OctoInformBeta

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

Lemma 9.3: Under the continuum hypothesis, the adjoint is represented in the real norming completion and the operator has norm at most 1/101/10

Path cosets in the perturbation image.

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

The lemma concerns a perturbation of the quotient of a dual by the closed span of its coordinate functionals. The relevant vectors in this quotient are cosets of bounded functionals obtained from infinite paths.

Let Γ=ω1\Gamma=\omega_1, the first uncountable ordinal.✦ Fix a bijection θ:Γ→ℕ×Γ\theta:\Gamma\to\mathbb N\times\Gamma.✦ Its fibers CkC_k form the countable unbounded partition into colors; write c(a)c(a) for the first coordinate of θ(a)\theta(a) and ιk(a)=θ−1(k,a)\iota_k(a)=\theta^{-1}(k,a) for the embedding into CkC_k. A rational array on Γ\Gamma is a rational-valued function of finite support.✦ Write δa\delta_a for the array with coefficient one at aa and zero elsewhere.

The operation families are the pure families fkf_k, for k∈ℕk\in\mathbb N, and one mixed family fmf_{\mathrm m}.✦ Their parameters, family indices and source colors are specified in (1). fsfifcffm201fkk+2k+1k+1qf=sf+1sf,qf′=sf+1.\begin{array}{c|ccc} f&s_f&i_f&c_f\\ \hline f_{\mathrm m}&2&0&1\\ f_k&k+2&k+1&k+1 \end{array} \qquad q_f=\frac{s_f+1}{s_f},\qquad q'_f=s_f+1.(1) The primal exponent is sfs_f, the path exponent is qfq_f, and the source exponent is qf′q'_f. Two arrays have successive supports for ff when every coordinate in the first support precedes every coordinate in the second; for the mixed family, the corresponding colors must also strictly increase.✦ For a natural parameter ss, the Hölder exponent qsq_s is (s+1)/s(s+1)/s when s>0s>0, and q0=0q_0=0.

For a natural parameter ss, the depth sequence is dn=n+11d_n=n+11, with indices starting at zero.✦ Define the exponent scale ts,nt_{s,n} and the auxiliary sequence bs,nb_{s,n} by the recursion in (2); the pair (ts,n,bs,n)(t_{s,n},b_{s,n}) is the recursive parameter state. ts,0=d02+2d0+2,bs,0=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,Ls,n=2s(s+1)ts,n,ms,n=2(s−1)(s+1)ts,n+dn.\begin{aligned} t_{s,0}&=d_0^2+2d_0+2,& b_{s,0}&=0,\\ b_{s,n+1}&=b_{s,n}+2\cdot 2^{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,\\ L_{s,n}&=2^{s(s+1)t_{s,n}},& m_{s,n}&=2^{(s-1)(s+1)t_{s,n}+d_n}. \end{aligned}(2) Here s−1s-1 is truncated at zero. At positive weight jj, the family block length Lf(j)L_f(j) and normalization factor mf(j)m_f(j) are Lsf,j−1L_{s_f,j-1} and msf,j−1m_{s_f,j-1}. For a natural parameter ss and a zero-based index nn, the block length Ls,nL_{s,n} is given by (2). For a natural parameter ss and a zero-based index nn, the normalization factor ms,nm_{s,n} is given by (2).

A Type I expression for ff has positive weight jj and nonzero, successive rational children x0,…,xℓ−1x_0,\ldots,x_{\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=0ℓ−1xh.u=\frac{1}{m_f(j)}\sum_{h=0}^{\ell-1}x_h.(3)✦ For the pure family fkf_k, stage zero consists of zero and the arrays ±δa\pm\delta_a with a∈Cka\in C_k.✦ Each succeeding norming stage retains the previous stage and adjoins admitted Type I values and admitted values formed from finite paths, called Type II values: all children of a Type I expression, and all children of every original path piece in a Type II expression, must belong to the previous stage.✦ The pure norming set KkK_k is the union of these stages.✦ The mixed family starts from ⋃kKk\bigcup_k K_k; its stage union is the full norming set KK.✦ For each family ff, write KfK_f for its corresponding stage union.✦

