VOL. 39, 1953 MA THEMA TICS: F.

HIRZEBR UCH 951

equations. It is also applicable to various classes of multistage continuous

games, such as games of survival.
If-as is necessary in more realistic mathematical models dealing with
the production of capital goods-time lags are taken into account, the
complexity of the problem increases.
A complete and detailed treatment of the above problem will be pre-
sented subsequently, together with a discussion of extensions in the
directions just cited.
1 Bellman, R., "On the Theory of Dynamic Programming," PRoc. NATL. ACAD. ScI.,
38, pp. 716-719 (1952).
2 Bellman, R., Glicksberg, I., and Gross, O., "On Some Variational Problems Occur-

ring in the Theory of Dynamic Programming," Ibid., 39, 298-301 (1953).

3 Note added in proof: In the meantime, we have developed a method based upon
simultaneous consideration of the problem and its dual, which makes the verification
quite simple.

ON STEENROD'S REDUCED POWERS, THE INDEX OF INERTIA,

AND THE TODD GENUS
By FRIEDRICH HIRZEBRUCH
THE INSTITUTE FOR ADVANCED STUDY
Communicated by S. Lefschetz, July 3, 1953
Introduction.-This note is a preliminary report on some results con-
cerning the Steenrod reduced powers' in oriented manifolds Mm and the
index r of such a manifold.
Definition of T: If the dimension m of M is not divisible by 4, then
Tr(M) = 0. If m = 4k, then 7(M) is equal to the number of positive
squares minus the number of negative squares of the normalized quadratic
form defined by the cup-product x - x where x e H2 (M4k, R).
We apply the results to almost complex manifolds, in particular we
give a definition of the Todd2 genus of an almost complex manifold and
state properties of this genus. Since a special algebraic formalism will
be used throughout the note, we outline this formalism in the first section.
All manifolds occurring in this note are compact, differentiable, and
oriented unless stated to the contrary. Full details with further applica-
tions will appear elsewhere.
1. Multiplicative r-Sequences.-Let EI a1xI, ao = 1, be the power
i =o
series with the indeterminates aL as coefficients, and let r be a field. Let
{ Kj } be a sequence of polynomials, K, being of weight j in the af and having
952 MA THEMA TICS: F. HIRZEBR UCH PROC. N. A. S.

co

coefficients in r (Ko = 1). If C = E cix', co 1, is an arbitrary power

i=0 X
series, we denote by K(C) the series E Kj(cl, C2, ..., Cj)X'. We call
j =0
Kj } a multiplicative r-sequence provided that K is a homomorphism,
i.e., K(AB) = K(A)K(B), where A and B are power series with indeter-
minates ai, bi as coefficients (ao = bo = 1). We construct for every given
power series Q(x) = Eyixi, (7y e r, yo = 1), a multiplicative r-sequence.
Writing formally
1 + a1x + ... +axm= (1 + ClX)(1 + a2x) ... (mX),
(1 +
we express Q(aix)Q(a2x) ... Q(camx) as a power series with coefficients
which are polynomials in the a1:
co

Q(aix)Q(a2x) ... Q(amx) = jEi=0 Kj, r(an, ..., a1)xj.

One verifies easily that K>, m does not depend on m for j < m. We write
Kj, j = Kj and obtain the unique multiplicative r-sequence with
K(1 + x) = Q(x).
Now let r be the rationals. We denote by {T,} the multiplicative
sequence belonging to Q(x) = -x(ez - 1)-1 and call it the Todd
sequence.2
2T1 = a,, 12T2 = a 2 + a2, 24T3 = ala2, 720T4 = -a4 + a3a, + 3a2 +
4aa - al.l4
4a2a2
For a prime q > 2 the coefficients of q?T(q-1), are integers mod q (i.e., do
not contain q in the denominators).
We also consider the multiplicative sequence belonging to Q(x) =
x(tgh)-l which we denote by {L11. We have
3L1 = a,, 45L2 = 7a2 - al, 945L3 = 62a3- 13a2a + 2a3,I ...
For a prime q > 3 the coefficients of qTL3/,(q-1), are integers mod q.
2. Reduced Powers. '--Let Mm be a compact oriented manifold. The
reduced powers are defined for every odd prime q:
(P.t(m Hk+2(q qlrg
__O_

