\magnification \magstep1
\input amstex
\documentstyle{amsppt}
\baselineskip 15pt
\pagewidth{6.4 truein}
\pageheight{8.6 truein}
\NoRunningHeads
\define\vp{\varphi}
\define\cs{\Cal S}
\define\ve{\varepsilon}
\define\fs{\frak S}
\define\fw{\frak W}
\define\bk{\Bbbk}
\define\vt{\vartheta}
\define\ph{\varphi}
\define\Dt{{1\over i}{\partial\hfill\over\partial t}}
\topmatter
\title Dispersive Smoothing for
Schr\"odinger Equations \endtitle
\endtopmatter
\head Lev Kapitanski$\,^*$ \endhead
\centerline{ Department of Mathematics, Kansas State University}
\centerline{ Manhattan, Kansas 66506, U S A }
\vglue .5pc
\centerline{ and }
\head Yuri Safarov$\,^*$ \endhead
\centerline{ Department of Mathematics, King's College,
London University}
\centerline{ Strand, London WC 2R 2LS, England, U K}
\vglue 2pc
\footnote""{$^*\;$ On leave from St.Petersburg
Branch of Steklov Mathematical Institute, St. Petersburg, Russia}
The phenomenon of the {\it global\/} ({\sl in time\/})
{\it dispersive smoothing\/} for the ``free" Schr\"odinger evolution
can be described as follows:
For any distribution $\,f\,$
of compact support, the solution
$\,\psi(t,x)\,$ of the Cauchy problem
$\,(\Dt - \Delta)\,\psi(t,x) = 0$, $\,t>0$,
$\,\psi(0,x) = f(x)$, $\,x\in\Bbb R^n$,
is infinitely differentiable with respect to $\,t\,$ and $\,x$,
when $\,t>0\,$ and $\,x\in\Bbb R^n$.
This is equivalent to saying that
the corresponding {\it fundamental solution\/} (= the solution
$\,S_0(t,x,y)\,$ of the initial value problem
with $\,f(x) = \delta (x-y)$)
is infinitely differentiable with respect
to $\,t,\,x$ and $\,y$, when $\,t>0$. And we have, indeed,
$\,S_0(t,x,y)= e^{-in\frac \pi4} (4\pi t)^{-\frac n2}\,
\exp\{i\,|x-y|^2/ 4t\}$, with the only singularity at $t=0$.
One would expect that dispersive smoothing should
survive ``small'' perturbations of the free Hamiltonian
$\,\Cal H_0 = - \Delta$. The problem is to determine
what perturbations are ``small''.
The case when the perturbed Hamiltonian has the form
$\,\Cal H = \Cal H_0 + V\,$ with a potential $\,V = V(x)$,
has been examined in
\cite{Ze}, \cite{OF}, \cite{Ki}, \cite{CFKS}.
The dispersive smoothing
takes place, for example, if the potential
is infinitely differentiable, and it and
all its derivatives are bounded, \cite{Ze}, \cite{OF}.
On the other hand, if $\,V(x)\,$ grows
quadratically or faster at infinity, then the singularities
may resurrect, as the example of the quantum harmonic oscillator
and Mehler's formula show ([Ze], [We], [CFKS], [MF]).
If the perturbation affects the metric of the space,
i.e. if $\,\Cal H_0 = - \Delta\,$ is replaced by
$\,
\Cal H = -\sum\limits_{j,k = 1}^{n}
{\partial\hfill\over\partial x_j}\,a^{j,k}(x)\,
{\partial\hfill\over\partial x_k}$,
the problem apparently becomes more subtle.
Practicaly no information on global dispersive smoothing
in that case has been available until recently.
At present, however, the situation has changed.
In their very interesting paper \cite{CKS}, W. Craig, T. Kappeler
and W. Strauss prove, in particular, a result that we will now
describe.
Assume that the coefficients $\,a^{j,k}(x)\,$ of $\,\Cal H\,$
are real and $\,C^{\infty}$, the matrix $\,(a^{j,k})\,$
is symmetric and positive-definite, and
$\,|a^{j,k}(x) - \delta^{jk}|\to 0\,$ sufficiently fast
as $\,|x|\to\infty$. It will be convenient to view the
principal symbol
$\,H(x,\xi) := \sum\limits_{j,k = 1}^{n}a^{j,k}(x)\xi_j\xi_k\,$
of the operator $\,\Cal H\,$
as the classical Hamiltonian and denote by
$\,(q(s;x,\xi),p(s;x,\xi))$ the corresponding Hamiltonian
trajectory emanating from the point
$\,(x,\xi)\,$ of the phase space $\,\Bbb R^{2n} = T^*(\Bbb R^n)$.
The point $\,(x,\xi)\,$ is said to be not trapped forwards
(respectively, backwards) by the bicharacteristic flow if
$\,|q(s;x,\xi)|\to +\infty\,$ as $\,s\to +\infty\,$
(respectively, $\,s\to -\infty$).
Let $\,S(t,x,y)\,$ denote the fundamental solution
corresponding to $\,\Cal H$, i.e. the solution of the initial problem
$$
{1\over i}{\partial\over \partial t}S(t,x,y)
+ \sum\limits_{j,k = 1}^{n}
{\partial\hfill\over\partial x_j}\,a^{j,k}(x)\,
{\partial\hfill\over\partial x_k}S(t,x,y) = 0,
\quad t>0,\quad S(0,x,y) =\delta (x-y)\,.
$$
\proclaim\nofrills{\bf Theorem CKS\/}\
{\rm([CKS, Theorem 1.9]).\/}
\noindent The point
$\,(x,\xi, y,\eta) \in T^*(\Bbb R^n)\times T^*(\Bbb R^n)\,$
does not belong to the wave front set of the fundamental solution
$\,S(t,\cdot,\cdot)\,$ for all $\,t>0\,$ if either $\,(x,\xi)\,$
is not trapped backwards or $\,(y,\eta)\,$ is not trapped forwards.
\endproclaim
The authors of \cite{CKS} do not work with such initial data as
$\,\delta (x-y)$. Instead, they work with the class of functions
$\,f\,$ satisfying
$\,(1+|x|)^k f(x)\in L^2$, for all $\,k>0$. Thus, in order to prove
their Theorem 1.9, the readers are instructed to approximate
$\,\delta$-function by the functions $\,f\,$ of the above class,
and check that all the intermediate results on the
wave front sets of the solutions, and all the estimates,
survive passing to the limit.
