Pigeonhole and Extremal Arguments

Study Sheet

Pigeonhole and Extremal Arguments

Proving that something must exist without finding it

The Pigeonhole Principle

Concept
More pigeons than holes

If n+1n + 1 objects are placed in nn boxes, some box holds at least two. The whole art is choosing the boxes.

Pigeonhole proofs are existence proofs: they show two objects must collide without saying which. The boxes are usually remainders mod nn, pairs that sum to something, or pieces of a figure. A good sign you need pigeonhole: the problem says "among any n+1n + 1 …" or "some two of them". Write down three things explicitly — what the objects are, what the boxes are, and why there are more objects than boxes — and the proof is done.

Example
Socks in the dark

A drawer holds red, blue and black socks. How many must you take, without looking, to be sure of a matching pair? Objects: socks drawn; boxes: the 33 colors. Four socks in three boxes force two of one color. Three are not enough, since they could be one of each. So the answer is 44 — and "three are not enough" is part of the answer, because "sure" means the smallest number that always works.

Example
Two consecutive numbers
12345678910five boxes {1, 2}, {3, 4}, …, {9, 10}; six circled numberstwo in one box
The chosen numbers 1,3,6,8,9,101, 3, 6, 8, 9, 10: the box {9,10}\{9, 10\} is forced to hold two.

Among any n+1n + 1 integers chosen from {1,…,2n}\{1, \ldots, 2n\}, two are consecutive. Boxes: {1,2},{3,4},…,{2n−1,2n}\{1, 2\}, \{3, 4\}, \ldots, \{2n - 1, 2n\} — nn boxes for n+1n + 1 numbers, so two chosen numbers share a box, and numbers in the same box are consecutive. ■\blacksquare

Example
A subset sum divisible by nn

Among any nn integers a1,…,ana_1, \ldots, a_n, some non-empty block ai+1+⋯+aja_{i+1} + \cdots + a_j has sum divisible by nn. Consider the partial sums s1=a1s_1 = a_1, s2=a1+a2s_2 = a_1 + a_2, …, sns_n. If one leaves remainder 00 mod nn, done. Otherwise nn sums fall into the n−1n - 1 nonzero remainders, so two agree: si≡sjs_i \equiv s_j with i<ji < j, and sj−si=ai+1+⋯+ajs_j - s_i = a_{i+1} + \cdots + a_j is divisible by nn. ■\blacksquare

Side note
Where it goes wrong

Boxes that are not disjoint or do not cover all the objects. Say explicitly what the objects are, what the boxes are, and why there are more objects than boxes. "Some two of the numbers 1,…,2n1, \ldots, 2n are consecutive" with boxes {1,2},{2,3},…\{1, 2\}, \{2, 3\}, \ldots fails: the boxes overlap, so one number can sit in two boxes at once.

Try it
Try it: three with a sum divisible by 3

Prove that among any 55 integers, some 33 have a sum divisible by 33.

Show the answer

Sort the integers by remainder mod 33. If all three remainders 0,1,20, 1, 2 occur, take one of each: the sum leaves remainder 0+1+2=3≡00 + 1 + 2 = 3 \equiv 0. Otherwise at most two remainders occur, and 55 integers in 22 boxes put at least 33 in one box: three integers with the same remainder rr have sum leaving 3r≡03r \equiv 0. ■\blacksquare

Choosing the Boxes

Concept
The general form

If NN objects are placed in nn boxes, some box holds at least ⌈Nn⌉\left\lceil \dfrac Nn \right\rceil of them. Choosing the boxes is the proof: remainders for divisibility claims, pieces of a region for distance claims, possible values of a count for "two are the same" claims.

The general form is an average argument: NN objects in nn boxes average N/nN/n per box, and some box is at least average. Since boxes hold whole numbers of objects, "at least N/nN/n" means at least the next integer up. The design rule for boxes: two objects in the same box must automatically have the property you want. That forces the size of the boxes — for distance claims, every box must have diameter at most the distance in the conclusion.

Example
Birth weekdays

Among 3030 people, at least ⌈30/7⌉=5\lceil 30/7\rceil = 5 were born on the same day of the week. If each of the 77 days had at most 44 people, there would be at most 2828 people.

