Journal of Mathematical Sciences, VoL 72, No.

4, 1994

THE SOLUTION OF MALFATTrS PROBLEM

V. A. Zalgaller and G. A. Los' UDC 514.112.3

For the first time Malfatti' s old problem about arranging three non-overlapping circles of greatest total area
in a triangle is solved.

The original conception of this paper belongs to G. A. Los' [4]. However, [4] contained significant gaps. A lengthy
correspondence between the authors of the present paper enabled them to flesh out and fill in these gaps.
The History of Malfatti's Problem and the Main Result. In 1803 Malfatti [12] posed the following problem: how
to arrange in a given triangle three non-overlapping circles of greatest total area?
Algebraically this is a question of choosing nine variables (six coordinates of centers and three radii) that will
maximize the sum of the squares of the radii subject to twelve constraints in the form of inequalities. Nine of the inequaIities
are linear (conditions for placing the centers on a particular side from each of the sides of the triangle at a distance not tess
than the radius) and three are nonlinear (conditions insuring that the circles do not overlap). The algebraic approach to the
problem shows little promise. For a fixed triangle a maximal collection of circles (their radii lie in the closed interval [0, R],
where R is the radius of the inscribed circle) exists by reason of compactness. GeometricalIy, it is clear that there are no
circles of zero radius in the maximal collection. However, a better arrangement of the circles may depend on the type of the
triangle.
This problem of elementary geometry has until now remained unsolved, a fact noted on numerous occasions [11],
[13], [5, p. 93], [8, pp. 245-247], [2, pp. 23-25], [7, pp. 150-153]. The present paper presents its solution. Despite the
simplicity of the finat result, the proof required different approaches to the cases encountered and, at one stage, even
computer calculations.
We denote the angles of this triangle by 2c~, 2j3, 23, and the vertices corresponding to them by A, B, C. In the
formulation of the main result it is assumed that

o<a--¢~r<~. (1)

The Main Result. Three non-overlapping circles of maximum total area are arranged in a triangle in the following
way. A circle K 1is inscribed in triangle A B C . A circle K2touches sides AB, AC and circle K 1 . For sinc~ _< tanB/2 a circle
K3touches sides BA, BC and circle Kl(Diag. 1 in Fig. 1) and for sina _. tan~/2 it touches sides AB, AC and circle
K 2 (Diag. 2 in Fig. 1). For sine~ = tan#/2 there are two substantially different maximal arrangements.
Rigid Systems. A system of three non-overlapping circles in a triangle is called rigid if each of the circles, with the
position of the other two circles remaining unchanged, cannot be continuously displaced in the triangle by increasing its
radius while at the same time not violating the rule that the circles do not overlap. It suffices to look for the solution of
Malfatti's problem only in the class of rigid systems.
L E M M A . Every circle of a rigid system has at least three points of contact with the other circles or the sides of the
triangle; moreover, these points of contact do not fit on one closed semicircle of the boundary of the circle under consider-
ation.
Indeed, when these conditions are violated it is easy to indicate the direction in which when the circle is displaced its
radius admits increase. The Lemma implies that every circle of a rigid system touches one of the sides.

Translated from Ukrainskii Geometricheskii Sbornik, No. 35, pp. 14-33, 1992. Original article submitted November
19, 1990.