\vglue 1pc
Some time ago, not knowing of the research undertaken
by Craig, Kappeler and Strauss, the authors
of the present paper also obtained a result on dispersive
smoothing for the Schr\"odinger equations with variable
coefficients. Originally, this was done in the line
of our work on the parametrix for the Schr\"odinger equations
(\cite{KS}; the work was reported by
Yu.~S. at the conference in Saint Jean de Monts, France,
in the spring of 1990, and by
L.~K. at the Special Analysis Seminar at the Courant
Institute in the fall of 1990). The result on the singularities
of the fundamental solution that
we needed and proved at that time
was quite primitive compared to Theorem CKS.
Namely, we showed
that $\,S(t,x,y)\,$ has no singularities for $\,t>0\,$
provided all the points of the phase space $\,T^*(\Bbb R^n)\,$
are not trapped forwards.
Our approach is very different
from that of \cite{CKS}. Fortunately, we can handle some cases
and situations not covered by the technique of \cite{CKS};
for example, the initial {\it boundary-value\/} problems
and {\it systems\/} of equations.
Also, our method is simpler, and we believe
that it provides an important additional insight into the
basic properties of the Schr\"odinger equations.
\vglue 1pc
The essence of our method is in exploiting the correspondence
between the flows generated by the Schr\"odinger and
hyperbolic equations and then using the ``decay" of the local energy
of the latter.
The correspondence between the flows is described by
the following
relation between the fundamental solutions of the
Schr\"odinger and wave equations,
$$
\fs(t)\,=\,
{e^{- i{\pi\over 4}}\over
\sqrt{4\pi t}}\,
\int e^{i{\;\tau^2\over 4t}}\,\fw(\tau)\,d\tau\,,
\quad t>0,\tag 1
$$
where $\,\fs(t)\,$ is the fundamental solution
of the Schr\"odinger
equation,
$$
{1\over i}{\partial\hfill\over\partial t}
\,\fs(t)+A\,\fs(t) =0,\quad t>0,\qquad
\fs(0) = I,\tag 2
$$
and $\,\fw(\tau)\,$ is the fundamental solution
of the corresponding wave equation,
$$
{\partial^2\hfill\over\partial \tau^2}\,
\fw(\tau)+A\,\fw(\tau) =0, \quad \tau>0,\qquad
\fw(0)=I,\quad {\partial\hfill\over\partial t}\fw(0)=0.
\tag 3
$$
Here $\,A\,$ is an operator that in the simplest case
will be a perturbed Laplacian, and
$\,I\,$ is the identity operator.
An informal proof of (1) goes as follows.
First, notice that the scalar function
$$
\Bbbk(t,\tau)= {e^{- i{\pi\over 4}}\over
\sqrt{4\pi t}}\,
e^{i{\;\tau^2\over 4t}}
$$
is the fundamental solution of the 1-dimensional
Schr\"odinger equation,
$$
{1\over i}{\partial\hfill\over\partial t}
\,\Bbbk(t,\tau)-
{\partial^2\hfill\over\partial\tau^2}\,
\Bbbk(t,\tau) =0,\quad t>0, \qquad
\Bbbk(0,\cdot) =\delta(\cdot).
$$
Therefore,
$$\multline
\big({1\over i}{\partial\hfill\over\partial t}\,+\,A\big)
\int \Bbbk(t,\tau) \,\fw(\tau)\,d\tau =
\int \big({1\over i}{\partial\hfill\over\partial t}
\Bbbk(t,\tau)\,\fw(\tau) +
\Bbbk(t,\tau)\,A\,\fw(\tau)\big)\,d\tau \\
= \int \big({\partial^2\hfill\over\partial\tau^2}\,
\Bbbk(t,\tau) \,\fw(\tau) +
\Bbbk(t,\tau)\,A\,\fw(\tau)\big)\,d\tau
= \int \Bbbk(t,\tau) \big({\partial^2\hfill\over\partial\tau^2}
+ A\big)\,\fw(\tau)\,d\tau\,=\,0.
\endmultline\tag 4a
$$
Hence, the right hand side of (1) does indeed represent
the fundamental solution of the Schr\"odinger equation\ (2).
The argument showing that the fundamental solution $\,\fs(t)\,$
becomes regular for $\,t>0$ is based on the following
observation:
$$\multline
\hfill t^2 A\,\fs(t) = t^2 A \int \bk(t,\tau)\,\fw(\tau)\,d\tau =
t^2 \int \bk(t,\tau)\,A\,\fw(\tau)\,d\tau \hfill \\
\hfill =
t^2\int \bk(t,\tau)\cdot {d^2\hfill\over d\tau^2}\fw(\tau)\,d\tau =
t^2 \int {\partial^2\hfill\over \partial\tau^2}\bk(t,\tau)\cdot
\fw(\tau)\,d\tau \hfill \\
\hfill =\, i {t\over 2} \fs(t) -
{1\over 4} \int \bk(t,\tau) \tau^2 \fw(\tau)\,d\tau .\hfill
\endmultline\tag 4b
$$
In other words, multiplying by $\,t^2\,$ improves the
regularity in spatial directions. Since $\,\Dt\fs(t)=-A\fs(t)$,
it improves the regularity in $\,t\,$ as well.
Of course, we will have to justify the above manipulations,
and, in particular, the fact that the regularity of
$\,\int \bk(t,\tau) \tau^2 \fw(\tau)\,d\tau\,$
is the same as that of
$\,\int \bk(t,\tau) \fw(\tau)\,d\tau$.
In what follows, we first justify (1) in an abstract setting,
and then show how it works in applications.
\vglue 1pc
The existence, uniqueness and regularity properties
of solutions for the Schr\"odinger equation
will be viewed against the
backdrop of a scale of Hilbert spaces $\,H_s$, $\,s\in\Bbb R$.
We will also use the sets
$\,H_s^{\text{\rm comp\/}}\,$ of elements of $\,H_s\,$
with compact support, and the space $\,H_s^{\text{\rm loc\/}}\,$ --
the ``dual" of $\,H_{-s}^{\text{\rm comp\/}}$.
