Some commutation principles for optimization problems over transformation groups and semi-FTvN systems
M. Seetharama Gowda · 2025
Read Some commutation principles for optimization problems over transformation groups and semi-FTvN systems by M. Seetharama Gowda (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-08654.txt.
Introduction
Gowda and David Sossa set out two grades of commutativity for optimization over symmetry groups, in a 2025 paper. In many such problems, the best solution works well with the symmetries of the data. When variables commute, they mark that good fit.
They work in a broad setting that includes the classic matrix case. They show the stronger grade implies the weaker one. Both grades show up as conditions for the best solution in certain problems. That link is the main payoff of the paper. Each complete hyperbolic polynomial gives a system of the kind they need. From this, several principles for such problems follow. The frame also brings together scattered results from matrix analysis and polynomial optimization.
Catalog record
| CREATOR | M. Seetharama Gowda |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 84 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-08654
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Some commutation principles for optimization problems over transformation groups and semi-FTvN systems M. Seetharama Gowda Abstract We introduce the concepts of commutativity relative to a transformation group and strong commutativity in the setting of a semi-FTvN system and show their appearance as optimality conditions in certain optimization problems. In the setting of a semi-FTvN system (in particular, in an FTvN system), we show that strong commutativity implies commutativity and observe that in the special case of Euclidean Jordan algebra, commutativity and strong commutativity concepts reduce, respectively, to those of operator and strong operator commutativity. We demonstrate that every complete hyperbolic polynomial induces a semi-FTvN system. By way of an application, we describe several commutation principles. Key Words : Transformation group, semi-FTvN system, (strong) commutativity, commutation principle, Euclidean Jordan algebra, hyperbolic polynomial AMS Subject Classification: 15A27, 17C20, 46N10, 90C33. 1 Introduction The main objective of the paper is to describe two commutativity concepts and show their relevance in certain commutation principles. For the first commutativity concept, we consider a real inner product space {\cal V} and the corresponding space \mathrm{L({\cal V})} of all bounded linear transformations equipped with the operator norm. Within \mathrm{L({\cal V})} , let \mathrm{GL({\cal V})} denote the set of all invertible linear transformations. Given a closed subgroup {\cal G} of \mathrm{GL({\cal V})} , let \mathrm{Lie({\cal G}):=\{D\in\mathrm{L({\cal V})}:\exp(tD)\in{\cal G}\,\,\mbox{for all}\,\,t\in\mathbb{R}\}}, (1) where, for a given D , \mathrm{\exp(tD)} denotes the sum of the power series \sum_{k=0}^{\infty}\frac{t^{k}D^{k}}{k!} that converges uniformly on bounded sets of \mathbb{R} to an element in \mathrm{L({\cal V})} . (When {\cal V} is finite-dimensional, \mathrm{Lie({\cal G})} is the Lie algebra of the matrix Lie group {\cal G} .) Given elements a and b in {\cal V} , we say that a commutes with b relative to {\cal G} if \langle Da,b\rangle=0 for all \mathrm{D\in\operatorname{Lie}({\cal G})} . As we see below, in the setting of a Euclidean Jordan algebra {\cal V} with its (algebra) automorphism group {\cal G} , this definition is equivalent to that of operator commutativity. In particular, for real symmetric matrices (or complex Hermitian matrices) A and B , this reduces to the usual commutativity AB=BA . The motivation for introducing the above concept comes from the so-called commutation principles [ 30 , 26 , 9 , 8 , 19 ] , where commutativity appears as an optimality condition in certain optimization problems. The following result - to be elaborated and proved later - provides an illustration: Theorem 1.1 Consider {\cal V} and {\cal G} as above. Let \Theta:{\cal V}\rightarrow\mathbb{R} be a Fréchet differentiable function and F:{\cal V}\rightarrow\mathbb{R} be a {\cal G} -invariant function. Let E be a set in {\cal V} that is {\cal G} -invariant. If a is a local optimizer of the problem \underset{x\in E}{min/max}\quad\Theta(x)+F(x), then a commutes with \Theta^{\prime}(a) relative to {\cal G}. Our second commutativity concept is defined in the setting of a semi-FTvN system. A semi-FTvN system is a triple ({\cal V},{\cal W},\lambda) , where {\cal V} and {\cal W} are real inner product spaces and \lambda:{\cal V}\rightarrow{\cal W} is a mapping that is “norm-preserving” and ”inner