1: \RequirePackage[english]{babel}
2: \documentclass[11pt]{amsart}
3: \usepackage{hyperref}
4: \usepackage{a4wide}
5: \usepackage{amsmath}
6: \usepackage{amsthm}
7: %\usepackage[english]{babel}%moved to first line, fix for amsart/babel bug
8: \newcommand{\RR}{\mathbb{R}}
9: \newcommand{\CC}{\mathbb{C}}
10: \newcommand{\NN}{\mathbb{N}}
11: \newcommand{\ZZ}{\mathbb{Z}}
12: \DeclareMathOperator{\EE}{\mathbb{E}}
13: \newcommand{\PP}{\mathbb{P}}
14: %%
15: \newcommand{\cD}{\mathcal{D}}
16: \newcommand{\cV}{\mathcal{V}}
17: \newcommand{\G}{G}
18: \newcommand{\E}{E}
19:
20: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
21: %
22: % NEWCOMMANDS
23: %
24: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
25:
26: \DeclareMathOperator{\sign}{\mathrm{sign}}
27: \newcommand{\CG}{C^\Gamma}
28: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
29: %
30: % MATH-OPERATORS
31: %
32: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
33: \newcommand{\Tr}{{\mathop{\mathrm{Tr}}}}
34: \DeclareMathOperator{\vol}{vol}
35: \DeclareMathOperator{\supp}{\mathrm{supp}}
36:
37: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
38: %
39: % Theorems
40: %
41: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
42: \newtheorem{thm}{Theorem}
43: \newtheorem{lem}[thm]{Lemma}
44: \theoremstyle{definition}
45: \newtheorem{dfn}[thm]{Definition}
46: \theoremstyle{remark}
47: \newtheorem{rem}[thm]{Remark}
48:
49:
50: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
51:
52: \begin{document}
53: \title[Wegner estimates for Schr\"odinger operators
54: on metric graphs]{The modulus of continuity of Wegner estimates for random Schr\"odinger operators
55: on metric graphs}
56:
57: \author[M.~J.~Gruber]{Michael J.\ Gruber}
58: \address{TU Clausthal\\
59: Institut f\"ur Mathematik\\
60: 38678 Clausthal-Zellerfeld\\
61: Germany}
62:
63: \urladdr{\url{http://www.math.tu-clausthal.de/~mjg/}}
64:
65: \author[I.~Veseli\'c]{Ivan Veseli\'c}
66: \address{TU Chemnitz\\
67: Fakult\"at f\"ur Mathematik\\
68: 09107 Chemnitz\\
69: Germany}
70: \urladdr{\url{http://www.tu-chemnitz.de/mathematik/schroedinger/members.php}}
71:
72: \thanks{The authors were financially supported by the DFG under grant Ve 253/2-2 within the Emmy-Noether-Programme.}
73:
74: \thanks{\copyright 2007 by the authors. Faithful reproduction of this article,
75: in its entirety is permitted for non-commercial purposes. {\today, \jobname.tex}}
76:
77: \keywords{random Schr\"odinger operators, alloy type model, quantum graph, metric graph,
78: integrated density of states, Wegner estimate}
79:
80: \subjclass[2000]{}
81:
82: \begin{abstract}
83: We consider an alloy type potential on an infinite metric graph.
84: We assume a covering condition on the single site potentials.
85: For random Schr\"odingers operator associated with the alloy type potential
86: restricted to finite volume subgraphs
87: we prove a Wegner estimate which reproduces the modulus of continuity of
88: the single site distribution measure. The Wegner constant is independent of the energy.
89: \end{abstract}
90:
91: \maketitle
92:
93:
94: \section{Introduction}
95:
96: We study the distribution of eigenvalues of alloy-type random Schr\"odinger
97: operators on finite metric graphs. More precisely, we prove a
98: Wegner estimate on the average number of eigenvalues in a given
99: energy interval. Our (upper) bound is proportional to the
100: volume of the finite metric graph and reproduces the modulus of
101: continuity of the single site random coupling constants.
102: The volume is the sum of the lengths of all edges. In particular,
103: for equilateral graphs it is the number of edges times the single edge length.
104:
105:
106: In the case that there exists a selfaveraging integrated density of states
107: our result implies estimates on its modulus of continuity.
108:
109:
110: Wegner estimates have been studied for Anderson-type operators on $l^2(\ZZ^d)$
111: and alloy-type operators on $L^2(\RR^d)$ starting with the paper \cite{Wegner-81}.
