Problem 1
A convex quadrilateral ABCD has perpendicular diagonals. The perpendicular bisectors
of AB and CD meet at a unique point P inside ABCD. Prove that ABCD is cyclic if and
only if triangles ABP and CDP have equal areas.
Solution:
Let AC and BD intersect at E. Suppose by symmetry that P is in ABE. Denote by M
and N the respective feet of perpendiculars from P to AC and BD. Without assuming that
PA PB and PC PD, we express the areas ABP and CDP as follows:
2ABP 2ABE 2PAE 2PBE
AM PNBN PM AM PNPM BN PMPN
AM BN PM PN,
2CDP 2CDE 2PCE 2PDE
CM PNDN PM CM PNPM DN PMPN
CM DN PM PN.
Therefore
2ABP CDP AM BN CM DN.
We now assume that PA PB and PC PD. Suppose that ABCD is cyclic. Then the
uniqueness of P implies that it must be the circumcenter. So M and N are the midpoints
of AC and BD, respectively. Hence AM CM and BN DN,so(*)
implies ABP CDP. Conversely, suppose that ABP CDP. Then, by (*), we have
AM BN CM DN.IfPA PC, assume by symmetry that PA PC. Then AM CM
and also BN DN, because PB PD. Thus AM BN CM DN, a contradiction. Hence
PA PC, which implies that P is equidistant from A, B, C and D. We conclude then that
ABCD is cyclic.
Alternative Solution:
Let AC and BD meet at E again. Assume by symmetry that P lies in BEC and denote
ABE and ACD . The triangles ABP and CDP are isosceles. If M and N are the
respective midpoints of their bases AB and CD, then PM AB and PN CD. Note that M,
N and P are not collinear due to the uniqueness of P.
Consider the median EM to the hypotenuse of the right triangle ABE.Wehave
BEM , AME 2 and EMP 90
2. Likewise, CEN , DNE 2
and ENP 90
2. Hence MEN 90
, and a direct computation yields
NPM 360
EMP MEN ENP 90
MEN.
It turns out that, whenever AC BD, the quadrilateral EMPN has a pair of equal opposite
angles, the ones at E and P.
We now prove our claim. Since AB 2EM and CD 2EN,wehaveABP
CDP
if and only if EM PM EN PN,or,EM/EN PN/PM. On account
of MEN NPM, the latter is equivalent to EMN PNM. This holds if and only if
EMN PNM and ENM PMN, and these in turn mean that EMPN is a
parallelogram. But the opposite angles of EMPN at E and P are always equal, as noted
above. So it is a parallelogram if and only if EMP ENP;thatis,if
90
2 90
2. We thus obtain a condition equivalent to ,ortoABCD being
cyclic.
Another Solution:
Choose a coordinate system so that the axes lie on the perpendicular lines AC and BD,
and so that the coordinates of A, B, C and D are 0,a, b,0, 0,c and d,0, respectively.
By assumption, the perpendicular bisectors of AB and CD have a unique common point.
Hence the linear system formed by their equations 2bx 2ay b
2
a
2
and 2dx 2cy d
2
c
2
has a unique solution, and it is given by
x
0
cb
2
a
2
ad
2
c
2
2ad bc
, y
0
db
2
a
2
bd
2
c
2
2ad bc
.
Naturally, x
0
,y
0
are the coordinates of P. Since it is interior to ABCD, ABP and CDP
have the same orientation. Then ABP CDP if and only if
||
||
This is equivalent to ax
0
by
0
ab cx
0
dy
0
cd. Inserting the expressions for x
0
and y
0
, after the inevitable algebra work we obtain the equivalent condition
ac bda c
2
b d
2
0.
Now, the choice of the coordinate system implies that a and c have different signs, as well
as b and d. Hence the second factor is nonzero, so ABP CDP if and only if ac bd.
The latter is equivalent to AE CE BE DE, where E is the intersection point of the
diagonals. However, it is a necessary and sufficient condition for ABCD to be cyclic.
