantity 0 does not depend on the choice of the
neighbourhood V. This follows from the fact that it can be defined in a way that
does not require consideration of any special neighbourhoods of X. Namely, let
S(X, r) be the circle of radius r in (M, p) with centre X; I(X, r) is its length. Then
for sufficiently small r > 0, namely for r < h, we have I(X, r) = Or and so 0 =
lim I(X, r)/r. The quantity 0 is called the total angle at the point X of the mani-
' .... 0
fold M and will be denoted by O(X). We put ro(X) = 21t - O(X). The quantity
ro(X) is called the curvature of M at the point X E M. The point X is called a
vertex of M if its curvature is non-zero. We note that if a point Y E V is distinct
from X, then by what we said above about the structure of cones the point Y
has a neighbourhood isometric to a disc on the plane, that is, to a cone Q(21t, (5),
and so the curvature at the point Y is zero. Hence it follows, in particular, that
54 Yu.O. Reshetnyak

V does not contain vertices of M other than X, so all the points of the set of
vertices of M are isolated.
B. For any boundary point X of the manifold M we can find a neighbour-
hood V of this point that admits an isometric map cp onto some circular sector
A(O, h) (the numbers 0 > 0 and h > 0 depend on the point X) such that cp(X) is
a vertex of the sector A(O, h).
Let X be a boundary point of the manifold M, and Va neighbourhood of it
for which condition B is satisfied. The number 0 does not depend on the choice
of V; this is established in exactly the same way as in the case of interior points
of M. We put O(X) = 0, K(X) = n - O. The quantity K(X) is called the turn of the
boundary of the polyhedron M at the point X. O(X) is the total angle of the
polyhedron at the point X. The point X is called a boundary vertex of the
polyhedron if K(X) =F O. If a point Y E V is a boundary point and distinct from
X, then some neighbourhood of Y is isometric to a half-disc on the plane, that
is, to a circular sector A(n, 15), where 15 > 0 and so it is not a boundary vertex of
M. We deduce that V does not contain boundary vertices other than X.1f Y E V
is an interior point of M, then some neighbourhood of Y is isometric to a disc,
and hence Y is not a vertex of M, so the set V also does not contain vertices of
the polyhedron M.
A neighbourhood V of a point X of a polyhedron M that satisfies condition
A in the case when X is an interior point of M, and condition B in the case when
X is a boundary point, will be called a canonical neighbourhood of X.

A remark about terminological character. The word "vertex", used without

adjoining the word "boundary", implies that the point to which it is applied is
an interior point of the given polyhedron.
A polyhedron M will be called complete if the metric space (M, p) is complete
(that is, any fundamental sequence of points in it is convergent).
Any compact set on a polyhedron contains finitely many vertices. This fol-
lows from the fact that by a theorem of Borel a compact set is covered by finitely
many canonical neighbourhoods.
Numerous examples of two-dimensional manifolds with polyhedral metric
can easily be constructed. Suppose, for example, that M is the boundary of a
bounded convex polyhedron D in the space Ie, endowed with the metric in-
duced from 1E3. As a topological space, M is homeomorphic to the sphere S(O, 1)
(a topological map of M onto the sphere can be obtained by choosing arbi-
trarily an interior point P of the polyhedron D and projecting its boundary onto
the sphere S(P, 1) by rays starting from P). If X E M is an interior point of some
face of a convex polyhedron M, then some neighbourhood of it is isometric to a
disc on the plane. If X is an interior point of an edge of the polyhedral surface
M, then some neighbourhood of it in M splits into two parts, each of which is a
flat half-disc, and they have a common diameter. This neighbourhood is obvi-
ously isometric to a disc on the plane, so in this case the point X is not a vertex
of a manifold M with polyhedral metric. Let X be a vertex of the polyhedron D.
Consider the ball B(X, 6) in the space 1E3. For sufficiently small 6 > 0 the inter-
 I. Two-Dimensional Manifolds of Bounded Curvature 55

section B(X, e) (') M is the union of finitely many circular sectors that adjoin
each other along radii starting from X. This neighbourhood is isometric to the
circular cone Q(O, e), where 0 is the sum of the angles of the sectors at the point
X. By well-known results of the theory of convex bodies, 0 < 21t. Thus the
vertices of the convex polyhedron D are vertices of M as a manifold with poly-
hedral metric. The curvature at all the vertices of the manifold is positive.
Here we mention the following fact. Let X be a vertex of a convex poly-
hedron D. A plane P passing through X will be called the supporting plane of the
polyhedron D at the point X if D is contained in one of the two closed half-spaces
for which P serves as the common boundary. A vector v will be called an
outward normal vector of D at X if it is orthogonal to some supporting plane P
at X, directed into the half-space bounded by this plane that does not contain
D, and the length of the vector v is equal to 1. We shall assume that the outward
normal vectors are laid off from the fixed point 0 in 1E3. The set of their ends
lies on the sphere S(O, 1) and is called the spherical image of the point X. The
following assertion is true.

Theorem 5.2.1. The spherical image of a vertex of a convex polyhedron is a

convex spherical polygon on the sphere S(O, 1) whose area is equal to the curvature
of the vertex X.
The polygon mentioned in Theorem 5.2.1 is obtained as follows.
Let HI' H2, ••• , Hm be all the faces of the polyhedron D that meet at the point
X, and Vi the outward normal vector orthogonal to the plane of the face Hi. The
points VI' V 2 , ••. , Vm on the sphere S(O, 1) are the vertices of a polyhedron - the
spherical image of the point X - and there are no other vertices of it.
A proof of the theorem is given in Aleksandrov (195Oc). It is useful to com-
pare this theorem with a classical result of differential geometry, according to
which the area of the spherical image of a set on a surface is equal to the integral
curvature of this set.
A cone and an angular domain are obviously manifolds with a polyhedral
metric. An infinite right circular cylinder and a Mobius surface are polyhedra.
In the last two cases the curvature at each point of the polyhedron is zero.

A theorem on the triangulation of a manifold with polyhedral metric. The con-

cept of a manifold with polyhedral metric can also be defined in another way,
namely, the following assertion is true.
Theorem 5.2.2. Let M be a two-dimensional manifold with boundary. We as-
sume that M is connected and is endowed with a polyhedral metric. Then M admits
a triangulation K such that any triangle T E K is isometric to a triangle on the
plane 1E2.
We just give an outline of the proof.
First of all we observe that for any interior point X E M we can find a
domain G homeomorphic to a closed disc such that G is split by shortest curves
starting from X into parts, each of which is isometric to a planar triangle. If X
56 Yu.G. Reshetnyak

Fig. IS Fig. 19

is an interior point of M, then X is also an interior point of G, and if X E aM,

then X is a boundary point of G. (In the first case the splitting of G into triangles
looks like that shown in Fig. 18, and in the second case in Fig. 19). A domain G
satisfying these conditions will be called a T-neighbourhood of the point X. The
existence of a T-neighbourhood of any point X E M is established by means of
constructions relating to a cone. 2
Next we construct a finite or denumerable sequence G1 , G2 , ••• , Gy , ••• of
subsets, each of which is a T-neighbourhood of some point X E M such that any
compact subset of M intersects finitely many sets Gy (in particular, if M is
compact, the sequence Gy is finite). We subdivide each of the domains Gy into
domains isometric to planar triangles. As a result we obtain a covering of M by
some sequence of triangles 7;., J.t = 1, 2, ... , such that any compact subset of M
intersects finitely many triangles 7;.. For an arbitrary point X E M we denote by
Q(X) the intersection of all the triangles 7;. that contain X. The set Q(X) is either
a one-point set or a shortest curve or is isometric to a planar convex polygon.
Subdividing into triangles all the domains Q(X) that are isometric to polygons,
we obtain the required triangulation of M.
Pasting together of manifolds with polyhedral metric. The following proposi-
tion is true.
Theorem 5.2.3. Let (My), v = 1, 2, ... , be a finite or denumerable set of two-
dimensional manifolds with polyhedral metric. We assume that for this set there is
specified a pasting rule that satisfies all the conditions of 3.2, and M is a manifold
with intrinsic metric that arises as a result of the pasting. Then M is also a
two-dimensional manifold with polyhedral metric.

2 Let X be an interior point of M, Va canonical neighbourhood of X, and Q(O, h), 0 = O(X), a finite
circular cone to which V is isometric. The cone Q(O, h) is split by its generators into finitely many
circular sectors, so the angle of any of them is less than min(7t, 0/2). On each of these generators we
specify some point. We obtain a finite set of points Y1, Y2 , ... , Y.. , Y.. +1 = Y1 • We shall assume that
they are numbered so that Y1 and 1';+1 lie on the sides of one circular sector. Let Sj be the sector
bounded by the generators (!! 1'; and (!! 1';+1' and OJ the angle of this sector. Since OJ < 0/2, OJ < 7t, the
shortest curve in the cone Q(O, h) joining any two points on its sides goes inside the sector. Hence it
is obvious that Sj is isometric to a sector of a disc on the usual plane 1E2 and the part 1i of it bounded
by the shortest curves [ll!, 1';], [(!!, 1';+tl is isometric to a planar triangle. We thus obtain a finite
collection of triangles T1 , T2 , ••• , T.. , whose union G' constitutes a neighbourhood of the vertex of
the cone Q(O, h). Under an isometrjc map of Q(O, h) onto V the set G' is transformed into a domain
G having exactly those properties that we need. For the case when X is a boundary point of M the
constructions are similar.
 I. Two-Dimensional Manifolds of Bounded Curvature 57

The proof of the theorem reduces to a consideration of neighbourhoods of

points of M that arise as a result of pasting together boundary points of the
manifolds Mv. Suppose, say, that a point X is obtained by pasting together points
Xi E Mv., i = 1,2, ... , m. In Mv. the point Xi has a neighbourhood isometric to
some ci'rcular sector A(lJi , h). By pasting together these neighbourhoods we
obtain a neighbourhood of X. Since the pasting together of the circular sectors
along their generators gives as a result either a circular sector or a circular cone,
we deduce that X has a neighbourhood isometric to either a circular cone
K(lJ, h) or a circular sector A(lJ, h). We have proved that M is a manifold with
polyhedral metric.
Here we mention that, in particular, having pasted a manifold M with poly-
hedral metric to a second copy M' of it so that points of the boundary that
correspond under an isometric map of M to M' are identified, we again obtain
a two-dimensional polyhedron if (in the terminology introduced above, this is a
twice covered polyhedron M). The polyhedron if does not have boundary points,
and M is isometric to some subdomain of if endowed with the induced metric.
We assume that a point X E if is obtained by pasting together points X' E aM
and X" E aM'. Then lJ(X') = lJ(X") and lJ(X) = 2lJ(X'), and w(X) = 2K(X').
Polygonal lines on a polyhedron. Let M be an arbitrary two-dimensional
polyhedron. By Theorem 2.3.2 any point X E M has a neighbourhood, any two
points of which can be joined by a shortest curve.
Let L be a shortest curve on M, and X an arbitrary interior point of L. If X
is a vertex of M, then w(X) is negative; this follows from what we said in 5.1
about shortest curves in a cone. If X is a boundary vertex, then the rotation of
the boundary at this point is negative.
We shall call a simple arc L an interval on the polyhedron M if L is a shortest
curve and no internal point of it is either a vertex or a boundary vertex of the
polyhedron.
Any shortest curve on a polyhedron is a compact set and therefore contains
finitely many vertices. Hence it follows that a shortest curve on a polyhedron
consists of finitely many intervals.
A simple arc L on a manifold M is called a polygonal line if we can find on it
a finite sequence of points Xo, Xl' ... ' X m , where Xo and Xm are the end-points
of L, numbered in the order of disposition on L, such that each of the arcs
[Xi-I' X;J is a shortest curve. If L is a polygonal line, then, subdividing the
shortest curves [Xi - l X;J into intervals, we deduce that L is the union of finitely
many intervals.
We shall call a set K on a polyhedron M a simple polygonal line if K is a
simple curve in the sense of the definition in 2.1 (that is, K is a closed subset of
M and K is homeomorphic to either a circle or an interval of the number line)
and any simple arc contained in K is a polygonal line.
The following assertion is true.
Theorem 5.2.4 (Aleksandrov and Zalgaller (1962)). Let M be a two-
dimensional manifold with polyhedral metric, and Q a connected closed subset of
58 Yu.G. Reshetnyak

,,(
1,2.

M Q

LIj.

Fig. 20
> 1,3

M. We assume that Q is a two-dimensional manifold with boundary in the topology

induced from M, and that any component of the boundary of Q is a simple polygo-
nalline in M. Let Pa be the induced metric in Q. Then the metric space (Q, Pa) is
a two-dimensional polyhedron.
Remark. Under the conditions of the theorem the boundary of Q, as a two-
dimensional manifold, generally speaking does not coincide with the boundary
of Q as a subset of M.
The proof of the theorem reduces to a consideration of the structure of neigh-
bourhoods of the boundary points of Q (see Fig. 20).
The area of a set on a polyhedron. On any manifold with polyhedral metric
there is defined a set function called the area or two-dimensional Lebesgue
measure. Let us specify a triangulation K of a polyhedron M such that each of
the triangles T1 , T2 , ••• , r;" ... occurring in it is isometric to a planar triangle. A
set A c r;, is said to be measurable if the image of A is a measurable set under
an isometric map of 1'" into a plane. We denote by CT(A) the two-dimensional
Lebesgue measure of the image of A under an isometric map of r;, into a plane
and call it the area or two-dimensional Lebesgue measure of the set A. Let A c M
be an arbitrary set. Then A is said to be measurable if A n r;, is a measurable set
for each v and we put
CT(A) = L CT(A n 1',,).
•
It is not difficult to show that the class of measurable sets and the function CT do
not depend on the choice of triangulation K of the polyhedron M.

S.3. Curvature of a Set on a Polyhedron. Turn of the Boundary. The Gauss-

Bonnet Theorem. For two-dimensional manifolds with polyhedral metric we
define a set function that can be regarded as an analogue of the concept of the
integral curvature of a set in a Riemannian manifold.
Let M be a two-dimensional manifold with polyhedral metric. A set A c M
is said to be bounded if the closure of A is compact. The totality of all bounded
Borel sets on a polyhedron M will be denote by ~o(M). If a set A c M is
bounded, then by a theorem of Borel its closure is covered by finitely many
canonical neighbourhoods, and so A contains finitely many vertices. We shall
denote the totality of all Borel sets in M by ~(M).
 I. Two-Dimensional Manifolds of Bounded Curvature 59

For an arbitrary number x E IR we put x+ = x, x- = 0 for x ~ 0 and x+ = 0,

x- = -x for x < O. Obviously we always have x+ - x- = x, x+ + x- = Ixl.
Let X be a vertex of M. We put w+(X) = [w(X)]+' w-(X) = [w(X)r, Iwl(X) =
Iw(X)I. The numbers w+(X) and w-(X) are called the positive and negative parts
of the curvature of the point X.
Let A be an arbitrary Borel set on a polyhedron. We put
w+(A) = L w+(X), w-(A) = L w-(X),
XEA XEA

Iwl(X) = L Iwl(X) = w+(A) + w-(A).

XEA

The summation on both sides is carried out over the set of all vertices of the
polyhedron that belong to A. Infinite values of the sum are allowed. The num-
bers w+(A) and w-(A) are called the positive and negative parts of the curvature
of the set A, and Iwl(A) is called its absolute curvature.
If A is a bounded set in M, then, as we mentioned above, A contains finitely
many vertices of M, and so the sum
w(A) = L w(X)
XEA

is defined and finite. It is called the curvature of the set A.

Let us consider some examples. If a polyhedron M is a cone Q(O), then it has
a unique vertex and the curvature of a set A c Q(O) is equal to zero if the vertex
o of the cone Q(O) does not belong to A, and is equal to 2n - 0 if 0 E A. In
particular, w(M) = 2n - O.
Let M be the surface of a cube. Then M has eight vertices. The total angle at
each of them is equal to 3n/2, so it follows that all the curvatures of the vertices
of the cube are equal to n/2, and so w(M) = 8(n/2) = 4n. Similarly, if M is the
surface of a regular tetrahedron, then M has four vertices and the total angle at
each of them is equal to 3(n/3) = n. In accordance with this the curvatures of all
the vertices of the tetrahedron are equal to nand w(M) = 4n. We see that in
both cases the curvature of the polyhedron M is equal to the same number,
namely 4n. This observation is not accidental. Moreover, the following general
assertion is true.
Theorem 5.3.1. Let M be a compact two-dimensional manifold without bound-
ary and with polyhedral metric. Then
w(M) = 2nX(M),
where X(M) is the Euler characteristic of M.
Proof We specify arbitrarily a triangulation K of the polyhedron M such
that each triangle of K is isometric to a planar triangle. The existence of such a
triangulation follows from Theorem 5.2.2. Let no be the number of vertices of K,
n 1 the number of edges, and n2 the number of faces. Then according to Euler's
formula no - n1 + n2 = X(M). We observe that 2nl = 3n2' from which we ob-
tain X(M) = no - tn 2 . Let 1';, i = 1,2, ... , n2' be all the triangles that form the
60 Yu.G. Reshetnyak

triangulation K. We denote the angles of the triangle 'Ii by !Xi' Pi and Yi' We have
ai + Pi+ Yi = n for any i. Hence
L (!Xi + Pi + yJ
"2

nn2 =
i=1

We split the last sum into groups so that in each of these groups there occur
the terms that are the angles at one vertex of the polyhedron. The sum of the
angles corresponding to a vertex X of the triangulation K is equal to O(X) =
2n - w(X). Hence we conclude that nn 2 = 2nno - w(M), and so w(M) =
2n(no - tn 2).As we showed above, no - tn2
= X(M), so the theorem is proved.
To each boundary vertex X of M there corresponds a number K(X), the
turn of the boundary at this point. This enables us to define some additive
functions specified on the boundary of the polyhedron. Let ~(oM) denote the
totality of all Borel subsets of oM, and ~o(oM) the totality of all bounded Borel
subsets of oM. For E E ~o(oM) we put
K(E) = L K(X).
XeE

For E E ~(oM) let

IKI(E)= L K(X), K+(E)= L [K(X)]+, K-(X) = L [K(X)r·
XeE XeE XeE

We shall call K(E) the turn of the boundary on the set E, and IKI(E), K+(E) and
K-(E) the absolute turn, the positive part ofthe turn and the negative part ofthe
turn of the boundary on the set E respectively.
If, in particular, the boundary of M is compact, then there is defined a num-
ber K(oM), the turn of the boundary of M. The number K(oM) is an analogue of
integral geodesic curvature.
The next assertion follows from Theorem 5.3.1.
Theorem 5.3.2 (the Gauss-Bonnet theorem for polyhedra). Let M be a two-
dimensional polyhedron. We assume that M is a compact manifold and the bound-
ary of M is not empty. Then
w(M) + K(oM) = 2nX(M),
where X(M) is the Euler characteristic of M.
Proof Let M be a twice convered manifold M. Then X(M) = 2X(M). In fact,
we specify arbitrarily a triangulation K of M. The manifold Mis obtained by
pasting together two different copies of M. From the triangulation of M there
naturally arises a triangulation K of M. Let no be the number of vertices, n 1 the
number of edges, n2 the number of faces of K, n~ the number of vertices of K
belonging to the boundary of M, and n~ the number of edges of this triangula-
tion contained in oM. Then obviously the number iio of vertices ofK is equal to
2no - n~, the number ii1 of edges ofthis triangulation is equal to 2n1 - nl' and
the number ii2 of faces of K is equal to 2n2' Each connected component of the
boundary of M is a simple curve. Hence it follows that on it there lie as many
 I. Two-Dimensional Manifolds of Bounded Curvature 61

edges of the triangulation as vertices, and so no = nJ.. This enables us to con-

clude that

X(M)'= fio - fi1 + fi2 = (2no - no) - (2n1 - nJ.) + 2n2

= 2(no - n 1 + n2) = 2X(M).

To any interior vertex X of a polyhedron M there correspond two vertices of

M, and at each of them the curvature is equal to w(X). To each boundary vertex
X there corresponds one vertex of M, whose curvature is equal to 2"(X). Hence
it follows that w(M) = 2w(M) + 2,,(aM). According to Theorem 5.3.1, w(M) =
2nX(M) = 4nx(M), and so the theorem is proved.

5.4. A Turn of a Polygonal Line on a Polyhedron. Let us specify arbitrarily a

two-dimensional manifold M with polyhedral metric.
Let L be a polygonal line on a polyhedron M; L is a simple arc. We assume
that L does not contain boundary points of M. We orient L in an arbitrary way
and specify along L a definite orientation of the manifold M. We have thus
defined the concepts of the left-hand and right-hand sides of L. With each
interior point X of L we associate two numbers, which we shall call the left and
right angles at a point of L. Let us describe the construction by means of which
they are determined. Let A be the beginning and B the end of L, and Xo = A,
Xl' ... , X m- 1 , Xm = B the points of L such that each of the arcs [Xi - 1 Xa is an
interval. (The points Xi are assumed to be numbered in the order of disposition
on L.) We assume that X lies inside one of the arcs [Xi - 1 Xa, i = 1, 2, ... , m.
Then some sufficiently small neighbourhood of the point X is isometric to a disc
on the plane, and under an isometric map into the plane the part of L lying in
this neighbourhood goes over to an interval. In this case we assume that the left
and right angles at X are equal to n. We now assume that X is one of the points
Xi' 0 < i < m. Let V be a canonical neighbourhood of Xi' We assume that this
neighbourhood is sufficiently small so that the intersection V n L consists of
intervals (X'Xa, [XiX"), where X' lies between X i- 1 and Xi' and X" is between
Xi and Xi+1' The polygonal line L splits V into two components, one of which
lies to the left of L (that is, it is contained in some left half-neighbourhood of L).
We denote this component by v,. The other component V\L, which we denote

°
by v" lies to the right of L. Under an isometric map of V onto the circular cone
Q(O, e), where = O(x), e > 0, the arcs [X'Xa, [XiX"] go into some generators
of this cone, and the domains V, and V, are mapped into sectors of the cone
Q(O, e). Let O,(X, L) be the angle of the sector corresponding to v" and or (X, L)
the angle of this sector corresponding to v,. Obviously O,(X, L) and 0r(X, L) do
not depend on the choice of the neighbourhood V. We shall call O,(X, L) the
angle at X on the left-hand side of L (or to the left of L), and 0r(X, L) the angle
at X on the right-hand side of L (or to the right of L). Obviously O,(X, L) +
Or (X, L) = O(X). We put ",(X, L) = n - O,(X, L), "r(X, L) = n - 0r(X, L). We
shall call ",(X, L) and "r(X, L) the left and right turns of the polygonal line L at
the point X, respectively. We have ",(X, L) + "r(X, L) = 2n - O(X) = w(X). In
62 Yu.G. Reshetnyak

particular, we deduce that if the point X = Xi is not a vertex of the polyhedron

M, then K,(X, L) = -K,(X, L). In the case when X does not coincide with one
of the points Xi' K,(X, L) = K,(X, L) = O. We put
m-l m-l
K,(L) =L K,(X i , L), K,(L) =L K,(Xi' L).
i=l i=l

We shall call K,(L) and K,(L) the left and right turns of L, respectively. The sum
K,(L) + K,(L) is equal to the sum of the curvatures at the vertices of the poly-
hedron that are interior points of L, that is, this sum is equal to the curvature of
the open arc (AB) = L\ {A, B}. We thus have
K,(L) + K,(L) = w[(AB)]. (5.1)

If L is a shortest curve, then it follows from what we said in 5.1 about shortest
curves on a cone that O,(X, L) ~ 1t, O,(X, L) ~ 1t at each point X E L, and so
K,(X, L) ~ 0, K,(X, L) ~ O. In particular, we deduce that K,(L) ~ 0 and K,(L) ~
O. We note that in the case of polyhedra, generally speaking, we cannot assert
that a turn of a shortest curve is equal to zero, as this holds for shortest curves
in a Riemannian manifold. It is easy to verify this by considering a shortest
curve on a cone K(O), where 0 > 21t, passing through its vertex O. It is obvious
from the arguments of 5.1 that the right angle and also the left angle of the
shortest curve at the vertex of the cone can take any values from 1t to 0 - 1t;
according to this, the left and right turns of a shortest curve at a vertex of an
open polygon can take any values from the interval (21t - 0, 0) = (w(O), 0).
Let us mention one consequence of Theorem 5.3.2. Let L be an oriented
polygonal line on a polyhedron M, contained in an open set U c M homeo-
morphic to an open disc in 1R2. We denote by A and B the beginning and end of
L. Along the arcs L we specify a definite orientation of M. Let K be an oriented
open polygon with beginning A and end B that does not have any other points
in common with L and is contained in U. We assume that K lies to the left of L.
Let G be a domain between K and L, homeomorphic to a closed disc, and GO
the totality of all its interior points. Then

(5.2)

where IX and p are the angles of G at A and B.

For the proof it is sufficient to consider the manifold obtained if we introduce
the induced metric in G and apply Theorem 5.3.2. The Euler characteristic of a
disc is equal to 1. The boundary of G consists of the arcs K and L. Since L lies
to the right of K, at each boundary vertex of G lying inside K the turn of the
boundary of G is equal to K,(X, K). At each boundary vertex of G lying inside L
the turn ofthe boundary of G is equal to K,(X, L). At A and B the turns are equal
to 1t - IX and 1t - p. Hence we conclude that K(OG) = 21t - IX - P+ K,(K) +
K,(L). The equality (5.2) now follows directly from Theorem 5.3.2.
Let K and L be open polygonal lines satisfying all the conditions listed above.
We have K,(K) = w(K\ {A, B}) - K,(K). Observing that (K\ {A, B}) u GO = G\L,
 I. Two-Dimensional Manifolds of Bounded Curvature 63

we obtain the following relation:

",(L) = ",(K) + IX + P- w(G\L). (5.3)

Similarly we obtain
"r(K) = "r(L) + IX + P- w(G\K). (5.4)

5.5. Characterization of the Intrinsic Geometry of Convex Polyhedra. Here

we give a result that is in a certain sense the beginning of the whole problem
area considered in the present article. We have in mind a theorem of A.D.
Aleksandrov on the realization of a convex polyhedron with a given metric
(see Aleksandrov (1948a». In differential geometry there was known a problem
posed by Weyl in 1918. If R is a closed convex surface in three-dimensional
Euclidean space that satisfies the usual regularity conditions in differential ge-
ometry, then by introducing in R the metric induced from 1E3 we obtain a two-
dimensional Riemannian manifold homeomorphic to a sphere and such that at
each point of it the Gaussian curvature is non-negative. We now assume that
we are given a two-dimensional Riemannian manifold M homeomorphic to a
sphere and such that at each point of it the Gaussian curvature is non-negative.
Weyl's problem is as follows: can we assert that there is a closed convex surface
R such that M is isometric to this surface or, as we shall say later, on which
there is realized a given Riemannian metric defined on a sphere? In Weyl's
original statement analytic Riemannian manifolds were considered, that is, it
was assumed that the transition functions for admissible charts are real-analytic,
and the coefficients of the metric tensor are real-analytic functions. In addition,
it was assumed that the Gaussian curvature is positive everywhere. In this state-
ment, Weyl's problem was solved by H. Lewy (Lewy (1935».
Aleksandrov (Aleksandrov (1948a» took a purely geometrical approach to
the solution of Weyl's problem. The main part of it forms an analogue of the
Weyl-Lewy theorem for convex polyhedra. Namely, Aleksandrov proved the
following proposition.
Theorem 5.5.1 (Aleksandrov (1948a». Let M be a two-dimensional manifold
with polyhedral metric, homeomorphic to a sphere. Then for M to be isometric to
the surface of a closed convex polyhedron in 1E3 it is necessary and sufficient that
the curvature of each vertex of M should be non-negative.
It is necessary to state the following remark making the formulation of the
theorem more precise. To the list of convex polygons we have to add doubly
covered flat convex polygons-polyhedra that are degenerate in a certain
sense.
Now let P be an arbitrary Riemannian metric of positive curvature on the
sphere. Then we can construct a sequence (PII)' n = 1, 2, ... , of polyhedral me-
tries of positive curvature that converges uniformly to P as n -+ 00. By Theorem
5.5.1, for each n there is a closed convex polyhedron RII on whose surface the
metric PII is realized (that is, such that the sphere with metric PII is isometric
64 Yu.G. Reshetnyak

to the surface of RII). From the sequence of polyhedra (R II ) we can extract a

subsequence (RIIJ, n1 < n2 < ... < nk < ... , that converges to some convex body
R. The surface of this body is the required one: on it there is realized the spe-
cified Riemannian metric p.
To obtain a complete proof of the Weyl-Lewy theorem by the method de-
scribed, however, some additional work is required. The fact is that in this
theorem we consider analytic Riemannian metrics and prove that they are real-
ized on convex surfaces that are also analytic. The limit of the sequence of
convex polyhedra can be an arbitrary convex body. A priori such a situation is
not excluded, when a given analytic Riemannian metric can be realized not only
for the surface whose existence follows from Lewy's theorem but also on some
other surfaces. The completion of the given approach to a solution of Weyl's
problem is contained in deep research byPogorelov on the unique determina-
tion of convex surfaces and on the degree of regularity of a convex surface
depending on how regular its intrinsic metric is (Pogorelov (1969)).
In connection with the method adopted by Aleksandrov for the solution of
Weyl's problem, there naturally arises the problem of examining all intrinsic
metrics on a sphere, each of which is the limit of polyhedral metrics of positive
curvature. The manifolds with intrinsic metric that arise in this way are special
cases of two-dimensional manifolds of bounded curvature. Namely, they are
manifolds whose curvature as a set function is non-negative.

5.6. An Extremal Property of a Convex Cone. The Method of Cutting and

Pasting as a Means of Solving Extremal Problems for Polyhedra. Here we shall
consider two-dimensional manifolds with polyhedral metric, homeomorphic to
a closed disc. Later we shall state a theorem from which it follows that for a wide
class of extremal problems about such polyhedra the solution is achieved on
polyhedra of sufficiently special type. For the latter the corresponding extremal
problem reduces essentially to a problem from elementary geometry.
A two-dimensional polyhedron R will be called a convex cone if it is homeo-
morphic to a closed disc, it has at most one vertex, and at each boundary vertex
of R the turn is positive and the curvature of R is non-negative.
If a convex cone does not have vertices, which is allowed by the definition,
then it is isometric to a convex polygon on a plane. It is clear that conversely
any such polygon, in the sense of the given definition, is a convex cone. Let D be
an arbitrary bounded convex polygon on a plane in 1E 3 , and 0 a point lying
outside this plane and such that the foot of the perpendicular drawn from it to
the plane of D belongs to the polygon. Then the lateral surface of the pyramid
with vertex 0 and base D is a convex cone.
Let OAB and O'A'B' be two equal triangles on a plane, where 10AI = 10'A'I,
lOBI = 10'B'I and IABI = IA'B'I, and the angles at A and B are acute. Pasting
these triangles together along the sides OA and O'A', OB and O'B' so that the
points 0 and 0' are identified, we obtain a polyhedron, which is a convex cone.
One can show that all possible types of convex cones are exhausted by the
examples mentioned.
 I. Two-Dimensional Manifolds of Bounded Curvature 65

Let K1 and K2 be rectifiable simple closed curves in the metric spaces M1 and
M 2, and cp a map of K1 onto K 2. We shall say that cp maps K1 onto K2 with the
lengths of arcs preserved if cp maps K 1 onto K 2 one-to-one and takes any arc of
K1 into an arc of K2 of the same length.
Suppose we are given metric spaces (M1' P1) and (M2' P2)' We shall call a
map cp: M1 -+ M2 contracting iffor any two points X, Y E M1
P2[CP(X), cp[Y)] ~ P1(X, Y).
Theorem 5.6.1. (Reshetnyak (1961a)). Let R be a two-dimensional polyhedron
homeomorphic to a closed disc and such that w+(R) < 2n. Then there is a convex
cone Q and a map cp: Q -+ R such that the Jollowing conditions are satisJied:
1) the curvature oj Q does not exceed w+(R);
2) cp is a contracting map;
3) the boundary oj the cone Q is transJormed by the map cp into the boundary
oj R with the lengths oj arcs preserved.
The theorem shows that if w+(R) < 2n, then a polyhedron of the type consid-
ered here can be deformed in such a way that the distances between its points
are not decreased, the length of the boundary remains fixed, and the whole
curvature proves to be collected at one point. The positive part of the curvature
is not increased, and the negative part becomes zero.
The properties of the cone Q defined in Theorem 5.6.1 are described by the
following theorem.
Theorem 5.6.2 (Reshetnyak (1961a)). Let Q and R be polyhedra homeomorphic
to a closed disc. We assume that there is a contracting map cp oj Qonto R such that
the boundary oj Q is mapped onto the boundary of R with lengths preserved. Then
a) if some arc K c oR is a shortest curve, then the arc cp -l(K) is also a shortest
curve;
b) the lengths oj the curves oR and oQ are the same;
c) the area oj R does not exceed the area of Q;
d) K+(K) ~ K+(cp(K)) Jor any simple arc K c oQ.
Theorems 5.6.1 and 5.6.2 turn out to be useful for solving different kinds of
extremal problems. Here we give an example of a problem of this kind. The
given theorems admit a generalization to the case of arbitrary two-dimensional
manifolds of bounded curvature. Other examples in which they find application
are more appropriately considered later in a section devoted to extremal prob-
lems in manifolds of bounded curvature.
Let R be a polyhedron homeomorphic to a disc, and Xl' ... , Xn all the
boundary vertices. Each of the arcs [X 1X 2], [X 2X 3 ], ... , [X n X 1] is an interval.
Let 11 , ••• , In be the lengths of these intervals.
Theorem 5.6.3 (Aleksandrov (1945)). In the set oj all polyhedra R homeomor-
phic to a disc and such that w + (R) ~ Wo < 2n, the boundary oj which consists oj
intervals whose lengths are 11 , 12 , ••• , In' the greatest area is that oj a polyhedron
that is a convex cone whose curvature is equal to Wo and whose vertex is equidis-
tant Jrom all its boundary vertices.
66 Yu.G. Reshetnyak

The proof of the theorem is as follows. Let R be a polyhedron satisfying the

conditions of the theorem, Q the convex cone corresponding to R in accordance
with Theorem 5.6.1, and Yl' Y2 , ••• , Y" the points that belong to the boundary
of Q and are transformed into the boundary vertices Xl' X 2 , ••• , Xn of R by the
contracting map ({J mentioned in Theorem 5.6.1. None of the arcs [Y1 Y2 ], ••• ,
[y"-1 Y,,], [Y" Y1] contains boundary vertices of Q, because otherwise the rota-
tion of this arc would be positive, which is impossible by virtue of assertion
d) of Theorem 5.6.2, according to which 1(+([Y;-1 Y;]) ~ 1(+([Xi - 1X;]) = O.
Thus the boundary of Q consists of intervals whose lengths are equal to
11' ... , In·
We have 0 ~ W(Q) ~ w +(R) ~ W0 By assertion c) of Theorem 5.6.2 the area
0

of Q is not less than the area of R. Thus Theorem 5.6.3 reduces to the case when
Q is a convex cone. For this case the proof is carried out by means of elementary
geometry. Cutting the cone along the shortest curve joining its vertex to one of
the boundary vertices, we obtain a polyhedron P isometric to a flat polyhedron
OZl Z2 ... ZnZ~ constructed as follows (see Fig. 21). The points Zl and Z~ lie on
the sides of the angle Zl OZ~ at an equal distance from the point O. The value of
this angle is equal to the total angle at the vertex of the cone Q and so it is not
less than 2n - Wo, and the open polygon ZlZ2 ... ZnZ~ goes inside it. We have
IZIZ21 = 11' IZ2Z31 = 12 , ... , IZnZ~ 1= In. The area of the polyhedron P is equal
to the area of the cone Q. The problem reduces to that of finding, among all such
polyhedra P, the one that has greatest area. This is a problem of elementary
geometry. It is easy to establish that the area of P is maximal when the angle
Zl OZ~ is equal to 2n - Wo, and the points Zl' Z2' ... , Zn' Z~ are at the same
distance from the point o. Pasting the extremal polyhedron P along the sides
OZl and OZ~, we obtain a convex cone constructed as described in Theorem
5.6.3, and the theorem is proved.
The proof of Theorem 5.6.1 uses the method of cutting and pasting sug-
gested by Aleksandrov. In its main features it consists in using the following two
devices.
We assume that some domain G on the polyhedron R is isometric in the
induced metric to a planar non-convex quadrangle H = ABCD with non-
convex vertex D (see Fig. 22). We regard the vertex D as lying opposite to A and
assume that the quadrangle H is endowed with the induced metric. On the plane
we construct a triangle T = A'B'C' with sides A'B' = AB, A'C' = AC, B'C' =
BD + CD. Such a triangle exists because obviously AB + BC > BD + DC.

~n

Fig. 21
 I. Two-Dimensional Manifolds of Bounded Curvature 67

B
D'
C B'~------""C'

A'
A
Fig. 22

Fig. 23

The first device, which constitutes the method of cutting and pasting, consists
in cutting out from R a domain G and instead of it pasting in the triangle T. The
sides [A'B'], [A'C] and [B'C] of the triangle must be pasted to the arcs corre-
sponding to the sides [AB], [AC] and the arc [BC] on the boundary of H. In
the polyhedron R' which is obtained as a result the total angle at the vertex
corresponding to the point D turns out to be less than that of the original
polyhedron R. By virtue of this the given method can be used to remove vertices
with negative curvature. By means of this we also get rid of boundary points at
which the turn of the boundary is negative.
The triangle T in the construction we have described has a contracting map
onto H under which the boundary of T is mapped into the boundary of H with
arc lengths preserved. The existence of such a map is established by the follow-
ing construction. In the space 1E3 we construct a pyramid whose base is the
quadrangle H, and the point D is the foot of the perpendicular drawn from the
vertex of the pyramid to the plane of H (Fig. 23). Let t be the height of the
pyramid, and 9(t) the total angle at its vertex. As t -+ 0, 9(t) tends to a limit equal
to 1t + (x, where (X > 1t is the angle of H at D. Obviously 9(t) -+ 0 as t -+ 00. Since
1t + (X > 21t and 9(t) is a continuous function of t, there is a to such that 9(t o) =
21t. When t = to the lateral surface of the given pyramid is obviously isometric
to the triangle T. The map of the orthogonal projection of the lateral surface
of the pyramid corresponding to t = to is the required contracting map of T
onto H.
The second device enables us to go over to the case when the positive part of
the curvature of a polyhedron is concentrated at one point. It consists in the
following. Let X and Y be two vertices of the polyhedron R such that at each of
them the curvature is positive, and w(X) + w(Y) < 21t. We join X and Y by
68 Yu.G. Reshetnyak

a shortest curve L. For simplicity we assume that L does not contain boun-
dary points of R. We cut R along this shortest curve. We obtain a polyhedron
homeomorphic to a circular annulus. We paste the hole formed in R as a result
of the cut by a polyhedron constructed as follows. On the plane we construct a
triangle X'Y'Z' such that IX'Y'I = p(X, Y) = I(L), and the angles at X' and Y'
are equal to (X = lW(X) and P= lW(Y) respectively. Since (X + P< n:, such a
triangle exists. Let X"Y"Z" be a triangle equal to X'Y'Z' with IX'Y'I = IX"Y"I,
IX'Z'I = IX"Z"I, IY'Z'I = IY"Z"I. We paste the given triangles along the sides
starting from Z' and Z". We paste X'Z' to X"Z" and Y'Z'to Y"Z" so that Z' and
Z" are identified. As a result we obtain a polyhedron Q with a single vertex
Z obtained by pasting together Z' and Z". The total angle at its vertex Z is
equal to 2(n: - (X - P) = 2n: - w(X) - w(Y). Hence the curvature at Z is equal
to w(X) + w( Y). The points obtained by pasting together X' and X", Y' and Y"
will be denoted by X and Y respectively. The boundary of Q is split into two
parts by X and Y, and the length of each is equal to PR(X, Y) = I(L). Now we
paste Q to the polyhedron R cut along the shortest curve L so that one of the
arcs [X Y] forming the boundary of Q is pasted to one bank of the cut, and the
second to the other bank. The point X on Q must be pasted to the point X in R,
and Yon Q to the point Yin R. As a result we obtain a polyhedron R'. The total
angle in R' at X will be equal to 2n: - w(X) + 2(X = 2n: and similarly the total
angle in R' at Y is also equal to 2n:. Thus the points X and Y have ceased to be
vertices of the polyhedron. The curvature at Z is equal to w(X) + w(Y). It is not
difficult to see that no new vertices with positive curvature arise. Thus the
curvature of X and Y turns out to be collected at one point - the point Z. The
resulting polyhedron R' has a contracting map onto R. To obtain this it is
sufficient to be able to construct a contracting map of Q onto an interval oflength
PR(X, Y) under which each arc contained in one ofthe arcs [XY] into which the
boundary of Q is divided by X and Y goes over to an arc of the same length.
Without difficulty the given construction can be extended to the case when the
shortest curve L joining the points X and Y contains boundary points of R.

5.7. The Concept of a K.Polyhedron. A polyhedron in the sense of the defini-

tion given above is a two-dimensional manifold with intrinsic metric, pasted
together from some set of polyhedra taken on the usual Euclidean plane. In
some questions it turns out to be useful to consider polyhedra pasted together
from finitely many polygons lying on a surface of constant curvature.
Suppose we are given a real number K. The symbol E K will denote a sphere
in 1E3 of radius r = IIJK in the case K > 0, the usual Euclidean plane 1E2 in
the case K = 0, and the Lobachevskij plane with Gaussian curvature K when
K<O.
We first define the concept of a cone of class K (K is an arbitrary real num-
ber) over a metric space (M, pl. In the case K = 0 a cone of class K is a cone QM
in the sense of the definition in 5.1.
Suppose that K > O. We put u = n:IJK and consider the direct product of
the set M and the interval I t1 = (0, u), that is, the set of all pairs of the form (x, r),
 I. Two-Dimensional Manifolds of Bounded Curvature 69

where r E (0, 0"). For Z = (x, r) E M x fa we put r(z) = r. To the product M x fa

we add two elements 0 and 0', for which we put r(O) = 0, r(O') = 0" = n/JK.
Formally we can represent 0 as the totality of all pairs of the form (x, 0), where
x E M, and 0' as the set of pairs of the form (x, 0"). We shall denote the set
(M x fa) U {O, O'} by QK(M). In the set QK(M) we introduce a metric p~. If
Zl = Z2' we put P~(Zl' Z2) = o. If Zl = 0, Z2 =f:. 0, then P~(Zl' Z2) = P~(Z2' zd =
r(z2). If Zl = 0', Z2 =f:. 0', then P~(Zl' Z2) = P~(Z2' zd = 0" - r(z2). In particu-
lar, p~(O, 0') = 0". Let Zl = (Xl' rd, Z2 = (X2' r2)' where 0 < r l < 0", 0 < r 2 <
0". Then if P(Xl' X2) ~ n, we put P~(Zh Z2) = r l + r2. If P(Xl' X2) < n, then
P~(Zl' Z2) is defined as follows. On the sphere EK we construct a spherical trian-
gle ABC for which IABI = rl' IACI = r2' and L,.ABC = P(Xl' X2). In this case
we put P~(Zl' z) = IBCI.1t is easy to verify that the function p~ of pairs of points
of the set QK(M) obtained in this way is a metric.
Suppose that K < O. In this case the construction is exactly similar to the
one we used in the case K = O. We first construct the set QM' as we did
in 5.1. We define a metric p~ in the set QM' but not the same as in 5.1.
Namely, we put P~(Zl' Z2) = 0 in the case Zl = Z2. If Zl = 0, Z2 =f:. Zl' we put
P~(Zl' Z2) = P~(Z2' zd = r(z2)· Finally, if Zl = (Xl' rd, Z2 = (X2' r2), where
r l > 0, r2 > 0, then P~(Zl' Z2) is defined as follows. On the Lobachevskij plane
EK we construct a triangle ABC with sides IABI = r l , IACI = r 2, and
LBAC = p(x l , x 2). We put pHZ1' Z2) = IBCI. We can show that in this case
also p~ is a metric in QM. We shall also denote the metric space that arises in
this way by QK(M).
For a point x E M we shall denote the totality of all points (x, r) E QM by y(x).
In the case K > 0 we also assume that the points 0 and 0' belong to y(x). In the
case K < 0 we assume that 0 E y(x). We call the set y(x) the generator of the
cone QK(M) corresponding to the point x of the space M.
All that we said in 5.1 about intervals of generators extends automatically to
the case of cones QK(M) for arbitrary K.
In the case K > 0 the points 0 and 0' are called vertices of the cone QK(M).
We shall call 0 the upper vertex and 0' the lower vertex of QK(M). In the
case K < 0 the point 0 is called the vertex of QK(M). Suppose we are given a
number h > o. We shall denote the set of all points Z E QK(M) for which
r(z) < h by QK(M, h) and the set of all points Z E QK(M) such that r(z) = h
by rK(M, h).
Suppose that M is an interval [a, b], where a < b, of the number line IR and
that p(x, y) = Ix - yl for x, y EM. We put fJ = b - a. We shall denote the cone
QK(M) corresponding to the given space M by AK(fJ). The notation is
reasonable, since the cones QK([ah bl ]) and QK([a2' b2]) for which bl - a l =
b2 - a2 = fJ are isometric. We shall call fJ the total angle of the cone AK(fJ).
Suppose that M = Ce, where Ce is a circle in the plane 1E2, and that the metric
in Ce is defined as in 5.1. The cone QK(C9) is denoted by QK(fJ).
The results of 5.1 about the construction of cones Q(fJ) and A(fJ) by pasting
together a finite collection of cones A(fJi ) extends in an obvious way to the case
of cones QK(fJ) and AK(fJ).
70 Yu.G. Reshetnyak

By analogy with what we did in 5.1, we define cones QK(O, h), AK(O, h) and
curves TK(O, h) in QK(O) and AK(O). We have the relation
I[TK(O, h)] _ 0
h
as h -0.
In the case 0 = 2n the cone QK(O) is isometric to a plane E K' Any point of the
cone QK(O) other than its vertices has a neighbourhood isometric to a disc on
the plane E K. Any point of the cone AK(O) that does not belong to its boundary
rays also has a neighbourhood isometric to a disc on the plane E K. If X E AK(O)
lies on one of the boundary rays of AK(O) and is distinct from its vertex, then
some neighbourhood of X is isometric to a half-disc on the plane E K'
All that we said in 5.1 about shortest curves on the cone Q extends in an
obvious way to the case of the cones QK(O) (with the stipulation that in the case
K > 0, 0 > 2n if we extend a shortest curve beyond the vertex of the cone we
must require that the length of the resulting curve is not greater than (j =
nlft, otherwise it ceases to be a shortest curve).
Let M be a two-dimensional manifold with boundary, and K a real number.
We assume that an intrinsic metric is specified in M. We shall say that the metric
space (M, p) is a polyhedron of class K, or briefly a K-polyhedron, if the following
conditions are satisfied:
a) each internal point X of M has a neighbourhood V that for some 0 > 0,
e > 0 admits an isometric map ({J onto the cone QK(O, e) such that ((J(X) is a
vertex of this cone;
b) each point X E oM has a neighbourhood V isometric to the cone AK(O, e)
for some 0 > 0, e > O. Under the isometry map of V onto AK(O, e) the point X
goes over to a vertex of this cone.
As in the case of ordinary polyhedra, the neighbourhood V of X that satisfies
one of the conditions, a) or b), is called a canonical neighbourhood of this point.
The quantity 0 is called the total angle at X and is denoted by O(X). If X is an
interior point of M, we put w(X) = 2n - O(X). We shall call w(X) the curvature
of the point X of the K-polyhedron M. If X is a boundary point of M, we put
n - O(X) = K(X). The number K(X) is called the turn of the boundary at X.
Then X E M is called a vertex of the K -polyhedron M if it is an interior point of
M and its curvature is not equal to zero. A point X E M is called a boundary
vertex of K if X belongs to the boundary of M and the turn of the boundary at
X is not equal to zero.
All that we said in 5.2 about the structure of a polyhedron close to points
that are not vertices or boundary vertices automatically extends to the case of
K-polyhedra.
The theorem about triangulation of polyhedra is also true in the case of
K-polyhedra, except that instead of triangles on a Euclidean plane we should
take triangles on the plane E K •
The theorem about pasting together manifolds with polyhedral metric (Theo-
rem 5.2.3) also has a complete analogue for K-polyhedra. The concepts of a
 I. Two-Dimensional Manifolds of Bounded Curvature 71

polygonal line on a K-polyhedron and the area are defined in complete analogy
with the case considered in 5.2.
Let M be an arbitrary K-polyhedron and suppose that a set A c M is such
that its closure is compact (that is, in the terminology of 5.3, A is bounded). Then
A contains finitely many vertices. We shall denote the sum of the curvatures at
all the vertices of the polyhedron that belong to A by wK(A) and call it the excess
of the curvature with respect to K. A complete analogue of the curvature of a set
on a K-polyhedron is the set function w defined by
w(A) = wK(A) + Ku(A),
where u(A) is the area of the set.
Theorems on the total curvature of a compact polyhedron (Theorem 5.3.1),
like the Gauss-Bonnet theorem (Theorem 5.3.2), remain valid for polyhedra of
class K.
Aleksandrov's theorem on the characterization of closed convex polyhedra in
Euclidean space has its analogue for arbitrary spaces of constant curvature. We
do not give the exact statement here, since this result is contained in a general
theorem of Pogorelov (Pogorelov (1957».
The method of cutting and pasting can be successfully used also in the study
of extremal problems for K-polyhedra. An analogue of Theorem 5.6.1 for the
case K < 0 was established in Belinskij (1975).

Chapter 2
Different Ways of Defining Two-Dimensional
Manifolds of Bounded Curvature

§ 6. Axioms of a Two-Dimensional Manifold of Bounded

Curvature. Characterization of such Manifolds by Means
of Approximation by Polyhedra

6.1. Axioms of a Two-Dimensional Manifold of Bounded Curvature.

A. Preliminary remarks. We first make some remarks about Riemannian
manifolds. Let (R, p) be a two-dimensional manifold with intrinsic metric. We
assume that its metric is Riemannian. Let us take a point Xo E R arbitrarily.
Then, as we said above, there is a neighbourhood U of Xo such that any two
points X, Y E U can be joined by a shortest curve, and this shortest curve is
unique. We choose a neighbourhood U that satisfies this condition. We shall
assume that U is homeomorphic to a disc and that the quantity

Iwl(U) = f'K(X)' du(X)

u
72 Yu.G. Reshetnyak

is finite. (We recall that K(X) is the Gaussian curvature of the manifold at the
point X, and that the additive set function u(E) is its area.) Obviously there is a
neighbourhood U of the point Xo that satisfies all these conditions.
Let A, Band C be three distinct points belonging to U, and K, Land M the
shortest curves joining the points Band C, C and A, A and B respectively. We
assume that K, Land M are contained in U. If A, Band C do not lie on one
shortest curve, then K, Land M have no points in common other than their
ends. In fact, suppose that any two of the given shortest curves have points in
common other than A, Band C. Suppose, for example, that K and M have a
point X in common, other than B. The shortest curves K and M are contained
in U and since any two points of U are joined by a unique shortest curve the
arcs [BX] of K and M coincide. The shortest curves K and M have a common
tangent at X. Since through a point of a Riemannian manifold there passes a
unique shortest curve with a given tangent, it follows that one of the shortest
curves K and M is contained in the other and we see that the points A, Band C
lie on one shortest curve, contrary to our assumption. The shortest curves K, L
and M form a simple closed curve, which bounds a domain T homeomorphic to
a closed disc. We shall call this domain a geodesic triangle. Let IX, p and y be the
angles ofthe triangle Tat A, Band C respectively. Then from the Gauss-Bonnet
formula it follows that

t5(T) = IX + p+ y - 1t = w(T) = ff K(X) du(X).

T

The quantity <5(T) is called the excess of the triangle T.

Let T1 , T2 , ••. , Tm be an arbitrary set of geodesic triangles contained in U and
such that no two have interior points in common. Then
m m m
L t5(T;) = L w(T;) ~ L w+(T;) ~ w+(U) = Do < 00.
i=1 i=1 i=1
(6.1)

For any system of pairwise non-overlapping geodesic triangles contained in U

the sum of their excesses is thus not greater than some constant Do < 00. We
observe that we can make the sum on the left-hand side of (6.1) arbitrarily close
to w+(U) by covering the set of those X where K(X) > 0 by sufficiently small
geodesic triangles.
Two-dimensional manifolds of bounded curvature are defined as two-
dimensional manifolds with intrinsic metric that satisfy an additional condi-
tion relating to excesses of geodesic triangles. The statement of this condition
is suggested by the remark made above about two-dimensional Riemannian
manifolds.
In order to give a definition oftwo-dimensional manifolds of bounded curva-
ture it is necessary to give an exact meaning to the concept of the angle between
curves. For a start it is sufficient to be able to do this when the curves are
shortest curves starting from one point. Having the concept of angle, we can
define the concept of the excess of a triangle.
 I. Two-Dimensional Manifolds of Bounded Curvature 73

B. Definition of the angle between curves in a space with intrinsic metric. Let
(M, p) be an arbitrary metric space with intrinsic metric, and K and L simple
arcs with a common origin 0 in the space M. We specify points X E K and
¥ E L arbitrarily. On the plane 1E2 we construct a triangle O'X'¥' with sides

10'X'1 = p(O, X), 10'¥'1 = p(O, ¥), IX'¥'I = p(X, ¥),

and let y(X, ¥) be the angle at the vertex 0' of this triangle. We put p(O, X) = x,
p(O, ¥) = Y and p(X, ¥) = z. Then we have
x 2 + y2 _ Z2
y(X, ¥) = arccos --2-:----
xy

Using the notation we have introduced, we can now state the definition.

Definition 6.1. The limit

lim y(X, ¥), (6.2)
x .... o. y .... 0
if it exists, is called the angle between the curves K and L at the point O.
The limit (6.2) may not exist in very simple cases, even if the curves K and L
are sufficiently good from the viewpoint of the geometry of the space M (for
example, they are shortest curves).
Suppose, for example, that (M, p) is a Banach space. Then in order that
between any rectilinear intervals starting from an arbitrary point of this space
there should exist an angle in the sense of the given definition it is necessary and
sufficient that (M, p) should be a Hilbert space.
There naturally arises the idea, suggested by one of the conditions defining a
manifold of bounded curvature, of introducing the requirement that between
any two shortest curves starting from one point there should exist an angle in
the sense of Definition 6.1. However, there is no need to do this. We introduce
quantities that are always defined for an arbitrary pair of curves starting from
one point.

Definition 6.2. Let K and L be simple arcs in a space with intrinsic metric,
starting from some point O. Then in the notation introduced above we put
y= lim y(X, ¥), y = lim y(X, ¥). (6.3)
x .... o.y .... 0 x .... o.y .... 0
The quantities y and yare called the upper and lower angles between the curves
K and L at the point O.
Formally y and y represent the following. For h > 0 we denote by y(h) and
y(h) the least upper -and greatest lower bounds of y(X, ¥) on the set of all pairs
(X, ¥) such that X E K, ¥ E L, 0 < p(O, X) ~ h, 0 < p(O, ¥) ~ h. The function
y(h) is non-decreasing, the function E(h) is non-increasing, and
y = lim y(h), y = lim y(h).
h .... O - h .... O-
74 Yu.G. Reshetnyak

We have 0 ~ y ~ y ~ n, and y = y if and only if there exists a limit (6.2),

which is then equal to I = y. -
Definition 6.3. Let K be a simple arc starting from a point 0 of a space with
intrinsic metric p. Then if there exists a limit

lim y(X, Y)(X E K, Y E K), (6.4)

x .... o. y .... o

we shall say that the curve K has a direction at the point O.

Obviously, if the limit (6.4) exists, then it is equal to zero. The assertion "the
curve K has a direction at the point 0" is equivalent to the following: "there
exists an angle between the curves K and L at the point 0 in the case when
K=L".
If K is a shortest curve and 0 is one of its end-points, then for any points
X, Y E K either p(O, X) = p(O, Y) + p(Y, X) or p(O, y) = p(O, X) + p(X, Y)
and so y(X, Y) = O. Hence it follows that in this case lim y(X, Y) exists, so
X-O,y ...... o
any shortest curve has a direction at each of its end-points.
Let L be a simple arc of the space 1E 3 , and 0 its origin. It is easy to verify that
in order that L should have a direction at 0 in the sense of Definition 6.3 it is
necessary and sufficient that it should have a tangent at this point, that is, that
the ray OX with origin 0 passing through an arbitrary point X E K should
converge to some limiting arc as X ~ O.
The definitions of angle mentioned here have various modifications; see
Aleksandrov and Zalgaller (1962).

C. Axioms of two-dimensional manifolds of bounded curvature. The main ob-

ject of the investigation is defined as a metric space (R, p) satisfying Axioms A
and B below.
Axiom A. (R, p) is a two-dimensional manifold without boundary and its
metric p is intrinsic.
Henceforth we shall always assume that R is a space for which Axiom A is
satisfied.
A triangle in the manifold R is a figure T consisting of three points A, B, C
and three shortest curves K, L, M that join the points Band C, C and A, A and
B, respectively. The points A, B, C are called the vertices of the triangle T, and
the shortest curves K = [BC], L = [CA], M = [AB] its sides. A triangle with
vertices A, B, C will be denoted by ABC. We note that since two points in
a space with intrinsic metric can be joined by several and even infinitely many
shortest curves, a triangle is not uniquely determined by specifying its vertices.
Henceforth, from the context it will always be clear whether ABC means a
specific triangle with vertices A, B, C or an arbitrary one of such triangles.
Let G c R be a domain in R that is homeomorphic to an open disc on a
plane. Then we shall say that the triangle T is contained in the domain G if all
its sides lie in G.
 I. Two-Dimensional Manifolds of Bounded Curvature 75

Fig. 24 Fig. 25

A triangle T is said to be homeomorphic to a disc if its sides form a simple

closed curve that bounds a domain homeomorphic to a disc (Fig. 24). We shall
identify this domain with the triangle T. In accordance with this we shall say
that its points are points of the triangle T, the interior points of the domain are
interior points of the triangle, and its boundary points are boundary points of
the triangle T.
A triangle T is said to be simple if it is homeomorphic to a disc and satisfies
the following condition. No two boundary points of T can be joined by a curve
that lies outside the triangle and is shorter than a simple arc with ends at these
points that forms that part of the boundary of the triangle enveloped by the
given curve (see Fig. 25).
Simple triangles Tl and T2 contained in G will be said to be non-overlapping
if they do not have interior points in common.
Suppose we are given a triangle T = ABC, and that K = [BC], L = [CA],
M = [AB] are its sides. Let ii be the upper angle between the simple arcs Land
M at the point A, Pthe upper angle between K and M at the point B, and y the
upper angle between Land K at the point C. The quantity
g(ABC) = g(T) = ii + P+ y - 1[

is called the upper excess of the triangle T.

Axiom B. For any point X E R there is a neighbourhood G homeomorphic
to an open disc in \R 2 and a constant Q( G) < 00 such that for any system Tl , T2 ,
... , Tm of pairwise non-overlapping simple triangles contained in G we have the
inequality
m
L g(1/) :;:; Q(G) < 00.
i=l
(6.5)

Definition 6.4. A metric space satifying Axioms A and B is called a two-

dimensional manifold of bounded curvature.
Henceforth Axiom B will be called the axiom of boundedness of curvature.
We note that in the given statement of Axiom B we have considered sums of
excesses of triangles, and not the sums of their absolute values, as was done in
the original version of Axiom B given in Aleksandrov (1948b), in which it was
m
required that the sums L Ig(1I)1 should be bounded. The boundedness of the
i=l
sums of the absolute values of the excesses in a suitable neighbourhood of a
76 Yu.G. Reshetnyak

point X follows from Axioms A and B. The proof of this is obtained, however,
as a result of rather deep constructions that form the main part of the theory of
two-dimensional manifolds of bounded curvature.
We note that between any two shortest curves in a two-dimensional manifold
of bounded curvature starting from one point there exists an angle in the sense
of Definition 6.1. The proof of this is also obtained at a rather late stage of the
constructions (see 7.1).
In order that we can conclude from Definition 6.4 that a two-dimensional
Riemannian manifold is a manifold of bounded curvature it is necessary to
prove that for arbitrary shortest curves K and L starting from one point 0 the
quantity y defined by (6.3) is equal to the angle between K and L in the sense of
the definition given previously. This is actually so, but the proof of the given asser-
tion is not trivial. It can be obtained by an application of the classical apparatus
of differential geometry. Later we shall give its proof, which relies on certain
results relating to two-dimensional manifolds of bounded curvature (see 7.1).

6.2. Theorems on the Approximation of Two-Dimensional Manifolds of

Bounded Curvature by Manifolds with Polyhedral and Riemannian Metric. A
number of the main results of the theory of two-dimensional manifolds of
bounded curvature are theorems on the approximation of two-dimensional
manifolds of bounded curvature by manifolds with polyhedral metric and two-
dimensional Riemannian manifolds. These theorems give a complete character-
ization of two-dimensional manifolds of bounded curvature. By means of them
we can define the class of manifolds of the type under consideration as a result
of closing the class of Riemannian manifolds (or manifolds with polyhedral
metric) witl~ respect to a suitable limiting process.
Let R be an arbitrary two-dimensional manifold, and U a domain in R
homeomorphic to a closed disc. A metric p defined in U will be said to be
polyhedral if this metric is intrinsic and the metric space (U, p) is polyhedral
in the sense of the definitions in 5.2. Similarly, if (U, p) is a two-dimensional
Riemannian manifold with boundary, we shall say that p is a Riemannian metric
in U. If p is either a polyhedral or a Riemannian metric defined in U, then the
w w:
symbols P' IWpl, denote, respectively, the integral curvature, the absolute
curvature and the positive part of the curvature as set functions in the manifold
(U, pl. The symboIIICpl(aU) denotes the absolute turn of the boundary of the
polyhedron (U, pl. In the case when (U, p) is a Riemannian manifold, IICpl(aU)
denotes the integral of the modulus of the geodesic curvature of the boundary
with respect to arc length.
Let (p,,), n = 1, 2, ... , be an arbitrary sequence of metrics defined in an open
set U of a two-dimensional manifold R. We shall say that as n -+ 00 the metrics
P.. converge to a metric p specified in U if P.. (X, Y) -+ p(X, Y) as n -+ 00 for any
X, Y E U, and the convergence is uniform.
We recall that a set U in a topological space is called a neighbourhood of a
point X of this space if X is an interior point of U.
We shall call the following theorem the first theorem on approximation.
 I. Two-Dimensional Manifolds of Bounded Curvature 77

Theorem 6.2.1. Let (R, p) be a two-dimensional manifold of bounded curvature.