In case k + 2(q - 1)r = m, there exists by Poincar6 duality an element

se H2(Q-l)r(pm Zq) such that
(Pqu = Squ for all u e Hk(Mm. ZQ).
Let PI, P2, ... be the Pontrjagin classes3 of Mm where pi e H4O(Mm, Z).
In the notation of Wu3 Pi = P041.
VOL. 39, 1953 MA THEMA TICS: F. HIRZEBR UCH 953

THEOREM 2.1. The class sq can be expressed as a polynomial in the

Pontrjagin classes:
q = qrLl,1(q-1)r('Pl P2. . .. P1/2(q-l)r)mod q.
THEOREM 2.2. If Mm is an almost complex manifold of n complex dimen-
sions (m = 2n), then we can express Sr as a polynomial in the Chern classes
ct where ci e H2t(Mm, Z):
q qrT(q c)r(C,C2
=
- C(q. .,
-l)r) mod q.

We have assumed that the prime q is odd, but in the case q = 2 we

may consider the Steenrod squares Sqt. For a manifold Mm (not neces-
sarily oriented) WU4 defined the class Ui e H1(Mm, Z2) by Sqtv = UPv
for all v e Hm - '(Mm, Z2).
THEOREM 2.3. Let wi e H(Mm, Z2) be the Stiefel-Whitney classes of
Mm. We have
Ui = 2tT(w1, w2, ..., wi) mod 2.
Remark:4 T2, + i is divisible by a,. Hence w, = 0 implies U2 + 1 = 0.
THEOREM 2.4. If Mm is almost complex, we have
U21 +1 = 0, U2' = 2'Tj(ci, c2, ..., ct) mod 2.
The proofs of Theorems 2.1-2.4 are based on the "diagonal" method
of Thom" and WU,4' 6 and on a topological interpretation of the multi-
plicative sequences which uses the Borel-Serre7 method of regarding the
classes of Stiefel-Whitney, Chern, Pontrjagin as elementary symmetric
functions, and which also uses the Whitney duality theorem. The new
point in the Theorems 2.1-2.4 lies in the explicit construction of the
polynomials and in the fact that for all primes q these polynomials are
obtainable from one and the same (rational) multiplicative sequence by
reduction mod q.
3. The Index.-For an oriented M4 we can regard the class Lk(pl,
., pk) e H4k(M4k, R) as a rational number. By using a strong theorem
of Thom8 (p. 1735, Theorem 7) we obtain:
THEOREM 3.1. We have T(M4k) = Lk(pb, ..., Pk). Hence Lk for an
M4 is always an integer. For example,
3 i,(M4) = Pi, 45 r(M8) = 7P2 - P', 945 T(M'2) = 62p3 - 13P2P1 + 2p3.
Remarks: It was known to Thom that r can be expressed as a poly-
nomial in the Pontrjagin classes.9 Theorem 3.1 implies for an M4: p1-
0(3) (see Wu6) and for an M8: 7P2 - p0O(45). From Theorem 2.1
we obtain Pi0(3) and 7P2 - pO0(15). Analogously for all dimensions.
In case M4k is almost complex, we can express the Pontrjagin classes by
the Chern classes3
954 .914A THEMA TICS: F. HIRZEBR UCH PROC. N. A. S.

