OctoInformBeta

OpenAI preprint, September 2026 · §9. Changing the predual

Lemma 9.2: One operator for each pure or mixed family, with prescribed values on coordinate vectors

Path cosets admit contractive sequence synthesis.

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

Infinite path functionals determine cosets modulo the closed span of coordinate functionals. The synthesis problem asks which coefficient sequences can be assigned linear combinations of these cosets in the dual quotient.

Let Γ=ω1\Gamma=\omega_1, the first uncountable ordinal.✦ Fix a bijection Γ→ℕ×Γ\Gamma\longrightarrow\mathbb N\times\Gamma, and let c(a)c(a) be its first coordinate at aa. The color classes Ck={a:c(a)=k}C_k=\{a:c(a)=k\} are unbounded and have cardinality ℵ1\aleph_1. A rational array uu on Γ\Gamma is a finitely supported function Γ→ℚ\Gamma\to\mathbb Q.✦

There is a pure operation family fkf_k for each color kk, and a mixed family fmf_{\mathrm m}.✦ Their parameters are s(fk)=k+2s(f_k)=k+2 and s(fm)=2s(f_{\mathrm m})=2.✦ The exponent of a family ff is qf=(s(f)+1)/s(f)q_f=(s(f)+1)/s(f); its path sequence space uses this finite positive exponent.✦

Parameters and Norming Arrays

The depth sequence is dn=n+11d_n=n+11, with indices starting at zero.✦ For a natural parameter ss, the recursive parameter state is the pair (ts,n,bs,n)(t_{s,n},b_{s,n}) defined by (1). The exponent scale ts,nt_{s,n} for a natural parameter ss and a zero-based index nn is the first coordinate of the recursive parameter state. (ts,0,bs,0)=(d02+2d0+2,0),bs,n+1=bs,n+2⋅2s(s+1)ts,n,ts,n+1=ts,n+(bs,n+1+1)(dn+1+1)+1+dn+1+dn+12.\begin{aligned} (t_{s,0},b_{s,0})&=(d_0^2+2d_0+2,0),\\ b_{s,n+1}&=b_{s,n}+2\cdot2^{s(s+1)t_{s,n}},\\ t_{s,n+1}&=t_{s,n}+(b_{s,n+1}+1)(d_{n+1}+1) +1+d_{n+1}+d_{n+1}^2. \end{aligned}(1)✦ The block lengths and normalization factors are given by (2). A positive weight jj uses the underlying parameter index j−1j-1; weight zero also uses index zero. The general exponent function is qs=(s+1)/sq_s=(s+1)/s for s>0s>0, and q0=0q_0=0.✦ Ls,n=2s(s+1)ts,n,ms,n=2(s−1)(s+1)ts,n+dn,Lf(j)=Ls(f),max⁡{j−1,0},mf(j)=ms(f),max⁡{j−1,0}.\begin{aligned} L_{s,n}&=2^{s(s+1)t_{s,n}},& m_{s,n}&=2^{(s-1)(s+1)t_{s,n}+d_n},\\ L_f(j)&=L_{s(f),\max\{j-1,0\}},& m_f(j)&=m_{s(f),\max\{j-1,0\}}. \end{aligned}(2) Here subtraction in s−1s-1 is truncated at zero when s=0s=0.

Two arrays u,vu,v are successive for ff when every a∈supp⁡ua\in\operatorname{supp}u precedes every b∈supp⁡vb\in\operatorname{supp}v. For the mixed family this also requires c(a)<c(b)c(a)<c(b) for every such pair. A Type I expression AA in a family ff is a tuple consisting of a positive weight jj and nonzero successive children v0,…,vℓ−1v_0,\ldots,v_{\ell-1}, where 1≤ℓ≤Lf(j)1\leq\ell\leq L_f(j).✦ Its value is the normalized sum in (3). u=1mf(j)∑h<ℓvh.u=\frac{1}{m_f(j)}\sum_{h<\ell}v_h.(3)✦

For a set BB of rational arrays on Γ\Gamma and an operation family ff, the norming construction stages Kf,n(B)K_{f,n}(B) are sets of rational arrays defined recursively as follows. Stage zero is BB. The pure base of color kk is Bk={0}∪{δa,−δa:a∈Ck}B_k=\{0\}\cup\{\delta_a,-\delta_a:a\in C_k\}, where δa\delta_a is the unit rational array at aa.✦ The pure and full norming sets in (4) distinguish the pure constructions from the mixed one. Kk=⋃n≥0Kfk,n(Bk),Bm=⋃k≥0Kk,Kfk=Kk,Kfm=K=⋃n≥0Kfm,n(Bm).\begin{aligned} K_k&=\bigcup_{n\geq0}K_{f_k,n}(B_k),& B_{\mathrm m}&=\bigcup_{k\geq0}K_k,\\ K_{f_k}&=K_k,& K_{f_{\mathrm m}}=K&=\bigcup_{n\geq0}K_{f_{\mathrm m},n}(B_{\mathrm m}). \end{aligned}(4)

