IMO 1983
Read the problems at the official link. The 1983 IMO problems are Β© the International Mathematical Olympiad, so we don't reproduce them here. Open the 1983 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 Functional equation with decay at infinity: the reciprocal
- Hint 1. Put $x=1$ to get $f(f(y))=y\,f(1)$, and find $f(1)$.
- Hint 2. For every $x$, the number $x f(x)$ is a fixed point of $f$. Which fixed points are possible?
Our solution (not official)
Answer: $f(x)=\dfrac1x$
Key idea: $x f(x)$ is always a fixed point, and the set of fixed points is closed under squaring and taking reciprocals. Decay at infinity then allows only the fixed point $1$.
Let $c=f(1)$.
- $x=1$: $f(f(y))=cy$ for all $y$.
- $x=y=1$: $f(c)=c$. Then $f(f(c))=f(c)=c$, but also $f(f(c))=c\cdot c$. So $c^2=c$ and $c=1$.
Fixed points. With $y=x$: $f(xf(x))=xf(x)$, so $u=xf(x)$ is a fixed point for every $x$. If $u$ is a fixed point, then:
- $f(u^2)=f(u\,f(u))=u\,f(u)=u^2$, so $u^{2^k}$ is a fixed point for every $k$;
- with $x=\frac1u$, $y=u$: $f\left(\tfrac1u\cdot u\right)=u\,f\left(\tfrac1u\right)$, so $f\left(\tfrac1u\right)=\tfrac1u$ is a fixed point.
If some fixed point $u\ne1$ existed, then one of $u,\frac1u$ would be a fixed point $v>1$, and $f\left(v^{2^k}\right)=v^{2^k}\to\infty$, against $f(x)\to0$. So the only fixed point is $1$, and $xf(x)=1$ for all $x$.
Check. $f(x)=\frac1x$ gives $f\left(\frac xy\right)=\frac yx=y\,f(x)$, and $\frac1x\to0$.
How we checked it: algebra only: substitutions
P2 Two circles, two tangents: equal angles at an intersection
- Hint 1. Let $S$ be the point where the two common tangents meet, on the line of centers. Where do $M_1$ and $M_2$ lie?
- Hint 2. Show $SA^2=SM_1\cdot SO_2=SM_2\cdot SO_1$ using the homothety at $S$ and the power of $S$. Then find similar triangles.
Our solution (not official)
Key idea: with $S$ the meeting point of the common tangents, $SA^2=SM_1\cdot SO_2=SM_2\cdot SO_1$. That gives two pairs of similar triangles, which move each angle at $A$ to an angle at a center.
If the radii are equal, the common tangents are parallel to $O_1O_2$, so $M_1=O_1$ and $M_2=O_2$, and the two angles are the same angle. From now on let the radii be $r_1<r_2$, and let $k=\frac{r_2}{r_1}$. The two common tangents of intersecting circles meet at the external center of similitude $S$, on line $O_1O_2$, beyond $O_1$. The homothety $h$ with center $S$ and ratio $k$ maps $C_1$ to $C_2$, $P_1\to P_2$ and $O_1\to O_2$.
Where $M_1$, $M_2$ are. The chord $P_1Q_1$ is perpendicular to $SO_1$ (by symmetry in the line of centers), so its midpoint $M_1$ is on segment $SO_1$, and in the right triangle $SP_1O_1$ the foot of the altitude gives $SM_1\cdot SO_1=SP_1^2$. Likewise $SM_2\cdot SO_2=SP_2^2$. With $SP_2=k\,SP_1$ and $SO_2=k\,SO_1$: $$SM_1\cdot SO_2=k\,SP_1^2,\qquad SM_2\cdot SO_1=\frac{k^2SP_1^2}{k}=k\,SP_1^2 .$$
$SA^2=k\,SP_1^2$. The image $h(A)$ lies on $C_2$ and on line $SA$, and $h(A)\ne A$ because $k\ne1$. So line $SA$ meets $C_2$ at $A$ and $h(A)$, and the power of $S$ with respect to $C_2$ is $$SA\cdot S\,h(A)=k\,SA^2=SP_2^2=k^2SP_1^2 .$$ So $SA^2=k\,SP_1^2=SM_1\cdot SO_2=SM_2\cdot SO_1$.
Similar triangles. From $\frac{SA}{SM_1}=\frac{SO_2}{SA}$ and the common angle at $S$, triangles $SAM_1$ and $SO_2A$ are similar, so $\angle SAM_1=\angle SO_2A$. In the same way $\angle SAM_2=\angle SO_1A$.
Finish. The points $S,O_1,O_2$ lie in this order on a line, so in triangle $AO_1O_2$ the exterior angle at $O_1$ gives $\angle SO_1A=\angle O_1AO_2+\angle SO_2A$. Hence $\angle SAM_2-\angle SAM_1=\angle O_1AO_2>0$. Both $M_1$ and $M_2$ lie on the ray from $S$ through $O_1$, so the rays $AM_1$ and $AM_2$ are on the same side of $AS$, and $$\angle M_1AM_2=\angle SAM_2-\angle SAM_1=\angle O_1AO_2 .$$
How we checked it: script: random intersecting circles; the two angles agree to 1e-15
P3 Largest number not representable with three pairwise coprime parts
- Hint 1. For the number itself: reduce a supposed representation modulo $a$, $b$ and $c$, and add $1$ to each coefficient.
- Hint 2. For larger numbers: choose $x<a$ and $y<b$ to fix the remainders modulo $a$ and $b$; the rest is a multiple of $ab$.
Our solution (not official)
Key idea: modulo $a$, only the term $xbc$ matters, so remainders pin down $x$ and $y$; the size bound decides whether $z\ge0$.
Let $N=2abc-ab-bc-ca$.
$N$ is not representable. Suppose $xbc+yca+zab=N$ with $x,y,z\ge0$. Modulo $a$: $xbc\equiv-bc$, and $bc$ is prime to $a$, so $x\equiv-1\pmod a$, hence $x+1\ge a$. Likewise $y+1\ge b$ and $z+1\ge c$. But the equation says $$(x+1)bc+(y+1)ca+(z+1)ab=2abc,$$ while the left side is at least $abc+abc+abc=3abc$. Contradiction.
Every $n>N$ is representable. Choose $x\in\{0,\dots,a-1\}$ with $xbc\equiv n\pmod a$, and $y\in\{0,\dots,b-1\}$ with $yca\equiv n\pmod b$ (possible since $bc$ is prime to $a$ and $ca$ to $b$). Then $m=n-xbc-yca$ is divisible by $a$ (as $yca\equiv0$) and by $b$ (as $xbc\equiv0$), so by $ab$: $m=zab$. And $$m\ge n-(a-1)bc-(b-1)ca>N-(a-1)bc-(b-1)ca=-ab ,$$ so $z>-1$, that is, $z\ge0$. Hence $n=xbc+yca+zab$.
How we checked it: script: for five coprime triples, N is not representable and the next 500 numbers are
P4 Two-coloring a triangle’s outline forces a right triangle
- Hint 1. Mark the points that cut each side in the ratio $1:2$. In the triangle they form, each side is perpendicular to a side of $ABC$.
- Hint 2. Two of the three marked points share a set. Then look at the side of $ABC$ that their segment is perpendicular to.
Our solution (not official)
Answer: Yes: one of the two sets always contains a right triangle
Key idea: the points at one third of each side form a triangle whose sides are perpendicular to the sides of $ABC$. Two of them share a set, and then any point of the matching side either completes a right triangle or the other set fills that side and gets one.
Let the side be $3$. Take $P$ on $BC$ with $BP=1$, $Q$ on $CA$ with $CQ=1$, $R$ on $AB$ with $AR=1$. In triangle $BPR$, $BP=1$, $BR=2$ and $\angle B=60^\circ$, so $PR^2=1+4-2=3$ and $BP^2+PR^2=BR^2$: the angle at $P$ is right, so $RP\perp BC$. In the same way $PQ\perp CA$ and $QR\perp AB$.
Two of $P,Q,R$ are in the same set; by the symmetry of the figure under turning, say $P,R\in S_1$.
- If some $X\ne P$ on side $BC$ is in $S_1$, then $R,P,X$ is a right triangle in $S_1$ (right angle at $P$). Done.
- Otherwise all of $BC$ except $P$ is in $S_2$.
- If $Q\in S_2$: its foot $F$ on $BC$ has $CF=\frac12\ne2=CP$, so $F\in S_2$, and $Q,F,B$ is a right triangle in $S_2$.
- If $Q\in S_1$: then $Q,R\in S_1$ and $QR\perp AB$. If some $Y\ne R$ on $AB$ is in $S_1$, then $Q,R,Y$ is a right triangle in $S_1$. Otherwise all of $AB$ except $R$ is in $S_2$, including $A$. The midpoint $D$ of $BC$ is not $P$, so $D\in S_2$, and $C\in S_2$. Then $A,D,C$ is a right triangle in $S_2$ (right angle at $D$).
In every case one set contains the vertices of a right triangle.
How we checked it: algebra only: case analysis
P5 Many numbers below 100,000 with no three in progression
- Hint 1. Think in base $3$. What happens when you add two numbers whose base-$3$ digits are all $0$ or $1$?
- Hint 2. How many such numbers have at most $11$ digits, and how big is the largest?
Our solution (not official)
Answer: Yes
Key idea: numbers whose base-$3$ digits are only $0$ and $1$ add without carries, so $x+z=2y$ forces $x=z$.
Let $T$ be the set of numbers whose base-$3$ representation, with $11$ digits, uses only $0$ and $1$. It has $2^{11}=2048$ elements, from $0$ up to $1+3+\dots+3^{10}=\frac{3^{11}-1}2=88573$.
No three in progression. Suppose $x,y,z\in T$ with $x+z=2y$. In base $3$, digit by digit, $x_i+z_i\le2$, so adding $x$ and $z$ involves no carries, and the digits of $x+z$ are $x_i+z_i$. The digits of $2y$ are $2y_i\in\{0,2\}$, also without carries. So $x_i+z_i=2y_i$ for every $i$, which with $x_i,z_i\in\{0,1\}$ forces $x_i=z_i=y_i$. So $x=y=z$.
The answer. Add $1$ to every element of $T$; this keeps the property (a progression shifted is still a progression). Now all $2048$ numbers are positive and at most $88574\le10^5$. Any $1983$ of them work. So the answer is yes.
How we checked it: script: the 2048 numbers contain no three-term progression; the largest is 88,574
P6 Cyclic inequality for triangle sides
- Hint 1. Substitute $a=y+z$, $b=z+x$, $c=x+y$ with $x,y,z>0$.
- Hint 2. The expression becomes $2\cdot\left(x y^3+y z^3+z x^3-x y z(x+y+z)\right)$. Divide by $xyz$.
Our solution (not official)
Answer: Equality exactly for equilateral triangles
Key idea: the substitution from the touch points of the incircle turns the inequality into $\frac{y^2}{z}+\frac{z^2}{x}+\frac{x^2}{y}\ge x+y+z$, which is AMβGM term by term.
For a triangle there are positive $x,y,z$ (the tangent lengths from the vertices to the incircle) with $a=y+z$, $b=z+x$, $c=x+y$. Expanding (a routine check), $$a^2b(a-b)+b^2c(b-c)+c^2a(c-a)=2\left(xy^3+yz^3+zx^3-xyz(x+y+z)\right).$$ Divide by $2\,x\,y\,z>0$. We need $$\frac{y^2}{z}+\frac{z^2}{x}+\frac{x^2}{y}\ge x+y+z .$$ By AMβGM, $\frac{y^2}z+z\ge2y$, $\frac{z^2}x+x\ge2z$ and $\frac{x^2}y+y\ge2x$. Adding these gives the claim.
Equality needs $y=z$, $z=x$ and $x=y$, so $x=y=z$, which means $a=b=c$. And for an equilateral triangle every term is $0$.
How we checked it: script: sympy confirms the expansion exactly