Unitary normalizers in finite-dimensional inclusions Keshab Chandra Bakshi and Silambarasan C Abstract We study regular inclusions of finite-dimensional von Neumann algebras from a matrix-theoretic perspective. To this end, we introduce a new combinatorial invariant of an inclusion, called the normalizer matrix , which encodes the structure of the normalizer purely at the level of the inclusion matrix. Using this invariant, we obtain a complete characterization of regular inclusions of finite-dimensional von Neumann algebras. As consequences, we show that every regular inclusion decomposes into finite direct sums and tensor products of basic building blocks, and that regular inclusions are necessarily of depth two. We further investigate the existence of unitary orthonormal bases in the sense of Pimsner–Popa and prove that, under a natural spectral condition, regularity is equivalent to the existence of a unitary orthonormal basis contained in the normalizer. These results provide a unified description of regularity, unitary bases, and depth through the normalizer matrix formalism. 1 Introduction Normalizers occupy a central position in the structure theory of operator algebras, reflecting internal symmetries of inclusions and governing their associated crossed-product and groupoid constructions. Their importance is already evident in Dixmier’s seminal classification of maximal abelian subalgebras (MASAs) into singular, semi-regular, and regular (Cartan) types, where regularity is defined by generation of the ambient algebra by the normalizer. This perspective was further developed by Feldman and Moore [ 10 ] , who showed that Cartan subalgebras give rise to measured equivalence relations, and later, Renault ( [ 16 ] ) discovered a C^{*} -algebraic analogue of this result in the language of groupoids. In subfactor theory, normalizers play an important role (see, for instance, [ 12 , 6 , 2 , 3 , 15 , 7 ] ). Regular subfactors form a distinguished class encompassing crossed-product inclusions by finite groups and are closely related to depth-two phenomena and quantum symmetry. In particular, finite-index regular subfactors are known to admit unitary Pimsner–Popa bases and to arise from actions of weak Hopf algebras or quantum groupoids (see [ 3 , 9 ] ). Related classification results have also been obtained in [ 4 ] for finite-index regular inclusions of simple unital C^{*} -algebras, where regularity forces depth two and a crossed-product-type structure. Motivated by these developments, the present article undertakes a systematic study of regular inclusions of finite-dimensional von Neumann algebras. Despite the simplicity of the finite-dimensional setting, a complete and purely structural characterization of regularity has so far been missing. Our goal is to provide such a classification and to relate regularity to combinatorial invariants of the inclusion and to the existence of unitary orthonormal bases. Our approach is matrix-theoretic in nature. Recall that any inclusion \mathcal{B}\subset\mathcal{A} of finite-dimensional von Neumann algebras is completely determined, up to isomorphism, by its inclusion matrix together with the dimension vectors of the simple summands. We introduce a new combinatorial invariant of an inclusion, called the normalizer matrix (see 3.2 ), which captures the structure of the normalizer purely at the level of the inclusion matrix. This invariant isolates the precise combinatorial obstructions to regularity and allows us