1072-3374/94/7204-3163512.50 ©1994 Plenum Publishing Corporation 3t63

 2 3

4 5

Fig. 1

We call two rigid systems combinatorially identical if to the sides of one triangle and the circles inscribed in it we
can put in correspondence the sides of a second triangle and the circles inscribed in it so that the correspondence is one-to-
one and the existence or lack of a common point of a pair "side and circle" or "two circles" is preserved under this corre-
spondence.
T H E O R E M . There exist only fourteen combinatorially not identical rigid systems. They are depicted in Fig. 1 and
Fig. 2.
Proof. 1. Assume that in a rigid system there is a circle touching all three sides of the triangle (Fig. 1). The area of
the triangle free from the circle splits into three regions. If the two remaining circles lie in different regions, then we have
the situation depicted in Diag. 1. If both of the remaining circles lie in one region, then either each of them touches two sides
of the triangle (Diag. 2), or one touches two sides of the triangle and the other touches only one side (Diag. 3 or Diag. 4,
depending on whether the first two circles touch each other or not), or each touches only one of the sides of the triangle
(Diag. 5).
2. Assume that each of the three circles touches exactly two sides of the triangle. If the centers of the circles lie upon
different bisectors, the either all three circles touch pairwise (Diag. 6, Fig. 2) or one touches two circles that do not touch
each other (Diag. 7). If the centers of two circles lie upon one bisector, then we arrive at the situation depicted in Diag. 8.
The positioning of all three centers upon one bisector is excluded in assumption 2.
3. Assume that each of two circles touches two sides of the triangle and that the third circle touches only one side.
Then either the first two circles touch each other (Diag. 9 and 10) or they do not (Diag. 11 and 12).
4. If one circle touches two sides and the remaining circles touch only one side of the triangle, then we only have the
situation of Diag. 13. If each of the circles touches only one side then we have the situation depicted in Diag. 14.
Comparison of Diag. 1 and 2. In Diag. 1 it is expedient to place circles K2and K3in the smaller angles of the
triangle while in Diag. 2 both K2and K3are squeezed into the smallest angle. If a circle K of radius Ris inscribed into
angle 2 t , then the circle inscribed into the same angle, but closer to the vertex and touching K , has radius x which is
easily calculated. From Fig. 3 we find that ( R / s i n 6 - R - x = x/sin& whence

x = R
1-sina
~ = l~
(1-tg~) (2)
l+s_n~ (l+tg~) 2
The areas of circle K 1of radius R and circle K 2 are the same in Diag. 1 and Diag. 2. Diag. 1 yields a greater area
of the circles than Diag. 2 if and only if the radius of K 3 is greater in Diag. 1 than in Diag. 2, that is,
l_t4 2 l-sin~

which, taking (1) into account, is equivalent to sins _> tanB/2.

Now to prove the main result we need to rule out the possibility that the absolute maximum of the total area of the
three circles is attained in the arrangements depicted in Diag" 3-14.

3164
 • i 6 7 8

lO 11

12 , t3 !4

Fig. 2

Excluding Diag. 3. Let us consider an isosceles triangle (Fig. 4) with circles K 1, K 2, K 3 and two auxiliary circles
K 4, K 5 arranged in it. For their radii

_ l+sin~ = R~ctgZ0,
Ra = ~tf-s-~
RIR2 ~

_ l-sina
R, = eeq-g~-fi-~,
1-sinH_ a .rr e'~
r~s = " , , ~ - r q t g (a " zJ

By reason of similarity, we assume R 1 = 1.

Let ce = 0.462. For such ~ we have R 2 = 2.608, R 3 = 0.381, R 4 = 0.383, R 5 = 0.809, that is, R 4 > R 3, R 5 >
R 3. Diminishing ~ , radius R3decreases while radius R4increases. Therefore, for c~ _ 0.462, the inequality R 4 > R 3 is
preserved. Let us verify that for ce >_ 0.462, the inequality R 5 > R 3 is preserved. Indeed,

1 _ 1-x

..... t g ( 4 " Z l+ctga I+Z I+2X-X i'

where x = tanS/2. For ee >_ 0.462 we have x 0 = tan(Tr/8-~/4) <_ tan0r/8-0.1155) = 0.2845. However for 0 _< x <_ x o

l-x 2x
> 0,
l+x l+2x-~ 2

therefore R 5 > R 3.
Suppose now that & , t B C is not necessarily isosceles. For the sake of definiteness, assume that Z B _< Z C. Then for
c~ _< 0.462 the area increases by replacing K3with Kaand for c~ _> 0.462 by replacing K3with K~.
Diag. 4 is excluded below together with Diag. 11.
Excluding Diag. 5. Let us turn to Fig, 5. Owing to the choice of notation, we assume that Z B _ Z C. Let Z A >_
~/3- Then / B ___ / A and K1can be reflected in the bisector of angle Cand then enlarged.

3165
 Fig. 3

A 2a K3 K5 B

Fig. 4

Assume now that / A < 7r/3. Let us draw a common tangent T to K 1, K 2. In view of / A < r/3, at least one of the
angles ~, * will be greater than / A . Suppose, for the sake of definiteness, ~o > / A. In that case K 1can be reflected in the
bisector of ~r and then enlarged.
Excluding Diag. 6. T H E O R E M . The sum of the areas of circles arranged in a triangle according to Diag. 6 is
always less than the area of circles arranged according to Diag. 1.
This assertion, confirmed by computer, was given in [10] and simplified in [9]. But the existence of the specially
written notice [3] enables us to confine ourselves to a quite short calculation.
Circles arranged according to Diag. 6 are called Malfatti circles (see [1, Appendix E] and [6, p. 11]. If 2a, 2/3, 27
are angles of the triangle and R is the radius of the inscribed circle, then, as shown in [11] (and, independently, in [4]), the
radii of the Malfatti circles are

