OctoInformBeta

OpenAI preprint, September 2026 · §4. Finiteness of approximate representation matches

Lemma 4.1: Powers-Størmer estimates for matrices and finite-spectrum trace data

Square roots of reduced densities are controlled.

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

A change in a tensor vector changes its reduced density operator, and the quantity of interest is the Hilbert-Schmidt distance between the square roots. The ambient Hilbert spaces may be infinite-dimensional; the tensor vectors considered here are algebraic tensors.

A finite positive trace τ\tau on a complex C*-algebra AA is a real-linear functional τ:A→ℝ\tau\colon A\to\mathbb R that is positive on positive elements and cyclic under multiplication, as in (1): τ(a)≥0(a≥0),τ(ab)=τ(ba)(a,b∈A).\tau(a)\geq 0\quad(a\geq 0), \qquad \tau(ab)=\tau(ba)\quad(a,b\in A).(1)✦ For self-adjoint a∈Aa\in A, write |a|=(a2)1/2|a|=(a^2)^{1/2} and let σℝ(a)\sigma_{\mathbb R}(a) denote its real spectrum. For matrices, Tr\operatorname{Tr} denotes the ordinary matrix trace.

Tensor Operators and Density Roots

Let EE and KK be complex Hilbert spaces, with inner products linear in the second variable. Write E′E' and K′K' for their continuous complex-linear duals, and ⊙\odot for the algebraic tensor product with its Hilbert tensor norm. The Hilbert-Schmidt space 𝒮2(K)\mathcal S_2(K) is the completion in (2); its norm is written ∥⋅∥2\|\cdot\|_2: 𝒮2(K)=K⊙K′ˆ.\mathcal S_2(K)=\widehat{K\odot K'}.(2)✦ The tensor operator map TK:K⊙K′→ℬ(K)T_K\colon K\odot K'\to\mathcal B(K), where ℬ(K)\mathcal B(K) is the space of bounded operators on KK, is the bounded linear map specified by (3).✦ Its continuous extension to 𝒮2(K)\mathcal S_2(K) is denoted by JKJ_K: TK(x⊗f)ξ=f(ξ)x(x,ξ∈K,f∈K′).T_K(x\otimes f)\xi=f(\xi)x \qquad(x,\xi\in K,\ f\in K').(3)

For z∈E⊙Kz\in E\odot K, the rectangular tensor operator Cz:E′→KC_z\colon E'\to K is the bounded operator defined linearly from elementary tensors in (4): Cx⊗y(f)=f(x)y(x∈E,y∈K,f∈E′).C_{x\otimes y}(f)=f(x)y \qquad(x\in E,\ y\in K,\ f\in E').(4)✦ The reduced density operator ρz∈ℬ(K)\rho_z\in\mathcal B(K) is defined by (5): ρz=CzCz*.\rho_z=C_zC_z^*.(5)✦ The reduced density root az∈𝒮2(K)a_z\in\mathcal S_2(K) is a chosen Hilbert-Schmidt element representing the positive square root of ρz\rho_z, as in (6): JK(az)=ρz1/2.J_K(a_z)=\rho_z^{1/2}.(6)✦

For a surjective linear isometry V:H0→H1V\colon H_0\to H_1 between normed complex spaces, the unitary extension Vˆ\widehat V is the continuous extension of VV from the completion of H0H_0 to the completion of H1H_1.✦ For a unitary vv on KK, the contragredient unitary v′v' on K′K' is defined by (7): v′f=f∘v−1(f∈K′).v'f=f\circ v^{-1}\qquad(f\in K').(7)✦ The Hilbert-Schmidt conjugation 𝒰v\mathcal U_v on 𝒮2(K)\mathcal S_2(K) is the unitary extension of v⊗v′v\otimes v', specified on elementary tensors by (8): 𝒰v(x⊗f)=vx⊗v′f.\mathcal U_v(x\otimes f)=vx\otimes v'f.(8)✦ For a unitary uu on EE, write u⊗vu\otimes v for the induced unitary on E⊙KE\odot K.

The Claim

Theorem 1 (Powers-Størmer and density roots). For positive semidefinite complex matrices x,yx,y of the same finite size, the matrix Powers-Størmer estimate is (9): Re⁡Tr⁡((x−y)2)≤Re⁡Tr⁡(|x2−y2|).\operatorname{Re}\operatorname{Tr}\bigl((x-y)^2\bigr) \leq \operatorname{Re}\operatorname{Tr}\bigl(|x^2-y^2|\bigr).(9)✦

Let τ\tau be a finite positive trace on AA, and let x,y∈Ax,y\in A be positive. If σℝ(x−y)\sigma_{\mathbb R}(x-y) is finite, the trace estimate is (10): τ((x−y)2)≤τ(|x2−y2|).\tau\bigl((x-y)^2\bigr)\leq\tau\bigl(|x^2-y^2|\bigr).(10)✦

For z,w∈E⊙Kz,w\in E\odot K, the reduced density roots satisfy (11): ∥az−aw∥22≤(∥z∥+∥w∥)∥z−w∥.\|a_z-a_w\|_2^2 \leq (\|z\|+\|w\|)\|z-w\|.(11)✦

For unitaries uu on EE and vv on KK, and z∈E⊙Kz\in E\odot K, the unitary estimate is (12): ∥𝒰v(az)−az∥22≤2∥z∥∥(u⊗v)z−z∥.\|\mathcal U_v(a_z)-a_z\|_2^2 \leq 2\|z\|\,\|(u\otimes v)z-z\|.(12)✦

In the paper, the result reads: for positive Hilbert-Schmidt operators X,YX,Y on any Hilbert space, ∥X−Y∥22≤∥X2−Y2∥1\|X-Y\|_2^2\leq\|X^2-Y^2\|_1, where ∥⋅∥1\|\cdot\|_1 is the trace norm and ∥⋅∥2\|\cdot\|_2 the Hilbert-Schmidt norm.

What Is Checked

The matrix estimate (9) is stated over positive semidefinite matrices; the trace estimate (10) is stated over finite positive traces, with finite real spectrum required of x−yx-y. The density estimates (11) and (12) concern algebraic tensors in arbitrary complex Hilbert spaces, with unitary actions on both factors for the latter.

The density root existence theorem supplies the norm identity (13) for the chosen roots: ∥az∥2=∥z∥(z∈E⊙K).\|a_z\|_2=\|z\|\qquad(z\in E\odot K).(13)