Introduction
Let K ⊂ S 3 K \subset S^3 K ⊂ S 3 be an oriented knot, X K = S 3 ∖ int N ( K ) X_K = S^3 \setminus\operatorname{int} N(K) X K = S 3 ∖ int N ( K ) its exterior, and
Δ K ( t ) ∈ Z [ t , t − 1 ] , Δ K ( t − 1 ) = Δ K ( t ) , Δ K ( 1 ) = 1 , (1) \Delta_K(t) \in\mathbb{Z}[t,t^{-1}], \qquad\Delta_K(t^{-1}) = \Delta_K(t), \qquad\Delta_K(1) = 1, \tag*{(1)} Δ K ( t ) ∈ Z [ t , t − 1 ] , Δ K ( t − 1 ) = Δ K ( t ) , Δ K ( 1 ) = 1 , (1)
the symmetrically normalized Alexander polynomial. Three different settings produce the polynomial. Fox calculus and Reidemeister torsion recover Δ K / ( t − 1 ) \Delta_K/(t-1) Δ K / ( t − 1 ) from the rank-one local systems on X K X_K X K ; Abelian BF theory is a field-theoretic realization of corresponding torsion data; and the work of Ozsváth and Szabó identifies Δ K \Delta_K Δ K with the Alexander-graded Euler characteristic of knot Floer homology [17 , 9 , 18 ] .
Friedl, Juhász and Rasmussen identify sutured Floer Euler characteristics with torsion [11 ] ; Hom, Lidman and Watson categorify Donaldson’s Alexander-module TQFT using bordered Floer theory [16 ] . Cellular BF theory and its gluing formalism are developed by Cattaneo, Mnev and Reshetikhin [10 ] . These results provide the starting point for the comparisons below.
We first ask which Floer Euler characteristics can be recovered from a single twisted torsion value. Evaluation at a primitive prime-order root of unity retains all Galois conjugates when the value is known exactly as an algebraic number. The genus bound and the normalization Δ K ( 1 ) = 1 \Delta_K(1) = 1 Δ K ( 1 ) = 1 then give a sharp coefficient-reconstruction formula. The corresponding local system is acyclic on the exterior and its boundary.
The absolute torsion is Δ K ( t ) / ( t − 1 ) \Delta_K(t)/(t-1) Δ K ( t ) / ( t − 1 ) , whereas the Floer Euler characteristic is Δ K ( t ) \Delta_K(t) Δ K ( t ) . The factor t − 1 t-1 t − 1 has a geometric interpretation: it removes the torsion contribution of a meridional annulus. Relative cellular BF theory with this boundary condition therefore produces the Alexander polynomial directly, with the usual ambiguity from cellular lifts. The comparison concerns based complexes and determinant lines.
Theorem 1.1 (Meridional relative BF torsion and Floer decategorification). Let ( X K , γ μ ) (X_K,\gamma_\mu) ( X K , γ μ ) denote the sutured complement of a knot, with two oppositely oriented meridional sutures, and let R − ⊂ ∂ X K R_-\subset\partial X_K R − ⊂ ∂ X K be its negative annulus. For z ∈ C × ∖ { 1 } z\in\mathbb{C}^{\times}\setminus\{1\} z ∈ C × ∖ { 1 } satisfying Δ K ( z ) ≠ 0 \Delta_K(z)\ne0 Δ K ( z ) = 0 , let L z L_z L z denote meridional holonomy z z z . Then the twisted cochain complexes of X K X_K X K , R − R_- R − , the whole boundary, and the relative pair ( X K , R − ) (X_K,R_-) ( X K , R − ) are acyclic. With compatible cellular lifts, the following equality holds modulo ± z \pm z ± z :
τ ( X K , R − ; L z ) = τ ( X K ; L z ) τ ( R − ; L z ) ≐ Δ K ( z ) . (2) \tau(X_K,R_-;L_z)=\frac{\tau(X_K;L_z)}{\tau(R_-;L_z)}\doteq\Delta_K(z). \tag*{(2)} τ ( X K , R − ; L z ) = τ ( R − ; L z ) τ ( X K ; L z ) ≐ Δ K ( z ) . (2) It is the determinant-line invariant of a finite-dimensional, linear Abelian BF system with complementary relative boundary conditions. After fixing the Alexander symmetric normalization, its Laurent polynomial agrees with both the sutured Floer Euler-characteristic torsion of ( X K , γ μ ) (X_K,\gamma_\mu) ( X K , γ μ ) and ∑ a χ ( H F K ^ ( K , a ) ) t a \sum_a\chi(\widehat{HFK}(K,a))t^a ∑ a χ ( H F K ( K , a )) t a .
The terms “determinant-line invariant” and “relative BF system” are spelled out in § 3 –§ 4 . The qualifier modulo units is essential: the rank-one system L z L_z L z is generally non-unimodular, so lifts, Euler structures, and normalizations cannot be suppressed. The established cellular BV–BFV gluing theorem of [10 , 7 ] is not invoked without checking its coefficient hypotheses.
Theorem 1.2 (Cyclotomic BF–Floer reconstruction). Let K ⊂ S 3 K\subset S^3 K ⊂ S 3 be a knot with g ( K ) ≤ g g(K)\le g g ( K ) ≤ g , and let p ≥ 2 g + 1 p\ge2g+1 p ≥ 2 g + 1 be an odd prime. Write ζ = e 2 π i / p \zeta=e^{2\pi i/p} ζ = e 2 π i / p and denote by L ζ L_\zeta L ζ the rank-one complex local system sending a positively oriented meridian of K K K to ζ \zeta ζ . Then:
(i) H ∙ ( X K ; L ζ ) = H ∙ ( ∂ X K ; L ζ ) = 0 H^\bullet(X_K;L_\zeta)=H^\bullet(\partial X_K;L_\zeta)=0 H ∙ ( X K ; L ζ ) = H ∙ ( ∂ X K ; L ζ ) = 0 , so the specialized Reidemeister torsion is defined and nonzero.
(ii) Fix the symmetric torsion normalization
τ K s y m ( t ) = Δ K ( t ) t − 1 . (3) \tau_K^{\mathrm{sym}}(t)=\frac{\Delta_K(t)}{t-1}. \tag*{(3)} τ K sym ( t ) = t − 1 Δ K ( t ) . (3) For every a ∈ Z a\in\mathbb{Z} a ∈ Z with − g ≤ a ≤ g -g\le a\le g − g ≤ a ≤ g , one has
χ ( H F K ^ ( K , a ) ) = 1 + Tr Q ( ζ ) / Q ( ζ − a ( ζ − 1 ) τ K s y m ( ζ ) ) p . (4) \chi(\widehat{HFK}(K,a))=\frac{1+\operatorname{Tr}_{\mathbb{Q}(\zeta)/\mathbb{Q}}\left(\zeta^{-a}(\zeta-1)\tau_K^{\mathrm{sym}}(\zeta)\right)}{p}. \tag*{(4)} χ ( H F K ( K , a )) = p 1 + Tr Q ( ζ ) / Q ( ζ − a ( ζ − 1 ) τ K sym ( ζ ) ) . (4) (iii) The entire symmetrically normalized Alexander polynomial, and hence every Alexander-graded Floer Euler characteristic, is uniquely determined by the single exact algebraic value τ K s y m ( ζ ) \tau_K^{\mathrm{sym}}(\zeta) τ K sym ( ζ ) and the genus bound g g g .
