Best approximants relative to a C ∗ -subalgebra, joint numerical range and subdifferentials
Tamara Bottazzi, Alejandro Varela · 2026
Read Best approximants relative to a C ∗ -subalgebra, joint numerical range and subdifferentials by Tamara Bottazzi, Alejandro Varela (2026) 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-2604-16041.txt.
Introduction
Bottazzi and Varela study when an n-by-n Hermitian matrix is minimal with respect to a C-star-subalgebra in the spectral norm. They broaden the moment of a subspace and relate it to the joint numerical range and subdifferentials of the maximum eigenvalue. They extend results known for diagonal operators and describe the subdifferential of the maximum eigenvalue via the moment of the matching eigenspace. They also describe minimality via moments and subdifferentials, with examples.
Catalog record
| CREATOR | Tamara Bottazzi, Alejandro Varela |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56394 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.16041
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Best approximants relative to a C ∗ -subalgebra, joint numerical range and subdifferentials Tamara Bottazzi, Alejandro Varela Abstract We study the minimality of n × n n\times n Hermitian matrices A A respect to a C ∗ C^{*} -subalgebra ℬ \mathcal{B} of M n ( ℂ ) M_{n}(\mathbb{C}) in the spectral norm, that is ‖ A ‖ ≤ ‖ A + B ‖ , for every B ∈ ℬ . \|A\|\leq\|A+B\|,\ \text{ for every }B\in\mathcal{B}. We generalize the notion of the moment of a subspace and relate it to the joint numerical range and the subdifferentials of the maximum eigenvalue. We extend results previously known for the subalgebra of diagonal operators and describe the subdifferential of the maximum eigenvalue in terms of the moment of the corresponding eigenspace. We also characterize ℬ \mathcal{B} -minimality via moments and subdifferentials, and provide examples. 2020 MSC: (PRIMARY) 15A60, 41A50, 47B15, (Secondary) 47A12, 47A30 Keywords: Minimal operators, Subdifferential of eigenvalues, Moment of a subspace, best approximation of matrices 1 Introduction and prelimininaries Let M_{n}(\mathbb{C}) be the space of complex square matrices of n\times n , which is a C^{*} -algebra. Consider a C^{*} -subalgebra \mathcal{B} of M_{n}(\mathbb{C}) . We call A\in M_{n}(\mathbb{C}) a \mathcal{B} -minimal matrix if \|A\|\leq\|A+B\|,\text{ for all }B\in\mathcal{B}. \mathcal{B} -minimality is closely related with best approximation, since A is \mathcal{B} -minimal if and only if dist(A,\mathcal{B})=\|A\|. If A is Hermitian, then dist(A,\mathcal{B})=dist(A,\mathcal{B}^{h}), where the superscript denotes the subset of Hermitian (self-adjoint) elements. This holds by Remark 2.1 [ 2 ] , \|A+\text{Re}(B)\|=\|\text{Re}(A+B)\|\leq\|A+B\| for all B\in\mathcal{B} . So, we will consider only the Hermitian elements of \mathcal{B} . Our interest in this subject is related to the following context. Let \mathcal{A} be a C ∗ -algebra and \mathcal{B} a C ∗ -subalgebra sharing the same unit. The description of minimal self-adjoint elements B\in\mathcal{B} with respect to a self-adjoint element A\in\mathcal{A} ( \mathcal{B} -minimality) allows a concrete description of geodesics in a natural Finsler metric induced