Alphabeta Math
Pipeline-generated
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

Grid homology and combinatorial knot Floer theory

1 · Prerequisites

2 · Logical flowchart

3 · Definitions, theorems and proofs

DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-6-luna)verified 2026-10-22 (gpt-6.1-sol)Open item page →

Admissible multipointed Heegaard data for knot Floer complexes

Definition

Work in the smooth category inherited from the oriented-link convention. In particular, an embedded bridge witness disk means the image of a smooth rank-two embedding of a planar disk, with ordinary smooth boundary or smooth transverse corners at its two marked endpoints; smoothness at a boundary or corner means local smooth extension in its charts. Witness parametrizations can be chosen for the finite disk families. This category convention does not require the knot arcs to meet the Heegaard surface transversely: tangential smooth crossings remain permitted.

Let K⊂S3 be a smooth oriented knot. Let Σ be a closed oriented surface of genus h, let k≥1, and put m=h+k−1. A handlebody is a compact three-manifold that is a regular neighborhood of a finite connected graph. A 2k-pointed Heegaard diagram for S3 consists of a splitting S3=Uα∪ΣUβ into handlebodies, two systems α=(α1,…,αm) and β=(β1,…,βm) of pairwise disjoint simple closed curves on Σ that bound pairwise disjoint properly embedded disks in Uα and Uβ, respectively, cutting each handlebody into exactly k three-balls, and two ordered sets of distinct points w=(w1,…,wk) and z=(z1,…,zk) disjoint from the curves. Curves from different systems intersect transversely in finitely many points. Each component of Σ∖α and of Σ∖β contains exactly one w-point and exactly one z-point. The diagram also includes trivial arc systems in the two handlebodies: the α-arcs join wi to zi, and the β-arcs join zi to wi−1, with indices modulo k. Each system is disjoint from its compression disks and is simultaneously boundary parallel: its arcs cobound pairwise disjoint embedded disks with arcs in the corresponding components of the attaching-system complement on Σ. Thus each ball obtained by compression contains one unknotted arc. Together the arcs form an oriented knot, oriented from wi to zi in Uα and from zi to wi−1 in Uβ. The diagram is compatible with K when this knot is orientation-preservingly ambient isotopic to K. For a grid diagram, compatibility uses the row and column orientation fixed in the grid definition, with w-points at O and z-points at X. For a grid representing one knot, the basepoint pairs can be numbered cyclically in reverse order of travel along the oriented knot so that wi,zi lie in the same component Ai of Σ∖α and wi,zi+1 lie in the same component Bi of Σ∖β.

Comparisons of these embedded diagrams allow orientation-preserving diagram isomorphisms: an orientation-preserving diffeomorphism of S3 carrying the two handlebodies, their oriented Heegaard surface, the two attaching systems, the ordered w,z points, and the two simultaneously boundary-parallel arc systems to the corresponding data. Equivalently one may specify the surface isomorphism together with its compatible collared extensions over both handlebodies carrying those arcs. The attaching systems are carried as systems; reordering their curve factors induces the corresponding product identification. Comparisons also allow arc normalizations in the fixed splitting: a smooth isotopy of the represented oriented knot, supported near the ordered marks and fixing each mark, during which every open connecting arc stays in its original handlebody and the two simultaneous boundary-parallel witness systems are retained or replaced by embedded witnesses for those same arcs. The surface, attaching curves, marked points and compression disks stay fixed. Such a normalization may replace a tangential marked crossing by a transverse one; the original data are not required to have neat arcs. This is a change of knot representative relative to the fixed surface, not a diagram isomorphism carrying both objects. Analytic and assigned grading data depending on the fixed surface diagram can be kept identical; the corresponding generator and disk-count identification is then the identity. Existence of normalizations with the required witnesses is a separate geometric assertion, not part of this definition. Such a comparison transports the represented oriented knot and any specified Seifert surface as well. For analytic data, it transports j, the surface area form, all chosen divisor neighborhoods and Jt by the induced symmetric-product diffeomorphism. The comparison maps on generators and disk classes are their pushforwards under this map; compatibility with the count differential is a mathematical property to be proved. No identity action is part of this convention, even when the ambient diffeomorphism is isotopic to the identity.

The m-fold symmetric product Sym⁡m(Σ)=Σm/Sm is the space of unordered m-tuples of points of Σ, where Sm permutes coordinates. Let q:Σm→Sym⁡m(Σ) be the quotient map. The product tori are the images Tα=q(α1×⋯×αm) and Tβ=q(β1×⋯×βm). Their intersection points are the generators. A Whitney disk class ϕ∈π2(x,y) is a relative homotopy class of maps from a disk to Sym⁡m(Σ) whose two boundary arcs lie on Tα and Tβ and whose endpoints are x and y. Its domain is the integral two-chain of complementary regions with coefficients equal to local multiplicities. Write nwi(ϕ) and nzi(ϕ) for its local multiplicities at the marked points. An almost complex structure J on Sym⁡m(Σ) is a smooth endomorphism of the tangent bundle with J2=−1. Fix a complex structure j and an area form on Σ. For m>0, choose a smooth path J=(Jt)t∈[0,1] of nearly symmetric almost complex structures, constant near t=0,1: each Jt agrees with the induced integrable structure Sym⁡m(j) on a fixed neighborhood of the diagonal and of every divisor {p}+Sym⁡m−1(Σ) for a chosen interior point p of every complementary region, including every marked point, and tames the form descended from the product area form off that neighborhood. The path is a sufficiently small perturbation in this class. For m=0 the symmetric product is a point. Local multiplicities are intersection numbers with these divisors; positivity is a mathematical property to be proved, not an additional choice of disk counts. A J-holomorphic strip representative is a smooth map u:R×[0,1]→Sym⁡m(Σ) satisfying ∂τu+Jt(u)∂tu=0, with u(τ,0)∈Tα, u(τ,1)∈Tβ, and limits x,y as τ→−∞,+∞. Translation in τ defines an R-action. The Maslov index μ(ϕ) is the Fredholm index of the linearized Cauchy–Riemann operator at a representative of ϕ; the relative Maslov grading convention is M(x)−M(y)=μ(ϕ)−2∑i=1knwi(ϕ), up to one overall additive constant.

A periodic domain is an integral two-chain in the complementary regions whose boundary is an integral sum of full α- and β-curves; its local multiplicity at a marked point is the coefficient of the containing region. The pointed diagram is admissible when every nonzero periodic domain with nwi=0 for all i has at least one positive and at least one negative region coefficient. A path J is regular for a set of disk classes when the linearized Cauchy–Riemann operator at every holomorphic-strip representative in those classes is surjective. For a path regular on the classes used, let MJ(ϕ) be the space of strip representatives. If, for each generator x, only finitely many pairs (y,ϕ) with MJ(ϕ)≠∅, μ(ϕ)=1, and nwi(ϕ)≥0 for every i occur, and each quotient is finite, the multipointed full filtered minus module is the free module over F2[U1,…,Uk] generated by Tα∩Tβ, with differential ∂−x=∑y∑ϕ∈π2(x,y),MJ(ϕ)≠∅μ(ϕ)=1,nwi(ϕ)≥0 ∀i#2(MJ(ϕ)/R) U1nw1(ϕ)⋯Uknwk(ϕ)y. Here #2 is the count modulo two of the zero-dimensional quotient when it is finite. Quotienting by all Ui leaves the fully blocked multipointed chain complex (the hat chain complex when k=1), with terms only from classes satisfying nwi=0 for every i; its Alexander-filtered associated-graded complex further retains only classes with every nzi=0. The relative Alexander grading convention is A(x)−A(y)=∑inzi(ϕ)−∑inwi(ϕ), with an overall additive constant. Each Ui lowers Maslov grading by two and Alexander filtration by one. In a grid diagram, Σ=T2, h=1, k=n, and m=n; the horizontal circles are the α-curves, the vertical circles are the β-curves, and the w-points and z-points are the O- and X-markings respectively.

DefinitionDefinition: Literature-sourcedProof: Not applicableverified 2026-10-22 (gpt-6.1-sol)Open item page →

Toroidal grid diagrams and gradings for oriented knots

Definition

Let K⊂S3 be a smooth oriented knot, meaning one oriented component of a finite smooth oriented link in the standard oriented three-sphere S3=R3∪{∞}. The grid orientation below fixes the sign convention for its Alexander grading.

For an integer n≥2, put Tn=R2/(nZ)2, with horizontal circles αi={y=i} and vertical circles βj={x=j} for i,j∈Z/nZ. A toroidal grid diagram of grid number n is a pair of disjoint sets O,X of cell centers (j+12,i+12), read modulo n, such that every row and every column contains exactly one point of each set. In the fundamental square [0,n)×[0,n), join the two marks in each row and each column by the straight segment between their centers. At each crossing of the projected link, the vertical segment is over and the horizontal segment is under. Orient each horizontal segment from its O-marking to its X-marking and each vertical segment from its X-marking to its O-marking. The markings and these segments determine an oriented link. Following each component from an O along its row to an X and then along that column to the next O defines a permutation of the n O-markings; its cycles are the link components. The grid represents a knot precisely when this permutation is one n-cycle.

A grid state is a subset x⊂⋃i,j(αi∩βj) containing exactly one intersection on each horizontal and each vertical circle. After numbering the circles, grid states correspond bijectively to permutations of {1,…,n}, so there are n! states. For states x,y, a rectangle from x to y is an embedded torus rectangle whose sides lie in grid circles, whose northeast and southwest corners lie in x, whose northwest and southeast corners lie in y, and whose remaining state points agree. Rectangles are subsets of the torus, so their lifts may cross a chosen fundamental-domain edge. Such a rectangle is empty when its interior contains no point of x; write Rect⁡∘(x,y) for the empty rectangles from x to y.

To define the gradings, fix the fundamental square [0,n)×[0,n). For finite signed point sets P,Q in this square, set I(P,Q)=#{(p,q):p1<q1, p2<q2} and J(P,Q)=(I(P,Q)+I(Q,P))/2, extended bilinearly to signed point sets. Regard x, O, and X as signed point sets and define

MO(x)=J(x−O,x−O)+1,MX(x)=J(x−X,x−X)+1.

For a knot grid, define the rational Alexander value

AG(x)=MO(x)−MX(x)2−n−12.

The lemma establishing integrality and the rectangle-change formula is a well-definedness supplier for this grading. For any rectangle r, let Oi(r) and Xi(r) be 1 if the interior of r contains the corresponding marking and 0 otherwise; let nO(r)=∑iOi(r) and nX(r)=∑iXi(r).

For illustration, this 3×3 marking pattern has one marking of each type in every row and column; its cell centers are read modulo three on the torus.

OXOXOX123
DefinitionDefinition: AI-adaptedProof: Not applicableverified 2026-10-22 (gpt-6.1-sol)Open item page →

Hat Heegaard Floer complex of a closed three-manifold

Definition

Let Y be a closed connected oriented smooth three-manifold. A pointed genus-g Heegaard diagram for Y consists of an embedded splitting Y=Uα∪ΣUβ, where Σ=∂Uα is a closed oriented genus-g surface and each U is a handlebody, meaning a regular neighborhood of a finite connected graph; ordered systems α=(α1,…,αg) and β=(β1,…,βg) of pairwise disjoint smooth simple closed curves bounding specified disjoint proper compression disks in the respective handlebodies, each disk system cutting its handlebody into one ball; and one point z off the curves. Opposite-system curves meet transversely in finitely many points. This is closed-manifold data; it includes neither a knot nor a second marked point.

Write q:Σg→Sym⁡g(Σ)=Σg/Sg for the permutation quotient. The product tori are the images Tα=q(α1×⋯×αg),Tβ=q(β1×⋯×βg). The group Sg acts on the ambient product, rather than on a fixed ordered product of distinct attaching circles. A generator is an intersection x∈Tα∩Tβ; equivalently it selects one intersection point on each curve of each system. There are finitely many generators. For g=0 the symmetric product and both empty products are one point, with the empty generator.

A Whitney class ϕ∈π2(x,y) is a relative homotopy class of maps of a closed disk with two marked boundary points into the symmetric product, taking its two boundary arcs into Tα,Tβ and its marked points to x,y. For g>0, its domain is the integral two-chain of complementary regions; its coefficient at an interior point p is the intersection multiplicity np(ϕ) with {p}+Sym⁡g−1(Σ). In particular nz(ϕ) is the basepoint coefficient. A periodic domain is an integral region two-chain whose boundary is an integral sum of full attaching circles. The domain of a class ψ∈π2(x,x) is periodic. Such a class is normalized when nz(ψ)=0. Write μ(ϕ) for the real Fredholm index of the disk linearization, before quotienting longitudinal translation; well-definedness and its topological index formula are properties to be proved in the analytic and index theory.

For g=0, the target is a point and the only Whitney class is constant: define its domain and every np(ϕ) to be zero and its index to be zero. Formal periodic domains are mΣ, with coefficient np(mΣ)=m at every point. Thus the only basepoint-normalized periodic domain is zero, and every sector weak-admissibility test is vacuous. No negative symmetric power or point-divisor intersection is used in this rank.

For a periodic domain P, cap its full attaching-circle boundary with the specified compression disks, using the boundary multiplicities and orientations, to obtain the integral class H(P)∈H2(Y;Z). A Spin-c structure s is a lift of the oriented tangent frame bundle through Spin⁡c(3)→SO(3); c1(s) is the first Chern class of its determinant line, with determinant character [s,u]↦u2. Weak admissibility in the specified Spin-c sector s means that EVERY nonzero integral periodic domain P with nz(P)=0 and ⟨c1(s),H(P)⟩=0 has both positive and negative region coefficients. This condition is imposed even when the sector has no diagram generators. It is a condition on domains and the intended sector, not only on periodic classes based at existing generators.

The ordinary sector module CF^(H,s) uses generators with Morse Spin-c label sz(x)=s. The label is defined by compatible handle Morse data: replace the gradient in disjoint tubes around the generator's index-one/index-two trajectories and the basepoint's index-zero/index-three trajectory by nonsingular fields, and take the Spin-c lift of their oriented perpendicular planes. Its actual construction, independence properties and periodic-index formula are obligations of the Spin-c and index theory. In particular the equality between the index of a REALIZED normalized periodic class based at x and the Chern evaluation in sz(x) is not assumed as its own proof. A realized-generator test alone is not substituted for the stated domain condition.

An admissible diagram for the full single-diagram hat module satisfies this sector condition for EVERY Spin-c structure on Y, including the sectors with no generators. The full module below uses all generators and hence all intended sectors; an empty sector contributes a zero summand but is not omitted from the admissibility test. Under Countable Choice, this full-diagram condition is equivalent to global weak admissibility: every nonzero periodic domain with nz=0 has both signs. Global weakness implies every sector test by restriction; conversely, the zero-Chern Spin-c sector supplied under Countable Choice by Complete Heegaard data have a zero-Chern Spin-c sector tests every normalized periodic domain. No generator in that sector is needed. For a SINGLE specified sector, ordinary weak admissibility remains the stated Chern-kernel condition and need not imply global weakness. No assertion of the full group from a diagram tested only at its represented sectors is made.

For g>0 choose a complex structure j and positive area form on Σ. Analytic data include a smooth path Jt on the symmetric product, constant on actual neighborhoods of t=0,1. On a fixed neighborhood of the diagonal and the point-divisors for a chosen interior point of every complementary region, including z, it agrees with the integrable structure Sym⁡g(j); away from that neighborhood it is tame for the form descended from the product surface area form. The path is chosen in a sufficiently small nearly symmetric class. Product-integrable normal forms near the finitely many transverse generators may be included among its specified auxiliary choices. Existence of such choices and of the regular perturbations needed below remains a mathematical obligation. A holomorphic strip is an actual smooth finite-Dirichlet-energy map u:R×[0,1]→Sym⁡g(Σ) satisfying us+Jt(u)ut=0,u(s,0)∈Tα,u(s,1)∈Tβ, with uniform generator limits at both longitudinal ends. The space in class ϕ is MJ(ϕ), and translation in s acts on it. Regular means that its actual linearized operator is surjective at the solutions being counted. Local multiplicity positivity, Fredholmness, the class index and compactness are results to be proved, not choices of which maps to count.

The hat diagram-level module CF^(H;F2) is free over F2 on the generators. For regular data for which the displayed quotient sets are finite and, for each x, only finitely many nonempty pairs (y,ϕ) occur, its candidate count operator is ∂^x=∑y ∑ϕ∈π2(x,y)μ(ϕ)=1, nz(ϕ)=0#2(MJ(ϕ)/R)y. Here #2 is parity of an actual finite zero-dimensional quotient. For g=0 the operator is zero on the empty generator. That the required regular finite counts exist, that ∂^2=0, and that its homology is invariant under changing pointed admissible diagrams and auxiliary choices are separate analytic and move theorems. Once these hold, the common homology is denoted HF^(Y;F2). No filtered chain-homotopy invariant, integral orientations or absolute grading is asserted by this definition.

Comparisons of embedded diagrams permit orientation-preserving isomorphisms carrying Y, its two handlebodies, their oriented surface, the attaching systems and z to the specified counterparts, equivalently compatible collared extensions of the surface map. Curve-factor reordering gives the corresponding product identification. Analytic data are transported by the induced symmetric-product map, including j, the area form, protected neighborhoods and Jt. The intended count identification is the actual pushforward of generators and disk classes with these transported data; its compatibility with the operator requires proof. An ambient isotopy is not assumed to act as the identity. Comparison of independently chosen perturbations likewise requires the continuation theorem.

Remarks

The multipointed knot construction has additional knot-arc and marked-pair data and is specialized to knots in the three-sphere. Its terminology is contextual here; no closed-manifold analytic conclusion is deduced from it.

The knot-specialized pointed convention is defined independently with additional knot-arc and marked-pair data; it does not alter this closed-manifold definition.

LemmaStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Integrality and rectangle changes of the grid Alexander grading

Statement

Let G be a toroidal grid of grid number n representing one oriented knot, and let MO, MX, and AG be the functions defined from its finite point sets and the pair-counting function J. Then MO and MX are integer-valued, and every rectangle r from a grid state x to a grid state y satisfies MO(x)−MO(y)=1−2nO(r)+2#(x∩Int⁡r) and MX(x)−MX(y)=1−2nX(r)+2#(x∩Int⁡r). Consequently AG is integer-valued and AG(x)−AG(y)=nX(r)−nO(r).

Facts & Assumptions

Given: A knot grid G with grid number n and the functions in the preceding definition.

[F1]

Toroidal grid diagrams and gradings for oriented knots defines grid states as one point on each horizontal and vertical circle, defines J(P,Q)=(I(P,Q)+I(Q,P))/2 by a finite pair count, and sets MO(x)=J(x−O,x−O)+1 and MX(x)=J(x−X,x−X)+1.

[F2]

Toroidal grid diagrams and gradings for oriented knots requires one O and one X in each row and column, and a one-component grid has a single n-cycle in the induced O→X→O marking permutation.

Proof

1.1F1givenalgebra

Each I(P,Q) is an integer count. For P=Q, J(P,P)=I(P,P), so the two self-pair terms in the formulas for MO and MX are integers; the cross terms occur doubled. Hence both Maslov functions are integer-valued.

1.2F1constructalgebra

Suppose a rectangle r is contained in the chosen fundamental square. Write its southwest and northeast corners as x1,x2∈x and its northwest and southeast corners as y1,y2∈y; all other state points are common. In J(x,x)−J(y,y) the common-point terms cancel, the ordered corner pair contributes 1, and each point of x strictly inside r contributes twice, once against each changed corner. Thus this difference is 1+2#(x∩Int⁡r). In 2J(x,O)−2J(y,O) the same cancellation leaves twice the number of O-markings in r. Substituting these two counts into [F1] gives MO(x)−MO(y)=1−2nO(r)+2#(x∩Int⁡r). Replacing every O by X gives the stated formula for MX.

