The matrix that is a graph

A Laplacian that is not symmetric

Point the edges and the matrix stops being symmetric. Its row sums are still exactly zero, so the null vector survives; everything built on the quadratic form does not, and the eigenvalues need a plane rather than a line. Asymmetry permits that and does not force it, which the smallest case here demonstrates by being asymmetric and real.

Worth reading first: A matrix with no numbers in it · A matrix that depends on its own eigenvalue · Symmetry is worth more than precision.

Every graph in this field so far has been undirected, and the reason the field stopped there rather than continuing is a specific one: an arrow breaks two properties that all seventeen of those essays rest on, and neither break is repaired by being careful.

Point the edges. The adjacency matrix stops being symmetric, so the Laplacian D_out − A does too, and the first thing to go is not an approximation or an accuracy — it is a theorem.

directed cycle on 18: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a directed cycle on 18, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 7.53·10⁻¹⁶. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 0.9848 and the matrix is 100 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. The ring is the closed form 1 − exp(2πik/n), which every computed eigenvalue lands on to 1.3·10⁻¹⁵.Re λIm λwhat survives the arrowsvertices18arcs18worst row sum0worst column sum0largest |Im λ|0.98asymmetry1the null vector is still exactand nothing else about the spectrum is real
Fig. 1 The Laplacian of a directed cycle on eighteen vertices. Its eigenvalues are 1 − exp(2πik/n), a circle of radius one centred at one, so all but the zero have an imaginary part and no ordering of them exists.

The picture is the first thing that changes and it is not a presentational choice. An undirected Laplacian’s spectrum is a list of real numbers and every figure in this field has drawn it on an axis — smallest to largest, with the gap after the zeros marked. There is no axis here. The eigenvalues are points in a plane, and the closest thing to an ordering is a distance from the origin, which is a choice rather than a fact.

What survives

One thing does, and it is the property the field’s first essay is named for. Every row of D_out − A is the vertex’s out-degree minus the weights of the arcs leaving it, so every row sums to exactly zero by cancellation of integers, and L·1 is the zero vector with no rounding anywhere. 1 is still a right null vector, 0 is still an eigenvalue, and it is still known before anything runs.

The columns do not. A column sums to the out-degree minus the in-degree, which is zero only when the vertex is balanced, and the left null vector is therefore something else entirely — it is the stationary distribution of the random walk, and computing it is a whole problem rather than a lookup.

That asymmetry between the two null vectors is the field’s new fact in its smallest form. In the undirected case they are the same vector and nobody has to say which one is meant.

It is worth noticing which of the two is the cheap one, because it is the opposite of what an application usually wants. The right null vector — the constant — is free and says nothing: every digraph has it, it carries no information about the graph, and the only thing it certifies is that the rows were built by subtraction. The left null vector is the one that describes the graph, and it is the one that costs an iteration. The undirected case gave both away at once and hid which was which.

What does not

The rest of the undirected apparatus goes together, because it is all one argument.

A symmetric Laplacian has a quadratic form xᵀLx = Σ over edges of (xᵢ − xⱼ)², which is what makes λ₂ the minimum of a Rayleigh quotient over vectors orthogonal to 1, which is what makes λ₂ a relaxed cut, which is what makes a sweep over the second eigenvector a partitioning algorithm, which is what Cheeger’s inequality is about. Remove the symmetry and the first link goes: there is no quadratic form, because xᵀLx and xᵀLᵀx are different numbers and neither is a sum over anything.

The loss propagates further than it first looks. The sweep cut is not merely unjustified, it is undefined: there is no real vector to sort. The spectral layout every picture in this field uses places a vertex at its second and third eigenvector entries, and those entries are complex. The component count reads the multiplicity of the zero eigenvalue, which survives — but the gap it is read against is a distance in a plane rather than a difference on a line.

So there is no second-smallest eigenvalue either, and the reason is more basic than the loss of a theorem. The eigenvalues are complex. There is no ordering of the complex numbers compatible with arithmetic, so “the second smallest” is not a quantity that exists — and every ranking of them by modulus or by real part is a choice somebody made rather than a fact about the matrix.

random strongly connected on 24: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a random strongly connected on 24, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 1.54·10⁻¹⁵. The column sums are not: the worst is 4, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1.176 and the matrix is 68.2 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices24arcs58worst row sum0worst column sum4largest |Im λ|1.2asymmetry0.68the null vector is still exactand nothing else about the spectrum is real
Fig. 2 A random strongly connected digraph, whose spectrum has no structure to it at all: the largest imaginary part is 1.18, the matrix is 68 per cent asymmetric, and the eigenvalue at the origin is computed at 1.5·10⁻¹⁵.

The claim that had to be dropped, and it improved the essay

The natural pair of statements is that a symmetric digraph’s Laplacian has a real spectrum and an asymmetric one’s does not. The first is a theorem. The second is false, and the measurement refused it before any of this was written.