(One may think of $\,H_s\,$
as the usual Sobolev space $\,H^s(\Bbb R^n)$, then
$\,H_s^{\text{\rm comp\/}}\,$ and $\,H_s^{\text{\rm loc\/}}\,$
have the natural meaning.) In order to localize the elements of
$\,H_s^{\text{\rm loc\/}}\,$ we use the ``cut-off functions".
Let $\,\Upsilon\,$ denote the set of cut-off functions
or operators of multiplication by such functions
(in practice, $\,\Upsilon\approx C^{\infty}_0(\Bbb R^n)$).
Naturally, we assume that for every
$\,f\in H_s^{\text{\rm comp\/}}\,$ there is $\,\zeta\in\Upsilon\,$
such that $\,f=\zeta\, f$.
The operator $\,A\,$ will be thought of as
a (second order) linear operator that maps
$\,H_s \,$ continuously
into $\,H_{s-2}\,$ for any $\,s\in\Bbb R$. We make an additional
assumption that $\,A\,$ and its adjoint, $\,A^*$, map
$\,H_s^{\text{\rm comp\/}}\,$ into $\,H_{s-2}^{\text{\rm comp\/}}$.
We will assume that the hyperbolic problem
has a solution $\,\fw(t)\,$ in the following sense.
First,
$\,{d^{\ell}\hfill\over dt^{\ell}}\fw
\in C([0,t_1]\to \text{Hom} (H_s,H_{s-\ell}))$,
for all $\,t_1>0$, $\,\ell = 0,\,1,\,\dots$, and
for every $\,s\in\Bbb R$.
Here $\,\text{Hom} (H_s,H_{s-\ell})\,$ is the space of
bounded linear operators from $\,H_s\,$ into $\,H_{s-\ell}$,
and the continuity with respect to
$\,t\,$ is understood in the strong operator topology.
Second, given $\,f\in H_s$, the function $\,w(t) = \fw(t) f\,$
satisfies $\,w(0)=f$, $\,w_t(0)=0$, and
$\,(w_{tt}(t) + A w(t),\eta) = 0$, for all
$\,\eta\in H_{-s+2}$.
In addition, we make the following hypothesis
motivated by the known results on the long-term behavior
of solutions of hyperbolic problems (see, e.g., \cite{V 4},
Theorems 10.3.4 and 10.3.6,
and \cite{Rau 1}, Theorem 3).
\proclaim{Hypothesis ($\bigstar$)} For any pair of cut-off
functions $\,\zeta_1,\,\zeta_2\in\Upsilon$, there exists $\,T>1\,$
such that for $\,\tau\ge T-1\,$ the operator
$\,\zeta_2\cdot\fw(\tau)\cdot\zeta_1\,$ can be written in
the following form:
$$
\zeta_2\cdot\fw(\tau)\cdot\zeta_1\,=\,
\sum\limits_{j=1}^{N-1} \phi_j(\tau)\,e^{-i \lambda_j \tau}\,
W_j\,+\,
\tilde{W}_N(\tau),\tag 5a
$$
where the $\,\lambda_j$'s are
complex numbers with non-negative imaginary parts,
the scalar functions $\,\phi_j(z)\,$ are holomorphic
in the half-plane $\,\text{\rm Re}\,z>T-1\,$ and satisfy there
the following condition: for every finite interval
$\,[a,b]\,$ there exists an integer $\,\ell_j\ge 0\,$ such that
$$
\underset{a\le\theta\le b}\to\sup \;\;{1\over \tau^{\ell}}\,
\sum\limits_{m=0}^{\ell}
\,|{d^m\hfill\over d\tau^m}\phi_j(\tau+i\theta)|\,=\,O(\tau^{-2}),
\quad\text{as}\quad \tau\to\infty,
\qquad\forall \ell\ge\ell_j.\tag 5b
$$
The operators $\,W_1,\dots,W_{N-1}\,$ do not depend on
$\,\tau\,$ and are {\rm smoothing\/} in the sense that each
$\,W_j\,$ is bounded from $\,H_{s_1}\,$ to $\,H_{s_2}\,$
for any pair of $\,s_1,\,s_2\in\Bbb R$.
Finally, the operator $\,\tilde{W}_N(\tau)\,$ and all the derivatives
$\,{d^{\ell}\hfill\over d\tau^{\ell}}\,
\tilde{W}_N(\tau)$, are
smoothing for all $\,\tau\ge T-1$.
Moreover, for any $\,M>0\,$ there is an integer
$\,\kappa_M\ge 0\,$ such that, given $\,s_1,\,s_2\in\Bbb R$,
$$
\|{d^{\ell}\hfill\over d\tau^{\ell}}\,
\tilde{W}_N(\tau)\|_{H_{s_1}\to H_{s_2}}\,=\,O(\tau^{-M}),
\quad\text{\rm as\/}\quad \tau\to\infty,\quad
\ell = \kappa_M,\,\kappa_M+1,\dots ,\tag 5c
$$
where $\,\|\cdot\|_{H_{s_1}\to H_{s_2}}\,$ denotes the norm
of an operator from $\,H_{s_1}\,$ to $\,H_{s_2}$.
\endproclaim
Several situations where Hypothesis ($\bigstar$) is valid
will be described later.
\proclaim{Proposition 1}
Under the above assumptions on $\,\fw(t)$,
there exists $\,\fs(t)\,$ that solves (2) in the
following sense. First,
$$
{d^{\ell}\hfill\over dt^{\ell}}\fs \in
C([0,t_1]\to
\text{Hom} (H_s^{\text{\rm comp\/}},H_{s-2\ell}^{\text{\rm loc\/}})),
\qquad\forall t_1>0,\;\ell = 0,\,1,\,\dots ,
$$
for every $\,s\in\Bbb R$. Second, given
$\,f\in H_s^{\text{\rm comp\/}}$, the function
$\,u(t)=\fs(t) f\,$ satisfies $\,u(0)=f\,$ and
$\,({1\over i}{\partial\hfill\over \partial t} u(t)
+ A u(t), \eta) = 0$, for all
$\,\eta\in H_{-s+2}^{\text{\rm comp\/}}$ and all $\,t>0$.