1.3F1F2constructalgebra

For a marking set P∈{O,X}, write MP(x)=J(x−P,x−P)+1 and let NW(Pi) denote the northwest grid vertex of the cell containing the ith mark, read on the bottom edge when that cell is in the top row. Let xNWP be the state consisting of these vertices. Expanding the J formula in [F1] gives MP(xNWP)=#{(i,j):NW(Pi)<NW(Pj)}−#{(i,j):NW(Pi)<Pj}−#{(i,j):Pi<NW(Pj)}+#{(i,j):Pi<Pj}+1. For distinct i,j with neither mark in the top row, all four inequalities are equivalent and their signed contributions cancel. If Pj is in the top row, the second and fourth inequalities are equivalent and the first and third are false; if Pi is in the top row, the first and second inequalities are equivalent and the third and fourth are false. These contributions also cancel. For i=j, the only inequality that can hold is NW(Pi)<Pi, and it holds exactly for the unique mark in the top row. The displayed count is therefore −1+1=0, proving MO(xNWO)=MX(xNWX)=0.

2.1F2step 1.2constructalgebra

For two complementary rectangles r and r′ with the same vertical edges, let v be their common width, counted in column strips. The one-O- and one-X-per-column conditions in [F2] give nO(r)+nO(r′)=v and nX(r)+nX(r′)=v. The state has one point on each of the v−1 vertical circles strictly between those two edges, so #(x∩Int⁡r)+#(x∩Int⁡r′)=v−1. If the formula of step 1.2 holds for r from x to y, these three equalities give the formula for r′ from y to x by subtraction. The same argument applies to complementary rectangles sharing horizontal edges, counting rows instead of columns. Of the four torus rectangles for the two states, one lies in the fundamental square; successive horizontal or vertical complementations reach each of the other three, so the formulas hold for every torus rectangle.

3.1step 1.2step 2.1algebra

Subtract the MX change from the MO change and divide by two. The constant terms and state-interior terms cancel, giving MO(x)−MX(x)2−MO(y)−MX(y)2=nX(r)−nO(r). The common normalization −(n−1)/2 in AG cancels, proving the rectangle-change formula and that all differences of AG are integers.

3.2F1F2step 1.2step 2.1step 1.3choosealgebra

The row and column connections from O to X and back form one n-cycle by [F2]. Equivalently, the relative permutation between xNWX and xNWO is an n-cycle. Factor that cycle into n−1 transpositions and apply them successively to obtain grid states between the two states. Two states that differ by one transposition have four intersection corners: the two row circles and two column circles cut the torus into four rectangles, and one of those rectangles has the first state at its southwest and northeast corners and the second state at its northwest and southeast corners. Thus each transition is realized by a torus rectangle, whether empty or not. Each rectangle changes MX by an odd integer by steps 1.2 and 2.1, so MX(xNWO)≡n−1(mod2), using the zero normalization at xNWX from step 1.3. Since MO(xNWO)=0, the numerator MO−MX−(n−1) defining twice AG(xNWO) is even. Hence this value is integral.

4.1step 3.1step 3.2constructalgebra∎

Any two grid states correspond to permutations, and a finite product of transpositions carries one to the other. Each transposition gives a rectangle transition. Step 3.1 shows that AG changes by an integer at every transition; since one state has integral Alexander value by step 3.2, all grid states do.

LemmaStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

A knot grid diagram is admissible

Statement

Every toroidal grid diagram representing a knot is an admissible 2n-pointed Heegaard diagram in the stated knot Floer convention: each nonzero periodic domain with zero local multiplicity at every O-marking has both positive and negative region coefficients.

Facts & Assumptions

Given: A grid diagram of grid number n representing one oriented knot.

[F1]

Toroidal grid diagrams and gradings for oriented knots identifies the O-to-X row matching followed by the X-to-O column matching as a single n-cycle, and every row and column has exactly one marking of each type.

[F2]

Admissible multipointed Heegaard data for knot Floer complexes defines admissibility by requiring every nonzero periodic domain with zero multiplicity at all w-basepoints to have both positive and negative region coefficients; in a grid, w=O and z=X.

Proof

1.1F1F2givenconstruct

Regard the grid torus as the standard Heegaard torus separating the two solid tori of S3, with horizontal circles meridians of the first and vertical circles meridians of the second. The n parallel meridian disks in either solid torus cut it into n balls, one for each row or column annulus. Push the row O-to-X arcs into the first solid torus and the column X-to-O arcs into the second. Their pushing disks lie in disjoint row or column annuli and are disjoint from the compression disks, so each arc system is simultaneously boundary parallel as required by [F2]. The over/under convention is vertical over horizontal, exactly [F1]'s link. Thus these data represent the same oriented knot and satisfy the topological part of the multipointed definition; it remains to prove its admissibility condition.

1.2F1F2construct

Let pij be the coefficient of cell Rij in a periodic domain P. Because ∂P is a sum of whole horizontal and vertical circles, the coefficient jump across a vertical circle is independent of the row, and the jump across a horizontal circle is independent of the column. Fix reference indices i0,j0 and set ai=pi,j0 and bj=pi0,j−pi0,j0. Summing the constant jumps from the reference row and column gives pij=ai+bj for every cell. Since the marking permutation in [F1] is one cycle, number its basepoint pairs in reverse order of knot travel so that row region Ai contains wi,zi and column region Bi contains wi,zi+1, as allowed by the convention in [F2]. Then P=∑iaiAi+∑ibiBi, with local multiplicities nwi(P)=ai+bi and nzi(P)=ai+bi−1.

2.1step 1.2algebra

Suppose P is periodic and all nwi(P)=0. Then bi=−ai, so nzi(P)=ai−ai−1 and ∑inzi(P)=0. If one nzi(P) is nonzero, these integer multiplicities include both a positive and a negative value; each is a region coefficient at the corresponding X-cell, so P has both signs.

3.1F2step 1.2step 2.1algebra∎

If all nzi(P) also vanish, then all ai are the same integer c and all bi=−c. Hence P=c(∑iAi−∑iBi)=0, because each of the two sums is the whole torus with coefficient one. Thus a nonzero periodic domain with zero O-multiplicity cannot have all X-multiplicities zero, and step 2.1 proves it has both signs. By [F2], the grid diagram is admissible.

LemmaStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-6-luna)verified 2026-10-22 (gpt-6.1-sol)Open item page →

The standard genus-zero hat diagram of the unknot has rank one

Statement

Let U be the standard oriented unknot and take its genus-zero one-pointed Heegaard diagram: Σ=S2, each handlebody is a three-ball, there are no α- or β-curves, and one w-point and one z-point are joined by trivial arcs in the two balls. The associated-graded hat chain complex has one generator, zero differential, and homology of rank one over F2. This is a computation for the stated diagram; it makes no invariance claim for other diagrams.

Facts & Assumptions

Given: The genus-zero, one-pointed diagram specified in the statement.

[F1]

Admissible multipointed Heegaard data for knot Floer complexes defines Sym⁡m(Σ), the product tori, and their intersection points as generators; its hat specialization sets the basepoint variable to zero.

[F2]

Admissible multipointed Heegaard data for knot Floer complexes defines the differential by counts of classes of Maslov index 1.

Proof

1.1F1givenconstruct

Here h=0 and k=1, so m=h+k−1=0. The empty products of α- and β-curves both map to the unique point of Sym⁡0(S2), which is a singleton. By [F1], this intersection is the entire generator set, so the chain group has dimension one.

2.1F1F2step 1.1algebra∎

Every map into a singleton is constant, so there is only the constant class in π2(x,x). Its pullback tangent bundle is zero-dimensional; hence the linearized Cauchy–Riemann operator has zero kernel and cokernel and Maslov index 0. Thus there is no class of Maslov index 1 to contribute under [F2]. The differential is zero, and the homology equals the one-dimensional chain group.

TheoremStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Grid moves generate isotopy of oriented knots

Statement

Assume countable choice. Let G0,G1 be finite planar grid diagrams, with vertical segments over horizontal segments and orientation from O to X in rows and X to O in columns, representing the same oriented knot in the standard oriented three-sphere. A cyclic permutation moves the first row or column to the last position. A row commutation exchanges adjacent rows when the closed intervals between their respective O- and X-mark abscissae are disjoint or one lies in the interior of the other; a column commutation is defined by the same condition on the intervals between marking heights. A stabilization chooses a marked cell and temporarily removes its marking together with the other markings in its row and column. It splits that row and column into two, places the three removed markings, and adds one new O-mark and one new X-mark so that every row and column again has one of each. In the four-cell subdivision of the chosen cell, exactly three cells are marked: the chosen marking type appears twice, the other type once, and any one corner may be the empty cell; the two other markings removed from the original row and column are placed outside this block in the new row and column. A destabilization is the inverse operation. Then a finite sequence of cyclic permutations, commutations, stabilizations, and destabilizations carries G0 to G1.

Facts & Assumptions

Given: Countable choice and the two finite grids representing the same oriented knot, with the full marked move conventions in the Statement.

[F1]

Every grid has a finite bridge-like refinement by grid moves refines any nonempty grid to a bridge-like grid by actual marked stabilizations and legal commutations, preserving the decorated planar-isotopy class after smoothing.

[F2]

Reidemeister's theorem for oriented diagrams connects regular diagrams of equivalent oriented links by finitely many planar isotopies and oriented R1/R2/R3 moves under countable choice.

[F3]

Planar isotopy preserves the grid-move class of bridge rectifications connects arbitrary bridge rectifications of planar-isotopic bridge diagrams by finitely many permitted grid moves.

[F4]

Relative grid realizations of Reidemeister moves connects arbitrary rectifications when bridge representatives of the endpoints of a local Reidemeister move have been supplied; it covers arbitrary exterior and every oriented sign variant.

[F5]

External input: Ozsvath, Stipsicz, and Szabo, Grid Homology for Knots and Links, Appendix B.4, Definition B.4.4, says every diagram can be approximated arbitrarily closely in C2 by bridge diagrams; the proof of Proposition B.4.3 begins by isotoping the diagram to bridge position. For a regular diagram, choose this small planar isotopy so its finite transverse crossings and their decorations persist. I use this source only to supply bridge representatives; the local arbitrary-exterior move realization remains [F4]. I inspected Definition B.4.4 and the surrounding Proposition B.4.3 and Theorem B.4.1 discussion.

Proof

1.1F1givenconstruct

A knot grid is nonempty. Apply [F1] to each Gi, obtaining a finite grid-move sequence to a bridge-like grid Bi. Smooth Bi as in that supplier to a bridge diagram Di. Its crossing decorations and strand orientation agree with the original grid diagram, and the supplier proves planar isotopy after smoothing. Thus D0,D1 represent the same oriented knot. The three-mark splits and strict interval commutations in [F1] are among those specified here: the retained exterior row and column partners provide the two markings outside the stabilization block in the Statement. No unlisted shared-endpoint switch has been used.

2.1F2F5step 1.1construct

By [F2] choose a finite sequence of planar isotopies and oriented Reidemeister moves from D0 to D1. Compose successive planar isotopies, and regard each remaining Reidemeister move as an edge between its two decorated planar-isotopy classes. The sequence now traverses finitely many such classes, with the first and last represented by D0,D1. If there is no Reidemeister edge, these two diagrams are planar isotopic. Otherwise every intermediate representative is a regular diagram and has a bridge representative in its decorated planar-isotopy class by [F5]; choose the perturbation small enough that its crossings and decorations persist. Choose one bridge representative for each occurrence of a class in this finite sequence, keeping D0,D1 at its ends, and choose a bridge rectification there. The choices are finite; no stronger choice principle is invoked.

3.1F3F4step 2.1constructcases

Across a Reidemeister edge, the chosen bridge representatives need only be planar isotopic to that edge's original endpoint diagrams. This is exactly the relative-representative interface of [F4], so their chosen grid rectifications are joined by actual grid moves, with every exterior strand included. Across an interval with only planar isotopy, [F3] joins the chosen rectifications. This also treats the sequence with no Reidemeister edge. The endpoint rectifications may be taken to be precisely B0,B1, because both suppliers apply to arbitrary rectifications. Concatenating the finitely many constituent finite sequences gives a sequence of allowed commutations, stabilizations and destabilizations from B0 to B1.

4.1F1step 1.1step 3.1construct∎

Prepend the refinement from G0 to B0 and append the reversed refinement from B1 to G1. Reversing a stabilization gives a destabilization and reversing a legal commutation gives a legal commutation, so all operations remain allowed. This produces the required finite sequence from G0 to G1. The proof needs no cyclic permutation, but allowing those additional moves in the Statement does not change the conclusion. The local suppliers cover all marking types and strand orientations, so the sequence preserves the prescribed oriented-knot convention throughout.

LemmaStatement: Literature-sourcedProof: AI-generatedjudge pass (gpt-6-luna)verified 2026-10-22 (gpt-6.1-sol)Open item page →

Regularity, compactness, and gluing for pointed Heegaard disks

Statement

Assume countable choice. Let H be any admissible 2k-pointed Heegaard diagram for an oriented knot in S3, with transverse attaching curves, product tori, and marked points as specified in the pointed-data definition. Admissibility means every nonzero periodic domain with all w-multiplicities zero has both positive and negative region coefficients. There is a smooth path J=(Jt)t∈[0,1], constant near its endpoints and chosen among perturbations for which all holomorphic-strip representatives used in the differential have nonnegative local multiplicity at every marked point, such that the linearized Cauchy–Riemann operator is surjective for every Maslov-index-1 strip class with all w-multiplicities nonnegative. Each corresponding moduli space modulo translation is finite. For every Maslov-index-2 class with all w-multiplicities nonnegative, the compactification of the one-dimensional moduli space modulo translation has as codimension-one boundary broken pairs of index-one strips meeting at an intermediate generator and disk bubbles whose entire boundary lies on one of the product tori (the boundary-degeneration strata). Gluing gives a collar at every broken end, and the boundary-degeneration contributions cancel modulo two in the differential-square count. No sphere bubbling or escape through a nonzero periodic domain gives an additional codimension-one end. Consequently the disk-count differential in the pointed-data definition squares to zero and preserves its Alexander filtration.

Facts & Assumptions

[F1]

For the actual balanced S3 diagram, the full-marked count theorem constructs ONE integrable background and genuine time-collar path with marked positivity, no nonconstant strip of index at most zero, regularity of every actual index-one and index-two strip, and finitely many rigid translation orbits across ALL generators/classes/nonnegative marked degrees. It also supplies finite regular endpoint unit disks: their parities are one for every alpha and beta cut component if k≥2, and zero for both components if k=1, with actual pointed sphere avoidance in that latter case. All positive z multiplicities are included. (Balanced knot data admit full-marked square-zero holomorphic counts)

[F2]

For THE path constructed in [F1], the full-marked compactification theorem gives an actual compact one-manifold for each pair of generators and pair of nonnegative marked vectors. Its finite boundary consists EXACTLY of broken rigid pairs and the unit boundary disks attached to a constant generator. All have actual exhaustive one-sided gluing collars. Its proof first uses actual bounded-energy bubble extraction and index additivity and then the nonpositive-index exclusion to check limiting-strip regularity; it excludes the singly pointed sphere alternative by the actual sphere avoidance. (Full-marked index-two compactification gives the boundary-disk curvature identity)

[F3]

Actual unbubbled strip neck cutoffs and regular gluing preserve Whitney concatenation. The actual boundary-disk attachment theorem constructs corrected maps in the class of the source plumbing, with exhaustive collars. Boundary-preserving local addition gives actual relative homotopies between sufficiently close compact cores. (Actual unbubbled strip breaking and regular gluing preserve Whitney concatenation, A regular boundary disk attachment has an exhaustive nodal half-collar, A boundary-preserving local addition flat near transverse generators)

[F4]

The actual relative Alexander grading satisfies A(x)−A(y)=∑inzi(u)−∑inwi(u) for every realized Whitney class; each Ui lowers filtration by one. The original full minus differential counts every z degree with weight Unw. (Balanced three-sphere knot generators admit consistent relative gradings, Admissible multipointed Heegaard data for knot Floer complexes)

[A1]

Countable Choice is assumed. The diagram, its complete cut systems, product tori and marked points are the actual data of the Statement; no zero-z restriction is added. (The Axiom of Countable Choice (ACω))

Proof

Given: Countable Choice and the admissible balanced pointed S3 knot diagram in the Statement.

1.1F1A1givenconstruct

Choose the actual background and path from [F1]. Its fixed endpoint neighborhoods are genuine positive time collars, and all marked local multiplicities of its actual strips are nonnegative. Each index-one solution is nonconstant, since a constant transverse generator has index zero. Its actual linearization is onto by [F1]. The globally finite list of actual rigid orbits in [F1] in particular makes every specified class's translation quotient finite, including the empty quotient. This uses the full-marked construction, not the zero-z associated-graded branch. Every actual index-two strip is regular as well, so its translation quotient has the indicated one-dimensional interval charts.

2.1F2F3step 1.1construct

Verify the original PER-WHITNEY-CLASS compactification. Fix an index-two class ϕ and its endpoints. If any marked coefficient is negative, its actual moduli is empty by step1.1, and no positive-degree nodal configuration can have that class. Otherwise let W,Z be its two marked vectors and take the actual compact one-manifold M(x,y;W,Z) in [F2], initially across classes. The Whitney-class label is locally constant in its interior: a sufficiently small local-addition chart gives the straight relative homotopy, with the same ends and physical boundary. Extend the label over each of its finitely many added endpoints using its actual collar. That collar's connected open interval has a single class. For a broken pair [F3]'s scalar real neck cut-and-cap homotopy identifies it with the concatenation of the two classes. For a disk endpoint, use the ACTUAL attachment source plumbing: cut out its small physical half-disk, retain the fixed generator chart on the annulus, and cap its outer boundary by the constant generator. The inner component is the node-fixed disk, while the outer strip is constant. The preglue represents precisely this boundary-disk insertion. Its Newton correction is homotopic to that preglue by the local-addition straight section homotopy preserving the two torus boundaries and generator ends. Scalar contraction on the remaining small annular chart proves the same class for every sufficiently small collar member; no holomorphicity of this homotopy is claimed. Exhaustion in [F2,F3] makes this label locally constant at the endpoint too. Thus the fixed-class preimage is clopen in the compact one-manifold. It is itself compact and inherits exactly the interval/half-interval charts and collars with that class label. This proves the stated compactification for ϕ, rather than replacing it by the marked-vector union.

3.1F1F2F3step 2.1construct

The boundary classification on this fixed-class part is EXACT: its endpoints are the matching broken index-one pairs and boundary-degeneration disks in [F2]. No other end can appear upon restricting a clopen subset. The actual energy/index argument in [F2] captures all limits before restricting: a nonconstant limiting strip has positive index by [F1], so total index two makes it index one or two and hence regular. Nonconstant boundary bubbles have index at least two, and spheres have index at least 2k. This leaves the listed broken and unit-disk cases when k≥2; for k=1 the only further unit-sphere possibility passes through the generator and is excluded by the chosen actual sphere avoidance. The finite-domain uniform energy bound and energy-complete extraction in [F1,F2] leave no escaping periodic-domain contribution. Every listed matching broken end has the exhaustive collar in [F2,F3], and so does each actual disk end.

