Given a block matrix M = [[A, B],[C, D]] with A invertible, the Schur complement of A in M is S = D − C A^{-1} B; it reduces block elimination and expresses determinants and inverses via block formulas.

Principle

Encodes the effect of eliminating block A by Gaussian elimination or conditioning in Gaussian distributions; connects blockwise inversion, determinant factorization det(M)=det(A)det(S), and Schur complement positivity with matrix definiteness.

Demonstration

For M = [[A, B],[C, D]] with invertible A, the block inverse is [[A^{-1}+A^{-1} B S^{-1} C A^{-1}, −A^{-1} B S^{-1}],[−S^{-1} C A^{-1}, S^{-1}]]. In multivariate Gaussian covariance Σ partitioned as [[Σ_11, Σ_12],[Σ_21, Σ_22]], the conditional covariance of the second block given the first is the Schur complement Σ_22 − Σ_21 Σ_11^{-1} Σ_12.

Misapplication

Using the Schur complement formula when the pivot block A is singular without replacing A^{-1} by a generalized inverse or reordering blocks, which yields invalid algebraic manipulations.

Consequence

Provides efficient block solvers, criteria for positive (semi)definiteness via Schur complements, and probabilistic interpretations as conditional covariances that enable dimension reduction and stability analysis.

Reversal

Forming the Schur complement of D instead of A (when D is invertible) yields S' = A − B D^{-1} C; choosing the opposite pivot reverses elimination order and leads to complementary block formulas.

Boundary

Requires invertibility of the chosen pivot block for the standard formula; extensions use pseudoinverses, limits, or rank‑factorizations for singular blocks. Applicability is to finite-dimensional block matrices and operators where block inversion makes sense.

Semantic Tension

Confused with Schur decomposition (unitary triangularization) or with matrix minors; the Schur complement is a blockwise algebraic reduction, distinct from similarity or spectral factorizations named after the same mathematician.

Synthesis

The Schur complement is the block-level residual after eliminating a pivot block: an algebraic construct that encodes conditional structure, enables block inversion and determinant factorization, and underpins definiteness tests and block Gaussian elimination.