The operator $\,\fs(t)\,$ has the following smoothing property:
$$
\left\{
\aligned
&\text{\rm for every integer $\,k>0$,}\\
&\text{\rm the
operator-valued function
$\,t\mapsto t^{2k} A^k \fs(t)\,$ is continuous in $\,t\,$}\\
&\text{\rm with values in
$\,\text{Hom} (H_s^{\text{\rm comp\/}},
H_{s}^{\text{\rm loc\/}})$.}
\endaligned
\right. \tag 6
$$
The operator $\,\fs(t)\,$ can be expressed in terms
of the operator $\,\fw\,$ as follows. For arbitrary
$\,\zeta_1,\,\zeta_2\in\Upsilon$, there exists
$\,T>1\,$ so that for any $\,C^{\infty}\,$ function
$\,\vartheta_0(\tau)\,$ on $\,\Bbb R\,$ that equals $\,1\,$
for $\,|\tau|\le T\,$ and $\,0\,$ for $\,|\tau|\ge T+1$,
there exists a smoothing operator $\,\frak C(t)\,$
(with infinitely smooth Schwartz kernel $\,\frak C(t,x,y)$)
such that
$$
\zeta_1\cdot \fs(t)\cdot\zeta_2
= \int \bk(t,\tau)\,\vartheta_0(\tau)
\zeta_1\cdot \fw(\tau)\cdot\zeta_2\,d\tau \,
+ \,\frak C(t). \tag 7
$$
\endproclaim
\demo{Sketch of the proof}
We regularize integral (1) by inserting
an extra factor $\,e^{-\ve\tau^2}$, $\,\ve>0$, into the integrand
and then letting $\,\ve\to 0$. Denote
$$
\fs_{\ve}(t) =
\int \bk(t,\tau) e^{-\ve\tau^2}\,\fw(\tau)\,d\tau.
$$
Our plan is to show first
that, given $\,f\in H_s^{\text{\rm comp\/}}$,
the limit
$$
\underset{\ve\downarrow 0}\to\lim\,(\fs_{\ve}(t)f,\eta)
\,=\,(\fs(t)f,\eta)
$$
exists for every $\,\eta\in H_{-s}^{\text{\rm comp\/}}\,$
and correctly defines an operator
$\,\fs(t) :\,H_s^{\text{\rm comp\/}}\to H_{s}^{\text{\rm loc\/}}$.
Second, we have to check that $\,\psi(t)=\fs(t) f\,$ is a
distributional solution of the problem
$\,\Dt \psi + A \psi = 0$, $\,\psi(0)=f$.
Finally, we have to justify (6) and (7).
The first step is the most involved. We pick two arbitrary
cut-off functions $\,\zeta_{1,2}\in\Upsilon$. By Hypothesis
($\bigstar$), there is $\,T>1\,$ so that (5a) holds for
$\,t\ge T-1$. Let $\,\vartheta_0(\tau)\,$ be a smooth real function
that equals $\,1\,$ for $\,\tau\le T\,$ and $\,0\,$ for
$\,\tau\ge T+1$. Noting
that $\,\fw(\tau)\,$ is an even function of $\,\tau$,
we write
$$\multline
\hfill\zeta_1\cdot\fs_{\ve}(t)\cdot\zeta_2 =
2\,\int\limits_0^{\infty}
\bk(t,\tau) e^{-\ve\tau^2}\,\vartheta_0(\tau)\,
\zeta_1\cdot\fw(\tau)\cdot\zeta_2\,d\tau\hfill \\
\hfill + 2\,\int\limits_0^{\infty}
\bk(t,\tau) e^{-\ve\tau^2}\,(1-\vartheta_0(\tau))\,
\zeta_1\cdot\fw(\tau)\cdot\zeta_2\,d\tau
\,=\,\fs^{(0)}_{\ve}(t)\,+\,\fs^{(1)}_{\ve}(t).\hfill
\endmultline
$$
The integral $\,\fs^{(0)}_{\ve}(t)\,$ converges to
$\,\fs^{(0)}(t)=2\,\int\limits_0^{\infty}
\bk(t,\tau)\,\vartheta_0(\tau)\,
\zeta_1\cdot\fw(\tau)\cdot\zeta_2\,d\tau\,$ in the strong operator
topology in any of the spaces $\,H_s$, as is obvious.
To treat $\,\fs^{(1)}_{\ve}(t)$, we use (5). The part of
$\,\fs^{(1)}_{\ve}(t)\,$ corresponding to $\,\tilde{W}_N\,$
in (5a) can be reguralized by repeatedly using the equality
$$
\exp\{({i\over 4t}-\ve) \tau^2\}=
\{ 2\tau({i\over 4t}-\ve)\}^{-1} {\partial\hfill\over\partial\tau}
\exp\{({i\over 4t}-\ve) \tau^2\},
$$
and integration by parts.
The other terms in (5a) lead to the integrals of the form
$$
\int\limits_0^{\infty}
\bk(t,\tau) e^{-\ve\tau^2}\,e^{-i\lambda_j\tau}\,
\phi_j(\tau)\,(1-\vartheta_0(\tau))\,d\tau\,W_j,
$$
where $\,\text{Im}\,\lambda_j\ge 0$. We treat these integrals
essentially in the same fashion as textbooks handle the
convergence of $\,\int_{-\infty}^{+\infty} e^{i\tau^2}\,d\tau\,$
(by moving the contour, etc.).
Again, integration by parts and use of (5b) take care of
regularization.
The same technique works to show that, given
$\,f\in H_s^{\text{\rm com\/}}$,
$$
\lim_{\ve\searrow 0}\,\int\limits_0^{\infty}
\big( \fs_{\ve}(t)f,\,(\Dt + A^*) \eta(t)\big)\,dt\,=\,
{1\over i}\,\big( f, \eta(0)\big),
$$
for all
$\,\eta\in C^{\infty}_0 (\Bbb R; H_{-s+2}^{\text{\rm com\/}})\,$
that vanish for large $\,t$.
The smoothing property is proved following the computation (4b)
except for that we start with $\,t^2\,A\,\fs_{\ve}(t)\,$
and then at the final step we let $\,\ve\searrow 0$.
We skip the details.