by the norm of \mathcal{A} on the flag manifolds U_{\mathcal{A}}/U_{\mathcal{B}} (a homogeneous space under the left action of U_{\mathcal{A}} ) associated with the quotient space \mathcal{A}/\mathcal{B} (see [ 9 ] ). In this setting, if Z is a Hermitian matrix that is minimal in the spectral norm, then the curve \gamma(t)=e^{itZ}Ae^{-itZ} is length-minimizing in U_{\mathcal{A}}/U_{\mathcal{B}} between \gamma(0) and \gamma(t) for |t|\leq\frac{\pi}{2\|Z\|} (see [ 9 ] ). \mathcal{B} -Minimality is closely related to Birkhoff-James orthogonality in the normed space \left(M_{n}(\mathbb{C}),\|\cdot\|\right) [ 4 , 11 ] : two matrices A,B\in M_{n}(\mathbb{C}) are Birkhoff-James orthogonal if and only if \|A\|\leq\|A+\lambda B\| , for every \lambda\in\mathbb{C} . In particular, A\in M_{n}(\mathbb{C}) is a \mathcal{B} -minimal matrix if and only if A is Birkhoff-James orthogonal to every B\in\mathcal{B} . This is equivalent to state that A is Birkhoff-James orthogonal to \mathcal{B} [ 10 , 6 ] . The case where \mathcal{B} is the C^{*} -subalgebra of diagonal matrices (respect to a fixed basis of \mathbb{C}^{n} ) has been extensively studied in [ 1 , 5 , 12 , 7 , 8 ] . In the present work, we are focused to study and generalize some of the results that involve relations between \mathcal{B} -minimality, joint numerical range (usually called algebraic joint numerical range) and subdifferentials, as we did for diagonal matrices in [ 13 , 7 , 8 ] . In this sense, in [ 8 ] we relate minimal operators to the subdifferentials of the maximum eigenvalue and the norm. We connect these notions with the moments of the eigenspaces corresponding to the largest and smallest eigenvalues, as well as with certain joint numerical ranges; these ideas were developed in [ 13 ] for matrices, and [ 7 ] for compact operators on a Hilbert space. In particular, one of our main results provides an explicit characterization of the subdifferential of the largest eigenvalue of a Hermitian matrix A(x) with variable real diagonal x , \partial\big(\lambda_{max}(A(x))\big)=\Diag(\partial\lambda_{max}(A(x)))=m_{S_{max}}, in terms of m_{S_{max}}=\text{conv}\{(|v_{1}|^{2},...,|v_{n}|^{2}):v=(v_{1},...,v_{n})\in S_{\lambda_{max}},\|v\|=1\} , the moment of the eigenspace related to the largest eigenvalue \lambda_{max}(A(x)) . Now, considering any other C^{*} -subalgebra \mathcal{B} of M_{n}(\mathbb{C}) , we need to generalize the definition of moment of a subspace in order to study its relation with the joint numerical range of a certain family and the subdifferentials of the maximum eigenvalue of a variable matrix A(x)=A_{0}+\sum_{i=1}^{\dim(\mathcal{B})}x_{i}B_{i} with A_{0}\in M_{n}^{h}(\mathbb{C}) , x_{i}\in\mathbb{R} and \{B_{i}\}_{i=1}^{\dim(\mathcal{B})} any fixed basis of \mathcal{B} . The results presented in this paper are divided into three parts. In Section 2 , we establish the appropriate definitions for the generalization we wish to develop and obtain several results that were previously known only for the algebra of diagonal operators. These include the moment set of an eigenspace, the joint numerical range of certain matrices defined with respect to a chosen basis \mathbb{B} of \mathcal{B} , the support of a pair of subspaces and their relation with \mathcal{B} -minimality. In Section 3 we study the subdifferential of the maximum eigenvalue of matrices and we link it with the moment of its eigenspace. Moreover, we state an equivalence between \mathcal{B} -minimality and the condition 0\in\partial\lambda_{min}+\partial\lambda_{max} . Finally, Section 4 is dedicated to show different examples, including one related to quantum information theory. 