to characterize regular inclusions in terms of elementary matrix data. Our first main result provides a complete characterization of regularity in finite dimension. Theorem A: (See Theorem 3.17 ) The inclusion of finite dimensional von Neumann algebras, \mathcal{B}\subset\mathcal{A} is regular if and only if 1. The inclusion matrix is an normalizer matrix , and 2. For each 0\leq i\leq s-1 , if j is in the i -th row support of the inclusion matrix, then the corresponding summands of \mathcal{B} have equal dimension. As a consequence, any regular inclusion is isomorphic to tensor product and direct sum of inclusions M_{k}(\mathbb{C})\subset M_{k}(\mathbb{C}), \mathbb{C}\subset\oplus_{i=0}^{v-1}M_{l_{i}}(\mathbb{C}) and \mathbb{C}^{t}\subset M_{t}(\mathbb{C}) . A second motivation for this work comes from the theory of unitary orthonormal (Pimsner–Popa) bases. Such bases play a central role in subfactor theory [ 14 ] and quantum information theory, where they are closely related to unitary error bases and to protocols such as teleportation and dense coding. While the existence of unitary orthonormal bases is automatic for several basic inclusions, a complete characterization for general finite-dimensional inclusions has remained open. In a seminal work, Werner [ 18 ] established a one-to-one correspondence between tight teleportation schemes for tripartite systems M_{n}(\mathbb{C})\otimes M_{n}(\mathbb{C})\otimes M_{n}(\mathbb{C}) and orthonormal bases of unitaries in M_{n}(\mathbb{C}) . This correspondence was subsequently extended by Conlon et al. [ 8 ] to inclusions of the form N\subset M_{n}(\mathbb{C}) , demonstrating how unitary orthonormal bases on general multi-matrix algebras can be used to study quantum teleportation schemes and quantum graphs. In these settings, tight teleportation schemes are shown to arise from orthonormal unitary Pimsner–Popa bases, generalizing Werner’s original framework. In all such applications, it is essential that the unitary orthonormal basis be contained in the normalizer. In this direction, we precisely characterize when a regular inclusion admits a unitary orthonormal basis contained in the normalizer. Unlike in the subfactor setting, regularity by itself does not ensure the existence of a unitary orthonormal basis in the normalizer; an additional spectral condition on the inclusion matrix, introduced in [ 1 ] , is required. Our second main result establishes a sharp equivalence. Theorem B: (See Theorem 4.2 , Corollary 4.3 and 4.4 ) Let (\mathcal{B}\subset\mathcal{A},E) be an inclusion of finite-dimensional von Neumann algebras, where E denotes the unique conditional expectation preserving the trace given in equation 4.8 . Then the following statements are equivalent: (i) There exists a unitary orthonormal basis of \mathcal{A} over \mathcal{B} contained in the normalizer \mathcal{N}_{\mathcal{A}}(\mathcal{B}) ; (ii) The inclusion \mathcal{B}\subset\mathcal{A} is regular and the inclusion matrix satisfies the spectral condition. Moreover, if either \mathcal{B} or \mathcal{A} is a matrix algebra, then the inclusion is regular if and only if it admits a unitary orthonormal basis contained in \mathcal{N}_{\mathcal{A}}(\mathcal{B}) . Finally, we show that every regular inclusion has depth two, linking our results to (weak) Hopf-algebraic structures. More precisely, if the inclusion \mathcal{B}\subset\mathcal{A} is regular, then the corresponding inclusion matrix has depth 2 in the sense of [ 5 ] .To see the connection