Distance and best approximations in operator norm and trace class norm
Saikat Roy · 2025
Read Distance and best approximations in operator norm and trace class norm by Saikat Roy (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-2505-12059.txt.
Introduction
Saikat Roy, writing solo in 2025, takes up a plain question: given a target and a tidy set of stand-ins, how near can one come, and which comes nearest? His stage is the space of operators, the math objects that stretch and turn spaces, under two ways of measuring size. For each way he derives the gap outright.
Finite cases drawn from quantum-style algebras get the same treatment. Worked examples show the formulas earning their keep: where a search would grind, the closed forms answer at once. Best-fit problems of this kind sit under signal work, control, and quantum theory, so a clean gap formula has reach beyond its own page.
Catalog record
| CREATOR | Saikat Roy |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-12059
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Distance and best approximations in operator norm and trace class norm Saikat Roy Abstract . We study the best approximation and distance problems in the operator space ℬ ( ℋ ) \mathcal{B}(\mathcal{H}) and in the space of trace class operators ℒ 1 ( ℬ ( ℋ ) ) \mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) . Formulations of distances are obtained in both cases. The case of finite-dimensional C ∗ C^{*} -algebras is also considered. The computational advantage of the results is illustrated through examples. 2020 Mathematics Subject Classification Primary 46B28, 46B02 Secondary 47L05. 1. Introduction Throughout the article \mathcal{H} denotes an infinite-dimensional separable Hilbert space over \mathbb{C} , equipped with an inner product \langle\cdot,\cdot\rangle . The symbol S_{\mathcal{H}} stands for the unit sphere of \mathcal{H} . For any sequence (\zeta_{n})\subset\mathcal{H} , we write \zeta_{n}\xrightarrow{w}\zeta_{0} to mean (\zeta_{n}) weakly converges to \zeta_{0} . Let \mathcal{B}(\mathcal{H}) denote the C^{*} -algebra of bounded linear operators on \mathcal{H} , \mathcal{K}(\mathcal{H}) , the ideal of compact operators, and \mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) , the ideal of trace class operators of \mathcal{B}(\mathcal{H}) . If \mathcal{V} is a finite-dimensional subspace of \mathcal{B}(\mathcal{H}) , we define N_{\mathcal{V}}:=\{a\in\mathcal{B}(\mathcal{H}):~Tr(ay)=0,~y\in\mathcal{V}\}=\{a\in\mathcal{B}(\mathcal{H}):~Tr(ay_{j})=0,~1\leq j\leq k\} for any basis \{y_{1},y_{2},\dots,y_{k}\} of \mathcal{V} . For any two vectors \xi and \zeta in \mathcal{H} , \xi\otimes\zeta denotes the rank-one operator \eta\mapsto\langle\eta,\zeta\rangle\zeta ( \eta\in\mathcal{H} ). spectrum of an element x\in\mathcal{B}(\mathcal{H}) is denoted by \sigma(x) . 1.1. Motivation A fundamental result due to Holmes and Kripke [ 4 ] asserts that \mathcal{K}(\mathcal{H}) is a proximinal subspace of \mathcal{B}(\mathcal{H}) and for any x\in\mathcal{B}(\mathcal{H})\setminus\mathcal{K}(\mathcal{H}) , the distance from x to \mathcal{K}(\mathcal{H}) is given by dist(x,\mathcal{K}(\mathcal{H}))=\Delta(x) , where \Delta(x):=\sup\{\lim_{n}\sup~\|x\zeta_{n}\|:~(\zeta_{n})\subset S_{\mathcal{H}},~\zeta_{n}\xrightarrow{w}\mathbf{0}\} (see also [ 1 , 3 ] ). It is natural to ask whether dist(x,\mathcal{V})=\Delta(x) , if we consider a proper subspace \mathcal{V} of \mathcal{K}(\mathcal{H}) instead of the whole space \mathcal{K}(\mathcal{H}) . The answer to this question is negative, as shown in Example 2.1 . Thus, the quantity \Delta(x) becomes a lower bound of dist(x,\mathcal{V}) . This leads to the question: what is the exact formulation of dist(x,\mathcal{V}) for x\in\mathcal{B}(\mathcal{H})\setminus\mathcal{K}(\mathcal{H}) and a proper subspace \mathcal{V} of \mathcal{K}(\mathcal{H}) ? In this article, we formulate dist(x,\mathcal{V}) in two cases. In the first case, we consider \mathcal{V} to be any finite-dimensional subspace of \mathcal{B}(\mathcal{H}) , and this answers the above question in the general framework of \mathcal{B}(\mathcal{H}) which in particular applies to the case of \mathcal{K}(\mathcal{H}) . In the second case, we consider the distance problem in the trace class norm when x\in\mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) and \mathcal{V} is any finite-dimensional subspace of \mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) . 