OctoInformBeta

OpenAI preprint, September 2026 · §2. Preliminaries

2.1 Rordam-Winter dimension-drop criterion: Unital **-homomorphisms from I(N,N+1)I(N, N+1) via cutoff rescaling under elementwise Cuntz subequivalence, with Gram and square-zero weighted units

The Rørdam-Winter criterion yields dimension-drop algebra homomorphisms.

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 unital C*C^*-algebra AA containing a family of N×NN \times N elements (N≥2N \ge 2) whose products satisfy the matrix-unit relations scaled by a positive contraction h∈Ah \in A. When is the defect 1A−h1_A - h small enough relative to a diagonal entry of this family to support a unital **-homomorphism from the dimension-drop algebra into AA?

These notes go through the dimension-drop criterion answering this question and the definitions it is built on.

Dimension-drop Algebras and Cuntz Subequivalence

For a natural number k≥1k \ge 1, let MkM_k denote the C*C^*-algebra of k×kk \times k complex matrices with identity 1k1_k; for a positive natural number NN, index MNM_N, MN+1M_{N+1}, and MN(N+1)M_{N(N+1)} by {0,…,N−1}\{0, \dots, N-1\}, {0,…,N}\{0, \dots, N\}, and {0,…,N−1}×{0,…,N}\{0, \dots, N-1\} \times \{0, \dots, N\}, respectively. The left and right matrix inclusions ιL:MN→MN(N+1)\iota_L : M_N \to M_{N(N+1)} and ιR:MN+1→MN(N+1)\iota_R : M_{N+1} \to M_{N(N+1)} are the unital **-homomorphisms given by ιL(a)=a⊗1N+1\iota_L(a) = a \otimes 1_{N+1} for a∈MNa \in M_N and ιR(b)=1N⊗b\iota_R(b) = 1_N \otimes b for b∈MN+1b \in M_{N+1}. For t∈[0,1]t \in [0,1], let ev⁡t:C([0,1],MN(N+1))→MN(N+1)\operatorname{ev}_t : C([0,1], M_{N(N+1)}) \to M_{N(N+1)} be the unital **-homomorphism of point evaluation f↦f(t)f \mapsto f(t); the dimension-drop algebra I(N,N+1)I(N, N+1) is the unital **-subalgebra of C([0,1],MN(N+1))C([0,1], M_{N(N+1)}) defined by I(N,N+1)={f∈C([0,1],MN(N+1)):ev⁡0(f)∈ιL(MN),ev⁡1(f)∈ιR(MN+1)}.I(N, N+1) = \{f \in C([0,1], M_{N(N+1)}) : \operatorname{ev}_0(f) \in \iota_L(M_N),\; \operatorname{ev}_1(f) \in \iota_R(M_{N+1})\}.(1)

A Pauli matrix system of rank N≥2N \ge 2 is a finite family (ea)a∈Λ(e_a)_{a \in \Lambda} in MNM_N with |Λ|=N2|\Lambda| = N^2 and a distinguished index 0∈Λ0 \in \Lambda such that e0=1Ne_0 = 1_N, each eae_a is self-adjoint with ea2=1Ne_a^2 = 1_N, Tr⁡(eaeb)=Nδab\operatorname{Tr}(e_a e_b) = N \delta_{ab} for all a,b∈Λa, b \in \Lambda (where δab\delta_{ab} is the Kronecker delta), and every a≠0a \ne 0 admits b∈MNb \in M_N with b2=1Nb^2 = 1_N and bea=−eabb e_a = -e_a b.✦

An element aa of a unital C*C^*-algebra AA is elementwise Cuntz subequivalent to b∈Ab \in A, written a≾ba \precsim b, if aa lies in the norm closure of {x*bx:x∈A}\{x^* b x : x \in A\}.✦ For an element aa of a unital C*C^*-algebra AA and δ∈ℝ\delta \in \mathbb{R}, the positive cut-down (a−δ)+(a - \delta)_+ is the element of AA given by the positive part (a−δ1A)+(a - \delta 1_A)^+ of a−δ1Aa - \delta 1_A.✦