4.1F1F2F4step 1.1step 2.1step 3.1constructalgebra

To verify cancellation in the ORIGINAL differential-square count, first keep all marked variables: assign a rigid strip weight UnwVnz. The matrix is finite by step1.1. Boundary parity for [F2]'s actual compact one-manifolds identifies the coefficient of each UWVZy in its square with the actual disk-end count, since its other endpoints are precisely the ordered rigid pairs with that monomial. If k≥2, an alpha unit disk in the component containing wi,zi has count one and weight UiVi, and a beta unit disk in the component containing wi,zi+1 has count one and weight UiVi+1, by [F1]. Hence D2=(∑iUi(Vi+Vi+1))I. If k=1, both actual unit counts are zero and the sphere alternative was excluded, so D2=0. Now set EVERY Vi=1. This specialization is exactly the full filtered minus differential in the original definition, still counting every positive z multiplicity. Its curvature becomes ∑i(Ui+Ui)=0 in characteristic two, and hence d2=0. The cancellation is in each polynomial coefficient of this differential-square count; it does not falsely assert that the alpha and beta disk insertions have the same Whitney class or the same full z vector.

5.1F1F4step 1.1step 4.1algebra

For an actual term from x to y, [F4] gives A(Unwy)=A(y)−∑inwi=A(x)−∑inzi≤A(x). The last inequality is the actual marked positivity in step1.1. Thus the finite full filtered minus differential preserves the Alexander filtration. An additive shift in the relative Alexander values has no effect on this inequality; no absolute normalization or invariance is needed.

6.1F1F2F3F4A1step 1.1step 2.1step 3.1step 4.1step 5.1construct∎

When the target is a point there are no nonconstant strips and every asserted differential and compactification is empty, so the same conclusion holds directly. For every positive-dimensional balanced diagram the preceding chosen path supplies all the surjectivity, finite quotients, perclass compactifications, exhaustive broken collars, boundary cancellation, absence of additional sphere/periodic ends and filtration statements. The proof has retained all w,z degrees and the full m=h+k−1 scope. It proves this chosen-data analytic lemma and no diagram-move, absolute-grading or combinatorial comparison theorem.

DefinitionDefinition: Literature-sourcedProof: AI-generatedjudge pass (gpt-6-luna)verified 2026-10-22 (gpt-6.1-sol)Open item page →

The Alexander filtration and tau level

Definition

Let K be a smooth oriented knot in the standard oriented three-sphere. Let F be a compact connected oriented Seifert surface with boundary K, and let H=(Σ,α,β,w,z) be an admissible doubly pointed Heegaard diagram representing (K,F). The associated one-pointed hat chain complex CF^(H) is the finite-dimensional chain complex over F2 obtained from the minus complex by setting its basepoint variable to zero; its differential counts index-one disks with zero multiplicity at w. Assume for this specified data that the differential is nonincreasing in the Alexander filtration, that the absolute Alexander formula below takes integer values, and that H(CF^(H))≅F2. Give a generator x its absolute Alexander filtration AF(x)=12⟨c1(sw,z(x)),[F^]⟩, where sw,z(x) is the homology class of the nowhere-vanishing vector field assigned to x by the pointed Morse construction, with its prescribed boundary data near the two basepoints; c1 is the first Chern class of the oriented plane field perpendicular to that vector field. To form the zero-framed surgery, remove a tubular neighborhood of K and glue in a solid torus whose meridian is the longitude induced by F. The surface F^ is obtained by deleting a boundary collar of F and capping its longitude by the meridian disk of that solid torus. This is the standard integer normalization, with the one-generator diagram of the standard unknot at filtration zero. The subcomplex FsCF^(H) is spanned by generators of Alexander filtration at most s.

More generally, a finite integer-filtered chain complex is a finite-dimensional chain complex C over F2 with subcomplexes FsC such that FsC=0 for all sufficiently small integers s and FsC=C for all sufficiently large integers s. If H(C)≅F2, define τ(C):=min⁡{s∈Z:im⁡(H(FsC)⟶H(C))≠0}. For the fixed diagram and surface above, write τH,F(K)=τ(CF^(H),F∗CF^(H)). Once diagram and surface independence is proved, their common value is the knot invariant τ(K). Until that comparison is proved, τH,F(K) denotes only the value for the specified data.

Facts & Assumptions

Given: A finite integer-filtered chain complex C over F2 with H(C)≅F2.

[A1]

The filtration is exhaustive, bounded below, and each FsC is a subcomplex, as required in the Definition.

[A2]

The total homology is nonzero and one-dimensional, as required in the Definition.

[A3]

The Alexander levels are integers, as required in the Definition.

[F1]

The well-ordering principle: every nonempty subset of N0 has a least element.

Verification

1.1A1A2givenalgebra

Since FsC=0 for all sufficiently small s, the inclusion-induced map H(FsC)→H(C) is zero at those levels. Since FsC=C for all sufficiently large s, this map is the identity on H(C) at those levels and is nonzero by [A2].

2.1A1A3F1step 1.1constructalgebra∎

Let S={s∈Z:H(FsC)→H(C) has nonzero image}. Step 1.1 shows S is nonempty, and [A1] gives an integer lower bound m for S. Then S−m is a nonempty subset of N0; by [F1] it has a least element k. For every s∈S, s−m≥k, so m+k is the least element of S. This proves that τ(C) exists and is unique.

LemmaStatement: AI-adaptedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Analytic foundations for disk and polygon counts

Statement

Assume ACω. Let a finite pointed Heegaard multidiagram have actual complete handlebody disk systems for closed oriented three-manifolds, with a basepoint in every one-system cut component. Ordered polygons use distinct labeled, pairwise transverse attaching systems; when an operation repeats a geometric system, use its separately labeled small transverse auxiliary copies. For the full hat operations impose global weak multidiagram admissibility: every nonzero full-curve multiperiodic domain vanishing at all basepoints has both signs. For operations in a specified total Spin-c family, ordinary weak admissibility is also allowed: every nonzero formal periodic domain difference in the capped Spin-c kernel, normalized at every mark and with Chern evaluation zero, has both signs. The same formal sector condition is imposed on each relevant boundary pair, including intended sectors with no generators. In this ordinary weak branch all basepoint multiplicities are fixed to zero, the required arities and compact auxiliary parameter families are finite, and counts use the corresponding fixed actual index bands. Strict all-mark normalized strong admissibility is a further allowed branch; it is not substituted for ordinary weakness. For a single diagram the full-module admissibility convention tests all intended Spin-c sectors, including empty sectors, and implies the required global weak domain condition.

In positive symmetric rank choose, among the auxiliary data, an arbitrarily small pointed Hamiltonian representative with no triple product-torus intersection, following its pair generators and transporting its domains, actual Spin-c labels and admissibility. Fix that representative. Compatible generic source-dependent perturbations, coherent over the required conformal faces and auxiliary parameter families, then make the disk and polygon spaces entering the hat differential, continuation, handleslide, stabilization and surgery operations regular with their actual expected dimensions. The common protected tensor and arbitrary constant product-angle end charts can be prescribed, and coherent perturbations can be arbitrarily small about a supplied compatible reference model. Auxiliary polygon and anchored-continuation parameter families here keep the autonomous end paths fixed and regular on the required energy budget; regular compatible parameter-boundary data may be retained. No slice-regularity assertion is made for arbitrary parameter-varying autonomous end paths. This does not assert regularity by almost-complex variations alone on an arbitrary fixed triple, or invariance under the preliminary curve choice.

Dimensions include every allowed conformal and auxiliary family parameter. Quotient only actual source automorphisms: longitudinal translation acts freely on autonomous nonconstant two-ended strips; an anchored source-dependent continuation strip and a three-marked triangle have no translation quotient. Every total zero-dimensional space used in a mod-two count is finite, across all its literal realized classes and with its specified labels. The one-dimensional spaces entering the chain-map, composition and homotopy identities compactify to actual compact one-manifolds. At fixed data their boundaries are precisely the matching rigid polygon–polygon conformal splittings and strip–polygon breakings. In an auxiliary family the broken configurations are the rigid joint fiber configurations at one common parameter, together with the prescribed parameter-boundary strata; they need not be products of individually rigid parameterized polygon spaces. Every such broken end has an exhaustive gluing collar. For the zero-basepoint operations no nonconstant sphere or one-torus boundary disk survives; thus the specified boundary-degeneration terms vanish, and there is no further escape or uncontrolled periodic-domain end. Boundary parity gives the differential square-zero identity and the polygon chain-map, composition and homotopy identities.

At symmetric rank zero retain the original empty systems and use the point target directly: the differential is zero, the matching constant triangle is counted once, higher rigid operations vanish, and the conformal and parameter boundary identities are the actual source-space identities. No triple-torus exclusion or free translation quotient is imposed on that point map.

This is the analytic count theorem. It does not construct a subordinate surgery diagram, prove that a named top generator is a cycle, identify the homology of a surgery total complex with a filling, or prove Heegaard-move invariance. Those remain duties of the respective operation constructions.

Facts & Assumptions

Given: The actual finite pointed multidiagram, complete cut systems, the stated operation-specific domain admissibility, any fixed total Spin-c label, finite required arities and auxiliary parameter spaces, and the zero-basepoint hat count convention.

[F1]

The full single-diagram definition quantifies weak admissibility in every intended Spin-c sector, even empty ones. Complete Heegaard data admit an actual zero-Chern Spin-c sector under ACω, so that full-module condition is equivalent to global mixed signs of normalized periodic domains. A single specified-sector weak condition is not asserted to imply this. (Hat Heegaard Floer complex of a closed three-manifold, Complete Heegaard data have a zero-Chern Spin-c sector)

[F2]

Positive-rank small triple-free representatives exist, with pair generators followed and actual source-dependent transports of polygon map spaces. The finite Morse Spin-c construction carries the actual lift, including torsion, and the capped-domain lattice under attaching-curve isotopies. In an ordinarily weak specified sector the nonnegative domains at each fixed entire mark vector and finite relative-index set are finite; the transfer to actual indices uses [F7]. Strict admissibility additionally gives a lower strip index and finite domains in every bounded upper-index family; that additional conclusion is not used for the ordinary weak branch. (Pointed transverse multi-diagrams admit small triple-free curve representatives, The Spin-c difference of Heegaard polygons is the capped periodic domain, Strict polygon admissibility bounds indices and positive domains)

[F3]

On a globally weak multidiagram, coherent source-dependent choices exist with all-index regular zero-marked end strips and polygons, genuine endpoint collars, actual arbitrary product-angle charts, prescribed compatible protected geometry and arbitrarily small perturbations about a compatible reference. The proved finite-count theorem gives stable-source compactness, actual index budgets, exhaustive nonlinear gluing, compact Hausdorff one-manifolds and all zero-marked A-infinity boundaries, including the direct rank-zero model and relative parameter construction. (Globally weak multidiagrams admit coherent all-index regular polygon data, Globally weak zero-marked polygon counts have the actual A-infinity boundaries)

[F4]

For strictly strongly admissible specified Spin-c families on general balanced intermediate three-manifolds, the full-marked theorem supplies good end paths with no nonconstant nonpositive-index strips, index-one regularity, coherent all-index polygon regularity, finite low-dimensional total counts and actual exhaustive boundaries. Its proof first obtains the finite lower-index band before taking limits, and retains all actual Spin-c extension and marked-degree data. (Strictly admissible triple-free polygons have full-marked low-dimensional counts)

[F5]

Actual fixed-background source-dependent continuation solutions have anchored compactness, all-index regular perturbations, finite counts, exhaustive collars and compact one-manifold parameter boundaries. The central component is not divided by translation; its autonomous escaping end levels are. (Fixed-background zero-marked continuation counts give chain homotopy equivalences)

[F6]

The actual point-divisor bundle admits a smooth connection flat in its canonical frame near the boundary tori, with bounded curvature and the degree identity na=−(2πi)−1∫u∗F. Actual finite-ended polygon charts, source-local universal regularity, domain-energy bounds, stable-source compactness and regular-tree gluing are proved without global or strong admissibility as premises; those assumptions enter only their applications to energy bounds. The bounded-energy autonomous argument uses dense time-slice injective sites and its free-image test. (Divisor degrees split over actual bubble trees and zero-marked strips have no bubbles, Source-dependent nearly symmetric polygons admit universal regularity under countable choice, Nearly symmetric polygons have positive domains and degree energy bounds, Bounded-energy polygons have stable-source, bubble and strip limits, Regular transverse polygon breakings have exhaustive gluing collars, Time-slice injective points of a finite-energy strip are open and dense)

[F7]

For actual same-corner, same-Spin-c polygons, common real-plane corner caps identify the relative Maslov number with Chern evaluation plus twice the entire marked degree. For finite-ended operators in the protected nearly symmetric tame class, its boundary-preserving comparison homotopy identifies that number with the difference of their real Fredholm indices. The two-edge version includes strips; common source and external parameters cancel in differences. Transverse-node indices add before regularity. (Relative Heegaard polygon Maslov variation is Chern evaluation plus marks, Relative finite-ended indices are compact determinant windings)

Proof

1.1F1F2F3F6F7givenconstructalgebra

The admissibility inputs are actual domain bounds, not a formal assumption that the desired moduli are compact. For a full single-diagram hat module, [F1] supplies a zero-Chern sector independently of whether it has generators; the required all-sector test therefore tests EVERY normalized periodic domain and gives the global weak pair condition. For several systems the multiperiodic condition in the Statement is imposed on the actual chosen auxiliary multidiagram; pairwise conditions alone are not substituted for it. Its nonnegative zero-mark domains are finite before any index or regularity choice, as proved in [F3]. For the ordinary weak specified-sector branch, fix the finitely many required corner tuples, arities and count dimensions zero and one. If a source has d inputs and p allowed family parameters, its fixed-map index is its count dimension minus d−2+p; the finite strata of a compact parameter family therefore require only a finite list of actual indices. Strips and anchored profiles have their stated respective index shifts. For each nonempty same-corner, same-Spin-c slice choose one actual smooth reference and identical transverse real-plane caps for it and every comparison map. The protected nearly symmetric comparison homotopy in [F7] preserves those boundary planes and caps. Consequently the actual fixed-map index difference is ν(D(u)−D(u0))=⟨c1(s),H(D(u)−D(u0))⟩ at zero entire mark vector. Thus each fixed actual index gives a fixed relative-index slice after subtracting the reference index. The common conformal and external parameter columns cancel; their finite dimension shifts have already been included. Apply [F2] to these relative-index slices to obtain finitely many nonnegative domains, independently of coefficients. Include the actual continuous concatenations representing the total classes of the required boundary identities: their total index is the same prescribed finite list by actual cap additivity, their marks are zero, and their total Spin-c extensions are the specified ones. Thus their domains belong to those same affine slices, not to a hypothetical finite class list. There are finitely many slices, so selecting their references uses only finite choice, and their union is finite. The domain-energy estimate of [F6] supplies a uniform energy bound E0 before the coefficient choices, on a small family with uniformly bounded geometric constants. Enlarge the finite list for the finitely many separated compositions and relative homotopies required, and choose a strict energy budget E>E0. In the strict branch [F2] separately supplies its stronger finite upper-index domains and lower strip index. No lower index bound is claimed for ordinary weak data.

1.2F3givenconstructalgebra

Treat rank zero by [F3]'s direct model. Its map operator is the zero operator between zero spaces and is onto of index zero. A triangle has one matching constant map, no conformal parameter and no continuous automorphism. A higher stable polygon has its actual positive source dimension and no rigid contribution. The four-marked source is an interval whose two ends are the two constant triangle products, and its parity is zero over the coefficient field. A matching one-parameter triangle family is its parameter interval; other assigned labels or nonzero marked vectors give empty sets. There are no nonconstant strips or bubbles. This proves the point-target count identities without excluding an allowed genus-zero input.

1.3F2F3constructalgebra

For positive rank apply [F2], protecting the basepoints and a witness in each old region. Pair generators persist without births, and the swept-annulus integral lattice transport preserves old sign witnesses. The actual collar extensions carry the compression disks, capped classes and Spin-c lifts. Consequently the prescribed admissibility and labels hold on the chosen representative. No constant stable polygon can occur: its at least three distinct boundary labels would give a common point of three product tori. This curve choice is explicit and necessary; the coefficient variation at a constant map is zero because its derivative is zero, so a J-only genericity assertion on the original arbitrary fixed triple would be false. Nothing in this step asserts count invariance under the curve change.

2.1F6step 1.1constructalgebra

Reconstruct the reverse divisor bound needed for ordinary weak components. For each chosen region point a, its actual divisor bundle has the canonical nonzero section σ on a neighborhood of the compact union of the fixed boundary tori. If a permitted compact family of boundary tracks is used, take its whole union and require the chosen divisor to avoid it; here the analytic coefficient families keep the chosen attaching curves fixed. Choose U1⋐U0 about that union, disjoint from the divisor, and use the flat connection in frame σ on U0. Cover the remaining compact target by finitely many line frames with flat local connections; a finite smooth partition, whose U0 coefficient is one on U1, patches them to a connection ∇. In a frame write ∇e=Ae. The transformation A′=A+g−1dg makes F=dA a global smooth bounded two-form, zero on U1. In a pulled-back disk frame write σ=fe along the boundary. There A=−f−1df, so Stokes gives ∫u∗F=−2πiwind⁡f=−2πina(u), with the actual relative degree convention in [F6]. For a finite-ended polygon or anchored strip truncate every end in its common exact generator chart inside U1, multiply its coordinate map by a real cutoff and cap at the generator. Scalar multiplication preserves both product boundary planes, with arbitrary specified angles. All modified pieces and caps have F=0, so the same integral formula holds on the uncapped source, independently of those choices. Pointwise ∣u∗F∣≤C∣du∣2/2, hence ∣na(u)∣≤CaEg(u). This is not a reversal of energy bounded by domain area. Finitely many region points simultaneously bound EVERY component domain with energy at most E, regardless of its split Spin-c label. The same curvature estimate on omitted necks and the actual energy identity give integer degree additivity: core integrals converge, remaining integrals tend to zero, and convergence of the resulting integers is eventual equality. This applies at nodes on divisors and to contained compact bubbles, without a zero-set avoidance assumption.

3.1F3F6step 1.1step 2.1constructcontradiction

Choose genuine end paths good on the energy budget without a lower-index assumption. Start in any prescribed smooth neighborhood of the compatible reference paths. Use the actual source-free, time-slice-injective universal argument in [F6]. A nonconstant strip has a free site outside the protected divisor and torus neighborhoods, since an entire image in the latter misses EVERY region divisor and has zero energy. A time-slice injective point and adjoint unique continuation localize a time-only variation there. The explicit countable Hilbert smoothing series with flat time-end jets and ACω Sard give auxiliary paths onto at EVERY nonconstant zero-mark strip, over ALL actual indices and component sectors. This uses no global finite-domain assertion. Cut the auxiliary paths to their fixed endpoint values in shrinking time collars. Flatness gives smooth convergence. If nononto zero-mark strips of energy at most E existed for infinitely many late cuts, [F6]'s actual compactness would give a finite auxiliary limiting chain. Degree additivity and positivity exclude every zero-mark sphere and one-torus disk, because every one-system cut component contains a basepoint. All retained nonconstant levels are onto, with the same exact transverse end operators. Patch their actual right inverses and whole transverse neck inverses by squared cutoffs. Fixed-core operator differences tend to zero; each neck commutator is O(1/T) and exact-generator tail errors tend to zero. Taking component-core transitions long, then the subsequence late, makes the finite sum of errors smaller than one; Neumann correction gives a right inverse, a contradiction. No bound on the unknown indices of these strips was used. If the entire limit is constant, energy tends to zero; step2.1 makes every integer region degree eventually zero, and the domain-energy estimate forces the original strip to be constant. Its exact transverse operator is invertible and cannot be a counterexample. Thus a late genuine-collar path is onto at every zero-mark solution of energy at most E. The same compactness argument proves openness on this budget; starting the auxiliary choice in any specified neighborhood proves density. Choose these paths for the finitely many end pairs.

