Matrix Algebras of Endomorphisms

• Matrix algebras of endomorphisms are algebraic structures that represent endomorphism rings via full, block, or constrained matrices, encoding preserved subspace data.
• They bridge settings from finite-dimensional vector spaces to operator algebras by linking coordinate representations with invariant substructures such as flags and gradings.
• These algebras facilitate the study of rigidity, automorphism, and decomposition properties, providing practical insights into modular representation theory and computational frameworks.

Searching arXiv for recent and foundational papers relevant to matrix algebras of endomorphisms. Matrix algebras of endomorphisms arise whenever endomorphism rings are expressed in coordinates, but the phrase covers several genuinely different phenomena. In the classical finite-dimensional setting, $\operatorname{End}_K(V)$ is the full matrix algebra $M_n(K)$. In more structured settings, endomorphisms form block-triangular or other constrained matrix subalgebras determined by invariant subspaces, gradings, or multiplicity data. In modular representation theory and related nonsemisimple contexts, natural endomorphism algebras may fail to be matrix algebras altogether and instead become commutative or nilpotent quotient algebras. Across these settings, the central problem is to identify which ambient structure on the underlying object is encoded by the resulting endomorphism algebra and, conversely, how matrix form controls endomorphism-theoretic invariants (Gupta et al., 2024, Besleaga et al., 2018, Kochhar, 2014).

Setting	Endomorphism algebra form	Matrix status
Finite-dimensional $K$-vector space	$\operatorname{End}_K(V)\cong M_n(K)$	Full matrix algebra
Generalized flag / preorder	$M(p,k)\cong \operatorname{End}(F)$	Structural matrix algebra
$\mathbb Z_p\times \mathbb Z_{p^m}$	$E_{p,p^m}$	Matrix-like, not full
Linear block code $\mathcal C(n,k)$	$\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$	Conjugate subalgebra of $M_n(\mathbb F_q)$
Young module $M_n(K)$0 in characteristic $M_n(K)$1	Square-zero commutative quotient	Generally not a matrix algebra
$M_n(K)$2 or $M_n(K)$3	Block endomorphisms from partial isometries	Matrix-block operator algebra

1. Full matrix algebras as the basic endomorphism model

For an $M_n(K)$4-dimensional $M_n(K)$5-vector space $M_n(K)$6, choosing a basis identifies $M_n(K)$7 with the full matrix algebra $M_n(K)$8. A particularly explicit realization uses a Galois extension $M_n(K)$9 of degree $K$0: if $K$1 as $K$2-vector spaces, then

$K$3

In this model, the trace pairing

$K$4

identifies $K$5 with $K$6, and the tensor isomorphism

$K$7

gives a canonical description of rank-one endomorphisms as maps

$K$8

The operator trace is then computed by the field trace through

$K$9

(Gupta et al., 2024).

This realization does not change the underlying algebraic object: it remains the full matrix algebra. What changes is the coordinate system. The Galois model replaces arbitrary matrix coordinates by formulas expressed through field trace, dual bases, and Galois conjugates. Basis criteria, rank-one operators, and cyclic linear-independence tests are therefore encoded by canonical determinant expressions rather than by an arbitrary choice of matrix entries (Gupta et al., 2024).

A related rigidity phenomenon appears for linear maps from $\operatorname{End}_K(V)\cong M_n(K)$0 into $\operatorname{End}_K(V)\cong M_n(K)$1. When such a linear map satisfies sufficiently strong root-of-unity or characteristic-polynomial conditions, the images of the primitive idempotents are forced to become pairwise orthogonal idempotents summing to the identity, so the map factors through an algebra homomorphism. In matrix terms, this means the image is conjugate to a block-scalar diagonal algebra inside $\operatorname{End}_K(V)\cong M_n(K)$2, again recovering the classical endomorphism picture from internal relations among matrices (Kulkarni et al., 2015).

