On the series solutions of
integral equations in scattering

If the series

is convergent then obviously we will have $A^{j}\Psi\rightarrow 0$ strongly in $\mathfrak{H}$. Now assume that $A^{j}\Psi$ converges strongly to zero in $\mathfrak{H}$, and focus on the nontrivial opposite direction.

The main observation of Suzuki is that the convergence of the series (9) depends more on the specific local behavior of the operator $A$ relative to the given vector $\psi$ rather than its global properties on the whole space $\mathfrak{H}$ which requires that $\textrm{Spr}(A)<1$. Indeed consider the Hilbert subspace $\mathfrak{H}_{0}\subset\mathfrak{H}$ space generated by the vectors $A^{j}\Psi,\,j\in\bm{N}$, that is

Clearly $\mathfrak{H}_{0}$ is invariant by $A$, and since $\Psi$ lies in $\mathfrak{H}_{0}$ to prove that the series (9) strongly in $\mathfrak{H}_{0}$ it is sufficient to show that $A_{0}$ the restriction of $A$ to $\mathfrak{H}_{0}$ verifies $\textrm{Spr}(A_{0})<1$. Remark that since $\mathfrak{H}_{0}\subset\mathfrak{H}$ we have $\textrm{Spr}(A_{0})\leq\textrm{Spr}(A)$.

Let $\mathfrak{M}$ be the linear manifold formed by the vectors $v\in\mathfrak{H}_{0}$ satisfying $A^{j}v$ tends strongly to zero as $j\to\infty$. We first remark that the fact $j\to A^{j}\Psi$ tends strongly to zero, $\mathfrak{M}$ contains all the vectors $A^{j}\Psi,\,j\in\bm{N}$, and consequently is dense in $\mathfrak{H}_{0}$.

Let $\sigma(A_{0})$ denotes the spectrum of $A_{0}$, and set $\sigma_{-}(A_{0})=\{\lambda\in\Sigma(A_{0});\;|\lambda|<1\}$ and $\sigma_{+}(A_{0})=\{\lambda\in\Sigma(A_{0});\;|\lambda|\geq 1\}$. Since $A_{0}$ is compact $\sigma_{+}(A_{0})$ is finite, in addition there exists a rectifiable Jordan curve $\mathcal{C}_{+}$ in the resolvent set surrounding $\sigma_{+}(A_{0})$ and does not contain other eigenvalues of $\sigma(A_{0})$. Similarly there exists a rectifiable Jordan curve $\mathcal{C}_{-}$ in the resolvent set surrounding only $\sigma_{-}(A_{0})$. Then following [18], the spectral projections

verify the following identities

Recalling that $\textrm{Spr}(A_{0})=\sup_{\lambda\in\sigma(A_{0})}{|\lambda|}.$ Since $\sigma(A_{0})$ is a sequence of complex values that may converge to zero, proving that $\textrm{Spr}(A_{0})<1$ is equivalent to show that $\sigma_{+}(A_{0})$ is an empty set. Let now $v\in P_{+}\mathfrak{H}_{0}$. By the density of the set $\mathfrak{M}$ in $\mathfrak{H}_{0}$, there exists a sequence $(v_{n})_{n\in\bm{N}_{0}}\in\mathfrak{M}$ that converges strongly to $v$. $\bm{N}_{0}$ here is the set $\bm{N}\setminus\{0\}$. Denote $v_{n,\pm}=P_{\pm}v_{n}$. Therefore $v_{n}=v_{n,+}+v_{n,-}$. Remarking that $Av_{n,-}=AP_{-}v_{n}=P_{-}Av_{n}$ converges strongly to $P_{-}v=0$ leads to $v_{n,-}\in\mathfrak{M}$. Hence $v_{n,+}=v_{n}-v_{n,-}$ lies in fact in $\mathfrak{M}\cap P_{+}\mathfrak{H}_{0}$. This shows that $\mathfrak{M}\cap P_{+}\mathfrak{H}_{0}$ is indeed dense in $P_{+}\mathfrak{H}_{0}$. But since $\sigma_{+}(A_{0})$ is finite $P_{+}\mathfrak{H}_{0}$ is finite dimensional space and consequently $\mathfrak{M}\cap P_{+}\mathfrak{H}_{0}=P_{+}\mathfrak{H}_{0}$. This is clear not correct if $P_{+}\mathfrak{H}_{0}$ is not trivial (take any eigenvector of $A_{0}$ associated to $\lambda\in\sigma_{+}(A_{0})$, it obviously does not belong to $\mathfrak{M}$). Then $\sigma_{+}(A_{0})$ is an empty set, and finally $\textrm{Spr}(A_{0})<1$, which achieves the proof.

∎

Since $\psi\in\mathfrak{M}$ we deduce from Proposition 1 that the Neumann series (9) is convergent. On the other hand we deduce from (12) $A^{j}\psi=A^{j}P_{-}\psi=A_{0}^{j}\psi$. Therefore

Let $n\in\bm{N}$ be fixed. The fact that $\psi\in\mathfrak{M}$ implies that $A^{n+1}\psi\in\mathfrak{M}$. Applying inequality (13) to the vector $A^{n+1}\psi$ leads to

with $C=\|(I-A_{0})^{-1}\|\|A_{0}\|$, which finishes the proof of the Theorem.

