OctoInformBeta

OpenAI preprint, September 2026 · §3. A mixed trace inequality

§3 (setting): cyclic row stars and Gram matrices: Block spaces for row stars and Gram matrices, and for a finite complex square matrix RR, equal matrix energy of R*RR^* R and RR*R R^*

Cyclic row stars and energy invariance under Gram flip.

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

Given three square matrices on a finite index set, one assembles them and their adjoints into a 3×33 \times 3 block array to extract the cyclic vertex row stars and their Gram matrices. For a finite complex square matrix RR, the accompanying trace identity asks whether R*RR^* R and its Gram flip RR*R R^* have the same matrix energy.

Real Trace and Matrix Energy

The real trace tr⁡ℝ(A)∈ℝ\operatorname{tr}_{\mathbb{R}}(A) \in \mathbb{R} of a complex square matrix A∈ℂι×ιA \in \mathbb{C}^{\iota \times \iota} over a finite index set ι\iota is the real number tr⁡ℝ(A)=Re⁡tr⁡(A).\operatorname{tr}_{\mathbb{R}}(A) = \operatorname{Re}\operatorname{tr}(A).(1)✦

The matrix energy Φ(P)∈ℝ\Phi(P) \in \mathbb{R} of a complex square matrix P∈ℂι×ιP \in \mathbb{C}^{\iota \times \iota} over a finite index set ι\iota is the real trace (1) of PPP \sqrt{P}, where P\sqrt{P} is the square root of PP defined by the continuous functional calculus, Φ(P)=tr⁡ℝ(PP).\Phi(P) = \operatorname{tr}_{\mathbb{R}}\bigl(P \sqrt{P}\bigr).(2)✦

Block Matrices and Vertex Projections

The triangle vertex set VV is the three-element cyclic group of residue classes modulo 33, V=ℤ/3ℤ={0,1,2},V = \mathbb{Z}/3\mathbb{Z} = \{0, 1, 2\},(3)✦ with cyclic arithmetic v+1v + 1 and v−1=v+2v - 1 = v + 2 for v∈Vv \in V.

The block index space Λι\Lambda_\iota of an index set ι\iota is the Cartesian product of the vertex set VV and ι\iota, Λι=V×ι.\Lambda_\iota = V \times \iota.(4)✦

The block matrix Blk⁡(A)∈ℂΛι×Λι\operatorname{Blk}(A) \in \mathbb{C}^{\Lambda_\iota \times \Lambda_\iota} of a 3×33 \times 3 matrix A=(Av,w)v,w∈VA = (A_{v,w})_{v,w \in V} with entries Av,w∈ℂι×ιA_{v,w} \in \mathbb{C}^{\iota \times \iota} over an index set ι\iota is the square matrix on Λι\Lambda_\iota with scalar entries (Blk⁡(A))(v,i),(w,j)=(Av,w)i,j\bigl(\operatorname{Blk}(A)\bigr)_{(v,i),(w,j)} = (A_{v,w})_{i,j}(5)✦ for all v,w∈Vv, w \in V and i,j∈ιi, j \in \iota.

The triangle block array Δblk(W)\Delta_{\mathrm{blk}}(W) of a triple W=(Wv)v∈VW = (W_v)_{v \in V} of matrices Wv∈ℂι×ιW_v \in \mathbb{C}^{\iota \times \iota} over an index set ι\iota is the 3×33 \times 3 matrix of ι×ι\iota \times \iota matrices Δblk(W)=(0W0W2*W0*0W1W2W1*0).\Delta_{\mathrm{blk}}(W) = \begin{pmatrix} 0 & W_0 & W_2^* \\ W_0^* & 0 & W_1 \\ W_2 & W_1^* & 0 \end{pmatrix}.(6)✦

The triangle matrix Δ(W)∈ℂΛι×Λι\Delta(W) \in \mathbb{C}^{\Lambda_\iota \times \Lambda_\iota} of a triple W=(Wv)v∈VW = (W_v)_{v \in V} of matrices Wv∈ℂι×ιW_v \in \mathbb{C}^{\iota \times \iota} over an index set ι\iota is the matrix on Λι\Lambda_\iota formed via (5) from the triangle block array (6), Δ(W)=Blk⁡(Δblk(W)).\Delta(W) = \operatorname{Blk}\bigl(\Delta_{\mathrm{blk}}(W)\bigr).(7)✦

The block-diagonal matrix Diag⁡(D)∈ℂΛι×Λι\operatorname{Diag}(D) \in \mathbb{C}^{\Lambda_\iota \times \Lambda_\iota} of a triple D=(Dv)v∈VD = (D_v)_{v \in V} of matrices Dv∈ℂι×ιD_v \in \mathbb{C}^{\iota \times \iota} over an index set ι\iota is the matrix on Λι\Lambda_\iota formed via (5) from the 3×33 \times 3 diagonal array with diagonal blocks D0,D1,D2D_0, D_1, D_2, Diag⁡(D)=Blk⁡(diag⁡(D0,D1,D2)).\operatorname{Diag}(D) = \operatorname{Blk}\bigl(\operatorname{diag}(D_0, D_1, D_2)\bigr).(8)✦

The vertex projection Pv∈ℂΛι×ΛιP_v \in \mathbb{C}^{\Lambda_\iota \times \Lambda_\iota} at a vertex v∈Vv \in V over an index set ι\iota is the diagonal matrix on Λι\Lambda_\iota given by Pv=diag⁡(𝟏w=v)(w,i)∈Λι,P_v = \operatorname{diag}\bigl(\mathbf{1}_{w = v}\bigr)_{(w,i) \in \Lambda_\iota},(9)✦ where 𝟏w=v\mathbf{1}_{w = v} is 11 when w=vw = v and 00 otherwise.

