Putnam 1998
Read the problems at the official link. The 1998 Putnam problems are Β© the Mathematical Association of America, so we don't reproduce them here. Open the 1998 problems (PDF) from the Putnam Archive.
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.
A1 Solid geometry: a cube standing inside a cone
- Hint 1. The top corners of the cube touch the cone. How far are they from the axis?
- Hint 2. The cone’s radius shrinks linearly with height: at height $h$ it is $1-\frac h3$.
Our solution (not official)
Answer: $\dfrac{9\sqrt2-6}{7}$
Key idea: the top four corners of the cube lie on the cone, at height $s$ and distance $s/\sqrt2$ from the axis.
Let the side be $s$. By symmetry the cube is centered on the axis, so its top corners are at height $s$ and at distance half the diagonal, $\frac{s}{\sqrt2}$, from the axis. At height $h$ the cone has radius $1-\frac h3$. The corners touch the cone, so $$\frac s{\sqrt2}=1-\frac s3\quad\Longrightarrow\quad s=\frac{1}{\frac1{\sqrt2}+\frac13}=\frac{3\sqrt2}{3+\sqrt2}=\frac{9\sqrt2-6}{7}\approx0.961 .$$
How we checked it: script: s/β2 and 1 β s/3 both equal 0.679623 for this s
A2 Areas: two regions under a circular arc
- Hint 1. Describe the arc by angles $\alpha<\beta$ and write each area as an integral of $\sqrt{1-t^2}$.
- Hint 2. An antiderivative of $\sqrt{1-t^2}$ is $\frac12\big(t\sqrt{1-t^2}+\arcsin t\big)$.
Our solution (not official)
Key idea: compute both areas exactly; the triangle terms cancel and the sector terms add to the arc length.
Let the arc run from angle $\alpha$ to angle $\beta$, with $0\le\alpha<\beta\le\frac\pi2$; its length is $\beta-\alpha$. Use $G(t)=\frac12\big(t\sqrt{1-t^2}+\arcsin t\big)$, an antiderivative of $\sqrt{1-t^2}$.
The region below the arc spans $\cos\beta\le x\le\cos\alpha$: $$A=G(\cos\alpha)-G(\cos\beta)=\tfrac12\big(\cos\alpha\sin\alpha-\cos\beta\sin\beta\big)+\tfrac12\big(\beta-\alpha\big),$$ using $\arcsin(\cos\theta)=\frac\pi2-\theta$ on $[0,\frac\pi2]$. The region left of the arc spans $\sin\alpha\le y\le\sin\beta$: $$B=G(\sin\beta)-G(\sin\alpha)=\tfrac12\big(\sin\beta\cos\beta-\sin\alpha\cos\alpha\big)+\tfrac12\big(\beta-\alpha\big).$$ Adding, the first parts cancel and $A+B=\beta-\alpha$, the length of the arc.
(Geometrically: each region is a sector of angle $\beta-\alpha$, area $\frac{\beta-\alpha}2$, plus one right triangle minus another; the triangles in $A$ and $B$ are the same two with opposite signs.)
How we checked it: script: numerical A + B equals Ξ² β Ξ± for random arcs
A3 Derivatives: a product of four derivatives somewhere nonnegative
- Hint 1. If the product is always negative, none of the four functions ever vanishes, so each has a fixed sign.
- Hint 2. Show: if $h$, $h'$ and $h''$ never vanish, then $h$ and $h''$ have the same sign.
Our solution (not official)
Key idea: a function that never vanishes, with first and second derivatives that never vanish, must have the same sign as its second derivative. Apply this to $f$ and to $f'$.
Lemma. If $h\in C^2(\mathbb R)$ and $h,h',h''$ never vanish, then $hh''>0$.
Proof. Each of $h,h',h''$ has constant sign. Replacing $h$ by $-h$ if needed, assume $h''>0$, so $h$ is convex and lies above each tangent line: $h(x)\ge h(x_0)+h'(x_0)(x-x_0)$. Since $h'(x_0)\ne0$, the right side tends to $+\infty$ as $x\to+\infty$ or $x\to-\infty$. So $h$ takes positive values, and since it never vanishes, $h>0$. Thus $hh''>0$.
The problem. Suppose $f f'f''f'''<0$ everywhere. Then none of $f,f',f'',f'''$ vanishes. The lemma for $h=f$ gives $ff''>0$, and for $h=f'$ (using $f'''$ continuous) gives $f'f'''>0$. Then $$ff'f''f'''=(ff'')(f'f''')>0,$$ a contradiction. So the product is $\ge0$ at some point.
How we checked it: reasoning only (a proof)
A4 Divisibility: concatenated numbers modulo 11
- Hint 1. Mod $11$, $10\equiv-1$, so appending a block of $L$ digits multiplies by $(-1)^L$.