The two-block family at twelve vertices is 25.8 per cent asymmetric in the Frobenius norm and has every eigenvalue real, to exactly zero. The graph is small enough that the few arcs crossing between its two rings all happen to be reciprocated in the parts of the matrix that matter, and a non-symmetric matrix is perfectly entitled to a real spectrum — every triangular matrix has one.

So asymmetry permits a complex spectrum and does not force one. What is asserted instead is the closed form: the directed cycle’s eigenvalues are 1 − exp(2πik/n) exactly, computed by the real Schur form and matching to 1.2·10⁻¹⁵, and its largest imaginary part is 0.985 at n = 9. That is a construction rather than an observation, which is what lets it carry a refusal — and the refusal is the one this essay publishes, fed a graph rather than a hope.

The two directions of the same graph

There is a second measurement worth taking before any theory, because it separates two words that get used as though they were one.

A digraph is symmetric when every arc appears both ways with equal weight, which makes its Laplacian the undirected Laplacian and its spectrum real. A digraph is balanced when every vertex’s in-degree equals its out-degree, which is a much weaker condition — a directed cycle is balanced and is as asymmetric as a matrix gets.

The measurement: the balanced family at twenty-four vertices is 82.5 per cent asymmetric and has a largest imaginary part of 1.51. So balance is not a weak form of symmetry, and the fact that it recovers some of the undirected theory — which the next essay is about — is not because it is nearly symmetric. It is because it makes one particular vector, the stationary distribution, equal to one particular other vector, the degrees.

balanced digraph on 24: where a walk spends its time, against where the arcs areTwo distributions over the 24 vertices of a balanced digraph on 24. The taller bar at each vertex is the stationary distribution of the random walk, computed by 434 power iterations with no subtraction anywhere and checked against Pᵀφ = φ at 1.07·10⁻¹⁴; the narrower one is the out-degree divided by the total, which is what an undirected intuition reaches for. On this family they agree to 2.77·10⁻¹⁴, because the digraph is balanced and the closed form says they must. That is the second route the whole construction is checked by.stationary distribution · out-degree sharevertexthe quantity that replaces the degreevertices24balanced1worst gap2.8·10⁻¹⁴as a share of max φ3.6·10⁻¹³Pᵀφ − φ1.1·10⁻¹⁴the degree is localand how often a walk visits is not
Fig. 3 A balanced digraph: eighty-two per cent asymmetric, and its stationary distribution equal to its degree distribution to 10⁻¹⁴. Two quantities that are the same object on an undirected graph, drawn on one where the matrix is not symmetric and they still coincide.

Why this was left until the field’s eighteenth essay

The graph field’s own record says it stopped at undirected graphs rather than running out of material, and the reason is visible now: the directed case is not a variation on the undirected one, it is a different subject that shares a picture.

Every result in the field so far factors through the quadratic form. Two Laplacians of one graph is about which cut each of two symmetric matrices minimises; a partition decided in the last digit is about the eigenvector of a symmetric matrix with a repeated eigenvalue; a graph with a tenth of the edges is about preserving a quadratic form. None of those questions can even be asked here without first building something new.

What is not lost is the field’s central claim, and it is worth checking that it survives: a discrete answer computed by continuous arithmetic is still what a partition is, and a rounding still makes it a different answer rather than a worse one. That part transfers unchanged. It is the machinery for getting the answer that has to be rebuilt.

The walk is the thing that has to be computed

On an undirected graph the stationary distribution of the random walk is the degree distribution. That is a one-line fact and it is why nothing in this field has ever had to compute one: the ranking essay computes a Perron vector because PageRank’s walk is not the graph’s walk, and every other essay reads a degree.

On a digraph the two are different, and the difference is not small. On the two-block family the worst per-vertex gap between the stationary distribution and the out-degree share is 0.062 — six per cent of the total probability mass sitting on the wrong vertex, on a graph with twenty-four of them.

That matters because everything the next essay needs is built from the stationary distribution, and it has three properties the degree does not:

  • It is global. A vertex’s degree is a property of its own arcs. How often a walk visits it depends on the whole graph, so a local change anywhere moves every entry.
  • It might not exist. A walk on a graph that is not strongly connected has more than one closed communicating class and therefore more than one stationary distribution — the essay on a chain with no stationary vector is about exactly that failure, arrived at from the Markov side.
  • And it has to be computed to an accuracy. The degrees are integers, read off the graph with no arithmetic in them at all. The stationary distribution is the fixed point of an iteration, and the figure above reports 824 power iterations to reach 10⁻¹⁴ — so a normalised Laplacian built from it carries an error the undirected one does not have, before any eigenvalue of it is computed.

