The minimax principle and related topics in the Jordan setting (Q2197210)

The result nowadays called Courant-Fischer minimax theorem proves an estimation of the eigenvalues of a Hermitian compact operator \(A\) on a Hilbert space \(H\) without any explicit use of its eigenvectors nor its characteristic polynomial, concretely, if \((\lambda_n (A))_n\) denotes the sequence of all eigenvalues of \(A\), arranged in decreasing order and counting multiplicity and extended with zeros in case that it is finite, then for each natural \(n\) we have \begin{align*} \lambda_n (A) &= \max_{M} \min_{\xi\in M,\ \|\xi\| =1} \langle A(\xi), \xi\rangle \\ &= \min_{N} \max_{\xi\in M,\ \|\xi\| =1}\langle A(\xi), \xi\rangle, \end{align*} where in the first maximum, \(M\) runs in the set of all \(n\)-dimensional subspaces of \(H\), while in the minimum, \(N\) runs in the set of all \((n-1)\)-codimensional subspaces of \(H\). In the finite-dimensional case, the result is due to \textit{E. Fischer} [Monatsh. Math. Phys. 16, 234--249 (1905; JFM 36.0166.01)], while a subsequent generalization for Hermitian compact operators on an infinite-dimensional Hilbert space was established by \textit{R. Courant} [Math. Z. 7, 1--57 (1920; JFM 47.0455.02)]. The Courant-Fischer minimax principle laid the groundwork for some interesting results like Cauchy's interlacing theorem and Weyl's inequalities. The latter is the starting point of the study of several inequalities involving eigenvalues and singular values of compact operators enriched by G. Polya, A. Horn and K. Fan. For example, the Ky Fan maximum principle [\textit{K. Fan}, Proc. Natl. Acad. Sci. USA 35, 652--655 (1949; Zbl 0041.00602)] provides an extremum property of the sum of the first \(k\) eigenvalues of a self-adjoint compact operator on a Hilbert space. We have already mentioned singular values, actually, the minimax principle can be stated for a not necessarily self-adjoint compact operator, \(A\), by just replacing the eigenvalues of \(A\) and the values of \(\langle A(\xi), \xi\rangle\) with its singular values (i.e., the eigenvalues of \((A^*A)^{1/2}\)) and values of \(\|A(\xi)\|\), respectively. \textit{U. Hirzebruch} [Math. Ann. 186, 65--69 (1970; Zbl 0183.31103)] generalized the Courant-Fischer min-max theorem on eigenvalues of symmetric matrices to the case of formal-real Jordan algebras, or equivalently, finite-dimensional \(JB\)-algebras in the standard terminology. A Weyl perturbation inequality and a Cauchy-Poincaré interlacing theorem in Euclidean Jordan algebras were obtained by \textit{M. S. Gowda} et al. [Linear Algebra Appl. 430, No. 8--9, 1992--2011 (2009; Zbl 1168.15003)] and by \textit{M. S. Gowda} and \textit{J. Tao} [Linear Multilinear Algebra 59, No. 1--3, 65--86 (2011; Zbl 1259.17024)]. The novelties in the Jordan setting include tools based on the Peirce decomposition of an element in a simple Euclidean Jordan algebra and the bounds for the inequalities are constructed by taking reduced traces of elements constructed from the set of primitive idempotents of the algebra. In the present note, the author extends the study to not necessarily self-adjoint elements in the wider setting of weakly compact \(JB^*\)-algebras and \(JB^*\)-triples. There are several natural obstacles arising in the Jordan setting, for example, unlike in the case of finite-dimensional \(C^*\)-algebras, these results for non-self-adjoint elements cannot be directly derived from their analogs for self-adjoint elements due to the lack of an optimal notion of modulus (\((A^*A)^{1/2}\)) in the Jordan setting. This was not an obstacle in the case of the \(JB^*\)-algebra of all symmetric compact operators on a (separable) Hilbert space, where a minimax principle was derived by \textit{J. Danciger} [Linear Algebra Appl. 412, No. 1, 22--29 (2006; Zbl 1082.15017)] in the finite dimensional case and by \textit{J. Danciger} et al. [Math. Nachr. 281, No. 6, 786--802 (2008; Zbl 1140.47003)] in the infinite-dimensional case. The new contributions in this paper allow us to enlarge our scope to the setting of elements in a weakly compact \(JB^*\)-triple. Theorem 3.5 contains a generalized minimax principle for weakly compact \(JB^*\)-triples where the singular values of an element, arranged in decreasing order and counting multiplicity, are characterized by tripotents and minimal tripotents, respectively, instead of subspaces and elements of the Hilbert space. Briefly speaking, for each element \(x\) in a weakly compact \(JB^*\)-triple \(\mathcal{U}\) and each natural \(n\), we have \begin{align*} \lambda_n (x) &= \max_{e} \inf \{ \|x\|_v : v\in \mathcal{U}_2 (e) \hbox{ minimal tripotent} \} \\ &= \min_{f} \sup \{ \|x\|_v : v\in \mathcal{U}_0 (f) \hbox{ minimal tripotent} \}, \end{align*} where \(e\) (respectively, \(f\)) runs in the set of tripotents in \(U\) of rank \(n\) (respectively, \(n-1\)), \(\lambda_n (x)\) is the \(n\)-th singular value in the triple spectral resolution of \(x\), and \( \|x\|_v\) stands for the preHilbertian seminorm associated with the the tripotent \(v\) appearing, for example, in the definition of the strong\(^*\) topology and Grothendieck's inequalities. A Weyl-type inequality and a Cauchy-Poincaré interlacing theorem are obtained as consequences of this result. The study in this paper is much more ambitious, and will serve as motivation for further developments. For the sake of brevity, we shall simply add that Section~4 is entirely devoted to develop a generalized Ky Fan maximum principle for weakly compact \(JB^*\)-triples. The problem of the continuity of the spectral resolutions is treated in Section~5. The last section of this paper contains a generalization of a recent result for finite-dimensional \(C^*\)-algebras, by showing that every weakly compact \(JB^*\)-triple satisfies the property (\(co\)), that is, informally speaking, each perturbation of a convex combination of elements in its closed unit ball coincides with a convex combination of perturbations of the elements also in its closed unit ball.