Example
Remainders as boxes

Among any 66 integers, two differ by a multiple of 55. Boxes: the 55 remainders on division by 55. Six integers, five boxes: two share a remainder, and then their difference leaves remainder 00, that is, it is a multiple of 55. ■\blacksquare

Example
Ten points in a triangle
3side 1 pieces:diameter 1
Nine pieces of side 11; ten points force two into one shaded piece.

Ten points lie in an equilateral triangle of side 33. Prove two of them are at distance at most 11. Boxes: lines parallel to the sides cut the triangle into 99 equilateral triangles of side 11. Ten points, nine pieces: two points lie in the same piece. The greatest distance between two points of a side-11 equilateral triangle is 11 (the length of a side), so those two points are at most 11 apart. ■\blacksquare

Side note
Where it goes wrong

Boxes of the wrong size. Cutting the triangle into 44 triangles of side 1.51.5 gives a collision easily, but only proves the two points are within 1.51.5. The box has to be small enough that "same box" already means "close enough".

Try it
Try it: five points in a square

Five points lie in a square of side 22. Prove that two of them are at distance at most 2\sqrt2.

Show the answer

Cut the square into four 1×11 \times 1 squares. Five points, four boxes: two points lie in the same small square. The greatest distance between two points of a unit square is its diagonal, 12+12=2\sqrt{1^2 + 1^2} = \sqrt2. ■\blacksquare

Extremal Arguments

Concept
Look at the biggest or smallest

Choose the object that is extreme — the largest number, the shortest distance, the player with the most wins — and argue that its extremity forces a property, or that a "more extreme" object would exist otherwise (contradiction).

The extreme object carries information the others do not: nothing is larger, so every comparison with it goes one way. That one-sided information is usually exactly what the proof needs. The precondition is that the extreme object exists — guaranteed for a finite non-empty set, and for any non-empty set of positive integers by well-ordering, but NOT for infinite sets of reals.

Example
Each number the average of its neighbors

Numbers are placed at the vertices of a polygon so that each equals the average of its two neighbors. Prove they are all equal. Take the largest value MM. Its two neighbors average to MM and are each at most MM, so both equal MM. Repeat around the polygon: all equal MM. ■\blacksquare

Example
An empty triangle
ABCevery other pointis farther from AB
Any point inside triangle ABCABC would be closer to line ABAB than CC is.

A finite set of at least three points in the plane, no three on a line. Prove some three of them form a triangle containing no other point of the set. Take any two points A,BA, B, and among the remaining points choose CC at the smallest distance from the line ABAB. Suppose some other point XX of the set lay in triangle ABCABC. Moving from CC toward the side ABAB only brings a point closer to line ABAB, so XX would be strictly closer to ABAB than CC — contradicting the choice of CC. ■\blacksquare

Side note
Where it goes wrong

Choosing an extreme object that does not exist. "Let xx be the smallest number in the interval (0,1)(0, 1)" — there is none. Before "take the largest", check that the set is finite and non-empty, or consists of positive integers.

Try it
Try it: a tournament king

In a tournament every pair of players plays once and there are no draws. Prove there is a player KK such that every other player either lost to KK or lost to someone who lost to KK.

Show the answer

Let KK be a player with the most wins. Suppose some QQ is not reached: QQ beat KK, and QQ also beat everyone KK beat (otherwise one of them beat QQ and reaches QQ in two steps). Then QQ has all of KK's wins plus the win over KK — more wins than KK, contradicting the choice of KK. ■\blacksquare

Well-Ordering and Infinite Descent

Concept
Every non-empty set of positive integers has a least element

This is the well-ordering principle. It gives two tools: the smallest counterexample (if a claim fails, take its smallest failure and show a smaller one exists) and infinite descent (a process producing ever-smaller positive integers must stop).

Well-ordering is what makes positive integers different from positive rationals or reals: you cannot keep going down forever. If an argument shows that every solution of some equation produces a strictly smaller positive solution, the equation can have no positive solutions at all — there would have to be a smallest one, and it would produce a smaller one.