2 General framework of \mathcal{B} -minimality Let \mathcal{B} be a C ∗ -subalgebra of M_{n}(\mathbb{C}) , and the subspace \mathcal{B}_{\text{tr}}^{\perp}=\{X\in M_{n}(\mathbb{C}):\text{tr}(XB)=0,\forall B\in\mathcal{B}\} with \dim(\mathcal{B}_{\text{tr}})^{\perp}=n^{2}-\dim(\mathcal{B}) . Recall the following result from [ 18 , Proposition 2.3, Lemma 2.5 and Theorem 2.6] . Theorem 1 . Let \mathcal{B} be a C ∗ -subalgebra of M_{n}(\mathbb{C}) . Then \begin{split}A\in M_{n}^{h}(\mathbb{C})&\text{ is }\mathcal{B}\text{-minimal }\Leftrightarrow\\ &\Leftrightarrow\text{ there exists a Hermitian }X\in\mathcal{B}_{\text{tr}}^{\perp}\setminus\{0\}\\ &\qquad\text{ such that }AX=\|A\|\ |X|.\end{split} (2.1) Moreover, if A\in M_{n}^{h}(\mathbb{C}) is \mathcal{B} -minimal, and \mathcal{B} shares the identity I=I_{n} with M_{n}^{h}(\mathbb{C}) , then \pm\|A\|\in\sigma(A) . With the hypothesis and notations of Theorem 1 plus the condition that \mathcal{B} has the same identity that M_{n}(\mathbb{C}) , we can consider the spectral projections E_{+} , E_{-} , I-E_{+}-E_{-} and the orthogonal decomposition of \mathbb{C}^{n}=\text{ran}(E_{+})\oplus\text{ran}(E_{-})\oplus\big(\text{ran}(E_{+})\oplus\text{ran}(E_{-})\big)^{\perp} in terms of the eigenspaces of A corresponding to the eigenvalues of -\|A\| and \|A\| . Hence, supposing that \|A\|=1 , we can write A=\begin{pmatrix}I&0&0\\ 0&-I&0\\ 0&0&A_{3,3}\end{pmatrix}\ \text{ and }\ X=\begin{pmatrix}X_{1,1}&0&0\\ 0&X_{2,2}&0\\ 0&0&0\end{pmatrix}. (2.2) with X_{1,1}>0 and X_{2,2}<0 (see Case 1 in Theorem 2.7 of [ 18 ] and note that X_{1,1} and X_{2,2} can be diagonal if certain basis is chosen). Here X\geq 0 denotes a positive semidefinite matrix, that is, \langle Xu,u\rangle\geq 0 for all u\in\mathbb{C}^{n} . We write X>0 (respectively X<0 ) to indicate that X is positive (respectively negative) definite. Let us denote as \mathcal{D} the set of density matrices, that is \mathcal{D}=\{X\in M_{n}^{h}(\mathbb{C}):\ X\geq 0,\ \text{tr}(X)=1\}. Now suppose that \{B_{1},B_{2},\dots,B_{r}\} is a fixed orthonormal basis of \mathcal{B} . Then, for any Y\in M_{n}(\mathbb{C}) , it is clear that \begin{split}\operatorname{tr}(YB_{k})&=0\ \text{for all }k=1,\dots,r\ \Leftrightarrow\\ &\Leftrightarrow\operatorname{tr}(YB)=0\ \text{for all }B\in\mathcal{B}\ \Leftrightarrow\ Y\in\mathcal{B}^{\perp}_{\operatorname{tr}}.\end{split} Remark 1 . If I\in\mathcal{B} and \dim\mathcal{B}=t we can consider an orthonormal basis of \mathcal{B} as \mathbb{B}=\{\frac{1}{\sqrt{n}}I,B_{2},\dots,B_{t}\} with B_{j} Hermitian. Observe that then it must hold that \text{tr}(B_{j})=0 for j=2,\dots,t . \mathbb{B} is also a orthonormal basis of \mathcal{B}^{h} as a real subspace. See for example the Gell-Mann generalized basis used in [ 7 , proof of Proposition 9] . In particular, taking the X of Theorem 1 (see