112: The strategy presented here follows the line of argument of \cite{HundertmarkKNSV-06}.
113: There the same type of Wegner bound as in this note was proven
114: for discrete operators on $l^2(\ZZ^d)$. For alloy-type operators on $L^2(\RR^d)$
115: a similar upper bound was derived. There however, a logarithmic correction term
116: due to the (possible) singularity of the spectral shift function appears.
117:
118: In the recent \cite{CombesHK} a Wegner estimate for
119: random Schr\"odinger operators on $L^2(\RR^d)$ without the logarithmic
120: correction in the energy interval length is given. The approach chosen there
121: does not use the spectral shift function and thus avoids the logarithmic term.
122: For Anderson models on $l^2(\ZZ^d)$ with H\"older continuous
123: distribution of the potential in \cite{Krishna} an alternative proof of the results in
124: \cite{HundertmarkKNSV-06} was presented.
125:
126:
127: The bounds proven in this paper for quantum graph operators have a Wegner constant which is
128: independent of the position of the energy interval. The same situation is encountered for
129: discrete operators, whereas for alloy-type operators on $L^2(\RR^d)$ the Wegner constant
130: grows with the energy. Note that this uniformity in our case is only possible since
131: we assume a covering condition on the single site potentials.
132: For site potentials of small support the uniformity does not
133: hold, cf.~\cite{HelmV}.
134:
135: For more background on the general theory of random Schr\"odinger operators see for instance \cite{CarmonaL-90,PasturF-92}
136: and for more information on Wegner estimates and the integrated density of states \cite{KirschM-07,Veselic-06b}, e.g.
137:
138: The next section contains the results of the paper and the last section the proofs.
139:
140: \section{Model and results}
141:
142: \begin{dfn}
143: Let $V$ and $E$ be countable sets and $\CG$ a map
144: \[
145: \CG \colon E \to V\times V \times [0,\infty), \quad e \mapsto
146: (\iota(e),\tau(e),l_e).
147: \]
148: We call the triple $G=(V,E,\CG)$ a metric graph, elements of $V=V(\G)$ vertices, elements of $E=E(\G)$ edges,
149: $\iota(e)$ the initial vertex of $e$, $\tau(e)$ the terminal vertex of $e$ and
150: $l_e$ the length of $e$. Both $\iota(e)$ and $\tau(e)$ are called endvertices of $e$,
151: or incident to $e$. The two endvertices of an edge are allowed to coincide.
152: The number of edges incident to the vertex $v$ is called the
153: \textit{degree of $v$}. We assume that all vertices have finite degree.
154: \end{dfn}
155: \smallskip
156:
157: Each edge $e$ will be identified with the open interval $\,(0,l_e)$,
158: where the point $0$ corresponds to the vertex $\iota(e)$ and $l_e$
159: to $\tau(e)$.
160: The identification of edges by intervals allows us to define in a
161: canonical way the length of a path between two points in $G$.
162: Taking the infimum over the lengths of
163: paths connecting two given points in $G$, one obtains a distance
164: function $d \colon G\times G\to [0,\infty)$. Since we assumed that
165: each vertex of $G$ has bounded degree, the map $d$ is indeed a
166: metric, cf.~for instance Section 2.2 in \cite{Schubert-06}. Thus
167: we have turned $G$ into a metric space $(G,d)$.
168:
169:
170: For a finite subset $\Lambda \subset E$ we define the subgraph
171: $\G_\Lambda $ by deleting all edges $e \in E\setminus \Lambda$
172: and the arising isolated vertices. We denote the set of vertices of
173: $G_\Lambda$ by $V_\Lambda$, the set of vertices $v\in V_\Lambda$ with $\deg_{G_\Lambda}v<\deg_G v$
174: by $V_\Lambda^\partial$, and its complement $V_\Lambda \setminus V_\Lambda^\partial$
175: by $V_\Lambda^i$. Elements of $V_\Lambda^\partial$
176: are called boundary vertices of $G_\Lambda$
177: and elements of $V_\Lambda^i$ interior vertices of $G_\Lambda$.