A norming set NN on Γ\Gamma is a set of rational arrays containing every δa\delta_a and satisfying |u(a)|≤1|u(a)|\leq1 for every member uu and coordinate aa.✦ The full set KK has these properties. Its bounded function space is ℓ∞(K)\ell_\infty(K) with the supremum norm.✦ The coordinate evaluation vector va∈ℓ∞(K)v_a\in\ell_\infty(K) is the function va(u)=u(a)v_a(u)=u(a).✦ Let c00(Γ)c_{00}(\Gamma) denote the real finitely supported functions on Γ\Gamma; the finite-support inclusion JJ and the real norming completion EE are defined in (5), for finitely supported x:Γ→ℝx:\Gamma\to\mathbb R. Jx=∑a∈Γx(a)va,E=J(c00(Γ))¯ℓ∞(K).Jx=\sum_{a\in\Gamma}x(a)v_a, \qquad E=\overline{J(c_{00}(\Gamma))}^{\,\ell_\infty(K)}.(5) For u∈Ku\in K, the norming evaluation functional uˆ\widehat u on EE is point evaluation, uˆ(z)=z(u)\widehat u(z)=z(u).✦ In particular, the coordinate functional at aa is ea*=δaˆe_a^*=\widehat{\delta_a}.✦ The coordinate predual and the dual quotient are the spaces in (6), with norm closure in the continuous real dual E*E^* and the quotient norm on WW. X0=span⁡ℝ{ea*:a∈Γ}¯E*,W=E*/X0.X_0=\overline{\operatorname{span}_{\mathbb R} \{e_a^*:a\in\Gamma\}}^{\,E^*}, \qquad W=E^*/X_0.(6)

Paths and Their Functionals

A predecessor closure family GG at β∈Γ\beta\in\Gamma is a family of finite sets Gn(a)⊆ΓG_n(a)\subseteq\Gamma, one for each a<βa<\beta and n≥0n\geq0.✦ A predecessor closure family GG at β∈Γ\beta\in\Gamma is coherent when, for each a<βa<\beta, the sets Gn(a)G_n(a) form an increasing sequence of finite subsets of [0,a][0,a] containing aa and exhausting [0,a][0,a], and Gn(b)=Gn(a)∩[0,b]G_n(b)=G_n(a)\cap[0,b] whenever b∈Gn(a)b\in G_n(a).✦ A coherent endpoint extension HH of a predecessor closure family GG at β∈Γ\beta\in\Gamma is an increasing sequence of finite sets Hn⊆[0,β]H_n\subseteq[0,\beta] containing β\beta and exhausting [0,β][0,\beta], with Hn∩[0,a]=Gn(a)H_n\cap[0,a]=G_n(a) whenever a<βa<\beta belongs to HnH_n.✦ The endpoint closure construction assigns a finite-set sequence to a predecessor family. Applied recursively to the already constructed families below each endpoint, it defines the coherent finite closure system Rn(β)R_n(\beta).✦

The positive-indexed finite closure is Fp(β)=Rmax⁡{p−1,0}(β)F_p(\beta)=R_{\max\{p-1,0\}}(\beta), so that closure index zero also selects the first system entry.✦ For a≤ba\leq b, the closure entry rank ρ(a,b)\rho(a,b) is the first positive pp for which a∈Fp(b)a\in F_p(b); it is zero when a>ba>b.✦

A raw infinite path rr consists of nonzero rational arrays uiu_i, declared natural weights jij_i, and natural metadata pip_i, all indexed from zero.✦ Its prefix support SnS_n, endpoint βn\beta_n, and prefix closure DnD_n are specified in (7). Sn=⋃i=0nsupp⁡ui,βn=max⁡Sn,Dn=Fpn(βn).S_n=\bigcup_{i=0}^{n}\operatorname{supp}u_i, \qquad \beta_n=\max S_n, \qquad D_n=F_{p_n}(\beta_n).(7) For a finite set D⊆ΓD\subseteq\Gamma, the order position of aa in DD is pos⁡D(a)=|{b∈D:b<a}|\operatorname{pos}_D(a)=|\{b\in D:b<a\}|.✦ The positional rational array code AD(u)A_D(u) of a rational array uu on Γ\Gamma relative to a finite set D⊆ΓD\subseteq\Gamma is the finite set given by AD(u)={(pos⁡D(a),u(a)):a∈D}A_D(u)=\{(\operatorname{pos}_D(a),u(a)):a\in D\}, including zero coefficients.✦ A finite path code cc is a finite list of triples recording a weight, metadata, and a positional rational array code.✦ The prefix code cn(r)c_n(r) is the list [(ji,pi,ADn(ui))]i=0n[(j_i,p_i,A_{D_n}(u_i))]_{i=0}^{n}, using the common closure DnD_n.✦ For a finite code cc, let M(c)M(c) be its largest metadata entry, with M(⌀)=0M(\varnothing)=0.✦ The coding map σ(c)\sigma(c) pairs the canonical natural-number code of cc with M(c)+1M(c)+1; it is injective and exceeds every metadata entry of cc.✦

The raw path rr is admissible when its pieces have successive ordinal supports, its metadata are strictly increasing and positive, and the conditions in (8) hold at every index nn. 1≤jn≤pn,ρ(a,b)≤pn(a,b∈Sn,a≤b),jn+1=σ(cn(r)).\begin{gathered} 1\leq j_n\leq p_n,\qquad \rho(a,b)\leq p_n\quad(a,b\in S_n,\ a\leq b),\\ j_{n+1}=\sigma(c_n(r)). \end{gathered}(8)✦ A valid prefix of length ℓ\ell of a raw infinite path rr is its initial segment of length ℓ\ell, with successive ordinal supports, strictly increasing positive metadata, and the conditions in (8) at indices below ℓ\ell, with the coding rule required only when n+1<ℓn+1<\ell; entries beyond the prefix are unrestricted.✦

A finite path PP in a family ff consists of raw data rr with pieces uiu_i and weights jij_i, a positive length ℓ\ell with valid prefix, and Type I expressions at positions i<ℓi<\ell whose values are uiu_i and whose weights are jij_i. Their values are successive for ff. The finite path sum of a finite path PP in a family ff, with length ℓ\ell and pieces uiu_i, is ∑i<ℓui\sum_{i<\ell}u_i.✦ An infinite path PP in a family ff consists of admissible raw data rr with pieces uiu_i and weights jij_i, and a Type I expression at every index whose value is uiu_i and whose weight is jij_i. Its values are successive for ff, and every child of every expression belongs to KfK_f. The finite prefix P|nP|_n retains the raw data and the first n+1n+1 original expressions.✦

The next norming construction stage retains the preceding stage and adds Type I values whose children lie there, together with finite path sums whose every child of every original uncropped piece lies there.✦

The path index set IfI_f consists of all raw infinite paths occurring in infinite paths of ff.✦ For each a∈Ifa\in I_f, choose an infinite path PaP_a with exactly those raw data, retaining the original rational pieces as well as their weights and metadata.✦ For an infinite path PP, let Φn(P)\Phi_n(P) be norming evaluation at its finite prefix sum.✦ The bounded path functional FP∈E*F_P\in E^* of an infinite path PP is the chosen pointwise limit in (9). Φn(P)=∑i=0nuiˆ,FP(x)=limn→∞Φn(P)(x)(x∈E).\Phi_n(P)=\widehat{\sum_{i=0}^{n}u_i}, \qquad F_P(x)=\lim_{n\to\infty}\Phi_n(P)(x) \quad(x\in E).(9)✦ The path coset zaz_a indexed by a∈Ifa\in I_f is given by za=FPa+X0∈Wz_a=F_{P_a}+X_0\in W.✦

For a finite exponent qfq_f, the real sequence space ℓqf(If)\ell_{q_f}(I_f) consists of families λ\lambda with ∑a∈If|λa|qf<∞\sum_{a\in I_f}|\lambda_a|^{q_f}<\infty, with its usual norm. The finitely supported sequence embedding sends each member of c00(If)c_{00}(I_f), the real vector space of finitely supported families on IfI_f, to the same coordinate family in this space.✦ Finite path synthesis is the real-linear map Sf:c00(If)→WS_f:c_{00}(I_f)\to W defined by (10). Sfλ=∑a∈Ifλaza.S_f\lambda=\sum_{a\in I_f}\lambda_a z_a.(10)✦

The Claim

Theorem 1. Let ff be a pure or mixed operation family. Finite path synthesis extends along the finitely supported sequence embedding to a continuous real-linear map Tf:ℓqf(If)→WT_f:\ell_{q_f}(I_f)\to W.✦ It satisfies (11) and (12), where a∈Ifa\in I_f, t∈ℝt\in\mathbb R, and eae_a is the unit coordinate vector of ℓqf(If)\ell_{q_f}(I_f). ∥Tf∥≤1.\|T_f\|\leq1.(11)✦ Tf(tea)=tza.T_f(t e_a)=t z_a.(12)✦

What Is Checked

The statement is over the concrete norming completion EE, its coordinate predual X0X_0, and the path indices of each pure or mixed family. Every raw prefix support is nonempty.✦ For an infinite path, each finite prefix sum belongs to KK✦, and the associated path functional has norm at most one. The quotient WW is complete.✦