OctoInform Beta

OpenAI preprint, September 2026 · §3. Reduced HNN adjunctions · §3

Lemma 3.5 [Limits of adjunctions]: The algebra generated by AA and two letters is H(A)H(A) when a faithful state matches the traces over a dense directed family

Limits of reduced Higman-Neumann-Neumann extensions are again reduced extensions.

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 AA be a C*C^*-algebra with a faithful trace τ\tau and two decreasing sequences of projections, each peaking to a state, the only state that takes the value one at every term of the sequence, and let H(A)H(A) be its reduced Higman-Neumann-Neumann (HNN) extension: the C*C^*-algebra generated by AA and a stable unitary vv that conjugates the first sequence onto the second, with a trace extending τ\tau. In a transfinite construction AA is reached as the closure of a union of smaller algebras carrying the same sequences, and their extensions are built first, inside one C*C^*-algebra CC and with one stable unitary. The question is whether the C*C^*-algebra generated in CC by AA and that unitary is then H(A)H(A), with the same base, stable unitary and trace.

The library answers it for a directed family of algebras mapping into AA, and, inside a single algebra, for an increasing chain of closed subalgebras indexed by the ordinals.

Bridge Data

A state of a C*C^*-algebra AA is a bounded linear functional on AA of norm one that is real and nonnegative on every a*aa^*a.✦ A sequence (pn)n≥0(p_n)_{n \ge 0} in AA peaks to a state ff of AA when the pnp_n are decreasing projections, f(pn)=1f(p_n) = 1 for every nn, and every state gg of AA with g(pn)=1g(p_n) = 1 for every nn equals ff.✦ The atoms of such a sequence are the differences dn=pn−pn+1d_n = p_n - p_{n+1}.✦

Bridge data on AA consist of a positive functional τ\tau on AA that is tracial, τ(ab)=τ(ba)\tau(ab) = \tau(ba), normalized, τ(1)=1\tau(1) = 1, and faithful, τ(a*a)=0\tau(a^*a) = 0 only for a=0a = 0, and of two sequences p1p^1, p2p^2 in AA, the peaks, peaking to states f1f_1, f2f_2, such that for i=1,2i = 1, 2 and every nn p0i=1,τ(dni)>0,limm→∞τ(pmi)=0,τ(pni)=τ(pn1).p^i_0 = 1, \qquad \tau(d^i_n) > 0, \qquad \lim_{m \to \infty} \tau(p^i_m) = 0, \qquad \tau(p^i_n) = \tau(p^1_n).(1)✦

Let cc be the C*C^*-algebra of convergent complex sequences, the continuous functions on ℕ∪{∞}\mathbb{N} \cup \{\infty\}, and write x∞x_\infty for the limit of xx in cc.✦ For i=1,2i = 1, 2 the peaks give a unital **-homomorphism δi:c→A\delta_i \colon c \to A and a linear map Ei:A→cE_i \colon A \to c of norm at most one, which sends aa to its sequence of corner states: δi(x)=x∞1+∑n(xn−x∞)dni,Ei(a)=(τ(dniadni)τ(dni))n≥0.\delta_i(x) = x_\infty 1 + \sum_n (x_n - x_\infty)\, d^i_n, \qquad E_i(a) = \Bigl(\frac{\tau(d^i_n a d^i_n)}{\tau(d^i_n)}\Bigr)_{n \ge 0}.(2)

The Reduced HNN Extension

The library builds the reduced HNN extension H(A)H(A) of bridge data on AA as a corner of an algebra of operators on a Fock space. Two matrix algebras enter, each with a faithful normalized trace and atoms, pairwise orthogonal projections of positive trace whose traces sum to one: M2(A)M_2(A), with the normalized matrix trace tr⁡(M)=12(τ(M11)+τ(M22))\operatorname{tr}(M) = \tfrac12\bigl(\tau(M_{11}) + \tau(M_{22})\bigr) and the atoms dnieiid^i_n e_{ii}, where the eije_{ij} are the matrix units, and M2(c)M_2(c), with the normalized matrix trace of τ∘δ1\tau \circ \delta_1.✦ Both contain c⊕cc \oplus c: in M2(A)M_2(A) as the matrices diag⁡(δ1(x1),δ2(x2))\operatorname{diag}(\delta_1(x_1), \delta_2(x_2)), with the expectation M↦(E1(M11),E2(M22))M \mapsto (E_1(M_{11}), E_2(M_{22})), and in M2(c)M_2(c) as the matrices diag⁡(x1,x2)\operatorname{diag}(x_1, x_2). The atoms of M2(c)M_2(c) are E1(dn1)eiiE_1(d^1_n)\, e_{ii} for i=1,2i = 1, 2 and n≥0n \ge 0. The two factors are amalgamated over c⊕cc \oplus c in the following sense. For x=(x1,x2)x = (x_1, x_2) in c⊕cc \oplus c and the atom ee of index (i,n)(i, n), the factor element of xx satisfies ex=(xi)nee\, x = (x_i)_n\, e. The expectations onto c⊕cc \oplus c are bounded, fix c⊕cc \oplus c and are linear over c⊕cc \oplus c from the left, and every element aa with zero expectation has zero trace in every atom corner, tr⁡(eae)=0\operatorname{tr}(e a e) = 0.

