Content-Type: multipart/mixed; boundary="-------------0206010108398"
This is a multi-part message in MIME format.
---------------0206010108398
Content-Type: text/plain; name="02-249.comments"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment; filename="02-249.comments"
21 pages, no figures. Submitted to Journal de Mathematiques Pures et Appliquees.
---------------0206010108398
Content-Type: text/plain; name="02-249.keywords"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment; filename="02-249.keywords"
time-dependent Hartree-Fock, TDHF, Slater closure, mean-field dynamics, fermions, propagation of chaos
---------------0206010108398
Content-Type: application/x-tex; name="TDHF1.tex"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline; filename="TDHF1.tex"
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% Bardos-Golse-Gottlieb-Mauser
%
% "Rigorous derivation of TDHF"
%
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% version Alex, Vienna, 07.Nov.01
% Alex, Los Angeles, 3.April.02
% updated, Francois, May.02
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Voici des remarques sur TDHF
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\documentclass[11pt]{article}
\usepackage{amsmath, amssymb}
\newtheorem{lemma}{Lemma}[section]
\newtheorem{theorem}{Theorem}[section]
\newtheorem{corollary}[lemma]{Corollary}
\newtheorem{remark}{Remark}[section]
\newtheorem{definition}{Definition}[section]
\newtheorem{proposition}{Proposition}[section]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def \AA {{\cal A}}
\def \BB {{\cal B}}
\def \FF {{\cal F}}
\def \CC {{\cal C}}
\def \DD {{\cal D}}
\def \EE {{\cal E}}
\def \GG {{\cal G}}
\def \HH {{\cal H}}
\def \KK {{\cal K}}
\def \II {{\cal I}}
\def \LL {{\cal L}}
\def \LA {{{\cal L}^\dagger}}
\def \MM {{\cal M}}
\def \NN {{\cal N}}
\def \OO {{\cal O}}
\def \PP {{\cal P}}
\def \QQ {{\cal Q}}
\def \RR {{\cal R}}
\def \SS {{\cal S}}
\def \TT {{\cal T}}
\def \UU {{\cal U}}
\def \VV {{\cal V}}
\def \BbR {{I\!\!R}}
\def \BbN {{I\!\!N}}
\def \BbH {{I\!\!H}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% macros for calculus %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def \do {d\omega}
\def \dx {dx}
\def \dv {dv}
\def \inttwo {\int \! \! \! \int }
\def \intthree{\int \! \! \! \int \! \! \! \int }
\def \del {\partial}
\def \GRAD {\nabla\!_x}
\def \DIV {\nabla\!_x \! \cdot \!}
\def \LAP {\Delta_x}
\def \DOT {\! \cdot \!}
\def \DDOT {\! : \!}
\def \VEE { \vee }
\def \vv {{v \! \vee \! v}}
\def \vvv {{v \! \vee \! v \! \vee \! v}}
\def \vvvv {{v \! \vee \! v \! \vee \! v \! \vee \! v}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% macros for operations %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def \Langle {\big\langle \! \! \big\langle}
\def \Rangle {\big\rangle \! \! \big\rangle}
\def \tr {\text{tr}}
\def \Null {\text{N}}
\def \Range {\text{R}}
\def \adj {\dagger}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%Claude's macros
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newcommand{\XN}{{X_N}}
\newcommand{\YN}{{Y_N}}
\newcommand{\ZN}{{Z_N}}
\newcommand{\WN}{{W_N}}
\newcommand{\Xn}{{X_n}}
\newcommand{\Yn}{{Y_n}}
\newcommand{\Xnpone}{{X_{n+1}}}
\newcommand{\Ynpone}{{Y_{n+1}}}
\newcommand{\Wn}{{W_n}}
\newcommand{\Zn}{{Z_n}}
\newcommand{\Xm}{{X_m}}
\newcommand{\Ym}{{Y_m}}
\newcommand{\Zm}{{Z_m}}
\newcommand{\Wm}{{W_m}}
\newcommand{\XnN}{X^{n}_N}
\newcommand{\YnN}{Y^{n}_N}
\newcommand{\ZnN}{Z^{n}_N}
\newcommand{\ZnoneN}{Z^{n+1}_N}
\newcommand{\WnN}{W^{n}_N}
\newcommand{\VnN}{V^{n}_N}
\newcommand{\rhoN}{\rho_N}
\newcommand{\rhon}{\rho_n}
\newcommand{\rhoone}{\rho_1}
\newcommand{\rhotwo}{\rho_2}
\newcommand{\rhoNn}{\rho_{N,n}}
\newcommand{\rhoNnpone}{\rho_{N,n+1}}
\newcommand{\rhonpone}{\rho_{n+1}}
\newcommand{\rhoNN}{\rho_{N,N}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
\title{Mean field dynamics of fermions and the time-dependent Hartree-Fock equation}
%%
%% AUTHORS :
%%
\author{
%------------------
Claude BARDOS%
\footnote{ Univ. Paris 7 and LAN (Univ. Paris 6), France
(bardos@math.jussieu.fr). },
Fran\c cois GOLSE%
\footnote{
ENS-Ulm and LAN (Univ. Paris 6), France (Francois.Golse@ens.fr).
},
Alex D.\ GOTTLIEB %
\footnote{
Wolfgang Pauli Inst. c/o Inst.\ f.\ Mathematik, Univ. Wien, Strudlhofg.\ 4, A--1090 Wien, Austria
(alex@alexgottlieb.com).
} \\
and
Norbert J.\ MAUSER%
\footnote{
Wolfgang Pauli Inst. c/o Inst.\ f.\ Mathematik, Univ. Wien, Strudlhofg.\ 4, A--1090 Wien, Austria
(mauser@courant.nyu.edu).
}
}
\maketitle
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Introduction}
In this article we consider the Hamiltonian dynamics of systems of
fermions and derive the time-dependent Hartree-Fock equation in
the mean field limit. We follow the approach of Spohn, who
derived a mean field dynamical equation (the time-dependent
Hartree equation) for mean field systems of distinguishable
particles, remarking at the time that ``the convergence of the
mean field limit with statistics included is an open
problem"\cite{Sp}.
In Spohn's theory the initial $N$-body density operator $D_N$
is assumed to be a product state $D^{\otimes N}$, i.e.,
the particles are statistically independent and identically distributed.
The mean field limit is investigated in the Schr\"odinger picture,
where $D_N(t)$ obeys the von Neumann equation
\begin{equation}
i\hbar \frac{d}{dt}D_N(t) \ = \ \sum_{1 \le j \le N} \big[ L_j, D_N(t)
\big] \ + \ \frac{1}{N-1} \sum_{1 \le i0$.
This phenomenon could be called the {\it propagation of Slater closure}
because it is like the ``propagation of chaos" mentioned above. Since $\{D_N(t)\}$ has Slater closure,
the two-body density operator
$D_{N:2}(t)$ is approximately equal to $(D_{N:1}(t) \otimes D_{N:1}(t)) \Sigma_2$ when $N$ is large,
where $\Sigma_2$ is the two-body operator
defined by
\[
\Sigma_2 (x\otimes y) \ = \ x\otimes y - y \otimes x \ .
\]
Substituting $(D_{N:1}(t) \otimes D_{N:1}(t)) \Sigma_2$ for $D_{N:2}(t)$
in the BBGKY hierarchy leads one to conjecture that,
when $N$ is large,
the single-body density operator should nearly obey
the time-dependent Hartree-Fock (TDHF) equation
\begin{eqnarray*}
i\hbar \frac{d}{dt}F(t)
& = &
\left[ L, \ F(t)\right] \ + \ \left[ V,\ (F(t)\otimes
F(t)) \Sigma_2 \right]_{:1} \\
F(0)
& = & D_{N:1}(0) .
\end{eqnarray*}
Theorem~\ref{Main result} confirms this conjecture.
Theorem~\ref{Main result} states that the distance in the trace
norm between $D_{N:1}(t)$ and the corresponding solution $F(t)$ of
the TDHF equation tends to $0$ as $N$ tends to infinity. The
trace norms of $D_{N:1}(t)$ and $F(t)$ are separately equal to
$1$, so it is significant that their difference $D_{N:1}(t)-F(t)$
converges to $0$ in the trace norm. A crucial detail of the proof
is Lemma~\ref{lemma1}, which states that the {\it operator} norm
of $D_{N:1}$ tends to $0$ if $\{D_N \}$ has Slater closure. Much
of the rest of the proof lies in bounding the {\it trace norm} of
$D_{N:1}(t)-F(t)$ by an expression involving the {\it operator
norm} of $D_{N:1}(0)$.
The use of the trace norm to measure the distance between two
density operators is quite natural. A density operator $D$
corresponds to a quantum state through the assignment $ B \mapsto
\hbox{Tr}(DB)$ of expectation values to bounded observables $B$.
Thus, two density operators $D$ and $D'$ are within $\varepsilon$
of one another in the trace norm if and only if they correspond to
quantum states that give expectations differing by no more than
$\varepsilon$ for all observables $B$ with $\|B\| \le 1$.
In this article, we assume that the two-body potential $V$ is a bounded operator. We
find the error in approximating
$D_{N:1}$ by the solution of the
TDHF equation to be (at worst) proportional to $\|V\|$.
Because of this, our estimates are not of much use for real
$N$-particle systems (where there is no mean field scaling),
for then the error becomes proportional to
$N\|V\|$ and this is not likely to be small. It would be better,
from a physical point of view, to prove that the accuracy of the
TDHF approximation is proportional to the {\it average}
interaction energy $\hbox{Tr}(D_N V)$ rather than the {\it
maximum} interaction energy $\|V\|$.
Recent work on the time-dependent Schr\"odinger-Poisson equation
\cite{BEGMY,EY} suggests that it may be possible to prove a
theorem similar to our Theorem~\ref{Main result} when $V$ is the
Coulomb potential. This work shall be published in a separate
paper.
This rest of this article is organized as follows:
The next
section discusses fermionic density operators and defines Slater
closure. The N-particle Hamiltonian and the associated
time-dependendent Hartree-Fock equation are described in
Section~3. This section concludes with the statement of our main
result, Theorem~\ref{Main result}, whose proof spans Sections~4,
5, and 6. Sections~7 is an appendix relating the von Neumann form
of the TDHF equation, used throughout this paper, to the
formulation of the TDHF equation as a coupled system of wave
equations, which may be more familiar to some readers.
\section{Fermionic density operators and Slater closure}
Let $\BbH$ be a Hilbert space, supposed to be the space of wavefunctions for a
certain type of quantum system (a ``component" or ``particle"). Then the
Hilbert space of wavefunctions for a system consisting of $N$ distinguishable
components or particles of that type is
$\BbH_N = \BbH^{\otimes N}$. If the components are not distinguishable, but
obey
Fermi-Dirac statistics, then the appropriate Hilbert space of wavefunctions
is the antisymmetric subspace $\AA_N \subset \BbH_N$.
To define this subspace, it is convenient first to define unitary {\it
transposition} and {\it permutation} operators
on $\BbH_N$.
The transposition operator $U_{(ij)}$ is defined by
extending the following isometry defined on {\it simple tensors}
\[
U_{(ij)}(x_1\otimes x_2 \otimes \cdots x_i \cdots x_j \cdots \otimes x_N)
\ = \
x_1\otimes x_2 \otimes \cdots x_j \cdots x_i \cdots \otimes x_N
\]
to all of $\BbH_N$. For any $\pi$ in the group $\Pi_N$ of
permutations of $\{1,2,\ldots,N\}$, one may define the permutation operator
$ U_{\pi} $ as $ U_{(i_k j_k)} \cdots U_{(i_2 j_2)} U_{(i_1 j_1)}$,
where $(i_k j_k)
\cdots (i_2 j_2)(i_1 j_1)$ is any product of transpositions that equals $\pi$.
The antisymmetric subspace may now be defined as
\[
\AA_N \ = \
\left\{ \psi \in \BbH_N \ : \ U_{\pi}\psi =
\hbox{sgn}(\pi)\psi \quad \forall \pi \in \Pi_N \right\} .
\]
One may verify that
$$P_{\AA_N} \ =
\ \frac{1}{N!} \sum_{\pi \in \Pi_N} \hbox{sgn}(\pi) U_{\pi}$$
is the orthogonal projector whose range is $\AA_N$.
The {\it pure states} of an $N$-fermion system correspond to the
orthogonal projectors $P_{\psi}$ onto one-dimensional subspaces of $\AA_N$.
That is, a pure state is given by
$$
P_{\psi}(\phi) \ = \ \langle \phi, \psi \rangle \psi
$$
for some $\psi \in \AA_N$ of unit length. The {\it statistical
states} of the $N$-fermion system are the positive trace class
operators or {\it density operators} $D$ on $\AA_N$ of trace $1$.
These can be identified with density operators $D$ on all of
$\BbH_N$ whose eigenvectors lie in $\AA_N$, i.e., such that
$$
D \ = \ \sum_{i=1}^{\infty} \lambda_i P_{\psi_i}
$$
for some orthonormal system $\{\psi_i\}$ in $\AA_N$ and a family
of positive numbers $\lambda_i$ that sum to $1$. It follows that
these {\it fermionic densities} are those density operators that
satisfy
\begin{equation}
\label{antisymmetric}
D U_{\pi} \ = \ U_{\pi} D \ = \ \hbox{sgn}(\pi) D
\qquad \qquad
\forall \pi \in \Pi_N.
\end{equation}
If a density operator $D$ on $\BbH_N$ commutes with every
permutation operator $U_{\pi}$ then it is {\it symmetric}. In
particular, fermionic densities are symmetric by
(\ref{antisymmetric}).
If $\{e_j\}_{j \in J}$ is an orthonormal basis of $\BbH$ then
$$
\left\{ e_{j_1}\otimes e_{j_2} \otimes \cdots \otimes e_{j_N}: \
j_1,j_2,\ldots,j_N \in J \right\}$$ is an orthonormal basis of
$\BbH_N$. Since $\AA_N$ is the range of $P_{\AA_N}$ and since
$$
P_{\AA_N}(e_{j_1}\otimes e_{j_2} \otimes \cdots \otimes e_{j_N}) \
= \ 0$$ unless all of the indices $j_i$ are distinct, the set
$$ \left\{
P_{\AA_N}(e_{j_1}\otimes e_{j_2} \otimes \cdots \otimes e_{j_N})\!
: \ j_1, j_2 , \ldots, j_n \ \hbox{all distinct}\right\}
$$
is a spanning set for $\AA_N$. In fact it is an orthogonal basis
for $\AA_N$, each vector having norm $1/\sqrt{N!}$. Vectors of
the form $\sqrt{N!}\ P_{\AA_N}(e_{j_1}\otimes e_{j_2} \otimes
\cdots \otimes e_{j_N})$ are known as {\it Slater determinants}.
The trace class operators on a Hilbert space $\BbH$ form a Banach
space $\TT(\BbH)$ with the norm $\Vert T\Vert_{tr} =
\hbox{Tr}(|T|)$. The important inequality
\begin{equation}
\Vert T B \Vert_{tr} \ \le \ \Vert T \Vert_{tr} \Vert B
\Vert
\label{basic inequality}
\end{equation}
holds whenever $B$ is a bounded operator of norm $\Vert B \Vert$
and $T \in \TT(\BbH)$.
It is this basic inequality that will produce our key estimates.
For $n \le N$, the $n^{th}$
{\it partial trace} is a contraction from
$\TT(\BbH^{\otimes N})$ onto $\TT(\BbH^{\otimes n})$. The $n^{th}$
partial trace of $T$ will be denoted $T_{:n}$, and may be defined
as follows: Let $\OO$ be any orthonormal basis of $\BbH$. If $T
\in \TT(\BbH^{\otimes N})$ and $n < N$ then
\begin{eqnarray}
\left \ =&\nonumber
\\
\sum_{z_1,\ldots,z_{N-n} \in \ \OO}
&\Big<\ T( w \otimes z_1 \otimes \cdots \otimes z_{N-n}),\ x
\otimes z_1 \otimes \cdots \otimes z_{N-n} \Big>
\label{partial trace}
\end{eqnarray}
for any $w,x \in \BbH^{\otimes n}$. If a trace class operator $T
\in \TT(\BbH^{\otimes N})$ satisfies (\ref{antisymmetric}) then so
does $T_{:n}$, i.e., the partial trace defines a positive
contraction from $\TT(\BbH^{\otimes N})$ to $\TT(\BbH^{\otimes
n})$ that carries fermionic densities to fermionic densities.
In the following definition, and throughout this article, we use
the superscript $^{\otimes n}$ to denote the $n^{th}$ tensor power
of an operator, and we use the notation $\Sigma_n$ for
$n!P_{\AA_n}$, i.e.,
\[
\Sigma_n \ = \ \sum_{\pi \in \Pi_n} \hbox{sgn}(\pi)
U_{\pi} \ .
\]
The $n^{th}$
{\it tensor power} of an operator $A$ on $\BbH$ is the operator
$A^{\otimes n}$ on $\BbH_n$ defined on simple tensors by
\[
A^{\otimes n}(x_1\otimes x_2 \otimes \cdots \otimes x_n)
\ = \
Ax_1 \otimes Ax_2 \otimes \cdots \otimes Ax_n\ .
\]
\begin{definition}
\label{closure}
For each $N$, let $D_N$ be a symmetric density operator on $\BbH_N$.
The sequence $\left\{D_N\right\}$ {\bf has Slater closure} if, for each fixed
$n$,
\[
\lim_{N \rightarrow \infty} \big\| D_{N:n} \ - \
D_{N:1}^{\otimes n} \Sigma_n \big\|_{tr} \ = \ 0.
\]
\end{definition}
This terminology is motivated by the observation that, if $\Psi_N$
is a Slater determinant in $\AA_N$ and $P_{\Psi_N}$ denotes the
orthogonal projector onto the span of $\Psi_N$ then
\begin{equation}
\left(P_{\Psi_N}\right)_{:n} \ = \ \frac{N^n(N-n)!}{N!}
\left(P_{\Psi_N}\right)_{:1}^{\otimes n} \sum_{\pi \in \Pi_n}
\hbox{sgn}(\pi)U_{\pi} \ ,
\label{SlaterFormula}
\end{equation}
for this implies the following:
\begin{proposition}
\label{Slater case}
For each $N$ let $\Psi_N$ be a Slater determinant in $\AA_N$, and let
$P_{\Psi_N}$ denote the orthoprojector onto the span of $\Psi_N$.
Then $\left\{ P_{\Psi_N} \right\}$ has Slater closure.
\end{proposition}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{The time-dependent Hartree-Fock equation}
We are going to prove that, in the mean field limit, the
time-dependent Hartree-Fock equation describes the time-evolution
of the single-particle state in systems of fermions. We state
our theorem in this section and go on to prove it in the three
subsequent sections.
First we describe the $N$-particle Hamiltonian. Let $iL^{(N)}$ be
a self-adjoint operator on $\BbH$, where $L^{(N)}$ may depend on
$N$ in an arbitrary manner. The free motion of the $j^{th}$
particle is governed by
\[
L^{(N)}_j = I^{\otimes j-1}\otimes L^{(N)} \otimes I^{\otimes N-j},
\]
where $I$ denotes the identity operator on $\BbH$. The
interaction between the particles has the form $1/(N-1)$ times the
sum over pairs of distinct particles of a two-body potential $V$.
Let $V$ be a bounded Hermitian operator on $\BbH \otimes \BbH$
that commutes with the transposition operator $U_{(12)}$. Define
the operator $V_{12}$ on $\BbH_N$ by
$$
V_{12}\left( x_1\otimes x_2 \otimes \cdots \otimes x_N \right)
\ = \
V(x_1\otimes x_2) \otimes x_3 \otimes \cdots \otimes x_N
$$
and for each $1 \le i < j \le N$ define $V_{ij}= U^*_{\pi} V_{12}
U_{\pi}$ where $\pi$ is any permutation with $\pi(i)=1$ and $\pi(j)=2$. Let
\begin{equation}
\label{Ham}
H_N \ = \ \sum_{1 \le j \le N}L^{(N)}_j \ + \ \frac{1}{N-1} \sum_{1 \le i 0$.
\end{theorem}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Two hierarchies, and their difference}
Consider the $N$-particle von Neumann equation (\ref{N-particle Schrodinger}).
From now on we will suppose that the initial $N$-particle density operator
$D_N(0)$ is symmetric, i.e., that
\[
U_{\pi}^* D_N(0) U_{\pi} = D_N(0)
\] for all $\pi \in \Pi_N$. (Recall that, in particular, fermionic densities are
symmetric.) The symmetry of the Hamiltonian (\ref{Ham})
ensures that $D_N(t)$ remains symmetric for all $t$.
From (\ref{N-particle Schrodinger}) and the
symmetry of $D_N(t)$, it follows that the partial trace
$D_{N:n}(t)$ satisfies
\begin{eqnarray}
i\hbar \frac{d}{dt}D_{N:n}(t)
& = &
\sum_{1 \le j \le n} \big[ L^{(N)}_j,D_{N:n}(t) \big] \ + \
\frac{1}{N-1} \sum_{1 \le i1$. When $F$ depends on $t$ we write $F^-_n(t)$ instead
of $F(t)^-_n$. The notation $F^-_2(t)$ has already been used in
the TDHF equation (\ref{TDHF}).
\begin{proposition}
\label{Prop1}
If $F(t)$ is a strong solution of the TDHF equation (\ref{TDHF}) then
\[
i\hbar \frac{d}{dt}F^-_n(t) \ = \
\sum_{j=1}^n \big[ L^{(N)}_j, F^-_n(t) \big] \ + \
\sum_{j=1}^n \left[ V_{j,n+1},\ F^-_{n+1}(t)\right]_{:n} \ +\
\RR_n(F(t))
\]
where $\RR_n$ is defined on trace class operators by $\RR_1(X) = {\bf 0}$ (the zero
operator) and
\begin{equation}
\label{Rn}
\RR_n(X) \ = \ \sum_{j=1}^n \Big[V_{j,n+1}, \ X^{\otimes n + 1}
\sum_{k \ne j}
U_{(k,n+1)}\Big]_{:n}\Sigma_n
\end{equation}
for $n > 1$.