∎

Let $V_{q}(k_{0})\varphi=\lambda\varphi$ in $\bm{R}^{d}$ with nonzero $\lambda$ and with $\varphi$ not identically zero outside any bounded ball in $\bm{R}^{d}$. Then

for $d=1$, and

for $d\in\{2,3\}$. Thus, for sufficiently large $R>0$ we have

as well as $(\Delta+k_{0}^{2})\varphi=0$ in $\{|x|>R\}$. In the case $d=1$ we readily find that

while in the case $d\in\{2,3\}$ we can follow the argument in, e.g., [32, Theorem 2.13, p. 38] to find

Hence

and this in conjunction with (3) gives

so $\Im\lambda>0$ and finally $\Re(e^{i\alpha}(1-\lambda))>0$ if $\alpha\in]-\pi/2,\pi/2[$ satisfies

where $\sigma(V_{q}(k_{0}))$ is the spectrum of $V_{q}(k_{0})$. The existence of the maximum in (17) follows from the fact that the eigenvalues of the compact operator $V_{q}(k_{0}):L^{2}(B)\rightarrow L^{2}(B)$ reside in the closed ball $\{\lambda^{\prime}\in\bm{C},\,\,\,|\lambda^{\prime}|\leq\|V_{q}(k_{0})\|_{L^{2}(B)\rightarrow L^{2}(B)}\}$ and can accumulate only at zero. These facts also imply that there exists $\varepsilon>0$ such that $|\gamma(1-\lambda^{\prime})-1|<1$ for all $\lambda^{\prime}\in\sigma(V_{q}(k_{0}))$, where $\gamma=\varepsilon e^{i\alpha}$. It remains to notice that each eigenvalue $\mu$ of $M$ is of the form $\mu=1-\gamma+\gamma\lambda^{\prime}$, where $\lambda^{\prime}$ is some eigenvalue of $V_{q}(k_{0})$, and finally that the equation $(I-M)u=\gamma\psi$ is equivalent with the equation $(I-V_{q}(k_{0}))u=\psi$. ∎

Assume $\lambda\in\bm{C}\setminus\{0\}$ and $\varphi\in L^{2}(]0,L[)$, $\varphi\not\equiv 0$, satisfy

Integration by parts readily shows the equivalence of the Lippmann-Schwinger equation (18) with the Helmholtz system

where $s=\left(1+q_{0}/\lambda\right)^{1/2}$. The eigenvectors of the Laplacian on $]0,L[$ are generally of the form

and we then readily find that (19) is equivalent with (20) together with

Next define the constant $T(k_{0}L)>0$ by $T(k_{0}L)\sinh(k_{0}LT(k_{0}L))=1$. Using the last condition in (21), we find that, necessarily,

which in turn implies

As an example, Figure 1 shows the set $\Lambda(q_{0},k_{0}L)$, as well as the numerically computed spectrum, for two sets of parameter values for $k_{0}$, $L$, and $q_{0}$.

Since $\Im\lambda>0$ and since the eigenvalues of $V_{q}(k_{0})$ accumulate precisely at zero, we have

for

here $s(t)$ is given by the above definition of $S_{\pm}(k_{0}L)$. Furthermore, if $\Re\omega(t)<1$ then $(\Re\omega(t)-1)/\Im\omega(t)<0$, while $\Re\omega(t)\geq 1$ implies

the latter can be seen by rewriting $\Re\omega-1\geq 0$ as

and noting that $(\Re s)^{2}-(\Im s)^{2}-1>0$ and $(\Re s)^{2}(\Im s)^{2}>0$. Now since also

we have

so $\Re\omega(t)\geq 1$ implies

that is, we get a similar estimate on $(\Re\omega(t)-1)/\Im\omega(t)$ as we do on $\|V_{q}(k_{0})\|$. Next, define

and

With $\alpha\in(0,\pi/2)$, $\varepsilon>0$, and $\gamma=\varepsilon e^{i{\alpha}}$, we have $\arg(\gamma(1-\omega))\in[\xi_{-},\xi_{+}]$ for all $\omega\in\Lambda(q_{0},k_{0}L)$, and if

then $|\gamma(1-\omega)-1|<1$ for all $\omega\in\Lambda(q_{0},k_{0}L)$, and specifically $|\gamma(1-\lambda^{\prime})-1|<1$ for all eigenvalues $\lambda^{\prime}$ of $V_{q}(k_{0})$; the condition (24) can be deduced by requiring $\Im(\gamma(1-\omega))<|\tan(\pi-2\arg(\gamma(1-\omega)))|$ and using $|1-\omega|\leq 1+\|V_{q}(k_{0})\|$. Note that we must choose $\alpha<\pi/2$ rather than $\alpha=\pi/2$ since