4.1F2F3F4F6step 1.1step 1.3step 3.1constructalgebra

In the global branch use [F3]'s all-index end choices and coherent polygon data. Its finite face induction preserves any supplied compatible protected tensor and every supplied product angle, and its source-local Hilbert variations are positive at every finite neck length and flat at the conformal faces. Thus genericity is genuine even at large finite necks. In the strict branch use [F4]. Its finite lower-index band and unconditional bubble index budget first exclude every nonconstant strip of index at most zero; its index-one regularity then suffices for every limiting strip which can occur in a low-dimensional polygon. In the ordinary weak branch use step3.1 on the chosen budget and repeat the source-local Hilbert coefficient construction: neither universal surjectivity nor the finite coherent face induction uses global weakness or a lower index bound. It makes nonconstant polygon vertices regular at every actual index. Their relevant limiting strip levels inherit energy at most E0<E by the actual energy identity, so those strips are onto and have index at least one from their nonzero translation vector. No regularity outside that energy budget is presumed. All three branches fix genuine time-constant endpoint collars and include every allowed source parameter in the actual operator.

5.1F3F6step 1.3step 3.1step 4.1constructalgebra

For a compact auxiliary parameter manifold B with finitely many boundary charts, construct regularity jointly at its conformal faces; do not assign its dimension separately to different vertices. The autonomous end paths are fixed. On a prescribed planar face with vertices v, use the configuration space B×∏vBv and the direct-sum section (∂ˉb,vuv)v, with each strip phase treated as in [F6]. Its derivative is ((Dvξv+Cvb˙)v), with ONE shared vector b˙∈TbB. At stable vertices the ordered boundary-label sublists of a finite ordered polygon face are distinct. Thus their source-local coefficient bumps are independent even at a common target value; these are variations of different ordered coefficient systems. At each nonconstant polygon vertex the universal bump argument of [F6] kills its obstruction, while every autonomous end level within the energy budget is already onto by step3.1. Hence the joint universal derivative is onto. Its parameter projection has Fredholm index ∑vav+dim⁡B, where av is its fixed-parameter component count index, including its conformal parameters and the actual autonomous strip phase. Apply the countable Hilbert Sard construction simultaneously over integer-index strata, corner tuples and finite face trees up to the fixed maximum arity. There are countably many such map charts and only finitely many label lists and tree types. Coefficient bumps are flat at conformal faces but positive in every vertex's interior; the finite arity induction therefore retains lower prescribed face data and achieves these joint conditions. On boundary faces of B retain already jointly regular compatible data and induct over their finite face strata; source/parameter bumps vanish there, and interior parameter regularity is obtained as in the actual relative construction. This produces joint regularity, not regularity of every individual fixed-parameter polygon slice at a critical parameter.

6.1F6step 5.1constructalgebra

The joint regular zero chart also supplies actual common-parameter gluing when an individual polygon vertex is not onto at fixed b. Take one bounded right inverse Q of ((ξv),b˙)↦(Dvξv+Cvb˙)v on the finite limiting tree; its finite-dimensional parameter vector is shared, not copied. For a forcing on the glued source, use the squared cutoffs to split it into the vector of component forcings, apply this single Q, and cut and glue the resulting section components while retaining its ONE output b˙. On component cores the resulting composition is the identity by the defining joint equation for Q. Generator-chart parameter derivatives vanish, since protected geometry is fixed, and the remaining parameter coefficients decay along the ends by the rapid coefficient bounds. Consequently the commutator and mismatch estimate is exactly the finite sum C/Te+o(1) of [F6], with the extra parameter term bounded by its vanishing tail coefficient times the common finite energy norm. Neumann correction gives the augmented glued right inverse. The two cutoff maps on kernels are inverse for long necks by the same error estimate, now retaining b˙ once; thus the local zero-chart dimensions and injectivity are the actual ones. Apply the proved Newton correction in the map/source charts with that shared parameter and the neck variables. Its derivative estimates, and the converse closeness of an actual convergent sequence to its pregluing, are unchanged because the parameter ranges in a compact finite-dimensional chart. This gives an exhaustive collar modeled on the joint component zero space times its neck coordinates. In particular a zero-dimensional joint face is one broken configuration with a fixed common parameter, even when its two polygon components are not separately rigid universal spaces.

7.1F3F4F6step 1.3step 2.1step 3.1step 4.1step 5.1step 6.1constructalgebra

Use [F6] with step1.1's uniform total energy bound in the ordinary weak branch, or the proved applications [F3] and [F4] in the other branches. Degrees add at every actual region divisor. A zero-basepoint sphere has its constant surface coefficient zero; a one-torus disk has constant coefficient on each one-system cut component, each of which contains a basepoint. Their domain-area inequalities force zero energy, so no nonconstant such bubble remains, including when it is entirely contained in a protected divisor or the diagonal. Constant compact ghosts are deleted only when unstable; no stable polygon vertex is deleted by that convention. The latter cannot be constant here by step1.3. At fixed coefficients the total count dimension is the sum of the regular vertex count dimensions plus the number q of nodes, with one longitudinal phase removed only for an actual autonomous nonconstant strip. In a common-parameter family it is instead the JOINT component count dimension in step5.1 plus q, by actual cap and source-tree additivity. That joint dimension is nonnegative by regularity, even when an individual vertex has negative fixed-parameter index. Consequently dimension zero is unbroken; a non-interior dimension-one limit has exactly one node and a rigid joint two-component configuration. With no auxiliary parameter these are the two ordinary rigid components. This rules out every extra multi-level, multi-conformal or bubble escape in the dimensions used by the asserted identities.

8.1F3F4F6step 1.1step 2.1step 4.1step 5.1step 6.1step 7.1constructalgebra

Finiteness and collars are topological conclusions of the proved analytic construction, not consequences of a merely formal dimension count. An unbroken convergent sequence converges in the global map chart by the exact end cutoff inverse. A rigid total set is discrete and sequentially compact across all its literal classes, hence finite under ACω by selecting ordered finite distinct tuples and flattening them. Each matching fixed-data pair, or matching common-parameter joint configuration, has its actual nonlinear chart and unique exhaustive half-interval collar by step6.1. All components inherit the total energy bound, independently of their possible split Spin-c labels; step2.1 bounds their domains without reversing the area inequality. The rigid joint boundary sets themselves are finite by the same compactness and joint dimension argument: a further splitting would have another node and negative joint face dimension. Thus no separate assumption of finitely many component classes or separately rigid parameterized vertices is required. Distinct positive-energy cores and strip phases separate distinct collar endpoints. The resulting Hausdorff second-countable sequentially compact one-manifold is compact; its finite interval boundary has even parity. The actual class, total Spin-c label and degrees are locally constant in map charts and carried by explicit neck sewing, so imposing any such label selects the corresponding open-and-closed part and does not assume a class/domain bijection.

9.1F3F4F6step 1.2step 4.1step 5.1step 6.1step 7.1step 8.1algebra

For a polygon with d inputs the count dimension is the actual fixed-map index plus d−2 conformal parameters; a triangle has no such parameter and no translation quotient. The rigid operations are therefore the index-one strip differential and the polygon counts of index 2−d. Enumerating the actual one-node ends in step8.1 gives precisely the finite A-infinity sum, including the output differential and every input differential. This proves differential square-zero, the triangle chain-map equation and the quadrilateral composition homotopy. For a compact auxiliary family use the actual joint construction in steps5.1–6.1. Its broken boundaries are the zero-dimensional common-parameter fiber configurations, not in general products of individually rigid parameterized polygon spaces; its prescribed parameter boundary strata are included as well. The parameter is counted once. Fixed-data A-infinity products remain precisely the ordinary products already enumerated. In particular universal index-minus-one triangles are rigid and universal index-zero triangles have their endpoint and strip-breaking boundaries. Counts of those boundaries give the asserted homotopy identities. Zero-basepoint boundary-degeneration terms are zero by step7.1, hence no omitted contribution remains to cancel. The analytic theorem supplies these identities for the actual admissible diagrams and inputs; computing named top cycles, constructing surgery multi-diagrams, identifying filling homology and proving move inverses require their own proofs and have not been assumed here.

10.1F5F6step 1.1step 1.2step 2.1step 3.1step 5.1step 6.1step 7.1step 8.1step 9.1constructalgebra∎

For continuation retain the compact-transition source profile as part of the equation. The global branch is [F5]. For ordinary weak specified-sector data reconstruct its analytic proof on step1.1's energy budget rather than apply its global-admissibility Statement. The actual anchored compactness has one unshifted central component and autonomous escaping levels; zero marks exclude bubbles by step2.1's degree argument. Universal source-dependent bumps make the central augmented operator onto on all index strata. Fixed ends only need regularity within the energy budget, supplied in step3.1; constants in a protected generator chart have the exact invertible transverse operator. For a relative parameter family fix its regular endpoints and use parameter-interior source bumps; endpoint zero charts are already regular. The flat time and rapid source smoothing series can be cut to genuine time collars and compact source transitions: a nononto sequence of energy at most E has an anchored auxiliary limit with onto augmented central operator and onto autonomous end levels. The squared-cutoff inverse retains the one central parameter output and has error O(1/T)+o(1); its parameter forcing is zero on generator necks and uniformly vanishing on escaping cores. Neumann correction contradicts nononto-ness. An entirely constant limit is handled by step2.1 and the zero-domain energy estimate, and its constant operator is already invertible. Thus a late genuine profile or family is regular at every relevant energy-bounded zero. The central rigid continuation operator has index zero with NO translation quotient. A count-one fixed profile has a rigid central zero and one autonomous rigid end strip; a relative one-parameter count has a fixed-map-index-minus-one central zero and the parameter endpoints. Actual anchored product collars, global tail convergence and endpoint parity prove the continuation equations exactly as in [F5], now with these bounded-energy hypotheses checked. For separated compositions retain the two unshifted regular profiles and send their separation to infinity. Every relevant total class was included in step1.1, so the same energy bound gives its actual two-central-component limit; the cutoff inverse and joint kernel matching give the actual glued products and exhaust them. Parameter comparison with a direct profile uses the same bounded-energy relative argument. The constant longitudinal profile has only constant index-zero solutions because every nonconstant autonomous solution within the budget is onto and has its nonzero translation vector. Its exact constants count the identity. This proves the required continuation chain maps, composition and inverse homotopies within the specified sector without importing a global index bound. Together with the polygon and direct point-target arguments, it proves the full analytic assertion; named top cycles, filling identification and Heegaard-move invariance remain independent operation duties.

AlgorithmAlgorithm: Literature-sourcedProof: Not applicableverified 2026-10-22 (gpt-6.1-sol)Open item page →

Finite grid complexes for knot Floer theory

Definition

The grid-complex algorithm takes a finite encoded grid and a mode from {tilde,minus-associated-graded,minus-filtered}. It requires an integer grid number n≥2, disjoint O- and X-markings with exactly one of each in every row and column, and an induced component permutation that is one n-cycle. If n<2 or the marking conditions fail, return invalid-grid; if the component permutation has more than one cycle, return link-input-not-supported. No chain complex is returned on either branch.

For a valid knot grid, number the horizontal and vertical circles 1,…,n and enumerate all permutations of {1,…,n} in lexicographic order. The state for a permutation σ consists of the n intersections (i,σ(i)). For each ordered pair of states x,y, if they agree in other than exactly n−2 points, add no rectangle terms. Otherwise enumerate the at most two torus rectangles with the four nonshared state points as corners, retain those whose interiors contain no point of x, and apply the selected mode. In the full filtered mode, each term drops filtration by nX(r); the associated-graded mode retains precisely the terms with zero drop, which are the rectangles with no X-marking:

  1. Tilde mode. Work over F2. Count a retained rectangle with coefficient 1 exactly when its interior contains no O- or X-marking.
  2. Associated-graded minus mode. Work over F2[U1,…,Un], with each Ui of Maslov degree −2 and Alexander filtration degree −1. Count a retained rectangle exactly when its interior contains no X-marking, with coefficient ∏i=1nUiOi(r).
  3. Full filtered minus mode. Work over F2[U1,…,Un], with each Ui of Maslov degree −2 and Alexander filtration degree −1. Count every retained rectangle, regardless of its X-markings, with coefficient ∏i=1nUiOi(r). Use the grid Alexander value for each state and assign filtration drop nX(r) to this term.

Return the ordered state list, the mode and coefficient ring, the exact differential matrix obtained by summing coefficients modulo two, the state Maslov and Alexander values, and the filtration data for the full filtered mode. The algorithm returns this finite presentation only; its identification with knot Floer homology and its independence of the chosen grid are separate theorems. The factor V⊗(n−1) is not multiplied into the output.

LemmaStatement: Literature-sourcedProof: AI-generatedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Knot Floer duality and exchange of the two filtrations

Statement

Assume countable choice. Let K⊂S3 be a smooth oriented knot and let C(K)=CFK∞(K) be its normalized bifiltered complex over F2[U,U−1]. The complex of the reverse mirror −K is filtered-chain-homotopy equivalent to the dual complex C(K)∗=Hom⁡F2[U,U−1](C(K),F2[U,U−1]), with dual differential dϕ(x)=ϕ(∂x) and dual basis elements at negated bifiltration coordinates. There is also a Maslov-graded filtered-chain-homotopy equivalence from C(K) to the complex obtained by exchanging the algebraic and Alexander filtrations. These are distinct symmetries: duality negates both filtration coordinates, while conjugation exchanges them.

Facts & Assumptions

Given: The smooth oriented knot, its normalized complex, and the ACω disk-count branch stated in the frontmatter. All coefficient modules below use F2.

[F1]

Actual admissible doubly pointed S3 diagrams have the full filtered minus count complex, retaining positive multiplicities at BOTH marks. The differential counts actual index-one classes with coefficient Unw; its degree is minus one. The accepted analytic construction supplies finite counts and square zero. Admissible multipointed Heegaard data for knot Floer complexes Regularity, compactness, and gluing for pointed Heegaard disks

[F2]

Every two generators of a balanced S3 diagram are joined by an actual Whitney class, and the integer relative gradings satisfy M(x)−M(y)=μ(ϕ)−2nw(ϕ) and A(x)−A(y)=nz(ϕ)−nw(ϕ). The Seifert normalization is the actual NEGATIVE modified Morse field with its zero-relative-Euler cap; it is carried by oriented diagram data. Total zero-w hat homology is one-dimensional and its sole class has normalized Maslov degree zero. Balanced three-sphere knot generators admit consistent relative gradings Doubly pointed knot Morse fields have a zero-framed Seifert extension Relative three-sphere Maslov gradings have a unique hat homology normalization

[F3]

Between actual doubly pointed diagrams of the SAME oriented knot in S3 there are full minus filtered equivalences with filtered inverse homotopies, preserving the absolute Seifert origin and normalized Maslov degree. Every term Uay has a≥0 and A(y)−a≤A(x); thus localization is nonincreasing in both bifiltration coordinates. Doubly pointed three-sphere moves preserve the Seifert-normalized filtration

[F4]

Write C:Sym⁡g(Σ,j)→Sym⁡g(−Σ,−j) for the map retaining the same unordered physical points. A conjugate surface coordinate is zˉ, so in unordered-root charts C conjugates every elementary symmetric coefficient. It is therefore a smooth real diffeomorphism even at collisions, and is antiholomorphic for the standard induced complex structures. The strip-equation, linearized-operator and marked-degree correspondences for the two transformations are proved directly in steps3.1 and4.2.

[F5]

Reversing the ambient orientation and the knot orientation is exactly the reverse-mirror convention in the Statement. Reverse mirror of an oriented knot

Proof

1.1F1F2givenconstructalgebra

The actual bifiltered model. Choose one supplied doubly pointed diagram H=(Σ,α,β,w,z) and good count data in [F1]. Write R=F2[U,U−1]. Its finite free localized module has generator x in coordinates (0,A(x)), and Ukx in (−k,A(x)−k) of Maslov degree M(x)−2k. Equivalently write [x,i,j] with j−i=A(x) and degree M(x)+2i. Its differential is the actual sum of [y,i−nw(ϕ),j−nz(ϕ)] over index-one classes ϕ:x→y. This dictionary follows directly from [F2], and records both positive marked degrees. All ensuing comparisons concern this finite free module and actual classes, not an abstractly postulated symmetry.

2.1F1F2F5step 1.1construct

Geometric diagrams and their knot labels. Swapping the two attaching systems while keeping the oriented surface gives H∨=(Σ,β,α,w,z). Physically the two handlebodies exchange their roles and both of their ambient orientations reverse, so this is (−S3,−K): the old beta bridge is now traversed from w to z and the old alpha bridge from z to w. After the fixed orientation-reversing ambient reflection, it is the reverse mirror [F5]. The compatible Morse function changes from f to −f. The diagram Hc=(−Σ,β,α,z,w) instead has the ORIGINAL ambient orientation (surface reversal cancels the handlebody exchange), and the ORIGINAL knot orientation (mark exchange cancels the bridge reversal). Explicitly, the new alpha arc is the OLD beta arc directed from old z to old w, and the new beta arc is the OLD alpha arc directed from old w to old z. Both are their ORIGINAL directions; only the starting point of the closed traversal changed. The surface reversal restores the original orientations on both physical handlebodies after their role exchange, so their identity gluing identifies the ambient oriented manifold with the original one. Thus this is the same oriented (S3,K) with its carried bridge disks, not an assumed oriented isotopy from K to its reverse. All complete compression data remain present. Their normalized periodic domains have zero coefficient at one mark precisely as before, and reversing labels or region orientations transports the admissibility sign tests. Use the transported count data rather than require the same arbitrary almost-complex path on every diagram.

3.1F1F4step 2.1constructalgebra

Actual dual classes and counted operators. For ϕ:x→y represented by u(s,t) put u∨(s,t)=u(−s,1−t) and Jt∨=J1−t. Its ends are y,x and its boundary lies on beta at t=0 and alpha at t=1. The equation becomes us∨+Jt∨ut∨=−(us+J1−tut)=0. The source map reverses both real coordinates and is orientation preserving; every individual divisor degree nw,nz is unchanged. Precomposition is an invertible map of the actual weighted domains and targets of the linearized real operators, so kernels, cokernels and Fredholm indices agree. It reverses the translation parameter but induces a bijection of its orbits, hence of the actual zero-dimensional counted moduli spaces. It also gives the inverse on literal continuous classes and identifies broken ends. Thus if ∂x contains Uay with coefficient e∈F2, the actual dual diagram differential contains Uax∨ in ∂y∨ with the SAME coefficient. This is exactly the R-linear transpose dϕ(x)=ϕ(∂x), without discarding any positive marked terms.

4.1F2F5step 2.1step 3.1constructalgebra

Dual absolute origins. On the physical exterior, f↦−f negates the selected negative Morse field while the ambient orientation also reverses. Its perpendicular plane therefore retains its physical orientation: its orientation is determined by the ambient orientation and the field direction, and both changed sign. Its determinant and Euler obstruction on the physical surface are unchanged. The reverse-mirror Seifert surface has the opposite surface orientation, so the capped pairing changes sign. The zero-Euler cap stays zero under this transport (a constant reversal of its boundary frame has winding zero). Hence A∨(x∨)=−A(x) with NO affine shift. The relative index equation and [F2]'s Whitney connectivity give M∨(x∨)=−M(x)+d for one integer d. After setting U=0, step3.1 is the finite vector-space dual of the actual TOTAL hat complex, including its positive-z differential. The homology of this dual is the dual of the original homology: a functional is a cycle exactly when it vanishes on boundaries, and quotient by functionals vanishing on cycles gives the dual of cycles modulo boundaries. Its sole homology class has degree zero when the basis degrees are −M(x). Uniqueness of [F2]'s hat normalization therefore gives d=0. Sending x∨ to the R-dual basis functional x∗ is now an actual graded chain isomorphism with dual basis coordinates (0,−A(x)). An arbitrary shifted dual basis has coordinates (−i,−j), exactly as stated.

4.2F1F2F4step 2.1step 3.1constructalgebra

Actual conjugate classes. For Hc use uc(s,t)=C(u(s,1−t)) and Jtc=−C∗J1−t. Direct substitution gives usc+Jtcutc=C∗(us+J1−tut)=0. This time the ends x,y are unchanged. Source t reflection and target complex conjugation each reverse the relevant orientation, so their two signs cancel in every divisor degree. Since the mark labels were exchanged, nwc(ϕc)=nz(ϕ) and nzc(ϕc)=nw(ϕ). The same invertible real-operator conjugacy as step3.1 gives the actual index and count equality; C is defined by conjugating the coefficients of each unordered-root polynomial, not by the generally nonsmooth symmetric power of an arbitrary surface map. The protected integrable divisor neighborhoods, product generator charts and taming forms are transported to the conjugate surface, so the chosen data remain in the allowed class. On the physical exterior f changes to -f with the SAME ambient orientation, so the negative selected field reverses and its perpendicular plane is conjugated. The knot and Seifert surface orientations are unchanged by step2.1. The cap again has relative Euler zero. Thus its actual capped Chern pairing gives Ac(xc)=−A(x).

5.1F2step 1.1step 4.2constructalgebra

The raw swapped map and its one possible Maslov shift. Define S[x,i,j]=[xc,j,i]. Step4.2 and Ac=−A make this an R-linear chain isomorphism: each term of the target differential lowers its first coordinate by nz and second by nw. It and its inverse interchange the two filtrations EXACTLY. The actual relative index equations give Mc(xc)−Mc(yc)=M(x)−M(y)−2(A(x)−A(y)). By [F2]'s actual Whitney connectivity there is ONE integer c with Mc(xc)=M(x)−2A(x)+c. Hence S has homogeneous Maslov degree c, since Mc(xc)+2j=M(x)+2i+c. It is not legitimate simply to declare c=0 from relative gradings; the next step fixes it using the actual normalized equivalences. For genus zero the point complex already has A=M=0 and the assertion is immediate.

6.1F2F3step 1.1step 2.1step 5.1constructalgebra

A square fixes the absolute Maslov shift without grid input. By step2.1 and [F3], choose an actual Maslov-degree-zero bifiltered chain homotopy equivalence E:C(Hc)→C(H) with its bifiltered inverse and homotopies. Put Φ=ES on the ORIGINAL module. It is a coordinate-swap-filtered equivalence of homogeneous degree c. Its inverse S−1E−1 has the same swap filtration rule and degree -c. Therefore Φ2 and its inverse preserve the ORIGINAL bifiltration, and their inverse homotopies do also: compose the given homotopies with the swap maps on both sides, so two swaps cancel. In particular they restrict to the algebraic sublevel i≤0, the actual full minus module of step1.1; quotienting by U gives inverse chain homotopy equivalences of the actual total hat complex. The square is homogeneous of degree 2c. Its induced homology map is an isomorphism of a one-dimensional vector space supported only in degree zero by [F2]. Consequently 2c=0 as an INTEGER, and c=0. This uses neither a grid identification nor an unproved equality of the two independently normalized w/z Maslov origins. It proves that Φ itself is the required Maslov-graded coordinate-swap equivalence.

7.1F1F2F3F4F5step 4.1step 4.2step 5.1step 6.1algebra∎

Finally apply [F3] to replace the chosen actual diagrams by any normalized models of K and of its reverse mirror; localizing the finite full-minus maps and their finite homotopies preserves all their identities and both filtration inequalities. Over a finite free R-module the R-dual of such a filtered homotopy equivalence is again one: dualize its explicit maps and homotopies and negate the dual basis coordinates. Thus step4.1 proves the FULL reverse-mirror duality, including its normalized grading and negated bifiltration; step6.1 proves the distinct Maslov-graded filtration exchange. Their generator maps, literal Whitney-class bijections, index-one counts, inverses and absolute origins were constructed above. No grid/analytic equality, four-dimensional degree formula, concordance-map nonvanishing or numerical knot computation is a premise. Both symmetries hold with all the original scope and the stated ACω analytic branch.

LemmaStatement: Literature-sourcedProof: AI-generatedjudge pass (gpt-6-luna)verified 2026-10-22 (gpt-6.1-sol)Open item page →

The tau level is preserved by filtered chain homotopy equivalence

Statement

Let C and D be finite integer-filtered chain complexes over F2, each with total homology isomorphic to F2. A filtered chain homotopy equivalence consists of filtration-preserving chain maps f:C→D and g:D→C such that gf and fg are chain homotopic to the respective identity maps by homotopies that preserve every filtration level. Then τ(C)=τ(D).

Facts & Assumptions

Given: C,D and filtered chain maps f,g satisfying the filtered chain homotopy equivalence hypotheses in the Statement.

[F1]

The Alexander filtration and tau level: for a finite filtered complex with one-dimensional total homology, τ(C) is the least integer s for which H(FsC)→H(C) has nonzero image.

Proof

1.1F1givenalgebra

Put t=τ(C). By [F1], some class in H(FtC) maps to a nonzero class in H(C). Since f preserves filtration, its restriction sends FtC to FtD. The square formed by these restricted maps and the inclusions into total complexes commutes, so the induced map sends that class to its image under f∗. The map f∗ is an isomorphism because g∗ is its homotopy inverse; hence the resulting class in H(D) is nonzero. Therefore H(FtD)→H(D) has nonzero image and τ(D)≤t.

2.1F1step 1.1algebra∎

The same argument using the filtration-preserving map g gives τ(C)≤τ(D). Together with step 1.1 this yields τ(C)=τ(D).

LemmaStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Rectangle-boundary pairings and grading changes in grid complexes

Statement

Let G be a toroidal grid representing a knot, and use the three differentials defined by the grid-complex algorithm. For any fixed source and target states, every non-annular composite of two counted rectangles has exactly two ordered rectangle decompositions; the two decompositions have equal O-monomial weights and are either both counted or both excluded in each mode. Returning composites consist of thin annuli, one horizontal annulus for each row and one vertical annulus for each column. Every such annulus contains one X and one O, so it is excluded in tilde and associated-graded minus modes. In the full filtered minus mode, each horizontal annulus contributes the variable of its unique O-marking, and each vertical annulus contributes the variable of its unique O-marking; every O-variable occurs once in each family and therefore twice in total. Thus all returning terms cancel over F2. Every counted rectangle term has Maslov degree −1 and lowers Alexander filtration by exactly the number of X-markings in its interior; the modes excluding X preserve Alexander grading.

Facts & Assumptions

Given: A knot grid G, its finite set of states, and one of the three differential modes in the grid-complex algorithm.

[F1]

Toroidal grid diagrams and gradings for oriented knots gives MO(x)=J(x−O,x−O)+1, the analogous MX formula, and the one-O/one-X-per-row-and-column conditions.

[F2]

Finite grid complexes for knot Floer theory defines the tilde mode by excluding both markings, the associated-graded minus mode by excluding X and weighting each Oi by Ui, and the full filtered minus mode by allowing X and weighting each Oi by Ui; each Ui lowers Maslov grading by two and Alexander filtration by one.

[F3]

Integrality and rectangle changes of the grid Alexander grading proves the Maslov rectangle-change formulas and AG(x)−AG(y)=nX(r)−nO(r) for every rectangle from x to y.

Proof

1.1F1givenalgebra

For two successive rectangles r1,r2, let ψ=r1+r2 be their cellwise sum as a two-chain, and let x,z be the initial and terminal states. The state permutation changes by one transposition at each rectangle. If x≠z, their relative permutation is either a product of two disjoint transpositions, so four points move, or a 3-cycle, so three points move. These are the only nonidentity products of two transpositions.

1.2F2F3algebra

For a counted rectangle r the interior contains no state point, so [F3] gives MO(x)−MO(y)=1−2nO(r). The coefficient ∏iUiOi(r) lowers Maslov degree by 2nO(r) by [F2]; hence the full term has degree −1.

2.1step 1.1construct

In the four-point case, the two transpositions are disjoint and commute. Exchanging their order gives the alternate intermediate state; the first rectangle in one decomposition has the same support as the second rectangle in the other, and conversely. Thus both decompositions have the same two rectangle interiors and are empty together. No third decomposition is possible, because the four distinct changing corners must be paired into the same two disjoint transpositions fixed by the endpoint permutation and the domain boundary.

2.2step 1.1construct

In the three-point case, following the boundary of ψ gives a six-corner region with five right-angle corners and one reflex corner of angle 270∘; every cell coefficient is zero or one. At the reflex corner the boundary can be cut in exactly two ways into two rectangles. Each cut gives the required intermediate grid state, and the resulting interiors contain no additional state point because any such point would already lie in the interior of one of the original empty rectangles. The six-corner boundary has no other concave corner at which to make a cut, so these are exactly the two decompositions.

2.3F1F2step 1.1algebra

If x=z, the two rectangles reverse the same state transposition and their union is a periodic annulus. If the annulus were not thin, its transverse width would include at least two grid strips; the interior would then contain a state point on an intervening grid circle, contradicting emptiness. Hence every returning domain is one of the n horizontal thin annuli, each occupying one row strip, or one of the n vertical thin annuli, each occupying one column strip. Conversely, the two state points on the boundary circles of any one-row strip split that strip into two empty rectangles, and the same holds for every one-column strip, so all 2n listed annuli occur as returning composites. Each contains exactly one O and one X by the row and column marking conditions, so it is excluded in tilde and associated-graded minus modes. In the full filtered minus mode, the horizontal annulus through the row of Oi contributes Ui, and the vertical annulus through its column also contributes Ui. Every Oi therefore occurs twice among all returning domains, and the total is zero over F2.

3.1F2step 2.1step 2.2algebra

In either non-annular case the two decompositions have the same cellwise domain ψ. A forbidden O- or X-marking occurs in one rectangle of a decomposition exactly when its cell has positive coefficient in ψ; hence a mode excluding that marking accepts either both decompositions or neither. For each Oi, the exponent in the product coefficient is its multiplicity in ψ, so the two accepted terms have the same monomial. They cancel over F2.

4.1F2F3algebra∎

The Alexander change in [F3] is AG(x)−AG(y)=nX(r)−nO(r). Multiplication by the coefficient lowers filtration by nO(r), so the output filtration is AG(x)−nX(r). Thus the full filtered term drops by exactly nX(r), and the tilde and associated-graded modes, which count only rectangles with nX(r)=0, preserve Alexander grading.

TheoremStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Termination and exact output of the grid-complex algorithm

Statement

For every finite encoded grid and every mode in the allowed mode set of the grid-complex algorithm, execution terminates. If the grid is invalid, the procedure returns exactly the specified invalid-grid or link-input-not-supported branch. If the input is a valid one-component grid, it returns the stated coefficient ring, all grid states in the declared order, and exactly the matrix entries prescribed by the selected rectangle mode, with the declared gradings and filtration data. This theorem asserts no topological invariance or identification with a Floer homology group.

Facts & Assumptions

Given: A finite encoded grid, a selected mode, and the exact validation and enumeration instructions in the algorithm.

[F1]

Finite grid complexes for knot Floer theory checks the grid number and disjoint one-per-row/column markings, determines the knot component count, and uses the allowed mode to filter the finite state and empty-rectangle lists and compute the output coefficients exactly as stated.

[F2]

Toroidal grid diagrams and gradings for oriented knots identifies the state set with the permutations of {1,…,n} and defines the torus rectangles used by the algorithm.

Proof

1.1F1construct

The validation stage scans the finite n×n marking array, counts the markings in each row and column, and traces a permutation on the finite set of n markings. Each scan and each cycle check terminates. A failed size, marking, or disjointness condition returns invalid-grid; a valid marking array with more than one component cycle returns link-input-not-supported, exactly the two specified rejection branches.

1.2F1F2construct

On a valid knot input, the algorithm enumerates the n! permutations. There are finitely many ordered state pairs; a pair has at most two candidate torus rectangles by [F2]. Testing whether a candidate is empty and counting its finitely many markings takes finitely many array operations. Every mode filter and coefficient calculation is therefore finite, so all matrix entries and grading lists are produced in finite time.

2.1F1F2step 1.2algebra∎

By [F2], every grid state appears exactly once in the permutation enumeration and every rectangle from a pair is among the enumerated torus rectangles. The empty-state test is the definition of Rect⁡∘; the mode-specific marking test and coefficient are the instructions in [F1]. Summing the accepted terms modulo two therefore gives exactly the specified differential matrix, and the remaining output fields are computed from the stated grading formulas.

TheoremStatement: Literature-sourcedProof: AI-generatedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Hat Heegaard Floer homology is a closed three-manifold invariant

Statement

Assume ACω. For every closed connected oriented smooth three-manifold Y, the homology of the hat Heegaard Floer complex over F2 is independent, up to isomorphism, of the pointed admissible Heegaard diagram used to define it. The assertion is invariance of the group HF^(Y;F2); it does not assert a diagram-independent filtered chain homotopy type.

Facts & Assumptions

Given: Countable Choice, Y and two pointed admissible diagrams in the corrected full-module sense. No torsion or rational-homology-sphere hypothesis is imposed on Y.

[F1]

Full-module admissibility tests EVERY intended Spin-c sector, even an empty-generator one. The finite Heegaard Spin construction supplies a determinant-trivial sector, so this is equivalent to the global mixed-sign condition. Every supplied complete geometric diagram has the actual finite winding preparation to a globally weak one. Actual globally weak chosen hat complexes and their square-zero finite differentials exist. (Hat Heegaard Floer complex of a closed three-manifold, Complete Heegaard data have a zero-Chern Spin-c sector, Globally weak closed Heegaard data admit a finite chosen hat complex, Complete pointed Heegaard diagrams admit globally weakly admissible isotopies)

[F2]

The actual circular top cycles, top-triangle products and quadrilateral counts give handleslide chain maps with the displayed explicit inverse homotopies. Actual attaching isotopies have regular-interval transport and birth/death equivalences. Their proof supplies the uniform paired-dual winding preparation and explicit unit-pair cancellation. (Globally weak hat handleslides have explicit inverse homotopies, Globally weak hat attaching isotopies have actual inverse maps)

[F3]

Ordinary stabilization in a disk in the basepoint region has literal inverse chain isomorphisms x↦x×c and x×c↦x, with zero inverse homotopies. Its separator proof is applied only in that region. (Hat Heegaard stabilization has explicit inverse chain maps)

[F4]

Transported diagram data identify actual counts by inverse chain isomorphisms. Actual fixed-background continuation now allows independently supplied generator jets and full boundary tensors. After that operation aligns chosen canonical data, the stated fixed-background full-boundary and changing-background theorems give actual equivalences with their common generator prescriptions. (Closed pointed diagram isomorphisms identify the actual hat count operators, Fixed-background hat complexes with different full boundary values are equivalent, Changing complex backgrounds gives actual zero-marked continuation equivalences, Fixed-background hat continuations allow independent generator data)

[F5]

Any two supplied complete pointed splittings of this Y are joined by the actual finite general-manifold based geometric move theorem. It makes no assertion that its intermediate diagrams are admissible. (Based Heegaard moves generate closed three-manifold diagram equivalence)

[F6]

Broad auxiliary disks have same-boundary neat preparations and smooth standard cut balls. The finite graph model embeds an abstract handlebody in S3 for this preparation. Smooth Schoenflies there supplies actual smooth balls. Marked planar cap transfers are actual disk slides, and two rooted-tree reductions of one extended ball cut are connected by actual marked slides. (Proper compression disks admit neat witnesses with standard cut balls, Smooth Schoenflies for an embedded two-sphere in the three-sphere, Marked planar cut-surface regions are realized by disk slides, Empty cut-ball births erase to balanced disk slides)

Proof

1.1F1F4givenconstructalgebra

By [F1] both full-module admissible endpoints are globally weak. This implication uses the zero-Chern sector whether or not it is represented by a generator; testing existing generators alone would be insufficient. Use the actual generic hat data of [F1], retaining any independently supplied generic endpoint count choices. Generator charts are auxiliary parts of the construction, and need not have been agreed in advance. The independent-generator operation in [F4] first compares arbitrary supplied data at its OWN fixed background and area form to the canonical choices used in the finite comparisons below. Transported charts on actual diagram isomorphisms still give literal operator identifications. All subsequent comparisons are the actual operations in [F4]. Every comparison below is between actual finite count complexes. No knot-specialized result is substituted for a closed-manifold operation.

1.2F6constructalgebra

First prove the based complete-disk comparison needed on ONE handlebody. Prepare two systems as neat auxiliary witnesses and make their intersections transverse. A handlebody is irreducible using exactly [F6], without any irreducibility assumption about Y: carry its finite graph neighborhood to the standard handlebody model in S3. An interior sphere bounds the two Schoenflies balls. The connected boundary of the handlebody lies on one side, so the ball on the other side misses that boundary and lies inside the handlebody, since its collar at the sphere does. Remove closed disk intersections as follows. Take a circle innermost on the first family over the entire second family. Its cap misses that second family. Replace the second disk's interior cap by a parallel of the first cap and round. Their sphere bounds the just proved interior ball. No other proper disk lies inside that ball, since its boundary is on the exterior handlebody boundary. Thus the boundary-fixed cap transfer preserves the whole second system's topological ball cuts and reduces the circle count. Only auxiliary witnesses change, with the same boundary curves. If using the topological cap transfer, the NEW smooth cut spheres have their actual smooth parametrizations from [F6]; no ambient smooth tangency change or arbitrary TOP smoothing is asserted. Iterate finitely.

2.1F6step 1.2constructalgebra

An outermost arc on a second-family disk now bounds a cap B missing the first family. Surgery of its first-family disk Di along B gives two disks pushed disjoint from that family. Cut the other first-family disks; the remaining manifold is a solid torus. Exactly one surgery disk is an essential meridian: their oriented relative classes add to the primitive old meridian, while an essential simple disk boundary is a primitive meridian and an inessential one bounds a ball by step 1.2. Two nonzero primitive classes could sum only to zero or twice a primitive class. Replacing Di by the essential disk Di′ is therefore a complete system. In the original cut ball Di′ separates the two faces of Di; otherwise it would cut off a ball after regluing those faces and would be inessential. Choose its boundary cap not containing z. It contains exactly one Di face and finitely many other faces. Apply [F6]'s planar supplier successively with exactly two inner faces, rather than once to this whole cap. Choose a finite tree of disjoint joining bands in the cap, rooted at its unique Di face, and process it outward from the root. At each stage the frontier of a small neighborhood of the processed subtree is the moving rim. A thin neighborhood of that moving face, the next unused face and their joining band is a disk with two holes whose interior misses every other current face and z. Its two labels are the current moving disk and a retained Dj with j≠i, so the supplier's distinct-disk hypothesis holds with r=1. Its completed slide preserves the cut ball and the retained disks. Use its finger-retraction identification, supported at that face and band, to carry the unprocessed cap and its remaining faces to the new cut-ball boundary; outside that neighborhood the unused faces are unchanged. Thus the next stage has the same description. If both faces of a retained disk occur, the second remains an unused face of that retained disk and is processed by a later two-face application; it is never paired with its other face in one application. A tiny auxiliary trivial arc near z, with second endpoint w, can be fixed before this construction and missed by all these cap neighborhoods, so the supplier's marked arc condition is retained. After finitely many stages the moving rim is the frontier of the whole tree neighborhood, and the remaining empty annulus in the cap gives the isotopy to ∂Di′. This is an actual based slide word, not just disk surgery or homology. The replacement uses one old subdisk and a parallel of B, so it strictly decreases the total remaining arc count. After finitely many exchanges the two families are disjoint. Their union cuts the first ball into balls: each proper disk has two rounded smooth sphere boundaries in the standard ball model, and [F6] gives both balls. Choose a tiny trivial arc near z missing this finite union, with a second endpoint w. The two original systems are two rooted-tree reductions of that common extended ball cut. [F6] compares them by slides fixing z,w and the arc. Forget the auxiliary w and arc. This proves complete-system comparison with no genus change, including empty genus-zero systems.