\qed\enddemo
\vglue 1pc
\remark{Remark} Note that we make no assumption that the operator
$\,A\,$ be self-adjoint. It is known, however,
that if $\,A\,$ is not self-adjoint,
the initial value problem (2) may turn out to be ill-posed,
at least in the $\,L^2$-based Sobolev spaces,
see \cite{Tak}, \cite{Mi}, and \cite{Ich}, for the
discussion of necessary and of sufficient conditions
for well-posedness. Moreover, if
$\,A\,$ differs from $\,A^*\,$
by an operator of order $\ge 1$, then, in general,
there are no energy estimates available for problem (2),
and already the existence of solutions becomes problematic.
{\it Nonetheless\/}, our Proposition 1 provides a construction
for the solutions corresponding to the initial data
of compact support.
\endremark
\vglue 1pc
Let us now turn to examples. The
Hypothesis ($\bigstar$) is evidently the most
restrictive ingredient of our approach. That hypothesis is
motivated by the uniform {\it local energy decay\/} theorems
for the wave equation in ``non-trapping" exterior domains.
The systematic study of that phenomenon has been initiated
by C. S. Morawetz, and P. D. Lax and R. S. Phillips,
and taken up by many other mathematitians
(see, e.g., \cite{Mo}, \cite{LP}, \cite{LMP}, \cite{Ral}, \cite{V},
\cite{S}, \cite{Rau 1}, \cite{Tay}, \cite{Me}, \cite{MS}).
In the situations
that were studied first, Hypothesis ($\bigstar$)
holds with functions
$\,\phi_j\,$ and exponents $\,k_j\in\Bbb C\,$ such that all
$\,|D_\tau^{\ell}\phi_j(\tau) e^{-i k_j \tau}|\,$
do actually decay as $\,\tau\to+\infty$. Consider the following example.
Let $\,\Omega\subset\Bbb R^n$, $\,n\ge 2$,
be a non-compact connected domain
with compact, $\,C^{\infty}$ boundary
$\,\partial\Omega$. Let $\,S(t,x,y)\,$ be the Schwartz kernel
of the solution operator
$\,S(t):\,f(\cdot)\mapsto u(t,\cdot)\,$ corresponding to the
initial boundary-value problem
$$\align
{1\over i}\,{\partial\hfill\over\partial t} u(t,x)
- \Delta u(t,x) & = 0,\;\quad\qquad t\in\Bbb R,\;x\in\Omega,\tag 8a \\
\Cal B u(t,x) & = 0,\;\quad \qquad x\in\partial\Omega,\tag 8b \\
u(0,x) & = f(x), \qquad x\in\Omega\,,\tag 8c
\endalign
$$
where $\,\Cal B\,$ is the boundary operator corresponding to
either the Dirichlet or Neumann boundary condition.
\proclaim{Corollary 2} Assume that $\,\Omega\,$ is
{\sl non-trapping\/} in the sense that for every sufficiently large
$\,R>0\,$ there exists $\,T_R>0\,$ such that no generalized geodesic
of length $\,T\ge T_R\,$
lies completely within the ball $\,\{|x|\le R\}$. (The generalized
geodesics are the projections into $\,\Omega\,$ of the generalized
bicharacteristics, see \cite{MS 2}, Definition 3.1.)
Then the fundamental solution S(t,x,y) of problem (8)
is $\,C^{\infty}\,$ in
$\,(0,\infty)\times{\overline{\Omega}}\times{\overline{\Omega}}$.
\endproclaim
\demo{Proof} We choose the scale of Hilbert spaces generated
from $\,H_0 = L^2(\Omega)\,$ by the powers of the square root
of the operator $\,A$, which, in our case, is $\,(-\Delta)\,$
with the boundary condition $\,\Cal B$. The set $\,\Upsilon\,$
is comprised of the restrictions to $\,\Omega\,$ of the functions
from $\,C^{\infty}_0(\Bbb R^n)$. That Hypothesis ($\bigstar$)
is true follows from \cite{Me},
\cite{Ral 2}, and \cite{V 3}.
In particular, the functions
$\,\phi_j(\tau),\;j=1,\dots,N-1$, in (5a) can be omitted, i.e. $N=1$.
Given $\,s_1,\,s_2$, the $\,H_{s_1}\to H_{s_2}$-norm of
$\,{d^{\ell}\over d\tau^{\ell}}\tilde{W}_1(\tau)\,$ behaves like
$\,{d^{\ell}\over d\tau^{\ell}}\tilde{\phi}_1(\tau)\,$ for large $\,\tau$,
where $\,\tilde{\phi}_1(\tau)= e^{-r \tau}\,$ with $\,r>0$, in the case
of odd $\,n$, and $\,\tilde{\phi}_1(\tau) = \tau^{-n}$, if
$\,n\,$ is even.
\qed\enddemo
Our next example is the Schr\"odinger equation with variable
coefficients in $\,\Bbb R^n$, $\,n\ge 1$. Let
$$
A\,=\,-\,a^{j,k}(x)\,{\hfill\partial^2\hfill\over\partial x_j
\partial x_k}\,+\,
b^j(x)\,{\partial\hfill\over\partial x_j}+b^0(x)\,,\tag 9a
$$
with summation over the repeated indices assumed.
The coefficients $\,a^{j,k}$, $\,b^j\,$ and $\,b^0\,$
are supposed to be infinitely differentiable, the matrix
$\,\{a^{j,k}(x)\}\,$ is real, symmetric and positive definite,
uniformly with respect to $\,x\in\Bbb R^n$.
The coefficients $\,b^j\,$ and $\,b^0\,$ are, in general,
complex-valued. In addition, we assume that
$\,a^{j,k}$, $\,b^j\,$ and $\,b^0\,$
stabilize outside some ball:
$$
a^{j,k}(x) = a^{jk}_{\infty},\quad
b^j(x) = b^j_{\infty},\quad
b^0(x) = b^0_{\infty},
\quad\text{for}\quad |x|\ge R.\tag 9b
$$
Let $\,(q(s;x,\xi),\,p(s;x,\xi))\,$ be the Hamiltonian
trajectory corresponding
to the Hamilton function $\,H(q,p)=a^{j,k}(q)p_jp_k\,$ and
starting at $\,q(0;x,\xi)=x$, $\,p(0;x,\xi)=\xi$.
\proclaim{Corollary 3} Assume that $\,|q(s;x,\xi)|\to\infty\,$
as $\,s\to +\infty$, for all
$\,x\in\Bbb R^n\,$ and $\,\xi\in\Bbb R^n\setminus 0\,$
({\sl the non-trapping condition\/}).