- Hint 2. The lengths are Fibonacci numbers; their parity repeats with period $3$.
Our solution (not official)
Answer: $n\equiv1\pmod 6$
Key idea: modulo $11$, concatenation is $A_n\equiv(-1)^{L_{n-2}}A_{n-1}+A_{n-2}$, where $L$ is the digit length; the residues repeat with period $6$.
Let $L_n$ be the number of digits of $A_n$. Then $L_1=L_2=1$ and $L_n=L_{n-1}+L_{n-2}$, so $L_n$ is even exactly when $3\mid n$. Since $A_n=A_{n-1}\cdot10^{L_{n-2}}+A_{n-2}$ and $10\equiv-1\pmod{11}$, $$A_n\equiv(-1)^{L_{n-2}}A_{n-1}+A_{n-2}\pmod{11}.$$ Compute residues:
- $A_1\equiv0$, $A_2\equiv1$;
- $A_3\equiv-1+0=-1$, $A_4\equiv1+1=2$, $A_5\equiv2-1=1$ ($L_3$ even), $A_6\equiv-1+2=1$, $A_7\equiv-1+1=0$, $A_8\equiv0+1=1$ ($L_6$ even).
So $(A_7,A_8)\equiv(A_1,A_2)$. The rule only depends on $n\bmod3$ through the parity of $L_{n-2}$, so from here everything repeats with period $6$. In one period the residues are $0,1,-1,2,1,1$, and only the first is $0$. Hence $11\mid A_n$ exactly when $n\equiv1\pmod6$.
How we checked it: script: computed A_n exactly for n β€ 25; divisible by 11 for n = 1, 7, 13, 19, 25
A5 Covering: Vitali’s lemma for finitely many discs
- Hint 1. Choose greedily: take the largest disc, throw away every disc that meets it, and repeat.
- Hint 2. If a disc $D$ meets a chosen disc $D_j$ whose radius is at least that of $D$, then $D\subseteq3D_j$.
Our solution (not official)
Key idea: pick discs greedily from largest to smallest; every discarded disc touches a chosen one at least as large, so it fits inside three times that disc.
Greedy choice. Let $D_1$ be a disc in $\mathcal F$ of largest radius. Remove from $\mathcal F$ every disc that meets $D_1$ (including $D_1$). From what is left, let $D_2$ be one of largest radius, remove all discs meeting $D_2$, and so on until nothing is left. The chosen discs are pairwise disjoint, because each was chosen from discs that do not meet the earlier ones.
Covering. Every disc $D\in\mathcal F$ was removed at some step $j$, because it meets $D_j$ (possibly $D=D_j$). At that step $D$ was still available, so its radius $r$ is at most the radius $r_j$ of $D_j$. If $P,P_j$ are the centers and $z\in D\cap D_j$, then $|PP_j|<r+r_j$, and for any $x\in D$, $$|xP_j|\le|xP|+|PP_j|<r+(r+r_j)\le3r_j .$$ So $D\subseteq3D_j$. Hence $E\subseteq\bigcup\mathcal F\subseteq\bigcup_j3D_j$.
How we checked it: reasoning only (a proof)
A6 Lattice geometry: a tight inequality forces a square corner
- Hint 1. Twice the area of a lattice triangle is an integer, and so are the squared side lengths.
- Hint 2. Use $2[ABC]\le|AB|\,|BC|$ and $|AB|^2+|BC|^2\ge2|AB|\,|BC|$.
Our solution (not official)
Key idea: split $(a+c)^2=a^2+c^2+2ac$; integrality squeezes $a^2+c^2$ down to $2K$, which forces $a=c$ and a right angle.
Let $a=|AB|$, $c=|BC|$ and $K=2[ABC]$. For lattice points, $a^2$, $c^2$ and $K$ are integers (the shoelace formula). Also $K=ac\sin B\le ac$.
The hypothesis is $a^2+c^2+2ac<4K+1$. Since $2ac\ge2K$, $$a^2+c^2<2K+1,\qquad\text{so}\qquad a^2+c^2\le2K$$ because both sides are integers. On the other hand $a^2+c^2\ge2ac\ge2K$. So equality holds throughout: $$a^2+c^2=2ac\ \Rightarrow\ a=c,\qquad ac=K=ac\sin B\ \Rightarrow\ \sin B=1 .$$ (If $K=0$ the points are collinear; then $(a+c)^2<1$, impossible for distinct lattice points.)
So $AB=BC$ and the angle at $B$ is a right angle. Then $A$, $B$, $C$ are three vertices of the square whose fourth vertex is $A+C-B$.
How we checked it: script: brute force over all lattice triples in a 7Γ7 grid; every triple satisfying the inequality is a right isosceles corner
B1 Algebra: minimizing a ratio built from x + 1/x
- Hint 1. The numerator is a difference of squares: $x^6+x^{-6}+2=(x^3+x^{-3})^2$.