Problem 2
Let ABCD be a cyclic quadrilateral. Let E and F be variable points on the sides AB and
CD, respectively, such that AE : EB CF : FD.LetP be the point on the segment EF
such that PE : PF AB : CD. Prove that the ratio between the areas of triangles APD and
BPC does not depend on the choice of E and F.
Solution:
We first assume that the lines AD and BC are not parallel and meet at S.SinceABCD is
cyclic, ASB and CSD are similar. Then, since AE : EB CF : FD, ASE and CSF are
also similar, so that
DSE CSF.
Moreover, we have
SE
SF
SA
SC
AB
CD
PE
PF
,
which implies that ESP FSP. Thus, ASP BSP,andsoP is equidistant from
the lines AD and BC. Therefore,
APD : BPC AD : BC,
which is a constant.
Next, assume that the lines AD and BC are parallel. Then ABCD is an isosceles
trapezoid with AB CD, and we have BE DF.LetM and N be the midpoints of AB and
CD, respectively. Then ME NF and clearly, E and F are equidistant from the line MN.
Thus P, the midpoint of EF, lies on MN. It follows that P is equidistant from AD and BC,
and hence
APD : BPC AD : BC.
Alternative Solution:
Let AE : EB CF : FD a : b,wherea b 1. Since PE : PF AB : CD,we
have
d
P,AD
CD
AB CD
d
E,AD
AB
AB CD
d
F,AD
,
d
P,BC
CD
AB CD
d
E,BC
AB
AB CD
d
F,BC
,
where dX,YZ stands for the distance from the point X to the line YZ. Thus, we obtain
APD
CD
AB CD
AED
AB
AB CD
AFD
a CD
AB CD
ABD
b AB
AB CD
ACD,
BPC
CD
AB CD
BEC
AB
AB CD
BFC
b CD
AB CD
BAC
a AB
AB CD
BDC.
Next, since A, B, C and D areconcyclic,wehavesinBAD sinBCD and
sinABC sinADC. Thus,
APD
BPC
a CD ABD b AB ACD
b CD BAC a AB BDC
a CD AB AD sinBAD b AB CD AD sinADC
b CD AB BC sinABC a AB CD BC sinBCD
AD
BC
a sinBAD b sinADC
b sinABC a sinBCD
AD
BC
.
Problem 3
Let I be the incenter of triangle ABC.LetK,L and M be the points of tangency of the
incircle of ABC with AB,BC and CA, respectively. The line t passes through B and is
parallel to KL. The lines MK and ML intersect t at the points R and S. Prove that RIS is
acute.
Solution:
Since the lines KL and RS are parallel, we have, in BKR ,
BKR 90° A/2,
KBR 90° B/2,
BRK 90° C/2.
Hence, by the law of sine,
BR
cosA/2
cosC/2
BK. #
Similarly, we have, in BLS,
BLS 90° C/2,
BSL 90° A/2,
LBS 90° B/2.
so that
BS
cos
C/2
cos
A/2
BL
cos
C/2
cos
A/2
BK. #
Notice now that BI RS and IK AB. Then, on account of (1) and (2), we obtain
IR
2
IS
2
RS
2
BI
2
BR
2
BI
2
BS
2
BR BS
2
2
BI
2
BR BS
2
BI
2
BK
2
2IK
2
0.
So,bythelawofcosine,RIS is acute.
Alternative Solution:
Let W be the midpoint of KL and Q the midpoint of KM. Then Q AI, W BI,
AIKM and BIKL. We first prove that AWRI and CWSI.
Since RBI RQI 90
, the points R, B, I, Q are concyclic, which implies
BRI BQI. Also, in the right triangles AIK and BIK,wehave
IQ IA KI
2
IW IB,
so that IQ/IW IB/IA. Hence AIW and BIQ are similar. It follows that
BRI BQI AWI.SinceBR IW, this implies that RIAW.Byasimilar
argument, we can prove that SICW. Thus, RIS 180
AWC. It remains to prove
that AWC is obtuse.
Let T be the midpoint of AC.Then2
WT WC WA LC KA.SinceLC and KA are
not collinear, we have
WT
LC KA
2
CM AM
2
AC
2
.
This implies that W is inside the circle with diameter AC,andsoAWC 90°.
Therefore, RIS is acute.
Comment:
Another proof for the fact AW RI is as follows. Since RBI RQI 90°, RBIQ
is cyclic and its circumcircle is orthogonal to the diameter RI. Consider the inversion with
respect to the incircle of ABC.Since takes B and Q into W and A, respectively, it takes
the circumcircle of RBIQ into the line AW.Since leaves the line RI invariant, we have
AW RI.
Problem 4
Let M and N be points inside triangle ABC such that
MAB NAC MBA NBC.
Prove that
AM AN
AB AC
BM BN
BA BC
CM CN
CA CB
1.
Solution:
Let K be the point on the ray BN such that BCK BMA. Note that K is outside
ABC, because BMA ACB.SinceMBA CBK,wehaveABM KBC;so,
AB
BK
BM
BC
AM
CK
. #
Then, since ABK MBC,andAB/KB BM/BC, we see that ABK MBC. Hence
AB
BM
BK
BC
AK
CM
. #
Now we have CKN MAB NAC. Consequently, the points A, N, C and K are
concyclic. By Ptolemy’s theorem, AC NK AN CK CN AK,or
AC
BK BN
AN CK CN AK. #
From (1) and (2), we find CK AM BC/BM, AK AB CM/BM and BK AB BC/BM.
Inserting these expressions in (3), we obtain
AC
AB BC
BM
BN
AN AM BC
BM
CN AB CM
BM
,
or
AM AN
AB AC
BM BN
BA BC
CM CN
CA CB
1.
Alternative Solution:
Let the complex coordinates of A, B, C, M and N be a, b, c, m and n, respectively. Since
the lines AM, BM and CM are concurrent, as well as the lines AN, BN and CN, it follows
from Ceva’s theorem that
sinBAM
sinMAC
sinCBM
sinMBA
sinACM
sinMCB
1. #
sinBAN
sinNAC
sinCBN
sinNBA
sinACN
sinNCB
1. #
By hypotheses, BAM NAC and MBA CBN. Hence BAN MAC
and NBA CBM. Combined with (1) and (2), these equalities imply
sinACM sinACN sinMCB sinNCB.
Thus,
cos
NCM 2ACM
cosNCM cos
NCM 2NCB
cosNCM.
and hence ACM NCB.
Since BAM NAC, MBA CBN and ACN MCB, the following
complex ratios are all positive real numbers:
m a
b a
:
c a
n a
,
m b
a b
:
c b
n b
and
m c
b c
:
a c
n c
.
Hence each of these equals its absolute value, and so
AM AN
AB AC
BM BN
BA BC
CM CN
CA CB
m a
n a
b a
c a
m b
n b
a b
c b
m c
n c
b c
a c
1.
Comment:
Contestants who are familiar with the notion of isogonal conjugate points may skip
over the early part of the proof.
Problem 5
Let ABC be a triangle, H its orthocenter, O its circumcenter, and R its circumradius. Let
D be the reflection of A across BC, E be that of B across CA,andF that of C across AB.
Prove that D,E and F are collinear if and only if OH 2R.
Solution:
Let G be the centroid of ABC,andA
, B
and C
be the midpoints of BC, CA and AB,
respectively. Let A
B
C
be the triangle for which A, B and C are the midpoints of B
C
,
C
A
and A
B
, respectively. Then G is the centroid and H the circumcenter of A
B
C
.
Let D
, E
and F
denote the projections of O on the lines B
C
, C
A
and A
B
,
respectively.
Consider the homothety h with center G and ratio 1/2. It maps A, B, C, A
, B
and C
into
A
, B
, C
, A , B and C, respectively. Note that A
D
BC, which implies
AD : A
D
2:1 GA : GA
and DAG D
A
G. We conclude that hD D
.
Similarly, h
E
E
and h
F
F
. Thus, D, E and F are collinear if and only if D
, E
and F
are collinear. Now D
, E
and F
are the projections of O on the sides B
C
, C
A
and A
B
, respectively. By Simson’s theorem, they are collinear if and only if O lies on the
circumcircle of A
B
C
. Since the circumradius of A
B
C
is 2R, O lies on its
circumcircle if and only if OH 2R.
Alternative Solution:
Let the complex coordinates of A, B, C, H and O be a, b, c, h and 0, respectively.
Consequently, aa
bb
cc
R
2
and h a b c.SinceD is symmetric to A with
respect to line BC, the complex coordinates d and a satisfy
d b
c b
a b
c b
,or
b
c
d
b c
a
bc
b
c
0. #
Since
b
c
R
2
b c
bc
and bc
b
c
R
2
b
2
c
2
bc
,
by inserting these expressions in (1), we obtain
d
bc ca ab
a
k 2bc
a
.
d
R
2
a b c
bc
R
2
h 2a
bc
.
where k bc ca ab. Similarly, we have
e
k 2ca
b
, e
R
2
h 2b
ca
, f
k 2ab
c
andf
R
2
h 2c
ab
.
Since
dd
1
ee
1
ff
1
e de
d
f df
d
ba
k2ab
ab
R
2
ab
h2c
abc
ca
k2ca
ca
R
2
ac
h2b
abc
R
2
c a
a b
a
2
b
2
c
2
ck 2abc
h 2c
bk 2abc
h 2b
R
2
b c
c a
a b
hk 4abc
a
2
b
2
c
2
and h
R
2
k/abc, it follows that
D,EandFarecollinear
0
hk 4abc 0
hh
4R
2
OH 2R.
Problem 6
Let ABCDEF be a convex hexagon such that B D F 360
and
AB
BC
CD
DE
EF
FA
1.
Prove that
BC
CA
AE
EF
FD
DB
1.
Solution:
Let P be the point such that FEA DEP and EFA EDP, where the angles
are directed. Then FEA and DEP are similar. Hence
FA
EF
DP
DE
. #
EF
ED
EA
EP
. #
It follows from (1) and the given conditions that ABC PDC and
AB
BC
DE FA
CD EF
DP
CD
.
Therefore, ABC and PDC are similar, which gives BCA DCP and
CB
CD
CA
CP
. #
Since FED AEP, (2) implies that FED and AEP are similar. Analogously, since
BCD ACP, (3) yields the similarity of BCD and ACP. Hence
FD
EF
PA
AE
and
BC
DB
CA
PA
.
Multiplying these together, we have the desired result.
Alternative Solution:
Let the complex coordinates of A,B,C,D,E and F be a,b,c,d,e and f, respectively.
Since ABCDEF is convex, B, D and F are the arguments of the complex numbers
a b
/
c b
,
c d
/
e d
and
e f
/
a f
, respectively. Then the condition
B D F 360
implies that the product of these three complex numbers is a
positive real number. It is equal to the product of their absolute values AB/BC,CD/DE and
EF/FA.Since
AB/BC
CD/DE
EF/FA 1, we have
a b
c b
c d
e d
e f
a f
1.
On the other hand, since
a bc de f c be da f
b ca ef d a cf eb d,
we deduce immediately that
b c
a c
a e
f e
f d
b d
1.
Taking absolute values on both sides gives
BC
CA
AE
EF
FD
DB
1.
Comment:
Considering the arguments of the complex numbers on both sides of the equality
b c
a c
a e
f e
f d
b d
1.
it shows that BDF AEF ACB.
Problem 7
Let ABC be a triangle such that ACB 2ABC.LetD be the point on the side BC
such that CD 2BD. The segment AD is extended to E so that AD DE. Prove that
ECB 180
2EBC.
Solution:
Let H be the midpoint of CD,sothatABEH is a parallelogram. Extend BC to G so that
CG CA.TakeBD DH HC a/3,CA b,AB c,BE AH x,AD DE y
and CE z.Since2ABC ACB CAG CGA 2CGA, ABG and CAG
are similar. Hence AB/BG CA/AG,or
c
2
ba b. #
Applying Apollonius’ theorem to the triangles ACD,ABH and CDE,wehave
b
2
y
2
2x
2
2a
2
9
, #
x
2
c
2
2y
2
2a
2
9
, #
y
2
z
2
2c
2
2a
2
9
. #
Eliminating y from (2) and (3), we have
x
2
c
2
2b
2
4x
2
2a
2
3
.
