OctoInformBeta

OpenAI preprint, September 2026 · §8. Path functionals on arbitrary subspaces

Lemma 8.1: Closed infinite-dimensional subspaces admit ordinal cuts and bounded lifts for pure or mixed families

Subspace reductions that preserve branch evaluations.

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 reduction problem for a closed infinite-dimensional subspace of the constructed space EE asks for an ordinal cut and a pure or mixed coordinate projection with a bounded lift into that subspace. Finite rational arrays and constructed branch paths supply the evaluations to be retained.

The coordinate set is Γ=ω1\Gamma=\omega_1, the first uncountable ordinal. Fix a bijection Γ→ℕ×Γ\Gamma\to\mathbb N\times\Gamma, and let c(a)c(a) be its first component at aa. The color classes Ck={a∈Γ:c(a)=k}C_k=\{a\in\Gamma:c(a)=k\} are unbounded and have cardinality ℵ1\aleph_1. A rational array dd on Γ\Gamma is a rational-valued function with finite support; write c00(Γ,ℚ)c_{00}(\Gamma,\mathbb Q) for these arrays and δa\delta_a for the unit array at aa.✦ The operation families are the pure families pure(k)\mathrm{pure}(k) and the mixed family mixed\mathrm{mixed}.✦

Norming Arrays and Their Completion

For a natural parameter ss, the depth dnd_n and recursive parameter state (tn,Bn)(t_n,B_n) are given by (1); the first component tnt_n is the exponent scale. All sequence indices start at zero. dn=n+11,(t0,B0)=(d02+2d0+2,0),Bn+1=Bn+2⋅2s(s+1)tn,tn+1=tn+(Bn+1+1)(dn+1+1)+1+dn+1+dn+12,Ls(n)=2s(s+1)tn,ms(n)=2(s−1)(s+1)tn+dn.\begin{aligned} d_n&=n+11, & (t_0,B_0)&=(d_0^2+2d_0+2,0),\\ B_{n+1}&=B_n+2\cdot2^{s(s+1)t_n},\\ t_{n+1}&=t_n+(B_{n+1}+1)(d_{n+1}+1)+1+d_{n+1}+d_{n+1}^2,\\ L_s(n)&=2^{s(s+1)t_n}, & m_s(n)&=2^{(s-1)(s+1)t_n+d_n}. \end{aligned}(1) Here subtraction in a natural parameter stops at zero. The exponent index of an operation family ff is the natural number sf=k+2s_f=k+2 for f=pure(k)f=\mathrm{pure}(k) and sf=2s_f=2 for f=mixedf=\mathrm{mixed}.✦ For an operation family ff and a positive weight jj, the block length and normalization are the natural numbers Lf(j)=Lsf(j−1)L_f(j)=L_{s_f}(j-1) and mf(j)=msf(j−1)m_f(j)=m_{s_f}(j-1): weight jj uses underlying parameter index j−1j-1. For the natural parameter ss, the block-length sequence is Ls(n)L_s(n) in (1), with nn starting at zero. The normalization factors for the natural parameter ss are ms(n)m_s(n) in (1), indexed from zero.

Rational arrays d,d′∈c00(Γ,ℚ)d,d^\prime\in c_{00}(\Gamma,\mathbb Q) have successive supports for an operation family ff when every coordinate in the support of dd precedes every coordinate in the support of d′d^\prime and, in the mixed family, the color of every coordinate in the first support precedes the color of every coordinate in the second.✦ A Type I expression ee for ff consists of a weight j≥1j\geq1 and nonzero successive children d0,…,dℓ−1d_0,\ldots,d_{\ell-1}, where 1≤ℓ≤Lf(j)1\leq\ell\leq L_f(j).✦ Its value is the normalized sum in (2). val⁡(e)=1mf(j)∑h<ℓdh.\operatorname{val}(e)=\frac{1}{m_f(j)}\sum_{h<\ell}d_h.(2)✦

For a base set BB of arrays, let Sf(B,n)S_f(B,n) be the norming construction stages, with Sf(B,0)=BS_f(B,0)=B. A successor stage retains the preceding stage and adds the values of Type I expressions whose children belong to it, together with the values of operations built from the finite paths described below, called Type II path operations, whose children in every original path piece belong to it. The pure base consists of zero and the signed coordinates of its color.✦ The pure norming sets KkK_k, family norming sets KfK_f, and full norming set KK are defined by (3). Bk={0}∪{δa,−δa:a∈Ck},Kk=⋃n≥0Spure(k)(Bk,n),Bpure(k)=Bk,Bmixed=⋃kKk,Kf=⋃n≥0Sf(Bf,n),K=Kmixed.\begin{aligned} B_k&=\{0\}\cup\{\delta_a,-\delta_a:a\in C_k\},\\ K_k&=\bigcup_{n\geq0}S_{\mathrm{pure}(k)}(B_k,n),\\ B_{\mathrm{pure}(k)}&=B_k, & B_{\mathrm{mixed}}&=\bigcup_k K_k,\\ K_f&=\bigcup_{n\geq0}S_f(B_f,n), & K&=K_{\mathrm{mixed}}. \end{aligned}(3) The base norming arrays of the mixed family form the union of the pure norming sets over all colors.