The bounded function space ℓ∞(K)\ell_\infty(K) carries the supremum norm. For a∈Γa\in \Gamma, the coordinate evaluation vector eae_a in ℓ∞(K)\ell_\infty(K) is given by ea(u)=u(a)e_a(u)=u(a), for u∈Ku\in K. The finite evaluation map sends a finitely supported real vector xx to the linear combination of these vectors in (4); its range closure is the real norming completion EE. xˆ(u)=∑a∈Γx(a)u(a),E={xˆ:x∈c00(Γ,ℝ)}¯⊆ℓ∞(K).\widehat x(u)=\sum_{a\in\Gamma}x(a)u(a),\qquad E=\overline{\{\widehat x:x\in c_{00}(\Gamma,\mathbb R)\}} \subseteq\ell_\infty(K).(4) The finite support inclusion into EE is the same map with its target restricted to EE.✦ For u∈Ku\in K, the norming evaluation functional is ϕu(x)=x(u)\phi_u(x)=x(u) on EE✦; the coordinate functional at aa is ea*=ϕδae_a^*=\phi_{\delta_a}.✦ The coordinate predual and dual quotient are defined in (5), with QQ the quotient map and WW carrying the quotient norm. X0=span⁡ℝ{ea*:a∈Γ}¯⊆E*,W=E*/X0,Q:E*→W.X_0=\overline{\operatorname{span}_{\mathbb R}\{e_a^*:a\in\Gamma\}} \subseteq E^*,\qquad W=E^*/X_0,\qquad Q:E^*\longrightarrow W.(5)

Paths and Their Functionals

A predecessor closure family ℋ\mathcal H at β<ω1\beta<\omega_1 is a family of finite sets ℋn(a)\mathcal H_n(a) indexed by a<βa<\beta and zero-based indices nn.✦ A predecessor closure family at β<ω1\beta<\omega_1 is coherent when its finite sets at each a<βa<\beta contain aa, lie in [0,a][0,a], increase with the zero-based index nn and exhaust [0,a][0,a], and restriction to any member bb gives the set at bb at the same index.✦ A coherent endpoint extension of a predecessor closure family at β<ω1\beta<\omega_1 is a family of finite sets at β\beta that contain β\beta, lie in [0,β][0,\beta], increase with the zero-based index nn, exhaust [0,β][0,\beta] and restrict to the predecessor family at every member below β\beta.✦ The endpoint construction extends the families recursively along ω1\omega_1 to the coherent finite closure system Hn(β)H_n(\beta).

The positive-indexed finite closures Fp(β)F_p(\beta) and their entry rank ρ\rho are defined by (6). Positive closure index pp uses underlying index p−1p-1; index zero also uses H0(β)H_0(\beta). Fp(β)=Hmax⁡{p−1,0}(β),ρ(a,b)={min⁡{p≥1:a∈Fp(b)},a≤b,0,a>b.F_p(\beta)=H_{\max\{p-1,0\}}(\beta),\qquad \rho(a,b)= \begin{cases} \min\{p\geq1:a\in F_p(b)\},&a\leq b,\\ 0,&a>b. \end{cases}(6)

Raw infinite path data RR consist of nonzero rational arrays unu_n, declared weights jnj_n and metadata pnp_n, indexed from zero.✦ Their prefix support DnD_n, endpoint βn\beta_n and prefix closure AnA_n are given by (7). Dn=⋃i=0nsupp⁡ui,βn=max⁡Dn,An=Fpn(βn).D_n=\bigcup_{i=0}^{n}\operatorname{supp}u_i,\qquad \beta_n=\max D_n,\qquad A_n=F_{p_n}(\beta_n).(7) For a finite coordinate set AA, the order position pos⁡A(a)\operatorname{pos}_A(a) of aa is the number of members of AA below aa. The positional rational array code of a rational array uu on Γ\Gamma relative to a finite coordinate set AA is the set of pairs (pos⁡A(a),u(a))(\operatorname{pos}_A(a),u(a)) for a∈Aa\in A, retaining zero coefficients.✦ A finite path code is a finite list of triples consisting of a weight, metadata and such a finite set of pairs.✦ The prefix code cn(R)c_n(R) lists these triples for i=0,…,ni=0,\ldots,n, coding every uiu_i relative to the common closure AnA_n.✦ The coding map σ\sigma pairs the natural-number code of a finite path code with one plus its maximum metadata, taken as zero for the empty code.✦ It is injective, and its value exceeds every metadata entry.

Raw infinite path data RR are admissible when the supports of unu_n are successive in ordinal order, the positive metadata pnp_n strictly increase, and the requirements in (8) hold. 1≤jn≤pn,ρ(a,b)≤pn(a,b∈Dn,a≤b),jn+1=σ(cn(R)).1\leq j_n\leq p_n,\qquad \rho(a,b)\leq p_n\quad(a,b\in D_n,\ a\leq b),\qquad j_{n+1}=\sigma(c_n(R)).(8) A valid prefix of raw infinite path data is a prefix of positive length ℓ\ell satisfying the admissibility conditions in (8), ordinal succession of supports and strict increase of positive metadata only at indices below ℓ\ell, with the coding rule required when n+1<ℓn+1<\ell. A finite path for ff is a valid prefix of raw infinite path data together with Type I expressions whose values and weights are its original uncropped arrays and declared weights; these arrays must also be successive for ff.✦ Its finite path sum is ∑i<ℓui\sum_{i<\ell}u_i.✦

An infinite path for ff consists of admissible raw data and Type I expressions at every index whose values and weights are the original uncropped arrays and declared weights, whose arrays are successive for ff and whose children all belong to KfK_f.✦ Its finite prefix through nn retains the raw data and the first n+1n+1 expressions, so index zero includes one piece.✦ Write 𝒫f\mathcal P_f for the set of raw data occurring in such infinite paths, and choose an infinite representative P̃\widetilde P for every P∈𝒫fP\in\mathcal P_f.

For an infinite path TT, the path partial sum at nn is the norming evaluation functional associated with the finite sum of its pieces through nn.✦ The bounded path functional FT∈E*F_T\in E^* is the chosen functional whose value at each x∈Ex\in E is the limit of these partial sums evaluated at xx.✦ The path coset associated with a raw path index is defined in (9). πP=QFP̃∈W(P∈𝒫f).\pi_P=QF_{\widetilde P}\in W \qquad(P\in\mathcal P_f).(9)✦

The Perturbation Operator

Assume the continuum hypothesis (CH), |ℝ|=ℵ1|\mathbb R|=\aleph_1. For each ff, choose the base path labelling bf:𝒫f↪Γb_f:\mathcal P_f\hookrightarrow\Gamma.✦ The colored path labelling λf\lambda_f is defined by (10), using the embeddings into the family index color and then into the source color. λf(P)=ιcf(ιif(bf(P))).\lambda_f(P)=\iota_{c_f}\bigl(\iota_{i_f}(b_f(P))\bigr).(10)✦