Example
Every n≥2n \ge 2 has a prime divisor

Among the divisors of nn that are greater than 11 (there is at least one, nn itself), let dd be the smallest. If dd were not prime, it would have a divisor aa with 1<a<d1 < a < d, and aa would also divide nn — a smaller divisor greater than 11. So dd is prime. ■\blacksquare

Example
No infinite descent

Find all integer solutions of x2=2y2x^2 = 2y^2. If there is a solution with x≠0x \ne 0, take one with the smallest ∣x∣>0|x| > 0. Then x2x^2 is even, so x=2x1x = 2x_1, giving 4x12=2y24x_1^2 = 2y^2, so y2=2x12y^2 = 2x_1^2 and y=2y1y = 2y_1; then x12=2y12x_1^2 = 2y_1^2, a solution with 0<∣x1∣<∣x∣0 < |x_1| < |x|. Contradiction. The only solution is (0,0)(0, 0). ■\blacksquare

Example
A descent through three variables

Prove the only integer solution of x2+y2=3z2x^2 + y^2 = 3z^2 is x=y=z=0x = y = z = 0. Suppose not. If z=0z = 0 then x2+y2=0x^2 + y^2 = 0 forces x=y=0x = y = 0, so a non-zero solution has z≠0z \ne 0; choose one with ∣z∣|z| as small as possible. Since 3∣x2+y23 \mid x^2 + y^2, both xx and yy are multiples of 33 (Direct Proof, squares mod 33). Then 9∣x2+y2=3z29 \mid x^2 + y^2 = 3z^2, so 3∣z23 \mid z^2 and 3∣z3 \mid z. Now (x/3,y/3,z/3)(x/3, y/3, z/3) is a solution with a smaller non-zero ∣z∣|z|. Contradiction. ■\blacksquare

Side note
Where it goes wrong

Descending in a set that has no floor. "Take the smallest positive rational solution" fails because positive rationals have no smallest element. Descent needs positive integers (or something that maps to them, like the size of a denominator).

Try it
Try it: binary

Use a smallest counterexample to prove that every positive integer is a sum of distinct powers of 22.

Show the answer

Suppose not, and let mm be the smallest positive integer that is not such a sum. m≠1=20m \ne 1 = 2^0. Let 2a2^a be the largest power of 22 with 2a≤m2^a \le m; then m<2a+1m < 2^{a+1}, so r=m−2a<2ar = m - 2^a < 2^a. If r=0r = 0, m=2am = 2^a is such a sum. Otherwise 0<r<m0 < r < m, so rr is a sum of distinct powers of 22, each at most r<2ar < 2^a. Adding 2a2^a, which is not among them, writes mm as a sum of distinct powers of 22 — contradiction. ■\blacksquare

Why It Works

Suppose NN objects are placed in nn boxes and every box holds at most ⌈N/n⌉−1\left\lceil N/n \right\rceil - 1. Since ⌈N/n⌉<N/n+1\lceil N/n\rceil < N/n + 1, each box holds fewer than N/nN/n objects, so the boxes together hold fewer than n⋅N/n=Nn\cdot N/n = N. That contradicts there being NN objects. ■\blacksquare The principle is nothing more than "the largest is at least the average", made exact for whole numbers.

A finite non-empty set of numbers has a largest element: by induction on its size, a set of k+1k + 1 numbers is a set of kk numbers (which has a largest, MM) plus one more number xx, and the larger of MM and xx is the largest of all. A non-empty set of positive integers has a smallest element by well-ordering, which is equivalent to induction. Every extremal argument quietly uses one of these two facts, and every extremal argument that fails has used neither.

Example
A failed attempt and the right boxes

Claim: among any 77 integers, two have a sum or a difference divisible by 1010. Failed attempt: "use the 1010 remainders mod 1010 as boxes" — 77 objects in 1010 boxes guarantees nothing. Right boxes: {0},{5},{1,9},{2,8},{3,7},{4,6}\{0\}, \{5\}, \{1, 9\}, \{2, 8\}, \{3, 7\}, \{4, 6\} — six boxes for the remainders. Seven integers in six boxes: two land in the same box. If they have the same remainder, their difference is divisible by 1010. If their remainders are rr and 10−r10 - r, their sum is. ■\blacksquare The boxes were built so that "same box" already means "the property holds".