between depth 2 in subfactor theory and regularity, we refer to [ 13 ] , [ 3 ] . Thus, our results may be viewed as a finite-dimensional analogue of the structure theory of regular inclusions developed in the subfactor and C^{*} -algebraic settings. Organization of the paper. In Section 2 we recall preliminaries on finite-dimensional inclusions and unitary orthonormal bases. Section 3 introduces normalizer matrices and establishes the classification of regular inclusions. In Section 4 we study unitary orthonormal bases and show that regular inclusions necessarily have depth two. 2 Preliminaries Throughout this article, we work with a triple (\mathcal{B}\subseteq\mathcal{A},E) where \mathcal{A},\mathcal{B} are finite dimensional von Neumann algebras with \mathcal{B} being a unital subalgebra of \mathcal{A} and E:\mathcal{A}\to\mathcal{B} is a conditional expectation map. Such a triple will be referred to, following [ 1 ] , as a subalgebra system. Definition 2.1 . Let (\mathcal{B}\subseteq\mathcal{A},E) be a subalgebra system. A family \{W_{j}:0\leq j\leq(d-1)\} (for some d\in\mathbb{N} ) of elements of \mathcal{A} is called a (right) Pimsner-Popa basis for (\mathcal{B}\subseteq\mathcal{A},E) if X=\sum_{j=0}^{d-1}W_{j}E(W_{j}^{*}X),~~\forall~X\in\mathcal{A}. It is said to be an unitary orthonormal basis if W_{j} is a unitary for every j , and E(W_{j}^{*}W_{k})=\delta_{jk},~~\forall~0\leq j,k\leq(d-1). Throughout, we will take \mathcal{A}=M_{n_{0}}\oplus M_{n_{1}}\oplus\cdots\oplus M_{n_{s-1}}; (2.1) \mathcal{B}=M_{m_{0}}\oplus M_{m_{1}}\oplus\cdots\oplus M_{m_{r-1}}. (2.2) The associated dimension vectors of \mathcal{A} and \mathcal{B} are n^{\prime}=\begin{bmatrix}n_{0}&n_{1}&\cdots&n_{s-1}\end{bmatrix}^{t} and m^{\prime}=\begin{bmatrix}m_{0}&m_{1}&\cdots&m_{r-1}\end{bmatrix}^{t} respectively. Notation: The block diagonal form \begin{bmatrix}A_{0}&\\ &\ddots\\ &&A_{n}\end{bmatrix} will be denoted by \text{bl-diag}(A_{1},\cdots,A_{n}) . Definition 2.2 . The inclusion matrix of \mathcal{B} in \mathcal{A} is given by an s\times r matrix: A=[a_{ij}]_{0\leq i\leq(s-1);0\leq j\leq(r-1).} where the algebra M_{m_{j}} of \mathcal{B} appears a_{ij} times in the algebra M_{n_{i}} of \mathcal{A}. Any inclusion of finite dimensional von Neumann algebras \mathcal{B}\subseteq\mathcal{A} is completely determined up to isomorphism by this triple (A,m^{\prime},n^{\prime}) of one inclusion matrix and two dimension vectors. Note that by dimension counting, \sum_{j=0}^{r-1}a_{ij}m_{j}=n_{i},~~\forall~0\leq i\leq(s-1) (2.3) or in matrix notation: n^{\prime}=Am^{\prime} . So, throughout this article unless otherwise stated, we consider inclusions of the following form \displaystyle i:\oplus_{j=0}^{r-1}M_{m_{j}} \displaystyle\rightarrow\oplus_{i=0}^{s-1}M_{n_{i}} \displaystyle\oplus_{j=0}^{r-1}X_{j} \displaystyle\longmapsto\oplus_{i=0}^{s-1}\text{bl-diag}(X_{0}\otimes\mathbf{1}_{a_{i0}},X_{1}\otimes\mathbf{1}_{a_{i1}},\cdots,X_{r-1}\otimes\mathbf{1}_{a_{ir-1}}) (2.4) corresponding to the inclusion matrix A=[a_{ij}]_{0\leq i\leq(s-1);0\leq j\leq(r-1)} . Let Q_{j} be the projection onto the j- th summand in the decomposition \mathcal{B}=\oplus_{j=0}^{r-1}M_{m_{j}} and similarly let P_{i} be the projection onto the i- th summand in the decomposition \mathcal{A}=\oplus_{i=0}^{s-1}M_{m_{i}} for 0\leq j\leq(r-1);0\leq i\leq(s-1). Note that Q_{j},P_{i} are minimal central projections of \mathcal{B}~\text{and}~\mathcal{A}, respectively. Definition 2.3 . Consider the inclusions \mathcal{B}_{1}\subseteq\mathcal{A}_{1} and \mathcal{B}_{2}\subseteq\mathcal{A}_{2} with inclusion matrix A_{1} and A_{2} respectively. Then the inclusion matrix of the direct sum \mathcal{B}_{1}\oplus\mathcal{B}_{2}\subseteq\mathcal{A}_{1}\oplus\mathcal{A}_{2} is given by bl-diag (A_{1},A_{2}) and the inclusion matrix of the tensor product \mathcal{B}_{1}\otimes\mathcal{B}_{2}\subseteq\mathcal{A}_{1}\otimes\mathcal{A}_{2} is given by A_{1}\otimes A_{2} . Definition 2.4 . A subalgebra system (\mathcal{B}\subseteq\mathcal{A},E) is said to have the U-property if it admits a unitary orthonormal basis. Theorem 2.5 . [ 1 ] Let (\mathcal{B}\subseteq\mathcal{A},E) be a subalgebra system, with Am^{\prime}=n^{\prime} , where A is the inclusion matrix, and m^{\prime},n^{\prime} are dimension vectors. Suppose (\mathcal{B}\subseteq\mathcal{A},E) admits a unitary orthonormal basis with d -unitary elements. Then A^{t}n^{\prime}=dm^{\prime} . Definition 2.6 . The normalizer of the inclusion \mathcal{B}\subseteq\mathcal{A} is defined as \mathcal{N}_{\mathcal{A}}(\mathcal{B}):=\{U\in\mathcal{U}(\mathcal{A}):U\mathcal{B}U^{*}=\mathcal{B}\}. where \mathcal{U(A)} denotes the group of unitaries of \mathcal{A} .Equivalently, the normalizer can also be defined as \mathcal{N}_{\mathcal{A}}(\mathcal{B}):=\{U\in\mathcal{U}(\mathcal{A}):Ad_{U}\in Aut(\mathcal{B})\}.\\ where Ad_{U}(x)=UxU^{*}. The inclusion \mathcal{B}\subseteq\mathcal{A} is called regular if \mathcal{A}=\text{span}\;\mathcal{N}_{\mathcal{A}}(\mathcal{B}) . The following are the three important examples of regular inclusions that admit a unitary orthonormal basis in the normalizer: 1. Trivial Inclusion: (M_{n}(\mathbb{C})\subset M_{n}(\mathbb{C}),E_{0}\;) where E_{0} is the unique conditional expectation preserving the usual trace on M_{n}(\mathbb{C}) In this case, E_{0}=Id , \mathcal{N_{A}(B)}=\mathcal{U(A)} and the identity operator forms the basis. 2. Scalar Inclusion: (\mathbb{C}\subset\oplus_{i=0}^{s-1}M_{n_{i}}(\mathbb{C}),E_{1}) where E_{1} is the unique conditional expectation preserving the state \varphi(\oplus X_{i})=\frac{1}{\sum_{i=0}^{s-1}n_{i}^{2}}\sum_{i=0}^{s-1}n_{i}~\mbox{trace}~(X_{i}),~~X_{i}\in M_{n_{i}},0\leq i\leq(s-1). (2.5) In this inclusion, one can see that \mathcal{N_{A}(B)}=\mathcal{U(A)} and its admits unitary orthonormal basis as proved in [ 1 ] Theorem 5.3. and [ 9 ] Theorem 2.1. 3. Diagonal Inclusion: (\mathbb{C}^{n}\subset M_{n}(\mathbb{C}),E_{2}) where E_{2} is the unique conditional expectation preserving the normalized trace on M_{n}(\mathbb{C}) . Let \{e_{0},e_{2},\cdots,e_{n-1}\} be the standard basis of \mathbb{C}^{n} and let U be the translation unitary defined by Ue_{k}=e_{k+1(mod\;n)} . Then \{U^{k}:k\in\mathbb{Z}_{n}\} is contained in the normalizer of \mathbb{C}^{n} and form a unitary orthonormal basis. For further details, see [ 8 ] Example 2.2. Tensor products and direct sums: Suppose (\mathcal{B}_{i}\subseteq\mathcal{A}_{i},E_{i}) for i=1,2 are two finite dimensional inclusions with U -property. Suppose \{U_{1}(i),\ldots,U_{d_{i}}(i)\} is a unitary o.n.b. in \mathcal{N}_{\mathcal{A}_{i}}(\mathcal{B}_{i}) for (\mathcal{B}_{i},\mathcal{A}_{i},E_{i}) for i=1,2. Take \mathcal{B}=\mathcal{B}_{1}\otimes\mathcal{B}_{2} . Consider it as a subalgebra of \mathcal{A}=\mathcal{A}_{1}\otimes\mathcal{A}_{2} with the inclusion matrix being the tensor product of inclusion matrices. Then it is seen easily that E=E_{1}\otimes E_{2} is a conditional expectation map from \mathcal{A} to \mathcal{B}. For this inclusion, (\mathcal{B}_{1}\otimes\mathcal{B}_{2}\subseteq\mathcal{A}_{1}\otimes\mathcal{A}_{2},E_{1}\otimes E_{2}) , we can observe that \{U_{j}(1)\otimes U_{k}(2):0\leq j\leq d_{1};0\leq k\leq d_{2}\} is a unitary orthonormal basis in \mathcal{N}_{\mathcal{A}_{1}\otimes\mathcal{A}_{2}}(\mathcal{B}_{1}\otimes\mathcal{B}_{2}) as \mathcal{N}_{\mathcal{A}_{1}}(\mathcal{B}_{1})\otimes\mathcal{N}_{\mathcal{A}_{2}}(\mathcal{B}_{2})\subseteq\mathcal{N}_{\mathcal{A}_{1}\otimes\mathcal{A}_{2}}(\mathcal{B}_{1}\otimes\mathcal{B}_{2}) . Under the same setup, if d_{1}=d_{2} , the subalgebra system (\mathcal{B}_{1}\oplus\mathcal{B}_{2}\subseteq\mathcal{A}_{1}\oplus\mathcal{A}_{2},E_{1}\oplus E_{2}) has U -property. In fact, \{U_{j}(1)\oplus U_{j}(2):0\leq j\leq d_{1}\} is a unitary orthonormal basis in \mathcal{N}_{\mathcal{A}_{1}\oplus\mathcal{A}_{2}}(\mathcal{B}_{1}\oplus\mathcal{B}_{2}) as \mathcal{N}_{\mathcal{A}_{1}}(\mathcal{B}_{1})\oplus\mathcal{N}_{\mathcal{A}_{2}}(\mathcal{B}_{2})\subseteq\mathcal{N}_{\mathcal{A}_{1}\oplus\mathcal{A}_{2}}(\mathcal{B}_{1}\oplus\mathcal{B}_{2}) . 3 Regular inclusions in finite dimension In this section, we develop a comprehensive framework for understanding regular inclusions of finite-dimensional von Neumann algebras. In Subsection 3.1, we introduce the notion of a normalizer matrix, a combinatorial invariant that encodes the structure of the normalizer entirely at the level of the inclusion matrix, and show that every normalizer matrix can be brought to a block-diagonal form reflecting the natural partitioning of row supports. Subsection 3.2 focuses on regular inclusions within matrix algebras, where we provide a detailed description of normalizers and establish that regularity imposes strict homogeneity conditions on the summands and equality of inclusion matrix entries. Finally, in Subsection 3.3, we extend these results to the general setting, proving that an inclusion is regular if and only if its inclusion matrix is a normalizer matrix and the dimensions of summands corresponding to each row support are equal. As a consequence, every regular inclusion decomposes into direct sums and tensor products of the fundamental building blocks identified in Section 2. This exposition lays the foundation for the subsequent analysis of unitary orthonormal bases. 3.1 Normalizer matrices To analyze the internal structure of regular finite-dimensional inclusions, we introduce the notion of a normalizer matrix. This matrix encodes the combinatorial structure of the normalizer. Our main goal is to show that every normalizer matrix can be permuted into a block-diagonal form, where each diagonal block has rows with equal non-zero entries and all off-diagonal blocks are zero. This reflects the natural partition of row indices according to identical support patterns. The block-diagonal form provides a convenient framework for understanding the combinatorial structure of normalizers . As we will see, this concept will be used crucially in later sections to establish structural results, construct unitary orthonormal bases, and describe decompositions into fundamental building blocks. Let A=[a_{ij}]_{0\leq i\leq(s-1),\,0\leq j\leq(r-1)} be an s\times r matrix. For each i\in\{0,1,\ldots,(s-1)\} , we define the support of the i -th row of A by Y_{i}:=\{\,j\in\{0,1,\ldots,(r-1)\}\mid a_{ij}\neq 0\,\}. We equip Y_{i} with the natural order inherited from \{0,1,\ldots,r-1\} ; that is, if Y_{i}=\{x_{1},x_{2},\ldots,x_{n}\}, then x_{1}