178:
179:
180:
181:
182: For any $\Lambda \subset E$ the Hilbert spaces $L_2(\G_\Lambda)$
183: have a natural direct sum representation
184: $L_2(\G_\Lambda)=\oplus_{e\in \Lambda}L_2(0,l_e)$. In particular
185: for $\Lambda = E$ we have $L_2(\G)= \oplus_{e\in \E}L_2(0,l_e)$,
186: and for $\tilde\Lambda\subset \Lambda \subset E$ we have
187: $L_2(\G_{\tilde\Lambda})=\oplus_{e\in
188: {\tilde\Lambda}}L_2(0,l_e)\subset L_2(\G_\Lambda)=\oplus_{e\in
189: \Lambda}L_2(0,l_e)$.
190:
191: For a function $\phi\colon \G \to \CC$ and an edge $e\in E$ we denote by
192: $\phi_e:=\phi|_e$ its restriction to $e$ (which is identified with $(0,l_e)$).
193: We denote by $C(\G)$ the space of continuous, complex-valued functions on the metric space $(\G,d)$.
194: Similarly, $C(\G_\Lambda)$ denotes the space of continuous, complex-valued functions on the metric sub-space $(\G_\Lambda,d)$.
195: For each $v \in V$, any edge $e$ incident to $v$, and function $f\in W^{2,2}(e) \subset C^1(e)\cong C^1(0,l_e)$
196: we define the boundary value $f(v)$ by continuity and the derivatives by
197: \begin{align}
198: \partial_e f(v) &:= \partial_e f(0) := \lim_{\epsilon\searrow 0} \frac{f(\epsilon)-f(0)}{\epsilon} & \text{ if } v = \iota(e)
199: \intertext{and}
200: \partial_e f(v) &:= \partial_e f(l_e) := \lim_{\epsilon\searrow 0} \frac{f(l_e-\epsilon)-f(l_e)}{\epsilon} & \text{ if } v = \tau(e).
201: \end{align}
202: Note that, since $f|_e\in W^{2,2}(e)$ the function $f$ is not only continuously differentiable
203: on the open segment $(0,l_e)$, but also its derivative has
204: well defined limits at both boundaries $0$ and $l_e$.
205: Our sign convention ensures that $\partial_e f(v)$ is the inward normal derivative,
206: and is independent of the orientation of the edge induced by $\iota,\tau$.
207:
208: For any $\Lambda \subset E$ it will be convenient to use the
209: following Sobolev space
210: \begin{align*}
211: W^{2,2}(\Lambda):= \oplus_{e\in \Lambda } W^{2,2}(e) \subset
212: C^1(\Lambda) := \oplus_{e\in \Lambda } C^{1}(e)
213: \\
214: \text{ with the norm }\|\phi\|^2_{W^{2,2}(\Lambda)}:= \sum_{e\in \Lambda} \|\phi_e\|^2_{W^{2,2}(0,l_e)} .
215: \end{align*}
216: Note that this space is defined on the edge set only and does not
217: see the graph structure of $G$.
218:
219: For $f\in W^{2,2}(\Lambda)$ and each vertex $v$ we gather the boundary values
220: $f_e (v)$ over all edges $e$ adjacent to $v$ in a vector
221: $f(v):= \{f_e (v) : e \in E, v \text{ incident to } e\}$.
222: Similarly, we gather the boundary values of $\partial_e f_e (v)$ over all
223: edges $e$ adjacent to $v$ in a vector $\partial f (v)$.
224:
225: Given the boundary values of functions, we can now dicuss the concept
226: of boundary condition. Here we use material from \cite{KostrykinS-99b,Harmer-00} to which we
227: refer for further details and proofs. A single-vertex boundary condition at $v\in V$
228: is a choice of subspace $S_v$ of $\CC^{\deg v}\times\CC^{\deg v}$ with
229: dimension $\deg v$ such that
230: $$
231: \eta((s,s'),(t,t')):=\langle s',t\rangle - \langle s, t'\rangle
232: $$
233: vanishes for all $(s,s'), (t,t')\in S_v$. An $f\in W^{2,2}(\Lambda)$ is said to
234: satisfy the single-vertex boundary condition $S_v$ at $v$ if $(f(v),\partial f(v))$ belongs
235: to $S_v$.
236: A field of single-vertex boundary conditions $S:=\{S_v : v\in V_\Lambda\}$ will be called
237: boundary condition. Given such a field, we obtain a selfadjoint realization $\Delta^\Lambda$
238: of the Laplacian $\Delta$ on $L^2 (E^\Lambda)$ by choosing the domain
239: $$
240: \cD(\Delta^\Lambda) :=\{f\in W^{2,2} (\Lambda) :\forall v : (f(v),\partial f(v))\in S_v\}.