product expanding”: ||\lambda(x)||=||x||\,\,\mbox{and}\,\,\big\langle x,y\big\rangle\leq\langle\lambda(x),\lambda(y)\big\rangle\quad(\forall\,x,y\in{\cal V}). (2) In a semi-FTvN system ({\cal V},{\cal W},\lambda) , we define strong commutativity of elements x and y in {\cal V} by the condition \langle x,y\rangle=\langle\lambda(x),\lambda(y)\rangle. (3) For commutativity in this system, we consider the group \mathrm{Aut({\cal V},{\cal W},\lambda)} of all invertible bounded linear transformations A:{\cal V}\rightarrow{\cal V} satisfying the condition \lambda(Ax)=\lambda(x) for all x\in{\cal V} . As \lambda is norm-preserving, this is a closed subgroup of the orthogonal group of {\cal V} . Considering a closed subgroup {\cal G} of \mathrm{Aut({\cal V},{\cal W},\lambda)} , we define \mathrm{Lie({\cal G})} by ( 1 ) and formulate the definition of commutativity relative to {\cal G} as we did earlier. We shall see that • In a semi-FTvN system, strong commutativity implies commutativity (relative to {\cal G} ). How do semi-FTvN systems arise? First, a semi-FTvN system is a generalization of the so-called FTvN system. An FTvN system (short for Fan-Theobald-von Neumann system) is a triple ({\cal V},{\cal W},\lambda) , where {\cal V} and {\cal W} are real inner product spaces and \lambda:{\cal V}\rightarrow{\cal W} is a mapping satisfying the condition \max\Big\{\!\left<c,x\right>:\,x\in[u]\Big\}=\left<\lambda(c),\lambda(u)\right>\quad(\forall\,c,u\in{\cal V}) (4) with [u]:=\{x\in{\cal V}:\lambda(x)=\lambda(u)\} denoting the so-called \lambda -orbit of u . It is shown in [ 11 ] , Section 3, that in such a system, the mapping \lambda satisfies ( 2 ); thus, every FTvN system is a semi-FTvN system. Numerous examples and properties of FTvN systems are described in the articles [ 7 , 11 , 18 , 17 ] . A simple example of an FTvN system is the triple ({\cal V},\mathbb{R},\lambda) , where {\cal V} is a real inner product space and \lambda(x)=||x|| . Non-trivial examples include Euclidean Jordan algebras , systems induced by complete and isometric hyperbolic polynomials , and normal decomposition systems (Eaton triples) , see [ 7 ] . In the setting of a Euclidean Jordan algebra (carrying the trace inner product), the strong commutativity relation ( 3 ) reduces to the strong operator commutativity of x and y which means that x and y have their spectral decompositions with respect to a common Jordan frame where their coefficients (eigenvalues) appear in the decreasing order [ 1 ] . This commutativity is stronger than the operator commutativity where (we merely require that) two elements have their spectral decompositions with respect to a common Jordan frame. We show, by rephrasing a recent result in [ 19 ] , that • In a Euclidean Jordan algebra, commutativity relative to the (algebra) automorphism group is equivalent to operator commutativity. The concept of strong commutativity ( 3 ) was initially studied in the setting of FTvN systems (where it was just called commutativity). For example, it was shown in [ 7 ] , Section 3.1, that in a FTvN system ({\cal V},{\cal W},\,\lambda) , for any c\in{\cal V} , \phi:{\cal W}\rightarrow\mathbb{R} , and any spectral set S (which is a set of the form \lambda^{-1}(Q) for some Q\subseteq{\cal W} ) in {\cal V} , \sup_{x\in S}\Big\{\!\left<c,x\right>+(\phi\circ\lambda)(x)\Big\}=\sup_{u\in\lambda(S)}\Big\{\!\left<\lambda(c),u\right>+\phi(u)\Big\} (5) with the attainment of one supremum implying the attainment of the other and additionally implying a relation of the form ( 3 ). Furthermore, it was shown in [ 18 ] that under certain conditions, ( 4 ) is equivalent to the Fenchel conjugate formula (\phi\circ\lambda)_{S}^{\ast}(z)=\phi_{\lambda(S)}^{\ast}\big(\lambda(z)\big) and to the subdifferential relation y\in\partial_{S}(\phi\circ\lambda)(\overline{x})\Longleftrightarrow\lambda(y)\in\partial_{\lambda(S)}\phi\big(\lambda(\overline{x})\big)\text{ and $y$ strongly commutes with $\overline{x}$}. (6) Semi-FTvN systems also arise from complete hyperbolic polynomials. We recall [ 5 , 2 ] that given a finite-dimensional real vector space {\cal V} and a nonzero element e\in{\cal V} , a real homogeneous polynomial p of degree n on {\cal V} is said to be hyperbolic relative to e if p(e)\neq 0 and for every x\in{\cal V} , the roots of the univariate polynomial t\rightarrow p(te-x) are real. Given such a polynomial p , for any x\in{\cal V} , let \lambda(x) denote the vector of roots of the polynomial t\rightarrow p(te-x) with entries written in decreasing order. We say that p is complete if \lambda(x)=0\Rightarrow x=0 . Based on a construction due to Bauschke et al [ 2 ] , we show that • Every complete hyperbolic polynomial induces a semi-FTvN system. In addition to Theorem