co co o\

i=o (-l)ip X2i = E

i=o CtX) Z
i=o (_1)ixi)
and obtain polynomials for r in the Chem classes.
We now state two theorems about the index which follow from formal
properties of the polynomials Lj.
THEOREM 3.2. If the manifold Mm is fibred in complex projective spaces
Pn of n complex dimensions with the group of all projective transformations
as structure group and the manifold Bm - 2n as base (m > 2n), then
r(Mm) = r(Bm - 2n)7(P )
Remarks: T(Pn) = 1, if n is even; 'r(Pn) = 0, if n is odd (see the Intro-
duction). For the direct product of two manifolds M, M' we have
r(M X M') = T(M) * T(M').
Consider a manifold M4k + 2. Every element x EH2(M4k + 2, Z) can
be represented by a subvariety V4k of M4k + 2 (Thom8, p. 573, Theorem 2).
The index r( V4k) only depends on x and may be denoted by r(x). If
Xl, ... . Xr E H2(M4k + 2, Z), then xi can be represented by a subvariety V4k
of M4k + 2, the restriction of x2 to V4" can be represented by a subvariety
V4k - 2 of V4k, etc. Finally, the restriction of Xr to V4k - 2r + 4 can be
represented by a subvariety V4k - 27 + 2 of V4k - 2r + 4 The index of
V4k - 2r + 2 only depends on the non-ordered r-tpl (xi, ..., xr) and may
be denoted by r(xl, ... , Xr)
THEOREM 3.3. We have for xi, X2 e H2(M4k + 2, Z)
'r(xi + x2) = r(xi) + r(X2) - T(XI, X2, XI + X2).
4. The Todd Genus.-Let Mn be an almost complex manifold of n
complex dimensions and ci e H21(M., Z) its Chern classes. We can
regard Tn(cl, . . . , cn) e H2'(M,, R) as a rational number which we denote
by T(Mn). We call T(Mn) the Todd genus of Mn. Kodaira1I proved
for all algebraic varieties M. which are a complete non-singular inter-
section of hypersurfaces in some projective space that
T(M,) = 1 -
gl + g2- . + (l)"gn
where gi is the number of linearly independent i-pl differentials of the
first kind attached to Mn. From the results of Todd,2 Hodge," and
Kodaira-Spencer12 it seems very likely that the last formula is true for
all non-singular algebraic varieties. But in the present moment it is not
known, even for algebraic varieties, whether T(M.) is always an integer.
Therefore it seems to be interesting that one can prove by the Theorems
2.4 and 3.1:
VOL. 39, 1953 MA THEMA TICS: F. HIRZEBR UCH 955

THEOREM 4.1. The number 2n - 'T(M.) is an integer for every almost

complex manifold. This means for n = 1, 2, 3, 4:
M1: cl 0(2), M2: cl + C2-0(6), M3: cl c2- 0(6),
M4: -C4 + C3C1 + 3C2 + 4c2A - C4O (90).
We state two theorems about the Todd genus which follow from formal
properties of the polynomials Tj and which are analogous to the Theorems
3.2 and 3.3.
THEoREM 4.2. If the almost complex manifold M,, is fibred in complex
projective spaces Pk of k complex dimensions with the group of all projective
transformations of Pk as structure group and the almost complex manifold
Bn - k as base, the fibering being compatible with the almost complex structures
of Mn, Pk and B. - k then
T(Mn) = T(Bn k) T(Pk) = T(Bn -k))
-