3.1F2step 2.1constructalgebra

Prepare a finite same-system slide/isotopy word while its opposite system is fixed. Choose paired dual-loop annuli for that fixed complete system as in [F2], avoiding the initial region representatives. There is ONE winding number making every prescribed diagram and every isotopy interval in this finite word globally weak. At finitely many diagram vertices, test evaluations are rational linear functions of the opposite boundary coefficients, whose injectivity follows from the complete moving system. For each compact isotopy interval choose the test points to be regular values of its finite moving traces. Their finite preimage counts bound all partial traces uniformly. Across a slide there are only finitely many new endpoint relation matrices; take the maximum of these finitely many linear bounds. Thus the same ±N calculation in [F2], with N larger than that maximum, supplies strict opposite signs for EVERY nonzero periodic chain throughout the word. The initial winding path remains globally weak because its supports miss every initial region representative. On an isotopy interval apply the actual maps of [F2]. At a slide use its actual circular representative, which is globally weak by that theorem; it is isotopic as a whole family to the prescribed slid system. Both endpoints of this last isotopy are globally weak for the wound opposite family, so its actual isotopy equivalence aligns them. A joint own-family isotopy is factored into single-curve ones by a finite subdivision into graphs in disjoint annuli. This proves actual equivalences along the prepared word, without counting on an inadmissible slice.

4.1F2F4step 2.1step 3.1constructalgebra

Consequently any two globally weak diagrams on the SAME embedded splitting have equivalent chosen complexes. Apply step 2.1 to their alpha systems. Wind the first beta system to prepare that finite alpha word by step 3.1, and perform it, ending with the target alpha and the wound beta, globally weak. Now apply step 2.1 to that beta and the target beta. Wind the fixed target alpha to prepare this beta word, and perform it. Its result is (α′,βtarget), globally weak, where alpha-prime is isotopic to the target alpha. The target diagram itself is globally weak. The two-endpoint isotopy theorem [F2] therefore removes this last alpha winding while keeping the target beta fixed. The same construction covers mere changes of auxiliary disk witnesses: their boundary curves did not change, and all counts depend on those curves rather than the chosen interior preparation. Every map and its inverse homotopy in this comparison is the actual one in [F2] or [F4].

5.1F3F4step 2.1step 4.1constructalgebra

The location restriction in [F3] causes no gap. For a geometric stabilization elsewhere, choose a path on the old surface carrying its small disk to a disk in the z-region, avoiding z. A compact supported surface flow and its two-sided collar extension carry the stabilizing ball and its torus bubble along that path. This is an actual ambient transport of the stabilized splitting fixing z and carrying both handlebodies. It need NOT preserve their attaching systems. After transport, those systems and the standard z-region stabilized systems are complete systems in the SAME two handlebodies. Step 2.1 gives their based same-genus slide/isotopy comparison, and step 4.1 supplies its actual equivalences whenever globally weak chosen representatives are used. Hence [F3] is applied only at a z-region disk; all other location changes use actual same-splitting moves. In particular no assertion that nq=0 at a point in an arbitrary old region, and no unproved target-degeneration gluing, has been inserted into the product stabilization proof.

6.1F1F3F4F5step 4.1step 5.1constructalgebra

Take the finite geometric word of [F5]. For every supplied intermediate splitting choose a complete globally weak diagram on that splitting by the finite winding theorem in [F1]; at the two ends use the given admissible diagrams. This preparation changes the disk systems, not the splitting or Y. A genus-preserving step or ambient diagram transport now compares diagrams on the same splitting after the actual transport, so step 4.1 applies. For a genus increase first stabilize its chosen lower diagram at a z-region disk by [F3]. Step 5.1 identifies its higher splitting with the prescribed higher splitting; the actual transported data are identified by [F4]. Step 4.1 then compares this higher global diagram with the chosen global diagram there. For a genus decrease use exactly the reverse equivalence, including the literal inverse P of [F3]. Thus there is an actual chain homotopy equivalence for each step of the word. Intermediate geometric diagrams themselves were never assumed admissible and received no count matrix.

7.1F1F2F4step 1.1step 6.1constructalgebra

Match the actual analytic choices, including independently supplied generator data and area forms. Fix one set of smooth separate-root generator charts for each of the finitely many geometric diagrams. Construct a canonical reference on its off-diagonal locus as follows. In each two-dimensional surface factor choose a smooth positive tensor with the required two real axes at its crossings. Relative to j take its Cayley matrix separately in each root factor, multiply by a symmetric configuration cutoff equal to one near the two tori and zero near the protected discriminant and point divisors, and retain the block diagonal matrix. It is permutation-equivariant, is the standard symmetric tensor near the protected set, and extends smoothly there because the cutoff vanishes on a neighborhood. Each two-dimensional block is positive; hence this reference is tame for the product form of EVERY positive surface area form. It need not be asserted to be a symmetric power of a nonholomorphic surface map. For the finite list of forms used by the actual small-triangle operations, choose generic perturbations sufficiently close to this reference to retain all their positive tameness margins and its prescribed generator and boundary tensors. First apply [F4]'s independent-generator continuation to EACH arbitrary endpoint at its OWN fixed j and area form, obtaining such a canonical complex. This step requires no original endpoint to be tame for another area form. Now canonical choices at a shared diagram are tame for one common selected form and have the common prescribed generator charts. Align their full torus values at fixed background by [F4] if needed, and apply its changing-background comparison with the common compatible boundary tensor, retained charts and ONE fixed form throughout that comparison. The raw-identity trivialization near the tori keeps the prescribed tensor there. Model operation data may likewise be chosen close to their block product-integrable references, and the same first continuation compares them to these canonical endpoint data at their own form. Thus the maps in step 6.1 compose on the actual endpoint complexes. This supplies the independent-chart and finite-area junctions explicitly, instead of extending a common-chart theorem by omission.

8.1F1F2F3F4step 6.1step 7.1constructalgebra∎

Compose these finitely many actual equivalences. Reverse their inverse maps in the reverse order; adding the transported finite chain homotopies gives inverse homotopies for the composites, by direct expansion of gf+I=dK+Kd. The resulting chain maps induce inverse isomorphisms on homology. This identifies the two hat groups for the given arbitrary closed oriented Y. Genus zero has the empty generator and zero differential by [F1]; the same stabilization and inverse maps include it. The conclusion is the asserted isomorphism class of the full group, with no diagram-independent filtered chain type, naturality or absolute grading inferred. The empty-generator loophole is excluded by the actual full-all-sector admissibility condition in step 1.1, not by assuming a generator must exist.

TheoremStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-6-luna)verified 2026-10-22 (gpt-6.1-sol)Open item page →

Invariance of the multipointed Heegaard Floer complex

Statement

Assume countable choice. Let K be an oriented knot in S3, let F be a compact connected oriented embedded surface with boundary K, and let H0,H1 be admissible multipointed Heegaard diagrams for K. For i=0,1, the diagram Hi has a closed genus-hi Heegaard surface and mi=hi+ki−1 curves in each attaching system, with ki ordered w-points and ki ordered z-points; each component of either attaching-system complement contains one point of each type, and the connecting arcs represent K. A curve isotopy is an isotopy of one attaching curve through embedded curves in its attaching system, disjoint from the other attaching curves and all marked points. A handleslide replaces one attaching curve by its band sum with another curve in the same attaching system, along an embedded arc disjoint from the other attaching curves and the marked points. A Heegaard stabilization forms connected sum with the genus-one splitting of S3 in a disk disjoint from all curves, marked points, and knot arcs; it raises the genus by one and leaves k fixed. A basepoint-pair (index-zero/three) stabilization keeps the genus fixed and adds one curve to each system and one w,z pair along the represented knot. Its inverse removes the following elementary configuration, after relabeling the cyclic pairs: curves α0,β0 bound disks A0,B0 on the surface, are disjoint from every opposite-system curve except each other, and intersect transversely in two points; A0∩B0 is a bigon containing just w0, while A0∖B0 contains just z0 and B0∖A0 contains the next z-point along the cyclic pairing. Remove α0,β0,w0,z0 and concatenate the adjacent trivial knot arcs. The surviving marked points stay fixed. Thus the move inserts a pair along an existing knot arc rather than adjoining an isolated marked component in an empty disk. Both descriptions include the simultaneous boundary-parallel arc systems and require the stated one-w, one-z condition in each attaching complement. Destabilizations are the respective inverse moves. Comparisons may apply the orientation-preserving diagram isomorphisms of the pointed-data definition, carrying the ordered marked points, attaching systems and simultaneous boundary-parallel arcs, before or after the listed geometric moves. Their complex identifications are induced by pushforward of generators, disk classes and almost complex data, rather than by assuming an isotopy acts as the identity. Comparisons also allow the fixed-splitting arc normalizations in that definition before or after the listed moves, retaining the simultaneous bridge witnesses and all ordered marks. They leave the surface attaching and analytic data fixed, so their identification of the actual count operators is the identity; they do not assert a diffeomorphism of a tangential and transverse surface-knot pair. With these identifications, any two such diagrams are related by a finite sequence of the listed moves through admissible multipointed diagrams, preserving the represented oriented knot.

For a diagram with j basepoint pairs, regard its full filtered minus complex as a complex over F2[U] by restriction of scalars along U↦Ui for any one chosen basepoint variable. The resulting complexes of H0 and H1 are filtered chain homotopy equivalent over F2[U], preserving relative Maslov grading and Alexander filtration, even when the diagrams have different numbers of basepoint pairs. The choice of the single variable action does not affect this type. Restriction retains the independent variables in the chain module; it is different from quotienting by U1=⋯=Uj. The latter ordinary quotient is the fully collapsed complex and has filtered chain homotopy type CFK−(K)⊗V⊗(j−1), where V has zero differential and bidegrees (0,0) and (−1,−1). Equivalently, a basepoint stabilization before quotienting is modeled by the cone of Uj+1+Ui on the polynomial extension of the old complex; after the simultaneous quotient this is a zero-map cone and contributes the displayed extra factor.

For an admissible doubly-pointed diagram H1 for (K,F), define HFK^(K) to be the homology of the Alexander-associated-graded complex obtained from its full filtered minus complex by quotienting by its sole variable U. For a generator x use the Alexander grading AF(x)=12⟨c1(sw,z(x)),[F^]⟩, where sw,z(x) is the relative Spin⁡c structure assigned by the pointed Heegaard construction from the gradient trajectories through x and the chosen basepoint pair, and F^ is obtained by removing a tubular neighborhood of K from F and capping the resulting longitude with the meridian disk of the solid torus attached so that this longitude bounds it (zero-framed surgery). The vector-field representative is extended across small balls around the critical points, and its first Chern class is the Chern class of the orthogonal oriented plane field. Use the absolute Maslov normalization in which the unique generator of HF^(S3) has degree zero.

For a diagram H with j basepoint pairs, let C~(H) be the Alexander-associated-graded complex over F2 obtained from its full filtered minus complex by setting every Ui=0; its differential therefore counts only classes with all nwi=nzi=0. Its homology is HFK^(K)⊗V⊗(j−1), where V=F2⟨v0,v1⟩ has bidegrees (M,A)(v0)=(0,0) and (M,A)(v1)=(−1,−1).

Facts & Assumptions

Given: The two actual balanced diagrams, their ordered marks and simultaneous bridge data, the oriented knot identification, the surface F, and Countable Choice. Restriction of scalars and ordinary quotient have the different meanings in the Statement.

[F1]

The actual balanced knot geometric word includes the stated elementary marked-pair operation, fixed-splitting arc normalizations and carried diagram isomorphisms. Broad smooth witnesses are prepared without discarding tangential knot inputs. Every balanced S3 diagram in that word is admissible by its actual periodic-domain calculation. (Balanced multipointed knot Heegaard moves generate diagram equivalence, Balanced multipointed knot diagrams in the three-sphere are admissible)

[F2]

Actual finite full-marked square-zero count data exist, with no nonconstant nonpositive-index strips, regular indices one/two and all positive w,z multiplicities included. Their separate V-weighted matrix has the actual boundary-curvature identity. Relative M,A differences are defined on the actual connected generator set. (Regularity, compactness, and gluing for pointed Heegaard disks, Balanced knot data admit full-marked square-zero holomorphic counts, Balanced three-sphere knot generators admit consistent relative gradings)

[F3]

Actual full-minus auxiliary and attaching-move comparisons have finite polynomial filtered inverse maps and inverse homotopies, including all tangency-pair cancellations, not only a zero-z specialization. Their same-handlebody top cycle and top product have actual zero marked degree. (Full minus auxiliary continuations are filtered equivalences for balanced knots, Balanced knot attaching moves have full minus inverse maps, Same-handlebody model top counts have no positive marked terms)

[F4]

Ordinary stabilization has actual full-minus inverse maps at every permitted knot-disjoint location. Its changed-target torus matching solves the mean obstructions, proves regularity and exhausts all rigid counts; it is not inferred from the hat separator count. (Ordinary knot Heegaard stabilization has full minus inverse maps)

[F5]

The actual marked-pair operation is the polynomial cone of U_new+U_i with shifted source copy (-1,-1), including its old singly pointed horizontal component. Actual divided-difference cone retractions, the general variable-dependent differential variant, and the triangular conjugation for a nullhomotopic cone map are finite and filtered. (Marked-pair stabilization is the full minus variable cone, Basepoint cones, a single variable action, and ordinary collapse)

[F6]

Carried diagram isomorphisms identify the actual generator, class, section and count data. The permitted fixed-splitting arc normalization keeps the surface and analytic count problem fixed. (A pointed diagram isomorphism identifies the actual filtered count complexes, Admissible multipointed Heegaard data for knot Floer complexes)

[F7]

The actual negative selected-Morse field and zero-relative-Euler filling construct the Seifert-cap Alexander origin. Its difference is n_z-n_w. The finite trajectory/cap calculation preserves that origin under the actual doubly pointed move word, and the total-hat normalization fixes the Maslov shift. (Doubly pointed knot Morse fields have a zero-framed Seifert extension, Relative three-sphere Maslov gradings have a unique hat homology normalization, Doubly pointed three-sphere moves preserve the Seifert-normalized filtration)

[F8]

The actual zero-w closed hat theory is invariant for S3 and its genus-zero point model has one-dimensional homology. Its arbitrary independent auxiliary data comparison is included. (Hat Heegaard Floer homology is a closed three-manifold invariant, The standard genus-zero hat diagram of the unknot has rank one)

Proof

1.1F1F2F6givenconstructalgebra

Use [F1]'s actual finite word for the represented oriented knot. It includes precisely the move configurations and identifications in the Statement. Its witness preparation retains the broad smooth inputs; it does not assume initially neat crossings or bridge interiors missing the compression faces. Each intermediate diagram is still balanced, so [F1]'s S3 periodic calculation gives admissibility at every vertex. Choose actual good analytic data there by [F2]. There are finitely many vertices and operations; Countable Choice is more than sufficient for these choices. For isomorphisms use [F6]'s pushforward; for the allowed fixed-splitting arc normalizations use the identical surface count problem. Neither operation is silently treated as an isotopy acting trivially on unrelated generators.

1.2F2constructalgebra

Prove that the chosen variable does not matter on a FIXED diagram. For k>=2 retain [F2]'s finite polynomial matrix D(U,V). At V=1 put L_j=(partial D/partial V_j)|(V=1). Differentiating the finite curvature identity gives dL_j+L_jd=U_j+U(j-1), with indices cyclic. Each L_j has Maslov degree -1 and filtration degree at most -1: its terms have at least one z multiplicity and derivative removes one formal V factor, while their actual weighted outputs have filtration loss at least one. Sums of consecutive L_j therefore give the same homotopy for any U_i+U_l. These are actual finite count matrices; no multiplication action on homology is substituted for a chain homotopy. For k=1 no comparison is needed.

2.1F3F4step 1.1constructalgebra

For an attaching isotopy or handleslide apply [F3], and for an ordinary genus change apply [F4] or its inverse. They are equivalences over the unchanged independent polynomial ring, before restriction or collapse. All positive z terms remain in their finite matrices. Their inverse identities are actual polynomial identities and their homotopies are Alexander filtered of Maslov degree plus one. Choose the local count data supplied by each operation and use [F3]'s auxiliary comparisons at its two ends to connect to the vertex data chosen in step 1.1. This includes independently chosen background structures, area forms, full boundary tensors and generator charts. In the ordinary operation the arbitrary-location comparison uses the knot-preserving same-splitting disk word in [F4], not a zero-w neck separator at an arbitrary region point.

2.2F3F5step 1.1constructalgebra

At a marked-pair insertion use [F5]'s actual cone over R[U_new]. Its source copy has shift (-1,-1), and its matrix is the old d on each diagonal, U_new+U_i on the indicated off-diagonal and zero on the reverse diagonal after V=1. In particular the reverse full positive-z bigons cancel; they have not been omitted. Restrict to any old chosen variable. The monic-variable retraction evaluates U_new=U_i and gives the old complex, retaining its OTHER independent variables. If the selected variable is the new one, divide instead in the old U_i and apply the general polynomial-differential retraction; its quotient is the old complex with this action renamed. These retractions and their inverse homotopies are filtered, with finite divided differences. Their assertion is restriction of scalars, not the ordinary quotient before taking this cone. The inverse removal has the reverse equivalence.

2.3F5step 1.2constructalgebra

Here is the actual chosen-action equivalence, including its infinite-rank restricted module. Introduce a polynomial variable T of degrees (-2,-1) and put E=C[T]. The two cones of T+U_i and T+U_l are filtered chain isomorphic by the triangular map (a,b) to (a,b+L a), where dL+Ld=U_i+U_l from step 1.2; its own inverse has the same formula. Regard the first cone over the ring containing T and all old variables EXCEPT U_i. The general monic-variable retraction in [F5], dividing U_i, retracts it onto C with U_i replaced by T. Restricting to F_2[T] is exactly the old U_i action, not the simultaneous collapse of the remaining variables. The second cone similarly retracts to the U_l action. Compose these finite identities to obtain an F_2[T]-linear filtered homotopy equivalence. The proof is valid even though the restricted module is not finite over F_2[T]; each polynomial and each image remains finite. Renaming T as U proves independence of the chosen action.

3.1F3F4F5step 1.1step 2.1step 2.2step 2.3constructalgebra

Choose an action along the finite word, using step 2.3 whenever the label must change. Steps 2.1–3.1 supply an equivalence at every operation over that single-variable ring. Compose the forward maps and the inverse maps in reverse order. Expanding the finite chain-homotopy identities gives inverse homotopies for the composites. Their degrees and filtration bounds survive composition. This proves the full single-action comparison even when k changes, with independent variables retained throughout. It proves no equality between that module and its ordinary collapsed quotient.