A norming set NN on Γ\Gamma is a set of rational arrays containing every δa\delta_a and having every coefficient bounded in absolute value by 11.✦ For a norming set NN on Γ\Gamma and a coordinate a∈Γa\in\Gamma, the coordinate evaluation vector xaNx_a^N is the bounded function d↦d(a)d\mapsto d(a) on NN .✦ The space ℓ∞(K)\ell_\infty(K) consists of bounded real-valued functions on the full norming set KK, equipped with the supremum norm; the finite-support inclusion ι\iota and its closed evaluation span give the real normed completion in (4). ι(v)(d)=∑av(a)d(a),E=ι(c00(Γ,ℝ))¯ℓ∞(K).\iota(v)(d)=\sum_a v(a)d(a),\qquad E=\overline{\iota(c_{00}(\Gamma,\mathbb R))}^{\,\ell_\infty(K)}.(4) For d∈Kd\in K, the norming evaluation functional is φd(x)=x(d)\varphi_d(x)=x(d).✦ The coordinate functional is ea*=φδae_a^*=\varphi_{\delta_a}, and the coordinate predual X0X_0 is the norm closure in E*E^* of their real linear span. The rational coordinate combination of an arbitrary rational array dd is ∑ad(a)ea*∈X0\sum_a d(a)e_a^*\in X_0; its evaluation on x∈Ex\in E is the rational linear map (5). ⟨d,x⟩=∑ad(a)ea*(x).\langle d,x\rangle=\sum_a d(a)e_a^*(x).(5)✦

Projections and Reductions

An ordinal and color rectangle A=(I,D)A=(I,D) consists of order-convex sets I⊆ΓI\subseteq\Gamma and D⊆ℕD\subseteq\mathbb N .✦ The support set of an ordinal and color rectangle A=(I,D)A=(I,D) is the coordinate set II for a pure family and I∩c−1(D)I\cap c^{-1}(D) for the mixed family.✦ An ordinal interval crop is an ordinal and color rectangle A=(I,D)A=(I,D) with D=ℕD=\mathbb N ; a color interval crop is such a rectangle with I=ΓI=\Gamma, and a single color crop is such a rectangle with D={k}D=\{k\} . A coordinate set S⊆ΓS\subseteq\Gamma is crop-stable for KK when restricting each member of KK to SS again gives a member of KK.✦

For a norming array dd in KK and a crop-stable rectangle A=(I,D)A=(I,D), the cropped normer is the rational array d|I∩c−1(D)d|_{I\cap c^{-1}(D)}.✦ The ambient crop on ℓ∞(K)\ell_\infty(K) and its restriction PAP_A to EE are given by (6). (PˆAx)(d)=x(d|I∩c−1(D)),PA=PˆA|E.(\widehat P_Ax)(d)=x\bigl(d|_{I\cap c^{-1}(D)}\bigr),\qquad P_A=\widehat P_A|_E.(6) An ordinal cut γ\gamma is an ordinal at most ω1\omega_1, with initial segment Iγ={a∈Γ:a<γ}I_\gamma=\{a\in\Gamma:a<\gamma\}.✦ Write P<γP_{<\gamma} for the initial ordinal projection and PCkP_{C_k} for the single color projection. The family projection Qγ,fQ_{\gamma,f} is defined by (7). Qγ,pure(k)=PCkP<γ,Qγ,mixed=P<γ.Q_{\gamma,\mathrm{pure}(k)}=P_{C_k}P_{<\gamma},\qquad Q_{\gamma,\mathrm{mixed}}=P_{<\gamma}.(7)✦ The initial ordinal crop at γ\gamma is the rectangle (Iγ,ℕ)(I_\gamma,\mathbb N).✦

