Probability — page 3

A collision that finds a factor

A walk through the remainders modulo a number must eventually repeat, and it repeats modulo each hidden prime factor long before it repeats modulo the number. Pollard saw that the earlier repeat can be detected without knowing the prime — and that its timing is the birthday problem, so the cost is the square root of the factor.

Two children and the sentence about one of them

A family has two children and at least one is a boy. The chance that both are boys is one in three — or one in two, or anything from one in three to certainty — and every one of those answers is right for some way the sentence could have come to be said. There is no host and no door, and the protocol is still the whole problem.

One coin, counted by runs and by wakings

Beauty is put to sleep and a fair coin is tossed. Heads, she is woken once; tails, twice, with the first waking erased from her memory. Each time she wakes she is asked how likely heads is. One half, say some; one third, say others; and unlike every earlier puzzle of this kind, stating the protocol exactly does not end the argument.

The cells a permutation must miss

A derangement is a permutation that misses the diagonal of a square grid. Forbid any other set of cells instead and inclusion–exclusion still counts what is left — driven entirely by one list of numbers, the ways to place non-attacking rooks on the forbidden cells. Boards that look nothing alike can share that list, and rooks on a staircase turn out to count the ways to split a set.

A round table with no couple together

Seat n couples round a table, men and women alternating, so that nobody sits beside their partner. Once the women are placed the men face a board of forbidden cells that bends round a corner — and that corner is the whole difficulty. The forbidden cells form a cycle, a count of non-adjacent points on a cycle finishes the problem, and the chance of a good seating creeps towards e^(−2) far more slowly than the hat problem reaches 1/e.

Fourteen people within a day

Twenty-three people probably include two with the same birthday. Fourteen probably include two whose birthdays are at most a day apart, and seven, two within a week. The near miss has an exact formula, found by a trick that takes k days away after every birthday and turns the question back into the plain one on a shorter year, and the pattern behind every threshold is a single square root: a window of k days either side makes each pair 2k + 1 times as likely to collide.

A room where nobody is alone

Twenty-three people probably include two who share a birthday. How many are needed before every single person shares a birthday with somebody else in the room? The answer is 3,064 — more than it takes for every day of the year to be somebody's birthday — and the reason is a count of loners, which rises as the room fills, peaks at 134 when the room is the size of the year, and then falls so slowly that the last loner lingers for thousands of arrivals.

The ground a walk covers

A random walk of a thousand steps visits far fewer than a thousand places: in one dimension about fifty, in the plane about four hundred, in space about six hundred and sixty. The share of steps that land on new ground is exactly the chance of never coming home — so the number that decides whether a walker returns also decides how much of the world it sees.

A walk that may not step where it has been

Forbid a walk on the square grid from ever revisiting a site and the number of possible n-step walks grows like 2.638ⁿ instead of 4ⁿ — a number nobody can write down exactly. On the honeycomb it is exactly √(2 + √2), proved in 2010. And the walks spread out like n to the three-quarters, faster than any ordinary walk, which physicists have used since 1949 and mathematicians still cannot prove.

The median of many small averages

Knowing only that a quantity has a finite spread, the plain average of n samples can be promised to within σ/√(nδ) with confidence 1 − δ, and no better — Chebyshev's bound is tight, and rare large jumps achieve it. Cut the same samples into a dozen blocks, average each block, and take the median of the averages, and the promise improves to within about σ√(log(1/δ)/n). Nothing about the data has been assumed beyond the spread; only the way of combining it has changed.

A sphere that is nearly all equator

On an ordinary globe, the band within a tenth of the radius of the equator holds a tenth of the surface. On a sphere in a thousand dimensions the same band holds 99.85% of it, and the band of a fifth holds all but about two parts in ten billion. Almost every point of a high-dimensional sphere is near every equator at once — and so any function that cannot change quickly is, over almost all of the sphere, almost constant.

A coin that lets the first player win

On a fair coin the second player in Penney's game always has a better pattern than the first, and the first can hold them to no worse than two to one. Bend the coin and every overlap is paid for in the letters it uses: the replies change, the first player's share swings between a third and a half, and past a heads chance of 1/∛2 the first player simply names HHH and wins.

