Concept

Eigenvalue

The factor by which a linear map stretches one of the directions it does not turn. It is a root of the characteristic polynomial, and its size decides whether repeated application of the map grows or shrinks that direction.

Named by 33 essays across 7 fields — each of them below, with the objects they name alongside it.

The directions the map leaves alone. Unit vectors and their images under the map. On the two marked lines the image points the same way as the original, stretched by 3.00 and 1.00.

The directions a map leaves alone

Almost every arrow comes out of a transformation pointing somewhere else. A few come out pointing exactly where they went in, only longer or shorter. Those few decide nearly everything the map does.

algebra · Eigenvectors
The unit square, mapped: area × 5. The unit square and the parallelogram it becomes under a linear map, with the area of that parallelogram computed from its own corners and set against ad − bc.

The number that says how much room is left

A linear map takes the unit square to a parallelogram. The area of that parallelogram is one number, it is computable from the four entries of the matrix, and almost everything the determinant is used for is a restatement of that sentence.

algebra · Determinant
The polynomial whose roots are the stretches. The determinant of A − λI plotted against λ for the map [2, 1, 1, 2], with its roots at 3 and 1 marked.

The polynomial whose roots are the stretches

Finding the directions a map leaves alone means finding the numbers at which it crushes something to nothing. Those numbers are the roots of one quadratic, and everything the map does is written in its two coefficients.

algebra · Eigenvectors
The same map, written in the basis of its own eigenvectors. Three panels: the map [2, 1, 1, 2] on the standard grid, the diagonal stretch by 3 and 1 it becomes on the eigenvector grid, and the two put back together.

The same map in a better basis

Measured along its own invariant directions, a linear map stops shearing and becomes two independent stretches. Nothing about the map has changed; the grid it is described against has.

algebra · Eigenvectors
The level curve, and the axes the matrix chooses. The curve xᵀAx = 1 for the matrix [2, 0.8, 0.8, 1.4], drawn by solving for the radius at each angle, with the two eigen-directions marked; they cross at a right angle and are the axes of the curve.

Symmetry forces a right angle

A matrix equal to its own reflection across the diagonal always has real stretches and always has perpendicular directions to stretch along. Neither is true of matrices in general, and both follow from one line of algebra.

algebra · Eigenvectors
The flow of a linear equation, and the matrix that runs it for one unit of time. Paths of points moving so that their velocity is [0.25, −1.2, 1.2, 0.25] applied to their position, with the position after time 1 marked on each; the matrix taking start to finish is e^A.

The exponential of a square

The series for e makes perfect sense with a matrix in it. What comes out solves a system of equations the way the ordinary exponential solves one, and a skew matrix exponentiates into a rotation with no trigonometry anywhere.

analysis · The exponential
How fast a chain forgets where it started. The total variation distance to the stationary distribution plotted logarithmically against the number of steps, for each of 3 starting states. The curves are straight lines of equal slope.

How long until it forgets

The essays before this one settle where a chain ends up and how much time it spends there, and none of them asks how long the settling takes. That question has an exact answer, it is a single number, and it is the only thing any practical use of a chain depends on.

probability · Markov chains
One sign decides which curve it is. 3 conics drawn from the general quadratic, each labelled with its discriminant B² − 4AC and the curve that sign names, checked against how many times the curve meets a large circle.

One sign decides which curve

The general quadratic in two variables has six coefficients and draws a conic. Which of the four it draws is settled by a single combination of three of them, and the other three cannot change the answer however they are chosen.

geometry · Conic sections
A map drawn through the cycle 0 → 1/3 → 1, and the graph its pieces make. The graph of a map made of two straight pieces through a cycle of three points, with the cycle drawn as a staircase, beside a two-node graph showing which piece may follow which and the matrix of that graph.

A matrix that counts the returns

Draw a map straight through the cycle 0 → 1/3 → 1 and its two pieces carry each other in a fixed pattern: the left piece only across the right, the right across both. The orbits' words are then walks on a two-node graph, and the number of points that come back after n steps is the trace of that graph's matrix to the nth power — 1, 3, 4, 7, 11, 18 — each one checked by solving for the points exactly.

dynamics · Symbolic dynamics
Four shapes of perimeter 300 between their inner and outer circles. A square, an ellipse, a Reuleaux triangle and a stadium, each drawn with the largest circle inside it and the smallest circle around it, the ring between the two shaded, with the ring's width and the widest ring allowed.

Nearly the most means nearly round

A shape that holds almost as much as a circle of the same perimeter must almost be a circle. Bonnesen made that exact: the ring between a convex shape's largest inscribed circle and smallest enclosing circle is never wider than √(L² − 4πA)/π. Three quite different shapes holding 99% of the circle's area all have rings under 9.55 wide, and not one of 200 random convex shapes breaks the bound.

geometry · Isoperimetric
9 corners of the 4-cube: some corner always has 2 chosen neighbours. The 4-dimensional cube with 9 of its corners chosen so that no chosen corner has more than 2 chosen neighbours, the fewest possible, with the edges between chosen corners drawn heavy.

Half the cube and √n neighbours

Choose more than half the corners of an n-dimensional cube, any way at all, and some chosen corner has at least √n chosen neighbours. That statement about a cube settled a thirty-year question about how sensitive a truth function must be to its inputs, and its proof is a matrix of plus and minus ones whose square is n times the identity. A search over every choice for the 4-cube finds the bound exactly: nine corners, and some corner always has two chosen neighbours.

logic · Truth functions
Discs from the rows, and the eigenvalues inside them. The complex plane with one disc per row of a 3×3 matrix — centred on its diagonal entry, with radius the sum of the sizes of the rest of the row — and the 3 eigenvalues marked: 4.14, −5.09, −1.06.

Discs that fence in the eigenvalues

Draw one disc for each row of a square matrix, centred on the diagonal entry, with a radius equal to the sum of the sizes of everything else in that row. Every eigenvalue lies inside one of the discs, and a group of discs set apart from the rest holds exactly as many eigenvalues as it has discs. Nothing is solved to find them.

algebra · Eigenvectors
The largest value on each plane, over every direction of space. A longitude–latitude map of directions u in three dimensions, each shaded by the largest value of xᵀAx on the plane perpendicular to u. The smallest such value is the middle eigenvalue 1.829, reached at ±v₁; the largest is 3.622, reached along a great circle.

The highest point on the sphere is an eigenvalue

For a symmetric matrix, walk a unit arrow over every direction and record the value of xᵀAx. The highest value reached is the largest eigenvalue, the lowest is the smallest, and every eigenvalue in between is a saddle height, a minimum of maxima. From that one description comes a theorem no formula for the roots could give: delete a row and its column, and every eigenvalue of what is left sits between two of the original's.

algebra · Eigenvectors
Powers of a matrix that shrink in the end. Three curves of the norm of the n-th power of a two-by-two matrix against n on a logarithmic axis: one decays steadily, two rise to peaks of about 18 and 7 before decaying.

A geometric series whose ratio is a matrix

1 + r + r² + … adds to 1/(1 − r) when r is smaller than one. Put a matrix in place of r and the same formula holds, with the inverse matrix in place of the fraction — but what must be smaller than one is not the matrix's size. It is its largest eigenvalue. A matrix whose eigenvalues are 0.9 and 0.8 can stretch vectors ten times over before its powers begin to shrink, and the series still converges, after a detour the eigenvalues say nothing about.

analysis · Geometric series
The trace–determinant plane and the flows it sorts. The plane of trace against determinant, divided by the horizontal axis and the parabola tr² = 4 det into saddle, node, spiral and centre regions, with 6 matrices marked and their phase portraits drawn alongside.

Two numbers decide the flow

A linear system in the plane has four coefficients, and what its solutions do forever afterwards — spiral in, race out, swing round, or split along two lines — is decided by two of the numbers made from them. The plane of trace against determinant is a complete map of the possibilities, and the only places it cannot decide are the lines where it changes its mind.

analysis · The exponential
Which real 2 × 2 matrices have a real logarithm, read off their eigenvalues. Eigenvalues of 7 matrices plotted in the complex plane with the negative real axis emphasised, beside a table saying for each whether a real logarithm exists and why: A yes, B yes, C yes, D no, E no, F yes, G no.