Weighted Matrix Units and Cutoff Rescaling

Let AA be a unital C*C^*-algebra and let ι\iota be a finite index set. A family of weighted matrix units W=(h,(xij)i,j∈ι)W = (h, (x_{ij})_{i,j \in \iota}) indexed by ι\iota in AA consists of a positive contraction h∈Ah \in A (h≥0h \ge 0, ∥h∥≤1\|h\| \le 1) and elements xij∈Ax_{ij} \in A (i,j∈ιi, j \in \iota) satisfying xij*=xji,xijxkl=δjkhxil,∑i∈ιxii=hx_{ij}^* = x_{ji}, \qquad x_{ij} x_{kl} = \delta_{jk}\, h x_{il}, \qquad \sum_{i \in \iota} x_{ii} = h(2)✦ for all i,j,k,l∈ιi, j, k, l \in \iota. A family of weighted matrix units W=(h,(xij)i,j∈ι)W = (h, (x_{ij})_{i,j \in \iota}) in a unital C*C^*-algebra AA has size NN when ι={0,…,N−1}\iota = \{0, \dots, N-1\}.

Given a family (si)i∈ι(s_i)_{i \in \iota} in a unital C*C^*-algebra AA indexed by a finite set ι\iota and an element c∈Ac \in A satisfying sisj*=δijcs_i s_j^* = \delta_{ij} c for all i,j∈ιi, j \in \iota and ∥∑i∈ιsi*si∥≤1\|\sum_{i \in \iota} s_i^* s_i\| \le 1, the Gram construction is the family of weighted matrix units (h,(xij)i,j∈ι)(h, (x_{ij})_{i,j \in \iota}) in AA defined by h=∑i∈ιsi*si,xij=si*sj(i,j∈ι).h = \sum_{i \in \iota} s_i^* s_i, \qquad x_{ij} = s_i^* s_j \qquad (i, j \in \iota).(3)✦

For an element ww of a unital C*C^*-algebra AA satisfying w2=0w^2 = 0 and ∥w∥≤1\|w\| \le 1, the square-zero construction is the family of weighted matrix units Ww=(hw,(xij(w))i,j∈{0,1})W_w = \bigl(h_w, (x^{(w)}_{ij})_{i,j \in \{0,1\}}\bigr) of size 22 in AA defined by hw=(w*w)1/2+(ww*)1/2,(xij(w))i,j∈{0,1}=((w*w)1/2w*w(ww*)1/2).h_w = (w^* w)^{1/2} + (w w^*)^{1/2}, \qquad \bigl(x^{(w)}_{ij}\bigr)_{i,j \in \{0,1\}} = \begin{pmatrix} (w^* w)^{1/2} & w^* \\ w & (w w^*)^{1/2} \end{pmatrix}.(4)✦

For weighted matrix units W=(h,(xij)i,j∈ι)W = (h, (x_{ij})_{i,j \in \iota}) in AA, a self-adjoint element r∈Ar \in A commuting with every xijx_{ij}, and l=rhl = r h with l≥0l \ge 0 and ∥l∥≤1\|l\| \le 1, the rescaling of WW by rr and ll is the family of weighted matrix units (l,(rxij)i,j∈ι)(l, (r x_{ij})_{i,j \in \iota}) in AA.✦ The continuous functions j,f:ℝ→ℝj, f : \mathbb{R} \to \mathbb{R} are defined by j(t)=max⁡(t,12)−1,f(t)=tj(t).j(t) = \max(t, \tfrac{1}{2})^{-1}, \qquad f(t) = t\, j(t).(5) For weighted matrix units W=(h,(xij)0≤i,j<N)W = (h, (x_{ij})_{0 \le i,j < N}) of size NN in AA, the cutoff rescaling W̃=(h̃,(x̃ij)0≤i,j<N)\widetilde{W} = (\widetilde{h}, (\widetilde{x}_{ij})_{0 \le i,j < N}) is the rescaling of WW by r=j(h)r = j(h) and l=f(h)l = f(h), formed via continuous functional calculus from (5).✦

The Claim

Theorem 1. Let N≥2N \ge 2 be a natural number whose matrix algebra MNM_N admits a Pauli matrix system, let AA be a unital C*C^*-algebra, and let W=(h,(xij)0≤i,j<N)W = (h, (x_{ij})_{0 \le i,j < N}) be a family of weighted matrix units of size NN in AA satisfying 1A−h≾(x00−12)+1_A - h \precsim (x_{00} - \tfrac{1}{2})_+. Then there exists a unital **-homomorphism Φ:I(N,N+1)→A,\Phi : I(N, N+1) \to A,(6)✦ and for the cutoff rescaling W̃=(h̃,(x̃ij)0≤i,j<N)\widetilde{W} = (\widetilde{h}, (\widetilde{x}_{ij})_{0 \le i,j < N}) of WW there exists v∈Av \in A such that v*v=1A−h̃,x̃00v=v.v^* v = 1_A - \widetilde{h}, \qquad \widetilde{x}_{00} v = v.(7)✦