1.2. Approach A general formulation of distance from a point x to a subspace Y in a Banach space X is given by the classical duality principle [ 12 ] . In [ 10 ] , an alternative approach is presented for real reflexive Banach spaces, sometimes under the assumption of strict convexity, which is focused on norm attainment of functionals. Such an approach is computationally advantageous as pointed out in [ 10 , Section 4] , [ 11 ] . Here we consider the duality relations between \mathcal{K}(\mathcal{H}) , \mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) and \mathcal{B}(\mathcal{H}) , and utilize the \sigma -WOT ( \sigma -weak operator topology) compactness of the closed unit ball of \mathcal{B}(\mathcal{H}) to prove our results. Our methods are based on Banach space theoretic and elementary operator algebraic techniques. The space \mathcal{B}(\mathcal{H}) is identified to the dual of \mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) through the isometric isomorphism \Omega:\mathcal{B}(\mathcal{H})\to\mathcal{L}^{1}(\mathcal{B}(\mathcal{H}))^{*} , defined as \Omega(x)=\rho_{x},\quad x\in\mathcal{B}(\mathcal{H}),~\text{where}~\rho_{x}(a)=Tr(ax),\quad a\in\mathcal{L}^{1}(\mathcal{B}(\mathcal{H})). Similarly, the space \mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) can be identified to the dual of \mathcal{K}(\mathcal{H}) through the isometric isomorphism \Gamma:\mathcal{L}^{1}(\mathcal{B}(\mathcal{H}))\to\mathcal{K}(\mathcal{H})^{*} , defined as \Gamma(x)=\tau_{x},\quad x\in\mathcal{L}^{1}(\mathcal{B}(\mathcal{H})),~\text{where}~\tau_{x}(a)=Tr(ax),\quad a\in\mathcal{K}(\mathcal{H}). Adapting such an approach, we formulate the distance between a point and a subspace in \mathcal{B}(\mathcal{H}) both under the usual operator norm and trace norm, in terms of trace. Theorem 1.1 . Let \mathcal{H} be a Hilbert space and \mathcal{V} be a finite-dimensional subspace of \mathcal{B}(\mathcal{H}) . Let x\in\mathcal{B}(\mathcal{H})\setminus\mathcal{V} . Then (1.3) \displaystyle dist(x,\mathcal{V}) \displaystyle=\sup\{|Tr(ax)|:~a\in N_{\mathcal{V}}\cap\mathcal{L}^{1}(\mathcal{B}(\mathcal{H})),~\|a\|_{1}=1\} \displaystyle\geq\sup\{|Tr((\xi\otimes\zeta)x)|:~(\xi\otimes\zeta)\in N_{\mathcal{V}},\|\xi\|=\|\zeta\|=1\}. Applying Theorem 1.1 we get a characterization of best approximation to x out of \mathcal{V} . Theorem 1.2 . Let \mathcal{H} be a Hilbert space and \mathcal{V} be a finite-dimensional subspace of \mathcal{B}(\mathcal{H}) . Let x\in\mathcal{B}(\mathcal{H})\setminus\mathcal{V} . Then z\in\mathcal{V} is a best approximation to x out of \mathcal{V} if and only if \Omega(x-z) is a Hahn-Banach extension of \Omega(x)\big|_{N_{\mathcal{V}}} . For trace class operators, we have the following formulation of distance. Theorem 1.3 . Let \mathcal{H} be a Hilbert space and \mathcal{V} be a finite-dimensional subspace of \mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) with dim~\mathcal{V}=n . Let x\in\mathcal{L}^{1}(\mathcal{B}(\mathcal{H}))\setminus\mathcal{V} . Then there exist scalars \lambda_{1},\dots,\lambda_{k}\in[0,1] (1\leq k\leq 2n+1) with \sum_{i=1}^{k}\lambda_{i}=1 such that (1.4) \displaystyle dist_{1}(x,\mathcal{V})= \displaystyle\sup\{|Tr(ax)|:~a\in\mathcal{K}(\mathcal{H})\cap N_{\mathcal{V}},~\|a\|=1\} \displaystyle= \displaystyle\max\{|Tr(ax)|:~a\in\mathcal{B}(\mathcal{H})\cap N_{\mathcal{V}},~\|a\|=1\} \displaystyle= \displaystyle\max\{\sum_{i=1}^{k}\lambda_{k}Tr(u_{k}x):~\sum_{i=1}^{k}Im~\lambda_{i}Tr(u_{i}x)=0,~\sum_{i=1}^{k}\lambda_{i}Tr(u_{i}w)=0,~v\in\mathcal{V} \displaystyle u_{i}^{*}u_{i}=Id_{\mathcal{H}},~\text{or}~u_{i}u_{i}^{*}=Id_{\mathcal{H}},~\text{for each}~i\in\{1,\dots,k\}\}. The