\end{proposition}
\noindent {\bf Proof}: \qquad
For any trace class operator $X$,
\begin{eqnarray}
&&\sum_{j=1}^n\big[ V_{j,n+1}, \ X^-_{n+1}\big]_{:n}
\nonumber\\
& = &
\sum_{j=1}^n \Big[ V_{j,n+1}, \ X^{\otimes n + 1} \Big( I -
\sum_{k=1}^n U_{(k,n+1)} \Big)
\Sigma_n \otimes I_{\BB(\BbH)} \Big]_{:n}
\label{FrancoisWantsNumberHere} \\
& = &
\sum_{j=1}^n \Big[ V_{j,n+1}, \ X^{\otimes n + 1} \big( I - \sum_{k=1}^n
U_{(k,n+1)} \big)
\Big]_{:n} \Sigma_n \ .
\nonumber
\end{eqnarray}
The first equality in (\ref{FrancoisWantsNumberHere}) holds
because
\[
\Sigma_{n+1}=\Big( I - \sum_{k=1}^n U_{(k,n+1)} \Big)
\Sigma_n \otimes I_{\BB(\BbH)} ,
\]
and the second equality in (\ref{FrancoisWantsNumberHere}) holds
because $\Sigma_n\otimes I_{\BB(\BbH)}$ commutes with
$\sum\limits_{j=1}^n V_{j,n+1}$. From the TDHF equation
(\ref{TDHF}) we calculate that
\begin{eqnarray}
&& i\hbar \frac{d}{dt}F^-_n(t)
=
i\hbar \frac{d}{dt} F(t)^{\otimes n}\Sigma_n \nonumber\\
& = &
i\hbar \Big\{
\sum_{j=1}^n F(t)^{\otimes j-1} \otimes \frac{d}{dt}F(t) \otimes
F(t)^{\otimes n-j} \Big\} \Sigma_n
\nonumber \\
& = &
\sum_{j=1}^n \left\{
\big[ L^{(N)}_j, F(t)^{\otimes n} \big] +
\left[ V_{j,n+1}, \ F(t)^{\otimes n + 1}\left(I -
U_{(j,n+1)}\right) \right]_{:n}
\right\} \Sigma_n
\nonumber \\
& = &
\sum_{j=1}^n \big[ L^{(N)}_j, F^-_n(t) \big] \ + \ \RR_n(F(t))
\label{AlexPutNumberHere} \\
& & + \
\sum_{j=1}^n \Big[ V_{j,n+1}, \ F(t)^{\otimes n + 1} \Big( I -
\sum_{k=1}^n U_{(k,n+1)} \Big)
\Big]_{:n} \Sigma_n \ .
\nonumber
\end{eqnarray}
By the identity (\ref{FrancoisWantsNumberHere}), the last sum in
(\ref{AlexPutNumberHere}) equals $\sum\limits_{j=1}^n \big[
V_{j,n+1}, \
F^-_{n+1}(t)\big]_{:n}$, proving the proposition.
\hfill $\square$
Now let $D_N(t)$ be a solution of the N-particle von Neumann equation
(\ref{N-particle Schrodinger}) and let $F(t)$ be a solution of the TDHF
equation (\ref{TDHF}).
For $1 \le n \le N$ define the $n^{th}$ {\it difference}
\begin{equation}
E_{N,n}(t) \ = \ D_{N:n}(t) - F^-_n(t) .
\label{difference}
\end{equation}
From the N-particle hierarchy equations (\ref{N-particle hierarchy
rewritten}) and (\ref{N-particle hierarchy stuff}) and
Proposition~\ref{Prop1}, it follows that
\begin{eqnarray}
i\hbar \frac{d}{dt} E_{N,n}(t)
& = &
\LL^{(N)}_n( E_{N,n}(t)) \ + \
\sum_{j=1}^n \left[ V_{j,n+1},\ E_{N,n+1}(t)\right]_{:n}
\nonumber \\
& &
+ \ \EE_n(t,N,D_N(0)) \ - \ \RR_n(F(t))
\label{Hierarchy of differences}
\end{eqnarray}
for $n=1,2,\ldots,N-1$.
The characters $\EE$ and $\RR$ were chosen to evoke the words ``error" and
``remainder." Indeed, in the next section we find bounds on these error terms
under conditions on $D_N(0)$ and $F(0)$. The rest of this section is
devoted to showing
how such bounds lead to an upper bound on the differences $E_{N,n}(t)$.
To this end, let us define
\begin{equation}
\hbox{Err}(t,N, n) \ = \ \EE_n(t,N,D_N(0)) \ - \ \RR_n(F(t)).
\label{err}
\end{equation}
Let $U^{(N)}_{n,t}$ denote the unitary operator
$\exp\big(\frac{it}{\hbar}\sum_{j=1}^n L_j^{(N)}\big) $ on
$\BbH_n$ and define isometries $\UU^{(N)}_{n,t}$ on the trace
class operators by
\[
\UU^{(N)}_{n,t}(\ \cdot \ ) \ = \ e^{\frac{it}{\hbar}
\LL^{(N)}_n}(\ \cdot \ ) \ = \ U^{(N)}_{n,t}(\ \cdot \ )
U^{(N)}_{n,-t} \ .