Our solution (not official)
Answer: $6$
Key idea: the numerator factors as a difference of squares, and one factor cancels the denominator.
Let $u=(x+\frac1x)^3$ and $w=x^3+\frac1{x^3}$. Since $w^2=x^6+\frac1{x^6}+2$, the numerator is $u^2-w^2=(u-w)(u+w)$, and the denominator is $u+w>0$. So the expression equals $$u-w=\Big(x+\frac1x\Big)^3-\Big(x^3+\frac1{x^3}\Big)=3\Big(x+\frac1x\Big).$$ For $x>0$, $x+\frac1x\ge2$ with equality at $x=1$. The minimum is $6$.
How we checked it: script: evaluated on a grid of x in (0.5, 2.5); minimum 6 at x = 1
B2 Optimization: shortest triangle touching two lines
- Hint 1. Reflect the given point across each of the two lines.
- Hint 2. The perimeter equals a path between the two reflections.
Our solution (not official)
Answer: $\sqrt{2(a^2+b^2)}$
Key idea: reflecting the fixed vertex across both lines turns the perimeter into a broken path between two fixed points.
Let $P=(a,b)$, and let $X$ be on the $x$-axis and $Y$ on the line $y=x$. Reflect $P$ across the $x$-axis to $P_1=(a,-b)$ and across $y=x$ to $P_2=(b,a)$. Then $|PX|=|P_1X|$ and $|PY|=|P_2Y|$, so $$|PX|+|XY|+|YP|=|P_1X|+|XY|+|YP_2|\ge|P_1P_2|=\sqrt{(a-b)^2+(a+b)^2}=\sqrt{2a^2+2b^2}.$$ Equality is reached. $P_1$ lies below both lines and $P_2$ lies above both (since $a>b>0$), so the segment $P_1P_2$ crosses the $x$-axis and the line $y=x$. Take $X$ and $Y$ to be those crossings. They are different points, because the segment misses the origin ($P_1$, $P_2$ and $O$ are not collinear: $a\cdot a-(-b)\cdot b=a^2+b^2\ne0$). For this triangle the perimeter is exactly $|P_1P_2|$.
So the minimum perimeter is $\sqrt{2(a^2+b^2)}$.
How we checked it: script: random search over X and Y for a = 3, b = 1.2 approaches 4.5695 = β(2(aΒ²+bΒ²))
B3 Surface area: a hemisphere above an inscribed pentagon
- Hint 1. Subtract the area above the five circular segments outside the pentagon.
- Hint 2. Archimedes: on a sphere of radius $1$, the zone between two parallel planes at distance $h$ has area $2\pi h$.
Our solution (not official)
Answer: $5\pi\cos\frac\pi5-3\pi$
Key idea: remove the five “caps” outside the pentagon; by Archimedes’ hat-box theorem each cap’s area depends only on its width.
The hemisphere has area $2\pi$. The part over the disk but outside the pentagon lies over five congruent circular segments. Each is cut off by a side at distance $\cos\frac\pi5$ from the center.
Take one side and turn the picture so the side is the line $x=\cos\frac\pi5$. The part of the whole unit sphere with $x\ge\cos\frac\pi5$ is a cap between parallel planes at distance $1-\cos\frac\pi5$, so its area is $2\pi(1-\cos\frac\pi5)$. The plane $z=0$ cuts it into two mirror-image halves, so the part over our segment (with $z\ge0$) has area $\pi(1-\cos\frac\pi5)$.
Therefore the area over the pentagon is $$2\pi-5\pi\Big(1-\cos\frac\pi5\Big)=5\pi\cos\frac\pi5-3\pi\approx3.283 .$$ In the requested form this is $A\sin\alpha+B\cos\beta$ with, for example, $A=3\pi$, $\alpha=-\frac\pi2$, $B=5\pi$, $\beta=\frac\pi5$.
How we checked it: script: Monte Carlo over 4 million points on the hemisphere gives 3.2831, matching 3.2832
B4 Floor sums: when alternating signs cancel exactly
- Hint 1. Pair index $i$ with $mn-1-i$. How does the exponent change?
- Hint 2. If $m$ and $n$ are both even, group the indices in pairs $2j,2j+1$ to relate the sum to the one for $m/2$, $n/2$.
Our solution (not official)
Answer: Exactly when $m$ and $n$ have different powers of $2$ in them
Key idea: a reflection kills the sum when $m+n$ is odd; halving both relates the even case to a smaller one; when both are odd the sum has an odd number of $\pm1$ terms.
Let $S(m,n)=\sum_{i=0}^{mn-1}(-1)^{\lfloor i/m\rfloor+\lfloor i/n\rfloor}$, and let $v(k)$ be the exponent of $2$ in $k$. We show $S(m,n)=0$ exactly when $v(m)\ne v(n)$.