A second result shows that the bound on the prime cannot generally be lowered.
Theorem 1.3 (Sharpness). For every odd prime p p p and every integer g ≥ ( p + 1 ) / 2 g\ge(p+1)/2 g ≥ ( p + 1 ) /2 , there is a fibered knot J p ⊂ S 3 J_p\subset S^3 J p ⊂ S 3 of genus ( p + 1 ) / 2 (p+1)/2 ( p + 1 ) /2 such that
τ J p s y m ( ζ ) = τ U s y m ( ζ ) = 1 ζ − 1 , \tau_{J_p}^{\mathrm{sym}}(\zeta)=\tau_U^{\mathrm{sym}}(\zeta)=\frac{1}{\zeta-1}, τ J p sym ( ζ ) = τ U sym ( ζ ) = ζ − 1 1 , where U U U denotes the unknot. The Alexander polynomials of J p J_p J p and U U U are different. Consequently, prime-order torsion evaluation is not injective on the class of knots of genus at most g g g when p < 2 g + 1 p<2g+1 p < 2 g + 1 .
Our reconstruction has a compatible multiplicative form. If K = K 1 # K 2 K=K_1\#K_2 K = K 1 # K 2 , then the torsion of the meridional annulus corrects the product of torsions by exactly one factor of t − 1 t-1 t − 1 . Under the arithmetic reconstruction this becomes the convolution of graded Floer Euler characteristics. We also recover classical branched-cover homology from a cyclotomic norm.
Sections 11–13 compare linear BV–BFV interface gluing with Floer tensor pairings after decategorification and apply the staircase classification to positive L-space knots. The standard twisted BF complex is acyclic, so it cannot be quasi-isomorphic to the associated graded knot Floer complex.
Section 14 studies derived scalar extension for the degree-zero torus algebra. Its projective pairing matrix ( 2 3 1 2 ) \begin{pmatrix}2 & 3 \\ 1 & 2\end{pmatrix} ( 2 1 3 2 ) obstructs independent single-field tensor factorizations. Reduction retaining both idempotents is conservative on perfect modules, but its pairing defect is measured by the chord ideal and a three-layer filtration. Section 15 resolves the torus algebra by six generators and realizes its representation differential in a classical one-dimensional cotangent BV–BFV model. The resulting equivalence is algebraic. Section 16 computes a fixed nonresonant quantum sector of Abelian BFV theory and identifies a Hochschild obstruction to a Morita comparison with the degree-zero torus algebra. A geometric comparison of boundary categories and manifold modules requires further constructions.
Twisted complexes, torsion, and Abelian BF theory
Local systems and the Fox complex
The canonical abelianization
ϕ : π 1 ( X K ) ⟶ H 1 ( X K ; Z ) ≅ Z \phi:\pi_1(X_K)\longrightarrow H_1(X_K;\mathbb{Z})\cong\mathbb{Z} ϕ : π 1 ( X K ) ⟶ H 1 ( X K ; Z ) ≅ Z
sends every oriented Wirtinger meridian to 1 1 1 . Write Λ = Z [ t , t − 1 ] \Lambda=\mathbb{Z}[t,t^{-1}] Λ = Z [ t , t − 1 ] , F = Q ( t ) F=\mathbb{Q}(t) F = Q ( t ) , and regard F F F as a Z [ π 1 ( X K ) ] \mathbb{Z}[\pi_1(X_K)] Z [ π 1 ( X K )] -module through γ ↦ t ϕ ( γ ) \gamma\mapsto t^{\phi(\gamma)} γ ↦ t ϕ ( γ ) . An associated Wirtinger presentation
π 1 ( X K ) = ⟨ x 1 , … , x n ∣ r 1 , … , r n − 1 ⟩ \pi_1(X_K)=\langle x_1,\ldots,x_n\mid r_1,\ldots,r_{n-1}\rangle π 1 ( X K ) = ⟨ x 1 , … , x n ∣ r 1 , … , r n − 1 ⟩
gives, after abelianized Fox differentiation, the finite based complex
0 ⟶ F n − 1 → d 2 F n → d 1 F ⟶ 0 , d 1 = ( t − 1 , … , t − 1 ) . (5) 0\longrightarrow F^{n-1}\xrightarrow{d_2}F^n\xrightarrow{d_1}F\longrightarrow0,\qquad d_1=(t-1,\ldots,t-1). \tag*{(5)} 0 ⟶ F n − 1 d 2 F n d 1 F ⟶ 0 , d 1 = ( t − 1 , … , t − 1 ) . (5)
Here the matrix of d 2 d_2 d 2 is the transpose of the ( n − 1 ) × n (n-1)\times n ( n − 1 ) × n Fox matrix ( ϕ ( ∂ r i / ∂ x j ) ) i j (\phi(\partial r_i/\partial x_j))_{ij} ( ϕ ( ∂ r i / ∂ x j ) ) ij . It follows from the Fox identity that d 1 d 2 = 0 d_1d_2=0 d 1 d 2 = 0 .
For a column j j j let A j ( t ) A_j(t) A j ( t ) denote the square matrix obtained by removing column j j j of the Fox matrix. The Alexander–Fox theorem says that
det A j ( t ) = ε t m Δ K ( t ) , ε ∈ { ± 1 } , m ∈ Z , (6) \det A_j(t)=\varepsilon t^m\Delta_K(t),\qquad\varepsilon\in\{\pm1\},\quad m\in\mathbb{Z}, \tag*{(6)} det A j ( t ) = ε t m Δ K ( t ) , ε ∈ { ± 1 } , m ∈ Z , (6)
with the usual choice of Wirtinger presentation. Over F F F , the complex (5) is acyclic.