Then there exists a one-parameter family
$\,S_{{}_{A}}(t)$, $\,00$, if
$\,n\,$ is odd,
and $\,\tilde{\phi}_1(\tau) = \tau^{-n}$, if
$\,n\,$ is even. This follows from
\cite{V 4}, chapters 9 and 10, and \cite{Rau 1};
for the case of the formally selfadjoint
$\,A\,$ see also \cite{PoS}.
\qed\enddemo
A generalization to systems of Schr\"odinger equations is
straightforward. Consider, for example, the following
version of the Schr\"odinger evolution in a locally curved
space with a gauge magnetic field.
Let $\,\bold g = g_{jk}(x)\,dx_j\,dx_k\,$ be a smooth Riemannian
metric in $\,\Bbb R^n$, flat outside a ball:
$\,g_{jk}(x)=\delta_{jk}$, if $\,|x|\ge R>0$. Let
$\,A_1(x),\dots,\,A_n(x)\,$ be a given $\,n$-tuple of smooth
functions of $\,x\,$ with values in the Lie algebra
$\,g\ell (m,\Bbb C)\,$
of all complex $\,m\times m$ - matrices, $\,m\ge 1$.
We assume that $\,A_1(x),\dots,\,A_n(x)\,$ are constant outside
of the ball
$\,|x|\le R\}\,$ and satisfy there the following commutator
condition:
$$
[\,A_j+A_j^\dag\,,\,A_k+A_k^\dag\,]\,=\,0,\qquad\forall j,k,\tag 10a
$$
where $\,A_j^\dag\,$ is the Hermite conjugate of $\,A_j$.
The Schr\"odinger field will be represented by
a column-vector $\,\vec\psi(t,x)\in\Bbb C^m$. We denote
$\,\nabla_j =\, i\,{\partial\hfill\over\partial x_j} + A_j(x)\,$
the $\,j$-th covariant derivative, and $\,\nabla_j^*\,$ - its
conjugate with respect to the $\,L^2$-product
$\,(\vec \xi,\vec\eta)=
\int g^{jk}(x)\,\xi_j(x)\, \overline{\eta_k(x)}\,\sqrt{g}\,dx$.
Finally, denote
$$
A\,=\,\frac12\,\sum\limits_{j=1}^n
(\,\nabla_j^*\,\nabla_j \,+\nabla_j\,\nabla_j^* \,),\tag 10b
$$
and consider the Cauchy problem
$$
\Dt \vec\psi + A\vec\psi = 0,
\quad \vec\psi\, |_{{}_{t=0}}= \vec\psi^{\,0}.\tag 10c
$$
\proclaim{Corollary 3} If the metric $\,\bold g\,$ has no
trapped geodesics then the (matrix-valued) fundamental solution
$\,S_{{}_{\Dt + A}}(t,x,y)$, corresponding to {\rm (10c)\/}, is
infinitely differentiable for $\,t>0$, $\,x,y\in\Bbb R^n$.
\endproclaim
%$\hfill\square$%
\vglue 1pc
\noindent{\bf Remark 4.\/} In many cases,
including those considered above for $\,n\ge 3$,
the localized solving operator for the hyperbolic problem,
$\,\zeta_2\cdot\fw(\tau)\cdot\zeta_1$, does, in fact, decay
as $\,\tau^{-3}\,$ when $\,\tau\to\infty$, which makes the justification of
the abstract scheme, given in the opening paragraphs,
quite trivial. Also, then we get
a simpler representation for $\,\zeta_2\cdot\fs(t)\cdot\zeta_1$,
namely,
$$
\zeta_1\cdot \fs(t)\cdot\zeta_2
= \int \bk(t,\tau)\,
\zeta_1\cdot \fw(\tau)\cdot\zeta_2\,d\tau.\tag 7b
$$
As a serendipitous consequence, we obtain the following {\it time-decay\/}
result for the perturbed Schr\"odinger evolution:
$$
\|\zeta_1\cdot \fs(t)\cdot\zeta_2\|_{{}_{H_s\to H_s}}\,
\le\,{c(s,\zeta_1,\zeta_2)\over \sqrt{t}},\tag 11
$$
for all $\,s\,$ and any pair of cut-off functions $\,\zeta_{1,2}$.
In particular, {\it if $\,n\ge 3$, then at least
$\,1/\sqrt t$-decay rate holds for
each of the three examples considered above.\/}
The time-decay for the solutions of
$\,\Dt\psi-\Delta\psi+V \psi=0\,$ has been studied by
several authors for different classes of the potentials $\,V$,
see \cite{Ste}, \cite{Rau 2}, \cite{J}, \cite{JK}, \cite{Mu}.
It has been shown, in particular, that the $\,1/\sqrt t$-decay
is generic, while for special classes of potentials
the rate of decay may be higher.
\vglue 1pc
\noindent{\bf Concluding remarks.\/} Thus we have shown that
the simple relation (1) between the Schr\"odinger and wave evolutions
allows us to prove the smoothness of the fundamental solution for
the Schr\"odinger equation, and also, the local time decay.
The method works in the cases where the local energy decay for
the hyperbolic evolution is known. We think that (1),
though transparent, is important.
It should be mentioned that (1) was implicitly used
in \cite{BdeM} in the study of the propagation of singularities
for equations with double characteristics.
Comparing our results with those of Craig, Kappeler and Strauss,
\cite{CKS}, one notices that we avoid microlocal statements. The
reason is that there is no microlocal version of the
local energy decay for hyperbolic problems available.
\proclaim{Question} Under what conditions on the operator
{\rm (9a)\/} and the {\sl pseudo-differential operators\/}
$\,\zeta_{1,2}(x,D_x)\,$ does the operator
$\,\zeta_2\cdot\fw(\tau)\cdot\zeta_1\,$ satisfy {\rm (5)\/}
for large $\,\tau$?
\endproclaim
We conjecture that at least the following holds.
Let $\,K\subset\Bbb R^n_x\,$ be a compact set and
$\,\Gamma\subset\Bbb R^n_{\xi}\,$ be a cone.
Let $\,A\,$
be a uniformly elliptic operator of the form (9a) with smooth
coefficients that stabilize at infinity so that
$\,A = - \Delta\,$ for $\,|x|\ge R>0$.
