Traces and quasi-traces on the Boutet de Monvel algebra. (Q1774075)

The authors' aim is to recognize a natural definition of trace for the Boutet de Montel algebra, (say (BdM)-algebra), i.e., a suitable algebra for pseudodifferential operators on compact, finite dimensional smooth manifolds with boundary, containing also explicit informations coming from the boundary problems. Here for trace one denotes a nontrivial linear functional tr\(:{\mathcal A}\to \kappa\), on some \(\kappa\)-algebra \({\mathcal A}\), \(\kappa={\mathbb R}, {\mathbb C}\), which vanishes on commutators: \(tr([A,B])=0\), \(\forall A,B\in{\mathcal A}\). (This fits the standard definition for linear operators on finite dimensional vector spaces.) The authors state that their approach is a generalization of the previous one by Kontsevich and Vishik (KV), given for a suitable class of pseudodifferential operators defined on closed finite dimensional smooth manifolds. (Let us denote by TR\((A)\) the (KV) trace.) This last approach goes over the definition of trace introduced by Wodzicki and, independently, by Guillemin that identified this with the residue, res\((A)\). In fact, res\((A)\) is determined from finitely many terms in the asymptotic expansion of the symbol of \(A\) (local character), and it vanishes on operators of noninteger order or of order \(<-n\), where \(n\) is the dimension of the manifold. Instead TR\((A)\) is well defined and nontrivial when \(A\) is of noninteger order or of order \(<-n\), and it vanishes on commutators of these types. Furthermore, TR\((A)\) depends on the full operator and cannot be determined from the terms in the asymptotic expansion of the symbol (global character), and equals the standard trace when the order is \(<-n\), it is moreover defined on integer order operators with certain properties. The authors remark that the (KV)'s approach, based on studies on generalized zeta functions \(\zeta(A,P,s)\equiv \text{TR}(AP^{-s})\), where \(P\) is an auxiliary elliptic operator, can be extended to compact manifolds with boundary, by considering the asymptotic expansion of the resolvent trace TR\((A(P-\lambda)^{-N})\). In particular, on an \(n\)-dimensional smooth compact manifold \(X\), with boundary \(\partial X\equiv X'\), they consider as an auxiliary operator \(P\) the (invertible) Dirichlet realization of a strongly elliptic principally scalar second-order differential operator, and an elliptic operator on the boundary. Then, they identify with a suitable coefficient \(C_0(A,P)\), in this expansion, the ``canonical trace'', and relate it to the zero-power coefficient in the Laurent series for the zeta function TR\((AP^{-s})\) at \(s=0\), when \(P\) is invertible. Really \(C_0(A,P)\) is a quasi-trace in the sense that it vanishes on commutators modulo local terms, and has a specific value independent of the auxiliary operator \(P\), modulo local terms. \(C_0(A,P)\) becomes a trace, i.e., the local ``errors'' vanish, when \(A\) is a singular Green operator of noninteger order, or of integer order with a certain parity. Remark. Let us emphasize that the generalization of the trace given by the authors has been possible thanks to the fact that they work on the (BdM)-algebra, that contains also informations on the boundary problem. So the translation of a boundary problem in a suitable algebra is really the very important fact. This has been also, independently and in a completely different framework, proved in the Dirac quantization of PDE's. In fact, such a quantization can be algebraically characterized by means of a suitable noncommutative algebra, built with the informations coming from some boundary problem. (See recent works by the reviewer of this article.)