A bounded linear operator T:Y→ZT:Y\to Z between normed spaces YY and ZZ is strictly singular when, for every closed infinite-dimensional subspace M⊆YM\subseteq Y and every c>0c>0, some x∈Mx\in M satisfies ∥Tx∥<c∥x∥\|Tx\|<c\|x\|.✦ A branch reduction R=(γ,f,H,U)R=(\gamma,f,H,U) of a linear subspace F⊆EF\subseteq E consists of an ordinal cut γ\gamma, an operation family ff, a closed infinite-dimensional subspace H⊆EH\subseteq E, and a bounded linear map U:H→FU:H\to F satisfying (8).✦ Denote the collection of these reductions by ℛ(F)\mathcal R(F). for every a<γ there is b∈Γ with a<b<γ,P<β|H is strictly singular(β<γ),f=mixed⇒PCk|H is strictly singular(k∈ℕ),Qγ,fUh=h(h∈H).\begin{aligned} &\text{for every }a<\gamma\text{ there is }b\in\Gamma \text{ with }a<b<\gamma,\\ &P_{<\beta}|_H\text{ is strictly singular}\quad(\beta<\gamma),\\ &f=\mathrm{mixed}\ \Longrightarrow\ P_{C_k}|_H\text{ is strictly singular}\quad(k\in\mathbb N),\\ &Q_{\gamma,f}Uh=h\quad(h\in H). \end{aligned}(8)

Block Approximations

An approximation state for a branch reduction R=(γ,f,H,U)R=(\gamma,f,H,U) is a pair of thresholds (β,l)(\beta,l) with β∈Γ\beta\in\Gamma, β<γ\beta<\gamma, and l∈ℕl\in\mathbb N.✦ At stage nn, a block approximation step consists of a unit vector h∈Hh\in H and a nonzero finitely supported real vector vv satisfying (9), together with a next threshold β′<γ\beta'<\gamma beyond every coordinate in supp⁡v\operatorname{supp}v. ∥h−ι(v)∥<εn,εn=1640002−n,supp⁡v⊆{Iγ∩[β,ω1)∩Ck,f=pure(k),Iγ∩[β,ω1)∩c−1([l,∞)),f=mixed.\begin{aligned} \|h-\iota(v)\|&<\varepsilon_n, &\varepsilon_n&=\frac{1}{64000}\,2^{-n},\\ \operatorname{supp}v&\subseteq \begin{cases} I_\gamma\cap[\beta,\omega_1)\cap C_k,&f=\mathrm{pure}(k),\\ I_\gamma\cap[\beta,\omega_1)\cap c^{-1}([l,\infty)),&f=\mathrm{mixed}. \end{cases} \end{aligned}(9)✦ The recursively chosen steps give a block sequence zR=(zn)z_R=(z_n) for ff satisfying the following admissibility conditions. An admissible block sequence z=(zn)z=(z_n) for an operation family ff is a sequence of nonzero finitely supported real vectors on Γ\Gamma with successive ordinal supports and, in the mixed family, successive colors; in the pure family of color kk, every support lies in CkC_k.✦ The ordinal and color block crop determined by γ,f,β,l\gamma,f,\beta,l is the rectangle with ordinal interval Iγ∩[β,ω1)I_\gamma\cap[\beta,\omega_1) and color interval {k}\{k\} for f=pure(k)f=\mathrm{pure}(k), or [l,∞)[l,\infty) for f=mixedf=\mathrm{mixed}.✦ The geometric approximation error at stage nn is εn=64000−12−n\varepsilon_n=64000^{-1}2^{-n}, with the stages indexed from zero.✦

For any admissible block sequence zz, its initial support Sz(m)S_z(m) and algebraic tail span Vz(a)V_z(a) are the finite union and real linear span in (10). The tail index aa starts at zero. Sz(m)=⋃n<msupp⁡zn,Vz(a)=span⁡ℝ{zn+a:n≥0}.S_z(m)=\bigcup_{n<m}\operatorname{supp}z_n,\qquad V_z(a)=\operatorname{span}_{\mathbb R}\{z_{n+a}:n\geq0\}.(10) The vector support crop of a nonzero finite vector vv is the rectangle whose ordinal interval runs from the minimum to the maximum of its support and whose color interval runs from the minimum to the maximum of the colors of its support.✦

For an admissible block sequence zz in an operation family ff, an exact pair (v,e)(v,e) at tail index aa and weight JJ consists of a nonzero vector v∈Vz(a)v\in V_z(a) and a Type I expression ee of weight JJ satisfying (11). Every child of ee belongs to KfK_f, its value is supported in the vector support crop for ff, and, when f=pure(k)f=\mathrm{pure}(k), the vector vv is supported on CkC_k. ∥ι(v)∥≤500,⟨val⁡(e),ι(v)⟩≥12.\|\iota(v)\|\leq500,\qquad \langle\operatorname{val}(e),\iota(v)\rangle\geq\tfrac12.(11) An exact-pair tree node TT for an admissible block sequence zz consists of a starting block index aa, a positive weight JJ, an exact pair (v,e)(v,e) at tail index aa and weight JJ, a finishing index m≥am\geq a with supp⁡v⊆Sz(m)\operatorname{supp}v\subseteq S_z(m), and an auxiliary index p≥Jp\geq J.✦

