IMO 1973
Read the problems at the official link. The 1973 IMO problems are ยฉ the International Mathematical Olympiad, so we don't reproduce them here. Open the 1973 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 โReport a problemโ at the bottom of the page.
P1 Odd number of unit vectors in a half-plane
- Hint 1. Induct on the odd number $n$, removing the two outermost vectors.
- Hint 2. The sum of the middle vectors points somewhere inside the angle spanned by the two outer ones. Compare its direction with the sum of the outer pair.
Our solution (not official)
Key idea: the two outermost vectors add up to a vector that makes an angle of at most $90^\circ$ with the sum of all the others. Adding it cannot make that sum shorter.
Induct on odd $n$. For $n=1$ the length is $1$. Let $n\ge3$ and measure each vector by its angle $\theta\in[0^\circ,180^\circ]$ from the ray of $g$. Sort them: $\theta_1\le\theta_2\le\dots\le\theta_n$.
Let $v$ be the sum of the middle $n-2$ vectors. By induction $|v|\ge1$. Each middle vector lies in the angle $[\theta_1,\theta_n]$, which is convex because it is at most $180^\circ$ wide. A sum of vectors from a convex angle stays in it, so the direction $\psi$ of $v$ satisfies $\theta_1\le\psi\le\theta_n$.
The outer pair adds up to $s=u_1+u_n$, which points along the bisector $\frac{\theta_1+\theta_n}2$ (or is $0$). The angle between $s$ and $v$ is at most $\frac{\theta_n-\theta_1}2\le90^\circ$, so $s\cdot v\ge0$. Hence $$|s+v|^2=|s|^2+2\,s\cdot v+|v|^2\ge|v|^2\ge1 .$$
How we checked it: script: 20,000 random odd families have smallest sum length 1.000 (attained when vectors cancel in pairs)
P2 Space point set where every segment has a parallel twin
- Hint 1. Try a small cubic grid. A shift by a coordinate unit vector often moves a segment to a parallel one.
- Hint 2. The only segments that cannot be shifted are the long diagonals of the big cube. Use half of one of them.
Our solution (not official)
Answer: Yes. For example the $27$ points of the grid $\{0,1,2\}^3$
Key idea: in the grid $\{0,1,2\}^3$, a segment can almost always be slid one step along some axis. The four long diagonals, which cannot, have a parallel half-length twin.
Let $G=\{0,1,2\}^3$; its $27$ points are not coplanar. Take two points $A\ne B$ of $G$ and let $d=B-A$. We find another pair of $G$ on a parallel but different line.
- Some coordinate of $d$ is $0$, say the $i$-th. Then $A$ and $B$ have the same $i$-th coordinate $t$. Pick $t'\in\{0,1,2\}$ with $|t'-t|=1$ and move both points by $(t'-t)e_i$. Both stay in $G$, the segment stays parallel, and the line moves, because $e_i$ is not parallel to $d$.
- Some coordinate of $d$ is $\pm1$, say the $i$-th, and the others are nonzero. The $i$-th coordinates of $A,B$ are $\{0,1\}$ or $\{1,2\}$; shift both by $+e_i$ or $-e_i$ to get $\{1,2\}$ or $\{0,1\}$. Again both stay in $G$ and the line moves.
- All coordinates of $d$ are $\pm2$. Then $A$ and $B$ are opposite corners of the big cube. Let $s=\pm e_1$ point from $A$’s first coordinate toward $1$, and take $A'=A+s$, $B'=A+\frac d2+s$. Then $A'$ has first coordinate $1$ and its other coordinates in $\{0,2\}$; $B'$ has coordinates $1$, $1\pm1$, so both lie in $G$. The segment $A'B'$ is parallel to $d$, and its line differs from line $AB$ because $s$ is not parallel to $d$.
So the answer is yes.
How we checked it: script: every one of the 351 pairs in {0,1,2}^3 has a parallel, non-collinear partner pair
P3 Palindromic quartic with a real root: smallest coefficient size
- Hint 1. A root is not $0$. Divide by $x^2$ and set $t=x+\frac1x$. What values can $t$ take?
