The time slicing approximation of the fundamental solution for the Schrödinger equation with electromagnetic fields (Q1389868)

The authors construct, for a short time \(| t-s| \), by a time slicing approximation, the fundamental solution \(K(\hbar, t, s, x, y)\) of the Schrödinger equation \[ i \hbar \partial_t u(t,x)= [\tfrac{1}{2} (- i \hbar \partial_x - A(t,x))^2 + V(x)] u(t,x), \quad t \in {\mathbb{R}}, \;x \in {\mathbb{R}}^d. \] For the vector and scalar potentials \(A(t,x)= (A_1(t,x),\dots, A_d(t,x))\) and \(V(x)\) and the magnetic fields \(B(t,x)= (B_{jk}(t,x)):= (\partial_{x_k} A_j(t,x) - \partial_{x_j} A_k(t,x))\), it is assumed that for every multi-index \(\alpha\), \[ | \partial_x^{\alpha} A(t,x)| + | \partial_x^{\alpha}\partial_t A(t,x)| \leq C_{\alpha}, \quad \partial_x^{\alpha} B(t,x)| \leq C_{\alpha} (1 + | x|)^{-1-\varepsilon}, \quad | \alpha| \geq 1, \] \[ | \partial_x^{\alpha} V(x)| \leq C_{\alpha}, \quad | \alpha| \geq 2, \quad (t,x) \in {\mathbb{R}} \times {\mathbb{R}}^d, \] with constants \(C_{\alpha}\) and some \(\varepsilon > 0\). The approximation is based on the idea of Feynman path integral that \(K(\hbar, t, s, x, y)\) be given by a formal ``integral'' \[ \int_{\{\gamma: \gamma(s)= y, \gamma (t)= x\}} e^{iS(\gamma)/\hbar} {\mathcal D}[\gamma], \quad\text{where}\quad S(\gamma):= \int_s^t L(\tau, {\dot \gamma}(\tau), \gamma(\tau)) d\tau \] is the action for the path \(\gamma: [s,t] \rightarrow {\mathbb{R}}^d\) with Lagrangian \(L(\tau, {\dot \gamma}, \gamma)= (1/2) | {\dot \gamma}| ^2 + A(\tau,\gamma){\dot \gamma} - V(\gamma)\), and \({\mathcal D}[\gamma]\) is a uniform ``measure'' on the space of all the paths \(\gamma\) joining \((s,y)\) and \((t,x)\). For each division \(\Delta: s= \tau_0 < \tau_1 < \cdots < \tau_L= t\) of the interval \([s,t]\) and any \(x_j \in {\mathbb{R}}^d\), \(j= 0, 1, \dots, L\), but with \(x_L= x, x_0= y\), let \(q_{\Delta}\) be the piecewise classical path joining the \((\tau_j, x_j)\), and put \(S(q_{\Delta})= S(x_L, \dots, x_0)\). Then the main result of this paper is, for \(| t-s| \) small enough, not only to prove that as \(| \Delta| \equiv \max_{1 \leq j \leq L} (\tau_j - \tau_{j-1})\) tends to zero, so that \(L\) tends to infinity, \[ K(\Delta; \hbar, t, s, x, y):= \prod_{j=1}^L\frac{1}{(2 \pi i \hbar (\tau_j - \tau_{j-1}))^{d/2}} \int_{{\mathbb{R}}^{d(L-1)}} e^{i \hbar^{-1} S(x_L, \dots,x_0)} \prod_{j=1}^{L-1} dx_j \] converges to \(K(\hbar, t, s, x, y)\) uniformly on \({\mathbb{R}}^d \times {\mathbb{R}}^d\), though all its \((x, y)\)-derivatives converge locally uniformly on \({\mathbb{R}}^d \times {\mathbb{R}}^d\), but also to give such an expression for \(K(\hbar, t, s, x, y)\) as \[ K(\hbar,t,s,x,y)= \frac {e^{i\hbar^{-1} S(t,s,x,y)}} {(2\pi i \hbar (t - s))^{d/2} (1+ (t-s)d(t, s, x, y))^{1/2}} (1+ r(\hbar, t, s, x, y)). \] Here \(d(t,s,x,y)\)= \(O(| t-s|+ | x-y|)\) for small \((| t-s|+ | x-y|)\), and further both \(d(t,s,x,y)\) and \(\hbar^{-1} | t-s| ^{-2} r(\hbar, t, s, x, y)\), together with all their \((x,y)\)-derivatives, are uniformly bounded in \(x, y, t, s\) with \(| t-s| \) small, and also in \(\hbar\) for the latter. It is also shown that \(1+ (t-s)d(t, s, x, y)= \det_2 (I+ GY)e^{{\text Tr} GZ}\). In this formula, \(G\) is the Green operator for \(-d^2/d\tau^2\) on \([s,t]\) with Dirichlet boundary condition, and \(Y= B(\tau, q(\tau)) d/d\tau+ Z(\tau)\), where \(Z(\tau) =(Z_{jk}(\tau))\), \[ Z_{jk}(\tau)= \sum_{m=1}^d \partial_{x_k} B_{jm}(t,q(\tau)){\dot q}_m(\tau)- \partial_{x_k} \bigl(\partial_t A_j(t, q(\tau))+ \partial_{x_j}V(q(\tau))\bigr) \] with \(q\) the classical path joining \((s,y)\) and \((t,x)\). If \({\mathcal L}= L^2([s,t];{\mathbb{R}}^d)\) and \({\mathcal H}= H_0^1([s,t]; {\mathbb{R}}^d)\), one sees that \(G\) is a Hilbert-Schmidt operator of \({\mathcal L}\) to \({\mathcal H}\) and \(Y\) a bounded operator of \({\mathcal H}\) to \({\mathcal L}\), so that \(GY\) and \(GZ\) are operators of Hilbert-Schmidt and of trace class, respectively, on \({\mathcal H}\), denoting the regularized determinant by \(\det_2\) and the trace by Tr. \textit{D. Fujiwara} [Lect. Notes Math. 1540, 39-53 (1993; Zbl 0799.46052)] first proved the case where the vector potential \(A(t,x)\) is absent, expressing, in his previous related work [Nagoya Math. J. 124, 61-98 (1991; Zbl 0741.41027)] on the stationary phase method on a space of large dimension, \(K(\Delta; \hbar, t, s, x, y)\) as a main term plus a remainder term for large \(L\). In this case one has \(Y= Z= \bigl(- \partial_{x_k} \partial_{x_j}V(q(\tau))\bigr)\), so that \(GY= GZ\) is trace class, then having only to use the usual determinant `\(\det\)' but not `\(\det_2\)' as in the above formula. The paper under review extends his result to the case where both vector and scalar potentials are present, using \textit{T. Tsuchida}'s extension [Nagoya Math. J. 136, 157-189 (1994; Zbl 0822.35113)] of Fujiwara's latter work to the present case. The new point is that the authors find how \(Y\) and \(Z\), as in the above formula, depend on the magnetic fields and consequently use the regularized determinant. Thus the issue has been well settled by the paper. For some previous related works on construction of the fundamental solution for the Schrödinger equation, refer to \textit{D. Fujiwara} [J. Anal. Math. 35, 41-96 (1979; Zbl 0418.35032); Duke Math. J. 47, 559-600 (1980; Zbl 0457.35026)] in case without vector potential and \textit{K. Yajima} [J. Anal. Math. 56, 29-76 (1991; Zbl 0739.35083)] in case with both vector and scalar potentials. The conditions on the vector potential used in the present paper is in fact due to Yajima in his work just mentioned.