Combining this with (1), we have
x
2
b
2a
3
b
a
3
. #
Eliminating y from (3) and (4), we have
x
2
c
2
2z
2
4c
2
2a
2
3
.
Combining this with (1) and (5), we have z b 2a/3. From this and (5), we obtain
x
2
zz a,orBE
2
CECE BC CE EP,
where P is the point on CE such that CP BC. Then BE/CE EP/BE. Hence BEP and
CEB are similar since BEP CEB. It follows that
ECB EBP EBC
1
2
180
ECB,
which simplifies to ECB 180
2EBC.
Alternative Solution:
Lemma. In a triangle PQR,ifS lies on the side QR and divides it in the ratio k :1,
then
k 1
cotPSR cotPQR kcotPRQ.
Proof. Let PSR , PQR and PRS .Thenwehave
k
QS
SR
QSP
SRP
PQsin
PRsin
sinsin
sinsin
sinsincos sincossin
sinsincos sincossin
cot cot
cot cot
.
The result follows at once.
Proof of the main result.
Put a BC,b CA and c AB.LetABC and ACB 2.LetH be the
midpoint of CD. Then ABEH is a parallelogram. We have
cotACB
1 tan
2
2tan
1
2
r
1
r
,
where r cot. Note that r 0 because must be acute. By the lemma with k
1
2
on
ABC,wehave
3
2
cotADC r
1
4
r
1
r
,
cotADC
1
6
3r
1
r
.
By the lemma with k 1onACD,wehave
2cotAHC cotADC cotACB,or
cotAHC
1
12
3r
1
r
1
4
r
1
r
1
3r
,
cotEBC cotAHB
1
3r
0.
By the lemma with k 2onECB,wehave
3cotEDB cotECB 2cotEBC,
cotECB
3
6
3r
1
r
2
3r
3
2
r
1
6r
1
2cotEBC
1
2
cotEBC
cot2EBC
cot
2EBC 180
.
Since 0 ECB 180° and 90° EBC 180°, we conclude that
ECB 2EBC 180°.
Another Solution:
Let the complex coordinates of A, B, C, D and E be a, b, c, d and e, respectively. Then
d
2b c
/3 and e 2d a.SinceACB 2ABC, the ratio
a b
c b
2
:
b c
a c
is real and positive. It is equal to AB
2
AC/BC
3
. On the other hand, a direct computation
shows that the ratio
e c
b c
:
c b
e b
2
is equal to
1
b c
3
b a
2
c a
3
2
4
b a
c a
3
4
27
b a
2
c a
b c
3
4
27
AB
2
AC
BC
3
,
which is a real number. Hence the arguments of e c/b c and c b
2
/e b
2
,
namely, ECB and 2EBC, differ by an integer multiple of 180°. We easily infer that
either ECB 2EBC or ECB 2EBC 180
, according as the ratio is positive or
negative. To prove that the latter holds, we have to show that AB
2
AC/BC
3
is greater than
4/27. Choose a point F on the ray AC such that CF CB.
Since CBF is isosceles and ACB 2ABC,wehaveCFB ABC. Thus, ABF
and ACB are similar and AB : AF AC : AB.SinceAF AC BC,
AB
2
AC
AC BC
.LetAC u
2
and AC BC v
2
. Then AB uv and BC v
2
u
2
.
From AB AC BC, we obtain u/v 1/2. Thus
AB
2
AC
BC
3
u
4
v
2
v
2
u
2
3
u/v
4
1 u
2
/v
2
3
1/2
4
1 1/4
3
4
27
.
and the conclusion follows.
Problem 8
Let ABC be a triangle such that A 90
and B C. The tangent at A to its
circumcircle meets the line BC at D.LetE be the reflection of A across BC, X the foot of
perpendicular from A to BE,andY the midpoint of AX. Let the line BY meet again at Z.
Prove that the line BD is tangent to the circumcircle of triangle ADZ.