241: $$
242:
243: Particularly relevant boundary conditions are Dirichlet boundary
244: conditions with subspace $S^D$ consisting of all those $(s,s')$ with $s=0$, Neumann conditions with
245: subspace $S^N$ consisting of all those $(s,s')$
246: with $s'=0$, and Kirchhoff (also known as free or standard) boundary conditions $S^K$ consisting
247: of all $(s,s')$ with $s$ having all components equal and $s'$ having the sum over its components equal to $0$.
248:
249: We define the linear operator
250: \[
251: -\Delta^\Lambda \colon \cD(\Delta^\Lambda) \to L^2(\Lambda)
252: \]
253: by the rule
254: \[
255: (-\Delta^\Lambda f)(x) := -\frac{\partial^2 f_e(x)}{\partial x^2 }
256: \]
257: if $x \in G^\Lambda$ is contained in the edge $e$. This way the function
258: $-\Delta^\Lambda f$ is defined on the set $E^\Lambda\subset G^\Lambda$, whose complement $V^\Lambda=G^\Lambda\setminus E^\Lambda$
259: in the metric space $G^\Lambda$ has Hausdorff measure zero.
260:
261: In our application for the Wegner estimate, we will start with boundary conditions defined on the graph $G$ and then restrict
262: to the induced subgraph $G_\Lambda$ for a finite subset $\Lambda\subset E(G)$ as described above.
263: On $V_\Lambda^i$ the boundary conditions will be induced by those on $V(G)$, but on $V_\Lambda^\partial$ there is no canonical choice.
264: We choose to put Dirichlet conditions on $V_\Lambda^\partial$ in order to define the restriction $-\Delta^\Lambda$ of $-\Delta^{E(G)}$ unambiguously.
265:
266:
267: An \emph{alloy-type potential} is a stochastic process $\cV\colon \Omega\times G \to \RR$
268: of the form $\cV_\omega=\sum_{e\in E} \omega_e \, u_e$, with the conditions
269: outlined in the following.
270:
271: The
272: \emph{coupling constants} $\omega_e, e \in E$, are a sequence
273: of bounded random variables which are independent and identically
274: distributed with distribution $\mu$. We call $\mu$ the \emph{single site
275: distribution} since in our model each edge $e$ is the site of a single
276: perturbation controlled by $\omega_e$. Note that $u_e$ is not necessarily supported
277: on $e$, see below. The expectation of the product measure
278: $\PP:=\bigotimes_{e\in E} \mu$ is denoted by $\EE$.
279:
280: The family of single site potentials $u_e, e\in E$, is assumed to fulfill a covering condition and a summability condition:
281: \begin{dfn}
282: The family of single site potentials $u_e, e\in E$, is said to fulfill a \emph{covering condition}
283: with lower bound $\kappa>0$ if,
284: for each finite set of edges $\Lambda$, there is a finite set of edges $\Lambda^u$ such that
285: \[
286: \sum_{e\in \Lambda^u} u_e \ge \kappa
287: \]
288: holds on the graph $G_{\Lambda}$.
289: \end{dfn}
290: For the following definition, recall that for a metric graph $\tilde G$ with finite set of edges $\tilde E$ and length function $e \mapsto l_e$
291: the volume is given by $\vol \tilde G=\sum_{e\in\tilde E} l_e$. In contrast to this, $|\Lambda|$ denotes the number of edges in
292: $\Lambda \subset E$.
293: \begin{dfn}
294: Denote by $\Lambda_e$ the minimal set of edges containing the support of $u_e|_\Lambda$
295: and by $V_e^\partial$ the boundary vertices of the induced subgraph $G_{\Lambda_e}$.
296: (Here for simplicity we suppress the dependence of $V_e^\partial$ on $\Lambda$.)
297: Then, the family of single site potentials $u_e, e\in E$ is called \emph{summable} if there are constants $C_j,j=1,2,3$, such that
298: \begin{equation} \label{eqn:summable}
299: \begin{aligned}
300: \sum_{e\in\Lambda^u} \sum_{v\in V_e^\partial}\deg v &\leq C_1 |\Lambda| \quad\text{(finite degree property)}, \\
301: \sum_{e\in\Lambda^u} \sqrt{\|u_e\|_\infty} \vol G_{\Lambda_e} &\leq C_2 |\Lambda| \quad\text{($L^2$-boundedness)}, \\
302: \sum_{e\in\Lambda^u} |\Lambda_e| &\leq C_3 |\Lambda| \quad\text{(volume growth)}
303: \end{aligned}
304: \end{equation}
305: for each finite set of edges $\Lambda$.