P= = 2 !+tg~

PB 2 l+tg~

R (l+tg~)~l+tg~.
P~ = 2 l+tg~

Let us note that

? l+sin~ - l+cos2a 4
[l+tg 2 = 2 l+cos= 2 i%sin2a = (I+-~E~'

where 2o~ = 7r/2-0~. Hence

2 = R2 (l+tga) ~
P~ (l+tg~)Z(1+tg~)" (3)
where & = ~r/4-od2;/~ = 7r/4-/3/2; ~, = ~r/4-2//2.
If after placing the inscribed circle in a triangle, we draw a maximal circle in each of the "free" angles of this
triangle, then the radii of these circles are
= R 1-sin~. l-sin~. R~ -- R l-sin7
R~ z+--Vffi~' R ~ = R ~ , ~ .

3166
 A B

Fig. 5

Note that Rc~ = Rtan2&.

To prove the theorem it suffices to verify that pa 2 + pB2 + p72 < R 2 + 2/3(R~ 2 + RB2 + R 2 ) , that is, that

(1+tga)
•
"+ (l+tgBl~+ ( i + ~ ) ~ < 1+ 2 ( t g ~ + t g ~ + t g ~ ) (4)

where, according to (3), 2&, 2~, 2? are the angles of an acute triangle. However (4) follows from the following inequality
that was proved in [3] for acute triangles:

(1+tg~)(l+tga)
'*(l+tgg)~+(l*tgr)
' (iTtg~
~ (l+tg~) r ~ c+~ (t¢~+tg'~÷tg%'),
where C = 9(~f3 + 1 ) - 2 - 2 / 9 = 0.98355 < 1.
Excluding Diag. 7. Assume that the maximum of the area of the circles is attained when they are arranged as shown
in Diag. 7 (Fig. 6). Radius x can be varied. Radii y and z are determined by the value of x. In addition, x and y are related
by + 2 @ + yctg/3 = c~, whence

~ct.g,÷);
y'-- ................ < O,
4 zgctg8 +x

y"= > O,
2xJ-/~y(~J~-~ctgB+rl

Analogously, s" > 0. Now for the sum of the squares of the radii L(x) = x 2 + y(x) 2 + Z(X)2 we have dZL/dx 2 = 2(1
+ y,2 + yy,, + zz") > 0. The fact that dZL/dx2 is positive contradicts the assumption that the area is maximal.
Excluding Diag. 8. Assume that the maximm-n of the area is attained when the circles are arranged as in Diag. 8
(Fig. 7). Here (as in the case of Diag. 7) y" > 0. In addition, z = x ( 1 - s i n c d l + sine0, whence z" = 0. Therefore, for
L(x) = x 2 + y(x) 2 + z(x)2 we again have d2L/dx2 > 0, which contradicts the assumption on the maximum.
Excluding Diag. 9. T H E O R E M , The total area of the three circles cannot attain absolute maximum when the circles
are arranged as shown in Diag. 9.
Proof. 1. Assume that such a maximum is attained when three circles K 1, K 2, K 3 are arranged as in Diag. 9 (Fig. 8).
Let us introduce coordinate axes x, y, placing them as shown in Fig. 8. By reason of similitude, we assume the radius of the
first circle to be fixed: R I = 1. All the configurations can be characterized by angles c~, i3, "y and one linear dimension OlD,
which it is convenient for us to denote by OlD = 2t2 - 1 .
Let us consider auxiliary circles K 4, K 5, K 6 (Fig. 8). (The possibility that K 4 and K 3 overlap and that point C is
located below the x axis, which is perpendicular to A C , is not ruled out.) The assumed maximality implies

R3 > R4, R3 > Rs, R3 > R6, (5)

3167
 A c B

Fig. 6

A C B
Fig. 7

otherwise replacing K 3 by a larger (or equal) circle from among K4, Ks, K 6 leads to the increase of the area of the three
circles (or to the already excluded diagrams).
2. The rigidity of Diag. 9 implies that center O3is located below the xaxis, otherwise we could enlarge K3bY dis-
placing it upwards. In addition, R 3 < R 1. Indeed, in sector / BAC after K 2 lie the two circles K 1 and K 3 and if we had R 3 _>
R 1, then we could enlarge K 3 by displacing it along AC. Therefore R 3 < R1, R 1 < OlD = 27'2-1 < R 1 + 2R3, that is,

