OctoInformBeta

OpenAI preprint, September 2026 · §2. A matrix energy and its Hessian

Lemma 2.3: With rectangular RR and square DD for right-hand side R*DRR^* D R, alongside energy and norm bounds for square RR and Hermitian LL

Sylvester solvability and Hessian cost bounds for inserted matrices.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Take a complex matrix RR with Gram matrix R*RR^* R, together with a complex square matrix DD and a Hermitian matrix LL of compatible dimensions. The question is whether the Sylvester equation for the square root of R*RR^* R admits a solution with right-hand side R*DRR^* D R, and how the Hessian quadratic cost of R*LRR^* L R at R*RR^* R is controlled by the matrix energy of R*RR^* R and by the norms of LL and RR.

Definitions

Fix an index set ι\iota, and let ∥⋅∥\|\cdot\| denote the norm on complex square matrices indexed by ι\iota. Write A*A^* for the conjugate transpose of any complex matrix AA, and write P1/2P^{1/2} for the square root of a complex square matrix PP indexed by ι\iota.

Given a complex square matrix AA indexed by ι\iota, its real trace tr⁡ℝ(A)\operatorname{tr}_{\mathbb{R}}(A) is the real number Re⁡tr⁡(A)\operatorname{Re}\operatorname{tr}(A).✦ The real trace functional tr⁡ℝ\operatorname{tr}_{\mathbb{R}} on the complex square matrices indexed by ι\iota is the continuous real-linear functional sending AA to tr⁡ℝ(A)\operatorname{tr}_{\mathbb{R}}(A), with operator norm ∥tr⁡ℝ∥\|\operatorname{tr}_{\mathbb{R}}\|.✦ For two complex square matrices XX and YY indexed by ι\iota, the real Hilbert-Schmidt inner product ⟨X,Y⟩HS\langle X, Y \rangle_{\mathrm{HS}} is the real number tr⁡ℝ(X*Y)\operatorname{tr}_{\mathbb{R}}(X^* Y).✦

The matrix energy Φ(P)\Phi(P) of a complex square matrix PP indexed by ι\iota is the real number Φ(P)=tr⁡ℝ(PP1/2).\Phi(P) = \operatorname{tr}_{\mathbb{R}}(P P^{1/2}).(1)✦

Associated with each complex square matrix AA indexed by ι\iota is the Sylvester operator SAS_A, the map on complex square matrices indexed by ι\iota defined by SA(X)=AX+XA.S_A(X) = A X + X A.(2)✦ The Sylvester penalty qA(X)q_A(X) of complex square matrices AA and XX indexed by ι\iota is the real number ⟨X,SA(X)⟩HS\langle X, S_A(X) \rangle_{\mathrm{HS}}.✦ For complex square matrices AA, KK, and XX indexed by ι\iota, the Sylvester variational functional VA,K(X)V_{A,K}(X) is the real number VA,K(X)=2⟨K,X⟩HS−qA(X).V_{A,K}(X) = 2 \langle K, X \rangle_{\mathrm{HS}} - q_A(X).(3)✦

The root variational cost b̃A(K)\tilde{b}_A(K) of complex square matrices AA and KK indexed by ι\iota is the real number b̃A(K)=32supXVA,K(X),\tilde{b}_A(K) = \frac{3}{2} \sup_{X} V_{A,K}(X),(4)✦ where XX ranges over all complex square matrices indexed by ι\iota. For complex square matrices PP and KK indexed by ι\iota, the Hessian quadratic cost bP(K)b_P(K) is the real number bP(K)=b̃P1/2(K).b_P(K) = \tilde{b}_{P^{1/2}}(K).(5)✦

The Claim

Theorem 1. Let κ\kappa be a finite index set, and let RR be a κ×ι\kappa \times \iota complex matrix and DD a κ×κ\kappa \times \kappa complex matrix. Then there exists a complex square matrix YY indexed by ι\iota satisfying the Sylvester equation (6): S(R*R)1/2(Y)=R*DR.S_{(R^* R)^{1/2}}(Y) = R^* D R.(6)✦ Moreover, for any complex square matrices RR and LL indexed by ι\iota with LL Hermitian, the Hessian quadratic cost satisfies the energy bound (7), bR*R(R*LR)≤34∥L∥2Φ(R*R),b_{R^* R}(R^* L R) \le \frac{3}{4} \|L\|^2 \Phi(R^* R),(7)✦ and the norm bound (8), |bR*R(R*LR)|≤34∥tr⁡ℝ∥∥L∥2∥R∥3.|b_{R^* R}(R^* L R)| \le \frac{3}{4} \|\operatorname{tr}_{\mathbb{R}}\| \|L\|^2 \|R\|^3.(8)✦

In the paper, the result reads: for a linear map R:E→FR \colon E \to F between finite-dimensional complex Hilbert spaces and a Hermitian operator L:F→FL \colon F \to F, the matrix R*LRR^* L R is supported on R*RR^* R and satisfies bR*R(R*LR)≤34∥L∥op2Φ(R*R)b_{R^* R}(R^* L R) \le \frac{3}{4} \|L\|_{\mathrm{op}}^2 \Phi(R^* R), where ∥L∥op\|L\|_{\mathrm{op}} denotes the operator norm. In place of linear maps between EE and FF and the operator norm ∥L∥op\|L\|_{\mathrm{op}}, the library formulates Sylvester solvability (6) for a κ×ι\kappa \times \iota matrix RR and a κ×κ\kappa \times \kappa matrix DD, and states the bounds (7) and (8) for complex square matrices RR and LL indexed by ι\iota with LL Hermitian using ∥⋅∥\|\cdot\| and ∥tr⁡ℝ∥\|\operatorname{tr}_{\mathbb{R}}\|.

What Is Checked

The first claim (6) is stated over the index set ι\iota, any finite index set κ\kappa, any κ×ι\kappa \times \iota complex matrix RR, and any κ×κ\kappa \times \kappa complex matrix DD, in terms of the Sylvester operator and the square root; the remaining two claims (7) and (8) are stated over any complex square matrices RR and LL indexed by ι\iota with LL Hermitian, in terms of the norm ∥⋅∥\|\cdot\|, the operator norm ∥tr⁡ℝ∥\|\operatorname{tr}_{\mathbb{R}}\|, the matrix energy, and the Hessian quadratic cost.