Reflection. For $0\le i<mn$, $\lfloor\frac{mn-1-i}{m}\rfloor=n-1-\lfloor\frac im\rfloor$, and similarly for $n$. So the term at $mn-1-i$ is $(-1)^{m+n}$ times the term at $i$. If $m+n$ is odd, $mn$ is even, the pairing $i\leftrightarrow mn-1-i$ has no fixed point, and the terms cancel: $S=0$.
Both odd. Then $S$ is a sum of $mn$ (an odd number of) terms $\pm1$, so $S$ is odd and $S\ne0$.
Both even, $m=2m'$, $n=2n'$. For $i=2j$ or $2j+1$, $\lfloor\frac i{2m'}\rfloor=\lfloor\frac j{m'}\rfloor$. So $$S(m,n)=2\sum_{j=0}^{2m'n'-1}(-1)^{\lfloor j/m'\rfloor+\lfloor j/n'\rfloor}.$$ Shifting $j$ by $m'n'$ adds $n'$ and $m'$ to the two floors, so the second half of this sum is $(-1)^{m'+n'}$ times the first half: $$S(m,n)=2\big(1+(-1)^{m'+n'}\big)\,S(m',n').$$
Conclusion by induction on $\min(v(m),v(n))$. If $v(m)=v(n)=0$, both are odd and $S\ne0$. If exactly one is odd, $S=0$. If both are even: when $m'+n'$ is odd we get $S=0$, and indeed $v(m')\ne v(n')$; when it is even, $S=4S(m',n')$ and $v(m')-v(n')=v(m)-v(n)$, so the induction applies.
How we checked it: script: computed the sum for all 1 β€ m, n < 60; it is zero exactly when the powers of 2 differ
B5 Decimals: a digit of the square root of a repunit
- Hint 1. $N=\frac{10^{1998}-1}{9}$, so $\sqrt N=\frac{10^{999}}3\sqrt{1-10^{-1998}}$.
- Hint 2. Use $1-\frac y2-\frac{y^2}2\le\sqrt{1-y}\le1-\frac y2$ for small $y>0$.
Our solution (not official)
Answer: $1$
Key idea: $\sqrt N$ is just below $\frac{10^{999}}3=333\ldots3.333\ldots$, by about $\frac16\cdot10^{-999}$, which changes the 1000th decimal digit from $3$ to $1$.
$N=\frac{10^{1998}-1}9$, so with $y=10^{-1998}$, $$\sqrt N=\frac{10^{999}}{3}\sqrt{1-y}.$$ For $0<y<1$, $1-\frac y2-\frac{y^2}{2}\le\sqrt{1-y}\le1-\frac y2$ (square the bounds). So $$\sqrt N=\frac{10^{999}}3-\varepsilon,\qquad \frac{10^{-999}}6\le\varepsilon\le\frac{10^{-999}}6+10^{-2996}.$$ The fractional part of $\frac{10^{999}}3$ is $0.333\ldots$ (all threes). Subtracting $\varepsilon\approx0.1666\ldots\times10^{-999}$: $$0.\underbrace{33\ldots3}_{999}\,3333\ldots-0.\underbrace{00\ldots0}_{999}\,1666\ldots=0.\underbrace{33\ldots3}_{999}\,1666\ldots$$ The tiny extra error ($10^{-2996}$) cannot reach the 1000th place. So the 1000th digit after the decimal point is $1$.
How we checked it: script: Python's decimal module with 2100 digits gives ...3333331666... with digit 1000 equal to 1
B6 Number theory: a cubic cannot always be a square
- Hint 1. Squares are $0$ or $1$ mod $4$, so two squares never differ by $2$ mod $4$.
- Hint 2. Compute $P(1)-P(3)$ and $P(2)-P(4)$ modulo $4$.
Our solution (not official)
Key idea: modulo $4$, one of two specific differences of values is $2$, but differences of squares never are.
Let $P(n)=n^3+an^2+bn+c$, and suppose $\sqrt{P(n)}$ is an integer for every positive $n$. Then each $P(n)$ is a perfect square, so $P(n)\equiv0$ or $1\pmod4$, and any difference $P(n)-P(m)$ is $\equiv 0$, $1$ or $-1\pmod 4$, never $2$.
Now compute: $$P(1)-P(3)=-26-8a-2b\equiv2-2b\pmod4,$$ $$P(2)-P(4)=-56-12a-2b\equiv-2b\pmod4 .$$
- If $b$ is even, the first difference is $\equiv2\pmod4$.
- If $b$ is odd, the second difference is $\equiv2\pmod4$.
Either way we have a contradiction. So for some $n\in\{1,2,3,4\}$, $\sqrt{P(n)}$ is not an integer.
How we checked it: script: for all 64 residue choices of a, b, c mod 4, some P(1..4) is 2 or 3 mod 4