2. Structural matrix algebras and generalized flags

A structural matrix algebra is obtained from a preorder $\operatorname{End}_K(V)\cong M_n(K)$3 on $\operatorname{End}_K(V)\cong M_n(K)$4 by imposing the zero-pattern condition

$\operatorname{End}_K(V)\cong M_n(K)$5

The key theorem is that $\operatorname{End}_K(V)\cong M_n(K)$6 is itself an endomorphism algebra, not of a plain vector space, but of a generalized flag determined by the preorder. Writing $\operatorname{End}_K(V)\cong M_n(K)$7 when $\operatorname{End}_K(V)\cong M_n(K)$8 and $\operatorname{End}_K(V)\cong M_n(K)$9, one obtains a poset $M(p,k)\cong \operatorname{End}(F)$0 of equivalence classes. A $M(p,k)\cong \operatorname{End}(F)$1-flag is then an $M(p,k)\cong \operatorname{End}(F)$2-dimensional vector space $M(p,k)\cong \operatorname{End}(F)$3 together with subspaces $M(p,k)\cong \operatorname{End}(F)$4 arising from a basis partitioned by these classes so that

$M(p,k)\cong \operatorname{End}(F)$5

is a basis of $M(p,k)\cong \operatorname{End}(F)$6. The preserving endomorphisms

$M(p,k)\cong \operatorname{End}(F)$7

satisfy

$M(p,k)\cong \operatorname{End}(F)$8

(Besleaga et al., 2018).

This identification generalizes the familiar equality $M(p,k)\cong \operatorname{End}(F)$9. The full matrix algebra corresponds to the trivial one-step flag, whereas upper triangular and upper block triangular algebras correspond to ordinary flags and block flags. The relation $\mathbb Z_p\times \mathbb Z_{p^m}$0 records exactly when a basis vector in class $\mathbb Z_p\times \mathbb Z_{p^m}$1 may map to one in class $\mathbb Z_p\times \mathbb Z_{p^m}$2 without violating preservation of the subspaces $\mathbb Z_p\times \mathbb Z_{p^m}$3 (Besleaga et al., 2018).

The flag viewpoint also reorganizes the internal structure of $\mathbb Z_p\times \mathbb Z_{p^m}$4. Its invariant subspaces become the subspaces $\mathbb Z_p\times \mathbb Z_{p^m}$5 indexed by subsets, equivalently antichains, of $\mathbb Z_p\times \mathbb Z_{p^m}$6. Automorphisms of the algebra are described by a combination of inner automorphisms, automorphisms of the poset $\mathbb Z_p\times \mathbb Z_{p^m}$7 preserving block sizes, and transitive scalar rescalings of matrix units. If the $\mathbb Z_p\times \mathbb Z_{p^m}$8-flag is equipped with a group grading, then

$\mathbb Z_p\times \mathbb Z_{p^m}$9

becomes a graded algebra, and the matrix units satisfy

$E_{p,p^m}$0

Thus good gradings on structural matrix algebras are induced by graded flags, and under graph-theoretic conditions on the Hasse diagram of $E_{p,p^m}$1, every good grading arises in this way (Besleaga et al., 2018).

3. Matrix-like endomorphism rings beyond the full matrix case

Many endomorphism rings are matrix-like without being full matrix algebras over a single coefficient ring. A basic example is the finite abelian $E_{p,p^m}$2-group

$E_{p,p^m}$3

Its endomorphism ring is isomorphic to

$E_{p,p^m}$4

with entrywise addition and a multiplication adapted to the mixed moduli in the four corners. This ring is not $E_{p,p^m}$5, not $E_{p,p^m}$6, and, by the stated theorem, cannot be embedded into matrices over any commutative ring. Its lower-left entry is forced into the ideal $E_{p,p^m}$7, reflecting the fact that $E_{p,p^m}$8 consists exactly of elements killed by $E_{p,p^m}$9 (Liu et al., 2016).

The matrix model makes arithmetic explicit. Invertibility is characterized by the conditions

$\mathcal C(n,k)$0

where $\mathcal C(n,k)$1 is the lowest $\mathcal C(n,k)$2-adic digit of the lower-right entry $\mathcal C(n,k)$3. Every element satisfies a quadratic relation

$\mathcal C(n,k)$4

with

$\mathcal C(n,k)$5

so the algebra has a Cayley–Hamilton-type structure despite not being a full matrix ring (Liu et al., 2016).

An analogous but linear-algebraic phenomenon appears for endomorphisms of a linear block code $\mathcal C(n,k)$6. Writing an endomorphism as an ambient matrix $\mathcal C(n,k)$7 satisfying $\mathcal C(n,k)$8, the transformation matrices are exactly

$\mathcal C(n,k)$9

where $\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$0 is a code characterization matrix and

$\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$1

Because $\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$2 is closed under addition, scalar multiplication, and multiplication, $\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$3 is a unital $\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$4-subalgebra of $\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$5. The same paper encodes all such endomorphism matrices as a larger linear code

$\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$6

defined by the parity-check matrix

$\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$7

(Mandelbaum et al., 2024).

These examples show that matrix algebras of endomorphisms often arise not as full matrix rings but as coordinate algebras preserving a built-in decomposition: torsion filtration in one case, code subspace structure in the other. A plausible implication is that “matrix algebra of endomorphisms” is best understood as a preservation algebra, with the zero pattern determined by the allowable images of distinguished subobjects (Liu et al., 2016, Mandelbaum et al., 2024).

4. Endomorphism algebras that are not matrix algebras

The identification of an endomorphism ring with a matrix algebra can fail completely in modular representation theory. For the symmetric group $\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$8, over a field $\mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}$9 of characteristic $M_n(\mathbb F_q)$0, and partitions $M_n(\mathbb F_q)$1 with at most two parts, the endomorphism algebra of the Young module $M_n(\mathbb F_q)$2 is obtained as

$M_n(\mathbb F_q)$3

where $M_n(\mathbb F_q)$4 is commutative in this two-part characteristic-$M_n(\mathbb F_q)$5 setting and $M_n(\mathbb F_q)$6 is a primitive idempotent constructed from binary data (Kochhar, 2014).

The resulting algebra is generated by the elements

$M_n(\mathbb F_q)$7

and the Orthogonality Lemma implies that these generators all have square zero. If $M_n(\mathbb F_q)$8, then

$M_n(\mathbb F_q)$9

is a quotient of

$M_n(K)$00

More precisely, if $M_n(K)$01 has dimension $M_n(K)$02 and $M_n(K)$03, then

$M_n(K)$04

where $M_n(K)$05 is a truncation ideal killing sufficiently large square-free monomials involving $M_n(K)$06. The isomorphism type depends only on $M_n(K)$07 (Kochhar, 2014).

This has a direct consequence for the matrix question. Since $M_n(K)$08 is commutative, the corner algebra $M_n(K)$09 is also commutative. A full matrix algebra $M_n(K)$10 is commutative only when $M_n(K)$11. Therefore, except in the one-dimensional case, these endomorphism rings are not matrix algebras. They are finite-dimensional commutative quotient algebras generated by square-zero elements, and hence are highly nonsemisimple (Kochhar, 2014).

This example corrects a common heuristic. Indecomposability of the underlying module does not force its endomorphism ring to resemble a full matrix algebra; in the Young-module setting it instead leads to a local, nilpotent, dimension-controlled commutative algebra (Kochhar, 2014).

5. Matrix-block endomorphisms in operator algebras

In operator algebra, matrix algebras of endomorphisms naturally appear for block-diagonal von Neumann algebras

$M_n(K)$12

Each $M_n(K)$13 is a Type I factor, so finite direct sums

$M_n(K)$14

and more general sums of $M_n(K)$15 are special cases. The corresponding endomorphisms are described by graph and correspondence data. If $M_n(K)$16 is a representation of a graph Toeplitz algebra, the associated endomorphism on

$M_n(K)$17

is

$M_n(K)$18

with strong-operator convergence (Gipson, 2017).

The converse theorem gives a full block-matrix description. If

$M_n(K)$19

is a countable sum of Type I factors and $M_n(K)$20 is a normal $M_n(K)$21-endomorphism of $M_n(K)$22, then there exists a graph $M_n(K)$23 and a representation $M_n(K)$24 such that

$M_n(K)$25

Writing

$M_n(K)$26

each nonzero block map $M_n(K)$27 has multiplicity $M_n(K)$28 and is implemented by isometries

$M_n(K)$29

through

$M_n(K)$30

The integers $M_n(K)$31 are exactly the graph adjacency numbers, so the graph records the matrix-block multiplicity pattern of the endomorphism (Gipson, 2017).

