Variance
Named by 36 essays across 8 fields — each of them below, with the objects they name alongside it.
How far from the average a thing can be
Knowing only an average and a spread — nothing about the shape, nothing about the number of outcomes, nothing about symmetry — the chance of landing three standard deviations out is at most one in nine. And there is a distribution that lands there exactly that often, so the bound cannot be improved.
The average settles and the wobble does not
Two theorems are usually met a page apart and sound as though one is a sharper version of the other. They are the same sums looked at through two different magnifying glasses: divide by the number of them and everything collapses to a point, divide by its square root and a shape appears.
The walk that becomes a curve
Shrink the steps of a random walk and it disappears. Shrink them while stretching the time in the right proportion — space by the square root of whatever time is divided by — and something is left behind, which is a curve nobody could draw.
The shape that averaging leaves alone
Adding independent quantities blurs their distributions together, and rescaling restores the width. Almost every shape is changed by that operation. Exactly one is returned unaltered, and that is why sums of unrelated things keep arriving at it.
An average that never settles
The average of many independent quantities is supposed to steady as their number grows. For one famous distribution it does not steady at all — the average of a thousand draws has exactly the same distribution as a single draw, and no amount of further averaging changes it.
How fast the bell arrives
The limit theorem says a standardised sum approaches the bell curve and says nothing about when. The rate is one over the square root of the number of terms, the constant in front is made of the third moment, and both are visible.
The error that does not care how many dimensions
A grid gets rapidly better in one dimension and hopelessly worse in twenty. Random points get better at the same slow rate whatever the dimension, which is why a method that is bad everywhere ends up being the only one that works.
Sampling where the answer lives
Monte Carlo error cannot be made to fall faster than the square root, so the only thing left to attack is the constant in front of it. Drawing points where the integrand is large, and dividing by how often they were drawn, leaves the answer alone and can shrink the noise many times over.
Too many orders to list
The rule is an average over every order the players could have arrived in. At seven players that is five thousand orders and at twenty it is more than there are seconds in the age of the universe — so the average is sampled, and the error falls at a rate that can be measured.
Finding a threshold with two moments
Every monotone property of a random graph has a threshold, and locating one is nearly always the same two calculations — count what the property needs, and check the count does not concentrate on rare cases. The triangle is where the method is cleanest.
The coefficient that is a polynomial
Add a second variable to track a statistic and each coefficient stops being a number. Set the new variable to one and the old count comes back untouched; leave it in and the mean of the statistic is a derivative rather than an average.
The bound is the answer to a search
Chebyshev's inequality is not a clever estimate that happens to be sharp. It is the exact answer to a maximisation over all distributions with a stated mean and variance, and the polynomial that proves nothing beats it is the certificate a search of that kind always produces.
No single input can move it far
Independence was never the hypothesis doing the work. A quantity built from many separately drawn inputs concentrates whenever changing one of them moves it only a little — and that covers quantities which are not sums of anything and have no formula at all.
Any unevenness brings the match sooner
Real birthdays are not spread evenly across the year, and every such departure pushes the famous twenty-three down rather than up. The proof is one move on two days at a time, and what it leaves behind is a single number — the one ecologists use to count species.
A coin in front of every term
Put all plus signs in front of 1, 1/2, 1/3, … and the sum runs off to infinity; alternate them and it settles on log 2. Toss a fair coin for each sign instead, and the sum settles — every time, on a different number. Where it tends to settle has a smooth, flat-topped shape, and at the value 2 that shape takes a height that agrees with one eighth to forty-two decimal places and is not one eighth.
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.
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.
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 numbers in one jagged record
A measured record comes with no rule, so its dimension has to be estimated from a finite stretch of samples — and two things go wrong that never trouble a graph built from a formula. The popular estimator depends on the units the record is written in, and the dimension, which describes the record up close, turns out to be independent of its memory, which describes it from far away. For a self-affine path the two are tied by D = 2 − H; for a record, they are two numbers.
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.
An urn forgets its start only below one half
Let each draw from an urn add balls of both colours in fixed amounts, and the long run depends on a single ratio of two eigenvalues. Below one half the urn behaves like a coin, its fluctuations spread like the square root of the draws and settle into a bell. Above one half the first few draws decide most of the outcome, the spread grows faster, and the shape that results is not a bell and depends on how the urn began.
Regression is a square completed halfway
A bell in two variables, sliced at a fixed x, is a bell in y — and completing the square in y alone says where it is centred: at ρx, on a line flatter than the ellipse's own axis. That gap is regression to the mean, and the same completion says it runs backwards in time just as well as forwards.
Counting a population by its repeats
The birthday problem runs forwards from a known year to the chance of a match. Run it backwards and the matches measure the year: count how many pairs among the draws came out the same, and the number of kinds is about the number of pairs divided by that. It costs the square root of the population rather than a census of it — and when the kinds are unevenly common, what the repeats measure is not how many kinds there are but how many they behave like.
Two-thirds of a step past the line
Add random numbers between nought and one until the total passes one, and it takes e of them on average. Ask the same about passing any t and the answer is an exact finite sum that settles, within a few units, onto the line 2t + 2/3. The slope is one over the average step. The two-thirds is the interesting part: it is what the one step that crosses the line charges for being chosen by its length, and it is set entirely by how much the steps vary.
