# Hardy-Muckenhoupt Bounds for Laplacian Eigenvalues

We present two graph quantities Psi(G,S) and Psi_2(G) which give constant factor estimates to the Dirichlet and Neumann eigenvalues, lambda(G,S) and lambda_2(G), respectively. Our techniques make use of a discrete Hardy-type inequality.

## Authors

• 4 publications
• 2 publications
• 4 publications
04/19/2021

### Computing eigenvalues of the Laplacian on rough domains

We prove a general Mosco convergence theorem for bounded Euclidean domai...
10/08/2018

### On the semiclassical spectrum of the Dirichlet-Pauli operator

This paper is devoted to semiclassical estimates of the eigenvalues of t...
07/13/2020

### Integer Laplacian eigenvalues of strictly chordal graphs

In this paper, we establish the relation between classic invariants of g...
05/18/2019

### A note on variance bounds and location of eigenvalues

We discuss some extensions and refinements of the variance bounds for bo...
12/21/2021

### Asymptotic analysis and numerical computation of the Laplacian eigenvalues using the conformal mapping

We consider the Dirichlet and Neumann eigenvalues of the Laplacian for a...
11/25/2021

### Graph recovery from graph wave equation

We propose a method by which to recover an underlying graph from a set o...
07/18/2019

### Bounds on Spreads of Matrices related to Fourth Central Moment. II

We derive some inequalities involving first four central moments of disc...
##### This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

## 1 Introduction

Let be a vertex and edge weighted undirected connected graph, i.e. forms a connected graph and , are positive weight functions on the edges and vertices respectively. We will think of our graphs as spring mass systems where vertex has mass and edge has spring constant . Let be the weighted adjacency matrix, let be the weighted degree matrix, and let be the Laplacian matrix. Let be the diagonal mass matrix. Then, the generalized eigenvalues of with respect to have a nice interpretation. Specifically, solutions of the generalized eigenvalue problem

 Lx=λMx

correspond to modes of vibration of the spring mass system. When the spring mass system is connected, is the fundamental mode of vibration111 The quantity is referred to in the literature under various names: the algebraic connectivity, the Fiedler value, the fundamental eigenvalue, etc. In this paper we will refer to as the Neumann eigenvalue to emphasize the boundary assumptions and to parallel our development in the Dirichlet case. . For an introduction to spring mass systems and the Laplacian, see chapter 5 of [strang2006linear].

The following result, known as Cheeger’s inequality, can be traced back to [alon1985lambda1, cheeger1969lower, dodziuk1984difference]. Define the isoperimetric quantity222 The quantity is often referred to as the conductance of the graph or the Cheeger constant. In this paper we will refer to as the isoperimetric constant and reserve the term conductance for the conductance of an edge. of to be

 Φ(G)=minA{∑e∈E(A,¯A)cemin(μ(A),μ(¯A))∣∣ ∣∣A,¯A≠∅}.

Then we can bound by

 λ22≤Φ≤√2λ2maxidiμi.

In this paper, we introduce the Neumann contentof a graph. Roughly, the Neumann content, , is the minimum ratio over subsets of the conductance between and and the minimum mass of either set. Thus, noting that is the conductance between and , the isoperimetric constant is roughly equal to the Neumann content where the minimization is restricted to sets . We will show how to use to give a constant factor estimate of . Along the way we will also define the Dirichlet content, , which allows us to estimate the Dirichlet eigenvalue. In particular, we prove the following theorems.

###### Theorem.

Let be a vertex and edge weighted connected graph with boundary set , a proper nonempty subset of . Let be the Dirichlet eigenvalue and let be the Hardy quantity of . Then

 Ψ4≤λ≤Ψ.
###### Theorem.

Let be a vertex and edge weighted connected graph. Let be the Neumann eigenvalue and let be the Neumann content of . Then,

 Ψ24≤λ2≤Ψ2.

