Let $X$ be an $n$-dimensional manifold with nonempty boundary $\partial X$ and $n\geq 2$. Proposition 4.1 of this paper by Schrohe states that it is "not very difficult" to show:

Proposition: A bounded operator on $L^2(X)$ with range in $H^{n+1}(X)$ is trace class.

Tragically, I cannot figure out a proof. I've tried generalizing the approach outlined in chapter 8 of Roe's Elliptic Operators, Topology and Asymptotic Methods for the boundaryless case, I've thought about Mercer's Theorem but the setting seems quite different, and looked around in Hörmander's PDEs book III and I don't think it's in there. A hint, proof outline or reference would be greatly appreciated, thank you!

  • 1
    $\begingroup$ consider posting this on MO if nothing happens once the bounty expires, it is maybe quite technical $\endgroup$
    – FShrike
    Feb 19 at 18:29

1 Answer 1


At the beginning of the section where Schrohe states Proposition 4.1 (top of page 14 in your link), Schrohe cites Nest, R. & Schrohe, E., Dixmier’s trace for boundary value problems, Manuscripta Mathematica, 96(2), 203–218 (1998), available at https://doi.org/10.1007/s002290050062 (but paywalled), which contains this (at page 209):

Proposition 2.1. A bounded operator on $L^2(X)$ with range in $H^n(X)$ is an element of $\mathcal{L}^{1,\infty}(L^2(X))$; if its range even is contained in $H^{n+1}(X)$ then it is trace class.

Proof: Let $A$ be bounded on $L^2(X)$ with range in $H^n(X)$. It is well-known that there is an invertible pseudodifferential operator $R$ of order $−n$ on $M$ such that $R_{+}: L^2(X) \to H^n(X)$ is an isomorphism with inverse equal to $(R^{−1})_{+}$, for a proof see e.g. [7, Theorem 3.2.14]. We then may write $A = R_{+}(R^{−1})_{+} A$. The composition $(R^{−1})_{+} A$ yields a bounded operator on $L^2(X)$, since $(R^{−1})_{+}: H^n(X) \to L^2(X)$ is bounded. On the other hand the singular values of $R_{+}$ on $L^2(X)$ can be estimated in terms of the singular values of $R$ on $L^2(M)$ and the norms of the operators $e^{+}: L^2(X) \to L^2(M)$ and $r^+: L^2(M) \to L^2(X)$. Since $R$ is of order $−n$, it is an element of $\mathcal{L}^{1,\infty}(L^2(M))$ according to Theorem 1.1. So we get the first assertion. The second statement is proven similarly, noting that operators of order $−n−1$ are trace class.


  • The reference [7] is Grubb, G., Functional Calculus of Pseudodifferential Boundary Problems [mis-cited in the text as "Functional Calculus for Boundary Value Problems"], Progress in Mathematics, vol. 65. Birkhäuser, Boston (2d ed. 1996), available at https://doi.org/10.1007/978-1-4612-0769-6 (but paywalled), and
  • "Theorem 1.1" is credited to Connes, and stated as follows: "Let $P \in \Psi(M)$ have order $−n$. Then $P: L^2(M, E) \to L^2(M, E)$ defines an element of $\mathcal{L}^{1,\infty}(L^2(M,E))$ and $\operatorname{Tr}_{\omega}(P) = \frac{1}{(2 \pi)^n n} \operatorname{res} P$, independent of $\omega$." For proof, the paper refers to Connes, A. The action functional in non-commutative geometry, Comm. Math. Phys. 117, 673–683 (1988), available at this link (no paywall).

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .