A Poisson Formula for the Wave Propagator on Schwarzschild-de Sitter Backgrounds
Extracted source text · arXiv 2512.16054
\documentclass[reqno,12pt,letterpaper]{amsart}
\usepackage{amsmath,amssymb,amsthm,graphicx,mathrsfs,url}
\usepackage[usenames,dvipsnames]{color}
\usepackage[colorlinks=true,linkcolor=Red,citecolor=Green]{hyperref}
\usepackage{amsxtra,esint}
\usepackage{mathtools,wasysym}
%\usepackage{showkeys} %only for draft mode
\mathtoolsset{showonlyrefs,showmanualtags}
\usepackage{float}
\usepackage[font=small,labelfont=bf]{caption}
\usepackage{hyperref,cleveref,amssymb,mathtools}
\usepackage{dsfont}
\usepackage[font=small,labelfont=bf]{caption}
\usepackage[style=alphabetic, backend=biber,doi=false,isbn=false,url=false,eprint=false,giveninits=true,maxbibnames=99]{biblatex}
\addbibresource{references.bib}
\newcommand{\mar}[1]{{\marginpar{\sffamily{\tiny #1}}}}
\setlength{\marginparwidth}{0.6in}
\def\?[#1]{\textbf{[#1]}\marginpar{\Large{\textbf{??}}}}
\def\smallsection#1{\smallskip\noindent\textbf{#1}.}
\setlength{\textheight}{8.50in} \setlength{\oddsidemargin}{0.00in}
\setlength{\evensidemargin}{0.00in} \setlength{\textwidth}{6.08in}
\setlength{\topmargin}{0.00in} \setlength{\headheight}{0.18in}
\setlength{\marginparwidth}{1.0in}
\setlength{\abovedisplayskip}{0.2in}
\setlength{\belowdisplayskip}{0.2in}
\setlength{\parskip}{0.05in}
\renewcommand{\baselinestretch}{1.10}
\newcommand{\indentalign}{\hspace{0.3in}&\hspace{-0.3in}}
\newtheorem*{theorem*}{Theorem}
\newtheorem{theorem}{Theorem}
\newtheorem{prop}{Proposition}[section]
\newtheorem{defi}[prop]{Definition}
\newtheorem{conj}{Conjecture}
\newtheorem{lemma}[prop]{Lemma}
\newtheorem{corr}[prop]{Corollary}
\newtheorem{rem}{Remark}[section]
\newtheorem{ex}{Example}[section]
\newtheorem{claim}{Claim}[section]
\numberwithin{equation}{section}
\newtheorem{question}{Question}
\newtheorem{hyp}{Hypothesis}
\newcommand{\abs}[1]{\left | #1 \right| }
\DeclareMathOperator{\tr}{Tr}
\DeclareMathOperator{\Id}{Id}
\newcommand*{\dd}{\mathop{}\!\mathrm{d}}
\DeclareMathOperator{\Tr}{Tr}
\DeclareMathOperator{\Spec}{Spec}
\DeclareMathOperator{\sech}{sech}
\DeclareMathOperator{\diag}{diag}
\DeclareMathOperator{\supp}{supp}
\DeclareMathOperator{\dist}{dist}
\DeclareMathOperator{\Span}{Span}
\DeclareMathOperator{\Op}{Op}
\DeclareMathOperator{\Res}{Res}
\newcommand{\ip}[2]{\left \langle#1,#2 \right \rangle}
\newcommand{\mat}[1]{\begin{pmatrix} #1 \end{pmatrix}}
\newcommand{\pr}[1]{ \left ( #1 \right ) }
\newcommand{\set}[1]{ \left \{ #1 \right \}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\p}{\partial}
\newcommand{\dbar}{\overline{\p}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\T}{\mathbb{T}}
\newcommand{\E}{\mathbb{E}}
\newcommand{\e}{\varepsilon}
\renewcommand{\epsilon}{\varepsilon}
\newcommand{\de}{\delta}
\newcommand{\R}{\mathbb{R}}
\newcommand{\Q}{\mathbb{Q}}
\renewcommand{\SS}{\mathbb{S}}
\newcommand{\norm}[1]{ \left \| #1 \right \| }
\newcommand{\jpb}[1]{ \left \langle #1 \right \rangle }
\newcommand{\C}{\mathbb{C}}
\renewcommand{\phi}{\varphi}
\renewcommand{\P}{\mathbb{P}}
\newcommand{\F}{\mathcal{F}}
\newcommand{\1}{\mathds{1}}
\newcommand{\comment}[1]{%
\text{\phantom{(#1)}} \tag{#1}
}
\renewcommand{\Re}[1]{{\rm{Re}} \left ( #1\right ) }
\renewcommand{\Im}[1]{{\rm{Im}} \left ( #1 \right ) }
\newcommand{\quest}[1]{{\color{red}(#1)}\index{question} \marginpar{\Large{ \color{red} \textbf{??}}}}
\newcounter{step}
\newcommand{\Step}{
\refstepcounter{step}
\noindent \textbf{Step \thestep.}
}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\title[Poisson formula for SdS]{A Poisson formula for the wave propagator on Schwarzschild-de Sitter backgrounds}
\author{Izak Oltman}
\address[Izak Oltman]{Department of Mathematics, Northwestern University, 2033 Sheridan Rd, Evanston, IL 60208}
\email{ioltman@northwestern.edu}
\author{Ben Pineau}
\address[Ben Pineau]{Courant Institute for Mathematical Sciences\\
New York University
} \email{brp305@nyu.edu}
\begin{document}
\begin{abstract}
This paper proposes a Poisson formula for the wave propagator of the Schwarzschild--de Sitter (SdS) metric.
That is done by proving a Poisson formula relating wave propagators and scattering resonances for a class of non-compactly supported potentials on the real line.
That class includes the Regge--Wheeler potentials obtained from separation of variables for SdS.
The novelty lies in allowing non-compact supports -- all exact Poisson formulae of Lax--Phillips, Melrose, and other authors required compactness of the support of the perturbation.
\end{abstract}
\maketitle
\section{Introduction}
{Quasinormal modes of black holes are supposed to measure the ringdown of gravitational waves -- see \cite{berti2009quasinormal} for a survey, \cite{cardoso2018quasinormal,jaramillo2021pseudospectrum,yang2025black} for more recent work and references.
Their study has been of interest to mathematicians since the work of Bachelot--Motet-Bachelot \cite{BMB} -- see Hintz--Xie \cite{hintz2021quasinormal},
Hitrik--Zworski \cite{hitrik2024overdamped}, and J\'ez\'equel \cite{jezequel2022upper} for some recent advances. }
{The quasinormal modes\footnote{We follow the informal convention, common in math and physics, of referring to both the frequencies and the corresponding states as modes; the former should be more correctly called quasinormal frequencies.} can be considered as scattering resonances for linear wave equations arising in general relativity -- see \cite[Chapter 5]{dyatlov2019mathematical} and references given there. One of the features of the theory of scattering resonances, going back to the work of Lax--Phillips, are Poisson formulas. They generalized the Poisson formula for the wave equation on compact Riemannian manifolds $ ( M, g ) $,
\begin{equation}
\label{eq:Poisson_comp}
\tr \cos t \sqrt { - \Delta_g } = \tfrac12 \sum_{ \lambda_j^2 \in \Spec ( - \Delta_g ) }
e^{ i \lambda_j t } .
\end{equation}
Here $ - \Delta_g $ is the Laplace--Beltrami operator for the metric $ g $ and
the traces are meant in the sense of distributions in $ t $. The formula \eqref{eq:Poisson_comp} is, of course, an immediate consequence of the spectral decomposition of $ - \Delta_g $.
In the non-compact setting, the left-hand side
is not distributionally of trace class and it has to be renormalized.
The right-hand side should then be replaced by a sum over scattering resonances with terms $ e^{ - i \lambda_j |t| } $ (since $ \Im {\lambda_j } < 0$) and the formula is typically not valid at $ t = 0 $ -- see \cite[\S\S 2.6, 3.10, 7.4]{dyatlov2019mathematical} for an introduction to this subject and references. }
{To investigate a possible Poisson formula for quasinormal modes of black holes, that is, a relation between (renormalized) traces of wave groups and sums over the modes, we consider the simplest case of Schwarzschild--de Sitter black holes.
There, a standard Regge--Wheeler reduction gives quasi-normal modes as the union of scattering resonances of a family of potentials parametrized by angular momenta.
}
Motivated by this, we compute here the trace of the propagator for the wave equation $\p_t^2 - \p_x^2 + V(x)$ for $ V $ which include the Regge--Wheeler potentials and their perturbations -- see \S \ref{s:RW}.
\subsection{Trace formula for a class non-compactly supported potentials.}
\label{s:trace}
We consider potentials satisfying the following hypothesis.
There exist constants $ A_\pm > 0 $ and $ R>0$ such that
\begin{equation}
\label{hyp:1}
V ( x ) |_{ \pm x > R } = F_\pm (\exp ( \mp x A_\pm ) ),
\end{equation}
where $ z \mapsto F_\pm ( z ) $ are holomorphic in a neighborhood of $ z = 0$ and $F_\pm (0)= 0$.
In particular this means there exist $\set{V^\pm_j \in \R : j\in \Z_{\ge 1}}$ and constants $A,C>0$ so that
\begin{equation*}
V(x)|_{ \pm x > C } = \sum_{j=1}^\infty \left(e^{\mp xA_\pm} \right)^jV^\pm_j
\end{equation*}
with $\abs{V^\pm _j } \le A^j$.
The first result, which is implicit in \cite[\S 3]{dyatlov2011quasi} and earlier works, is the meromorphy of the resolvent of the corresponding Schr\"odinger operator (Green function):
\begin{prop}
\label{p:0}
The resolvent of $ P_V \coloneq D_x^2 + V ( x ) $, $D_x \coloneq \p_x /i ,$
\[ R_V ( \lambda ) \coloneq ( P_V - \lambda^2 )^{-1} \colon L^2 \to L^2, \ \ \ \Im
\lambda \gg 1 , \]
continues meromorphically to
\[ R_V ( \lambda ) \colon L^2_{\rm{c}} ( \mathbb R) \to L^2_{\rm{loc} } ( \mathbb R), \ \ \ \lambda \in \mathbb C.\]
\end{prop}
In other words, for any $ f, g \in L^2_{\rm{c}} $,
$ \lambda \mapsto \langle R_V ( \lambda ) f , g \rangle $ is a
meromorphic function with poles independent of $ f $ and $ g $.
The poles are called {\em scattering resonances} and we denote their set (with elements included according to their multiplicity -- see \Cref{def:multi}) $ {\rm{Res}} (V) $.
For recent interesting developments in 1D scattering, see the work of Korotyaev \cite{Ko04,Ko05,Ko14}.
Define $\Box_V \coloneq \p_t^2 + P_V $ and let $U_V(t)\coloneq \cos (t\sqrt {P_V})$ be the cosine wave propagator with respect to $\Box_V$.
That is, if $u\in C_0^\infty (\R)$:
\begin{align}\label{eq:205}
\begin{cases}
\Box_V (U(t)u(x))(t,x) = 0, \\
U(0)u(x) = u(x),\\
(\p_t (U(t) u(x) )\big|_{t=0} = 0.
\end{cases}
\end{align}
Our main result relates the trace of the wave propagator minus the free wave propagator to a sum over resonances, analogous to \eqref{eq:Poisson_comp}.
\begin{theorem*}[Main result]
Suppose $V\in C^\infty (\R ;\R)$ satisfies \eqref{hyp:1} and $U_V(t)$ is the wave propagator for $\Box_V$.
Then for $t >0$,
\begin{align}
\Tr(U_V(t) - U_0(t)) = \tfrac{1}{2}\sum_{\lambda_j \in {\rm{Res}}(V)} e^{-i\lambda_j t} + \mathcal{A}_+ ( t ) + \mathcal{A}_ -( t ) - \tfrac 1 2, \label{eq:218}
\end{align}
in the sense of distributions. Here ${\rm{Res}}(V)$ is the set of resonances of $-\p_x^2 + V$ and
\[ \mathcal{A}_\pm ( t ) \coloneq -\frac{1}{2(e^{t A_\pm / 2} -1)} .\]
\end{theorem*}
\begin{ex}[P\"oschl-Teller potential]
The scattering matrix, $S(\lambda)$, for the P\"oschl-Teller potential
\begin{align}
V_{\rm{PT}}(x) \coloneq \frac{\ell^2 + 1/4}{\cosh^2 (x)}
\end{align}
can be explicitly computed \cite{cevik2016resonances} and the resonances are given by $$\set{\lambda_j^\pm = -i(j+1/2)\pm \ell : j\in \Z_{\ge 0}}.$$
The Birman--Kre\u{\i}n trace formula can be applied (exactly as will be done in \S \ref{s:birman}, but with the poles of $\frac{d}{d\lambda} \log \det S(\lambda)$ explicit).
It can then be shown that
\begin{align}
\Tr (U_{V_{\rm{PT}}}(t) - U_0(t)) &= \tfrac 1 2 \sum_{\lambda_j \in \Res( V_{\rm{PT}}) } e^{-i\lambda_j t } \underbrace{- \frac{1}{e^{t}-1}}_{\mathcal{A_+}+\mathcal A_-} - \tfrac 1 2 \\
&= \frac{\cos(\ell t)}{2\sinh(t/2)}- \frac{e^{-t/2}}{2\sinh(t/2)}- \tfrac 1 2\\
&=\tfrac{1}{2} \left(\frac{\cos(\ell t)-e^{-t/2}}{\sinh(t/2)} - 1\right), \ \ \ t > 0 . \label{eq:243}
\end{align}
We can compare this to a numerically computed approximation of the flat trace of $U_{V_{\rm{PT}}}(t)$, which is presented in \Cref{fig:theonlyfigure}.
\begin{figure}
\centering
\includegraphics[scale=.35]{fig1.jpg}
\caption{Comparison between a numerical approximation of the flat trace for the Pöschl–Teller potential (with $\ell = 1$) and the analytic expression in \eqref{eq:243}.
The numerical trace is computed in MATLAB by truncating and discretizing the spatial domain, evolving a normalized Gaussian centered at each grid point, $x_i$, under the cosine propagator $\cos\big(t\sqrt{P_{V_{\mathrm{PT}}}}\big)$, subtracting the free wave evolution of the same Gaussian, and evaluating at $x_i$.
This approximates $U_{V_{\rm{PT}}}(t,x_i,x_i)-U_{0}(t,x_i,x_i)$ (the Schwartz kernels).
Computing the sum of these terms, multiplied by $x_{i} - x_{i-1}$ will approximate the flat trace.}
\label{fig:theonlyfigure}
\end{figure}{
Note that as $t\to 0^+$, the right-hand side of \eqref{eq:243} goes to zero.
This is a classical result for compactly supported potentials (see Smith \cite{smith2023trace} and references given for a recent account).
}
\end{ex}
{When $ V$ is compactly supported, Proposition \ref{p:0} is classical -- see \cite[\S 2.2]{dyatlov2019mathematical}, and we have the Poisson formula
\begin{equation}
\label{eq:Melr} \Tr(U_V(t) - U_0(t)) = \tfrac{1}{2} {\rm{p.v.} } \sum_{\lambda_j \in {\rm{Res}}(V)} e^{-i\lambda_j |t|} - |{\rm{chsupp}} V | \delta_0 ( t ) - \tfrac12 , \end{equation}
see \cite[Theorem 2.21]{dyatlov2019mathematical}.
Both sides are defined distributionally on $ \mathbb R $ and $ \rm{p.v.} $ denotes the distributional principal value obtained by summing over $ |\lambda_j | \leq R $ and taking the $ R \to \infty $ limit.
The notation $ \rm{ch} $ denotes
the convex hull.
The presence of the $|{\rm{chsupp}} V | \delta_0 ( t ) $ term explains the need for the renormalization terms $ \mathcal{A}_\pm $ in the non-compact case.
For $ t > 0 $, \eqref{eq:Melr} was proved by Melrose \cite{melrose1982scattering} and for $ t > 2 {\rm{diam}}(\supp V ) $ by Lax--Phillips \cite{LaxPhillips1978}.}
The proof of such trace identities often relies on the Eisenbud–-Wigner delay operator, $\partial_\lambda S(\lambda) S(\lambda)^{-1}$.
In the compactly supported setting, this operator is linked to the Breit-–Wigner series $\sum_{\lambda_j \in \Res(V)\setminus{0}} \Im {\lambda_j} |\lambda_j|^{-2}$ via the Breit-–Wigner approximation (cf. \cite[Theorem 2.20]{dyatlov2019mathematical}).
For non-compactly supported potentials, however, the convergence or divergence of this approximation remains an open question; see Backus \cite{backus2020breit} for a thorough discussion.
If we define $\iota \colon (t,x)\mapsto (t,x,x)$, $\pi\colon (t,x)\mapsto t$, and denote the kernel of $U_V(t)$ by $U_V (t,x,y)$, as a distribution on $(0,\infty)\times\R^2$, {with the pairing using the Lebesgue measure}, then the \textit{flat trace} is defined as
\begin{align}
\tr^\flat U_V(t) \coloneq \pi _* \iota^*U_V \label{eq:flat_trace}
\end{align}
(which is well-defined thanks to the usual wave front set \cite[\S 29.1]{hormander4}).