The iteration is a power iteration rather than a call to an eigensolver, deliberately. Every step is a probability vector times a stochastic matrix, so every entry is a sum of nonnegative terms and no subtraction happens anywhere — which means a nonnegative vector cannot come back with a negative entry. The Perron vector with four negative entries is the measurement that made that a rule here rather than a preference.

two blocks with 1 arc back: where a walk spends its time, against where the arcs areTwo distributions over the 24 vertices of a two blocks with 1 arc back. The taller bar at each vertex is the stationary distribution of the random walk, computed by 824 power iterations with no subtraction anywhere and checked against Pᵀφ = φ at 9.69·10⁻¹⁵; the narrower one is the out-degree divided by the total, which is what an undirected intuition reaches for. They differ by up to 0.0616, which is 62 per cent of the largest entry. The difference is what makes a directed normalised Laplacian a global object: the degree of a vertex is local and how often a walk visits it is not.stationary distribution · out-degree sharevertexthe quantity that replaces the degreevertices24balanced0worst gap0.062as a share of max φ0.62Pᵀφ − φ9.7·10⁻¹⁵the degree is localand how often a walk visits is not
Fig. 4 The unbalanced case: the walk’s distribution against the degree share, differing by up to six per cent of the mass. These are the two candidate weights for a normalised Laplacian and only one of them gives a theorem.

The eigenvectors go too, and that is the part with a cost

The eigenvalues being complex is the visible loss. The eigenvectors being non-orthogonal is the expensive one, and it is worth pricing because it decides which algorithms are available.

A symmetric matrix has an orthonormal eigenbasis. That is what makes the Jacobi sweep this field uses possible, it is what makes an eigenvalue’s condition number exactly 1 — so the essay on eigenvalue sensitivity has nothing to report on a symmetric matrix — and it is what lets a subspace be described by a projector.

A directed Laplacian has none of that. Its eigenvectors are the columns of some invertible matrix, that matrix has a condition number, and the eigenvalues inherit it: an eigenvalue’s sensitivity is 1/|yᵀx| for its left and right eigenvectors, which for a symmetric matrix is 1 and here is not. So a computed directed spectrum carries a perturbation that a computed undirected spectrum does not, and the essay on eigenvalues that are not there describes what that does at its worst.

The eigenvalue at the origin is the exception and it is worth saying why. It is not sensitive, because it is not computed from the matrix’s numerical structure at all: the right null vector is 1 by construction and the row sums cancel as integers. Everything about the zero survives; everything about the rest is now a numerical question.

What is left to build on

The situation after this essay is worth stating plainly, because it is unusually bare.

There is a matrix. It has an exact null vector on one side and an unknown one on the other. Its eigenvalues are complex, unordered, and attached to eigenvectors that are not orthogonal. There is no quadratic form, no variational characterisation, no Cheeger inequality, no sweep cut, and no second-smallest eigenvalue.

What there is instead is a walk, which exists whenever the graph is strongly connected and which carries all the information the degrees carried in the undirected case and more. Chung’s construction symmetrises the Laplacian with respect to it and recovers the whole apparatus — a real spectrum in [0, 2], a variational characterisation, a sweep cut and a Cheeger inequality. The price is that the quantity the inequality is about is no longer the weight of the arcs crossing a cut, and that substitution is the next essay.

Chung's Laplacian against the undirected one, on four familiesThe worst entrywise difference between the directed normalised Laplacian and the undirected normalised Laplacian of the same graph with its arrows removed. On the balanced and symmetric families the difference is at the rounding level at every size — 10⁻¹⁴ and below — because a balanced digraph's stationary distribution is its degree distribution and the construction then collapses to the undirected one. On the two-block family with a single arc back the difference is 0.163 at n = 30, which is the size of what the direction adds. The two curves are the same computation and the flat one is a closed form, so this figure is the field's second route rather than a comparison.711151923273110⁻¹⁷10⁻¹⁴10⁻¹¹10⁻⁸10⁻⁵10⁻²verticesworst entrywise differencetwo blocks · random strongbalanced · symmetrica closed form, and what it excludesbalanced, worst3.9·10⁻¹⁴symmetric, worst1.1·10⁻¹⁶two blocks0.16random strong0.28balanced is undirected in disguiseand nothing else is
Fig. 5 What the symmetrisation is checked against: on a balanced digraph it must reduce exactly to the undirected normalised Laplacian, and it does, at 10⁻¹⁴ across every size.

The one closed form the directed case supplies

Every field on this site needs at least one object whose answer is known in advance, because a measurement checked only against another measurement is checked against nothing. The undirected field had several — the path’s spectrum, the cycle’s, the hypercube’s, Foster’s identity, the exact spanning-tree count. The directed field has one, and it is the cycle.

The Laplacian of a directed cycle on n vertices is I − S, with S the cyclic shift, and S’s eigenvalues are the n-th roots of unity. So the spectrum is

λₖ  =  1 − exp(2πik/n),      k = 0 … n − 1

a circle of radius one centred at 1, passing through the origin. Every eigenvalue but λ₀ has a nonzero imaginary part, the whole spectrum is determined by n, and there is nothing approximate about it.