R3<*, R~>t2-*. x<t<~-{. (6)

3. Let us draw a lower common tangent T to K 1 and K 3 (Fig. 8). The angle ~ that it forms with the y axis satisfies
the inequality ~b since for ~ < 23,, we would have had ~b ___ 23,. From R 4 _> R 3 it follows that o~ < 3, and, therefore, o~ <
7r/4.
4. Our immediate goal is to derive from (5) and the rigidity of the diagram as narrow constraints on angles c~,/3, 3,
and parameter t as possible. Beforehand we express R2, R3, R4, Rs, R 6 in terms of o~, /3, % t.
5. Let us prove that

R = r t - s in~ a
2 • (7)

From the trapezoid O1Q1Q202 we have Q1Q2 = 2~/R 2 = 2~]-R2. On the other hand, Q1Q2 = AM-MQ1-AQ2 = ( 2 t 2 - 1
+ 1/cos2oO/sin2o~-tan2o~-Rzctanec Equating these two values of QaQ2 yields a quadratic equation for ~R2, whence we
obtain (3).
Analogously,

(t-sinai a
R.-- L J" (8)
In addition,

l-sina It-sin~] 2 (9)

Rs = RZl+sin~ = '

l-sin~
R, = R1i+sin- tga [ 4 - ~ 3 " (10)

3168
 Fig, 8

D C
Fig. 9

6. The equality

~ t-'sin~ ]z
R= =t2 kl+cos~ts in=9 (1i)

is more difficult to prove. From 1 < t < ~r2 and 0 < a < 7r/4 it follows that

t-gln~>O, l+cos~-tsin~>O. (12)

Let us prove ( I i ) . From AOIO3D 3 we find 03D 3 = ((R t + R3)2-(O1D-R3)2) 1/2 = 2t(R3-t2 + I) 1/2. From the
trapezoid 02P2P303 we have P2P3 = 2(R2R3) I/2. On the other hand,

=2t(l-tsin=) 2
P2P3=DA-AP2-DP3=MDctg2~-R2ctu~-O~D3
. . . . 2t~.
C08~

Equating the two expressions for PzP3, we get

(13)

where tz = x/R3, m I = t-since, m 2 = tsin~. After squaring (13),

3169
 TABLE 1

l t ~D(t) i[ ~o(t)
e

" i,oo 0,464 I 0,841 i 0,841 >~/2

I •. 1,05
!,I0
0,423 i 0,848
0,385 ] 0,855
0,719 >~/2
0,602 >~/2
I 1,1s I 0,350 ; 0,861 ! 0,489 .~/2
t0~1,198342 ,0,318832510,8668877 0,3850767 ~/2
1,20 f 0,318 i 0,867 0,3S2 <~!2
1,25 [ 0,288 ! 0,873 0,280 <~/2
! oo ! 0,259 I 0,878 0,184 ~W/2

*Most of the values have been rounded off to the third decimal sufficient
for derivation.

TABLE 2

t %(t). ~,(t) ~o(t) ~(t)

1,19s342 0,31883 0,31883 0,,38508 0,38508
1,20 0,31779 0,32~-11 0~38161 0,38593
1,205 0,31467 0,33196 0,37119 0,38849
1,21 0,31157 0,34176 0,36083 0,39103
ii215 0,30849 0,35150 0,35053 0,39355
1,22 0,30543 0,36119 0,34029 0,39605
1,225 0,30239 0,37082 0,33010 0,39853
1,23 0,29938 0,38040 0,31988 0,40100
1,235 0,29638 0,38992 0,30991 0,40345
1,24 0,29341 I 0,39939 0,29991 0,40588
1,24509 0,29040 0¢40897 0,28978 0,40834
i,25 0,28752 0,39479 0,28007 0~41070
1,25 ~ 0,28461 0,38051 0,27025 0~41308
1,26 0,28172 0,366.40 0,26048 0,41544
1,262019i 0,28056 0,36075 0,25655 0,41639
1,265 0,27885 0,35246 0,27095 ! 0,41779
1,27 0,27599 i 0,33865 0,29491 0,42012
1,275 0,27315 0,32501 0,31866 0,42244
1,28 0,27034 0,31152 0,34220 0,42474
1,285 0,26755 0,29818 0,36551 0,42702
1,29 0,26477 0,28500 0,38862 0,42929
1,295 0,28201 0,27196 0,41153 0,43153
1,30000 0,259~7 O~25927 0,43389 0,43389

z 2 (m~-t 2cosaa) -2%m2z÷nv~2+ (t ~- 1) cosaa-O. (14)