Because $\tr^\flat U_0(t)=0$ and $\tr^\flat (U_V - U_0) = \tr(U_V-U_0)$, \eqref{eq:218} can be written
\begin{align}
\Tr^\flat U_V (t) = \tfrac{1}{2}\sum_{\lambda_j \in {\rm{Res}}(V)} e^{-i\lambda_j t} + \mathcal{A}_+ ( t ) + \mathcal{A}_ -( t ) -\tfrac{1}{2}.\label{eq:231}
\end{align}
\subsection{Regge--Wheeler potentials}
As a consequence of our main theorem, we propose a trace formula for the wave propagator in the Schwarzschild-de Sitter metric.
It necessarily involves a renormalization of the trace -- not the flat trace -- which can be used to pass from a trace formula for each Regge--Wheeler potential (parameterized by angular momenta), to a trace formula for the full potential.
Because $\mathcal{A}_{\pm}(t)$ are independent of angular momentum, simply summing over angular momentum is too naive.
On the other hand, thanks to recent work of
J\'ez\'equel \cite{jezequel2022upper} the sum of $e^{-i\lambda_j t}$ over all quasinormal modes (that is, the sum over $ {\rm{Res}} ( V_\ell ) $) is a well-defined distribution.
\label{s:RW}
Schwarzschild-de Sitter spacetime can be constructed in the following way (c.f. \cite[\S 3.1]{jezequel2022upper}).
Define $M \coloneqq (r_- , r_+) _r \times \SS^2$
where $0 < r_- < r_+$ are the positive roots of
\begin{align}
G(r) \coloneqq 1 - \frac{2m}{r} - \frac{\Lambda r^2 }{3} \label{eq:120}
\end{align}
(for $0 < g/m^2 < 1 $).
Here $m >0$ is the \textit{black hole mass} and $\Lambda \in (0,(9m^2)^{-1})$ is the \textit{cosmological constant}.
Define the Schwarzschild-de Sitter spacetime as $\hat M \coloneqq \R_t \times M$ and let $g$ be the Lorentzian metric:
\begin{align}
g \coloneqq -G \dd t^2 + G^{-1} \dd r^2 + r^2 g_\SS
\end{align}
with $g_\SS$ the standard metric on $\SS^2$.
With this Lorentizan metric, we have the wave operator $\Box_g $, which satisfies, for $u\in C_0^\infty((r_- , r_+) \times \SS^2)$
\begin{align}
\Box_g u = - \frac{1}{G} D_t^2 u + r^{-2} D_r (r^2 G D_r u ) + r^{-2} \Delta _{\SS^2} u
\end{align}
(cf. \cite[Chapter 4]{chandrasekhar1998mathematical}).
Let $U_g(t)$ be the Dirichlet wave propagator with respect to $\Box_g$ (i.e., $U_g(t)$ satisfies \eqref{eq:205} with $\R$ replaced by $M$ and $\Box_V$ replaced by $\Box_g$).
Following \cite[\S 4]{barreto1997distribution}, write:
\begin{align}
\Box_g = -G^{-1} D_t^2 + G^{-1} P \label{eq:boxg}
\end{align}
where
\begin{align}
P \coloneqq G r^{-2} D_r (r^2 G) D_r + G r^{-2} \Delta _{\SS ^2} .
\end{align}
Let $\tilde P \coloneq r Pr^{-1}$ and define $x= x(r)$ such that $x'(r) = G^{-1}$ so that
\begin{align}
\tilde P = D_x^2 + G r^{-2}\Delta _{\mathbb{S}^2} + G r^{-1} (\p_r G). \label{eq:161}
\end{align}
Decomposing a function $u(x,\omega)$ by spherical harmonics: $u(x,\omega) = \sum_{\ell,k} a_{\ell,k} (x) Y_\ell^k (\omega) $ (where $\Delta _\omega Y_\ell^k(\omega) = \ell (\ell +1 ) Y_\ell^k (\omega)$) we get that $u$ is a generalized eigenfunction of $\tilde P$ with eigenvalue $\lambda^2$ if and only if:
\begin{align}
(D_x^2 + \underbrace{G r^{-2} ( \ell (\ell +1 ) +r \p_r G) }_{\coloneqq V_\ell (x)} ) a_{\ell,k} = \lambda^2 a_{\ell,k}. \label{eq:165}
\end{align}
Note for each $\ell$, $V_\ell(x(r))$ is a smooth function on the interval $r\in (r_-,r_+)$ vanishing at $r= r_\pm$.
Now, if $U_{\tilde V}(t)$ is the propagator for $-D_t^2 + \tilde P$ and $U_{V_\ell}(t)$ is the propagator for $-D_t^2 + D_x^2 + V_\ell (x)$, then for $u_0\in C^\infty _0(\R_x\times \mathbb S ^2 _\omega) $
\begin{align}
U_{\tilde V}(t) u_0 = \sum_{\ell=0}^\infty \sum_{k = -\ell}^\ell Y_\ell ^k U_{V_\ell}\ip{u_0}{Y^k_\ell}_{L^2 _\omega (\mathbb S ^2 )}. \label{eq:179}
\end{align}
Here $\ip{u_0}{Y^k_\ell}_{L^2 _\omega (\mathbb S ^2 )}$ is integration of $u_0$ against $Y^k_\ell$ in the $\omega$ variable, so the resulting object is a function of $x$ alone.
Using this, we can rewrite the propagator for $\Box_g$ (defined in \eqref{eq:boxg}) as
\begin{align}
[U_g ( t ) F] ( r , \omega ) = r^{-1} [ U_{\tilde V } (t) ( r F )] ( r, \omega ) , \ \
F = F ( r,\omega ) . \label{eq:185}
\end{align}
It is straightforward to verify that $V_\ell (x)$ satisfies \eqref{hyp:1} with $A_\pm = \abs{\p_r G(r_\pm)}$.
Indeed, by fixing $r_0 \in (r_- , r_+)$ we can define
\begin{align}
x(r) = \int_{r_0}^r (G(s))^{-1} \dd s
\end{align}
so that $x(r_\pm ) = \pm \infty $. Near $r = r_+$, we have:
\begin{align}
G(r) = (r_+ - r) f(r)
\end{align}
where $f(r)$ is nonzero and holomorphic near $r = r_+$.
Taylor expanding $f$ and integrating, we get for $r$ near $r_+$
\begin{align}
x(r) = -A_+^{-1}\log(r_+ - r) +b(r)\label{eq:182}
\end{align}
for $b$ holomorphic near $r_+$.
Therefore $w\coloneq \exp(-A_+ x)=(r-r_+)\exp(b(r))$.
By the inverse function theorem, we get (for $w$ near $0$) $V_\ell (x(w))=F_+ (w)$ with $F$ holomorphic near $w= 0$.
We similarly get, for the change of coordinates $w = \exp(A_- x)$, $V_\ell (x(w)) = F_-(w)$ is holomorphic near $w = 0$.
\begin{theorem*}[Trace formula for each spherical harmonic]
For each $\ell \in \Z_{\ge 0}$, $t >0$
\[ \begin{split}
\Tr_{L^2 ( \mathbb R )} (U_{V_\ell} (t) - U_0(t)) & = \tfrac{1}{2}\sum_{\lambda_j \in \Res(V_\ell)} e^{-i\lambda_j t} \\
& \ \ \ \ \ \ \ \ - \frac{1}{2(e^{tA_-/2} - 1)} - \frac{1}{2(e^{t A_+ / 2} - 1)}-\tfrac{1}{2} \label{eq:369}
\end{split} \]
in the sense of distributions, where $U_{V_\ell}$ is the propagator for $V_\ell$ (defined in \eqref{eq:165}), $\Res(V_\ell)$ is the set of resonances of $V_\ell$, $A_\pm = \abs{\p_r (G(r_\pm))}$.
\end{theorem*}
\begin{rem}
The constants $A_\pm$ are related to the surface gravity of the event and cosmological horizon, $\kappa_-$ and $\kappa_+$, respectively, by the simple relation $A_\pm = 2 \kappa_\pm$ (cf. \cite[\S 12.5]{wald2024general}).
We can then write the terms in \eqref{eq:369} as
\begin{align}
\frac{1}{e^{tA_\pm /2}- 1} = - \sum_{m=0}^\infty e^{- m t\kappa_\pm } = -\sum_{m=0}^\infty e^{-i(-i m \kappa_\pm )t}.
\end{align}
This suggests that these terms can be included in the sum over resonances.
Interestingly, on the de Sitter space, there are resonances at $\set{-im\kappa_\pm : m \in \Z_{\ge 0}}$ (cf. \cite{hintz2021quasinormal}).
Making sense of this connection could lead to a formulation of the Poisson formula for other metrics.
\end{rem}
The trace on the left-hand side of \eqref{eq:369} is written in the $x$ variable.
Changing variables $x$ to $r$, the left-hand side of \eqref{eq:369} is
\begin{align}
\Tr_{L^2 ((r_- ,r_+), G(r)^{-1} \dd r)} (U_{V_\ell }(t) - U_0(t)). \label{eq:392}
\end{align}
Recalling the notion of flat trace \eqref{eq:flat_trace}, \eqref{eq:392} can be written
\begin{align}
\Tr^\flat_{L^2 ((r_- ,r_+), G(r)^{-1} \dd r)} (U_{V_\ell }(t) ). \label{eq:396}
\end{align}
Conjugating $U_{V_\ell}$ by $r$ allows us to write \eqref{eq:396} as
\begin{align}\label{eq:409}
\Tr^\flat_{L^2 ((r_- ,r_+), r^2G(r)^{-1} \dd r)} (r^{-1}U_{V_\ell }(t) r) .
\end{align}
Let $\Pi_\ell$ be the orthogonal projection in the $\omega$ variable onto the space spanned by $\set{Y_\ell ^k (\omega) : k = -\ell, \ell + 1,\dots ,\ell}$.
Then by \eqref{eq:179} and \eqref{eq:185}, we can rewrite \eqref{eq:409} as
\begin{align}
\Tr^\flat_{L^2 ((r_-,r_+)\times \mathbb S^2 ,r^2 G^{-1}\dd r \dd \omega)} ( \Pi _\ell U_g (t) \Pi_\ell ).
\end{align}
So if we now define the \textit{gravitationally renormalized trace} on $$\text{SdS}\coloneq L^2((r_- ,r_+)\times \mathbb S ^2 , G(r)^{-1}r^2 \dd r \dd \omega) $$ as
\begin{align}
\widetilde \Tr ^\flat_{\rm{SdS}} (A(t)) \coloneq \sum_{\ell =0}^\infty\left ( \Tr^\flat_{\rm{SdS}}\Pi _\ell A(t) \Pi_\ell +\frac{1}{2(e^{tA_- /2}-1)} + \frac{1}{2(e^{tA_+/2} - 1) }+\tfrac{1}{2}\right )
\end{align}
then \eqref{eq:369} gives rise to the following global trace formula.
\begin{theorem}[Global trace formula]
The gravitationally renormalized trace of the wave propagator on the Schwarzschild-de Sitter metric is
\begin{align}
\widetilde \Tr ^\flat_{\rm{SdS}} (U_g(t)) = \tfrac 1 2 \sum_{\lambda \in \Res_g} e^{-it\lambda}, \quad t> 0
\end{align}
where $\Res_g \coloneq \bigcup_{\ell} \Res(V_\ell)$ (and the term on the right-hand side is a well-defined distribution by \cite[Theorem 3]{jezequel2022upper}).
\end{theorem}
\subsection{Outline of proof}
An outline of a proof of our main result is presented below.
The order of steps below is \textit{not} in the chronological order of the paper, but presented in a way that motivates the required analysis of \S \ref{s:analysisofincoming}.
\begin{enumerate}
\item The Birman--Kre\u{\i}n trace formula (which we prove for our class of potentials in \Cref{appendix}) relates the trace of $U_{V}(t)-U_0(t)$ to a term depending on negative eigenvalues of $P_V$, a term involving the resonance of $V$ at $0$ (if it exists), and an integral involving the logarithmic derivative of the scattering matrix $S(\lambda)$ for $V$ (see \eqref{BK1}).
This is presented in \S \ref{s:birman}.
\item To apply the Birman--Kre\u{\i}n trace formula, we first need to prove the resolvent $(P_V -\lambda^2)^{-1}$ (defined for $\Im \lambda \gg 1$) can be meromorphically continued to $\C$.
This is done by defining and estimating incoming and outgoing solutions $u_\pm (\lambda,x)$ to $(P_V -\lambda^2 )u = 0$ (\S \ref{s:incoutgo})
\item The logarithmic derivative of the scattering matrix is $ \frac{d}{d\lambda} \log(T(\lambda)/T(-\lambda))$ where $T(\lambda)$ is the transmission coefficient.
\item The transmission coefficient is computed by taking the Wronskian of the incoming and outgoing solutions (see \eqref{eq:598}).
\item The term involving the logarithmic derivative of the scattering matrix will have singularities coming from $T(\lambda)$, which are decoupled into three sets (see \eqref{eq:887}).
Two of these end up giving the terms $\mathcal{A}_\pm(t)$, while the third term is related to the resonances of $V$ (a Lemma relating zeros of the Wronskian of $u_\pm$ to resonances of $V$, with multiplicity, is required -- which is proven in Appendix \ref{s:proof of lemmainduct}).
\end{enumerate}
\smallsection{Notation}
We use the following notation throughout this paper. We write $D_x \coloneq -i \partial_x $.
We let $L^2_c$ denote the space of $L^2$ functions with compact support and $L^2_{\rm{loc}}$ the space of locally $L^2$ functions
We let $\norm{\cdot}_1$ denote the trace norm.
For functions $f(x)$ and $g(x)$, we write $f(x) \lesssim g(x)$ if there exists a constant $C>0$ such that $f(x) \le C g(x)$ for all $x$.
If $C$ depends on a parameter $\alpha$, we write $f(x) \lesssim_\alpha g(x)$.
We write $f(x) \gg 1$ to mean $f(x) > C $ for some sufficiently large $C >0$.
If $g(x)$ is positive and $N \in \Z$, we write $f = \mathcal{O}(g^N)$ to mean there exists $C_N > 0$ such that $|f(x)|\le C_N g(x)^N$.
We write $f = \mathcal{O}(g^{ \infty})$ if $f = \mathcal{O}(g^N)$ for all $N> 0$ (and similarly $f = \mathcal{O}(g^{- \infty}) \iff f = \mathcal{O}(g^{-N}) \ \forall \ N \in \Z_{>0}$).
\smallsection{Acknowledgments}
We thank Maciej Zworski for suggesting this project and numerous helpful discussions (in particular, his help in generalizing the Birman--Kre\u{\i}n trace formula).
We also thank Peter Hintz for helpful comments about the trace formula in the Schwarzschild-de Sitter setting.
Conversations with Malo J\'ez\'equel, Ethan Sussman, and Jared Wunsch were also very useful.
The first author was supported by NSF grant DMS-2136217. The second author was supported by a fellowship
in the Simons Society of Fellows.
\section{Analysis of incoming and outgoing solutions}\label{s:analysisofincoming}
In this section, we first construct and analyze the behavior of the incoming and outgoing solutions to
\begin{align}
(D_x^2 + V (x)) u(x) = \lambda ^2 u(x)
\end{align}
where $V$ has the asymptotics described by \Cref{hyp:1} and $D_x \coloneq - i \p_x$. We will then use this to show that the corresponding resolvent operator $(P_V-\lambda^2)^{-1}$ can be meromorphically extended to the complex plane.
\subsection{Incoming and outgoing solutions} \label{s:incoutgo}
We begin by recalling the definition for incoming and outgoing solutions for $P_V\coloneq (D_x^2+V(x)-\lambda^2)$. In the sequel, we assume $V$ satisfies \Cref{hyp:1} with $A_\pm $.
\begin{defi}[Outgoing solution]\label{def:outgoing}
We say $u(x)= u(x,\lambda)$ is an outgoing solution for the operator
\begin{align}
P_\lambda \coloneqq (D_x^2 + V (x) - \lambda^2) \label{eq:plambdadef}
\end{align}
at $+\infty $ if $P_\lambda u = 0$ and there exists a globally defined, smooth profile $v_+ (w) = v_+(w,\lambda)$ that is holomorphic in a neighborhood of zero, such that
\begin{align}
u (x) = e^{i \lambda x }v_+ (e^{-A_+ x}).
\end{align}
Analogously, $u(x) = u(x,\lambda)$ is an outgoing solution for $P_\lambda$ at $-\infty $ if $P_\lambda u = 0$ and there exists a globally defined, smooth profile $v_- (w)$ that is holomorphic in a neighborhood of zero, such that
\begin{align}
u (x) = e^{-i \lambda x }v_- (e^{A_- x}).
\end{align}
\end{defi}
We also have a similar definition for incoming solutions.
\begin{defi}[Incoming solution]\label{def:incoming}
We say $u(x) = u(x,\lambda)$ is an incoming solution to $P_\lambda$ at $+\infty $ if $P_\lambda u = 0$ and there exists a globally defined, smooth profile $w_+ (w)$ that is holomorphic in a neighborhood of zero, such that
\begin{align}
u (x) = e^{-i \lambda x }w_+ (e^{-A_+ x}).
