A simple counterexample related to the Lie-Trotter product formula (Q444681): Difference between revisions

The ``addition formula'' \[ e^{A + B} = \lim_{n \to \infty} \Big( e^{A /n} e^{B/n}\Big)^n \] for matrix exponentials is due to Sophus Lie. Its generalization to semigroups \[ U(t)u = \lim_{n \to \infty} \Big( S \Big( {t \over n} \Big) T \Big( {t \over n} \Big) \Big)^n u \eqno(1) \] was given by \textit{H. F. Trotter} [Proc. Am. Math. Soc. 10, 545--551 (1959; Zbl 0099.10401)]; here, \(A\) (resp., \(B)\) is the infinitesimal generator of the strongly continuous semigroup \(S(t)\) (resp., \(T(t))\) and \(U(t)\) is the semigroup generated by \(\overline{A + B}\). In the language of differential equations in Banach spaces, formula (1) is the basis of the splitting method of assembling the solution \(u'(t) = (A + B)u(t)\) from the solutions of \(u'(t) = Au(t)\) and \(u'(t) = Bu(t).\) The validity of Trotter's formula (1) requires additional assumptions. In a slightly generalized version, it is assumed that \[ \Big \| \Big(S\Big( {t \over n} \Big) T \Big( {t \over n} \Big) \Big)^n \Big \| \leq Me^{\omega t} \quad (t \geq 0, \;n = 1, 2, \dots) \eqno(2) \] and that \(\overline{A + B}\) is an infinitesimal generator; in another version, this last requirement follows from other hypotheses. The question arises: does the sole assumption that \(\overline{A + B}\) is a semigroup generator guarantee (1)? The answer is known to be ``no'' through a counterexample of \textit{F. Kühnemund} and \textit{M. Wacker} [Semigroup Forum 60, No.~3, 478--485 (2000; Zbl 0976.47021)]. The authors provide a simpler counterexample using operator matrices; the limit in (1) is zero, while \(U(t)u \neq 0\).