For any point Xo of it we can construct a neighbourhood U of this point that
is homeomorphic to a closed disc and a sequence of metrics (Pn), n = 1, 2, ... ,
defined in U such that (U, Pn) is a polyhedron homeomorphic to a disc for each n,
Pn(X, Y) -+ p(X, Y) as n -+ 00 in the domain U, and there is a constant C < 00
such that for all n
IWpnl(U) + IKPnl(OU) ~ C.
A proof of Theorem 6.2.1 is given in Aleksandrov and Zalgaller (1962). We
note that in that work the theorem is stated in a weaker form than the one here
(boundedness is guaranteed only for absolute curvatures), but from the argu-
ments given there it follows that the given statement of the theorem is true.
Remark. Theorem 6.2.1 (Aleksandrov and Zalgaller (1962)) remains true if in
its statement instead of polyhedral metrics we consider Riemannian metrics.
The given theorem establishes a necessary condition that is satisfied by any
two-dimensional manifold of bounded curvature. This condition is also suffi-
cient. In the part that touches on the sufficiency, the conditions imposed on the
sequence of polyhedral metrics Pn can be weakened. Namely, the following theo-
rem, which we shall call the second theorem on approximation, is true.
Theorem 6.2.2. Let (R, p) be a two-dimensional manifold without boundary,
endowed with an intrinsic metric p. We assume that for any point Xo E R we can
find a neighbourhood U of this point and a sequence of polyhedral metrics Pn'
n = 1, 2, ... , specified in U that converges to the metric P and is such that the
sequence (w:JU)), n = 1, 2, ... , is bounded. Then (R, p) is a two-dimensional
manifold of bounded curvature.
Theorem 6.2.1 establishes the possibility of a local approximation of the
metric of a two-dimensional manifold of bounded curvature by polyhedral or
Riemannian metrics in a neighbourhood of an arbitrary point of the manifold.
There arises the question, is a similar kind of theorem true about the approxi-
mation of a metric by polyhedral or Riemannian metrics defined on the whole
manifold? The answer to this question is positive. The required approximations
are easily constructed by means of theorems on the analytic representation of a
two-dimensional manifold of bounded curvature, to which § 7 of this article is
devoted.

6.3. Proof of the First Theorem on Approximation. The complete proof of

Theorem 6.2.1 presented in Aleksandrov and Zalgaller (1962) turns out to be
rather extensive, and here we give only an outline of it, mentioning some of
the most important parts of it. The arguments on which the proof of Theorem
6.2.1 is based are conceptually a generalization of the technique worked out by
Aleksandrov in his investigation of two-dimensional manifolds of non-negative
curvature. We derive an analogue of Theorem 6.2.1, touching on approximation
by Riemannian metrics, from a theorem on the analytic representation of a
two-dimensional manifold of bounded curvature presented in § 7.
78 Yu.G. Reshetnyak

A *. Estimate of the distortion of the angles of a triangle under transition to a

development of it. Henceforth R denotes a metric space with intrinsic metric p.
We shall assume that R is locally compact and that all the later constructions
refer to a small domain G of R in which any two points can be joined by a
shortest curve.
For points X, Y E G instead of p(X, Y) = I we shall simply write IXYI = I.
In the arguments on which the proof of Theorem 6.2.1 relies, an essential role
is played by a method that consists in comparing a triangle in the space (R, p)
with a flat triangle having the same lengths of sides. Let T be a triangle in the
space (R, p), A, B, C its vertices, and the shortest curves K = [BC], L = [CA],
M = [AB] its sides. Let a = IBq, b = ICAI, c = IABI be the lengths of the sides
of T. On the plane 1E2 we construct a flat triangle A'B'C', the lengths of whose
sides are IB'C'I = a, IC'A'I = b, IA'B'I = c. Such a triangle can certainly be con-
structed, since we only need it to satisfy the inequalities a ~ b + c, b ~ c + a,
C ~ a + b. These inequalities obviously hold. We shall call A'B'C' a development
of the triangle T. Henceforth we shall assume that to each vertex and each side
of T there corresponds a vertex or side of the same name of the triangle A'B'C',
that is, the vertex A corresponds to the vertex A', the side [Be] to [B'C'], and
so on. Let oc o, Po, Yo be the angles at the vertices A', B', C' ofthe triangle A'B'C'.
We observe that A'B'C' may degenerate into an interval. In this case one of the
numbers OC o, Po, Yo is equal to 7t and the other two are equal to zero. Let ii, p, Y
be the upper angles between the sides of the triangle T at its vertices A, B, C.
Then there is defined the number S(T)=ii + p+ y - 7t=(ii - oc o) + (P - Po) +
(y - Yo), the upper excess of the triangle T.
Suppose we are given a triangle T with vertices A, B, C in the space R. We
arbitrarily take points X and Y on the sides AB and AC of the given triangle,
other than A. We shall join X and Y by a shortest curve XY. We assume that T
is contained in a domain, any two points of which can be joined by a shortest
curve, and so such shortest curves exist. Let S(AXY) be the upper excess of the
triangle AXY. We put

v1(T) = sup [inf S(AXY)].

XeAB,YeAC [XYj

The greatest lower bound (in square brackets) is taken with respect to the set of
all triangles whose sides are the intervals [AX] and [AY] of the sides AB and
AC of T and an arbitrary shortest curve joining X and Y (there may be many
such shortest curves). Similarly we define the quantities v:(T) and vt(T). We
put v+(T) = max{v1(T), v: (T), vt(T)}.
Lemma 6.3.1. In the notation introduced above, we have
ii - OC o ~ v1(T), P- Po ~ v: (T), y - Yo ~ vt(T).
We note that the lemma refers to an arbitrary metric space and in its formu-
lation it is not assumed that the given manifolds are manifolds of bounded
curvature.
The proof of Lemma 6.3.1 will be carried out in 6.4.
 I. Two-Dimensional Manifolds of Bounded Curvature 79

K L

x
L

Fig. 26

B*. Some methods of transforming shortest curves. In the constructions de-

scribed later, two methods are of essential use. The first is called the removal of
superfluous intersections and consists in the following. We assume that the
shortest curves K and L have at least two points in common. Let X and Y be
the leftmost and rightmost of these points on the curve K. The arcs [XY] on K
and L have the same length, because otherwise by replacing of the arc [XY] on
one of them by the arc [XY] taken on the other we could reduce the length of
one of the curves K and L without changing its ends, which contradicts the fact
that they are shortest curves. Replacing the arc [XY] of K by the arc [XY] of
L, we obtain a shortest curve K'. The common points of K' and L form a simple
arc lying on each of them. Replacing the pair of shortest curves K, L by the pair
K', L, we can remove the singularities in a mutual arrangement of the type of
those shown in Fig. 26.
We assume that there is a finite collection of shortest curves, no two of which
have common intersections, that is, the common part oftwo of them is either an
empty set, or it consists of one point, or it is a simple arc. If to this collection
we add a new shortest curve, then by replacing some arcs of it we can arrange
that the resulting collection of shortest curves does not have superfluous
intersections.
We assume that R is a two-dimensional manifold with intrinsic metric. Let L
be a shortest curve in R joining the points A, B E R. We orient L so that A is its
beginning and B is its end. We shall assume that L lies in a domain G of R
homeomorphic to an open disc, and for any point X E L the distance to the
boundary is greater than the length of L. We assume that G is oriented. Then we
can define what is meant by L being left or right. We shall say that L is the
leftmost (rightmost) shortest curve joining the points A and B if any simple arc
whose ends are arbitrary points X, Y ELand which lies to the left (respectively,
right) of L has length greater than that of the arc [XY] of L. For any points A,
BEG, each of which is distant from the boundary of G by a distance greater
than p(A, B), there are leftmost and rightmost shortest curves joining A and B.
The leftmost shortest curve joining A and B is unique. The same is true for the
rightmost shortest curve.
The concepts of leftmost and rightmost shortest curves make it possible
to select uniquely, in the set of all shortest curves joining given points, one
80 Yu.G. Reshetnyak

definite shortest curve. The use of these concepts constitutes the second method
of transforming shortest curves.
C*. A lemma about partitioning on a triangle. The next step in the proof of
Theorem 6.2.1 consists in establishing that each point X of a two-dimensional
manifold of bounded curvature has a neighbourhood that can be partitioned
into arbitrarily fine simple triangles.
As a preliminary we introduce certain concepts. Next we consider domains
homeomorphic to a closed disc in a two-dimensional manifold R with intrinsic
metric of bounded curvature.
Let G be such a domain. We shall say that G is convex if for any two points
X, Y E G there is a shortest curve contained in G that joins them. If any shortest
curve joining two arbitrary points X, Y EGis contained in G, then G is said to
be completely convex.
A domain G is said to be boundedly convex if its boundary is rectifiable and
G can be enclosed in a domain U homeomorphic to an open disc so that the
distance from the boundary of G to the boundary of the domain U exceeds the
perimeter of G by more than four times, and the following condition is satisfied.
For any two points X, Y E iJG there is no simple arc with ends X and Y that
goes outside G and is shorter than the part of the boundary of G bounded by
this arc.
The domain G is said to be absolutely convex if it is completely convex and
boundedly convex.
We shall call a point X ERa transit point if there is a shortest curve for which
X is an interior point.
Transit points in a manifold R form an everywhere dense set. In fact, let
X E R. Then in any neighbourhood of X there is a shortest curve, and hence also
transit points of the manifold. On any simple arc and any simple closed curve in
R the transit points form an everywhere dense set. In fact, let L be a simple arc,
and X an arbitrary interior point of L. Then for any neighbourhood U of X we
can find a neighbourhood V c U of this point such that L partitions V. Let W
be a neighbourhood of X such that any two points Y, Z E W can be joined by a
shortest curve, and any such shortest curve is contained in V. We choose points
Y, Z E W that lie on different sides of L. The shortest curve joining them inter-
sects L. The point of intersection of this shortest curve with L is a transit point
contained in the neighbourhood U of the point X.
The result of this step is included in the following proposition.
Lemma 6.3.2. Suppose that a domain G in a two-dimensional manifold of
bounded curvature, homeomorphic to a closed disc, is absolutely convex. We as-
sume that G is a polygon, that is, the boundary of G is a closed polygonal line.
Then for any e > 0 we can find a finite system of pairwise non-overlapping simple
triangles such that G is their union, the diameter of any of these triangles does not
exceed e, and the vertices of all these triangles other than boundary vertices of G
are transit points and at each of them any side is less than the sum of the other
two.
 I. Two-Dimensional Manifolds of Bounded Curvature 81

The proof of Lemma 6.3.2 in its technical aspect is rather cumbersome. It

is essential to use the axiom of boundedness of curvature. To its full extent
the reader can become acquainted with it in the monograph Aleksandrov and
Zalgaller (1962). Here we just outline the proof of Lemma 6.3.2 in very general
terms.
We first prove the following proposition.
Lemma 6.3.3. For any point X of a two-dimensional manifold of bounded
curvature, whatever the values of 8 > 0, 15 > 0, there is a closed neighbourhood V
of this point contained in the disc B(X, 15) that is absolutely convex and homeo-
morphic to a closed disc on a plane, where the boundary of V is a simple closed
polygonal line whose length is less than 8, and all the vertices are transit points.
In order to prove Lemma 6.3.3 it is first necessary to establish that for any
8> 0 and 15 > 0 we can construct a neighbourhood U of the point X lying in the
disc B(X, 15) that is bounded by a simple closed curve whose length is less than
8. The proof of this assertion uses the axiom of bounded ness of curvature, and if
this axiom is not satisfied, then generally speaking the assertion is not true, as
the following example shows.
On the plane 1R2 we consider the disc B(O, 1). Let M be the topological
manifold obtained from 1R2 by identifying points of B(O, 1) so that to the disc
B(O, 1) there corresponds one point of M, which we denote by O. Obviously M
is a two-dimensional manifold homeomorphic to 1R2. In M we introduce an
intrinsic metric p so that in a neighbourhood of any point X =F 0 the metric p
coincides with the natural metric of a plane. The metric p is uniquely determined
by the given condition. It is not difficult to see that in the two-dimensional
manifold with intrinsic metric that we have constructed any closed curve inside
which the point 0 lies has length not less than 2n.
After we have established that any point has a small neighbourhood with
arbitrarily small perimeter, the validity of Lemma 6.3.3 is established as follows.
Suppose we are given 8 > 0 and 15 > O. We first construct a neighbourhood U
lying in the disc B(X, 15) and such that the boundary of U is a simple closed
curve r whose length is less than 8 1 , Here 8 1 is determined by 8 and 15. Transit
points lying on the curve r form on it an everywhere dense subset, and so it can
be partitioned into finitely many arbitrarily small arcs whose ends are transit
points. Let us specify such a partitioning arbitrarily. Joining the ends of the arcs
of the partition by shortest curves so that no superfluous intersections arise (see
B*), we obtain a closed polygonal line r that encircles the point X. Without
loss of generality we can assume that r is simple (if not, then by discarding
loops that are part of r it is converted into a simple closed polygonal line). If
the points on r that were joined to obtain r are situated sufficiently densely on
r, then r is contained in B(X, 15). All the vertices of r are transit points. We
shall consider closed curves in R that envelop r and lie in B(X, 15). Among them
there are curves of shortest length. If 8 1 is sufficiently small, we can guarantee
that such a "minimal" curve does not touch the boundary of B(X, 15). In this case
such a curve is a polygonal line, and its vertices are vertices of r. Among the
82 Yu.O. Reshetnyak

closed curves of shortest length that envelop r there is an outermost one. It

bounds the required neighbourhood of the point X.
The proof of Lemma 6.3.2 is obtained as follows. Suppose we are given 6 > O.
For an arbitrary point X E G we denote by V(X) a neighbourhood of X that
satisfies all the conditions of Lemma 6.3.3 and corresponds to the given 6 and
() = 6/2. By a theorem of Borel, G is covered by finitely many neighbourhoods
V(X). Let VI' V2 , ... , Vm be these neighbourhoods. All the possible intersections
of the domains V. with the set G form a finite collection of convex polygons
whose union is G. Subdividing these polygons into triangles, we obtain the
partition whose existence is asserted by Lemma 6.3.2.
In conclusion, a few words about the proof of Lemma 6.3.3. As we mentioned
above, it reduces to proving that any point X of a two-dimensional manifold of
bounded curvature has arbitrarily small neighbourhoods with an arbitrarily
small perimeter. The existence of such a neighbourhood is proved as follows.
We first establish that for any r > 0 there is a finite collection of triangles T1 ,
T2 , ••• , Tm contained in the disc B(X, r) that satisfies the following conditions.
The point X is a vertex of each of the triangles Th T2 , ... , Tm , and the sides of a
triangle 7;, i = 1, 2, ... , m, starting from X have no common points other than
X. The common part of two adjacent triangles is an arc beginning at X and
lying on a side of each of them (see Fig. 27). Finally, it is still required that the
side of 7; opposite to X is shorter than each of the other two for any i = 1,2, ... ,
m. From the fact that 7; c B(X, r) for any i it follows that the lengths of the sides
of the triangles 7; starting from X do not exceed r.
The construction of the collection of triangles T1 , T2 , ... , Tm satisfying the
conditions listed requires rather laborious arguments. However, we use only the
fact that R is a two-dimensional manifold with intrinsic metric. The axiom of
boundedness of curvature is not used.
We shall say that a triangle 7; is boundedly convex at a point X if there is a
neighbourhood V of X such that any shortest curve joining points P and Q that
belong to V and lie on sides of 7; is contained in 7;.
We assume that among the triangles T1 , T2 , ... , Tm there is one thatis not
boundedly convex. Let 7; be this triangle, and Yand Z its vertices other than X.
Then in any neighbourhood V of X there are points P E [XY] and Q E [XZ]
that can be joined by a shortest curve lying outside the triangle 7;. This shortest
curve together with the intervals [PY] and [QZ] on the sides of 7; and its side

Fig. 27
 I. Two-Dimensional Manifolds of Bounded Curvature 83

x
Fig. 28 Fig. 29

[YZ] form a closed polygonal line r. If P and Q are sufficiently close to X, the
situation represented in Fig. 28 cannot hold, and r encircles the point X, as
represented in Fig. 29. (Let,., = min{IXPI, IXQI}; in the case represented in Fig.
29 the shortest curve [PQ] goes out from the disc of radius,., and so its length
is not less than 2,., -IXPI -IXQI. If IXPI < ,.,/2, IXQI < ,.,/2, then the situation
becomes impossible, because otherwise it turns out that S([PQ]) > IXPI +
IXQI.) The length of r does not exceed the perimeter of the triangle 7;. This, in
turn, does not exceed 3r. Choosing r = 8/3, we deduce that r is the required
polygonal line.
Henceforth we shall assume that each of the triangles T1 , T2 , ••• , Tm is
boundedly convex at the point X. We assume that the upper angle of each of
them at X is less than n. We specify a sufficiently small h > 0 and layoff on the
sides of the triangles 1'; starting from X intervals equal to h. Joining the ends of
these intervals by shortest curves, we obtain a closed polygonal line. Thanks to
the condition of bounded convexity at the point X for sufficiently small h, each
of these shortest curves goes into the triangle 7;. By virtue of the fact that the
upper angle of 7; is less than n for small h, this shortest curve does not go
through the point X. It is not difficult to see that the polygonal line thus con-
structed is the required one if h is sufficiently small.
Difficulties arise when the triangle 7;, being boundedly convex at the point X,
has angle at this point equal to n. We cannot guarantee that for points P and Q
sufficiently close to X and Y, taken on the sides of 7; starting from X, the
shortest curve [PQ] does not go through X. The condition of boundedness of
curvature enables us to prove that the number of such triangles 7; does not
exceed some constant depending only on the point X. Detailed analysis of this
requires a consideration of several possibilities. Let us restrict ourselves to the
case that is the principal one. We observe that the angle of development of
7; corresponding to X does not exceed n/3, since this is the smallest of the
angles of development. Since ii = n, on the basis of Lemma 6.3.1 we deduce that
vt(7;) ~ 2n/3. Hence it follows that among the triangles XPQ, where P and Q
are points on the sides of the triangle 7; starting from X, there is one whose
excess is greater than n/3. In the basic case when the side PQ of this triangle goes
inside the triangle 7; we obtain a simple triangle with excess greater than n/3.
84 Yu.G. Reshetnyak

Fig. 30

We can assume that all the constructions are carried out in a neighbourhood G
of the point X such that the sum of the excesses of any finitely many non-
overlapping simple triangles lying in this neighbourhood does not exceed some
constant Q < 00. Hence it follows that the number of triangles T; for which
ii = n does not exceed the number 3Q/n. The desired polygonal line is now
constructed as follows (Fig. 30). Let us specify a sufficiently small number h > O.
For each i = 1, 2, ... , m we define an arc L j lying in the triangle T; and not
containing the point X. If the angle of T; at X is different from n, then L j is a
shortest curve joining points on the sides of T; lying at a distance h from the
point X. If iij = n, then L j consists of the side of T; lying opposite to X, and
intervals of two other sides consisting of points P lying on these sides and such
that IPXI ~ h. The arcs L t , L 2 , ... , Lm form a closed polygonal line r* encircl-
ing the point X. We note that s(Lj) ::s;; 2h in the case iij < nand s(L j ) ::s;; 3r in the
case iij ~ n (we recall that the lengths of the sides of T; starting from X do not
exceed r, and the side lying opposite to X is very short). Hence we conclude that
s(L) ::s;; 2mh + 9Qr/n. Since h > 0 and r > 0 are arbitrary, it follows that s(L) can
be made less than any preassigned 6 > 0, as required.

D*. Construction of an approximating sequence of polyhedra. We now de-

scribe the final part of the proof of Theorem 6.2.1. Since we intend to explain
only the idea of the proof, some details of a technical character are omitted here.
Let G be a neighbourhood of the point X homeomorphic to a closed disc,
such that its boundary is a simple closed polygonal line. We shall assume that G
is absolutely convex and such that for any system of pairwise non-overlapping
simple triangles contained in G the sum of their excesses does not exceed some
constant Q < 00. The existence of a neighbourhood G satisfying all these condi-
tions follows from Lemma 6.3.3.
Let At, A 2 , ... , Am be all the vertices of the closed polygonal line r that
bounds the domain G.
Let J be a partition of G into pairwise non-overlapping simple triangles Tt ,
T2 , ..• , T, such that each of the points Ai is a vertex of one of these triangles, and
the vertices of the triangles T; other than those that lie on the boundary of G are
transit points, and in each of the triangles any side is less than the sum of the
other two. Let d(J) be the largest diameter of the triangles T;, i = 1, 2, ... , r.
According to Lemma 6.3.2, whatever the number 6 > 0 the partition J can be
 I. Two-Dimensional Manifolds of Bounded Curvature 8S

chosen so that d(A) is less than E:. Let 1;* be a development of the triangle
1;. Thanks to the condition that in 1; each side is less than the sum of the
other two, the triangle 1;* does not degenerate into an interval. Pasting together
the triangles 1;* in the order in which the triangles 1; adjoin each other in
the composition of the partition A, we obtain a two-dimensional polyhedron,
which we denote by G4 • We map the polyhedron G4 topologically onto G
so that the image of the triangle 1;*, i = 1, 2, ... , r, is the triangle 1; and
the sides of 1;* are mapped isometrically onto the corresponding sides of 1;.
For arbitrary points X, Y E G we denote by P4(X, Y) the distance on the
polyhedron G4 between the points that go into X and Y under the given
topological map.
Specifying a sequence AI' A 2 , ••• , Ay of partitions of G such that d(Ay) -+ 0 as
v -+ 00, by means of the construction we have described we obtain a sequence of
polyhedral metrics py = P4, defined in G. As v -+ 00 these metrics converge to
the metric p of the domain G. The proof of the last assertion uses the axiom of
boundedness of curvature. Since this assertion by itself is quite likely, we shall
not dwell on how it can be proved.
The partition A of G into triangles obtained by applying Lemma 6.3.2 can-
not, generally speaking, be a triangulation of G in the sense of the definition in
3.1, since the vertices of the triangles defined may be interior points ofthe sides
of other triangles (see Fig. 31). (In the given case our terminology differs from
that used in Aleksandrov and Zalgaller (1962).) In this connection we need to
define more exactly what such a vertex of the partition A is. Let S be the union
of the sides of the triangles forming the partition. A point XES will be called a
vertex of A if it is either a vertex of one of the triangles forming A or such that
no neighbourhood of it with respect to S is a simple arc.
To complete the proof of Theorem 6.2.1 it is sufficient to establish that
the absolute curvature and absolute turn of the boundary of G4 are bounded
above by a constant that depends only on G. The main instrument for con-
structing the required estimate is the Gauss-Bonnet formula for polyhedra. In
addition, we use the inequalities of Lemma 6.3.1 and the singularities of the
partition A of G, namely the fact that all the vertices of the triangles that consti-
tute A and are not boundary points of G are transit points.
Namely, the following propositions are true.

/Fig.31
86 Yu.G. Reshetnyak

I. For the triangles Tl , T2 , ••• , Tm that constitute A we have

m
L v+('1I) ~ .Qo·
i=l
(6.6)

II. Let X be a vertex of one ofthe triangles of the partition A, and ii l , ii 2 , ... ,
iin the angles at X of all the triangles of the partition that meet at X. If for some
i the point X is an interior point of a side of a triangle '11 of the given partition,
we put iii = n. Let

O,iX) = ii l + ii2 + ... + iin • (6.7)

If X is an interior point of G, then
0AX) ~ 2n. (6.8)
If X belongs to the boundary of G, and X is distinct from each of the points A l ,
A 2 , ... , Am, then
0AX) ~ n. (6.9)
We shall carry out the proof of Propositions I and II later. For now we just
mention that the fulfilment ofthe inequalities (6.8) and (6.9) follows from the fact
that the points X to which they refer are transit points.
Using Propositions I and II, we now show how to obtain the required esti-
mates for Iwl(G A) + IKI(t3GA ).
We use the Gauss-Bonnet formula for polyhedra. Let us apply it to the
polyhedron GA, which is homeomorphic to a disc, and so its Euler characteristic
is equal to 1. On the basis of Theorem 5.3.2 we have
w(G A ) + ,,(aG A ) = 2n.
Moreover,
w(G A ) = w+(G A ) - w-(G A ),
,,(t3GA ) = ,,+(t3GA ) - ,,-(t3GA )·
Hence
w-(G A ) + K-(aG A ) = w+(G A ) + ,,+(t3GA ) - 2n. (6.10)
To obtain upper bounds for w-(G A ) and ,,-(t3G A) it is sufficient to obtain
upper bounds for w+(G A ) and K+(aG A ). These bounds are established by means
of Propositions I and II stated above.
Let us find a bound for w+(G A ). Let X be an arbitrary vertex of the poly-
hedron GA , and X the corresponding vertex of the partition~. Let 11 1 , 112 , . . . ,
11k be all the triangles of the partition A on whose boundary X lies. Let iii be the
angle at X of the triangle 11., and (Xi the angle at X of the triangle 11~. If X is an
.
interior point of a side of 11*, we assume that iii = -(Xi = n. In this case X is an
interior point of a side of 11. and iii - (Xi = O. If X is a vertex of 11., then it
follows from Lemma 6.3.1 that iii - (Xi ~ v+ ('11). Summing over i = 1, 2, ... , k,
 I. Two-Dimensional Manifolds of Bounded Curvature 87

we obtain
8AX) - 8(X) ~ L; ,,+(1;), (6.11)

where in the sum on the right we take only those triangles 7; for which the point
X is a vertex. If X is an interior vertex of the paritition A, then by Proposition
II we have 8AX) ~ 2n, from which it follows that in this case
w(X) = 2n - 8(X) ~ 8AX) - 8(X) ~ L; "+(7;).
Since the right-hand side of the last inequality is non-negative, this enables us to
conclude that

w+(X) ~ L; "+(7;).
Let us sum the given inequality over the set of all interior vertices of the poly-
hedron G,1. For each i = 1, 2, ... , r the triangle 7; in the sum obtained on the
right occurs as many times as it has vertices lying inside G, that is, at most three
times, and as a result we arrive at the following inequality:
r
w+(G,1) ~ 3 L "+(7;).
;=1

By Proposition I the sum on the right does not exceed Qo, and we obtain the
following bound for w+(G,1):
(6.12)
Let X be an arbitrary boundary vertex of the partition A, and X, as above,
the point corresponding to it on the polyhedron G,1. We again use the inequality
(6.11). If X is not one of the points A;, then by Proposition II we have 8AX) ~ n,
and so in this case we have
(6.13)
(the summation is taken over all triangles 7; for which X is a vertex). If X = A;,
then 8,1 (X) ~ 0 and we obtain the bound
K(X) = n - 8(X) ~ n + 8,1(X) - 8(X) ~ L "+(7;) + n. (6.14)
i

The right-hand sides of(6.13) and (6.14) are non-negative, so it follows that these
inequalities remain true if the expression on the left is replaced by [K(X)] +.
Summing over the set of all boundary vertices of G,1, we obtain

K+(OG,1) = L [K(X)]+ ~ mn + L "+(7;) ~ mn + 3Q o (6.15)

i i

and the required bound for K+(OG,1) is established. From the equality (6.10) and
the inequalities (6.12) and (6.15) it then follows that
w-(G,1) + K-(OG,1) ~ 6Q o + (m - 2)n.
88 Yu.G. Reshetnyak

Hence, finally, we obtain the required bound

Iwl(G.:!) + 1"1 (oG.:!) ~ 12Do + (2m - 2)n.
To complete the proof of Theorem 6.2.1 it remains to establish the validity of
Propositions I and II.
Proof of Proposition I. Let T = ABC be a simple triangle in a two-
dimensional manifold. We assume that v': (T) > O. We shall consider all possible
pairs of points X E [AB], Y E [AC], X =F A, Y =F A, for which inf (j(AXY) is
IXY]

°
positive. To find the quantity v': (T) only such pairs need to be taken into
consideration. Let X and Y be such that (j(AXY) > for any shortest curve
[XY]. Then none of these shortest curves passes through A. By virtue of the
bounded convexity of the triangle T (we recall that this condition occurs in the
definition of a simple triangle), there is a shortest curve that joins the points X
and Y and lies in T. Let us find the rightmost of such shortest curves (the
rightmost in relation to the arc [X A] u [A Y]). It is not difficult to show that the
triangle AXY, which has as sides this shortest curve and the intervals [AX] and
[AY] of sides of the triangle T, is simple. We have (j(AXY) ~ inf (j(AXY). This
IXY]
enables us to conclude that v': (T) does not exceed the least upper bound of the
excesses of the simple triangles AX Y contained in T. Moreover, it follows that
v+(T) does not exceed the least upper bound of the excesses of the simple
triangles contained in T. From what we have said it obviously follows that for
the triangles Tl , T2 , ••• , T, that constitute the partition A not only the sum of
their excesses, but also the sum

does not exceed the constant Do. Thus Proposition I is proved.

Proof of Proposition II. Let us restrict ourselves to an assertion relating to
the case when X is an interior point of a side of the polyhedron G. Let [XXo],
[XX1 ], ••• , [XXm] be the sides of triangles of the partition A for which X is a
common vertex. Let iii be the angle between the shortest curves [XXi- l ] and
[XXJ at the point X. On the plane we specify a point 0 arbitrarily and con-
struct non-overlapping triangles Ai-10A i, i = 1, 2, ... , m, such that IOAil =
lXXii, i = 0, 1, ... , m, and LAi-10A i = iii (Figs. 32,33). We assume that iil +
ii2 + ... + iim < n. We denote by (j the smallest of the lengths of the intervals
[OAJ. Let Bo and Bm be the points on the rays OAo and OAm that are at a
distance (j from the point o. The rays OA i , i = 1, 2, ... , m - 1, lie inside the
angle AoOAm. By the assumption we have made, the value of this angle is less
than n, and so it is convex. Consequently, the interval joining Bo and Bm inter-
sects each ofthese arcs. Let Bi be the point of the arc OA i that lies on the interval
[BoBm]. We put di = IOBil, i = 0, 1,2, ... , m, do = dm = c5. Let 1= IBoBml, 'i =
IBi-1Bil, I = 11 + 12 + ... + 1m. Obviously we have I < do + dm(the inequality is
strict!). We put do + dm-I = 8. On the shortest curve [XXa, i = 0,1,2, ... , m,
 I. Two-Dimensional Manifolds of Bounded Curvature 89

Fig. 32

A1

a
Fig. 33

we mark the point li,n such that IXli,nl = ddn (n is a natural number). Let Ai,n
be the length of the shortest curve [li-l,n li,n]. We construct a development of
the triangle Xli-l,n li,n. Let Yi,n be the angle of this development at the vertex
corresponding to the point X. We have
(6.16)
n .... oo

We construct the simple triangle obtained from the development of the triangle
Xli-l,n li,n by a similarity transformation with coefficient of expansion n. Two
sides of this triangle are equal to dj-l and di, and the third is equal to nAi,n and
the angle opposite it is equal to Yi,n' In the triangle OBi- 1 Bi the sides [OBj-l]
and [o.BJ are equal to di - 1 and di respectively, and the angle between them is
equal to ai' The inequality (6.16) enables us to conclude that Ii = IBj-lBd ~
lim n).i,n' So there is a number no such that nAi,n < Ii + (elm) when n ~ no for
n .... oo
each i = 1, ... , m. Hence we conclude that when n ~ no
I. e 1 1
L Ai,n < i=lL ~n + -n = -(I
m m
IYO,n Ym,nl ~ + e) = -(do + dm) = IYo,nOI + IOYm,nl.
i=l n n
We obtain a contradiction with the fact that the arc [XoX] u [XXm] is a
shortest curve. For the case when X is a boundary vertex Proposition II is
proved.
In the case when X is an interior vertex the triangles 1i are situated as shown
in Fig. 34. The dotted line denotes the shortest curve passing through the point
X (X is a transit point!). This shortest curve splits the neighbourhood of X into
90 Yu.G. Reshetnyak

L
A B

Fig. 34 Fig. 35

two semicircles. Applying the inequality we have proved to each of these semi-
circles, we conclude that in the given case al + a2 + ... + am ~ 2n.

6.4. Proof of Lemma 6.3.1.

Lemma 6.3.1 contains an estimate of the difference between the angles of a
triangle in a space with intrinsic metric and the corresponding angles of its
development. It is obvious from the preceding text that this estimate is used
substantially in the proof of the first approximation theorem.
Next, R denotes a locally compact metric space with intrinsic metric, and p
denotes this metric. All the arguments relate to a domain of the space R in which

°
any two points can be joined by a shortest curve.
Lemma A. Let L = [AB] be a shortest curve in R, X(s), ~ s ~ I, its para-
metrization, where the parameter s is arc length, and Y a point outside L. For
s E [0, I] we put z(s) = IYX(s)l. We fix arbitrarily a point X = X(s), where <
s ~ I, and a shortest curve L, and join X and Y by a shortest curve. We denote
°
the greatest lower bound of the upper angles at X between the arc [AX] and all
possible shortest curves [XY] by ~(s). For any s E [0, /] we have
dZ)
( -d () I. z(s) - z(s - h) J( )
.
s = 1m h ~ cos., s .
S l.l.l h- +0

Proof (see Fig. 35). We fix arbitrarily the value s E (0, /] and the shortest
curve K = [XYJ, where X = X(s). Let (sn), n = 1, 2, ... , be an arbitrary se-
quence of points of the interval (0, s) such that Sn --+ s as n --+ 00. We put s - Sn =
hn' X(sn) = Xn and on the shortest curve K we specify a sequence of points (y")
such that Y" --+ X as n --+ 00 and at the same time hn/lXY"I--+ as n --+ 00. We
put IXY"I = kn' IXn Y"I = In. By the triangle inequality we have
°
I YY"I ~ I YXnl - IXn y"1,
that is, z(s) - kn ~ z(sn) - In. Hence
z(s) - z(sn) kn - In
h ~-h-·
n n

I short for: left lower

 I. Two-Dimensional Manifolds of Bounded Curvature 91

Let ~n be the angle of the development of the triangle Xn Y,.X at the vertex
corresponding to the point X. We have
): _ h; + k; - I; _ kn - In h; - (k n - In)2
cos '>n - 2h k - -h- + 2h k
n n II n n

Furthermore, Ik n - lnl ~ hn, and so

kn - In hn
- h - ~ cos ~n - 2k .
n n

From the condition hn/kn -+ 0 as n -+ 00 we obtain

· z(s) - z(sn)
I1m I· kn - In I· ):
~ 1m - h - ~ 1m cos '>n·
11-00 S - 8n 11-00 n 11-+00

We have

lim cos
11-00
~n = cos (lim ~n) ~ cos ~,
11-+00

where ~ is the upper angle between the shortest curves K and L at the point X.
Since the sequence Sn -+ S - 0 is taken arbitrarily, it follows that

(dZ)
ds (s) 1.1. ~ cos ~.
-
The shortest curve K = [XY] was chosen arbitrarily. The left-hand side of the
last inequality does not depend on the choice of K. Going over to the greatest
lower bound over the set of all shortest curves joining Y and X, we obtain the
required result.
Lemma B. In the notation of Lemma A let a(s) and ~o(s) be the angles of the
development of the triangle AX(s) Y at the vertices corresponding to the points A
and X = X(s) respectively. If sin ~o(s) # 0, then

( da) (s) = lim a(s) - a(s - h) ~ ~. cos ~(s! - cos ~o(s) .

ds 1.1. h-++O h s sm ~o(s)
Proof. Suppose we are given a planar triangle with sides x, y, z. Let a be the
angle opposite the side z. We have
Z = Jx 2 + y2 - 2xy cos a.
Fixing y, we regard Z as a function of the variables x and a. Obviously z is a
·n- . bl f
d luerentIa . 0 f x an d a. We have dz = x - y cos a dx
e unctIOn xy . ada.
+ -sm
z z
): d enote th
Let '>0 e ang ei · t h·d
OpposIte e SI e y. Then we b tam
. -
sin-a= - sin-~o. 0
z y
Moreover, x - y cos a = z cos ~o and finally dz = cos ~o dx + x sin ~o da. Hence
it follows that if we give x and a increments Ax and Aa, then the increment of z
92 Yu.O. Reshetnyak

can be represented as
Az = cos eoAx + x sin eoAcx + O(Ax, Acx)(IAxl + IAcxl),
where O(Ax, Acx) -+ 0 as Ax -+ 0 and Acx -+ O. We put x = s, z = z(s), Y = z(O),
cx= cx(s). Obviously cx(s) is a continuous function of s. We put As = -h, Az =
z(s - h) - z(s), Acx = cx(s - h) - cx(s). After obvious transformations we obtain
z(s) - z(s - h) . cx(s) - cx(s - h)
h = cos eo + [s sm eo - O"O(h, Acx)] h - O(h, Acx).
(6.17)
Here 0" = sign Acx. As h -+ 0, cx(s) - cx(s - h) -+ 0 and so O(h, Acx) -+ O. We choose
.. cx(s) - cx(s - hm )
a sequence of quantities hm -+ + 0, m = 1, 2, ... , such that h -+

(~:) 1.1.
(s) as m -+ 00. Then, putting h = hm in (6.17) and proceeding t~ the limit,
we obtain

(~:)I.I. (s)s sin eo + cos eo

. z(s) - z(s - hm)
11m l' z(s) - z(s - h) ~( )
= ~ 1m ~cos .. s
m-+Q() hm h-++O h
and so the lemma is proved.
Relying on Lemmas A and B, we can now prove Lemma 6.3.1.
Despite what we have proved, we assume that
Ii - cx o ~ vl + 3e, (6.18)
where e > O. Since ABC is one of the triangles AX Y considered in the definition
of vl, we can choose the shortest curve BC so that

(Ii - cxo) + (p - Po) + (y - Yo) ~ vl + e. (6.19)

In view of (6.18), it follows from (6.19) that at least one of the quantities Po -p
and Yo - Yis not less than e.
We put IABI = Xo, IACI = Yo, IBCI = zo0 For points X E [AB] and Y E [AC]
such that IAXI = x > 0, IAYI = Y > 0 we denote by cx(x, y) the angle of the
development of the triangle AXY at its vertex corresponding to A. We prove
that there are numbers Xl > 0, YI > 0 such that Xl ~ xo, YI ~ Yo, XIYI < XoYo
and

(6.20)

where M = sin 2 ; > O. We assume that Po -p ~ e. Since Po -p > 0, we have

Po> O. Moreover, in this case Po "# n, because otherwise the angle p would be
 I. Two-Dimensional Manifolds of Bounded Curvature 93

equal to 11:. We now use the inequality of Lemma B, putting s = X o, ~o(s) = P,

~(s) = Pin it. We obtain
aa (
( -Xo,Y)) 1 cos P- cos Po' 2 . P+ Po . Po - P
~-'. ~-·sm--·sm--.
ax 1.1. Xo sm Po Xo 2 2
By hypothesis, > Po ~ P+ e, P> O. Hence it follows that 11: - e ~ p, Po ~ e
11:

and so 11: -
e P + Po e 11: Po - P e
2 ~ - 2 - ~ 2' 2 ~ - 2 - ~ 2' Hence we conclude that

2 sin2~
( a~ a (xo, Y) ) ~ --. (6.21)
uX 1.1. Xo

By the definition of the left lower derivative,

( -aa (xo,Y )) -_ l' a(xo, y) - a(x, y)

1m
ax 1.1. X.... Xo-o Xo - x
We have
.
11m In Xo -In x
. (6.22)
x .... xo-o Xo - x
From (6.21) and (6.22) it follows that there is an Xl E (0, xo) such that

a(xo, Yo) - a(xI' Yo) > sin2~(ln Xo -In Xl),

and we see that (6.20) is satisfied if we put YI = Yo' By construction Xl ::::; Xo,
YI ~ Yo, XIYI < XoYo'
Similarly we conclude that if Yo - Y ~ e there is a Y I E (0, Yo) such that

ao(x o, Yo) - ao(x o, YI) > MIn Yo.

YI
Putting Xl = Xo, we obtain a pair (Xl' YI) for which (6.20) is satisfied, and Xl ~
Xo, YI ~ Yo, XIYt < xoYo' Let t ~ 0 be the greatest lower bound of XIYI on the
set ofall pairs (Xl, YI) for which (6.20) is satisfied, and 0 < Xl ~ Xo, 0 < YI ~ Yo,
XIYI < XoYo. We find a sequence of values X = Xn, Y = Yn' n = 1, 2, ... , for
which (6.20) is satisfied such that Xn --t X, Yn --t Y as n --t 00, where 0 ~ x ~ Xo,
o ~ y ~ Yo, XnYn --t t. For each n we have
(6.23)

Hence it follows that x > 0, y > 0, since otherwise the right-hand side of (6.23)
tends to 00 as n --t 00, which is impossible, since its left-hand side does not
exceed 11:. Proceeding to the limit, we obtain

- y)
ao(xo, Yo) - ao(x, - ~ M In--.
XoYo (6.24)
t
94 Yu.G. Reshetnyak

Since t < XoYo, the right-hand side of (6.24) is positive. Consider a triangle
AB' C' with vertex B' on the shortest curve AB and vertex C' on the shortest
curve BC such that p(A, B') = X, p(A, C') = y. The quantity v1, calculated from
the triangle AB'C', obviously does not exceed VA. for the triangle ABC, tXo(.x, y) ~
tXo(xo, Yo), hence it follows that (6.18) is preserved for the triangle AB'C'. Similar
arguments applied to AB'C' enable us to conclude that there are numbers x' > 0,
y' > 0 for which x' ~ x, y' ~ y, where at least one of the inequalities is strict and

tXo(x, y) - tX(x', y') > Min ~y,. . (6.25)

xy

Adding (6.24) and (6.25), we obtain

tXo(xo, Yo) - tX(x', y') > M In x~y? .

xy

At the same time, x'y' < xy = t. This obviously contradicts the definition of t.
The resulting contradiction proves the lemma.

6.5. Proof of the Second Theorem on Approximation. The proof of the second
theorem on approximation (Theorem 6.3.2), which we give here, follows the
paper Reshetnyak (1962). It relies on an analogue of Theorem 5.6.1 for the
case of manifolds of bounded curvature and is simpler than the proof given in
Aleksandrov and Zalgaller (1962).
We shall say that a domain G in a metric space (R, p) with intrinsic metric is
convex if G is metrically connected and PG(X, Y) = p(X, Y) for any X, Y E G.
We shall call a two-dimensional manifold Q with intrinsic metric that is homeo-
morphic to a closed disc a convex cone if Q is isometric to a convex domain on
the cone Q(9), where 9 < 2n.
Let (R, p) be a two-dimensional manifold with intrinsic metric, and U c Ran
open domain in R. We assume that in U there is specified a sequence ofmetrics
(Pn), n = 1,2, ... , such that the following conditions are satisfied:
1) as n -+ 00 the metrics Pn converge to a metric Pu;
2) the space (U, Pn) is a polyhedron;
3) there is a constant A < 00 such that w:JU) < A for all n;
4) as n -+ 00 the set functions w:JE) converge weakly to some set function
wo(E).
Condition 4 implies the following: a) wo(A) ~ lim w:JA) for any closed set
n-+ao
A c U; b) wo(V) ~ lim w:JV) for any open set V c U.
n-+ao

Lemma 6.5.1. Suppose that all the conditions 1-4 listed above are satisfied.
Let r be a simple closed curve contained in U, and G c U the open set those
boundary is the curve r. We assume that wo(G) < 2n and that the curve r is
rectifiable with respect to the metric p. Then there is a convex cone Q such that
w(Q) ~ wo(G), and there is a contraction cp of the cone Q onto the domain G u r
which maps the boundary of the cone one-to-one and preserves arc lengths.
 I. Two-Dimensional Manifolds of Bounded Curvature 95

The lemma is obtained from Theorem 5.6.1 by a limiting process. We outline

a proof of it that does not go into details of a technical character.
We first construct a sequence of simple closed curves (r,,), n = 1, 2, ... , that
converges to the curve r and is such that r" for each n is a polygonal line on a
manifold with polyhedral metric (U, P.), and spJr,,) -+ sp(r) as n -+ 00. Let G. be
the domain bounded by the simple closed curve r". It is easy to verify that

wo(G u r) ~ lim w:JG. u r,,) .

..... 00

If wo(r) = 0, it is not difficult to verify that, whatever the value of B > 0,

w:JG.) ~ w:JG. u r,,) < wo(G) + B for sufficiently large n. In particular,
w:(G.) < 2n for sufficiently large n. Having constructed a convex cone Q. and
its contraction map <P. onto the domain G. u r" that satisfies the conditions of
Theorem 5.6.1, by means of a limiting process we then obtain the required
convex cone Q.
However, it may happen that wo(r) #- O. In this case we cannot conclude
that w:JG.) < wo(G) + B for sufficiently large n. The quantity wo(r) will be
non-zero if there occurs an accumulation of the positive part of the curvature
of G. close to its boundary. The difficulty that arises in this way can be over-
come as follows. Suppose we are given h > O. We denote by r,,(h) the set of all
points of G. for which P.(X, r,,) < h, and suppose that Gn(h) = Gn \Kn(h), G(h) =
{X E Glpo(X, r) ~ h}. Then
wo(G(h)) ~ lim w:JG.(h))
n .... oo

and so, whatever the value of B > 0, for sufficiently large n we have
w:JGn(h)) < wo(G(h)) + B ~ wo(G) + B.
This shows that the excess of the curvature w:
n ' because of which we cannot

assert that w:JG) does not exceed w(G) + B for sufficiently large n, is concen-
trated in the domain r,,(h). It is impossible to apply Theorem 5.6.1 directly to the
domain Gn(h), since its topological nature may be rather complicated, and what
is more we cannot say anything about the length of the boundary of Gn(h). We
therefore proceed as follows. We construct a sequence of values hn -+ 0, hn > 0
for all n, such that lim w:JG.(hn)) ~ wo(G). Such a sequence is obtained by the
n .... oo
standard diagonal process. We now transform each of the polyhedra (Gn, Pn) as
follows. We consider all its vertices whose curvature is positive and which lie in
the band r,,(h.). We join each of these vertices to the curve r" by a shortest curve.
We cut Gn along these shortest curves and to each of the cuts we paste a circular
sector whose radius is equal to the distance from the corresponding vertex to r",
and the angle is equal to the curvature at this vertex. After this procedure the
curvature of each of those vertices of the polyhedron (Gn, Pn) lying in the band
r,,(h.) at which it is positive becomes equal to zero, and the length of the bound-
ary increases by no more than hnw:JGn) ~ Mh•. As a result w~btain a se-
quence of polyhedra (G~, p~) homeomorphic to a disc such that lim w:JG~) ~
n .... oo
96 Yu.G. Reshetnyak

wo(G), and the length of the boundary of G~ has the limit spo(F). Having con-
structed a convex cone Qn satisfying the conditions of Theorem 5.6.1, for each
of the polyhedra G~ by a suitable limiting process we obtain a convex cone
satisfying all the conditions of the lemma.
By means of Lemma 6.5.1, Theorem 6.3.2 is proved as follows.
We assume that a two-dimensional manifold R with intrinsic metric p satis-
fies the conditions of the lemma. Let X be an arbitrary point of R, let U be a
neighbourhood of this point homeomorphic to a disc, and let Pn be a sequence
of polyhedral metrics defined in U that converges to P as n -+ 00 and is such that
w:JU) ~ Q < 00 for all n. Without loss of generality we can assume that the set
functions w:JE) converge weakly in U to some set function wo(E). This can
always be achieved by replacing the sequence Pn by a suitable subsequence.
Let T be an arbitrary simple triangle contained in U, let A, B, C be its
vertices, and K = [BC], L = [AC], M = [AB] its sides. The sides of T form a
simple closed curve r. Let G be the interior of the triangle, and ii, p, Ythe upper
angles of the triangle at its vertices A, B, C. Let us prove that
b(T) = ii + p+ y - n ~ wo(G).
If wo(G) ~ 2n the inequality is obvious, since each of the numbers ii, p, Y does
not exceed n. Suppose that wo(G) < 2n. Then by the lemma there is a convex
cone Q that has a contraction map onto the domain G such that the boundary
of Q is mapped onto the boundary of G with arc lengths preserved, and w(Q) ~
wo(Q). Let A o, Bo, Co be points that belong to the boundary of Q and are
such that qJ(Ao) = A, qJ(Bo) = B, qJ(Co) = C. Each of the arcs [AoBo], [BoCo],
[CoAoJ that form the boundary of Q is a shortest curve. In fact, we have
1([AoBoJ) = 1([ABJ) by virtue of the fact that qJ maps oQ onto the boundary
of G with arc lengths preserved. Hence 1([AoBo]) = p(A, B) ~ p(Ao, Bo); since
1([AoBoJ) ~ p(Ao, Bo), we have 1([AoBo]) = p(Ao, Bo). Let 0(0' Po, Yo denote the
bounding angles of the cone Q at the points A o, Bo, Co. Then by the Gauss-
Bonnet formula (Theorem 5.3.2) 0(0 + Po + Yo - n = w(Q) ~ wo(G). Let us
prove that
0(0 ~ ii, Po ~ p, Yo ~ y. (6.26)
Let X E [AB] and Y E [AC] be arbitrary points on the sides of the triangle,
and X' and Y' the corresponding points of the boundary of Q, so that qJ(X') =
X and qJ(Y') = Y. We put zo = p(X', Y'), z = p(X, Y). Then zo ~ z. On the
plane we construct triangles with sides x, y, z and x, y, zo respectively. Let
y(X, Y) and yo(X, Y) be the angles of these triangles at the vertices opposite to
the sides z and zo respectively. Since zo ~ z we have yo(X, Y) ~ y(X, Y). It re-
mains ,to observe that if the points X and Yare sufficiently close to A, then
Yo(X, Y) = cxo. Hence it obviously follows that
ii = lim y(X, Y) ~ cxo;
X-+A
Y-+A

the other two inequalities in (6.26) are proved similarly.

 I. Two-Dimensional Manifolds of Bounded Curvature 97

From (6.26) it follows that in the given case

8(T) = ii + P+ y - 7t ~ w(Q) ~ wo(G).
Let T1 , T2 , ... , Tm be an arbitrary finite collection of pairwise non-
overlapping triangles homeomorphic to a disc and lying in the domain U. Let Gi
be the interior of the triangle 7;. By what we have proved, for each i we have
c5(7;) ~ wo(G;).
Hence

i~ c5(7;) ~ i~ WO(G i) = W CQ G ~ D.
i)

Here we have used the fact that the sets Gi are pairwise disjoint. Thus, for any
system of pairwise non-overlapping triangles homeomorphic to a disc and lying
in U the sum oftheir excesses does not exceed D. We have thus established that
for the two-dimensional manifold under consideration the axiom of bounded-
ness of curvature is satisfied (even in a form somewhat stronger than that given
in 6.2, we do not require the bounding convexity of the triangles). Thus Theorem
6.3.2 is proved.

§ 7. Analytic Characterization of Two-Dimensional Manifolds

of Bounded Curvature

7.1. Theorems on Isothermal Coordinates in a Two-Dimensional Manifold of

Bounded Curvature. As we mentioned above, in a neighbourhood of each point
of a two-dimensional Riemannian manifold we can introduce a local coordinate
system in which the metric quadratic form of the manifold is
(7.1)
Such a coordinate system is called isothermal. Finding an isothermal coordinate
system reduces to the construction of a conformal map of a domain of a Rie-
mannian manifold into the plane 1R2. We note that the size of the domain in a
Riemannian manifold in which we can introduce a coordinate system of the
stated type depends only on the topological structure of the Riemannian mani-
fold. In particular, in any domain of a Riemannian manifold that is homeo-
morphic to an open disc on the plane we can introduce an isothermal coordi-
nate system (the truth of this follows from some general results of the theory of
functions of a complex variable, namely from the principle of uniformization,
whose formulation will be given later).
Considering isothermal coordinate systems, it is advisable to identify the
plane 1R2 with the set of complex numbers C, and a point (x, y) E 1R2 with the
complex number z = x + iy. In this notation the differential quadratic form (7.1)
is ds 2 = l(z) Idzl 2 •
98 Yu.G. Reshetnyak

The function A(Z) == A(X, y) on the right-hand side of (7.1) is expressed in

terms of the integral curvature of the Riemannian manifold by the formula

A(Z) = exp UL In Iz ~ CI w(dC) + h(Z)}, (7.2)

where G c C is the range of values for the given coordinate system and h(z) is a
harmonic function. As we shall show later, the concept of integral curvature can
be defined for an arbitrary two-dimensional manifold of bounded curvature.
For the latter it is a completely additive set function. In this connection there
naturally arises the conjecture that in a neighbourhood of each point of a two-
dimensional manifold of bounded curvature we can introduce a local coordi-
nate system in which the metric of the manifold is defined by a line element of
the form (7.1), and the function A(Z) is expressed by a formula of the form (7.2)
with the difference that in contrast to a Riemannian manifold w is an additive
set function of sufficiently general form. In particular, it can be not absolutely
continuous with respect to area, as happens in the case of a Riemannian mani-
fold. In favour of such a conjecture, in particular, is the circumstance that the
possibility of introducing an isothermal coordinate system in a domain of a
Riemannian manifold depends only on its topological structure. Let us explain
this idea. To obtain a coordinate system of some special type in a neighbour-
hood of a point of a two-dimensional manifold of bounded curvature it is natu-
ral to consider first a sequence of Riemannian metrics p" defined in a neigh-
bourhood of the given point and converging to the metric of the manifold in this
neighbourhood. For each of the Riemannian metrics we construct in a neigh-
bourhood of the given point a coordinate system of the form of interest to us.
After this we shall attempt to obtain the original coordinate system in the
manifold under consideration by a limiting process. This attempt may turn out
to be untenable, in particular because the size of the domain where the coordi-
nate system in question is defined for the metric p" tends to zero as n --+ 00. For
an isothermal coordinate system of this kind the "vanishing" of the domain of
definition cannot happen.
Here we give statements of theorems on the representation of a two-
dimensional manifold of bounded curvature by an isothermal line element. The
later parts of the present section will be devoted to a proof ofthem. We note that
the different features of the proof have independent significance as a means of
investigating two-dimensional manifolds of bounded curvature.
We first give some definitions. Let A be a Borel subset of the complex plane
C. We shall call an arbitrary totally additive set function defined on the totality
of all Borel subsets A simply a measure in the set A (thus, it is not assumed that
the measure is non-negative). For any measure w there are defined non-negative
measures w + and w -, specified in A and such that w = w + - W -. The measure
w + is called the positive part or upper variation of wand is defined by the
relation
w + (E) = sup w(H),
HeE
 I. Two-Dimensional Manifolds of Bounded Curvature 99

where the supremum is taken over the totality of all Borel sets H c E. Corre-
spondingly, w- is called the negative part or lower variation of the measure w.
The sum w+ + w- is denoted by Iwl and called the total variation of the measure
w.
Let G be a domain in Co We assume that in G there are specified a measure
wand a harmonic function h. We put

A(Z; w, h) = exp{~ fa ln lz ~ "w(dO + h(Z)}. (7.3)

Here the integral is understood in the sense of Lebesgue and Stieltjes. The
function A(Z; w, h) is defined and is finite for almost all Z E G.
Functions that have a representation of the form (7.3) are closely connected
with a class of functions that plays an important role in the theory of partial
differential equations, namely, with the class of subharmonic functions. Refer-
ring the reader to the corresponding literature (Privalov (1937), Hayman and
Kennedy (1978)) for the details, we observe that a function u(z) defined in a
bounded domain G on the plane IC is subharmonic if and only if it has the
representation

u(Z) == u(z; fJ., h) = ~

n
fG In Iz - "fJ.(dO + h(z),
where fJ. is a non-negative measure in the domain G, and h(z) is a harmonic
function. From (7.2) we obtain
In A(Z) = In A(Z; w, h) = U1(z) - U2(Z),
where u1(z) = u(z; w-, h), U2(Z) = u(z; w+, 0). Thus, In A(Z) is the difference of
two subharmonic functions or, as we say, In A{Z) is a b-subharmonic function.
Let G be a bounded domain on the plane and A(Z) == A(Z; w, h) a function
defined in G. Let L be an arbitrary rectifiable simple arc lying in the domain G,
and z{s), 0 ~ s ~ I, a parametrization of it, where the parameter s is arc length.
We consider the function A[Z(S)]. We observe that the function A(Z) is only
defined almost everywhere in the domain G. A rectifiable simple arc is a set of
measure zero. Therefore a priori it is not clear whether the quantity A[Z(S)] is
defined for a set of values of s of positive measure. However, we can show that
A[Z(S)] is defined for almost all s E [0, I]. This follows from the next lemma.
Lemma 7.1.1 (Reshetnyak (1960)). Let u(z) be a subharmonic function in a
domain G. Then u(z) is defined for all Z E G, and -00 ~ u(z) < 00. Let E be the
set of those Z for which u(z) = -00. Then for any e > 0 and any IX > 0 there is a
sequence of open discs (Bm = B(c m, r m )), m = 1,2, ... , such that

and
00

L r::' < e.
m=l
100 Yu.G. Reshetnyak

The result of the lemma (which was known long before the cited work
of Reshetnyak) can be restated as follows. For any (X> 0 the (X-dimensional
Hausdorff measure of the set of those Z E G for which u(z) = -00 is equal to
zero.
From Lemma 7.1.1 it follows, in particular, that if the function A(Z) has a
representation of the form (7.3), then the set of those Z for which A(Z) is not
defined (they can be points for which the functions ut(z) and u2(z) in (7.3) become
-00 simultaneously) can be covered by a sequence of discs for which the sum of
the radii is less than e, whatever e > 0 is.
Thus, for almost all s E [0, I] the quantity A[Z(S)] is defined and is finite. The
function A[Z(S)] is non-negative and measurable, and so the integral

s.«L) = LfiWldzl I == JA[Z(S)] ds

is defined. We shall call s.«L) the length of the simple arc L with respect to the
line element ds 2 = A(X, y)(dX2 + dy2) = A(z)ldzI 2, where A(Z) = A(Z; w, h). We
note that s.«L) may be equal to 00.
Let Zl and Z2 be two arbitrary points in the domain G, where Zt "# Z2' We
denote the greatest lower bound of s.«L) on the set of rectifiable simple arcs
joining Zl and Z2 by p.«zt> Z2)' We put p.«z, z) = O. It can happen that
p.«zo, z) = 00 for some points Zo E G, whatever the point Z EGis, where Z "# Zoo
If Zo has this property, then we shall call it a point at infinity of G with respect to
the line element A(z)ldzI 2. If Zo is a point at infinity, then w({zo}) ~ 2n. Con-
versely, if w( {zo}) > 2n, then Zo is a point at infinity with respect to the line
element A.(z) Idzl 2 • If w( {zo}) = 2n, the point Zo may be at infinity or may not be.
The role of the number 2n becomes clear if we observe that by representing the

f f
integral on the right-hand side of (7.2) in the form
1 1 w({zo}) 1
- In-I-Ylw(d() = 2 In-I- + -1 1
In-I-Ylw(d(),
n G Z - .. n Z - Zo n G\{zol Z - ..
we deduce that fiW can be represented in the form
Ili:\ w({zo})
v A(Z) = Iz - zol Ao(Z), (X =
-<I
2n .

The factor Ao(Z) is such that the main contribution to the behaviour of the
function fiW as Z -+ Zo is given by the factor IZ - Zo 1-<1. If w( {zo} ) > 2n, then
(X> 1, and in this case the integral of fiW along any rectifiable curve passing
through Zo is equal to 00. In the case w( {zo}) = 2n the convergence of the
integral depends on the behaviour of Ao(Z) as Z -+ Zoo By the properties of totally
additive set functions, all the points at infinity with respect to the line element
A.(z) IdzI 2 are isolated.
Theorem 7.1.1 (Reshetnyak (1960». Suppose that a function A(Z) == A(Z; w, h)
is specified in a domain G on the plane C. Let Gbe the domain obtained from G by
excluding points at infinity with respect to the given line element A(z)ldzI 2. Then
 I. Two-Dimensional Manifolds of Bounded Curvature 101

the function P;,(Zl' Z2) is the intrinsic metric in Gcompatible with the topology of
Gas a subset of C, and the metric space (G, P;.) is a two-dimensional manifold of
bounded curvature.
Let G be a domain on the plane C, and P = P;. a metric in this domain that
can be defined from some function A(Z) that has a representation of the form
(7.3). For brevity we shall say that P is a subharmonic metric in the domain G.
Theorem 7.1.2. Let M be a two-dimensional manifold of bounded curvature.
Any point p E M has a neighbourhood U which in the induced metric Pu is iso-
metric to some flat domain G with sub harmonic metric P;..
We make two remarks about the theorems we have stated.
1. In Reshetnyak (1960), where we present a proof of Theorems 7.1.1 and
7.1.2, the metric P;,(Zl' Z2) is defined somewhat differently from the way we have
done here, namely as the greatest lower bound of s;.(L) on the set of all simple
arcs that satisfy a stronger condition then rectifiability, namely the condition
that the variation of turn is bounded (the exact statement of this condition is
given in 7.2). However, by virtue of the results in Reshetnyak (1963a) such a
definition of P;,(Zl, Z2) is equivalent to the one given above.
2. Let U be a domain in a two-dimensional manifold of bounded curvature,
isometric to a flat domain G with some subharmonic metric P;.. Any isometric
map of(U, Pu) onto (G, P;.) is called an isothermal coordinate system in U.
Suppose we are given an open set U c C and a map f: U -+ C, f(z) =
u(z) + iv(z) for any Z E U, where u and v are real functions. We shall say that the
map f is coriformal if one of the functions f(z) and J(z) = u(z) - iv(z) is an ana-
lytic function of z. This means that the functions u = Ref and v = Imf satisfy
either the system of equations Ux = vy , uy = - Vx or the system of equations
U x = -v y, uy = V x ' In both cases we obtain

1f'(zW = 1/2([ux(Z)]2 + [U y (Z)]2 + [vx(Z)]2 + [V,(Z)]2).