Solution:
Let G be diametrically opposite A,andH the point of intersection of AE and BD. Note
that B and G are on the same side of AE because of the condition B C. We claim
that G, H and Z are collinear. Indeed, AEG 90
AXB and, in addition,
AGE ABE ABX. Hence AGE and ABX are similar. It follows that
GAE BAX and GA/BA AE/AX. Moreover, AE/AX AH/AY,sinceH and Y are
the midpoints of AE and AX, respectively. So AGH and ABY are also similar. Hence
AGH ABY,andifGH meets at Z
, then ABZ
AGZ
ABY ABZ.
This implies that Z and Z
coincide, both of them being on the minor arc AE. This justifies
the claim.
We now prove that DAZ ZDH.SinceG, H and Z are in a line,
AZH AZG 90
.SinceDA and DE are tangents to ,wehaveDAZ AEZ
and DEZ EAZ. Hence DHZ 90
AHZ EAZ DEZ,sothatZHED
is a cyclic quadrilateral. It follows that ZDH ZEA DAZ,andDB is indeed
tangent to the circumcircle of ADZ.
Alternative Solution:
As in the solution above, we prove that G, H and Z are collinear. Next, the
circumcircle of ADZ and BD are tangent if and only if so are their images under the
inversion with respect to . Note that leaves A and Z unchanged, as well as the line BD.
Also, it takes D to the midpoint H of the chord AE.Sotheimageof under the inversion is
the circumcircle
of AHZ. It always has a point in common with BD, namely, H.
Besides, AH BD. Hence
and BD are tangent if and only if AH is a diameter of
,
which is equivalent to AZH 90
. On the other hand, AZG 90
,sinceAG is a
diameter of . We infer that is tangent to BD if and only if G, H and Z are collinear,
which was already proven.
Another Solution:
Let the rectangular coordinates of A, B and C be A
a
1
,a
2
, B
0,0
and C
c,0
,
respectively. Then the equation of the circumcircle of ABC is x
2
y
2
cx.SinceA lies
on the circumcircle, we have a
1
2
a
2
2
ca
1
. Since the equation of the tangent line of the
circumcircle at A is 2a
1
x 2a
2
y c
x a
1
, the coordinates of D are
D
ca
1
/
2a
1
c
,0
or
D
a
1
a
1
2
a
2
2
a
1
2
a
2
2
,0 .
Since E is symmetric to A across the x-axis, its coordinates are E
a
1
,a
2
. Then, the
equation of line BE is a
2
x a
1
y 0, and the equation of the perpendicular AX is
a
1
x a
2
y a
1
2
a
2
2
. Solving these two equations, we obtain the coordinates of the point X
as follows : X
a
1
a
1
2
a
2
2
/
a
1
2
a
2
2
,a
2
a
1
2
a
2
2
/
a
1
2
a
2
2
, and the coordinates of the
midpoint of AX are Y
a
1
3
/
a
1
2
a
2
2
,a
2
3
/
a
1
2
a
2
2
. Thus, the equation of line BY is
a
2
3
x a
1
3
y 0, and the coordinates of the point of intersection of line BY and the
circumcircle are
Z
a
1
5
a
1
2
a
2
2
a
1
6
a
2
6
,
a
1
2
a
2
3
a
1
2
a
2
2
a
1
6
a
2
6
.
The equation of the circle through A, D and Z is
x
2
y
2
2a
1
a
1
2
a
2
2
a
1
2
a
2
2
x
a
2
a
1
2
a
2
2
2
a
1
2
a
2
2
2
y
a
1
2
a
1
2
a
2
2
2
a
1
2
a
2
2
2
0.
which is obviously tangent to the line BD (the x-axis) at D.
Comment:
Minor changes in the proofs show that the result remains true if B C. Also, one
may actually claim (see the inversive solution) that: if Z is a point on the minor arc
AC
of , then the circumcircle of ADZ is tangent to BD if and only if BZ passes through the
midpoint of AX, the perpendicular from A to BE. Again, no new ideas are involved in
achieving this. It therefore seems reasonable not to add artificial difficulty, and leave the
problem the way it was proposed. The synthetic proofs are deceivingly short; they do
require time and thinking.
Problem 9
Let a
1
,a
2
,,a
n
be positive real numbers such that a
1
a
2
a
n
1. Prove that
a
1
a
2
a
n
1 a
1
a
2
a
n
a
1
a
2
a
n
1 a
1
1 a
2
1 a
n
1
n
n1
.
Solution:
Set a
n1
1 a
1
a
2
a
n
. Clearly a
n1
0; we thus obtain n 1 positive
numbers a
1
,a
2
,,a
n1
adding up to 1. With this notation, the inequality takes the form
n
n1
a
1
a
2
a
n
a
n1
1 a
1
1 a
2
1 a
n
1 a
n1
.
Using the AM-GM inequality, we have for each i 1,2,,n 1,
1 a
i
a
1
a
i1
a
i1
a
n1
n
n
a
1
a
i1
a
i1
a
n1
n
n
a
1
a
2
a
n
a
n1
a
i
.
Multiplying these n 1 inequalities, we conclude that
1 a
1
1 a
2
1 a
n
1 a
n1
n
n1
n
a
1
a
2
a
n
a
n1
n1
a
1
a
2
a
n
a
n1
n
n1
a
1
a
2
a
n
a
n1
,
as desired. If n 2, the equality occurs if and only if a
1
a
2
a
n
a
n1
;thatis,if
and only if the original numbers a
1
,a
2
,,a
n
are all equal to 1/n 1.Inthecasen 1,
however, we have equality for any a
1
0,1.
Problem 10
Let r
1
,r
2
,,r
n
be real numbers greater than or equal to 1. Prove that
1
r
1
1
1
r
2
1
1
r
n
1
n
n
r
1
r
2
r
n
1
.
Solution:
The case n 1 is clear. We now prove the inequality for all n of the form n 2
k
,
k 1,2, , by induction on k.Fork 1, we have
1
r
1
1
1
r
2
1
2
r
1
r
2
1
r
1
r
2
1 r
1
r
2
2
r
1
1r
2
1 r
1
r
2
1
0.
To make the inductive step, it suffices to prove that if the claim holds for any n numbers,
then it also holds for any 2n numbers. Indeed, if r
1
,r
2
,,r
2n
1, we have
i1
2n
1
r
i
1
i1
n
1
r
i
1
in1
2n
1
r
i
1
n
n
r
1
r
2
r
n
1
n
n
r
n1
r
n2
r
2n
1
2n
2n
r
1
r
2
r
2n
1
.
The induction is complete.
To prove the inequality for any n,letk be a positive integer such that m 2
k
n. Then
put r
n1
r
n2
r
m
n
r
1
r
2
r
n
to obtain
1
r
1
1
1
r
n
1
m n
n
r
1
r
2
r
n
1
m
n
r
1
r
2
r
n
1
.
This completes the proof.
Comment:
Here is another approach to this problem. Let r
i
e
a
i
. Note that a
i
0, since r
i
1.
Then the inequality is equivalent to
1
e
a
1
1
1
e
a
2
1
1
e
a
n
1
n
e
1
n
i1
n
a
i
1
.
Consider the function
fx
1
e
x
1
.
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
-4 -2 2 4
x
The graph of fx
1
e
x
1
We compute
f
x e
x
1
2
e
x
,
f
x e
x
1
3
e
x
e
x
1.
Since f
x 0forx 0, fx is convex on 0,. Thus,
1
n
fa
1
f a
2
fa
n
f
a
1
a
2
a
n
n
;
that is,
1
e
a
1
1
1
e
a
2
1
1
e
a
n
1
n
e
1
n
i1
n
a
i
1
.
Problem 11
Let x,y and z be positive real numbers such that xyz 1. Prove that
x
3
1 y1 z
y
3
1 z1 x
z
3
1 x1 y
3
4
.
Solution:
The inequality is equivalent to the following one:
x
4
x
3
y
4
y
3
z
4
z
3
3
4
x 1y 1z 1.
In fact, a stronger inequality holds true, namely
x
4
x
3
y
4
y
3
z
4
z
3
1
4
x 1
3
y 1
3
z 1
3
.
(It is indeed stronger, since u
3
v
3
w
3
3uvw for any positive numbers u,v and w.) To
represent the difference between the left- and the right-hand sides, put
ft t
4
t
3
1
4
t 1
3
, gt t 14t
2
3t 1.
We have ft
1
4
t 1gt.Also,g is a strictly increasing function on 0,,takingon
positive values for t 0. Since
x
4
x
3
y
4
y
3
z
4
z
3
1
4
x 1
3
y 1
3
z 1
3
fx fy fz
1
4
x 1gx
1
4
y 1gy
1
4
z 1gz,
it suffices to show that the last expression is nonnegative.
Assume that x y z; then gx gy gz 0. Since xyz 1, we have x 1
and z 1. Hence x 1gx x 1gy and z 1gy z 1gz.So,
1
4
x 1gx
1
4
y 1gy
1
4
z 1gz
1
4
x 1 y 1 z 1gy
1
4
x y z 3gy
1
4
3
3
xyz 3gy 0,
because xyz 1. This completes the proof. Clearly, the equality occurs if and only if
x y z 1.
Alternative Solution:
Assume x y z so that
1
1 y1 z
1
1 z1 x
1
1 x1 y
.
Then Chebyshev’s inequality gives that
x
3
1 y1 z
y
3
1 z1 x
z
3
1 x1 y
1
3
x
3
y
3
z
3
1
1 y1 z
1
1 z1 x
1
1 x1 y
1
3
x
3
y
3
z
3
3 x y z
1 x1 y1 z
.
Now, setting x y z/3 a for convenience, we have by the AM-GM inequality
1
3
x
3
y
3
z
3
a
3
,
x y z 3
3
xyz 3,
1 x1 y1 z
1 x 1 y 1 z
3
3
1 a
3
.
It follows that
x
3
1 y1 z
y
3
1 z1 x
z
3
1 x1 y
a
3
3 3
1 a
3
.
So, it suffices to show that
6a
3
1 a
3
3
4
;
or, 8a
3
1 a
3
. This is true, because a 1. Clearly, the equality occurs if and only if
x y z 1. The proof is complete.
Comment:
None of the solutions above actually uses the condition xyz 1. They both work,
provided that x y z 3. Moreover, the alternative solution also shows that the
inequality still holds if the exponent 3 is replaced by any number greater than or equal to 3.
Problem 12
Let n k 0 be integers. The numbers cn,k are defined as follows:
cn,0 cn,n 1n 0;
cn 1,k 2
k
cn,k cn,k 1n k 1.
Prove that c n,k cn,n k for all n k 0.
Solution:
The assertion is clear for n 0:
c0,0 c0,0 0 1.
The general case follows by induction on n. Suppose
cm,j cm,m j for all n m j 0.
Then, by the given recurrence relation and the induction hypothesis, we have
cn 1,k 2
k
cn,k cn,k 1,
cn 1,n 1 k 2
n1k
cn,n 1 k cn,n k
2
n1k
cn,k 1 cn,k.
To finish the proof we show that:
2
k
1cn,k 2
n1k
1cn,k 1. #
Note that the induction hypothesis implies that
2
k
1cn 1,k 2
nk
1cn 1,k 1, #
2
k1
1cn 1,k 1 2
nk1
1cn 1,k 2. #
We compute
2
k
1cn,k 2
n1k
1cn,k 1
2
k
1
2
k
cn 1,k cn 1,k 1
2
n1k
1
2
k1
cn 1,k 1 cn 1,k 2
2
k
2
nk
1cn 1,k 1 2
k
1cn 1,k 1
2
n1k
1
2
k1
cn 1,k 1 cn 1,k 2
2
n
2
k
2
k
1 2
n
2
k1
cn 1,k 1
2
n1k
1cn 1,k 2
2
k1
1cn 1,k 1 2
n1k
1cn 1,k 2
0
where we have used the given recurrence relation in the first step, used (2) in the second
step, and used (3) in the last step. Thus, (1) holds, which completes the proof.
Alternative Solution:
Consider the sequence of polynomials:
P
n
x
j0
n1
x 2
j
k0
n
an,kx
k
, n 1,2,3,,
and notice the two recurrence relations:
P
n1
x P
n
xx 2
n
,
P
n1
x 2
n
x 1P
n
x/2.