Cyclic Row Stars and Gram Matrices

The vertex star matrix R̃v(W)∈ℂΛι×Λι\widetilde{R}_v(W) \in \mathbb{C}^{\Lambda_\iota \times \Lambda_\iota} of a triple W=(Wu)u∈VW = (W_u)_{u \in V} of matrices Wu∈ℂι×ιW_u \in \mathbb{C}^{\iota \times \iota} at a vertex v∈Vv \in V over a finite index set ι\iota is the matrix on Λι\Lambda_\iota defined from (9) and (7) by R̃v(W)=PvΔ(W).\widetilde{R}_v(W) = P_v \Delta(W).(10)✦

The star Gram matrix T̃v(W)∈ℂΛι×Λι\widetilde{T}_v(W) \in \mathbb{C}^{\Lambda_\iota \times \Lambda_\iota} associated with a finite index set ι\iota, a triple W=(Wu)u∈VW = (W_u)_{u \in V} of matrices Wu∈ℂι×ιW_u \in \mathbb{C}^{\iota \times \iota}, and a vertex v∈Vv \in V is the matrix on Λι\Lambda_\iota formed from (10) by T̃v(W)=R̃v(W)*R̃v(W).\widetilde{T}_v(W) = \widetilde{R}_v(W)^* \widetilde{R}_v(W).(11)✦

The inserted star Gram matrix T̃v(W,D)∈ℂΛι×Λι\widetilde{T}_v(W, D) \in \mathbb{C}^{\Lambda_\iota \times \Lambda_\iota} for a finite index set ι\iota, a triple W=(Wu)u∈VW = (W_u)_{u \in V} of matrices Wu∈ℂι×ιW_u \in \mathbb{C}^{\iota \times \iota}, a second triple D=(Du)u∈VD = (D_u)_{u \in V} of matrices Du∈ℂι×ιD_u \in \mathbb{C}^{\iota \times \iota} through (8), and a vertex v∈Vv \in V is the matrix on Λι\Lambda_\iota defined from (10) by T̃v(W,D)=R̃v(W)*Diag⁡(D)R̃v(W).\widetilde{T}_v(W, D) = \widetilde{R}_v(W)^* \operatorname{Diag}(D) \widetilde{R}_v(W).(12)✦

The center Gram matrix S̃v(W)∈ℂι×ι\widetilde{S}_v(W) \in \mathbb{C}^{\iota \times \iota} of a triple W=(Wu)u∈VW = (W_u)_{u \in V} of matrices Wu∈ℂι×ιW_u \in \mathbb{C}^{\iota \times \iota} at a vertex v∈Vv \in V over a finite index set ι\iota is the ι×ι\iota \times \iota matrix defined from row vv of (6) by S̃v(W)=∑w∈V(Δblk(W))v,w(Δblk(W))v,w*.\widetilde{S}_v(W) = \sum_{w \in V} \bigl(\Delta_{\mathrm{blk}}(W)\bigr)_{v,w} \bigl(\Delta_{\mathrm{blk}}(W)\bigr)_{v,w}^*.(13)✦

The Claim

Theorem 1. Let ι\iota be a finite index set and let R∈ℂι×ιR \in \mathbb{C}^{\iota \times \iota} be a complex square matrix. Then the matrix energy (2) satisfies Φ(R*R)=Φ(RR*).\Phi(R^* R) = \Phi(R R^*).(14)✦

In the paper, the result reads: for finite-dimensional complex Hilbert spaces HvH_v indexed by v∈ℤ/3ℤv \in \mathbb{Z}/3\mathbb{Z} and linear maps Wv:Hv+1→HvW_v : H_{v+1} \to H_v, setting Rv=[WvWv−1*]R_v = [W_v \;\; W_{v-1}^*], Tv=Rv*RvT_v = R_v^* R_v, and Sv=RvRv*S_v = R_v R_v^* gives Φ(Tv)=Φ(Sv)\Phi(T_v) = \Phi(S_v). In place of vertex-dependent spaces HvH_v, the library works over a single finite index set ι\iota, builds R̃v(W)\widetilde{R}_v(W) and T̃v(W)\widetilde{T}_v(W) on Λι\Lambda_\iota alongside S̃v(W)\widetilde{S}_v(W) on ι\iota, and formulates (14) for an arbitrary square matrix.

What Is Checked

The claim (14) is stated over any finite index set ι\iota, any complex square matrix R∈ℂι×ιR \in \mathbb{C}^{\iota \times \iota}, and the matrix energy (2) defined via the real trace (1) and the square root from the continuous functional calculus, in the setting of the cyclic vertex set (3), block index space (4), block matrices (5), triangle constructions (6) and (7), block diagonal matrices (8), vertex projections (9), vertex stars (10), star and inserted star Gram matrices (11) and (12), and center Gram matrices (13).