4.1F5step 1.2step 3.1constructalgebra

To compute the ordinary collapsed quotient, keep one selected U and impose the other relations U_j+U=0 successively. After any earlier quotient the underlying module is still free on the generator basis over the remaining polynomial variables, and the identities in step 1.2 specialize literally. The filtered triangular cone isomorphism and general divided-difference retraction in [F5] therefore give C/(U_j+U) equivalent to C tensor V over the surviving ring at that stage. Iterate to get C_coll equivalent, over the chosen F_2[U], to C_restricted tensor V^(k-1). Step 3.1 identifies C_restricted with an admissible doubly pointed complex, denoted CFK^-(K). The factors V have exactly the source-copy shift (-1,-1) in [F5]. No flatness assertion or commutation of homology with tensor product is required: each quotient is applied to actual finite chain and homotopy identities.

4.2F1F7F8step 2.1step 3.1constructalgebra

Fix the absolute doubly pointed origin by [F7]. It is the actual relative Spin-c field with the zero-framed Seifert cap stated here, and gives the displayed half-Chern evaluation and relative n_z-n_w law. For two doubly pointed diagrams use the k=1 word in [F1]; it needs no marked-pair equalization. [F7]'s finite exterior trajectory/meridian-cap calculation fixes zero affine shift for the actual attaching maps, while its cancelling-ball calculation fixes zero for ordinary stabilization at any allowed location. Thus this word's full filtered maps preserve A_F, not just an arbitrarily shifted relative grading. Every differential term has A(U^a y)=A(x)-n_z, including n_z>0. The total U=0 complex is the closed zero-w hat complex, with the z point merely protected; [F8] gives its one-dimensional total homology. [F7]'s Maslov normalization places that class in degree zero and forces each homogeneous equivalence to have degree zero. Its inverse homotopies have degree plus one. This supplies the original absolute normalization without claiming a general rational cobordism grading theorem or absolute multipointed cone centering.

5.1F2F3F4F5F7F8step 3.1step 4.1step 4.2constructalgebra∎

Fix the multipointed origins explicitly before reading the displayed bidegrees. Start with an [F7]-normalized doubly pointed model and transport its origins along a chosen actual move word of step 3.1, using the actual smooth comparison references in [F3, F4] and the specified (-1,-1) source shift at each cone. The connected-generator relative grading in [F2] shows that any two such transported choices differ by ONE common pair of integer shifts. Specialize step 4.1 by U=0. Its left side sets EVERY independent U_i to zero; its right side is the doubly pointed total hat complex tensor V^(k-1). The maps and homotopies remain filtered identities because identities of maps are specialized, without any flatness assertion about homology. Passing to the Alexander associated graded preserves those identities. Its differential counts precisely the classes with n_w=n_z=0, by positivity and the relative formula in [F2]. This is a FINITE complex with bounded finite filtration. Its associated-graded homology is nonzero: if it vanished, a total cycle's leading filtration term would be a boundary in its graded piece; subtract that boundary and repeat down the finite filtration to make every total cycle a boundary. This contradicts the nonzero total homology HF^(S3)⊗V⊗(k−1) from [F8] and the preceding actual equivalence. This is the elementary finite-filtration spectral-sequence argument. Compare two transported choices using the normalized doubly pointed comparison in step 4.2 and their actual quotient/associated-graded equivalences. They give an isomorphism of the same finite nonzero bigraded homology with itself shifted by their common (M,A) difference. Its support is a finite nonempty subset of Z2. Such a subset cannot equal a nonzero translate: maximizing or minimizing a coordinate with nonzero translation component gives a contradiction. Hence both shifts are zero. The multipointed origins and the resulting bidegrees are therefore independent of the move word; no new absolute multipointed Spin-c theorem was used. The doubly pointed associated-graded homology is the stated HFK^(K), with step 4.2's normalization. Tensoring over F2 with zero-differential V is an actual direct sum of its degree shifts. Its homology is thus HFK^(K)⊗V⊗(k−1) with the displayed bidegrees. This proves the tilde assertion, the claimed independence, and its distinction from chosen-action restriction.

LemmaStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-6-luna)verified 2026-10-22 (gpt-6.1-sol)Open item page →

The three grid rectangle differentials square to zero

Statement

For every one-component toroidal grid and each of the tilde, associated-graded minus, and full filtered minus modes, the differential defined by the grid-complex algorithm satisfies ∂2=0 over its stated coefficient ring.

Facts & Assumptions

Given: A one-component toroidal grid G and one of the three differential modes defined by the algorithm.

[F1]

Finite grid complexes for knot Floer theory defines each differential as a finite sum over empty-state rectangles, with the stated mode-specific marking restriction and coefficient.

[F2]

Rectangle-boundary pairings and grading changes in grid complexes proves that every non-annular two-rectangle composite has two equal contributions and that the only possible annular contributions either vanish by mode restriction or occur twice with the same monomial.

Proof

1.1F1F2given

Fix initial and terminal states x,z. The coefficient of z in ∂2x is the sum over all ordered pairs of counted empty rectangles with an intermediate state. By [F2], every non-annular composite contributes in a pair of decompositions with equal coefficient, and so its total is zero in F2.

2.1F2step 1.1algebra∎

If z≠x, all composites are non-annular and step 1.1 gives zero coefficient. If z=x, the only additional composites are thin annuli; [F2] shows these are excluded in the first two modes and cancel in the full filtered mode. Thus every matrix coefficient of ∂2 is zero, for every mode.

LemmaStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

The 2-by-2 unknot grid has tilde homology of rank two

Statement

Let G2 be the toroidal 2×2 grid whose bottom row has O in its left cell and X in its right cell, and whose top row has X in its left cell and O in its right cell. With the row and column orientation convention for grid diagrams, this is a one-component unknot grid. The tilde grid complex over F2, whose differential counts empty-state rectangles disjoint from both marking sets, has homology of rank two.

OXXObottomrowtoprow

Facts & Assumptions

Given: The specified 2-by-2 marking array on a torus.

[F1]

Toroidal grid diagrams and gradings for oriented knots traces each component by following the row from O to X and the column from X to the next O; it identifies the components with cycles of this finite marking permutation and gives exactly two grid states for a 2-by-2 knot grid.

[F2]

Finite grid complexes for knot Floer theory defines the tilde differential by counting empty-state rectangles whose interiors contain no O- or X-marking.

[F3]

The three grid rectangle differentials square to zero establishes that the tilde differential is a chain-complex differential.

Verification

1.1F1givenconstruct

Starting at the bottom-left O, the oriented route follows its row to the bottom-right X, then follows that column to the top-right O, then that row to the top-left X, then that column back to the starting O. Thus the link has one component. Its planar projection has no transverse self-intersection: the four segments meet only at these marked turning points, and smoothing the four corners gives a simple closed planar curve, which bounds a disk. Therefore it is an unknot.

1.2F1F2construct

There are exactly two grid states: x={(0,0),(1,1)} and y={(0,1),(1,0)}, with coordinates read modulo two. The rectangles from x to y are the bottom-left and top-right cells, each containing one O; the rectangles from y to x are the bottom-right and top-left cells, each containing one X. No such rectangle is counted in tilde mode, and there are no rectangles from a state to itself. Hence the tilde differential is zero.

2.1F3step 1.2algebra∎

The tilde chain group has its two grid states as a basis, and its differential is zero by step 1.2. The chain condition is supplied by [F3], so its homology equals this two-dimensional chain group and has rank two.

LemmaStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

A 3-by-3 unknot grid has tilde homology of rank four

Statement

Use row indices r=0,1,2 from bottom to top and column indices c=0,1,2 from left to right. Let G3 have O-markings in cells (r,c)=(0,0),(1,1),(2,2) and X-markings in cells (0,1),(1,2),(2,0). This toroidal grid represents the unknot. Its tilde grid complex over F2, counting empty-state rectangles disjoint from both marking sets, has homology of rank four.

OXOXOXr=0r=1r=2c=0c=1c=2

Facts & Assumptions

Given: The 3×3 marking pattern and row/column indices in the statement.

[F1]

Toroidal grid diagrams and gradings for oriented knots defines the O→X→O component permutation, identifies grid states with permutations, and defines empty torus rectangles.

[F2]

Finite grid complexes for knot Floer theory defines the tilde differential by counting empty-state rectangles whose interiors avoid both O- and X-markings.

[F3]

The three grid rectangle differentials square to zero proves that this differential squares to zero.

Proof

1.1F1givenconstruct

The marking permutation sends the O in row 0 to the X in column 1 and then to the O in row 1; it continues from row 1 through column 2 to row 2, and from row 2 through column 0 back to row 0. Thus the grid has one component. In its planar projection the oriented path has vertices (0.5,0.5),(1.5,0.5),(1.5,1.5),(2.5,1.5),(2.5,2.5),(0.5,2.5) and returns to (0.5,0.5). These six axis-parallel edges form a simple polygon and meet only at consecutive endpoints, so the projection has no crossings and is an unknot diagram.

1.2F1F2construct

The six grid states, in lexicographic order as row-to-column permutations, are 012,021,102,120,201,210. The only cells containing neither marking are (r,c)=(0,2),(1,0),(2,1); each is isolated from the other two by marked cells across every side, including the torus identifications. Every marking-free rectangle made of more than one cell contains a pair of side-adjacent cells and hence a marking. Therefore the only possible tilde rectangles are these three single-cell rectangles.

1.3F1F2step 1.2algebra

For state 201, the SW and NE corners of cell (r,c)=(0,2) lie in that state and the other two corners give state 021; cell (1,0) similarly gives 201→210, and cell (2,1) gives 201→102, with the row interval in the last cell wrapping around the torus. No other state has the SW/NE pair of corners of any of these empty cells. Thus the differential matrix, with rows and columns ordered 012,021,102,120,201,210 and rows indexing targets, has rows 000000,000010,000010,000000,000000,000010. It sends only the basis vector for state 201 to the sum of the three distinct target basis vectors, so its rank is one.

2.1F3step 1.3algebra∎

By [F3] this is a chain complex. Rank-nullity gives dim⁡ker⁡∂=6−1=5 and dim⁡im⁡∂=1; hence the homology has dimension 5−1=4.

LemmaStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Basepoint variables act chain homotopically on a knot grid

Statement

Let G be a toroidal grid representing one oriented knot, and let ∂− be the full filtered minus differential over F2[U1,…,Un]. Call Oi and Oj consecutive when the X-marking in the row of Oi lies in the column of Oj. Let Xℓ be that marking and define the module map Hℓ(x)=∑y∑r∈Rect⁡∘(x,y)Xℓ∈Int⁡(r)(∏a=1nUaOa(r))y. Then ∂−Hℓ+Hℓ∂−=(Ui+Uj)id⁡. The map Hℓ has Maslov degree −1 and does not raise the Alexander filtration. Consequently multiplication by any two variables Ui,Uj is filtered chain homotopic, since the consecutive relation follows the single cycle of the knot grid.

Facts & Assumptions

Given: A knot grid G, a consecutive pair Oi,Oj, and the associated marking Xℓ.

[F1]

Toroidal grid diagrams and gradings for oriented knots gives one O and one X in each row and column and a single cyclic O→X→O marking route for a knot grid.

[F2]

Finite grid complexes for knot Floer theory defines ∂− by summing all empty-state rectangles with coefficient ∏aUaOa(r); each Ua has Maslov degree −2 and Alexander filtration degree −1.

[F3]

Rectangle-boundary pairings and grading changes in grid complexes proves that each non-annular two-rectangle domain has two decompositions with equal total O-weight and that the only returning domains are thin horizontal or vertical annuli.

[F4]

Integrality and rectangle changes of the grid Alexander grading proves AG(x)−AG(y)=nX(r)−nO(r) and the Maslov rectangle-change formula.

[F5]

The three grid rectangle differentials square to zero establishes that ∂− is a differential.

Proof

1.1F2F4givenconstructalgebra

The sum defining Hℓ is finite because there are finitely many states and at most two torus rectangles for each ordered state pair. It is F2[U1,…,Un]-linear by its coefficient formula. For each counted rectangle r, [F4] gives M(x)−M(y)=1−2nO(r); the coefficient has Maslov degree −2nO(r) by [F2]. Therefore each term of Hℓ(x) has Maslov degree M(x)−1. Also [F4] gives A(y)−nO(r)=A(x)−nX(r), and Xℓ∈Int⁡(r) implies nX(r)≥1. Thus Hℓ does not raise the Alexander filtration.

1.2F2F3givenconstructalgebra

Expand ∂−Hℓ+Hℓ∂− and group its terms by their two-rectangle domain ψ. If ψ is non-annular, [F3] pairs its two decompositions. In each decomposition, when Xℓ has multiplicity zero neither ordered term is selected; when it has multiplicity one exactly one of the two rectangles contains it and exactly one ordering is selected; when it has multiplicity two both rectangles contain it and the two possible placements of Hℓ cancel in characteristic two. The paired decomposition has the same domain, O-weight, and multiplicity at Xℓ, so every non-annular contribution cancels.

2.1F1F2F3step 1.2algebra

A returning two-rectangle domain is a thin horizontal or vertical annulus by [F3]. The only thin annuli containing Xℓ are the horizontal one in its row and the vertical one in its column. In the horizontal annulus the unique O-marking is Oi, so its coefficient is Ui; in the vertical annulus the unique O-marking is Oj, so its coefficient is Uj. Each annulus has one decomposition and contributes its coefficient times the input generator. Hence the returning contribution is (Ui+Uj)x. Together with step 1.2 this proves the chain homotopy identity.

3.1F1F5step 2.1constructalgebra∎

The marking route in [F1] orders the variables cyclically, and each adjacent pair in that cycle is consecutive as defined in the statement. Any two indices are joined by a finite path of adjacent pairs. By [F5], each multiplication map is a chain map for the differential; adding the homotopy maps from step 2.1 along the path gives a filtered chain homotopy between multiplication by Ui and multiplication by Uj.

TheoremStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Commutation and stabilization invariance of grid complexes

Statement

Let G and G′ be one-component grid diagrams related by one of the following local moves. A cyclic permutation changes the planar cut of the toroidal grid by moving the first row or column to the last. A row commutation exchanges consecutive rows whose closed horizontal marking intervals are disjoint or strictly nested; a column commutation uses the same condition on vertical marking intervals. A stabilization replaces a marked cell by a two-by-two block with the original marking type in two opposite cells and the other type in a third cell, leaves the fourth cell empty, and inserts one row and one column, relocating the old companion markings so each new row and column has one mark of each type. Any of the four empty corners is allowed, and destabilization is the inverse move. Regard each full filtered minus complex returned by the grid-complex algorithm as a complex over F2[U] by letting U act as one chosen marking variable Ui, retaining the other independent variables. The resulting complexes are filtered chain homotopy equivalent over F2[U], with equivalences preserving the Maslov grading and Alexander filtration. For each move, finite grid-domain counts give filtered chain maps in both directions and filtered chain homotopies from both composites to the identity. For a stabilization, its multivariable target is first compared with the cone of Un+1+Ui on the polynomial extension of the original complex. After instead taking the simultaneous ordinary quotient U1=⋯=Un=U, a stabilization contributes a zero-map cone, equivalently an extra zero-differential factor V; those fully collapsed complexes are compared with this factor retained. Under commutation the tilde homologies are isomorphic; under a stabilization from grid number n to n+1, the tilde homology changes by tensoring with the two-dimensional bigraded vector space V with one generator in bidegree (0,0) and one in bidegree (−1,−1).

Facts & Assumptions

Given: The two one-component grids and the single grid move in the statement. All polynomial variables initially remain independent and are matched by their individual markings.

[F1]

Filtered pentagon maps and hexagon inverse homotopies gives the weighted pentagon maps, hexagon inverse homotopies, and their filtered, associated-graded, and tilde versions for commutations and switches.

[F2]

The filtered destabilization-domain map and its cone target defines the filtered incoming-domain map for a southwest X stabilization and proves its chain-map and grading identities by all domain cases.

[F3]

Explicit inverse homotopies for filtered grid destabilization supplies an explicit inverse and both homotopies for that map by a finite cone-contraction formula, and derives its retained-variable and tilde comparisons.

[F4]

Changing the chosen polynomial action on a filtered knot grid gives explicit filtered equivalences and inverse homotopies for changing the chosen variable action along the knot's single cycle.

[F5]

Basepoint cones, a single variable action, and ordinary collapse proves that a cone of the new variable plus an old one retracts onto the smaller complex over the old-variable ring, while its simultaneous ordinary collapse is a zero-map cone and contributes V.

Proof

1.1givenconstructalgebra

A cyclic permutation is a change of fundamental cut on the same grid torus. It identifies states, rectangles, and marking monomials bijectively. The finite pair-count gradings also agree under this change. To check this directly for a column cut, use the signed set P=x−O in MO=I(P,P)+1. The wrapping column contains one state point of sign +1 and one O of sign −1. Only their comparisons with nonwrapping points change. At an integer state height the signed number of all points below is zero, and at the height of a marking in row r it is one, since the states have heights 0,…,n−1 and the markings have one height in each open row. Thus the signed difference of the below/above counts for either wrapping point, excluding itself, is one. Multiplying by the two signs and adding gives zero; the comparisons between the two wrapping points cancel from this sum and do not change. Hence I(P,P) is unchanged. Apply the same count to x−X, and interchange coordinates for a row cut. Both Maslov functions and the Alexander formula are unchanged. The torus-identification map is therefore a filtered chain isomorphism, with inverse the opposite cut change and zero inverse homotopies.

1.2F1given

For a commutation, use the maps and homotopies in [F1]. Their weights retain every O variable separately, and their filtration drop is the number of X in the polygon. Thus they are filtered equivalences over the full ring; restricting the ring action to any chosen marking variable preserves these identities. Their associated-graded maps retain precisely polygons avoiding X. Setting the variables to zero gives the tilde equivalence. The identical conclusion holds for the switches used below.

1.3F2F3F4F5construct

For a southwest X stabilization, [F2] supplies the map to the cone of U1+U2 on the polynomial extension of the smaller complex. By [F3], it has a finite matrix inverse and both filtered inverse homotopies. Compose with the explicit retraction in [F5]. Its inclusion, divided-difference homotopy, and evaluation are linear over every old variable, so the resulting equivalence and homotopies are over any chosen retained old-variable action. If either grid uses another chosen variable, including the newly inserted one, first or last compose with the chosen-variable equivalences of [F4]. Hence the assertion holds for arbitrary choices of the single variable action, without setting any independent variable equal to another.

2.1givenstep 1.1construct

The other local stabilization types reduce to this one using the following finite operations. An O stabilization introduces a vertical segment joining marks in consecutive rows. Commute this short segment horizontally toward the next X endpoint; while passing another vertical segment, its length-one interval is either disjoint or nested, since no marking height lies strictly between its two endpoints. At a shared endpoint level the exchange is a switch. Stop at the next X, where deleting that short segment is an X destabilization back to the original grid. Tracking the empty corner gives the pairs O:NE with X:SW, O:SE with X:NW, O:NW with X:SE, and O:SW with X:NE. Among the four X types, exchanging the two new adjacent rows moves the empty corner across a horizontal side, and exchanging the two new adjacent columns moves it across a vertical side; these are switches because the central opposite-type marking shares a column or row with the adjacent same-type marking. One or two such switches carry any empty corner to the southwest corner. These operations concern only the added row, added column, and their endpoint segments; after deleting them the original row and column orders are unchanged. Thus every stabilization is a southwest X stabilization followed by finitely many commutations, switches, and cut changes.