### 1.1 Related work

A very recent independent paper [schild2018schur] introduced a quantity specifically in the case of the normalized Laplacian, i.e., when . In this setting, the Neumann content  is equivalent to the definition of up to constant factors: . In [schild2018schur], it is proved that

 ρ25600≤λ2≤2ρ.

This parallels our Theorem 6.5 in the normalized Laplacian case with different constants.

The application of the Hardy-Muckenhoupt inequality to estimating the Dirichlet eigenvalue was noted in [miclo1999example]. In that paper, the authors showed how to bound the Dirichlet eigenvalue on an infinite path graph by the (infinite path analogue of) . Specifically,

 Ψ4≤λ≤2Ψ.

This parallels our Theorem 4.5 in the case of a vertex and edge weighted path graph with different constants.

Other methods for estimating have been proposed. A method for lower bounding based on path embeddings is presented in [GuLeMi99, GuatteryMiller2000, kahale1997semidefinite]. In this method, a graph with known eigenstructure is embedded into a host graph. Then the fundamental eigenvalue of the host graph can be estimated in terms of the eigenstructure of the embedded graph and the “distortion” of the embedding. For a review of path embedding methods, see the introduction in [GuatteryMiller2000].

### 1.2 Applications

The Laplacian matrix, and in particular its eigenstructure, finds many applications in computer science, physics, numerical analysis, and the social sciences. Computing the Neumann eigenvector of a graph has become a standard routine used in image segmentation

[shi2000normalized] and clustering [von2007tutorial, ng2002spectral]. The eigenstructure of the Laplacian is used to model virus propagation in computer networks [wang2003epidemic] and design search engines [brin1998anatomy]. In numerical analysis and physics, the Laplacian matrix is used to approximate differential equations such as heat flow and the wave equations on meshes [quarteroni2009numerical].

In section 2, we set notation and discuss background related to weighted graphs, Laplacians, the eigenvalue problems, and electrical networks. In section 3, we introduce Muckenhoupt’s weighted Hardy inequality. In section 4, we introduce the Hardy quantity and the Dirichlet content and show how Muckenhoupt’s result can be used to bound the Dirichlet eigenvalue on a path graph. In section 5, we extend the bounds on the Dirichlet eigenvalue from path graphs to arbitrary graphs. Finally in section 6, we introduce the two-sided Hardy quantity and the Neumann content and extend the bounds on the Dirichlet eigenvalue on a graph to the Neumann eigenvalue on a graph.

## 2 Preliminaries

### 2.1 Vertex and edge weighted graphs

Let be an undirected connected graph with vertex set and edge set . The mass of vertex is and the conductance333 As we are dealing with spring mass systems, perhaps it would be better to refer to these quantities as spring constants and compliances. Nonetheless, we have chosen to refer to these quantities as conductances and resistances as this is the terminology most commonly found in the spectral graph theory literature. of edge is . We will assume our graphs are connected and that all masses and conductances are positive.

### 2.2 Laplacians

Let be the degree of vertex and let be the degree matrix. Let be the adjacency matrix of , i.e. if and otherwise. The Laplacian matrix corresponding to is . Note that the quadratic form associated with is

 x⊤Lx=∑(u,v)∈Eκ(u,v)(xu−xv)2.

### 2.3 The generalized Laplacian eigenvalue problem

Let be the diagonal matrix of masses.

###### Definition 2.1.

The Neumann problem on is to find

 λ2(G)=minx∈RV{x⊤Lxx⊤Mx∣∣∣x⊤M1=0,x≠0}.

We will refer to the minimum value as the Neumann eigenvalue.

By the Courant-Fischer min-max principle, we can rewrite this quantity as

 λ2 =minWmaxx∈Wx⊤Lxx⊤Mx

where varies over the two dimensional subspaces of .

At times we will consider the Laplacian eigenvalue problem with extra boundary conditions. This corresponds to fixing the value of at a given set of vertices to zero.

###### Definition 2.2.

Let be a graph and let be a proper nonempty subset of . The Dirichlet problem on with boundary set is to find

 λ(G,S) =minx∈RV{x⊤Lxx⊤Mx∣∣∣x↾S=0,x≠0}.

We will refer to the minimum value as the Dirichlet eigenvalue.

###### Remark 2.3.

Letting be the principal submatrix of indexed by vertices in and letting be the restriction of onto the corresponding coordinates, we have

 L¯Sx¯S=λ(G,S)x¯S.

In other words is an eigenvector of . We caution that itself is, in general, not an eigenvector of .

### 2.4 Graphs as electrical networks and effective resistance

Given an edge-weighted graph, we can think of its edges as electrical conductors with conductance . Thinking of as an assignment of voltages to the vertices of our electrical network, we have that

 x⊤Lx=∑(u,v)∈Eκ(u,v)(xu−xv)2

is the “power dissipated in our system”. Then drawing inspiration from physics, we define the effective resistance between two sets of vertices in terms of the minimum power required to maintain a unit voltage drop.

###### Definition 2.4.

Given nonempty disjoint sets , the effective resistance between and , denoted , is the quantity such that

 1R(A,B) =minx∈RV{x⊤Lx∣∣x↾A=1,x↾B=0}.

If is a single element, we will opt to write instead of the more cumbersome . Similarly we will write or where appropriate.

###### Remark 2.5.

When and are singleton sets, this definition agrees with the standard definition . In general, we can define in a different way. Consider contracting all vertices in to a single Vertex and all vertices to a single vertex . Then is the effective resistance between and in the new graph. This is the definition given in [schild2018schur].

### 2.5 Miscellaneous notation

In our paper does not contain .

## 3 Weighted Hardy inequalities

The following theorem, due to Muckenhoupt [muckenhoupt1972hardy], relates the norm of the “running integral” of a function to its norm.444The original theorem deals more generally with norms and Borel measures — see [muckenhoupt1972hardy]. We refer to this inequality as the Muckenhoupt-Hardy inequality.

###### Theorem 3.1.

[Muckenhoupt 1972] Let , be functions from to . Let be the smallest (possibly infinite) constant such that for all ,

 ∫∞0μ(x)(∫x0f(t)dt)2dx ≤C∫∞0κ(x)f(x)2dx.

Let

 B=supr>0(∫∞rμ(x)dx)(∫r01κ(x)dx).

Then . In particular, is finite if and only if is finite.

Letting for some function with and dividing through by the constant and the term on the left, we can reinterpret the Muckenhoupt-Hardy inequality as a bound on the Dirichlet eigenvalue on the nonnegative line. In the next section we will make this statement formal and give a proof of the rephrased theorem in the finite, discrete case. Our proof will be stated in the language of graph Laplacians but closely follows the structure of [miclo1999example, muckenhoupt1972hardy] and is only included for completeness.

## 4 The Dirichlet problem on path graphs

Throughout this section, let be a vertex and edge weighted connected path graph. Let the vertices be and let the boundary set be . Let and let edge have conductance . Let vertex have mass .

### 4.1 The Hardy quantity and the Dirichlet content

For , let .

Let be a set of vertices disjoint from the boundary. Consider the graph consisting of two vertices and let the boundary set be . Let have mass and let the edge has conductance . Then the Dirichlet eigenvalue of this two node system is given by . We will define the Dirichlet content  to be the minimum such quantity and, for historical reasons, we will define the Hardy quantity to be .

###### Definition 4.1.

Define the Dirichlet content, , to be

 Ψ=minA⊆V{R(S,A)−1μ(A)∣∣∣A≠∅,A∩S=∅}.
###### Definition 4.2.

Define the Hardy quantity to be , i.e.

 H=maxA⊆V{R(S,A)μ(A)|A≠∅,A∩S=∅}.

In a path graph, we may choose to optimize over tail sets. This gives us a second characterization of (and thus ) on path graphs.

###### Lemma 4.3.

Let be the tail set beginning at . Then

 H =max1≤k≤N(k∑i=11κi)μ(Ak).
###### Proof.

Let . Let be the minimum element in . Then and . Note also that on a path graph . ∎

### 4.2 Bounding the Dirichlet eigenvalue

###### Theorem 4.4.

Let be a vertex and edge weighted connected path graph. Let be the Dirichlet eigenvalue and let be the Hardy quantity of . Then,

 14H≤λ≤1H.

We reiterate that the below proof has been known since [muckenhoupt1972hardy] and is included only for completeness.

###### Proof.

We begin by proving the upper bound. Note that if , then . Applying this bound to , we note that the numerator of the Rayleigh quotient becomes an effective resistance term.

 λ =minx{x⊤Lxx⊤Mx∣∣∣x0=0,x≠0} ≤min1≤k≤Nminx{x⊤Lxx⊤Mx∣∣∣x0=0,x↾Ak=1} ≤min1≤k≤N1μ(Ak)minx{x⊤Lx∣∣x0=0,x↾Ak=1} =min1≤k≤NR(S,Ak)−1μ(Ak) =H−1.

On the other hand, let

be an arbitrary nonzero vector with

. Applying Cauchy-Schwarz to the voltage drops,

 N∑i=1μix2i =N∑i=1μi(i∑j=1(xj−xj−1))2 =N∑i=1μi⎛⎜⎝i∑j=1(xj−xj−1)κ1/2jR(x0,xj)1/41κ1/2jR(x0,xj)1/4⎞⎟⎠2 ≤N∑i=1μii∑j=1(xj−xj−1)2κjR(x0,xj)1/2i∑j=11κjR(x0,xj)1/2.

We use the following inequality: for , . Note that , thus the above inequality allows us to bound the second summation as a telescoping series.

 i∑j=11κjR(x0,xj)1/2 ≤2i∑j=1R(x0,xj)1/2−R(x0,xj−1)1/2 =2R(x0,xi)1/2.

Comparing this to the Hardy quantity, we have that . We complete the bound of the original expression by substituting in our estimate of the second summation, switching the order of summation, then applying a second time.

 N∑i=1μix2i ≤2H1/2N∑i=1μiμ1/2(Ai)i∑j=1(xj−xj−1)2κjR(x0,xj)1/2 =2H1/2N∑j=1(xj−xj−1)2κjR(x0,xj)1/2N∑i=jμiμ1/2(Ai) ≤4H1/2N∑j=1(xj−xj−1)2κjR(x0,xj)1/2(N−1∑i=j(μ1/2(Ai)−μ1/2(Ai+1))+μ1/2N) ≤4H1/2N∑j=1(xj−xj−1)2κjR(x0,xj)1/2μ1/2(Aj) ≤4HN∑j=1κj(xj−xj−1)2.

Rearranging, we have that for all with ,

 14H≤∑Ni=1κi(xi−xi−1)2∑Ni=1μix2i.

Then minimizing over such concludes the proof. ∎

The following theorem follows as a corollary.

###### Theorem 4.5.

Let be a vertex and edge weighted connected path graph. Let be the Dirichlet eigenvalue and let be the Dirichlet content of . Then,

 Ψ4≤λ≤Ψ.

## 5 The Dirichlet problem on general graphs

Throughout this section, let be a vertex and edge weighted connected graph. Let the boundary set, , be a proper nonempty subset of .

### 5.1 Bounding the Dirichlet eigenvalue

###### Theorem 5.1.

Let be a vertex and edge weighted connected graph with boundary set , a proper nonempty subset of . Let be the Dirichlet eigenvalue and let be the Hardy quantity of . Then

 14H ≤λ≤1H.

The proof of the upper bound in the graph case is the same as the proof of the upper bound in the path case.