\end{align}
Analogously, $u(x) = u(x,\lambda)$ is an incoming solution to $P_\lambda$ at $-\infty $ if $P_\lambda u = 0$ and there exists a globally defined, smooth profile $w_- (w)$ that is holomorphic in a neighborhood of zero, such that
\begin{align}
u (x) = e^{i \lambda x }w_- (e^{A_- x}).
\end{align}
\end{defi}
Our first result in this section shows that for each $\lambda$, we can construct outgoing solutions with the asymptotics described above.
\begin{prop}\label{prop:existenceofoutgoing}
For each $\lambda\in \C$, there exist outgoing solutions for $P_\lambda$ of the form
\begin{align}\label{eq:336}
u_\pm (x) =e^{\pm i \lambda x } v_\pm (e^{\mp A_\pm x})
\end{align}
where $v_\pm(w) = v_\pm (w,\lambda)$ are holomorphic near $w = 0$ in a neighborhood that is independent of $\lambda$ and such that
\[ v_\pm (0) = \frac{1}{\Gamma(1 - 2 i \lambda A_\pm^{-1} )} . \]
Moreover, there exists $C,M_\pm > 0$, independent of $\lambda$, such that for $x\in \R$
\begin{equation}
\abs{\partial_x^j v_\pm (e^{\mp A_\pm x})} \le C \exp\Bigl(C \langle x-M_\pm\rangle\langle\lambda\rangle\log \langle\lambda\rangle \Bigr) , \ \ \ j=0,1. \label{eq:boundonvp}
\end{equation}
where we write $\langle z \rangle\coloneq (1+|z|^2)^{\frac{1}{2}}$.
\end{prop}
\begin{proof} We will provide the details for the case $u_+$ as the other case follows from analogous reasoning.
\Step We begin by making an ansatz
\begin{equation*}
u_+(x) = e^{i\lambda x}v_+(e^{-A_+ x}).
\end{equation*}
It will be convenient in the forthcoming calculations to write as shorthand $w\coloneq e^{-A_+x}$. A straightforward calculation shows that solving $P_{\lambda}u_+=0$ is equivalent to $v_+$ solving the equation
\begin{align}
2 \lambda D_x v_+ (w)+ D_x^2 v_+(w)+ V (x)v_+(w)=0. \label{eq:121}
\end{align}
We can write this equation in terms of of $D_w$ by noting the simple identities $D_x = (-A_+ w ) D_w$ and $D_x^2 = A_+^2 (w D_w )^2 $. Hence, we find that $ (D_x^2 + V(x) -\lambda^2 ) u_+(x) =0$ if and only if
\begin{align}
-2\lambda A_+w D_w v_+ +A_+^2 (wD_w)^2 v_+ + V v_+ = 0. \label{eq:251}
\end{align}
To construct $v_+$, we will begin by formally expanding $v_+$ and $V$ in a power series near $w=0$:
\begin{align}
v_+(w) = \sum_{j=0}^\infty v_j w^j , \ \
\ V(w) = \sum_{j=1}^\infty V_j w^j, \label{eq:510}
\end{align}
where the coefficients $V_j$ are defined as in \eqref{hyp:1}. We note that the potential does not have a constant term in this expansion (i.e, $V$ vanishes to at least first order at $w=0$).
Substituting this into \eqref{eq:251} and matching powers of $w$, we obtain the follow recursion relation for each $j \ge 1$:
\begin{align}
j A_+^2 (j - 2 i A_+^{-1} \lambda) v_j = \sum_{\ell = 1}^j V_{\ell}v_{j - \ell}. \label{eq:258}
\end{align}
\Step We now prove the $v_j$'s (the coefficients in the Taylor expansion of $v_+$) have sufficient decay to ensure that $v_+$ is analytic and has a nontrivial radius of convergence that is uniform in $\lambda$. We will begin by showing that the coefficients $v_j$ defined by the recursion \eqref{eq:258} are entire functions of $\lambda$.
To simplify notation, in the forthcoming calculations, we set
\begin{align}
\alpha \coloneq 2iA_+^{-1}\lambda. \label{eq:defofalpha}
\end{align}
We can write \eqref{eq:258} as a $j\times j$ system as follows. For each $j$, we have
\begin{align}
({\bf{\Lambda}}_j - \textbf{A}_j) \mat{v_1 \\ v_2 \\ \vdots \\ v_j} = \mat{V_1 v_0 \\ V_2 v_0 \\ \vdots \\ V_j v_0}
\end{align}
where $({\bf{\Lambda}}_j) \coloneq \diag(( kA_+^2(k-\alpha))_{k=1}^{j})$ and
\begin{align}
\bf{A}_j \coloneqq \mat{0 & \cdots& \cdots & 0 &0 \\
V_1 & 0 & \cdots & \cdots & 0 \\
V_2 & V_1 & \cdots & \cdots &\vdots \\
\vdots & \ddots & \ddots & \ddots & 0 \\
V_{j-1} & \cdots & V_2 & V_1 & 0 \\}.
\end{align}
By Cramer's rule, as long as $\alpha \neq 1,2,\dots, j$, we can invert ${\bf \Lambda_j} - \textbf{A}_j$ as:
\begin{align}
({\bf\Lambda}_j - \textbf{A}_j)^{-1} = \frac{\textbf{B}_j(\alpha)}{\det({\bf\Lambda}_j - \textbf{A}_j)} = \frac{\textbf{B}_j(\alpha)}{A_+^{2j} j! (j-\alpha) \cdots (1- \alpha)}
\end{align}
where $\textbf{B}_j(\alpha)$ is a $j\times j$ matrix with entries holomorphic in $\alpha$.
If we set $v_0 \coloneq \Gamma(1 - \alpha)^{-1}$, then for all $\alpha$ (including integer values),
\begin{align}
\mat{v_1 \\ \vdots \\ v_j} &= \frac{\textbf{B}_j (\alpha)}{j! (j-\alpha ) \cdots (1- \alpha )\Gamma(1-\alpha) } \mat{V_1 \\ \vdots \\ V_j}\\
&= \frac{\textbf{B}_j (\alpha)}{j! \Gamma(j+1 - \alpha)} \mat{V_1 \\ \vdots \\ V_j}
\end{align}
so that $v_j$ are entire functions of $\alpha$ (and thus $\lambda$).
\Step For each $j$, we want to obtain a strong enough growth bound for $v_j(\alpha)$ for $\alpha \in \C$. We will do this in two steps. First, we will estimate $v_j(\alpha)$ for $\alpha$ away from $\N$, and then extend this to a global bound by using the maximum principle.
We now use the assumption that $|V_j|\le A^j$ for some positive constant $A$. The following lemma will give us the required estimate away from $\mathbb{N}$.
\begin{lemma}\label{lemma:1}
Suppose $\e \in (0,A_+^{-2})$ and $\Omega_\e \coloneq \set{z \in \C : \dist(z,\N) > \e}$, then for $\alpha \in \Omega_\e$,
\begin{align}
\abs{v_j(\alpha)} \le \frac{A^{j}\abs{v_0}}{\e^{j}A_+^{2j}} .
\end{align}
\end{lemma}
\begin{proof}
We prove this by induction. For $j=0$ this is clear. We then have
\begin{align}
\abs{v_j(\alpha) } &\le \frac{1}{j A_+^2|j-\alpha|} \abs{\sum_{k=0}^{j-1} v_k V_{j-k} }\le \frac{\abs{v_0}}{j A_+^2 \e} \sum_{k=0}^{j-1} A^{k} \e^{-k} A_+^{-2k}A^{j- k}\\
&= \frac{ |v_0| A^{j}}{jA_+^{2}\e} \sum_{k=0}^{j-1} \e^{-k}A_+^{-2k}\le \frac{|v_0|A^j}{j A_+^2 \e} \sum_{k=0}^{j-1}\e^{-(j-1)} A_+^{-2(j-1)} = \frac{|v_0|A^j}{\e^j A_+^{2j}}.
\end{align}
The first inequality follows from \eqref{eq:258}, while the second inequality follows from the induction hypothesis, and the third inequality follows from the choice of $\e$.
\end{proof}
Next, we extend \Cref{lemma:1} to a global estimate. We fix $\e _ 0\in (0,\min(A_+^{-2}, 1))$.
By Lemma \ref{lemma:1}, and the choice of $v_0 = \Gamma(1-\alpha)^{-1}$ and the above steps, we have that $v_j(\alpha)$ are entire functions, and for $\alpha\in \Omega_{\e_0}$, we have
\begin{align}\label{eq:28}
\abs{v_j(\alpha)}\le \frac{ A^{j}}{\e_0^{j} A_+^{2j} \abs{\Gamma(1-\alpha)}}.
\end{align}
To obtain a global bound for $v_j(\alpha)$, we use the following standard bound of the reciprocal of the Gamma function. It follows from the factorization of $ \Gamma( z)$ and, for instance, \cite[Theorem 1.11]{Hayman1964}.
\begin{lemma}\label{lemma:gammabound}
There exist a constant $C > 0$ such that for $z\in \C $,
\begin{align}
\abs{\frac{1}{\Gamma(z)}} \le C\exp\left( C \abs z \log (1+ \abs{z}) \right) .
\end{align}
\end{lemma}
Now, for $k\in \N$, we apply the maximum principle and Lemma \ref{lemma:gammabound} to get that:
\begin{align}
\max_{|\alpha - k| \le \e_0 } \abs{v_j (\alpha) }&\le \max_{|\alpha - k| = \e_0} \abs{\frac{A^{j}}{\e_0^{j}A_+^{2j}\Gamma(1-\alpha)}}\le \max_{|\alpha - k| = \e_0} \frac{ CA^{j}}{A_+^{2j}\e_0^{j}} e^{C |1-\alpha| \log (|1-\alpha| +1 )} \\
&\le \frac{C A^{j}}{A_+^{2j}\e_0^{j}} e^{C (k+1) \log(k+2)} \le \frac{C A^{j}}{A_+^{2j}\e_0^{j}} e^{C k \log(k)}\label{eq:394}
\end{align}
where the constant $C$ may change in each inequality.
We can then use this to get a global bound on $v_j$. We have by \eqref{eq:28} that for $\dist(\alpha,\N) > \e_0$:
\begin{align}
\abs{v_j (\alpha) }\le \frac{ A^{j}}{\e_0^jA_+^{2j}} e^{C |1 - \alpha | \log (|1-\alpha | + 1))}
\end{align}
and by \eqref{eq:394}, for $|\alpha - k |\le \e_0$ (for some $k\in \N$):
\begin{align}
\abs{ v_j(\alpha)} \le \frac{CA^j}{A_+^{2j}\e_0^j} e^{C k \log (k)}.
\end{align}
Therefore, there exists a $C>0$ such that for all $\alpha \in \C$:
\begin{align}
\abs{v_j (\alpha )}\le \frac{CA^j}{A_+^{2j}\e_0^j} e^{C (|\alpha| + 1) \log (|1-\alpha| +1)}.
\end{align}
This shows that $v_+(w)$ as defined by \eqref{eq:510} is well-defined in some neighborhood of zero with radius of convergence that is independent of $\lambda$.
\\
\Step Apriori, the above construction only ensures that $v_+$ is defined in a neighborhood of $w=0$ (uniformly in $\lambda$). We now carry out a standard energy-type estimate for $P_{\lambda}$ to show that $v_+$ can be extended to a global (in $w$) smooth solution of \eqref{eq:251}. Along the way, we also show the bound \eqref{eq:boundonvp}. We begin by fixing $\delta > 0$ less than the radius of convergence of $\sum_0^\infty v_j w^j$ (which is independent of $\alpha$), to ensure that there exists a $C = C(\delta )$ (independent of $\alpha$) such that
\begin{align}
\abs{(\partial_w^j v_+)(\delta) } \le C e^{ C (|\alpha | + 1 ) \log (|1-\alpha | +1)}\label{eq:476}
\end{align}
for $j = 0,1$. We recall that in terms of the original variable $x$, we have $x(\delta) = A_+^{-1} \log(1/\delta)$.
We will now use a simple energy estimate to estimate the growth of $v_+$ and its derivatives.
By \eqref{eq:121}, $v_+$ satisfies the the ODE in the $x$ variable:
\begin{align}
2 i \lambda \p_x v_+(w) + \p_x^2 v_+(w) - V(x) v_+(w) = 0.
\end{align}
Here, we can rewrite the above equation as a $2\times 2$ system
\begin{equation}\label{matrixeqn}
\p_x \mat{u_1(x) \\ u_2(x)} = \underbrace{\mat{0 & 1 \\ {V(x)} & -2i\lambda }}_{\coloneqq \textbf{A}(x)}\mat{u_1(x) \\ u_2(x)}.
\end{equation}
where $u_1 (x)\coloneq v_+(w(x))$ and $u_2 (x)\coloneq \p_x (v_+(w(x))$. Let $\textbf{U}(x) \coloneq (u_1 (x),u_2(x))^t$, so that $\p_x \textbf{U} = \textbf{A}\textbf{U}$.
This gives the simple energy identity
\begin{align}
\frac{d}{dx} \abs{\textbf{U}}^2 = 2 \Re{\ip{\textbf{A} \textbf{U}}{\textbf{U}}}.
\end{align}
By the Cauchy--Schwartz inequality
\begin{align}
\frac{d}{dx} \abs{\textbf{U}(x)}^2 \ge -C\langle\lambda\rangle\abs{\textbf{U}(x)}^2
\end{align}
for some positive constant $C$ (independent of $\lambda$). This gives us that:
\begin{align}
\frac{d}{dx} \left ( e^{C\langle\lambda\rangle x} \abs{\textbf{U}(x)}^2\right ) \ge 0
\end{align}
so that integrating from $x$ to some $x_0>x$, we get that:
\begin{align}
\abs{\textbf{U}(x)}^2 \le e^{C\langle\lambda\rangle(x_0 - x)} \abs{\textbf{U}(x_0)}^2.
\end{align}
Letting $x_0 = x(\delta)$, we get, by \eqref{eq:476} that there exists a $C>0$ such that for $j= 0,1$
\begin{align}
\abs{\p_x^j (v_+ (w(x)) } \le C e^{C\langle\lambda\rangle (x(\delta)- x) + C(|\alpha| + 1 ) \log(|1-\alpha| + 1)}.
\end{align}
This shows that $v_+$ can be continued to a global $C^1$ (in $w$) solution of $P_{\lambda}v_+=0$ and that (by elementary estimates) $v_+$ satisfies the bound \eqref{eq:boundonvp}. To show that $v_+$ is smooth, one can simply apply derivatives to \eqref{matrixeqn} and inductively apply the above argument.
\setcounter{step}{0}
\end{proof}
\subsection{Meromorphic continuation of the resolvent}\label{s:meromorphic continuation}
We now prove a more precise version of Proposition \ref{p:0} which meromorphically extends the resolvent $R_V (\lambda)$ to the entire complex plane.
The poles of the meromorphically continued resolvent will be the zeros of the Wronskian of outgoing solutions to $P_V -\lambda^2$.
This Wronskian is entire in $\lambda$.
A technical Lemma is first required to prove that the Wronskian is not identically zero.
\begin{prop}\label{prop:Wronskian}
There exist purely imaginary $\lambda$ in the upper half plane such that the outgoing solutions $u_+(x,\lambda)$ and $u_-(x,\lambda)$ from Proposition \ref{prop:existenceofoutgoing} are linearly independent.
In particular, the Wronskian of $u_+(x,\lambda)$ and $u_-(x,\lambda)$
\begin{align}
W[u_+,u_-](\lambda) \coloneq (\p_x u_+(x,\lambda) )u_-(x,\lambda) - u_+(x,\lambda) (\p_x u_-(x,\lambda)) \label{eq:wronk}
\end{align}
is not identically zero in $\lambda$.
\end{prop}
\begin{proof}
Take $\lambda=i\mu$ for some sufficiently large real $\mu>0$.
It suffices to show that $u_+(x,\lambda)$ does not vanish as $x\to- \infty$ (since, by definition, we have $\lim_ {x\to -\infty}u_-(x,\lambda)= 0$).
Without loss of generality, we can assume $u_+$ is real-valued (as $\lambda^2\in\mathbb{R}$, and thus $\operatorname{Re}(u_+)$ will also satisfy the equation $P_\lambda u=0$). Our first aim will be to show that $u_+$ is non-negative and decreasing. We define
\begin{equation*}
M \coloneq \sup\{x\in\mathbb{R}: u_+(x)\leq 0\}.
\end{equation*}
By definition of $u_+$ (as in Proposition \ref{prop:existenceofoutgoing}), we know that $-\infty\leq M<\infty$.
It will suffice to show that $M=-\infty$ and that $u_+$ is decreasing.
We begin by selecting $\mu$ large enough so that
\begin{equation}\label{lowercomparisonbound}
V (x)-\lambda^2=V (x)+\mu^2\geq 1,\hspace{5mm}\forall x\in\mathbb{R}.
\end{equation}
Because $P_{V}u_+=0$, we have
\begin{equation*}
\partial_x^2u_+ (x)=(V (x)+\mu^2)u_+ (x)\geq u_+(x),\hspace{5mm}x\in (M,\infty).
\end{equation*}
Hence, on $(M,\infty)$, we have
\begin{equation*}
\partial_x^2u_+> 0.