The matrix that has no logarithm

Every square matrix has an exponential, and a matrix exponential is always invertible. The converse fails, and it fails in a way that can be read straight off the eigenvalues — a real matrix with eigenvalues −1 and −2 is no exponential at all, while minus the identity is the exponential of a whole continuum of matrices that do not even commute with each other.

analysis · The exponential
The narrowest door in two 6-cliques joined by one edge, and the gap it pins down. two 6-cliques joined by one edge, with the vertex set of smallest conductance coloured and the 1 edges leaving it thickened. Beside it a logarithmic ruler marks half the conductance squared, the spectral gap and twice the conductance, in that order from the bottom.

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.

probability · Markov chains
Two different graphs with the same adjacency matrix eigenvalues. a star with four arms: adjacency matrix eigenvalues 2, 0³, −2; a square and a lone point: adjacency matrix eigenvalues 2, 0³, −2. The characteristic polynomials are identical.

Two graphs the eigenvalues cannot tell apart

A graph's matrix has eigenvalues, and they count a surprising amount of the drawing: its edges, its triangles, every closed walk of every length. They do not count everything. A star with four arms and a square beside a lone point have the same eigenvalues exactly, although one of them is in two pieces — and on six points ten of the 156 graphs have a twin of this kind.

algebra · Linear maps
The Hoffman–Singleton graph: five pentagons, five pentagrams. Fifty vertices of degree seven and girth five, built from five pentagons and five pentagrams joined by the rule j of pentagon h to h·i + j of pentagram i.

As many points as two steps allow

In a graph where every point has d neighbours and every point is within two steps of every other, there can be at most d² + 1 points — one, its d neighbours, and d(d − 1) more reached through them. Graphs that meet the bound exactly are rare to the point of absurdity. The pentagon does it for d = 2, the Petersen graph for d = 3, a fifty-point graph found in 1960 for d = 7, and an eigenvalue argument proves there is nothing else — except possibly one graph with 3,250 points and 57 neighbours each, which nobody has found or ruled out.

discrete · Extremal graphs
How many three-state chains constant rates can produce. s 0.05: 94.9%; s 0.1: 90.2%; s 0.2: 79.6%; s 0.3: 70.3%; s 0.5: 47.7%; s 0.7: 22.9%; s 1: 4.3%.

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.

analysis · The exponential
The lowest mode of the regular pentagon. Level lines of the first Dirichlet eigenfunction of a regular 5-gon; λ times area 18.9191.

Three sides and four are proved

Among all shapes of a given area, the disc has the smallest lowest eigenvalue. Among triangles it is the equilateral one, among quadrilaterals the square — both proved by sliding chords to an axis. For five sides the regular pentagon wins every computation, every nudge raises its value by the square of the nudge, and there is still no proof.

geometry · Isoperimetric
Five urns, each drawn as walks. Simulated walks of 2000 draws for urns with replacement matrices (0,1,1,0), (2,1,1,2), (3,1,1,3), (7,1,1,7), (1,0,0,1).

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.

probability · Random walk
One form completed three ways, and the same straddle each time. The plane coloured by the sign of 2x² + 6xy + 3y². Three pairs of lines, one for completing with x first, one for y first and one along the eigenvectors; their coefficients are 2 and −3/2; 3 and −1; 5.541 and −0.541, and every pair has one line in a positive wedge and one in a negative wedge.

The signs no completion can change

A quadratic in several variables can be completed square by square in many orders, and each order hands back different coefficients. What no order changes is how many come out positive and how many negative — and that count, read off a completion of A − tI, says how many eigenvalues lie below t without finding a single one.

algebra · Completing the square
A random sequence that shrinks, grows or wanders, by one parameter. Runs of tₖ₊₁ = tₖ ± β·tₖ₋₁ at β = 0.6, 0.70258, 0.8 on a logarithmic scale: shrinking, wandering and growing.

A threshold no average can see

Damp the random Fibonacci sequence — add or subtract β times the term before last, by a coin — and somewhere between β = 0.6 and β = 0.8 almost every run stops shrinking and starts growing. The switch is at β ≈ 0.70258. The average of the squares, which can be computed exactly, grows for every β there is, so the one number that decides the sequence's fate is invisible to the one number anybody can compute.

dynamics · Sensitive dependence
Two chaotic maps, coupled weakly and strongly. The gap between two coupled copies of the logistic map at full chaos: at ε = 0.15 it stays large; at ε = 0.3 it falls to nothing.

Two chaotic orbits made to agree

Chaos means two copies of a system started a hair apart soon have nothing in common. Couple them — let each take a fraction ε of its next step from the other — and above a definite strength they agree to every digit for ever, while each stays exactly as chaotic as before. The strength needed is read off the same number that measured the chaos: the Lyapunov exponent.

dynamics · Sensitive dependence
Strings that overlap themselves leave more behind. 00: 253.293; 01: 230.283; 02: 230.299; 03: 230.315; 04: 230.330; 05: 230.345; 06: 230.361; 07: 230.376; 08: 230.390; 09: 230.405; 10: 220.887; 11: 244.784; 12: 222.442; 13: 223.049; 14: 223.575; 15: 224.033; 16: 224.438; 17: 224.798; 18: 225.120; 19: 225.410; 20: 225.673; 21: 225.913; 22: 249.163; 23: 226.334; 24: 226.520; 25: 226.692; 26: 226.852; 27: 227.001; 28: 227.140; 29: 227.270; 30: 227.393; 31: 227.508; 32: 227.617; 33: 250.747; 34: 227.817; 35: 227.910; 36: 227.998; 37: 228.081; 38: 228.161; 39: 228.237; 40: 228.310; 41: 228.380; 42: 228.446; 43: 228.510; 44: 251.598; 45: 228.631; 46: 228.688; 47: 228.743; 48: 228.796; 49: 228.847; 50: 228.896; 51: 228.944; 52: 228.990; 53: 229.035; 54: 229.078; 55: 252.146; 56: 229.161; 57: 229.201; 58: 229.239; 59: 229.277; 60: 229.313; 61: 229.348; 62: 229.383; 63: 229.416; 64: 229.449; 65: 229.481; 66: 252.538; 67: 229.543; 68: 229.572; 69: 229.601; 70: 229.630; 71: 229.657; 72: 229.684; 73: 229.711; 74: 229.737; 75: 229.762; 76: 229.787; 77: 252.837; 78: 229.835; 79: 229.859; 80: 229.882; 81: 229.904; 82: 229.927; 83: 229.948; 84: 229.970; 85: 229.991; 86: 230.011; 87: 230.031; 88: 253.077; 89: 230.071; 90: 230.090; 91: 230.109; 92: 230.128; 93: 230.146; 94: 230.164; 95: 230.182; 96: 230.199; 97: 230.216; 98: 230.233; 99: 253.275.

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.

analysis · Harmonic series
A ring of identical clocks locked into a twist. Ring of 40, nearest-neighbour coupling, settled into a 2-twisted state; largest deviation 2.6e-11; order parameter 2.5e-16.

A ring of clocks settles into a twist

Put identical oscillators round a ring, each pulled only by its neighbours, and start them at random. They always lock — every one ends at the same frequency — but most of the time not in step: the angles wind round the ring a whole number of times and stay wound. The winding number a random start chooses is spread like a bell whose width grows like the square root of the ring, and only coupling to about a third of the ring makes full synchrony certain.

dynamics · Sensitive dependence
Eigenvalues that will not crowd together. Unfolded spacings, eigenvalues: 0.36 1.52 1.15 0.88 1.23 1.13 0.34 0.75 2.05 0.65 1.66 0.64 1.07 1.01 0.86 1.23 0.69 1.30 0.77 0.94 0.99 0.71 0.92 2.49 0.32 1.07 0.62 1.34 0.07 0.51 0.98 2.25 0.64 0.78 0.35 0.95 1.36 1.89 0.21 0.79 0.46 0.84 1.12 1.52 0.96 0.65 1.09 0.29 1.30 1.06 1.83 1.50 0.28 0.43 1.19 1.01 0.52 1.54 1.51 1.56. Independent points: 0.34 0.01 0.31 3.72 0.25 2.29 0.98 2.14 1.99 0.54 0.15 0.58 0.96 0.22 0.26 1.11 0.04 0.38 0.55 0.74 0.89 0.11 1.47 0.52 0.06 2.11 1.29 3.04 0.44 0.28 2.49 3.97 0.87 1.04 0.83 0.77 0.79 0.69 1.76 1.82 0.14 0.07 1.03 0.88 1.05 0.37 0.06 0.82 0.70 1.72 0.90 1.24 0.32 0.25 1.07 0.55 1.59 3.23 0.36 0.40. Gaps under a quarter of the mean: 2 against 9.

Eigenvalues that keep their distance

Fill a large symmetric matrix with independent random numbers and its eigenvalues do something no list of independent numbers does: they refuse to crowd together. The reason is a count of conditions in a two-by-two matrix, and its consequences — a semicircle, gaps that vanish at nought, a spectrum almost as rigid as a lattice — reach from heavy nuclei to the zeros of the zeta function.

algebra · Eigenvectors
Fewer tables have rates behind them as states are added. n=2: 0.02:1.000 0.05:1.000 0.1:1.000 0.2:1.000 0.3:1.000 0.5:1.000 0.7:0.847 1:0.489; n=3: 0.02:0.975 0.05:0.945 0.1:0.881 0.2:0.797 0.3:0.690 0.5:0.453 0.7:0.244 1:0.043; n=4: 0.02:0.941 0.05:0.865 0.1:0.735 0.2:0.520 0.3:0.329 0.5:0.108 0.7:0.019 1:0.001; n=5: 0.02:0.896 0.05:0.755 0.1:0.526 0.2:0.275 0.3:0.130 0.5:0.020 0.7:0.001 1:0.000; n=6: 0.02:0.807 0.05:0.575 0.1:0.388 0.2:0.090 0.3:0.025 0.5:0.000 0.7:0.000 1:0.000

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.

analysis · The exponential
The sizes three sentences allow. every point has exactly one neighbour: 2,4,6,8,10,12,14,16,18,20,22,24,26,28,30; every two points have exactly one common neighbour: 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29; every two players are both beaten by a third: 1,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30

The sizes at which a sentence can come true

Write a sentence about a relation and ask for which numbers of points some relation makes it true. "Every two points have exactly one common neighbour" comes true at every odd size and no even one, and every graph that satisfies it is a windmill of triangles. "Every two players are both beaten by some third" fails for every tournament of two to six players and holds from seven on. A search can settle one size at a time; only a proof settles all of them, and the two sentences need different kinds.

logic · Quantifiers
Three arrows in and three out: counting on prisms of the cubic grid. Entropy per point of Eulerian orientations on cubic prisms: 2x2: 2.79369, 2x3: 2.68151, 3x3: 2.58713; small tori 2 × 2 × 2 3.14618, 2 × 2 × 3 2.95513, 2 × 3 × 3 2.81100, 3 × 3 × 3 2.69925; Pauling 2.5.

Three arrows in and three out

Put an arrow on every edge of the cubic grid so that every point has three arrows in and three out. On the square grid the same question — two in, two out — has an exact answer, found by Elliott Lieb in 1967: about 1.5396 arrangements per point. On the cubic grid nobody knows the answer. Pauling's guess of 2.5 is a proven lower bound, and exact counts on prisms of the grid come down towards it from above — 2.794, 2.682, 2.587 as the cross-section widens — without reaching it.

discrete · Eulerian paths
Six lines all at one angle. The six diagonals of the icosahedron: every pair meets at cosine 0.447214, an angle of 63.43 degrees.

Six lines all at one angle

Take the six long diagonals of an icosahedron: every pair meets at the same angle. Squaring each line into the matrix that projects onto it removes every sign and leaves a table no vector can kill, so at most d(d + 1)/2 lines can do this in d dimensions. The bound is reached at 2, 3, 7 and 23, and at no other dimension anyone can find.

algebra · Inner product
A table with harmonic rows. The Hilbert table of size 8 and its largest stretch 1.6959390, against pi.

A table that never reaches π

Fill a square table with 1/1, 1/2, 1/3 along the rows, each row starting one place further on. Every row is a piece of the harmonic series, and yet the table stretches no vector by as much as π. At a thousand rows the stretch is 2.42, at a trillion it is near 2.94, and one integral explains both the number and the failure to arrive.

analysis · Harmonic series

Named alongside it

The objects these essays reach for when they reach for this one.

MatrixDeterminantEigenvectorExhaustive searchTraceMatrix exponentialBasisCharacteristic polynomialLogarithmMarkov chainQuadratic formStability

All concepts