Definition 2.1 (Acyclic chain torsion: our convention). Let 0 → C N → ∂ N ⋯ → ∂ 1 C 0 → 0 0\to C_N\xrightarrow{\partial_N}\cdots\xrightarrow{\partial_1}C_0\to0 0 → C N ∂ N ⋯ ∂ 1 C 0 → 0 be an acyclic complex of finite-dimensional vector spaces over a field, each equipped with a specified ordered basis c i c_i c i . Write B i = im ( ∂ i + 1 ) B_i=\operatorname{im}(\partial_{i+1}) B i = im ( ∂ i + 1 ) , choose an ordered basis b i b_i b i of B i B_i B i , and lift b i − 1 b_{i-1} b i − 1 to a tuple b ~ i − 1 \widetilde b_{i-1} b i − 1 in C i C_i C i . If [ v / c i ] [v/c_i] [ v / c i ] is the determinant of the transition matrix from c i c_i c i to the concatenated basis v v v , define
τ ( C ∗ , c ∗ ) = ∏ i = 0 N [ b i b ~ i − 1 / c i ] ( − 1 ) i + 1 , b − 1 = b N = ∅ . (7) \tau(C_*,c_*)=\prod_{i=0}^{N}\left[b_i\widetilde b_{i-1}/c_i\right]^{(-1)^{i+1}},\qquad b_{-1}=b_N=\varnothing. \tag*{(7)} τ ( C ∗ , c ∗ ) = i = 0 ∏ N [ b i b i − 1 / c i ] ( − 1 ) i + 1 , b − 1 = b N = ∅ . (7)
This scalar is independent of the chosen boundary bases and lifts. In particular, the two-term complex 0 → F → u F → 0 0\to F\xrightarrow{u}F\to0 0 → F u F → 0 in degrees 1 , 0 1,0 1 , 0 has torsion u − 1 u^{-1} u − 1 .
Verification of independence. Changing b i b_i b i by a matrix M M M changes the determinant in degree i i i by det M \det M det M and that in degree i + 1 i+1 i + 1 by the same factor; the exponents in (7) are opposite, so they cancel. Changing a lift by an element of B i B_i B i amounts to a block-triangular basis change with diagonal blocks equal to the identity, and leaves the determinants unchanged.
Proposition 2.2 (Torsion normalization). For the acyclic complex (5) , the Milnor torsion convention in which the circle 0 → F → t − 1 F → 0 0\to F\xrightarrow{t-1}F\to0 0 → F t − 1 F → 0 has torsion ( t − 1 ) − 1 (t-1)^{-1} ( t − 1 ) − 1 gives
τ ( X K ; F ϕ ) = Δ K ( t ) t − 1 . (8) \tau(X_K;F_\phi)=\frac{\Delta_K(t)}{t-1}. \tag*{(8)} τ ( X K ; F ϕ ) = t − 1 Δ K ( t ) . (8)
where ≐ \doteq ≐ means equality up to ± t m \pm t^{m} ± t m . Equation (3) selects the unique symmetrized numerator with value 1 1 1 at t = 1 t=1 t = 1 .
Proof. Choose j = n j=n j = n for simplicity. A basis for the image of d 1 d_1 d 1 is ( t − 1 ) e 0 (t-1)e_0 ( t − 1 ) e 0 and a lift is e n e_n e n in F n F^n F n . The columns of d 2 d_2 d 2 together with e n e_n e n form a basis of F n F^n F n , whose determinant is ± det A n ( t ) \pm\det A_n(t) ± det A n ( t ) . In the stated alternating-basis convention the torsion is their determinant divided by t − 1 t-1 t − 1 , up to sign. Equation (6) finishes the calculation. □
Remark 2.3. A gauge-fixed Gaussian BF integral may produce the torsion, its inverse, or a density in a determinant line, depending on field parity, the chosen polarization and the integration convention. Throughout this paper τ K s y m \tau_K^{\mathrm{sym}} τ K sym means the explicitly defined torsion expression (3) ; it is not asserted to be a convention-independent partition function. The factor t − 1 t-1 t − 1 is indispensable, including in connected-sum formulas.
The twisted Abelian BF model
For a complex character z ∈ C × z\in\mathbb{C}^{\times} z ∈ C × , let L z L_z L z be the flat line associated to γ ↦ z φ ( γ ) \gamma\mapsto z^{\varphi(\gamma)} γ ↦ z φ ( γ ) and L z ∨ L_z^{\vee} L z ∨ its dual. A formal twisted three-dimensional Abelian BF action is
S z [ A , B ] = ∫ X K ⟨ B ∧ d L z A ⟩ , A ∈ Ω 1 ( X K ; L z ) , B ∈ Ω 1 ( X K ; L z ∨ ) . (9) S_z[A,B]=\int_{X_K}\langle B\wedge d_{L_z}A\rangle,\qquad A\in\Omega^1(X_K;L_z),\quad B\in\Omega^1(X_K;L_z^{\vee}). \tag*{(9)} S z [ A , B ] = ∫ X K ⟨ B ∧ d L z A ⟩ , A ∈ Ω 1 ( X K ; L z ) , B ∈ Ω 1 ( X K ; L z ∨ ) . (9)
The infinitesimal bulk gauge symmetries are A ↦ A + d L z λ A\mapsto A+d_{L_z}\lambda A ↦ A + d L z λ and B ↦ B + d L z ∨ η B\mapsto B+d_{L_z^{\vee}}\eta B ↦ B + d L z ∨ η , with boundary terms requiring a boundary condition or a BFV extension. Formally, the variation includes
δ S z ∣ ∂ X K = ∫ ∂ X K ⟨ B ∧ δ A ⟩ , \delta S_z\big|_{\partial X_K}=\int_{\partial X_K}\langle B\wedge\delta A\rangle, δ S z ∂ X K = ∫ ∂ X K ⟨ B ∧ δ A ⟩ ,
which supplies the canonical boundary one-form of the BFV description. The BV resolution uses the twisted de Rham complex and its dual. Upon cellular reduction and a choice of gauge fixing, determinant data of these complexes give torsion-type amplitudes, as in the cellular BF approach of [10 ] . This background motivates our terminology BF torsion sector ; it is not used as an unproved functional-integral identity.
Lemma 2.4 (Acyclic boundary sector). If z ≠ 1 z\ne1 z = 1 , then H ∙ ( ∂ X K ; L z ) = 0 H^{\bullet}(\partial X_K;L_z)=0 H ∙ ( ∂ X K ; L z ) = 0 .
Proof. Choose meridian μ \mu μ and preferred longitude λ \lambda λ on ∂ X K ≅ T 2 \partial X_K\cong T^2 ∂ X K ≅ T 2 . Their holonomies are z z z and 1 1 1 , respectively. The twisted cellular cochain complex of the torus has the form
0 ⟶ C → ( z − 1 ) C 2 → ( 0 z − 1 ) C ⟶ 0 0\longrightarrow\mathbb{C}\xrightarrow{\left(\begin{smallmatrix}z-1\end{smallmatrix}\right)}\mathbb{C}^2\xrightarrow{\left(\begin{smallmatrix}0&z-1\end{smallmatrix}\right)}\mathbb{C}\longrightarrow0 0 ⟶ C ( z − 1 ) C 2 ( 0 z − 1 ) C ⟶ 0 (up to a harmless sign in the last differential). For z ≠ 1 z\ne1 z = 1 it is exact. □
Lemma 2.5 (Prime characters are nonresonant). For every knot K K K and every primitive p p p th root ζ \zeta ζ of unity, with p p p prime, one has Δ K ( ζ ) ≠ 0 \Delta_K(\zeta)\ne0 Δ K ( ζ ) = 0 . Consequently H ∙ ( X K ; L ζ ) = 0 H^{\bullet}(X_K;L_\zeta)=0 H ∙ ( X K ; L ζ ) = 0 .
