Feynman path integrals for magnetic Schrödinger operators on infinite weighted graphs (Q1996487)

On an infinite weighted graph \((X,b,m)\) where \(X\) is a countable set with a positive Radon measure \(m\) and \(b\) is a graph on \(X\) which is a nonnegative symmetric function on \(X\times X\) with the properties \(b(x,x)=0\) and \(\sum_{y \in X}b(x,y) <\infty\) for all \(x \in X\), the paper under review considers the Cauchy problem for the Schrödinger equation: \( \tfrac{d}{dt} \psi(t) = -iL_{v,\theta}\,\psi(t), \; \psi(0) = f, \) having not only \textit{electric potential} [\(v: X \rightarrow {\mathbb R}\)] but also \textit{magnetic potential} [\(\theta: \{(x,y)\in X \times X\,;\; b(x,y)>0\} \rightarrow {\mathbb R}\)], to represent its solution as a Feynman path integral formula. The discrete Schrödinger operator \(L_{v,\theta}\) on the graph concerned is given formally as \[ L_{v,\theta}\,f(x) = \tfrac1{m(x)}\sum_{y \in X}b(x,y)(f(x)- e^{i\theta(x,y)}f(y)) + v(x)f(x), \quad x \in X, \] when applied to a function \(f\) in the linear space \(C_c(X)\) of functions on \(X\) with finite support. Here the magnetic potential \(\theta\) is an antisymmetric function satisfying \(\theta(x,y) = - \theta(y,x),\; x,y \in X\), and the electric potential \(v\) is an arbitrary function on \(X\). Then \(L_{v,\theta}\) can make a quadratic form with form domain \(C_c(X)\) in the Hilbert space \(\ell^2(X,m) = \ell^2(X,b,m)\). If this quadratic form is semi-bounded from below and closable, one can, through the methods of quadratic forms, redefine the operator \(L_{v,\theta}\) as \textit{the} self-adjoint operator associated with the closure of this quadratic form. The main result is the following Feynman path integral formula representing the integral kernel \(e^{-itL_{v,\theta}}(x,y)\) of this discrete Schrödinger unitary group \(e^{-itL_{v,\theta}}\) for all \(t\geq 0,\, x,\, y \in X\): \begin{align*} e^{-itL_{v,\theta}}(x,y) &= \tfrac1{m(y)}\int_{\{\omega \in \Omega\,;\; \omega(0)=x,\,\omega(t)=y,\, N_t(\omega)<\infty\}}\, i^{N_t(\omega)} e^{{\mathcal A}_t(v,\theta|\omega)}\, d{\mathbb P}_x(\omega), \\ \text{ with } {\mathcal A}_t(v,\theta|\omega) :&= i\int_0^t \theta(d\omega(s)) -i\int_0^t\big(v(\omega(s)) +\text{deg}(\omega(s))\big)ds + \int_0^t\text{deg}(\omega(s))ds, \quad t\geq 0\,, \end{align*} when \(\theta\) and \(v\) are such that the quadratic forms on the form domain \(C_c(X)\) associated with both the operators \(L_{v,\theta}\) and \(L_{-\text{deg},0}\) are semi-bounded from below and closable. -- The notations used here are briefly explained as follows. \(\Omega\) is the space of right-continuous jump paths \(\omega: [0,t] \rightarrow X\). \(\{{\mathbb P}_x\}_{x \in X}\) is the Markov family of probability measures on \(\Omega\), each \({\mathbb P}_x\) on \(\{\omega \in \Omega;\,\omega(0)=x\}\); this family is to be induced through the quadratic form (here probably better said as ``through the regular Dirichlet form'', cf. [\textit{M. Fukushima} et al., Dirichlet forms and symmetric Markov processes. 2nd revised and extended ed. Berlin: Walter de Gruyter (2011; Zbl 1227.31001)]) associated with the operator \(L_{v,\theta}|_{v=0, \theta=0}\) having no electric and magnetic potentials. \(N_t(\omega) \in {\mathbb N}\cup\{\infty\}\) is the number of jumps of the path \(\omega\) until the time \(t\). \(\int_0^t \theta(d\omega(s)): \Omega \ni \omega \rightarrow {\mathbb R}\) is the line integral of \(\theta\) along the path \(\omega\)\,. For \(x \in X\), \(\text{deg}(x)\) \(\equiv\) \(\text{deg}_m(x) := \tfrac1{m(x)}\sum_{y\in X} b(x,y)\) is what is called \textit{weighted degree function} on \((X,b,m)\), which means, e.g. if \(m\equiv 1\), \(\text{deg}_1(x)\) is the number of edges emerging from the vertex \(x\). The result will be interesting, since a measure-theoretic path integral approach is in general not allowed to the Schrödinger unitary group. For the heat semigroup which is well-known to possess the \textit{Feynman-Kac-Itô formula} as a genuine path integral formula with the Wiener measure on the space of \textit{Brownian motion paths}, a similar treatment on graphs has also been done by \textit{B. Güneysu} et al. [Probab. Theory Relat. Fields 165, No. 1--2, 365--399 (2016; Zbl 1341.81027)].