OctoInformBeta

OpenAI preprint, September 2026 · §5. Sampling and orthogonal packing · §5

§5 (proof of Theorem 1.3): the packing rounds: Orthogonal PP, SS with energy at most EN(η)E_N(\eta), from nuclearity or positive diagonals, and halving error bounded by remaining trace and balance

Packing rounds build orthogonal projection pairs with explicit bounds.

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

Let A⊆DA\subseteq D be a unital inclusion of C*^*-algebras and KK a finite set of unitaries of AA. Write T(D)T(D) for the set of tracial states of DD. One looks for a projection PP of DD that nearly commutes with every x∈Kx\in K in the trace 2-norm ∥d∥2,τ=τ(d*d)1/2\|d\|_{2,\tau}=\tau(d^*d)^{1/2} of each τ∈T(D)\tau\in T(D), and whose uniform halving error sup⁡τ|τ(Px)−12τ(x)|\sup_\tau|\tau(Px)-\frac12\tau(x)| is small for every x∈Kx\in K.

The packing builds such a PP together with a projection SS orthogonal to it over NN rounds, and bounds at every trace the commutator energy of PP and SS against KK, the trace of the uncovered part 1−P−S1-P-S and the balance |τ(Px)−τ(Sx)||\tau(Px)-\tau(Sx)|. A round works inside blocks, the ranges of rows of partial isometries on which every element of KK, once compressed, is close to a complex matrix tensored with the row’s projection.

The library states a packing step on one row, a sampled round over a weighted family of blocks, the finite packing under nuclearity of AA or under positive diagonals for AA, the halving estimate for an orthogonal pair, and the existence of a real number η\eta with 0<η<10<\eta<1 and a number NN of rounds that make the final bounds smaller than any ε>0\varepsilon>0.

Hypotheses on the Algebras

A C*^*-algebra DD has real rank zero when for every self-adjoint a∈Da\in D and every ε>0\varepsilon>0 there is a self-adjoint b∈Db\in D with finite spectrum and ∥a−b∥<ε\|a-b\|<\varepsilon.✦ An element gg of DD is full when the closure of the two-sided ideal generated by gg is DD.✦ We say that DD has arbitrarily small full projections when for every δ>0 there is a full projection g∈D with supττ(g)<δ.\text{for every }\delta>0\text{ there is a full projection }g\in D\text{ with } \sup_\tau\tau(g)<\delta.(1)

A C*^*-norm on a complex algebra RR with involution is an algebra norm on RR with ∥x*x∥=∥x∥2\|x^*x\|=\|x\|^2 for all x∈Rx\in R.✦ A C*^*-algebra AA is nuclear when, for every C*^*-algebra CC, the algebraic tensor product A⊙CA\odot C carries at most one C*^*-norm.✦ A bistochastic Kraus family in AA is a finite family (v1,…,vk)(v_1,\dots,v_k) of elements of AA with ∑ivi*vi=1=∑ivivi*\sum_iv_i^*v_i=1=\sum_iv_iv_i^*.✦

A positive diagonal for a finite set KK of unitaries of AA with tolerance μ\mu consists of weights w1,…,wm≥0w_1,\dots,w_m\ge0 with ∑iwi=1\sum_iw_i=1 and coefficients v1,…,vm∈Av_1,\dots,v_m\in A of norm at most 11 such that ∥1−∑iwivivi*∥<μand∥∑iwi(xvi⊗vi*−vi⊗vi*x)∥<μfor every x∈K.\Big\|1-\sum_iw_iv_iv_i^*\Big\|<\mu \quad\text{and}\quad \Big\|\sum_iw_i\,(xv_i\otimes v_i^*-v_i\otimes v_i^*x)\Big\|<\mu \quad\text{for every }x\in K.(2)✦

Rows, Blocks and Rounds

A projection row UU over a projection qq of DD, indexed by a finite set II, is a family (ui)i∈I(u_i)_{i\in I} of elements of DD with ui*ui=qu_i^*u_i=q for every ii and ui*uj=0u_i^*u_j=0 for i≠ji\neq j.✦ The row matrix coefficients of an element a∈Da\in D are the elements uiauj*u_iau_j^*✦; for a complex matrix s=(sij)i,j∈Is=(s_{ij})_{i,j\in I} the row matrix combination of aa is U(s⊗a)U*=∑i,jsijuiauj*U(s\otimes a)U^*=\sum_{i,j}s_{ij}u_iau_j^*✦, and the row diagonal sum of aa is U(1⊗a)U*=∑iuiaui*U(1\otimes a)U^*=\sum_iu_iau_i^*✦. The row diagonal sum of qq is the range p=UU*p=UU^* of UU.