The concept of a conformal map has a single geometrical meaning, which we
do not dwell on, to save space, referring the reader to textbooks where the
beginnings of the theory of functions of a complex variable are presented.
Theorem 7.1.3 (Huber (1960)). Let cp: U -+ C, t/I: V -+ C be isothermal coordi-
nate systems in a two-dimensional manifold of bounded curvature. We assume that
the set Un V is not empty, and let G = cp(U n V), H = t/I(U n V). Then G and H
are open sets in C and the function 8(z) = t/I[cp-l(Z)] is a conformal map of G onto
H. If ds 2 = A(Z) Idzl 2 is the line element corresponding to the coordinate system cp,
and da 2 = ll(w)ldwI 2 is the line element corresponding to the coordinate system t/I,
then for all z E G we have
A(Z) = 1l[8(z)]I8'(zW. (7.4)
The assertion that the map 8 = t/I 0 cp-l is conformal is established in Huber
(1960) and Reshetnyak (1963b). The rule for transforming a line element on
going over from one isothermal coordinate system to another, which can be
defined by (7.4), is proved in Huber (1960).
102 Yu.G. Reshetnyak

As a corollary of Theorem 7.1.3 we obtain an analytic characterization of an

arbitrary two-dimensional manifold of bounded curvature. For this purpose
we require the concept of a Riemann surface. Suppose we are given a two-
dimensional manifold M. Then we shall say that M is a Riemann surface if there
is specified in M a set ~ of local coordinate systems whose domains of definition
cover M, and for any two charts qJ: U --+ C and 1/1: V --+ C that belong to ~ and
are such that Un V is non-empty the maps qJ 0 1/1-1 and 1/1 0 qJ-l are conformal.
Charts belonging to the set ~ are called basic charts of the Riemann surface M.
If for any overlapping basic charts qJ: U --+ C, 1/1: V --+ C of a Riemann surface M
the maps qJ 0 1/1-1 and 1/1 0 qJ-1 are holomorphic functions, then M is said to be
oriented.
Any domain on the plane C naturally represents a Riemann surface if as a
unique basic chart we take the identity map of this domain into the plane.
Any two-dimensional Riemannian manifold turns into a Riemann surface if
for the basic charts we take all possible isothermal coordinate systems of this
manifold. In fact, from the formulae for transforming the components of the
metric tensor of a Riemannian manifold on going over from one local coordi-
nate system to another it follows easily that when these coordinate systems are
isothermal the components of the transition function (u, v) satisfy either the
Cauchy-Riemann system of equations u", = vy, uy = -v", or the system u", =
-vy, uy = v",.
In particular, a sphere and more generally any surface in three-dimensional
Euclidean space that satisfies the regularity conditions adopted in differential
geometry is a Riemann surface.
Let M 1 and M 2 be arbitrary Riemann surfaces, and f: M 1 --+ M 2 a continu-
ous map. Then we shall say that the map f is conformal if for any basic chart
qJ: U --+ C of M1 and any basic chart 1/1: V --+ C of M2 the map 1/1 ofo qJ-1 is
conformal. We shall call 1/1 0 f 0 qJ-1 a representation of the map f by means of
the charts qJ and 1/1. The Riemann surfaces M 1 and M 2 are said to be equivalent
if there is a topological map of M 1 onto M 2 that is conformal.
The principle of unijormization, of which we spoke earlier, is expressed by the
following statements. Any Riemann surface homeomorphic to an open disc in C
is conformally equivalent to either a disc or the plane C. A Riemann surface
homeomorphic to a sphere is conformally equivalent to a sphere.
Let M be a two-dimensional Riemann surface. Then we shall say that on M
there is determined a quadratic differential a if for any basic chart qJ: U --+ C
there is defined a differential quadratic form ds 2 = A.(z)ldzI 2 , specified on the set
G = qJ(U) and called a representation of the quadratic differential a in the local
coordinate system qJ, and the differential quadratic forms A.(z)ldzI 2 and J.t(w)ldwI 2
corresponding to two arbitrary overlapping basic charts qJ and 1/1 are connected
by (7.4).
From Theorem 7.1.3 it obviously follows that if M is a two-dimensional
manifold of bounded curvature, then M is a Riemann surface for which the
isothermal local coordinate systems form a set of basic charts. To any isother-
mal coordinate system qJ: U --+ C of the manifold M there corresponds a differ-
 I. Two-Dimensional Manifolds of Bounded Curvature 103

ential quadratic form ds 2 = A.(z)ldzI 2 , where the function A.(z) is defined on the
set G = qJ(U) and has a representation of the form (7.2). If A.(z)ldzI 2 and
Jl(w) Idwl 2 are the differential quadratic forms that correspond to two isothermal
coordinate systems y: U -+ C and .p: V -+ C, then the functions A. and Jl are
connected by (7.4). Thus, on the Riemann surface M there is defined a quadratic
differential (T.
The converse assertion is true. Namely, we have the following theorem.
Theorem 7.1.4 (Huber (1960». Let M be a two-dimensional Riemann surface.
We assume that on M there is specified a quadratic differential (T such that for any
basic local coordinate system of the surface the coefficient A.(z) in the representa-
tion of the differential (T in this coordinate system has a representation of the form
(7.3). If M is connected, then on it we can define an intrinsic metric Pa such that
for any basic coordinate system qJ: U -+ C any point p E U has a neighbourhood
V c U having the property that for any Pl' P2 E V we have
Pa(Plo P2) = p;.[qJ(pd, qJ(P2)]'
where P;. is the metric defined in the domain G = qJ(U) by the line element ds 2 =
A.(z)ldzI 2 , the representation of (T in the coordinate system qJ, as described above.
The metric Pa satisfying the given condition is unique and the metric space (M, Pa)
is a two-dimensional manifold of bounded curvature.
Theorem 7.1.4 enables us to establish a connection between the theory of
manifolds of bounded curvature and the theory of functions of a complex vari-
able. This gives a way of studying the structure in the large of two-dimensional
manifolds of bounded curvature by the methods of the theory of func-
tions. Some examples on the application of this theorem to the study of two-
dimensional Riemannian manifolds of bounded curvature can be found in
Huber (1954), for example.
If the function co in (7.2) is an indefinite integral of some function Jt"(z),
continuous and satisfying a Holder condition with exponent IX, 0 < IX ~ I, inside
G, then the function A.(z) has continuous partial derivatives of the first and
second order 2 , and these derivatives also satisfy a Holder condition with
exponent 3 IX. In this case the metric P;. is Riemannian and for the point
z = x + iy the number Jt"(z) is the Gaussian curvature of the metric at this point.
Thus, from Theorem 7.1.1 it follows that a two-dimensional Riemannian mani-
fold is a manifold of bounded curvature. It is true that we can still only prove

z We say that a function oX'": G -+ iii satisfies a HOlder condition with exponent IX, where 0 < a. :E; 1, if
there is a constant L such that O:E; L < 00 and loX'"(zt> - oX'"(zz)1 :E; Liz, - zzl" for any z" Zz E G.
We shall say that oX'" satisfies a Holder condition with exponent a., 0 < a. :E; 1, inside G if for any com-
pact A c G there is an L(A) < 00 such that loX'"(zt> - oX'"(zz)1 :E; L(A)lz, - zzl" for any z" Zz of A.
3Continuity of the function K(z) is insufficient to guarantee the existence of continuous second
derivatives of l(z), since, as we know, the solution of the Poisson equation Lfu(z) = f(z), where f(z)
is an arbitrary continuous function, may be a function that does not have continuous second
derivatives (in this case the equation itself must be understood in a generalized sense, Petrovskij
(1961». '
104 Yu.G. Reshetnyak

this under certain assumpions about a Riemannian metric. Here it is relevant to

mention the following result. If in a domain U of a Riemannian manifold we
introduce a local coordinate system such that the coefficients of the metric
tensor of the manifold in this coordinate system are functions that have continu-
ous second derivatives satisfying a Holder condition with exponent a, then in
this domain we can introduce an isothermal coordinate system, and the coeffi-
cient A.(z) of the metric tensor in this coordinate system will have continuous
second derivatives satisfying a Holder condition.
As an example of the application of Theorem 7.1.4 we mention a theorem
about isothermal coordinates on polyhedra.
Let M be a two-dimensional manifold with intrinsic metric p. We shall say
that M is a manifold of type I if M is complete and homeomorphic to a sphere
with finitely many points deleted.
We assume that M is a manifold of type I, M' is a set homeomorphic to M
on the sphere S(O, 1) c ~3 obtained by excluding certain points ai' a2' ... , an'
and lp is a topological map of M' onto M. We take a point Yo E M arbitrarily
and let B(yo, r) be the disc in M with centre Yo and radius r. Since M is a
complete two-dimensional manifold with intrinsic metric, the set B(yo, r)
is compact, whatever the value of r > 0. Hence it follows that the set
lp-i [B(yo, r)] eM' is compact, and so none ofthe points ai' a2' ... , an is a limit
point. This enables us to conclude that for any sequence (x m ), m = 1, 2, ... , of
points of the sphere S(O, 1) that converge to one of the points aj, for the points
Ym = lp(xm) we have p(Yo, Ym) -+ 00 whatever the point Yo E M. In this connec-
tion we shall say later that to the point aj there corresponds a point at infinity
of the manifold M.
We now consider the case when M is a two-dimensional manifold of type I
with a polyhedral metric, and the set of its vertices is finite. It is convenient to
regard points at infinity of the manifold as vertices of it.

Theorem 7.1.5. Let M be a polyhedron of type I satisfying all the con-

ditions listed above. Then in M there is an isothermal coordinate system whose
domain of definition is the whole manifold M except possibly for one point.
For any such coordinate system the range of values is the whole plane except
for a finite set of points, and the function A.(z) in the representation ds 2 =
A.(z) IdzI 2 of the line element of the polyhedron in this coordinate system has the
representation
w. 1
L -11t In-I- I + C,
n
In A.(z) = (7.5)
j=i Z - Zj

where Z l' Z 2, ••. , Zn are the points corresponding to the vertices of the polyhedron,
and C is a constant (we recall that to the number of vertices we also assign points
at irifinity). If Zj is an ordinary vertex of M, then Wj is the curvature of this vertex.
We have
(7.6)
 I. Two-Dimensional Manifolds of Bounded Curvature 105

The equality (7.5) can be rewritten in the form

n f
In A.(z) = 1 c In 1z _1 "w(dO + C,

where w is the measure, concentrated at the points Zl' Z2' .•. , Zn' and w({zJ) =
Wj'
We observe that if in the isothermal coordinate system an ordinary point of
M corresponds to the point 00 in the plane C, then the equality sign holds in
(7.6).
We mention the special case when the complete polyhedral manifold M is
homeomorphic to a plane and has a unique finite vertex. In this case M is a
cone. Choosing an isothermal coordinate in M so that to a point at infinity
there corresponds the point 00 of the extended complex plane C, we deduce that
the matrix of M is defined by the line element

(7.7)

where w < 2n is the curvature of the cone, and C is a constant. By changing the
variables it is easy to arrange that the constant C in (7.7) is equal to 1.
The plane C with line element

(7.8)

is isometric to the surface of an infinite right circular cylinder. In this case

the constant C has the following geometrical meaning: the quantity 2nft is
equal to the length of the section of a cylinder by a plane perpendicular to its
generators.

7.2. Some Information about Curves on a Plane and in a Riemannian Mani-

fold. Here we describe some auxiliary concepts that can be used in the proof of
theorems stated in 7.1 on the representation of a two-dimensional manifold of
bounded curvature by means of an isothermal line element.
Let (p, q) be an arbitrary pair of non-zero vectors on the plane Co We assume
that the vectors p and q do not lie on one line. The symbol (p,q) denotes the
angle between them, taken with the + sign if the pair of vectors (p, q) is right,
and with the - sign if the pair is left 4 . If the vectors p and q lie on one line, we
put (p~) = 0 if p and q have the same direction, and (p,q) = n if they have
opposite directions.

4Let p = (P"P2)' q = (q"q2)' [p, q] = P,q2 - P2q,. The pair (p, q) is right if [p, q] > 0, and left if
[p, q] < O. The angle (p-:q) is formally defined as the number cp such that -1t < cp ~ 1t and [p, q] =
Ipl'lql' sin cpo
106 Yu.G. Reshetnyak

A polygonal line on the plane C is any finite sequence L of points Xo, Xl' ... ,
Xm and vectors Pi = X i- 1 Xi such that Xi-l =1= Xi for any i = 1, 2, ... , m. We call
the points Xi' i = 0, 1, ... , m, the vertices of the polygonal line L, and the vectors
Pi = X i- 1 X i its links. We put
m m-l /'..
s(L) = L Ipil,
i=l
IC(L) = L (Pi' pi+d,
i=l

The quantity s(L) is the length of L, and we shall call x(L) the rotation of L, and
Ixl (L) the absolute rotation of L.
One of the main concepts of the theory presented here is that of the turn of a
simple arc in a two-dimensional manifold of bounded curvature. If a simple arc
K in such a manifold has a definite direction at each of its end-points, then we
can define for it two numbers IC,(K) and IC,(K), called the left and right turns of
the arc K. In the case when the manifold is the Euclidean plane with its usual
metric, and the simpler arc K is a polygonal line, IC,(K) = IC(K), and IC,(K) =
-IC(K). In the given case the terms "left turn" and "right turn" can be justified
as follows. Let (Xi and Pi be the angles between the rays Xi X i - 1 and X iXi+l on
the left and right of L respectively (see Fig. 36). Then (Xi + Pi = 2n and IC,(L) =
m m
L (n -
i=l
(Xd, ICr(L) = -IC(L) = L (n - Pd·
i=l
Let F be a simple closed polygonal line. It bounds a domain D homeo-
morphic to a disc. Let us orient Fpositively. Let Xl' X 2 , ••• , Xm be successive
vertices of it. We put X mX 1 = Pl' X 1 X 2 = P2' ... , X m- 1 X m = Pm. Let (Xi be the
angle of the domain D at the point Xi. By the Gauss-Bonnet theorem for poly-
hedra we have
m m /'..
IC(F) = L (n - (Xi) = L (Pi' pi+d = 2n (7.9)
i=l i=l
(in the second sum we put Pm+l = pd.
Let K and L be two oriented simple arcs with common beginning A and
common end B, having no points in common other than A and B. We assume
that K and L are polygonal lines. The simple arcs K and L bound some domain
D. Let (X and Pbe the angles of this domain at the points A and B. We assume
that L lies to the left of K. The polygonal lines K and L form a simple closed

XL-1
x;"

Fig. 36
 I. Two-Dimensional Manifolds of Bounded Curvature 107

polygonal line r, which we orient positively. Then by (7.9) we have K(r) =

K(K) + (n - IX) + (n - f3) - K(L) = 2n. Hence K(K) = IX + f3 + K(L). This equal-
ity can also be written as follows:
(7.10)
We have KI(K) = -Kr(K), KI(L) = -Kr(L). Hence Kr(L) = IX + f3 + Kr(K).
Changing the notation, we deduce that if L lies to the left of K, then (7.10) is
satisfied, and if L lies to the right of K, then
(7.11)
Let L be a simple arc on the plane. Then we shall say that L is a curve with
bounded rotation if the least upper bound of the absolute rotations of polygonal
lines inscribed in L is finite. This least upper bound will be denoted by IKI(L)
and called the absolute rotation of L.
Lemma 7.2.1. Any simple arc L on a plane such that IKI (L) < 00 is rectifiable,
and
d(L)
s(L) ~ -2-[IKI(L) + n], (7.12)

where d(L) is the diameter of L.

If a sequence of simple arcs (Lm), m = 1,2, ... , converges to the simple arc L
as m -+ 00, then

If the absolute rotations of the curves Lm, m = 1, 2, ... , are bounded in aggre-
gate, that is, IKI(Lm) ~ H = const < 00 for all m, then the limiting curve L is
rectifiable, and as m -+ 00 the lengths of Lm converge to the length of L.
Let L be a rectifiable simple arc on the plane. We shall say that L is one-sidedly
smooth if it has a parametrization z(t), a ~ t ~ b, such that for any t E (a, b]
there is a left derivative z;(t) and for any t E [a, b) there is a right derivative z;(t),
and Iz;(t)1 # 0, Iz;(t)1 # 0 at each point at which the corresponding derivative is
defined, and the following conditions are also satisfied: if a < t ~ b, then
z;(t) = lim z;(u) = lim z;(u),
U-+I-O U-+I-O

and if a ~ t < b, then

z;(t) = lim z;(u) = lim z;(u).
u-+I+O U-+I+O

Any unilaterally smooth simple arc is rectifiable, and if '(s), 0 ~ s ~ 1, is its

parametrization, where the parameter s is arc length, then the function '(s) also
satisfies all the conditions stated in the definition of a unilaterally smooth simple
arc. We have le;(s)1 = 1, ";(s)1 = 1 at each point at which the corresponding
derivatives are defined. We shall call the vectors tl(S) = ';(s) and tr(s) = ';(s)
respectively the left and right tangent vectors at the point '(s) of the simple arc L.
108 Yu.G. Reshetnyak

Lemma 7.2.2. Any simple arc with bounded rotation is unilaterally smooth. If
L is a unilaterally smooth curve and (s), 0 ~ s ~ I, is its parametrization, where
the parameter s is arc length, then in order that L should be a curve with bounded
rotation it is necessary and sufficient that at least one of the functions cas) and
(;(s) (we put (1(0) = (;(0), (;(1) = (1(/» should be a function of bounded variation.
We observe that if one of the derivatives mentioned in the statement of the
lemma is a function of bounded variation, then the other is automatically a
function of bounded variation.
Let L be an oriented simple arc on the plane, p a point not belonging ro L,
and z(t), a ~ t ~ b, an arbitrary right parametrization of L. Then we can define
a continuous function O(t), a ~ t ~ b, such that O(t) is the polar angle of the
vector z(t) - p for each t. We put cp(t, p, z) = O(t) - O(a). We shall call O(t) the
angular function of the simple arc L with respect to the point p and the param-
etrization z. We put cp(b, p, z) = cp(p, L). The quantity cp(p, L) does not depend
on the choice of right parametrization z of the simple arc L, and we shall call it
the angle at which L is seen from the point p.
Lemma 7.2.3. Let L be a simple arc on the plane, p a point not belonging to L,
and cp(t) = cp(t, p, z), a ~ t ~ b, the angular function of L with respect to the point
p and the parametrizaion z of L. Then we have
b
Var cp(t) ~ IKI(L) + n. (7.13)
a

Proofs of Lemmas 7.2.2 and 7.2.2 can be found in Reshetnyak (1960) and
Reshetnyak (1963b). for example. Lemma 7.2.3 was obtained by Radon in his
research into potential theory. Its proof reduces to a consideration of the case
when the simple arc L is a polygonal line. For this case the proof of the un-
equality (7.13) is a problem in elementary geometry.
Let K be a simple arc on the plane. Let us orient K, and let A and B the its
beginning and end respectively. We assume that K has tangents at A and B.
Then with the arc K we can associate a number K(K), which we call the rotation
of K. Let us construct arbitrarily a simple polygonal line L with ends A and B
lying to the left of K and having no points in common with K other than A and
B. Such a polygonal line exists, since K has tangents at A and B. The polygonal
line L and the curve K bound a domain D homeomorphic to a disc. Let (X and P
be the angles of this domain at A and B. We put
K(K) = K,(K) = (X + P+ K,(L). (7.14)
By means of (7.8) it is easy to show that the right-hand side of (7.14) does not
depend on the choice of the polygonal line L, so it follows that the given defini-
tion is reasonable. We put Kr{K) = - K{K). It is easy to verify (7.10) and (7.11) if
K and L are not polygonal lines, but arbitrary simple arcs having tangents at A
and B. The quantity K(K) will be called the rotation of the simple arc K, and
the quantities K,(K) = K(K) and K,(K) = - K(K) the left and right turns of K
respectively.
 I. Two-Dimensional Manifolds of Bounded Curvature 109

Let K be a one-sidedly smooth simple arc on the plane, and x(t), a ~ t ~ b, a

right parametrization of it. We denote the rotation of the arc [x (a), x(t)] of the
curve K by K(t). Then we have
b
IKI(K) = Var K(t),
a

which establishes a connection between the rotation and absolute rotation of a

simple arc.
Let us consider the case when the simple arc L is of class C 2 and z(s), 0 ~
s ~ I, is a parametrization of it, where s is arc length. Then z'(s) :F 0 for all
s E [0, I], and there is defined a number k(s), the curvature of L at the point z(s).
We have ZIl(S) = k(s)v(s), where v(s) is the unit normal vector at the point z(s),
chosen so that the pair of vectors (z'(s), v(s)) on the plane is a right pair. In this
case the turn ofthe simple arc L is equal to

K(L) = I k(s) ds.

In the given case the quantity IKI(L) is equal to

I Ik(s)1 ds.

The concepts and results concerning plane curves mentioned here form part
of a general theory constructed in the monograph of Aleksandrov and Reshetn-
yak (1989).
Representation of the turn of a curve in a Riemannian manifold in an isothermal
coordinate system. Henceforth we shall require an explicit expression for the
geodesic curvature and the turn of a curve in a Riemannian manifold. For
our purposes it is sufficient to consider the case when the coordinate system
specified in the Riemannian manifold is isothermal.
We assume that in a flat domain G there is specified a Riemannian metric
with line element
(7.15)
We shall assume that the function A.(x, y) has all the partial derivatives of the
first and second orders, and that these derivatives are continuous. Then, as we
mentioned above, the function A.(x, y) = A.(z) has the representation

In A.(z) = ~ f fin Iz ~ CI dw(C) + h(z), (7.16)

G

where the set function w(E) is the integral curvature and h(z) is a harmonic
function.
Let L be a smooth (of class C 2 ) oriented simple arc contained in the domain
G, and z(s) = x(s) + iy(s), 0 ~ s ~ I, a parametrization of it, where the parameter
110 Yu.G. Reshetnyak

s is arc length in the sense of the geometry of the plane [R2. The function z(s) has
a continuous second derivative and IZ'(s)1 = 1 for all s E [0, 1]. We denote by
k(s) the curvature in the sense of the geometry of [R2 at the point z(s) of the arc
L. The geodesic curvature with respect to the line element (7.15) at the point z(s)
of L is then expressed by the formula
k(s) 1
kg(s) = /1 +
/1[ -(lnA)yX'(S) + (In A)xY'(S)].
y A 2y A
Multiplying both sides of this equality by JJ.
and integrating with respect to s,
we deduce that the turn of the arc L with respect to the Riemannian metric with
line element (7.15) is equal to

Ki L ) = I k(s) ds +~ I[ -(In A)y dx + (In A)x dy].

Let v(s) be the unit normal vector at the point z(s) of the arc L, directed so that
the pair of vectors (t(s), v(s)) is a right pair (that is, such that [t(s), v(s)] > 0). We
have t(s) = z'(s) = (x'(s), y'(s)), v(s) = (- y'(s), x'(s)), from which it follows that

- (In A)y dx + (In A)x dy =

o
- ov {In A} ds.

Hence we obtain

Ki L ) = K(L) - ~ I :v {In A(Z)} ds.

Substituting into the integrand the value ofln l(z) from (7.16), we have

Ki L ) = K(L) - 2n
1 ff {II
G
0
0( I)}
ov In Iz(s) _ ~I ds w(d() -
1II Tv
2: 0
oh(z)
ds.

We now observe that for the function

1
)"(z) = In-I- Y = -lnJ(x - ~)2 - (y - ,,)2 (7.17)
Z - .. I

(we assume, as usual, that z = x + iy, , = ~ + i,,) we have

(y - ,,) dx - (x - ~) dy
(In A)y dx - (In )")x dy =- ( ~)2 (
X - + y - ,,)
2

x - ~)
= d ( arctan xy -_ ,,) (
~ = d arctan y _" .

Hence it is clear that (In A)y dx - (In A)x dy for the function A(Z) defined by (7.17)
is the angle at which the infinitesimal element (dx, dy) is seen from the point
( = (~, ,,), and so the integral

I (0I

o ov
-In 1)
Iz(s) - "
ds
 I. Two-Dimensional Manifolds of Bounded Curvature 111

is equal to qJ«(, L), the angle at which L is seen from the point (. Finally we
obtain

Kg(L) = K(L) - 21n II

G
qJ«(, L) dw«() - ~ L!~ (z) ds. (7.18)

This formula is fundamental for the investigation of the differential proper-

ties of shortest curves on a manifold of bounded curvature in isothermal
coordinates.
The equality (7.18) also enables us to get rid of the function h(z) in the repre-
sentation of the function A(Z), which is given by (7.16), as the following proposi-
tion shows.
Lemma 7.2.4. Let us assume that in the disc B(O, 1) on the plane ~2 there is
specified a Riemannian metric p defined by the line element ds 2 = A(z)(dx2 + dy2),
and that the following conditions are satisfied:
1) the boundary of the disc B(O; 1) is a smooth curve of class C2 with respect to
the metric p, and the absolute turn of the circle r(O, 1) in this metric does not
exceed some constant M 1 < 00;
2) there is a constant M2 < 00 such that Iwl(B(O, ~ M 2. 1»
Then the function A(Z) has the representation

A(Z) = exp {~f-

n B(O.l)
In I
Z
~ YI dw(O + c},
..

where C is a constant, and wis a measure in the closed disc B(O, 1) such that
Iwl(.8(O, 1» ~ (Ml + M 2) + 2n
and w(E) = w(E) for any set E lying in the open disc B(O, 1).
The measure w is thus obtained from the measure w by adding the measure
concentrated on the circle r(O, 1).
The proof of the lemma is based on the use of the following well-known
representation of a harmonic function in terms of the value of its normal deriva-
tive on the boundary of a circle:

1
h(z) = h(O) + -
f ah (0 In-I-y11dn
-a 1 (7.19)
n 1"(0.1) v Z - ..

On the plane we introduce polar coordinates (r, qJ) and let Fe be the arc of the
circle nO, 1) corresponding to values qJ E [0, 0]. We put

u(O) = I: !~ (e iql ) dqJ = Ir, !~ (0 Id(l.

The equality (7.15) enables us to estimate the variation of the function u(O) in
terms of the variations of the functions

Kg(Fe) = I: kg(e iql ) dqJ, K(Fe) = ()

112 Yu.G. Reshetnyak

and

v(O) = 21 f_ qJ(C, 18) den(C).

n JB(O.l)

To estimate the variation of the function v(O) we must take into account that
qJ(C, 18) as a function of the variable 0 is monotonic when CE B(O, 1), and 0 ~
qJ(C, 18) ~ 2n.

7.3. Proofs of Theorems 7.1.1, 7.1.2 and 7.1.3. The proofs of these theorems
are based on a proposition which we call a theorem on the convergence of
metrics (Reshetnyak (1960)).
Let G be a closed domain on the plane, bounded by finitely many simple
closed curves of bounded rotation. Further, we consider various measures en
defined in G and satisfying the following condition: the integral

n
In A(Z; en) == 1 f G
1 " en(dC)
In Iz _

is defined and finite for almost all ZED.

Lemma 7.3.1. Let en be a measure defined in a domain Gee and let A(Z) =
),(z; en). Then for any
Zl' Z2 E G the quantity P;..G(Zl' Z2) is equal to the greatest
lower bound of the integral

L A(Z) Idzi (7.20)

on the set of all curves K with bounded rotation contained in G and joining the
points Zl and Z2' On the set G, obtained from G by excluding points at infinity in
the sense of the metric P;., the metric P;. is intrinsic and compatible with the natural
topology of G.
We observe that by definition P;,(Zl' Z2) is the greatest lower bound of the
integral (7.20) on the set of all rectifiable curves joining the points Zl and Z2 and
contained in G.
Theorem 7.3.1. Let (en~) and (en;), n = 1, 2, ... , be sequences of non-negative
measures defined in the domain G and weakly converging to the measures en l and
en 2 respectively as n -+ 00. Let enll = en~ - en;, en = enl - en2, AII(Z) = A[Z,WIIJ,
A(Z) = A(Z, en). Then as n -+ 00 the functions 15;. (z, C) converge to the function
15;.(z, C) uniformly on any closed set A c G not containing points Z such that
en l (z) ~ 2n.
Proof of Theorem 7.1.1. Let G be a domain on the plane in which there is
specified a function A defined by (7.2), and suppose that the metric P;,(Zl' Z2) is
defined with respect to A as stated above. It is required to prove that the domain
G, endowed with the metric P;., is a two-dimensional manifold of bounded cur-
vature. The required proof is obtained by applying Theorem 6.2.2 and consists
 I. Two-Dimensional Manifolds of Bounded Curvature 113

roughly in the following. We first establish that in a neighbourhood U of

an arbitrary point Zo E G the function A(Z; w, h) can be represented in the

1
form

A(Z; w, h) = C exp 1 n G
1 " w(do,
In Iz _ (7.21)

where C > 0 is a constant (it is necessary to distinguish the cases w( {zo}) < 2n
and w( {zo}) = 2n; see below). We then construct sequences of non-negative
measures (w;) and (w;), each of which is concentrated on a finite set, and as
m -+ 00 they converge weakly to w+ and w- respectively. Let wn = w; - w; and

An(Z) = C exp Ut In Iz ~ "Wn(dO}.

Since the measure Wn is concentrated on a finite set, the metric PAn is polyhedral.
By Theorem 7.3.1, as n -+ 00 the metrics PAn converge to the metric PA in a
neighbourhood of the point Zoo The absolute curvature of the polyhedron
(U, p;.J is equal to w;(U) + w;(U) ~ M = const < 00. Application of Theo-
rem 6.2.2 now enables us to conclude that (G, PA) is a manifold of bounded
curvature.
We now show how to realize the given programme in detail. First of all we
show how to represent the function A(Z; w, h) in the form (7.21).
We take a point Zo E G arbitrarily. Let b > 0 be such that the closed ball
Bo = B(zo, b) c G. We have

In A(Z; w, h) = ~n Jr
Bo
ln l
Z
~ YIwed,) +
\,
1
G\Bo
ln l
Z
~ Ylw(dO + h(z).
\,

We denote the second integral on the right by hl(z). The function hl(z) is har-
monic in the disc Bo. Using the well-known representation of a harmonic func-
tion in a disc in terms of the values of its normal derivative along a circle, for
Z E B(zo, b) we obtain

(7.22)

1
where vo(O) = v(O) + VI (0), C = const > 0,

v(O) =b 1ov
0
9 oh.
(Zo + be'''') d<p, VI (0) =b
9
0
oh l
Tv(Zo
.
+ be'''') d<p.
The quantity VI (0) has the representation

VI (0) = 1 <pc"~
G\Bo
To)w(dO,

where To is the arc {zlz = Zo + be i"', 0 ~ <p ~ O} of the circle {zllz - zol = b},
and <pC', To) is its angular function. Obviously v(O) and VI (0) are functions of
bounded variation. Because of this the representation for the function A(Z) in the
114 Yu.G. Reshetnyak

disc Bo can be written as follows:

In A.{z) = ~ Lo In Iz ~ CI w{d() + In C.
Here w{E) = w{E) + O'{E), where 0' is a measure concentrated on the circle
{zllz - zol = t5}.
We now construct a sequence of polyhedral metrics defined in the disc Bo and
convergent in a neighbourhood of Zo to the metric P;.. We first assume that
w{ {zo}) < 2n. On the basis of Theorem 6.2.2 this will prove that the domain G
with metric P;. is a two-dimensional manifold of bounded curvature.
Let {A 1 ,A 2, . .. , Arn} be a partition of the disc Bo into pairwise disjoint Borel
sets, the diameter of each of which does not exceed lin, where n is an arbitrarily
specified natural number. For each i = 1, 2, ... , r" we choose a point Zi E Ai
arbitrarily and let w! and w; be measures in the disc Bo concentrated on a finite
set {Zl' Z2' ... , zrJ, and let w!{{z;}) = W+{Ai)' w;{{z;}) = w-{A;), It is easy to
show that as n - t 00 the set functions w! and w; defined in this way converge
weakly to the set functions w+ and w- respectively. We put w" = w! - w; and
define A.,,{z) by putting
A.,,{Z) = CA.{z; w,,),
where C is the constant on the right-hand side of (7.22). The metric P;. defined by
the function A.,,{z) on the plane is polyhedral. By Theorem 7.3.1, as n - t 00 the
functions P;'n converge uniformly to the metric P;',Bo on any closed set A c Bo
not containing points z with w+ ( {z}) ~ 2n. Since the metric P;. is intrinsic, there
is an 6 E (0,1) such that P;.,Bo(Zl' Z2) = P;,(zl' Z2) for Zl' Z2 E B(zo, e). We thus
have a sequence of polyhedral metrics P;'n converging to the metric P;. in the
neighbourhood B(O, e) of the point Zo = 0. The absolute curvatures of the poly-
hedral metrics PAn are bounded in aggregate. The sequence of metrics that we
have constructed actually satisfies the conditions of Theorem 6.2.3.
It was assumed above that w( {zo}) < 2n. We now consider the case when
w({zo}) = 2n and Zo is not a point at infinity with respect to the metric P;.. In
the given case the approximating polyhedral metric is constructed in two steps.
We first replace the set function w(E) by the one obtained from it by a uniform
distribution of the load of the set function w at the point 0 over a small neigh-
bourhood. The function A. corresponding to the modified function w determines
the metric P;.. We then approximate the metric P;. by a polyhedral metric, as
indicated above. Formally the construction is carried out as follows. We have

In A.(z) = -1
n
ff
Bo\{zol
1 w{d()
In-I-YI
Z - ..
+ 21n 1Z - 1 Zo 1 < 21n 1Z - 1 Zo 1+ In A.o(z)

i
and let

1 2 ,. 1
Yh{Z) = -
n 0
In 1Z - Zo -
he i"'l dcp.
 I. Two-Dimensional Manifolds of Bounded Curvature 115

1 1
Then 'Yh(Z) = 21n I when Iz - zol > h, and 'Yh(Z) = 21n-h when Iz - zol ~
Iz - Zo
h. We denote by c5n(E) a measure uniformly distributed along the circle
{zllz - zol = h}, where its value on the whole circle is equal to 2n. Then

'Yh(Z) = 1nfIn Iz _1 " c5h(dn

We define Ah(Z) by putting

In Ah(Z) = ~ ff In Iz ~ ,,&(dO + 'Yh(Z) = ~ f fin Iz ~ "wh(dO,

Bo\{O} Bo

where the measure wh(E) is defined by

wh(E) = &(E\ {zo}) + c5h (E).
Obviously Ah(Z) = A(Z) when Izi ~ hand Ah(Z) ~ A(Z) for all z. Hence we conclude
that as h -+ 0 the metric P;'h converges to P;. uniformly in the disc Bo. For t: > 0
we choose hE (0,1) such that Ip;'h(Zl' Z2) - PA(Zl' z2)1 < t:12 for all Zl, Z2 E Bo·
For the function Wh we have Wh({ZO}) = O. From what we have proved there
is a polyhedral metric PI' such that IpAh(Zl,Z2)-pizl,Z2)1<t:12 for any
Zl, Z2 E Bo, and the absolute curvature of the metric PI' does not exceed

IWhl(Bo) ~ 2n + 1&I(Bo\{zo}) = &(Bo)·

Thus, in a neighbourhood of each point of the domain G that is not a point
at infinity in the sense of the metric PA there is a sequence of polyhedral metrics
satisfying the conditions of Theorem 6.2.2. We have thus proved that (G, PA) is
a two-dimensional manifold of bounded curvature.
In particular, we finally obtain a proof of the fact that any two-dimensional
Riemannian manifold is a manifold of bounded curvature.
The proof of Theorem 7.1.2 relies on Theorem 6.2.1. Let X be a point of a
two-dimensional manifold of bounded curvature. Then according to Theorem
6.2.1 the point X has a boundedly convex neighbourhood U, homeomorphic to
a closed disc, in which we can specify a sequence of polyhedral metrics Pn,
n = 1, 2, ... , that converges uniformly in U to the metric of the manifold and
such that the sums of the absolute turns of the boundary and of the absolute
curvatures of the polyhedra (U, Pn) do not exceed some constan"t M < 00 that
does not depend on n. In the statement of Theorem 6.2.1 that we gave above we
spoke of the possibility of approximation by Riemannian metrics. However, up
to now nothing has been said about how to obtain such an approximation. We
now have at our disposal all the necessary methods for constructing an approxi-
mation of the metric by Riemannian metrics. This is obtained by smoothing the
polyhedral metrics that converge to the metric of the manifold in the given
domain U.
We first show how to smooth the metric of a circular cone and a circular
sector. For this we use an isothermal coordinate system. As we mentioned
116 Yu.G. Reshetnyak

above, the plane C, endowed with the metric P;. generated by the line element
A(Z, ex)ldzI 2, where A(Z, ex) = (l/lz!)"'/", OJ < 2n, is isometric to the cone
Q(2n - OJ). Let C+ be the half-plane {zlRe Z ~ O}. The half-plane C+, endowed
with the same metric, is isometric to the angular domain A(n - (co/2)). Smooth-
ing the metric of the cone is brought about, roughly speaking, by spreading the
curvature over some neighbourhood of the point O. This is done as follows.
Let tjJ be the function of the real variable t defined by the following condition:
ljJ(t) = 0 when t ~ 0, tjJ(t) = exp( -l/t) when t > O. The function IjJ belongs to
the class Coo. We put H(z) = H(x, y) = CtjJ((1/4) - IzI2), where C > 0 is a con-
stant. The function H is non-negative and belongs to the class Coo, H(z) = 0
when Izl > 1/2, and H(O) > O. We define the constant C from the condition

ff R2
H(x, y) dx dy = 1.

t
Let us specify a number h > 0 arbitrarily and put

)'h(Z; OJ) = exp{~ ln lz ~ (I' :2H(D de d1J}.

The function )'h(Z) == Ah(Z; OJ) belongs to the class Coo, Ah(Z) = (l/1z!)"'/" when
Izi > h/2. In C we consider the Riemannian metric Pl h ' As h ~ 0 this metric
converges to the metric of the cone Q(2n - OJ). The Gaussian curvature of the
metric P)'h at the point Z is equal to :z H(z/h)/Ah(Z). Hence it follows that the
absolute curvature of the plane in the metric Pl is equal to Iwl. The same metric
P),h' considered in the half-plane C+ as h ~ 0, converges uniformly to the metric
of the angular domain A(n - (OJ/2)). In view of the fact that the function H(z)
depends only on Izl, the metric P;'h is invariant under rotations around the point
O. Any line passing through the point 0 is a geodesic. The absolute curvature of
the half-plane C+ in the metric Pl is equal to

IOJI
Ji2 fc+ H
(z)11 dx dy = 2'
IOJI

The disc B(O, r) on the plane C with metric Pl is isometric to a finite circular
cone. When h < r the metric Pl h outside the disc B(O, h/2) coincides with the
metric Pl and as h ~ 0 these metrics converge uniformly to the metric of the
cone Q(2n - co, r).
Let P be an arbitrary polyhedron homeomorphic to a closed disc, and 0 > 0
the smallest of the distances between its vertices. Let X be an arbitrary vertex of
P, and U = B(X, r), where r < 012, its circular neighbourhood on the poly-
hedron P. If X is an interior vertex of P, then U is isometric to the cone Q(O, r),
where () = (}(X), and if X is a boundary vertex, then U is isometric to the circular
sector A((}, r). Let us map U isometrically into a plane, endowed with the line
element (1/Iz!)"'/"ldzI 2 so that to the point X there corresponds the point O. In
the case when X is an interior vertex, the image of U is a disc, and if X is a
 I. Two-Dimensional Manifolds of Bounded Curvature 117

boundary vertex the image of U is a half-disc. We shall assume that the diameter
of this half-disc lies on the line Re z. We now construct the approximating
Riemannian metric P;'h' where h is sufficiently small, h < ho. Transferring the
metric P;'h back to the disc B(X, r), we obtain a Riemannian metric on P that is
specified in a neighbourhood of the point X. Carrying out the given construc-
tion for all the vertices of P, we obtain on it a family of Riemannian metrics Ph
that depend on a parameter h E (0, ho). As h -+ 0 these metrics converge uni-
formly to the metric of the polyhedron P. It is not difficult to see that the
absolute curvature of each ofthem is equal to Iwl(P) + 1"1 (oP). The turn of the
boundary in the metric Ph is equal to zero.
We now consider the original manifold of bounded curvature. We have a
point X, its neighbourhood G homeomorphic to a closed disc, and a sequence
of polyhedral metrics (p,,) defined in G and converging as n -+ 00 to the metric
of G such that the quantities IWpnl(G) + I"Pnl(G) are bounded in aggregate.
Approximating the metrics p" by a Riemannian metric, as we mentioned above,
we obtain a sequence of Riemannian metrics (r,,), n = 1, 2, ... , converging in G
to the metric of the manifold and such that the sequence of absolute curvatures
{lwrnl(G)), n = 1, 2, ... , is bounded, and the turn of the boundary in each of
these metrics is equal to zero.
By the theorem on the existence of isothermal coordinates in a Riemannian
manifold, for each n in the disc iit = ii(O, 1) we can introduce a Riemannian
metric P" defined by the line element ds 2 = A,,(Z) Idzl 2 such that the disc iit in this
metric is isometric to the domain G with metric r". We can assume that to the
point 0 there corresponds by isometry a given point X. The function Aiz) has
the representation

A,,(Z) = exp UII In Iz ~ CI w,,(d() + h,,(z)} ,

where h" is a harmonic function and the sequence (lw"I(Bd), n = 1, 2, ... , is
bounded. The turn of the circle nO, 1) in each of the Riemannian metrics P;. is
equal to zero. Because of this the function A" has the representation n

A,,(Z) = y"i,,(z) = y" exp {~ LI In Iz ~ CI w,,(dC)} ,

where y" > 0 is a constant and the variations of the set functions w,,(E) are
bounded in aggregate.
From the sequence of set functions (w,,) we choose a subsequence for which
the functions w: and w;; converge weakly to certain set functions WO and w t
respectively. Let Wo = WO - wt. We have i,,(z) = A(Z, w,,). Let io(z) = A(Z, wo).
For simplicity we shall assume that the chosen sequence coincides with the
original one. On the basis of Theorem 7.3.1 the Riemannian metrics Pi. con-
verge uniformly to the metric Pi.o on any closed subset of the disc iit that does
not contain points at infinity with respect to the
_ metric Pi. 0 . The metric P;.n differs
from Pin by a constant factor y". The disc B t with metric P;'n is isometric to
(G, rIll. Since the metrics p" converge to some metric P;. as n -+ 00, this enables us
118 Yu.G. Reshetnyak

to conclude that the factors Yn converge as n -+ 00 to some finite non-zero limit

Yo, and the metric Pi. o does not have points at infinity. Hence it follows that
_ the
metrics P;'n converge to the metric P;'o = YoPi. o as n -+ 00. Since the disc B1 with
metric P;'n is isometric to (G, rn ), it follows that the disc B1 with metric P;'o is
isometric to the domain G endowed with the natural metric of the manifold.
Thus Theorem 7.1.2 is proved.
The proof of Theorem 7.1.3 is based on an additional assumption, which is
expressed by Lemma 7.3.2 stated below. As a preliminary we describe some
constructions that are necessary for the statement of this lemma.
Let G be a domain in C, and A.(z) = A.(z; w, h) a function defined in G. We
specify a point Zo E G arbitrarily. For r > 0 we put 1,(z) = A.(zo + rz). The
domain of definition of the function 1,(z) is the set (G, r) obtained from G by
the transformation z -+ z - zo. Obviously there is an ro > 0 such that when
r
0< r < ro the domain G, contains the disc K2 = B(O, 2). We define the con-
stant Y, > 0 from the condition

r
llzl=1
Jy,l,(z)ldzl = 2n

and let A.,(z) = y,l,(z).

For z E C we put lp,(z) = Zo + rz. We have A.(z) = A.(z; w, h). Let w, be the
measure defined by the condition w,(E) = w[lp,(E)].1t is not difficult to see that
).,(z) = ).(z, w" C,h,), where C, = In y" h,(z) = h(zo + rz). We shall assume that
0< r < roo In this case, in the disc K2 there is defined a metric P;'r,K 2 • We put
wo = w({zoH and let ,l.o(z) = Izl-woi". As r -+ 0 the set functions w,(E) converge
weakly to the measure 0 defined by the condition woo(E) = 0 if 0 ¢ E and
woo(E) = Wo if 0 E E. We have A. == A.(z; w, h). The theorem on convergence of
metrics (Theorem 7.3.1) enables us to conclude that the following proposition is
true.
Lemma 7.3.2. If the point Zo E G and the functions A.(z) == A.(z; w, h) are de-
fined in G ~ that w( {zo}} < 2n, then as r -+ 0 the metric P;'.,K 2 converges in the
disc K2 = B(O, 2) to the metric P;'o,K 2 •
Let Zo be an arbitrary point on the plane C, and (rm), m = 1,2, ... , a sequence
of simple closed curves such that the point Zo lies inside each of them. We put
rm = inf Iz - zol, Rm = sup Iz - zol. We shall say that the sequence of curves
zer". zerm
(rm) converges regularly to the point Zo if Rm -+ 0 as m -+ 00 and the ratio rm/Rm
tends to 1 as m -+ 00.
Theorem 7.3.2 (Men'shov (1948». Let G and H be open subsets of the plane C,
and lp: G -+ H a topological map of G onto H. We assume that for all points z E G,
excluding points that generate a no more than countable set, the following condi-
tion is satisfied: for any sequence (rm) of simple closed curves lying in the domain
G that converge regularly to the point z their images lp(rm) converge regularly to
the point w = lp(z). Then the map lp is conformal.
 I. Two-Dimensional Manifolds of Bounded Curvature 119

We assume that in G there is specified a metric p and let (Fm), m. = 1,2, ... , be
a sequence of simple closed curves such that a given point Zo of G lies inside each
of them. We shall say that the sequence (Fm), m = 1,2, ... , converges regularly
to the point Zo in the sense of the metric p if the ratio rm/Rm, where rm =
inf p(z, zo), Rm = sup p(z, zo), tends to 1 as m ~ 00.
zeFm zerm
Let p be the metric p)" where A = A(Z; w, h). Lemma 7.3.2 enables us to con-
clude that the sequence of closed curves (rm) converges regularly to a point
Zo E G such that w( {zo}) < 2n if and only if it converges to Zo regularly in the
sense of the metric p),.
Let G and H be domains on the plane C, and let p), and Pp. be metrics
defined in G and H respectively by line elements A(Z) IdzI 2 and Jl(w) Idwl 2 , where
A(Z) = A(Z; w, h), Jl(w) = Jl(w; e, g), wand e are measures defined in G and H
respectively, and hand g are harmonic functions. We assume that the metric
spaces (G, p),) and (H, Pp.) are isometric and that qJ is an isometric map of(G, p),)
onto (H, P,J. The map qJ is topological. Any sequence of closed curves converg-
ing to the point Z E G regularly in the sense of the metric p), is transformed by
the map qJ into a sequence of closed curves converging to the point W = qJ(z)
regularly in the sense of the metric Pp.' By the previous remark, a sequence of
curves that converge to the point Z regularly in the sense of the metric p), con-
verges regularly to this point. An exception is formed by points Z for which
w( {z}) ;:: 2n, e( {w}) ;:: 2n, where W = qJ(z). Consequently, the map qJ satisfies
the conditions of Men'shov's theorem, and so it is conformal. This proves the
assertion of Theorem 7.1.3 about the conformal property of transition functions
of two distinct isothermal coordinate systems in a two-dimensional manifold of
bounded curvature.
Formula (7.4) on the transformation of a line element on going over from one
isothermal cordinate system to another is proved by arguments similar to those
used above on deriving the formula for transforming the coefficients of the
metric tensor of a Riemannian manifold on going over from one local coordinate
system to another. The fact that in the given case A(Z) and Jl(w) may be discon-
tinuous functions somewhat complicates the arguments. However, it is easy to
overcome the difficulties of technical character that arise here.
Theorem 7.1.4, like Theorem 7.1.5, is a simple consequence of the preceding
results contained in Theorems 7.1.1-7.1.3. For this reason we shall not dwell on
its proof.

7.4. On the Proof of Theorem 7.3.1. Theorem 7.3.1, a theorem on the conver-
gence of metrics, is an important instrument for the proof of theorems on the
analytic representation of two-dimensional manifolds of bounded curvature by
means of an isothermal line element. It is therefore necessary to say a few words
about its proof.
Suppose that the conditions of Theorem 7.3.1 are satisfied, that is, we are
given a domain G whose boundary is the union of finitely many simple closed
curves of bounded rotation. We assume that in G there are specified sequences
120 Yu.G. Reshetnyak

of non-negative measures (W~), (W;), n = 1,2, ... , that converge weakly to cer-
tain measures WI and w 2 respectively. We put Wn = w~ - w;, w = WI - w 2, and
let An(Z) = )..(z, wn), )..(z) = )..(z, w). Metrics PAn and PA are defined in G. We need
to prove that on any compact set A c G that does not contain points Z such that
WI ({ z}) ~ 211:, PAJZ 1, Z2) converges uniformly to PA(Z l' Z2) as n -+ 00.
For the proof of this it is sufficient to show that if (z In), (Z2n), n = 1,2, ... , are
two arbitrary sequences of points of G that converge as n -+ 00 to points Z 1 and
Z2 such that w l ( {ZI}) < 211: and w l ( {Z2}) < 211:, then PAjzln' z2n) -+ PA(ZI' Z2) as
n -+ 00. We shall prove this last assertion.
We recall that PA(ZI' Z2) is the greatest lower bound of the quantity

sA(K) = LJVz) ds

on the set of all curves K with finite absolute rotation that join Z 1 and Z 2 and are
contained in the domain G.
We shall say that a measure w defined on the plane C is regular if it is the
indefinite integral of a function of class COO. We assume that the measures w!
and w; in the statement of Theorem 7.3.1 are regular (we suppose that measures
defined in G can be extended to the plane C by means of the stipulation that
w(E) = 0 if E 11 G = 0). In this case the metrics PA are Riemannian. It is suffi-
cient to establish the truth of Theorem 7.3.1 for this case. In fact, if this is done,
then in the general case we can proceed as follows. For each n we approximate
the measures w~ and w; by regular measures w~ and w;. By hypothesis, for the
case when the measures w~, i = 1,2, ... , are regular Theorem 7.3.1 is true. Let
us construct regular measures w~ and w; sufficiently close in the sense of weak
topology to the measures w! and w; respectively. We put in = )..(z, w! - w;).
For each n there is defined in the domain G a Riemannian metric Pi . If w! and
w; are sufficiently close to the measures w! and w;, then as n -+ 00 the measures
w~ converge weakly to the measure WI' and w;to w 2. Since Theorem 7.3.1 is
assumed to be proved for the case when the measures w! and w; are regular,
this enables us to conclude that Pi (zln' Z2n) -+ PA(ZI' Z2) as n -+ 00 (here in =
}.(z, w! - w;)). Using again the fact that Theorem 7.3.1 is assumed to be proved
for the case of regularity of the measures w~, i = 1, 2, ... , we can assert that
if the measures w~, i = 1, 2, ... , are chosen sufficiently close to w~, then
IP;)Zln' z2n) - PAjz ln' z2n)l-+ 0 as n -+ 00. Since Pij z ln, Z2n) -+ PA(ZI' Z2)' it
follows that PA (zln' z2n) -+ PA(ZI' Z2) as n -+ 00.
Thus, the g~neral case reduces to the case when the measures w! and w;,
n = 1,2, ... , are regular, so the metrics PAn are Riemannian. Henceforth we shall
assume that this condition is satisfied.
The subsequent arguments rely on the next two propositions.
Lemma 7.4.1. Suppose that the sequences of measures (w!), (w;), n = 1,2, ... ,
satisfy all the conditions listed above. We assume that we are given a sequence of
curves (Kn) lying in G that converges to a curve K that also lies in G. If the
absolute curvatures of the curves Kn are bounded in aggregate, l"I(K n) ~ M =
 I. Two-Dimensional Manifolds of Bounded Curvature 121

const < 00 for all n and the curve K does not contain points z for which
wl({z}) ~ 2n, then

as n --+ 00.

The proof of this lemma is based on arguments of purely technical character.

Lemma 7.4.2. Let w~, w~, n = 1, 2, ... , be sequences of regular measures

defined in a closed domain G and satisfying all the conditions stated above. Then
for any point z E G such that WI ( {z}) + w 2 ( {z}) < 2n there are numbers fJ > 0
and M < 00 having the property that for any curve that lies in the set G n B(z, fJ)
and is a shortest curve, in at least one of the Riemannian metrics P;.. the absolute
rotation does not exceed M. "

In Lemma 7.4.2 shortest curves containing boundary points are admitted.

The proof of the lemma is based on the use of formula (7.18) for the tum of a
curve in an isothermal coordinate system. We assume that curve L is a shortest
curve in the metric P;.., where A. = A.(z; w), and lies in the disc B(zo, r) c G. We
assume that the measure w is regular. Let z(t), a ~ t ~ b, be a parametrization
of L, and let K(t) and cp(t, 0 be the rotation and angular function of the arc of L
corresponding to values of the parameter lying in the interval [a, t]. The geode-
sic tum of L is equal to zero, and so by (7.18) we have

K(t) = 2~ L cp(t, C)w(dE,). (7.23)

(We are considering the case when L does not contain boundary points of G.)

i
Hence

IKI(L) = Vb K(t) ~
1
-2 Vb cp(t, Olwl(dE,). (7.24)
a n G a

b
The quantity V cp(t, 0 is estimated as follows. On the one hand, we have
a
b
V cp(t, 0 ~ K(L) + n. (7.25)
a

On the other hand, if the point Cis at a distance h from L, then

b (Y)
V s(L) 4r[K(L) + n]
cpt,., ~-~ . (7.26)
a h nh
The result of Lemma 7.4.2 is obtained by a combination of these two esti-
mates. We assume that L does not overlap the boundary of G. If this is not so,
then on the right-hand side of (7.23) there appears a term that depends only on
G and r; otherwise the arguments do not differ from the case when L does not
contain boundary points of G.
122 Yu.G. Reshetnyak

The proof of Theorem 7.3.1 is now realized as follows. We specify e > 0

arbitrarily. and specify a curve K of bounded curvature joining the points Zl and
Z2 and such that s;.(K) < P;,(ZI' Z2) + e. On the curve K there may be points Z
for which w l ( {z}) ~ 2n. Replacing a small arc of K containing such a point Z by
an arc of a circle with centre z. we can arrange that K does not pass through
such points. Let Kn be the curve obtained from K by joining the intervals
[Zl' Zln] and [Z2. z2nl As n -+ 00 the curves Kn converge to the curve K and
IKI(Kn) ~ IKI(K) + 2n for each n. so the absolute rotations of the curves Kn are
bounded in aggregate. By Lemma 7.4.1
s;.JK n) -+ s;.(K) < P;,(Zl' Z2) +e
as n -+ 00. For each n

and so

hence. since e > 0 is arbitrary. we conclude that

lim P;. n (Z In. Z2n) ~ P;.(Z 1. Zz)· (7.27)
n.... oo

We now show that

(7.28)
n.... oo

We single out all points Z for which wl({z}) + w 2 ({z}) ~ 2n and which are not
points at infinity in the sense of the metric P;.. The set E of such points Z is finite.
We shall assume that Zl ¢ E and Zz ¢ E. We specify e > 0 arbitrarily and for
each point Z E E we construct a disc with centre Z so that these discs do not
overlap and the sum of the lengths of their circles in the metric P;. is less than e.
Let Kn be the shortest curve in the metric p;'njoining the points Ztn and ZZn. We
have
s;.JKn) = P;.JZtn. Z2n)·
The curve Kn can be inside some of the discs constructed above. Replacing
each arc of Kn contained in such a disc by an arc of the circle of the disc. we
obtain a curve K~joining the points Ztn and Z2n. For sufficiently large n we shall
have

Applying Borel's theorem on covering and using Lemma 7.4.2. it is not difficult
to show that the absolute rotations of the curves K~ are bounded in aggregate.
Without loss of generality we can assume that as n -+ 00 the curves K~ converge
to some curve K joining the points Zl and Zz. As n -+ 00 we have
P;.JZln. zzn) ~ sdK~) - 2e -+ s;.(K) - 2e ~ P;,(Zl' zz) - 2e.
 I. Two-Dimensional Manifolds of Bounded Curvature 123

Hence
lim p).JZln, Z2n) ~ pAZl' Z2) - 28
n-+oo

and since 8 > 0 is arbitrary the inequality (7.28) follows. The required relation
follows from (7.27) and (7.28):
P)'(Zl' Z2) = lim p).JZln' Z2n)'
n-+oo

In the proof of (7.28) we have assumed that (Ol({Z}) + (02({Z}) < 2n for
points Z = Z 1 , Z2' If this condition is not satisfied, then the required result can be
deduced from what we have proved by a small shift of the points Zl and Z2'

Chapter 3
Basic Facts of the Theory of Manifolds
of Bounded Curvature

§ 8. Basic Results of the Theory of Two-Dimensional

Manifolds of Bounded Curvature

8.1. A Turn of a Curve and the Integral Curvature of a Set. The concepts of a
turn have been defined for curves in a Riemannian manifold and on a manifold
with polyhedral metric. Next for such manifolds we also defined the concept of
the integral curvature of a set. Our aim is to extend these concepts to the case of
arbitrary two-dimensional manifolds of bounded curvature. Here, in contrast to
the monograph of Aleksandrov and Zalgaller (1962), we shall rely on the ana-
lytic apparatus at our disposal - the representation of the metric of a manifold
of bounded curvature by the isothermal line element ds 2 = A(z)ldzI2. We have
A(Z) = A(Z; (0, h). We establish the geometrical meaning of the set function (0 and
then solve the problem of introducing the set function known as integral curva-
ture. This is preceded by a study of the concept of the turn of a simple arc in a
manifold whose metric is defined by an isothermal line element. We first define
the turn formally by analogy with the case of curves in a Riemannian manifold.
Having established the geometrical meaning of the quantities introduced for-
mally by means of isothermal coordinates, we obtain an answer to the question
of how to regard the turn of a curve in an arbitrary two-dimensional manifold
of bounded curvature.
The total angle and curvature at a point of a two-dimensional manifold of
bounded curvature. Let M be a two-dimensional manifold of bounded curvature,
U a domain in M homeomorphic to a disc, and qJ: U -+ C an isothermal coordi-
nate system in M. Let G = qJ(U). Then in G there is defined a function A(Z) ==
124 Yu.G. Reshetnyak

A(Z; w, h) and a metric P;. in it such that cp is an isometric map of (U, Pu) onto
(G, P;.).
Let K be a simple arc on a plane, a its beginning and b its end. We take a
point Zo E K arbitrarily. Let Z be a point of K other than zoo We denote by
l(zo, z) the ray with beginning at Zo and passing through z. If l(zo, z) converges
to some limiting ray 10 when z tends to Zo from the right (left), we shall say that
the curve K has right (left) half-tangents at the point zoo We shall denote the left
half-tangent of K at Zo by l,(zo, K) and the right half-tangent by l,(zo, K).
Let K be a simple arc in a domain U of a manifold of bounded curvature.
Then we shall say that K belongs to the class J if its image in the isothermal
coordinate system has half-tangents at its end-points. The property that a curve
has left and right half-tangents at any point of it is preserved by conformal
maps. So clearly the property that a curve belongs to the class J does not
depend on the choice of isothermal coordinate system in the domain U. As we
shall show later, J coincides with the class of curves that have a definite direc-
tion at their end-points.
Let X be an arbitrary point of the domain U, and z = cp(X) the corre-
sponding point on the plane C in the given isothermal coordinate system. In the
domain G = cp(U) there is defined a set function w. We show that w({z}) does
not depend on the choice of isothermal coordinate system.
Let F(X, r) be a circle in a two-dimensional manifold of bounded curvature,
X E U its centre, and r its radius.

Lemma 8.1.1. Let X E U, Z = cp(X). Then there is a number (jl > 0 such that
if 0 < r <b 1 , then F(X, r) is a simple closed curve. Let u(X, r) be the length of the
curve F(X, r). Then
u(X, r)
- - -+ O(X) = 2n - w( {z} ) (8.1)
r

as r -+ O.

We shall denote the quantity w( {z}) (see the statement ofthe lemma) by w(X)
and call it the curvature of the manifold M at the point X. We shall call O(X) =
2n - w(X) the total angle at the point X.
We shall call a point X E U a peak point if w(X) = 2n.
There arises the question of studying the structure of a circle of arbitrary
radius in a two-dimensional manifold of bounded curvature. The papers of
Zalgaller (1950b) and Burago and Stratilatova (1965) are devoted to this. If the
radius of the circle is sufficiently large, then generally speaking the circle is not
a simple closed curve. However, this may hold also in the case of Riemannian
manifolds. Burago and Stratilatova gave a definition of the length of a circle in
a two-dimensional manifold of bounded curvature for the case when the circle
is not a simple closed curve. They also mentioned a formula that gives an
expression for the length of a circle and the area of the disc bounded by it in
terms of other characteristics of the manifold.
 I. Two-Dimensional Manifolds of Bounded Curvature 125

We note that in the case when a two-dimensional manifold of bounded cur-

vature is a cone of curvature Wo with vertex X, then, as we showed above,
/1(Xo, r) = (211: - wo)r,
so the definition given here agrees with the definition that we have in the case
when the manifold is a cone, and hence also in the case when it is a polyhedron.
Let Kl and K2 be two simple arcs in a domain U that have a common
starting point X. We assume that Kl and K2 have no common points other
than X, and that the arcs Ll = lI'(Kd, L2 = lI'(K 2) have right half-tangents at
the point z = lI'(X). The curves Kl and K2 split a neighbourhood of X into two
sectors Sl and S2. Let Xl and X2 be the images of these sectors under the map
lI': U ..... Co The curves Ll and L2 have right half-tangents at the point z. Let '1
be the half-tangent of Ll and '2 the half-tangent of L 2. The rays 11 and '2 split
the plane C into two angular domains V1 and V2 • The value of one of them is
naturally called the angle of the sector Xl' and the value of the other is called
the angle of the sector X2 (see Fig. 37). Let (x' be the angle of the sector Xl, and
(x" the angle of the sector X 2 • Let Wo = w(X). We put

(Xl = (1 211:o)(x,
- w , (X2 = (1 211:o)"
- w (x. (8.2)

We shall call (Xl and (X2 the angles of the sectors Sl and S2.
The next proposition establishes the geometrical meaning of the quantities (Xl
and (X2.
Lemma 8.1.2. We assume that the simple arcs Kl and K2 starting from a point
X EU are such that the curves Ll = lI'(Kd and L2 = lI'(K 2) have right half-
tangents at the point z = lI'(X) and that Kl and K2 have no points in common
other than X. Then there is a number O2 > 0 such that when 0 < r < O2 the circle
F(X, r) in the manifold M is a simple closed curve and is split by Kl and K2 into
two arcs, one of which lies in the sector Sl and the other in the sector S2. Let /11(r)
be the length of the first are, and /12(r) the length of the second. Then

(8.3)

Fig. 37
126 Yu.G. Reshetnyak

Lemma 8.1.2 shows that the value of IXI and 1X2 is determined by the geo-
metrical properties of a small neighbourhood of the point Zo in a manifold of
bounded curvature.
The formal turn of a curve in isothermal coordinates. As above, let U be a
domain homeomorphic to a disc in a two-dimensional manifold of bounded
curvature, cp: U ~ C an isothermal coordinate system defined in U, G = cp(U),
A(z)ldzI 2 the line element of the manifold M in this coordinate system, and
).(z) == A(Z; ro, h).
For an oriented simple arc K on the plane C there is defined an angular
function Z 1--+ cp(K, z). Up to now we have assumed that cp(K, z) is defined only
at points not belonging to K. Now we need to define it in the case z E K also.
This definition is based on the following proposition.
Lemma 8.1.3. Let K be an oriented simple arc on the plane, Zo an arbitrary
internal point of it, a its beginning and b its end. If the sequence (zn), n = 1, 2, ... ,
converges to the point zo, having kept on one side of K, then there is a finite limit
(8.4)
n-+oo

We assume that K belongs to the class .J and let (an), (bn), n = 1, 2, ... , be se-
quences of points of the curve K such that a < an < bn < b, and let Kn be the arc
[anbn] of K. Then there are finite limits
(8.5)
n-+oo n-+oo

If the sequence (zn) converges to the point Zo E K on the right with respect to
K, then the limit (8.4) will be denoted by cp,(K, zo), and if Zn ~ Zo on the left of
K, then the limit (8.4) will be denoted by cp,(K, zo). Clearly, the value of the limit
(8.4) depends only on which side of K the sequence (zn), n = 1, 2, ... , converges.
The existence of the limits (8.5) is guaranteed by the fact that K has half-tangents
at a and b. We put cp(K, a) and cp(K, b) equal to the limits (8.5) respectively.
If Z is not an interior point of K, we put
cp,(K, z) = cp,(K, z) = cp(K, z).
We thus obtain functions z 1--+ cp,(K, z) and z 1--+ cp,(K, z) defined at all points of
the plane C. We note that if z is an interior point of K, then we have
cp,(K, z) - cp,(K, z) = 2n. (8.6)
We assume that in the representation A(Z) = A(Z; ro, h) the measure ro is regu-
lar, that is, it is the indefinite integral of a function of class Coo and K is a smooth
curve of class C2 in the domain G. Let z(s), 0 ~ s ~ I, be a parametrization of the
curve K, where the parameter s is arc length. In the given case the metric p.. is
Riemannian, and as we showed above (formula (7.18» the turn K;.(K) of the

f
curve K in this metric is expressed by the formula

1
K;.(K) = K(K) - 2n G
r
1 iJh
cp(K, z)ro(dz) -:2 JK iJv (z)ldzl·
 I. Two-Dimensional Manifolds of Bounded Curvature 127

Let K be a simple arc lying in the domain G, and h(z) a harmonic function in
G. We specify arbitrarily a neighbourhood V of the curve K homeomorphic to
a disc and contained in G. Let g be a harmonic function defined in V and
conjugate to h, that is, such that
og oh og oh
ox (z) = oy (z), oy (z) =- ox (z).