Proof. If Δ K ( ζ ) = 0 \Delta_K(\zeta)=0 Δ K ( ζ ) = 0 , then the minimal polynomial Φ p ( t ) = 1 + t + ⋯ + t p − 1 \Phi_p(t)=1+t+\cdots+t^{p-1} Φ p ( t ) = 1 + t + ⋯ + t p − 1 divides t N Δ K ( t ) ∈ Z [ t ] t^N\Delta_K(t)\in\mathbb{Z}[t] t N Δ K ( t ) ∈ Z [ t ] for a sufficiently large N N N , by Gauss’s lemma. Evaluation at t = 1 t=1 t = 1 gives p ∣ Δ K ( 1 ) = 1 p\mid\Delta_K(1)=1 p ∣ Δ K ( 1 ) = 1 , a contradiction. Thus an Alexander–Fox minor is invertible at t = ζ t=\zeta t = ζ , and the specialized complex (5) is exact. The knot exterior has the homotopy type of the corresponding two-dimensional spine. □
BV–BFV boundary data and the meridional relative model
Boundary data
For clarity, we first work with smooth forms and then with finite cellular cochains. On an oriented three-manifold X X X with a flat rank-one system L z L_z L z , introduce the total (ghost-extended) fields
A ∈ Ω ∙ ( X ; L z ) [ 1 ] , B ∈ Ω ∙ ( X ; L z ∨ ) [ 1 ] . (10) \mathcal{A}\in\Omega^{\bullet}(X;L_z)[1],\qquad\mathcal{B}\in\Omega^{\bullet}(X;L_z^{\vee})[1]. \tag*{(10)} A ∈ Ω ∙ ( X ; L z ) [ 1 ] , B ∈ Ω ∙ ( X ; L z ∨ ) [ 1 ] . (10)
A k k k -form component of A \mathcal{A} A or B \mathcal{B} B has ghost number 1 − k 1-k 1 − k ; the degree-zero components are the classical 1 1 1 -form fields A A A and B B B . The bracket [ 1 ] [1] [ 1 ] records the degree shift, not a tensor power. The fields include ghosts, antifields and their differential-form partners. The natural pairings use L z ⊗ L z ∨ → C L_z\otimes L_z^{\vee}\to\mathbb{C} L z ⊗ L z ∨ → C .
Up to the standard Koszul sign convention, the odd symplectic form, action, and cohomological vector field are
ω X = ∫ X ⟨ δ B ∧ δ A ⟩ , deg ω X = − 1 , (11) \omega_X=\int_X\langle\delta B\wedge\delta A\rangle,\qquad\deg\omega_X=-1, \tag*{(11)} ω X = ∫ X ⟨ δ B ∧ δ A ⟩ , deg ω X = − 1 , (11)
S X = ∫ X ⟨ B ∧ d z A ⟩ , Q X A = d z A , Q X B = d z − 1 B . (12) S_X=\int_X\langle B\wedge d_zA\rangle,\qquad Q_XA=d_zA,\quad Q_XB=d_{z^{-1}}B. \tag*{(12)} S X = ∫ X ⟨ B ∧ d z A ⟩ , Q X A = d z A , Q X B = d z − 1 B . (12)
Here d z 2 = d z − 1 2 = 0 d_z^2=d_{z^{-1}}^2=0 d z 2 = d z − 1 2 = 0 because the background connections are flat. An integration by parts in δ S X \delta S_X δ S X creates the boundary term
α ∂ X = ∫ ∂ X ⟨ B ∂ ∧ δ A ∂ ⟩ , ω ∂ X = δ α ∂ X . (13) \alpha_{\partial X}=\int_{\partial X}\langle B_{\partial}\wedge\delta A_{\partial}\rangle,\qquad\omega_{\partial X}=\delta\alpha_{\partial X}. \tag*{(13)} α ∂ X = ∫ ∂ X ⟨ B ∂ ∧ δ A ∂ ⟩ , ω ∂ X = δ α ∂ X . (13)
The boundary two-form has degree zero and Q ∂ A ∂ = d z A ∂ Q_{\partial}A_{\partial}=d_zA_{\partial} Q ∂ A ∂ = d z A ∂ , Q ∂ B ∂ = d z − 1 B ∂ Q_{\partial}B_{\partial}=d_{z^{-1}}B_{\partial} Q ∂ B ∂ = d z − 1 B ∂ . It is Hamiltonian, with BFV charge represented by S ∂ = ∫ ∂ X ⟨ B ∂ ∧ d z A ∂ ⟩ S_{\partial}=\int_{\partial X}\langle B_{\partial}\wedge d_zA_{\partial}\rangle S ∂ = ∫ ∂ X ⟨ B ∂ ∧ d z A ∂ ⟩ . The BV and BFV structures satisfy the usual bulk–boundary compatibility relation ι Q X ω X = δ S X + π ∗ α ∂ X \iota_{Q_X}\omega_X=\delta S_X+\pi^*\alpha_{\partial X} ι Q X ω X = δ S X + π ∗ α ∂ X , up to the global convention for the sign of the boundary primitive. For a detailed account and the modified quantum master equation, see [6 , 7 ] .
It is important that the boundary pairing in (13) is between dual local systems. Replacing L z V L_z^V L z V by L z L_z L z without a trivialization would in general make the integral meaningless.
The sutured annuli as boundary conditions
Choose two disjoint, oppositely oriented meridional sutures on ∂ X K ≅ T 2 \partial X_K\cong T^2 ∂ X K ≅ T 2 . They cut the torus into two annuli; one is R − R_- R − , with oriented meridional core, and the closure of its complement is an annulus R − c R_-^c R − c . For defining the relative complex we only require the embedded annulus R − R_- R − ; we do not confuse R − c R_-^c R − c with the sutured R + R_+ R + after reserving annular neighborhoods of the sutures. They differ by boundary collars and have the same homotopy type.