Call M2(A)M_2(A) and M2(c)M_2(c) the factors. The Fock space is ℓ2\ell^2 of the reduced paths: finite sequences of letters, consecutive letters from different factors, each letter ending at the atom where the next one begins.✦ A letter of a factor MM from an atom qlq_l to an atom qrq_r is a vector of a chosen Hilbert basis of the centered part of qlMqrq_l M q_r, the vectors of its closure in L2(M,tr⁡)L^2(M, \operatorname{tr}) orthogonal to qrq_r.✦ The factors act on the Fock space, M2(A)M_2(A) by πA\pi_A and M2(c)M_2(c) by πc\pi_c; the model algebra ℳ\mathcal{M} is the norm closure of the algebra they generate, and the model trace is the vector functional on ℳ\mathcal{M} of the vacuum vector (tr⁡(qr))r(\sqrt{\operatorname{tr}(q_r)})_r, carried by the empty paths at the atoms qrq_r of M2(A)M_2(A). A factor MM acts on the Fock space as follows. A reduced path that does not begin with a letter of MM is a tail tt starting at an atom qrq_r. The vacuum together with the letters of MM from any atom qlq_l to qrq_r index the chosen Hilbert basis of the closure of MqrMq_r in L2(M,tr⁡)L^2(M, \operatorname{tr}), and joining the vacuum or a letter to tt identifies this closure with a subspace of the Fock space. The Fock space is the direct sum of these subspaces over all tails tt, and MM acts on each of them by the Gelfand-Naimark-Segal (GNS) representation, that is, by left multiplication.

Then H(A)H(A) is the corner πA(e11)ℳπA(e11)\pi_A(e_{11})\, \mathcal{M}\, \pi_A(e_{11}).✦ It contains AA through a↦πA(ae11)a \mapsto \pi_A(a e_{11})✦, its stable letters are v1=v*v_1 = v^* and v2=vv_2 = v, where v=πA(e12)πc(e21),v = \pi_A(e_{12})\, \pi_c(e_{21}),(3)✦ and its trace τ̃\tilde\tau is twice the model trace on the corner.

Reduced Words and Presentations

An HNN word over AA is a finite sequence (i1,a1),…,(in,an)(i_1, a_1), \dots, (i_n, a_n), n≥1n \ge 1, of pairs of an index ik∈{1,2}i_k \in \{1, 2\} and a coefficient ak∈Aa_k \in A✦; its coefficients lie in a subset SS of AA when every aka_k lies in SS.✦ For a map ψ\psi from AA to an algebra CC and elements w1w_1, w2w_2 of CC, the word evaluates to wi1ψ(a1)wi2ψ(a2)⋯winψ(an).w_{i_1}\psi(a_1)\, w_{i_2}\psi(a_2) \cdots w_{i_n}\psi(a_n).(4)✦ For bridge data on AA, an HNN word over AA is reduced when Eik+1(ak)=0E_{i_{k+1}}(a_k) = 0 whenever ik≠ik+1i_k \ne i_{k+1}, that is, when every coefficient standing between two different indices has vanishing sequence of corner states for the second index.

For a closed **-subalgebra SS of AA containing all the peaks and an element zz of AA, a closed **-subalgebra TT of AA presents SS along zz when SS carries bridge data with the peaks p1p^1, p2p^2 and the trace τ|S\tau|_S, and there is a **-isomorphism e:H(S)→Te \colon H(S) \to T that is the identity on SS, sends v1v_1 and v2v_2 to zz and z*z^*, and satisfies τ∘e=τ̃S\tau \circ e = \tilde\tau_S, where τ̃S\tilde\tau_S is the trace of H(S)H(S).✦

The Claim

Theorem 1. Let AA carry bridge data with trace τ\tau and peaks p1p^1, p2p^2.

  1. Let (Bj)j∈J(B_j)_{j \in J} be a nonempty family of C*C^*-algebras with unital **-homomorphisms ιj:Bj→A\iota_j \colon B_j \to A whose ranges are directed under inclusion, each BjB_j carrying bridge data whose peaks ιj\iota_j maps to p1p^1, p2p^2 and whose trace is τ∘ιj\tau \circ \iota_j. Let CC be a C*C^*-algebra with a unital **-homomorphism ψ:A→C\psi \colon A \to C, elements w1w_1, w2w_2, and unital **-homomorphisms Fj:H(Bj)→CF_j \colon H(B_j) \to C that agree with ψ∘ιj\psi \circ \iota_j on BjB_j and send the stable letters of H(Bj)H(B_j) to w1w_1, w2w_2. Let gg be a bounded linear functional on CC with g∘Fj=τ̃jg \circ F_j = \tilde\tau_j, the trace of H(Bj)H(B_j), for every jj, and let A∪A_\cup be the union of the ranges of the ιj\iota_j. Then for every reduced HNN word (i1,a1),…,(in,an)(i_1, a_1), \dots, (i_n, a_n) with coefficients in A∪A_\cup and every a0∈A∪a_0 \in A_\cup, g(ψ(a0)wi1ψ(a1)⋯winψ(an))=0.g\bigl(\psi(a_0)\, w_{i_1}\psi(a_1) \cdots w_{i_n}\psi(a_n)\bigr) = 0.(5)✦