This framework classifies equality and conjugacy of induced endomorphisms by coherent unitary equivalence of the underlying correspondences. It also distinguishes the unital case, where the representation factors through the Cuntz–Pimsner algebra $M_n(K)$32. The central point is that endomorphisms of direct sums of Type I factors are governed by the same kind of multiplicity data that governs endomorphisms of $M_n(K)$33, but now arranged across blocks rather than along a single Hilbert-space multiplicity space (Gipson, 2017).

A related localized matrix-block phenomenon appears in the Cuntz algebra $M_n(K)$34. Its finite core pieces

$M_n(K)$35

parametrize localized endomorphisms $M_n(K)$36 via unitaries $M_n(K)$37, and permutation unitaries yield permutative endomorphisms whose invertibility can be tested by nilpotency criteria or by rooted-tree combinatorics. Thus finite matrix blocks inside $M_n(K)$38 act as coordinate charts for substantial endomorphism families, even though the full endomorphism theory extends beyond these localized models (Conti et al., 2011).

6. Endomorphisms of matrix algebras and matrix-built endomorphism families

The algebra $M_n(K)$39 is itself an object of endomorphism theory. A linear endomorphism

$M_n(K)$40

satisfying

$M_n(K)$41

belongs either to the classical Frobenius family

$M_n(K)$42

with $M_n(K)$43, or to a singular family obtained from a full non-singular $M_n(K)$44-dimensional subspace $M_n(K)$45, an isomorphism $M_n(K)$46, and a nonzero vector $M_n(K)$47, via

$M_n(K)$48

The singular case exists exactly when $M_n(K)$49 admits a division algebra structure over $M_n(K)$50, equivalently when $M_n(K)$51 contains a full non-singular $M_n(K)$52-dimensional subspace (Pazzis, 2010).

This result is notable because it classifies endomorphisms of a matrix algebra by the geometry of its large singular subspaces. The non-singular case preserves the ambient matrix-algebra structure; the singular case collapses $M_n(K)$53 onto an $M_n(K)$54-dimensional subspace all of whose nonzero elements remain invertible (Pazzis, 2010).

A different matrix-built family comes from Yang–Baxter theory. A unitary $M_n(K)$55-matrix

$M_n(K)$56

defines a Cuntz endomorphism $M_n(K)$57 of $M_n(K)$58. The Yang–Baxter equation is equivalent to

$M_n(K)$59

and finite-dimensional matrix blocks

$M_n(K)$60

control the relative commutants

$M_n(K)$61

The paper further proves that the left and right partial traces of an $M_n(K)$62-matrix coincide and are normal, that the partial trace is an invariant of the associated braid-group character, and that upper and lower bounds on the minimal and Jones indices can be read off from finite matrix data such as $M_n(K)$63 (Conti et al., 2019).

In large endomorphism rings, additive decomposition results also show how matrix arguments propagate to infinite settings. For a free $M_n(K)$64-module $M_n(K)$65 of infinite rank,

$M_n(K)$66

and the endomorphism ring is a sum of three nilpotent subrings of nilpotency index $M_n(K)$67. Likewise, every bounded operator on an infinite-dimensional complex Hilbert space is a sum of four automorphisms of order $M_n(K)$68. These statements are proved by first decomposing $M_n(K)$69 matrices using commutators and then transferring the result through the self-similarity of infinite-dimensional endomorphism rings (Breaz et al., 2022).

Taken together, these results suggest a broad taxonomy. Some endomorphism algebras are literally full matrix algebras; some are structural or stabilized matrix algebras determined by preserved subspaces, blocks, or cosets; and some, especially in nonsemisimple representation theory, are explicitly non-matrix algebras. The modern theory therefore studies not a single model but a spectrum of matrix realizations, each controlled by the structure preserved by endomorphisms in the given category (Gupta et al., 2024, Besleaga et al., 2018, Kochhar, 2014, Gipson, 2017).