306: \end{dfn}
307: In particular, this holds if $u_e$ is supported on $e$, uniformly bounded above and away from $0$
308: (so that $\Lambda^u=\Lambda$, $\Lambda_e=e$) and there are uniform bounds on vertex degrees and edge lengths.
309: But our definition is much more general. For instance, decreasing edge lengths can compensate for potential growth and vice versa.
310:
311: In the following we consider for a finite subset $\Lambda \subset E$
312: and an alloy-type potential whose family of single site potentials fulfills the covering condition and is summable
313: the restriction $\cV_\omega^\Lambda = \cV_\omega \chi_\Lambda$.
314: On $\cD(\Delta^\Lambda)$ we define a
315: \emph{random Schr\"odinger operator of alloy-type} by $H_\omega^\Lambda =-\Delta^\Lambda +\cV_\omega^\Lambda$.
316: Since the potential is bounded, $H_\omega^\Lambda $ is selfadjoint with boundary conditions described
317: by the field $\{S_v : v\in V_\Lambda\}$ and lower semi-bounded. In the following we will be dealing exclusively with Schr\"odinger operators
318: $H_\omega^\Lambda $ on \emph{finite} edge sets $\Lambda$.
319:
320: \begin{rem}
321: For random Schr\"odinger operators on the whole, infinite graph $G$ one needs to impose
322: more restrictive conditions if one wants to ensure that the Schr\"odinger operator is lower semi-bounded.
323: For instance, Dirichlet, Neumann, standard/free/Kirchhoff conditions and others with ``$L_+=0$'' lead to positive graph Laplacians (see \cite{KostrykinS-06}).
324: \end{rem}
325:
326:
327: Finally, we introduce the \emph{modulus of continuity} of the distribution of the single site distribution, for
328: $\varepsilon>0$, as
329: \begin{equation}
330: \label{definition-s-mu-epsilon}
331: s(\mu,\varepsilon)=\sup\{\mu([\lambda-\varepsilon,\lambda +\varepsilon]) \mid \lambda \in \RR\} .
332: \end{equation}
333:
334: With these definitions, we can formulate
335: \begin{thm} \label{thm:wegner}
336: \label{t-WE} Let $\cV_\omega$ be an alloy-type potential.
337: Then there exists a constant $C_W$ such that for all $\lambda\in\RR$,
338: all finite sets of edges $\Lambda$ and all $\varepsilon \le 1/2$
339: \begin{equation}
340: \label{e-WE}
341: \EE\{\Tr [ \chi_{[\lambda-\varepsilon,\lambda +\varepsilon]}(H_\omega^\Lambda) ]\}
342: \le C_W \ s(\mu,\varepsilon) \, |\Lambda| \, .
343: \end{equation}
344: \end{thm}
345: \begin{rem}[integrated density of states]
346: For the discussion in this remark we assume that all edge lengths $l_e \equiv l$ are equal,
347: that all single site potentials $u_e$ have the same shape, that the boundary condition $S_v$ at the vertex $v\in V$
348: depends only on the degree $\deg (v)$, and that there is a uniform bound
349: $d_+ := \sup_{v \in V} \deg(v)< \infty$ on the vertex degree.
350: In that case $H_\omega^\Lambda$ is
351: selfadjoint and lower semi-bounded even for $\Lambda = E$, and the domain of the operator is independent of $\omega \in \Omega$.
352: For an energy $\lambda\in \RR$ and an exhaustion
353: $\Lambda_1 \subset \Lambda_2 \subset \Lambda_3 \subset\dots$
354: of the edge set $E$ consider the sequence of random variables
355: \[
356: N_\omega^n(\lambda):= \frac{1}{\vol G_{\Lambda_n}}\Tr [ \chi_{(-\infty,\lambda ]}(H_\omega^{\Lambda_n}) ].
357: \]
358:
359: Under certain additional conditions one can show that the sequence of distribution functions $N_\omega^n(\lambda), n \in \NN$
360: converges and that the resulting limiting distribution function $N$ is independent of $\omega \in \Omega$ almost surely.
361: If this is the case Theorem \ref{thm:wegner} implies the following continuity property of $N$:
362: \[
363: \forall \varepsilon \in[0,1/2], \forall \lambda \in \RR \quad : \quad
364: N(\lambda+ \epsilon) -N(\lambda) \le C_W \ s(\mu,\varepsilon)
365: \]
366: Thus the integrated density of states inherits the continuity modulus of the single site measure $\mu$.
