Documentation

TauCeti.Analysis.Semigroups.BoundedGenerator.Perturbation

Comparing commuting bounded-generator semigroups #

This file gives the perturbation estimate needed to compare the bounded semigroups in the Yosida approximation. If bounded operators A and B commute and their exponentials are bounded by M and N at nonnegative times, then

‖exp (t A) x - exp (t B) x‖ ≤ M * N * t * ‖(A - B) x‖.

The proof applies Duhamel's formula to the difference of the two exponentials. Commutativity moves A - B through the second exponential, after which the two exponential factors contribute their bounds. This pointwise estimate is sharper than the generic Banach-algebra bound involving exp (t ‖A‖) and exp (t ‖B‖); that generic bound is useless for Yosida approximations because their operator norms grow with the approximation parameter.

The two constants are kept separate rather than fixed to one: the Yosida approximations of a dissipative operator have contractive exponentials, but under the Hille--Yosida hypotheses for a general growth constant M they only satisfy ‖exp (t A_lambda)‖ ≤ M, and the resulting comparison carries the factor M ^ 2.

Main result #

References #

Engel--Nagel, One-Parameter Semigroups for Linear Evolution Equations, Section II.3.5; Pazy, Semigroups of Linear Operators and Applications to Partial Differential Equations, Chapter 1.

theorem TauCeti.Semigroups.norm_exp_smul_sub_exp_smul_apply_le_of_commute {X : Type u_1} [NormedAddCommGroup X] [NormedSpace X] [CompleteSpace X] {M N : } (A B : X →L[] X) (hcomm : Commute A B) (hA : ∀ (s : ), 0 sNormedSpace.exp (s A) M) (hB : ∀ (s : ), 0 sNormedSpace.exp (s B) N) {t : } (ht : 0 t) (x : X) :
(NormedSpace.exp (t A)) x - (NormedSpace.exp (t B)) x M * N * t * (A - B) x

If A and B commute and their bounded-generator semigroups obey the uniform bounds M and N at nonnegative times, then their orbits differ by at most

‖exp (t A) x - exp (t B) x‖ ≤ M * N * t * ‖(A - B) x‖ for t ≥ 0.

This is the bounded Duhamel estimate in the commuting case; the contraction case is M = N = 1.