Spectral shift function in the large coupling constant limit (Q5937219)

Let \(H_0\) be a selfadjoint operator and \(H(\alpha) = H_0 + \alpha V\) its perturbed selfadjoint operator in a Hilbert space, where \(\alpha\) is a positive coupling constant. If the operator \(V\) is of trace class, Krein and Lifshits introduced the spectral shift function \(\xi(\lambda, \alpha) = \xi(\lambda; H(\alpha), H_0)\), which is integrable and satisfies \[ \text{Tr} [\varphi(H(\alpha)) - \varphi(H_0)] = \int_{\mathbf R} \xi(\lambda, \alpha) \varphi'(\lambda) d\lambda \] for \(\varphi\) in a suitable class of functions. Hence we have formally \[ \xi(\lambda, \alpha) = \text{Tr} (E_{H_0}(-\infty, \lambda) - E_{H(\alpha)}(-\infty, \lambda)). \] Here and below \(E_M(\delta)\) denotes the spectral projection of a selfadjoint operator \(M\) corresponding to an interval \(\delta\) in \({\mathbf R}\). In the paper under review, in case \(V\) is not necessarily of trace class, the author deals with the following representation: \[ \xi(\lambda; H(\alpha), H_0) = \int_{\mathbf R} \text{ind} (E_{J+\alpha(A_{\lambda}+ tB_{\lambda})}(-\infty, 0), E_J(-\infty, 0)) \frac{dt}{\pi(1+t^2)}, \] a.e. \(\lambda \in {\mathbf R}\), as a new definition of \(\xi(\lambda; H(\alpha), H_0)\), calling it the generalized spectral shift function. Here \( \text{ind}(P, Q)\) is the Fredholm index of the operator \(P-Q-I\) for projections \(P\) and \(Q\): \(\text{ind}(P, Q) = \text{dim} \text{Ker} (P-Q-I) - \text{dim} \text{Ker} (P-Q+I)\), and \(V = V_+ - V_-\) with \(2V_{\pm} := |V|\pm V \geq 0\) and \(J\) the sign of \(V\). The following conditions are assumed: (1) \(|V|^{1/2}|H_0+iI|^{-1/2}\) is a compact operator; (2) The limit \(X_{\lambda} := \lim_{z\rightarrow \lambda + i0}|V|^{1/2}(H_0-zI)^{-1}|V|^{1/2}\) exists in operator norm for a.e. \(\lambda \in {\mathbf R}\); (3) Putting \(A_{\lambda} = \text{Re} X_{\lambda}\) and \(B_{\lambda} = \text{Im} X_{\lambda}\), \(B_{\lambda}\) satisfies \(\int_0^{\infty} \text{rank} E_{B_{\lambda}^*B_{\lambda}}(s^2, +\infty) ds < \infty\). This representation was found, in case \(V\) is of trace class and sign-definite, i.e. \(V \geq 0\) or \(V \leq 0\), by \textit{A. B. Pushnitskij} [St. Petersbg. Math. J. 9, No. 6, 1181-1194 (1997); translation from Algebra Anal. 9, No. 6, 197-213 (1997; Zbl 0916.47011)]. Then this is extended to the cases \(V\) is not necessarily sign-definite and not necessarily of trace class in \textit{F. Gesztesy} and \textit{K. A. Makarov} [J. Anal. Math. 81, 139-183 (2000; Zbl 0964.47013)] and \textit{A. Pushnitski} [J. Funct. Anal. 183, No. 2, 269-320 (2001; Zbl 0998.47003)], and there it is also observed that this \(\xi(\lambda; H(\alpha),H_0)\) coincides with the spectral shift function under pretty wide conditions on \(H_0\) and \(V\). However, as the author writes, it seems to be still unclear whether the above definition leads to a new class of perturbations. The main result is to have given an asymptotic behaviour of the leading term of this \(\xi(\lambda; H(\alpha), H_0)\) as \(\alpha \rightarrow \infty\). It does not depend on \(\lambda\), as is especially noted. For this purpose, two additional conditions are required for a point \(\mu\) in the resolvent set of \(H_0\): (4) For some \(p>0\), \(\xi(\mu; H_f(\alpha), H_0)\) with an auxiliary operator \(H_f(\alpha) := H_0+ \alpha f(J)|V|\) in place of \(H(\alpha)\), where \(f\) is a real-valued function on \({\mathbf R}\) with \(\pm f(\pm 1) >0\), has asymptotics \((C_+|f(1)|^p - C_-|f(-1)|^p)\alpha^p\) as \(\alpha \rightarrow \infty\), where the constants \(C_{\pm}\) are independent of \(f\) but may depend on \(V_{\pm}\); (5) The operator \(\alpha(A_{\lambda}+tB_{\lambda}-X_{\mu})\), a sort of variation when passing from \(\lambda\) to \(\mu\), is small in a certain sense. It is mentioned that condition (4) is satisfied in most of applications. The results are applied to the Dirac operator.