Let p,z,y,tp,z,y,t be elements of DD, c,η,εc,\eta,\varepsilon real numbers and KK a finite set of unitaries of DD. The quadruple (p,z,y,t)(p,z,y,t) satisfies the step bounds at (c,η,ε;K)(c,\eta,\varepsilon;K) when, for every τ∈T(D)\tau\in T(D), every x∈Kx\in K and every a∈Da\in D with ∥a∥≤1\|a\|\le1, |τ(y)−cτ(p)|≤ε,|τ(zy)−cτ(zp)|≤ε,τ(y)−3τ(zy)≤τ(t)≤τ(y),∥t−y∥2,τ2≤5τ(zy),|τ((t−y)a)|≤2τ(y)1/2∥t−y∥2,τ,|τ(yx)−cτ(px)|≤η(τ(y)+cτ(p))+2ε,∥[y,x]∥2,τ2≤c∥[p,x]∥2,τ2+6η(τ(y)+cτ(p))+10ε,\begin{aligned} &|\tau(y)-c\,\tau(p)|\le\varepsilon,\qquad|\tau(zy)-c\,\tau(zp)|\le\varepsilon,\\ &\tau(y)-3\,\tau(zy)\le\tau(t)\le\tau(y),\qquad\|t-y\|_{2,\tau}^2\le5\,\tau(zy),\\ &|\tau((t-y)a)|\le2\,\tau(y)^{1/2}\,\|t-y\|_{2,\tau},\\ &|\tau(yx)-c\,\tau(px)|\le\eta\,(\tau(y)+c\,\tau(p))+2\varepsilon,\\ &\|[y,x]\|_{2,\tau}^2\le c\,\|[p,x]\|_{2,\tau}^2+6\eta\,(\tau(y)+c\,\tau(p)) +10\varepsilon, \end{aligned}(3)✦ where τ(p)\tau(p), τ(y)\tau(y), τ(t)\tau(t), τ(zy)\tau(zy) and τ(zp)\tau(zp) stand for their real parts.

A weighted block family FF under an element rr of DD consists of a finite index set JJ and, for each j∈Jj\in J, a weight wjw_j with 0<wj≤10<w_j\le1, a nonzero projection qj≤rq_j\le r, a number of columns kj≥1k_j\ge1 and a projection row UjU_j over qjq_j indexed by {1,…,kj}\{1,\dots,k_j\}.✦ The block range projections of a weighted block family FF are the ranges pj=UjUj*p_j=U_jU_j^* of its rows UjU_j.✦

Let FF be such a family, z0z_0 an element of DD, KK a finite set of unitaries of DD and η,δ\eta,\delta real numbers. A sampled packing round for FF, z0z_0, KK, η\eta, δ\delta consists of a projection zz of DD, the final projection, and, for each a=(j,ℓ)a=(j,\ell) with j∈Jj\in J and ℓ∈{0,1}\ell\in\{0,1\}, an occupied projection zaz_a, a sampled projection yay_a and an inserted projection tat_a, such that za≤z,tata′=0(a≠a′),z0ta=0,z=z0+∑ata,z_a\le z,\qquad t_at_{a'}=0\ \ (a\neq a'),\qquad z_0t_a=0,\qquad z=z_0+\sum_at_a,(4)✦ and each (pj,za,ya,ta)(p_j,z_a,y_a,t_a) satisfies the step bounds (3) at (ηwj/2,η,δ;K)(\eta w_j/2,\eta,\delta;K).

Orthogonal Pairs and Packing Bounds

An orthogonal projection pair (P,S)(P,S) in DD is a pair of projections P,SP,S of DD with PS=0PS=0✦; its total projection is P+SP+S✦. For τ∈T(D)\tau\in T(D) and x∈Dx\in D, the commutator energy, the remaining trace and the balance of the pair are ℰτ(x)=∥[P,x]∥2,τ2+∥[S,x]∥2,τ2,ρ(τ)=τ(1−P−S),βτ(x)=|τ(Px)−τ(Sx)|.\mathcal E_\tau(x)=\|[P,x]\|_{2,\tau}^2+\|[S,x]\|_{2,\tau}^2,\qquad \rho(\tau)=\tau(1-P-S),\qquad \beta_\tau(x)=|\tau(Px)-\tau(Sx)|.(5)

For a real number η\eta and N∈ℕN\in\mathbb N, the packing energy, the packing remainder and the packing balance are EN(η)=η((1+1000η)N−1),RN(η)=(1−η)N+3Nη2+3EN(η),BN(η)=N(7η2+2η(60EN(η)+75η)1/2).\begin{aligned} E_N(\eta)&=\eta\,\big((1+1000\eta)^N-1\big),\qquad R_N(\eta)=(1-\eta)^N+3N\eta^2+3E_N(\eta),\\ B_N(\eta)&=N\big(7\eta^2+2\eta\,(60E_N(\eta)+75\eta)^{1/2}\big). \end{aligned}(6)

Given a unital inclusion A⊆DA\subseteq D, a finite set KK of unitaries of AA, a real number η\eta and N∈ℕN\in\mathbb N, an orthogonal projection pair (P,S)(P,S) in DD satisfies the packing bounds for KK, η\eta, NN when, for every τ∈T(D)\tau\in T(D) and every x∈Kx\in K, ℰτ(x)≤EN(η),ρ(τ)≤RN(η),βτ(x)≤BN(η).\mathcal E_\tau(x)\le E_N(\eta),\qquad\rho(\tau)\le R_N(\eta),\qquad \beta_\tau(x)\le B_N(\eta).(7)✦

The Claim

Theorem 1. Let DD be a C*^*-algebra; in (a) to (d), DD has real rank zero and arbitrarily small full projections (1).