A mixed linear boundary condition is
A ∣ R − = 0 , B ∣ R − c = 0. (14) A|_{R_-}=0,\qquad B|_{R_-^c}=0. \tag*{(14)} A ∣ R − = 0 , B ∣ R − c = 0. (14)
Consequently α ∂ X \alpha_{\partial X} α ∂ X vanishes on variations satisfying (14) : at each point of the boundary, either δ A \delta A δ A or B B B vanishes. The boundary condition is preserved by Q X Q_X Q X since the exterior covariant differential commutes with restriction. In a cellular description with dual CW structures, the two complementary complexes are naturally paired by Poincaré–Lefschetz duality:
C k ( X K , R − ; L z ) ⊗ C 3 − k ( X K , R − c ; L z V ) ⟶ C . (15) C^k(X_K,R_-;L_z)\otimes C^{3-k}(X_K,R_-^c;L_z^V)\longrightarrow\mathbb{C}. \tag*{(15)} C k ( X K , R − ; L z ) ⊗ C 3 − k ( X K , R − c ; L z V ) ⟶ C . (15)
The precise chain-level pairing can be constructed with a dual cellular decomposition. The corresponding gauge-fixed quadratic BF model has the underlying twisted differential C ∙ ( X K , R − ; L z ) C^\bullet(X_K,R_-;L_z) C ∙ ( X K , R − ; L z ) . Its one-loop determinant-line contribution is the torsion of this relative complex; this is the finite-dimensional manifestation of the BV–BFV boundary-polarization principle. It is not a claim that the unregularized continuum path integral is an ordinary number.
The annuli meet along meridional circles. Promoting (14) to a fully functorial quantum BV–BFV theory with polarized interfaces requires the treatment of corners and coherent state spaces. We work at the cellular torsion level , where the relative complex and all of its gluing maps are defined without appealing to a conjectural corner quantization.
Acyclic boundary and corner sectors
Lemma 3.1 (No boundary or corner residual cohomology). Let z ≠ 1 z\ne1 z = 1 and R ⊂ ∂ X K R\subset\partial X_K R ⊂ ∂ X K be a meridional annulus. The rank-one twisted cellular complexes of R R R , of its two meridional boundary circles, and of the entire torus ∂ X K \partial X_K ∂ X K are acyclic. If also Δ K ( z ) ≠ 0 \Delta_K(z)\ne0 Δ K ( z ) = 0 , then H ∗ ( X K , R ; L z ) = 0 H^\ast(X_K,R;L_z)=0 H ∗ ( X K , R ; L z ) = 0 .
Proof. A meridional circle has the cellular cochain complex
0 ⟶ C → z − 1 C ⟶ 0 , (16) 0\longrightarrow\mathbb{C}\xrightarrow{z-1}\mathbb{C}\longrightarrow0, \tag*{(16)} 0 ⟶ C z − 1 C ⟶ 0 , (16)
which is exact. An annulus retracts onto its meridional core. For the torus, use a one-vertex, two-edge, one-face cellulation with meridional holonomy z z z and longitudinal holonomy 1 1 1 :
0 ⟶ C → ( z − 1 0 ) C 2 → ( 0 z − 1 ) C ⟶ 0. (17) 0 \longrightarrow\mathbb{C} \xrightarrow{\begin{pmatrix} z^{-1} \\ 0 \end{pmatrix}} \mathbb{C}^{2} \xrightarrow{\begin{pmatrix} 0 & z-1 \end{pmatrix}} \mathbb{C} \longrightarrow0. \tag*{(17)} 0 ⟶ C ( z − 1 0 ) C 2 ( 0 z − 1 ) C ⟶ 0. (17)
It is exact. If Δ K ( z ) ≠ 0 \Delta_K(z) \ne0 Δ K ( z ) = 0 , the Fox complex of X K X_K X K is also exact. Finally, the long exact sequence of the pair ( X K , R ) (X_K,R) ( X K , R ) then proves that its relative twisted cohomology vanishes. □
The lemma means that the reduced boundary and corner classical cohomologies vanish in these sectors. It does not mean that all unreduced boundary BFV fields disappear, or that one has constructed a canonical equivalence with the nontrivial bordered Floer categories. In fact, this acyclicity helps explain why the scalar theory cannot retain Maslov-graded Floer information.
Determinant lines, torsion and the normalization issue
A self-contained algebraic definition is useful here. For a finite based cochain complex C ∙ C^\bullet C ∙ of vector spaces over a field E E E , put
Det C ∙ = ⨂ k ( det C k ) ( − 1 ) k . (18) \operatorname{Det} C^\bullet= \bigotimes_k (\det C^k)^{(-1)^k}. \tag*{(18)} Det C ∙ = k ⨂ ( det C k ) ( − 1 ) k . (18)
The decomposition into cycles, boundaries and cohomology induces a canonical determinant-line isomorphism Det C ∙ ≃ Det H ∙ ( C ) \operatorname{Det} C^\bullet\simeq\operatorname{Det} H^\bullet(C) Det C ∙ ≃ Det H ∙ ( C ) . If C C C is acyclic and cellular lifts specify bases, this map yields a nonzero scalar, its Reidemeister torsion; changing from cochains to chains or reversing the convention may invert this scalar. We fix the Milnor chain convention from Proposition 2.2 for all numerical formulas. The Gaussian/BV pushforward is being compared with the corresponding chosen torsion branch, not given a new normalization-independent definition.
For a non-unimodular rank-one representation, replacing the chosen lift of a k k k -cell by a deck translate changes the basis by a power of z z z . The torsion therefore changes by a unit ± z n \pm z^n ± z n ; this is precisely the familiar ± t n \pm t^n ± t n ambiguity of Alexander torsion. The cellular BV–BFV construction in [10 ] is stated for coefficient systems with determinant holonomy ± 1 \pm1 ± 1 . The specialization L ζ L_\zeta L ζ typically does not meet this hypothesis. In this paper we instead make fixed-lift comparisons of the finite-dimensional quadratic complexes, followed by the stated symmetric Alexander normalization. This restriction is substantial: the full non-unimodular quantum BV–BFV functor, with Euler-structure coherence, is not constructed here.
Proposition 3.2 (Boundary normalization as relative torsion). For every knot K K K and every z ≠ 1 z \ne1 z = 1 such that Δ K ( z ) ≠ 0 \Delta_K(z) \ne0 Δ K ( z ) = 0 , let R − R_- R − be a meridional annulus of its sutured exterior. In the Milnor torsion convention one has
τ ( X K , R − ; L z ) ≐ Δ K ( z ) , τ ( X K ; L z ) τ ( X K , R − ; L z ) = 1 z − 1 . (19) \tau(X_K,R_-;L_z) \doteq\Delta_K(z), \qquad\frac{\tau(X_K;L_z)}{\tau(X_K,R_-;L_z)} = \frac{1}{z-1}. \tag*{(19)} τ ( X K , R − ; L z ) ≐ Δ K ( z ) , τ ( X K , R − ; L z ) τ ( X K ; L z ) = z − 1 1 . (19) Both equalities are statements about torsion of acyclic based complexes; a normalized value is obtained by symmetrizing the Laurent polynomial.
