--- title: Matrix Algebras of Endomorphisms url: https://www.emergentmind.com/topics/matrix-algebras-of-endomorphisms type: topic --- # Matrix Algebras of Endomorphisms 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 [2405.18121], [1802.03427], [1412.2538]. | 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 \(Y^\mu\) in characteristic \(2\) | Square-zero commutative quotient | Generally not a matrix algebra | | \(\bigoplus_i B(H_i)\) or \(\bigoplus_i M_{n_i}(\mathbb C)\) | Block endomorphisms from partial isometries | Matrix-block operator algebra | ## 1. Full matrix algebras as the basic endomorphism model For an \(n\)-dimensional \(K\)-vector space \(V\), choosing a basis identifies \(\operatorname{End}_K(V)\) with the full matrix algebra \(M_n(K)\). A particularly explicit realization uses a Galois extension \(L/K\) of degree \(n\): if \(V\cong L\) as \(K\)-vector spaces, then \[ \operatorname{End}_K(V)\cong \operatorname{End}_K(L)\cong M_n(K). \] In this model, the trace pairing \[ (a,x)\mapsto \operatorname{Tr}_{L/K}(ax) \] identifies \(L^*\) with \(L\), and the tensor isomorphism \[ L\otimes_K L^*\cong \operatorname{End}_K(L) \] gives a canonical description of rank-one endomorphisms as maps \[ x\mapsto \operatorname{Tr}_{L/K}(ax)\,u. \] The operator trace is then computed by the field trace through \[ \operatorname{tr}(u\otimes \operatorname{Tr}\cdot a)=\operatorname{Tr}_{L/K}(au) \] [2405.18121]. 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 [2405.18121]. A related rigidity phenomenon appears for linear maps from \(k^d\) into \(\operatorname{End}_k(V)\). 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 \(M_m(k)\), again recovering the classical endomorphism picture from internal relations among matrices [1507.08361]. ## 2. Structural matrix algebras and generalized flags A structural matrix algebra is obtained from a preorder \(p\) on \(\{1,\dots,n\}\) by imposing the zero-pattern condition \[ M(p,k)=\{(a_{ij})\in M_n(k)\mid a_{ij}=0 \text{ whenever } (i,j)\notin p\}. \] The key theorem is that \(M(p,k)\) is itself an endomorphism algebra, not of a plain vector space, but of a generalized flag determined by the preorder. Writing \(i\sim j\) when \(i\,p\,j\) and \(j\,p\,i\), one obtains a poset \(C\) of equivalence classes. A \(p\)-flag is then an \(n\)-dimensional vector space \(V\) together with subspaces \((V_\alpha)_{\alpha\in C}\) arising from a basis partitioned by these classes so that \[ \bigcup_{\beta\le \alpha} B_\beta \] is a basis of \(V_\alpha\). The preserving endomorphisms \[ \operatorname{End}(F)=\{f\in \operatorname{End}_k(V)\mid f(V_\alpha)\subseteq V_\alpha \text{ for all }\alpha\in C\} \] satisfy \[ \operatorname{End}(F)\cong M(p,k) \] [1802.03427]. This identification generalizes the familiar equality \(M_n(k)\cong \operatorname{End}_k(k^n)\). 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 \(i\,p\,j\) records exactly when a basis vector in class \(\hat j\) may map to one in class \(\hat i\) without violating preservation of the subspaces \(V_\alpha\) [1802.03427]. The flag viewpoint also reorganizes the internal structure of \(M(p,k)\). Its invariant subspaces become the subspaces \(V_D\) indexed by subsets, equivalently antichains, of \(C\). Automorphisms of the algebra are described by a combination of inner automorphisms, automorphisms of the poset \(C\) preserving block sizes, and transitive scalar rescalings of matrix units. If the \(p\)-flag is equipped with a group