Analysis — page 4

A dust that almost every line misses

Keep the four corner squares of a square, then the four corners of each of those, and so on. What is left has a length, in the sense that measures length, and yet almost every straight line misses it entirely. Stage by stage the chance that a random line hits the dust falls — to 48% by the eighth stage — and how fast it falls to nothing is one of the few questions about a simple picture that is still open.

A series that waits on π

Add 1/(n³ sin² n) for n = 1, 2, 3, … and the terms are mostly tiny, except where n is almost a multiple of π and sin n is almost nought. Ten million terms add to 30.3145, four-fifths of it from the single term at n = 355. Whether the sum is finite depends on how closely fractions can approach π — on a number called its irrationality measure — and the best proof available says only that the measure is below 7.1, where the series needs it below 2.5.

The exponent a staircase shares with its set

The Cantor function rises from nought to one on a set of length nought, and it is Hölder continuous with exponent log 2/log 3 — the same number as the dimension of that set. It is not a coincidence: both numbers say that an interval of width r carries mass r to the power 0.6309, and that one inequality proves the dimension and the smoothness at once. Tilt the weights of the construction and the two numbers separate, which shows exactly what the coincidence was measuring.

When a table of moves came from steady rates

A process that jumps between states at constant rates, watched once a year, produces a table of yearly moves that is the exponential of its rates. Most tables that anyone could write down are not — a random three-state table is only about one time in twenty-three — and the few that are can have two different sets of rates behind them, but only once the process has forgotten where it started.

The best nodes have no formula

Interpolating through n points magnifies any error in the data by at most the Lebesgue constant of the points. Chebyshev's points keep it near (2/π)·log n; stretching them to the ends of the interval brings it within two hundredths of the best possible; and the best possible points, characterised in 1978 by having every bump of the error curve the same height, have never been given a formula.

Only one curve through the factorials bends the right way

Infinitely many smooth curves pass through 1, 1, 2, 6, 24, 120, and some of them even obey the factorial's own rule, f(x + 1) = (x + 1)·f(x), at every x. Ask that the logarithm of the curve bend upwards everywhere and exactly one survives — Euler's integral. A wiggle of any size breaks the condition somewhere, the smaller the wiggle the further out, and the proof that nothing else survives is a squeeze that computes √π along the way.

The sine of a figure eight

Walk round a circle and read the height against the distance walked, and the reading is the sine. Do the same on Bernoulli's figure eight and the reading is a new function, with an arc integral that has t⁴ where the circle's has t², a length Gauss found inside an average, two periods instead of one — and exactly the same list of equal divisions that compass and straightedge can draw.

The sine, rebuilt from its zeros

A polynomial is a product of factors, one for each root. Euler treated the sine the same way — one factor for each place the wave crosses the axis — and the product he wrote down is correct, though the zeros alone do not justify it. Multiplied out, it hands over the sum of the reciprocal squares, π²/6, and every even power after it.

Settling in order settles everywhere at once

A sequence of functions can settle at every point and never settle uniformly — the largest gap stays put while each point escapes it. Dini found the circumstances in which that cannot happen: continuous functions, a continuous limit, every member above the next, and a closed interval. Under those four conditions convergence at each point is convergence everywhere at once, and Weierstrass's iteration for the square root becomes a sequence of polynomials converging uniformly to |x|.

Close in height and nowhere close in slope

Uniform convergence carries continuity to the limit and carries integrals to the limit. It carries slopes nowhere. The functions sin(n²x)/n flatten to nothing while their slopes grow without bound; smooth curves converge uniformly to a corner; a Fourier series converges and its differentiated series diverges. What does carry a slope is uniform convergence of the slopes themselves — and the reason is that integration averages wiggles away while differentiation multiplies them.

The tangent counts zigzags

Arrange the numbers 1 to n so that they go up, down, up, down. There are 1, 1, 2, 5, 16, 61, 272 such arrangements, and those are, multiplied by factorials, the Taylor coefficients of tan x and sec x — the odd lengths from the tangent, the even from the secant. Désiré André proved it in 1879 by splitting a zigzag at its largest number, and the split turns out to be the derivative of the tangent written out term by term.

Seven times half of π, and then not

The area under sin x / x, out to infinity, is π/2. Multiply in sin(x/3)/(x/3) and the area is still π/2; multiply in the waves for 5, 7, 9, 11 and 13 and it is still π/2, exactly. Add the wave for 15 and the area falls short — by about 2.3 × 10⁻¹¹. Nothing about the curves changes at the eighth factor. What changes is that 1/3 + 1/5 + … + 1/15 has just passed 1.