Denote by
$\,(q(s;x,\xi), p(s;x,\xi))$,
as before, the Hamilton trajectories
corresponding to $\,H(q,p)= a^{j,k}(q) p_j p_k$.
We say that the conic set
$\,K\times\Gamma\,$ is {\it non-trapping\/} if all the rays
starting from the points in $\,K\times(\Gamma\setminus 0)\,$
go to infinity: $\,|q(t;x,\xi)|\to\infty\,$ as $\,t\to\infty$.
Let $\,\upsilon(x,\xi)\,$ be a (full) symbol from
$\,S^0(\Bbb R^n_x\times\Bbb R^n_{\xi})$, which has support
in the conic set $\,K\times\Gamma$. Denote
$\,\zeta_1(x,D_x)= \text{Op}\,\upsilon$.
Let $\,\zeta_2\in C^{\infty}_0(\Bbb R^n)$.
As before,
$\,\fw(\tau) :\,f\mapsto w(\tau)\,$ will be the solution operator
of the hyperbolic problem
$\,w_{\tau\tau}+ A w = 0$, $\,w(0)=f$, $\,w_\tau(0)=0$.
\proclaim{Conjecture} If $\,K\times\Gamma\,$ is non-trapping, then for all
sufficiently large $\,t>0\,$ the operator
$\,\zeta_2\cdot\fw(\tau)\cdot\zeta_1\,$
is smoothing and
its $\,\text{Hom} (H^{s_1}, H^{s_2})$-norms decay
as $\,\tau\to\infty$.
\endproclaim
\noindent{\bf ACKNOWLEDGMENTS.\/} The authors are grateful
to Walter Craig and Walter Strauss for providing the manuscript of
the paper \cite{CKS} and for sharing the enthusiasm about the
Schr\"odinger equation.
\Refs
\widestnumber\key{CFKS}
\ref\key BdeM\by L. Boutet de Monvel
\paper Propagation des singularit\'es des solutions
d'\'equations analogues \`a l'\'equa-tion de Schr\"odinger
\inbook Fourier Integral Operators
and Partial Differential Equations
\bookinfo (Lecture Notes in Math., No. 459)
\publ Springer-Verlag
\publaddr Berlin-Heidelberg-New York
\yr 1975
\endref
\ref\key CKS\by W. Craig, T. Kappeler \& W. Strauss
\paper Microlocal dispersive smoothing for
the Schr\"odinger equation
\paperinfo draft of a paper\yr March 1994
\endref
\ref\key CFKS\by H. L. Cycon, R. G. Froese, W. Kirsch, and B. Simon
\book Schr\"odinger operators
\publ Springer-Verlag
\publaddr Berlin $\cdot$ Heidelberg
\yr 1987
\endref
\ref\key I 1\by W. Ichinose
\paper The Cauchy problem for Schr\"odinger type equations
with variable coefficients
\jour Osaka J. Math.
\vol 24 \yr 1987\pages 853-886
\endref
\ref\key I 2\by W. Ichinose
\paper On $L^2$ well posedness of the Cauchy problem
for Schr\"odinger type equations on the Riemannian manifold
and the Maslov theory
\jour Duke Math. J.
\vol 56\issue 3\pages 549-588\yr 1988
\endref
\ref\key J 1\by A. Jensen
\paper Local decay in time of solutions to Schr\"odinger's
equation witha dilation-analytic interaction
\jour Manuscripta Math.
\vol 25
\yr 1978
\pages 61-77
\endref
\ref\key J 2\by A. Jensen
\paper Spectral properties of Schr\"odinger operators
and time-decay of the wave functions. Results in
$\,L^2(\Bbb R^m),\,m\ge 5$
\jour Duke Math. Journal
\vol 47
\yr 1980
\pages 57-80
\moreref Results in
$\,L^2(\Bbb R^4)$:
\jour J. Math. Anal. Appl.
\vol 101
\yr 1984
\pages 397-422
\endref
\ref\key JK\by A. Jensen and T. Kato
\paper Spectral properties of Schr\"odinger operators
and time-decay of the wave functions
\jour Duke Math. Journal
\vol 46
\yr 1979
\pages 583-611
\endref
\ref\key KS\by L. Kapitanski and Yu. G. Safarov
\paper Global fundamental solution of the Schr\"odinger evolution
equation
\paperinfo in preparation
\endref
\ref\key Ki 1\by H. Kitada
\paper On a construction of the fundamental solution for Schr\"odinger
equations
\jour J. Fac. Sci. Univ. Tokyo
\vol IA 27 \pages 193-226
\yr 1980
\endref
\ref\key Ki 2\by H. Kitada
\paper Fundamental Solutions and eigenfunction expansions
for Schr\"odinger operators: I. Fundamental solutions
\jour Math. Z.\vol 198\pages 181-190\yr 1988
\endref
\ref\key LP \by P. D. Lax and R. S. Phillips
\book Scattering theory
\publ Academic Press
\publaddr New York$\,\cdot\,$Boston
\bookinfo 1st edition: 1967,
2nd edition: 1989
\endref
\ref\key LMP\by P. D. Lax, C. S. Morawetz, and R. S. Phillips
\paper Exponential decay of solutions of the wave equation in
exterior of a star-shaped obstacle
\jour Comm. Pure Appl. Math.
\vol 16 \pages 477-486 \yr 1963
\endref
\ref\key MF\by V. P. Maslov and M. V. Fedoriuk
\book Semi-classical approximation in quantum
mechanics
\publ D. Reidel Publ. Co.
\publaddr Dordrecht, Holland
\yr 1981
\endref
\ref\key Me \by R. B. Melrose
\paper Singularities and energy decay in acoustical
scattering
\jour Duke Math. J.\vol 46\issue 1\pages 43-59\yr 1979
\endref
\ref\key MS\by R. B. Melrose and J. Sjostrand
\paper Singularities of boundary problems, \rom{I}
\jour Comm. Pure Appl. Math.
\vol 31 \pages 593-617 \yr 1978
\moreref\rom{II}
\jour Comm. Pure Appl. Math.