367: Let us stress that the Wegner constant $C_W$ is energy independent.
368:
369: For a particular example of a random Schr\"odinger operator on a metric graph with $\ZZ^d$-structure the construction of the
370: integrated density of states has been carried out in \cite{HelmV} following the strategy of \cite{KirschM-82c}.
371: \end{rem}
372:
373: Wegner estimates play an important role in the proof of spectral localization for random Schr\"odinger operators
374: via the so called multiscale analysis.
375: For an alloy type model on a $\ZZ^d$-metric graph localization has been proven in \cite{ExnerHS}
376: using (weaker) Wegner estimates and multiscale analysis.
377:
378: \section{Proofs}
379: Consider a pair of selfadjoint, lower semi-bounded operators $H_1,H_2$. If the spectrum of both $H_1$ and $H_2$ is
380: purely discrete, the spectral shift function (SSF) $\xi(\cdot) = \xi(\cdot,H_2,H_1)$ is defined as the difference
381: of the eigenvalue counting functions, i.e.
382: \[
383: \xi(\lambda):= \Tr[\chi_{(-\infty,\lambda]}(H_2) -\chi_{(-\infty,\lambda]}(H_2)] .
384: \]
385: If we merely assume that the difference $H_2-H_1$ is trace class,
386: then there is still a unique function $\xi$ such that \emph{Krein's trace identity}
387: \begin{equation}
388: \label{e-KTI}
389: \Tr \left [ \rho (H_2)-\rho (H_1) \right ]
390: = \int \rho'(\lambda) \, \xi(\lambda, H_2,H_1) \, d\lambda
391: \end{equation}
392: holds for all $\rho \in C^\infty$ with compactly supported derivative.
393: In the case of operators with discrete spectrum both definitions of the function $\xi$ coincide.
394: We can weaken the assumption on the operator pair further. Assume that there exists
395: a monotone, smooth function $g\colon \RR \to [0,\infty)$ which is bounded on the spectra
396: of $H_1$ and $H_2$ and such that $g(H_2)-g(H_1)$ is trace class. In that case the definition
397: \begin{equation}
398: \label{e-IP}
399: \xi(\lambda, H_2, H_1)
400: :=
401: \sign(g')\ \xi\big(g(\lambda), g(H_2), g(H_1)\big) .
402: \end{equation}
403: makes sense and is independent of the choice of the function $g$.
404:
405: \begin{lem} \label{Potential-Graph}
406: Let ${\tilde G}$ be a finite or infinite metric graph, ${\tilde\Lambda}$ a finite subset of its edges,
407: $-\Delta$ a selfadjoint realization of the Laplacian on $L^2({\tilde G})$ and
408: $W_1,W_2$ two potentials acting as bounded operators on $L^2({\tilde G})$ such that $\supp (W_2-W_1) \subset \tilde G_{\tilde\Lambda}$.
409: Set $H_j=-\Delta +W_j,j=1,2,$ and assume that the SSF $\xi_{H_1,H_2} $ is well defined.
410: Denote the restriction of $H_j$ to $L^2({\tilde\Lambda})$ by $h_j, j=1,2$. Then we have
411: \[
412: \left|\xi_{H_1,H_2} (\lambda) \right|
413: \le \sum_{v \in V_{\tilde\Lambda}^\partial} \deg(v) +\left|\xi_{h_1,h_2} (\lambda) \right|.
414: \]
415: \end{lem}
416: \begin{proof}
417: The basic idea is to decouple the interior of ${\tilde G}_{\tilde\Lambda}$ from the exterior by choosing appropriate
418: boundary conditions on $V_{\tilde\Lambda}^\partial$; this is in the spirit of the statements around equation (9) -- (11) in \cite{HelmV}.
419: So, let $H_j^D$ be $H_j$ but with Dirichlet conditions in $V_{\tilde\Lambda}^\partial$.
420: Then $\left| \xi_{H_j, H_j^D}(\lambda)\right| \leq \sum_{v \in V_{\tilde\Lambda}^\partial} \deg(v)$
421: according to Corollary 11 and Lemma 13 of \cite{GruberLV}.
422: Now, the $H_j^D$ decompose into a direct sum of exterior and interior parts, the former coinciding by assumption,
423: the latter being given by $h_j$. This proves the assertion.