In the paper, the result reads: if BB is a unital C*C^*-algebra, a1,a2∈Ba_1, a_2 \in B are orthogonal positive contractions with a1=v*va_1 = v^* v and a2=vv*a_2 = v v^* for some v∈Bv \in B, and 1B−a1−a2≾(a1−ε)+1_B - a_1 - a_2 \precsim (a_1 - \varepsilon)_+ for some ε>0\varepsilon > 0, then there exists a unital **-homomorphism I(2,3)→BI(2,3) \to B. The theorem generalizes I(2,3)I(2,3) to I(N,N+1)I(N, N+1) for arbitrary N≥2N \ge 2 whose matrix algebra MNM_N admits a Pauli matrix system, takes weighted matrix units directly (supplied for N=2N = 2 by (3) or (4)), fixes ε=1/2\varepsilon = 1/2, and uses elementwise Cuntz subequivalence.

What Is Checked

The claims (6) and (7) are stated over a natural number N≥2N \ge 2 whose matrix algebra MNM_N admits a Pauli matrix system, the dimension-drop algebra (1), a unital C*C^*-algebra AA, elementwise Cuntz subequivalence ≾\precsim, a family W=(h,(xij)0≤i,j<N)W = (h, (x_{ij})_{0 \le i,j < N}) of weighted matrix units (2) of size NN in AA, the positive cut-down (x00−12)+(x_{00} - \tfrac{1}{2})_+ with 1A−h≾(x00−12)+1_A - h \precsim (x_{00} - \tfrac{1}{2})_+, and the cutoff rescaling W̃\widetilde{W} of WW. The constructions (3) and (4) are stated over a finite index set ι\iota, a family (si)i∈ι(s_i)_{i \in \iota} in AA, and c∈Ac \in A satisfying sisj*=δijcs_i s_j^* = \delta_{ij} c and ∥∑i∈ιsi*si∥≤1\|\sum_{i \in \iota} s_i^* s_i\| \le 1, and over w∈Aw \in A satisfying w2=0w^2 = 0 and ∥w∥≤1\|w\| \le 1.

For any w∈Aw \in A, ww*w=ww*ww \sqrt{w^* w} = \sqrt{w w^*}\, w, and when w2=0w^2 = 0, w*ww=0\sqrt{w^* w}\, w = 0, www*=0w \sqrt{w w^*} = 0✦, and w*www*=0\sqrt{w^* w} \sqrt{w w^*} = 0.✦ For weighted matrix units W=(h,(xij)i,j∈ι)W = (h, (x_{ij})_{i,j \in \iota}) in AA, the entries induce a complex-linear map from the algebra of ι×ι\iota \times \iota complex matrices to AA sending (mij)(m_{ij}) to ∑i,j∈ιmijxij\sum_{i,j \in \iota} m_{ij} x_{ij}✦, and 1A−h1_A - h is self-adjoint✦, yielding the non-unital **-subalgebra (1A−h)A(1A−h)¯\overline{(1_A - h)A(1_A - h)} of AA. When WW has size NN, the cone Csc([0,1],MN)={f∈C([0,1],MN):f(0)∈ℂ1N}C_{\mathrm{sc}}([0,1], M_N) = \{f \in C([0,1], M_N) : f(0) \in \mathbb{C} 1_N\} is **-isomorphic to the unitization of N×NN \times N matrices over {f∈C([0,1],ℂ):f(0)=0}\{f \in C([0,1], \mathbb{C}) : f(0) = 0\}✦, and WW induces a unital **-homomorphism Csc([0,1],MN)→AC_{\mathrm{sc}}([0,1], M_N) \to A✦ restricting on {f∈Csc([0,1],MN):f(1)=0}\{f \in C_{\mathrm{sc}}([0,1], M_N) : f(1) = 0\} to a **-homomorphism into (1A−h)A(1A−h)¯\overline{(1_A - h)A(1_A - h)}.✦ Any pair of weighted matrix unit families in AA of sizes NN and N+1N+1 whose entries commute elementwise and whose weights sum to 1A1_A induces a unital **-homomorphism I(N,N+1)→AI(N, N+1) \to A.✦