Since m12m22-(m12-t2cOs2ot)(rrt22 + (t2-1)c0s2o0 = m12cos2o~,

m, (.~±cosa)
z~' "r~-cacos~a (15)

Then ml(m2-mlz) --- zt2cos2o~ -T- mlcoso~, but (13) implies that rr~-mlz > 0. This obliges us to choose the minus sign in
(15). Therefore,

.4(%-eosa) 2
R3=~ ,, , (16)
(.r~_t~cos~a ) 2'

The identity rnl2-t2cos2rx = (m2-cosa)(1 + cosa-tsino0 (different from the standard factorization of a difference
of squares) holds, therefore (16) implies (11). When deriving (16) we assume the leading coefficient in (14) to be different
from zero. By continuous dependance of both sides of (11) on t and a , this equality is also preserved for those pairs t, a for
which m12-t2cos2o~ = 0.
7. From R 5 < R3, taking (9), (11), and (12) into account, we have 1 + cosa-tsinc~ < t(1 + sina). This implies
that for a fixed t

3170
 TABLE 3

1,198342
1,20
1,205 1,285 ! 5,67 5,25
I, I 9;~oI 9;41I 1,29 I 8,III 7,88
1,215 7,25[ 7 , 1 9 I 1,295 Ii0,66 10,57
1,22 I, 3000113,30 113,30
1,225

TABLE 4

-- -- Vs 7~' Vs i. V6 Vs I V~ i Vs V6
I .,------z----- ......i
1,225 19'ii 8,9018,32 8,27 8,50{8,25 7,6417,551
1.2~ 18,187,81i~,~s 6,8317,61'7,2~ 6,5016,nI
ii,235 16,9716,42.5,70 5,49- 6,51~5,93 5,3715,09
1,24 !5,6814,8914,41 4,o915,3614,5614,2413,86
1,24509 [4,27'3,18i3,06 2,611'4,13-3,0513,11 2,50
! ,25 t14,68' 3,48 '3,48 . '
~,9914,2213)0513,~512,70
1,255 5,07 3,8013,94 3,42{4,32 3,07}3,4412,88
1,=a 15,4~.4,1014,43 3,901.4,4213,n13,70.3,14
1,26201915,56 4,21,4',63 4,10 4,45j3,14i3,83!3,27
1,265 6~I0 5,02 5 , 2 1 4 , 7 5 t 4 ; 9 5 t 3 , 8 9 1 4 , 3 3 1 3 , 8 4
1,27 17,01 6,2616t2215,85i5,84i5,0915,25~4,86
1,~75 7,94.7,44i7,2716,991~,79.16,2816,~15,9~

Fig. 10

t-i I (17)
a>~ o (t) = a r c c o s . - - arccos--,

where e~o(t) is a root of 1 + cosc~-tsin~ = t(i + since).

8. For a fixed t the value of P~ attains the maximum R3(t) when a is smallest, that is,

t(t-.in%(t))
(18)
R~"{7=~+eosao(tl_t,in=,(t )
9. For a fixed t

(19
cos~"
It is geometrically evident (see Fig. 9) that for a fixed OlD = 2 t 2 - 1 to the movement from top to bottom of point Calong
the line p there corresponds the increase of 7 from 0 to n / 2 . In this case R4first decreases (until O4falls onto the line
OlD) and then increases. The attained mintmum is minR4 = (OLD-1)/2 = t~ - 1.

3171
 B

A <~-c

Fig. 11

• 2f3
B
A 2a
Fig. 12

Fig. 13

//
The angle alies in the range %(t) < c~ < ~l"(t), where a l ( t ) i s the value of a t o which (for a fixed t ) corre-
sponds the position of the center of K30n OlD. For a = al"(t) the radius R 3 = t 2 - 1 . Therefore we find from (11):

tz_ t 2 ~ _ l ' (20)

The requirement %(t) < oq"(t) gives rise to an additional upper bound for t: t < t 1 = 1.3. This follows from the fact that
fit)
= ~ l " ( t ) - % ( t ) , see (20) and (17), strictly decreases on 1 _< t _ ~/2 vanishing when t 1 = 1.299896.
For t E (1, tl), (19) implies

J Ra(ti "" t (21)

70 (t ) = a r c c o s arccos

where

3172
 C A

Fig. 14

K~ LK

Fig. 15

~I (~) = a r e c o s + areeo~

4 ~ dR3(t) +1

are two roots of t - s i n 7 = RR~3(t3(t)cos3`.