Three patterns in a circle

Race three coin patterns at once and the gamblers' accounting still gives each one's chance of arriving first — one fairness equation per pattern. What it does not give is any way to read the three-way result off the two-way ones. HHHT, TTHH and HTTH beat one another in a circle, and HHH loses both its head-to-head races and still finishes ahead of one of the patterns that beat it.

Evidence measured in decibans

Write a probability as odds and take the logarithm, and every piece of evidence becomes a length. A positive result on a good test is thirteen decibans; a negative one is minus twenty. Lay the lengths end to end from the prior and the posterior is where they stop, in any order. The rule fails in exactly one way — when two pieces of evidence share a cause — and Turing built a code-breaking method on the arithmetic.

The envelope that always looks better

Two envelopes, one holding twice as much as the other. Open one, see an amount, and reason that the other holds double or half with equal chance — so switching gains a quarter on average. By symmetry the same argument says switch back. The step that fails is not the arithmetic; it is the claim that double and half are equally likely whatever amount is seen, which no honest prior allows — and there is one prior under which the other envelope really does look better at every amount.

Points on a lattice that see almost nothing

Average a function over the points of a carefully tilted lattice and the error can fall like one over the square of the number of points — far faster than random sampling, and faster than the most evenly spread sequences. The reason is that a lattice rule is blind to only a thin set of frequencies, its dual lattice, and a smooth periodic function has almost nothing there.

An error bar for points that are not random

Evenly spread points integrate far better than random ones and give no error bar; random points give an error bar and integrate badly. Randomise the even points themselves — shift a lattice by a random vector, or scramble the digits of a Sobol' sequence — and both are kept: an unbiased estimate, a confidence interval from a handful of repeats, and an error that falls faster than any deterministic set's.

The narrowest door sets the pace

How fast a chain forgets is an eigenvalue, and nobody can compute the eigenvalues of a chain worth studying. Cheeger's inequality trades the eigenvalue for a picture — the narrowest door in the state space — and pins the one between the square of the other and twice it. Both ends of that range are reached, on graphs small enough to search completely.

The forgetting that happens all at once

A single small chain forgets its start gradually, a little more with every step. A family of large ones can do something different — stay almost perfectly informed about where it began, and then lose all of it inside a window far shorter than the wait. That cliff is the cutoff phenomenon, and it is why "seven shuffles" is an answer rather than a convention.

When every value comes from the same hat

A rule that sees values one at a time and must keep or discard each on the spot can guarantee half of what a prophet collects, and no more, when the values come from different distributions. When they all come from the same one, the guarantee rises to 0.745 — and a single fixed threshold, set so that each value crosses it with chance 1/n, already secures 1 − 1/e. For bounded values the best rule collects nearly everything; only a heavy tail, where one enormous value carries the prize, keeps the gap open.

The rank that remembers every value

Values arrive one at a time, each must be kept or discarded on the spot, and the aim is to keep one whose rank among all of them is low on average. Told only who is leading, the best rule gets 3.87. Shown the values, a rule gets below 2.33 — and how much lower the best possible rule goes is not known, because the rank of what is kept depends on every value seen, and the best rule may need to remember all of them.

Drawn without putting back

Every concentration bound on this shelf assumes the draws are independent. A real sample is not: a pollster does not ring the same person twice, and every ball taken from an urn changes what is left in it. The dependence runs the helpful way. Wassily Hoeffding proved in 1963 that a sample drawn without replacement is at least as concentrated as one drawn with it, for every convex measure of spread at once — and the variance falls by an exact factor that reaches zero when the whole urn is taken.

Charged for the variance, not the range

Hoeffding's inequality knows one thing about each term of a sum: the interval it lies in. For ten thousand coins that each land heads once in a thousand, that makes it promise almost nothing — a 92% chance of thirty heads, when the truth is two in ten million. Tell the bound each term's variance as well and it changes character: Bernstein's and Bennett's inequalities decay like the normal curve while the deviation is small and like a Poisson tail beyond, and for rare events they are millions of times sharper.

Two constants that do not walk at random

Pollard's factoring method trusts x² + c modulo a prime to repeat as soon as a random function would, after about √p steps. For c = 1 and c = 3 it does. For c = 0 and c = −2 it runs twelve to eighteen times longer on average, with almost no tail and enormous cycles, because those two maps are multiplication in disguise and their cycles are set by the order of 2 rather than by chance. Every other constant walks at random — including in the one respect in which x² + c is plainly not random, that it is two-to-one.