Pólya’s theorem

6 Treatment of the integral in the Fourier inversion

In this part, we analyze the integral

\begin{align} I_r = \iint _{[-\pi ,\pi ]^d} \Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{r} (\theta ) \big) \; \mathrm{d}^d \theta , \end{align}

where \(\widehat{\mathrm{G}}_{r} \colon \mathbb {R}^d \to \mathbb {C}\) is the Fourier transform of the regularized Green’s function of a random walk \(X\) on \(\mathbb {Z}^d\). By Corollary 5.7, the finiteness of this integral in the limit \(r \nearrow 1\) characterizes expectation transience of \(X\).

6.1 Decomposition of the integral

The main integral \(I_r\) can be decomposed into two parts: an easy "high frequency part", which contains the contributions of \(\theta \) away from \(0\), and a more interesting "low frequency part", which contains the contributions of \(\theta \) near \(0\).

Definition 6.1
#

For any \(\delta {\gt} 0\), define the integrals

\begin{align} J_r^{(\delta )} = \; & \iint _{[-\pi ,\pi ]^d\setminus B_\delta } \Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{r} (\theta ) \big) \; \mathrm{d}^d \theta \\ K_r^{(\delta )} = \; & \iint _{B_\delta } \Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{r} (\theta ) \big) \; \mathrm{d}^d \theta , \end{align}

where \(B_\delta := \left\{ \theta \in \mathbb {R}^d \, \big| \, \| \theta \| {\lt} \delta \right\} \) is the ball of radius \(\delta \) centered at \(\vec{0} \in \mathbb {R}^d\).

Lemma 6.2
#

For any \(0 {\lt} \delta \le \pi \) we can write

\begin{align*} I_r = J_r^{(\delta )} + K_r^{(\delta )} . \end{align*}
Proof

Obvious, since \([-\pi ,\pi ]^d= ([-\pi ,\pi ]^d\setminus B_\delta ) \cup B_\delta \) is a disjoint union.

6.2 Dominated convergence away from the origin

TODO: Define non-degenerate step distribution (essentially \(\sum _{u\in \mathbb {Z}^d} p(u) e^{\mathrm{i}u \cdot \theta } \ne 1\) for \(\theta \ne 0\) modulo periodicity).

Lemma 6.3

If \(X\) is a time-homogeneous Markovian random walk with suitable non-degeneracy conditions on its step distribution (to be written down more precisely), then for any \(0 {\lt} \delta \le \pi \) the limit

\begin{align*} \lim _{r \nearrow 1} J_r^{(\delta )} \end{align*}

exists and is finite (limit in \(\mathbb {R}\)).

Proof

Under the nondegeneracy conditions, on the compact set \([-\pi ,\pi ]^d\setminus B_\delta \), the continuous integrand \(\theta \mapsto \widehat{\mathrm{G}}_{r}(\theta )\) is bounded (and therefore dominated by a constant function) and it has the pointwise limit \(\lim _{r \nearrow 1} \widehat{\mathrm{G}}_{r}(\theta ) = \widehat{\mathrm{G}}_{1}(\theta )\). It therefore follows from the dominated convergence theorem that \(\lim _{r \nearrow 1} J_r^{(\delta )} = J_1^{(\delta )} \in \mathbb {R}\).

6.3 Monotone convergence near the origin

TODO: Think about the best conditions for step distribution under which monotone convergence can be applied (real-valuedness requires symmetricity of the step-distribution?!?).

Lemma 6.4

If \(X\) is a time-homogeneous Markovian random walk with suitable symmetricity and integrability conditions on its step distribution (to be written down more precisely), then there exists a \(\delta _0 {\gt} 0\) such that for any \(0 {\lt} \delta \le \delta _0\), the limit

\begin{align*} \lim _{r \nearrow 1} K_r^{(\delta )} \end{align*}

is increasing and exists in \([0,+\infty ]\).

Proof

We can now rephrase the recurrence criterion in terms of only the low frequency integral.

A nice random walk \(X= \big(X(t)\big)_{t \in \mathbb {N}}\) on \(\mathbb {Z}^d\) is expectation recurrent if and only if for any small \(\delta {\gt}0\) \(\lim _{r \nearrow 1} \, K_r^{(\delta )} = +\infty \). In other words, \(X\) is expectation transient if and only if for some small \(\delta {\gt}0\) we have \(\lim _{r \nearrow 1} \, K_r^{(\delta )} \, {\lt} \, +\infty \).

Proof

6.4 Characterizing finiteness of the integral for simple random walk

For the simple random walk \(X= \big(X(t)\big)_{t \in \mathbb {N}}\) on \(\mathbb {Z}^d\), there exists a small \(\delta _0 {\gt} 0\) (something like \(\delta _0 = \frac{\pi }{8}\)) such that for all \(\theta \in B_{\delta _0} \setminus \left\{ 0 \right\} \) we have \(\Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{r}(\theta ) \big) \uparrow \Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{1}(\theta ) \big)\) as \(r \uparrow 1\) and

\begin{align*} \Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{1}(\theta ) \big) \asymp \frac{1}{\| \theta \| ^{2}} . \end{align*}
Proof

Positivity estimates and Taylor series approximation of the Fourier transform of the Green’s function given explicitly in Lemma 5.4. …

Lemma 6.7

For the simple random walk \(X= \big(X(t)\big)_{t \in \mathbb {N}}\) on \(\mathbb {Z}^d\) with \(d {\gt} 2\), we have \(\lim _{r \nearrow 1} \, K_r^{(\delta )} \, {\lt} \, +\infty \) for \(\delta {\gt} 0\) small enough.

Proof

By the increasing limit part of Lemma 6.6 and monotone convergence theorem, we have

\begin{align*} K_r^{(\delta )} \uparrow K_1^{(\delta )} = \iint _{B_\delta } \Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{1} (\theta ) \big) \; \mathrm{d}^d \theta . \end{align*}

By the second part of Lemma 6.6, \(\Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{1}(\theta ) \big) \ge C_1 \| \theta \| ^{-2}\) (for some positive constant \(C_1\)), so using this lower bound in integration and switching to radial coordinates, we have

\begin{align*} K_1^{(\delta )} \; \ge \; \iint _{B_\delta } \frac{C_1}{\| \theta \| ^2} \; \mathrm{d}^d \theta \; = \; C_1 \, A_d \int _{0}^{\delta } \rho ^{d-3} \; \mathrm{d}\rho \; = \; + \infty . \end{align*}
Lemma 6.8

For the simple random walk \(X= \big(X(t)\big)_{t \in \mathbb {N}}\) on \(\mathbb {Z}^d\) with \(d \le 2\), we have \(\lim _{r \nearrow 1} \, K_r^{(\delta )} \, = \, +\infty \) for any \(\delta {\gt} 0\) small.

Proof

By the increasing limit part of Lemma 6.6 and monotone convergence theorem, we have

\begin{align*} K_r^{(\delta )} \uparrow K_1^{(\delta )} = \iint _{B_\delta } \Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{1} (\theta ) \big) \; \mathrm{d}^d \theta . \end{align*}

By the second part of Lemma 6.6, \(\Re \mathfrak {e}\big( \widehat{\mathrm{G}}_{1}(\theta ) \big) \le C_2 \| \theta \| ^{-2}\) (for some positive constant \(C_2\)), so using this upper bound in integration and switching to radial coordinates, we have

\begin{align*} K_1^{(\delta )} \; \le \; \iint _{B_\delta } \frac{C_2}{\| \theta \| ^2} \; \mathrm{d}^d \theta \; = \; C_2 \, A_d \int _{0}^{\delta } \rho ^{d-3} \; \mathrm{d}\rho \; {\lt} \; + \infty . \end{align*}