Measured against the real Schur form the agreement is 1.2·10⁻¹⁵ at n = 9, which is the rounding level for a computation of that size. That is the ring in the hero and it is the only assertion in this essay that compares a computed spectrum with a formula rather than with a property.

It is also the reason the cycle is the family the refusal uses. A claim that a graph Laplacian’s eigenvalues are real has to be refused by an instance, and an instance whose spectrum is known is worth more than one whose spectrum was computed — because if the computation were wrong, an instance of the second kind would refuse the claim for the wrong reason.

directed cycle on 12: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a directed cycle on 12, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 8.65·10⁻¹⁶. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1 and the matrix is 100 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. The ring is the closed form 1 − exp(2πik/n), which every computed eigenvalue lands on to 1.3·10⁻¹⁵.Re λIm λwhat survives the arrowsvertices12arcs12worst row sum0worst column sum0largest |Im λ|1asymmetry1the null vector is still exactand nothing else about the spectrum is real
Fig. 6 Twelve vertices, twelve eigenvalues, all of them on the ring the closed form draws.

At other settings

symmetric digraph on 20: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a symmetric digraph on 20, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 2.16·10⁻¹⁵. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 0 and the matrix is 0 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices20arcs112worst row sum0worst column sum0largest |Im λ|0asymmetry0the null vector is still exactand nothing else about the spectrum is real
Fig. 7 The degenerate case: every arc both ways, every eigenvalue real to exactly zero, and the picture is a line inside a plane.
balanced digraph on 24: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a balanced digraph on 24, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 3.89·10⁻¹⁶. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1.505 and the matrix is 82.5 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices24arcs45worst row sum0worst column sum0largest |Im λ|1.5asymmetry0.82the null vector is still exactand nothing else about the spectrum is real
Fig. 8 Balanced and 82 per cent asymmetric, which is the pair of words this field has to keep apart.
two blocks with 1 arc back: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a two blocks with 1 arc back, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 3.09·10⁻¹⁵. The column sums are not: the worst is 2, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 0.1044 and the matrix is 23.6 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices30arcs66worst row sum0worst column sum2largest |Im λ|0.1asymmetry0.24the null vector is still exactand nothing else about the spectrum is real
Fig. 9 Two blocks with one arc back, where the imaginary parts are small and not zero.
directed cycle on 36: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a directed cycle on 36, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 4.69·10⁻¹⁶. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1 and the matrix is 100 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. The ring is the closed form 1 − exp(2πik/n), which every computed eigenvalue lands on to 1.3·10⁻¹⁵.Re λIm λwhat survives the arrowsvertices36arcs36worst row sum0worst column sum0largest |Im λ|1asymmetry1the null vector is still exactand nothing else about the spectrum is real
Fig. 10 The closed form at thirty-six vertices, filling in the same circle.
directed cycle on 8: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a directed cycle on 8, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 1.46·10⁻¹⁶. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1 and the matrix is 100 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. The ring is the closed form 1 − exp(2πik/n), which every computed eigenvalue lands on to 1.3·10⁻¹⁵.Re λIm λwhat survives the arrowsvertices8arcs8worst row sum0worst column sum0largest |Im λ|1asymmetry1the null vector is still exactand nothing else about the spectrum is real
Fig. 11 And at eight, where the eight eigenvalues are visibly the eighth roots of unity reflected.
random strongly connected on 32: where a walk spends its time, against where the arcs areTwo distributions over the 32 vertices of a random strongly connected on 32. The taller bar at each vertex is the stationary distribution of the random walk, computed by 142 power iterations with no subtraction anywhere and checked against Pᵀφ = φ at 7.92·10⁻¹⁵; the narrower one is the out-degree divided by the total, which is what an undirected intuition reaches for. They differ by up to 0.071, which is 79.4 per cent of the largest entry. The difference is what makes a directed normalised Laplacian a global object: the degree of a vertex is local and how often a walk visits it is not.stationary distribution · out-degree sharevertexthe quantity that replaces the degreevertices32balanced0worst gap0.071as a share of max φ0.79Pᵀφ − φ7.9·10⁻¹⁵the degree is localand how often a walk visits is not
Fig. 12 A random strongly connected digraph, where the walk and the degrees disagree everywhere.