Proof. Choose a CW decomposition of the pair ( X K , R − ) (X_K,R_-) ( X K , R − ) with compatible lifts and bases. The restriction or quotient of cellular chains provides a short exact sequence
0 ⟶ C ∙ ( R − ; L z ) ⟶ C ∙ ( X K ; L z ) ⟶ C ∙ ( X K , R − ; L z ) ⟶ 0. (20) 0 \longrightarrow C_\bullet(R_-;L_z) \longrightarrow C_\bullet(X_K;L_z) \longrightarrow C_\bullet(X_K,R_-;L_z) \longrightarrow0. \tag*{(20)} 0 ⟶ C ∙ ( R − ; L z ) ⟶ C ∙ ( X K ; L z ) ⟶ C ∙ ( X K , R − ; L z ) ⟶ 0. (20)
All three complexes are acyclic by Lemma 3.1 . The multiplicativity property of torsion for a based short exact sequence therefore yields τ ( X K ; L z ) = τ ( R − ; L z ) τ ( X K , R − ; L z ) \tau(X_K;L_z)=\tau(R_-;L_z)\tau(X_K,R_-;L_z) τ ( X K ; L z ) = τ ( R − ; L z ) τ ( X K , R − ; L z ) . The annulus has the homotopy type of (16) , whose torsion in our convention is ( z − 1 ) − 1 (z-1)^{-1} ( z − 1 ) − 1 . Proposition 2.2 gives τ ( X K ; L z ) = Δ K ( z ) / ( z − 1 ) \tau(X_K;L_z)=\Delta_K(z)/(z-1) τ ( X K ; L z ) = Δ K ( z ) / ( z − 1 ) . Division produces (3.10). □
Definition 3.3 (Symmetrized relative BF torsion character). For a knot exterior, define B K ( t ) ∈ Z [ t , t − 1 ] \mathcal{B}_K(t)\in\mathbb{Z}[t,t^{-1}] B K ( t ) ∈ Z [ t , t − 1 ] to be the unique representative of the meridional relative torsion class [ τ ( X K , R − ; F ϕ ) ] ∈ F × / ( ± t Z ) [\tau(X_K,R_-;F_\phi)]\in F^\times/(\pm t^{\mathbb{Z}}) [ τ ( X K , R − ; F ϕ )] ∈ F × / ( ± t Z ) that is symmetric in t ↔ t − 1 t\leftrightarrow t^{-1} t ↔ t − 1 and equals 1 1 1 at t = 1 t=1 t = 1 . At
any nonresonant character z z z , its value B K ( z ) \mathcal{B}_K(z) B K ( z ) is called the symmetrized relative BF torsion character .
Remark 3.4. Existence and uniqueness in Definition 3.3 follow from Proposition 3.2 and the normalization of Δ K \Delta_K Δ K . “Character” here denotes a function of the abelian holonomy, not a character of a Floer-category representation. The symmetric normalization is extra data added after cellular torsion: it is not a canonical value of a bare Gaussian integral.
Corollary 3.5. For every knot K K K one has the identity of normalized Laurent polynomials
B K ( t ) = Δ K ( t ) . (21) \mathcal{B}_K(t)=\Delta_K(t). \tag*{(21)} B K ( t ) = Δ K ( t ) . (21) For every nonresonant z z z , this is an equality of the symmetrically normalized determinant-line contribution of the mixed cellular BF boundary model with the Alexander polynomial evaluated at z z z .
Proof. This is Proposition 3.2 together with the uniqueness of the symmetric Alexander normalization.
The Floer side: sutures and decategorification
Sutured and knot Floer homology
4.1. Sutured and knot Floer homology. The knot Floer invariant H F K ^ m ( K , a ) \widehat{HFK}_m(K,a) H F K m ( K , a ) is a finitely generated group (or vector space after passing to a field) with a Maslov grading m m m and an Alexander grading a a a . It is computed from a doubly pointed Heegaard diagram by counting suitable holomorphic disks. The groups themselves depend on substantially more than an Euler-characteristic polynomial. In our comparison all Floer groups are taken over F 2 \mathbb{F}_2 F 2 and signs in their Euler characteristics use the Maslov parity fixed by the usual absolute normalization for knots in S 3 S^3 S 3 .
The balanced sutured manifold ( X K , γ μ ) (X_K,\gamma_\mu) ( X K , γ μ ) obtained using two oppositely oriented meridional sutures satisfies S F H ( X K , γ μ ) ≅ H F K ^ ( K ) SFH(X_K,\gamma_\mu)\cong\widehat{HFK}(K) S F H ( X K , γ μ ) ≅ H F K ( K ) as an ungraded sum, with the relevant Alexander grading recovered from the relative Spin c \operatorname{Spin}^c Spin c decomposition and a choice of Seifert surface [8 ] . The theorem of Friedl–Juhász–Rasmussen [11 ] identifies the relative torsion class of a balanced sutured manifold with the Euler-characteristic class of its sutured Floer homology, up to the conventional affine H 1 H_1 H 1 unit and sign. The familiar Ozsváth–Szabó identity then gives the normalized formula
∑ a χ ( H F K ^ ( K , a ) ) t a = Δ K ( t ) = B K ( t ) . (22) \sum_a \chi\bigl(\widehat{HFK}(K,a)\bigr)t^a=\Delta_K(t)=\mathcal{B}_K(t). \tag*{(22)} a ∑ χ ( H F K ( K , a ) ) t a = Δ K ( t ) = B K ( t ) . (22)
The left equality is the original knot Floer categorification statement; the right equality is the explicit meridional relative torsion calculation. Their coexistence is our precisely defined decategorified BV–BFV/Floer comparison .
Theorem 4.1 (Relative BV–BFV/Floer comparison and its exact scope). Fix a knot K K K , a meridional sutured exterior, and the Milnor and Alexander normalization conventions above. The following triangle commutes after passing to the common Laurent-polynomial decategorification:
BF r e l ( K ) → torsion τ ( X K , R − ) normalize ↓ ↓ normalize Δ K ( t ) = ∑ a χ ( H F K ^ ( K , a ) ) t a \begin{CD}
\operatorname{BF}_{\mathrm{rel}}(K) @>{\text{torsion}}>> \tau(X_K,\mathbb{R}_{-}) \\
@V{\text{normalize}}VV @VV{\text{normalize}}V \\
\Delta_K(t)=\sum_a\chi\bigl(\widehat{HFK}(K,a)\bigr)t^a
\end{CD} BF rel ( K ) normalize ↓ ⏐ Δ K ( t ) = a ∑ χ ( H F K ( K , a ) ) t a torsion τ ( X K , R − ) ↓ ⏐ normalize The two routes compute the same decategorified invariant. No quasi-isomorphism, natural transformation, or pairing theorem between the BV–BFV boundary-state complex and a bordered or sutured Floer chain complex follows from this triangle.
Proof. The determinant-line assignment is the finite-dimensional acyclic torsion construction of (18) . Its relative meridional evaluation is computed in Proposition 3.2, and its normalized representative is B K = Δ K \mathcal{B}_K=\Delta_K B K = Δ K by Corollary 3.5. The identification with the graded Euler-characteristic polynomial is (22) , with the cited Floer theorems as input. □
Remark 4.2 (The pre-existing categorification result). Hom, Lidman and Watson proved a genuine categorification of Donaldson’s Alexander-module TQFT using bordered Floer theory [16 ] . Their theorem provides substantially stronger Floer data than our triangle; it does not, as stated, identify that bordered theory with a BV–BFV state. Claiming such an identification would require an independent construction and proof.