\end{equation*}
Since $\partial_xu_+\to 0$ as $x\to\infty$, we obtain by integrating from $x\in (M,\infty)$ to $\infty$,
\begin{equation*}
\partial_xu_+<0,\hspace{5mm}x\in (M,\infty).
\end{equation*}
This implies $u_+$ is decreasing on $(M,\infty)$. This further implies both that $M=-\infty$ and that (with strict inequality)
$ \liminf_{x\to-\infty} u_+(x)>0$.
\end{proof}
As a consequence of the proposition, we have the following:
\begin{prop}\label{prop:meroresolv}
For $\Im{\lambda} \gg 1$, there exists an operator
\begin{align}
R_V(\lambda) \coloneq (D_x^2 + V(x) -\lambda^2) ^{-1}\colon L^2 (\R) \to L^2 (\R)
\end{align}
which can be extended to a meromorphic family of operators
\begin{align}
R_V (\lambda) \coloneq (D_x^2 + V(x) -\lambda^2)^{-1} \colon L^2_{\rm{c}} (\R) \to L^2_{\rm{loc}} (\R)
\end{align}
for all $\lambda \in \C$.
Moreover, if $u_\pm(x,\lambda)$ are outgoing solutions, as constructed in Proposition \ref{prop:existenceofoutgoing}, then the poles of the resolvent are the $\lambda \in \C$ such that $W[u_+,u_-](\lambda)=0$ (defined in \eqref{eq:wronk}), which are notably independent of $x$.
\end{prop}
\begin{proof}
If $f \in L^2(\R)$ and $\Im \lambda \gg 1$, then using the variation of parameters formula, a solution to
\begin{align}
(D_x^2 + V(x) - \lambda^2 ) u = f
\end{align}
can be written as
\begin{align}
u(x) = \int_\R R(x,y,\lambda) f(y) \dd y
\end{align}
where:
\begin{align}
R(x,y,\lambda) \coloneq \frac{u_+(x,\lambda) u_-(y,\lambda) H(x-y)}{W[u_+,u_-](\lambda)} + \frac{u_-(x,\lambda) u_+(y,\lambda) H(y-x)}{W[u_+,u_-](\lambda)}.\label{eq:535}
\end{align}
Here, $u_\pm$ are solutions to $ (D_x^2 + V(x) - \lambda^2 ) u_\pm =0 $ as constructed in Proposition \ref{prop:existenceofoutgoing}, and $H(x)$ is the Heaviside function.
By \Cref{prop:Wronskian}, the Wronskian for $u_{\pm}$ is not identically zero. Moreover, because the Wronskian is entire in $\lambda$, it follows that the resolvent can be extended to a meromorphic function of $\lambda$ (with poles where the Wronskian is zero).
\end{proof}
\subsection{The Scattering Matrix}
Because $V$ decays exponentially, we have the existence of the scattering matrix $S(\lambda)$ \cite[\S 5]{melin1985operator}, whose components we write as
\begin{align}
S(\lambda) \coloneqq \mat{T(\lambda) & R_+(\lambda) \\ R_-(\lambda) & T(\lambda)}\label{eq:645}
\end{align}
where $T(\lambda)$ is the transmission coefficient, $R_+(\lambda)$ is the right reflection coefficient, and $R_-(\lambda)$ is the left reflection coefficient.
{We recall that for compactly supported solutions to $ ( P_V - \lambda^2 ) u = 0$, $ \lambda \in \mathbb R \setminus \{ 0 \} $ have the form
\[ u|_{ x \ll - 1 } = A_- e^{ i \lambda x } + B_- e^{ - i \lambda x} , \ \
u|_{ x \gg 1 } = A_+ e^{ i \lambda x } + B_+ e^{ - i \lambda x} , \]
and $ S ( \lambda ) $ takes the incoming terms to the outgoing terms, with the
convention that for $ V = 0 $ the scattering matrix is the identity:
\[ S( \lambda ) \begin{pmatrix} A_- \\ B_+ \end{pmatrix} = \begin{pmatrix} A_+ \\ B_- \end{pmatrix} .
\]
In our situation the same definition applies but now in an asymptotic sense as
$ x \to \pm \infty $.}
Let $u_\pm = u_\pm(x,\lambda)$ be outgoing solutions at $\pm \infty$ and let $w_\pm = w_\pm (x,\lambda)$ be incoming solutions at $\pm \infty$ (recalling Definitions \ref{def:outgoing} and \ref{def:incoming}) so that
\begin{align}
u_\pm (x)|_{\pm x \gg 1} = e^{\pm i\lambda x} v_\pm (e^{\mp A_\pm x}) , \ \ w_\pm (x)|_{\pm x \gg 1} = e^{\mp i\lambda x} \tilde v_\pm (e^{\mp A_\pm x}).
\end{align}
For now, we will suppress the $\lambda$ in the arguments of $u_\pm$ and $w_\pm$.
Because $u_+$ and $w_+$ are linearly independent, there exist constants $a,b\in \C$ (depending on $\lambda$) such that
\begin{align}
u_-(x) = a u_+(x) + bw_+(x). \label{eq:613}
\end{align}
The scattering matrix will then satisfy
\begin{align}
S(\lambda) \mat{0 \\ b\tilde v_+(0)} = \mat{{ }av_+(0)\\v_- (0)}\label{eq:661}
\end{align}
so that
\begin{align}
T(\lambda) b\tilde v_+(0) = v_-(0).\label{eq:625}
\end{align}
The transmission coefficient $T(\lambda)$ can be related to the Wronskian of $u_+$ and $u_-$.
Indeed, using \eqref{eq:613},
\begin{align}
W[u_+,u_-](\lambda) &= b W[u_+,w_+] (\lambda)= b \lim_{x\to \infty} W[e^{i\lambda x} v_+ (e^{-A_+ x}),e^{-i\lambda x }\tilde v_+(e^{-A_+ x})]\\
&= b2i\lambda v_+(0) \tilde v_+(0)= \frac{2i \lambda v_+(0)v_-(0)}{T(\lambda)}\label{eq:672}
\end{align}
where in the last equality we used \eqref{eq:625}.
Finally using \Cref{prop:existenceofoutgoing}, we can rewrite \eqref{eq:672} as
\begin{align}
T(\lambda) = \frac{2i\lambda }{\Gamma(1- 2i \lambda A_+^{-1} )\Gamma(1- 2i \lambda A_-^{-1})W[u_+,u_-](\lambda)}\label{eq:598}
\end{align}
which is a meromorphic function and has poles only at the zeros (in $\lambda$) of $W[u_+,u_-](\lambda)$.
\begin{rem}
Observe that, besides $\lambda = 0$, the only zeros of $T(\lambda)$ are at the poles of $\Gamma(1- 2i \lambda A_+^{-1} )$ and $\Gamma(1- 2i \lambda A_-^{-1})$ where $W[u_-,u_+](\lambda)\neq 0$, which are along the negative imaginary axis at the points
\begin{align}
\set{-\frac{1}{2} A_\pm i (k+1 ) : k \in\Z_{\ge 0}}.
\end{align}
By unitarity of the scattering matrix, we see that these points provide ``false'' poles of the scattering matrix along the positive imaginary axis at the points:
\begin{align}
\set{\frac{1}{2} A_\pm i (k+1 ) : k \in\Z_{\ge 0}}\setminus \set{\lambda \in \C : W[u_-,u_+](-\lambda ) = 0}.
\end{align}
The existence of such poles contradicts Heisenberg's original hypothesis that all poles of the scattering matrix correspond to resonances.\footnote{See for instance \cite[\S 12.1.2]{newton2013scattering} for a discussion of false poles in the context of Jost functions. See also \cite{Pais1995} for a bibliographical account of Jost which includes a brief discussion, with references, of ``false'' poles.
As explained by Graham--Zworski \cite{graham2003scattering}, these ``false'' poles also play an important role in scattering on asymptotically hyperbolic spaces and its relation to the conformal structure of the boundary at infinity.
}
These ``false'' poles will give rise to the $\mathcal{A}_\pm (t)$ appearing in our trace formula \eqref{eq:218}.
\end{rem}
\section{Applying the Birman--Kre\u{\i}n trace formula}\label{s:birman}
We are now prepared to apply the Birman--Kre\u{\i}n trace formula to prove our main result.
Recall we are computing $\Tr(U_V(t) - U_0(t))$, where $U_V(t) \coloneq \cos (t\sqrt{P_V})$ is the wave propagator for a potential $V(x)$.
We fix $\phi \in C_0^\infty ((0,\infty))$\footnote{There is a singularity at $t=0$ which requires us to choose $\phi$ supported in positive time.}, so that
\begin{align}\label{traceapp}
\Tr(U_V(t) - U_0(t)) (\phi) &= \Tr(\cos(t\sqrt{P_V}) - \cos(t\sqrt{P_0}) )(\phi) \\
&= \Tr(f(P_V) - f(P_0) ) \label{eq:763}
\end{align}
where $f(\lambda^2 ) = \frac{1}{2} (\hat \phi (\lambda) + \hat \phi (-\lambda))$. Equation \eqref{eq:763} follows by writing
\begin{align*}
\cos\left(t \sqrt{P_V}\right ) (\phi) &= \tfrac{1}{2}\int_{-\infty}^\infty \left( e^{it \sqrt{P_V} } + e^{-it\sqrt{P_V}}\right) \phi(t) \dd t\\
&= \tfrac{1}{2} (\hat \phi (\sqrt{P_V}) + \hat \phi(-\sqrt{P_V})) = f(P_V).
\end{align*}
By the Birman--Kre\u{\i}n trace formula (\Cref{t:BK}), we obtain
\begin{align}
\begin{split}
\Tr(f(P_V) - f(P_0)) = \frac{1}{2\pi i } &\int_0^\infty f(\lambda^2 ) \Tr (S(\lambda)^{-1} \p_\lambda S(\lambda )) \dd \lambda \\
&+ \sum_{j=1}^k f(E_j) + \tfrac{1}{2} (m_R(0) - 1) f(0)
\end{split} \label{BK1}
\end{align}
where $E_j$ are the negative eigenvalues of $D_x^2 + V(x)$.
Note that if $V$ is positive (which we do not necessarily assume here), there are no negative eigenvalues.
We now focus on the first term on the right-hand side of \eqref{BK1}.
The term we are integrating against $f(\lambda)$ is
\begin{align}
\mathcal{G}(\lambda)\coloneq \Tr \left ( S(\lambda)^{-1} \p_\lambda S(\lambda ) \right) & = \frac{d}{d\lambda} \log \det (S(\lambda))=\frac{d}{d\lambda} \log \left( T(\lambda)/T(-\lambda)\right)\label{eq:682}
\end{align}
where the first equality follows from Jacobi's formula and the second is a standard fact about the scattering matrix (see, for instance, \cite[\S 2.10 Exercise 3]{dyatlov2019mathematical}).
Moreover, we have by \eqref{eq:598}
\begin{align*}
T(\lambda) = \frac{2i\lambda}{\Gamma(1-\alpha_-)\Gamma(1-\alpha_+)F(\lambda)}
\end{align*}
where
\begin{align}
F(\lambda) \coloneq W[u_+ (\cdot ,\lambda ) ,u_-(\cdot ,\lambda)]\label{def:F}
\end{align}
is entire in $\lambda$, with $u_+$ and $u_-$ constructed in \Cref{prop:existenceofoutgoing} and $\alpha_\pm = 2i\lambda A_{\pm}^{-1}$.
Therefore, we can expand
\begin{equation} \label{eq:740}
\begin{split}
\log(T(\lambda) / T(-\lambda) ) &= \log(\Gamma(1+\alpha_-)) - \log(\Gamma(1 - \alpha_-))+\log(\Gamma(1+\alpha_+)) \\
& \qquad \qquad- \log(\Gamma(1-\alpha_+)) + \log(F(-\lambda)) - \log (F(\lambda)).
\end{split}
\end{equation}
Computing the derivative, we obtain
\begin{equation}
\mathcal{G}(\lambda) = \frac{d}{d\lambda}\log(T(\lambda)/T(-\lambda))(\lambda)=\mathcal{G}_{\Gamma}^+(\lambda)+\mathcal{G}_{\Gamma}^-(\lambda)+\mathcal{G}_F(\lambda) \label{eq:887}
\end{equation}
where we define
\begin{align}
\mathcal{G}_\Gamma^\pm(\lambda) & \coloneq \frac{d}{d\lambda} \left( \log(\Gamma(1+\alpha_\pm)) - \log(\Gamma(1 - \alpha_\pm))\right)\\
&= \frac{2iA_\pm^{-1} \Gamma'(1 + \alpha_\pm)}{\Gamma(1 + \alpha_\pm)} + \frac{2iA_\pm^{-1}\Gamma'(1-\alpha_\pm))}{\Gamma(1-\alpha_\pm)} \label{eqa2}
\end{align}
and
\begin{align}
\mathcal{G}_F(\lambda)\coloneq \frac{d}{d\lambda}(\log(F(-\lambda)) - \log (F(\lambda))). \label{eq:751}
\end{align}
Inserting this into the first term on the right-hand side of \eqref{BK1} gives
\begin{equation}\label{firsttermcontributions}
\begin{split}
&\frac{1}{2\pi i}\int_{0}^{\infty}f(\lambda^2)\Tr(S(\lambda)^{-1}\partial_{\lambda}S(\lambda))\dd \lambda\\
=&\frac{1}{2\pi i}\int_{0}^{\infty}f(\lambda^2)\left(\mathcal{G}_{F}(\lambda)+\mathcal{G}_{\Gamma}^+(\lambda)+\mathcal{G}_{\Gamma}^-(\lambda)\right)\dd \lambda
\end{split}
\end{equation}
It will turn out to be relatively straightforward to analyze the contribution of $\mathcal{G}_{\Gamma}^\pm$ in \eqref{firsttermcontributions} using the well-known series expansion of the digamma function.
On the other hand, to analyze $\mathcal{G}_F$, we will need some precise quantitative information about the distribution of the zeros of $F$.
Our starting point is to observe that, thanks to \eqref{eq:boundonvp} and the fact that the Wronskian is constant in $x$, we have the upper bound
\begin{equation*}
|F(\lambda)|\lesssim e^{C|\lambda|\log(1+|\lambda|)}.
\end{equation*}
Since $F$ is entire in $\lambda$, this yields the Hadamard factorization (cf. \cite[Theorem 1.9]{Hayman1964})
\begin{align}
F(\lambda) = e^{a\lambda + b }\lambda^m \prod_{j=1}^\infty\left( 1 - \frac{\lambda }{\lambda_j} \right) e^{\lambda / \lambda_j}.
\end{align}
for some $a,b\in\mathbb{C}$. Here $m$ is the multiplicity of the zero at $0$ and the $\lambda_j$ are the nonzero roots of $F$. We then observe the following (formal at this point) expansion of $\log(F(\lambda))$,
\begin{align}
\log(F(\lambda)) = (a\lambda + b) + m \log (\lambda) + \sum_{j=1}^\infty \left( \log\left(1 - \frac{\lambda}{\lambda_j} \right) + \frac{\lambda}{\lambda_j} \right)
\end{align}
so that
\begin{align}
\frac{d}{d\lambda} \log(F(\lambda)) &= a + \frac{m}{\lambda} + \sum_{j=1}^\infty \left( \frac{1}{\lambda-\lambda_j} + \frac{1}{\lambda_j} \right).
\end{align}
Similarly, we have
\begin{align}
\frac{d}{d\lambda} \log(F(-\lambda))&= -a + \frac{m}{\lambda} + \sum_{j=1}^\infty \left( \frac{1}{\lambda_j +\lambda} - \frac{1}{\lambda_j}\right).
\end{align}
Therefore, we have the following formal expansion of $\mathcal{G}_F$ (defined in \eqref{eq:751}),
\begin{align}
\mathcal{G}_F(\lambda) = -2a + \sum_{j=1}^\infty \left( \frac{1}{\lambda_j +\lambda} + \frac{1}{\lambda_j - \lambda} - \frac{2}{\lambda_j}\right). \label{AFdef}
\end{align}
We observe in particular that $\mathcal{G}_F$ is an even function in $\lambda$. In order to compute the contribution of this term in \eqref{firsttermcontributions} and rigorously justify the forthcoming manipulations, we will need the following proposition, which gives a precise description of the location and density of the zeros of $F$.
\begin{prop}\label{Fprop}
The roots of $F(\lambda) = W[u_+,u_-](\lambda)$ satisfy the following properties
\begin{enumerate}
\item \label{F:prop1} There exists $C>0$ such that:
\begin{align}
\set{F^{-1}(0)} \subset \set{ i y : y\in [0,C]} \cup \set{z \in \C : \Im{z} < 0}.
\end{align}
\item \label{F:prop2} If $F(\lambda) = 0$, then $F(-\bar\lambda)= 0$.\footnote{In fact for all $z\in \C$, $F(z) = \overline{F(-\bar z)}.$}
\item \label{F:prop3} There exists $\delta>0$ such that $F$ does not vanish anywhere on the strip $S_{\delta}\coloneq\{\lambda \in \C : -\delta<\operatorname{Im}(\lambda)<0\}$.\footnote{We remark that \cite{martinez2002resonance} can be used to show that $F$ does not vanish on a logarithmic region, however this is not needed in this article.}
\item \label{F:prop4} For each $\e>0$, there exists a $C>0$ such that for all $r >0$
\begin{align}
\#\set{z \in \C : F(z) =0 , |z| \le r}\le C (1+r^{1+\e}).
\end{align}
\end{enumerate}
\end{prop}
For ease of exposition, we postpone the proof of this proposition until the end of the section.
We now compute the contribution of $\mathcal{G}_{F}$ in \eqref{firsttermcontributions}. Using that $\mathcal{G}_{F}$ is even, we observe the expansion
\begin{align}
\frac{1}{2\pi i} \int_0^\infty f(\lambda^2 )\mathcal{G}_{F}(\lambda)\dd \lambda&= \frac{1}{4\pi i} \int_0^\infty (\hat \phi(\lambda) + \hat \phi (-\lambda)) \mathcal{G}_F(\lambda)\dd \lambda\\
&= \frac{1}{4\pi i} \int_{-\infty}^\infty \hat \phi (\lambda)\mathcal{G}_F(\lambda)\dd \lambda \label{eq:619}
\end{align}
To compute the integral in the second line, our first objective will be to establish the uniform summability bound
\begin{lemma}\label{summabilityestimate}
Let $\mathcal{G}_F^j$ denote the $j$th summand in \eqref{AFdef}. Then, for every $\epsilon>0$ sufficiently small, we have the uniform bound
\begin{equation}\label{unifbound}
\sum_{j=1}^{\infty}|\mathcal{G}_F^j(\lambda)|\leq C_{\epsilon}\langle\lambda\rangle^{1+\epsilon},\hspace{5mm}\lambda\in\mathbb{R}
\end{equation}
with $C_\e > 0$.
\end{lemma}
The above lemma asserts, in particular that for each $\lambda$, $\mathcal{G}_F^j$ is absolutely summable with a sub-polynomial upper bound in $\lambda\in\mathbb{R}$. We note importantly that this bound holds for real $\lambda$.
\begin{proof}
Without loss of generality, we assume $\lambda_j$ are indexed in ascending order of magnitude.
That is,
\begin{equation*}
|\lambda_1|\leq |\lambda_2|\leq...\leq|\lambda_k|\leq\cdots.
\end{equation*}
Since $F$ is entire and $\lambda_j\neq 0$, we have the lower bound
\begin{equation}\label{lowerlambdabound}
\inf_{j\in\mathbb{N}}|\lambda_j|>c_0
\end{equation}
for some $c_0>0$ depending on the profile of $F$.
Moreover by Property \ref{F:prop4} in \Cref{Fprop}, we have
\begin{equation}\label{lambdajbound}
|\lambda_j|\geq Cj^{1-\epsilon}
\end{equation}
where $C>0$ depends on $\epsilon$.\footnote{
Indeed, if \eqref{lambdajbound} were not true, then for every $\tilde C >0$, there would exist a subsequence $j_k \to \infty$ such that $\abs{\lambda_{j_k}} < \tilde C (j_k )^{1-\e}$.
But because $\lambda_j$ are ordered by absolute value, we have that $j_k \le \# \set{|\lambda_{j}| \le (j_k)^{1-\e} \tilde C }\le C(1 + ((j_k)^{1-\e} \tilde C)^{1+\e})$.
This implies that $j_k \le C (1 + (j_k)^{1-\e^2} \tilde C ^{1+\e})$ which cannot hold as $k \to \infty$.
}
As above, in the sequel, we will adopt the convention that $C>0$ denotes a generic constant (which can change from line to line) depending on $\epsilon$ and on $c_0$, but is otherwise universal.
Now we consider separately the regions $|\lambda_j|> 2|\lambda|$ and $|\lambda_j|\leq 2|\lambda|$. When $|\lambda_j|\geq 2|\lambda|$, we have by straightforward algebraic manipulation and \eqref{lambdajbound},
\begin{equation*}
\begin{split}
\abs{\frac{1}{\lambda_j+\lambda}+\frac{1}{\lambda_j-\lambda}-\frac{2}{\lambda_j}}&=\abs{\frac{\lambda}{\lambda_j(\lambda_j-\lambda)}-\frac{\lambda}{\lambda_j(\lambda_j+\lambda)}}
\\
&\leq \abs{\frac{\lambda}{\lambda_j(\lambda_j-\lambda)}}+\abs{\frac{\lambda}{\lambda_j(\lambda_j+\lambda)}}
\\
&\leq C\frac{|\lambda|}{j^{2(1-\epsilon)}}
\end{split}
\end{equation*}
Above, in the last estimate, we used that $|\lambda_j+\lambda|\approx |\lambda_j-\lambda|\approx |\lambda_j|$. Hence,
\begin{equation*}
\sum_{|\lambda_j|\geq 2|\lambda|}|\mathcal{G}_F^j(\lambda)|\leq C|\lambda|.
\end{equation*}
In the region $|\lambda_j|\leq 2|\lambda|$, we may assume $|\lambda|\geq c_0 / 2$, otherwise this region is empty and we are done. We begin by showing the bound
\begin{equation}\label{summandbound}
\frac{1}{|\lambda_j\pm\lambda|}+\frac{1}{|\lambda_j|}\leq C.
\end{equation}
We already have the required bound for $\abs{\lambda_j}^{-1}$ in view of (\ref{lowerlambdabound}). To estimate $|\lambda_j\pm\lambda|$, we have for some $\delta>0$ (in view of the fact that $\lambda$ is real),
\begin{equation*}
|\lambda_j\pm\lambda|\geq |\operatorname{Im}(\lambda_j)|\geq \delta.
\end{equation*}
In the second inequality, we used Property \ref{F:prop3} of \Cref{Fprop}. Therefore, we have \eqref{summandbound}, and thus (in light of Property \ref{F:prop4} of \Cref{Fprop}), we have
\begin{equation*}
\sum_{|\lambda_j|\leq 2|\lambda|}|\mathcal{G}_F^j(\lambda)|\leq C\#\set{j: |\lambda_j|\leq 2|\lambda|}\leq C(1+|\lambda|^{1+\epsilon})\leq C\langle\lambda\rangle^{1+\epsilon}.
\end{equation*}
Combining everything yields (\ref{unifbound}), as desired.
\end{proof}
Now, we return to our analysis of \eqref{eq:619}. Using \Cref{summabilityestimate}, that $\hat{\varphi}$ is Schwartz, $\varphi(0)=0$, and Fubini's Theorem, we compute
\begin{equation*}
\begin{split}
\frac{1}{4\pi i } \int_{-\infty}^\infty \hat{\varphi}(\lambda)\mathcal{G}_F(\lambda)\dd \lambda&=\frac{1}{4\pi i}\sum_{j=1}^{\infty}\int_{-\infty}^{\infty}\hat{\varphi}(\lambda)\mathcal{G}_F^j(\lambda)\dd \lambda.
\end{split}
\end{equation*}
We note that the term $-2a$ from \eqref{AFdef} does not contribute to the sum as $\int_{-\infty}^\infty \hat \phi (\lambda ) \dd \lambda = 0$.
It remains to compute each of the above summands. For this, we will carry out a simple contour deformation. We let $R_n\to\infty$ be a sequence to be chosen. Clearly we have for each $j$
\begin{equation*}
\int_{-\infty}^{\infty}\hat{\varphi}(\lambda)\mathcal{G}_F^j(\lambda)\dd \lambda=\int_{-R_n}^{R_n}\hat{\varphi}(\lambda)\mathcal{G}_F^j(\lambda)\dd \lambda+O(R_n^{-\infty}).
\end{equation*}
Let $\gamma_1^n$ be the line from $-R_n$ to $-R_n-iR_n$, $\gamma_2^n$ be the line from $-R_n-iR_n$ to $R_n-iR_n$, and $\gamma_3^n$ be the line from $R_n-iR_n$ to $R_n$. We denote by $\gamma^n$, the union of these three lines oriented clockwise and let $D_n$ denote the rectangle enclosed by $\gamma^n$ and the real axis. Choosing $R_n$ such that the contour does not intersect any of the poles of $\mathcal{G}_F$, we have by the residue theorem,
\begin{equation}\label{contourdef2}
\int_{-R_n}^{R_n}\hat{\varphi}(\lambda)\mathcal{G}_F^j(\lambda)\dd\lambda=-\int_{\gamma^n}\hat{\varphi}(\lambda)\mathcal{G}_F^j(\lambda)\dd \lambda-2\pi i \1_{ D_n}(\lambda_j) \text{Res}(\hat{\varphi}(\lambda) \mathcal{G}_F^j(\lambda),\lambda_j).
\end{equation}
Here $\1_{D_n}(\lambda_j)$ is $1$ if $\lambda_j \in D_n$ and $0$ otherwise.
By Property \ref{F:prop4} of \Cref{Fprop} and pigeonholing, we can further arrange for the sequence $R_n$ to satisfying the following properties:
\begin{enumerate}
\item (Approximate unit spacing). The sequence $R_n$ satisfies
\begin{equation}\label{Rnapprox}
|R_n-n|<1.
\end{equation}
\item (Quantitative avoidance of poles) For $\lambda \in \gamma^n$, we have
\begin{equation}\label{lowerbound}
\inf_{j}|\lambda_j\pm\lambda|\geq Cn^{-2}.
\end{equation}
\end{enumerate}
\begin{rem}
$n^{-2}$ is not optimal but will suffice for our purposes.
\end{rem}
It remains to estimate the contributions of $\gamma_i^n$ for each $i=1,2,3$. First, for $\lambda$ in $\gamma_1^n$ or $\gamma_3^n$, we compute from \eqref{lowerbound} and the fact that the length of $\gamma^n$ is on the order of $n$,
\begin{equation*}
\int_{\gamma^n}|\hat{\varphi}(\lambda)||\mathcal{G}_F^j(\lambda)|\dd \lambda\leq Cn^3\sup_{\lambda\in \gamma^n}|\hat{\varphi}(\lambda)|.
\end{equation*}
In view of the support properties of $\varphi$ and that $\lambda$ is in the lower half plane, we have for $\lambda \in \gamma _1^n\cup \gamma_3^n$
\begin{equation*}
n^4|\hat{\varphi}(\lambda)|\leq C|\lambda|^4|\hat{\varphi}(\lambda)|\leq C\int_{0}^{\infty}|\varphi^{(4)}(t)|e^{t\operatorname{Im}(\lambda)}\dd t\leq C\|\varphi^{(4)}\|_{L^1(\mathbb{R})}.
\end{equation*}
For $\lambda$ on the contour $\gamma_2^n$, we use the fact that $\Im \lambda =-R_n$ to obtain
\begin{equation*}
|\hat{\varphi}(\lambda)|\leq Ce^{-R_n}\|\varphi\|_{L^1(\mathbb{R})}.
\end{equation*}
Combining the above, we obtain
\begin{equation}\label{vanishingcontour}
\int_{\gamma^n}\hat{\varphi}(\lambda)\mathcal{G}_F^j(\lambda)\dd\lambda\to 0,\hspace{5mm}\text{as}\hspace{2mm}n\to\infty.
\end{equation}
It remains to compute the sum of the second term on the right-hand side of \eqref{contourdef2} over $j$. Indeed, we have
\begin{align}
\sum_{j=1}^\infty \int_{-\infty}^\infty \hat \phi (\lambda ) \mathcal{G}_F^j (\lambda ) \dd \lambda &= \sum_{j=1}^\infty \lim_{n\to \infty} \int_{-R_n}^{R_n} \hat \phi (\lambda) \mathcal G_F^j (\lambda ) \dd \lambda\\
&= -2\pi i \sum_{j=1}^\infty\lim_{n\to\infty} \1_{D_n} (\lambda_j)\mathrm{Res}(\hat{\varphi}(\lambda) \mathcal{G}_F^j(\lambda),\lambda_j)\\
&= -2\pi i \left( \sum_{\Im{\lambda_j}>0}\hat \phi (-\lambda _j)- \sum_{\Im{\lambda _j} <0} \hat \phi (\lambda _j) \right)\label{eq:1119}
\end{align}
From the definition of $f$ and the fact that the $\lambda_j$ with positive imaginary part satisfy $\operatorname{Re}(\lambda_j)=0$ (by \Cref{Fprop} Property \eqref{F:prop1}), we can further write
\begin{equation}
\sum_{\operatorname{Im(\lambda_j)}>0}\hat{\varphi}(-\lambda_j)=2\sum_{\Im {\lambda_j } > 0}f(\lambda_j^2)-\sum_{\operatorname{Im}(\lambda_j)>0}\hat{\varphi}(\lambda_j) \label{eq:1130}
\end{equation}
Let $\tilde E_j = \lambda_j^2$ for $\Im{\lambda_j}>0$.
By \Cref{Fprop}, there are finitely many $\tilde E_j$ (whose cardinality we denote by $k'$) which are negative real numbers (we will later show they are the negative eigenvalues of $P_V$).
We can now rewrite \eqref{eq:1119} using \eqref{eq:1130} as
\begin{equation*}
\sum_{j=1}^\infty \int_{-\infty}^\infty \hat \phi (\lambda ) \mathcal{G}_F^j (\lambda ) \dd \lambda =-2\pi i\left( 2 \sum_{j=1}^{k'} f(\tilde E_j) - \sum_{j=1}^\infty \hat \phi (\lambda _j) \right)
\end{equation*}
Thanks to \eqref{contourdef2}, we can insert this into \eqref{BK1} to obtain
\begin{align}
\begin{split}
\tr(f(P_V) - f(P_0)) =&\frac{1}{2}\sum_{j=1}^{\infty}\hat{\varphi}(\lambda_j)+\frac{1}{4\pi i } \int_{-\infty}^\infty \hat{\varphi}(\lambda)(\mathcal{G}_{\Gamma}^+(\lambda)+\mathcal{G}_{\Gamma}^-(\lambda))\dd \lambda \\
&+ \frac{1}{2} (m_R(0) - 1) f(0) + \sum_{j=1}^k f(E_j) - \sum_{j=1}^{k'} f(\tilde E_j).
\end{split} \label{BK2}
\end{align}
\
It remains to compute the contributions of $\mathcal{G}_{\Gamma}^+$ and $\mathcal{G}_{\Gamma}^-$. We recall from \eqref{eqa2} the formula,
\begin{equation*}
\begin{split}
\mathcal{G}_\Gamma^\pm(\lambda) = \frac{2iA_\pm^{-1} \Gamma'(1 + \alpha_\pm)}{\Gamma(1 + \alpha_\pm)} + \frac{2iA_\pm^{-1}\Gamma'(1-\alpha_\pm))}{\Gamma(1-\alpha_\pm)}
\end{split}
\end{equation*}
where $\alpha_\pm=2i\lambda A^{-1}_\pm$. The main ingredient here is the following well-known formula for the digamma function,
\begin{align}
\Psi(z) \coloneq \frac{\Gamma'(z)}{\Gamma(z)} = -\gamma + \sum_{j=0}^\infty \left ( \frac{1}{j+1} - \frac{1}{j+z} \right),
\end{align}
where $\gamma$ is the Euler–Mascheroni constant. Using this expansion, one can carry out a contour deformation argument as in the analysis of the term $\mathcal{G}_F(\lambda)$ to write
\begin{align}
&\frac{1}{2\pi i}\int_0^\infty f(\lambda^2 ) (\mathcal G_\Gamma ^+ (\lambda)+G_\Gamma ^-(\lambda)) \dd \lambda \\&= \tfrac{1}{4\pi i }\int_{-\infty}^\infty \hat{\varphi}(\lambda)(\mathcal{G}_{\Gamma}^+(\lambda)+\mathcal{G}_{\Gamma}^-(\lambda))\dd \lambda\\
&=-\tfrac{1}{2}\sum_{j=0}^{\infty}\left(\hat{\varphi}\left(-iA_+\frac{j+1}{2}\right)+\hat{\varphi}\left(-iA_-\frac{j+1}{2}\right)\right).
\end{align}
The sum can be simplified as
\begin{align}
&\sum_{n=0}^{\infty}\left( \hat \phi\left( \frac{-A_- (n+1)i}{2}\right) + \hat \phi\left( \frac{-A_+ (n+1)i}{2}\right)\right)\\
\qquad \qquad \qquad \qquad \qquad &= \sum_{n=0}^\infty\int_{-\infty}^\infty \left( e^{ -tA_-(n+1)/2} +e^{-tA_+ (n+1)/2}\right) \phi(t) \dd t\\
&=\int_{-\infty}^\infty \phi(t) \sum_{n=0}^\infty \left( e^{-tA_-/2} e^{-tA_-n/2} + e^{-tA_+/2 } e^{-tA_+n/2} \right) \dd t\\
&=\int_{-\infty}^\infty \phi(t) \left( \frac{ e^{-tA_-/2}}{1-e^{-tA_-/2}} +\frac{ e^{-tA_+/2 }}{ 1 - e^{-tA_+/2}} \right) \dd t\\
&= \int_0^\infty \phi(t) \left( \frac{ 1}{e^{tA_- /2}-1} + \frac{1}{e^{tA_+ /2}-1} \right) \dd t
\end{align}
Combining this with \eqref{BK2} and \eqref{traceapp} and using that $\varphi$ is supported on $(0,\infty)$, we obtain
\begin{equation}\label{finaltrace}
\begin{split}
\Tr(U(t) - U(0)) &=\tfrac{1}{2}\sum_{j=1}^{\infty}e^{-i\lambda_jt}- \tfrac{1}{2} \left(\frac{ 1}{e^{tA_- /2}-1} + \frac{1}{e^{tA_+ /2}-1} \right)\\
&+ \tfrac{1}{2} (m_R(0) - 1) + \sum_{j=1}^k f(E_j) - \sum_{j=1}^{k'} f(\tilde E_j)
\end{split}
\end{equation}
in the sense of distributions. To conclude the proof of \eqref{eq:218}, we need to show that the $\lambda_j$ appearing in the first term on the right-hand side correspond to non-zero resonances of $V$ and that the $\tilde{E}_j$ correspond to the negative eigenvalues for $V$.