directed cycle on 18: where a walk spends its time, against where the arcs areTwo distributions over the 18 vertices of a directed cycle on 18. The taller bar at each vertex is the stationary distribution of the random walk, computed by 1 power iterations with no subtraction anywhere and checked against Pᵀφ = φ at 0; the narrower one is the out-degree divided by the total, which is what an undirected intuition reaches for. On this family they agree to 1.39·10⁻¹⁷, because the digraph is balanced and the closed form says they must. That is the second route the whole construction is checked by.stationary distribution · out-degree sharevertexthe quantity that replaces the degreevertices18balanced1worst gap1.4·10⁻¹⁷as a share of max φ2.5·10⁻¹⁶Pᵀφ − φ0the degree is localand how often a walk visits is not
Fig. 13 The directed cycle, where every vertex has one arc in and one out and the two distributions are both uniform.
One sweep cut, two conductances: 0.1235 and 0.04Every prefix of the sweep over the second eigenvector of Chung's Laplacian, on two blocks with 1 arc back at 24 vertices, with two quantities plotted for each. The circulation conductance is the probability that a step of the stationary walk crosses the cut, divided by the smaller side's stationary mass; the arc conductance is the weight of arcs crossing it, divided by the smaller side's degree. At the best cut they are 0.1235 and 0.04, a factor of 3.09 apart. The shaded band is Cheeger's, λ₂/2 = 0.03309 to √(2λ₂) = 0.3638 with λ₂ = 0.06619, and it is a statement about the first quantity only. Reading it as a statement about the second is the substitution this figure exists to separate.Cheeger's band, for the circulationcirculation · arcsvertices on the smaller sidewhich of the two the theorem is aboutλ₂ of 𝓛0.066λ₂/20.033√(2λ₂)0.36circulation Φ0.12arc conductance0.04their ratio3.1the inequality holdsabout a quantity nobody counts
Fig. 14 What the symmetrisation buys, and what it renames.
two blocks 40: 40 vertices, 223 edges, and a matrix built from themThe two blocks 40 laid out at its own second and third Laplacian eigenvectors, so the picture is the same object the measurements are about. The Laplacian L = D − A is built by subtraction of integers, so every row sums to exactly zero — measured at 0, not at the rounding level — and L·1 is the zero vector with no arithmetic error anywhere in it. Its 1 zero eigenvalue counts the connected components, which breadth-first search also puts at 1; the smallest non-zero eigenvalue is 1.458 and the largest computed zero is 4.26·10⁻¹⁶, a gap of 3.42·10¹⁵. The quadratic form xᵀLx is Σ over edges of (xᵢ − xⱼ)², which at the alternating vector is 444.the matrix, measuredvertices40edges223‖L·1‖∞0zero eigenvalues1components, by search1λ₂1.5laid out at its own eigenvectorsand the row sums are exactly zero
Fig. 15 The undirected object this field spent seventeen essays on, for comparison.
Twenty eigenvalues, one of them zero, and the graph has one pieceThe Laplacian spectrum of two ten-vertex paths joined by a single edge of conductance 10⁻⁴, plotted on a logarithmic axis so the zero end is visible. The dashed line is the threshold 10⁻¹⁰‖L‖ = 1.02·10⁻⁹, and the number of eigenvalues below it is the spectral count of connected components: 1. Breadth-first search, which performs no arithmetic on any weight, says 1. The largest computed zero is 9.42·10⁻¹⁷ and the smallest eigenvalue above the threshold is 1.999·10⁻⁵. A count of components is therefore a rank decision at a threshold, and the bridge's conductance decides which side of it λ₂ falls on — at nine orders above the rounding level, not at it.13579111315171910⁻¹⁹10⁻¹⁶10⁻¹³10⁻¹⁰10⁻⁷10⁻⁴10⁻¹indexeigenvalue of L10⁻¹⁰‖L‖, the thresholda count that is a thresholdeigenvalues below it1components, by search1largest computed zero9.4·10⁻¹⁷smallest above it2·10⁻⁵ratio2.1·10¹¹the graph is connectedat every stop of the slider
Fig. 16 An undirected Laplacian’s spectrum, which is a list of real numbers on a line.
169 steps to forget where it started, and λ₂ said 178A random walk on a sixty-vertex graph with two blocks joined at a density of 0.02, started at one vertex, measured as the total variation distance from its stationary distribution. It reaches 10⁻⁶ after 169 steps. The dashed line is |λ₂|ᵏ with λ₂ = 0.92517 from the walk matrix, and the number of steps that predicts is 178. The distance is not the eigenvalue: for the first several steps the walk is leaving one block and the decay is faster than the asymptote, and the two curves become parallel only once the slowest mode is all that is left. The eigenvalue is a rate, not a schedule, and the difference between them is 8.63 steps.02448729612014416810⁻⁶10⁻⁵10⁻⁴10⁻³10⁻²10⁻¹1steps of the walkdistance from stationarya rate read as a timeλ₂ of the walk0.93steps measured169steps predicted178ratio0.95the dashed line is the eigenvaluethe curve is the walk
Fig. 17 The walk whose stationary distribution this essay needs, in the field’s undirected version of it.
4 entries of a strictly positive vector are negativeThe Perron eigenvector of a twelve-vertex clique with a tail of 24, normalised to its largest entry, plotted as absolute values on a logarithmic axis. Perron–Frobenius says every entry of this vector is strictly positive, and it is a theorem about a connected nonnegative matrix rather than a statement about a good day. Each step along the tail divides the entry by about the Perron root 11.0077, so the entries fall by a decade a step and reach the rounding level after sixteen. Beyond it 4 of them are negative — the smallest at -1.507·10⁻²⁰ — and they are drawn as open circles. The bracket printed beside it is Collatz–Wielandt: for any strictly positive x, the Perron root lies between the smallest and largest of (Ax)ᵢ/xᵢ, so it is a two-sided bound that is proved, from one product, and after forty steps it is 1.2·10⁻⁴ wide.1611162126313610⁻²²10⁻¹⁹10⁻¹⁶10⁻¹³10⁻¹⁰10⁻⁷10⁻⁴10⁻¹vertex, clique first then tail|entry| relative to the largestone rounding of the largest entrya bound that is provedPerron root11bracket, low11bracket, high11bracket width1.2·10⁻⁴smallest entry-1.5·10⁻²⁰entries below zero4every entry is positiveand the picture disagrees
Fig. 18 The Perron vector whose negative entries made power iteration the rule here.
directed cycle on 26: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a directed cycle on 26, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 6.16·10⁻¹⁶. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 0.9927 and the matrix is 100 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. The ring is the closed form 1 − exp(2πik/n), which every computed eigenvalue lands on to 1.3·10⁻¹⁵.Re λIm λwhat survives the arrowsvertices26arcs26worst row sum0worst column sum0largest |Im λ|0.99asymmetry1the null vector is still exactand nothing else about the spectrum is real
Fig. 19 Twenty-six vertices on the ring the closed form draws.
random strongly connected on 36: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a random strongly connected on 36, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 7.06·10⁻¹⁶. The column sums are not: the worst is 4, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 0.9311 and the matrix is 65.6 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices36arcs86worst row sum0worst column sum4largest |Im λ|0.93asymmetry0.66the null vector is still exactand nothing else about the spectrum is real
Fig. 20 A larger random strongly connected digraph, filling the plane.
balanced digraph on 18: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a balanced digraph on 18, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 2.77·10⁻¹⁶. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1.617 and the matrix is 81.9 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices18arcs34worst row sum0worst column sum0largest |Im λ|1.6asymmetry0.82the null vector is still exactand nothing else about the spectrum is real
Fig. 21 Balanced at eighteen, still complex and still not symmetric.
symmetric digraph on 30: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a symmetric digraph on 30, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 1.75·10⁻¹⁵. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 0 and the matrix is 0 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices30arcs268worst row sum0worst column sum0largest |Im λ|0asymmetry0the null vector is still exactand nothing else about the spectrum is real
Fig. 22 Every arc both ways at thirty vertices, and a spectrum on a line.
balanced digraph on 18: where a walk spends its time, against where the arcs areTwo distributions over the 18 vertices of a balanced digraph on 18. The taller bar at each vertex is the stationary distribution of the random walk, computed by 298 power iterations with no subtraction anywhere and checked against Pᵀφ = φ at 8.78·10⁻¹⁵; the narrower one is the out-degree divided by the total, which is what an undirected intuition reaches for. On this family they agree to 2·10⁻¹⁴, because the digraph is balanced and the closed form says they must. That is the second route the whole construction is checked by.stationary distribution · out-degree sharevertexthe quantity that replaces the degreevertices18balanced1worst gap2·10⁻¹⁴as a share of max φ2·10⁻¹³Pᵀφ − φ8.8·10⁻¹⁵the degree is localand how often a walk visits is not
Fig. 23 The closed form at eighteen vertices, agreeing to 10⁻¹⁴.
two blocks with 1 arc back: where a walk spends its time, against where the arcs areTwo distributions over the 32 vertices of a two blocks with 1 arc back. The taller bar at each vertex is the stationary distribution of the random walk, computed by 1088 power iterations with no subtraction anywhere and checked against Pᵀφ = φ at 9.5·10⁻¹⁵; the narrower one is the out-degree divided by the total, which is what an undirected intuition reaches for. They differ by up to 0.053, which is 75.9 per cent of the largest entry. The difference is what makes a directed normalised Laplacian a global object: the degree of a vertex is local and how often a walk visits it is not.stationary distribution · out-degree sharevertexthe quantity that replaces the degreevertices32balanced0worst gap0.053as a share of max φ0.76Pᵀφ − φ9.5·10⁻¹⁵the degree is localand how often a walk visits is not
Fig. 24 The unbalanced case at thirty-two, where the gap does not shrink.
Chung's Laplacian against the undirected one, on four familiesThe worst entrywise difference between the directed normalised Laplacian and the undirected normalised Laplacian of the same graph with its arrows removed. On the balanced and symmetric families the difference is at the rounding level at every size — 10⁻¹⁴ and below — because a balanced digraph's stationary distribution is its degree distribution and the construction then collapses to the undirected one. On the two-block family with a single arc back the difference is 0.1473 at n = 24, which is the size of what the direction adds. The two curves are the same computation and the flat one is a closed form, so this figure is the field's second route rather than a comparison.71115192310⁻¹⁷10⁻¹⁴10⁻¹¹10⁻⁸10⁻⁵10⁻²verticesworst entrywise differencetwo blocks · random strongbalanced · symmetrica closed form, and what it excludesbalanced, worst3.7·10⁻¹⁴symmetric, worst1.1·10⁻¹⁶two blocks0.15random strong0.28balanced is undirected in disguiseand nothing else is
Fig. 25 The check that the symmetrisation collapses when it should.
One sweep cut, two conductances: 0.07599 and 0.1515Every prefix of the sweep over the second eigenvector of Chung's Laplacian, on two blocks with 3 arcs back at 30 vertices, with two quantities plotted for each. The circulation conductance is the probability that a step of the stationary walk crosses the cut, divided by the smaller side's stationary mass; the arc conductance is the weight of arcs crossing it, divided by the smaller side's degree. At the best cut they are 0.07599 and 0.1515, a factor of 1.99 apart. The shaded band is Cheeger's, λ₂/2 = 0.03821 to √(2λ₂) = 0.391 with λ₂ = 0.07643, and it is a statement about the first quantity only. Reading it as a statement about the second is the substitution this figure exists to separate.Cheeger's band, for the circulationcirculation · arcsvertices on the smaller sidewhich of the two the theorem is aboutλ₂ of 𝓛0.076λ₂/20.038√(2λ₂)0.39circulation Φ0.076arc conductance0.15their ratio2the inequality holdsabout a quantity nobody counts
Fig. 26 What the symmetrisation buys, at thirty vertices.
The lower bound is attained and the upper one is out by 21.5×Cheeger's inequality on nine graphs of about 24 vertices each: for the normalised Laplacian's λ₂, λ₂/2 ≤ φ ≤ √(2λ₂), where φ is the conductance the sweep cut actually achieves. Each row shows the two bounds as a bar and the measured conductance as a dot inside it. The lower bound is tight on the complete at a ratio of 1. The upper bound is loosest on the barbell — by a factor of 21.5 — which is the graph in the census with a real bottleneck, and therefore the shape the inequality is always quoted about. The square root is what makes it loose: it is the price of turning a spectral quantity into a combinatorial guarantee, and it is paid where the guarantee is wanted.10⁻³10⁻²10⁻¹110¹12345678910conductance, and the two bounds on itpathcyclegridbarbelltwo blockshypercubepreferentialstarcompletethe bar is the inequalitythe dot is the graph
Fig. 27 The undirected inequality this field has to rebuild.
One real vector, 39 candidate cuts, and the best of them is number 20The Fiedler vector of the two blocks 40, its entries sorted, drawn as the pale rising curve against the right-hand scale; and against the logarithmic left-hand scale, the conductance of the cut that takes the first k vertices in that order. λ₂ = 1.4579. The eigenvector is a real vector and the answer wanted is a subset, so something has to round it: the sweep takes every prefix and keeps the best, which here is k = 20 at a conductance of 0.08491 against a worst prefix of 1 — a factor of 11.8 between the best cut this vector offers and the worst. Nothing in the eigenvalue problem chose k; the sorting did.071421283510⁻¹1vertices on the smaller sideconductance of the prefix cut0.0849, the best prefixthe rounding stepλ₂1.5cuts considered39best conductance0.085at k =20worst prefix1the dashed curve is the eigenvectorthe solid one is what it costs
Fig. 28 The undirected sweep cut, which needs a real eigenvector.
directed cycle on 48: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a directed cycle on 48, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 9.44·10⁻¹⁷. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1 and the matrix is 100 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. The ring is the closed form 1 − exp(2πik/n), which every computed eigenvalue lands on to 1.3·10⁻¹⁵.Re λIm λwhat survives the arrowsvertices48arcs48worst row sum0worst column sum0largest |Im λ|1asymmetry1the null vector is still exactand nothing else about the spectrum is real
Fig. 29 Forty-eight vertices on the ring.
two blocks with 1 arc back: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a two blocks with 1 arc back, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 4.39·10⁻¹⁶. The column sums are not: the worst is 2, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 0.08805 and the matrix is 23.5 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices40arcs88worst row sum0worst column sum2largest |Im λ|0.088asymmetry0.23the null vector is still exactand nothing else about the spectrum is real
Fig. 30 Two blocks at forty, where the imaginary parts are small and not zero.
random strongly connected on 12: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a random strongly connected on 12, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 1.81·10⁻¹⁵. The column sums are not: the worst is 3, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1.033 and the matrix is 65.7 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices12arcs29worst row sum0worst column sum3largest |Im λ|1asymmetry0.66the null vector is still exactand nothing else about the spectrum is real
Fig. 31 A small random strongly connected digraph.
balanced digraph on 36: a Laplacian whose eigenvalues need a planeThe spectrum of L = D_out − A for a balanced digraph on 36, drawn in the complex plane. The row sums are exactly zero — measured at 0, not at the rounding level — so 1 is a right null vector and 0 is an eigenvalue, computed here at 1.09·10⁻¹⁵. The column sums are not: the worst is 0, and the left null vector is the stationary distribution rather than 1. The largest imaginary part is 1.784 and the matrix is 82.6 per cent asymmetric in the Frobenius norm — and those two numbers are not the same statement: asymmetry permits a complex spectrum and does not force one, which this family demonstrates at small sizes by being asymmetric and real. There is no closed form for this family; the degenerate check is the symmetric one, where every imaginary part is exactly zero.Re λIm λwhat survives the arrowsvertices36arcs67worst row sum0worst column sum0largest |Im λ|1.8asymmetry0.83the null vector is still exactand nothing else about the spectrum is real
Fig. 32 Balanced at thirty-six, and as asymmetric as ever.
random strongly connected on 18: where a walk spends its time, against where the arcs areTwo distributions over the 18 vertices of a random strongly connected on 18. The taller bar at each vertex is the stationary distribution of the random walk, computed by 98 power iterations with no subtraction anywhere and checked against Pᵀφ = φ at 4.61·10⁻¹⁵; the narrower one is the out-degree divided by the total, which is what an undirected intuition reaches for. They differ by up to 0.0811, which is 63.7 per cent of the largest entry. The difference is what makes a directed normalised Laplacian a global object: the degree of a vertex is local and how often a walk visits it is not.stationary distribution · out-degree sharevertexthe quantity that replaces the degreevertices18balanced0worst gap0.081as a share of max φ0.64Pᵀφ − φ4.6·10⁻¹⁵the degree is localand how often a walk visits is not
Fig. 33 The walk against the degrees on a random digraph.
symmetric digraph on 24: where a walk spends its time, against where the arcs areTwo distributions over the 24 vertices of a symmetric digraph on 24. The taller bar at each vertex is the stationary distribution of the random walk, computed by 51 power iterations with no subtraction anywhere and checked against Pᵀφ = φ at 3.53·10⁻¹⁵; the narrower one is the out-degree divided by the total, which is what an undirected intuition reaches for. On this family they agree to 2.23·10⁻¹⁵, because the digraph is balanced and the closed form says they must. That is the second route the whole construction is checked by.stationary distribution · out-degree sharevertexthe quantity that replaces the degreevertices24balanced1worst gap2.2·10⁻¹⁵as a share of max φ3.4·10⁻¹⁴Pᵀφ − φ3.5·10⁻¹⁵the degree is localand how often a walk visits is not
Fig. 34 And on a symmetric one, where they are the same object.
Chung's Laplacian against the undirected one, on four familiesThe worst entrywise difference between the directed normalised Laplacian and the undirected normalised Laplacian of the same graph with its arrows removed. On the balanced and symmetric families the difference is at the rounding level at every size — 10⁻¹⁴ and below — because a balanced digraph's stationary distribution is its degree distribution and the construction then collapses to the undirected one. On the two-block family with a single arc back the difference is 0.1683 at n = 38, which is the size of what the direction adds. The two curves are the same computation and the flat one is a closed form, so this figure is the field's second route rather than a comparison.7111519232731353910⁻¹⁷10⁻¹⁴10⁻¹¹10⁻⁸10⁻⁵10⁻²verticesworst entrywise differencetwo blocks · random strongbalanced · symmetrica closed form, and what it excludesbalanced, worst9.7·10⁻¹⁴symmetric, worst1.1·10⁻¹⁶two blocks0.17random strong0.28balanced is undirected in disguiseand nothing else is
Fig. 35 The closed-form check out to thirty-eight.
One sweep cut, two conductances: 0.06114 and 0.1364Every prefix of the sweep over the second eigenvector of Chung's Laplacian, on two blocks with 2 arcs back at 20 vertices, with two quantities plotted for each. The circulation conductance is the probability that a step of the stationary walk crosses the cut, divided by the smaller side's stationary mass; the arc conductance is the weight of arcs crossing it, divided by the smaller side's degree. At the best cut they are 0.06114 and 0.1364, a factor of 2.23 apart. The shaded band is Cheeger's, λ₂/2 = 0.04358 to √(2λ₂) = 0.4175 with λ₂ = 0.08717, and it is a statement about the first quantity only. Reading it as a statement about the second is the substitution this figure exists to separate.Cheeger's band, for the circulationcirculation · arcsvertices on the smaller sidewhich of the two the theorem is aboutλ₂ of 𝓛0.087λ₂/20.044√(2λ₂)0.42circulation Φ0.061arc conductance0.14their ratio2.2the inequality holdsabout a quantity nobody counts
Fig. 36 The quantity the symmetrisation renames.

What links here

Computed from the collection, not written here: the essays that point at this one.

Shares its objects with

Essays that name at least two of the same things, and that neither author linked.

Named objects

A flat tag is an object no other essay names yet.

Complex eigenvaluesDirected graphGraph laplacianNull spaceQuadratic formStationary distributionSymmetry