Finite color synthesis and finite path synthesis send finitely supported real coefficients on 𝒫f\mathcal P_f to the sums in (11). Their continuous linear extensions are respectively Tf:ℓqf′(𝒫f)→ET_f:\ell_{q'_f}(\mathcal P_f)\to E and Sf:ℓqf(𝒫f)→WS_f:\ell_{q_f}(\mathcal P_f)\to W. Tfc=∑P∈𝒫fc(P)eλf(P),Sfc=∑P∈𝒫fc(P)πP(c∈c00(𝒫f,ℝ)).T_fc=\sum_{P\in\mathcal P_f}c(P)e_{\lambda_f(P)},\qquad S_fc=\sum_{P\in\mathcal P_f}c(P)\pi_P \qquad(c\in c_{00}(\mathcal P_f,\mathbb R)).(11) These extensions use the finitely supported sequence embedding, which takes each coefficient family to the same coordinate function in its ℓp\ell_p space.✦ For conjugate exponents p,p′p,p' and a continuous linear map T:ℓp(I)→ET:\ell_p(I)\to E, let eie_i be the coordinate unit vectors in ℓp(I)\ell_p(I); for a functional g∈E*g\in E^*, the dual analysis operator is ATg=(g(Tei))i∈I∈ℓp′(I)A_Tg=(g(Te_i))_{i\in I}\in\ell_{p'}(I), and the dual coefficients of a continuous real linear functional hh on ℓp(I)\ell_p(I) form the vector (h(ei))i∈I∈ℓp′(I)(h(e_i))_{i\in I}\in\ell_{p'}(I). The color analysis operator is Af=ATfA_f=A_{T_f}✦, and the family perturbation operator is Df=SfAfD_f=S_fA_f.✦

The enumeration 𝔣0=fm\mathfrak f_0=f_{\mathrm m}, 𝔣k+1=fk\mathfrak f_{k+1}=f_k lists the mixed family followed by the pure family of each color kk.✦ With this enumeration, the perturbation weights and the norm-convergent operator sum defining B:E*→WB:E^*\to W are (12). τn=132(12)n,B=∑n=0∞τnD𝔣n.\tau_n=\frac1{32}\left(\frac12\right)^n,\qquad B=\sum_{n=0}^{\infty}\tau_nD_{\mathfrak f_n}.(12) The canonical embedding JE:E→E**J_E:E\to E^{**} is given by (JEv)(g)=g(v)(J_Ev)(g)=g(v), and B*w=w∘BB^*w=w\circ B for w∈W*w\in W^*.

The Claim

Theorem 1. Assume CH. For every pure or mixed family ff and every P∈𝒫fP\in\mathcal P_f, πP∈B(X0).\pi_P\in B(X_0).(13)✦ Let ZZ be a closed real linear subspace of WW containing every πP\pi_P with P∈𝒫fP\in\mathcal P_f. For every a∈ℓqf(𝒫f)a\in\ell_{q_f}(\mathcal P_f), Sfa∈Z.S_fa\in Z.(14)✦ The range of BB satisfies B(E*)⊆B(X0)¯.B(E^*)\subseteq\overline{B(X_0)}.(15)✦ For every w∈W*w\in W^*, there is v∈Ev\in E such that B*w=JEv.B^*w=J_Ev.(16)✦ The operator norm satisfies ∥B∥≤110.\|B\|\leq\frac1{10}.(17)✦

In the paper, with VV denoting the closed linear span of all path cosets, the result reads: B(X0)¯=span⁡ℝ{πP:P∈𝒫f,f a family}¯=V\overline{B(X_0)} =\overline{\operatorname{span}_{\mathbb R} \{\pi_P:P\in\mathcal P_f,\ f\text{ a family}\}}=V, B(E*)⊆VB(E^*)\subseteq V and B*(W*)⊆JE(E)B^*(W^*)\subseteq J_E(E).

What Is Checked

The assertions (13), (14), (15), (16) and (17) are stated over the real norming completion, its coordinate predual and dual quotient, with the CH labellings of all pure and mixed path families; (14) allows any closed linear subspace ZZ with the stated family of path cosets.

The full norming set contains every δa\delta_a✦, and every coefficient of every array in KK has absolute value at most one.✦ Every raw prefix support DnD_n is nonempty.✦ The finite sum of each infinite path prefix belongs to KK✦, and the pointwise limit defining its path functional exists with norm at most one.✦ For every family, the source and path exponents in (1) are Hölder conjugates.✦ The general dual analysis operator satisfies ∥AT∥≤∥T∥\|A_T\|\leq\|T\|, and the quotient WW is complete.✦