A function g satisfying these conditions exists since V is simply-connected. We

assume that the simple arc K is smooth. Let a be its beginning and b its end, and
let z(s), 0 ~ s ~ 1, be a parametrization of K, where the parameter s is arc length.
Then

L~~ (z) Idzl = L! (g[z(s)]) ds = g[z(l)] - g[z(O)] = g(b) - g(a).

We use the given relation to give a meaning to the integral

L~~(Z)ldZI
in the case when the simple arc K is not smooth. Namely, in this case we put

f K
oh
ov (z)ldzl = g(b) - g(a), (8.7)

where g is a harmonic function defined in a neighbourhood of K and conjugate

to h. It is easy to show that the difference g(b) - g(a) does not depend on the
choice of neighbourhood of K and the function g. (If the neighbourhood is
chosen, then g(z) is defined up to a constant term.)
We now assume that L is an arbitrary simple arc of class A in the domain U,
and K = cp(L). The left turn of the curve L in the coordinate system cp: U -+ C is
the quantity

1
K,(L) = K(K) - 2n
fG CPr(K, z)w(dz) - 21 fK oh
ov (z) Idzl. (8.8)

The right turn of L is the quantity

Kr(L) = -K(K) 1
+ 2n f G cp,(K, z)w(dz) + 21 f
K
oh
ov (z)ldzl. (8.9)

The second integral on the right-hand side of (8.8) and (8.9) is defined, as we
described above, in accordance with (8.7).
We shall not indicate the coordinate system cp in the notation for the quant-
ities K,(L) and Kr(L) since, as we shall show later, they do not depend on the
choice of this coordinate system.
Let KO be the totality of all interior points of the simple arc K = cp(L).
Adding (8.8) and (8.9) term by term and taking (8.7) into consideration, we obtain

K,(L) + Kr(L) = 21
n
f
KO
2nw(dO = w(KO). (8.10)
128 Yu.G. Reshetnyak

We now mention some simple properties of ",(L) and "r(L) that follow di-
rectly from the definition.
Theorem 8.1.1. Let LeU be a simple arc of class J, X an interior point of L,
and L1 and L2 the arcs into which K is split by the point X. We assume that the
arcs K1 = qJ(L 1), K2 = qJ(L 2) have half-tangents at the point z = qJ(X), and let
lJ,(X, K) and lJr(X, K) be the angles in the sense of the metric p.. at the point Y
between the arcs L1 and L2 on the left and right of L, respectively. Then

",(L) = ",(Ld + ",(L 2) + (n - lJ,(X, L)),

",(L) = ",(Ld + ",(L 2) + (n - lJr(X, L)).

The proof of the theorem follows directly from the representations of ",(L)
and ",(L), which are given by (8.8) and (8.9).
Next we put

n - lJ,(X, L) = ",(X, L), n - lJr(X, L) = ",(X, L)

and the quantities ",(X, L) and ",(X, L) will be called the left turn and right turn
of the curve L at the point X.
We assume that the domain U is oriented by the stipulation that the iso-
thermal coordinate system qJ: U -+ C specified in it is right. For E c U we put
w(E) = w[qJ(E)].

Theorem 8.1.2. Let L1 and L2 be two simple arcs of class J with common
beginning A and common end B, lying in the domain U. We assume that L1 and L2
have no points in common other than A and B, and that L2 lies to the right of L 1.
Let D be the domain included between the curves K and L, and let Q( and Pbe the
angles of D at the points A and B. Then

",(L 2 ) = ",(L 1 ) + Q( + P- w(D u KO),

",(L 2 ) = ",(L 1 ) - Q( - P+ w(D u KO).
Corollary. Let Land L n, n = 1, 2, ... , be simple arcs of class J joining the
points A, BE U, where LeU and Ln c U for all n. We assume that for each n
the curves Land Ln have no points in common other than A and B. Let Q(n and Pn
be the angles of the domain Dn included between Land Ln at the points A and B
respectively. If the curves Ln converge to the curve L on the right, then
",(L) = lim (",(L n) + Q(n + Pn).
n-+oo

Similarly, if Ln -+ L on the left, then

",(L) = n-+oo
lim (",(L n) + + Pn).Q(n

Theorem 8.1.2 follows directly from the expressions for ",(L) and "r(L), which
are given by (8.8) and (8.9).
 I. Two-Dimensional Manifolds of Bounded Curvature 129

For the proof of the corollary we observe that if Ln for each n lies to the right
of L, then, according to the theorem,
Kr(Ln) + OCn + Pn = Kr(L) + iiJ(Dn U L~).
By virtue of the property of complete additivity of a measure, w(Dn n L~) -+ 0
as n -+ 00, so it follows that Kr(L) = lim (Kr(Ln) + OCn + Pn). The assertion about
n-+oo
the left turn is proved in exactly the same way.
Geometrical definition of the turn of a simple arc. Here we shall establish the
geometrical meaning of the quantities K,(L) and Kr(L) introduced above in
a formal way by means of the isothermal coordinate system qJ: U -+ C defined
in a domain U of a two-dimensional manifold of bounded curvature. The
main instrument will be the following theorem, which also has independent
significance.
Theorem 8.1.3. Suppose we are given a metric PA(Zh Z2), where l(z) ==
l(z; w, h), in an open domain G on the plane C. We assume that the curve K is a
shortest curve in G in the sense of the metric PA. If K does not contain peak points
of the metric PA (that is, such that w( {z}) = 21l), then K is a plane curve of bounded
rotation. The left and right turns of K in the metric PA (defined by (8.8) and (8.9»
are non-positive.
The proof of Theorem 8.1.3 is based on an approximation of the metric PA
specified in the domain G by Riemannian metrics. Here we shall not dwell on
the details, referring the reader to the article Reshetnyak (1963b). Let us make
some remarks about Theorem 8.1.3.
First of aH we observe that the limitation that consists in the absence of peak
points of the metric PI. on the curve is due to an important matter. In fact, it is
easy to construct an example of a manifold of bounded curvature for which the
two shortest curves K and L starting from the point A with w(A) = 21l are
arranged as in Fig. 38. The shortest curve K on approaching the point A makes
an infinite set of coils. Under any topological map of an arbitrary neighbour-
hood of the point A into a plane the image of at least one of the curves K and L
will not be a curve of bounded rotation.
In the case when the metric PI. is Riemannian, by virtue of classical results of
Riemannian geometry we can assert that the left and right turns of a shortest
curve are equal to zero. In the general case it is impossible to assert this; it is
easy to convince oneself in this just example of polyhedra.

Fig. 38
130 Yu.G. Reshetnyak

Let K be a shortest curve in a two-dimensional manifold of bounded curva-

ture that does not contain peak points. Any arc L of the curve K is also a
shortest curve, and so by Theorem 8.1.3 its left and right turns are non-positive.
Applying (8.10), we deduce that for any arc L ofthe curve K we have w(LO) ~ O.
This enables us to conclude that for any set E c KO that is open with respect to
K we have w(E) ~ 0, from which it follows that in general for any Borel set
E c KO the quantity w(E) is non-positive. Moreover, the relations "I(L) ~ 0,
",(L) ~ 0, "I(L) + ",(L) = w(KO) enable us to conclude that for any shortest
curve L that does not contain peak points we have

I"I(L)I ~ Ico(LO)1 = Icol(LO),

(8.11)
1",(L) 1~ Ico(LO)1 = Icol(Lo).
If the measure co in the domain Gee is non-negative, then points Z such
that co( {z} ) = 2n are points at infinity with respect to the metric P;., where
A(Z) == A(Z; co, h), so in this case the metric P;. does not have peak points. For
suppose that the point Zo is such that co( {zo}) = 2n. Then in the disc B(zo, r),
where 0 < r < 1/2, we have

lnA(z)=-1
n
fB(zo.,) Z
1I
I n_Ylco(dEc)+u(z),
-
..

where u(z) is a harmonic function. When Z E B(zo, r), 'E B(zo, r) we have

Iz - " < 2r < 1 and so In Iz ~ el > O. Hence we conclude that in the disc
B(zo, r/2)

In A(Z) ;?; 2ln, 1 I+

Z - Zo
i
B(zo.')\{ZO}
1 co(dEc) + u(z)
In -,-;:,
Z - ..
;?; 2ln, 1 , + C,
Z - Zo
where C is a constant, and so
A(Z);?; eC'lz - zor 2

for any Z E B(zo, r/2). Hence it is clear that for any curve K starting from the
point Zo we have s;.(K) = 00, as we needed to prove.
If the set function co is non-negative, then by what we have said any curve K
that is a shortest curve in the metric P;., where A(Z) == A(Z; co, h), does not contain
peak points ofthe metric P;., and so by Theorem 8.1.3 K is a curve with bounded
rotation on the plane. We have "I(K) ~ 0, ",(K) ~ 0, co(KO) ;?; O. The equality
(8.10) enables us to conclude that in this case "I(K) = ",(K) = O. Thus we obtain
the following result.
Theorem 8.1.4. If the measure co in the domain G is non-negative, A(Z) ==
A(Z; co, h), then any curve K that is a shortest curve with respect to the metric P;. is
a curve with bounded rotation and its left and right turns in the metric P;. are equal
to zero.
 I. Two-Dimensional Manifolds of Bounded Curvature 131

Let LeU be a shortest curve in the domain U of a two-dimensional mani-

fold of bounded curvature M. We assume that L does not contain peak points.
Then it follows from Theorem 8.1.3 that the shortest curve L, like any arc of it,
is a curve of class J. We take an interior point X of the shortest curve L
arbitrarily. Let (An), (Bn), n = 1, 2, ... , be sequences of points of L that converge
to X on the left and right respectively. Let Ln and Mn be the arcs [An X] and
[XBn] of the curve L, and Kn the arc [AnBn]. Let "I(X, L) and ",(X, L) be the
left and right turns of L at X. By Theorem 8.1.1 "I(K,,) = "1(Ln ) + "I(M,,) +
"I(X, L). According to Theorem 8.1.3 "1(Kn ) ~ O. As n -+ 00, "I(L,,) -+ 0 and
"1(Mn ) -+ O. This enables us to conclude that "I(X, L) ~ O. Similarly we can
show that ",(X, L) ~ O. We have thus established the following proposition.
Theorem 8.1.5. If a shortest curve L in a domain U of a manifold of bounded
curvature does not contain peak points, then at any point of it the left and right
turns are non-positive.
We shall define a geodesic polygonal line in a domain U of a two-
dimensional manifold of bounded curvature as any finite sequence L of oriented
simple arcs L 1, L 2 , ••. , L m, each of which is a shortest curve, where the end of
the arc L1 coincides with the beginning of the arc Li+1 when i = 1, 2, ... , m - 1
and the union of the arcs L l' L 2 , ••• , Lm is a simple arc i. The arc I is called
the carrier ofthe polygonal line L, the arcs L j , i = 1, 2, ... , m, are called its links,
and their end-points are called the vertices of L. We note that by dividing any of
the arcs L j into two we obtain another geodesic polygonal line.
If the carrier I of a geodesic polygonal line L does not contain peak points,
then the image K of I in the isothermal coordinate system cp: U -+ C is a curve
with bounded rotation. We denote by ",(L) the sum of the right turns of I at
interior points of it that are vertices of L. Similarly, let "I(L) be the sum of the
left turns of I at these points. We shall call ",(L) and "I(L) the conditional right
and left turns of L. Let X j - 1 and X j be the beginning and end of the shortest
curve L j • By Theorem 8.1.1 we have
_ m m-1 m

"I(L) = L "1(L + L
j=1
j)
j=1
"1(Xj , L j ) = L "1(L + "I(L).
j=l
j)

Since L j is a shortest curve, by (8.11) it follows that

I"I(I) - "I(L) 1 ~ Iwl(Lo)· (8.12)
Similarly,
(8.13)
Theorem 8.1.6. Let LeU be a simple arc of class J with end-points A and B.
Then there is a sequence of geodesic polygonal lines (L,,), n = 1, 2, ... , joining A
and Band coverging to L on the right (left) (that is, such that the carriers In of
these polygonal lines converge to a curve K on the right, respectively left). Let (L,,),
n = 1, 2, ... , be such a sequence, and !x" and p" the angles of the domain D" at the
points A and B. If the polygonal lines Ln converge to the curve K on the right,
132 Yu.O. Reshetnyak

then K,(L n) - (Xn - Pn -+ K,(L). If Ln -+ K to the left, then ",(L n) + (Xn + Pn -+ K,(L)
as n -+ 00.
Theorem 8.1.6 is proved by applying the corollary of Theorem 8.1.2 and the
inequalities (8.12), (8.13).
Theorem 8.1.6 establishes the geometrical meaning ofthe quantities K,(L) and
K,(L). In particular, it implies the equivalence of the definition of turn given in
the monograph Aleksandrov and Zalgaller (1962) and turn in the sense of the
definition of the present work.
Curvature as a set function in a manifold of bounded curvature. Let M be a
two-dimensional manifold of bounded curvature, U a domain in M homeo-
morphic to a disc, and qJ: U -+ C an isothermal coordinate system in this do-
main. We shall say that a simple arc L in M belongs to the class A if any partial
arc Ll of it belongs to the class J. This is equivalent to the fact that for any
isothermal coordinate system defined in a neighbourhood of L the image of Lis
a simple arc on the plane that has left and right tangents at each of its points.
Similarly we shall say that a simple closed curve r belongs to the class A if any
simple arc contained in r belongs to the class J.
As above, let G = qJ(U) and let A.(z) == A.(z; co, h) be a function defined in U
such that the domain G with metric PA is isometric to the domain U with
induced metric Pu and qJ: U -+ G is an isometric map of(U, Pu) onto (G, PAl.
Let r be a simple closed curve of class A lying in U and A = qJ(r) its image
in a given isothermal coordinate system. We shall assume that U is oriented by
stipulating that the coordinate system qJ is right, and we orient the curve r
positively. Let D be the domain bounded by r. We take points X and Y on r
arbitrarily. They split r into two simple arcs r l and r 2, which we shall assume
to be oriented so that when we go round r in the positive direction we also go
round each of the arcs r l and r 2 in the positive direction. Let (X and Pbe the
angles of D at X and Y. We put
K(r) = K,(rl ) + K,(r2) + (n - (X) + (n - Pl.

It follows easily from Theorem 8.1.6 that K(r) does not depend on the choice
of the points X and Y. Let Zl = qJ(X), Z2 = qJ(Y), Al = qJ(rd, A2 = qJ(r2). For
the angular function qJ,(A, z) we have qJ(A, z) = qJ,(A l , z) + qJ,(A 2, z) for any
point z other than z 1 and z2' Let 1i and pbe the angles of the domain Q bounded
by the curve A at the points Zl and Z2' We have K(A l ) + K(A 2) + (n -iX) +
(n - P) = 2n. At the point Zl we have qJ(A, z) - qJ,(L l , z) - qJ,(L 2, z) = 2n - 1i
and similarly for the point Z2' Using this relation, after obvious transformations
we obtain

K(r) = 2n - ;n L qJ(r, z)co(dz). (8.14)

We now observe that qJ(r, z) = 2n if z belongs to Q and qJ(F, z) = 0 otherwise.

Hence we conclude that the integral on the right-hand side of (8.14) is equal to
 I. Two-Dimensional Manifolds of Bounded Curvature 133

2nw[Q], and so
,,(F) + w(Q) = 2n.
For an arbitrary Borel set E c U we put Q(E) = w[qJ(E)]. Obviously Q is
a measure defined in a domain of the two-dimensional manifold of bounded
curvature M.
For any positively oriented simple closed curve of class A contained in U we
have
,,(F) + t2(D) = 2n, (8.15)
where D is the domain bounded by the curve r. This property uniquely deter-
mines the set function Q, and as a result we arrive at the following theorem.
Theorem 8.1.7. On any two-dimensional manifold of bounded curvature M we
can define a measure Q such that for any simple closed curve r of class A lying in
a domain U of M homeomorphic to an open disc and bounding a domain D the
equality (8.15) is satisfied. The measure Q satisfying this condition is unique. For
any isothermal coordinate system qJ: U .... C specified in a domain U we have
Q(E) = w[qJ(E)], where E c U is a Borel set and W is the measure by means of
which the function l(z) = l(z; w, h) is defined in the representation ds 2 = l(z) Idzl 2
of the metric tensor of the manifold in the given coordinate system.
The measure Q defined in Theorem 8.1.7 is called the curvature of a set of a
given manifold of bounded curvature. The equality (8.15), by analogy with the
case of a Riemannian manifold or a manifold with polygonal metric, is called
the Gauss-Bonnet formula.
Curves with bounded variation of turn. Here we distinguish an important class
of curves in a two-dimensional manifold of bounded curvature.
Let L be an arbitrary simple arc of class A in a two-dimensional manifold of
bounded curvature M. We orient the arc L and specify a definite orientation of
M. Let X(t), a ~ t ~ b, be an arbitrary right parametrization of L, and L t the arc
of L corresponding in this parametrization to the interval [a, t], a < t. Then
there are defined the quantities
",(t)= ",(L t ), ",(t) = ",(Lt ), Q(t) = Q(L?).
We also put ",(a) = ",(a) = O. For all t E [a, b] we have ",(t) + ",(t) = Q(t).
Representing the measure Q in the form Q = Q+ - Q-, where Q+ and Q- are
its positive and negative parts, we obtain Q(t) = Q+(L?) - Q-(L?). The func-
tions Q+(L?) and Q-(L?) are obviously non-decreasing, and so Q(t) is a function
of bounded variation. We shall say that L is a curve of bounded variation of
turn if one of the functions ",(t) and ",(t) is a function of bounded variation. In
view of the equality ",(t) + ",(t) = Q(t) it is then obvious that the other is also
a function of bounded variation.
Theorem 8.1.8. If a curve L in a two-dimensional manifold of bounded curva-
ture does not contain peak points and is a curve of bounded variation of turn, then
134 Yu.G. Reshetnyak

it is rectifiable. We assume that qJ: U -+ C is an isothermal coordinate system

defined in a neighbourhood of L and let K = cp(L). We assume that L does not
contain peak points. Then for L to be a curve with bounded variation of turn it is
necessary and sufficient that K should be a curve with bounded rotation.
It was shown in Zalgeller (1965) that if a simple arc L with bounded variation
of turn has peak points, then it may turn out to be non-rectifiable. In this case
the image of a curve in an isothermal coordinate system may not be a curve of
bounded rotation.
Area in a two-dimensional manifold of bounded curvature. Let M be a two-
dimensional manifold of bounded curvature, and E a Borel set in M. We assume
that E is contained in the domain of definition of an isothermal coordinate
system cp: U -+ C in M. Let G = cp(U) and let A.(z) Idzl 2 be the line element in this
cordinate system. We put

CT(E) = ff
tp(E)
A.(z) dx dy.

The formula (7.4) for transforming the line element of a manifold on going over
from one coordinate system to another enables us to conclude that CT(E) does
not depend on the choice of coordinate system in M.
We assume that E is an arbitrary Borel set in M. Then E can be represented
as the union of a no more than countable set of pairwise disjoint Borel sets, each
of which is contained in the domain of definition of an isothermal coordinate
system of M. Let
E= UEk
k

be such a representation of the set E. It is easy to show that the sum L CT(Ek )
does not depend on the choice of this representation of E. We put It

L CT(Ek ) = CT(E).
k

Thus, on the totality of all Borel sets of M there is defined a non-negative set
function CT. It is easy to verify that CT is a measure in M. We shall call CT the area
in M. We shall call CT(E), where E is a Borel set in M, the area of the set E.
This definition of area is in some respects formal and relies on the analogy
with the case of a Riemannian manifold. We now state some results that enable
us to establish the geometrical meaning of the set function CT.
Theorem 8.1.9. (Aleksandrov and Zalgaller (1962». Let G be an open set in a
two-dimensional manifold of bounded curvature, and (PII)' n = 1,2, ... , a sequence
of Riemannian metrics defined in G that converges to the metric PG induced in G.
We assume that the absolute curvatures of the Riemannian manifolds (G, PII) are
bounded in aggregate and let CTiE) be the area of the Borel set E in the Rieman-
nian metric p". Then as n -+ 00 the set functions CT" converge weakly to the set
function CT.
 I. Two-Dimensional Manifolds of Bounded Curvature 13S

8.2. A Theorem on the Contraction of a Cone. Angle between Curves. Com-

parison Theorems. Using the results of the previous part, we can first of all make
the theorem on the contraction of a cone (Lemma 6.5.1) more precise. Namely,
the following proposition is true.
Theorem 8.2.1. Let M be a two-dimensional manifold of bounded curvature,
and r a simple closed curve in M that bounds an open domain D homeomorphic to
a disc. We assume that r is rectifiable and that Q+(D) < 2n. Then there is a
convex cone Q such that co(Q) :::;; Q+(D) and Q has a map qJ onto the set D u r such
that the following conditions are satisfied:
1) the boundary oQ of the cone is mapped one-to-one onto the curve r. The
image of any simple arc L c oQ is a simple arc of r whose length is equal to the
length of L;
2) for any points X, Y E Q we have
Pa(X, Y) ~ PD(qJ(X), qJ(Y».
For the proof of the theorem it is sufficient to observe that the sequence of
polyhedral metrics (Pn) considered in 6.5 can be chosen so that the set functions
co:n converge weakly to the function COo = Q+. The existence of such an approxi-
mating sequence follows easily from the results of 8.1 and § 7.
Let us mention some applications of Theorem 8.2.1.
Theorem 8.2.2 (the first comparison theorem, Aleksandrov and Zalgaller
(1962». Let T be a triangle homeomorphic to a disc in a two-dimensional manifold
of bounded curvature M, iX the upper angle at one vertex of T, and (Xo the angle at
the corresponding vertex of the development of the triangle T onto the plane. Then
iX - (Xo :::;; Q+(TO), (8.16)
where TO denotes the interior of the domain bounded by the sides of the triangle T.
A few words about the proof of Theorem 8.2.2. Let A, B, C be the vertices of
the triangle TO. The inequality (8.16) is obvious if Q+(TO) ~ 2n. We shall as-
sume that Q+(TO) < 2n. Let us construct a convex cone Q and a map qJ of Q
onto T that corresponds in the sense of Theorem 8.2.1 to the domain TO. We
denote the points on the boundary of the cone corresponding to A, B, C by A',
B', C' respectively. Let K = [Be], L = [CA], M = [AB] be the sides of the
triangle T, and K' = [B'C'], J.} = [C'A'], M' = [A'B'] the arcs of the boundary
of the cone Q that correspond to them. Each of the arcs K', J}, M' is a shortest
curve. For example, let R' be an arbitrary curve on the cone Q that joins the
points B' and C', and let R = qJ(R'). Then obviously s(R):::;; s(R'). The curve R
joins the points Band C, and since K is a shortest curve, s(K) :::;; s(R) :::;; s(R').
We have s(K) = s(K'), and so s(K') :::;; s(R'), that is, we see that any curve in Q
that joins the points B' and C' has length at least s(K'), and so K' is a shortest
curve.
Let (x' be the angle of the boundary of the cone Q at the point A'. We show
that
iX :::;; (X'. (8.17)
136 Yu.G. Reshetnyak

Let X ELand f E M be two arbitrary points on the sides of the triangle T

starting from its vertex A, where X -::f: A, f -::f: A. We put x 1= p(A, X), y =
p(A, f), Z = p(X, f). We construct a planar triangle with sides equal to x, y, z,
and let y(x, y) be the angle of this triangle lying opposite the side equal to z. We
put z' = PQ(X', f') and let R' be the shortest curve in Q that joins X' and f',
R = qJ(R'). We have s(R) ~ s(R') = z'. On the other hand, we obviously have
z ~ s(R), so z ~ z'. Let us construct a planar triangle with sides x, y, z'; let
y'(x, y) be its angle opposite the side equal to z'. Since z' ~ z we have y(x, y) ~
y'(x, y). Obviously y'(x, y) -+ IX' as x -+ 0, y -+ o. We have iX = lim y(x, y), and
so the inequality (8.17) is proved. x"'O,y...o
Since Q(Q) ~ Q+(TO), by virtue of (8.17) for the proof of Theorem 8.2.2 it is
sufficient to establish that

IX' - 1X0 ~ Q(Q). (8.18)

The simple arcs K', L', M' are shortest curves on the cone Q. The inequality
(8.18) is therefore a special case of (8.16) when M is the convex cone Q. The proof
of (8.18) is a problem of elementary geometry. Let us give a solution of it. If
Q(Q) = 0, then A'B'C' is an ordinary planar triangle: IX' = 1X0 and IX' - 1X0 = 0 ~
Q(Q), so in this case (8.18) is true. We shall assume that Q(Q) > O. Let 0 be the
vertex of the cone Q. We join A' to 0 by a shortest curve and then draw from 0
the generator OZ' (where Z' E aQ) of the cone Q such that the angles to the left
and right of 0 between the shortest curves OA' and OZ' are equal. Let (J be their
common value. We have 2(J = 2n - Q(Q). The points A' and Z' are joined in Q
by two shortest curves lying on opposite sides of Q (Fig. 39). On the cone Q we
obtain two triangles A'OZ'. Since the vertex of the cone Q lies on the boundary
of each of these triangles, the latter are isometric to planar triangles with the
e
same lengths of sides. Let and 11 be the angles of the triangles A'OZ' at the
points A' and Z' respectively. We have e + 11 = n - (J, so 2e + 211 = 2n - 2(J =
Q(Q). In particular, 2e ~ Q(Q). The point Z' lies on the shortest curve K' =
[B' C']. For if Z' were an interior point of the shortest curve M' = [A' B'], then
we could shorten M', since the angle between the arc Z'B' and the shortest curve
A'Z' on the other side of 0 with respect to the shortest curve M' is equal to
n - 211 < n. Similarly we can establish that Z' cannot be an interior point of the
o

B'

Fig. 39
 I. Two-Dimensional Manifolds of Bounded Curvature 137

c'

--
2'

8'

Fig. 40

shortest curve [A'e']. Cutting out from the cone Q the lune formed by the
shortest curves [A'Z'] and pasting these shortest curves together, we obtain
a planar quadrangle A'B'Z'C' for which IA'B'I = p(A, B), IA'C'I = p(A, C),
IB'Z'I + IZ'C'I = p(B, C). The angle of this triangle at A' is equal to (I.' - 2e.
Straightening the planar polygonal line B'Z'C' (Fig. 40), we obtain a planar
triangle that is the development of the original triangle T, so the angle at its
vertex corresponding to A' is equal to (1.0' On straightening the polygonal line
B'Z'C' the angle at the vertex A' increases, and so we deduce that (I.' - 2e :::;;; (1.0'
so (I.' - (1.0 :::;;; 2e :::;;; .Q(Q), and the inequality (8.18) and with it Theorem 8.2.2 are
proved.

Theorem 8.2.3 (Aleksandrov and Zalgaller (1962». There is a definite angle

between any two shortest curves in a two-dimensional manifold of bounded curva-
ture starting from one point of the manifold.
In other words, we assert the following. Let K and L be shortest curves in a
manifold of bounded curvature with common starting point 0, let X E K and
Y E L be points of these shortest curves such that p(O, X) = x > 0, p(O, Y) =
y > 0, and let y(x, y) be the angle of the development onto the plane of the
triangle OX Y corresponding to the vertex O. Then there exists the limit

lim y(x, y).

x ....O,y .... O

Here we shall not dwell on how Theorem 8.2.3 is proved. We only mention
that in the proof of the first basic theorem on approximation we use a special
case of Theorem 8.2.3. We did not mention this above, relating it to details of
technical character. In the special case that is necessary for the proof of the
first theorem on approximation, the required result is established by applying
Lemma 6.3.1 (an estimate of the difference IX - (1.0 in terms of v+(OXY». In the
general case the truth of Theorem 8.2.3 was established in Aleksandrov and
Zalgaller (1962) by means ofthe first theorem on approximation.
A proof of Theorm 8.2.3 different from that given in Aleksandrov and
Zalgaller (1962) can be obtained by using the estimate of Theorem 8.2.2.
138 Yu.G. Reshetnyak

Remark. In the conditions of Theorem 8.2.2 we denote by IX the angle of the

domain TO at the point A, and by (X the angle of the triangle T at this point.
Obviously (X ::s:;; IX. The estimate of Theorem 8.2.2 can be strengthened, namely we
have
IX - (Xo ::s:;; w+(TO). (8.19)
The proof of (8.19) can also be established by applying Theorem 8.2.1.
Theorem 8.2.4 (the second comparison theorem, Aleksandrov and Zalgaller
(1962)). Let T = ABC be a triangle homeomorphic to a disc in a two-dimensional
manifold of bounded curvature, TO its interior, (X the angle of the triangle T at the
point A, and (Xo the corresponding ·angle of the development of the triangle. Then
3
(X - (Xo ~ -w-(TO) - L ,,;,
i=l

where ,,;, i = 1,2,3, are the negative parts of the turns of the sides of the triangle
T from the side converted to the domain TO.
Remark. In Aleksandrov and Zalgaller (1962) Theorems 8.2.2 and 8.2.4 were
stated in more general form. This more general result can be obtained as a
consequence of Theorems 8.2.2 and 8.2.4, given here.
The proof of Theorem 8.2.4 in its basic part relies on a proposition analogous
to Lemma 6.3.1 which gives a lower estimate of the difference (X - (Xo (in con-
trast to Lemma 6.3.1, which enables us to estimate the difference from above).
Namely, the following lemma is true.
Lemma 8.2.1 (Aleksandrov and Zalgaller (1962». Suppose that for each pair
of points X, Yon the sides [AB] and [AC] of the triangle T = ABC in a two~
dimensional manifold of bounded curvature we choose a definite shortest curve
[XY], where in all cases the left convergence (that is, with the side of the point A)
X" --+ X or y" --+ Y implies the convergence [X" Y] --+ [XY], [Xy,,] --+ [XY].
Then
(8.20)
where (X is the angle of the triangle T at the point A, (Xo is the corresponding angle
of the development, and VA is the greatest lower bound of the excesses of the
triangles AXY.
The condition concerning the convergence of shortest curves in the statement
of the lemma can be established if we understand by [XY] the extreme (from the
side of the point A) shortest curve joining X and Y.
We mention that Aleksandrov and Zalgaller (1962) stated a more general
proposition relating to arbitrary metric spaces. In this case instead of (X we need
to take a lower limit. Excesses of triangles are understood in the corresponding
way. The proof of Lemma 8.2.1 (also, like the more general proposition, given in
Aleksandrov and Zalgaller (1962» is carried out by means of arguments similar
to those by means of which Lemma 6.3.1 is proved. The proof of Theorem 8.2.4
reduces to an estimate of VA'
 I. Two-Dimensional Manifolds of Bounded Curvature 139

Theorems 8.2.3 and 8.2.4 enable us to show, in particular, that a curve K

belongs to the class J if and only if it has definite directions at its ends.
As a preliminary we indicate some auxiliary propositions that also have a
definite independent interest.
Let 0 be a point in a two-dimensional manifold of bounded curvature M
such that Q( 0) < 2n. In a neighbourhood U of the point 0 we specify a definite
orientation of the manifold M. Let Lo = [XO] be a simple arc with bounded
variation of the turn lying in U, and L = [OY] any other simple arc with
bounded variation of the turn contained in U. We assume that Lo and L have
no points in common other than O. Then they form a curve K. Let us find the
angle on the right of the curve K at the point O. Whatever the number qJ E
[0, O(X)], where O(X) = 2n - Q(X), the arc L = [OY] can be chosen so that
the angle is equal to the given qJ. It is easy to verify the truth of this assertion
by using an isothermal coordinate system. We say that qJ E [0, O(X)] is non-
singular if there is a shortest curve L = [OY] such that the angle on the right
of the curve K = Lo u L at the point X is equal to qJ.
Lemma 8.2.2. In the notation introduced above, the set of non-singular values
is everywhere dense in the interval [0, O(X)].
Lemma 8.2.3. Let Lo = [AO], L = [OB] be arbitrary shortest curves starting
from a point 0 of a two-dimensional manifold of bounded curvature M. We assume
that Q(O) < 2n and that Lo and L have no points in common other than O. Then
for any e > 0 there is a neighbourhood U of the point 0 such that for any triangle
T = OXY lying on one side of the simple arc K = Lo u L the angle 0( of the
development of this triangle corresponding to the point 0 does not exceed qJ + e,
where qJ is the angle of the curve K at the point 0 from the side where the triangle
T lies.
The proof of Lemma 8.2.3 is carried out by a simple application of Theorem
8.2.4.
Let L be a curve of class J, and 0 its starting point. In a neighbourhood of
o we introduce an isothermal coordinate system qJ: U -+ C and let K = qJ(L).
We assume that qJ(O) = O. The curve K has a half-tangent at O. Let K' be a
segment of this half-tangent and L' the simple arc that goes over to K under the
map qJ. By Lemma 8.2.2 there are shortest curves Ll and L2 starting from 0
that lie on different sides of L and make an angle less than e/3 with 1:. The
curves Kl = qJ(L 1 ) and K2 = qJ(L 2 ) are curves with bounded rotation and a
small arc of the curve K is contained in the angle included between them.
Lemma 8.2.3 enables us to conclude that for any triangle OXY, where X E L,
Y E L, the angle y(X, Y) < ~e + 1
if X and Yare sufficiently close to O. Hence
it follows that lim y(X, Y) = 0, that is, L has a definite direction at o.
x-+o, Y-+O
The verification that if the curve L has a definite direction at one of its
end-points, then its image in an isothermal coordinate system has a half-tangent
at the corresponding point, is easily carried out.
140 Yu.G. Reshetnyak

Theorem 8.2.S. If the curvature of a two-dimensional manifold of bounded

curvature M, as a set function, is non-positive, then the angles of any triangle
in M homeomorphic to a disc do not exceed the corresponding angles of the
development.

Remark. The condition in the statement of the theorem that the triangle is
homeomorphic to a disc cannot be omitted. In fact, let M be the surface of a
right circular cylinder and r the section of M by a plane perpendicular to the
generators. On r we mark points X, Y, Z that divide r into three equal arcs.
The curve r is a geodesic, and the arcs into which it is split by the points X, Y,
Z are shortest curves. We thus obtain a triangle XYZ. The angles at its vertices
are all equal to n. At the same time the angles ofthe development ofthe triangle
XYZ are all equal to nl3 < n.
The next result follows from Theorem 8.2.4.
Theorem 8.2.6. Let M be a complete two-dimensional manifold of bounded
curvature, where the curvature of M, as a set function, is non-negative. Then for
any triangle in M the angles at its vertices are not less than the corresponding
angles of the development.

The given assertion follows from the fact that in a manifold of non-negative
curvature the turn of a shortest curve is equal to zero, and any two shortest
curves in such a manifold do not have points in common other than the end-
points. Using the results in Aleksandrov (1948a) about the topological structure
of complete two-dimensional manifolds of bounded curvature, it is not difficult
to conclude that any triangle in such a manifold is homeomorphic to a disc (so
long as its sides do not lie on one shortest curve). Theorem 8.2.6 is thus obtained
by a direct application of Theorem 8.2.4.
Let M be a two-dimensional manifold of bounded curvature, K a real
number, 0 the curvature and u the area of the manifold. We say that M is
a manifold of curvature not greater than K if for any Borel set E such that O(E)
is defined we have O(E) ~ Ku(E). If O(E) ~ Ku(E) for any E c M for which
O(E) is defined, then M is called a two-dimensional manifold of curvature not
less than K. Let IK denote, in the case K > 0, a sphere of radius 1/JK, in
the case K = 0 an ordinary Euclidean plane, and finally in the case K < 0 a
Lobachevskij plane with Gaussian curvature K. Let T = XYZ be a triangle in
an arbitrary metric space (M, p), and x = p(Y, Z), y = p(Z, X), z = p(X, Y) the
lengths of its sides. On the surface I K we construct a triangle T' with side
lengths x, y, z (in the case K > 0 this is possible only if x + y + z ~ 2nIJK).
The triangle T' will be called the K-development of the triangle T. Theorems
8.2.5 and 8.2.6 have analogues for the cases when M is a two-dimensional mani-
fold of curvature not greater than K (respectively, not less than K and instead
of an ordinary development we consider a K-development (see Aleksandrov
(1951), (1954), (1957b).
We note that for the case when M is a K-polyhedron (see §6), where K > 0,
an analogue of Theorem 8.2.1 was obtained in Belinskij (1975).
 I. Two-Dimensional Manifolds of Bounded Curvature 141

8.3. A Theorem on Pasting Together Two-Dimensional Manifolds of Bounded

Curvature. In 3.2 we described a general procedure for pasting together two-
dimensional manifolds with intrinsic metric. Here we mention a theorem that
establishes conditions under which the pasting together of two-dimensional
manifolds of bounded curvature gives as a result a manifold of bounded
curvature.
Let M be a two-dimensional manifold with boundary, endowed with an
intrinsic metric PM' We shall say that M is a two-dimensional manifold of
bounded curvature if M is isometric to a subset of some two-dimensional mani-
fold M' without boundary that is endowed with an intrinsic metric and is a
manifold of bounded curvature in the sense of the definitions in 6.1.
Similarly a two-dimensional manifold M with boundary will be called a
Riemannian manifold if it is isometric to a domain in a two-dimensional Rie-
mannian manifold in the sense of the definition given in § 4.
In order that the basic concepts of Riemannian geometry, the Gaussian cur-
vature, the geodesic curvature of a curve, and so on, can be introduced for a
Riemannian space it is sufficient that the metric tensor of the manifold should
belong to the class COO, If in the conditions of 3.2 the pasted manifolds are
Riemannian manifolds of class Coo, then the manifold obtained as a result of the
pasting will be a Riemannian manifold of the same smoothness class only if
restrictions on the pasted arcs are satisfied; these are very severe even in the case
when these arcs are "good" from the viewpoint of the geometry of the manifold.
In contrast to this, the conditions under which as a result of pasting together
manifolds of bounded curvature we obtain a manifold of the same type are
substantially weaker.
Theorem 8.3.1 (Aleksandrov and Zalgaller (1962». Let (Dk), k = 1,2, ... , be a
finite or denumerable set of two-dimensional manifolds with boundary, each of
which is endowed with an intrinsic metric. We assume that each of the manifolds
Dk is a two-dimensional manifold of bounded curvature and that for each k any
simple arc contained in ODk is a curve with bounded variation of turn. We assume
that for the manifolds Dk there is specified a law of pasting that satisfies conditions
1)-5) in 3.2. Then the two-dimensional manifold D obtained by pasting together the
Dk is a manifold of bounded curvature.
The next theorem gives a means of calculating the basic geometrical charac-
teristics of a manifold D from the corresponding characteristics relating to the
manifolds Dk •
Theorem 8.3.2. Suppose that the manifolds (Dk), k = 1, 2, ... , satisfy all the
conditions of the preceding theorem, and that D is the manifold obtained as a result
of pasting them together. Let Ak be the set of points of D corresponding to points
of Dk. Then the following assertions are true:
A) for any Borel set E eDits area is equal to the sum of the areas of the sets
Ek c Dk from which E arises under the pasting;
B) if a point XED belongs to A~ for some k and x' E Dk is the point from which
x is obtained under the pasting, then the total angle of D at x is equal to the total
142 Yu.G. Reshetnyak

angle of Dk at x'. If the point XED is obtained by pasting together the points
Xl E ODk" X2 E ODk2 , ... , xm E oDkm , then the total angle of D at x is equal to the
sum of the total angles of the manifolds Dk" Dk2 , ... , Dk mat the points Xl' x 2 , ... ,
Xm respectively;
C) if a simple arc LcD is obtained by pasting together the simple arcs Ll c
ODk" L2 C ODk2 , then L is a curve with bounded variation of turn. The turn of L
on the side of a domain L1k c D is equal to the turn of Li on the side of the
manifold Di . If L c L1~ for s~me k, then the left and right turns of the arc L in the
manifold D are equal to the left and right turns respectively of the arc z.: c Dk from
which L is obtained by the pasting;
D) let E c D be a Borel set. We assume that E c L1~ for some k. Then the
curvature of the set E in the manifold D is equal to the curvature in Dk of the set
E' that goes over to E under the pasting.
Theorems 8.3.1 and 8.3.2 indicate a method for constructing a large number
of specific examples of two-dimensional manifolds of bounded curvature.
Example. A collection of flat domains whose pasting together gives the sur-
face of the right circular cylinder shown in Fig. 6 satisfies the conditions of
Theorem 8.3.1. Hence it follows that the surface of the cylinder is a manifold of
bounded curvature. Assertions C and D of Theorem 8.3.2 enable us to conclude
that the curvature, as a set function of this manifold, is positive and concentrated
on the two circles that bound the base of the cylinder. If the set E is a simple arc
lying on one of these circles, then w(E) is equal to the rotation of this arc in the
plane of the base (we assume that the plane of the base and the bounding circle
are oriented so that the rotation of any arc of the circle is positive).
Another example. Let R be a convex open set on the plane 1E2 such that the
set M = 1E2\R is connected (this excludes the case when R is the strip included
between two parallel lines). Pasting together the manifold M and a second copy
of it, we obtain a two-dimensional manifold Nt (the twice covered manifold M).
By Theorem 8.3.1, Nt is a manifold of bounded curvature. Theorem 8.3.2 enables
us to conclude that its curvature, as a set function, is non-positive and concen-
trated on the curve that arises from the bounding curve of the domain R.

8.4. Theorems on Passage to the Limit for Two-Dimensional Manifolds of

Bounded Curvature. Let (M, p) and (Mn' Pn), n = 1, 2, ... , be metric spaces with
intrinsic metric. In accordance with the definition given above, we shall say that
the spaces (Mn' Pn) converge to the space (M, p) if for each n there is specified a
homeomorphism ({In of M onto Mn such that Pn (({In (X), ((In(Y)) -+ p(X, Y) uni-
formly as n -+ 00. Putting Pn(X, Y) = Pn(({Jn(X), ({In(Y)), we obtain in this case a
metric Pn on the set M. Since ({In is a homeomorphism, the metric Pn is compati-
ble with the topology of M. As n -+ 00 the metrics Pn converge uniformly to the
metric p. The general situation described above thus reduces to the following.
There is a topological space M on which there are specified intrinsic metrics P
and Pn' n = 1,2, ... , where each of these metrics is compatible with the topology
of M and Pn -+ P uniformly on the set M x Mas n -+ 00.
 I. Two-Dimensional Manifolds of Bounded Curvature 143

Next we assume to be fixed a two-dimensional manifold M and intrinsic

metrics p, p", n = 1, 2, ... , defined in M. The length of a curve L c M with
respect to the metric p" is denoted by s,,(L). The symbol s(L) denotes the length
of L with respect to the metric p. If each of the manifolds (M, p), (M, p,,) is a
two-dimensional manifold of bounded curvature, then for any Borel set E c M
whose closure is compact there are defined quantities Q(E) and Q,,(E), its curva-
tures with respect to the metrics p and p" respectively. If (M, p) is a manifold of
bounded curvature, we shall say for brevity that p is a metric of bounded curva-
ture in M.
Theorem 8.4.1. If the metrics p" on the manifold M converge to the metric p as
n -+ 00, then for any sequence {x,,(t), a ~ t ~ b} of parametrized curves that con-
verges as n -+ 00 to a parametrized curve x(t), a ~ t ~ b, we have
s(x; a, b) ~ lim s,,(x,,; a, b). (8.21)

A set E c M will be said to be bounded if its closure is compact. The totality

of all Borel sets E c M is denoted by ~(M). Henceforth ~o(M) denotes the
totality of all bounded Borel subsets of M. A measure in M will be any totally
additive real set function defined either on ~(M) or on ~o(M).
A function f: M -+ ~ is said to be compactly supported if there is a compact
set A c M such that f(x) = 0 for all x ~ A. The totality of all continuous com-
pactly supported functions f: M -+ ~ is denoted by Co(M). We shall denote by
C(M) the set of all bounded continuous functions f: M -+ ~.
If the measure a has the totality of sets ~(M) as domain of definition, then for
any function f e C(M) the integral

fM f(x)a(dx) (8.22)

is defined and finite. In the case when the measure a has the totality of sets
~o(M) as domain of definition, that is, a(E) makes sense only for a bounded
Borel set E, the integral (8.22) is defined for any compactly supported continu-
ous function.
Let (a,,: ~(M) -+ ~), n = 1, 2, ... , be a sequence of measures in M, each of
which is defined on all Borel subsets of M. Then we say that the sequence (a,,),
n = 1,2, ... , converges weakly to the measure a: ~(M) -+ ~ as n -+ 00 if for any
function f e C(M)

f M
f(x)a(dx) = lim
" .... 00
f
M
f(x)a,,(dx). (8.23)

We shall say that the sequence of measures a,,: ~o(M) -+ ~ converges locally
weakly to the measure a: ~o(M) -+ ~ as n -+ 00 if (8.23) holds for any function
fe Co(M).
Above we defined the concept of weak convergence for a sequence of non-
negative measures. In this case the definition given here is equivalent to the
previous one.
144 Yu.G. Reshetnyak

Theorem 8.4.2. We assume that the metrics PII converge to the metric P as
n - 00 and that (M, PII) are two-dimensional manifolds of bounded curvature,
where the quantities IDnl (M) are bounded in aggregate,
IDnl(M):S:;; C = const < 00 (8.24)
for all n. Then as n - 00 the set functions Dn converge locally weakly to the
function D.
For the proof see Aleksandrov and Zalgaller (1962).
Let (Pn), n = 1, 2, ... , be a sequence of metrics of bounded curvature in a
two-dimensional manifold M. We shall say that the sequence (Pn) converges
tamely to the metric P if D: - D+, D;; - D- as n - 00 in the sense of weak
convergence.
Theorem 8.4.3. We assume that the space (M, p) is isometric to a domain in a
two-dimensional manifold of bounded curvature that has compact closure and is
endowed with the induced metric. Then in M there is a sequence of Riemannian
metrics Pn that converges tamely to P as n - 00. Under the same assumptions it is
also possible to construct a sequence of polyhedral metrics (Pn) in the manifold M
that converges tamely to P as n - 00.
A proof of the theorem can be obtained, for example, by means of the theo-
rems on analytic representation of a two-dimensional manifold of bounded
curvature stated in § 7.
We note that the construction by means of which Theorem 6.2.1 on the
approximability of the metric of a manifold of bounded curvature by polyhedral
metrics was proved leads to a tamely convergent sequence of polyhedral metrics
(see Aleksandrov and Zalgaller (1962».
Theorem 8.4.4. We assume that PII are metrics of bounded curvature that con-
verge as n - 00 to a metric p, and that the quantities IDnl (M) are bounded in
aggregate, IDnl(M):S:;; C = const < 00 for all n. Let (L II ), n = 1,2, ... , be a se-
quence of simple arcs in the manifold M that converges to a simple arc L. We
assume that LII for each n is a curve with bounded variation of turn in the manifold
(M, PII). Let I",I (Ln) denote the total variation of left turn of the arc Ln in the
»,
metric Pn. Then if the sequence (I ",I (L II n = 1,2, ... , is bounded and the curve L
does not contain peak points in the sense of the metric p, it follows that L is a curve
with bounded variation of turn. We have
s(L) = lim sll(Ln). (8.25)

Remark. Under the conditions of the theorem the sequence of variations of

right turns of the curves LII in the metrics PII is also bounded.
For a proof of Theorem 8.4.4 see Aleksandrov and Zalgaller (1962).
In Burago and Zalgaller (1965) some results are cited that touch on the
relations between the variation of turn of a limiting curve and the variations of
turn of the curves that converge to it. In the same paper an analogue of Theo-
rem 8.4.4 was established for the case of curves that do not have simple arcs.
 I. Two-Dimensional Manifolds of Bounded Curvature 145

In Aleksandrov and Zalgaller (1962), Ch. VIII, § 2, there is discussed the

question of the convergence of domains, each of which is a polygon with a fixed
number of sides in the manifold (M, Pn). (We recall that by a polygon we under-
stand a domain whose boundary consists of finitely many shortest curves, and
each connected component of the boundary is the homeomorphic image of
either a circle or an open interval.)
The theorems stated here can be extended to the case when convergence of
metrics is understood in a weaker sense. We assume that in the manifold M
there is specified a sequence of domains (Mn), n = 1, 2, ... , such that for each n
the closure of Mn is compact and contained in Mn+1 and

U Mn=M.
00

n=1

Let us assume that for each n there is specified a metric Pn. We shall say that the
sequence of metrics (Pn), n = 1, 2, ... , converges locally to the metric P if for any
n the sequence of metrics (Pm), m = n, n + 1, ... , converges to P on the set Mn,
that is, Pm(X, Y) -+ p(X, Y) as m -+ 00 uniformly when X, Y E Mn, whatever the
value ofn.
The results of this section admit an extension to the case of a sequence of
metrics (Pn) that converges locally to the metric P as n -+ 00. The best possible
results are obtained as a simple consequence of the theorems given above. All
the necessary details can be found in Aleksandrov and Zalgaller (1962).
In the paper Burago (1965b) the author considered the question of the struc-
ture of metric spaces that are not two-dimensional manifolds and at the same
time admit an approximation by two-dimensional Riemannian manifolds with
absolute integral curvatures that are bounded in aggregate. It is easy to con-
struct simple examples of such spaces. Let P be a plane in the three-dimensional
space 1E 3, let 0 E P, and let [OA] be an interval perpendicular to P. We put
M = P u [OA] and introduce in M the intrinsic metric induced from 1E3. Obvi-
ously the set M is not a manifold. Let r,. be the circle in the plane P with centre
o and radius lin. We denote by Mn the surface in 1E3 formed by the part of P
outside r,. and the lateral surface of the right circular cone with vertex A for
which r,. is the base circle (see Fig. 41). It is natural to regard M as the limit of
a sequence of spaces Mn. Each of them is a manifold of bounded curvature and
the sequence 1.QI(Mn ) is bounded. Another example of this kind occurs when M

Fig. 41
146 Yu.G. Reshetnyak

Fig. 42

is formed by two spheres Sl and S2 in 1E3 and an interval [AB] joining them (see
Fig. 42). Replacing the interval by the cylindrical hub of width lin on whose axis
[AB] lies, we obtain a sequence of two-dimensional manifolds of bounded cur-
vature Mn , which is naturally regarded as approximating M, for which the
sequence IDI(Mn) is also bounded.
Let us refine the statement of the problem considered in Burago (1965b). The
function r(x, y) of pairs of points of the two-dimensional manifold M is said to
be semimetric if r(x, x) = 0, r(x, y) = r(y, x) and r(x, y) ~ r(x, z) + r(z, y) for any
x, y, z E M. Obviously r(x, y) is non-negative. In contrast to the metric, r(x, y) = 0
does not necessarily imply that x = y.
Such concepts as the length of a curve, a shortest curve, an intrinsic metric, a
triangle, and a triangle with convex boundary are defined for a semimetric word
for word the same as for an ordinary metric. The definition of the concept of the
upper angle between curves is also similar to the case of curves in a metric space.
A degenerate manifold of bounded curvature is a two-dimensional manifold
with intrinsic semimetric for which an axiom of boundedness of curvature is
satisfied that is similar to the corresponding axiom for the case of manifolds
with intrinsic metric with the difference, however, that it is necessary to require
that the sums of the absolute values of the excesses of the triangles are bounded
(this is due to the importance of the matter). The main result of Burago (see
Burago (1965b» relating to degenerate manifolds of bounded curvature is con-
tained in the following theorem.
Theorem 8.4.5. Let M be a two-dimensional manifold with boundary, homeo-
morphic to a closed disc, and r a semimetric in M. Then in order that (M, r) should
be a degenerate two-dimensional manifold of bounded curvature it is necessary and
sufficient that there should be a sequence of metrics (Pn), n = 1, 2, ... , defined in
M that converges uniformly to r as n -+ 00 and is such that (M, Pn) is a two-
dimensional manifold of bounded curvature for each n, and the quantities
IKnl(oM) + IDnl(M) are bounded in aggregate, where IKnl(oM) is the turn of the
boundary of M and IDnl(M) is the absolute curvature of M in the metric Pn'
Let M be a two-dimensional manifold endowed with a semimetric r(x, y). In
M we introduce an equivalence relation by putting x'" y ifr(x, y) = O. Let Mbe
the set of equivalence classes with respect to this relation. For X EM, Y EM we
put p(X, Y) = r(x, y), where x E X, Y E Y. Generally speaking, the metric space
(M, p) is not a manifold. In Burago (1965b) the author gave an exhaustive
description of the topological structure of metric spaces, to which we add by the
 I. Two-Dimensional Manifolds of Bounded Curvature 147

method described here the concept of a degenerate two-dimensional manifold of

bounded curvature. In that paper Burago established some compactness theo-
rems for semimetrics of bounded curvature in a two-dimensional manifold, and
proved the closure of the class of degenerate two-dimensional manifolds of
bounded curvature with respect to some naturally defined passages to the limit.

8.5. Some Inequalities and Estimates. Extremal Problems for Two-

Dimensional Manifolds of Bounded Curvature.
Application of the theorem on contraction of a cone to the solution of extremal
problems. Theorems 5.6.1 and 5.6.2 on the contraction of a convex cone enable
us to distinguish a class of extremal problems with respect to polyhedra, for
each of which the solution is a convex cone.
Consider the set of all two-dimensional polyhedra homeomorphic to a closed
disc. A function F(M) defined on this set is said to be increasing if for any
polyhedra M1 and M2 such that M2 admits a contracting map onto M1 we have
F(M 1 ) ~ F(M2)' If equality holds if and only if M1 and M2 are isometric, then
F is said to be strictly increasing.
Theorem 8.5.1. Let IDl = IDl(ll' 12, ... , 1m , wo) be the totality of all polyhedra
homeomorphic to a closed disc for which the boundary links have lengths equal to
11 ,12 , ••• , 1m respectively, and the positive part of the curvature does not exceed a
number wo, 0 ~ Wo < 2n, and F(M) is an increasing function of the polyhedra. If
F has a maximum in IDl, then among the manifolds MEIDl for which it takes its
greatest value there is at least one convex cone. If F is a strictly increasing
function, then the desired polyhedron that realizes the maximum can only be a
convex cone.
By Theorem 8.5.1 the solution of the problem about finding the maximum of
F comes down to consideration of the case when the polyhedron M is a convex
cone. Theorem 8.5.1 is an obvious consequence of Theorem 5.6.1.
From Theorem 8.2.1 there follows in an obvious way a result analogous to
Theorem 8.5.1, but relating to arbitrary two-dimensional manifolds of bounded
curvature.
An increasing (and even strictly increasing) funetion of a polyhedron is its
area a(M). Another example is the diameter d(M) of the polyhedron M, that is,
the least upper bound of he distances between points of the polyhedron.
The solution of the extremal problem in the class IDl(ll' 12 , ••• , 1m , wo) in the
case when F(M) is the area, F(M) = a(M), is given by Theorem 6.5.3. Theorem
8.2.1 enables us to extend it to the case of arbitrary two-dimensional manifolds
of bounded curvature. Namely, the following proposition is true.
Theorem 8.5.2. Let G be a domain in a two-dimensional manifold of bounded
curvature M, homeomorphic to a closed disc. We assume that the boundary of Gis
rectifiable and let p be its length. If W o = Q+(GO) < 2n, then
p2
a(G) ~ 2(2n _ wo) (8.26)
148 Yu.G. Reshetnyak

Here the equality sign holds if and only if the domain G with the induced metric is
isometric to the lateral surface of a right circular cone.
Theorem 8.5.2 is due to Aleksandrov (Aleksandrov (1945». We note that for
the special case of this theorem when M is Riemannian the inequality (8.26) was
obtained earlier by Fiala (Fiala (1941».
The assertion of Theorem 8.5.2 for the case of equality in (8.26) follows from
the fact that the area is a strictly increasing functional.
The extremal problem corresponding to the case when F(M) is the diameter
of the polyhedron, F(M) == d(M), was solved by Aleksandrov and Strel'tsov
(Aleksandrov and Strel'tsov (1965». The best possible result is given by the
following theorem.

Theorem 8.5.3. Let G be a domain in a two-dimensional manifold of bounded

curvature, homeomorphic to a closed disc, p < ex) the length of the boundary of G,
Q+ = Q+(GO), and d(G) the intrinsic diameter of G, that is, the exact limit of the
distances between points of G in the metric induced in G. Then when 0 ~ Q+ ~ rr.

P
d(G) ~ 2' (8.27)

and when rr. < Q+ < 2rr.

P (8.28)
d(G) ~ Q+.
2sinT

In both cases the inequality is exact. In (8.27) the equality sign holds in many
cases, for example, for lunes between two meridians on a sphere. In (8.28) equal-
ity is attained if and only if G with respect to its intrinsic geometry is an isosceles
triangle, pasted together along the lateral sides, with base p and angle at the
vertex 2rr. - Q+.
Theorem 8.2.1 and its special case - Theorem 5.6.1 - can also be applied as
an auxiliary means of investigation in the study of extremal problems that do
not immediately reduce to it. From the results proved by means of it we give
some theorems on the estimate of the length of a curve in a two-dimensional
manifold of bounded curvature.
Let L be a simple arc With bounded variation of turn in a two-dimensional
manifold of bounded curvature. In certain cases instead of the variation of left
and right turns of the arc L it is advisable to consider the quantity
1
K*(L) = 2(IKr l(L) + IK,I(L) -IQI(LO». (8.29)

Here IK,I and IKr I are the variations ofleft and right turns of the curve L, and L °
is the totality of all interior points of L. The quantity K*(L) is called the charac-
teristic turn or winding of the simple arc L. It does not depend on the choice of
orientation of L. We assume that L is split by an interior point X of it into two
 I. Two-Dimensional Manifolds of Bounded Curvature 149

arcs Ll and L 2, and let e and e2 be the left and right angles of L at the point X.
1
We put K*(X) = !(In - ell + In - e2 1-12n - e 2 1). Then
1 - (

K*(L) = K*(Ll) + K*(L2) + K*(X).

Using the given property of winding, we can define it for curves that are not
simple arcs. Namely, we shall say that L is a curve of bounded variation of turn
if it can be formed from finitely many simple arcs L 1 , L 2, ... , Lm, each of which
is a curve of bounded variation of turn, and the end of Li is the beginning of
Li+l' i = 1,2, ... , m - 1. Let Xi be this common point of Li and Li+l' Then there
is defined the quantity Kt(X;) = !(In - efl + In - e;'1 - 12n - ef - e;'I), where
ef and e;' are the angles of the sectors into which Li and Li+l split a small
neighbourhood of Xi' We put
m m-l
K*(L) = L K*(L;) + L K*(XJ (8.30)
i=l i=1

It is easy to show that the sum on the right does not depend on the choice of
splitting into simple arcs of the curve L.
Theorem 8.5.4 (Aleksandrov and Strel'tsov (1965)). Let G be a domain homeo-
morphic to a closed disc in a two-dimensional manifold, and L a curve lying in G.
If Q+(G) + K*(L) < 2n, then L either does not have multiple points or it consists
of a loop (a curve without multiple points and with coincident ends) and one or two
simple arcs extending the ends of the loop that do not have other points in common
with the loop and lie outside the region bounded by it.
Theorem 8.5.5 (Aleksandrov and Strel'tsov (1965)). Let G be a domain homeo-
morphic to a closed disc in a two-dimensional manifold of bounded curvature. We
assume that the length of the boundary of G is finite. Then for any curve LeG
the length s(L) is bounded above by the quantity So = s(p, Q+(GO), K*(L)) < 00,
where p is the length of the boundary of G.
The next theorem gives an exact estimate of the quantity s(L), which is true,
however, only under certain restrictions.
Theorem 8.5.6. Suppose that the conditions of Theorem 8.5.5 are satisfied. We
put Q+(GO) + K*(L) = 2(1)0' If 2(1)0 :::;; n, then

P
s(L) :::;; 1
+ cos (1)0 '
and if n :::;; 2w o < 2n, then

s(L):::;;~.-.
P
smw o
In both cases the estimate for s(L) is exact.

The proofs of Theorems 8.5.5 and 8.5.6 are contained in Aleksandrov and
Strel'tsov (1965).
150 Yu.G. Reshetnyak

In Aleksandrov and Strel'tsov (1965) Theorem 8.5.6 was established for the
case when G is a domain of more general type than in the statement of the
theorem given here. The description ofthis class turns out, however, to be rather
cumbersome; this is why we do not give Theorem 8.5.6 in the most general form
here. We note that in Aleksandrov and Strel'tsov (1965) cases are also described
in which each of the inequalities of Theorem 8.5.6 becomes an equality.

Inequalities of isoperimetric type. The inequality of Theorem 8.5.2 is a special

case of a whole class of inequalities known as isoperimetric.
Henceforth M denotes a two-dimensional manifold of bounded curvature.
Let G be a domain, that is, a connected open set in M. We denote by G the
metric space that arises from G if we cut M along aGo The set G is embedded in
a natural way in G. We put aG = G\ G. We shall say that G is simple if G is
compact and the set aG is non-empty and consists of finitely many connected
components, each of which is a simple closed curve of finite length. We denote
the sum of the lengths of the connected components of aG by p(G) and call it the
perimeter of G. The boundary of a simple domain is the union of finitely many
rectifiable curves and the perimeter of G can be understood as the length of the
boundary, taking account of multiplicity. We denote by X = X(G) the Euler
characteristic of G. Since Gis connected and the boundary of Gis non-empty,
X ~ 1.
Let k be a real number. We put Qk(E) = Q(E) - ku(E). Defined as a measure,
Q k is called the curvature with respect to a k-plane. We shall also consider the
measures Q: and Q;;, the upper and lower variations of Q.
If M is Riemannian and K(x) denotes the Gaussian curvature at a point
x E M, we have

Qk(E) = L (K(x) - k)u(dx),

Q:(E) = L
(K(x) - ktu(dx), Q;;(E) = L
(K(x) - kfu(dx).