  2. If, in addition, A∪A_\cup is dense in AA, gg is a faithful state, and ψ(A)\psi(A), w1w_1, w2w_2 generate CC as a C*C^*-algebra, then there is a **-isomorphism e:H(A)→Ce \colon H(A) \to C with e(a)=ψ(a)(a∈A),e(v1)=w1,e(v2)=w2,g∘e=τ̃.e(a) = \psi(a) \ (a \in A), \qquad e(v_1) = w_1, \qquad e(v_2) = w_2, \qquad g \circ e = \tilde\tau.(6)✦

  3. Let z∈Az \in A, and let (Sγ)(S_\gamma) and (Rγ)(R_\gamma) be families of closed **-subalgebras of AA indexed by the ordinals, (Sγ)(S_\gamma) increasing and every SγS_\gamma containing all the peaks. Let β>0\beta > 0 be an ordinal such that SβS_\beta is the closure of ⋃γ<βSγ\bigcup_{\gamma < \beta} S_\gamma and RγR_\gamma presents SγS_\gamma along zz for every γ<β\gamma < \beta. Then SβS_\beta carries bridge data with the peaks p1p^1, p2p^2 and the trace τ|Sβ\tau|_{S_\beta}, and there is a **-isomorphism ee of H(Sβ)H(S_\beta) onto the C*C^*-algebra C*(Sβ,z)C^*(S_\beta, z) generated by SβS_\beta and zz with e(a)=a(a∈Sβ),e(v1)=z,e(v2)=z*,τ∘e=τ̃Sβ.e(a) = a \ (a \in S_\beta), \qquad e(v_1) = z, \qquad e(v_2) = z^*, \qquad \tau \circ e = \tilde\tau_{S_\beta}.(7)✦

In the paper, the result reads: for a nonzero limit ordinal λ\lambda and an increasing family (Oβ)β<λ(O_\beta)_{\beta < \lambda} of unital C*C^*-algebras with compatible faithful tracial states, all containing fixed D1D_1, D2D_2 with a fixed θ\theta and compatible trace-preserving expectations onto the DiD_i, after identifying the H(Oβ)H(O_\beta) with the same stable unitary vv, the expectations extend to O→DiO \to D_i for the closure OO of ⋃β<λOβ\bigcup_{\beta < \lambda} O_\beta, and the closure RR of ⋃β<λH(Oβ)\bigcup_{\beta < \lambda} H(O_\beta) is H(O)H(O) with its specified generators and trace.

What Is Checked

The claim is stated over bridge data on AA, on the BjB_j and on the SγS_\gamma; over their reduced HNN extensions, built as corners of model algebras, with the embedded base algebra, the stable letters v1v_1, v2v_2 and the trace; over reduced HNN words and their evaluation (4); and over presentations along zz inside AA.

About these constructions the library proves that the atoms of a peaking sequence are pairwise orthogonal projections, that the corner states τ(dnadn)/τ(dn)\tau(d_n a d_n)/\tau(d_n) converge to f(a)f(a) for every aa when τ\tau is a state positive on every atom✦, and that each δi\delta_i is injective.✦ For the matrix algebras it proves that the normalized matrix trace of a tracial, normalized and faithful functional is again tracial, normalized and faithful, that corresponding atoms of M2(A)M_2(A) and M2(c)M_2(c) have the same trace✦, and that πA(e11)\pi_A(e_{11}) is a projection✦ whose corner contains vv.✦ It proves as well that a positive functional ff satisfies |f(a)|≤f(1)∥a∥|f(a)| \le f(1)\, \|a\|✦ and is a state when f(1)=1f(1) = 1✦, that the closure of MqMq in L2(M,tr⁡)L^2(M, \operatorname{tr}) is invariant under the GNS representation for every projection qq✦, that splitting a reduced path at its first letter is inverse to joining the letter back, and that qr/tr⁡(qr)q_r/\sqrt{\operatorname{tr}(q_r)} together with the chosen bases of the centered parts of the qlMqrq_l M q_r is an orthonormal family.✦ It proves also that the atoms eve_v of cc for the trace τ∘δ1\tau \circ \delta_1, indexed by the points vv of ℕ\mathbb{N}, satisfy evx=xveve_v x = x_v e_v for every xx in cc, and that the C*C^*-algebra generated in CC by the range of a **-homomorphism from AA and by two elements of CC is closed.