424: \end{proof}
425: We will apply this lemma for a set of edges ${\tilde\Lambda}:= {\Lambda}_e$ containing the support of $u_e$,
426: with finite $\tilde G=G_{\Lambda_e}$ such that the SSF exists automatically.
427:
428: \begin{lem} \label{Potential-Kante}
429: Let $-\Delta$ be a selfadjoint realization of the Laplacian on an arbitrary finite graph $\tilde G$
430: and let $W_1,W_2$ be bounded potentials on $L^2(\tilde G)$.
431: Set $H_j=-\Delta +W_j,j=1,2$.
432: Then
433: \[
434: \left|\xi_{H_1,H_2} (\lambda) \right| \leq \left(\sqrt{\|W_1\|}+\sqrt{\|W_2\|}\right) \frac{\vol \tilde G}\pi +5|E(\tilde G)|
435: \]
436: \end{lem}
437: where $\vol \tilde G= \sum_{e\in E(\tilde G)} l_e$ is the one-dimensional volume of $\tilde G$.
438: \begin{proof}
439: This is an extension of Lemma 14 in \cite{GruberLV}, where this is proved for metric graphs with edges
440: of length $1$. Using the same proof but keeping track of the lengths $l_e$ yields the desired estimate.
441: \end{proof}
442:
443:
444: \begin{proof}[Proof of Theorem~\ref{thm:wegner}]
445: Let $\rho$ be a smooth, monotone switch function
446: $\rho:=\rho_{\lambda,\varepsilon}\colon \RR \to [-1,0]$. \label{p-rho} By a switch function we mean that
447: for a positive $\varepsilon \le 1/2$, $\rho$ has the following properties: $ \rho\equiv -1$ on
448: $(-\infty,\lambda-\varepsilon]$, $\rho\equiv 0$ on $[\lambda+\varepsilon,\infty)$ and $\|\rho'\|_\infty \le
449: 1/\varepsilon$.
450: Then
451: \[
452: \chi_{[\lambda-\varepsilon,\lambda +\varepsilon]} (x) \le \rho(x+2\varepsilon) -\rho(x-2\varepsilon)
453: \]
454: We may assume without loss of generality $\sum_{e\in \Lambda^u} u_e\ge 1$, i.e.\ $\kappa=1$.
455: By the min-max principle for eigenvalues, we conclude
456: \[
457: \Tr [\rho (H_\omega^\Lambda+ \varepsilon)] \le \Tr\Big[\rho (H_\omega^\Lambda+ \varepsilon\sum_{e\in \Lambda^u} u_e) \Big] .
458: \]
459: Let $\Lambda^u$ be as above. Then $\sum_{e\in\Lambda^u}u_e(x)\geq1$ for $x\in G_\Lambda$.
460: $\Lambda^u$ contains $L:=|\Lambda^u|$ edges. We
461: enumerate the edges in $\Lambda^u$ by $ e\colon \{1, \dots, L\} \to \Lambda^u $, $n\mapsto e(n)$, and set
462: \[
463: W_0 \equiv 0, \quad W_n =\sum_{m=1}^{n} u_{e(m)}, \qquad n=1,2,\dots, L
464: \]
465: Thus
466: \begin{align}
467: \chi_{[\lambda-\varepsilon,\lambda +\varepsilon]} (H_\omega^\Lambda)
468: & \le \nonumber
469: \rho(H_\omega^\Lambda+2\varepsilon)-\rho(H_\omega^\Lambda-2\varepsilon)
470: \\ \label{e-project}
471: & \le
472: \rho(H_\omega^\Lambda-2\varepsilon+4\varepsilon W_{L})-\rho(H_\omega^\Lambda-2\varepsilon)
473: \\ \nonumber
474: & =
475: \sum_{n=1}^{L} \rho(H_\omega^\Lambda+2\varepsilon+4\varepsilon W_{n})-
476: \rho(H_\omega^\Lambda-2\varepsilon+4\varepsilon W_{n-1})
477: \end{align}
478: We fix $n \in \{1, \dots, L\}$, define
479: \[
480: \omega^\perp := \{\omega_e^\perp\}_{e \in \Lambda^u}, \qquad
481: \omega_e^\perp :=\begin{cases} 0 \quad &\text{if }e=e(n), \\
482: \omega_e \quad &\text{if } e\neq e(n), \end{cases}
483: \]
484: and set
485: \[
486: \phi_{n}(\eta) = \Tr\bigl[\rho(H_{\omega^\perp}^\Lambda-2\varepsilon +4\varepsilon W_{n-1}+\eta
487: u_{e(n)})\bigr], \quad \eta\in\RR.
