Content-Type: multipart/mixed; boundary="-------------0101221822705" This is a multi-part message in MIME format. ---------------0101221822705 Content-Type: text/plain; name="01-33.keywords" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="01-33.keywords" CLR estimate, weak type estimates, singular values ---------------0101221822705 Content-Type: application/x-tex; name="CLR22-12-00.TEX" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="CLR22-12-00.TEX" % This document is written in LaTeX2e and uses the % amsmath.sty file, the latexsymb.sty, % the amssymb.sty, and the amsfonts.sty %%%%%%%%%%%%%%%%%%%%%%%%%% % % Final version December 22, 2000 % % Dirk Hundertmark %%%%%%%%%%%%%%%%%%%%%%%%% %\documentclass[11pt]{article} %\usepackage{amsmath} %\usepackage{amsfonts} %\usepackage{amsthm} \documentclass[reqno,11pt]{amsart} \usepackage{amssymb} \usepackage{latexsym} %\usepackage[first,bottomafter]{draftcopy} \textwidth 5.1in \usepackage{amsmath} \usepackage{amsfonts} %\usepackage{showkeys} %\usepackage{draftcopy} \newcommand{\nz}{\mathbb{N}} % natural numbers \newcommand{\rz}{\mathbb{R}} % real numbers \newcommand{\rd}{\mathbb{R}^d} % R^d \newcommand{\rzplus}{\mathbb{R_{+}}} % non-negative real numbers \newcommand{\cz}{\mathbb{C}} % complex numbers \newcommand{\gz}{\mathbb{Z}} % Ganze Zahlen \newcommand{\e}{{\mathrm{e}}} % Euler \newcommand{\vareps}{\varepsilon} \newcommand{\rmL}{{\mathrm{L}}} \newcommand{\defeq}{:=} \newcommand{\slim}{\mathop{s\mathrm{-}\lim}} \newcommand{\calB}{{\mathcal{B}}} \newcommand{\cJ}{{\mathcal{J}}} \newcommand{\cK}{{\mathcal{K}}} \newcommand{\calK}{\cK} \newcommand{\cL}{{\mathcal{L}}} \newcommand{\calL}{\cL} \newcommand{\calS}{{\mathcal{S}}} \newcommand{\calF}{{\mathcal{F}}} \newcommand{\calG}{{\mathcal{G}}} \newcommand{\id}{\mathbf{1}} \newcommand{\idG}{\mathbf{1}_{\calG}} \newcommand{\tr}{{\mathop{\mathrm{tr}} \nolimits}} % trace \newcommand{\trG}{\tr_\calG} \newcommand{\Tr}{% {{\mathop{\mathrm{tr}}\nolimits}_{L^2(\rz^d,\calG )}}} \newcommand{\abs}[1]{\left\vert#1\right\vert} \newcommand{\norm}[1]{\left\Vert#1\right\Vert} \newcommand{\traceclass}{{\mathcal{S}^1}} \newcommand{\HS}{{\mathcal{S}^2}} \def\spec{\mathop{\mathrm{spec}}\nolimits} % spectrum \def\const{{\mathrm{const}}} % Konstante %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% TWO COMMANDS FOR MARKING CHANGES -- IN THE MARGIN \def\ch{ \marginpar{change} } \newcommand{\chn}[1]{\marginpar{ch-{#1}}} %% to invoke write: %% \chn{.INITIALS..} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Environments \newtheorem{theorem}{Theorem}[section] %\newtheorem{theorem}{Theorem} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \newtheorem{definition}[theorem]{Definition} %{\theorembodyfont{\rmfamily} { \newtheorem{remark}[theorem]{Remark}} { \newtheorem{remarks}[theorem]{Remarks}} \newtheorem{claim}[theorem]{Claim} % QED Symbol \renewcommand\qedsymbol{\hbox {\hskip 1pt \vrule width 6pt height 6pt depth 1.5pt \hskip 1pt}} \title[On the number of bound states for Schr\"odinger operators]{On the number of bound states for Schr\"odinger operators with operator-valued potentials} \author{Dirk Hundertmark } \thanks{Department of Mathematics 253--37, California Institute of Technology, Pasadena, CA 91125, U.S.A.; E-mail: dirkh@caltech.edu} \thanks{2000 Mathematics subject classification. Primary: 35P15, 47B10; Secondary: 81Q10, 47L20. Key words: CLR estimate, weak type estimates, singular values} \thanks{\copyright 2000 by the author. Reproduction of this article, in its entirety, by any means is permitted for non-commercial purposes.} %%%%%%%%%%%%%%%% \begin{document} %%%%%%%%%%%%%%%% \noindent \begin{abstract} \noindent Cwikel's bound is extended to an operator-valued setting. One application of this result is a semi-classical bound for the number of negative bound states for Schr\"odinger operators with operator-valued potentials. We recover Cwikel's bound for the Lieb--Thirring constant $L_{0,3}$ which is far worse than the best available by Lieb (for scalar potentials). However, it leads to a uniform bound (in the dimension $d\ge 3$) for the quotient $L_{0,d}/ L^{\text{cl}}_{0,d}$, where $L^{\text{cl}}_{0,d}$ is the so-called classical constant. This gives some improvement in large dimensions. \end{abstract} \maketitle %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Introduction}\label{sec:intro} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% The Lieb-Thirring inequalities bound certain moments of the negative eigenvalues of a one-particle Schr\"odinger operator by the corresponding classical phase space moment. More precisely, for ``nice enough" potentials one has \begin{equation}\label{eq:LT1} \tr_{L^2(\rd)} (-\Delta +V)_-^\gamma \le \frac{C_{\gamma,d}}{(2\pi)^d} \iint\limits_{\rz^d \rz^d}d\xi dx\,(\xi^2+V(x))_-^\gamma . \end{equation} Here and in the following, $(x)_- =\tfrac{1}{2}(\vert x\vert-x)$ is the negative part of a real number or a self-adjoint operator. Doing the $\xi$ integration explicitly with the help of scaling the above inequality is equivalent to its more often used form \begin{equation}\label{eq:LT2} \tr_{L^2(\rd)} (-\Delta +V)_-^\gamma \le L_{\gamma,d} \int_{\rz^d}dx\, V(x)_-^{\gamma+d/2} , \end{equation} where the Lieb-Thirring constant $L_{\gamma,d}$ is given by $L_{\gamma,d}= C_{\gamma,d} L_{\gamma,d}^{\text{cl}} $ with the classical Lieb-Thirring constant \begin{equation}\label{eq:LTcl} L_{\gamma,d}^{\text{cl}} = \frac{1}{(2\pi)^d}\int_{\rz^d}dp (1-p^2)_+^\gamma . \end{equation} This integral is, of course, explicitly given by a quotient of Gamma functions, but we will have no need for this. The Lieb-Thirring inequalities are valid as soon as the potential $V$ is in $L^{\gamma+d/2}(\rd)$. These inequalities are important tools in the spectral theory of Schr\"odinger operators and they are known to hold if and only if $\gamma\ge \tfrac{1}{2}$ if $d=1$, $\gamma >0 $ if $d=2$, and $\gamma \ge 0$ if $d\ge 3$. The bound for the critical case $\gamma =0$, that is, the bound for the number of negative eigenvalues of a Schr\"odinger operator in three or more dimensions is the celebrated Cwikel-Lieb-Rozenblum bound \cite{Cwikel,Lieb,Rozenblum}. Later, different proofs for this were given by Conlon and Li and Yau \cite{Conlon, LY}. The remaining case $\gamma = \tfrac{1}{2}$ in $d=1$ was settled in \cite{Weidl}. The well-known Weyl asymptotic formula $$ \lim_{\lambda\to\infty}\tr (-\Delta+\lambda V)_-^\gamma = L_{\gamma,d}^{\text{cl}}\int dx\, V(x)_-^{\gamma+d/2} $$ immediately gives the lower bound $C_{\gamma,d}\ge 1$. There are certain refined lower bounds \cite{LT2,HR} for small values of $\gamma$. In particular, one always has $C_{\gamma,d}>1$ for $\gamma<1$; see \cite{HR}. In one dimension this even happens for $\gamma<3/2$, and in two dimensions, one always has $C_{1,2} > 1$ \cite{LT2}. Depending on the dimension there are certain conjectures for the optimal value of the constants in these inequalities \cite{LT,LT2}. One part of the conjectures on the Lieb-Thirring constants is that, indeed, $C_{\gamma,d} =1$ for $d\ge 3$ and moments $\gamma \ge 1$. For the physically most important case $\gamma =1$, $d=3$ this would imply, via a duality argument, that the kinetic energy of fermions is bounded below by the Thomas-Fermi ansatz for the kinetic energy, which in turn has certain consequences for the energy of large quantum Coulomb systems \cite{Lieb,LT}. Laptev and Weidl \cite{LW} realized that a, at first glance, purely technical extension of the Lieb-Thirring inequality from scalar to operator-valued potentials already suggested in \cite{Laptev} is a key in proving at least a part of the Lieb-Thirring conjecture. It allowed them to show that $C_{\gamma,d} = 1$ for all $d\in\nz$ as long as $\gamma\ge 3/2$. To prove this they considered Schr\"odinger operators of the form $-\Delta\otimes \idG +V$ on the Hilbert space $L^2(\rd,\calG)$ where V now is an operator-valued potential with values $V(x)$ in the bounded self-adjoint operators on the auxiliary Hilbert space $\calG$. In this case the Lieb-Thirring inequalities (\ref{eq:LT1}) and (\ref{eq:LT2}) are modified to \begin{equation}\label{eq:LToper1} \Tr (-\Delta\otimes\idG+V)_-^\gamma \le \frac{C_{\gamma,d}}{(2\pi)^d} \iint\limits_{\rz^d \rz^d}dpdx\, \trG(p^2+V(x))_-^\gamma, \end{equation} or, again doing the $\xi$ integral explicitly with the help of the spectral theorem and scaling \begin{equation}\label{eq:LToper2} \Tr (-\Delta\otimes\idG+V)_-^\gamma \le L_{\gamma,d} \int_{\rz^d}dx\, \trG(V(x)_-^{\gamma+d/2}). \end{equation} Here we abused the notation slightly in using the same symbol for the constants as in the scalar case. But in the following, we will only consider the operator-valued case anyway. Laptev and Weidl realized that this extension of the Lieb-Thirring inequality gives rise to the possibility of an inductive proof for $C_{3/2,d}=1$ as long as one has the a priori information $C_{3/2,1}=1$ for operator-valued potentials. This idea together with ideas in \cite{HLT} was then later used in \cite{HLW} to prove improved bounds on $C_{\gamma,d}$ in the range $1/2 \le \gamma \le 3/2$; in particular, it was shown that $C_{1,d}\le 2$ uniformly in $d\in\nz$. Unlike the scalar case, however, the range of parameters $\gamma$ and $d$ for which (\ref{eq:LToper1}) or equivalently (\ref{eq:LToper2}) holds is not known. The results in \cite{HLW} only show that these inequalities are true for $\gamma\ge 1/2$ and all $d\in \nz$. This shortcoming has to do with the way the Lieb-Thirring estimates are proven for operator-valued potentials: First, the estimate is shown to hold in one dimension. Then a suitable induction proof, using the one-dimensional result, is set up to prove the full result in all dimensions. This turns out to give good estimates for the coefficients $C_{\gamma,d}$ in the Lieb-Thirring inequality, for example, they are independend of the dimension. However, moments below $1/2$ cannot be addressed with this method, since the a priori estimate fails already for scalar potentials. This led Ari Laptev \cite{Laptevprivate}, see also \cite{LW2}, to ask the question whether, in particular, the Cwikel-Lieb-Rozenblum estimate holds for Schr\"odinger operators with operator-valued potentials. In this note we answer his question affirmatively, that is, the Lieb-Thirring inequalities for operator-valued potentials are shown to hold also for $\gamma=0$ as long as $d\ge 3$ and then, by a monotonicity argument also for all $\gamma\ge 0$. More precisely, we want to show that Cwikel's proof of the Cwikel-Lieb-Rozenblum bound can be adapted to the operator-valued setting. However, the bound for $C_{0,d}$ is far from being optimal since we use Cwikel's approach. But, nevertheless, reasoning similar to Laptev and Weidl, any a priori bound on $C_{0,3}$ implies the bound $C_{0,d}\le C_{0,3}$ for $d\ge 3$, thus giving a \emph{uniform} bound in the dimension, whereas the best available bound in the scalar case due to Lieb \cite{Lieb} grows like $\sqrt{\pi d}$, see \cite{LT2}. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Statement of the results}\label{sec:results} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Let $\calG$ be a (separable) Hilbert space with norm $\norm{.}_\calG$, scalar product $\langle.,.\rangle_\calG$, and let $\idG$ be the identity operator on $\calG$. We follow the convention that scalar products are linear in the second component. Furthermore, $\calB(\calG)$ is the Banach space of bounded operators equipped with the operator norm $\norm{.}_{\calB(\calG)}$ and $\calK(\calG)$ the (separable) ideal of the compact operators on $\calG$. For a compact operator $A\in\calK(\calG)$, the singular values $\mu_n(A)$, $n\in\nz$ are the eigenvalues of $\vert A\vert\defeq (A^*A)^{1/2}$ arranged in decreasing order counting multiplicity. $A^*$ is the adjoint of $A$. $\calS^q(\calG)$ denotes the ideal of compact operators $A\in\calK(\calG)$ whose singular values are $q$-summable, that is, $\sum_n \mu_n(A)^q <\infty$. In particular, $\calS^1(\calG)$ and $\calS^2(\calG)$ are the trace class and Hilbert-Schmidt operators on $\calG$. We will often write $\calB$, $\calK$, and $\calS^q$ if there is no ambiguity. Of course, $A\in\calS^q$ if and only if $\trG(\vert A\vert^q) = \trG((A^*A)^{q/2})<\infty$, where $\trG$ is the trace on $\calG$. The Hilbert space $L^2(\rd,\calG)$ is the space of all measurable functions $\phi\!: \rd\to\calG$ such that $$ \norm{\psi}_{L^2(\rd,\calG)}^2\defeq \int_{\rd}dx\,\norm{\psi(x)}_{\calG}^2 <\infty $$ and the Sobolev space $H^1(\rd,\calG)$ consists of all functions $\psi\in L^2(\rd,\calG)$ with finite norm $$ \norm{\psi}_{H^1(\rd,\calG)}^2 \defeq \sum_{l=1}^d \norm{\partial_l \psi}_{L^2(\rd,\calG)}^2 + \norm{\psi}_{L^2(\rd,\calG)}^2 . $$ As in the scalar case, the quadratic form $$ h_0(\psi,\psi) \defeq \sum_{l=1}^d \norm{\partial_l \psi}_{L^2(\rd,\calG)}^2 $$ is closed in $L^2(\rd,\calG)$ on the domain $H^1(\rd,\calG)$. Naturally, this form corresponds to the Laplacian $-\Delta\otimes\idG$ on $L^2(\rd,\calG)$. $L^{q}(\rd,\calB(\calG))$ is the space of operator-valued functions $f\!:\rd\to\calB(\calG)$ with finite norm $$ \norm{f}_{q}^q = \norm{f}_{L^{q}(\rd,\calB(\calG))}^q \defeq \int_{\rd}dx\, \norm{f(x)}_{\calB(\calG)}^q $$ and $ L^q(\rd,\calS^r(\calG))$ the space of operator-valued functions $f$ whose norm $$ \norm{f}_{q,r}^q = \norm{f}_{L^{q}(\rd,\calS^r(\calG))}^q \defeq \int_{\rd}dx\, \trG(\vert f(x)\vert^r)^{q/r} $$ is finite. A potential is a function $V\in L^{q}(\rd,\calB(\calG))$ such that $V(x)$ is a symmetric operator for almost every $x\in\rd$. If \begin{equation}\label{eq:formbound} q\ge 1 \text{ for } d=1,\quad q>1 \text{ for } d=2\text{,} \quad\text{and } q\ge d/2 \text{ for } d\ge 3 \end{equation} one sees, using Sobolev embedding theorems as in the scalar case, that the real-valued quadratic form $$ v[\psi,\psi] \defeq \int_{\rd}dx\, \langle \psi(x), V(x)\psi(x)\rangle_{\calG} $$ is infinitesimally form-bounded with respect to $h_0$. Hence the form sum $$ h[\psi,\psi]\defeq h_0[\psi,\psi] + v[\psi,\psi] $$ is closed and semi-bounded from below on $H^1(\rd,\calG)$ and thus generates the self-adjoint operator $$ H= -\Delta\otimes\idG + V $$ on $L^2(\rd,\calG)$ by the KLMN theorem \cite{RSII}. It is easy to see that any potential $V\in L^{q}(\rd,\calB(\calG))$ satisfying (\ref{eq:formbound}) for which $V(x)\in\calK(\calG)$ for almost every $x\in\rd$ is relatively form compact with respect to $h_0$. Hence by Weyl's theorem for such potentials, the negative eigenvalues $E_0 \le E_1 \le E_3 \le \cdots\le 0$ are at most a countable set with accumulation point zero and their eigenspaces are finite-dimensional. In particular, this is the case for potentials $V\in L^{q}(\rd,\calS^r(\calG))$. Our first result is a generalized version of a basic observation of Laptev and Weidl: The two versions (\ref{eq:LToper1}) and (\ref{eq:LToper2}) of the Lieb-Thirring inequality give rise to two different monotonicity properties of $C_{\gamma,d}$ in $d$. \begin{theorem}[Sub-multiplicativity of $\mathbf{C_{\gamma,d}}$] \label{thm:subm} If, for dimensions $n$ and $d-n$, the Lieb-Thirring inequality holds for operator-valued potentials then it also holds in dimension $d$. Moreover, \begin{eqnarray} C_{\gamma,d} &\le& C_{\gamma,n} C_{\gamma,d-n} \qquad\text{ and} \label{eq:subm1}\\ C_{\gamma,d} &\le& C_{\gamma,n} C_{\gamma+n/2,d-n} . \label{eq:subm2} \end{eqnarray} \end{theorem} \begin{remarks}\rm i) In the scalar case Aizenman and Lieb \cite{AL} showed that the map $\gamma \to C_{\gamma,d}= L_{\gamma,d}/L^{\text{cl}}_{\gamma,d}$ is decreasing. This monotonicity holds also in the general case, so, in fact, (\ref{eq:subm2}) implies (\ref{eq:subm1}). The monotonicity in $\gamma$ is most easily seen in the phase space picture: By scaling one has, for $\gamma > \gamma_0\ge 0$, $$ \int_0^\infty (s+t)_-^{\gamma_0}\, t^{\gamma-\gamma_0 -1} dt = (s)_-^\gamma\, B(\gamma - \gamma_0,\gamma_0+1) , $$ where $B(\alpha,\beta)= \int_0^1 t^{\alpha-1}(1-t)^{\beta-1} dt$ is the Beta function. In other words, for each choice of $\gamma\! >\! \gamma_0\!\ge\! 0$ there exists a positive measure $\mu$ on $\rz_+$ with $(s)_-^\gamma = \int_{\rz_+}(s+t)_-^{\gamma_0}\, d\mu(t)$. Using this, the functional calculus, and the Fubini-Tonelli theorem, we immediately get \begin{equation*} \begin{split} \Tr(\Delta +V)_-^\gamma & = \int_0^\infty \Tr(\Delta +V +t)_-^{\gamma_0}\, d\mu(t) \\ &\le \frac{C_{\gamma_0,d}}{(2\pi)^d} \int_0^\infty d\mu(t) \iint d\xi dx\, \trG (\xi^2 +V(x) +t)_-^{\gamma_0} \\ &= \frac{C_{\gamma_0,d}}{(2\pi)^d} \iint d\xi dx \int_0^\infty d\mu(t) \, \trG (\xi^2 +V(x) +t)_-^{\gamma_0} \\ &= \frac{C_{\gamma_0,d}}{(2\pi)^d} \iint d\xi dx \, \trG (\xi^2 +V(x))_-^{\gamma} .\\[0.3em] \end{split} \end{equation*} ii) Theorem \ref{thm:subm} is a slight extension of a very nice observation of Laptev and Weidl \cite{Laptev,LW}. They used it to show $C_{\gamma,d}=1$ as long as $\gamma\ge 3/2$. Basically this follows immediately by induction and the above monotonicity from (\ref{eq:subm1}) for $n=1$ once one knows that $C_{3/2,1}=1$. The beauty of this observation is that this bound is well-known in the scalar case \cite{LT2} and Laptev and Weidl gave a proof for it in the general case. See also \cite{BL} for an elegant alternative proof which avoids the proof of Buslaev-Fadeev-Zhakarov type sum rules for matrix-valued potentials.\\[0.3em] iii) Using $C_{\gamma,d}=1$ for $\gamma\ge 3/2$ and (\ref{eq:subm2}), we get the bound $$ C_{\gamma,d} \le C_{\gamma, 3} $$ in $d\ge 3$ for all $\gamma\ge 0$. In particular, this implies a uniform bound (in $d$) for the constant in the Cwikel-Lieb-Rozenblum bound as soon as such an estimate is established in dimension three for operator-valued potentials. Below we will recover Cwikel's bound $C_{0,3}\le 3^4=81$, see Corollary \ref{cor:boundstates}. It is, already for scalar potentials, known, that $C_{0,3}\ge 8/\sqrt{3} > 4.6188$, \cite{GGMT} \cite[eq.\ (4.24)]{LT2} (see also the discussion in \cite[page 96--97]{Simon3}); in fact, it is conjectured to be the correct value \cite{GGMT,LT2,Simon2}. In the \emph{scalar case} Lieb's proof \cite{Lieb} of the CLR-bound gives by far the best estimate, $C_{0,3}^{\mathrm{scalar}}\le 6.87$. However, Lieb's estimate grows like $\sqrt{\pi d}$ for large dimensions \cite[eq. (5.5)]{LT2}. While we get a quite large bound on $C_{0,3}$ this at least furnishes the uniform bound $C_{0,d}\le 81$ for all $d\ge 3$. It would be nice to extend Lieb's or even Conlon's proof \cite{Conlon} of the CLR-bound to operator-valued potentials. \end{remarks} To state our second result, Cwikel's bound in the operator-valued case, we need some more notation: $L^q_w(\rd,\calB(\calG))$, the analog of the weak $L^q$-space $L^q_w(\rd)$, is given by all operator-valued functions $g\!:\rd\to\calB(\calG)$ for which $$ \norm{g}^*_{q,w} = \norm{g}^*_{L^q_w(\rd,\calB(\calG))} \defeq \sup_{t>0} \big(t \, \vert\{ \norm{g(\cdot)}_{\calB(\calG)}>t \}\vert^{1/q}\big) <\infty. $$ Here $\vert B\vert$ is the $d$-dimensional Lebesgue measure of a Borel set $B\subset\rd$. Note that $\norm{\cdot}^*_{q,w}$ is not a norm since it fails to obey the triangle inequality already for scalar $g$. But, as in the scalar case, one can give a norm on $L_w^q(\rd;\calB(\calG))$ which is equivalent to $\norm{\cdot}^*_{L^q(\rd;\calB(\calG))}$. However, we will not need this. With $p$ we abbreviate the operator $-i\nabla$ and similarly to the scalar case we define the operator $f(x)g(p)$ to be $$ \psi\to f(x)g(p)\psi (x)= f(x) \frac{1}{(2\pi)^{d/2}} \int \e^{ix\zeta} g(\zeta) \hat{\psi}(\zeta)\,d\zeta, $$ that is, $f(x)g(p)= M_{f}\calF^{-1}M_{g}\calF$ with $M_f$, $M_g$ the ``multiplication" operators by $f(x)$ and $g(\xi)$ and $\calF$ the Fourier transform. A priori, $f(x)g(p)$ is well-defined only for simple functions, but it will turn out to be a compact operator for rather general ``functions" $f$ and $g$. The extension of Cwikel's bound to the operator-valued case is \begin{theorem}[Cwikel's bound, operator-valued case] \label{thm:Cwikel} Let $f$ and $g$ be operator-valued functions on an auxiliary Hilbert space $\calG$. Assume that $f\in L^q(\rz^d\!,\calS^q(\calG)\!)$ and $g\in L_w^q(\rz^d,\calB(\calG)\!)$ for some $q>2$. Then $f(x)g(p)$ is a compact operator on $L^2(\rd,\calG)$. In fact, it is in the weak operator ideal $\calS^q_w(L^2(\rd,\calG))$ and, moreover, \begin{equation}\label{eq:cwikelbound} \norm{f(x)g(p)}^*_{q,w} := \sup_{n\ge 1}\, n^{1/q} \mu_n\big(f(x)g(p)\!\big) \le K_q \norm{f}_{q,q}\norm{g}_{q,w}^* \end{equation} where the constant $K_q$ is given by $$ K_q = (2\pi)^{-d/q} \frac{q}{2} \left(\frac{8}{q-2}\right)^{1-2/q} \left( 1+ \frac{2}{q-2}\right)^{1/q} . $$ \end{theorem} As in the scalar case Theorem \ref{thm:Cwikel} gives a bound for the number of negative eigenvalues of Schr\"odinger operators with operator-valued potentials. \begin{corollary}\label{cor:boundstates} Let $\calG$ be some auxiliary Hilbert space and $V$ a potential in $L^{d/2}(\rd,\calS^{d/2}(\calG))$. Then the operator $-\Delta\otimes\idG +V$ has a finite number $N$ of negative eigenvalues. Furthermore, we have the bound $$ N\le L_{0,d} \int_\rd \trG (V(x)_-^{d/2})\, dx $$ with $$ L_{0,d} \le (2\pi\, K_d)^d L^{\mathrm{cl}}_{0,d}, $$ that is, $C_{0,d}\le (2\pi\, K_d)^d$. \end{corollary} \begin{proof} For completeness we explicitly derive the estimate for the number of negative eigenvalues of $-\Delta\otimes\idG +V$ from Theorem \ref{thm:Cwikel}. Replacing $V$ with $-(V)_-$ if necessary and using the min--max principle, we can assume $V$ to be non-positive. Let $N$ be the number of negative eigenvalues of $-\Delta\otimes +V$ and put $Y\defeq \vert V\vert^{1/2}\,(\vert p\vert^{-1}\!\otimes\!\idG)$. By the Birman-Schwinger principle \cite{Birman,Schwinger,BS,Simon,RSIV} one has $$ 1 \le \mu_N(Y). $$ But $\xi\to \vert \xi\vert^{-1}\!\otimes\!\idG$ has weak $L^d(\rd,\calB(\calG))$-norm $\tau_d^{1/d}$, $\tau_d$ being the volume of the unit ball in $\rd$. With Theorem \ref{thm:Cwikel} we arrive at $$ 1\le K_d \tau_d^{1/d} \Vert\vert V\vert^{1/2}\Vert_{d,d} N^{-1/d}, $$ that is, $$ N\le K_d^d \tau_d \Vert\vert V\vert^{1/2}\Vert_{d,d}^d = (2\pi K_d)^d L^{\text{cl}}_{0,d} \int \trG(\vert V(x)\vert^{d/2})\, dx, $$ since $L^{\text{cl}}_{0,d} = \tau_d/(2\pi)^d$. \end{proof} \begin{remark}\rm Corollary \ref{cor:boundstates} gives the a priori bound $C_{0,d}\le (2\pi\, K_d)^d$ for $d\ge 3$. Using Theorem \ref{thm:subm} and the fact that $C_{\gamma,d}=1$ if $\gamma\ge 3/2$, \cite{LW}, we know that $C_{0,d}\le \min_{n=3,\ldots,d} C_{0,n}$. Since the a priori bound given in Corollary \ref{cor:boundstates} increases rather fast in the dimension, the best we can conclude is $C_{0,d}\le (2\pi\, K_3)^3=3^4=81$. \end{remark} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Proof of the sub-multiplicativity of the Lieb-Thirring constants} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% We proceed very similarly to \cite{LW}, but freeze the first $n =(x_{n+1},\ldots,x_d)$ and $\xi_<$, $\xi_>$ similarly defined. Put $$ W(x_<)\defeq (-\Delta_> +V(x_<,.))_{-}, $$ where $\Delta_>$ is the Laplacian in the $x_>$ variables. Clearly, by assumption on $V$, $W$ is a non-negative compact operator on $L^2(\rz^{d-n},\calG)$ for almost all $x_<\in\rz^{n}$ and, moreover, \begin{equation}\label{eq:freeze} \begin{split} \Tr (-\Delta +V)_-^\gamma &\le\tr_{L^2(\rz^n,L^2(\rz^{d-n},\calG))} (-\Delta_< -W)_-^\gamma\\ &\le \frac{C_{\gamma,n}}{(2\pi)^n} \iint\limits_{\rz^n \rz^n}d\xi_+V(x_<,.))_-^\gamma \\ \le \frac{C_{\gamma,d-n}}{(2\pi)^{d-n}} \iint\limits_{\rz^{d-n} \rz^{d-n}}d\xi_>dx_>\, \trG (\xi_<^2 +\xi_>^2+V(x_<,x_>))_-^\gamma. \end{multline*} This together with (\ref{eq:freeze}) and the Fubini-Tonelli theorem shows (\ref{eq:subm1}).\\ For the other inequality we use the more usual form (\ref{eq:LToper2}) of the Lieb-Thirring inequality. Again, freezing the first $n$ coordinates and proceeding as before, we immediately get \begin{equation}\label{eq:subm3} L_{\gamma,d}\le L_{\gamma,n} L_{\gamma+n/2,d-n}, \end{equation} where $L_{\gamma+n/2,d-n}$ enters now because in the first application of the Lieb-Thirring inequality (\ref{eq:LToper2}) the exponent is raised from $\gamma$ to $\gamma+n/2$. Using the definition (\ref{eq:LTcl}) for the classical Lieb-Thirring constant together with the Fubini-Tonelli theorem and scaling, one easily sees \begin{equation*} \begin{split} L_{\gamma,d}^{\text{cl}} &= \int_{\rd}\! dp\, (|p|^2-1)_{-}^\gamma \\ &= \int_{\rz^n}\! dp_{<}\, (\vert p_{<}\vert^2-1)_{-}^\gamma \int_{\rz^{(d-n)}}\! dp_{>}\, (\vert p_{>}\vert^2-1)_{-}^{\gamma+n/2} \\ &= L_{\gamma,n}^{\text{cl}} L_{\gamma+n/2,d-n}^{\text{cl}} . \end{split} \end{equation*} This together with (\ref{eq:subm3}) proves (\ref{eq:subm2}) and thus Theorem \ref{thm:subm}. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \section{Proof of Cwikel's bound}\label{sec:cwikel} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% The proof of Theorem \ref{thm:Cwikel} follows closely Cwikel's original proof. We first need a criterion for $f(x)g(p)$ to be a Hilbert-Schmidt operator. \begin{lemma}\label{lem:HS} Let $f\in L^2(\rd,\calS^2(\calG))$ and assume $g$ obeys $\norm{g(.)}_{\calB(\calG)}\in L^2(\rd)$. Then the operator $f(x)g(p)$ is Hilbert-Schmidt and we have the estimate \begin{equation*} \begin{split} \norm{f(x)g(p)}_{HS}^2 &= (2\pi)^{-d} \iint \trG [g^*(\xi)f(x)^*f(x)g(\xi)]\, dx d\xi\\ &\le (2\pi)^{-d} \int \trG [\vert f(x)\vert^2]\,dx \int \norm{g(\xi)}_{\calG}^2\, d\xi. \end{split} \end{equation*} \end{lemma} \begin{proof} In the scalar case this is well-known and is usually shown by noting that in this case $f(x)g(p)$ is a convolution operator. Another proof is by changing the basis: Let $\calF$ be the Fourier transform on $L^2(\rd,\calG)$, that is, $$ \calF u (\xi) \defeq (2\pi)^{-d/2} \int_{\rd} \e^{-i\xi\cdot x} u(x)\,dx . $$ Then the Hilbert-Schmidt norms of $f(x)g(p)$ and $M_f\calF^{-1}M_g$ are equal. The operator $M_f\calF^{-1}M_g$ has ``kernel" $(2\pi)^{-d/2}\e^{ix\cdot\xi}f(x)g(\xi)$ and thus by \cite[Theorem VI.23]{RSI} or \cite[Section III.9]{GK} \begin{equation*} \begin{split} \norm{f(x)g(p)}_{HS}^2 &= \norm{f(x)\calF^{-1}g(\xi)}_{HS}^2 \\ &= (2\pi)^{-d} \iint \trG(g(\xi)^*f(x)^*f(x)g(\xi))\,dx d\xi \\ &= (2\pi)^{-d} \iint \trG(\vert f(x)\vert^2 \vert g(\xi)^*\vert^2)\,dx d\xi \\ &\le (2\pi)^{-d} \iint \trG(\vert f(x)\vert^2) \norm{g(\xi)}_{\calG}^2)\,dx d\xi \\ &= (2\pi)^{-d} \int \trG (\vert f(x)\vert^2)\,dx \int \norm{g(\xi)}_{\calG}^2)\,d\xi . \end{split} \end{equation*} \end{proof} The first step rests on splitting the operator $f(x)g(p)$ (which is a priori only defined on simple functions) into manageable pieces. Fix $t\!>\!0$, $r\!>\!1$ and assume that $f$ and $g$ are non-negative, in particular, self-adjoint, operator-valued functions. For a Borel subset $B$ of $\rz$ let $\chi_{B}(f(x))$ and $\chi_{B}(g(\xi))$ be the spectral projection operators of $f(x)$ and $g(\xi)$, respectively. By the functional calculus we have \begin{equation}\label{eq:decomposition} \begin{split} f(x)&= \sum_{l\in\gz} f(x)\chi_{t(r^{l-1}, r^l]}(f(x)) = \sum_{l\in\gz} f_l(x)\\ g(\xi)&= \sum_{l\in\gz} g(\xi)\chi_{(r^{l-1}, r^{l}]}(g(\xi)) = \sum_{l\in\gz} g_l(\xi) , \end{split} \end{equation} where $f_l$ (resp., $g_m$) are mutually orthogonal operators. We use this decomposition of $f$ and $g$ to split the operator $f(x)g(p)$ into \begin{equation} f(X)g(P) = B_t +H_t \end{equation} with $B_t\defeq \sum_{l+m \le 1}f_l(x)g_m(p)$, $H_t\defeq \sum_{l+m > 1}f_l(x)g_m(p)$. Note that this decomposition of $f(x)g(p)$ is slightly different from the one used by Cwikel. We have \begin{lemma} \label{lem:bounds} Let $f$ and $g$ be non-negative operator-valued functions. If $q>2$ and $f\!\in\! L^q(\rd,\calS^q(\calG))$, $g\!\in\! L^q_{w}(\rd,\calB(\calG))$ with $\norm{f}_{q,q}=1$ and $\norm{g}_{q,w}^*=1$ then \noindent {\rm a)} $B_t$ is a bounded operator with operator norm bounded by $$ \norm{B_t}_{L^2(\rz^d,\calG)} \le t \frac{r}{1-r^{-1}}. $$ {\rm b)} $H_t$ is a Hilbert-Schmidt operator with Hilbert-Schmidt norm bounded by $$ \norm{H_t}_{\text{HS}}^2\le (2\pi)^{-d}t^{-(q-2)} \Big(1+\frac{2}{q-2} \Big). $$ \end{lemma} \begin{remarks}\rm i) Due to our choice of $B_t$, $H_t$ the bound in Lemma \ref{lem:bounds}.b) is \emph{independent} of $r$ and in a) it easy to see that the choice $r\!=\!2$ is optimal.\\[0.3em] ii) This lemma also shows that $f(x)g(p)$ is a compact operator since it is the norm limit for $t\to 0$ of the Hilbert-Schmidt operators $H_t$. \end{remarks} \begin{proof} Part a) follows completely Cwikel's original proof: Since the $f_l$ (resp., $g_m$) are orthogonal operators for different indices we get, for simple functions $\psi$ and $\phi$, say, \begin{equation*} \begin{split} \vert\langle\psi, & B_t \phi \rangle \vert \le \sum_{l+m\le 1} r^{l+m} \big\Vert r^{-l}f_l(x)\psi \big\Vert_2\, \big\Vert r^{-m}g_m(p)\phi \big\Vert_2 \\ &\le \sum_{s\le 1} r^s \sum_{m\in\nz} \big\Vert r^{-(s-m)}f_{s-m}(x)\psi\big\Vert_2 \big\Vert r^{-m}g_m(p)\phi \big\Vert_2 \\ &\le \sum_{s\le 1} r^s \left(\sum_{m\in\nz} \big\Vert r^{-(s-m)}f_{s-m}(x)\psi\big\Vert_2^2\right)^{1/2} \left(\sum_{m\in\nz} \big\Vert r^{-m}g_m(p)\phi\big\Vert_2^2\right)^{1/2} \\ &= \sum_{s\le 1} r^s \Big\Vert \sum_{m\in\nz}r^{-(s-m)}f_{s-m}(x)\psi\Big\Vert_2 \Big\Vert \sum_{m\in\nz} r^{-m}g_m(p)\phi\Big\Vert_2 \\ &\le r (1-r^{-1})^{-1} t \norm{\psi}\norm{\phi} , \end{split} \end{equation*} since $\sum_l r^{-l}f_l(x)\le t\idG$ and $\sum_{m}r^{-m}g_m(\xi) \le \idG$. Thus $B_t$ extends to a bounded operator on $L^2(\rd,\calG)$ with the given bound for its norm.\\[0.3em] To prove part b) observe that by Lemma \ref{lem:HS} and the cyclicity of the trace, we have $$ \norm{H_t}_{\text{HS}}^2 = \sum_{l+m >1} \iint \tr_{\calG}[f_l(x)g_m(\xi)^2 f_l(x)]\, dx d\xi . $$ Assume for $x,\xi\in\rd$ the operator inequality \begin{equation} \begin{split}\label{eq:operineq} \sum_{l+m >1}f_l(x)g_m(\xi)^2 f_{l}(x) &\le \Big( \norm{g(\xi)} f(x) \chi_{(t,\infty)}(\norm{g(\xi)}f(x)\!) \Big)^2 \\ &=: h(x,\xi)^2 \end{split} \end{equation} on the Hilbert space $\calG$. Note that the projection operator $\chi_{(t,\infty)}(\norm{g(\xi)}f(x))$ (on $\calG$) commutes with $f(x)$ for all $x,\xi\in\rd$. Let $\lambda_j(x)$ be the $j^{\text{th}}$ ordered eigenvalue of $f(x)$, and $E_{j}(\alpha)\defeq \{\norm{g(.)} \lambda_j(.)>\alpha \}$. Each $E_j$ has $2d$ dimensional Lebesgue measure $$ \vert E_j(\alpha)\vert_{2d} = \int \vert \{\norm{g(.)}>\alpha/\lambda_j(x) \}\vert_{d}\, dx \le \alpha^{-q} \int \lambda_j(x)^q \, dx , $$ since $\norm{g}_{q,w}^* =1$ by assumption. Thus we see \begin{equation*} \begin{split} \norm{H_t}_{HS}^2 &\le (2\pi)^{-d} \iint\trG[h(x,\xi)^2]\, dx d\xi \\ &= (2\pi)^{-d}\sum_{j} 2 \int_0^\infty \vert E_j(\max(\alpha, t))\vert_{2d} \alpha\, d\alpha \\ & = (2\pi)^{-d}\left( \sum_j 2 \int_0^{t} \vert E_j(t)\vert_{2d} \alpha\, d\alpha +\sum_j 2 \int_{t}^\infty \vert E_j(\alpha)\vert_{2d} \alpha\, d\alpha \right) \\ &\le (2\pi)^{-d} t^{-(q-2)} \big(1+ 2/(q-2) \big)\sum_j \int \lambda_j(x)^q\, dx \\ &= (2\pi)^{-d} t^{-(q-2)} \big(1+ 2/(q-2) \big), \end{split} \end{equation*} since $\sum_j \int \lambda_j(x)^q\, dx = \norm{f}_{q,q}^q =1$ by assumption. It remains to prove (\ref{eq:operineq}): Again, let $s=l+m$ and note that the $g_m(\xi)= g(\xi)\chi_{(r^{m-1}, r^{m}]}(g(\xi)\!)$ are orthogonal operators for different indices. As operators on $\calG$, \begin{equation*} \begin{split} \sum_{l+m>1} & f_l(x)g_m(\xi)^2f_l(x) = \sum_{l\in\gz}\sum_{s\ge 2} f_l(x) g_{s-l}(\xi)^2 f_l(x) \\ & = \sum_{l\in \gz} f_l(x) \big(\sum_{s\ge 2} g_{s-l}(\xi)^2\big) f_l(x) = \sum_{l\in\gz} f_l(x) g(\xi)^2 \chi_{(r^{1-l},\infty)}(g(\xi)\!) f_l(x) \\ &\le \sum_{l\in\gz} f_l(x) \norm{g(\xi)}_{\calG}^2 \chi_{(r^{1-l},\infty)}(\|g(\xi)\|) f_l(x) \\ &= f(x)^2 \norm{g(\xi)}_\calG^2 \sum_{l\in\gz} \underbrace{\chi_{(r^{1-l},\infty)}(\|g(\xi)\|) \chi_{t(r^{l-1},r^l]}(f(x)\!) }_{\le \chi_{ (t,\infty)}( \norm{g(\xi)}f(x)) \chi_{t(r^{l-1},r^l]}(f(x)\!)} \\ &\le f(x)^2 \norm{g(\xi)}_\calG^2 \chi_{ (t,\infty)}( \norm{g(\xi)}f(x)) \sum_{l\in\gz} \chi_{t(r^{l-1},r^l]}(f(x)\!) \\ &= f(x)^2 \norm{g(\xi)}_\calG^2 \chi_{ (t,\infty)}( \norm{g(\xi)}f(x)), \end{split} \end{equation*} which proves (\ref{eq:operineq}) and hence the lemma. \end{proof} Given the above bounds the proof of Theorem \ref{thm:Cwikel} is by now a standard interpolation argument. We give this argument for the sake of completeness: \begin{proof}[Proof of Theorem \ref{thm:Cwikel}] First, without loss of generality assume that $f$ and $g$ are non-negative operator-valued functions. Indeed, let $\calF$ be the Fourier transform and $M_f$ and $M_g$ the operators of ``multiplication" by $f$ and $g$ and note that $f(x)g(p)$ and $M_f\calF^{-1}M_g$ have the same singular values. With the polar decompositions $f(x)= U_1(x)\vert f(x)\vert$ and $g(\xi)= \vert g^*(\xi)\vert U_2^*(\xi)$ in the Hilbert space $\calG$ we have $$ M_f\calF^{-1}M_g = U_1 M_{\vert f\vert}\calF^{-1}M_{\vert g^*\vert} U_2^* , $$ where $U_j$, $j\in \{1,2\}$ are fibered partial isometries in the space $L^2(\rd,\calG)$, for example, $(U_1\psi)(x)= U_1(x)\psi(x)$ . Hence the singular values of $f(x)g(p)$ are bounded by the singular values of $M_{\vert f\vert}\calF^{-1}M_{\vert g^*\vert}$ and $\norm{\vert g^*\vert}_{q,w}^* = \norm{g}_{q,w}^*$. By one of the consequences of Ky Fan's inequality \cite{GK} we have \begin{equation*} \begin{split} \mu_n (f(x)g(p)) &= \mu_n(B_t+H_t) \le \mu_1 (B_t) + \mu_n(H_t) \le \norm{B_t} + \frac{1}{\sqrt{n}}\norm{H_t}_{HS} \\ &\le t \frac{r}{1-r^{-1}} + (2\pi)^{-d/2}t^{-(q-2)/2} \big(1+\frac{2}{q-2} \big)^{1/2} \frac{1}{\sqrt{n}} \end{split} \end{equation*} using Lemma \ref{lem:bounds}. Choosing $t$ and $r$ ($=2$) optimal gives \begin{equation*} \mu_n (f(x)g(p)) \le (2\pi)^{-d/q} \frac{q}{2} \left(\frac{8}{q-2}\right)^{1-2/q} \left( 1+ \frac{2}{q-2}\right)^{1/q} n^{-1/q} \end{equation*} which proves Theorem \ref{thm:Cwikel}. \end{proof} \noindent{\textbf{ Acknowledgment:}} It is a pleasure to thank Ari Laptev and Timo Weidl for stimulating discussions and Michael Solomyak for comments and discussions on an earlier version of the paper. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{thebibliography}{10} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \small{ \bibitem{AL} Aizenmann, M.\ and Lieb, E.~H., \textit{On semi-classical bounds for eigenvalues of Schr\"odinger operators.} Phys.~Lett.~\textbf{66A} (1978), 427--429. \bibitem{BL} Benguria, R.\ and Loss, M., \textit{On a theorem of Laptev and Weidl.} Math.\ Research Letters, \textbf{7} (2000), 195--203. \bibitem{Birman} Birman, M.~S., \textit{The spectrum of singular boundary problems.} Mat. Sb. \textbf{55} no.\ 2 (1961), 125--174, translated in Amer.~Math.~Soc.~Trans.\ (2), \textbf{53} (1966), 23--80. \bibitem{BS} Birman, M.~S.\ and Solomyak, M.~Z., \textit{Spectral theory of selfadjoint operators in Hilbert space.} Translated from the 1980 Russian original by S.\ Khrushchëv and V. Peller. Mathematics and its Applications (Soviet Series). D.\ Reidel Publishing Co., Dordrecht, 1987. xv+301 pp. \bibitem{Conlon} Conlon, J.~G., \textit{A new proof of the {C}wikel-{L}ieb-{R}osenbljum bound.} Rocky Mountain J.\ Math. \textbf{15} no.\ 1 (1985), 117--122. \bibitem{Cwikel} Cwikel, M., \textit{Weak type estimates for singular values and the number of bound states of Schr\"odinger operators.} Ann.\ Math.\ \textbf{106} no.\ 1 (1977), 93--100. \bibitem{GGMT} Glaser, V., Grosse, H., Martin, A., and Thirring, W., \textit{A family of optimal conditions for the absence of bound states in a potential.} Studies in Math.\ Phys., Essays in Honor of Valentine Bargmann, Princeton University Press, New Jersey (1976). \bibitem{GK} Gohberg, I.~C.\ and Krein, M.~C., \textit{Introduction to the theory of linear non-self-adjoint operators.} Trans.\ Math.\ Monographs vol.\ \textbf{18}, AMS (1969). \bibitem{HR} Helffer, B.,and Robert, D., \textit{Riesz means of bound states and semi-classical limit connected with a Lieb-Thirring conjecture. I, II.} Asymptotic Anal.\ \textbf{3} (1990), no. 2, 91--103. Ann. Inst. H. Poincar\'{e} Phys.\ Th\'{e}or.\ \textbf{53} no.\ 2 (1990), 139--147. \bibitem{HLW} Hundertmark, D., Laptev, A., and Weidl, T., \textit{New bounds on the Lieb-Thirring constant.} Invent.\ Math.\ \textbf{140} no.\ 3 (2000), 693--704 \bibitem{HLT} Hundertmark, D., Lieb, E.~H., and Thomas, L.~E., \textit{A sharp bound for an eigenvalue moment of the one-dimensional Schr\"odinger operator.} Adv.\ Theor.\ Math.\ Phys.\ \textbf{2}, no.\ 4 (1998), 719--731. \bibitem{Laptev} Laptev, A., \textit{Dirichlet and Neumann eigenvalue problems on domains in Euclidian spaces.} J.\ Funct.\ Anal.\ \textbf{151} (1997), 531--545. \bibitem{Laptevprivate} Laptev, A. Private communication. \bibitem{LW} Laptev, A.\ and Weidl, T., \textit{Sharp Lieb-Thirring inequalities in high dimensions.} Acta Math.\ \textbf{184} no.\ 1 (2000), 87--111. % \bibitem{LW2} Laptev, A.\ and Weidl, T., \textit{Recent results on Lieb-Thirring inequalities.} To appear in Journ\'ees \'Equation aux d\'eriv\'ees partielles. % % \bibitem{LY} Li, P.\ and S.-T.~Yau, S.~T., \textit{On the Schr\"odinger equation and the eigenvalue problem.} Comm.\ Math.\ Phys.\ \textbf{88} (1983), 309--318. \bibitem{Lieb} Lieb, E.~H., \textit{Bounds on the eigenvalues of the Laplace and Schr\"odinger operators.} Bull.\ Amer.\ Math.\ Soc., \textbf{82} (1976), 751--753. See also Proc.\ A.M.S.\ Symp.\ Pure Math. \textbf{36} (1980), 241--252. \bibitem{Lieb2} Lieb, E.~H., \textit{Thomas-Fermi and related theories of atoms and molecules.} Rev.\ Modern Phys.\ \textbf{53}, no.\ 4 (1981), 603--641. \bibitem{LT} Lieb, E.~H.\ and W.\ Thirring, W., \textit{Bound for the kinetic energy of fermions which proves the stability of matter.} Phys.\ Rev.\ Lett.\ \textbf{35} (1975), 687--689. Errata \textbf{35} (1975), 1116. \bibitem{LT2} Lieb, E.~H.\ and W.\ Thirring, W., \textit{Inequalities for the moments of the eigenvalues of the Schr\"odinger Hamiltonian and their relation to Sobolev inequalities.} Studies in Math.\ Phys., Essays in Honor of Valentine Bargmann, Princeton University Press, New Jersey (1976), \bibitem{Rozenblum} Rozenblum, G.~V., \textit{Distribution of the discrete spectrum of singular differential operators.} Dokl.\ Akad. Nauk SSSR, \textbf{202}, \textbf{N 5} (1972), 1012-1015. Translated in Soviet Math.\ Dok.\ \textbf{13} (1972), 245--249. See also Izv.\ VUZov, Matematika, \textbf{N.1}(1976), 75--86. \bibitem{RSI} Reed, M.\ and Simon, B., \textit{Methods of Modern Mathematical Physics I: Functional Analysis.} Revised and enlarged edition. Academic Press, New York (1980). \bibitem{RSII} Reed, M.\ and Simon, B., \textit{Methods of Modern Mathematical Physics II: Fourier Analysis, Self--Adjointness.} Academic Press, New York (1975). \bibitem{RSIV} Reed, M.\ and Simon, B., \textit{Methods of Modern Mathematical Physics IV: Analysis of Operators.} Academic Press, New York (1978). \bibitem{Schwinger} Schwinger, J., \textit{On the bound states of a given potential.} Proc.\ Nat.\ Acad.\ Sci.\ U.S.A.\ \textbf{47}, (1961), 122--129. \bibitem{Simon} Simon, B., \textit{Quantum Mechanics for Hamiltonians Defined as Quadratic Forms.} Princeton Series in Physics, Princeton University press, New Jersey (1971). \bibitem{Simon2} Simon, B., \textit{Bound states of two-body Schr\"odinger operators -- A review.} Studies in Math.\ Phys., Essays in Honor of Valentine Bargmann, Princeton University Press, New Jersey (1976). \bibitem{Simon3} Simon, B., \textit{Functional Integration and Quantum Physics.} Academic Press, New York, (1977). \bibitem{Weidl} Weidl, T., \textit{On the Lieb-Thirring constants $L_{\gamma,1}$ for $\gamma\geq 1/2.$} Comm.\ Math.\ Phys.\ \textbf{178} no.\ 1 (1996), 135--146. } \end{thebibliography} %%%%%%%%%%%%%%%% \end{document} %%%%%%%%%%%%%%%% ---------------0101221822705--