Theorem 8.S.7. Let G be a simple domain in a two-dimensional manifold of

bounded curvature M, p = p(G), Qt = Qt(G), and F = u(G) the area of G. Then
for any k we have
p2 + 2(Qt - 21tx)F + kF2 ~ O. (8.31)
For the proof of Theorem 8.5.7 see Burago (1989) and Ionin (1969a), (1969b),
(1972a), (1972b). The given theorem contains as a special case many other well-
known results. For example, Theorem 8.5.2 follows from Theorem 8.5.7 if we
put k = O.
In the article Aleksandrov and Strel'tsov (1965) and the monograph Burago
and Zalgaller (1980) other extremal problems of the theory of two-dimensional
manifolds of bounded curvature are also considered. Some unsolved problems
relating to the same range of equations are stated in the same article and mono-
 I. Two-Dimensional Manifolds of Bounded Curvature 151

graph. We should also mention the articles Aleksandrov (1945c), Aleksandrov,

Borisov and Rusieshvili (1975), Aleksandrov and Strel'tsov (1953), and Rusie-
shvili (1957), which are devoted to extremal problems.

§ 9. Further Results. Some Additional Remarks

In conclusion we give a survey of further research on the theory of two-

dimensional manifolds of bounded curvature.
A circle, an equidistant, and other special sets in a two-dimensional manifold of
bounded curvature. Let us specify arbitrarily a two-dimensional manifold of
bounded curvature M. Let A c M be an arbitrary non-empty subset of M. For
r > 0 we put S(A, r) = {x E Mlp(X, A) = r). Sets S(A, r) corresponding to differ-
ent r are called equidistants of the set A. In the special case when A consists of a
unique point OA = {O}, S(A, r) is the circle with centre 0 and radius r. The
question of the structure of a circle in an arbitrary two-dimensional manifold of
bounded curvature was considered in the articles Burgo and Stratilatova (1965)
and Zalgaller (1950b). The case of manifolds of positive curvatures was studied
in the second of these articles, and the general case was considered in the first.
In the case when A is a simple arc with bounded variation of turn, not con-
taining peak points of M, the question of the structure of the equidistant S(A, r)
for small r was considered in Borisov (1955) and Borisov (1965).
We give here some results about the structure of a circle and an equidistant
of a set, obtained in Burago and Stratilatova (1965). Apart from the significance
of a result, we are also guided in our choice by the simplicity of the statements.
In the general case a circle on a two-dimensional manifold of bounded curva-
ture can be arranged in a rather complicated way. In particular, it may not be a
simple closed curve. Moreover, a circle may even be a disconnected set.
Let G be a domain homeomorphic to an open disc in a two-dimensional
manifold of bounded curvature, and 0 an arbitrary point of G. We shall con-
sider circles with centre 0 contained in G.
It is natural to pose the following question: under which conditions is the
circle S(O, r) a simple closed curve? The answer is contained in the following
theorem.
Theorem 9.1 (Burago and Stratilatova (1965». If the circle S(O, r) lies in a
domain Go homeomorphic to a disc, and Q+(Go\ {O}) < n, then S(O, r) is a simple
closed curve.
Theorem 9.2 (Burago and Stratilatova (1965». If a connected component r of
a circle S(O, r) eGis not a one-point set, then it is a continuous image of the circle
Sl = S(O, 1) in 1R2. There is a continuous map x: Sl -+ M such that X(Sl) = rand
the following conditions are satisfied:
1) the path x: Sl -+ M does not have essential intersections, that is, by an arbi-
trarily small deformation of the map x we can obtain a topological map of Sl into M;
152 Yu.G. Reshetnyak

2) let R1(t) (R,(t» be the leftmost (rightmost) shortest curve joining the points°
and x(t), where t E Sl' Then, if t runs through Sl in one direction, the radius R,(t)
(R,(t» rotates monotonically around the point 0.
Remark 1. The rotation of the radius R1(t) of which we spoke in condition 2
of the theorem may take place in a discontinuous way.
Remark 2. We shall call the map x: Sl -+ r mentioned in Theorem 9.2 a
canonical parametrization of the component r of the circle S(O, r).
Other properties of a circle are described by the following theorem.
Theorem 9.3 (Burago and Stratilatova (1965». If the circle S(O, r) lies in a
domain G of the manifold M that is homeomorphic to a disc and is a connected set,
then it is a curve with bounded variation of turn. We have
IKI(S(O, r» ~ 0(0) + tr[B(O, r)\{O}],
K-(S(O, r» ~ ,Q+[B(O, r)\{O}].
(Here K denotes the turn from the side of the domain B(O, r) - the disc bounded
by the given circle.)
Remark 1. Since in the conditions of Theorem 9.3 the circle S(O, r) may not be
a simple closed curve, explanations of how its turn should be understood are
necessary. Let x(t), t E Sl' be an arbitrary parametrization of the circle S(O, r).
We take an arc [a, b] of Sl arbitrarily. We shall say that any such arc defines an
arc L of the circle S(O, r). The turn of this is defined as follows. On the arc [a, b]
we can specify a sequence of points to = a < tl < ... < tm- 1 < tm = b such that
Li = x([tj-l t;]), i = 1, 2, ... , m, is a simple arc with bounded variation of turn
that lies on the boundary of the disc B(O, r). The turn of the arc L is defined as
the sum of the turns of the arcs Li and the turns from the side of the arc at the
points Xi = x(t i ), i = 1,2, ... , m - 1.
Theorem 9.3 admits an extension to the case when the circle is disconnected.
The first of the estimates of Theorem 9.3 extends to this case.
In Burago and Stratilatova (1965) the authors gave a formula for the length
of the circle S(O, r). They established that the length of S(O, r) is a function of
bounded variation with respect to the variable r on any interval [0, I], where
1< 00. In that article they also studied the question of circles in metrics of
bounded curvature that converge to some limit metric. They established some
conditions under which the lengths of these circles converge to the length of a
circle in the limit metric.
Concerning equidistants in a two-dimensional manifold of bounded curva-
ture we give just the following result.
Theorem 9.4. Let L be a simple arc with bounded variation of turn in a two-
dimensional manifold of bounded curvature M that does not contain peak points.
Then there is a ~ > 0 such that when 0 < r < ~, S(L, r) is a simple closed curve with
bounded variation of turn that bounds a domain B(L, r) homeomorphic to a disc.
 I. Two-Dimensional Manifolds of Bounded Curvature 153

We have
I"I [S(L, r)] ~ IKII (L) + I"rl (L) + 101 [B(L, r)\L] + O(A) + O(B),
where A and B are the end-points of the curve L.
Theorem 9.4 is a consequence of the estimates contained in Theorem 9.3 and
a theorem established in Borisov (1965). We do not give a statement ofthe latter
theorem because it is rather cumbersome.
Along with a circle and an equidistant in a two-dimensional manifold of
bounded curvature, we can also consider various other loci. For example, we
can study the locus of points such that the sum of the distances from them to
two given points is a constant. The following questions arise: 1) under what
conditions is a given locus a simple arc or a simple closed curve? 2) will the curve
have bounded variation of turn? 3) how do we estimate the variation of turn or
the length of the curve?
In Stratilatova (1965) propositions were proved that establish sufficient con-
ditions under which the answer to the questions posed above is positive.
Quasigeodesic. As we know, a geodesic in a Riemannian manifold can be
characterized as a curve for which the geodesic curvature at each point is zero.
There naturally arises the idea of considering curves characterized by similar
properties in an arbitrary two-dimensional manifold of bounded curvature. For
a curve the concepts of left and right turns are defined. In contrast to the case of
a Riemannian manifold, their sum may be non-zero. In this connection we first
introduce a characterization of a curve in which the left and right turns of the
curve occur in an equivalent way.
Let L be a simple arc with bounded variation of turn. We put

K*(L) = ~ {IKrl(L) + IKII(L) -IOI(LO)}

(LO is the arc L with the end-points excluded). The quantity K*(L) is called the
characteristic turn or winding of the arc L.
The concept of winding is also defined for curves with multiple points (see
Aleksandrov and Burago (1965».
A curve L is called a quasigeodesic if its winding is equal to zero. In the case
when M is a Riemannian manifold the concepts of quasigeodesic and geodesic
coincide. In the general case this is not so. For example, let M be a cone with
vertex 0, where the curvature at this vertex is positive. We assume that the curve
L is composed of segments of generators of the cone that start from 0 and split
the total angle at M in half. Let 0 = 0(0), 0 < 2n. Then the left and right turns
of the curve L at the point 0 are equal to n - 0/2, O(L) = 2n - 0, and it is easy
to see that K*(L) = o. The curve L is thus a quasigeodesic. No arc L inside which
o lies is a shortest curve, so L is not a geodesic.
The concept of a quasigeodesic was introduced by Aleksandrov in Alek-
sandrov (1949b). Some known results about geodesics admit an extension to the
case of two-dimensional manifolds of bounded curvature if instead of geodesics
154 Yu.G. Reshetnyak

we consider quasigeodesics. In particular, generalizations of this kind have been

obtained for a famous theorem of Lyustemik and Shnirel'man on three closed
geodesics (Aleksandrov (1950a», a theorem of Cohn-Vossen (Cohn-Vossen
(1959» on the structure of a geodesic in a domain whose curvature is small
(Aleksandrov and Burago (1965», and some others.
We shall say that M is a two-dimensional manifold of specific curvature
bounded above if there is a number K such that for any Borel set E c M we have
D(E) ~ Ka(E).
Theorem 9.5 (Aleksandrov and Burago (1965)). In a manifold of specific cur-
vature bounded above, any quasigeodesic is a geodesic.
Theorem 9.6 (Aleksandrov and Burago (1965». If the metrics (PII)' n = 1,2,
... , of bounded curvature in a two-dimensional manifold converge tamely to the
metric P and the quasigeodesics in the metric P converge to some curve L o, then Lo
is a quasigeodesic in the metric p.
Therem 9.7 (Aleksandrov and Burago (1965)). From each point X of a two-
dimensional manifold of bounded curvature in each direction we can draw a qua-
sigeodesic. The quasigeodesic can be extended indefinitely beyond any of its ends.
Special coordinate systems in a two-dimensional manifold of bounded curva-
ture. Apart from an isothermal coordinate system in a two-dimensional mani-
fold of bounded curvature, we can also consider some others.
Let ~; be the quadrant {(X,Y)E~2Ix>O,y>O} of the plane ~2. We
assume that from ~; there is cut out a polygon Po homeomorphic to a disc and
that in its place there is pasted a polygon P, also homeomorphic to a disc, cut
out from a two-dimensional manifold of bounded curvature. The manifold M =
(~; \Po) u P formed under this pasting is in tum a two-dimensional manifold of
bounded curvature.
Henceforth we shall assume that IDI(M) < n12. This condition will be en-
sured, in particular, if for a polygon P taken on a manifold of bounded curva-
ture we have IDI(P) + IKI(oP) < n12.
Theorem 9.8 (Bakel'man (1961), Bakel'man (1965». Let M be the manifold
defined as above. Then there is a topological map cp of M onto ~; such that if we
introduce in ~! the line element
ds 2 = du 2 + 2 cos t(u, v) du dv + dv 2, (9.1)
where t(u, v) = nl2 - D[cp-l(Dyv )], and Dyv is the rectangle {(x, y) E ~210 <
x ~ u, 0 < y ~ v}, then cp is an isometric map of M onto the square ~! endowed
with the metric generated by the line element (9.1).
From Theorem 9.8 it follows that for any point X of a two-dimensional
manifold of bounded curvature such that IDI(X) < nl2 we can find a circle in
which the metric ofthe manifold is defined by the line element (9.1). The coordi-
nate system in the given manifold obtained in this way is called a Chebyshev
coordinate system.
 I. Two-Dimensional Manifolds of Bounded Curvature 155

The proof of Theorem 9.8 is based on an approximation of the metric of the

manifold M by polyhedral metrics. It is much simpler than the proof of the
theorem on isometric coordinates given in § 7.
The condition IQ I(M) < nl2 in the statement of the theorem cannot be
discarded.
Theorem 9.8 has a converse. Namely, we assume that in ~! we are given a
metric w such that Iwl(~!) < n12. We put

r(u, v) = nl2 - w(Duv)

and let p be the metric determined in ~! by the line element (9.1). Then, as
Bakel'man proved in Bakel'man (1965), the manifold ~! with metric p is a
manifold of bounded curvature and the function w is its curvature.
A so-called equidistant coordinate system was constructed in Borisov (1965).
Let L be a simple arc with finite variation of turn that does not contain peak
points of the manifold. The domain of definition of the equidistant coordinate
system is the set G bounded by the arc L, two shortest curves starting from its
ends, and an arc of the equidistant S(L, h), where h > O. We assume that when
o < r ::::; h the set S(L, r) is a simple closed curve. With each point X E G there
are associated two numbers rand t. Here r = p(X, L). The number t is deter-
mined from X in a more complicated way. We shall not describe the construc-
tion given in Borisov (1965), because it is rather cumbersome. We just mention
that in the case when the manifold M is Riemannian, and the curve L is suffi-
ciently smooth, t is determined as follows. Let W), 0 ::::; t ::::; 1, be a parametriza-
tion of L such that for any t1t2 E [0,1] the length of the arc [~(tdW2)] is equal
to s(L) 1t2 - t 11. Let n(X) be the point of L that is closest to X. Then t is defined
by the condition n(X) = ~(t).
As an application of the equidistant coordinate system a formula was given
in Borisov (1965) for the variation of the length of an arc with finite variation of
turn in a two-dimensional manifold of bounded curvature.
Suppose we are given numbers K1 and K 2, where K1 ::::; K 2. We shall say
that the specific curvature of a manifold of bounded curvature M lies between
K1 and K2 if for any Borel set E E ~o(M) we have K1 O'(E)::::; Q(E) ::::; K20'(E).

Theorem 9.9. If a two-dimensional manifold of bounded curvature is such that

its specific curvature at each point lies in some interval [K1' K 2], then for any
point 0 E M there is a number ro > 0 such that any point X lying in the disc
B(O, ro) is joined to 0 by a unique shortest curve. We fix the radius [OXo] of the
disc B(O, ro) arbitrarily, and for an arbitrary point X E B(O, ro) let r = p(X, 0)
and let fJ be the angle between the shortest curves [OXo] and [OX], taken in
a definite direction. Then the metric of M in the disc B(O, ro) is defined by
the line element of the form ds 2 = dr2 + b2(r, 0) d0 2, where the function b(r, 0)
ob
is such that b(O, 0) = 0, :;-(0, fJ) = 1 for all fJ E [0, 2n), for almost all (r, fJ)
ur 02b
such that 0 < r < ro, 0 < fJ < 2n there is a finite second derivative or2 (r, fJ),
156 Yu.G. Reshetnyak

where

and for an arbitrary Borel set E

Q(E) = -
ff
E
o2b
or2 (r, 0) dr dO.

A proof of Theorem 9.9 for the case when K 1 ~ 0 was given in Aleksandrov
(1948a). However, for the general case the arguments do not differ from those in
Aleksandrov (1948a).
Proportional convergence. In certain questions a concept of convergnce of
metrics different from the one used above turns out to be useful. We assume that
on the set M there are specified metrics p, Pn' n = 1,2, .... We shall say that as
n ~ 00 the metrics Pn converge proportionally to the metric P if for any e > 0 we
can find a number no such that for any n ~ no
1
-1-p(X, Y) ~ Pn(X, Y) ~ (1 + e)p(X, Y)
+e
for any X, Y E M.
Let M be a two-dimensional manifold with intrinsic metric, and A an arbi-
trary point of M. We shall say that M has a tangent cone at A if there is a cone
Q(O, r) that admits a topological map qJ onto a neighbourhood of A such that A
corresponds to the vertex 0 of the given cone and for any X, Y E Q(O, r) such
that X -=1= Y

PQ(X, Y) ~1
PM[qJ(X), qJ(Y)]
asX~O, Y~O.

Theorem 9.10 (Burago (1965c), Zalgaller (1950b)). A two-dimensional mani-

fold of bounded curvature has a tangent cone at any point A of it that is not a peak
point.
In Zalgaller (1950b) this theorem was proved for the case of manifolds whose
curvature is non-negative. In Burago (1965c) a proof was given for the general
case.

Theorem 9.11 (Reshetnyak (1959)). Let M be a compact two-dimensional man-

ifold with intrinsic metric p. We assume that M has a tangent cone at each point
of it. Then there is a sequence of polyhedral metrics (Pn), n = 1, 2, ... , defined in
M, that converges proportionally to the metric of the manifold.
Classes of surfaces with metric of bounded curvature. Let F be a two-
dimensional surface in three-dimensional Euclidean space. We assume that any
 I. Two-Dimensional Manifolds of Bounded Curvature 157

two points X, Y E F can be joined by a rectifiable curve lying on F and that the
intrinsic metric defined in a natural way on M is compatible with the natural
topology of F as a subset of 1E3. This condition will be satisfied under rather
weak assumptions about F. For example, it is satisfied if F is a smooth slH"face
of class C 1 • There arises the following question: under what additional c;ndi-
tions is a surface F endowed with an intrinsic metric a two-dimensional mani-
fold of bounded curvature? Let us quote some results in this direction.
First of all we mention that any convex surface in 1E3 is a two-dimensional
manifold of bounded curvature. The curvature of the manifold as a set function
is non-negative. As Aleksandrov showed (see Aleksandrov (1948a)) any two-
dimensional manifold with non-negative curvature homeomorphic to a sphere
is isometric to a closed convex surface. (For further generalizations of the given
theorem see Pogorelov (1957), Pogorelov (1969).) This theorem enables us to
conclude, in particular, that any two-dimensional manifold of non-negative cur-
vature can be realized at least locally in the form of a convex surface.
By analogy with the case of convex surfaces it is natural to pose the following
problem: to find a class 9l of surfaces in 1E3 determined by certain outwardly
geometrical conditions and such that if a surface F in 1E3 belongs to the class 9l,
then it is a two-dimensional manifold of bounded curvature, and conversely,
given any manifold of bounded curvature, for any point of it we can find a
neighbourhood isometric to some surface F E 9l. This problem seems rather
difficult, and probably does not have an effective solution. Some particular
classes of surfaces are known that satisfy the first of the conditions in the state-
ment of the problem, namely such that the surfaces that occur in this class are
manifolds of bounded curvature.
In this connection we mention, first of all, the surface representable by the
difference of convex surfaces, introduced by A. D. Aleksandrov (Aleksandrov
(1949a)). We shall say that F is a surface representable by the difference of
convex surfaces if in some orthogonal Cartesian coordinate system F can be
specified by the equation
z = fl(x, y) - f2(x, y),
where (x, y) E G, a convex domain of ~2, and fl (x, y) and f2(x, y) are convex
functions. (A function is said to be convex if the segment joining two arbitrary
points of its graph lies above the graph of the function.) It was proved in Alek-
sandrov (1949a) that if F is a surface representable by the difference of convex
surfaces, then F is a two-dimensional manifold of bounded curvature.
Another class consists of the surfaces of bounded extrinsic curvature, intro-
duced by Pogorelov in Pogorelov (1956). Let F be a smooth surface in 1E 3 , and
E its convex hull, that is, the smallest convex set containing F. Let E be the
totality of all internal points of F that belong to the boundary of E. If X E E,
then X is a smooth point of the boundary of the convex body E, and in it there
is defined a vector v(X), the outward normal vector of the convex body E at the
point X. Laying off the vector v(X) from a fixed point 0 in 1E 3 , we obtain a map
v of the set E into the unit sphere S(O, 1) of the space. We denote by 8(E) the
ISS Yu.G. Reshetnyak

area of the set v(E) on the sphere S(O, 1), c9(E) = u[v(E)). We shall say that F is a
surface of bounded extrinsic curvature if there is a constant 8 0 < 00 such that for
any finite system of pairwise non-overlapping domains FI , F2 , ••• , Fm of F we have

Any surface of bounded extrinsic curvature is a two-dimensional manifold of

bounded curvature, as Pogorelov showed in Pogorelov (1956).
Smooth surfaces of class wi were considered in Bakerman (1956). Let F be
a smooth surface in Ie. We shall say that F is smooth of class wi if any point
of it has a neighbourhood that admits a parametrization f(u, v) = [Xl (u, v),
X2(U, v), X3(U, v)], (u, v) E B(O, 1), such that the functions Xi have generalized (in
the sense of Sobolev) derivatives of the first and second orders in the disc
B(O, 1), where the derivatives of/au, af/av are continuous and linearly indepen-
dent at each point (u, v) E B(O, 1), and the second derivatives are square inte-
grable in B(O, 1). It was shown in Bakerman (1956) that any smooth surface of
class wi is a two-dimensional manifold of bounded curvature. This class of
surfaces is interesting in that most of the results in a classical course of differen-
tial geometry admit an extension to surfaces that belong to this class.
The class of saddle surfaces in 1E3 was introduced in Shefer (1964). A surface
F is called a saddle surface if no plane can cut off a crust from it, that is, for any
plane P, any connected component of the set F\P contains boundary points of
F. In the regular case the class of saddle surfaces coincides with the class of
surfaces of non-positive Gaussian curvature. The main result of Shefer (1964)
consists in proving that a saddle surface is a manifold of bounded curvature, and
its curvature as a set function is non-positive. A more detailed discussion of the
paper Shefer (1964), and of some others that are an extension of it, is given in
one of the books in the present series (Burago (1989)).
Generalizations of two-dimensional manifolds .of bounded curvature. Apart
from Riemannian geometry on a two-dimensional manifold, we can specify
various others, for example, pseudo-Riemannian geome.try, the geometry of
affine connection, and Finsler geometry. In each ofthese geometries the concept
of curvature is defined and there arises the question of the construction for them
of an analogue of the concept of a two-dimensional manifold of bounded curva-
ture. For the case of pseudo-Riemannian geometry the solution of this problem
is known. Namely, in Gurevich (1979a), (1979b), (1981a), (1981b) the concept of
plane kinematics of bounded curvature is introduced; this is a generalization of
the concept ofa manifold of bounded curvature to the pseudo-Riemannian case.
(The term "kinematics" is connected with a known physical interpretation of
pseudo-Riemannian geometry.)
Another question is how to regard the multidimensional analogue of the
theory presented in this article. In this setting there are individual results.
Namely, Aleksandrov introduced the concepts of metric spaces of curvature no
greater than some number K and a space of curvature bounded below by K.
 I. Two-Dimensional Manifolds of Bounded Curvature 159

Some questions of the theory of spaces of unilaterally bounded curvature

are considered in the article by Berestovskii and Nikolkaev (Part II of this
volume).
In the general case a theory of multidimensional manifolds that can be re-
garded as the complete analogue of two-dimensional manifolds of bounded
curvature has not been constructed.
On isothermal coordinates. The possibility of using isothermal coordinates to
solve certain problems of the theory of surfaces was observed quite a long time
ago. In this connection we mention the papers Lozinskij (1944), Beckenbach and
Rado (1933), Beurling (1952).

References *

Aleksandrov, A.D. [1944]: Intrinsic metric of a convex surface in a space of constant curvature.
Dok!. Akad. Nauk SSSR 45, 3-6 [Russian], Zb!.61,376
Aleksandrov, A.D. [1945a]: Isoperimetric inequalities on curved surfaces. Dok!. Akad. Nauk SSSR
47,235-238 [Russian], Zb!.61,376
Aleksandrov, A.D. [1945b]: Convex surfaces as surfaces of positive curvature. Dok!. Akad. Nauk
SSSR 50, 27-30 [Russian]
Aleksandrov, A.D. [1945c]: One isoperimetric problem. Dok!. Akad. Nauk SSSR 50, 31-34
[Russian]
Aleksandrov, A.D. [1947]: The method of gluing in the theory of surfaces. Dok!. Akad. Nauk 57,
863-865 [Russian], Zb!.29,228
Aleksandrov, A.D. [1948a]: Intrinsic Geometry of Convex Surfaces. Gostekhizdat, Moscow-
Leningrad [Russian], Zb!.38,352. German trans!.: Academie-Verlag. Berlin 1955, Zb!.65,151
Aleksandrov, A.D. [1948b]: Foundations of the intrinsic geometry of surfaces. Dok!. Akad. Nauk
SSSR 60,1483-1486 [Russian], Zb!.38,351
Aleksandrov, A.D. [1948c]: Curves on manifolds of bounded curvature. Dok!. Akad. Nauk SSSR
63, 349-352 [Russian], Zbl.41,508
Aleksandrov, A.D. [1949a]: Surfaces representable by the difference of convex functions. Izv. Akad.
Nauk KazSSR 60, Ser. Mat. Mekh. 3, 3-20 [Russian]
Aleksandrov, A.D. [1949b]: Quasigeodesics. Dok!. Akad. Nauk SSSR 69, 717-720 [Russian],
Zb1.39,390
Aleksandrov, A.D. [1950a]: Quasigeodesics on manifolds homeomorphic to a sphere. Dok!. Akad.
Nauk SSSR 70,557-560 [Russian], Zbl.39,391
Aleksandrov, A.D. [1950b]: Surfaces representable by differences of convex functions. Dok!. Akad.
Nauk SSSR 72,613-616 [Russian], Zb!.39,180
Aleksandrov, A.D. [l95Oc]: Convex Polyhedra. Gostekhizdat, Moscow-Leningrad [Russian],
Zbl.41,509
Aleksandrov, A.D. [1951]: A theorem on triangles in a metric space and some applications of it. Tr.
Mat. Inst. Steklova 38, 5-23 [Russian], Zb!.49,395.

* For the convenience of the reader, references to reviews in Zentralblatt flir Mathematik (Zb!.),
compiled using the MATH database, and Jahrbuch tiber die Fortschritte der Mathematik (Jbuch)
have, as far as possible, been included in this bibliography.
160 Yu.G. Reshetnyak

Aleksandrov, A.D. [1954]: Synthetic methods in the theory of surfaces. Convegno Int. Geometria
DilTerenziale, Italia 1953, 162-175, ZbI.55,409.
Aleksandrov, A.D. [1957a]: Ruled surfaces in metric spaces. Vestn. Leningr. Univ. 12, No.1, 5-26
[Russian], Zb1.96,166
Aleksandrov, A.D. [1957b]: Dber eine Verallgemeinerung der Riemannschen Geometrie. Schr. For-
schungsinst. Math. 1, 33-84, Zb1.77,357
Aleksandrov, A.D., Borisov, Yu.F. and Rusieshvili, G.I. [1975]: An extremal property of cones in
Lobachevskij space. Tr. Ped. Inst. GruzSSR 2, 3-27 [Russian]
Aleksandrov, A.D. and Burago, Yu.D. [1965]: Quasigeodesics. Tr. Mat. Inst. Steklova 76,49-63.
English transl.: Proc. Steklov Inst. Math. 76,58-76 (1965), ZbI.142,204
Aleksandrov, A.D. and Reshetnyak Yu.G. [1989]: General Theory of Irregular Curves, Kluwer
Academic Publishers, Dordrecht, Boston-London, ZbI.691.53002
Aleksandrov, A.D. and Strel'tsov, V.V. [1953]: Estimate of the length of a curve on a surface. Dok!.
Akad. Nauk SSSR 93, 221-224 [Russian], Zb1.52,182
Aleksandrov, A.D. and Strel'tsov, V.V. [1965]: The isoperimetric problem and estimates of the
length of a curve on a surface. Tr. Mat. Inst. Steklova 76,67-80. English transl.: Proc. Steklov
Inst. Math. 76,81-99 (1965), Zb1.147,224
Aleksandrov, A.D. and Zalgaller, V.A. [1962]: Two-dimensional manifolds of bounded curvature.
Tr. Mat. Inst. Steklov 63, English transl.: Intrinsic geometry of surfaces. Transl. Math. Mono-
graphs 15, Am. Math. Soc. (1967), Zb1.122,l70
Alekseevskij, D.V., Vinogradov, A.M. and Lychagin, V.V. [1988]: Basic ideas and notions of dilTer-
ential geometry. Itogi Nauki Tekh., Ser. Sovrem. Probl. Mat., Fundam. Napravleniya 28, 1-
299. English transl. in: Encycl. Math. Sci. 28, Springer, Berlin-Heidelberg-New York 1991,
ZbI.675.53001
Bakel'man, I.Ya. [1956]: The dilTerential geometry of smooth non-regular surfaces. Usp. Mat. Nauk
11, No.2, 67-124 [Russian], Zb1.70,392
Bakel'man, I.Ya. [1961]: Generalized Chebyshev nets and manifolds of bounded curvature.
Dok!. Akad. Nauk SSSR 138, 506-507. English transl.: SOy. Math., Dok!. 2, 631-632,
Zbl.l04,168
Bakel'man, I.Ya. [1965]: Chebyshev nets in manifolds of bounded curvature. Tr. Mat. Inst. Stek-
lova 76,124-129. English transl.: Proc. Steklov Inst. Math. 76, 154-160 (1967), Zbl.171,432
Beckenbach, E.F. and Rado, T. [1933]: Subharmonic functions and surfaces of negative curvature.
Trans. Amer. Math. Soc. 35, 662-674, ZbI.7,130
Belinskij, S.P. [1975]: A mapping of a cone onto a polyhedron for the Lobachevskij plane. Sib. Mat.
Zh. 16,203-211. English transl.: Sib. Math. J. 16, 159-164, ZbI.324.53044
Beurling, A. [1952]: Sur la geometrie metrique des surfaces Ii courbure totale o. Comm. Sem. Math.
Univ. Lund, Suppl., 7-11, Zb1.48,388
Borisov, Yu.F. [1949]: Curves on complete two-dimensional manifolds with boundary. Dok!. Akad.
Nauk SSSR 64, 9-12 [Russian], Zb1.36,234
Borisov, Yu.F. [1950]: Manifolds of bounded curvature with boundary. Dokl. Akad. Nauk SSSR
74,877-880 [Russian], Zb!.39,180
Borisov, Yu.F. [1955]: The geometry of the semi-neighbourhood of a curve in a two-dimensional
manifold of bounded curvature. Dok!. Akad. Nauk SSSR 103, 537-539 [Russian], Zb!.67,402
Borisov, Yu.F. [1965]: Semi-neighbourhoods and variation of the length ofa curve on a surface. Tr.
Mat. Inst. Steklova 76, 26-48. English trans!.: Proc. Steklov Inst. Math. 76, 31-57 (1967),
Zb!.173,504
Burago, Yu.D. [1965a]: Curves in convergent spaces. Tr. Mat. Inst. Steklova 76, 5-25. English
trans!.: Proc. Steklov Inst. Math. 76,3-29 (1967), Zb1.142,203
Burago, Yu.D. [1965b]: Closure of the class of manifolds of bounded curvature. Tr. Mat. Inst.
Steklova 76, 141-147. English trans!.: Proc. Steklov Inst. Math. 76, 175-183 (1967), ZbI.168,200
Burago, Yu.D. [1965c]: On the proportional approximation of a metric. Tr. Mat. Inst. Steklova 76,
120-123. English trans!.: Proc. Steklov Inst. Math. 76, 149-153 (1967), ZbI.142,204
Burago, Yu.D. [1978]: On the radius ofinjectivity on a surface with curvature bounded from above.
Ukr. Geom. Sb. 21, 10-14 [Russian], Zb!.438.53052
 I. Two-Dimensional Manifolds of Bounded Curvature 161

Burago, Yu.D. [1989]: Geometry of surfaces in Euclidean spaces. Itogi Nauki Tekh., Ser. Sovrem.
Probl. Math., Fundam. Napravleniya 48, 5-97. English. transl. in: Encycl. Math. Sc. 48,
Springer, Berlin-Heidelberg-New York 1992, ZbI.711.53003
Burago, Yu.D. and Stratilatova, M.B. [1965]: A circle on a surface. Tr. Mat. Inst. Steklova 76,
88-114. English transl.: Proc. Steklov Inst. Math. 76, 109-141 (1967), Zb1.142,203
Burago, Yu.D. and Zalgaller, V.A. [1965]: The isometric problem for domains with restricted width
on a surface. Tr. Mat. Inst. Steklova 76, 81-87. English transl.: Proc. Steklov Inst. Math. 76,
100-108 (1967)
Burago, Yu.D. and Zalgeller, V.A. [1980]: Geometric Inequalities. Nauka, Leningrad. English
transl.: Springer, Berlin-Heidelberg-New York 1988, ZbI.436.52009
Cohn-Vossen, S.E. [1959]: Some Problems of DitTerential Geometry. Gosudarstv. Izdat. Fiz.-Mat.
Lit., Moscow [Russian], Zb1.91,341
Fiala, F. [1941]: Le probleme des isoperimetres sur les surfaces ouvertes Ii courbure positive. Com-
ment. Math. Helv. 13,293-346, ZbI.25,230.
Gurevich, V.L. [1970]: Rotation of a longest curve in kinematics of bounded curvature. Deposited
at VINITI, No. 361-79 [Russian], ZbI.423.53007
Gurevich, V.L. (1979a): Kinematics of bounded curvature. Sib. Mat. Zh. 20, No.1, 37-48. English
transl.: Sib. Math. J. 20, 26-33 (1979), Zbl.412.53029
Gurevich, V.L. [1979b]: Two-dimensional pseudo-Riemannian manifolds of bounded curvature.
Dokl. Akad. Nauk SSSR 247, 22-25. English transl.: Sov. Math., Dokl. 20, 638-641 (1979),
Zb1.428.53030
Gurevich, V.L. [1981a]: Two-dimensional kinematics of bounded curvature. I. Sib. Mat. Zh. 22, No.
5,3-30. English transl.: Sib. Math. J. 22, 653-672 (1982), ZbI.492.53013
Gurevich, V.L. [1981b]: Two-dimensional kinematics of bounded curvature. II. Sib. Mat. Zh. 22,
No.6, 41-53; English transl.: Sib. Math. J. 22, 836-845 (1982), Zbl.492.53014
Hayman W.K., Kennedy P.B. [1976]: Subharmonic Functions, vol. I. Academic Press, London-
New York-San Francisco, ZbI.419.31001
Huber, A. [1954]: On the isoperimetric inequality on surfaces of variable Gaussian curvature. Ann.
Math., II. Ser. 60, 237-247, Zb1.56,158
Huber, A. [1957]: Zur isoperimetrischen Ungleichung auf gekriimmten Flachen. Acta Math. 97,
95-101, Zb1.77,358
Huber, A. [1960]: Zum potentialtheoretischen Aspekt der Alexandrowschen Flachentheorie. Com-
ment. Math. Helv. 34, 99-126, Zb1.105,161
Ionin, V.K. [1969a]: Riemannian spaces with Euclidean isoperimetric inequality. Dokl. Akad. Nauk
SSSR 188,990-992. English transl.: Sov. Math. Dokl. 10, 1225-1227 (1970), Zb1.208.248
Ionin, V.K. [1969b]: Isoperimetric and certain other inequalities for a manifold of bounded curva-
ture. Sib. Mat. Zh. 10,329-342. English transl.: Sib. Math. J. 10,233-243, Zb1.188,275
Ionin, V.K. [1972a]: Isoperimetric inequalities in simply-connected Riemannian spaces of non-
positive curvature. Dokl. Akad. Nauk SSSR 203,282-284. English transl.: Sov. Math. Dokl. 13,
378-381 (1972), ZbI.258.52011
Ionin, V.K. [1972b]: Isoperimetric inequalities for surfaces of negative curvature. Sib. Mat. Zh. 13,
933-938. English transl.: Sib. Math. J. 13,650-653 (1973), Zb1.241.52002
Ionin, V.K. [1982]: Upper estimate of the radius of the circumscribed disc of a domain on a
manifold of bounded curvature. Symposium on geometry in the large and foundations of the
theory of relativity. Novosibirsk, 49-50 [Russian], ZbI.489.53001
Kobayashi, S. and Nomizu, K. [1963]: Foundations of DitTerential Geometry. Vol. I. Interscience,
New York-London-Sydney, ZbI.119,375.
Kuratowski, K. [1966]: Topology. Academic Press, New-York-London, ZbI.158,408
Lewy, H. [1935]: A priori limitations for solutions of Monge-Ampere equatons. Trans. Am. Math.
Soc. 37,417-434, Zb1.11,350
Lozinskij, S.M. [1944]: Subharmonic functions and their applications to the theory of surfaces. Izv.
Akad. Nauk SSSR, Ser. Mat. 8,175-194 [Russian], Zb1.61,241
Men'shov, D.E. [1948]: Sur une generalisation d'un theoreme de M.H. Bohr. Mat. sb. (2) 44,
339-356
162 Yu.G. Reshetnyak

Petrovskij, I.G. [1961]: Lectures on Partial Differential Equations. 3rd ed. Gosudarstv. Izdat. Fiz.-
Mat. Lit., Moscow, Zb1.l15,81. English transl.: I1iffe Books Ltd., London 1967, Zb1.l63,1l7
Pogorelov, A.V. [1949]: Quasigeodesic lines on a convex surface. Mat. Sb., Nov. Ser. 25, 275-306.
English transl.: Am. Math. Soc., Transl., I. Ser. 6, 430-473, Zb1.41,89
Pogorelov, A.V. [1956]: Surfaces of Bounded Extrinsic Curvature. Izdat. Khar'kov. Gos. Univ.,
Khar'kov [Russian], Zbl. 74,176
Pogorelov, A.V. [1957]: Some Questions of Geometry in the Large in Riemannian Space. Izdat.
Khar'kov. Gos. Univ., Khar'kov, ZbI.80,149. German transl.: VEB Deutscher Verlag der
Wissenschaften, Berlin 1960, ZbI.90,125.
Pogorelov, A.V. [1961]: Some Results on Geometry in the Large. Izdat. Khar'kov. Gos. Univ.,
Khar'kov. English transl.: Adv. Math. 1, 191-264 (1964), Zb1.l29,357
Pogorelov, A.V. [1969]: Extrinsic Geometry of Convex Surfaces. Nauka, Moscow. English transl.:
Transl. Math. Monographs 35, Am. Math. Soc., Providence 1973, Zbl.31 1.53067
Privalov, 1.1. [1937]: Subharmonic Functions. Moscow-Leningrad [Russian]
Rado, T. [1925]: Uber den Begriff der Riemannschen Fliichen. Acta Szeged. 2, 101-121, Jbuch
51,273
Reshetnyak, Yu.G. [1954]: Isothermal coordinates in manifolds of bounded curvature. Dok\. Akad.
Nauk SSSR 94, 631-633 [Russian], Zb1.58,384
Reshetnyak, Yu.G. [1959]: Investigation of manifolds of bounded curvature by means of isothermal
coordinates. Izv. Sib. Otd. Akad. Nauk SSSR 1959, No. 10, 15-28 [Russian], Zbl.l15,164
Reshetnyak, Yu.G. [1960]: Isothermal coordinates in manifolds of bounded curvature. I, II. Sib.
Mat. Zh.1, 88-116, 248-276 [Russian], Zb1.l08,338
Reshetnyak, Yu.G. [l961a]: A special map ofa cone onto a manifold. Mat. Sb., Nov. Ser. 53, 39-52
[Russian], Zb1.l24,152
Reshetnyak, Yu.G. [1961 b]: On the isoperimetric property of two-dimensional manifolds of curva-
ture not greater than K. Vestn. Leningr. Univ.16, No. 19,58-76, Zb1.l06,365
Reshetnyak, Yu.G. [1962]: On a special map ofa cone into a two-dimensional manifold of bounded
curvature. Sib. Mat. Zh. 3, 256-272 [Russian], Zb1.l24,153
Reshetnyak, Yu.G. [1963a]: The length of a curve in a manifold of bounded curvature with an
isothermal line element. Sib. Mat. Zh. 4, 212-226 [Russian], Zb1.l28,159
Reshetnyak, Yu.G. [1963b]: Rotation of a curve in a manifold of bounded curvature with an
isothermal line element. Sib. Mat. Zh. 4, 870-911 [Russian], Zb1.l45,193
Rusieshvili, G.I. [1957]: Some extremal problems of the intrinsic geometry of manifolds in
Lobachevskij space. Vestn. Leningr. Univ.12, No. I, 76-79 [Russian], Zb1.94,352
Shefel', S.Z. [1964]: On the intrinsic geometry of saddle surfaces. Sib. Mat. Zh. 5, 1382-1396 [Rus-
sian], Zb1.l42,l89
Stratilatova, M.B. [1965]: Certain geometrical loci of points on a surface. Tr. Mat. Inst. Steklova 76,
115-119. English transl.: Proc. Steklov Inst. Math. 76, 142-148 (1965), Zb1.l42,204
Strel'tsov, V.V. [1952a]: Estimate of the length of an open polygon on a polyhedron. Izv. Akad.
Nauk KazSSR 1,3-36 [Russian].
Strel'tsov, V.V. [1952b]: Some extremal problems of the intrinsic; geometry of polyhedra. Izv. Akad.
Nauk KazSSR 2, 37-63 [Russian].
Strel'tsov, V.V. [1953]: General points of geodesics. Izv. Akad. Nauk KazSSR 3, 89-103 [Rus-
sian], Zb1.55,410
Strel'tsov, V.V. [1956]: A map of a surface of negative curvature. Izv. Akad. Nauk KazSSR 5, 29-44
[Russian], Zbl.72,174
Toponogov, V.A. [1969]: An isoperimetric inequality for surfaces whose Gaussian curvature is
bounded above. Sib. Mat. Zh. 10, 144-157. English transl.: Sib. Math. J. 10, 104-113 (1969),
Zb1.l86,558
Verner, A.L. [1965]: A semigeodesic coordinate net on tubes of non-positive curvature. Tr. Mat.
Inst. Steklova 76, 130-140. English transl.: Proc. Steklov Inst. Math. 76, 161-173 (1967),
Zbl.156,430
Zalgaller, V.A. [1950a]: Curves with bounded variation ofrotation on convex surfaces. Mat. Sb.,
Nov. Ser. 26, 205-214 [Russian], Zb1.39,181
 I. Two-Dimensional Manifolds of Bounded Curvature 163

Zalgaller, V.A. [1950b]: The circle on a convex surface. The local almost-isometry of a convex
surface to a cone. Mat. Sb., Nov. Ser. 26,401-424 [Russian], Zbl.37,250
Zalgaller, V.A. [1952]: On the class of curves with bounded variation of rotation on a convex
surface. Mat. Sb. Nov. Ser. 30, 59-72 [Russian], Zb!.49,395
Zalgaller, V.A. [1956]: Foundations of the theory of two-dimensional manifolds of bounded curva-
ture. Dokl. Akad. Nauk SSSR 108,575-576 [Russian], Zb!.70,392
Zalgaller, V.A. [1965]: Curves on a surface near to cusp points. Tr. Mat. Inst. Steklova 76, 64-66.
English trans!.: Proc. Steklov Inst. Math. 76,77-80 (1967), Zb1.l56,430
 II. Multidimensional Generalized
Riemannian Spaces
V.N. Berestovskij and I.G. Nikolaev
Translated from the Russian
by E. Primrose

Contents

Introduction .................................................. 168

0.1. Riemannian Spaces .................................... 168
0.2. Generalized Riemannian Spaces ......................... 168
0.3. Riemannian Geometry and Generalized Riemannian Spaces .. 170
0.4. A Brief Characterization of the Article by Chapters ......... 172
0.5. In What Sense Do the Stated Results Have Multidimensional
Character? ........................................... 174
0.6. Final Remarks on the Text .............................. 174
Chapter 1. Basic Concepts Connected with the Intrinsic Metric ....... 175
§ 1. Intrinsic Metric, Shortest Curve, Triangle, Angle, Excess of a Triangle 175
1.1. Intrinsic Metric ...................................... . 175
1.2. Shortest Curve ....................................... . 176
1.3. Triangle ............................................. . 176
1.4. Angle ............................................... . 176
1.5. Excess .............................................. . 178
§ 2. General Propositions about Upper Angles ................... . 178
§ 3. The Space of Directions at a Point. K -Cone. Tangent Space ..... . 179
3.1. Direction ............................................ . 180
3.2. K -Cone ............................................. . 180
3.3. Tangent Space ........................................ 181
§4. Remarks, Examples ........................................ 181
4.1. Intrinsic Metric, Shortest Curve, Angles ................... 181
4.2. An Assertion Completely Dual to Theorem 2.1, that is,
the Corresponding Lower Bound for the Lower Angle Does
not Hold ............................................. 183
4.3. Tangent Space, Space of Directions ....................... 183
166 V.N. Berestovskij and I.G. Nikolaev

Chapter 2. Spaces of Curvature ~ K (and ~ K') .................... 185

§ 5. Spaces of Curvature ~ K. The Domain RK and its Basic
Properties ................................................ 185
5.1. Definition of a Space of Curvature ~ K ................... 185
5.2. Basic Properties of the Domain RK .....•........•....••.. 186
5.3. The Domain PK ••••••••••••••••••••••••••••••••••••••• 188
§6. The Operation of Gluing ................................... 188
6.1. Gluing of Metric Spaces with Intrinsic Metric .............. 188
§ 7. Equivalent Definitions of Upper Boundedness of the Curvature ... 189
7.1. Conditions under which a Space of Curvature ~ K is the
Domain RK ..........................................• 190
7.2. Connection with the Riemannian Definition of Curvature .... 191
7.3. Definition of Upper Boundedness of Curvature ............. 192
7.4. Non-Expanding Maps in Spaces of Curvature ~ K ......... 192
7.5. Boundedness of the Curvature from the Viewpoint of
Distance Geometry .................................... 193
§ 8. Space of Directions, Tangent Space at a Point of a Space of
Curvature ~ K ............................................ 193
8.1. Conditions under which a Shortest Curve Goes out in each
Direction ............................................. 194
8.2. Intrinsic Metric in Dp .................................. 195
8.3. Tangent Space ........................................ 196
§9. Surfaces and their Areas .................................... 196
9.1. The Definition of the Area of a Surface .................... 196
9.2. Properties of Area ..................................... 197
9.3. Ruled Surfaces in RK .•................................. 197
9.4. Isoperimetric Inequality ................................ 198
9.5. Plateau's Problem ..................................... 198
§ 10. Spaces of Curvature both ~ K and ~ K' ...................... 198
10.1. Definition ofa Space of Curvat tire both ~ K and ~ K' ..... 199
10.2. Basic Properties of a Domain RK',K •••••••••••••••••••••• 199
10.3. Equivalent Definitions of Boundedness of Curvature ....... 201
§ 11. Remarks, Examples ........................................ 201
11.1. Spaces of Curvature ~ K as a Generalization of
Riemannian Spaces •.................................. 201
11.2. Polyhedral Metrics ................................... 204
11.3. Spaces of Curvature ~ K' .............................. 205

Chapter 3. Spaces with Bounded Curvature ........................ 206

§ 12. CO-Riemannian Structure in Spaces with Bounded Curvature .... 207
12.1. Definition of a Space with Bounded Curvature ............ 207
12.2. The Tangent Space at a Point of a Space with Bounded
Curvature ........................................... 208
12.3. Introduction of CO-Smooth Riemannian Structure ......... 209
 II. Multidimensional Generalized Riemannian Spaces 167

§ 13. Parallel Translation in Spaces with Bounded Curvature ......... 210

13.1. Construction of a Parallel Translation ................... 210
13.2. Statement of the Main Results .......................... 211
13.3. Plan of the Proof of the Main Results of § 13 .............. 213
§ 14. Smoothness of the Metric of Spaces with Bounded Curvature .... 216
14.1. Statement of the Main Result.. ...... ................... 216
14.2. Plan of the Proof of Theorem 14.1 ....................... 218
§ 15. Spaces with Bounded Curvature and Limits of Smooth
Riemannian Metrics ....................................... 220
15.1. Approximation ofthe Metric ofa Space with Bounded
Curvature by Smooth Riemannian Metrics ............... 220
15.2. A Space of Riemannian Manifolds with Sectional Curvatures
Bounded in Aggregate ................................. 222

Chapter 4. Existence of the Curvature of a Metric Space at a Point

and the Axioms of Riemannian Geometry .......................... 225
§ 16. The Space of Directions of an Arbitrary Metric Space ........... 226
16.1. Distance between Directions ........................... 226
16.2. Space of Directions ................................... 230
§ 17. Curvature of a Metric Space ................................ 230
17.1. Definition of Non-isotropic Riemannian Curvature ........ 230
17.2. Existence of Curvature at a Point ....................... 231
17.3. Geometrical Meaning of Sectional Curvature ............. 232
17.4. Isotropic Riemannian Curvature ........................ 235
17.5. Wald Curvature and its Connection with Isotropic
Riemannian Curvature ................................ 235
17.6. Continuity of Curvature ............................... 236
§ 18. Axioms of Riemannian Geometry ............................ 237
18.1. Synthetic Description of C 2 '''-Smooth Riemannian
Manifolds ........................................... 237
18.2. Synthetic Description of Cm'''-Smooth Riemannian Manifolds
(m = 3, 4, ... ) ........................................ 238
18.3. Isotropic Metric Spaces ............................. ,. 240

References .................................................... 241

168 V.N. Berestovskij and I.G. Nikolaev

Introduction

0.1. Riemannian Spaces. A Riemannian space is usually understood as a

space such that in small domains of it Euclidean geometry holds approximately,
to within infinitesimals of higher order in comparison with the dimensions of the
domain.
Distances in a Riemannian space are specified by means of a metric ten-
sor: from the metric tensor we calculate the lengths of piecewise-differentiable
curves, and the distance between a pair of points of the Riemannian space is
taken to be equal to the length of the "shortest curve" of this type that joins
these points.
Riemannian geometry traditionally considers spaces whose geometry differs
from Euclidean geometry by an infinitesimal quantity of the second order. The
difference in the second order of distances of Riemannian space from the corre-
sponding Euclidean distances is controlled by the sectional curvature, which
plays a key role in Riemannian geometry.
Classical Riemannian geometry assumes that the metric tensor has suffi-
ciently good differential properties. The minimal requirements on the smooth-
ness of the metric - the fact that it is twice differentiable - are due to the fact that
the sectional curvature of a Riemannian space is calculated in terms of the
second derivatives of the metric tensor.
Henceforth a Riemannian space, or as we shall more often call it, a Rieman-
nian manifold, will be understood as a Riemannian space whose metric tensor
is at least twice continuously differentiable.
In this article we shall be dealing with Riemannian metrics of inferior
smoothness that are "generalized" with respect to the traditional point of view.
We shall also call such spaces Riemannian manifolds, each time making precise
the smoothness class oftheir metric tensor.

0.2. Generalized Riemannian Spaces. The main concepts of Riemannian

geometry are "distance" and "curvature", which estimates the difference be-
tween the distances of a Riemannian space and Euclidean distances. However,
"curvature" can be defined by starting from distances. This gives the pos-
sibility of defining the "boundedness of curvature" of a metric space. The
"generalized Riemannian spaces" thus obtained are the main object of study in
this article.
Thus, the starting point is a metric space. To avoid pathological cases we
shall consider only those spaces whose metric is intrinsic. This means that the
distance between points of the space coincides with the length (measured in the
metric of the space) of the "shortest curve" joining these points.
At the basis of the definition of boundedness of curvature lies the following
observation.
It is known that the sectional curvature of a Riemannian manifold at a point
in a two-dimensional direction can be calculated in terms of the excesses of
 II. Multidimensional Generalized Riemannian Spaces 169

geodesic triangles in the following way:

K,AP) = lim b(T)/u(T), (0.1)

<1
T--P

where b(T) is the excess of the triangle T, that is, the number equal to the sum
of its angles minus n, u(T) is the "area" of the triangle T, that is, the number
equal to the area of a Euclidean triangle with the same lengths of sides as T, and
the triangles T are contracted to a point P so that in the limit they are tangent
to the plane element s at the point P (Cartan (1928».
For general metric spaces there is the concept of the angle between shortest
curves (the "upper angle", which always exists; for the definition see 1.4.1). To
calculate the angle between curves in the metric space, we need to do the follow-
ing, roughly speaking. We consider points lying on curves other than the origin.
From the distances between these points and the origin we calculate the angle
as if the curves were situated in Euclidean space, and then proceed to the limit,
making the points tend to the origin.
The role of shortest geodesic in a metric space is played by a shortest curve,
that is, a curve whose length is equal to the distance between its beginning and
end.
Thus, in a metric space we can consider triangles formed from shortest
curves.
By (0.1), it is natural to define the boundedness above of the curvature of a
metric space 9Jl by requiring that for an arbitrary sequence of triangles contract-
ing to any point we have

lim b(T)/u(T) ~ K, (0.2)

T--P

where b(T) and u(T) have in this case the same meaning as in (0.1) (it is neces-
sary to specify the case of degenerate triangles in (0.2) separately). If in addition
we require that

lim b(T)/u(T) ~ K',

T--P

then we specify the bilateral boundedness of the curvature of the metric space.
In the first case we arrive at the so-called spaces of curvature ~ K, and in the
second case at the spaces of curvature both ~ K and ~ K'.
These spaces were introduced by A.D. Aleksandrov in Aleksandrov (1951).
The theory of spaces of curvature ~ K (and ~ K') is based on those ideas that
lie at the basis of the intrinsic geometry of convex surfaces (Aleksandrov (1948».
Here the important features are the axiomatic method, in which the spaces
under consideration are defined by a minimal set of properties of their metric,
and the synthetic method, which goes back to Euclid and is based on geometri-
cal constructions in the spaces, and consideration of curves, triangles and other
figures in them.
170 V.N. Berestovskij and I.G. Nikolaev

0.3. Riemannian Geometry and Generalized Riemannian Spaces. In classical

Riemannian geometry, when the existence of any necessary analytic properties
of a Riemannian metric is implied, a series of naturally occurring metrics falls
out of consideration. Among those not included in classical Riemannian geome-
try are the metrics considered in this article of curvature ~ K or of curvature
both ~ K and ~ K'. Apart from the fact that they arise in a natural geometric
way, we should mention that consideration of them is useful in the framework
of Riemannian geometry itself.

0.3.1. Problems of Riemannian Geometry in which Generalized Riemannian

Spaces Arise. First of all there are problems in which we need to consider
"limits" of Riemannian metrics (extremal problems, problems connected with
the concept of stability, and so on).
In this article we mention one of these problems - Berger's theorem (Berger
(1983» - connected with "loosening" the conditions ofthe rigidity theorem (15.2
and 6.1). In Berger's theorem the question is whether we can assert that if
the curvature of a compact simply-connected Riemannian manifold of even
dimension has the estimate i - e ~ K" ~ 1 (in the rigidity theorem i ~ K" ~ 1),
then if the manifold is not homeomorphic to a sphere, it is diffeomorphic to
one of the symmetric spaces of rank one (in the rigidity theorem isometry was
asserted).
The method by which Berger proved this theorem is noteworthy. The proof
of the theorem reduces to the investigation of the "limit" space for a sequence
of Riemannian manifolds with estimates for the curvature i-en ~ K" ~ 1,
where en -+ 0 as n -+ 00. Using the specific character of the situation, one can
prove that the limit space (and this is precisely a space of curvature ~ 1 and ~ i
in the sense of Aleksandrov) is a COO-smooth Riemannian manifold, and so
the rigidity theorem can be applied to it, from which the required result is
obtained.
We mention that in this theorem the crucial step is to use a theorem proved
by Gromov on the compactness of a class of Riemannian manifolds (Gromov
(1981».
We observe that such a use of generalized Riemannian spaces recalls the
situation in differential equations when we first turn to a "generalized solution",
and then establish that it is a classical solution.

0.3.2. Generalized Riemannian Spaces and the "Closure" of Classes of Rie-

mannian Manifolds with Bounded Curvatures. The set of Riemannian metrics
on a given manifold forms a metric space (we can also consider the space of
metrics specified on different manifolds; see § 15). This space is not complete - its
completion may be rather poor. However, under natural restrictions, namely if
bounds for the sectional curvatures K' ~ K" ~ K (K', K = const) are satisfied,
and for the radius of injectivity of the manifold we have i(IDl) ~ const > 0, the
completion is obtained by adjoining to the Riemannian manifolds under consid-
eration Aleksandrov spaces of curvature ~ K and ~ K'.
 II. Multidimensional Generalized Riemannian Spaces 171

In exactly the same way we can consider classes of Riemannian manifolds

with upper bounds on the sectional curvatures, Ka ~ K, and a lower bound on
the radius of injectivity. The completion of such spaces leads to spaces of curva-
ture ~ K.
We emphasize that in the first case it has been proved that the completion
consists exactly of spaces of curvature both ~ K and ~ K' (more details about
this are given in 0.4).

0.3.3. Synthetic Description of Riemannian Geometry. Spaces of curvature

both ~ K and ~ K' give a synthetic description of the completion of the class of
Riemannian manifolds with bounds on the sectional curvatures K' ~ Ka ~ K.
The problem of a synthetic coordinate-free description of Riemannian geometry
is also very important.
In order to obtain such a description, a new concept is necessary for metric
spaces that is an analogue of the sectional curvature of a Riemannian manifold.
The corresponding concept ("non-isotropic Riemannian curvature" in § 17 of
this article) is copied from Riemannian geometry by means of formula (0.1). The
role of plane element is played by a pair of "directions" of the metric space,
which can be defined in the most general case (see § 3).
Briefly, the most important axioms imposed on a metric space so that it is a
e 2 -smooth Riemannian manifold are that at each point of the metric space for
arbitrary pairs of directions at this point (provided that the angle between them
is greater than zero and less than n) there should exist a non-isotropic Rieman-
nian curvature and it should have a "good" modulus of continuity (for example,
it satisfies a Holder condition; see § 18). In the definition of the "continuity" of
the non-isotropic Riemannian curvature we need to introduce a function that
specifies the distances between the directions at different points of the metric
space (see § 16).
It is curious that in this way we can give the synthetic axiomatics of em-smooth
(m = 2, 3, 4, ... ) and thereby e oo -smooth Riemannian manifolds (see § 18).
We also consider the isotropic case when the "curvature" of a metric space
does not depend on the directions. Such a curvature, defined in the metric space
in accordance with (0.1), we call an isotropic Riemannian curvature (§ 17).
We note that the isotropic Riemannian curvature that we have introduced
coincides with the Wald curvature, well known in distance geometry, which is
defined in terms of the embeddability into a two-dimensional space of constant
curvature of quadruples of points of a metric space that contain "linear" triples
of points.
Metric spaces at each point of which there is an isotropic Riemannian curva-
ture we shall call isotropic metric spaces. Of course, it makes sense to consider
the very concept of isotropy only in the multidimensional case.
The main result for isotropic metric spaces is the generalization of a well-
known theorem of Schur: under additional assumptions, which are satisfied if
the metric space under consideration is a manifold of dimension greater than
two, an isotropic metric space is isometric to a space of constant curvature.
172 V.N. Berestovskij and I.G. Nikolaev

0.4. A Brief Characterization of the Article by Chapters. In the first chapter

we describe the main concepts of intrinsic geometry: intrinsic metric, shortest
curve, angle, triangle, and excess of a triangle. Here we give the construction of
the space tangent to an arbitrary metric space.
The main demands on the concepts that we introduce are that they should be
associated with an arbitrary metric space and that in the Riemannian case they
should lead to standard concepts.
Chapter 2 is devoted to the study of spaces of curvature ~ K. At the end of
the chapter we consider spaces of curvature both ~ K and ~ K', and we also
discuss the case when the curvature is only bounded below.
The method by which the spaces under consideration are investigated in this
chapter is purely synthetic.
In Chapter 2 we discuss "boundedness of curvature" in some detail.
We mention different (but equivalent) points of view on the definition of
boundedness of curvature of a metric space and obtain consequences that imply
the boundedness of curvature of a metric space. In the main these consequences
are described by properties that are in common with Riemannian spaces:
local uniqueness of the shortest curve joining a given pair of points, convex-
ity of a small ball, the existence of an angle between shortest curves, and
so on.
As an illustration of the methods developed we mention the solution of Pla-
teau's problem on the existence of a surface of minimal area and establish an
isoperimetric inequality that generalizes Carleman's well-known inequality to
the case of spaces with curvature bounded above.
Without dwelling on these results in detail, we give the two most important
assertions for spaces of curvature ~ K, on which (and on the analogue of them
for spaces of curvature both ~ K and ~ K') is based the whole synthetic theory
of the generalized Riemannian spaces considered in this article and which, prop-
erly speaking, make possible the study of these spaces.
The first is Aleksandrov's theorem of comparison of the angles of a triangle:
For an arbitrary sufficiently small triangle in a space of curvature ~ Keach
of its upper angles has the bound

where ilK is the corresponding angle in a triangle with the same lengths of sides
as the original but on a surface of constant curvature K.
The second is the property of K-concavity:
Let Land M be two shortest curves in a small domain of a space of cur-
vature ~ K, starting from a point O. Consider points X and Y lying on the
shortest curves Land M respectively. We denote by ytM(X, y) (x = OX, y = OY)
the angle of a triangle on a surface of constant curvature K with sides x, y, z
(z = X Y), lying opposite the side of length z.
K-concavity means that for any two such shortest curves Land M (lying in a
small domain of the space) the angle ytM(X, y) is a non-decreasing function of x
andy.
 II. Multidimensional Generalized Riemannian Spaces 173

We observe that formally the property of K-concavity follows easily from the
theorem on comparison of angles. Its importance, however, is based on the
following three features: firstly, K-concavity enables us to characterize in a new
way the upper boundedness of curvature, including the case of a Riemannian
manifold; secondly, most of the properties of spaces of curvature ~ K have
K-concavity as their basis; thirdly, there are situations (for example, the case
of a space of curvature ~ K' in § 11) when the analogues of K-concavity
(K'-convexity) do not follow only from the theorem on comparison of angles
(but, for example, for spaces of curvature ~ K' significant results are based on
K' -convexity).
Chapter 3 is devoted to spaces of curvature both ~ K and ~ K'.
More significant assertions are obtained under the assumption that the space
of curvature ~ K and ~ K' under consideration is a topological manifold. It
is interesting that there is no necessity to assume in advance that the metric
space is a manifold. The condition of local compactness, local extendability of a
shortest curve (12.1) and bilateral boundedness of curvature already guarantee
that the space under consideration is a manifold (12.2.1). Such spaces we shall
call spaces with bounded curvature for short.
The central result proved for spaces of bounded curvature is the following:
1) A space with bounded curvature is a differentiable manifold; the intrinsic
metric of a space with bounded curvature is specified by means of a certain
metric tensor, specified on this manifold.
2) In a neighbourhood of each point of a space with bounded curvature we
can introduce a so-called harmonic coordinate system (for the definition see
14.1.3). Harmonic coordinate systems specify in a space with bounded curvature
an atlas of class C 3 ,1I, where for ex we can take an arbitrary number lying in the
interval (0, 1).
3) The components of a metric tensor in an arbitrary harmonic coordinate
system are continuous functions that have generalized second derivatives in the
sense of Sobolev (Sobolev (1950» that are summable with an arbitrarily large
degree. In particular, the first derivatives of the components of the metric tensor
satisfy a Holder condition with arbitrary ex E (0, 1).
Example 12.1 given in the text shows that we cannot expect two-fold con-
tinuous differentiability of the components of a metric tensor in a space with
bounded curvature.
The importance of the theorem we have proved (in the text it is Theorem
14.1) is that it gives information on the second derivatives of the metric tensor
of a space with bounded curvature. The enables us to state a theorem (it follows
from Theorem 14.1) which gives the possibility of regarding metrics of bounded
curvature as limits of Riemannian metrics (of class COO) with sectional curvatures
bounded in aggregate (see Theorem 15.1).
All this enables us to look on spaces with bounded curvature as Riemannian
spaces, but with "worse" differential properties of the metric tensor compared
with the generally accepted ones, and at the same time as spaces with metrics
that are the limits of smooth Riemannian metrics, where for them we preserve
174 V.N. Berestovskij and I.G. Nikolaev

the assertions of "smooth" Riemannian geometry that "depend on the maxi-

mum modulus of sectional curvatures"; compare with 0.3.2.
The investigation of spaces with bounded curvature in this chapter has mixed
synthetic-analytic character. The synthetic part, which is essentially used in the
proof of the main results, is related to parallel displacement, which is introduced
in Chapter 3 in a geometric way. Consideration of it has independent interest.
The analytic part is based on rather subtle results of the theory of functions with
generalized derivatives and the theory of elliptic equations.
In Chapter 4 it is a question of the synthetic description of Riemannian
geometry and of isotropic metric spaces. We have already discussed this in 0.3.3.
We only note that these results have a main theorem which asserts the existence
in a definite sense of "almost everywhere" non-isotropic Riemannian curvature.
More precisely, this theorem (in the text it is Theorem 17.1) asserts that there is
a set (!) E 9Jl of zero n-dimensional Hausdorff measure (n = dim M) such that for
each point P E 9Jl\ (!) and arbitrary "admissible" pair of directions e, (E D p 9Jl
the sectional curvature K,,(P), which is formally calculated from the metric
tensor at the point P in the direction of the two-dimensional element u of 9Jl p ,
e
specified by the bivector 1\ (, can also be calculated as the limit of the ratios
J(Tm)/u(Tm) for some sequence of triangles that contract to the point P "in the
direction of the pair (e, 0".