  1. (Packing step.) Let UU be a projection row over a projection qq of DD, indexed by a finite set II, with range pp; let zz be a projection of DD, KK a finite set of unitaries of DD, c∈[0,1]c\in[0,1], η<1\eta<1 and ε>0\varepsilon>0; and suppose that each x∈Kx\in K admits a complex matrix HH indexed by II with ∥pxp−U(H⊗q)U*∥<η\|pxp-U(H\otimes q)U^*\|<\eta. Then there are projections yy and tt of DD with t≤1−zand(p,z,y,t) satisfies the step bounds at (c,η,ε;K).t\le1-z\quad\text{and}\quad(p,z,y,t)\text{ satisfies the step bounds at } (c,\eta,\varepsilon;K).(8)✦

  2. (Sampled round.) Let z0z_0 be a projection of DD and FF a weighted block family under an element rr of DD, with weights wjw_j, projections qjq_j, numbers of columns kjk_j, rows UjU_j over qjq_j and ranges pjp_j; let KK be a finite set of unitaries of DD, 0<η<10<\eta<1, σ≤η\sigma\le\eta and δ>0\delta>0; and suppose that for every jj and every x∈Kx\in K there is H∈Mkj(ℂ)H\in M_{k_j}(\mathbb C) with ∥pjxpj−Uj(H⊗qj)Uj*∥<σ\|p_jxp_j-U_j(H\otimes q_j)U_j^*\|<\sigma. Then there is a sampled packing round for FF, z0z_0, KK, η\eta, δ\delta whose projections satisfy, for every a=(j,ℓ)a=(j,\ell), (pj,za,ya,ta) satisfies the step bounds at (ηwj/2,σ,δ;K).(p_j,z_a,y_a,t_a)\text{ satisfies the step bounds at }(\eta w_j/2,\sigma,\delta;K).(9)✦

  3. (Packing from nuclearity.) Let AA be a nuclear C*^*-algebra with a unital inclusion A⊆DA\subseteq D, and suppose that DD has a tracial state. Let 0<η<10<\eta<1, N∈ℕN\in\mathbb N, KK a finite set of unitaries of AA and 𝒲\mathcal W a finite set of bistochastic Kraus families in AA. Then there is an orthogonal projection pair (P,S)(P,S) in DD with $$\begin{gathered} (P,S)\text{ satisfies the packing bounds \eqref{eq:bounds} for }K,\eta,N,\\ \textstyle\sum_i\mathcal E_\tau(v_i)\le E_N(\eta)\text{ for every }(v_i)\in\mathcal W \text{ and every }\tau\in T(D). \end{gathered}$$(10)✦

  4. (Packing from positive diagonals.) Let A⊆DA\subseteq D be a unital inclusion and 0<η<10<\eta<1, and suppose that every finite set of unitaries of AA has a positive diagonal with tolerance η2\eta^2. Then for every N∈ℕN\in\mathbb N and every finite set KK of unitaries of AA there is an orthogonal projection pair (P,S)(P,S) in DD such that $$(P,S)\text{ satisfies the packing bounds \eqref{eq:bounds} for }K,\eta,N.$$(11)✦

  5. (Halving estimate.) Let (P,S)(P,S) be an orthogonal projection pair in DD, τ∈T(D)\tau\in T(D) and xx a unitary of DD. Then |τ(Px)−12τ(x)|≤12(ρ(τ)+βτ(x)).\big|\tau(Px)-\tfrac12\tau(x)\big|\le\tfrac12\big(\rho(\tau)+\beta_\tau(x)\big).(12)✦

  6. (Parameters.) For every ε>0\varepsilon>0 there are a real number η\eta with 0<η<10<\eta<1 and N∈ℕN\in\mathbb N such that EN(η)1/2<εand12(RN(η)+BN(η))<ε.E_N(\eta)^{1/2}<\varepsilon\quad\text{and}\quad \tfrac12\big(R_N(\eta)+B_N(\eta)\big)<\varepsilon.(13)✦

In the paper, the result reads: the packing builds orthogonal projections PP, SS with energy EN(η)=η((1+1000η)N−1)E_N(\eta)=\eta((1+1000\eta)^N-1), remaining mass and coefficient balance controlled round by round, and then |τ(Px)−12τ(x)|≤12(τ(1−P−S)+|τ(Px)−τ(Sx)|)|\tau(Px)-\frac12\tau(x)|\le\frac12(\tau(1-P-S)+|\tau(Px)-\tau(Sx)|) gives the halving estimate.

What Is Checked

The claims are stated over a C*^*-algebra DD and its tracial states, with real rank zero and arbitrarily small full projections (1) in (a) to (d), over projection rows, weighted block families and sampled packing rounds with the step bounds (3), over orthogonal projection pairs with the quantities (5) and the packing bounds (7), and over a nuclear AA or positive diagonals (2) for AA.