Paths and Branch Functionals

Let ℱβ(n)\mathcal F_\beta(n) be the coherent finite closure system on Γ\Gamma, indexed from zero. Its finite sets lie at or below their endpoint, contain that endpoint, increase with nn, and exhaust the segment at or below the endpoint. Coherence means that α∈ℱβ(n)\alpha\in\mathcal F_\beta(n) entails ℱα(n)=ℱβ(n)∩[0,α]\mathcal F_\alpha(n)=\mathcal F_\beta(n)\cap[0,\alpha] . A predecessor closure family 𝒞\mathcal C at an ordinal β∈Γ\beta\in\Gamma assigns a finite subset 𝒞α(n)\mathcal C_\alpha(n) of Γ\Gamma to every endpoint α<β\alpha<\beta and every zero-based natural index nn; a coherent endpoint extension 𝒞′\mathcal C^\prime is an extension to β\beta whose sets lie at or below β\beta, contain β\beta, increase with nn, exhaust the segment at or below β\beta, and agree with the predecessor sets at their members. The system is obtained recursively by these endpoint extensions. The positive closure convention and closure entry rank are specified in (12). Fp(β)=ℱβ(max⁡{p−1,0}),ρ(α,β)={min⁡{p≥1:α∈Fp(β)},α≤β,0,α>β.F_p(\beta)=\mathcal F_\beta(\max\{p-1,0\}),\qquad \rho(\alpha,\beta)= \begin{cases} \min\{p\geq1:\alpha\in F_p(\beta)\},&\alpha\leq\beta,\\ 0,&\alpha>\beta. \end{cases}(12)

A raw infinite path rr on Γ\Gamma is a sequence of nonzero rational pieces uiu_i, with natural weights jij_i and auxiliary indices pip_i , all indexed from zero.✦ For a raw infinite path rr on Γ\Gamma with pieces uiu_i and auxiliary indices pip_i , its prefix support is the finite coordinate set Sn=⋃i≤nsupp⁡uiS_n=\bigcup_{i\leq n}\operatorname{supp}u_i , its endpoint is the ordinal ηn=max⁡Sn\eta_n=\max S_n , and its closure is the finite coordinate set Dn=Fpn(ηn)D_n=F_{p_n}(\eta_n). For a finite coordinate set DD, the order position of aa is pos⁡D(a)=|{b∈D:b<a}|\operatorname{pos}_D(a)=|\{b\in D:b<a\}|.✦ For a rational array uu on Γ\Gamma and a finite coordinate set DD, the positional array code of uu is {(pos⁡D(a),u(a)):a∈D}\{(\operatorname{pos}_D(a),u(a)):a\in D\}, retaining zero coefficients.✦ A finite path code is a finite list of triples consisting of a weight, an auxiliary index, and a positional array code.✦ The prefix code cnc_n of a raw infinite path rr is the list of triples consisting of a weight, an auxiliary index, and the positional array code of the corresponding piece for i=0,…,ni=0,\ldots,n, using the common closure DnD_n.✦

The coding map σ\sigma pairs a fixed natural numbering of a finite code with one plus the maximum of its auxiliary indices, taking that maximum to be zero for the empty code.✦ It is injective and its value exceeds every auxiliary index in the code. A raw infinite path rr on Γ\Gamma is admissible when its pieces have successive ordinal supports, its auxiliary indices are strictly increasing and positive, and the conditions in (13) hold. 1≤jn≤pn,jn+1=σ(cn),ρ(α,β)≤pn(α,β∈Sn,α≤β).\begin{aligned} 1\leq j_n&\leq p_n, & j_{n+1}&=\sigma(c_n),\\ \rho(\alpha,\beta)&\leq p_n &&(\alpha,\beta\in S_n,\ \alpha\leq\beta). \end{aligned}(13)✦