Sectors that shrink with the primes
Hecke proved that the angles of the primes a² + b² are evenly spread, so every fixed sector eventually gets its share. Let the sector shrink as the primes grow, to width X^(−α), and the question is open beyond small α. Measured on the 581,517 such primes between twenty and forty million, wide shrinking sectors fluctuate far less than chance and thin ones as much as chance — and the thin sectors that go empty are not scattered: they gather beside the directions of small rational slope, most of all slopes like 1/3 and 1/1 whose two terms are both odd.
The longest run has no limit
Toss a fair coin n times and the longest run of heads is about log₂ n. The average settles: it is log₂ n − 0.667 to within a few millionths by a million tosses. The distribution does not. Because the run is a whole number and log₂ n is not, the chance of each value swings round the same cycle once every doubling of n, for ever, and computing it exactly to a million tosses shows the swing as large at the end as at the start.
Edges that crowd each other out
Pick a spanning tree of a graph uniformly at random, and the presence of one edge makes every other edge less likely — a theorem of Kirchhoff's electricity. Pick a forest instead, or a connected subgraph, and the same is believed and unproved. Checking every pair of edges in every labelled graph up to six vertices, 871,926 pairs for forests alone, finds no exception, and comes as close as 0.9994 to one.
The longest thread between two random strings
Write down two random strings of a thousand bits each and find the longest sequence that can be read in both, skipping freely. It is about 81% of each, and the exact share — the Chvátal–Sankoff constant — has resisted computation for fifty years. Its fluctuations are a smaller mystery with a sharper edge: the variance is known to be at most n/2, and over every length that can be computed it grows like n to the power 0.7, with nobody able to say what it does after that.
One, plus a quarter, plus a ninth
Give n workers n jobs with every cost drawn at random, average one, and find the cheapest way to pair them. However large n is, the cheapest total averages less than π²/6, and for every n it is exactly 1 + 1/4 + 1/9 + … + 1/n². Parisi guessed the formula in 1998 from n = 1, 2 and 3; thirteen sizes of random tables, solved exactly, land on it within sampling error.
The shape of a cell around a random point
Scatter points at random and give each the region nearer to it than to any other. The cells come in every shape from triangles to twelve-sided polygons, and almost nothing about them is fixed by a formula — except one thing: on a closed surface they have exactly six sides on average, every time, because three cells meet at every corner. Two hundred thousand cells measure the rest: 29.5 per cent hexagons, Lewis's law, and small cells surrounded by large ones.
How many new kinds the next sample will show
Count how many kinds were seen once, twice, three times. Good and Toulmin's alternating sum f₁t − f₂t² + f₃t³ − … then predicts, with no bias at all, how many new kinds a further sample t times as large will reveal. It works perfectly up to t = 1 and is worthless just beyond: past that point the kinds seen most often are multiplied by t to a high power, and the spread of the estimate passes 10⁴⁰ by t = 1.5. Truncating the sum at a random point rescues it, and each fourfold increase in the first sample buys about half a unit more of future.
One random path down a search tree
Before running a search that may take a week, it is worth knowing whether it will take a week or a century. Donald Knuth's answer, from 1975, is to walk down the search tree once at random, multiply the numbers of choices met on the way, and add up the products: the result is, on average, exactly the size of the tree. On the eight-queens tree one path guesses 769 against a true 2,057 and twenty thousand paths average to within a tenth of a per cent. On a tree built to hide its bulk, the same unbiased estimate is almost always wrong by a factor of thousands.
The middle value has a bell of its own
The central limit theorem is a theorem about sums, and the median of a sample is not a sum of anything. It has a bell all the same: the middle of n draws spreads like 1/(2f√n), where f is the height of the density at the middle — not the variance, which the median never consults. That one number decides when the median beats the average, by a factor of two for a two-sided exponential and outright for the Cauchy law; and where the density is nought at the middle, the bell fails, slowing to a fourth root or splitting in two.
A quarter turn quiets the echoes
Lay a sequence of signs on a shifted copy of itself and add up the squares of the overlaps. Random signs leave the total near n²/2, and the squares modulo a prime, started a quarter of the way along, cut it to a twelfth. Exhaustive search finds far quieter short sequences, and no infinite family is proved to do better than 6.342.
Numbers whose subset sums all differ
The powers of two 1, 2, 4, …, 64 have 128 subset sums and no two are equal. So do 20, 31, 37, 40, 42, 43 and 44, whose largest member is smaller, and an exhaustive search shows no seven numbers do it with less. A pigeonhole on the sums says the largest member must grow like 2ⁿ/n; a variance argument improves that to 2ⁿ/√n; the best known sets sit near 0.22 × 2ⁿ. Whether the truth is a constant times 2ⁿ is Erdős's five-hundred-dollar question, and the pigeonhole cannot reach it.
Named alongside it
The objects these essays reach for when they reach for this one.
ExpectationConvergence rateNormal distributionExhaustive searchScalingConcentration inequalityHeavy tailsSamplingIndependenceRandom walkTail boundCentral limit theorem