A sum of factorials that converges at every prime

1 + 1 + 2 + 6 + 24 + 120 + … grows faster than any geometric series and has no sum in the real numbers. Measure size by divisibility instead and the same series converges at every prime at once, to a different number each time. Whether any of those numbers ends in zero is a question Đuro Kurepa asked in 1971, and a search through every prime below 100,000 says no while a coin-toss model says it should have said yes about twice.

The slope a Gauss sum leaves behind

Riemann is said to have offered sin x + sin(4x)/4 + sin(9x)/9 + … as a continuous function with no derivative anywhere. It has one after all, at π and at every π times an odd number over an odd number, and the slope there is always exactly −1/2. Computed with a million terms, the function's behaviour at every fraction is read off a single number — a quadratic Gauss sum — and the slope appears exactly where that number is zero.

What a missing string of digits leaves behind

Strike from the harmonic series every term whose denominator contains 42 and the rest adds up to 228.45. Strike out 99 instead and it adds up to 253.28. The difference is not about which numbers are lost but about waiting: 99 overlaps itself, so it takes 110 random digits on average to turn up where 42 takes 100, and the sum is almost exactly ln 10 times that wait.

Where a random walk leaves the snowflake

Start a random walk at the centre of the Koch snowflake and record where it first touches the edge. The edge has dimension 1.26, but the places the walks arrive form a set of dimension one: 60,000 walks spread their entropy by ln 3 a level, not ln 4, and the busiest fifth of the edge takes nine tenths of them. Makarov proved in 1985 that it must be so for every simply connected region in the plane.

The length round an ellipse

An ellipse has an area anyone can write down, πab, and a perimeter nobody can: it is an integral with no expression in the functions of school mathematics. Kepler guessed it, Euler guessed better, and in 1914 Ramanujan wrote down a formula whose error falls like the fifth power of how far the ellipse is from a circle — off by four hundredths of a per cent even for a segment. Gauss had already found the way to compute it exactly, doubling the correct digits at every step.

Turning a slow series geometric

Leibniz's series 1 − 1/3 + 1/5 − 1/7 + … adds up to π/4, and it takes five billion terms to give ten decimal places. Euler found a rule that rewrites it, using exactly the same terms, as a series whose terms halve at every step — and forty of those give thirteen places. Aitken's and Shanks's later rules do better still. None of them works on every slow series, and the reason is that each one is a guess about the shape of the error.

The corners of a random hull

Drop ten thousand points at random into a triangle and stretch a band round them: about twenty points touch it. Drop them into a disc and about seventy-three do. The count grows like a logarithm in one shape and like a cube root in the other, and the reason is where the boundary bends.

A wave known only at its ticks

Read a sine wave nine cycles a second at ten instants a second and the readings describe a wave of one cycle a second instead. Read a wave with nothing faster than 3.7 cycles a second at more than 7.4 instants a second and the readings determine it exactly. Both facts are the same fold of the frequency axis, and it is why wagon wheels turn backwards on film.

A grid that packs the copies

Sampling a pattern on a grid makes copies of its frequencies, one for every point of a lattice, and the pattern can be read back from the samples only if the copies do not collide. For a pattern whose frequencies fill a disc the square grid wastes room between its copies. Searching every lattice of samples for the fewest that work finds a nearly hexagonal one every time — 124 samples where the square grid needs 169 — and the saving is the hexagonal packing's over the square, 13.4% in the limit. In three dimensions it is 29%, and in twenty-four the cubic grid needs 16,777,216 times as many samples as the best.

The route round that rates cannot avoid

A table of yearly moves between states came from steady rates only if it is the exponential of a rate matrix. With two states that is true of half of all tables, with three of one in twenty-four, with four of about one in two thousand, and of none of 400,000 tables drawn at random with five. Even a table that is nine parts staying put fails more and more often as states are added, and always for one reason: rates that carry 1 to 2 and 2 to 3 also carry 1 to 3, and a table whose direct move is rarer than that route round cannot have come from them.

A slope without a subtraction

The definition of a slope subtracts two nearly equal numbers and divides by a tiny one, and a computer that follows it loses half its digits however carefully the step is chosen. Two pieces of arithmetic avoid the subtraction altogether. Step sideways into the complex numbers by 10⁻²⁰⁰ and read off the imaginary part; or calculate with numbers a + bε in which ε² = 0, and the slope rides along exactly. Run backwards, the second gives all thousand slopes of a function of a thousand variables for under five times the cost of the function.

The zeros a partial sum cannot avoid

The exponential is never zero, and every partial sum 1 + z + z²/2! + … + zⁿ/n! of its series is zero n times. Divide those zeros by n and they line up on a single egg-shaped curve, |z e^(1−z)| = 1, found by Szegő in 1924. The curve is where two sizes tie — the exponential the series is reaching for, and the terms it has not yet reached — and the same tie, carried one step further, places every zero to within a few hundredths.