This is the content of the following lemma, which is proved in \Cref{s:proof of lemmainduct}.
\begin{lemma}\label{lemma:inductthing}
$\lambda_0$ is a pole of $R_V(\lambda)$ with multiplicity $ m $ if and only if $\lambda_0$ is an order $m$ zero of $F(\lambda) = W[u_+(\cdot,\lambda),u_-(\cdot ,\lambda)]$.
Moreover, if $0$ is a resonance (see \Cref{def:multi}) of $-\p_x^2 + V(x)$, then it is simple.
\end{lemma}
With this, we see that $\tilde E_j$ are the (negative) eigenvalues (corresponding to the poles of $R_V(\lambda)$ in the upper half plane) of $P_V$ so the sum of the last two sums in \eqref{finaltrace} are zero, and
\begin{align}
\tfrac{1}{2} \sum_{j=0}^\infty e^{-i\lambda _j t} + \tfrac{1}{2} (m_R(0) -1) = \tfrac{1}{2} \sum_{\lambda_j \in \Res(V)} e^{-i\lambda _j t} - \tfrac{1}{2}
\end{align}
which establishes the trace formula \eqref{eq:218}. It remains to finally prove \Cref{Fprop}, which we originally postponed.
\begin{proof}[Proof of \Cref{Fprop}]\phantom{ }
\noindent Properties \ref{F:prop1} and \ref{F:prop2} are standard results in scattering theory for real-valued potentials (see, for instance, \cite[\S 2.2]{dyatlov2019mathematical}).
Indeed, the roots of $F$ correspond to the resonances of $P_V$ by \Cref{lemma:inductthing}.
Because the spectrum of $P_V$ is contained in the positive real axis with possibly finitely many negative eigenvalues, the resolvent $(P_V -\lambda^2)$ is a meromorphic operator for $\Im \lambda >0$ on $L^2$ with finitely many poles along the imaginary axis corresponding to negative eigenvalues of $P_V$ (this proves \ref{F:prop1}).
Property \ref{F:prop2} follows by taking the complex conjugate of outgoing solutions to $P_V -\lambda ^2 $.
We next prove Property \ref{F:prop3}.
In view of Property \ref{F:prop1}, it suffices to show that there is some $\delta>0$ such that $F$ does not vanish anywhere on $S_{M,\delta}\coloneq S_{\delta}\cap \{\lambda: |\lambda|\geq M\}$ where $\delta>0$ and $M>0$ are some positive constants to be chosen. Moreover, thanks to \Cref{lemma:inductthing}, it is sufficient to show that $S_{M,\delta}$ is resonance-free. In view of this, we need to show that for every $\rho\in C_c^{\infty}(\mathbb{R})$ and $\lambda\in S_{M,\delta}$, there holds
\begin{equation}\label{L2L2bound}
\|\rho R_V\rho\|_{L^2\to L^2}<\infty.
\end{equation}
To set the stage, we use the standard formula for the resolvent, see for instance the discussion before \cite[Lemma 3.1]{froese1997asymptotic}.
For that we define
\[ V^{\frac12} \coloneq \text{sgn}( V ) |V|^{\frac12}, \ \ \ \ \mathbf R_V ( \lambda ) \coloneq V^{\frac12} R_0 ( \lambda )
|V|^{\frac12} \]
(note that $ V^{\frac12} | V|^{\frac12} = V $).
Multiplying the identity $R_0(\lambda) = R_V(\lambda) +R_0(\lambda )VR_V(\lambda)$ from the left and the right by $V^{1/2}$ and $\abs{V}^{1/2}$ respectively, we then have (omitting $\lambda$ for improved readability)
\[ (\text{Id} + \mathbf R_V ) V^{\frac12} R_V |V|^{\frac12} = \mathbf R_V \ \Longrightarrow \
V^{\frac12} R_V |V|^{\frac12} = (\text{Id} + \mathbf R_V )^{-1} \mathbf R_V , \]
provided $\text{Id} + \mathbf R_V$ is invertible. Inserting $ R_V = R_0 - R_V V R_0 $ into $ R_V = R_0 - R_0 V R_V $ we obtain
\[ R_V = R_0 - R_0 V R_0 + R_0 |V|^{\frac12} V^{\frac12} R_V |V|^{\frac12} V^{\frac12} R_0 , \]
and hence, if $\text{Id} + \mathbf R_V$ is invertible, we obtain the formal identity.
\begin{equation}\label{eq:resol}
R_V = R_0 - R_0 V R_0 +
R_0 |V|^{\frac12} (\text{Id} + \mathbf R_V )^{-1} \mathbf R_V V^{\frac12} R_0 .
\end{equation}
To establish \eqref{L2L2bound}, we need the following simple lemma, which will allow us to invert $\text{Id}+ \mathbf R_V$ on $S_{M,\delta}$ (and thus make use of \eqref{eq:resol}) using the Neumann series.
\begin{lemma}\label{inversionlemma}
There exist parameters $\delta,M>0$ such that for every $\lambda\in S_{M,\delta}$, there holds
\begin{equation}\label{Rv12bound}
\|\mathbf{R}_V\|_{L^2\to L^2}\leq \tfrac{1}{2}.
\end{equation}
\end{lemma}
\begin{proof}
From the representation formula for the free resolvent, it suffices to establish the estimate
\begin{equation}\label{youngineq}
|\lambda|^{-1}\sup_{x\in\mathbb{R}}\int_{\mathbb{R}}|V(x)|^{\frac{1}{2}}|V(y)|^{\frac{1}{2}}e^{|\Im\lambda||x-y|}\dd y\leq\tfrac{1}{2},\hspace{5mm}\lambda\in S_{M,\delta}.
\end{equation}
Since $V$ satisfies the sub-exponential bound $|V(x)|\leq Ce^{-\alpha|x|}$ for some $C,\alpha>0$, it follows that by taking $\delta$ small enough, there holds
\begin{equation*}
|\lambda|^{-1}\sup_{x\in\mathbb{R}}\int_{\mathbb{R}}|V(x)|^{\frac{1}{2}}|V(y)|^{\frac{1}{2}}e^{|\Im\lambda||x-y|}\dd y\leq \frac{C_{\alpha}}{M},\hspace{5mm}\lambda\in S_{M,\delta}.
\end{equation*}
The desired bound then follows by taking $M$ sufficiently large.
\end{proof}
By slightly modifying the above argument, we have for every $\rho\in C_c^{\infty}(\mathbb{R})$
\begin{equation}\label{freebounds}
\|\rho R_0 |V|^{\frac{1}{2}}\|_{L^2\to L^2}+\|V^{\frac{1}{2}}R_0\rho\|_{L^2\to L^2}\leq C_{\rho}
\end{equation}
for $\lambda\in S_{M,\delta}$ (where $\delta$ and $M$ are exactly as above, and do not depend on the choice of $\rho$). At this point, \eqref{L2L2bound} follows immediately from the formula \eqref{eq:resol}, \Cref{inversionlemma}, and \eqref{freebounds}. This completes the proof of \eqref{F:prop3}.
\medskip
We finally prove Property \ref{F:prop4}.
The bound is clear for $r\leq 1$ thanks to the fact that $F$ is non-constant and entire. We now assume $r\geq 1$. Thanks to \eqref{eq:boundonvp} and the fact that the Wronskian is constant in $x$, we have the upper bound
\begin{equation*}
|F(\lambda)|\lesssim e^{C|\lambda|\log(1+|\lambda|)}.
\end{equation*}
Moreover, since $F$ is non-constant, we can find $\lambda_0$ in the unit ball such that $F(\lambda_0)\neq 0$. Jensen's formula then yields
\begin{equation*}
\#\set{z \in \C : F(z) =0 , |z| \le r}\leq C|r|\log(1+|r|)\lesssim_{\epsilon} r^{1+\epsilon},\hspace{5mm}r\geq 1
\end{equation*}
for some constant $C$ depending on $F(\lambda_0)$, but not on $r$.
\end{proof}
\appendix
\section{Proof of \texorpdfstring{\Cref{lemma:inductthing}}{Lemma 3.3}}\label{s:proof of lemmainduct}
In proving \Cref{Fprop}, we used \Cref{lemma:inductthing}, which states that the poles of $R_V(\lambda)$ coincide (with multiplicity) with the zeros of $F(\lambda) = W[u_+ (\cdot,\lambda), u_-(\cdot,\lambda) ]$.
We define the multiplicity of a resonance in the usual way.
\begin{defi}\label{def:multi}
For a potential $V(x)$ satisfying \eqref{hyp:1}, a resonance is a pole of the meromorphic continuation of the resolvent $R_V(\lambda)$ (recall \Cref{prop:meroresolv}).
If $R_V(\lambda)$ has a pole at $\lambda _0 \in \C$, then we can write
\begin{align}
R_V(\lambda ) = \sum_{j=1}^J (\lambda - \lambda_0)^{-j}B_j +B_0 (\lambda)
\end{align}
for finite rank operators $B_j$, $B_0(\lambda)$ holomorphic near $\lambda = \lambda_0$, and $J \in \Z_{\ge 1}$ (the order of the pole).
The \textbf{multiplicity of the pole} $\lambda_0$ is the dimension of the space spanned by $B_j (L^2_c (\R))$ for $j=1,\dots ,J$.
\end{defi}
By \eqref{eq:535}, the poles of $R_V(\lambda)$ are exactly the poles of the resolvent function
\begin{align}
R(x,y,\lambda) = \frac{u_+(x,\lambda) u _-(y,\lambda) H(x-y) + u_-(x,\lambda)u_+(y,\lambda) H(y-x) }{F(\lambda)}\label{eq:810}
\end{align}
where $H(\cdot)$ is the Heaviside function. We now prove \Cref{lemma:inductthing}. We begin with the first part, where we show that $\lambda_0$ is a pole of $R_V(\lambda)$ with multiplicity $m$ if and only if $\lambda_0$ is a zero of order $m$ for $F(\lambda)$. Clearly, from the formula for $R(x,y,\lambda)$, if $F(\lambda_0)\neq 0$, then $\lambda_0$ is not a pole of the resolvent. Hence, it suffices to show that if $\lambda_0\in\mathbb{C}$ is a zero of order $m\geq 1$ for $F$, if and only if $\lambda_0$ is a pole of $R_V(\lambda)$ with multiplicity $m$. We will begin with the forward implication, which is more difficult. In our proof, we will need the following technical lemma relating the spaces spanned by the $\lambda$ derivatives of $u_+$ and $u_-$ at $\lambda=\lambda_0$.
\begin{lemma}\label{ordwronsk}
If $F(\lambda)$ vanishes to order $m\geq 1$ at $\lambda_0\in\mathbb{C}$, then there exists a polynomial $p_{m-1}(\lambda)$ of degree $m-1$ (depending only on $\lambda)$ such that the function
\begin{equation}\label{rdef}
r_{m-1}(x,\lambda)\coloneq u_+(x,\lambda)-p_{m-1}(\lambda)u_-(x,\lambda)
\end{equation}
vanishes to order $m$ at $\lambda=\lambda_0$. That is, $r_{m-1}$ satisfies
\begin{equation}\label{vanishingcond}
(\partial_{l}^{\lambda}r_{m-1})_{|\lambda=\lambda_0}=0,\hspace{5mm}0\leq l\leq m-1.
\end{equation}
In particular, we have
\begin{equation*}
\Span\{(\partial_{\lambda}^ju_+)_{|\lambda=\lambda_0} \}_{j=0}^{m-1}\subset\Span\{(\partial_{\lambda}^ju_-)_{|\lambda=\lambda_0}\}_{j=0}^{m-1}.
\end{equation*}
\end{lemma}
\begin{proof}
First, the case $m=1$ is obvious as the $W(u_+,u_-)(\lambda_0)=0$ implies that $u_+$ and $u_-$ are linearly dependent at $\lambda_0$.
For $m >1$, we construct $r_{m-1}$ by induction (for fixed $m$).
Assume that for $0\leq j<m-1$, we have constructed $r_0,...,r_j$ satisfying \eqref{rdef} and \eqref{vanishingcond} (the case $j=0$ follows easily).
We now aim to construct $r_{j+1}$.
First, we observe that there holds
\begin{equation}\label{wronskid}
\partial_{\lambda}^{j+1}W(u_+,u_-)_{|\lambda=\lambda_0}=\partial_{\lambda}^{j+1}W(r_{j},u_-)_{|\lambda=\lambda_0}=W(\partial^{j+1}_{\lambda}r_j,u_-)_{|\lambda=\lambda_0}=0
\end{equation}
where in the first equality, we used the formula \eqref{rdef} and in the second, we used the Leibniz rule and \eqref{vanishingcond}.
Moreover, we claim that $(\partial_{\lambda}^{j+1}r_j)(x,\lambda_0)$ satisfies $(P_V-\lambda_0^2 ) (\p_\lambda ^{j+j}r(x,\lambda_0))$.
To see this, we compute
\begin{equation*}
0=\partial_{\lambda}^{j+1}((P_V-\lambda^2)r_j)=(P_V-\lambda^2)\partial_{\lambda}^{j+1}r_j-2(j+1)\lambda\partial_{\lambda}^{j}r_j-j(j+1)\partial_{\lambda}^{j-1}r_j
\end{equation*}
and see by \eqref{vanishingcond} that the second two terms on the right-hand side vanish. This, with \eqref{wronskid}, shows that there exists a constant $c_j$ such that
\begin{equation*}
(\partial_{\lambda}^{j+1}r_j)(x,\lambda_0)=c_ju_-(x,\lambda_0).
\end{equation*}
We then define $r_{j+1}$ by correcting $r_j$ by a suitable degree $j+1$ monomial. We can simply take
\begin{equation*}
r_{j+1}(x,\lambda)\coloneq r_j(x,\lambda)-\frac{c_j}{(j+1)!}(\lambda-\lambda_0)^{j+1}u_-(x,\lambda).
\end{equation*}
or equivalently
\begin{equation*}
p_{j+1}(\lambda)\coloneq p_j(\lambda)+\frac{c_j}{(j+1)!}(\lambda-\lambda_0)^{j+1}.
\end{equation*}
It is straightforward to verify that $r_{j+1}$ satisfies \eqref{rdef} and $\eqref{vanishingcond}$. This completes the proof.
\end{proof}
Now, returning to the proof of \Cref{lemma:inductthing}, let us suppose $F$ has a zero of order $m\geq 1$ at $\lambda_0 \in \C$.
Then $F = (\lambda - \lambda_0)^m G(\lambda)$ for $G$ an entire function nonzero near $\lambda_0$.
The Schwartz kernel of the resolvent can be written
\begin{align}
R(x,y,\lambda) = \frac{u_+ (x,\lambda)u_- (y,\lambda ) H(x-y) + u _- (x,\lambda) u_+(y,\lambda) H(y-x)}{(\lambda - \lambda_0 )^m G(\lambda)}.\label{eq:1117}
\end{align}
By applying \Cref{ordwronsk} and multiplying \eqref{rdef} by $u_{-}$, we find that there exists a polynomial $p(\lambda)$
\begin{equation*}
\p_\lambda^j \big|_{\lambda_0} (u_+(y,\lambda)u_-(x,\lambda))=\p_\lambda^j \big|_{\lambda_0} (u_+(x,\lambda)u_-(y,\lambda))= \p_\lambda^j\big|_{\lambda_0}(p(\lambda)u_-(x,\lambda)u_-(y,\lambda)),
\end{equation*}
for $0\leq j\leq m-1$. Therefore, by Taylor expanding $u_+(x,\lambda)u_{-}(y,\lambda)$ and $u_-(x,\lambda)u_+(y,\lambda)$ about $\lambda=\lambda_0$ and using the identity $H(x-y)+H(y-x) = 1$, we can write the Schwartz kernel of the resolvent in \eqref{eq:1117} as
\begin{align}
R(x,y,\lambda) = \sum_{j=0}^{m-1} \frac{\p_\lambda^j \big|_{\lambda_0}\big(p(\lambda)u_-(x,\lambda)u_-(y,\lambda))}{j!G(\lambda)(\lambda -\lambda_0)^{m-j}} + \tilde{R}(x,y,\lambda)
\end{align}
with $\tilde{R}(x,y,\lambda)$ holomorphic in $\lambda$ near $\lambda_0$.
For notational convenience, we define $T_j(\lambda)$ as the operators with Schwartz kernel
\begin{align}
T_j(x,y,\lambda) \coloneq \p_\lambda^j(p(\lambda)u_-(x,\lambda)u_-(y,\lambda))
\end{align}
so that
\begin{align}
R(x,y,\lambda) = \sum_{j=0}^{m-1} \frac{T_j(x,y,\lambda_0)}{j!G(\lambda)(\lambda - \lambda _0)^{m-j}} + \tilde{R}(x,y,\lambda).
\end{align}
Let $B_j$ be the operator with kernel $T_j(x,y,\lambda_0)$.
It suffices to show that
\begin{align}
\dim (\Span(B_0(L^2_{c}),B_1(L^2_{c}),\dots,B_{m-1}(L^2_{c})) = m. \label{eq:1188}
\end{align}
To this end, we first observe that
\begin{align}
T_0 (x,y,\lambda) = p(\lambda) u_-(x,\lambda) u_-(y,\lambda)\label{eq:1192}
\end{align}
so that $B_0$ is a rank one operator and $\Span (B_0 (L^2_c)) = \Span(u_-(x,\lambda_0))$.
For $j\geq 1$, by the Leibniz rule, $T_j(x,y,\lambda)$ is of the form
\begin{equation}\label{eq:1209}
T_{j}(x,y,\lambda) = p(\lambda)\p_\lambda^{j} u_-(x,\lambda)u_-(y,\lambda) + \sum_{\substack{|\alpha |<j}} c_{\alpha} \p_\lambda^{\alpha_1}p(\lambda) \partial_\lambda^{\alpha_2} u_-(x,\lambda) \partial _\lambda^{\alpha_3} u_-(y,\lambda)
\end{equation}
for some coefficients $c_\alpha\in \Z_{\ge 0}$. To determine the rank of $B_j$, we need the following lemma.
\begin{lemma}\label{lemma:1222}
For each $j = 1,\dots, m$,
$\p_\lambda ^j\big|_{\lambda_0} u_-(x,\lambda)$ is not in the span of $$\set{\p_\lambda^0\big|_{\lambda_0} u_-(x,\lambda),\p_\lambda \big|_{\lambda_0} u_-(x,\lambda),\dots,\p_\lambda ^{j-1}\big|_{\lambda_0} u_-(x,\lambda)}.$$
\end{lemma}
\begin{proof}
Our first observation is to notice that by differentiating the equation $(P-\lambda^2)u_-=0$ a total of $1\leq \ell \le j$ times in $\lambda$, there holds
\begin{equation}\label{eq:1172}
(P-\lambda_0^2)(\p_\lambda ^l\big|_{\lambda_0}u_-(x,\lambda))=2l\lambda_0\p_\lambda ^{l-1}\big|_{\lambda_0}u_-+(l-1)l\p_\lambda ^{l-2}\big|_{\lambda_0}u_-
\end{equation}
with the convention that $(l-1)l\p_\lambda ^{l-2}\big|_{\lambda_0}u_-=0$ when $l=1$.
Note that the right-hand side of \eqref{eq:1172} involves derivatives of order at most $\ell -1$.
We can therefore inductively apply \eqref{eq:1172} to get that for each $\ell \in \Z_{\ge 0}$
\begin{align}
(P-\lambda_0^2)^{\ell + 1}(\partial_\lambda^\ell\big|_{\lambda_0} u_-(x,\lambda)) = 0. \label{eq:1177}
\end{align}
Let $j>1$ and suppose by contradiction that $\p_\lambda ^j\big|_{\lambda_0} u_-(x,\lambda)$ is in the span of the vectors $\set{\p_\lambda^\ell|_{\lambda_0} u_-(x,\lambda)}_{\ell=0}^{j-1}$.
By \eqref{eq:1177}, $(P-\lambda_0^2)^{j} (\p_\lambda ^j\big|_{\lambda_0}u_-(x,\lambda)) = 0$.
While on the other hand, by repeatedly using \eqref{eq:1172},
\begin{align}
0= (P-\lambda_0^2)^{j} \left(\p_\lambda ^j\big|_{\lambda_0}u_-(x,\lambda)\right) = c u_-(x,\lambda_0)
\end{align}
for some nonzero $c \in \C$.
Because $u_-(x,\lambda_0)$ is not identically zero, this gives a contradiction.
\end{proof}
We will also need the following lemma, which along with \Cref{lemma:1222} proves \eqref{eq:1188}
\begin{lemma}\label{lemma:1240}
For $j=0,\dots, m$:
\begin{equation}\label{eq:1241}
\begin{split}
&\Span(B_0(L^2_{c}),B_1(L^2_{c}),\dots,B_{j-1}(L^2_{c})) \\
&\qquad \qquad = \Span(\p^0_\lambda\big|_{\lambda _0}u_-(x,\lambda), \p^1_\lambda \big|_{\lambda _0}u_-(x,\lambda), \dots, \p^{j-1}_\lambda \big|_{\lambda _0}u_-(x,\lambda)).
\end{split}
\end{equation}
\end{lemma}
\begin{proof}
The case where $j=0$ was established by \eqref{eq:1192}.
We now assume \eqref{eq:1241} is true for a $j\leq m-1$.
Let $u \in L^2_c$ be such that $\int u(y) u_-(y,\lambda)\dd y\neq 0$, so by \eqref{eq:1209}
\begin{align}
B_{j} u = \sum_{\ell=0}^{j} c_\ell \p_\lambda^\ell|_{\lambda_0}u_-(x,\lambda)
\end{align}
for $c_\ell \in \C$ and $c_{j} \neq 0$.
Therefore, by the inductive hypothesis, we have $\p_\lambda^{j}|_{\lambda_0}u_-(x,\lambda)\in\Span(B_0(L^2_{c}),B_1(L^2_{c}),\dots,B_{j}(L^2_{c}))$, and thus
\begin{align}
&\Span(\p^0_\lambda |_{\lambda _0}u_-(x,\lambda), \p^1_\lambda |_{\lambda _0}u_-(x,\lambda), \dots, \p^{j}_\lambda |_{\lambda _0}u_-(x,\lambda)) \\
&\qquad\qquad\qquad\qquad\qquad\subset\Span(B_0(L^2_{c}),B_1(L^2_{c}),\dots,B_{j}(L^2_{c})).
\end{align}
But we also immediately have the converse inclusion as for $u\in L^2_c$, $B_{j} u$ is a linear combination of $\p^\ell_\lambda|_{\lambda_0}u_-(x,\lambda)$ for $\ell=0,\dots, j$.
\end{proof}
We have now shown that if $F(\lambda)$ is a zero of order $m$ at $\lambda_0$, then the resolvent $R(\lambda)$ has a pole of multiplicity $m$.
For the converse direction, we require the following Lemma.
\begin{lemma}\label{lemma:1281}
If the resolvent $R(\lambda)$ has a pole of order $m \in \Z_{>0}$ at $\lambda_0$, then $F(\lambda)$ has a zero of order $m$ at $\lambda_0$.
\end{lemma}
\begin{proof}
Given the hypothesis, we may write the Schwartz kernel of $R(\lambda)$ as
\begin{align}
R(\lambda ) = \sum_{j=0}^m\frac{R_j(x,y,\lambda)}{(\lambda - \lambda_0)^{m-j}}
\end{align}
with $R_j$ holomorphic near $\lambda_0$ and $R_0 \neq 0$.
This must agree with the expression of $R(\lambda)$ given in \eqref{eq:810}, which immediately implies that $F$ has a zero of order $m$ at $\lambda_0$.
\end{proof}
We now suppose that $R(\lambda)$ has a resonance of multiplicity $m$ at $\lambda_0$.
By \Cref{lemma:1281}, this implies the order of the resonance is $m$.
Indeed, if the order of the resonance at $\lambda_0$ was $m' \neq m$, then by \Cref{lemma:1281}, $F$ would have a zero of order $m'$ at $\lambda_0$.
We could then apply the forward direction to see that this implies the multiplicity of the resonance is $m'$, which is a contradiction.
\\
To conclude the proof of \Cref{lemma:inductthing}, it remains to show that if $0$ is a resonance, then it is simple. For this, we recall from \eqref{eq:598} that the transmission coefficient $T(\lambda)$ can be expressed in terms of $F(\lambda)$ as
\begin{align}
T(\lambda ) = \frac{2i\lambda}{\Gamma(1-\alpha_+)\Gamma(1-\alpha_-)F(\lambda)}.\label{eq:1184}
\end{align}
By \cite[Theorem 2.1]{MKlaus_1988}, for a large class of potentials (which include $V$), $T(0) \neq 0$.
If $0$ is a resonance of $-\p_x^2 + V(x)$, then by \eqref{eq:1184} and that $T(0)\neq 0$, $F(\lambda)$ must have a zero of order $1$.
So by the above argument, $0$ is a resonance of multiplicity $1$. This finally concludes the proof of \Cref{lemma:inductthing}.
\section{Birman-Kre\u{\i}n trace formula for exponentially decaying potentials}\label{appendix}
In this section we prove the Birman-Kre\u{\i}n trace formula for exponentially decaying potentials (i.e., where $\abs{V(x)} \le e^{-c|x|}$).
The assumption on the decay is stronger than necessary but allows an easy definition
of the multiplicity of the resonance at zero as then the resolvent continues meromorphically to a strip.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{theorem}[\bf Birman--Kre\u{\i}n trace formula in one dimension]
\label{t:BK}
Let $V \in C^\infty (\R ; \R)$ be such that $\abs{V(x)}\le e^{-c|x|}$ for some $c>0$, and set $P_V = D_x^2 + V(x)$.
Then for $ f \in \mathscr S ( \mathbb{R} ) $
the operator $ f ( P_V ) - f ( P_0) $ is of trace class and
\begin{equation}
\label{eq:BK1}
\begin{split} \tr \left( f ( P_V ) - f ( P_0 ) \right) = &
\, \frac{1}{ 2
\pi i } \int_0^\infty f ( \lambda^2 ) \tr \left( S ( \lambda)^{-1} \partial_\lambda S (
\lambda )\right) \dd \lambda \\ &
\ \ \ \ \ \ \ \ + \sum_{j=1}^Kk f ( E_j )
+ \tfrac12 ( m_R ( 0 ) - 1 ) f ( 0 )
\,,
\end{split}
\end{equation}
where $ S ( \lambda ) $ is the scattering matrix, $m_R(0)$ is the multiplicity of the resonance at $0$, and
$E_j$ are the (negative) eigenvalues of
$ P_V $.
\end{theorem}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
The proof of \Cref{t:BK} is a mild modification of the proof of the one-dimensional Birman-Kre\u{\i}n trace theorem for compactly supported potentials found in \cite[\S 2.6]{dyatlov2019mathematical}.
Our proof critically uses a determinant identity to analyze the scattering matrix.
A version of this identity was proven in \cite{froese1997asymptotic}, but for super-exponentially decaying potentials.
A minor adjustment to the proof will enable us to prove the identity for $\{\operatorname{Im}(\lambda)>-\delta\}\setminus\{0\}$.
This is a weaker result, but will suffice for our purposes.
The main technical input is the following mild adaptation of \cite[Lemma 7.6]{froese1997asymptotic} to our setting.
\begin{lemma}\label{Froeseadaptation}
Let $V\in L^{\infty}(\mathbb{R})$ be a function satisfying the exponential decay bound
\begin{equation*}
|V(x)|\leq C e^{-2c_0|x|}
\end{equation*}
for constants $C,c_0>0$.
Then for $\lambda\in\mathbb{C}$ such that $0<4|\operatorname{Im}(\lambda)|<c_0$, $\mathbf{R}_V (\lambda)$ is trace class and
\begin{align}\label{eq:R2S}
\det S ( \lambda ) = \frac{ \det (\rm{Id} + \mathbf R_V ( - \lambda ) )}{ \det ( \rm{Id} + \mathbf R_V ( \lambda ) ) },
\end{align}
where $\mathbf R_V ( \lambda ) \coloneq V^{\frac12} R_0 ( \lambda)|V|^{1/2}$ (which was introduced in the proof of \Cref{lemma:inductthing}).
\end{lemma}
The proof follows by making minor adjustments to the argument in \cite{froese1997asymptotic} (specifically \cite[Lemma 3.1 and 7.6]{froese1997asymptotic}). We include the short argument for the convenience to the reader.
\begin{proof}
\Step The first step is to show that
\begin{equation}\label{limittrace}
\lim_{L\to\infty}V^{\frac{1}{2}}\mathbf{R}_{\chi_L}(\lambda)|V|^{\frac{1}{2}}=\mathbf{R}_V(\lambda)
\end{equation}
where the limit is taken in the trace norm and $\chi_L$ denotes the indicator function on the interval $[-L,L]$.
The hypothesis on $V$ ensures that we have the bound $|V|^{\frac{1}{2}}\leq C e^{-c_0|x|}$. Our first goal will be to show that it suffices to establish \eqref{limittrace} with $V$ replaced by a simpler step function approximation $w$ of $C e^{-c_0|x|}$. To define $w$, we define the dyadic-like telescoping sequence $\mu_k$ by
\begin{equation*}
\mu_1\coloneq Ce^{-c_0},\hspace{5mm} \mu_k\coloneq C(e^{-c_0 k}-e^{-c_0 (k+1)})
\end{equation*}
where $k>1$. We then define $w$ by
\begin{equation*}
w(x)=\sum_{k=1}^{\infty}\mu_k\chi_k(x)
\end{equation*}
where $\chi_k$ is the indicator function on $[-k,k]$. As a consequence of the telescoping identity
\begin{equation*}
\sum_{k\geq j}\mu_k=Ce^{-c_0 j},
\end{equation*}
there holds $|V|^{\frac{1}{2}}w^{-1}\leq 1.$ Consequently, it suffices to establish \eqref{limittrace} with $V^{\frac{1}{2}}$ and $|V|^{\frac{1}{2}}$ replaced by $w$. To this end, we estimate
\begin{equation}\label{traceestseq}
\begin{split}
\|w\chi_LR_0\chi_Lw&-wR_0w\|_{1}
\\&=\|w\chi_LR_0(\lambda)\chi_Lw-wR_0(\lambda)\chi_Lw+wR_0(\lambda)\chi_Lw-wR_0(\lambda)w\|_1
\\
&\leq 2\|wR_0(\lambda)(1-\chi_L)w\|_1
\\
&\leq 2\|\sum_{i\geq 1}\mu_i\chi_iR_0(\lambda)\sum_{j>L}\mu_j\chi_j\|_1
\\
&\leq 2\sum_i\sum_{j>L}\mu_i\mu_j\|\chi_iR_0(\lambda)\chi_j\|_1
\\
&\leq 2\sum_{i\leq j}\sum_{j>L}\mu_i\mu_j\|\chi_jR_0(\lambda)\chi_j\|_1+2\sum_{i>j}\sum_{j>L}\mu_i\mu_j\|\chi_iR_0(\lambda)\chi_i\|_1
\\
&\leq \frac{C}{|\operatorname{Im}(\lambda)|}\left(\sum_{i\geq 1}\mu_i\sum_{j>L}je^{j(4|\operatorname{Im}(\lambda)|-c_0)}+\sum_{i\geq 1}ie^{i(4|\operatorname{Im}(\lambda)|-c_0)}\sum_{j>L}\mu_j\right)
\end{split}
\end{equation}
where in the last line, we used the elementary bound (see for instance, \cite [Corollary 7.5]{froese1997asymptotic})
\begin{equation*}
\|\chi_kR_0(\lambda)\chi_k\|_1\leq \frac{C}{|\operatorname{Im}(\lambda)|}ke^{4k|\operatorname{Im}(\lambda)|}.
\end{equation*}
Thanks to the hypothesis $|\operatorname{Im}(\lambda)|<\frac{c_0}{4}$ and the summability of $\mu_i$, it follows that the last line of \eqref{traceestseq} goes to zero as $L\to\infty$.
\Step \cite[Lemma 3.1]{froese1997asymptotic} follows without modification.
Indeed, by \cite[Proposition 7.2]{froese1997asymptotic},
\begin{align}
\textbf{R}_{\chi_1}(\lambda)=\textbf{R}_{\chi_1} (-\lambda)+F_2(\lambda)\label{eq:1599}
\end{align}
where $F_2(\lambda)$ is a rank two operator.
Conjugating \eqref{eq:1599} by unitary dilation $\psi (x) \mapsto L^{-1/2}\psi(x/L)$, multiplying the left by $V^{1/2}$ and right by $|V|^{1/2}$, and using \eqref{limittrace}, we get
\begin{align}
\textbf{R}_V(\lambda) = \textbf{R}_V(-\lambda)+F_V (\lambda)\label{eq:1604}
\end{align}
where $F_V(\lambda)$ is a rank two operator.
Rearranging \eqref{eq:1604}, and taking the Fredholm determinant gives \eqref{eq:R2S} (see \cite[eq. 3.4]{froese1997asymptotic}, and subsequent discussion).
\setcounter{step}{0}
\end{proof}
\begin{proof}[Proof of Theorem \ref{t:BK}]
For simplicity, we assume that there are {\em no} negative eigenvalues as their contribution is easy to analyze.
\Step We first show that $f(P_V) - f(P_0)$ is trace class.
Because $ P_V $ is self-adjoint, we have by Stone's formula
\begin{equation*}
\begin{split}
f ( P_V ) & = \frac{ 1 } {2 \pi i } \int_0^\infty f ( \lambda^2 ) ( R_V (
\lambda ) - R_V ( - \lambda ) ) 2 \lambda \dd \lambda \\
& = \frac{1}{ 4 \pi i } \int_\mathbb{R} f ( \lambda^2 ) ( R_V (
\lambda ) - R_V ( - \lambda ) ) 2 \lambda \dd \lambda,
\end{split}
\end{equation*}
using that the integrand is even in $ \lambda $.
The integral on the right-hand side should be interpreted as an operator $ L^2_{\rm{c} } \to L^2_{\rm{loc}} $.
We therefore have
\begin{align}
f(P_V)-f(P_0)&=\frac{1}{4\pi i } \int_\R f(\lambda ^2 ) (R_V(\lambda)-R_0 (\lambda ) -(R_V (-\lambda )-R_0 (\lambda ))2\lambda \dd \lambda\\
&= \frac{1}{4\pi i } \int_\R f(\lambda^2 ) (R_V(-\lambda)VR_0(-\lambda)-R_V(\lambda )VR_0(\lambda ))2\lambda \dd \lambda,\label{eq:1413}
\end{align}
where we use that
\begin{align}
R_V(\lambda )-R_0(\lambda ) = -R_V (\lambda )VR_0(\lambda).\label{eq:1474}
\end{align}
We now define the meromorhic family of operators
\begin{align}
B(\lambda ) \coloneq 2\lambda R_V (\lambda ) VR_0(\lambda ) \colon L^2_{\rm{c} } \to L^2_{\rm{loc}}.\label{eq:Bla}
\end{align}
We observe that this family is holomorphic in the closed upper half plane $\{\Im{\lambda} \ge 0\}$.
This is because $R_V (\lambda )VR_0(\lambda)$ has a simple pole at $\lambda =0$ (canceling the $\lambda$ is the definition of $B(\lambda)$).
This follows by using \eqref{eq:1474} and \Cref{lemma:inductthing}.
We can then rewrite \eqref{eq:1413} as
\begin{align}
f(P_V ) - f(P_0) = \frac{1}{4\pi i }\int_\R f(\lambda^2 ) (B(-\lambda) -B(\lambda) )\dd \lambda.\label{eq:fpV}
\end{align}
Fixing $\e>0$ sufficiently small, let $g \in \mathcal S(\C)$, $\supp g\subset \set{\lambda \in \C : \Im{\lambda }<\e}$, be an \textit{almost analytic extension} of $f(\lambda)$ (see for instance \cite[\S B.2]{dyatlov2019mathematical}):
\begin{align}
g(\lambda ) = f(\lambda^2 ) ,\ \lambda \in \R ,\ \dbar _\lambda g(\lambda ) = \mathcal{O}(\abs{\Im \lambda}^\infty).\label{eq:faag}
\end{align}
The support of $g$ can be made arbitrarily close to the real axis.
This is critical, as we will eventually use \Cref{Froeseadaptation} -- which is only valid in a neighborhood of the real axis (which depends on the decay of $V$ i.e., $A_\pm$).
The Cauchy--Green formula \cite[(D.1.1)]{dyatlov2019mathematical} applied to the right-hand
side of \eqref{eq:fpV} shows that
\begin{align}\label{eq:1489}
f(P_V ) - f(P_0) = \sum_{\pm}\pm\frac{1}{2\pi}\int_{\Im \lambda > 0} \dbar _\lambda g(\lambda) B(\pm\lambda) \dd m(\lambda).
\end{align}
Since there are no negative eigenvalues, the spectral theorem gives the bound
\begin{equation}
\label{eq:RVbound} \| R_V ( \lambda ) \|_{ L^2 \to L^2 } = \frac1 { d
( \lambda^2 , \mathbb{R}_+) } \leq \frac{1} { | \lambda | \Im \lambda }
\,, \ \ \Im \lambda > 0 \,. \end{equation}
In particular, when $ V = 0 $ we have the following $L^2\to H^2$ bound when $ \Im \lambda > 0 $,
\[ \begin{split} \| \lambda R_0 ( \lambda ) \|_{ L^2 \to H^2 } & \lesssim | \lambda | ( \| D_x^2 R_0
( \lambda ) \|_{ L^2 \to L^2 } + \| R_0 ( \lambda ) \|_{ L^2 \to L^2 }
) \\
& \lesssim | \lambda | ( 1 + |\lambda |^2 ) \| R_0 ( \lambda ) \|_{
L^2 \to L^2 } \\
& \lesssim \frac{( 1 + | \lambda|^2 )}{\Im \lambda} .
\end{split} \]
Combining this estimate with the sub-exponential bound $ |V | \leq e^{ - c|x| } $, ensures
$ V R_0 ( \lambda ) $ is a trace class operator for $ \Im \lambda > 0 $ and that
\[ \| \lambda V R_0 ( \lambda) \|_{ \mathcal L_1 } \lesssim \frac{ 1 + |\lambda|^2 } { \Im \lambda } , \ \ \Im \lambda > 0 . \]
From this bound, as well as \eqref{eq:Bla} and \eqref{eq:RVbound}, we find that for $\Im\lambda>0$, there holds
\begin{equation}
\label{eq:Blah} \| B( \lambda ) \|_{ {\mathcal L}_1} \lesssim
\frac{ 1 + |\lambda|^2 } { |\Im
\lambda |^2 |\lambda |} \lesssim \frac{ 1 + |\lambda|^2 } { |\Im
\lambda |^3}. \end{equation}
Using \eqref{eq:Blah} and \eqref{eq:faag} we conclude that for any $ N > 0 $, and in particular for $ N \geq 4 $,
\begin{align}
&\norm{ \int_{\pm\Im\lambda > 0} \dbar _\lambda g(\lambda) B(\pm \lambda ) \dd m(\lambda) }_{{\mathcal L}_1} \\
&\qquad\leq C_N
\int_{ 0 < \pm \Im \lambda < 1 } |\Im \lambda|^N
( 1 + |\lambda |)^{-N + 2 } | \Im \lambda |^{-3 } \dd { m}
( \lambda ) < \infty \,.
\end{align}
It follows that $f ( P_V ) - f ( P_0 )$ is trace class, as desired.
\medskip
\Step
We now relate $B(\lambda)$ to the scattering matrix.
Define the operator
\begin{align}
\mathcal{R}(\lambda) \coloneq -\p_\lambda \textbf{R}_V(\lambda) (\text{Id} + \textbf{R}_V(\lambda))^{-1}.
\end{align}
This operator is related to the scattering matrix.
Indeed, by taking the logarithmic derivative of both sides of \eqref{eq:R2S}, we get
\begin{equation}
\label{eq:Fsc} \tr ( \partial_\lambda S( \lambda ) S ( \lambda )^{-1}) = \tr \mathcal R (\lambda) + \tr \mathcal R (-\lambda).
\end{equation}
\begin{lemma}\label{lem:1560}
For $\Im \lambda > 0$, the operator $\mathcal{R}(\lambda)$ is trace class and $\tr \mathcal R (\lambda) = \tr B(\lambda)$.
\end{lemma}
\begin{proof}
{
We observe that for $\lambda \in \C$, $ \mathcal R ( \lambda ) $ is a meromorphic
family of operators in $ \mathcal L_1 ( L^2 ) $, and has no poles in the region $\Im \lambda > 0 $ (recalling our assumption that there are no negative eigenvalues).
We then have a direct analog of \cite[(2.2.33)]{dyatlov2019mathematical}:
\[ \text{Id} + \mathbf R_V ( \lambda ) = U_1 ( \lambda ) ( Q_0 + \lambda^{-1} Q_{-1} + \lambda Q_1 ) U_2 ( \lambda ) , \]
where $ U_j ( \lambda ) $ are invertible, holomorphic, and $Q_j$ are finite rank operators such that
\[ \text{rank}\hspace{1mm} Q_{-1} = 1 , \ \ \text{rank}\hspace{1mm} Q_1 = m_R ( 0 ) , \ \ Q_j Q_k = \delta_{jk} Q_j , \]
see \cite[Theorem C.10]{dyatlov2019mathematical}. }
Hence,
\begin{equation}
\label{eq:trF0} \tr \mathcal R ( \lambda ) = - \tr ( \lambda^{-1} Q_1 - \lambda^{-1} Q_{-1 } ) +
\varphi ( \lambda ) = \frac{ 1 - m_R ( 0 ) } \lambda + \varphi (
\lambda ) ,
\end{equation}
where $ \varphi ( \lambda ) $ is holomorphic in $ \Im \lambda \geq 0 $.
Equation \eqref{eq:trF0} follows by writing
\begin{align}
\tr \mathcal{R}(\lambda ) = &-\p_\lambda \log \det (U_1 (\lambda ) ) - \p _\lambda \log \det (U_2 (\lambda )) \\
&-\log (\det (Q_0 + \lambda ^{-1} Q_{-1} + \lambda Q_1)),
\end{align}
using that $Q_j Q_k = \delta_{jk}Q_j$ (and Jacobi's formula) to rewrite the third term as $$-\tr (\lambda ^{-1} Q_1 - \lambda ^{-1}Q_{-1}). $$
{
In order to control $ \varphi ( \lambda ) $, we first observe the $L^2\to H^2$ bound for $\mathbf{R}_V$
\[ \| \mathbf R_V ( \lambda ) \|_{ L^2 \to H^2 }
\leq C |\lambda| e^{ C ( \Im \lambda )_- } , \ \ |\lambda |
\geq 1 . \]
From the Cauchy formula, we also have that for $ \Im \lambda \geq 0 $, $ |\lambda | \gtrsim 1 $,
\[ \| \partial_\lambda \mathbf R_V ( \lambda ) \|_{ \mathcal L_1 }
\leq C \| \partial_\lambda \mathbf R_V ( \lambda ) \|_{ L^2 \to H^2 }
\leq C' | \lambda | . \]
Combining this bound with the definition of $ \mathcal R ( \lambda ) $, and the
invertibility of $ I + \mathbf R_V ( \lambda ) $ in the region
$ |\lambda | \gg 1 $, $ \Im \lambda \geq 0 $, we obtain for some $R_0>0$
\[ |\tr \mathcal R ( \lambda ) | \leq C | \lambda | , \ \
| \lambda | \geq R_0 , \ \ \Im \lambda \geq 0 . \]
Then, since $ \varphi ( \lambda ) $ is holomorphic in the region $ \Im \lambda \geq 0 $, we obtain the bound
\[ | \varphi (\lambda ) | \leq C ( 1 + | \lambda | ) , \ \
\Im \lambda \geq 0 . \]}
We next show that
\begin{equation}
\label{eq:FB0} \tr \mathcal R( \lambda ) = \tr B ( \lambda ) \,,
\end{equation}
where $ B ( \lambda ) $ is given by \eqref{eq:Bla}.
Observe by cyclicity of the trace (cf. \cite[Theorem B.4.9]{dyatlov2019mathematical})
\begin{align}
\tr (B(\lambda )) &= -2\lambda \tr (R_V (\lambda)V R_0(\lambda)) \\
&= -2\lambda \tr(R_0 (\lambda ) V R_0(\lambda)) - 2 \lambda \tr((R_V(\lambda) - R_0(\lambda) V R_0(\lambda))\\
&= -2\lambda \tr(V^{1/2} R_0^2(\lambda)|V|^{1/2} ) - 2 \lambda \tr((R_V(\lambda) - R_0(\lambda) V R_0(\lambda))\\
&= -2\lambda \tr(V^{1/2} R_0^2(\lambda)|V|^{1/2} ) + 2 \lambda \tr(R_0 (\lambda )V R_V (\lambda) V R_0(\lambda)) \label{eq:1589}
\end{align}
where in the last equality, we use the resolvent identity $(R_0 -R_V ) V R_0 = R_0 V R_V $.
To compute $\tr \mathcal{R}(\lambda)$, we first use the fact that $ R_0 ( \lambda ) $ is
bounded
on $ L^2 $ for $ \Im \lambda > 0 $ and hence
\[ \partial_\lambda ( \mathbf R_V ( \lambda ) ) = 2 \lambda V^{\frac12} R_0 (
\lambda)^2 |V|^{\frac12} \,. \]
Therefore $\tr \mathcal R (\lambda) $ is equal to
\begin{align}
&-2 \lambda \tr (V^{1/2} R_0 (\lambda )^2 |V|^{1/2} (1 + \textbf{R}_V(\lambda) ^{-1})\\
&=-2 \lambda \tr (V^{1/2} R_0(\lambda)^2 |V|^{1/2}) - 2 \lambda \tr (V^{1/2} R_0^2(\lambda ) |V|^{1/2}((1 + \textbf{R}_V(\lambda ))^{-1} -1)). \label{eq:1601}
\end{align}
The second term on the right-hand side can be written
\begin{align}
&-2 \lambda \tr (V^{1/2} R_0^2(\lambda ) |V|^{1/2}(1 + \textbf{R}_V(\lambda ))^{-1}(1 - (1 +\textbf{R}_V(\lambda))) ) \\
&\quad =2 \lambda \tr (V^{1/2} R_0^2(\lambda ) |V|^{1/2}(1 + \textbf{R}_V(\lambda ))^{-1}(\textbf{R}_V(\lambda)) )\\
&\quad = 2\lambda \tr (V^{1/2} R_0 ^2 (\lambda) V R_V (\lambda ) |V|^{1/2})=2 \lambda \tr (R_0(\lambda ) V R_V (\lambda )VR_0(\lambda)) \label{eq:1607}
\end{align}
where in the penultimate equality we use that $$(1 + \textbf{R}_V(\lambda ))^{-1}\textbf{R}_V(\lambda ) = V^{1/2} R_V (\lambda ) |V|^{1/2}$$ and in the final equality we use cyclicity of the trace.
Comparing \eqref{eq:1589} with \eqref{eq:1601} and \eqref{eq:1607} gives \eqref{eq:FB0}.
\end{proof}
\Step We lastly compute the trace of $f(P_V ) - f(P_0)$.
From \Cref{lem:1560} we get for $\e > 0$
\begin{align}
\tr\left(f(P_V) - f(P_0)\right) &= \frac{1}{2\pi}\sum_\pm \pm \int_{\set{\pm \Im \lambda > 0}\cap\set{|\lambda|>\e}} \dbar _\lambda g(\lambda ) \tr \mathcal R(\pm \lambda )\dd m(\lambda ) \\
&+\frac{1}{2\pi}\sum_\pm \pm \int_{\set{\pm \Im \lambda > 0}\cap\set{|\lambda|\le\e}} \dbar _\lambda g(\lambda ) \tr \mathcal R(\pm \lambda )\dd m(\lambda ).
\end{align}
The second term is $\mathcal O(\e^\infty)$ using that $\Tr \mathcal R (\lambda ) = \tr B(\lambda)$ \eqref{eq:FB0}, the trace norm bound on $B$ \eqref{eq:Blah}, and $\dbar$ bounds on $g$ \eqref{eq:faag}.
We next apply the Cauchy--Green formula (cf. \cite[(D.1.1)]{dyatlov2019mathematical}) on the first term to get
\begin{align}
&\frac{1}{2\pi}\sum_\pm \pm \int_{\set{\pm \Im \lambda > 0}\cap\set{|\lambda|>\e}} \dbar _\lambda g(\lambda ) \tr \mathcal R(\pm \lambda )\dd m(\lambda )\\
& \qquad \qquad \qquad \qquad =\frac{1}{4\pi i } \sum_\pm \ \int_{\gamma _\pm (\e)} g(\lambda ) \Tr \mathcal R (\pm \lambda ) \dd \lambda
\end{align}
where $\gamma_\pm(\e)$ is the contour going from $-\infty$ to $-\e$, then along the semicircle $|z| = \e$ to $+\e$ such that $\pm\Im {\gamma_\pm (\e)} \ge 0$, then from $+\e $ to $\infty$.
Note that
\begin{align}
\int_{\gamma_\pm (\e)} g(\lambda ) \Tr \mathcal R(\pm \lambda ) \dd \lambda &= \int_{\gamma_\pm (\e)\cap \R} g(\lambda ) \Tr \mathcal R(\pm \lambda ) \dd \lambda \\
&+ \int_{\gamma_\pm (\e)\setminus \R} g(\lambda ) \Tr \mathcal R(\pm \lambda ) \dd \lambda.\label{eq:1942}
\end{align}
The sum over $\pm$ of the second term can be estimated using \eqref{eq:trF0}
\begin{align}
\sum _\pm \int_{\gamma_\pm (\e)\setminus \R} g(\lambda ) \Tr \mathcal R(\pm \lambda ) \dd \lambda &= (m_R(0) - 1) f(0) \oint_{|\lambda| = 1} \frac{\dd \lambda}{\lambda } + \mathcal{O}(\e)\\
&= 2\pi i (m_R(0) - 1)f(0) + \mathcal{O}(\e).
\end{align}
The sum over $\pm$ of the first term of \eqref{eq:1942} can be written, using \eqref{eq:Fsc},
\begin{align}
\int_{-\infty}^\infty f(\lambda ^2 ) \Tr \p_\lambda S(\lambda ) S(\lambda)^{-1} \dd \lambda- \int_{-\e}^{\e} f(\lambda ^2 ) \Tr \p_\lambda S(\lambda ) S(\lambda)^{-1} \dd \lambda.
\end{align}
The second term is $\mathcal{O}(\e)$.
Sending $\e\to 0$, and using that $\Tr \p_\lambda S(\lambda)S(\lambda)^{-1}$ is even, gives the theorem.
\setcounter{step}{0}
\end{proof}
\printbibliography
\end{document}