Euler coefficients and Alexander support
Fix coefficients in F 2 \mathbb{F}_2 F 2 and keep the Maslov-parity convention from § 4 . For later arithmetic formulas, write the coefficient vector of the decategorified Floer invariant as
c a ( K ) = χ ( H F K ^ ( K , a ) ) = ∑ m ( − 1 ) m dim F 2 H F K ^ m ( K , a ) . (23) c_a(K)=\chi\left(\widehat{\mathrm{HFK}}(K,a)\right)=\sum_m(-1)^m\dim_{\mathbb{F}_2}\widehat{\mathrm{HFK}}_m(K,a). \tag*{(23)} c a ( K ) = χ ( HFK ( K , a ) ) = m ∑ ( − 1 ) m dim F 2 HFK m ( K , a ) . (23)
Ozsváth and Szabó’s Euler characteristic identity, together with the genus bound, gives
Δ K ( t ) = ∑ a ∈ Z c a ( K ) t a , c a ( K ) = 0 when ∣ a ∣ > g ( K ) . (24) \Delta_K(t)=\sum_{a\in\mathbb{Z}}c_a(K)t^a,\qquad c_a(K)=0\text{ when }|a|>g(K). \tag*{(24)} Δ K ( t ) = a ∈ Z ∑ c a ( K ) t a , c a ( K ) = 0 when ∣ a ∣ > g ( K ) . (24)
The symmetry c a = c − a c_a=c_{-a} c a = c − a and the identity ∑ a c a = 1 \sum_a c_a=1 ∑ a c a = 1 follow from the corresponding Alexander-polynomial properties. Notice that c a c_a c a does not determine H F K ^ ( K , a ) \widehat{\mathrm{HFK}}(K,a) HFK ( K , a ) as a graded group: cancellations in Euler characteristic discard information.
The established torsion–Floer connection is considerably richer than (24) . Friedl–Juhász–Rasmussen compare sutured Floer Euler characteristics with torsion; Hom–Lidman–Watson use bordered Floer theory to categorify Donaldson’s Alexander-module TQFT [11 , 16 ] . We do not construct a map between BV–BFV state complexes and bordered Floer modules here. Our contribution is a finite-character reconstruction statement for precisely the common decategorified information.
Exact cyclotomic reconstruction and optimality
A trace-orthogonality lemma
Let p p p be prime, ζ = e 2 π i / p \zeta=e^{2\pi i/p} ζ = e 2 π i / p , and E p = Q ( ζ ) E_p=\mathbb{Q}(\zeta) E p = Q ( ζ ) . The Galois group Gal ( E p / Q ) \operatorname{Gal}(E_p/\mathbb{Q}) Gal ( E p / Q ) consists of σ j ( ζ ) = ζ j \sigma_j(\zeta)=\zeta^j σ j ( ζ ) = ζ j for 1 ≤ j ≤ p − 1 1\le j\le p-1 1 ≤ j ≤ p − 1 .
Lemma 6.1 (Prime cyclotomic trace). For any integer m m m ,
Tr E p / Q ( ζ m ) = { p − 1 , p ∣ m , − 1 , p ∤ m . (25) \operatorname{Tr}_{E_p/\mathbb{Q}}(\zeta^m)=
\begin{cases}
p-1,&p\mid m,\\
-1,&p\nmid m.
\end{cases} \tag*{(25)} Tr E p / Q ( ζ m ) = { p − 1 , − 1 , p ∣ m , p ∤ m . (25) Proof. The trace is ∑ j = 1 p − 1 ζ j m \sum_{j=1}^{p-1}\zeta^{jm} ∑ j = 1 p − 1 ζ j m . If p ∣ m p\mid m p ∣ m , each summand is 1. Otherwise multiplication by m m m permutes the nonzero residue classes modulo p p p , so the sum is ∑ j = 1 p − 1 ζ j = − 1 \sum_{j=1}^{p-1}\zeta^j=-1 ∑ j = 1 p − 1 ζ j = − 1 . □
Proposition 6.2 (Exact extraction). Let P ( t ) = ∑ b = − g g c b t b ∈ Z [ t , t − 1 ] P(t)=\sum_{b=-g}^{g}c_b t^b\in\mathbb{Z}[t,t^{-1}] P ( t ) = ∑ b = − g g c b t b ∈ Z [ t , t − 1 ] satisfy P ( 1 ) = 1 P(1)=1 P ( 1 ) = 1 . If p ≥ 2 g + 1 p\ge2g+1 p ≥ 2 g + 1 is prime, then for every − g ≤ a ≤ g -g\le a\le g − g ≤ a ≤ g ,
c a = 1 + Tr E p / Q ( ζ − a P ( ζ ) ) p . (26) c_a=\frac{1+\operatorname{Tr}_{E_p/\mathbb{Q}}(\zeta^{-a}P(\zeta))}{p}. \tag*{(26)} c a = p 1 + Tr E p / Q ( ζ − a P ( ζ )) . (26) In particular, the map P ↦ P ( ζ ) P\mapsto P(\zeta) P ↦ P ( ζ ) is injective on the class of such normalized Laurent polynomials.
Proof. Since ∣ a − b ∣ ≤ 2 g < p |a-b| \le2g < p ∣ a − b ∣ ≤ 2 g < p , the only way p p p can divide b − a b-a b − a is b = a b=a b = a . Lemma 6.1 therefore gives
Tr ( ζ − a P ( ζ ) ) = ∑ b = − g g c b Tr ( ζ b − a ) = ( p − 1 ) c a − ∑ b ≠ a c b = p c a − P ( 1 ) = p c a − 1. \begin{aligned}
\operatorname{Tr}(\zeta^{-a}P(\zeta))=\sum_{b=-g}^{g}c_b\operatorname{Tr}(\zeta^{b-a}) \\
&=(p-1)c_a-\sum_{b\ne a}c_b=pc_a-P(1)=pc_a-1.
\end{aligned} Tr ( ζ − a P ( ζ )) = b = − g ∑ g c b Tr ( ζ b − a ) = ( p − 1 ) c a − b = a ∑ c b = p c a − P ( 1 ) = p c a − 1.
Rearrange to obtain (26) . Equality of evaluation values implies equality of every c a c_a c a .