, The value of 71corresponds to the situation in which side BC touches K3so that
K4coincides with K 3 . Further increase of y makes the substitution of K3by K4impossible.

10. The inequality tg(~- d21=~,<~"~< ~ 3 ~ ( t ) implies

(22)

11. Table 1 shows the nature of c~o, 70, ~o', e~o + flo' + 7o, in particular, their monotomcity. Let us denote by t o =
1.198342 the root of So(t) + ~o'(t) + 3,o(t) = 7r/t. Since s o + ~o' + 7o > ~r/2 contradicts a + fl + 3' = 7r/2, the validity
of (5) for rigid diagram 9 is possible only' when to < t < t 1.
Here C~o(t) < a < cq(t), ~o'(t) < ~ < fit(t), 7o(0 < 3' < 7t(0, where oq(t) = a ' / 2 - 3 ' o ( t ) - ! o ' ( t ) ; ill(t) =
~r/2-3,o(t)-ao(t), must be fulfilled. The two-sided bounds on c~ and ~3can be somewhat narrowed in part of the range of t .
The former bound a < oq"(t) is more exact than a < a1'(t) when 1.24509 _ t <_ 1.3. For such t we have ~ >
~o"(0 = 7 r / 2 - 7 1 ( t ) - a l " ( t ) • The bound ! > /30"(0 is more exact than ~ > fio'(t) when 1.262019 _ t _< 1.3.
12. Finally, for rigid diagram 9

3173
 to<t<tt, ~ ( t ) < = < . t ( t } , ao(t)<l~<a~(t), (23)

where Otl(t) = min{eq'(t), O~l"(t)} ; ~0(t) ----- max{/30'(t), /30"(0 }.

We call the triangles and the arrangement of K~, K~, K 3 satisfying (23) localized. Table 2 presents their numerical
characteristics.
13. To conclude the proof of the theorem it is enough to prove the following lemma.
L E M M A . In every localized arrangement the total area of K 1, K 2, K 3 increases if we replace them with circles
arranged as shown in Diag. 1, that is,

2 2 2 2 2 (24)
Yl=Pl +Pa+P3 - (Rz +Ra+R~) >0,

where p ~ is the radius of the inscribed circle and

14. The value of Pl is easily expressed in terms of R1 and R,2by comparing two representations of length A C AC
= RlCtg/3 + 2 x / ' / ~ 2 + R2ctg~, AC = plctgc~ + plctg~3. Let us take into account that R~ = 1 and denote V2 = V~(ctg~ +
ctg~3)2R2-~. It is enough to show that V2 > 0.

wherep(~,~)=I + tan4 [~- ;]+t~'[~ ~];

q(t,a)-
[~2Jtl+con"-tsin~j
-- . . . . .

15. We next verify V2 > 0 for various t, a E (ao, cq), 13 E (/3o, /31) along the following path. Let t be fixed and
suppose that % < a < ~x1, /30 < t3 < /31. In view of Op/Oot < O, Op/Ol3 < O, Oq/Oa > 0: p ( a , / 3 ) _> p = P(al,/31), q(o~,
/3) < q = q(t, a l ) , whence

. 2

V~>V~=C(R2, ot, ~ ) +p (R2ctg2a+2

.
ctcjactgB c t +Hag'fl
~ ) -a o(R~) ( ctg(*+ctg/~ ) 2t

r-. ctgp. I
where c (/~, a, 0) = 4 p (l+J/~2ct~a+-----) ; no (R~) = (i+q} R 2 + ~ ,
•

From (7) it follows that OR2/Oo~ < 0, therefore r 0 _< R 2 _< r t, where
2
t

whence C(R2, ~,/3) ~ C = 4p(1 + x/-roctgoq + ctg~l/X[rl),

V3~Y,---C- (n, (R2) etg2u+2a2 (R~) ctgactg@+a~ (R2) ctg2B) ,

where al(R 2) = ao(R2)-pR2; a2(R 2) = ao(Rz)-p; a3(Rz) = ao(Rz)P/R2. It is enough to show that V4 > 0.
16. In all cases of further computation it will turn out that X = ctgZeq + 2ctgc~lctg/31 -(p-1)ctg2/30 > 0, Therefore

' . - L ( c t - ,, + ctg ct - (p- ) ct¢ ) o.