grading, then \[ \operatorname{End}(F)=\bigoplus_{\sigma\in G}\operatorname{End}(F)_\sigma \] becomes a graded algebra, and the matrix units satisfy \[ \deg(e_{ij})=g_i g_j^{-1}. \] Thus good gradings on structural matrix algebras are induced by graded flags, and under graph-theoretic conditions on the Hasse diagram of \(C\), every good grading arises in this way [1802.03427]. ## 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 \(p\)-group \[ \mathbb Z_p\times \mathbb Z_{p^m}. \] Its endomorphism ring is isomorphic to \[ E_{p,p^m}=\left\{\begin{pmatrix} a & b \\ p^{m-1}c & d \end{pmatrix}: a,b,c\in \mathbb Z_p,\ d\in \mathbb Z_{p^m}\right\}, \] with entrywise addition and a multiplication adapted to the mixed moduli in the four corners. This ring is not \(M_2(\mathbb Z_{p^m})\), not \(M_2(\mathbb Z_p)\), and, by the stated theorem, cannot be embedded into matrices over any commutative ring. Its lower-left entry is forced into the ideal \(p^{m-1}\mathbb Z_{p^m}\), reflecting the fact that \(\mathrm{Hom}(\mathbb Z_p,\mathbb Z_{p^m})\) consists exactly of elements killed by \(p\) [1605.00805]. The matrix model makes arithmetic explicit. Invertibility is characterized by the conditions \[ a\neq 0 \quad\text{and}\quad u_0\neq 0, \] where \(u_0\) is the lowest \(p\)-adic digit of the lower-right entry \(d\in \mathbb Z_{p^m}\). Every element satisfies a quadratic relation \[ A^2+rA+sI=0, \] with \[ r=-(a+d)\bmod p^m,\qquad s=(ad-p^{m-1}bc)\bmod p^m, \] so the algebra has a Cayley–Hamilton-type structure despite not being a full matrix ring [1605.00805]. An analogous but linear-algebraic phenomenon appears for endomorphisms of a linear block code \(\mathcal C(n,k)\subseteq \mathbb F_q^n\). Writing an endomorphism as an ambient matrix \(\bm T\in M_n(\mathbb F_q)\) satisfying \(\bm T(\mathcal C)\subseteq \mathcal C\), the transformation matrices are exactly \[ \mathcal T_E(\mathcal C)=\bm A\mathcal Z\bm A^{-1}, \] where \(\bm A\) is a code characterization matrix and \[ \mathcal Z=\left\{\begin{bmatrix}\bm C & 0\\ \bm D & \bm E\end{bmatrix}\right\}. \] Because \(\mathcal Z\) is closed under addition, scalar multiplication, and multiplication, \(\mathcal T_E(\mathcal C)\) is a unital \(\mathbb F_q\)-subalgebra of \(M_n(\mathbb F_q)\). The same paper encodes all such endomorphism matrices as a larger linear code \[ \mathcal C_E\bigl(n^2,\,2kn-k^2\bigr), \] defined by the parity-check matrix \[ \bm H_E=\bm A_2^{\mathsf T}\otimes \bm\Omega_1 \] [2402.00562]. 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 [1605.00805], [2402.00562]. ## 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 \(\Sigma_r\), over a field \(K\) of characteristic \(2\), and partitions \(\lambda,\mu\) with at most two parts, the endomorphism algebra of the Young module \(Y^\mu\) is obtained as \[ \operatorname{End}_{K[\Sigma_r]}(Y^\mu)\cong e_{m,g}S_K(\lambda)e_{m,g}, \] where \(S_K(\lambda)=\operatorname{End}_{K[\Sigma_r]}(M^\lambda)\) is commutative in this two-part characteristic-\(2\) setting and \(e_{m,g}\) is a primitive idempotent constructed from binary data [1412.2538]. The resulting algebra is generated by the elements \[ T=\{e_{m,g}b(2^s): (m+2g)_s=0\}, \] and the Orthogonality Lemma implies that these generators all have square zero. If \(k=|T|\), then \[ \operatorname{End}_{K[\Sigma_r]}(Y^\mu) \] is a quotient of \[ K[x_1,\dots,x_k]/(x_i^2: i=1,\dots,k). \] More precisely, if \(A=\operatorname{End}_{K[\Sigma_r]}(Y^\mu)\) has dimension \(n\) and \(2^{k-1}