Logarithms and exponentials in Banach algebras (Q2341457)

This readable and interesting article catches our attention by promising an ``elementary'' proof of a classical result in functional analysis: If \(0\) is in the unbounded connected component of the complement of the spectrum of an invertible element of a complex Banach algebra, then that element is an exponential, and, hence, has a logarithm. By ``elementary'', the authors mean that the holomorphic functional calculus will be bypassed. The only requisite is given in Lemma 2.1, which demonstrates that, in a unital real or complex Banach algebra with a sub-multiplicative norm, an element \(f\) for which \(\|\mathbf{1}-f\|<1\) is an exponential. The first proof presented of the main result quickly fumbles the ball by requiring the reader to know how the spectrum of an element relative to a subalgebra is related to the spectrum in the larger algebra. Happily, this brief slip-up is immediately followed by the promised elementary version. Here, we are asked to consider a path, parameterized by \(\phi(t)\), that connects \(0\) to \(\infty\) and lies entirely outside the spectrum of the given invertible element \(a\). Lemma 2.1 implies that some subset of the associated path \(\Phi(t)=a-\phi(t)\cdot\mathbf{1}\) must consist of exponentials; thus, when we are shown that this preferred subset is both open and closed as a set, it follows that \(\Phi(0)=a\) must be an exponential. This seems about as uncluttered a proof as possible. The appeal of the paper is enhanced by the discussion in Sections 2 and 3, which focus on related results in real Banach algebras and their application to real (\(n\times n\)) matrices. In this setting, the \textit{real-symmetric spectrum} of an element \(x\) of a real, unital Banach algebra \(A\) is defined by \[ \sigma^*(x):=\{\lambda\in{\mathbb C}:(a-\lambda\mathbf{1})(a-\bar{\lambda}\mathbf{1})\;\text{not\;invertible\;in\;}A\}. \] Theorem 2.1 shows that, if \(0\) is in the unbounded connected component of \({\mathbb C}\backslash\sigma^*(x)\), then \(x^2\) is an exponential; moreover, if the interval \((-\infty,\,0]\) lies in this unbounded connected component, then \(x\) itself is an exponential. Since a real matrix has only a finite number of eigenvalues, the hypothesis automatically holds for any invertible real matrix. Thus, for instance, (Corollary 3.5) an invertible real matrix \(M\) can be written as \(M=e^B\), for some real matrix \(B\), if, and only if, \(M=A^2\) for some real matrix \(A\). Theorem 3.3 asserts that every real matrix with positive determinant is the product of two exponential matrices. A simple example illustrates, nonetheless, that such a matrix need not have a real logarithm. However (see Corollary 3.6), if an \(n\times n\) matrix \(A\) is invertible, then the \(2n\times 2n\) matrix \(M= A\oplus A\) is indeed the exponential of a real matrix.