IMO 1995
Read the problems at the official link. The 1995 IMO problems are Β© the International Mathematical Olympiad, so we don't reproduce them here. Open the 1995 problems (PDF) from the IMO's official site.
The topic labels, hints and solutions below are our own, written from scratch: our solutions, not official ones. If you spot a mistake, please use βReport a problemβ at the bottom of the page.
P1 Two circles on a line and a concurrence
- Hint 1. Since $P$ is on the radical axis, $PM\cdot PC=PN\cdot PB$. What does that say about $B,C,M,N$?
- Hint 2. Show that $A,M,N,D$ are concyclic, then look for a radical center.
Our solution (not official)
Key idea: $A,M,N,D$ lie on one circle. Then $AM$, $DN$ and $XY$ are the three radical axes of that circle and the two given circles, so they meet.
Let $\omega_1$ and $\omega_2$ be the circles on diameters $AC$ and $BD$. Line $XY$ is their radical axis, and $P$ is on it, so $PM\cdot PC=PN\cdot PB$ (signed). So $B,C,M,N$ are concyclic.
$A,M,N,D$ are concyclic. Use directed angles mod $180^\circ$. Since $AC$ is a diameter, $MA\perp MC$; since $BD$ is a diameter, $NB\perp ND$. Then $$\angle(MA,MN)=\angle(MA,MC)+\angle(MC,MN)=90^\circ+\angle(BC,BN),$$ using the circle $BCMN$ for the second term. Also $$\angle(DA,DN)=\angle(DB,DN)=\angle(DB,BN)+\angle(BN,DN)=\angle(BC,BN)+90^\circ,$$ because lines $DB$, $BC$ and $DA$ are all the line $\ell$ through the four points. So $\angle(MA,MN)=\angle(DA,DN)$, and $A,M,N,D$ lie on a circle $\omega_3$.
Radical center. The radical axes of the pairs among $\omega_1,\omega_2,\omega_3$ are $XY$, $AM$ (for $\omega_1,\omega_3$) and $DN$ (for $\omega_2,\omega_3$). Three radical axes meet at one point or are parallel. They are not parallel: $AM\parallel XY$ would mean $AM\perp\ell$, so $AM$ would be tangent to $\omega_1$ at $A$, forcing $M=A$; then $P$ would be on line $CA=\ell$, so $P=Z$, which is excluded. So $AM$, $DN$, $XY$ are concurrent.
How we checked it: numeric: 500 random configurations
P2 Cyclic inequality with product one
- Hint 1. Substitute $x=\frac1a$, $y=\frac1b$, $z=\frac1c$. What does each term become?
- Hint 2. Then use $\sum\frac{x^2}{y+z}\ge\frac{(x+y+z)^2}{2(x+y+z)}$.
Our solution (not official)
Key idea: after $x=1/a$ and so on, the terms become $\frac{x^2}{y+z}$, and the CauchyβSchwarz (Engel) form plus AMβGM finish.
Let $x=\frac1a$, $y=\frac1b$, $z=\frac1c$, so $xyz=1$. Then $$\frac1{a^3(b+c)}=\frac{x^3}{\frac1y+\frac1z}=\frac{x^3yz}{y+z}=\frac{x^2}{y+z}.$$ By CauchyβSchwarz in Engel form, $$\frac{x^2}{y+z}+\frac{y^2}{z+x}+\frac{z^2}{x+y}\ge\frac{(x+y+z)^2}{2(x+y+z)}=\frac{x+y+z}2\ge\frac{3\sqrt[3]{xyz}}2=\frac32 .$$
How we checked it: reasoning
P3 Triangle areas as sums of point weights
- Hint 1. For four points in convex position, the two ways to cut the quadrilateral along a diagonal give a linear relation between the weights.
- Hint 2. Among five points, look at the convex hull: a pentagon, a quadrilateral with one point inside, or a triangle with two inside.
Our solution (not official)
Answer: $n=4$ only
Key idea: the area rules give linear relations between the weights: for a convex quadrilateral, opposite corners have equal weight sums; for a point inside a triangle, the inside weight is tied to the outer ones. Five points always create a contradiction.
$n=4$ works: the corners of a unit square, with all $r_i=\frac16$.
Two relations. Write $[ijk]$ for the area of triangle $A_iA_jA_k$, equal to $r_i+r_j+r_k$.
- (Q) If $P,Q,R,S$ form a convex quadrilateral with diagonals $PR$ and $QS$, then $[PQR]+[PRS]=[PQS]+[QRS]$, so $r_P+r_R=r_Q+r_S$.
- (T) If $S$ is inside triangle $PQR$, then $[PQR]=[PQS]+[QRS]+[RPS]$, so $r_P+r_Q+r_R+3r_S=0$.
No five points. It is enough to rule out $n=5$. Look at the convex hull.
- Pentagon $A_1\dots A_5$. For each $i$, (Q) on $A_i,A_{i+1},A_{i+2},A_{i+3}$ and on $A_i,A_{i+1},A_{i+2},A_{i+4}$ gives $r_i+r_{i+2}=r_{i+1}+r_{i+3}=r_{i+1}+r_{i+4}$, so $r_{i+3}=r_{i+4}$. Hence all $r_i$ are equal and all triangles have the same area. Then $A_3,A_4,A_5$ are equally far from line $A_1A_2$, on the same side, so they are collinear. Contradiction.
- Quadrilateral $A_1A_2A_3A_4$ with $A_5$ inside. $A_5$ lies in one triangle of each diagonal split; relabel so it is in $A_1A_2A_3$ and $A_1A_2A_4$. (T) twice gives $r_1+r_2+r_3=r_1+r_2+r_4=-3r_5$, so $r_3=r_4$. Line $A_4A_5$ crosses side $A_1A_2$, so it separates $A_1$ from $A_3$; and $A_4,A_5$ lie on opposite sides of $A_1A_3$. So $A_1,A_4,A_3,A_5$ is convex with diagonals $A_1A_3$ and $A_4A_5$, and (Q) gives $r_1+r_3=r_4+r_5$, so $r_1=r_5$. In the same way, line $A_3A_5$ crosses side $A_1A_2$ and separates $A_2$ from $A_4$, so (Q) on diagonals $A_2A_4$, $A_3A_5$ gives $r_2+r_4=r_3+r_5$, so $r_2=r_5$. Now with $s=r_1=r_2=r_5$ and $t=r_3=r_4$, (T) reads $5s+t=0$. Area $[125]=3s>0$, so $t=-5s<0$ and $[341]=2t+s=-9s<0$. Contradiction.
- Triangle $A_1A_2A_3$ with $A_4,A_5$ inside. (T) twice gives $r_4=r_5$. Then $[124]=[125]$, so line $A_4A_5$ is parallel to $A_1A_2$ (both points are on the same side of it); likewise $[134]=[135]$ makes it parallel to $A_1A_3$. Contradiction.
How we checked it: reasoning; relations (Q) and (T) checked on examples
P4 A recurrence that halves or inverts, then returns
- Hint 1. Factor the relation: it says $(2x_i-x_{i-1})(x_ix_{i-1}-1)=0$.
- Hint 2. In terms of $t=\log_2x$, each step is $t\mapsto t-1$ or $t\mapsto-t$. Use parity.
Our solution (not official)
Answer: $x_0=2^{997}$
Key idea: each step either halves or inverts. In base-$2$ logarithms that is “subtract $1$” or “change sign”, and a parity count bounds the start.
Multiply the relation by $x_{i-1}x_i$ and rearrange: $$(2x_i-x_{i-1})\Big(1-\frac1{x_{i-1}x_i}\Big)=0 .$$ So each step is $x_i=\frac{x_{i-1}}2$ or $x_i=\frac1{x_{i-1}}$. With $t_i=\log_2x_i$, each step is $t\mapsto t-1$ or $t\mapsto-t$.
Let $r$ be the number of sign changes and $h=1995-r$ the number of halvings. Following the steps, $t_{1995}=(-1)^rt_0+c$, where $c$ is a sum of $h$ terms $\pm1$. So $|c|\le h$ and $c\equiv h\pmod2$.
- If $r$ is even, $h$ is odd, so $c\ne0$, and $t_{1995}=t_0+c\ne t_0$. Impossible.
- So $r$ is odd, $h\le1994$, and $t_0=-t_0+c$ gives $t_0=\frac c2\le\frac{1994}2=997$.
Hence $x_0\le2^{997}$. This is reached: from $2^{997}$, halve $1994$ times to get $2^{-997}$, then invert to return to $2^{997}$.
How we checked it: reasoning; the sequence for 2^997 checked by script
P5 Hexagon with two equilateral corners and a path bound
- Hint 1. Triangles $BCD$ and $EFA$ are equilateral. Reflect in line $BE$: where do $A$, $D$, $C$, $F$ go?
- Hint 2. A point $G$ with $\angle AGB=120^\circ$ on the right side satisfies $AG+GB=GC'$, where $ABC'$ is equilateral (Ptolemy).
Our solution (not official)
Key idea: reflecting in $BE$ swaps $A$ and $D$. The reflected corners $C'$ and $F'$ are apexes of equilateral triangles on $AB$ and $DE$, and Ptolemy turns $AG+GB$ into $C'G$. A broken line $C'GHF'$ is at least $C'F'=CF$.
Triangle $BCD$ has $BC=CD$ and a $60^\circ$ angle, so it is equilateral and $BD=AB$. Likewise $EFA$ is equilateral and $EA=ED$. So $B$ and $E$ are both equidistant from $A$ and $D$, and the reflection in line $BE$ swaps $A$ and $D$. Let $C'$ and $F'$ be the images of $C$ and $F$. Then $ABC'$ and $DEF'$ are equilateral, and $C'F'=CF$.
Sides. $C$ lies on the far side of line $BD$ from $E$, so $C'$ lies on the far side of line $AB$ from $E$, outside the hexagon. $G$ is inside the convex hexagon, on the same side of $AB$ as $E$. Since $\angle AGB=120^\circ$ and $\angle AC'B=60^\circ$, $G$ lies on the circle $ABC'$, on the arc $AB$ away from $C'$.
Ptolemy. For the cyclic quadrilateral $AGBC'$, $AG\cdot BC'+GB\cdot AC'=GC'\cdot AB$, and the three sides of $ABC'$ are equal, so $AG+GB=GC'$. In the same way $DH+HE=HF'$. Therefore $$AG+GB+GH+DH+HE=C'G+GH+HF'\ge C'F'=CF .$$
How we checked it: reasoning
P6 Half-size subsets with sum divisible by a prime
- Hint 1. Split $\{1,\dots,2p\}$ into $\{1,\dots,p\}$ and $\{p+1,\dots,2p\}$. Shift the part of a subset in the first half cyclically.
- Hint 2. If a subset meets the first half in $j$ elements with $0<j<p$, the shift by $s$ changes the sum by $js$ mod $p$.
Our solution (not official)
Answer: $\dfrac{\binom{2p}{p}-2}{p}+2$
Key idea: rotate the part of the subset that lies in $\{1,\dots,p\}$. Apart from the two halves themselves, subsets fall into groups of $p$, and in each group exactly one sum is divisible by $p$.
Let $X$ be a $p$-element subset other than $\{1,\dots,p\}$ and $\{p+1,\dots,2p\}$, and let $j=|X\cap\{1,\dots,p\}|$, so $0<j<p$. For $s=0,\dots,p-1$, let $X_s$ be obtained by shifting each element of $X\cap\{1,\dots,p\}$ by $s$ cyclically within $\{1,\dots,p\}$, keeping the rest of $X$.
- The sum of $X_s$ is the sum of $X$ plus $js$ mod $p$. Since $p\nmid j$, the $p$ sets $X_0,\dots,X_{p-1}$ have all $p$ residues, one each.
- These $p$ sets are distinct: a $j$-element subset of $\mathbb Z_p$ that is fixed by a nonzero shift is a union of cosets of the whole group, so $j\in\{0,p\}$.
So the $\binom{2p}p-2$ such subsets split into $\frac{\binom{2p}p-2}p$ groups, each with exactly one good subset. Both halves are good: their sums are $\frac{p(p+1)}2$ and $p^2+\frac{p(p+1)}2$, multiples of $p$ since $p$ is odd. The total is $$\frac{\binom{2p}p-2}p+2 .$$
How we checked it: script: brute force for p = 3, 5, 7 gives 8, 52, 492