3 2
8R~ R2 R2

This implies that it is enough to verify V4 > 0 for R 2 = r o and R e = r].

3174
 In addition, it will prove to be in all cases that the coefficients at(ro), a2(ro), a3(r0), al(rl), a2(rl), a3(r 1) are
positive. Therefore, it is enough to verify by direct calculation that V5 > 0, V6 > 0, where

Vs=C- (aI (to) ctg2~o+2a~ (to) etgaoctgSo+a~ (to)ctg~Bo) ;

vs=C" (al (rl) ctgz%+2a2 (/'I)ctg%ctg~o+a~ (rl) ctg28o) •
17. For tO < t <_ 1.225 and }.28 _< t _< tI k is enough to take for %, ~I, rio,/91 values from Table 2; the results
of the computations are given in Table 3. They confirm V 5 > 0 and V 6 > 0 in these ranges of t.
18. On the remaining interval 1.225 ~ t _< 1,28 it is enough to consider separately four combinations of possible
bounds:
I) aoS=S(ao+=,)!2, 0o~t3S(~o+B1)/2,
II) =o~=s (=o+cq) 1~-, (rio4~,) / 2"~s~t,
IH) (=o+a~)/2sas=1, 0os~s(~0÷~)/2,
IV) (=o+=I) 12s=s=~, ('~o+~)12s~sB1,
where ~0, °tl, ~0, ~1 are given in Table 2. The results of the computations are given in Table 4:
The positivity of all of the values of V5, V6 concludes the proof of the theorem.
Excluding Diag. 10. The smallest circle (Diag. t0, Fig. 2) can be replaced by a circle symmetric to it with respect
to the line connecting the centers of the two other circles and then enlarged without leaving &dBC.
Excluding Diag. 11 and 4. THEOREM. The total area of the circles cannot attain a maximum when they are
arranged as shown in Diag. 11.
Proof, Suppose that K1, K2, K3 are arranged in a triangle as shown in Diag. 11 (Fig. 10). Let us draw a line M N ,
the common tangent of Kland K 2 . The point Nlies to the right of A, otherwise K2would admit enlargement and the
diagram would not be rigid. For what follows it is immaterial whether Nlies on A C o r to the right of C. Let us consider
separately two possibilities.
1. Suppose that M N has no common points with K 3 (Fig. 10). Let us consider K3, K2 in ~AMN denoting their radii
by x, y. The value L(x) = x 2 + y(x)2 is a strictly convex function of x. Therefore, with a small change of x that would not
put K2, K3 out of &AMN, we can increase their total area.
2. Suppose that MNintersects K3(or touches K3) (Fig. 11). Let us draw the common tangent P Q o f K 2, K 3. If in
i / t P Q the angles / A _<_ Z Q, then by "moving" K2from Qinto A, it becomes possible to increase K I and with it the totat
area of the circles. Finally, if / X > / Q, then we slightly turn PQaround the center of K 2 . We turn it so slightly that,
firstly, in ~ , t P ' Q ' the inequality / A > / Q , is preserved and, secondly, the circle K~inscnbed into ~SAP Q 0t Is shown
t / ' ' l r " '

/
in Fig. 11 with broken line) does not force K~to leave 2wtBC. (The latter is essential for the case where Q/lies to the right
of C.) The sum L o f the squares of the radii of K 3, K 2 is less than that of K3',/(2'. This follows from the fact that in
zkAP'Q' the value L i s a convex function of radius x o f K3and the position of K3', K2' corresponds to the absolute maximum
of L in view of Z A > Z Q'. By replacing K3, K2 with K3', K 2' we increase the total area of the three circles.
THEOREM. The total area of the circles cannot attain a maximum when they are arranged as depicted in Diag. 4.
Proof. Diag. 11 becomes Diag. 4 when K 1is the inscribed circle. This particularity of Diag. 4 does not preclude us
from carrying out for it a proof that repeats verbatim the one given for Diag. 11.
Excluding Diag. 12. Assume that the maximum of the area is attained when the circles are arranged according to
Diag. 12 (Fig. 12). Let us vary x preserving the point of contact Dunchanged. Here radii x, y are related by 2.,/~ + yctg~
= DB. Hence,

y'=--* Y <0,

2 a y2 ~
2 (~ctg~+;~) y =2y+-----ctgS-y (X~yctgB+2X) >O,
,x[~y

that is, y" > O. Analogously, z" > 0 and, therefore, for L(x) = x 2 -I.- y(x)2 + Z(X)2 we have aeL/dx 2 > 0, which contradicts
the assumption on the maximum of the area.
Excluding Diag. 13. THEOREM. The absolute maximum of the total area of three circles cannot be attained when
they are arranged as depicted in Diag. 13.
Proof. Assume that such a maximum is attained when K 1, K 2, K 3 are arranged according to rigid diagram 13 (Fig.
13). In this case,