0.5. In what Sense Do the Stated Results Have Multidimensional Character?

The answer to this question is given by the theory of so-called two-dimensional
manifolds of bounded integral curvature, in which restrictions are imposed not
on the Gaussian curvature but the integral curvature (Aleksandrov and
Zalgaller (1962), Reshetnyak (1960a)) (the case of two-dimensional spaces for
which "curvature exists" at each point was considered by Aleksandrov in
Aleksandrov (1948)), since spaces of curvature ~ K and spaces of curva-
ture ~ K and ~ K' in the two-dimensional case have bounded integral curva-
ture in the sense of Aleksandrov.
For the multidimensional case we have not succeeded in introducing
the concept of a manifold of bounded integral curvature, which in the two-
dimensional case has much greater generality than a space of curvature ~ K.

0.6. Final Remarks on the Text. As a basis for writing the present article we
have taken the survey Aleksandrov, Berestovskij and Nikolaev (1986). However,
we have added sections with remarks and examples that comment on the mate-
rial set forth. In addition, in § 15 and Chapter 4 we place results not stated in
Aleksandrov, Berestovskij and Nikolaev (1986).
The contents of Chapters 1 and 2 reflect the fundamental work of
Aleksandrov. Chapter 3 follows the results of Berestovskij on the introduc-
tion of a Riemannian structure in spaces with bounded curvature (Berestovskij
(1975)) and of Nikolaev on the geometrical definition of parallel displacement
(Nikolaev (1980), (1983a)), and on the smoothness of the metric of a space with
bounded curvature (Nikolaev (1980), (1983b)).
 II. Multidimensional Generalized Riemannian Spaces 175

Chaptet 4 was written as a result of the work of Nikolaev, announced earlier

in Nikolaev (1987), (1989).
In conclusion, the authors express their gratitude to Yu.D. Burago and V.A.
Zalgaller for the attention they have shown to the work and their valuable advice.

Chapter 1
Basic Concepts Connected with the Intrinsic Metric

It is well known that such "Euclidean" concepts as straight line, angle, trian-
gle, and so on, make sense and are very useful in Riemannian geometry.
In this chapter we show how basic "Euclidean" concepts of this kind carry
over to the case of general metric spaces, and we describe their properties.
The main demands on the concepts introduced here are that they must admit
a definition in an arbitrary metric space and in the case of a Riemannian mani-
fold they must coincide with the corresponding concepts that exist there.
Henceforth the material of Chapter 1 will be a kind of "linguistic" basis for
the construction of the synthetic axiomatics of Riemannian geometry.

§ 1. Intrinsic Metric, Shortest Curve, Triangle, Angle,

Excess of a Triangle

1.1. Intrinsic Metric. We recall that in a Riemannian manifold IDl the lengths
of piecewise-differentiable curves are calculated by means of the metric tensor.
If IDl is connected, we can introduce a standard metric in the Riemannian mani-
fold IDl. The distance between points of IDl with respect to this metric is taken to
be equal to the infimum of the lengths of all piecewise-differentiable curves
L c IDl joining the points under consideration. The metric on IDl introduced in
this way will be intrinsic in accordance with the definition given below.
We now consider an arbitrary metric space with metric p. It will often be
denoted by (IDl, pl. For an arbitrary curve Lin IDl its length in the metric p is
defined:
m-l
lp(L) = sup L p(Xj , X j +1)'
j=l

where Xl, X 2 , ••• , Xm is an arbitrary sequence of points of L, numbered in the

order of their position on the curve, and the upper bound is taken over all such
sequences of points.
The metric p is called intrinsic if for any X, Y E IDl, p(X, Y) is equal to the
greatest lower bound of the lengths (measured in the metric p) of curves joining
X and Y.
176 V.N. Berestovskij and 1.0. Nikolaev

From the definition it follows, in particular, that rol is a metrically connected

space. This means that any two points of it are joined by a curve of finite length.

1.2. Shortest Curve. It is well known that a geodesic curve in a Riemannian

manifold is locally the shortest curve among all curves joining the same points.
In this connection the following definition is natural.
A curve L in a metric space (rol, p) joining points A, B E rol is called a shortest
curve if its length is equal to p(A, B). Both p(A, B) and the shortest curve with
ends A and B are often denoted by AB.
In what follows the concept of convexity turns out to be useful for us. A
subset V in (rol, p) is said to be convex if any two points of it are joined by a
shortest curve in rol and any such shortest curve lies in V.
We observe that often, in contrast to the definition of convexity that we
have given, the second condition is replaced by the condition that at least one
shortest curve joining points of the set lies in the set.

1.3. Triangle. As in Riemannian manifolds, in metric spaces we can consider

triangles.
A triangle T = ABC in a metric space (rol, p) (where A, B, C E rol) is a set
consisting of points of the shortest curves AB, BC, AC, called the sides of T. The
points A, B, C are called the vertices of T. The perimeter ofthe triangle is the sum
AB + BC + AC. A short notation for the triangle is LfABC.
Later on we define the measure of the difference between a triangle in a
metric space and a planar triangle with the same lengths of sides - the excess of
the triangle - which enables us to estimate the "distortion" of the metric space,
as is done in the case of Riemannian manifolds. However, it is sometimes conve-
nient to regard as the "model" space not only a Euclidean plane, but other
spaces of constant curvature. In this connection we introduce the following
notation. We denote by SK the Euclidean plane when K = 0, the Lobachevskij
plane of curvature K when K < 0, and a two-dimensional open hemisphere of
radius 11ft when K > 0; we denote the n-dimensional analogue of SK by SR:.
Let T = ABC be a triangle in a metric space; we denote by TK the triangle
AKBKC K on SK that has sides of the same length as T does (degeneracy of the
triangle TK into an interval is allowed). When K ~ 0 the triangle TK always
exists on account of the triangle inequality for the metric. When K > 0 it is
necessary to require that the perimeter of the triangle T is less than 21tlft.
When K > 0 it will be assumed everywhere in this chapter that this requirement
is satisfied.

1.4. Angle. In order to calculate the angle between geodesics of a Rieman-

nian manifold starting from a point 0 it is sufficient, as we know, to consider
their tangent vectors at 0 and by means of the metric tensor to calculate from
the well-known cosine formula the angles between these vectors. However, if we
take into account the "infinitesimal Euclidean property" of a Riemannian mani-
fold at a point, we can calculate the same quantity by starting only from the
 II. Multidimensional Generalized Riemannian Spaces 177

metric of the manifold. For this we can consider geodesic triangles, one vertex
of which is at 0 and the other two lie on geodesics starting from O. Next we
calculate from the sides of these triangles the angle at the vertex 0 by means of
the formulae of Euclidean geometry and proceed to the limit, contracting the
triangles to the point O. Now we give the definition of angle according to
Aleksandrov (Aleksandrov (1957b» in the metric space.
Let Land M be two curves in the metric space (IDl, p) that have common
starting point O. On Land M respectively we choose arbitrary points X and Y,
and let x = OX, Y = OY, z = XY. We denote by ytM(X, y) the angle opposite to
the side of length z in the triangle TK = OKXKyK on SK corresponding to the
triangle T = OX Y.

1.4.1. Upper Angle, Angle. The upper angle between the curves Land M is
by definition the quantity

x,)'~o

We observe that the upper angle between Land M always exists and does not
depend on K. We obtain the definition of the lower angle between curves if in
the definition given above the upper limit is replaced by the lower.
We shall say that there is an angle IX between Land M if iX = {! = IX.

1.4.2. Strong Upper and Lower Angles, Angle in the Strong Sense. A central
role in our presentation will be played by the upper angle and the angle. How-
ever, there are situations when such an approach to the definition of an angle
turns out to be unsatisfactory. Namely, in order to obtain a meaningful defini-
tion oflower boundedness of the curvature it is necessary to invoke the concepts
given in this subsection of strong lower angle and angle in the strong sense (see
4.2 and 11.3).
The strong upper angle between Land M is defined as the upper limit of
ytM(X, y) on condition that x or y tends to zero. In particular, we allow the
situation when x tends to zero but y is non-zero, or vice versa. Similarly we
define the strong lower angle (in the definition of the strong upper angle the
upper limit is replaced by the lower). The upper, lower and strong upper and
lower angles will be denoted by iX, ~, iX" ~, respectively. Obviously,

The existence of an angle in the strong sense IX, is determined in the same way
as the existence of an angle: ~, = iX,.
To conclude 1.4 we note again that in a Riemannian manifold there is an
angle between any two geodesics. The existence of an angle in the strong sense
can be guaranteed in the general case only for small geodesics (in contrast to iX
and ~, the strong lower angle ~, is not a local concept); see § 4, where the angles
we have introduced will be discussed in more detail. In an arbitrary metric
space, for shortest curves the angles IX and IX, may not exist.
178 V.N. Berestovskij and I.G. Nikolaev

1.5. Excess. We now introduce a quantity that characterizes the difference

between a triangle in a metric space and the corresponding planar triangle. We
need the following notation. If T = ABC is a triangle in a metric space, then the
angle ii at its vertex A is the upper angle between its sides AB and AC. We
denote by pand y the angles of AABC at the vertices Band C, and by a. K· PK, I'K
the angles of the corresponding triangle TK = AKBKCKat the vertices AK, BK, CK.
The K-area O'K(T) or O'K(AABC) of the triangle T is understood as the area of
the triangle TK; O'o(T) will be called the area of T and denoted by O'(T).
The absolute excess of the triangle T = ABC (or simply the excess) is the
quantity
~(T) = ii + p+ y - n.
We recall that by the Gauss-Bonnet theorem the excess and the K-area ofthe
triangle T on SK are connected by the equality
(j(T) = K· O'K(T).
It is also convenient to use the concept of the excess of a triangle with respect
to SK' or briefly the K-excess. The K-excess of a triangle T is the quantity
~K(T) = (ii + p+ y) - (a. K + PK + I'K)'
We mention the following equalities, which will be useful later:
~(T) = ~o(T), ~K(T) = ~(T) - (j(T K). (1.1)

§ 2. General Propositions about Upper Angles

The general propositions about upper angles stated in this section relate to
general metric spaces.
Proposition 2.1 (Aleksandrov (1948), Aleksandrov and Zalgaller (1962». Let
ii n , ii 13 , ii23 be the upper angles between curves L l , L 2 , L3 starting from one
point. Then

If Ll and L3 are branches of the same shortest curve, then ii13 = n. Therefore
the next result follows from Proposition 2.1.
Proposition 2.2. The sum of the upper adjacent angles is not less than n.
It turns out that for shortest curves the concepts of upper angle and strong
upper angle coincide.
Proposition 2.3 (Aleksandrov (1957b». The upper angle ii between two
shortest curves Land M starting from a common point is equal to the strong upper
angle between them, that is,
ii = lim yfM(X, y).
%-+0
 II. Multidimensional Generalized Riemannian Spaces 179

The proof is based on the following lemma.

Lemma (Aleksandrov (1951), (1957b)). For any number K

cos ytM(X, y) = (y - z)/x + e,

where e --+ 0 as x/y --+ O.
We note that, generally speaking, the strong lower angle does not coincide
with the lower angle.
Theorem 2.1, stated below, is a kind of foundation for the further develop-
ment of the theory. It enables us, under very general assumptions, to estimate
the difference between the upper angle in a triangle T = ABC and the corre-
sponding angle !XK in the triangle TK on SK' The only condition on T is that any
two points on its sides can be joined by a shortest curve.
Theorem 2.1 (Aleksandrov and Zalgaller (1962)). Let a be the upper angle at
the vertex A oj the triangle ABC, and !XK the corresponding angle in the triangle
on the K-plane with the same lengths oj sides (when K > 0 the perimeter oj AABC
must be less than 2n/J'K). We assume that any two points on the sides oj AABC
can be joined by a shortest curve. Then

where viA = sup{inf{bK(AAXY)}}; here the supremum is taken over all possible
points X E AB, Y E AC, and the inJimum is taken over all possible shortest curves
X Y joining X and Y.
The next theorem gives an estimate of the angles of a triangle that is in a
certain sense opposite to the estimate of Theorem 2.1.
Theorem 2.2 (Aleksandrov (1957b)). Let ABC be a triangle in some metric
space, any two points oj which can be joined by a unique shortest curve, where
between these shortest curves and the corresponding segments oj the sides there is
an angle in the strong sense. Suppose also that K' is an arbitrary number and ilK'
is the lower bound oj K'-excesses oj triangles AX Y with vertices X and Y on the
sides AB and AC oj the triangle ABC. Under these assumptions, Jor the angle !X
oj AABC at the vertex A and the corresponding angle !XK' oj the triangle A'B'C'
with the same lengths oj sides on the K'-plane we have

§ 3. The Space of Directions at a Point. K-Cone.

Tangent Space

In a Riemannian manifold 9Jl the space oj directions at a given point P E 9Jl

is naturally assumed to be the set of unit vectors touching the manifold at P
(that is, the unit sphere in 9Jl p ). At the same time the direction at a point P of a
180 V.N. Berestovskij and 1.0. Nikolaev

Riemannian manifold IDl can be represented as a set of paths in IDl beginning at

P and having initially a given unit velocity vector. This leads us to the definition
of the space of directions at a given point of a metric space in the sense of
Aleksandrov. Having the space of directions at a given point of a metric space,
we can easily construct the tangent cone to the metric space at the given point.
The concept of K-cone introduced here is auxiliary.

3.1. Direction (Aleksandrov and Zalgaller (1962), Aleksandrov (1957b». The

following definition of the space of directions is related to any metric space IDl.
Let us fix an arbitrary point P E IDl. A curve L c IDl starting from P has a
definite direction at this point if the upper angle that it makes with itself at P is
equal to zero. We shall also say that two curves Land M beginning at P have
the same direction at this point if the upper angle between them at P is zero (we
observe that in this case it follows from the inequality !Xll :::;; !Xl2 + !X21 that L
and M have a definite direction at P).
On the set of curves that have a definite direction at P we introduce an
equivalence relation by calling two curves equivalent if they have the same
direction at this point (by Proposition 2.1, the given relation is an equivalence
relation). Now we define the set of directions of IDl at P as the set of classes of
equivalent curves that have a direction at this point. The next result follows
easily from the triangle inequality for upper angles (Proposition 2.1).

Proposition 3.1. The directions at a given point form a metric space in which
the distance between directions is defined as the upper angle between them. Nota-
tion: (QpM, iX).

3.2. K-Cone (Aleksandrov, Berestovskij and Nikolaev (1986». At the basis of

the definition there lies (depending on K) the cosine theorem of hyperbolic,
Euclidean or spherical geometry.

3.2.1. K :::;; O. The K-cone CKIDl over the metric space IDl is homeomorphic to
the factor space of the space IDl x ~+ (~+ = [0, +00» obtained by contracting
the set IDl x 0 to one point 0, called the vertex ofthe cone CKIDl.If(X, t) E IDl x
~+, then [X, t] denotes the corresponding point in CKIDl. The metric in CKIDl is
introduced as follows:
If K = 0, then for X, Y E IDl, t, s E ~+

([ X ] [Y. ]) = {[t + S2 2ts cos p(X, if p(X, Y) :::;; n,

2 - y)]l/2
PCK ' t, ,s (t + s) if p(X, Y) > n.

In the case K < 0, for k = J -K, PcK([X, t], [Y, s]) is defined by the relations
k- l . arcosh(cosh kt· cosh ks - sinh kt· sinh ks' cos p(X, Y» if p(X, Y) :::;; n,
(t + s) if p(X, Y) > n.
 II. Multidimensional Generalized Riemannian Spaces 181

3.2.2. K > O. Topologically CK 9Jl is the factor space of the space 9Jl x
[0, n/ft] obtained by contracting the set 9Jl x 0 to one point, and the set
9Jl x n/ft to another (that is, suspension over 9Jl).
For k = ft, PcK([X, t], [Y, s]) is defined by the relations
k- 1 arccos(cos kt· cos ks+ sin kt· sin ks' cos p(X, Y)) if p(X, Y) ~ n.
min{(t + s), 2n/ft - (t + s)} if p(X, Y) > n.

3.3. Tangent Space (Aleksandrov, Berestovskij and Nikolaev (1986)). The

tangent space 9Jl p to the metric space 9Jl at a point P E 9Jl is by definition the
O-cone over the space of directions .Qp9Jl to 9Jl at the point P, that is,
9Jl p = Co .Qp9Jl·
The vertex of the O-cone is called the origin in 9Jlp. Sometimes 9Jlp is also
called a tangent cone.

§ 4. Remarks, Examples

4.1. Intrinsic Metric, Shortest Curve, Angles.

4.1.1. Extrinsic and Intrinsic Metric of a Surface. The concept of intrinsic
metric arose originally in the study of the metric properties of surfaces, where it
was particularly intuitive. The fact is that on a surface there is already a metric
- the extrinsic metric of the ambient space. However, it is natural to measure
the distance between given points of the surface along the surface, that is, to
understand it as the greatest lower bound of the lengths of all curves on the
surface that join given points, and this is the intrinsic metric. We should observe
that metrically disconnected spaces fall out of consideration, since on them
it is impossible to specify the intrinsic metric from an existing one. An exam-
ple of such a space is a cylinder with directrix in the form of a curve that
is not rectifiable on any part. We also observe that in the case of a metrically
connected space the topologies determined by the extrinsic metric and the in-
trinsic metric constructed from it may differ substantially. As an example we can
consider a cone with directrix in the form of a curve that is not rectifiable
on any part. From the viewpoint of its intrinsic metric this cone is isometric to
a continuum of isolated segments with one common end-point (Aleksandrov
and Zalgaller (1962)).

4.1.2. Shortest Curves in a Space with Intrinsic Metric. The example of an

open non-convex domain in En shows that, generally speaking, any two points
of a metric space with intrinsic metric are not necessarily joined by a shortest
curve. However, any two points of a compact metric space with intrinsic metric
are obviously joined by at least one shortest curve, in particular, if the metric
space with intrinsic metric is a manifold, then at least locally any two of its
182 V.N. Berestovskij and I.G. Nikolaev

points are joined by a shortest curve. In the general case the shortest curve
joining the points is not unique.
Example 4.1 (Busemann (1955». Suppose that on the set 1R2 = {(x, y) Ix, Y E IR}
we introduce the norm II(x, y)lll = Ixl + Iyl. Then any curve (x(t), y(t» in
(1R2, II II d, each "coordinate" of which x(t), y(t) varies monotonically, is a
shortest curve in (1R2, II 111). In particular, the curves y = x P , P > 0, in 1R2 are
shortest curves joining the points (0, 0) and (1, 1).

4.1.3. The Triangle Inequality for Lower Angles. The Existence of« and «•. In
(1R2, II 111) we consider the rays el = (t, t), e2 = (t, 0), e3 = (0, t), t ~ O. We denote
by a.jj the upper angle between the rays ej and ej, i, j = 1,2, 3. We denote by gjj,
a.fj, gfj the lower angle and the strong upper and lower angles between the
corresponding rays. A simple calculation shows that
a. 1 2 = a.13 = a.~2 = a.b = n/2; g12 = g13 = gh = gb = 0;
gh = g23 = a.23 = a.h = n.
Hence it is obvious that for lower angles (and strong lower angles) between the
shortest curves el> e2, e3 the triangle inequality if not satisfied (compare with
Proposition 2.1):
n = gh = g23 > g13 + g12 = gil + gh = O.
We also see that there is a strong angle between the shortest curves e2 and e3,
but no angle or strong angle between e1 and e2 or between e1 and e3 • Later in
Example 4.2 it will be shown that a strong angle may not exist even between
shortest curves on a sphere.

4.1.4. Strong lower angle. The upper and lower angles are local concepts. If
it is a question of the angle between shortest curves, then by Proposition 2.3 the
strong upper angle is equal to the upper angle, and so this is also a local concept.
However, this does not hold for the strong lower angle. In Example 4.3 given
below the lower angle at the vertex B is equal to n, and the strong lower angle
at this vertex is equal to t/I, where 0 < t/I < n/3. Thus, the definition of strong
lower angles has the defect that it is connected not with the movement of the
shortest curves Land M in a small neighbourhood of their origin 0, but with
the properties of whole parts of L or M. Moreover, in the general case the
shortest curve is not unique and the strong lower angle now describes some-
thing stronger than the angle between shortest curves (see Fig. 1). We also
observe that in two-dimensional manifolds of bounded curvature and even in
smooth manifolds, as Example 4.2 shows, the strong angle between certain
shortest curves may not exist. In this connection we note that there is another
definition of the strong lower angle g; (and correspondingly the strong angle ex;:
g; = a.;). This differs from the fact mentioned above that the limit is considered
not over all sequences of points X n, Y" such that Xn or Y" tends to 0, but only
over those such that as Xn -+ 0 there are shortest curves X n, Y" that converge to
 II. Multidimensional Generalized Riemannian Spaces 183

o B

Fig. 1 Fig. 2

part of the shortest curve M, and as y" --+ 0 they converge to part of the shortest
curve L (it is assumed that at least one such sequence exists).
Example 4.2 (Aleksandrov and Zalgaller (1962)). On a sphere we consider
two shortest curves starting from a point 0 and forming an angle qJ, 0 < qJ < n/2,
and the opposite ends of the shortest curves coincide at the point diametrically
opposite to O. Then obviously ~s = 0, but ~; = ~ = IX = IX; = qJ. Here between
the shortest curves under consideration there is an angle 0(; but not O(s'
We note that in two-dimensional manifolds of bounded curvature there is an
angle 0(; between shortest curves, but not O(s (Aleksandrov and Zalgaller (1962)).

4.2. An Assertion Completely Dual to Theorem 2.1, that is, the Corresponding
Lower Bound for the Lower Angle Does not Hold. Let us give the corresponding
example.
Example 4.3 (Aleksandrov and Zalgaller (1962)). On a plane we consider a
concave curvilinear triangle with angles at the vertices A, B, C equal to t/I, where
0< t/I < n/3 (see Fig. 2). We now consider the metric space consisting of the
three threads AB, BC, CA of equal length I, and infinitely many threads join-
ing points X E AB and Y E AC having the same length as the corresponding
shortest curve X Y in Fig. 2. The metric is determined from the lengths of curves.
Then in the resulting metric space, for example, the distance from A to a point
Z E X Y close to Y is equal to YZ + Y A, and ABC is a triangle. At its vertex A
we have ~ = t/I, 0(0 = n/3, ~ - 0(0 < O. However, for any triangle AXY we have
b(LlAXY) = (qJ + n + n) - n > O. Thus, the inequality ~ - 0(0 ~ inf bo(LlAXY),
X E AB, Y E AC, is impossible.
We observe that the lower bound can be obtained if we calculate the excesses
of the triangles AXY from the strong lower angles ~s (the same is true for strong
lower angles ~;) (Aleksandrov and Zalgaller (1962)).
Other definitions of angles occur in the literature. For this we refer the reader
to the appendix to Chapter 2 of Aleksandrov and Zalgaller (1962).

4.3. Tangent Space, Space of Directions. The concept of the tangent space
(cone) to an arbitrary metric space at a point P E IDl assumes that at P the
184 V.N. Berestovskij and I.G. Nikolaev

metrics IDl and IDl p coincide infinitesimally. However, in the general case this is
not so. Consider two shortest curves L, MeIDl with common origin P, and
points X E L, Y E M. Then

p = XY = (PX 2 + py2 - 2PX· Py· cos y2M(X, y»1/2.

In IDl p the corresponding distance p' is

p' = (PX 2 + py2 _ 2PX. PY'cos iX)1/2,

where iX is the upper angle between Land M. We see that p = p' + o(pjOj(X, Y»
if and only if there is an angle between Land M.

4.3.1. The space of directions of an arbitrary metric space is not necessarily

arranged more simply than the original space. Consider an arbitrary metric
space IDl of diameter less than n. Then obviously the space of directions of the
K-cone over IDl at the vertex of this cone is isometric to IDl: QoCKIDl = IDl.

4.3.2. In a normed vector space the existence of a direction of a curve at a

point does not always imply the existence of the corresponding half-tangent at
this point. A curve L in Euclidean space E" with origin at a point PEE" has a
definite direction at P if and only if there is a right derivative of it at P, that is,
in this case L has a right half-tangent at P. In an arbitrary normed space this is
not so.
Example 4.4. On a Euclidean plane we consider an orthogonal coordinate
system Oxy and the disc B: x 2 + y2 ~ 1. We denote by C(B) the space of contin-
uous functions on B with the usual sup-norm. It is well known that the map that
associates with an arbitrary point (xo, Yo) E B the function

f(xO,yo)(x, y) = [(x - x o)2 + (y - YO)2J1/ 2

is isometric. Consider in C(B) the image of the segment I: x = t, Y = 0, 0 ~ t ~ 1:
!,(x, y) = [(x - t)2 + y2]1/2.
Obviously the resulting curve in C(B) is a shortest curve, and so it has a definite
direction at the point !o(x, y) E C(B). However, the curve t H!, E C(B) does not
have a right derivative at t = 0, and so it does not have a right half-tangent. For
if it existed, then it could be calculated as a pointwise limit
drt.!,(x, Y)lt=o = lim {[(x - t)2 + y2]1/2 - (x 2 + y2)1/2}/t.
t-O+

It is easy to calculate that then drt.!,(x, Y)lt=o = -x/(x 2 + y2)1/2 when x 2 +

y2 # 0 and is equal to 1 when x 2 + y2 = O. However, the resulting function is
discontinuous at zero (that is, it does not belong to C(B» and so there cannot
be a right derivative in C(B).
We observe that a similar example can also be given in a finite-dimensional
normed space.
 II. Multidimensional Generalized Riemannian Spaces 185

Chapter 2
Spaces of Curvature ~ K (and ~ K')

In this chapter we shall consider metric spaces whose curvature is in a certain

sense either bounded above or bilaterally bounded. These spaces, namely spaces
of curvature ~ K and spaces of curvature both ~ K and ~ K', were introduced
by Aleksandrov (Aleksandrov (1951), (1957b)). As an example ofthe situation in
which spaces of curvature ~ K arise, we can consider metrics that are limits of
smooth Riemannian metrics with bound on the sectional curvatures K~ ~ K
(see 11.1.3).
A way of specifying restrictions on the curvature is based on a known ex-
pression for the sectional curvature of a Riemannian manifold in terms of the
excesses of geodesic triangles (see (0.1), (0.2)).
The upper boundedness of the curvature, that is, just the right inequality
being satisfied, enables us to prove the existence of many geometrical properties
that Riemannian manifolds have. We mention that the results presented in this
chapter have a purely synthetic character, and they have as their basis the angle
comparison theorem (Theorem 5.1) and the property of K-concavity (Theorem
5.2), which are satisfied in the general case in a small domain RK of a space
of curvature ~ K. At the end of the chapter we consider spaces of curvature
both ~ K and ~ K'.

§ 5. Spaces of Curvature::::;; K. The Domain RK and its

Basic Properties

5.1. Definition of a Space of Curvature ~ K. A space of curvature ~ K can

also be characterized by the fact that for each sufficiently small triangle Tits
K -excess bK(T) is non-positive. A space of curvature ~ K was originally defined
in this form in Aleksandrov (1951). Later, in § 7, we shall prove that the defini-
tion given at the beginning of the chapter and the one given here are equivalent.
Particular significance in what follows is played by a domain of space of curva-
ture ~ K in which the bounds for K-excesses of triangles mentioned above are
satisfied. We shall denote it by R K • Most of the results presented in this chapter
refer to the domain R K • A precise definition of a space of curvature ~ K is as
follows:
The domain RK is a metric space with the following properties:
a) any two points in RK can be joined by a shortest curve;
b) each triangle in RK has non-positive K-excess;
c) if K > 0, then the perimeter of each triangle in RK is less than 2n/..jK.
By a space of curvature ~ K we understand a metric space, each point of
which is contained in some neighbourhood of the original space, which is the
domain RK •
186 V.N. Berestovskij and I.G. Nikolaev

5.2. Basic Properties of the Domain R K •

5.2.1. The Angle Comparison Theorem for Triangles in R K • The next theorem
follows immediately from Theorem 2.1.

Theorem 5.1 (Aleksandrov (1957b». The upper angles ii, p, Y of an arbitrary

triangle T in RK are not greater than the corresponding angles aK' 13K' YK of the
triangle TK on SK'
In what follows, an important role is played by the following assertion of
elementary geometry.

Lemma 5.1 (Aleksandrov (1957b». Let Q = ABCD be a quadrangle on SK,

bounded by the triangle ABC (when K > 0 the perimeter of Q is less than
2n/JK). Then there is a triangle T on SK obtained from Q by rectifying the
polygonal line CDB, that is, a triangle with side lengths AB, AC, BD + DC. If Q
is not equal to T, then:
1. the angles of Q at the vertices A, B, C are less than the corresponding angles
ofT;
2. the area of Q is less than the area of T.
5.2.2. K-concavity (Aleksandrov (1957b». From Theorem 5.1 by means of
Lemma 5.1 we can obtain the following theorem.
Theorem 5.2 (Aleksandrov (1957b». For any two shortest curves Land M in
R K, starting from one point 0, the angle ytM(X, y) (see § 1) is a non-decreasing
function of x and y.
The proof of Theorem 5.2 consists in the following. Suppose that the shortest
curves Land M start from the point O. We take points Yand YI on M and a
point X on L so that OY = y, OYI = YI' OX = x,O < YI < y, x> 0 (see Fig. 3).
Let T = AOXY, TI = AOXY!> T2 = AXYI Y; TIK = AO'X'Y{, Tl = AX'Y{Y' be
triangles on SK corresponding to TI and T2 such that TIK and Tllie on opposite
sides of X' Y{ (when K > 0 the existence of such triangles follows from condition
c); see 5.1).
In consequence of Proposition 2.2 and Theorem 5.1, the angle of the quad-
rangle Q composed of the triangles TIK and Tl at the vertex Y{ is either equal to
n or reentrant, that is, Q is bounded by the triangle O'X'Y'. By rectifying the
polygonal line O'Y{ Y' we obtain a triangle TK. By Lemma 5.1 the angle of Qat
the vertex 0' (that is, ytM(X, YI» is not greater than the corresponding angle of
the triangle TK (that is, ytM(X, y», as required.

Fig. 3
 II. Multidimensional Generalized Riemannian Spaces 187

5.2.3. Consequences of K-Concavity.

Proposition 5.1 (Aleksandrov (1957b)). Let X and Y be points on the sides AB
and AC of the triangle T = ABC in R K, and X' and Y' the corresponding points
on the sides of the triangle TK = A'B'C' (that is, A'X' = AX, A'Y' = AY). Then
XY::::;X'Y'.
The next proposition, which follows immediately from the monotonicity of
the angle ytM(X, y), is particularly important for us. In Riemannian manifolds
locally between shortest curves there is an angle in the strong sense. However,
in Finsler spaces (even with infinitely differentiable metric tensor) the proposi-
tion stated below does not hold in any domain of the space (see 4.1.3).
Proposition 5.2 (Aleksandrov (1957b)). Between any two shortest curves in RK
starting from one point there is an angle in the strong sense.
We now state propositions that describe the local behaviour of shortest
curves in R K. It turns out that shortest curves in RK have a series of properties
common to the local properties of geodesics in Riemannian manifolds.
Proposition 5.3 (Aleksandrov (1957b)). In RK shortest curves depend continu-
ously on their ends.
Proof Consider the shortest curves AB and AnBn in R K, An ..... A, Bn ..... B. On
AB we take an arbitrary point C and on AnBn points Cn so that
ACjAB = AnCn/AnBn.
We now consider the shortest curve ABn (see condition a)) and take on it the
point Dn that divides ABn in the same ratio (see Fig. 4). Applying Proposition 5.1
to the triangle ABBn and taking into account that ACjAB = ADn/ABn = t, we
deduce that CDn ..... 0 as BBn ..... O. Similarly DnCn ..... 0 as AnA ..... O. Since CCn ::::;
CDn + DnCn, we obtain the required result.
Proposition 5.4 (Aleksandrov (1957b)). Any two points of RK are joined by a
unique shortest curve.
Proposition 5.5 (Aleksandrov, Berestovskii and Nikolaev (1986)). An open
ball in RK is convex (if its radius is less than rr./2jK when K > 0).
Bit
-- ..............
8

A
Fig. 4
188 V.N. Berestovskij and I.G. Nikolaev

5.3. The Domain PK (Reshetnyak (1960b). (1968). Nikolaev (1979». We give

a modification of the concept of the domain RK that occurs in the literature and
is useful in what follows. The domain PK is a metric space with the conditions a)
of 5.1 and
b/) for any triangle contained in PK whose perimeter is less than 2n/ft
when K > 0 the K-excess is non-positive.
Obviously the concepts of the domains PK and RK differ only when K > 0
(the closure of SK when K > 0 is PK but not RK)'

§ 6. The Operation of Gluing

In this section we shall describe the operation of gluing for spaces of curva-
ture ~ K along compact convex sets and give some examples.

6.1. Gluing of Metric Spaces with Intrinsic Metric. We consider only the
situation when the metric spaces that are glued contain isometric convex sets
and the gluing is carried out along these sets. The general case is quite compli-
cated and we do not discuss it.
Let (roll' pd and (rol 2• P2) be metric spaces with intrinsic metric. Vl S;;; roll
and V2 s;;; rol 2 isometric convex sets in these spaces. and ({): roll --+ rol 2 the corre-
sponding isometry. Consider the set roll U rol 2 and specify an equivalence rela-
tion on it. Namely.
X is equivalent to X if X E roll \ Vl •
Y is equivalent to Y if Y E rol 2 \ V2 •
Z is equivalent to (()(Z) ifZEVl ·
We denote by roll Utp rol 2 the set of classes of equivalent elements of
roll U rol 2 • We now introduce an intrinsic metric on roll Utp rol 2 :

X
Ptp ( •
Y) =
!
Pl(X, Y)
P2(X, Y)
if X, Y E roll'
if X, Y E rol 2,
inf {Pl(X, Z) + P2(Y, ({)(Z»}
ZeV,

where X, Y denote the points of roll U rol 2 contained in the corresponding

classes of equivalent elements X, YE roll Utp rol 2 •
The metric space with intrinsic metric (roll Utp rol 2 , Ptp) is the result of gluing
the metric spaces (roll' Pl) and (rol 2, P2) with intrinsic metric along the con-
vex sets VI and V2 by means of the isometry ({).

6.1.1. Gluing of Spaces of Curvature ~ K. It turns out that the result of gluing
spaces of curvature ~ K along isometric compact convex sets is a space of
curvature ~ K. More precisely, the following theorem holds.
 II. Multidimensional Generalized Riemannian Spaces 189

Theorem 6.1 (Reshetnyak (1960b». If the spaces IDl1 and IDl2' which are do-
mains of PK , are glued together along compact convex sets V1 s;;; IDl1' V2 s;;; IDl2 by
means of some isometry ({J, then the space IDl1 utp IDl2 that arises as a result of the
gluing is also a domain of PK •

6.1.2. Example. Let IDl1 be a closed convex set in SKI (for SKI see § 1), IDl2 a
closed convex set in SK2 ' where K 1, K2 ::s:;; K, and P E IDl1 and Q E IDl2 certain
points. Theorem 6.1 asserts that the space obtained by gluing IDl1 and IDl2 at
the points P and Q is a space of curvature ::s:;; K. Fig. 5 shows a space of curva-
ture ::s:;; 0 obtained by gluing two Euclidean triangles at the vertices 0' and 0".

6.1.3. K-Fan (Reshetnyak (1968». A K{an is a space of curvature ::s:;; K ob-

tained as a result of successive gluing of finitely many triangles of SK' In this
subsection, by a triangle we understand the corresponding convex domain of SK
bounded by a triangle in the original understanding of this term, and the trian-
gle may be degenerate.
Let 1i c SK' i = 1,2, ... , m, be a set of triangles. For them we introduce the
notation 1i = AjOjBj. We assume that in 1i and 1i+1 the sides OjBj and 0i+1Ai+1
have different lengths. We denote by ({Jj the isometry of OjBj onto OJ+1Aj+1
under which the point OJ goes over to 0i+1' Then the K-fan formed from the
triangles T1, ... , Tm is a metric space with intrinsic metric, glued together from
the 1i along OjBj and OJ+1 Ai+1' i = 1, ... , m - 1, that is, T1 Utpl T2 Utp2' •• utpm_1 Tm.
The resulting space is a space of curvature ::s:;; K. Obviously all the vertices OJ are
glued together at one point.
If we reject the condition ({Jj(Oj) = 0i+l in the definition given above, we
arrive at the concept of a generalized K{an. Figs. 6 and 7 show a K-fan and a
generalized K-fan respectively, glued together from four triangles.

§ 7. Equivalent Definitions of Upper Boundedness

of the Curvature

A definition of a space of curvature ::s:;; K was given in 5.1. In this section we

give different versions of the definition of upper boundednes of the curvature: we

A
Fig. 5 Fig. 6
190 V.N. Berestovskij and I.G. Nikolaev

Fig. 7

give the connection with the usual Riemannian definition of curvature, we state
a theorem about non-expanding maps in a space of curvature ~ K that is char-
acteristic for these spaces, and we give a definition of boundedness of curvature
from the positions of distance geometry. We note that these definitions in the
case of Riemannian manifolds describe the upper boundedness of the sectional
curvatures of a Riemannian manifold by the number K.

7.1. Conditions under which a Space of Curvature ~ K is the Domain R K •

In the definition of the domain RK we require that the K-excesses of arbi-
trary triangles are non-positive. It would be desirable that this condition should
be imposed only on small triangles. Naturally, this condition is necessary, but
generally speaking insufficient. For example, a circular cylinder in this sense has
non-positive curvature, and at the same time it is not the domain Ro, since some
pairs of points on it are joined by two shortest curves. If we require in addition
that a shortest curve depends continuously on its ends, we obtain conditions
that specify the domain R K •
Proposition 7.1 (Aleksandrov (1957b». Suppose that a domain G of a metric
space satisfies the following conditions:
1) G is convex, that is, any two points of G are joined by a shortest curve, and
any such shortest curve lies in G;
2) shortest curves in G depend continuously on their ends;
3) each point in G is contained in a neighbourhood in which the K -excess of
any triangle is non-positive;
4) when K > 0 the perimeter of any triangle in G is less than 2n/.jK.
Then G is the domain R K •

Let us give a plan of the proof. From the conditions of Proposition 7.1 it
follows that in G the conditions that determine the domain RK are locally satis-
fied, and so in small domains G the results of § 5 are valid. In particular, between
any two shortest curves lying in G the angle exists, and for small triangles the
angle comparison theorem for triangles is satisfied.
Now let T = ABC be an arbitrary triangle in the domian G. We prove that
each angle of T is not greater than the corresponding angle in the triangle TK
with the same lengths of sides. To this end we split T by shortest curves ADj into
"narrow" triangles 11 = ADjDi+l' as shown in Fig. 8. This is possible by virtue
of the conditions 1) and 2). In turn we split each "narrow" triangle 11 into small
 II. Multidimensional Generalized Riemannian Spaces 191

Fig. 8

triangles Tij = Ei-l,jEi.jEi-l.j+l (see Fig. 8) so that all the small triangles have
non-positive excess bK. On SK we consider the triangles 1if corresponding to 1ij.
By gluing together the 1if for fixed i in the order of adjoining the triangles 1ij
corresponding to them, we obtain a "narrow" generalized K-fan (6.1.3). We
observe that the sum of the angles adjoining one vertex (lying inside ADi or
ADi-d is not less than 7r. (Theorem 5.1, Proposition 2.2). Straightening the corre-
sponding reentrant angles of the K-fan, we deduce from Lemma 5.1, applied
successively under each straightening, that the angles of the "narrow" triangle 7;
are not greater than the angles of the corresponding triangle 7;K on SK' We
now glue together the K-fans A'B'D'l' ... , D~C'A' from the triangles 7;K. Again
straightening the reentrant angles at D~, D;, ... , D~ and using Lemma 5.1, we
obtain the required assertion.

7.2. Connection with the Riemannian Definition of Curvature. Let us estab-

lish the connection between the definition of a space of curvature ~ K and the
definition of sectional curvature, known from Riemannian geometry, as the
limit of the ratio of the excesses of geodesic triangles to their "areas".
Theorem 7.1 (Aleksandrov (1957b)). In order that a metric space should be a
space of curvature ~ K it is necessary and sufficient that the following conditions
are satisfied:
1) each point has a neighbourhood, any two points of which are joined by a
shortest curve;
2) for any sequence of triangles Tm contracted to a point

(7.1)
m"'OO

where b(Tm) is the excess of the triangle Tm, and u(Tm) is its area (see 1.5).
192 V.N. Berestovskij and I.G. Nikolaev

Remark. When u(Tm) = 0 it is assumed that c5(Tm) ~ 0, so (7.1) makes sense

in this case.
The proof of necessity is based on (1.1) and a known expression for the excess
of a triangle on SK in terms of its area (the Gauss-Bonnet theorem). The plan of
the proof of sufficiency is as follows. If (7.1) is satisfied, it is not difficult to show
that each point is contained in a neighbourhood R H £, where Il is any preas-
signed positive number. In R H £ the shortest curves depend continuously on
their ends (Proposition 5.3). It remains to apply Proposition 7.1 and the limiting
process as Il -+ o.
Starting from the geometrical meaning of the curvature of a Riemannian
manifold and Theorem 7.1, we obtain the following result.
Corollary 7.1. A Riemannian manifold with sectional curvature ~ K (at all
points in all two-dimensional directions) is a space of curvature ~ K.

7.3. Definition of Upper Boundedness of Curvature. In a space in which lo-

cally any two points are joined by a shortest curve, the condition c5K (T) ~ 0,
which is satisfied for sufficiently small triangles T, is equivalent to the angle
comparison theorem for triangles (that is, Theorem 5.1) for sufficiently small
triangles or the property of K-concavity in the small (5.2.2). Thus, each of these
properties can be adopted as the definition of upper boundedness of the curva-
ture of the space.
7.4. Non-Expanding Maps in Spaces of Curvature ~ K. Let (roll' pd and
(rol 2, P2) be metric spaces. We say that a map qJ: roll -+ rol 2 is non-expanding
(Reshetnyak (1968)) if
P2(qJ(X), qJ(Y» ~ Pl(X, Y), X, Y E roll.
qJ maps a rectifiable curve r l croll equilengthwise into a rectifiable curve
r2 c rol 2 if it takes any arc of r l into an arc of r2 ofthe same length.
A convex domain V on SK with bounding curve L majorizes a rectifiable
curve r in the metric space rol (it is assumed that the length of r is equal to the
length of L) if there is a non-expanding map of V into rol that maps L equi-
lengthwise onto r.
The following theorem holds for non-expanding maps in a space of
curvature ~ K.
Theorem 7.2 (Reshetnyak (1968». For any closed rectifiable curve r in PK
whose length is less than 21t/JK when K > 0 there is a convex domain on SK that
majorizes the curve r.
Corollary 7.2. In order that a metric space should be a space of curvature ~ K
it is necessary and sufficient that condition 1) of Theorem 7.1 should be satisfied
and that
2') for each SUfficiently small triangle T there is a convex domain on SK that
majorizes T (obviously the boundary of this domain is a triangle with the same
lengths of sides as T has).
 II. Multidimensional Generalized Riemannian Spaces 193

Theorem 7.2 is proved as follows. The condition on the length of the curve r
enables us to span on r a ruled surface fI' consisting of shortest curves OX
°
joining a fixed point E r and points X E r. In the curve r there is inscribed
a sequence of polygonal lines I',. = OX 1 X 2 ... Xm.,O, n = 1, 2, ... , that con-
verges to r. We then construct an "almost" non-expanding map "'" of the K-fan
d/J" = OX 1 X 2 ... Xmn into IDl, under which the contour of the fan d/I" is mapped
equilengthwise onto the polygonal line 1',.. By means of Lemma 5.1 there is
established the existence of a majorizing map cp" of the convex domain v" in SK
onto the fan d/I". The composition "'" 0 cp" "almost" majorizes the polygonal line
1',.. There is established the existence of a subsequence "'" 0 cp" that converges to
some map, majorizing r, of the convex domain V s;: SK onto the ruled surface fI'.

7.5. Boundedness of the Curvature from the Viewpoint of Distance Geome-

try. There is a connection between the theory of generalized Riemannian spaces
and the so-called distance geometry (Blumenthal (1970». Let us give the neces-
sary definitions.
We shall say that in a metric space (IDl, p) a point B lies between points A and
C if
p(A, C) = p(A, B) + p(B, C), p(A, B), p(B, C) > O.
A quadruple of distinct points in the metric space is said to be exceptional
if one of these points lies between distinct pairs of points of the quadruple,
but the quadruple does not have rectilinear arrangement, that is, it is not em-
bedded isometrically into the number line. Otherwise the quadruple is said to be
ordinary. Let us explain what we have said by an example. We turn to the
example of 6.1.2 (see Fig. 5). In this metric space, obtained by gluing together
two planar triangles at the vertices, the quadruple of points A, 0, C, D is obvi-
ously exceptional.
The next theorem characterizes the boundedness of curvature in terms of the
isometric embeddability of quadruples in the model space s1;.
Theorem 7.3 (Berestovksij (1986». A metric space is a space of curvature ~ K
if and only if it satisfies condition 1) of Theorem 7.1 and
2") each point has a neighbourhood d/I such that each ordinary quadruple of
points of d/J is embedded isometrically in s1;. (for the definition of s1;. see 1.3) with
K' ~ K, depending on the chosen quadruple.
We note that from the isometric embeddability of ordinary quadruples in s1;
there easily follows the property of K-concavity in d/I.

§ 8. Space of Directions, Tangent Space at a Point of a

Space of Curvature ~ K

The definitions of a space of directions and a tangent space were given in

3.1 and 3.3. In this section we discuss the properties of a space of directions
194 V.N. Berestovskij and I.G. Nikolaev

and a tangent space at an arbitrary point of a space of curvature ~ K. In order

to obtain significant results it is necessary to make additional assumptions.
These assumptions are satisfied, for example, in the case when the space of
curvature ~ K is a topological manifold. The main results of this section are
that the space of directions at a point of a space of curvature ~ K is a space
of curvature ~ 1, and the corresponding tangent space at the point has
curvature ~ 0 in the sense of Aleksandrov.

8.1. Conditions under which a Shortest Curve Goes out in each Direction. The
next proposition follows easily from the property of K-concavity in RK (see
5.2.2).
Proposition 8.1 (Aleksandrov (1957b). If in RK the shortest curves L n , M n ,
n = 1,2, ... , going out from a point P converge to shortest curves L, M, then the
angles a(Ln, Mn) converge to the angle a(L, M}. In particular, if Ln - L, then
a(Ln, L} -0.
Remark. In the case when the shortest curves Ln, Mn have starting point
Pn -# P and converge to shortest curves L, M respectively, we have
a(L, M} ~ lim a(Ln, Mn}.

From Proposition 8.1 it follows, in particular, that the directions specified by

shortest curves in RK with common starting point depend continuously on the
shortest curves.
We can now state a condition under which a shortest curve goes out in each
direction.
Proposition 8.2 (Aleksandrov (1957b). If a point P of a space of curvature
~ K has a compact neighbourhood and each shortest curve PX can be extended to
a shortest curve PX' of length r (where r is a fixed positive number), then a shortest
curve goes out in each direction at P. The space of directions at P is compact.
Remark. The example of a closed disc on a Euclidean plane that is a domain
Ro with P on its bounding circle shows that the condition of extendability of the
shortest curves PX is essential.

8.1.1. Plan of the Proof of Proposition 8.2. Suppose a curve has a definite
directio'n at a point P. From the existence of a direction (for the definition see
3.1) it follows that there is a sequence of shortest curves with starting point P
such that the angle between them tends to zero. Local compactness and extend-
ability of shortest curves enables us to go over to a sequence of shortest curves
that converge to a shortest curve different from the point P. Obviously the
original curve and the limiting shortest curve have the same direction.

8.1.2. The Case when the Point P Has a Neighbourhood Homeomorphic to a

Ball in a Finite-dimensional Euclidean Space. As the proposition stated below
 II. Multidimensional Generalized Riemannian Spaces 195

shows, the conditions of Proposition 8.2 are satisfied for topological manifolds
with metric curvature ~ K.
Proposition 8.3 (Aleksandrov (1957b)). If the point P is contained in a neigh-
bourhood IJIt that is a domain RK and is homeomorphic to a ball in En, and ro is the
distance from P to the boundary of 1JIt, then any shortest curve starting from P can
be extended to have length roo

8.2. Intrinsic Metric in !lp.

8.2.1. The Angle of the Cone of Directions. A curve in Q p is usually called a
cone of directions, and its length is called the angle of the cone of directions.

8.2.2. The Cone of Directions of a Surface Triangle. Let T = ABC be an

arbitrary triangle in RK • Let D, (0 ~ t ~ 1, D, E Be) be a parametrization of Be.
Consider all possible shortest curves AD" 0 ~ t ~ 1. The resulting set of shortest
curves forms a surface triangle, which we shall also denote by the letter T. We
denote by CT(t) the direction specified by the shortest curve AD,. The resulting
cone of directions CT is called the cone of directions of the surface triangle T.
The surface triangle T, with respect to the intrinsic metric induced from R K ,
is a space of curvature ~ K (see 9.3). We denote its angle at the vertex A by a.
We can prove that the angle of the cone CT is exactly equal to a. Hence it follows
(see Theorem 5.1) that a ~ aK. It was proved in Aleksandrov (1957b) that a = aK
if and only if the surface triangle T and the corresponding surface triangle TK
from SK are isometric.

8.2.3. Shortest Curves in !lp. Let e, ,E Q p be two arbitrary directions, the

angle a between which is less than n. Consider two shortest curves AB and AC,
e
starting from A and specifying the directions and, respectively. Let X E AB
and Y E AC be arbitrary points. By virtue of 8.2.2 the angle of the cone CT ,
where T = AAXY is arbitrarily small, is different from a. We make X and Y
tend to A. We obtain a sequence of cones whose angles are arbitrarily close to
the angle a of the triangle ABC at the vertex A. In the case when the conditions
of Proposition 8.2 (or 8.3) are satisfied, it is easy to prove that there is a limit
e
cone, which is obviously a shortest curve joining the directions and ,. From
Theorem 8.1 stated below it follows that there are no other shortest curves
e
joining and ,.

8.2.4. The Curvature of !lp.

Theorem 8.1 (Nikolaev (1978)). If a point P of a space of curvature ~ K
has a neighbourhood homeomorphic to a ball in En, n > 1, then the space of
directions QpWl at this point is a space of curvature ~ 1 (more precisely, any
domain in QpWl of diameter less than n is a domain P1 ).
Remark. In Theorem 8.1 the case n = 1 is excluded. In this case the space of
directions consists of two points.
196 V.N. Berestovskij and I.G. Nikolaev

8.3. Tangent Space. An immediate consequence of Theorem 8.1 is the follow-

ing theorem.

Theorem 8.2 (Aleksandrov, Berestovskij and Nikolaev (1986». If some neigh-

bourhood of a point P of a metric space IDl of curvature ~ K is a manifold of
finite dimension or the conditions of Proposition 8.2 are satisfied, then the tangent
space IDl p to IDl at the point P is a space of curvature ~ 0 (more precisely, IDl p is a
domain Ro).

§ 9. Surfaces and their Areas

A space of curvature ~ K can have a "rather poor" structure in compari-

son with a Riemannian manifold. Moreover, the situation is possible in which
this space is not even a topological manifold (see 6.1.2). Nevertheless, direct
synthetic methods enable us to solve interesting geometrical problems even in
spaces of curvature ~ K. In this section we touch on the following ones: the area
of a surface, Plateau's problem, and the "extrinsic" geometry of "ruled" surfaces
in R K •

9.1. The Definition of the Area of a Surface (Aleksandrov (1957b), Nikolaev

(1979». We denote by B a closed unit disc on a Euclidean plane. We understand
a parametrized surface f in the metric space IDl as any continuous map of the
disc B into IDt
By a non-parametrized surface we understand the image of a parametrized
surface (Aleksandrov (1957b».
A triangulation of the disc B is by definition a triangulation of an arbitrary
polygon inscribed in B. Suppose we are given a triangulation of B, that is, a
splitting into triangles of a polygon inscribed in B; we denote their vertices by
Ai, i = 1, 2, ... , I. Let f be a surface in PK • With each point Ai we associate a
point Ai E M, i = 1, 2, ... , I, where the points Ai and ~ coincide if and only if
the points f(Ai) and f(Aj) coincide. We now join the points Ai by shortest curves
in the same order as the vertices of triangulation of the disc are joined. We thus
obtain a set of triangles in PK , which we call a complex of the surface J. We
call the points Ai the vertices of the complex. We say that a sequence {4>",} of
complexes belongs to 4>(f) or approximates the surface f if the maximum of
the distances between Ai and f(Ai) and the largest side of the triangles tends to
zero with the growth of m.
If 4> is a complex, then as the area of 4> we take the number u(4)), which is
equal to the sum of the areas of Euclidean triangles with the same lengths of
sides as for the triangles of the complex 4>. Now the area u(J) of the surface f is
defined like the area of a surface in Euclidean space:
 II. Multidimensional Generalized Riemannian Spaces 197

This definition was given in Nikolaev (1979) and modifies the definition ofthe
area in Aleksandrov (1957b). In the case of Euclidean space the definition we
have given turns into the definition of area according to Lebesgue (see Cesari
(1956».
If F is a non-parametrized surface, then by the definition in Aleksandrov
(1957b)
a(F) = inf{a(f)}, f(B) = F.

9.2. Properties of Area. As in the case of Lebesgue area, for surfaces in

Euclidean space we can prove the following properties of area:
1) There is a sequence of complexes that approximate a given surface in PK ,
the limit of whose areas is the area of the given surface.
2) (Semicontinuity). If the sequence of surfaces {1m} in PK converges uni-
formly to the surface f, then
a(f) ~ lim a(fm).
3) (Kolmogorov's principle). If p is a non-expanding ,map from one domain
PK to another, and f is a surface in the first domain, then
a(p 0 f) ~ a(f).

9.3. Ruled Surfaces in R K • The general definition was given in Aleksandrov

(1957a). Here we consider a special case.
Let rbe a closed curve in RK (whose length when K > 0 is less than 27t/..jK),
C the bounding circle of the disc B, and f: C -+ RK a parametrization of the
curve r. Let 0 be an arbitrary point on C. Consider the surface whose para-
metrization f is specified as follows: for X E B, lying on the interval 0 Y, Y E C,
f(X) lies on the shortest curve O'Y' (0' = f(O), Y' = f(Y», and
O'f(X): O'Y' = OX: OY.
Because of the condition on the length of the curve r, O'Y' depends continu-
ously on Y' E rand f is a parametrized surface. In this case we shall say that a
ruled surface with vertex at 0' is spanned on f In the general case a ruled
surface is a surface formed by shortest curves. We note that in the case when r
r r
is a triangle, the surface spanned on at a vertex of is a surface triangle (see
8.2.2). The next theorem extends to spaces of curvature ~ K the known "extrin-
sic" geometrical properties of ruled surfaces in Riemannian manifolds.
Theorem 9.1 (Aleksandrov (1957a». A ruled surface in a space of curvature
~ K is itself a space of curvature ~ K in the sense of its intrinsic metric induced
from the space (so long as it has an intrinsic metric at all; see the corresponding
example in 4.1.1).
In other words, the "extrinsic curvature" of a ruled surface is non-positive.
Theorem 9.1 was proved in Aleksandrov (1967a) by the method of approximat-
ing the ruled surface by finite sequences of its generators and constructing the
geometry of finite sequences of intervals.
198 V.N. Berestovskij and I.G. Nikolaev

Here we mention the following application of the area of a surface triangle.

Proposition 9.1 (Aleksandrov (1957b». The area of a surface triangle Tin RK
is not greater than the area of the corresponding triangle TK on SK and is equal to
it if and only if T and TK are isometric.

9.4. Isoperimetric Inequality. From Proposition 9.1 we can derive the follow-
ing theorem.
Theorem 9.2 (Aleksandrov (1957b». The area of a ruled surface spanned on a
closed rectifiable curve r in RK (the length of r in the case K > 0 is less than
2n/ JK) with vertex at a point on r is not greater than the area of a disc on SK
with length of circle equal to the length of r. Equality holds only when the disc and
the ruled surface are isometric.

9.5. Plateau's Problem. Let rbe a closed curve in PK • We denote by S(F) the
set of those parametrized surfaces in PK whose restriction to the boundary of the
disc B is a closed curve equivalent in the sense of Frechet to the curve r. We
introduce the notation
a(F) = inf{u(fn, f E S(F).
Plateau's problem is posed as follows: in the class S(F) is there a surface
whose area is equal to a(F)? (A minimal surface.)
Theorem 9.3 (Nikolaev (1979». Suppose that PK is finitely compact (that is, in
PK each bounded closed set is compact), its diameter is less than n/2JK when
K > 0, and r is a closed Jordan curve in PK • Then there is an f E S(r) such that
u(f) = a(F).
Remark. By Theorem 9.2, a(F) < +00.

9.5.1. Plan of the Proof of Theorem 9.3.

The proof relies on an analogue of the Busemann-Feller theorem (about
projection onto a convex body in Euclidean space) and the lemma based on it
about the cutting of "crusts". The latter consists in the fact that from each
surface of a "minimizing" sequence of surfaces by means of "projection" onto
a ball in PK we "cut off" those "crusts" that interfere with the equicontinui-
ty of the surfaces. The area of each surface is not increased, by virtue of the
Kolmogrov principle and the generalized Busemann-Feller theorem. Applying
Arzehl's theorem, we obtain the required surface.

§ 10. Spaces of Curvature both ~ K and ~ K'

Here we consider spaces of curvature ~ K for which we assume in addition

that their curvature ~ K', where K' ~ K. For such surfaces, of couse, all the
properties of spaces of curvature ~ K are satisfied, and the lower bound on the
 II. Multidimensional Generalized Riemannian Spaces 199

curvature implies the existence of properties that are in a certain sense opposite
to those that hold by virtue of the upper boundedness of the curvature. For
example, together with the property of K-concavity (5.2.2) there is the property
of K'-convexity (10.2.1) and so on. Section 10 is devoted to the presentation of
these properties. We should mention that the results presented here are a basis
for obtaining significantly stronger assertions for spaces of curvature both ~ K
and ~ K' in what follows. Namely, it turns out that these spaces "almost" coin-
cide with ordinary Riemannian manifolds.
In conclusion we point out that it is possible to define and investigate spaces
whose curvature is only bounded below. We shall touch on this question in § 11
(11.3).

10.1. Definition of a Space of Curvature both ~ K and ~ 1('. Let K and K'
be real numbers with K' ~ K. A metric space is called a space of curvature
both ~ K and ~ K' if each point of it has a neighbourhood G for which the
following conditions are satisfied:
a) any two points of G are joined by a shortest curve, and this shortest curve
lies in G;
b) for any triangle Tin G its K- and K'-excesses have the bounds

bK(T) ~ 0, bdT) ~ 0;
c) when K > 0 the perimeter of each triangle of G is less than 2rt/JK.
Any neighbourhood of a metric space for which conditions a), b), c) are
satisfied is called a domain RK'.K'
Remark. In b) the lower boundedness of the curvature should have been
defined in terms of strong lower angles, and not upper angles (see 11.3). How-
ever, in the case that we consider this is not important, because from the upper
boundedness of the curvature there follows the existence of an angle in the
strong sense (see Proposition 5.2) and so the upper angle coincides with the
strong lower angle.

10.2. Basic Properties of a Domain RK'.K' A domain RK'.K is, in particular, a

domain R K, and the results of § 5 are satisfied for it. In particular, in RK'.K
between any shortest curves there is a strong angle. The angle comparison
theorem in RK'.K follows immediately from Theorem 2.2.
Theorem 10.1 (Aleksandrov (1957b)). The angle (X of any triangle Tin RK'.K
has the bounds

where (XK' and (XK are the corresponding angles of the triangles T K' and TK on SK'
and SK respectively.
Using Theorem 10.1, we can prove the following lemma:
200 V.N. Berestovskij and I.G. Nikolaev

Lemma 10.1 (Berestovskij (1975)). For any three shortest curves in RK',K with
common starting point the sum of the angles between pairs of them does not exceed
2n.

It follows from Lemma 10.1 that in RK',K the sum of adjacent angles is not
greater than n. Taking account of Proposition 2.2, we obtain the following
lemma.

Lemma 10.2. In RK',K the sum of adjacent angles is equal to n.

10.2.1. K'-Convexity. The angle comparison theorem in RK',K and Lemma

10.2 enable us to prove the following theorem.
Theorem 10.2 (Aleksandrov (1957b». For any two shortest curves Land M
in RK',K' starting from a common point 0, the angle yt~(x, y) (see 1.4) is a non-
increasing function of x and y (that is, yt~(x, y) ~ yt;,,(xo, Yo) when x ~
x o, y ~ Yo)·
The proof of Theorem 10.2 is completely analogous to the proof of Theorem
5.2. We only need to mention the following:
We keep the notation of Theorem 5.2. Then the angles rx' and P' of the
triangles Tt and Tf at the vertex Y{ are not greater than the angles rx and Pin
Tl and T2 at the vertex Y1 (Theorem 10.1). By Lemma 10.2, n = rx + P~ rx' + P'
and the quadrangle Q formed from Tt and Tf has angle at Y{ not greater than
n (see Fig. 9). Straightening the polygonal line O'Y{ Y', we obtain the required
bound.