3.1F1F3F4F5step 1.1step 1.2step 1.3step 2.1constructalgebra

Compose the equivalences for that finite sequence, using step 1.2 for commutations and switches and step 1.3 for its sole stabilization. Explicit inverse homotopies compose: for equivalences f1,f2 with inverses g1,g2 and homotopies gifi+1=dHi+Hid, a homotopy for g1g2f2f1+1 is H1+g1H2f1. The counterpart on the other complex has the analogous formula. Every summand remains filtered and has degree +1. Reversing the roles of the two grids proves the destabilization assertion as well. All map formulas are finite sums of composable grid-domain counts and the explicit polynomial formulas in [F3]–[F5].

4.1F1F3F5step 1.1step 1.2step 1.3step 2.1step 3.1algebra∎

The full polynomial equivalence with the stabilization cone in step 1.3 can also be tensored with the simultaneous ordinary coefficient quotient. Its cross map U1+U2 becomes zero, so [F5] identifies the quotient cone with the collapsed smaller complex tensored with V. The first summand's grading and filtration shift is (−1,−1), giving exactly the displayed bidegrees. Taking the associated-graded equivalence first and then setting every variable to zero gives the same direct sum for tilde homology. Commutation and switches preserve tilde homology by step 1.2, and cut changes do by step 1.1. These comparisons prove the tilde and collapsed assertions for every stabilization type.

Remarks

Grid moves generate isotopy of oriented knots addresses the separate global fact that these local moves connect any two grids of an oriented knot. The proof here constructs the equivalences for each specified local move.

TheoremStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

Grid homology is independent of the grid and computes knot Floer homology

Statement

Assume countable choice. Let K be an oriented knot in S3, and let F be a compact connected oriented surface embedded in S3 with boundary K and minimum possible genus; write g(K)=genus⁡(F). Let G be any toroidal grid diagram for K with grid number n. Let Cgrid−(G) be the full filtered minus complex produced by the grid-complex algorithm, and let CU−(G) denote this same complex regarded over F2[U] by letting U act as one chosen Ui, with the other variables retained independently. For an admissible 2n-pointed Heegaard diagram H compatible with (K,F), define CU−(H) by the same restriction to a single variable action. Write CFK−(K) for the full filtered complex of an admissible doubly-pointed diagram. The ordinary fully collapsed quotient, obtained instead by setting all Ui=U, has type CFK−(K)⊗V⊗(n−1); the restriction of scalars CU− has type CFK−(K).

For the absolute Alexander normalization, let H1 be an admissible doubly-pointed Heegaard diagram for (K,F). For a generator x of H1, set AF(x)=12⟨c1(sw,z(x)),[F^]⟩, where sw,z(x) is the relative Spin⁡c structure assigned to this generator by the pointed Heegaard construction: it is the homology class of the nowhere-vanishing vector field obtained from the gradient trajectories through x and the chosen basepoint pair, after extending the field across small balls around the critical points. Its first Chern class is that of the oriented plane field orthogonal to this vector field. The surface F^ is obtained by removing a tubular neighborhood N(K) from F and capping the resulting longitude with the meridian disk of the solid torus attached in zero-framed surgery on K. Let HFK^(K) be the homology of the Alexander-associated-graded complex obtained from the doubly-pointed complex CU−(H1) by quotienting by its sole variable U, with this AF normalization.

Then CU−(G) and CU−(H) are filtered chain homotopy equivalent, preserving relative Maslov grading and the AF filtration, and their Alexander-associated-graded minus complexes are chain homotopy equivalent. For any two grids of K, the resulting normalized filtered complexes are filtered chain homotopy equivalent. If GH~(G) denotes the homology of the tilde grid complex over F2, then GH~(G)≅HFK^(K)⊗V⊗(n−1), where V is the two-dimensional bigraded vector space with one generator in bidegree (0,0) and one in bidegree (−1,−1).

Facts & Assumptions

Given: The oriented knot, its minimum-genus Seifert surface, the grid and the analytic diagrams in the Statement. All polynomial variables remain independent until a quotient is explicitly taken.

[F1]

The grid grading formulas are integral, and every realized flat-grid Whitney class satisfies the full relative Maslov and Alexander formulas, including positive X multiplicity. Its index is nx(D)+ny(D), and its local domain equation is ΔD=1x−1y. Actual integrable full disk counts are precisely the finitely many regular empty-state rectangles, with all individual O weights. (Integrality and rectangle changes of the grid Alexander grading, Realized grid Whitney classes satisfy the full relative grading formula, Realized flat-grid index-one positive domains are empty rectangles, Integrable flat-grid rectangle counts give the full filtered grid differential)

[F2]

The purely combinatorial grid maps and inverse homotopies preserve the stated J-formula Maslov grading and Alexander filtration. Commutation preserves tilde homology; stabilization adds the zero-differential factor V of degrees (0,0),(-1,-1). Over one chosen variable action they are actual filtered equivalences with independent variables retained. Grid moves generate oriented knot isotopy. These results concern finite grid domains and do not invoke the grid-to-analytic identification proved here. (Commutation and stabilization invariance of grid complexes, Grid moves generate isotopy of oriented knots)

[F3]

The analytic full-minus auxiliary comparisons apply to any finite square-zero count endpoints with no nonconstant nonpositive-index strips and regular index-one strips. The actual balanced-knot diagram comparison, ordinary target-neck stabilization and marked-pair cone then give equivalence over one chosen action with a normalized doubly pointed complex. Ordinary collapse instead gives its tensor product with V to the number of extra pairs. All these comparisons retain positive z terms. The multipointed relative origins are fixed by transport from the doubly pointed Seifert-Chern and total-hat normalizations, with source cone shift (-1,-1). (Full minus auxiliary continuations are filtered equivalences for balanced knots, Invariance of the multipointed Heegaard Floer complex, Basepoint cones, a single variable action, and ordinary collapse)

[F4]

The normalized doubly pointed total zero-w hat homology is one-dimensional in Maslov degree zero. Analytic conjugation identifies its Alexander-associated-graded homology in degrees (m,a) and (m-2a,-a), independently of any grid comparison. Its Alexander origin is the negative modified-Morse field with zero-relative-Euler Seifert cap, hence the stated half-Chern pairing. (Relative three-sphere Maslov gradings have a unique hat homology normalization, Knot Floer duality and exchange of the two filtrations, Doubly pointed knot Morse fields have a zero-framed Seifert extension)

[F5]

The grid is admissible. The grid algorithm's full and associated-graded rectangle matrices square to zero and have the stated variable actions. (A knot grid diagram is admissible, The three grid rectangle differentials square to zero, Finite grid complexes for knot Floer theory)

Proof

1.1F1F5givenconstruct

Use the standard genus-one splitting of the oriented three-sphere: the horizontal circles bound parallel meridian disks in one solid torus, and the vertical circles bound parallel meridian disks in the other; their primitive meridians meet once. Taking all n parallel disks cuts each side into n balls. The row and column marking condition puts one O and one X in every corresponding cut ball. Push the horizontal O-to-X segments into the lower side and the vertical X-to-O segments into the upper side, using their evident rectangular bridge disks. These simultaneous arcs give exactly the oriented knot of the grid, since every vertical segment passes over every horizontal segment. Thus this is an actual balanced 2n-pointed diagram of K, not only an abstract torus count. Its analytic symmetric rank is 1+n−1=n, and w=O,z=X. Give it the actual flat integrable data of [F1]. The identity on states identifies its full polynomial count matrix with the algorithmic one.

2.1F1F3F5step 1.1constructalgebra

Check the remaining good-endpoint condition before applying [F3]. Positivity gives a nonnegative domain D for every actual integrable strip. Its index nx(D)+ny(D) is nonnegative. If it is zero, every cell touching either state has coefficient zero. The local equation ΔD=1x−1y then forces x=y, since at any changed vertex the four coefficients would have nonzero alternating sum. Consequently ΔD=0, and telescoping its row and column differences gives Dij=ai+bj. Because x is one point in every row and column, the n cells immediately northeast of its state points have total coefficient ∑iai+∑jbj=0. The sum of ALL cell coefficients is n times that sum, hence zero. Nonnegativity gives D=0. The actual ordered-root energy bound is then zero, so the strip is constant. Therefore no nonconstant index-at-most-zero strip exists. Finiteness, index-one regularity and square zero were already supplied by [F1,F5]; no index-two regularity of this special integrable path is assumed. The precise endpoint hypotheses of [F3] hold.

3.1F1F3step 1.1step 2.1constructalgebra

Apply the actual full-minus auxiliary and diagram equivalences of [F3] to this endpoint. They identify its chosen-action restriction with a doubly pointed full complex, and identify its ordinary collapse with that complex tensor V⊗(n−1). They also give the quotient-by-all-variables and associated-graded identities by specializing the actual finite maps and homotopies. At this stage use ONLY the relative formulas of [F1]: on the connected state set, its numerical J-gradings differ from the transported normalized analytic origins by one integer Maslov constant b and one integer Alexander constant c. In particular, if H denotes normalized doubly pointed associated-graded zero-U homology, the grid tilde homology has the relative identification H⊗V⊗(n−1) with common shifts (b,c). This constructs the comparison before determining either shift; no assertion of the theorem under proof is a premise.

4.1F2F3F4step 3.1constructalgebra

Fix b without using Alexander centering. List the O markings in order of increasing row, with distinct column indices o0,…,on−1. Replace X by X∗ having in row i the column oi+1, with cyclic indices. This is disjoint from O and is a one-component grid: its marking permutation sends row i to row i+1. Its vertical segments join successive row heights, except for the one segment from the last row to the first. The short vertical segments meet no other horizontal height, so every crossing lies on that sole long vertical segment, which passes OVER the complementary embedded planar arc. This is an unknot: raise the long segment into an upper ball and lower the complementary planar arc into a lower ball; both are boundary-parallel arcs, with the rectangular upper bridge disk and the planar lower bridge disk. Match their boundary chords on the separating plane and join the two disks along that chord to obtain an embedded spanning disk, rounding its finite corners. The FULL unfiltered rectangle matrix and M_O do not depend on X, so replacing X by X∗ changes neither this matrix nor its Maslov grading. By the purely combinatorial graded chosen-action equivalences and move generation in [F2], this same unfiltered complex is graded equivalent to a size-two unknot grid. In that grid the two states have M_O=0,-1; direct rectangle enumeration gives the cone F2[U1,U2]→U1+U2F2[U1,U2] with source degree -1 and target degree zero. The monic polynomial retraction in [F3] evaluates the other variable at the chosen one, so after quotienting that chosen U its total homology is one-dimensional in M_O degree zero. The combinatorial inverse homotopies specialize too, so this is true for the original grid's chosen-action total zero-U complex. Step 3.1 compares it to the analytic total hat homology, whose normalized degree is zero by [F4]. Its common Maslov shift is therefore b=0. This also proves nontriviality without treating the Alexander-associated-graded complex as total hat homology.

5.1F1F2step 4.1constructalgebra

Fix c by a finite reciprocal symmetry. Define the finite Laurent polynomial PG(t)=∑x(−1)MO(x)tAG(x) and put Dn(t)=(1−t−1)n−1. The Euler characteristic of a finite bigraded complex equals that of its homology degree by degree: split each finite degree into boundaries, a homology complement and a complement mapped isomorphically to boundaries; the latter pairs cancel with opposite Maslov signs. Thus [F2] implies that QG=PG/Dn is unchanged by every grid move, as a rational function: commutation preserves P and stabilization multiplies P by 1−t−1, precisely the Euler polynomial of V. Construct GT by transposing the square and INTERCHANGING O and X. The ambient rotation (x,y,z)↦(y,x,−z) has determinant plus one and changes old vertical overpasses to new horizontal underpasses. An old horizontal O-to-X arc becomes a new vertical X-to-O arc; an old vertical X-to-O arc becomes a new horizontal O-to-X arc. Hence it carries the SAME oriented knot to GT, rather than a mirror or assumed orientation reversal. Both grids have size n, so [F2] gives PGT=PG. Transposition preserves the strict pair relation defining J, while mark exchange swaps M_O and M_X. The literal formulas are therefore AGT(xT)=−AG(x)−(n−1) and MOT(xT)=MO(x)−2AG(x)−(n−1). Since A_G is integral, their signs give PG(t)=(−1)n−1t−(n−1)PG(t−1). Using Dn(t−1)=(−1)n−1tn−1Dn(t) yields QG(t)=QG(t−1). This derives reciprocity from actual J formulas and purely combinatorial moves; no analytic grading identification or unproved Alexander determinant formula has been used.

6.1F3F4step 3.1step 4.1step 5.1constructalgebra

Write QH(t)=∑m,a(−1)mdim⁡Hm,a ta. This is a finite Laurent polynomial. By [F4]'s ACTUAL conjugation, Hm,a is isomorphic to Hm−2a,−a. The degree difference is even, hence QH(t)=QH(t−1). For clarity, this symmetry is supplied by the normalized analytic coordinate-swap map: on the vertical corner (i,j)=(0,a) it lands on the horizontal corner (a,0), and multiplying by Ua identifies that corner with (0,−a) and lowers Maslov degree by 2a. It is a map of the corner subquotients, not a declaration that the full positive-z hat differential preserves A. Moreover QH(1)=1. Indeed the associated graded and total zero-U doubly pointed complexes have the same finite chain groups with their Maslov degrees, so their ordinary Euler characteristics coincide; the latter has exactly one homology class of normalized degree zero. Therefore Q_H is NONZERO. The relative comparison in step 3.1 and b=0 now give PG=tcDnQH, or QG=tcQH. Substituting the two reciprocal identities yields (tc−t−c)QH=0 in the Laurent polynomial domain Z[t,t−1]. Since Q_H is nonzero, c=0. This is precisely the missing absolute combinatorial Alexander anchor, with no appeal to grid equality to establish its own shift.

7.1F1F3F4step 1.1step 3.1step 4.1step 6.1constructalgebra∎

The identity in step 1.1 and the actual analytic equivalences in step 3.1 consequently preserve both normalized origins, by steps 4.1 and 5.1. Compose with [F3]'s normalized comparison to the diagram H in the Statement; its inverse and inverse homotopies remain finite and filtered. Taking the Alexander associated graded of these identities gives the claimed associated-graded minus equivalences. Each term retains every individual O weight until the explicitly stated restriction or quotient. The ordinary collapse and all-variable-zero associated-graded specialization are exactly the two actual constructions of [F3], yielding respectively CFK−(K)⊗V⊗(n−1) and HFK^(K)⊗V⊗(n−1), with the displayed bidegrees now absolutely centered. The chosen-action restriction yields CFK−(K) instead. Composing the comparisons for two grids gives their normalized filtered equivalence and inverse homotopies. Finally [F4] identifies the preserved Alexander origin with the zero-framed capped-Seifert half-Chern expression of the Statement. Its cap and negative-Morse convention are inherited from that supplier, not replaced by a winding-number convention. Every original conclusion follows with all positive z terms and all stated scalar distinctions retained.

LemmaStatement: Literature-sourcedProof: AI-adaptedverified 2026-10-22 (gpt-6.1-sol)Open item page →

The tau level is independent of the normalized Heegaard model

Statement

Assume the Axiom of Countable Choice. Let K be a smooth oriented knot, let F be a fixed minimum-genus oriented Seifert surface, and let H0,H1 be admissible doubly pointed Heegaard diagrams for the same pair (K,F) whose hat complexes have total homology F2 and the same absolute Alexander normalization. Their diagramwise thresholds agree. Moreover, every grid presentation of (K,F) gives this same value after the full filtered grid complex is converted to the normalized singly pointed model, with all basepoint-variable and stabilization factors accounted for. The common value is τ(K).

Facts & Assumptions

Given: The Axiom of Countable Choice, a knot K, a fixed minimum-genus Seifert surface F, and two admissible doubly pointed diagrams with the stated one-dimensional total homology and common absolute normalization.

[A1]

The Axiom of Countable Choice (ACω) states the axiom that every sequence of nonempty sets admits a choice function; it is inherited by the current diagram/move suppliers that require selected finite data.

[F1]

The tau level is preserved by filtered chain homotopy equivalence proves that a filtration-preserving chain homotopy equivalence between finite models with one-dimensional total homology preserves their tau thresholds.

[L1]

Invariance of the multipointed Heegaard Floer complex supplies the actual finite filtered chain homotopy equivalences for admissible analytic diagrams, with the specified single-variable restriction and the Seifert-normalized doubly pointed origin. Specializing the actual finite identities at U=0 gives the filtered hat comparisons; this does not supply the separate grid-to-analytic identification.

[L2]

Commutation and stabilization invariance of grid complexes gives the local filtered equivalences for commutations and stabilizations, using the stated polygon maps and the stabilization cone comparison. Its simultaneous-collapse statement retains the extra V factor; this local move comparison is separate from the analytic grid comparison in [L3].

[L3]

Grid homology is independent of the grid and computes knot Floer homology constructs the normalized comparison from the grid full filtered minus complex to the singly pointed analytic model. It distinguishes restriction to one variable action from simultaneous collapse, and accounts for the latter by V⊗(n−1); its argument also identifies the grid Alexander origin with the Seifert-normalized one.

[L4]

Grid moves generate isotopy of oriented knots gives a finite grid-move sequence for grids of the same oriented knot, by refining to bridge-like grids, applying an oriented Reidemeister sequence, and translating each local change using the stated arbitrary-exterior move supplier.

Proof

1.1F1L1givenconstructalgebra

Apply [L1] to the two analytic doubly pointed diagrams in the Statement. Its maps and inverse homotopies are finite polynomial identities preserving their Seifert-normalized Alexander filtration. Specialize those identities at U=0; no flatness assertion about homology is required. The resulting actual finite hat models are filtered chain homotopy equivalent and carry the nonzero total generators to each other. Therefore [F1] gives equality of their thresholds. This proves the analytic-diagram comparison in the first sentence.

1.2L1L3givenconstructalgebra

Fix a grid G with n marking pairs and choose one marking variable Ui. Use the actual normalized grid-to-analytic comparison in [L3], together with the pointed-move comparison in [L1], to convert its full filtered minus complex to the finite singly pointed model for (K,F). In this comparison first restrict scalars along U↦Ui and retain the other Uj as independent module variables. The supplied chain maps and inverse homotopies preserve the Alexander filtration and the Seifert-normalized origin. Specializing these actual identities at the chosen U=0 gives the corresponding filtered equivalence to the singly pointed hat endpoint. The restricted grid module can still have infinite rank over F2 after this specialization, so it is not itself an input to the finite-complex definition of τ; the grid value in the Statement is the threshold on the finite normalized singly pointed endpoint obtained by this conversion.

1.3L1L3givenalgebra

The separate simultaneous ordinary collapse sets all marking variables equal before taking the one-variable quotient. By [L3] and [L1] its filtered type is the singly pointed minus model tensored with V⊗(n−1), with each V carrying bidegrees (0,0) and (−1,−1). After quotienting the remaining variable by zero, this collapsed model has the extra V homology factors and is not the finite complex with one-dimensional total homology required in the definition of τ. Those factors are accounted for by the actual basepoint-pair cone and retraction comparison in [L1, L3]: use the retained-variable equivalence to the singly pointed model when defining the grid's value, and do not apply τ to the unreduced simultaneous-collapse complex. Thus no extra filtration shift from the V factors is included in the claimed value.

2.1A1F1L1L2L3L4step 1.1step 1.2step 1.3algebra∎

For any two grid presentations, [L3] supplies this conversion for each grid to a normalized singly pointed model. Compare those finite endpoints by the actual pointed-move equivalence [L1], specialize its maps and filtered inverse homotopies at U=0, and apply [F1]. This gives the same threshold as in step 1.1. The grid-move and grid-isotopy suppliers [L2, L4] describe additional combinatorial comparisons; they do not replace the filtered grid-to-analytic conversion or alter the variable-collapse distinction above.

4 · Examples, counterexamples and false statements

None yet.

Sources