C^{*} -homomorphism Id\otimes\mathbf{1}:\mathcal{B}(\mathcal{H})\to\mathcal{B}(\mathcal{H}\overline{\otimes}\ell_{2}(\mathbb{N})) , defined as id\otimes\mathbf{1}(x)=x\otimes\mathbf{1} isometrically embeds \mathcal{B}(\mathcal{H}) inside \mathcal{B}(\mathcal{H}\overline{\otimes}\ell_{2}(\mathbb{N})) . The pull-back of weak operator topology on \mathcal{B}(\mathcal{H}\overline{\otimes}\ell_{2}(\mathbb{N})) under the map Id\otimes\mathbf{1} is a locally convex topology, known as the \sigma -WOT topology on \mathcal{B}(\mathcal{H}) . In particular, the \sigma -WOT continuous functionals are described by the following lemma. Lemma 1.4 ( [ 13 ] , Chapter II) . Let \varphi:\mathcal{B}(\mathcal{H})\to\mathbb{C} be a linear functional. Then the following conditions are equivalent: (i) There exists a\in\mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) such that \varphi(x)=\varphi_{a}(x)=Tr(ax),\quad x\in\mathcal{B}(\mathcal{H}). (ii) \varphi is \sigma -WOT continuous. Thus, \sigma -WOT continuous functionals are nothing but the weak ∗ continuous functional on \mathcal{B}(\mathcal{H}) . Since \mathcal{B}(\mathcal{H}) is a von-Neumann algebra, its closed unit ball is \sigma -WOT compact. Therefore, for any x\in\mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) , the following collection M_{x}:=\{a\in\mathcal{B}(\mathcal{H}):~\|a\|=1,~\pi(x)(a)=\|x\|_{1}\} is always non-empty, where \pi:\mathcal{L}^{1}(\mathcal{B}(\mathcal{H}))\to\mathcal{L}^{1}(\mathcal{B}(\mathcal{H}))^{**} denotes the canonical embedding. This observation leads to the following characterization of best approximations in trace class norm. Theorem 1.5 . Let \mathcal{H} be a Hilbert space and \mathcal{V} be any finite-dimensional subspace of \mathcal{L}^{1}(\mathcal{B}(\mathcal{H})) . Let x\in\mathcal{L}^{1}(\mathcal{B}(\mathcal{H}))\setminus\mathcal{V} . Then y_{0}\in\mathcal{V} is a best approximation to x out of \mathcal{V} if and only if M_{x-y_{0}}\cap N_{\mathcal{V}}\neq\emptyset . The above result has an important consequence for smooth trace class operators. Finally, we consider the finite-dimensional C^{*} -algebra case to formulate the distance of a point from a subspace in the C^{*} -norm and in the trace norm. Theorem 1.6 . Let n_{1},n_{2},\dots n_{p} be natural numbers and \mathcal{M} be the finite-dimensional C^{*} -algebra M_{n_{1}}(\mathbb{C})\oplus_{\infty}M_{n_{2}}(\mathbb{C})\oplus_{\infty}\dots\oplus_{\infty}M_{n_{p}}(\mathbb{C}). Let \mathcal{V} be a subspace of \mathcal{M} with dim~\mathcal{V}=k . Let \mathbf{x}=(x_{1},x_{2},\dots,x_{p})\in\mathcal{M}\setminus\mathcal{V} . Then dist(\mathbf{x},\mathcal{V})=\max\left\{|\sum_{i=1}^{p}Tr(x_{i}a_{i})|:~\sum_{i=1}^{p}Tr(a_{i}y^{(j)}_{i})=0,~\sum_{i=1}^{p}\|a_{i}\|_{1}=1,~1\leq j\leq k\right\}, and dist_{1}(\mathbf{x},\mathcal{V})=\max\left\{|\sum_{i=1}^{p}Tr(x_{i}a_{i})|:~\sum_{i=1}^{p}Tr(a_{i}y^{(j)}_{i})=0,~\max\{\|a_{i}\|:~1\leq i\leq p\}=1,~1\leq j\leq k\right\}, for any basis \{\mathbf{y}_{j}\in\mathcal{M}:\mathbf{y}_{j}=(y^{(j)}_{1},y^{(j)}_{2},\dots,y^{(j)}_{p}),1\leq j\leq k\} of \mathcal{V} . The applications of the above results are illustrated through suitable examples. The following two results will be used in the sequel. For a normed linear space X , we denote the unit sphere of X by S_{X} and the continuous dual of X by X^{*} . Theorem 1.7 ( [ 12 ] Theorem 1.1, Page 170) . Let X be a complex normed linear space and G be an n -dimensional subspace of X . Let x\in X\setminus G and g_{0}\in G . Then the following conditions are equivalent: (i) g_{0} is a best approximation to x out of G . (ii) There exist k -number of extremal points f_{1},\dots,f_{k}\in S_{X^{*}} (1\leq k\leq 2n+1) and scalars \lambda_{1},\dots,\lambda_{k}\in[0,1] with \sum_{i=1}^{k}\lambda_{i}=1 such that \sum_{i=1}^{k}\lambda_{i}f_{i}(x-g_{0})=\|x-g_{0}\|\quad\text{and}\quad\sum_{i=1}^{k}\lambda_{i}f_{i}(g)=0,\quad g\in G. The above result is due to Singer and has been applied to study the best approximation problems in the space of compact operators on a Banach space in [ 7 ] . Lemma 1.8 ( [