10.2.2. Consequences of K'-Convexity. We shall give just two direct conse-

quences of K' -convexity.
Proposition 10.1. Let X and Y be points on the sides AB and AC of the triangle
T = ABC in R K ,,K and X' and Y' the corresponding points on the sides of T K' =
A'B'C' (that is, A'X' = AX, A'Y' = AY). Then XY ~ X'Y'.

X'

yl
1

Fig. 9
 II. Multidimensional Generalized Riemannian Spaces 201

From Proposition 10.1 it follows that in RK'.K the non-overlapping condition

for shortest curves is satisfied:
Proposition 10.2. If AC :::> AB and ACl :::> AB are satisfied for shortest curves
in RK'.K' then either AC l £; AC or AC £; AC l •

10.3. Equivalent Definitions of Boundedness of Curvature. We give variants

of restrictions on the curvature, similar to those in § 7, for spaces of curvature
both ~ K and ~ K'.
Theorem 10.3. Let 9R be a metric space in which points are locally joined by
shortest curves. Then the following assertions are equivalent:
1) 9R is a space of curvature both ~ K and ~ K';
2) for each point P E 9R and each sequence of triangles Tm of 9R that contract
in an arbitrary way to P we have
K' ~ lim l5(Tm)/a(Tm) ~ lim l5(Tm)/a(Tm) ~ K
Tm~P Tm~P

(in the case when a(Tm) = 0 it is required that l5(Tm) = 0);

3) each point of 9R has a neighbourhood, any quadruple of points of which
is isometrically embedded in Slo with K' ~ Ko ~ K, depending on the chosen
quadruple;
4) in 9R the condition of K-concavity and K'-convexity is locally satisfied.
Remark. The proof of the equivalence of 1) and 3) is contained In
Berestovskij (1986).

10.3.1. Comparison of the Angles of Triangles in RK'.K with the Angles of

the Corresponding Euclidean Triangles (Aleksandrov, Berestovskij and Nikolaev
(1986)).
Proposition 10.3. For angles of the triangle Tin RK'.K we have the bound
10( - 0(0 I ~ J1." a(T), (10.1)
where Jl is a positive constant that depends on K' and K.

§ 11. Remarks, Examples

11.1. Spaces of Curvature ~ K as a Generalization of Riemannian Spaces.

By Corollary 7.1, any Riemannian manifold whose sectional curvatures at all
points in all two-dimensional directions have the bound Ka ~ K is a space of
curvature ~ K in the sense of Aleksandrov. At the same time, in the definition
of a space of curvature ~ K there are no requirements on the smoothness of the
metric, moreover, the space may not be a manifold; see the example in 6.1.2 (or
it may be a manifold, but infinite-dimensional). Despite this, spaces of curva-
ture ~ K have a number of important properties in common with Riemannian
202 V.N. Berestovskij and I.G. Nikolaev

spaces; a local angle comparison theorem for triangles, the existence of angles
between shortest curves, the local uniqueness of shortest curves, and so on. We
observe that shortest curves in a space of curvature ~ K may "ramify", that is,
coincide on a certain part, and then diverge. For example, the shortest curves
AOC and AOD in Fig. 5 behave in this way. This effect occurs also for shortest
curves on a cone with total angle at the vertex> 2n, passing through the vertex
of the cone. Such a cone is a space of curvature ~ O.

11.1.1. Spaces of Curvature ~ K and Finsler Spaces.

The following assertion was mentioned by Aleksandrov in Aleksandrov
(1957a), p. 7:
A Finsler manifold of curvature ~ K is a Riemannian manifold.
Thus, the essence of the restrictions on the curvature of a space of cur-
vature ~ K is not in the differential properties of the metric (the metric tensor
of a Finsler space may even be infinitely differentiable), but in its infinitesimal
properties.

11.1.2. The Domain R K and the Radius of Injectivity of a Riemannian Mani-

fold. We recall that the radius of injectivity i m(P) of a Riemannian manifold 9Jl
at a point P E 9Jl is the supremum of all b such that the map expp , contracted
onto the ball in 9Jlp . with centre at the origin and radius b, is injective. The
number i(9Jl), which is equal to inf{im(p)}, P E 9Jl, is called the radius of injec-
tivity of the manifold 9Jl.
Proposition 11.1. Let WI be a complete Riemannian manifold for which:
1) the radius of injectivity i(WI) is positive;
2) the sectional curvatures of WI at all points in all two-dimensional directions
have the bound Ka ~ K.
Then an open ball in 9Jl with centre at an arbitrary point P E WI and radius ro,
equal to i(9Jl)/4 when K ~ 0 and min{n/3.jK, i(9Jl)/4} when K > 0, is a domain
RK •
Remark. In the case when the curvature of 9Jl is additionally subject to the
restriction Ka ~ K', the ball mentioned in Proposition 11.1 is a domain RK',K'
The proof of Proposition 11.1 consists in the following.
Obviously any two points of B(P, 2ro) are joined by a unique shortest curve.
But then, in view of the completeness of 9Jl, the shortest curves with ends in
B(P, 2ro) depend continuously on their ends. Obviously the perimeter of any
triangle with vertices in B(P, ro) is less than 2n/.jK when K > 0, and any
shortest curve with ends in B(P, ro) lies in B(P, 2ro). By virtue of the conditions
on the curvature of WI and Proposition 7.1, we deduce that for each triangle
with vertices in B(P, ro) the angle comparison theorem holds for the triangle.
Hence B(P, ro) is a domain R K •
We mention that from some bounds on the curvature it is impossible to ob-
tain a lower bound for the radius of injectivity, as, for example, on an arbitrarily
 II. Multidimensional Generalized Riemannian Spaces 203

"narrow" torus we can specify a planar metric (K == 0). The requirements that
9Jl is simply-connected are also insufficient. Namely, there is an example of
a compact Riemannian manifold diffeomorphic to a three-dimensional sphere
for which all sectional curvatures lie within the limits 0 < K" ~ 1 and on which
there is a closed geodesic of length < 2n, that is, in the manifold there is a
triangle with perimeter less than 2n and with angles equal to n (Gromoll,
Klingenberg and Meyer (1968), § 6.4, Remark 4) and the angle comparison theo-
rem is not true in the large on 9Jl, that is, 9Jl is not R l'
However, if there are restrictions on the volume, the diameter and the curva-
ture of a compact Riemannian manifold, then a positive bound on the radius of
injectivity exists.
Proposition 11.2 (see in Peters (1986». For each compact Riemannian manifold
9Jl for which
1) the diameter of 9Jl is not greater than d;
2) for the sectional curvatures of 9Jl at all points in all two-dimensional direc-
tions we have IK"I ~ Co;
3) the volume of 9Jl has the bound Vol(9Jl) ~ Vo > 0, the radius of injectivity
i(9Jl) has the bound i(9Jl) ~ io, where io is a positive constant depending only on d,
Co, Vo and dim 9Jl.

11.1.3. Spaces of Curvature ~ K and Limits of Riemannian Metrics. In many

cases metrics of curvature ~ K in the sense of Aleksandrov arise as the limit of
Riemannian metrics with curvatures ~ K, for example, such is the metric of a
cone with total angle at the vertex> 2n. We give conditions under which the
limit of Riemannian metrics of curvature ~ K is a metric of curvature ~ K. Let
us first recall the following concepts.
We say that a Riemannian metric P is specified on a set 9Jl if the metric space
(9Jl, p) is isometric to some Riemannian manifold 9Jlp with respect to its stan-
dard intrinsic metric; see 1.1.
A space is said to be finitely compact if each set in it that is bounded in the
sense of the metric of this space has compact closure. For Riemannian mani-
folds completeness is equivalent to finite compactness.
Proposition 11.3. We assume that on the set 9Jl there are specified Riemannian
matrics Pm defined in 9Jl by "one and the same topology", and that the following
conditions are satisfied:
1) the metric space (9Jl, Pm) is finitely compact;
2) the radius of injectivity im (9Jl) of 9Jl with respect to the Riemannian metric
Pm has the bound
i m (9Jl) ~ io > 0;
3) the sectional curvatures of 9Jl pm at all points in all two-dimensional direc-
tions have the bound K~ ~ K;
4) the metrics Pm converge to the metric P uniformly on each compact set of 9Jl.
Then the limit space (9Jl, p) is a space of curvature ~ K in the sense of
Aleksandrov.
204 V.N. Berestovskij and I.G. Nikolaev

Remark. If in Proposition 11.3 we replace 3) by the condition K' ~ K': ~ K,

then in the limit we obtain a space of curvature both ~ K and ~ K'.

The proof of Proposition 11.3 consists in the following.

By Lemma 7 of Aleksandrov and Zalgaller (1962) (§ 1 of Chapter 4) the metric
p of the space (rol, p) is intrinsic and any two points in (rol, p) can be joined
by a shortest curve. By Proposition 11.1 the ball Bm(p, ro) £; (rol, Pm), where
ro = io/4 when K ~ 0 and min{i3/4, n/3JK} when K > 0, is a domain R K.
Hence the condition of K-concavity is satisfied in Bm(P, ro) (see 5.2.2). By means
of a limiting process we deduce that the property of K -concavity is satisfied for
the ball B(P, ro) in (rol, pl. But then B(P, ro) is a domain RK in (rol, p) (see 7.3).
Since the point P is arbitrary, we deduce that (rol, p) is a space of curva-
ture ~ K.
In conclusion we mention the following: those spaces of curvature ~ K that
are two-dimensional topological manifolds are special cases of the well studied
two-dimensional manifolds of bounded integral curvature (Aleksandrov and
Zalgaller (1962». From the results of that work (see also Part I of this volume)
it follows that the metric of these spaces is in turn the limit of smooth Rieman-
nian metrics of curvature ~ K m, where Km -+ K as m -+ 00. It is not known
whether a similar assertion is true for the multidimensional case.

11.2. Polyhedral Metrics. We can say that a metric is polyhedral if it is

obtained by "gluing" finitely many I-dimensional polyhedra in s:C along iso-
metric (1- I)-dimensional faces.
For two-dimensional polyhedral metrics a simple criterion holds: a poly-
hedron formed from triangles SK is a space of curvature ~ K if and only if the
sum of its angles at each vertex not lying on the boundary is ~ 2n. We give an
example showing that for multidimensional polyhedral metrics the situation is
not so simple.

11.2.1. One Example of a Three-dimensional Polyhedral Metric. We give a

slightly modified example from Ionin (1972).
In E3 we consider a set R' included between a tetrahedron S' = A'B'C'D' and
a tetrahedron S~ = A'IB~ C~D~, whose base lies inside the triangle A'B'C' and
whose vertex D~ is strictly inside S~, with intrinsic metric induced from E 3 • We
now consider the set R obtained as a result of "doubling" the set R' with intrin-
sic metric defined in a similar way, that is, we consider a second copy R" of
the space R' and "glue together" R' and R" along the congruent tetrahedra
D~A'IB~ C~ and DrA~BrCr (see Fig. 10).
We denote the glued vertices A~, B~, C~, D~ and A~, Br, cr, Dr by AI' B l ,
Cl , Dl respectively. Obviously R has the following properties:
1) R can be obtained by gluing finitely many tetrahedra in E 3 , that is, the
intrinsic metric of R is a three-dimensional polyhedral metric;
2) with respect to the intrinsic metric arising as a result of the gluing, R is
homeomorphic to a closed ball in E 3 ;
 II. Multidimensional Generalized Riemannian Spaces 205

D' nil

e"
A'

8'

Fig. 10

3) the sum of the dihedral angles at each of the edges A1D1, B1D1o C1D1 is at
least 2n, and the sum of the solid angles at each of the vertices D1 , A 10 B1 , C1 is
greater than the area of a unit sphere, that is, 4n.
However, R is not a space of curvature ~ 0 (and is not a space of curvature
~ K for any K) in the sense of Aleksandrov, since in an arbitrarily small neigh-
bourhood of the vertex D~ there are pairs of points that are joined by two
shortest curves.

ll.3. Spaces of Curvature ~ K'. We can define a space whose curvature is

bounded below: a space of curvature ~ K'. In order that the angle comparison
theorem should be fulfilled, the lower bound on the curvature must be imposed
in terms of strong lower angles (see Example 4.3): each point of IDl has a neigh-
bourhood G such that the strong lower excess of an arbitrary triangle of G is
non-negative: ~. + P. + Y. - (OCK' + PK' + yd ~ 0 (Aleksandrov and Berestovskij
(1984), Aleksandrov, Berestovskij and Nikolaev (1986».
As in Theorem 5 on p. 28 of Aleksandrov and Zalgaller (1962), we can prove
the inequalities ~ ~ (J,K', P ~ PK" Y ~ YK' for the lower angles ~, p, y and the
property of non-overlappIng of shortest curves (see Proposition 1(r.2)~ In Alek-
sandrov and Berestovskij (1984) and Aleksandrov, Berestovskij and Nikolaev
(1986) (§ 8) it was erroneously asserted that there is an angle in the strong sense
between shortest curves lying in G (see Example 4.2, and also Aleksandrov
(1948), pp. 55-56). Under certain additional conditions (for example, assuming
that in G the sum of lower adjacent angles does not exceed n) we can prove that
the metric of Gis K' -convex (for the definition see 10.2.1), so it folows that there
is an ordinary angle between arbitrary shortest curves lying in G. Condition
(A;) in Aleksandrov, Berestovskij and Nikolaev (1986), § 8, is actually equiva-
lent to the K' -convexity of the metric. It is easy to prove that conditions (A +)
and (At> are equivalent if we add to what we said above for G the condition that
206 V.N. Berestovskij and I.G. Nikolaev

shortest curves lying in G depend continuously on their ends. The authors do

not know whether the conditions imposed on G in addition to the conditions on
the excesses of triangles in G are essential.
11.3.1. Global Lower Bound on the Angles of Triangles in a Space of Curva-
ture ~ K'. As we saw above, a bound for the angles of a triangle (X ~ (XK and so
on is fulfilled in the general case only for sufficiently small triangles. It is very
important, however, to know conditions that guarantee the fulfilment of such a
bound for all triangles of a space of curvature ~ K'.
For two-dimensional metrized manifolds the corresponding assertion was
first obtained by Aleksandrov (Aleksandrov (1948». In particular, by Alek-
sandrov's theorem on complete convex surfaces that are spaces of curvature ~ 0
the bound (X ~ (xo is satisfied for angles of arbitrary triangles.
In the multidimensional case a generalization of Aleksandrov's theorem
holds only for smooth Riemannian manifolds. Namely, a theorem due to
Toponogov (Toponogov (1959) asserts that the required bound on the angles of
a triangle is satisfied in the large in complete simply-connected Riemannian
manifolds whose sectional curvatures have the bound K" ~ K' at all points for
all plane elements (J in the tangent space at the corresponding point.
We observe that the proposition that Toponogov's theorem can be extended
to the case of the spaces with bounded curvature considered in Chapter 3 is very
likely (for this see Theorem 15.1).1

Chapter 3
Spaces with Bounded Curvature

In Chapter 3 we continue the study of properties of spaces of curvature both

~ K and ~ K'. The main question of interest to us here is the comparison
of these spaces with classical Riemannian spaces. We naturally restrict our
consideration to that of metrics of curvature both ~ K and ~ K' specified on
finite-dimensional topological manifolds. For this we add to the axioms of a
space of curvature both ~ K and ~ K' the conditions of local compactness and
local extendability of a shortest curve (for the definition see 12.1). As we shall
show, these conditions are necessary and sufficient for a space of curvature both
~ K and ~ K' to be a topological manifold of finite dimension. Spaces of
curvature both ~ K and ~ K' with the properties stated above we shall call
spaces with bounded curvature.
The spaces with bounded curvature that we investigate turn out to be Rie-
mannian, but in a certain generalized sense in comparison with the generally
accepted point of view: in accordance with 1.1 their intrinsic metric is specified

1 After our article had been published, the general Toponogov theorem for spaces of curvature
bounded below appeared in a paper by Yu. Burago, M. Gromov and G. Perel'man (Russian Math.
Survey 47: 2 (1992), 1-58).
 II. Multidimensional Generalized Riemannian Spaces 207

by a metric tensor; however, the differential properties of this tensor, generally

speaking, are worse than those usually required in Riemannian geometry (in the
general case the metric tensor of a space with bounded curvature is not twice
continuously differentiable). Our main results are concerned with determining
the differential properties of the metric of spaces with bounded curvature. The
results obtained in this direction enable us to assert that the metric of a space with
bounded curvature is in a certain natural sense the limit of smooth Riemannian
metrics (for example, of class COO) whose sectional curvatures at all points in all
two-dimensional directions are bilaterally uniformly bounded.

§ 12. CO-Riemannian Structure in Spaces with

Bounded Curvature

In this section we show that, as in the case of classical Riemannian manifolds,

the intrinsic metric of a space with bounded curvature is specified by means of
a certain metric tensor. There will be no question of the differential properties of
this tensor. Here we just mention that the metric tensor of a space with bounded
curvature is continuous. Hence in a space with bounded curvature we specify
the structure of a CO -smooth Riemannian manifold.

12.1. Definition of a Space with Bounded Curvature. We shall define a space

with bounded curvature as a metric space 9Jl with intrinsic metric in which the
following axioms are satisfied:
a) 9Jl is a locally compact metric space;
b) the condition of local extendability of a shortest curve is satisfied in 9Jl: for
each point P E 9Jl some open ball B(P, rp) has the property that any shortest
curve XY with ends X, Y E B(P, rp) can be extended to a shortest curve Xl Yl in
9Jl for which X and Yare internal points;
c) the curvature of 9Jl is locally bounded in the sense of Aleksandrov: each
point P E 9Jl is contained in some neighbourhood o/J £; 9Jl that is a domain RK',K
(see 10.1) for some K', K (where K' :::; K) that depend on o/J (or equivalently
conditions 2) and 3) of Theorem 10.3 are satisfied).
Remark. From conditions a), b), c) it will follow (see Theorem 12.1) that a
space with bounded curvature is a topological manifold of finite dimension. We
note that conversely a space of curvature both :::; K and ~ K' that is a topologi-
cal manifold of finite dimension satisfies conditions a), b), c), that is, it is a space
with bounded curvature. Obviously in the verification we need only condition
b), whose fulfilment follows from Proposition 8.3.
Example 12.1 (Peters (1986)). Part of a circular cylinder in E3 closed by two
hemispheres, considered with the intrinsic metric induced from E 3 , is a space
with bounded curvature (see Fig. 11). The components of the metric tensor of
this space in some coordinate system are not twice continuously differentiable
functions (otherwise the space under consideration would have constant curva-
208 V.N. Berestovskij and I.G. Nikolaev

Fig. 11

ture). Moreover, it is easy to construct a coordinate system in which the compo-

nents of the metric tensor are differentiable and their first derivatives satisfy a
Lipschitz condition. Thus, the space with bounded curvature under consider-
ation is a Riemannian manifold of class Cl,l but not C2 •

12.2. The Tangent Space at a Point of a Space with Bounded Curvature. In 8.3
we mentioned that the curvature of the tangent space rolp at a point P of a space
of curvature ~ K that is a topological manifold is not greater than zero. For
spaces with bounded curvature, using in addition the lower boundedness of
their curvature, we can deduce that the curvature of rolp is liot less than zero.
But then rolp is a space of zero curvature. Thus we have the following result.
Proposition 12.1 (Berestovskij (1975». The tangent space at an arbitrary point
P of a space with bounded curvature is isometric to a finite-dimensional Euclidean
space.

12.2.1. Exponential Map. The results of 12.2 enable us to define an exponen-

tial map in spaces with bounded curvature.
Consider the ball B(P, r), P E rol, where r (0 < r < rp) is so small that the
closed ball B(P, r) is compact. Then we can define the map
expp: B(O, r) c rolp -.. B(P, r) £; rol
(here B(O, r) is the ball in the Euclidean space rolp of radius r with centre at the
origin), which associates with each element [D, t] of B(O, r), where DE Dprol,
o~ t ~ r (for the definition of a tangent element see 3.3) the end of the unique
shortest curve of length t that specifies the direction D (see Proposition 8.2).
The uniqueness of the shortest curve is guaranteed by the non-overlapping of
shortest curves (Proposition 10.2) and Theorem 10.1 (comparison of angles). By
part b) of the definition of a space with bounded curvature, any shortest curve
with origin at P can be extended at least to length rp > t. The map expp is
obviously a bijective map of the ball B(O, r) onto B(P, r). From the results of § 8
we can deduce that expp is a continuous map, and so by what we said above it
is a homeomorphism. We thus arrive at the following theorem.
 II. Multidimensional Generalized Riemannian Spaces 209

Theorem 12.1 (Berestovskij (1975)). In a space 9.Jl with bounded curvature all
the tangent spaces 9.Jl p , P E 9.Jl, are Euclidean spaces of the same finite dimension.
9.Jl is a topological manifold of the same dimension.
Remark. If in the definition of a space with bounded curvature we reject local
compactness, then we can assert that 9.Jl is an infinite-dimensional manifold
(Berestovskij (1981».

12.3. Introduction of C°-8mooth Riemannian Structure.

12.3.1. Statement of the Main Result
Theorem 12.2 (Berestovskij (1975». A space 9.Jl with bounded curvature is a
CO-smooth Riemannian manifold of finite dimension.
More precisely, Theorem 12.2 asserts the following.
1) In a neighbourhood of each point of 9.Jl we can construct a coordinate
system of special form - a distance coordinate system (12.3.2) - and the distance
coordinates specify on 9.Jl the structure of a Cl-smooth differentiable manifold.
2) On 9.Jl we can specify a symmetric positively defined tensor field g of type
(2,0) of class Co. In the distance coordinate system g is written as a matrix
(gij(X»i.j=I ..... ", whose coefficients are continuous. For the length of an arbitrary
Cl-smooth curve y we have the standard formula

I(y) = f: L.tl /
giiy(t»'ji(t)'jj(t)J 2 dt, (12.1)

where y(t), oc ~ t ~ p, is a parametrization of the curve y, and j is the field of

vectors tangent to y with respect to the parametrization y(t).
3) The distance between points P, Q E 9.Jl in the original metric coincides
with the greatest lower bound of the lengths of all Cl-smooth curves joining P
and Q, calculated from (12.1), that is, it is equal to the infimum of the integrals
in (12.1).
Remark 1. In fact the distance coordinates specify on 9.Jl the structure of a
CI.I-smooth differentiable manifold, and the components ofthe metric tensor in
the distance coordinate system gij(x) satisfy a Lipschitz condition: gij E CO. l , see
Theorem 13.2.

12.3.2. Distance Coordinates and the Metric Tensor. We denote by n the

dimension of a space 9.Jl with bounded curvature. We shall say that points
Po, PI' ... , P" E RK'.K C 9.Jl are in general position if the vectors are linearly
independent.
Suppose Q is an arbitrary point and the points Pi' i = 0, 1, ... , n, are in
general position. We denote p(Q, P;) by ui(Q), i = 1, 2, ... , n. We. can prove
that the map u = (u l , u2 , ... , u") maps a small neighbourhood of Po homeo-
morphically onto a domain of E", that is, it is a local coordinate system in 9.Jl in
a neighbourhood of Po. We call a coordinate system of this kind a distance
coordinate system.
210 V.N. Berestovskij and I.G. Nikolaev

We denote by A(Q) the matrix

A(Q) = (cos 1+ (Xp,(Q), X piQ»)i.j=l.2 ..... ",
where 4 denotes the angle between corresponding vectors, and Q = Po. We can
verify directly (Berestovskij (1975» that the distance coordinates are connected
to each other by transition functions of class Cl. Direct calculation shows that
in distance coordinates the components of the metric tensor, calculated for the
standard formulae, have the form
G(u l , u2, ... , u") = [A(U-l(U l , ... , U,,))]-l. (12.2)
Thus the components of the metric tensor in distance coordinates are contin-
uous functions. Relying on (10.1), by means of (12.2) we can prove that formula
(12.1) and 3) hold.

§ 13. Parallel Translation in Spaces with Bounded Curvature

Parallel translation is concerned with a number of fundamental concepts of

differential geometry. By specifying on a manifold the operation of parallel
translation of vectors, we obtain the corresponding geometry on the manifold.
In Riemannian geometry a metric is an original concept, and a parallel transla-
tion consistent with the Riemannian structure is restored uniquely from the
metric (the Levi-Civita theorem).
In this section, by means of a purely geometrical construction we construct
an operation that subsequently turns out to be a parallel translation uniquely
corresponding to the metric.
In the case of a classical Riemannian manifold our construction gives the
parallel translation of Levi-Civita.

13.1. Construction of a Parallel Translation (Nikolaev (1980), (1983». We

shall carry out all the constructions in a sufficiently small convex domain
rlIt £; IDl, for which the bounds on the dimensions will be stated in the course of
the presentation.
At the basis of the construction of a parallel translation lies the symmetry
map.
Symmetry with respect to a point 0 E IDl is the map 9'0: rlIt -+ IDl that sets up a
correspondence between a point P E rlIt and the point P' E IDllying on the exten-
sion of the shortest curve PO beyond 0 so that P'O = PO. The points P and P'
are said to be symmetrical with respect to the point O. By virtue of the axioms
of a space with bounded curvature, the symmetry map is well defined in a small
domain rlIt £; IDl.
Consider a shortest curve AA' c rlIt and the tangent spaces to IDl at A and A'.
We construct a map ilAA , of the ball B(O, e) c IDlA of radius e > 0 with centre at
e
the origin (in IDlA ,). Let E B(O, e). From A we draw the shortest curve AH in
 II. Multidimensional Generalized Riemannian Spaces 211

H"

H'
Fig. 12

e e
the direction of the vector and of length I I (H = eXPA e). Let H' be the point
symmetrical to H about the midpoint ofthe shortest curve AA'. Then IlAA,(e) =
eXPA,1 H'. If the domain 0/1 is sufficiently small, and e is close to zero, then all our
constructions are obviously possible.
Now let AB be an arbitrary shortest curve lying in 0/1. We split it into 2m equal
intervals by points Ao = A, A 1 , ••• , A2 m = B. We denote the map IlA.Al+1 con-
structed above by Ill, 1+1 , and put hm = AB/2m.
e
Let us define a map Ilm : IDlA -+ IDl B. Let E IDlA be an arbitrary vector; we put
e' = hmeMI. Consider the map

Il~(e) = Il2m_1,2m 0'" 0 Il 1,2 0 Ilo,1(0

(for m = 1 see Fig. 12). When m ~ 1 we take the vector Ilm(e) equal to a vector
oflength lei directed like the vector Il~(e) (when m = 0 the direction of ll~(e) is
changed to the opposite). By means of (10.1) we can show that for each E IDlA e
the sequence Ilm(e) has a limit, which we denote by Il(e) (Nikolaev (1983a». We
call the resulting map Il: IDlA -+ IDlB a parallel translation along the shortest
curve AB. After this a parallel translation is defined along any rectifiable curve
in the space with bounded curvature.

Remark. E. Cartan considered symmetry in smooth Riemannian manifolds,

and by means of Cartan symmetry he interpreted parallel translation in Rie-
mannian manifolds (Cartan (1928), pp. 205, 206». Our construction is based on
this interpretation of Cartan.

13.2. Statement of the Main Results. The geometrically introduced parallel

translation II preserves the scalar product of vectors, that is, it is Riemannian.
Namely, we have the following result.
Proposition 13.1 (Nikolaev (1983a». In a space with bounded curvature a par-
allel translation Il along an arbitrary rectifiable curve is an isometric map of the
corresponding tangent spaces.
It is known that in the smooth case the curvature tensor is expressed in terms
of the ratio of the increment in the vector, on going round a closed contour lying
on a surface touching a given two-dimensional direction, and the area of this
212 V.N. Berestovskij and I.G. Nikolaev

surface. Therefore, in the non-smooth case the boundedness of curvature of a

space with a parallel translation specified in it (in terms of the parallel transla-
tion) is naturally understood in the sense that all such ratios are bounded.
Theorem 13.1 stated below connects the boundedness of curvature of a space
according to Aleksandrov and the boundedness of curvature in the sense men-
tioned above in terms of parallel translation.
Theorem 13.1 (Nikolaev (1983b». Any point P in a space with bounded curva-
ture 9R has a neighbourhood "Y c: 9R such that for an arbitrary closed rectifiable
e
curve yin "Y (P E y) the increment Ae of a vector E 9R p under a parallel transla-
tion fl along y has the bound
IAel ~ Jl(K', K)' u(P, y) 'Iel,
where Jl(K', K) is a positive constant depending on K' and K, and u(P, y) is the
area of the ruled surface spanned on y (see 9.3).
By means of the construction of a parallel translation we can prove the
following theorem, which is necessary for further examination of the parallel
translation fl that we have introduced.
Theorem 13.2 (Nikolaev (1983b». The components of the metric tensor (gij),
i,j = 1,2, ... , n, in an arbitrary distance coordinate sytem in a space with bounded
curvature satisfy a Lipschitz condition, that is,
Igij(X) - gij(y) I ~ Llx - YI,
where x and y belong to the range of distance coordinates, and L is a positive
constant.
Let G s;;; ~" be the range of distance coordinates. From Theorem 13.2 it
follows that at almost all points (in the sense of Lebesgue measure) of G there is
a total differential of the components of the metric tensor gij in the distance
coordinate system. This gives us the possibility of defining analytically the par-
allel translation fl' along "almost all" differentiable curves in 9R. We say that
a curve y c: 9R is differentiable if it is specified in the distance coordinates by
functions of class Cl. We assume that the differentiable curve y is such that at
almost all points of y the gij are differentiable (by Theorem 13.2 this is satisfied
for "almost all" differentiable curves). Then along y in the distance coordinate
system we can define the Christoffel symbols

r,!
IJ
=!2 gkl (88Xigjk + 8g8xkij _ 8x
8gij )
k ' (13.1)

We now define a parallel translation along y, fl': 9R Y(/o) -+ 9R Y(/)' by the

formula
er(t) = eo - it I;j"(y(s»ei(s)' yl(S) ds, (13.2)
J/o
where eA, ... , eo are the coordinates of the parallel displaceable vector
eo E 9R Y(/o) in the distance coordinate system, y 1 (s), ... , y"(s) are the coordinates
 II. Multidimensional Generalized Riemannian Spaces 213

of the vector tangent to the curve y for the value s of the parameter, and
(~l (t),
... , ~n(t)) is the resulting field of parallel vectors.
From the Lipschitz condition for giix) it follows easily that the integral
equation (13.2) has a unique solution ~r(s) on condition that ~r(o) = ~~.
The next theorem connects the geometrically defined parallel translation fl
and the parallel translation fl' introduced analytically.

Theorem 13.3 (Nikolaev (1983b)). Along any differentiable curve in 9Jl, at

almost all points of which the components of the metric tensor in the distance
coordinate system are differentiable, the geometrically defined parallel translation
fl and the analytically defined parallel translation fl' coincide.

Remark. From Theorem 14.1 stated in § 14 it follows that by using "harmonic"

coordinates we can define a parallel translation analytically (that is, by means of
(13.2)) for any differentiable curve in 9Jl and it also coincides with the geometri-
cally defined parallel translation n.

13.3. Plan of the Proof of the Main Results of § 13.

13.3.1. Isometry of the map n. First of all we mention the properties we need
for the symmetry map that lies at the basis of the construction of the map n.

Lemma 13.1. We consider arbitrary points P, Q E 9Jl and points P', Q' E IlIJ
symmetrical to them with respect to some point 0 E 1lIJ. Then
IPQ2 - P'Q'21 ~ jl'[max{PO, QO}J4,
where jl' is a positive constant depending on K' and K.
Lemma 13.2. We retain the notation of Lemma 13.1. We also consider a point
A E IlIJ and the point A' symmetrical to it with respect to O. Then
Icos 4. PAQ - cos 4. P'A'Q'I ~ JL"'M 2(1 + M2/m 2 + M 4 /m 4 ),
where M = max{AO, AP, AQ}, m = min{AP, AQ, A'P', A'Q'}, and jl" is a con-
stant depending on K' and K.
Lemmas 13.1 and 13.2 were proved in Aleksandrov, Berestovskij and
Nikolaev (1986). They follow easily from (10.1) and the Euclidean cosine
theorem.
From Lemmas 13.1 and 13.2 we obtain the following preliminary bound for
flAA ,·

Lemma 13.3 (Nikolaev (l983a)). Let A' A be a shortest curve of length h > 0,
L, and L2 (0 < Ll < L 2) certain fixed constants, and flAA ,: 9JlA -+ 9JlA , the map
constructed in 13.1. Then for arbitrary vectors ~, , E 9JlA whose lengths satisfy the
inequalities
214 V.N. Berestovskij and I.G. Nikolaev

we have the bound

14(~, C) - "4(llAA'(~)' llAAml :s:;; JJ"" h2,
where JJ'" is a constant depending on L l , L 2 , K' and K.
The proof of Lemma 13.3 is obtained from Lemma 13.2 by considering in IDlA
the orthonormal basis Xi' i = 1, 2, ... , n, and in IDlA' the basis Xi = llAA'(Xi )
(orthonormal to within O(h2)).
We now consider an arbitrary shortest curve AB c d/J and vectors ~, CE IDlA.
Then, by Lemma 13,3, for the maps ll~ constructed in 13.1 we have the follow-
ing bounds (the fulfilment of the bound required in Lemma 13.3 with constants
Ll and L2 depending on K' and K is guaranteed by Lemma 13.1):
14 (ll2'"-1. 2'" 0'" 0 llO.l(~)' ll2'"-1.2'" 0'" 0 llO.l(m
- "b,(ll2'"-2.2'"-1 0'" 0 llO.l(~)' ll2'"-2.2'"-1 0'" 0 llO.l(ml :s:;; jJ'''/4m,
14(ll2'"-2.2'"-1 0'" 0 llO.l(~)' ll2'"-2.2'"-1 0'" 0 llO.l(m
- 4(ll2'"-3.2'"-2 0'" 0 llO.l(~)' ll2'"-3.2'"-2 0'" 0 llO.l(ml :s:;; JJ"'/4m, ...
from which it follows that

Making m in the last inequality tend to infinity, we obtain the assertion of

Proposition 13.1.

13.3.2. Boundedness of the Curvature in Terms of Parallel Translation. We

observe that it is sufficient to prove Theorem 13.1 for the case when y is a
triangle. The general case reduces to consideration of a parallel translation
along a K-fan (see 6.1.3) inscribed in y, which in turn reduces to consideration
of a parallel translation along the triangles that form the K-fan.
We also observe that there is no need to consider triangles that can be arbi-
trarily close to degenerate ones. We can always "split" such "poor" triangles
into "proper" triangles whose angles differ from zero and 1t by some constant,
and represent a parallel translation along a "poor" triangle as a sum of parallel
translations along "proper" triangles.
In considering a parallel translation along a "proper" triangle T = OBC it is
sufficient to develop bounds for the maps llOB' llBC' llco constructed in 13.1.
These bounds can be obtained synthetically by using (10.1).

13.3.3. The Lipschitz Condition for the Components of the Metric Tensor in a
Distance Coordinate System. Let us state a lemma from which Theorem 13.2
follows directly. As a preliminary we introduce the following concept: let B E tft
be a fixed point. When Q E d/J\ {B} we denote by XB(Q) a vector of IDlQ such that
B = eXPQ XB(Q).
Lemma 13.4 (Nikolaev (1983b)). Let y: 1- dIJ be a differentible curve, and let
B be a fixed point of tft such that P9II(y(t), B) ~ ~ > 0 for any tel and some ~ > 0
 II. Multidimensional Generalized Riemannian Spaces 215

(here Pro/ denotes the intrinsic metric of IDl). Then for any t, to E I we have the
bound
I(XB):o - XB(to)1 ~ Jl(t5)'I(y; t, to)·
Here, as usual, (XB):o denotes the vector of IDly(to) obtained as a result of
parallel translation of the vector XB(t) along the curve y from the point y(t) to
the point y(t o), l(y; t, to) denotes the length of the arc of y corresponding to
values of the parameter lying between to and t, and Jl(t5) denotes a positive
constant depending only on t5, K' and K.
The proof of this lemma is based on the following geometrical assertion,
which can be obtained purely synthetically by using (10.1).
Consider the vector field XB(Q) for Q belonging to the shortest curve AC
(B ¢ AC). Let L be an arbitrary point for which AL = AC, and I.: the point
symmetrical to L with respect to the midpoint 0 of the shortest curve AC. We
have the bound
Icos 4 LAB + cos 4BCI.:1 ~ Jl(t5)·AC.
A direct consequence of Lemma 13.4 is the following lemma.
Lemma 13.5 (Nikolaev (1983b». Consider a convex domain "Y s; o/J and
points A, B E Oft\ "Y for which pro/(A, "Y), pro/(B, "Y) ~ t5 > 0 for some t5. Then for
the function cp(P) = (XA(P), XB(P» we have the bound
Icp(P) - cp(Q)1 ~ Jl(t5). PQ
for arbitrary P, Q E "Y. Here Jl(t5) is a constant depending on t5, K' and K.
For the points A and B we consider the points 1';, i = 1, 2, ... , n, that take
part in the construction of the distance coordinate system (see 12.3.2). Then,
taking account of (12.2), we obtain the assertion of Theorem 13.2 from Lemma
13.5.
13.3.4. Coincidence of Geometrically and Analytically Defined Parallel
Translations. It is known that to a parallel translation there corresponds the
concept of covariant differentiation of the vector field along a differentiable
curve. Namely, let y: 1--+ IDl be a differentible curve, and X the vector field along
y. We denote by (X):o the result of parallel translation of the vector X(t) along y
from the point y(t) to the point y(t o). We call the limit
lim [(X):o - X(to)]/(t - to),

if it exists, the covariant derivative of the vector field X at the point to along y
and denote it by J7yXlto.
Obviously in the case of a parallel translation defined analytically II'
specifies a Levi-Civita connection on IDl. It is known that on (IDl, p) there is only
one Levi-Civita connection (Gromoll, Klingenberg and Meyer (1968». There-
fore, to prove that the parallel translations II and II' coincide we need to prove
that II specifies a connection J7 on IDl that is a Levi-Civita connection. This
216 V.N. Berestovskij and I.G. Nikolaev

means that apart from the usual properties of a connection we need to establish
that V is a Riemannian connection (that is, the following rule for differentiating
the scalar product along y is satisfied:
(X, Y)' = (VyX, Y) + (X, VyY)
and the connection V has zero torsion: V x Y = VyX + [X, Y]).
It is sufficient to carry out the verification ofthe necessary properties of V for
vector fields of the form XB for arbitrary B E rot The geometrical character of
the definition of the fields XB (see 13.3.3) enables us to do this on the whole
synthetically, relying on the boundedness of the curvature of rol in the form
(10.1).

§ 14. Smoothness of the Metric of Spaces with

Bounded Curvature

In this section we describe the differential properties of the metric tensor of a

space with bounded curvature.
Example 12.1 shows that we cannot expect the metric tensor of a space with
bounded curvature to be twice continuously differentiable. Nevertheless we can
assert that the metric tensor of a space with bounded curvature has second
derivatives in some generalized sense. This turns out to be very important, since
it gives the possibility of writing the components of the curvature tensor of the
metric formally by means of the components of the metric tensor and its first
and second deriatives by well-known formulae, and in many cases to carryover
the apparatus of "smooth Riemannian geometry" almost automatically to the
case of a space with bounded curvature. In the next chapter we give results
showing that the formally written curvature coincides with the curvature de-
fined in the natural geometrical way.
The differential properties of the metric of spaces with bounded curvature are
described in terms of the Sobolev function spaces W~. This enables us to use the
apparatus of the theory of functions with generalized derivatives.

14.1. Statement of the Main Result.

14.1.1. Some Function Spaces. Let Q be a domain of ~n; we denote by Lp(Q),
p ~ 1, the normed space of all functions on Q that are Lebesgue integrable in
degree p with norm

IlfIILp(n) = [In IflPdxJIP.

We define the space of functions W~(Q), p, I ~ 1, as the subspace of those
functions of Lp(Q) that have in Q all generalized derivatives with respect to Xi,
i = 1, 2, ... , n, up to order I inclusive that are integrable in degree p on Q. For
 II. Multidimensional Generalized Riemannian Spaces 217

f E W~(.Q) we introduce the norm

Ilfllw~(Q) = [fa (lfl P+ J1 IS#k ID(S~flP) dxTIP,

where s = (S1' S2' ... ' sn), Si ~ 0, are integers.
In the case of a domain Q with smooth boundary, W~(Q) is the closure, in the
norm indicated above, of all smooth functions on Q.
We also need the class offunctions Wp1(Q). We denote by Wp1(Q) the closure
in the norm Wp1 of the set of all smooth functions that are compactly supported
inQ.
°
As usual, we understand by C·IX(Q), r = 0, 1,2, ... , < 0( < 1, the space of r
times continuously differentiable functions on Q, all r-th derivatives of which
satisfy a Holder condition which exponent 0(: ID(r~f(x) - D(ry(x o) I ~ Clx - xOI IX•
F or the function spaces under consideration we refer the reader to the mono-
graphs Sobolev (1950), Gol'dshtein and Reshetnyak (1983), and Ladyzhenskaya
and UraI'tseva (1964).

14.1.2. What Do We Understand by the Smoothness of the Metric of a Rie-

mannian Manifold? The metric of a Riemannian manifold belongs to some
function space H if for each point of 9Jl there is a local coordinate system of
differential structure given on 9Jl such that the components of the metric tensor,
written in this coordinate system, are functions of the class H.

14.1.3. Harmonic Coordinates. We recall that a coordinate system ~1, ~2, ... ,
~n,specified in a domain of an n-dimensional Riemannian manifold, is called
harmonic if
Ll2 ~' = 0, I = 1, 2, ... , n,
where Ll2 is the Laplace operator on 9Jl (see 14.2).
In the investigation of the question of the smoothness of the metric of a
Riemannian manifold, not all coordinate systems are in an equivalent position.
Thus, for example, in normal Riemannian coordinates there is a loss of two
derivatives of the components of the metric tensor. In many cases harmonic
coordinates are optimal in the sense of smoothness of the metric (for example,
in the case H = C·IX). Sometimes harmonic coordinates give the worst smooth-
ness of the metric in comparison with what it really is (Sabitov and ShefeI'
(1976), p. 924).
Harmonic coordinate systems were first used by Einstein (Einstein (1916)). In
Riemannian geometry harmonic coordinate systems were first applied to ques-
tions connected with the smoothness of the metric in Sabitov and ShefeI' (1976);
these results were proved again later in DeTurck and Kazdan (1981).

14.1.4. Statement of the Main Result.

Theorem 14.1 (Nikolaev (1983b)). Let 9Jl be a space with bounded curvature.
Then in a neighbourhood of each point of it we can introduce a harmonic coordi-
218 V.N. Berestovskij and I.G. Nikolaev

nate system. The components of the metric tensor in any harmonic coordinate
system in IDl are continuous functions of class w,,2, where for p we can take any
number not less than one. Harmonic coordinate systems specify on IDl an atlas of
class c 3, .. for any 0 < IX < 1.
Remark 1. From Sobolev's embedding theorem (see Ladyzhenskaya and
Ural'tseva (1964), Theorem 2.1 on p. 64) it follows that in harmonic coordinates
the components of the metric tensor belong to the class C 1, .. for any 0 < IX < 1.
Remark 2. The components of the metric tensor in a harmonic coordinate
system have an ordinary second differential almost everywhere (in the sense of
n-dimensional Lebesgue measure) (this follows from the fact that gij E Wp2 when
p > n and Theorem 5.2 in Chapter 2 of Gol'dshtein and Reshetnyak (1983».
14.2. Plan of the Proof of Theorem 14.1.
14.2.1. Harmonic Coordinates in Spaces with Bounded Curvature. The
Laplace operator ,12 on IDl has the form
.. il 2 u ij ilu
,1 2u = glJ ilxiilx j - g rJ ilx P.
By Theorem 13.2 the coefficients in ,12 for the higher derivatives belong to the
class Co, 1, and the coefficients for the lower derivatives are bounded with re-
spect to the distance coordinate system. In this connection, the equation ,12 = 0 e
ee
has solutions 1, 2 , ••• , e"
that satisfy at the initial point the condition
ile'
axi(X O) = lJI, i, I = 1,2, ... , n,

(see Theorem 2 in Chapter 5 of Bers, John and Schechter (1964); the assertion of
this theorem remains true when we require not continuity but only boundedness
of the lowest coefficients of the equation).
Now let gij' rJ
and so on be the components of the metric tensor and the
Christoffel symbols in a harmonic coordinate system. From the "harmonicity"
of the coordinates it follows that
gp'rp',J = O. (14.1)
From (14.1), by analogy with what was done in Sabitov and ShefeI' (1976), we
can derive an equation of elliptic type on the metric. Since e'(x) (the transition
functions from the distance coordinate system to the harmonic) belong to the
class Wp2 for any p ~ 1 (Ladyzhenskaya and UraI'tseva (1964), Theorem 15.1 of
Chapter III), we have
gij E w"l, Ii), rp"j E Lp- (14.2)
The conditions (14.2) enable us to define the generalized function

(Rij, rt) = -~ fa gP'(Djgp'· Drt - Dpgi/· Djrt

+ D,gij· Dprt - D,gpj· Dirt) de, i,j = 1,2, ... , n,

 II. Multidimensional Generalized Riemannian Spaces 219

that is, a functional on Wpl. Here D, = a/ae', and a is the range of the harmonic
coordinates. We observe that up to the first derivatives of gij' (Rij' ,,) in the
smooth case is the integral with respect to a of the product of" and the corre-
sponding Ricci curvature. Differentiating (14.1), we can obtain the equality

ft] gPID,gij' Dp" de = -(Rij + Rji , ,,).

If the generalized function (Rij' ,,) is represented in the form

(Rij ,l1) = LRij(e)'l1(e) de, (14.3)

then gij is the generalized solution of the equation

a~p (gPI ~~il) = 2Rij. (14.4)

Equation (14.4) is the required elliptic equation on the metric.

14.2.2. The Bound in L, ofthe Curvature of the Averaged Metric. In order to

prove that the generalized function (Rij''') that we have introduced can be
represented in the form (14.3), we need to prove that the norm of the functional
Rij is bounded. To this end we consider the Sobolev averaging of the metric gij
and prove that the curvature of the averaged metric is bounded.
Let a o be a convex domain, lying strictly inside the domain a, where the
harmonic coordinates are specified. We take the Sobolev average of the compo-
nents of the metric tensor gij (Gol'dshtein and Reshetnyak (1983)). Then for the
resulting metric gt, i, j = 1, 2, ... , n, we denote the Christoffel symbols by Iir,
k = 1,2, ... , n.
Proposition 14.1 (Nikolaev (1983b)). For each q ~ 1 we have the bound

aIjtk aIitk
II aei -
II
ae j L U2 l:::; JJ.(q), (14.5)
q o

where JJ.(q) is a constant depending on q, K', K (and the chosen harmonic coordi-
nate system).
The plan of the proof of Proposition 14.1 is as follows.
Consider a square in a whose sides are parallel to the i-th andj-th coordinate
curves in a and have length e > O. Let x denote the vertex of the square with the
least i-th and j-th coordinates. Let Ke,x,i,j denote the curve in 9Jl whose har-
monic coordinates form the square in a mentioned above. We now specify the
function L1~,ijl(X) for x E ao: we carry out a parallel translation of the I-th coordi-
nate vector E,(X) of the harmonic coordinates along K.,x,i,j' Then L1~,ijl is the
r-th coordinate of the increment of E,(X), By Theorem 13.1, L1~,ijl(X) = 0(e 2 ), on
the other hand, L1~,ijl(X) can be expressed in terms of the components of the
metric tensor gij by means of (13.2), from which we deduce (14.5).
220 V.N. Berestovskij and I.G. Nikolaev

14.2.3. Smoothness of the Metric. From Proposition 14.1 we deduce that the
generalized function (R u,") can be represented in the form (14.3) and that
Ri8) E Lp for any p ;;:: 1. Then the smoothness of the metric is obtained from
Theorem 15.1 in Ladyzhenskaya and Ural'tseva (1964); p. 236. The existence of
a C 3 •iI-smooth atlas follows from Sabitov and Shefel' (1976).

§ 15. Spaces with Bounded Curvature and Limits of

Smooth Riemannian Metrics

From Theorem 14.1 on the smoothness ofthe metric of a space with bounded
curvature we can deduce an important corollary that characterizes the metric of
such a space as the limit of smooth Riemannian metrics with uniformly bilater-
ally bounded sectional curvatures at all points in all two-dimensional directions.
As we mentioned in 11.1.3, if certain conditions are satisfied, the converse asser-
tion is also true: the uniform limit of such smooth Riemannian metrics is a
metric with bounded curvature.
It is convenient to illustrate the main results of this section on the class of
Riemannian manifolds rol(n, d, A, V) introduced below (see 15.2). Roughly speak-
ing, rol(n, d, A, V) is a class of n-dimensional compact Riemannian manifolds
whose curvatures have the bound IK.,I ~ A and for which certain conditions of
"normalizing" character are satisfied.
In rol(n, d, A, V) we introduce a metric in a natural way (see 15.2.1). With
respect to this metric rol(n, d, A, V) is not a complete metric space. Its comple-
tion consists precisely of the corresponding spaces with bounded curvature.
In this section we also discuss Gromov's compactness theorem for the class
of Riemannian manifolds rol(n, d, A, V) (Gromov (1981».
15.1. Approximation of the Metric of a Space with Bounded Curvature by
Smooth Riemannian Metrics.
15.1.1. A Local Variant of Approximation. Smooth metrics that converge to
the original metric in a small domain of a space with bounded curvature are
constructed by means of the operation of Sobolev averaging (Gol'dshtein and
Reshetnyak (1983».
ee
Let Q c ~n be a domain in which harmonic coordinates 1, 2 , • 00, areen
specified, and Q h the subdomain of Q at a distance h from the boundary of Q.
Then in Q h we define the averaged metric gZ:

gZ(e) = ~n l/ii(U)We ~ u) du,

where w: ~n --+ ~ is the averaging kernel, that is, a function for which the follow-
ing conditions are satisfied:
1) the support of w is contained in the unit ball B(O, 1) c ~n;
2) the function W is infinitely differentiable;
3) f nnW(X) dx = 1.
 II. Multidimensional Generalized Riemannian Spaces 221

The resulting metric ge is infinitely differentiable. From Theorem 17.1, and

also from the standard properties of the Sobolev averaging, it follows that the
curvatures K~ of the metric ge have the bounds
K' - e'(h) :s:; K~ :s:; K + e(h),
where K' and K are such that rol in the neighbourhood under consideration is a
space of curvature both :s:; K and ~ K', and e(h) and e'(h) tend to zero as h -+ O.
On the other hand, obviously Ilgij - gellw; -+ 0 as h -+ 0, and from Sobolev's
embedding theorems (Ladyzhenskaya and Ural'tseva (1964» it follows that ge
converge uniformly to gij in an arbitrary subdomain Do of the domain D.
Putting hm = 11m, we obtain the required local approximation of the metric
gij by smooth metrics g'ij = gem with sectional curvatures that are bilaterally
bounded in aggregate.

15.1.2. Global Approximation. The global approximation of the metric of a

space with bounded curvature can be constructed by means of the de Rham
averaging operator (de Rham (1955», which has as its basis the Sobolev averag-
ing operator. Necessary bounds for the de Rham averaging operator can be
obtained by starting from its form. Some of these bounds were obtained in
Gol'dshtein, Kuz'minov and Shvedov (1984).
Theorem 15.1. Let M be a space with bounded curvature for which the bounds
of the curvatures are Ko and K (that is, rol is a space of curvature both :s:; K and
~ Ko). Then for every K' > K and K~ < K on rol there is a sequence {gm} of
infinitely differentable Riemannian metrics for which
1) the sectional curvatures of the Riemannian manifold (IDl, gm) at all points in
all two-dimensional directions have the bounds
K~:S:; K';:S:; K';
2) the sequence {Pm} of intrinsic metrics of the Riemannian manifolds (rol, gm)
converges uniformly to the intrinsic metric P of the space rol on each compact
subset .91 c rol: Pm(P, Q) ~ p(P, Q), P, QEd.
Remark 1. In fact we can assert that the metrics gm obtained from the metric
9 by de Rham averaging converge to 9 in the norm W,2, p ~ 1.
Remark 2. As we have already mentioned, in a space rol with bounded curva-
ture there is a C3 '<Z-smooth differential structure. By a well-known theorem of
Whitney, this structure contains a COO-smooth atlas. We have in mind that the
metric tensors 9m belong to the class Coo with respect to the charts of this atlas.
Remark 3. If the space (rol, p) is complete, then, beginning with a certain
number, the spaces (rol, Pm) are also complete.
Remark 4. The authors do not know whether it is possible to approximate
the metric 9 by smooth Riemannian metrics 9m with the same bounds for the
sectional curvatures (that is, whether one can take Ko = K~ and K = K' in
Theorem 15.1).
222 V.N. Berestovskij and I.G. Nikolaev

15.2. A Space of Riemannian Manifolds with Sectional Curvatures Bounded in

Aggregate. We restrict ourselves to the consideration of compact Riemannian
manifolds whose sectional curvatures have the bound

IK"I ~A. (15.1 )

We observe that by mUltiplying the metric of an arbitrary compact manifold

by a suitable constant we can arrange that (15.1) is satisfied for a Riemannian
manifold conformally equivalent to the given one. To avoid this, we introduce a
restriction on the diameter of the manifold

d(Wl) ~ d. (15.2)

We also restrict ourselves to the consideration of manifolds of constant di-

mension n:

dim(Wl) = n. (15.3)
In the class of Riemannian manifolds that satisfy conditions (15.1)-(15.3) we
can find a sequence of manifolds that "converge" to a manifold of lower dimen-
sion as shown, for example, in Fig. 13. To remove the possibility of this situa-
tion, we introduce a condition on the volume V(Wl):

V(Wl) ~ v> 0 (15.4)

or on the radius of injectivity i(Wl) of Wl:
i(illl) ~ io > O. (15.4')
We denote by illl(n, d, A, V) the class of compact Riemannian manifolds that
satisfy conditions (15.1)-(15.4). Correspondingly, Wl(n, d, A, io) is the class of
compact Riemannian manifolds that satisfy conditions (15.1)-(15.4)'.
Remark. From Proposition 11.2 it follows that in the class Wl(n, d, A, V) for
all Wl E Wl(n, d, A, V) there is a uniform bound of type (15.4), on the radius of
injectivity i(Wl).
We have taken the above heuristic arguments for the definition of the classes
Wl(n, d, A, V) and Wl(n, d, A, io) from Peters (1987).
We now introduce a distance in the space Wl(n, d, A, V) that specifies a "natu-
ral" convergence of Riemannian manifolds.

Torus "Narrower torus" Circle

Fig. 13
 II. Multidimensional Generalized Riemannian Spaces 223

15.2.1. Lipschitz Distance. Let f: X --+ Y be the Lipschitz map of the metric
spaces X and Y. Then the dilatation of f is the quantity
dilf = sup p(f(x),J(x'))/p(x, x').
x#x'

The Lipschitz distance between the metric spaces X and Y is taken to be

dL(X, Y) = inf{ lIn dil fl + lIn dil f- 1 1},
where the infimum is considered over all bi-Lipschitz maps f: X --+ Y. If there is
no bi-Lipschitz map f: X --+ Y, we take the distance to be +00.

15.2.2. Hausdorff Distance. The classical Hausdorff distance between sub-

spaces of a given metric space is defined as follows.
Let Z be a metric space, and X and Y subspaces of Z. We denote the set
{z E Zlp(z, X) < e} by U.Z(X). Then d~(X, Y) = inf{eIU.Z(X);2 Y, U.Z(Y);2 X}.
Now the Hausdorff distance between arbitrary metric spaces X and Y is
taken to be inf{d~(f(X), g(Y))}, where the infimum is considered over all metric
spaces Z and all isometric embeddings f: X --+ Z and g: Y --+ Z. The notation is
dH(X, Y).

15.2.3 Connection between the Topologies Specified by the Lipschitz Distance

and the Hausdorff Distance. Obviously, in the general case these topologies do
not coincide (for an example, see Peters (1986)), namely, the topology specified
by the Hausdorff distance is coarser than the topology specified by the Lipschitz
distance. However, for the class of Riemannian manifolds 9Jl(n, d, A, V) the fol-
lowing assertion is true.
Proposition 15.1 (Gromov (1981)). In 9Jl(n, d, A, V) the topologies specified by
the Lipschitz and Hausdorff distances coincide.
In other words, convergence of a sequence of Riemannian manifolds of
9Jl(n, d, A, V) with respect to the Hausdorff distance is equivalent to the conver-
gence of the same sequence with respect to the Lipschitz distance.

15.2.4. Gromov's Compactness Theorem.

Theorem 15.2 (Gromov (1981)). The class of Riemannian manifolds
9Jl(n, d, A, V) with respect to the topology defined by the Lipschitz distance (or the
Hausdorff distance) is a relatively compact set in a wider class of C1 , I-smooth
n-dimensional manifolds with CO -smooth metric.
Thus, Gromov's compactness theorem asserts that from any sequence of
Riemannian manifolds in the class 9Jl(n, d, A, V) we can choose a sequence that
converges with respect to Lipschitz (or Hausdorff) distance to a Riemannian
manifold with continuous metric tensor.

15.2.5. Limiting metrics for 9Jl(n, d, A, V). As we have seen, Gromov's theo-
rem asserts only the continuity of the limiting metric tensor. In fact, the prop-
224 V.N. Berestovskij and I.G. Nikolaev

erties of the limiting metric are considerably better. This is caused by the fact
that the limiting manifold is a space with bounded curvature (see the remark to
Proposition 11.3).
Theorem 15.3 (Peters (1986». Suppose that 0 < 0( < 1. Then an arbitrary
sequence of IDl(n, d, A, V) contains a subsequence that converges to an n-
dimensional differentiable manifold IDl with metric g of Holder class C1 ....
Since the limiting manifold is a space with bounded curvature, we can
strengthen Theorem 15.3 by applying Theorem 14.1:
Theorem 15.4 (Peters (1987». The limiting Riemannian manifold in the com-
pactness theorem 15.2 is a space with bounded curvature, so the components of the
metric tensor of the limiting metric in a harmonic coordinate system belong to the
class Wp2 for any p ~ 1.
By Sobolev's embedding theorems, Theorem 15.3 obviously follows from
Theorem 15.4.

15.2.6. Why is it Important to Know the Smoothness of the Metric of the

Limiting Riemannian Manifold? To answer this question we give one of the
characteristic examples of the application of Gromov's compactness theorem.
First example. The rigidity theorem and <5-pinched manifolds with <5 < 1/4.
The rigidity theorem says that if IDl is a compact simply-connected Riemannian
manifold of even dimension n with <5-pinched curvature for <5 = 1/4 (that is, at
all points P E IDl for all plane elements G ~ IDl p the sectional curvatures K,,(P)
satisfy the bounds 1/4 ~ K,,(P) ~ 1), then IDl is either homeomorphic to an
n-dimensional sphere or is isometric to one of the symmetric spaces of rank 1:
the complex projective space CP", the quaternion projective space HP", or the
Cayley projective space CaP 2 (see Berger (1983».
In the case <5 < 1/4 there is the conjecture that there are no other differ-
entiable manifolds apart from the symmetric spaces of rank 1 mentioned above.
In this connection Berger proved the following theorem.

Theorem 15.s (Berger (1983». For each even n there is a number s(n) < 1/4
such that all simply-connected compact Riemannian manifolds that have an s(n)-
pinched metric are either homeomorphic to sn or diffeomorphic to one of the
symmetric spaces of rank 1.
There is great interest in the method of proof of Theorem 15.5. The proof can
be carried out by contradiction.
Suppose there is a sequence of sm(n)-pinched Riemannian manifolds (IDlm' gm),
m = 1, 2, ... , where 8",(n) -+ 1/4 as m -+ 00, for which the assertion of Theorem
15.5 does not hold.
By means of Cheeger's finiteness theorem (see Peters (1986» we can arrange
matters so that all the manifolds IDlm are diffeomorphic to some manifold IDl.
The theorems of Meyer and Klingenberg give the necessary bounds on the
 II. Multidimensional Generalized Riemannian Spaces 225

diameter and the radius of injectivity of the Riemannian manifolds (rol, gm).
Consequently, we can apply Gromov's compactness theorem to the sequence
(rol, gm). If the metric of the limiting manifold (rol, go) is sufficiently smooth, then
by applying the rigidity theorem to (rol, go) we obtain the necessary contradic-
tion. The main difficulty in the proof of Theorem 15.5 just consists in estab-
lishing that the metric go is sufficiently smooth (in Berger (1983) it was proved
that go belongs to the class COO).
Another example of this kind of the application of Gromov's theorem is given
by Brittain's theorem:

Theorem (Peters (1986». There is a positive number e, depending only on n,

maxlKWlI and a positive number Vo, such that if Ric(rol) ~ n - 1, Vol(rol) ~ Vo
and d(rol) ~ 1t - e (where d(rol) is the diameter of roll, then rol is diffeomorphic to
sn.
For applications of Gromov's compactness theorem we refer the reader to
the articles of Peters (Peters (1986), (1987».

Chapter 4
Existence of the Curvature of a Metric Space at a Point
and the Axioms of Riemannian Geometry

In this chapter we state the geometrical conditions that distinguish Rieman-

nian manifolds among general metric spaces. The most essential part of these
conditions is that "continuous curvature exists" for a given metric space.
Thus, an important feature is a suitable "Riemannian" definition of the cur-
vature at a point of the space. The curvature defined here, as in the case of
multidimensional Riemannian manifolds, depends not only on the point, but
also on a pair of directions at the point. We call it non-isotropic Riemannian
curvature.
The value of the curvature is calculated by means of the limit of the ratios of
the excesses of triangles and their "area", namely ;5(T)/u(T), that contract to a
given point, so that the directions of fixed pairs of sides of these triangles "con-
verge" to a given "admissible" pair of directions at the point. Continuity of
the non-isotropic Riemannian curvature is defined by means of the distance d,
which is introduced between the directions at different points of the space (see
§ 16).
On the other hand, at points of two-fold differentiability of a Riemannian
metric, from the metric tensor we can purely formally calculate the sectional
curvature K,,(P) of the space rol. Our main result involving the curvature that
we have introduced is that "almost everywhere" K,,(P) can be calculated by
means of the limit of the ratios ;5(T)/u(T).
226 V.N. Berestovskij and I.G. Nikolaev

We also introduce the isotropic Riemannian curvature, that is, the curvature
that depends only on the point, and not on the directions at this point.
At the end of the chapter we state a theorem asserting that isotropic metric
spaces (that is, metric spaces at each point of which the isotropic Riemannian
curvature exists) of dimension greater than two (the dimension can be under-
stood, for example, in the sense of Menger and Uryson, see Hurewicz and Wall-
man (1941)) are spaces of constant curvature.

§ 16. The Space of Directions of an Arbitrary Metric Space

In this section we define the function of the distance d between directions

specified at different points of a metric space. We shall call the set of all directions
of a metric space, considered together with the distance function d, the space of
directions of the given metric space.
In contrast to the space of directions at a point (§ 3), the space of directions is
not a metric space. The distance function d that we have defined is semimetric
(the triangle inequality is not satisfied for d). We give an example showing that
this is due to the essence of the question.
We need the concept of the space of directions to define the continuity of the
non-isotropic Riemannian curvature of a metric space.

16.1. Distance between Directions (Nikolaev (1987». The following require-

ments are imposed on the functions of distance between arbitrary directions:
this function makes sense for directions in an arbitrary metric space; the dis-
tance we have introduced "induces" in the case of a Riemannian manifold the
standard metric in the spherical bundle to the manifold (the Sasaki metric,
Sasaki (1958), (1962)). We recall that in the case of a Riemannian manifold the
set of directions coincides exactly with the spherical bundle of the manifold.

16.1.1. Heuristic Arguments. Let us recall how to define the Sasaki metric in
the spherical bundle. Consider two unit tangent vectors v and v + dv that touch
a Riemannian manifold 9Jl at points P and dP. Then the square of the distance
between these vectors is taken to be

da. 2 = ds 2 + dlP,
where ds denotes the distance between the points P and P + dP, and dO denotes
the angle between the vector v and the vector obtained as a result of parallel
translation of the vector v + dv along the shortest curve joining the points
P + dP and P (see Fig. 14).
An expression for the components of the metric tensor of the Sasaki metric in
terms ofthe metric ofthe Riemannian manifold was given in Sasaki (1962). Here
we just give a formula for the calculation of the lengths of differentiable curves
 II. Multidimensional Generalized Riemannian Spaces 227

v+ liv
v

P+ liP
P
Fig. 14

r
in the spherical bundle with respect to the Sasaki metric:

1(8) = «C, c) + (Vee, veewl2 dt. (16.1)

Here 8(t) = (c(t), e(t)), a ~ t ~ b, is the field of unit vectors defined along the
differentiable curve c(t) in IDl, that is, a differentiable curve in Q(IDl), c = c*(d/dt)
the field of tangent vectors to c(t), (, ) the scalar product in the tangent space
to IDl, and V the Levi-Civita connection (Gromoll, Klingenberg and Meyer
(1968)).
From what we have said it follows that the basis for the calculation of the
lengths of curves in Q(IDl) must be the distance, defined between close vectors as
d'(e, 0 = (PQ2 + 02(e, m1/2, eE QpIDl, 'E QQIDl.
As before, O(e, 0 here denotes the angle between the unit vector and the vector e
obtained as a result of parallel translation of the vector , along the unique
shortest curve joining P and Q.
Namely, in order to calculate the length of the curve E(t) in Q(IDl) we need to
split it by arbitrary points to = a < tl < t2 < ... < tm = b and consider the limit
over all such splittings of the interval [a, bJ:
m-l
Id·(8) = lim
m-+oo
L
i=O
d'(8(ta,8(t i +1)), (16.2)
max It/-tJI .... O

which coincides exactly with the quantity defined in (16.1). In this sense the
distance d' induces the Sasaki metric in Q(IDl).
We give an example showing that the distance d' is not a metric.
Example 16.1. On SK we consider a non-degenerate triangle T = ABC. Let
K = _k 2 < O. Consider the unit vectors 11> 12 , 13 touching the sides AB, AC, CB,
respectively, of the triangle T (see Fig. 15). Then obviously
0(11,/ 2) = Q:, 0(11' ' 3 ) = p, 0(/2,/ 3 ) = n - y.
But 0(12,/ 3 ) - 0(11,/ 2) - 0(11' 13 ) = -t5(T) > 0, that is, the triangle inequality
for 0 is not satisfied.
We now suppose that the triangle T is equilateral: AB = BC = AC = Co,
Q: = P = y = Q: o • We observe that Q:o is close to zero for very large k. Now taking
k sufficiently large, and Co sufficiently small, we deduce that 0(12' ' 3 ) - 0(11' ' 2)-
0(11' ' 3 ) is close to n, that is, the triangle inequality is violated for d'.
228 V.N. Berestovskij and I.G. Nikolaev

Fig. 15

To conclude this subsection we make the following important remark.

We assume that in D(IDl) we have defined the distance d, connected with d' as
follows:
Id(~, 0- d'(~, 01 :::; Jl. PQq, ~ E DpIDl, 'E DQIDl,
where Jl is a positive constant and q > 1.
Then obviously by calculating the length of the curve E(t) in D(IDl) from
(16.2), but with respect to the distance d, we obtain the same value. Thus, the
distances d and d' induce the same metric in D(IDl).
In an arbitrary metric space there is no concept of parallel translation. There-
fore, in order to introduce on the set of directions of an arbitrary metric space
the distance induced in the case of a smooth Riemannian manifold by the Sasaki
metric, we consider another distance d, which differs from d' by O(PQ2).
At the basis of the definition of the distance d is the formula for the cosine of
the angle between rays in Euclidean space.
Suppose that a ray I with starting point P passes through a point X, and that
a ray I' with starting point Q passes through a point Y. Then for the cosine of
the angle IX between the rays I and I' we have
cos IX = (XQ2 + yp2 - PQ2 - Xy 2)/(2PX· QY). (16.3)

16.1.2. Definition of the Distance d. We denote the set of directions of a

metric space IDl by D(IDl), that is,
D(IDl) = U DpIDl.
Pe WI

Suppose that ~, , E D(IDl), and that ~ E DpIDl, , E DQIDl, where P and Q are
points of IDl and P =P Q. We consider curves Land M, starting at P and Q
respectively, that specify the directions ~ and, (see 3.1).
Suppose that X E L, X =P P, and Y E M, Y =P Q, are arbitrary points. In ac-
cordance with (16.3) we introduce into consideration the function
hLM(X, Y) = (XQ2 + yp2 _ PQ2 - Xy 2)/(2PX·QY).
 II. Multidimensional Generalized Riemannian Spaces 229

To obtain a local characterization, we introduce the quantity

hLM = lim lim hLM(X, Y).
a~O+ X~P a"'PXIQr..,b
b~+oo r~Q

We observe that hLM , when P = Q, is equal to the cosine of the upper angle
between these curves (see Aleksandrov and Zalgaller (1962), Theorem 1 on
p.250).
Finally, to define the function h(e, ') depending only on the directions e and
" and not on each of the curves that specify them, we put h(e, ') = inf{hLM}'
where the infimum is considered over all curves Land M such that L specifies
the direction e and M specifies the direction ,.
We introduce into consideration the function

<p(e, 0 = { sup [h(e,,.,) - h(" ,.,)] + sup [h(e, ,.,') - h(" ",')]}/2,
"e Dp 9JI ,,' e Do 9JI

which coincides with Ie - " in the case when e and, are unit vectors in E".
Finally we define the distance between the directions e E Dp 9R and, E DQ9R in
the case of an arbitrary metric space as
d(e, ') = [PQ2 + <p2(e, ')]1/2.

16.1.3. Remarks on the Definition of the Function hLM • As we have seen, a

key role in the definition of the distance d is played by the function hLM . We
mentioned above that in Euclidean space hLM coincides with the cosine of the
angle between Land M. In the general case hLM can exceed one and even be
equal to infinity. Using the formulae for the first and second variation of the
length of a geodesic, we can prove, however, that in the case of a Riemannian
manifold hLM differs slightly, to the cosine of the angle between the vector e and
the vector, translated in a parallel way to the same point.

Proposition 16.1. Let G be a domain RK'.K of a C2-smooth Riemannian mani-

fold 9R whose diameter has the bound
diam G ~ min{1, (6nq-l},
where C = max{IKI, IK'I}, n = dim 9R. Let e E Dp 9R, 'E DQ9R be arbitrary
directions, where P, Q E G. Then
Ih(e, ') - cos O(e, ')1 ~ C(K', K)' PQ2,
where the constant C(K', K) depends only on K, K' and n.
From Proposition 16.1 it follows that we have the bound
Id(e,O - d'(e, 01 ~ 03(e, 0/24 + C(K', K)' PQ2,
where the constant C(K', K) depends on K, K' and n. Using the remark made
at the end of 16.1.1 and the last inequality, we can deduce that the semimetric d
also induces the Sasaki metric in D(9R).
230 V.N. Berestovskij and I.G. Nikolaev

We now discuss the behaviour of hLM in "non-Riemannian" metric spaces.

Namely, we give the following examples.
Example 16.2. Consider the normed space (~2, II 111) (see Example 4.1) and
in it the rays 11 = (0, t), 12 = (1, 1 + t), t ~ O. Then hLM = +00.
Example 16.3. On the set ~2 = {(x, y)lx, y E~} we specify the norm
lI(x, y)lIoo = max{lxl, Iyl}. In the resulting normed space we consider the rays
11 = (0, -t), 12 = (1,1 + t), t ~ O. Then hLM = -1.
Example 16.4. In the normed space (~2, II 1100) we consider the rays 11 =
~ O. Then hLM = O.
(0, t), 12 = (1, 1 - t), t
Examples 16.2-16.4 show that the distance we have introduced is slightly
connected with parallelism in the "non-Riemannian" case.

16.2. Space of Directions. Let us recall the concept of a semimetric space. A

set rol with a non-negative function d911 defined on a Cartesian square of rol is
called a semimetric space if d!JI satisfies the following conditions:
1) d'J)z(P, Q) is equal to zero if and only if P = Q;
2) d'J)l(P, Q) = d 9ll (Q, P);
see Blumenthal (1970), § 5. In contrast to Blumenthal (1970) we admit the case
d'J)z(P, Q) = +00. The function d'J)z is called a semimetric. The notation for a
semimetric space rol with semimetric d911 is (rol, d!JI). For the topology of semi-
metric spaces and a discussion of possible pathologies we refer the reader to
Blumenthal (1970).
Directly from the definition it follows that the function d introduced in 16.1.2
is a semimetric.
The semimetric space (D(rol), d) will be called the space of directions of the
metric space rol.

§ 17. Curvature of a Metric Space

Here we discuss questions connected with the determination of the existence

of the curvature at a point of a metric space, we discuss the concept of continuity
of the curvature in a metric space, and we state a theorem which asserts that
"almost everywhere" the formally written sectional curvature can be calculated
by means of the ratios b(Tm)/u(Tm).

17.1. Definition of Non-isotropic Riemannian Curvature (Nikolaev (1987)).

17.1.1. Convergence of Triangles to a Point in a Given Pair of Directions.
Let (rol, p) be a metric space, P E rol, and {Tm} a sequence of triangles in rol, one
of whose vertices coincides with P. Consider a pair of directions e, ,
E Dprol
satisfying the condition
(17.1)
 II. Multidimensional Generalized Riemannian Spaces 231

A sequence of triangles {Tm = PBmCm} converges to the point P in a pair of

directions (~, 0 (notation Tm ~ P) if
a) the vertices of the triangles Tm converge (in the sense of the metric p) to the
point P as m ~ 00;
b) the directions ~m' 'm E Dprol specified by the sides am and bm of the triangle
Tm that start from P converge (in the sense of the metric (i in 3.1) to the directions
~ and, respectively.

17.1.2. Non-isotropic Riemannian Curvature. Let P be a point of the metric

space (rol, p), and ~, , E Dprol a pair of directions satisfying (17.1). We say that at
the point P of the metric space (rol, p) in the direction of the pair (~, 0 there is a
non-isotropic Riemannian curvature K(P; ~, 0 if the following conditions are
satisfied:
a) there is a sequence {Tm = PBmCm} of non-degenerate triangles in (rol, p)
(that is, (J(Tm) =f. 0) that converge to P in the pair of directions (~, 0 for which a
limit exists:

m-+oo

b) iffor a sequence of non-degenerate triangles {T~ = PB~C~} that converge

to P in the pair of directions (~, 0 there exists a limit of the quantities
(5(T~)/(J(T~), then it coincides with K(P; ~, O.
Remark 1. In part b) of the definition above we can take into consideration
degenerate triangles «(J(Tm) = 0 for them). In this case the existence of the limit
of (5(Tm)/(J(Tm) is understood in the sense that from (J(Tm) = 0 for sufficiently
large m it follows that (5(Tm) = O.
Remark 2. It may turn out that there is no sequence of non-degenerate trian-
gles that converge to P in the pair of directions (~, O. In this case we assume that
K(P; ~, 0 does not exist.
Remark 3. In the case of classical Riemannian manifolds our definition of
non-isotropic Riemannian curvature obviously gives the sectional curvature of
a Riemannian manifold at a point in the direction of the plane element specified
by the directions ~ and ,.

17.2. Existence of Curvature at a Point. The example of a convex cone in E 3 ,

considered in the intrinsic metric induced from E 3 , shows that for the metric

'E
space to be Riemannian it is insufficient to require the existence of non-isotropic
Riemannian curvature for all pairs ~, Dprol satisfying (17.1) (the "specific
curvature" of the cone at its vertex is unbounded).
In this connection, below we make more precise how we should understand
the existence of the curvature of the metric space at a point of it.
Before giving a definition of the existence of curvature we mention the
upper and lower curvatures of a metric space, which are necessary for this
definition.
232 V.N. Berestovskij and I.G. Nikolaev

17.2.1. Upper (Lower) Curvature at a Point (Aleksandrov (1957a)). Let T be

a triangle in the metric space (IDl, p). We first define the "averaged" curvature
K(T) of the triangle T (Aleksandrov (1951), p. 6):
K(T) = J(T)/u(T) when u(T) -# 0,
K(T) = +00 when u(T) = 0 and J(T) > o.
K(T) = -00 when u(T) = 0 and J(T) ~ O.
Here J(T) is the excess of the triangle T, calculated from the upper angles.
Similarly we can define the "averaged" curvature K(1): K(T) = ~s(T)/u(T)
when u(T) -# 0; K(T) = +00 when u(T) = 0 and ~s(T) ~ 0; K(T) = -00 when
u(T) = 0 and ~s(T) < O.
Here ~s(T) is the excess of the triangle T, calculated from the strong lower
angles.
We define the upper (lower) curvature of a metric space at a point Pas
K 9JI (P) = lim K(T) (Km(P) = lim K(T)),
T~P T~P

where the limits are considered over all triangles that contract to P (in the case
when there is no sequence oftriangles that contract to P we put Km(P) = +00,
Km(P) = -00.

17.2.2. Existence of the Curvature of a Metric Space at a Point. We say that

the curvature exists for a metric space (IDl, p) at a point P E 9R if the following
conditions are satisfied:
a) at the point P of the metric space ,(9Jl, p) the upper and lower curvatures
satisfy the inequalities
-00 < K m(P), K m(P) < +00.

b) for arbitrary pairs of directions e, 'E D 9R satisfying (17.1) there is a

p
non-isotropic Riemannian curvature K(P; e, O.
Remark. In the case when there is no pair of directions ~, , E Dp 9R satisfying
(17.1) we assume that the metric space (9R, p) does not have curvature at P.

17.3. Geometrical Meaning of Sectional Curvature. As we have already men-

tioned, Theorem 14.1 enables us to calculate the sectional curvature K,,(P) "at
almost all points" of a space with bounded curvature from the standard for-
mulae of "smooth Riemannian geometry". However, in view of the fact that up
to now the restrictions on the curvature and its existence have been defined by
means of the excesses of triangles, it is important to establish that the formally
introduced sectional curvature has the same geometrical meaing: it admits cal-
culation by means of the limits of the ratios J(T)/u(T).
In classical Riemannian geometry this fact is well known and, as we have
already mentioned, it lies at the basis of the construction of the generalized
Riemannian spaces considered in this article.
 II. Multidimensional Generalized Riemannian Spaces 233

The geometrical meaning for sectional curvature mentioned above is per-

served for spaces with bounded curvature. However, in view of the fact that the
metric of these spaces has substantially "worse" differential properties, in the
proof we have to overcome substantial difficulties.
Thus our main result is the following theorem.

Theorem 17.1. Let m be a space with bounded curvature. Then there is a set
(92 em of zero n-dimensional Hausdorff measure (n = dim m) that includes the
set (91 of all points in m at which the metric tensor does not have second deriva-
tives. At each point P E m\ (92 the following condition is satisfied.
For arbitrary pairs of directions ~, ( E Dpm satisfying (17.1) there is a se-
quence {Tm = PBmCm) of non-degenerate triangles that contract to P in the pair
of directions (~, 0 (see 17.1.1) such that the limit of the ratios 6(Tm)/a(Tm) exists
and

Here K,,(P) denotes the sectional curvature of m, calculated at P in the direction

of the two-dimensional element of area a c mp given by the bivector ~ 1\ (.
Remark. By Remark 2 to Theorem 14.1, in a harmonic coordinate system the
components of the metric tensor have a second differential almost everywhere.
However, this is not sufficient to prove Theorem 17.1. The set of points at which
the sectional curvature can be calculated as the limit of the ratios 6(Tm)/a(Tm)
may differ from the set of points of two-fold differentiability of the metric tensor
by a set of zero n-dimensional Hausdorff measure.
In the two-dimensional case the assertion of Theorem 17.1 follows from the
theorem on differentability of measures. The multidimensional case differs
substantially.
We also note that the assertion of Theorem 17.1 can be strengthened (see
Nikolaev (1987)). However, the statement given here is sufficient for us.

17.3.1. Plan of the Proof of Theorem 17.1. Let us introduce the following
notation: we denote the angles of the triangle Tm = PBmCmat the vertices P, Bm,
Cm by am' 13m' Ym respectively, the corresponding angles in a planar triangle with
the same lengths of sides as Tm by ao m, 130m' YOm' and we denote a(Tm) by am.
For the proof of Theorem 17.1 it is sufficient to establish the existence of the
"almost everywhere" limits
(am - aOm)/am, (13m - 130m)/am,
(Ym - YOm)/(Jm
and the fact that they are equal to KAP)/3.
In Cartan (1928) (see Chapter IX) it was shown that in order to prove that
the last of the above limits is equal to K,,(P)/3 it is sufficient to verify
234 V.N. Berestovskij and I.G. Nikolaev

that

lim
Z; -z~m. 2
= - - Ka(P),
m.... oo lTm' Xm . Ym . SID Ym 3
where Zm = PBm, ZOm = x; + Y; - 2x m' Ym' cos OCm (Xm = BmCm, Ym = PCm)·
In the case of "smooth Riemannian spaces" the investigation of the above
limit can be carried out by using Taylor's formula and the formula for second
variation of length (Synge's formula, see Gromoll, Klingenberg and Meyer
(1968)). For the triangle T = PBC we need to consider the variation V, where
Yes, t), s, t E [0, 1), is a point X lying on the shortest curve joining P and Bt at a
distance s· PBt from P, and the point Bt in turn lies on the shortest curve BC at
a distance t . BC from B.
We observe that after transformations in the remainder term of the Taylor
expansion for Z2(t) = P13" taken in integral form, under the integral there will be
differences of the components of the tensor curvature taken at the points P and
Bt respectively.
In view of the fact that in the case of a space with bounded curvature these
differences, generally speaking, do not tend to zero when P - 13, -+ 0, we use the
following construction.
We consider a pair of vectors e, e' E rolo for which e /\ e' :F O. Here 0 is a
fixed point of rol.
Let Bm and Cm denote the points given by
Bm = expo(m- 1 . e), Cm = expo(m- 1 . e').
For sufficiently large m these points are well defined.
Let S(P) and S'(P) denote the vectors in rolp obtained from the vectors e and
e' as a result of parallel translation of these vectors along the shortest curve
joining 0 and P. In a small neighbourhood of 0 such a shortest curve is unique.
Finally, let Tm(P) denote the triangle with vertices
P, Bm(P) = expp(m- 1 • S(P)), Cm(P) = expp(m- 1 • S'(P)).
If P is sufficiently close to 0, then such a triangle is well defined.
Let us introduce the function

h (P) = I z;(P) - z~m(P) + ~K (p)1

m lTm(Tm(P))' xm(P). Ym(P)' sin Ym(P) 3 a ,

where the notations zm(P) and so on have the same sense for the triangle Tm(P)
as for the triangle Tm.
Relying on the integral bounds that follow from Theorems 13.2 and 14.1, we
can prove that

from which it follows that there is a subsequence {hmJ ofthe sequence {h m } for
which lim hmk(P) = 0 as k -+ 00 for almost all P E rol; from this we obtain the
assertion of Theorem 17.1.
 II. Multidimensional Generalized Riemannian Spaces 235

In the proof that Ihml L p --+ 0 as m --+ 00 we needed to use Synge's formula for
the variation V, considered in a small domain of a space with bounded curva-
ture. This formula is deduced by means of Theorems 13.2 and 14.1. Its specific
character in the case of a space with bounded curvature is that it can be applied
only to "almost all" triangles.

17.4. Isotropic Riemannian Curvature. Let us recall that a neighbourhood

of a point of a metric space is said to be linear if it can be isometrically em-
bedded in a straight line.
We shall say that a metric space (rol, p) at a point P E rol has isotropic Rie-
mannian curvature if
a) no neighbourhood of P is linear;
b) for each sequence of triangles {Tm} that contract to P in an arbitrary way
(that is, the vertices of Tm converge to P in the sense ofthe metric p) there exists
the limit

which does not depend on the choice of the sequence of triangles that converge
to P.
In the case when there is no such sequence of non-degenerate triangles {Tm}
we suppose that there is no isotropic Riemannian curvature at P.
We also observe that as in Remark 1 in 17.1.2 we need to mention the case of
degenerate triangles separately.
The curvature R(P) was introduced in Kirk (1964):
A metric space rol with intrinsic metric has curvature R(P) at an accumula-
tion point P if it satisfies condition a) of the previous definition and
b') for each e > 0 there is a number p > 0 such that the ball with centre at P
and radius p is a domain RK-e,K+e (see 10.1), where K = R(P).
Reyling on the connection between the boundedness of curvature in terms of
the ratios ;5(T)/a(T) and the domains RK',K (for the case when the curvature is
only bounded above, see Theorem 7.1) we can assert that in metric spaces with
intrinsic metric and condition a) at an accumulation point we have K(P) = R(P)
on condition that one of these curvatures exists.

17.5. Wald Curvature and its Connection with Isotropic Riemannian Curva-
ture (Wald (1935». Let us recall that a quadruple of points ofa metric space has
curvature of immersion equal to K if it is isometric to some quadruple in SK'
A metric space rol has Wald curvature Kw(P) at an accumulation point P ifrol
satisfies condition a) of 17.4 and
b") for each e > 0 there is a number p > 0 such that each quadruple of points
Q of the ball in M with centre P and radius p has curvature of immersion K(Q)
and
IK(Q) - Kw(P) I < e.
236 V.N. Berestovskij and I.G. Nikolaev

A
1\
/ \
/ \
/ \
/ \
/ \
/ \
/ \
~--------~
BCD
Fig. 16

In Wald (1935) it was shown that a non-linear quadruple of distinct points

has two unequal curvatures of immersion, while a quadruple of distinct points
that contains a linear triple of points (that is, a triple of points such that the
distances between them "form a degenerate triangle", see Fig. 16) has no more
than one curvature of immersion.
Wald weakened the original definition of the curvature Kw(P) by admitting
only quadruples of points that contain a linear triple of points. The resulting
curvature is also called Wald curvature and denoted by Kw(P).
In Kirk (1964) it was shown that Kw(P) = R(P). Thus for metric spaces with
intrinsic metric and with condition a) of 7.4 the existence of isotropic Rieman-
nian curvature at an accumulation point is equivalent to the existence of Wald
curvature Kw(P) at this point, and we have
K(P) = Kw(P).

17.6. Continuity of Curvature. Let (IDl, p) be a metric space. We say that the
curvature of (IDl, p) is continuous at a point P E IDl if
a) there is a number rp > 0 such that for all points Q E B(P, rp) c IDl of(IDl, p)
the curvature exists at Q (see 17.2.2);
b) for an arbitrary e> 0 we can find a number ~, 0 < ~ < rp, such that for
all points Q belonging to B(P,~) and all pairs of directions (E DpIDl, e,
e', ('
E DQIDl satisfying the inequality

maxi {d(e, n, d«(, C')} < ~

and condition (17.1) we have
IK(P; e, C) - K(Q; e', (')1 < e.
The continuity of the curvature of (IDl, p) on a subset d of IDl implies its
continuity at each point P E d.
The curvature of(IDl, p) satisfies a Holder condition with exponent 0( (0 < 0( < 1)
and constant L on a subset d £; IDl if it is continuous at each point P E d and
if the non-isotropic Riemannian curvature has the bound
 II. Multidimensional Generalized Riemannian Spaces 237

for arbitrary P, QEd and all pairs of directions ~, 'E DpIDl, ~', " E DQIDl
satisfying the condition (17.1).

§ 18. Axioms of Riemannian Geometry

More precisely speaking, it is a question of the axioms of classical Rieman-

nian geometry. We give a list of axioms of synthetic character that distinguish
among general metric spaces c m,,,,_ and COO-smooth Riemannian manifolds
(m = 2,3, ... , (X E (0, 1)).

18.1. Synthetic Description of Cl''''·Smooth Riemannian Manifolds.

Theorem 18.1. Let (IDl, p) be a metric space for which the following conditions
are satisfied:
1) p is the intrinsic metric;
2) (IDl, p) is a locally compact metric space;
3) in (IDl, p) the condition of local extendability of shortest curves is satisfied
(see 12.1);
4) the curvature exists at each point of (IDl, p) (see 17.2.2);
5) the curvature of (IDl, p) satisfies a Holder condition with exponent (X E (0, 1)
in a small neighbourhood of an arbitrary point of IDl (see 17.6).
Then we can specify the structure of a C4 ''''-smooth differentiable manifold on
IDl, and we can specify the metric p with respect to the charts of this structure by
means of a C 2 ''''-smooth metric tensor.
Remark 1. The above conditions are obviously necessary in order that the
given metric space should be a C 2 ''''-smooth Riemannian manifold.
Remark 2. The C4 ''''-smooth differentiable structure on IDl that occurs in The-
orem 18.1 contains an atlas formed by harmonic coordinate systems on (IDl, pl.
The metric tensor that specifies p has maximal possible smoothness.
Remark 3. In Peters (1987) (see p. 14) an example is constructed from which
it follows that, when (X = 0, in the harmonic coordinate system in (IDl, p) the
components ofthe metric tensor that specifies p may have discontinuous second
derivatives.
In this case, however, we can assert that in harmonic coordinates the metric
tensor of (IDl, p) belongs to the Sobolev class Wp2 for each 1 ~ p < +00, and the
definition of the curvature tensor of (IDl, p) is extended to a tensor specified at
each point of IDl whose components with respect to the harmonic coordinates
are continuous functions. Thus when (X = 0 we can assert that (IDl, p) is a Rie-
mannian manifold with continuous curvature tensor.

18.1.1. Plan of the Proof of Theorem 18.1. From the conditions of Theorem
18.1 it follows that (IDl, p) is a space with bounded curvature. By Theorem 17.1
238 V.N. Berestovskij and I.G. Nikolaev

the non-isotropic Riemannian curvature "at almost all points" of IDl coincides
with the sectional curvature, calculated formally from the metric tensor. In
particular, the sectional curvature also has a Holder bound.
Now, reverting to equation (15.4), we can make use of a theorem on the
smoothness of solutions of elliptic equations (Ladyzhenskaya and UraI'tseva
(1964». The existence of the required differential structure follows from the
results of Sabitov and ShefeI' (1976).

IS.2. Synthetic Description of C .., R-Smooth Riemannian Manifolds (m =

3, 4, ... ). As a preliminary we introduce the concepts of the tangent space T(IDl)
for an arbitrary metric space (IDl, p) and the i-th tangent space T(i)(IDl), which
play an important role in the synthetic description of Cm,lI-smooth Riemannian
manifolds.

IS.2.1. The Tangent Space T(IDl) and the Spaces T(i)(IDl).

The set
T(IDl) = U IDl p , P E IDl
is called the set of tangent elements of the metric space (IDl, p) (see 3.3). As in
16.1.2, for [~, s] E IDl p , [(, t] E IDlQ we introduce the function

<1>([~, s], [(, t]) = {sup

qe Up!Dl
[s· h(~, ,,) - t· h«(, ,,)]

+ sup [s'
q' e U(!!Dl
h(~, ,,') - t· h«(, ,,')]}/2.

The semimetric in T(IDl) between the tangent elements [~, s] E IDl p , [(, t] E
IDlQ is introduced by means of the constant c:
c([~, s], [(, t]) = [PQ2 + <1>2([~, s], [(, t])]1/2. (18.1)
We call the semimetric space (T(IDl), c) the tangent space of the metric space
(IDl, pl.
We now go over to the definition of the spaces T(i)(IDl).
We denote IDl by T(O)(IDl) and p by Co = Ilo. We denote the semimetric space
(T(IDl), c) by (T(1)(IDl), c l ). We assume that the semimetric C I induces the metric
III in T(1)(IDl) (see 16.1.1), that is, for an arbitrary pair of points of T(l)(IDl) the
infimum of the lengths of all curves (measured in the semimetric cl)joining these
points is defined and is a finite quantity, and it specifies a metric in T(l)(IDl).
Then for the metric space (T(I)(IDl), Ill) we can consider the semimetric space
(T(2)(IDl), c2), where T(2)(IDl) = T(T(l)(IDl», and the semimetric C2 is determined
from the metric Il~ by means offormula (18.1).
Suppose we have constructed the spaces (T(i)(IDl), Ci), i = 0, 1, 2, ... , k - 1,
and the semimetrics Ci induce the metrics Ili' i = I, 2, ... , k - 2. If Ck-l induces
the metric Ilk-I' then for the metric space (T(k-I)(IDl),llk_l ) we can again con-
sider the semimetric space (T(k)(IDl), Ck), where T(k)(IDl) = T(T(k-I)(IDl», and the
semimetric Ck is determined from the metric Ilk - l in accordance with (18.1).
 II. Multidimensional Generalized Riemannian Spaces 239

18.2.2. Statement of the Main Result. As a preliminary we introduce the

following concepts.
Let (IDl, p) be a metric space. We shall say that (IDl, p) has HOlder continuous
curvature with exponent <X E (0,1) if(IDl, p) satisfies conditions 4) and 5) of Theo-
rem 18.1.
If (IDl, p) satisfies all the conditions of Theorem 18.1, then from this theorem,
and also from the bounds of Proposition 16.1, we can deduce that the semi-
metric C 1 induces the Sasaki metric ~l in T(1)(IDl) (see 16.1). If in turn the metric
space (T(1)(IDl), ~1) has Holder continuous curvature, then similarly we deduce
that the semimetric C 2 induces in T(2)(IDl) the Riemannian metric ~2' which
coincides with the Sasaki metric for T(T(1)(IDl)), and so on.
Thus for a metric space (IDl, p) that satisfies the conditons of Theorem 18.1 we
can give the following definitions.
{T(i)(IDl)}, i = 1, 2, ... , m, is a sequence of spaces with Holder continuous
curvature with exponent <X (0 < <X < 1) if there is a sequence of numbers <Xi
(0 < <Xi < 1), i = 1, 2, ... , m, <Xm = <X, such that the metric spaces (T(i)(IDl), ~),
i = 1,2, ... , m, have Holder continuous curvarture with exponent <Xi'
In a similar way we define what is meant by an infinite sequence of spaces
{T(i)(IDl)}, i = I, 2, ... , being a sequence of spaces with Holder continuous
curvature.

Theorem 18.2. Let (IDl, p) be a metric space for which conditions 1)-5) of
Theorem 18.1 are satisfied, and suppose that in addition the following condition is
satisfied:
6) {T(i)(IDl)}, i = 1, 2, ... , m, is a sequence of spaces with Holder continuous
curvature with exponent <X E (0, 1).
Then on IDl we can specify the structure of a Cm+4.rz-smooth differentiable
manifold, and we can specify a metric p with respect to the charts of this structure
by means of a Cm+2.rz-smooth metric tensor, m = 1, 2, 3.
Corollary 18.1. If we replace condition 6) in Theorem 18.2 by the condition
6') {T(i)(IDl)}, i = 1,2, ... , is an infinite sequence of spaces with HOlder con-
tinuous curvature, then we can specify on IDl the structure of a COO-smooth differ-
entiable manifold, and we can specify a metric p with respect to the charts of this
structure by means of a COO-smooth metric tensor.

Remark 1. The conditions in Theorem 18.2 and Corollary 18.1 are also neces-
sary in order that a given metric space should be a Cm+2.rz-smooth or respec-
tively Coo -smooth Riemannian manifold.

Remark 2. The differential structure on IDl that occurs in Theorem 18.2

and Corollary 18.1 contains an atlas formed by harmonic coordinate systems
in IDl.

Remark 3. The proof of Theorem 18.2 is obtained as a consequence of Theo-

rem 18.1 and the results of Sasaki (1958), (1962).
240 V.N. Berestovskij and I.G. Nikolaev

18.3. Isotropic Metric Spaces. For isotropic metric spaces, that is, for metric
spaces at each point of which there is isotropic Riemannian curvature (see 17.4)
we have the following theorem, which extends the well-known theorem of Schur
to the case of metric spaces.
Theorem 18.3 (Nikolaev (1989». Let (IDl, p) be a locally compact metric space
with intrinsic metric, whose Menger-Uryson dimension is greater than two. We
assume that the condition of local extendability of shortest curves is satisfied in IDl
and that IDl is isotropic at all its points. Then (IDl, p) is isotropic to a Riemannian
manifold of constant curvature.

18.3.1. The Consequence for "Multidimensional" Metric Spaces at each Point

of which the Wald Curvature K,,(P) Exists. The next result follows directly from
Theorem 18.3 and the results of 17.5.
Theorem 18.4. Let (IDl, p) be a locally compact metric space with intrinsic
metric, whose Menger-Uryson dimension is greater than two. We assume that the
condition of local extendability of shortest curves is satisfied in IDl and that the
Wald curvature Kw(P) exists at each point P E IDl. Then (IDl, p) is isometric to a
Riemannian manifold of constant curvature.
Remark. Theorem 18.4 gives the answer to a question in Kirk (1964).

18.3.2. Plan of the Proof of Theorem 18.3. The standard proof of Schur's
theorem is based on the use of the Bianchi identity. Hence we assume that the
components of the metric tensor of a Riemannian manifold are at least thrice
differentiable. From the fact that the metric space is isotropic (and the remain-
ing conditions of the theorem) we can deduce that (IDl, p) is a space with
bounded curvature. Therefore by Theorem 14.1 the components of the metric
tensor of IDl in harmonic coordinates belong to the Sobolev class w,,2
(I :::; p < +(0). Thus we need to modify the proof of Schur's theorem in order to
bring into consideration the derivatives of a metric tensor of order not greater
than two.
This is achieved as follows. As we know (Cartan (1928» Bianchi's identity
implies that the absolute extrinsic derivative of the curvature form of a manifold
is equal to zero. As for the usual extrinsic differentation, for the absolute differ-
ential we can introduce the concept of generalized differentiation, and this in
turn enables us to rewrite Bianchi's identity in a generalized sense, not using the
third derivatives of the metric tensor. The truth of Bianchi's identity on (IDl, p)
is established by means of Theorem 14.1. Using the isotropy of the metric,
Bianchi's identity on (IDl, p) reduces to the vanishing of the derivative of the
distribution on IDl defined by K (Hormander (1983»:

fQ
o<p
K(x)~(x) dx 1 dx 2
UX k
... dx" = 0, k = 1,2, ... , n, (18.2)

where x: o/J c IDl -+ IR" (x(OlI) = Q) is an arbitrary chart on IDl, and ({J is an arbi-
trary smooth function compactly supported in Q.
 II. Multidimensional Generalized Riemannian Spaces 241

By virtue of the continuity of the function K(x) (this follows from the defini-
tion of isotropic curvature as in the two-dimensional case in Aleksandrov
(1948» and Theorem 3.1.4' in Hormander (1983), in view of the arbitrariness of
the chart x we deduce that the function K(x) is identically constant on 9)1
From a theorem of Aleksandrov (see Aleksandrov (1957b» it then follows
that (9)1, p) is isometric to a Riemannian manifold of constant curvature, as
required.

References *

Aleksandrov, A.D. [1948]: Intrinsic Geometry of Convex Surfaces. Gostekhizdat, Moscow-

Leningrad, ZbI.38,352. German transl.: Akademie Verlag, Berlin 1955, Zb1.65,151
Aleksandrov, A.D. [1951]: A theorem on triangles in a metric space and some applications of it. Tr.
Mat. Inst. Steklova 38, 5-23 [Russian], Zb1.49,395
Aleksandrov, A.D. [1957a]: Ruled surfaces in metric spaces. Vestn. Leningr. Univ. 12, No. 1,5-26
[Russian], Zb1.96,166
Aleksandrov, A.D. [1957b]: Uber eine Verallgemeinerung der Riemannschen Geometrie. Schr.
Forschungsinst. Math. 1,33-84, Zb!.77,357
Aleksandrov, A.D. and Berestovskij, V.N. [1984]: Generalized Riemannian space. Mathematical
Encyclopaedia, vo!' 4, 1022-1026. English transI.: Kliiwer Acad. Pub!., Dordrecht.
Aleksandrov, A.D., Berestovskij, V.N. and Nikolaev, I.G. [1986]: Generalized Riemannian spaces.
Usp. Mat. Nauk 41, No.3, 3-44. English transl.: Russ. Math. Surv. 41, No.3, I-54 (1986),
Zb!.625.53059
Aleksandrov, A.D. and Zalgaller, V.A. [1962]: Two-dimensional manifolds of bounded curvature.
Tr. Mat. Inst. Steklova 63. English transl.: Intrinsic geometry of surfaces. Trans!. Math. Mono-
graphs 15, Am. Math. Soc. (1967), Zb1.l22,170
Berestovskij V.N. [1975]: Introduction of a Riemannian structure in certain metric spaces. Sib. Mat.
Zh. 16, 651-662. English trans!': Sib. Math. J. 16,499-507 (1976), Zb!.325.53059
Berestovskij, V.N. [1981]: Spaces with bounded curvature. Dok!. Akad. Nauk SSSR 258, 269-271.
English trans!': SOy. Math., Dok!. 23, 491-493 (1981), Zb!.511.53074
Berestovskij, V.N. [1986]: Spaces with bounded curvature and distance geometry. Sib. Mat. Zh. 27,
No. I, 11-25. English trans!.: Sib. Math. J. 27, 8-19 (1986), Zb!.596.53029
Berger, M. [1983]: Sur les varietes Riemanniennes pincees juste au dessous de 1/4. Ann. Inst.
Fourier 33, 135-150, Zb!.497.53044
Bers, L., John, F. and Schechter, M. [1964]: Partial Differential Equations. Interscience, New York,
Zb1.l28,93 and Zb1.l28,94
Blumenthal, L.M. [1970]: Theory and Applications of Distance Geometry. 2nd. ed., Chelsea Pub!.
Co., New York, Zb!'50,385
Busemann, H. [1955]: The Geometry of Geodesics. Academic Press, New York-London,
Zbl.l12,370
Cartan, E. [1928]: Leyons sur la geometrie des espaces de Riemann. 2nd ed., Gauthier-Villars, Paris,
Jbuch 54,755
Cesari, L. [1956]: Surface Area. Princeton Univ. Press, Princeton, N.J., Zb!.73,41

* For the convenience of the reader, references to reviews in Zentralblatt fUr Mathematik (Zb!.),
compiled using the MATH database, and Jahrbuch iiber die Fortschritte der Mathematik (Jbuch)
have, as far as possible, been included in this bibliography.
242 V.N. Berestovskij and I.G. Nikolaev

DeTurck, D.M. and Kazdan, J.L. [1981]: Some regularity theorems in Riemannian geometry. Ann.
Sci. Ec. Norm. Super., IV. Ser. 14,243-260, Zb!.486.53014
Einstein, A. [1916]: Niiherungsweise Integration der Feldgleichungen der Gravitation. Sitzungsber.
Preussische Akad. Wiss. 1,688-696, Jbuch 46,1293
Gol'dshtein, V.M., Kuz'minov, V.I. and Shvedov, I.A. [1984]: A property of de Rham regularization
operators. Sib. Mat. Zh. 25, No.2, 104-111. English trans!.: Sib. Math. J. 25, 251-257 (1984),
Zb!.546.58002
Gol'dshtein, V.M. and Reshetnyak, Yu.G. [1983]: Introduction to the Theory of Functions with
Generalized Derivatives and Quasiconformal Mappings. Nauka, Moscow. English trans!.:
Quasiconformal Mappings and Sobolev's spaces. Math. and its App!., Sov. Ser. 54. Kluwer
Acad. Pub!., Dordrecht 1990, Zb!.591.46021
Gromoll, D., Klingenberg, W. and Meyer, W. [1968]: Riemannsche Geometrie im Grossen. Lect.
Not. Math. 55, Springer, Berlin-Heidelberg-New York, Zb1.155,307
Gromov, M. [1981]: Structures metriques pour les varietes Riemanniennes. Cedic, Paris,
Zb!.509.53034
Hiirmander, L. [1983]: The Analysis of Linear Partial DilTerential Operators. I. Distribution The-
ory and Fourier Analysis. Springer, Berlin-Heidelberg-New York, Zb!.521.35001
Hurewicz, W. and Wallman, H. [1941]: Dimension Theory. Princeton Univ. Press, Princeton, N.J.,
Zb!'60,398
Ionin, V.K. [1972]: Isoperimetric inequalities in simply-connected Riemannian spaces of non-
positive curvature. Dokl. Akad. Nauk SSSR 203, 282-284. English trans!.: Sov. Math., Dokl.
13,378-381 (1972), Zb!.258.52011
Kirk, W.A. [1964]: On the curvature of a metric space at a point. Pac. 1. Math. 14, 195-198,
Zb1.168,434
Ladyzhenskaya, O.A. and Ural'tseva, N.N. [1964]: Linear and Quasilinear Elliptic Equations.
Nauka, Moscow. English trans!.: Academic Press, New-York-London 1968, Zb1.143,336
Nikolaev, I.G. [1978]: Space of directions at a point in a space of curvature not greater than K. Sib.
Mat. Zh. 19, 1341-1348. English trans!.: Sib. Math. 1. 19,944-949 (1979), Zb!.405.53044
Nikolaev, I.G. [1979]: Solution of the Plateau problem in spaces of curvature not greater than K.
Sib. Mat. Zh. 20, 345-353. English transl.: Sib. Math. J. 20, 246-252 (1979), Zbl.406.53044
Nikolaev, I.G. [1980]: Parallel translation and smoothness of the metric of spaces with bounded
curvature. Dokl. Akad. Nauk SSSR 250, 1056-1058. English trans!.: Sov. Math., Dokl. 21,
263-265 (1980), Zb!.505.53015
Nilolaev, I.G. [1983a]: Parallel translation of vectors in spaces with curvature that is bilaterally
bounded in the sense of A.D. Aleksandrov. Sib. Mat. Zh. 24, No. I, 130-145. English trans!.:
Sib. Math. J. 24,106-119 (1983), Zb!.539.53030
Nikolaev, I.G. [1983b]: Smoothness of the metric of spaces with curvature that is bilaterally
bounded in the sense of A.D. Aleksandrov. Sib. Mat. Zh. 24, No.2, 114-132. English trans!.:
Sib. Math. 1. 24, 247-263, Zb!.547.53011
Nikolaev, I.G. [1987]: The curvature ofa metric space at a point. All-Union Conf. on Geometry "in
the large". Novoskbirsk [Russian]
Nikolaev, I.G. [1989]: Isotropic metric spaces. Dok!. Akad. Nauk SSSR 305, 1314-1317. English
trans!.: Sov. Math., Dokl. 39,408-410, Zb!,714.54031
Peters, S. [1986]: Konvergenz Riemannscher Mannigfaltigkeiten. Bonner Math. Schr. 169,
Zb!.648.53024
Peters, S. [1987]: Convergence of Riemannian manifolds. Compos. Math. 62, 3-16, Zb!.618,53036
Reshetnyak, Yu.G. [1960a]: Isothermal coordinates in manifolds of bounded curvature. I, II. Sib.
Mat. Zh. 1, 88-116, 248-276 [Russian], Zb1.108,338
Reshetnyak, Yu.G. [1960b]: On the theory of spaces of curvature not greater than K. Mat. Sb.,
Nov. Ser. 52, 789-798 [Russian], Zb1.101,402
Reshetnyak, Yu.G. [1968]: Non-expanding maps in spaces of curvature not greater than K. Sib.
Mat. Zh. 9,918-927. English trans!': Sib. Math. J. 9, 683-689, Zb1.167,508
Rham de, G. [1955]: Varietes dilTerentiables: Formes courants, formes harmoniques. Herrmann,
Paris, Zb!.65,324
 II. Multidimensional Generalized Riemannian Spaces 243

Sabitov, I.Kh. and Shefel', S.Z. [1976]: Connections between the order of smoothness of a surface
and that of its metric. Sib. Mat. Zh. 17, 916-925. English transl.: Sib. Math. J. 17, 687-694,
Zb1.386.53014
Sasaki, S. [1958]: On the differential geometry of tangent bundles of Riemannian manifolds. I.
Tohoku Math. J., II. Ser. 10,338-354, Zb1.86,150
Sasaki, S. [1962]: On the differential geometry of tangent bundles of Riemannian manifolds. II.
Tohoku Math. J., II. Ser. 14, 146-155, Zbl.109,405
Sobolev, S.L. [1950]: Some Applications of Functional Analysis to Mathematical Physics. Izdat.
Leningrad. Univ., Leningrad [Russian]
Toponogov, VA [1959]: Riemannian spaces of curvature bounded below. Usp. Mat. Nauk 14, No.
1,87-130. English transl.: Transl., II. Ser., Am. Math. Soc. 37, 291-336 (1964), Zb1.114,375
Wald, A. [1935]: Begriindung einer koordinatenlosen Differentialgeometrie der Flachen. Ergeb.
Math. KolIoq., Vol. 7, 24-46, Zb1.14,230
 Author Index

Aleksandrov, A.D. 1,7,9,10, 16,20,22,63, Levi-Civita, T. 210,215,227

64,66,71,77,140,148,153,156-158,169, Lewy, fI. 63
170,172,174,177,185,194,201-207,212, Lipschitz, L. 209,213,214,223
241
Menger, K. 226, 240
Bakel'man, I.Ya. 155 Men'shov, D.E. 119
Berestovskij, V.N. 159,174
Berger, M. 170,224 Nikolaev, I.G. 159, 174, 175
Bianchi, L. 240
Burago, Yu.D. 124,145-147,152,175 Peters, S. 225
Pogorelov, A.V. 64,157, 158
Carleman, T. 172
Cartan, E. 211 Radon, J. 108
Cayley, A. 224 Reshetnyak, Yu.G. 7
Rham de, G. 221
DeTurck, D.M. 217 Ricci, C. 219

Einstein, A. 217 Sabitov,I.Kh. 217

ShereI', S.Z. 158,217
Euclid 169
Sobolev, S.L. 173,219-221,237
Stratilatova, M.B. 124, 152
Fiala, F. 148
Strel'tsov, V.V. 148
Gromov, M.L. 220
Taylor, J.C. 234

flausdorff,F. 14,174,223,233 Uryson, P.S. 226, 240

fliilder, E. 171, 173,217,236-239
Wald, A. 171,235,236,240
Kazdan, J.L. 217 Weyl, fl. 63, 64
Klingenberg, W. 224
Kolmogorov, A.N. 198 Zalgaller, V.A. 175
 Subject Index

Angle 176 - of compact Riemannian manifolds

- between curves 73 9Jl(n, d, A, io) 222
- in the strong sense 177 - of Sobolev Wp 237
-, left 61 Completion of a space, Hausdorff 14
-, lower, between curves 73 Complex of a surface, 196
-, -, strong 177, 182 - of a surface, vertex of 196
- of a domain 46 K-concavity 172,186,187
-, polar 32 Condition of Holder 103, 104,236
-, right 61 - of Lipschitz 213,214
-, total 53,124 - of local extendability of shortest curves
-,-,ofacircularcone 51 207
-, -, of a polyhedron 54 - of non-overlapping of shortest curves 201
-, upper, between curves 73, 169, 177 Cone, circular 51
-, -, strong 177 -,-,finite 51
Arc of class J 124 -,convex 64,94
-, oriented 34 -,finite 50
-, simple 11 -, generator of 50
-, unilaterally smooth 107 - of class K 68
Area of a set 43, 58, 134 Cone of directions 195
- of a surface 196 - of directions, angle of 195
K-area 178 - over a metric space 50
Atlas of class Coo 36 - over a metric space, vertex of 50
Averaging ofa metric, Sobolev 219,221 -,tangent 156,181
Axiom of boundedness of curvature 75 K-cone 180
Axioms of Riemannian geometry 237 Connection, Levi-Civita 215,227
Continuity of curvature 236
Ball, closed 14 Convergence of a triangle in a pair of
-, open 13 directions 231
Band, Mobius 24 -, proportional 156
Boundary 19 K' -convexity 200
- rays of a cone 51 Coordinate system, Chebyshev 154
- -, distance 209
Characteristic, Euler 22, 59 - -, equidistant 155
Chart 36 - -, harmonic 217
-, admissible 36 - - , isothermal 7,48,97,101,159
-, basic 36 - - , local 36
-, -, of a Riemann surface 102 - - , special 19
Charts, overlapping 36 Coordinates, harmonic 217
- oriented in the opposite way 37 Cross-section of a cone 50
- - in the same way 37 Curvature, absolute 59
-, smoothly compatible 36 -, absolute integral 45
Circle 14, 151 -, averaged, of a triangle 232
Circular sector 51 -, boundedness of, in terms of parallel
Class of compact Riemannian manifolds translation 212
9Jl(n, d, A, V) 222 -, continuous, HOlder 239
 Subject Index 247

-, Gaussian, 7,44,47,68,72, 103, 116, 140, -, angular 50

141, 150, 158 -, boundedly convex 80
-, integral 45,49 -, completely convex 80
-, isotropic, Riemannian 226, 235 -,convex 80
-, left 44 - of definition of a chart 36
-, lower 232 -, simple 20, 45
-, non-isotropic, Riemannian 225,231 -, standard 19
-ofa cone 51 -, standard, centre of 19
- of a curve, left 44 -PK 188
- of a curve, right 44 - RK 185, 186, 190
- of a manifold 53, 124 -RK',K 199
- of a polyhedron at a point 53
- of a set, integral 45 End-points of an arc 12
- of a triangle K(T) 232 Equation, Poisson 49, 103
- of a triangle K(T) 232 Equidistant 151
- of immersion of a quadruple 235 Excess ofa triangle 47,72,178
- ofWald Kw(P) 235 - of a triangle, absolute 178
- of Wald K~(P) 236,240 - of a triangle, upper 75
-, negative part of 59 K-excess 178
-, positive part of 59 Existence of a direction of a curve at a
-, right 44 point 180
-R(P) 235 - of curvature at a point 231,232
-, upper 232
Curve lying to the left 34 Face of a triangulation 21
-lying to the right 34 K-fan 189
- of class .d 132 -, generalized 189
-, oriented 32,33 Formula of Euler 59
-, negatively oriented 31,33 - of Gauss and Bonnet 43, 45, 46, 72, 85, 86,
-, positively oriented 31,33 96,133
-, parametrized 11 - of Synge 234
-, simple 12 Function of bounded variation 133
-, simple closed II -, subharmonic 99
-, simple closed, oriented 33 Functions, transition 36
- with bounded rotation 107
- with bounded variation of turn 133,
Geodesic 8
149
- polygonal line 47, 131
Cutting a space with respect to a set 17
- polygonal line, vertex of 47
- of manifolds 30
- triangle 47
Derivative, covariant, of a vector field 215
-, left 107 Half-tangent, left 124
-, right 107 -, right 124
Development of a triangle 78, 140 Homeomorphism, left 134
Differential, quadratic 102 -, right 34
Dilatation 223
Direction 180 Image, spherical, of a point 55
- of a curve at a point 74 Index of a point with respect to a closed
Disc 14 path 33
Distance between directions 226, 229 Inequality, isoperimetric 198
- between points 39 Internal diameter of a set 30
- geometry 193 . Interval of a generator of a cone SO, 5 I
-, Hausdorff 223 - on a polyhedron 57
-, Lipschitz 223
Domain, absolutely convex 80 Jacobian of a map 37
248 Subject Index

Lebesgue measure of a set 58 Parametrization, left 34

Length of a curve 175 - of a simple arc 11
- of a path 12, 39 - , right 34
- of a path in a Riemannian manifold 39 Pasting rule 26
- of a simple arc 12 Path 11
- of a simple arc with respect to a line - , piecewise smooth, of class C' 38
element 100 - , rectifiable 13
-,simple 11
Manifold of bounded curvature (two- - , smooth, of class C' 37,38
dimensional) 7, 10,75 Plane, Euclidean, arithmetic 11
- of type E 104 - of Lobachevskij 39, 68, 69
- , orientable 37 - , supporting, of a polyhedron 55
- , Riemannian, two-dimensional 6,8 Point, admissible 16
- , twice covered 30 - at infinity with respect to a line element
- , two-dimensional, of specific curvature 100
bounded above 154 - , boundary 19
- , vertex of 53 - , interior 12, 19
- with boundary, two-dimensional 19 - , peak 124
Map, conformal 101 - , transit 80
- , contracting 65 Polygonal line 57
- , exponential 208 - - , links of 106
- , non-expanding 192 - - on the plane C 106
Measure, Hausdorff 100 - - , simple 57
- , Lebesgue 58,212 - - , vertices of 106
- , negative part of 99 Polyhedron, boundary vertex of 54
- , positive part of 98 - , complete 54
Method, analytic 7 - of class K 70
- , approximative 6 - , supporting plane of 55
- , axiomatic 6,169 K-polyhedron 70
- , synthetic 169 Principle of Kolmogorov 197
Metric 11,39 - of uniformization 102
- , extrinsic 181 Problem of Plateau 172, 198
- , induced intrinsic 14 -ofWeyl 63
- , intrinsic 14,175,181,195
- , natural, of a Riemannian manifold 39 Quadruple of points, exceptional 193
- of Sasaki 226-229,239 - of points, ordinary 193
- , polyhedral 53,76,204 Quasigeodesic 153
- , Riemannian 76,109,117,120,121,129
- , subharmonic 101 Radius of injectivity 202
- , polar 49
Neighbourhood, canonical 54, 70 Range of a chart 36
- , linear 235 Representation of a map 102
- of an arc, canonical 34 Rotation of a curve, absolute 107
- of a point 18,76 - of a polygonal line 106
- , special 19 - of a polygonal line, absolute 106
T-neighbourhood of a point 56
Semicontinuity of area 197
Operation of cutting 16 Semimetric 230
- of pasting (gluing) 22, 188, 189 - in T(IDl) 238
Operator, averaging, de Rham 221 Semineighbourhood, left 35
- , averaging, Sobolev 221 - , right 35
- of Laplace 218 Sequences of spaces {(T(;'(IDl»} with Holder
continuous curvature 239
Parallel translation 210-213 Set, Borel 42
 Subject Index 249

--,bounded 42,58 -- of A.D. Aleksandrov 206

--,convex 176 -- of Berger on rigidity 170, 224
--, measurable 58 -- of Borel 21, 54, 58, 82, 122
--, metrically connected 13 -- of Brittain 225
-- of tangent elements 238 -- of Brouwer 19
Shortest curve 16, 176, 195 -- of Busemann and Feller 198
Side ofa simple arc 21,31 -- of Cheeger 224
-- of a triangle 74 -- of Cohn-Vossen 154
Smoothness of a metric 220 -- of Gauss and Bonnet 178-192
Space, Banach 73 -- of Gauss and Bonnet for polyhedra 60,
--, finitely compact 203 71
--, generalized Riemannian 168, 170 -- of Gromov on compactness 223
--, Hausdorff 18 -- of Jordan 21
--, Hausdorff completion of 14 -- of Klingenberg 224
--, Hilbert 73 -- of Levi-Civita 210
--, linearly connected 11 -- of Lewy 64
--, metrically connected 13, 176 -- of Lyusternik and Shnirel'man 154
--, metric, complete 14 -- of Meyer 224
--, metric, isotropic 226,240 -- of Pogorelov 71
-- of curvature"" K 185, 202 -- of Rado 21
-- of curvature ;;. K' 205 -- of Schoenflies 21,31,35
-- of curvature "" K and ;;. K' 198, 199 -- of Schur 240
-- of directions 179,181,183,226,230 -- of Sobolev on embedding 218,224
-- of functions W~, IV!, C" 216,217 -- of Toponogov 206
--, semimetric 230 -- of Whitney 221
--, Sobolev 216 -- on contraction of a cone 135,147
--,tangent 181,183,196,208,238 -- on convergence of metrics 119
-- with bounded curvature 173, 206, 208 -- on pasting 56
-- with intrinsic metric 13 -- on triangulation 55
-- SK' SiC 176 Triangle 8, 74, 176
-- T(i)(!Dl) 238 --, area of 178
Sphere 14 --, K-area of 178
Surface, Mobius 25 --, geodesic 47
--, non-parametrized 196 -- homeomorphic to a disc 75
-- of bounded extrinsic curvature 157 --, perimeter of 176
--, parametrized 196 --, side of 176
-- representable by the difference of convex --, simple 75
surfaces 157 --,surface 195,198
--, Riemann 102 --, topological 21
--, ruled 197 --, vertex of 74,176
--, saddle 158 Triangles, non-overlapping 75
Symmetry with respect to a point 210 Triangulation of a disc 196
Synthetic description of C 2 "-smooth -- of a manifold 21
Riemannian manifolds 237 --, edge of 21
-- -- of Cm"-smooth Riemannian manifolds --, face of 21
238 --, vertex of 21
Turn of a curve, left, at a point 128
Tensor, metric, of a Riemannian manifold -- of a curve, right, at a point 128
39 -- of an arc, left (right) 44,47
Theorem, approximation, first 76 -- of a polygonal line, left 61,62
--, approximation, second 77,94 -- of a polygonal line on a polyhedron 61
--, comparison, first 135 -- of a polygonal line, right 61, 62
--, comparison of angles 172 -- of the boundary of a polyhedron 54
--, comparison, second 138 -- of the boundary on a set 60
250 Subject Index

Variation of a measure, lower Vector, outward normal 55

99 Vertex oca partition 85
- of a measure, total 99
- of a measure, upper 98 Winding oCan arc 153
 Encyclopaedia of Mathematical Sciences
Editor-in-Chief: R. V. Gamkrelidze

Geometry Volume 55: A.N.Parshin, 1.R.Shafarevich (Eds.)

Algebraic Geometry IV
Volume 28: R. V.Gamkrelidze (Ed.) Linear Algebraic Groups.
Invariant Theory
Geometry I
1993. Approx. 300 pp. 9 figs. ISBN 3-540-54682-0
Basic Ideas and Concepts
of Differential Geometry
1991. VII, 264 pp. 62 figs. ISBN 3-540-51999-8
Number Theory
Volume 29: E.B. Vinberg (Ed.)
Geometry II Volume 49: A.N.Parshin, I.R.Shafarevich (Eds.)
Geometry of Spaces Number Theory I
of Constant Curvature Fundamental Problems, Ideas and
1993. IX, 254 pp. 87 figs. ISBN 3-540-52000-7 Theories
1994. Approx. 340 pp. 16 figs.
Volume 48: Yu. D. Burago, V.A.Zalgailer (Eds.) ISBN 3-540-53384-2
Geometry III
Theory of Surfaces Volume 62: A. N.Parshin, I.R.Shafarevich (Eds.)
1992. VIII, 256 pp. ISBN 3-540-53377-X Number Theory II
Algebraic Number Theory
1992. VI, 269 pp.ISBN 3-540-53386-9

Algebraic Geometry
Volume 60: S.Lang
Volume 23: I.R.Shafarevich (Ed.) Number Theory III
Diophantine Geometry
Algebraic Geometry I
Algebraic Curves. Algebraic Manifolds 1991. XIII, 296 pp. 1fig. ISBN 3-540-53004-5
and Schemes
1993. Approx. 330 pp. 9 figs. ISBN 3-540-51995-5

Volume 35: I.R.Shafarevich (Ed.)

Algebraic Geometry II
Co homological Methods in Algebra.
Geometric Applications to Algebraic
Surfaces
1995. Approx. 270 pp. ISBN 3-540-54680-4

Volume 36: A.N.Parshin, I.R.Shafarevich (Eds.)

Algebraic Geometry III
Complex Algebraic Manifolds
1995. Approx. 270 pp. ISBN 3-540-54681-2
Springer
 Encyclopaedia of Mathematical Sciences
Editor-in-Chief: R. V. Gamkrelidze

AI ebra

Volume 11: A.I.Kostrikin, I.R.Shafarevich (Eds.) Volume 12: D. B. Fuks, S.P.Novikov (Eds.)
Algebra I Topology I
Basic Notions of Algebra General Survey. Classical Manifolds
1990. V, 258 pp. 45 figs. ISBN 3·540·17006·5 1994. Approx. 300 pp. 79 figs.
ISBN 3-540-17007-3

Volume 18: A.I. Kostrikin, I. R.Shafarevich (Eds.)

Volume 24: S.P.Novikov, V. A. Rokhlin (Eds.)
Algebra II
Noncommutative Rings. Identities
Topology II
Homotopies and Homologies
1991. VII, 234 pp. 10 figs. ISBN 3·540·18177-6
1994. Approx. 235 pp. ISBN 3-540-51996-3

Volume 38: A.I.Kostrikin, I.R.Shafarevich (Eds)

Volume 17: A. V.Arkhangel'skij,
Algebra V L.S.Pontryagin (Eds.)
Homological Algebra
General Topology I
1994. Approx. 230 pp. ISBN 3'540-53373-7 Basic Concepts and Constructions.
Dimension Theory
Volume 57: A.I.Kostrikin, I. R.Shafarevich 1990. VII. 202 pp. 15 figs. ISBN 3-540-18178-4
(Eds.)
Algebra VI Volume 50: A. V.ArkhangeJ'skij (Ed.)
Combinatorial and Asymptotic
Methods of Algebra
General Topology II
Compactness. Homdlogies of
1994. Approx. 260 pp. 4 figs. ISBN 3-540-54699-5
General Spaces
1994. Approx. 270 pp. ISBN 3-540-54695-2
Volume 58: A.N.Parshin, I.R.Shafarevich
(Eds.)
Volume 51: A. V.Arkhangel'skij (Ed.)
Algebra VII
Combinatorial Group Theory. General Topology III
Applications to Geometry Paracompactness. Metrization.
Coverings
1993. Approx. 260 pp. 38 figs. ISBN 3-540-54700-2
1994. Approx. 240 pp. ISBN 3-540-54698-7

Volume 73: A.I. Kostrikin, I.R.Shafarevich (Eds.)

Algebra VIII
Representations of Finite-
Dimensional Algebras
1992. VI, 177 pp. 98 figs. ISBN 3-540-53732-5

Springer