In one sentence
Put objects into boxes, and some box holds at least two. More generally, put objects into boxes, and some box holds at least .
The proof takes one line (by contradiction): if each box held at most , there would be at most objects, not enough room. The real difficulty is that the problem never hands you the boxes. You have to partition the objects into classes so that “two in the same class” is exactly the property the problem wants. That is precisely the partition of a set from Vol. 1.
Example: any 11 numbers chosen from 1, 2, …, 20 include one that is a multiple of another
Step 1 / 4- First, try to avoid it
Take : twice any of them is more than , so none is a multiple of another. numbers can avoid it.
Predict first
What is the largest number of numbers you can choose from so that none of them is a multiple of another?
: the numbers will do. is impossible, because there are only possible odd parts. “At most ” and “any must succeed” are two sides of the same fact.
Stop 1 · Families of subsets (Vol. 1)
Boxes do not have to hold numbers; they can hold subsets.
Example: among any 2ⁿ⁻¹ + 1 distinct subsets of {1, 2, …, n}, two are disjoint
Step 1 / 4- The objects are subsets
An -element set has subsets, and the problem takes of them: more than half.
Stop 2 · Periodic sequences (Vol. 5)
Infinitely many terms but only finitely many “states”: the pigeonhole principle forces a state to repeat, and the recurrence turns the repeat into a period.
Example: the remainders of the Fibonacci sequence mod m form a periodic sequence
Step 1 / 4- States
is determined by the two terms before it, so take the remainders of two neighboring terms, , as one state.
Stop 3 · Existence problems (Vol. 5)
“Some run of consecutive terms must add up to a multiple of ”: the boxes hold not the numbers themselves but their prefix sums.
One problem, several solutions
Given any integers , prove that some run of consecutive terms (at least one term) has a sum that is a multiple of .
Prefix sums into remainder boxes
Let (). If some , we are done.
Otherwise the prefix sums fall into the nonzero remainders, so for some , and then is a multiple of .
Example: (). The prefix sums leave remainders ; , so .
CostOne use of the pigeonhole principle, clean and quick.
Trying to construct it directly
For specific numbers you can find it by trial, but for general numbers no formula tells you which run it is.
That is the value of the pigeonhole principle: it proves the run exists without having to find it.
CostYou cannot say which run to take, only try them one by one: exactly the “existence only” proof the pigeonhole principle is good at.
Practice
What is the smallest number of lattice points (points whose coordinates are both integers) you must take in the plane to guarantee that the midpoint of some two of them is also a lattice point?
What is the smallest number of numbers you must choose from to guarantee that two of them are coprime?
A family of subsets of has the property that any two of its subsets intersect. At most how many subsets can it contain?
Given any distinct integers, prove that two of them have a sum or a difference that is a multiple of .