Remarks: T(Pk) = 1 for all k. Since { Tj } is a multiplicative sequence

we have for the direct product of two almost complex manifolds M, M'
T(M X M') = T(M).T(M').
This (in the algebraic case) was checked by Todd2 for dimensions not
exceeding 6.
For a class x e H2(M,,, Z) we can define a virtual Todd genus T(x),
which is a polynomial of weight n in x and the Chern classes of Mn, such
that T(x) is the Todd genus of every admissible almost complex sub-
variety Vn _. 1 of Mn representing x. Moreover, if xl, x2 e H2(Mn, Z) we
can define a virtual Todd genus T(x1, x2) with T(x1, x2) = T(x2, xI) such
that for every admissible almost complex subvariety Vn 1 representing
xI, the number T(x1, x2) is the virtual genus with respect to Vn - 1 of the
restriction of x2 to V,, - 1. The virtual genus T(x1, x2) is a polynomial of
weight n in xl, x2 and the Chern classes of Mn. The following theorem is
well known in algebraic geometry:
THEOREM 4.3. For xl, x2 e H2(Mn, Z), we have
T(x1 + X2) = T(x1) + T(x2) - T(x1, X2).
1 Steenrod, N. E., these PROCEEDINGS, 39, 213-223 (1953).
2 Todd, J. A., Proc. London Math. Soc., (2), 43, 190-225 (1937).
3 Wu, Wen-Tsun, and Reeb, G., "Sur les espaces fibr6s et les varietes feuilletees,"
Actual. sci. industr., 1183, (1952).
4 Wu, Wen-Tsun, Compt. rend. acad. sci., Paris, 230, 508-511 (1950).
6 Thom, R., Ann. sci. 1Acol. norm. sup. (3), 69, 109-182 (1952).
6 Wu, Wen-Tsun, "Sur les puissances de Steenrod," Colloque de Topologie de Stras-
bourg, 1951; (mimeographed notes).
7 Borel, A., and Serre, J. P., Compt. rend. acad. sci., Paris, 233, 680-682 (1951);
Borel, A., Ann. Math., 57, 115-207 (1953); Borel, A., and Serre, J. P., Am. J. Math.,
75, 409-448 (1953).
956 MA THEMA TICS: K. MENGER PROC. N. A. S.

8 Thom, R., Compt. rend. acad. sci., Paris, 236, 453, 573, 1733 (1953).
9 Thom, R., "Quelques proprietes globales des vari6t6s diff6rentiables"; (to appear
in Comm. Math. Helv.).
10 Kodaira, K., "The Theory of Harmonic Integrals and Their Application to Algebraic
Geometry," Notes, Princeton University, 1953.
1 Hodge, W. V. D., Proc. London Math. Soc., (3), 1, 138-151 (1951).
12 Kodaira, K., and Spencer, D. C., "On Arithmetic Genera of Algebraic Varieties,"
these PROCEEDINGS, 39, 641-649 (1953).

THE IDEAS OF VARIABLE AND FUNCTION

By KARL MENGER
ILLINOIS INSTITUTE OF TECHNOLOGY
Communicated by J. von Neumann, June 18, 1953
Investigation of the semantics of pure and applied analysis reveals that
there exist three different concepts of a variable, which may conveniently
be described as a logical, a scientific, and a mathematical concept. Fol-
lowers of Weierstrass -have treated the logical notion; French analysts,
the mathematical and scientific ideas. The confusion of the three con-
cepts accounts for the obscurity which marks the introductions to many
treatises on analysis. The lack of a distinction is also one of the reasons
that prompted Russell to call the notion of variable "one of the most
difficult with which logic has to deal." Severally, the concepts and their
mutual relations (which we propose to study within the realm of real
numbers) seem clear and simple.
The Logwcal Concepts of Arithmetized Amlysis.-A function is a non-
empty set of ordered pairs of numbers such that no two pairs of the set
contain equal first and unequal second elements. The set of all first
(second) elements of the pairs, referred to as arguments (values), is called
the domain (the range) of the function. In arithmnetized analysis (in
conformity with the usage of modem logic), a real variable is a symbol
which stands for any element of a certain set of real numbers, called the
range of the variable. E.g., the set of pairs (x, tan x), for all numbers x
which are not odd multiples of Tr/2, is a function which we shall denote
by tan. The letter x in the definition of tan and in theformula D tanx =
sec2 x is a variable having the domain of tan as range. The symbols log
and v,/ denote other functions.
For the set of all pairs (x, x), in spite of the paramount importance of
this function, no traditional symbol exists. We shall denote this identity
function by l, and write &n, 16. 2, and 16 for the functions (x, x"), (x, 16x2),
and (x, 16), respectively, which traditionally are referred to as the func-
tions x, xm, 16x2, and 16, that is, by their values for x. (Even references