488: \]
489: The function $\phi_{n}$ is continuously differentiable, monotone increasing and bounded.
490: By definition of $\phi_{n}$,
491: \[
492: \Tr [ \rho(H_\omega^\Lambda-2\varepsilon+4\varepsilon W_n))
493: -\rho(H_\omega^\Lambda-2\varepsilon+4\varepsilon W_{n-1})]
494: =
495: \phi_{n}(\omega_{e(n)}+4\varepsilon)-\phi_{n}(\omega_{e(n)})]
496: \]
497: since $\phi_{n}(\eta) = \Tr\bigl[\rho(H_{\omega}^\Lambda-2\varepsilon +4\varepsilon W_{n-1}+(\eta
498: -\omega_{e(n)})u_{e(n)})\bigr]$,
499: so that
500: \[
501: \EE_{\omega_{e(n)}} \{ \Tr [ \rho(H_\omega^\Lambda-2\varepsilon+4\varepsilon W_n))
502: -\rho(H_\omega^\Lambda-2\varepsilon+4\varepsilon W_{n-1})] \}
503: =
504: \int [\phi_{n}(\omega_{e(n)}+4\varepsilon)-\phi_{n}(\omega_{e(n)})]\, d\mu(\omega_{e(n)})
505: \]
506: where $\EE_{\omega_{e(n)}}$ denotes the expectation with respect to the random variable $\omega_{e(n)}$ only.
507: Let $\supp(\mu)\subset(a,b)$.
508: Using Lemma 6 in \cite{HundertmarkKNSV-06}
509: we have
510: \begin{align*}
511: \int [\phi_{n}(\omega_{e(n)}+4\varepsilon)-\phi_{n}(\omega_{e(n)})]\, d\mu(\omega_{e(n)})
512: &\le
513: s(\mu,4\varepsilon) [\phi_{n}(b)-\phi_{n}(a)]
514: \end{align*}
515: Denote by $\xi_{\Lambda,n}$ the SSF associated to the pair of operators $H_n(a),H_n(b)$ on $L^2(\Lambda)$
516: where $H_n(\eta)$ is given by $H_n(\eta):=H_\omega^\Lambda-2\varepsilon+4\varepsilon W_{n-1}+ (\eta-\omega_{e(n)}) u_{e(n)} $.
517: Then by the Krein trace identity and the normalization of $\rho$
518: \begin{align*}
519: \phi_{n}(b)-\phi_{n}(a)= \int_a^b \rho' \ \xi_{\Lambda,n} \ d\lambda
520: \le \|\xi_{\Lambda,n}\|_\infty
521: \end{align*}
522: Let $\Lambda_e$, $u_e$, $V_{e}^\partial$ and $G_{\Lambda_e} $ be as in the definition of summable potentials.
523: By $\xi_{\Lambda_{e(n)},n}$ we denote the SSF associated to the pair $H_n(a),H_n(b)$,
524: but now considered as operators on $L^2(\Lambda_{e(n)})$.
525: Apply Lemma \ref{Potential-Graph} to obtain:
526: \begin{align*}
527: \|\xi_{\Lambda,n}\|_\infty
528: \le \sum_{v \in V_{e(n)}^\partial} \deg(v) +\|\xi_{\Lambda_{e(n)},n}\|_\infty
529: \end{align*}
530:
531: Now apply Lemma \ref{Potential-Kante} successively $L$ times to obtain
532: \[
533: \begin{aligned}
534: \EE \{\Tr [\chi_{[\lambda-\varepsilon,\lambda +\varepsilon]} (H_\omega^\Lambda) ]\}
535: & \le s(\mu,4\varepsilon) \sum_{n=1}^{L} \left( \sum_{v \in V_{e(n)}^\partial} \deg(v)
536: + \sqrt{\|u_{e(n)}\|_\infty} \frac{\vol G_{\Lambda_{e(n)}}}\pi + 5 | \Lambda_{e(n)}| \right) \\
537: &\le s(\mu,4\varepsilon) (C_1+C_2/\pi+5C_3) |\Lambda|
538: \end{aligned}
539: \]
540: by the summability condition on the family of single site potentials.
541: \end{proof}
542:
543: \bibliographystyle{habbrv}
544: \bibliography{gv}
545:
546: \end{document}
547:
548:
549:
550: