# TRANSACTIONS OF THE Volume 00, Number 0, Pages pi.math. mhall/conformallaplacian.pdf Volume 00,...

date post

20-Feb-2021Category

## Documents

view

0download

0

Embed Size (px)

### Transcript of TRANSACTIONS OF THE Volume 00, Number 0, Pages pi.math. mhall/conformallaplacian.pdf Volume 00,...

TRANSACTIONS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 00, Number 0, Pages 000–000 S 0002-9947(XX)0000-0

CONFORMAL ENERGY, CONFORMAL LAPLACIAN, AND ENERGY MEASURES ON THE SIERPINSKI GASKET

JONAS AZZAM, MICHAEL A. HALL, AND ROBERT S. STRICHARTZ

ABSTRACT. On the Sierpinski Gasket (SG) and related fractals, we define a notion of conformal energy Eϕ and conformal Laplacian ∆ϕ for a given conformal factor ϕ, based on the corresponding notions in Riemannian geometry in dimension n 6= 2. We derive a differential equation that describes the dependence of the effective resistances of Eϕ on ϕ. We show that the spectrum of ∆ϕ (Dirichlet or Neumann) has similar asymptotics compared to the spectrum of the standard Laplacian, and also has similar spectral gaps (provided the function ϕ does not vary too much). We illustrate these results with numerical approximations. We give a linear extension algorithm to compute the energy measures of harmonic functions (with respect to the standard energy), and as an application we show how to compute the Lp dimensions of these measures for integer values of p ≥ 2. We derive analogous linear extension algorithms for energy measures on related fractals.

1. INTRODUCTION

In Riemannian geometry, a conformal change of metric induces a well-defined change in curvature, energy, and Laplacian. For example, when the dimension n = 2, the energy is unchanged. In all other dimensions, the new energy uniquely determines the conformal factor (it is simply the integral of the gradient squared with respect to a measure that is the old Riemannian measure multiplied by a power of the conformal factor).

In fractal analysis we have, as yet, no analogue of Riemannian metric or curvature, but we do have energy and Laplacian (see [18] for a discussion of these matters). So we will construct an analogue of “conformal change” by considering a change in energy obtained by multiplying all energy measures by a conformal factor ϕ(x). We neeed ϕ to be positive and measurable, and to avoid technicalities we will assume

(1.1) 0 < c1 ≤ ϕ(x) ≤ c2,

although in the future it might be desirable to drop these conditions. We definitely do not want to assume that ϕ is continuous, as our basic examples will be step functions.

Suppose, to be specific, we have a self-similar energy E on a p.c.f, self-similar fractal K ([8],[22]). For the most part we will deal with the standard energy on the Sierpinski Gasket SG. Recall that SG is the unique nonempty compact set in R2 satisfying

(1.2) K = ⋃ i

FiK

where Fi(x) = 12x+ 1 2qi and {q0, q1, q2} = V0 are the vertices of a triangle, (we call V0 the boundary of K). We

approximate K by a sequence of graphs Γm with vertices Vm and edge relation x ∼ m y defined inductively by applying

the m-fold iterations of {Fi} to V0. We define the renormalized graph energies

(1.3) Em(u) = (

5 3

)m ∑ x ∼

m y

(u(x)− u(y))2

for any function defined on Vm (typically the restriction of a function defined on K), and then

2000 Mathematics Subject Classification. Primary 28A80. The first and second authors were supported by the National Science Foundation through the Research Experiences for Undergraduates (REU)

program at Cornell. The third author was supported in part by the National Science Foundation, grant DMS-0140194.

c©1997 American Mathematical Society

1

2 JONAS AZZAM, MICHAEL A. HALL, AND ROBERT S. STRICHARTZ

(1.4) E(u) = lim m→∞

Em(u)

(the limit is always well-defined since {Em(u)} is always monotone increasing). We define the domain of energy domE to be the functions u with E(u) < ∞. It is known that domE ⊆ C(K) so these functions are determined by their restriction to V∗ =

⋃ m Vm, which is dense in K. The resistance metric R(x, y) is defined by

(1.5) R(x, y) = (

inf { E(u) : u(x) = 0, u(y) = 1

})−1 = sup

(u(x)− u(y))2

E(u)

It is known that R(x, y) is finite and defines a metric that is topologically equivalent to the Euclidean metric (it is metrically equivalent to a power of the Euclidean metric).

The pair (E , domE) satisfies the axioms for a Dirichlet form on K: domE modulo constants is a Hilbert space with respect to the inner product E(u, v) obtained from E(u) by polarization, and the Markov property holds. The energy is also local, meaning E(u, v) = 0 if u and v have disjoint support. It is known that the energy E(u) may be written as the integral over K of a positive measure νu, which may be characterized in two ways. First, if C = FwK is a cell (Fw = Fw1 ◦Fw2 ◦ ... ◦Fwm for any word w = (w1, w2, ..., wm)), then νu(C) is defined in the same manner as E(u), except that the sum is restricted to edges lying in C. Second, for any f ∈ domE,

(1.6) ∫ f dνu = E(u, fu)−

1 2 E(f, u2)

Similarly, we can define signed measures νu,v so that E(u, v) = ∫ dνu,v . An interesting observation of Kusuoka

([11], see [2] for another proof) is that these measures are all singular with respect to the standard self-similar measure µ on K (the normalized Hausdorff measure), but they are all absolutely continuous with respect to a single measure ν, called the Kusuoka measure, defined below. See [5] and [6] for more recent singularity results.

A function onK is called harmonic if it minimizes energy among all functions with the same boundary values. The space H0 of harmonic functions is 3-dimensional, as each harmonic function is determined by its boundary values. Indeed, there is a local linear harmonic extension algorithm

(1.7) h ∣∣ FwV0

= Aw(h ∣∣ V0

)

for Aw = Awm ...Aw1 , where Ai are explicit matrices,

(1.8) A0 =

1 0 025 25 15 2 5

1 5

2 5

andA1,A2 are obtained fromA0 by cyclic permutation of rows and columns. Of course constants are harmonic and

have zero energy, so one can find an explicit orthonormal basis {h1, h2} for harmonic functions in the energy inner product. Then we define

(1.9) ν = νh1 + νh2

It is easy to see that ν is independent of the choice of basis. So now, given a conformal factor ϕ, we will define a conformal energy by

(1.10) Eϕ(u) = ∫ ϕ dνu

and

(1.11) Eϕ(u, v) = ∫ ϕ dνu,v

CONFORMAL ENERGY, CONFORMAL LAPLACIAN, AND ENERGY MEASURES ON THE SIERPINSKI GASKET 3

Using the ideas in [9], it is straightforward to verify that Eϕ with domain domE is a Dirichlet form that is strongly local and for which the resistance metric is finite. This is actually a nontrivial fact. In [13] a description of all such Dirichlet forms is given, but it is very indirect, and does not allow for the construction of a large family of forms.

It follows from [9] that the energy Eϕ may by obtained as a monotone increasing limit of graph energies (Eϕ)m on Γm,

(1.12) (Eϕ)m(u) = ∑

x ∼ m

y

cm(x, y)(u(x)− u(y))2

for certain positive conductances cm(x, y). In fact (1.12) is just the restriction of Eϕ to Γm. That means that (1.12) is just the minimum of Eϕ(ũ) over all ũ that agree with u on Vm. The exact determination of the conductances is a difficult problem. One of the main results of this paper is that we can write a kind of differential equation that determines these conductances. Nevertheless, it is easy to find a different family of discrete energies (Ẽϕ)m, given by (1.12) with different conductances c̃m(x, y) so that

(1.13) Eϕ(u) = lim m→∞

(Ẽϕ)m(u)

for u ∈ domE, but not necessarily monotonic, at least in the case that ϕ is continuous or even piecewise continuous. One can take

(1.14) c̃(x, y) = 1 2 (ϕ(x) + ϕ(y))

( 5 3

)m or better still we can replace the discrete average 12 (ϕ(x) + ϕ(y)) with an integral average of ϕ over the m-cell

with vertices x and y, with respect to µ or ν. This is certainly adequate for numerical approximations, and is clearly consistent with (1.10).

To construct a Laplacian we need an energy form and a measure. For the standard Laplacian ∆ on SG we choose the energy (1.4) and the measure µ. We define u ∈ domE and ∆u = f to mean that u ∈ domE, f ∈ C(K), and

(1.15) −E(u, v) = ∫ fv dµ

for all v ∈ dom0E, where dom0E is defined to be the subspace of domE of functions vanishing on the boundary. Similarly we define domŁ2∆ with the only difference that we only require f ∈ L

2. Then −∆ on domL2∆ becomes a positive self-adjoint operator under Dirichlet boundary conditions with compact inverse (or under Neumann boundary conditions with compact resolvant), and so has a discrete spectrum, with eigenfunctions belonging to domD. In fact the nature of the eigenvalues and eigenfunctions are known explicity via the method of spectral decimation of Fukushima and Shima ([4],[15],[16]). For functions in domD there is also a pointwise formula for ∆u as a limit of a difference quotient

(1.16) ∆u(x) = lim m→∞

3 2 5m∆mu(x)

for x ∈ V∗ \ V0, where

(1.17) ∆mu(x) = ∑

y ∼ m

x

(u(y)− u(x))

To define a conformal Laplacian ∆ϕ we want to modify (1.15) by replacing E by Eϕ on the left side, and also changing the measure µ on the right side. In Riemannian geometry, when we make a conformal change of metric we also change the Riemannian measure by multiplying by a power of the conformal factor, so that leads us to expect that the new measure µϕ should be a power of ϕ multiplying µ. But what power? One way to answer this is

Recommended

*View more*