A finite path pp in an operation family ff is a positive-length prefix of raw path data satisfying the raw-path admissibility conditions within the prefix, with original uncropped Type I pieces having the declared values and weights and successive supports for ff.✦ Entries beyond the prefix are unrestricted. An infinite path pp in an operation family ff is an admissible raw infinite path with original uncropped Type I pieces at every index, having the declared values and weights and successive supports for ff, with every original child belonging to KfK_f.✦ Its prefix through index nn has length n+1n+1, so index zero includes the first piece.✦ The finite path sum of a finite path pp in an operation family ff is the rational array obtained by summing its original rational pieces.✦ For an infinite path pp in an operation family ff , the partial functional p(n)p^{(n)} and bounded path functional Φp∈E*\Phi_p\in E^* are the linear functionals given by (14). p(n)=φ∑i=0nui,Φp(x)=limn→∞p(n)(x)(x∈E).p^{(n)}=\varphi_{\sum_{i=0}^{n}u_i},\qquad \Phi_p(x)=\lim_{n\to\infty}p^{(n)}(x)\quad(x\in E).(14)

The exact pair tree on zz has indexed nodes TjT_j. Its binary indexing has root 00 and children 2j+12j+1 and 2j+22j+2 at node jj; the ancestral list runs from the root to jj, including both endpoints. The ancestral path code of a node in an exact pair tree is the raw prefix code along the list of nodes from the root to that node, including both endpoints.✦ The requested weight of a node in an exact pair tree is the natural number 11 for the first node and otherwise σ\sigma of its parent’s ancestral code.✦ For a binary sequence b∈{0,1}ℕb\in\{0,1\}^{\mathbb N} , let νb(0)=0\nu_b(0)=0 and let νb(n+1)\nu_b(n+1) be the child of νb(n)\nu_b(n) chosen by b(n)b(n) .✦ The branch path pb(z)p_b(z) of a binary sequence bb through the exact pair tree on an admissible block sequence zz in an operation family ff is the infinite path in ff whose original piece, weight, and auxiliary index at position nn come from Tνb(n)T_{\nu_b(n)}.✦ For a branch reduction RR, write pb=pb(zR)p_b=p_b(z_R).

The Claim

Theorem 1. Let F⊆EF\subseteq E be a closed infinite-dimensional real linear subspace. Then ℛ(F)≠⌀.\mathcal R(F)\ne\varnothing.(15)✦ Let γ\gamma be an ordinal cut, let ff be an operation family, and let x∈Ex\in E. Let dd be a rational array supported below γ\gamma and, if f=pure(k)f=\mathrm{pure}(k), supported on CkC_k. Then ⟨d,Qγ,fx⟩=⟨d,x⟩.\langle d,Q_{\gamma,f}x\rangle=\langle d,x\rangle.(16)✦ Let R=(γ,f,H,U)R=(\gamma,f,H,U) be a branch reduction of FF, let b∈{0,1}ℕb\in\{0,1\}^{\mathbb N} be a binary sequence, and let x∈Ex\in E. Then Φpb(Qγ,fx)=Φpb(x).\Phi_{p_b}(Q_{\gamma,f}x)=\Phi_{p_b}(x).(17)✦

In the paper, the result reads: for a closed infinite-dimensional F⊆EF\subseteq E there are a nonzero limit ordinal γ≤ω1\gamma\leq\omega_1, a closed infinite-dimensional H⊆P<γEH\subseteq P_{<\gamma}E, and a bounded linear map U:H→FU:H\to F such that either HH lies on one color CkC_k and PCkP<γUh=hP_{C_k}P_{<\gamma}Uh=h, or P<γUh=hP_{<\gamma}Uh=h and every PCk|HP_{C_k}|_H is strictly singular; in both cases every P<β|HP_{<\beta}|_H for β<γ\beta<\gamma is strictly singular, and every actual path pp supported below γ\gamma, and on CkC_k in the first case, satisfies Φp(Uh)=Φp(h)\Phi_p(Uh)=\Phi_p(h).

What Is Checked

The statements (15), (16), and (17) concern the full norming completion EE, its bounded ordinal and color projections, and the branch paths built from the block approximations of a branch reduction; the existence assertion assumes that FF is closed and infinite-dimensional.

The full norming set contains every positive coordinate array✦, and each of its coefficients has absolute value at most 11.✦ Initial ordinal segments are order convex.✦ Restricting any member of KK to the mixed support set of any ordinal and color rectangle again gives a member of KK.✦ For any norming set and any coordinate set whose restrictions preserve normers, the ambient crop preserves the closed evaluation span.✦

Every allowed approximation state and stage admit a block approximation step.✦ Raw prefix supports are nonempty.✦ Every finite prefix of an infinite path has its sum in KK✦, and the resulting path functional has norm at most 11.✦ Each coordinate functional belongs to the coordinate predual X0X_0✦, and the prescribed next weight is positive for every finite list of exact pair tree nodes.✦