\vol 35 \pages 129-168 \yr 1982
\endref
\ref \key Mi\by S. Mizohata
\book On the Cauchy problem
\bookinfo Notes and reports in Mathematics in Science and Engineering, 3
\publ Academic Press, Inc., Orlando, FL, and Science Press
\publaddr Beijing
\yr 1985
\endref
\ref\key Mo 1\by C. S. Morawetz
\paper The decay of solutions of the exterior initial-boundary
value problem for the wave equation
\jour Comm. Pure Appl. Math.
\vol 14 \pages 561-568 \yr 1961
\endref
\ref\key Mo 2\by C. S. Morawetz
\paper The limiting amplitude principle
\jour Comm. Pure Appl. Math.
\vol 15 \pages 349-361 \yr 1962
\endref
\ref\key Mo 3\by C. S. Morawetz
\paper Exponential decay of solutions of the wave equation
\jour Comm. Pure Appl. Math.
\vol 19 \pages 439-444
\yr 1966
\endref
\ref\key Mo 4\by C. S. Morawetz
\paper Time decay for the nonlinear Klein-Gordon equation
\jour Proc.Roy.Soc.\vol A. 306\pages 291-296 \yr 1968
\endref
\ref\key Mo 5\by C. S. Morawetz
\paper Decay for solutions of the exterior problem
for the wave equation
\jour Comm. Pure Appl. Math. \vol 28
\pages 229-264 \yr 1975
\endref
\ref\key MRS\by C. S. Morawetz, J. V. Ralston, and W. Strauss
\paper Decay of solutions of the wave equation outside
nontrapping obstacles
\jour Comm. Pure Appl. Math.
\vol 30 \pages 447-508 \yr 1977
\endref
\ref\key Mu\by M. Murata
\paper Rate of decay of local energy and spectral properties
of elliptic operators
\jour Japan J. Math.
\vol 60
\yr 1980
\pages 77-127
\endref
\ref\key OF\by T. A. Osborn and Y. Fujiwara
\paper The evolution kernels: uniform asymptotics expansions
\jour J. Math. Phys.
\vol 24 \issue 5
\yr 1983
\pages 1093-1103
\endref
\ref\key PoS\by G. S. Popov and M. A. Shubin
\paper Asymptotic expansion of the spectral function
for second-order elliptic operators in $\Bbb R^n$
\jour Functional Anal. Appl.
\vol 17\issue 3\pages 193-200 \yr 1983
\endref
\ref\key Ral 1\by J. Ralston
\paper Solutions of the wave equation with localized energy
\jour Comm. Pure Appl. Math. \vol 22
\pages 807-823 \yr 1969
\endref
\ref\key Ral 2\by J. Ralston
\paper Note on the decay of acoustical waves
\jour Duke Math. J. \vol 46 \pages 799-804
\yr 1979
\endref
\ref\key Rau 1\by J. Rauch
\paper Asymptotic behavior of solutions to hyperbolic
partial differential equations with zero speeds
\jour Comm. Pure Appl. Math.
\vol 31\pages 431-480\yr 1978
\endref
\ref\key Rau 2\by J. Rauch
\paper Local decay of scattering solutions
to Schr\"odinger's equation
\jour Comm. Math. Phys.
\vol 61
\yr 1978
\pages 149-168
\endref
\ref\key Ste \by S. Steinberg
\paper Local time decay for solutions of the Schr\"odinger
equation and the wave equation
\jour Archive for Rational Mech. Anal.
\vol 54
\issue 2
\yr 1974
\pages 134-147
\endref
\ref\key Str \by W. A. Strauss
\paper Dispersal of waves vanishing on the boundary
of an exterior domain
\jour Comm. Pure Appl. Math.
\vol 28\pages 265-278\yr 1975
\endref
\ref\key Tak \by J. Takeuchi
\paper A necessary condition for the well-posedness
of the Cauchy problem for a certain class of evolution equations
\jour Proc. Japan Acad. Ser.A Math. Sci.
\vol 50\pages 133-137\yr 1974
\endref
\ref\key Tay \by M. Taylor
\paper Grazing rays and reflection of singularities
of solutions to wave equations
\jour Comm. Pure and Appl. Math.
\vol 29
\pages 1-38 and 463-481
\yr 1976
\moreref\rom{II} (systems)
\jour Comm. Pure and Appl. Math.
\vol 29
\pages 463-481
\yr 1976
\endref
\ref\key V 1\by B. R. Vainberg
\paper Behavior of the solution of the Cauchy problem
for a hyperbolic equation as $\,t\to\infty$
\jour Math. USSR-Sbornik
\vol 7\issue 4 \yr 1969\pages 533-567
\endref
\ref\key V 2\by B. R. Vainberg
\paper On exterior elliptic problems polynomially depending
on a spectral parameter, and the asymptotic behavior
for large time of solutions of nonstationary problems
\jour Math. USSR-Sbornik
\vol 21\issue 2\yr 1973 \pages 221-239
\endref
\ref\key V 3\by B. R. Vainberg
\paper On the short wave asymptotic behavior of solutions
of stationary problems and the asymptotic behavior as
$\,t\to\infty\,$ of solutions of non-stationary
problems
\jour Russian Math. Surveys
\vol 30\issue 2\pages 1-58\yr 1975
\endref
\ref\key V 4\by B. R. Vainberg
\book Asymptotic methods in equations of mathematical
physics
\lang Russian
\publ Moscow University Press
\yr 1982
\transl\nofrills English translation:
\publ Gordon and Breach Science Publishers
\publaddr New York - London
\yr 1989
\endref
\ref\key We\by A. Weinstein
\paper A symbol class for some Schr\"odinger equations
on $\,\Bbb R^n$
\jour Amer. Journ. Math.\vol 107\pages 1-23\yr 1985
\endref
\ref\key Wi \by C. Wilcox
\paper The initial-boundary value problem for the wave equation
in an exterior domain with spherical boundary
\jour Amer. Math. Soc. Not., Abstract No.564-20
\vol 6
\yr 1959
\endref
\ref\key Za \by E. C. Zachmanoglou
\paper The decay of solutions of the initial-boundary value
problem
for hyperbolic equations
\jour J. Math. Anal. Appl. \vol 13
\pages 504-515 \yr 1966
\endref
\ref\key Ze\by S. Zelditch
\paper Reconstruction of singularities for solutions of
Schr\"odinger equations
\jour Comm. Math. Phys. \vol 90\yr 1983\pages 1-26
\endref
\endRefs
\vfill
\pagebreak
\end