###### Proof of upper bound..

Note,

 λ =minx{x⊤Lxx⊤Mx∣∣∣x↾S=0,x≠0} ≤minA⊆V,x{x⊤Lxx⊤Mx∣∣∣A≠∅,A∩S=∅,x0=0,x↾Ak=1} ≤minA⊆V{R(S,Ak)−1μ(Ak)∣∣∣A≠∅,A∩S=∅} =H−1.

Before proving the lower bound, we state a useful fact.

###### Fact 5.2.

Let be an edge with conductance and let such that . Consider splitting the edge into segments, , with conductance by inserting zero mass vertices. Let be the original graph and let be the new graph. Then . In particular, given let be the linear extension of , then .

To prove the lower bound, we use the above fact to reduce the Dirichlet problem on a graph to the Dirichlet problem on a path.

###### Proof of lower bound..

We construct a new graph from as follows. Let be a solution to the Dirichlet problem corresponding to . Let be the distinct values of . Without loss of generality, suppose . For each edge such that , split into segments such that in the minimum energy extension of , the new vertices on take on all intermediate values (this is possible by Fact 5.2). Let be the minimum energy extension of .

Let , let . Let be the conductance between and . Let . Then applying Theorem 4.5,

 λ(G,S) =λ(G′,S) =minz∈RN{∑Ni=1~κi(zi−zi−1)2∑Ni=1~μiz2i∣∣ ∣∣z0=0,z≠0} ≥14min1≤k≤N1μ′(~Ak)∑ki=11~κi ≥14min1≤k≤N1μ′(~Ak)R′(S,~Ak) ≥14minA′⊆V′{R′(S,A′)−1μ′(A′)∣∣∣A′≠∅,A′∩S=∅}.

Finally, let . Then and . Thus,

 λ(G,S) ≥14minA′⊆V′{R′(S,A′)−1μ′(A′)∣∣∣A′≠∅,A′∩S=∅} ≥14minA⊆V{R(S,A)−1μ(A)∣∣∣A≠∅,A∩S=∅} =14H.

The following theorem follows as a corollary.

###### Theorem 5.3.

Let be a vertex and edge weighted connected graph with boundary set , a proper nonempty subset of . Let be the Dirichlet eigenvalue and let be the Hardy quantity of . Then

 Ψ4≤λ≤Ψ.

## 6 The Neumann problem on general graphs

Throughout this section, let be a vertex and edge weighted connected graph.

### 6.1 The two-sided Hardy quantity and the Neumann content

Let be disjoint nonempty sets. Consider the graph consisting of two vertices where vertex has mass , vertex has mass and the edge has conductance . Then the Neumann eigenvalue of this two node system is given by . We will define the Neumann content  to be the minimum such quantity and, for historical reasons, we will define the two-sided Hardy quantity to be .

###### Definition 6.1.

Define the Neumann content to be

 Ψ2=minA,B⊆V{μ(A)−1+μ(B)−1R(A,B)∣∣∣A,B≠∅,A∩B=∅}.
###### Definition 6.2.

Define the two-sided Hardy quantity to be , i.e.

 H2=maxA,B⊆V{R(S,A)(μ(A)−1+μ(B)−1)−1∣∣A,B≠∅,A∩B=∅}.

We note that the isoperimetric constant of a weighted graph is closely related to . Recall

 Φ(G)=minA⊂V{∑e∈E(A,¯A)κemin(μ(A),μ(¯A))∣∣ ∣∣A,¯A≠∅}.

Noting that and , we can rewrite

 Φ(G)=minA⊂V{max(μ(A)−1,μ(¯A)−1)R(A,¯A)∣∣∣A,¯A≠∅}.

Thus, up to constant factors, can be thought of as the Neumann content where and are required to partition the vertices.

### 6.2 Bounding the Neumann eigenvalue

In this section we show how to extend the bounds on the Dirichlet eigenvalue to the Neumann eigenvalue.

We will bound the Neumann eigenvalue by applying Courant-Fischer to a carefully chosen two-dimensional subspace. In particular, we will split our graph into two parts sharing a common boundary. We will then take our two-dimensional subspace to be the linear span of solutions to the Dirichlet problem on either side of this boundary.

Let such that takes on both positive and negative values. We will write this concisely as . We will “pinch” the graph at the zero level set of to create a new graph : for every edge such that , insert a new vertex such that the minimum energy extension of assigns . Let .

Abusing notation we will also let be the minimum energy extension of to . Let , let and . Similarly define and note that has no edges between and .

We have the following lemma regarding the optimal “pinch.”

###### Lemma 6.3.
 λ2(G) =minf{max(λ(G′,F≤0),λ(G′,F≥0))∣∣±f∉RV≥0}.
###### Proof.

Let denote the quantity on the right hand side.

We begin by showing that . Let take on both positive and negative values. Note that . Let be solutions to the two Dirichlet problems with Dirichlet eigenvalues and respectively. Note that and that , thus . Applying Courant-Fischer to the subspace generated by and ,

 λ2(G) =λ2(G′) ≤maxx∈span(y,z)x⊤L′xx⊤M′x =max(α,β)≠0α2y⊤L′y+β2z⊤L′zα2y⊤M′y+β2z⊤M′z =max(λ(G′,F≤0),λ(G′,F≥0)).

Next we show that . We will exhibit a choice of taking on both positive and negative values such that . This will additionally imply that the minimum is achieved.

Let be a solution to the Neumann problem of . We will pick . Abusing notation, also let be the minimum energy extension of to . Note that . Let be with the and coordinates zeroed out respectively. Note that agrees with on the support of and that agrees with on the support of . Thus . Then,

 λ(G′,F≤0) ≤y⊤L′yy⊤M′y =λ2(G).

Similarly, . ∎

###### Lemma 6.4.

Let and . Then,

 R′(A,F0)+R′(B,F0)≤R′(A,B).
###### Proof.

Let be an assignment of voltages such that , and . Let

Let be an assignment of voltages such that , and . Let .

Note that is zero on and , thus .

Let . Note that is an assignment of voltages with a voltage drop of across and . Thus

 1R(A,B) ≤(αy+(1−α)z)⊤L′(αy+(1−α)z) =α2Y+(1−α)2Z =Y2Z+YZ2(Y+Z)2 =YZY+Z =1R′(A,F0)+R′(F0,B).

Rearranging terms completes the proof. ∎

###### Theorem 6.5.

Let be a vertex and edge weighted connected graph. Let be the Neumann eigenvalue and let be the two-sided Hardy quantity of . Then

 14H2 ≤λ2≤1H2.
###### Proof.

For , , let if for all .

We begin by deriving the upper bound. We express in its “pinch-point” characterization (Lemma 6.3), then apply Theorem 5.1 to each Dirichlet problem.

 λ2(G) =minf{max(λ(G′,F≤0),λ(G′,F≥0))∣∣±f∉RV≥0} ≤minA,B⊆V,f∈RV{max(R′(A,F0)−1μ(A),R′(B,F0)−1μ(B))∣∣ ∣∣±f∉RV≥0,A

Note that given , disjoint and nonempty, we can pick , taking both positive and negative values, such that and . Picking such an , the two terms in the maximum are equal.

 λ2(G) ≤minA,B⊆V{μ(A)−1+μ(B)−1R′(A,B)∣∣∣A,B≠∅,A∩B=∅} =H−12.

Next we derive the lower bound. Again, we express in its “pinch-point” characterization and apply Theorem 5.1 to each Dirichlet problem.

 λ2(G) =minf{max(λ(G′,F≤0),λ(G′,F≥0))∣∣±f∉RV≥0} ≥14minA,B⊆V,f∈RV{max(R′(A,F0