3175
 1. The inequality K3 must be fulfilled. Otherwise we can reflect K 3 in the bisector of 2 y after which we can enlarge
K 2 . Therefore,

fi - r., '*+0 > 2~. (25)

2. The rays l 1, l2 of the common tangents of K1, K 2 and K 1, K3 reach, respectively, the rays ACand ABand,
therefore, the common tangent T also reaches ACand AB forming with them the angles q0and ~ .
3. The inequality

~<ze (26)

must be fulfilled. Otherwise K 2 can be reflected in the bisector of 0 after which it can be enlarged (if ~ > 2~) or we can
arrive at the already excluded Diag. 11 (if ¢~ = 2c0.
4. Each of K2, K 3 must be larger than K 4 .
5. If K2, K 3 are replaced with two smaller circles equal to K 4 , preserving the contact between them as well as the
contact of the lower of the two with side ACbut waiving the contact of the upper one with side AB(Fig. 14), then the new
angle q~/will be smaller than the previous one:

(27)
to' <~0.

It is easy to verify the validity of (27) if we break up the process of decreasing K2, K3 into the following steps. First, we
decrease K 2 to the needed size preserving the point of its contact with T. Next, we move it until it touches K 1and AC; this
requires us to turn Tand thus decrease qo. Next, we decrease K 3to the needed size preserving the point of its contact with
K 1 . The angle ~pagain decreases. Finally, we "roll" the decreased K 3 "over K 1 " toward the decreased K 2 . The angle
tp again decreases.
6. Now, using (27), (25), and Fig. 14 we have

~>~'>~-7>2a, (28)

which contradicts (2).

Excluding Diag. 14. Suppose that the circles are arranged as shown in rigid diagram 14 (Fig. 15). By reason of the
choice of notation we assume / A to be the smallest of the angles of the triangle. Therefore, / A < 7r/3. In view of the
rigidity of the diagram, the rays of the common tangents LN, LM reach AC, AB and, therefore, the common tangent Tof K I,
K2 reaches AC, AB. Suppose that for the sake of definiteness ~ _> ~.
In view of the fact that / A < 7r/3, we have ~ > / A. Reflecting K I in the bisector of 0 , we can then enlarge K.

REFERENCES

1. J. Hadamard, Elementary Geometry Pt. I: Plane Geometry [Russian translation], Moscow (1957).
2. G. D. Balk and M. B. Balk, "Probability test," Kvant, No. 1, 20-25 (1972).
3. V. A. Zalgaller, "An inequality for acute triangles," Ukr. Geom. Sb., No. 34, 10-25 (1991).
4. G. A. Los', Malfatti's Optimization Problem [in Russian], Dep. Ukr. NIINTI July 5, 1988.
5. T. Saaty, Integer Optimization Methods and Related Extremal Problems [Russian translation], Moscow (1973).
6. A. B. Kharazishvili, Introduction to Combinatorial Geometry [in Russian], Tbilisi (1985).
7. H. D6rrie, Triumph der Mathematik (Hundert bertthmte Probleme aus zwei Jahrtausend mathematischer Kultur),
Physica-Verlag, Wtirzburg (1958). [English translation: 100 Great Problems of Elementary Mathematics, Dover,
New York (1965)].
8. H. Eves, in: A Survey of Geometry. 2, Allyn & Bacon (1965), pp. 50-66.
9. H. Gabai and E. Liban, "On Goldberg's inequality associated with the Malfatti problem," Math. Mag., 41, No. 5,
251-252 (1967).

3•76
10. M. Goldberg, "On the original Malfatti problem," Math. Mag., 40, No. 5, 241-247 (1967).
11. H. Lob and H. W. Richmond, "On the solutions of the Malfatti problem for a triangle," Proc. London Math. Soc.,
2, No. 30, 287-301 (1930).
12. C. Malfatti, "Memoria sopra una problema stereotomico," Memoria di Matematica e di Fisica della Societa ttaliana
della Scienze, 10, No. 1,235-244 (1803).
13. A. Procissi, "Questioni connesse col problema di Malfatti e bibliografia," Period. Math., 4, No. 12, 189-205 (1932).

3177