15.15 Tensor Symmetric Matrix Case

Tensor Symmetric Matrix Case is the specialization of the general theory of symmetric tensors to order two, in which the abstract notion of a symmetric tensor reduces to the familiar object of a symmetric matrix, and every general definition, invariant, and decomposition question about symmetric tensors acquires a concrete, classically understood answer.

Why Order Two Is Singled Out

Tensors versus Matrices

A tensor of order d over an n-dimensional vector space is, in coordinates, an array indexed by d separate indices, each running over the n basis directions. When d equals two, this array is exactly a square matrix, and the entire apparatus of linear algebra, developed independently of and long before general tensor theory, becomes directly applicable. The Matrix Case is the name given to this overlap: it is the region of symmetric tensor theory where classical linear algebra supplies complete, exact answers to questions that remain open or intricate for tensors of order three and above.

Scope Within the Symmetric Tensor Hierarchy

Within the broader classification of symmetric tensors by order, the Matrix Case sits at the base of the hierarchy, beneath the order-three and higher cases addressed by decomposition theory, apolarity, and the Rank Relation. Understanding the Matrix Case in full is a prerequisite for appreciating what breaks down, and what generalizes cleanly, once the order is increased.

The Governing Constraint

Component-Level Symmetry

The defining requirement of the Matrix Case, elaborated in full as the Component Constraint, is that the two-index array of components be unchanged under swapping its indices:

which is equivalent to the array, read as a matrix, being equal to its own transpose. Because the symmetric group on two elements has only two members, the identity and the single transposition, this is the entire symmetry requirement; there is no distinction, as there is for higher orders, between different partial symmetry types.

Dimension of the Space of Symmetric Matrices

The space of symmetric order-two tensors over an n-dimensional space has dimension equal to n choose two plus n, or equivalently n times n plus one over two,

matching the number of independent entries on and above the main diagonal.

Structural Results Available Only in This Case

The Spectral Theorem

Every real symmetric matrix is orthogonally diagonalizable, meaning there exists an orthonormal basis of eigenvectors in which the matrix takes diagonal form with real eigenvalues on the diagonal. This result, the spectral theorem, has no exact counterpart for symmetric tensors of order three or higher, where an eigenvector-like decomposition (in the sense of tensor eigenvalues) exists but does not, in general, give an additive decomposition into orthogonal rank-one pieces with the same simplicity or uniqueness.

Equality of Symmetric Rank and Ordinary Rank

As detailed in the Rank Relation, the symmetric rank of a symmetric tensor can, for orders three and above, be strictly larger than its ordinary tensor rank. In the Matrix Case, this gap never opens: the two ranks always coincide, and both equal the number of nonzero eigenvalues obtained from the spectral theorem. This coincidence is one of the cleanest facts distinguishing the Matrix Case from the general theory.

Sylvester's Law of Inertia

Over the real numbers, symmetric matrices are classified up to congruence (change of basis) not merely by rank but by signature, the numbers of positive and negative eigenvalues, a classification with no direct higher-order analogue of comparable simplicity. This connects the Matrix Case tightly to the classification of symmetric bilinear and quadratic forms, formalized as the Bilinear Form Relation.

Decomposition in the Matrix Case

Eigendecomposition as Symmetric Decomposition

The spectral theorem supplies, immediately, a decomposition of a symmetric matrix into pure power forms: writing the matrix using its orthonormal eigenvectors u_1 through u_n and eigenvalues lambda_1 through lambda_n,

discards the terms with zero eigenvalue to leave a minimal decomposition, recovering the symmetric rank directly as the count of nonzero eigenvalues, with no further search or optimization required, unlike Reconstruction procedures needed for higher orders.

Uniqueness of the Decomposition

Whenever all nonzero eigenvalues are distinct, the eigendecomposition, and hence the minimal Term Set, is unique up to the sign ambiguity of each eigenvector. Repeated eigenvalues introduce genuine non-uniqueness, since any orthonormal basis of the corresponding eigenspace yields an equally valid decomposition, an early and simple instance of the positive-dimensional families of term sets that recur, in more complicated form, at higher orders.

Applications Anchored in the Matrix Case

Quadratic Forms and Optimization

Symmetric matrices, through the Bilinear Form Relation, correspond to quadratic forms, and the sign pattern of the eigenvalues (definiteness) governs the local behavior of critical points in optimization and the classification of conic sections and quadric surfaces in geometry.

Covariance and Metric Structures

Positive semidefinite symmetric matrices arise as covariance matrices in probability and statistics and as metric tensors in geometry; their eigendecomposition, guaranteed by the results specific to the Matrix Case, underlies principal component analysis, the geometry of ellipsoids, and the definition of distances in curved spaces.