\]
Then $Z_{N,n}(t) = \UU^{(N)}_{n,t}(E_{N,n}(t))$ has the same
trace norm as $E_{N,n}(t)$ and satisfies
\begin{equation}
\label{Z hierarchy}
\frac{d}{dt} Z_{N,n}(t) \ = \ -\frac{i}{\hbar}
\sum_{j=1}^n \left[ V_{j,n+1},\ Z_{N,n+1}(t)\right]_{:n} -\frac{i}{\hbar}
\hbox{Err}(t,N, n)
\end{equation}
for $n=1,2,\ldots,N-1$. From (\ref{Z hierarchy}) it follows that
\begin{eqnarray*}
\big\| E_{N,n}(t) \big\|_{tr} & = &
\big\| Z_{N,n}(t) \big\|_{tr}
\\
& \le &
\big\| Z_{N,n}(0) \big\|_{tr} \ + \ \frac{2\|V\|n}{\hbar}
\int_0^t \big\| Z_{N,n+1}(s) \big\|_{tr}ds\\
& + & \ \frac{1}{\hbar} \int_0^t \big\| \hbox{Err}(s,N,n)
\big\|_{tr}ds
\end{eqnarray*}
for $n=1,2,\ldots,N-1$. Recalling that $\|Z_{N,n+1}(s)\|_{tr} =
\|Z_{N,n+1}(s)\|_{tr}$,
the preceding inequality becomes
\begin{equation}
\big\| E_{N,n}(t) \big\|_{tr} \ \le \
\varepsilon(N,n,t) \ + \ \frac{2\|V\|n}{\hbar} \int_0^t \big\|
E_{N,n+1}(s) \big\|_{tr}ds
\label{iterate me}
\end{equation}
if we define
\begin{equation}
\varepsilon(N,n,t) \ = \ \big\| E_{N,n}(0) \big\|_{tr} \ +
\ \frac{1}{\hbar} \int_0^t \big\| \hbox{Err}(s,N,n) \big\|_{tr}ds.
\label{eps}
\end{equation}
Beginning from (\ref{iterate me}) and iterating the inequality $m$ times (for
some $m \le N-n-1$)
we obtain our desired bound on the trace norm of $E_{N,n}(t)$:
\begin{eqnarray}
\big\| E_{N,n}(t) \big\|_{tr}
& \le &
\sum_{k = 0}^{m} \binom{n+k-1}{n-1} \left(\frac{2\|V\|t }{\hbar}
\right) ^k \varepsilon(N,n+k,t) \nonumber \\
& + &
\binom{n+m-1}{n-1} \left(\frac{2\|V\|t }{\hbar} \right) ^{m}
\sup_{s \in [0,t]}\Big\{ \big\| E_{N,n+m+1}(s) \big\|_{tr} \Big\} .
\nonumber \\
\label{Bardos inequality}
\end{eqnarray}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Error estimates}
In this section we collect the error estimates that will be used
to prove Theorem~\ref{Main result}.
If $D_N(0)$ is a density operator then the solution $D_N(t)$ of the N-particle
von Neumann equation (\ref{N-particle Schrodinger}) is a
density operator for all $t>0$, and it is clear from (\ref{N-particle
hierarchy stuff}) that
\begin{equation}
\| \EE_n(t,N,D_N(0)) \|_{tr}
\ \le \
\frac{4n(n-1)}{N-1} \| V \| \
\label{estimates for N-particle hierarchy stuff}
\end{equation}
for all $t$.
\begin{lemma}
\label{lemma1}
If $\left\{D_N\right\}$ has Slater closure then
\[
\lim\limits_{N \rightarrow \infty} \|D_{N:1}\| = 0.
\]
\end{lemma}
\noindent {\bf Proof}: \qquad The trace norm of $D_{N:1}^2$ equals the
sum of the squares of the eigenvalues of $D_{N:1}$. Since the
operator norm of $D_{N:1}$ equals its largest eigenvalue, it
follows that $\big\| D_{N:1} \big\| \le \left\| D_{N:1}^2
\right\|_{tr}^{1/2}$. But $D_{N:1}^2 = \left\{ D_{N:1}^{\otimes
2}U_{(12)}\right\}_{:1}$, whence
\[
\left\| D_{N:1}^2 \right\|_{tr} \ = \ \Big\| \Big\{D_{N:2} -
D_{N:1}^{\otimes 2}(I - U_{(12)}) \Big\}_{:1} \Big\|_{tr}
\ \le \ \Big\| D_{N:2} - D_{N:1}^{\otimes
2}\Sigma_2\Big\|_{tr}.
\]
The Slater closure of $\left\{D_N\right\}$ implies that the
right-hand side of the preceding inequality tends to $0$ as $N
\longrightarrow \infty$. \hfill $\square$
\begin{lemma}
If $F$ is a density operator then $\big\| F_n^- \|_{tr} \le 1$ for all $n$.
\label{estimate for Fn-}
\end{lemma}
\noindent {\bf Proof}: \qquad Since $ \Sigma_n =\big(\Sigma_n\big)^* =
\frac{1}{n!} \big(\Sigma_n \big)^2 $ commutes with $F^{\otimes n}$, it
follows that
$
F_n^- = F^{\otimes n}\Sigma_n = \frac{1}{n!}
\Sigma_n \left( F^{\otimes n} \right) \Sigma_n $ is a nonnegative
operator. Thus, the trace norm of $F_n^-$ equals its trace. This
trace is
\[
\sum_{j_1,\ldots,j_n \in J} \Big< e_{j_1} \otimes \cdots
\otimes e_{j_n},\ F^{\otimes n} \Sigma_n(e_{j_1} \otimes \cdots
\otimes e_{j_n})\Big>
\]
where $\{e_j\}_{j \in J}$ is an orthonormal basis for $\BbH$. This sum may be
taken over distinct indices $j_1,\ldots,j_n \in J$,
since $\Sigma_n$ annihilates all tensor products $e_{j_1} \otimes \cdots
\otimes e_{j_n}$ with repeating factors, so that
\begin{eqnarray*}
\hbox{Tr}\left(F_n^- \right)
& = &
\sum_{\stackrel{distinct}{j_1,\ldots,j_n \in J} }
\Big< e_{j_1} \otimes \cdots \otimes e_{j_n},\ F^{\otimes n}
\Sigma_n(e_{j_1} \otimes \cdots \otimes e_{j_n})\Big> \\
& = &
\sum_{\stackrel{ distinct}{j_1,\ldots,j_n \in J}}
\left< e_{j_1} \otimes \cdots \otimes e_{j_n},\ F^{\otimes n}
(e_{j_1} \otimes \cdots \otimes e_{j_n})\right> \\
& \le & \hbox{Tr}\left(F^{\otimes n} \right) \ = \ 1
\end{eqnarray*}
as asserted. \hfill $\square$
The next lemma provides a bound on the trace norm of the
remainder term $\RR_n(F)$ when $F$ is a density operator.
The bound is proportional to the {\it operator} norm of $F$.
\begin{lemma}
Let $\RR_n$ be as in (\ref{Rn}) and let $F$ be a density
operator. Then
\begin{equation}
\Vert\RR_n(F) \Vert _{tr} \ \le \ 2n(n-1)\| V \| \ \|F\|.
\label{Estimate for R}
\end{equation}
\end{lemma}
\noindent {\bf Proof}: \qquad From (\ref{Rn}) we see that $\RR_n(F)$ equals
\[
\Bigg\{
\sum_{\stackrel{j,k =1}{j \ne k}}^{n} \Big(
V_{j,n+1} F^{\otimes n+1}U_{(k,n+1)} \ - \ F^{\otimes
n+1}U_{(k,n+1)} V_{j,n+1} \Big) \left( \Sigma_n \otimes
I_{\BB(\BbH)}\right)
\Bigg\}_{:n}.
\]
Since $U_{(k,n+1)} $ commutes with $F^{\otimes n+1}$ and since
$\Sigma_n \otimes I_{\BB(\BbH)}$ commutes with $\sum_{j,k: j \ne k}
U_{(k,n+1)} V_{j,n+1} $, it follows that $\RR_n(F)$ equals
\[
\Bigg\{ \sum_{\stackrel{j,k =1}{j \ne k}}^{n}
V_{j,n+1} U_{(k,n+1)} \big(F_n^- \otimes F\big)
\Bigg\}_{:n}
\ - \
\Bigg\{
\big(F_n^- \otimes F\big)
\sum_{\stackrel{j,k =1}{j \ne k}}^{n} U_{(k,n+1)} V_{j,n+1}
\Bigg\}_{:n}\ .
\]
Since the trace norm of a trace class operator equal the trace norm of its
adjoint, it follows that
\begin{eqnarray}
\| \RR_n(F) \|_{tr}
& \le &
2 \ \Bigg\| \sum_{\stackrel{j,k =1}{j \ne k}}^{n}
\Big\{ V_{j,n+1} U_{(k,n+1)} \big(F_n^- \otimes F\big)
\Big\}_{:n}
\Bigg\|_{tr} \nonumber \\
& \le &
2n(n-1) \Big\| \Big\{ V_{n-1,n+1} U_{(n,n+1)} \big(F_n^- \otimes F\big)
\Big\}_{:n}
\Big\|_{tr} \ .
\label{inequality}
\end{eqnarray}
But one may verify directly that
\begin{equation}
\Big\{ V_{n-1,n+1} U_{(n,n+1)} \big(F_n^- \otimes F\big)
\Big\}_{:n}
\ = \
(I^{\otimes n-1} \otimes F)V_{n-1,n}F_n^- ,
\label{claim}
\end{equation}
so that, by (\ref{basic inequality}) and Lemma~\ref{estimate for Fn-},
\[
\Big\| \Big\{ V_{n-1,n+1} U_{(n,n+1)} \big(F_n^- \otimes F\big)
\Big\}_{:n}
\Big\|_{tr}
\ \le \
\| F \| \ \| V \| \ \|F_n^- \|_{tr} \ \le \| F \| \ \| V \|.
\]
Substituting this in (\ref{inequality}) yields (\ref{Estimate for
R}).
To verify (\ref{claim}), choose an orthonormal basis $\{e_j\}_{j \in J}$
for
$\BbH$ and check that
the operators on both sides of (\ref{claim}) have the same matrix
elements relative to the basis $\big\{e_i \otimes e_j : \ i,j \in J \big\}$.
\hfill $\square$
Let $F(t)$ be a solution of the TDHF equation (\ref{TDHF}). Since the
(operator) norm of $F(t)$ is constant, it follows from
Lemma~\ref{Estimate for R} that
\[
\Vert\RR_n(F(t)) \Vert _{tr} \ \le \ 2 n(n-1) \Vert V \Vert
\ \|F(0)\|
\]
for all $ t \ge 0$.
Thus, $\hbox{Err}(t,N, n)$ of equation (\ref{err}) satisfies
\[
\| \hbox{Err}(t,N, n) \|_{tr} \ \le \ 2 n(n-1) \Vert V \Vert
\left(
\frac{2}{N-1} \ + \ \|F(0)\|
\right)
\]
and $\varepsilon(N,n,t)$ of equation (\ref{eps}) satisfies
\begin{eqnarray}
\varepsilon(N,n,t)
& = &
2n(n-1)\|V\| \frac{t}{\hbar}
\left(
\frac{2}{N-1} \ + \ \|F(0)\|
\right)
\nonumber \\
& & + \
\big\| E_{N,n}(0) \big\|_{tr} .
\label{estimate for eps}
\end{eqnarray}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Proof of the theorem}
Equipped with the estimates of the preceding sections, we proceed to the proof
of Theorem~\ref{Main result}.
So, let us assume that $D_N(0)$ is a symmetric density for each $N$ and that
the sequence $\{D_N(0)\}$ has Slater closure. Let $D_N(t)$ be the solution
of (\ref{N-particle Schrodinger})
with initial value $D_N(0)$, and let $F^{(N)}(t)$ be the solution of the TDHF
equation (\ref{TDHF}) whose
initial value is $F^{(N)}(0)=D_{N:1}(0)$. Let
$\left\{F^{(N)}\right\}^-_n(t)$ denote
$\big\{F^{(N)}(t)\big\}^{\otimes n}\Sigma_n$ and let $E_{N,n}(t)$
denote the difference between $D_{N:n}(t)$ and
$\left\{F^{(N)}\right\}^-_n(t)$.
We have the upper bound (\ref{Bardos inequality}) for the trace norm of
$E_{N,n}(t)$, into which we now substitute the estimates (\ref{estimate for
eps}). In the same stroke, we will replace the binomial coefficients
$\binom{n+k-1}{n-1}$ with the larger quantities $(n+k)^n/n!$
and we will use the fact that $\sup\limits_{s \in [0,t]}\big\{ \big\|
E_{N,n+m+1}(s) \big\|_{tr} \big\} \le 2$ by Lemma~\ref{estimate for Fn-}.
Also, let us set $T = 2\|V\|t / \hbar.$
We obtain
\begin{eqnarray}
\big\| E_{N,n}(t) \big\|_{tr}
& \le &
\frac{1}{n!} \sum_{k = 0}^{m} (n+k)^n \big\|
E_{N,n+k}(0)
\big\|_{tr} T^k \nonumber \\
& + &
\frac{1}{n!}
\sum_{k = 0}^{m} (n+k)^{n+2}
\Big(
\frac{2}{N-1} \ + \ \big\|F^{(N)}(0)\big\|
\Big)
T^{k+1} \nonumber \\
& + &
\frac{2}{n!} (n+m)^n T^m
\label{Bardos revisited}
\end{eqnarray}
for $m \le N-n-1$. Fix $T$ to be less than $1$, i.e., fix $t <
\frac{\hbar}{2\|V\|}$. For fixed $n$, consider the limit of the
right-hand-side of (\ref{Bardos revisited}) as $N$ and $m$ tend to
infinity. The individual terms (fixed $k$) tend to $0$, for
$\big\|F^{(N)}(0)\big\|$ tends to $0$ by Lemma~\ref{lemma1} and
$\big\| E_{N,n+k}(0) \big\|_{tr}$ tends to $0$ thanks to the
hypothesis that $\{D_N(0)\}$ has Slater closure (recall that
$F^{(N)}(0)=D_{N:1}(0)$). On the other hand, the series on the
right-hand-side of (\ref{Bardos revisited}) are dominated,
uniformly with respect to $m$, by a series that converges
absolutely for $T < 1$.
It follows that
\begin{equation}
\lim_{N \rightarrow \infty} \big\| E_{N,n}(t) \big\|_{tr} \ = \ 0
\label{error tends to zero}
\end{equation}
if $t < \frac{\hbar}{2\|V\|}$.
When $n=1$, this shows that
$
\lim\limits_{N \rightarrow \infty} \big\| D_{N:1}(t) - F^{(N)}(t)
\big\|_{tr} = 0
$
and consequently
\[
\lim_{N \rightarrow \infty} \Big\| D_{N:1}^{\otimes n}(t)
\Sigma_n \ - \ \big\{F^{(N)}\big\}^-_n(t) \Big\|_{tr} \ = \ 0
\]
for $n>1$ and $t < \frac{\hbar}{2\|V\|}$. From (\ref{error tends to zero})
again it follows that, for any $n$ and any $t < \frac{\hbar}{2\|V\|}$,
\[
\lim_{N \rightarrow \infty} \Big\| D_{N:n}(t) \ - \
D_{N:1}^{\otimes n}(t) \Sigma_n \Big\|_{tr} \ = \ 0,
\]
i.e., $\{D_N(t)\}$ has Slater closure. This proves the theorem up to $t =
\frac{\hbar}{2\|V\|}$.
Let $\tau = \frac{\hbar}{3\|V\|}$. At time $\tau$, it is no longer the case that
$D_{N:1}(\tau) = F^{(N)}(\tau)$. However, $\big\| E_{N,n+k}(\tau) \big\|_{tr}$ and
$\big\|F^{(N)}(\tau)\big\|$ still tend to $0$, and an argument nearly identical
to the one above shows that the theorem holds through time $2\tau >
\frac{\hbar}{2\|V\|}$. This argument may be repeated to establish the
conclusion of the theorem for all $t>0$.
\hfill $\square$
\section{Appendix: TDHF equations for wavefunctions}
The main body of this text describes time-dependent Hartree-Fock
equations in the language of density matrices and operator
calculus. In another formulation --- which may be more familiar
to some readers --- the TDHF equations are written as a system of
coupled Schr\"odinger equations for $N$ time-dependent orbitals.
This appendix explains how to recast the wavefunction formulation
of the TDHF equations into the language of density operators used
in this paper.
The starting point in this discussion is the linear $N$-body
Schr\"odinger
\begin{equation}
\label{N-Schrod} i\hbar \frac{\partial}{\partial t} \Psi_N \ = \
-\frac{\hbar^2}{2} \sum_{k=1}^N\Delta_{x_k}\Psi_N
+ \frac{1}{N-1} \sum_{1\le k