- Hint 2. The condition becomes $at+b=2-t^2$, a line in the $(a,b)$-plane. Use CauchyโSchwarz.
Our solution (not official)
Answer: $\dfrac45$, at $(a,b)=\left(\mp\dfrac45,-\dfrac25\right)$
Key idea: the substitution $t=x+\frac1x$ turns the quartic into a line in the $(a,b)$-plane, and $a^2+b^2$ is the squared distance to the origin.
A real root $x$ is not $0$. Dividing by $x^2$ and putting $t=x+\frac1x$ gives $x^2+\frac1{x^2}=t^2-2$, so $$t^2+at+b-2=0,\qquad |t|\ge2,$$ since $|x+\frac1x|\ge2$ for real $x\ne0$. Conversely, any $t$ with $|t|\ge2$ comes from a real $x$.
So $(a,b)$ is admissible exactly when $at+b=2-t^2$ for some $|t|\ge2$. By CauchyโSchwarz, $(at+b)^2\le(a^2+b^2)(t^2+1)$, so $$a^2+b^2\ge\frac{(t^2-2)^2}{t^2+1}.$$ With $s=t^2\ge4$, the function $h(s)=\frac{(s-2)^2}{s+1}$ has $h'(s)=\frac{(s-2)(s+4)}{(s+1)^2}>0$, so $h(s)\ge h(4)=\frac45$.
Equality needs $t^2=4$ and $(a,b)$ parallel to $(t,1)$. For $t=2$: $(a,b)=\lambda(2,1)$ with $5\lambda=-2$, so $(a,b)=\left(-\frac45,-\frac25\right)$, and $x=1$ is a root. For $t=-2$ we get $\left(\frac45,-\frac25\right)$ with root $x=-1$. The minimum is $\frac45$.
How we checked it: script: 200,000 random values of t give a minimum of 0.80008 for (tยฒ-2)ยฒ/(tยฒ+1)
P4 Shortest sweep of a triangle with a short-range detector
- Hint 1. Any good path must come within detector range $r$ (half the altitude) of both far vertices. Which one does it reach first?
- Hint 2. Bound $AP+PB$ for points $P$ at height at least $r$ above line $AB$ (Minkowski’s inequality). For coverage, use that the distance to a segment is a convex function.
Our solution (not official)
Answer: Go to the midpoint of the altitude from one far vertex, then head straight for the other far vertex and stop at detector range from it. Length $\frac{2\sqrt7-\sqrt3}{4}$ times the side.
Key idea: a path must come within range of both far corners. A sharp lower bound comes from the sum of distances from a point to two fixed points, and the path that meets it covers the triangle because distance to a segment is convex.
Let the triangle be $ABC$ with side $s$, altitude $h=\frac{\sqrt3}2s$, start at $A$, and detector range $r=\frac h2$. Put $M$ at the midpoint of $AB$, so $CM$ is an altitude.
Lower bound. The path must pass a point $P$ with $PC\le r$ and a point $Q$ with $QB\le r$. The triangle is symmetric in its altitude from $A$, which swaps $B$ and $C$, so we may assume the path reaches $P$ no later than $Q$. Then $$\text{length}\ge AP+PQ\ge AP+PB-QB\ge AP+PB-r .$$ Now $A$ and $B$ are mirror images in line $CM$. Put coordinates with $M=(0,0)$, $A=(-\frac s2,0)$, $B=(\frac s2,0)$, $C=(0,h)$. Since $PC\le r$, the point $P=(x,y)$ has $y\ge h-r=\frac h2$. By Minkowski’s inequality $$AP+PB=\sqrt{(x+\tfrac s2)^2+y^2}+\sqrt{(\tfrac s2-x)^2+y^2}\ge\sqrt{s^2+4y^2}\ge\sqrt{s^2+h^2}.$$ With $h^2=\frac34s^2$ this is $\frac{\sqrt7}2s$, so $$\text{length}\ge\frac{\sqrt7}{2}s-\frac{\sqrt3}{4}s=\frac{2\sqrt7-\sqrt3}{4}\,s\approx0.890\,s .$$
A path with this length. Let $P_0$ be the midpoint of $CM$ (the equality point above) and $Q_0$ the point of segment $P_0B$ at distance $r$ from $B$. The path $A\to P_0\to Q_0$ has length $AP_0+P_0B-r$, which is the bound.
It sweeps the whole triangle. The distance from a point to a fixed segment is a convex function, so on a triangle it is largest at a vertex.
- On triangle $AMC$, the distance to segment $AP_0$ is $0$ at $A$, at most $MP_0=r$ at $M$, and at most $CP_0=r$ at $C$.
- On triangle $BMC$, the distance to segment $P_0Q_0$ is at most $CP_0=r$ at $C$, at most $MP_0=r$ at $M$, and exactly $BQ_0=r$ at $B$.
These two triangles make up $ABC$, so every point is within range. The mirror-image path (via the midpoint of the altitude from $B$, ending near $C$) is equally short.
How we checked it: script: grid check that the path covers the triangle (largest gap 0), and random sampling of the lower bound
P5 A group of linear maps, each with a fixed point
- Hint 1. Which maps $x\mapsto x+b$ can belong to the set?
- Hint 2. The commutator $g\circ f\circ g^{-1}\circ f^{-1}$ has slope $1$. What does that say about $f$ and $g$?
Our solution (not official)
Key idea: the only translation in the set is the identity, so all commutators are trivial. Commuting maps share the fixed point of any non-identity element.
Translations. A map $x\mapsto x+b$ with $b\ne0$ has no fixed point. So the only element of slope $1$ in $G$ is the identity.
Everything commutes. For $f,g\in G$, the map $h=g\circ f\circ g^{-1}\circ f^{-1}$ is in $G$ by (a) and (b), and its slope is the product of the slopes $a_g\,a_f\,a_g^{-1}a_f^{-1}=1$. So $h$ is the identity, which means $g\circ f=f\circ g$.
A common fixed point. If every element of $G$ is the identity, any $k$ works. Otherwise pick $f\in G$ that is not the identity; its slope $a\ne1$, so it has exactly one fixed point $k=\frac{b}{1-a}$. For any $g\in G$, $$f(g(k))=g(f(k))=g(k),$$ so $g(k)$ is a fixed point of $f$, and therefore $g(k)=k$.
How we checked it: algebra only: slope bookkeeping
P6 Dominating sequence with controlled ratios and total
- Hint 1. Spread each $a_j$ over all positions with geometric weights $q^{|k-j|}$.
- Hint 2. Compare the weights at positions $k$ and $k+1$: each ratio is $q$ or $\frac1q$.
Our solution (not official)
Key idea: let each $a_j$ leak into its neighbors with weight $q$ per step. Ratios of neighbors then stay between $q$ and $\frac1q$, and the total is at most a two-sided geometric series.
If $n=1$, take $b_1=\frac{a_1}{1-q}$: then $a_1<b_1<\frac{1+q}{1-q}a_1$, and (b) is empty. For $n\ge2$ set $$b_k=\sum_{j=1}^n a_j\,q^{|k-j|}\qquad(k=1,\dots,n).$$
(a) $b_k$ contains $a_k$ (weight $q^0=1$) plus at least one more positive term, so $b_k>a_k$.
(b) For each $j$, the weight ratio $q^{|k+1-j|}/q^{|k-j|}$ is $q$ if $j\le k$ and $\frac1q$ if $j\ge k+1$. Both lie in $[q,\frac1q]$, so $q\,b_k\le b_{k+1}\le\frac1q\,b_k$. The inequalities are strict: the term $j=k+1$ has ratio $\frac1q>q$, and the term $j=k$ has ratio $q<\frac1q$.
(c) Adding over $k$, $$\sum_k b_k=\sum_j a_j\sum_{k=1}^n q^{|k-j|}<\sum_j a_j\left(1+2\sum_{m=1}^\infty q^m\right)=\frac{1+q}{1-q}\sum_j a_j .$$
How we checked it: script: 2,000 random cases (n โค 8) satisfy all three conditions