Proof of Theorem 1.2. Lemma 2.4 and Lemma 2.5 give assertion (i). Proposition 2.2 gives Δ K ( ζ ) = ( ζ − 1 ) τ K s y m ( ζ ) \Delta_K(\zeta)=(\zeta-1)\tau_K^{\mathrm{sym}}(\zeta) Δ K ( ζ ) = ( ζ − 1 ) τ K sym ( ζ ) , equivalently B K ( ζ ) = Δ K ( ζ ) \mathcal{B}_K(\zeta)=\Delta_K(\zeta) B K ( ζ ) = Δ K ( ζ ) by Corollary 3.5; apply Proposition 6.2 to (24) to obtain (4) . Assertion (iii) follows from injectivity in Proposition 6.2.
Corollary 6.3 (Arithmetic integrality constraints). Under the hypotheses of Theorem 1.2, let α = ( ζ − 1 ) τ K s y m ( ζ ) = Δ K ( ζ ) \alpha=(\zeta-1)\tau_K^{\mathrm{sym}}(\zeta)=\Delta_K(\zeta) α = ( ζ − 1 ) τ K sym ( ζ ) = Δ K ( ζ ) . Then
Tr E p / Q ( ζ − a α ) ≡ − 1 ( m o d p ) ( − g ≤ a ≤ g ) , (27) \operatorname{Tr}_{E_p/\mathbb{Q}}(\zeta^{-a}\alpha)\equiv-1\pmod p\qquad(-g\le a\le g), \tag*{(27)} Tr E p / Q ( ζ − a α ) ≡ − 1 ( mod p ) ( − g ≤ a ≤ g ) , (27)
and these traces determine the entire Euler-characteristic polynomial. In particular, α \alpha α is an algebraic integer, fixed by complex conjugation.
Proof. The congruence is exactly Tr ( ζ − a α ) = p c a ( K ) − 1 \operatorname{Tr}(\zeta^{-a}\alpha)=pc_a(K)-1 Tr ( ζ − a α ) = p c a ( K ) − 1 . Integrality is clear from α ∈ Z [ ζ ] \alpha\in\mathbb{Z}[\zeta] α ∈ Z [ ζ ] ; symmetry of Δ K \Delta_K Δ K gives α ‾ = Δ K ( ζ − 1 ) = α \overline{\alpha}=\Delta_K(\zeta^{-1})=\alpha α = Δ K ( ζ − 1 ) = α .
Remark 6.4 (What “one value” means). A single exact element of E p E_p E p is not the same as an unstructured complex floating-point approximation to one embedding of that element. Formula (4) requires the exact field trace, equivalently all Galois conjugates determined by an exact representation of α \alpha α . Small numerical error in a single complex approximation is not by itself sufficient to recover an exact algebraic number without further height or coefficient bounds.
A sharp counterexample below the genus threshold
We now prove that the inequality p ≥ 2 g + 1 p\ge2g+1 p ≥ 2 g + 1 is the precise uniform threshold when one allows all knots of genus at most g g g . Define
R p ( t ) = t − ( p − 1 ) / 2 Φ p ( t ) ( t + t − 1 − 2 ) , P p ( t ) = 1 + R p ( t ) . (28) R_p(t)=t^{-(p-1)/2}\Phi_p(t)(t+t^{-1}-2),\qquad P_p(t)=1+R_p(t). \tag*{(28)} R p ( t ) = t − ( p − 1 ) /2 Φ p ( t ) ( t + t − 1 − 2 ) , P p ( t ) = 1 + R p ( t ) . (28)
Lemma 6.5. For any odd prime p p p , the Laurent polynomial P p P_p P p satisfies
P p ( t − 1 ) = P p ( t ) , P p ( 1 ) = 1 , P p ( ζ ) = 1 , P_p(t^{-1})=P_p(t),\qquad P_p(1)=1,\qquad P_p(\zeta)=1, P p ( t − 1 ) = P p ( t ) , P p ( 1 ) = 1 , P p ( ζ ) = 1 , and is monic, with highest and lowest exponents equal to $\pm(p+1)/2.
Proof. The normalized cyclotomic polynomial t − ( p − 1 ) / 2 Φ p ( t ) t^{-(p-1)/2}\Phi_p(t) t − ( p − 1 ) /2 Φ p ( t ) is invariant under t ↔ t − 1 t\leftrightarrow t^{-1} t ↔ t − 1 ; so is t + t − 1 − 2 t+t^{-1}-2 t + t − 1 − 2 . The latter factor vanishes at t = 1 t=1 t = 1 , while Φ p ( ζ ) = 0 \Phi_p(\zeta)=0 Φ p ( ζ ) = 0 . The highest exponent in R p R_p R p is ( p − 1 ) / 2 + 1 = ( p + 1 ) / 2 (p-1)/2+1=(p+1)/2 ( p − 1 ) /2 + 1 = ( p + 1 ) /2 and its coefficient is 1. The same holds at the lowest exponent by symmetry, and adding 1 does not change these extreme terms.
Proof of Theorem 1.3. By Lemma 6.5, P p ( t ) P_p(t) P p ( t ) is an admissible monic knot Alexander polynomial. A realization theorem of Stoimenow [21 ] produces a fibered knot J p J_p J p whose normalized Alexander polynomial is P p P_p P p . For fibered knots, the genus equals the maximal exponent in the normalized Alexander polynomial; therefore g ( J p ) = ( p + 1 ) / 2 g(J_p)=(p+1)/2 g ( J p ) = ( p + 1 ) /2 . Since P p ( ζ ) = 1 = Δ U ( ζ ) P_p(\zeta)=1=\Delta_U(\zeta) P p ( ζ ) = 1 = Δ U ( ζ ) , division by ζ − 1 \zeta-1 ζ − 1 yields equality of their symmetrically normalized torsions at ζ \zeta ζ , although P p ≠ 1 P_p\ne1 P p = 1 . This knot lies in the genus-at-most-g g g class when g ≥ ( p + 1 ) / 2 g\ge(p+1)/2 g ≥ ( p + 1 ) /2 , equivalently p < 2 g + 1 p<2g+1 p < 2 g + 1 for odd p p p .
Remark 6.6. The obstruction does not come from a zero of Δ K ( ζ ) \Delta_K(\zeta) Δ K ( ζ ) : prime cyclotomic Alexander evaluations never vanish. Rather, Φ p ( t ) \Phi_p(t) Φ p ( t ) lies in the kernel of evaluation at ζ \zeta ζ , and multiplying it by the symmetric factor t + t − 1 − 2 t+t^{-1}-2 t + t − 1 − 2 preserves P ( 1 ) = 1 P(1)=1 P ( 1 ) = 1 . The genus bound rules out exactly this type of aliasing when p ≥ 2 g + 1 p\ge2g+1 p ≥ 2 g + 1 .
Cyclotomic norms and branched covers
Let Σ p ( K ) \Sigma_p(K) Σ p ( K ) be the p p p -fold cyclic branched cover of S 3 S^3 S 3 along K K K . The classical Fox resultant formula states that, whenever the resulting group is finite,
∣ H 1 ( Σ p ( K ) ; Z ) ∣ = ∣ ∏ j = 1 p − 1 Δ K ( ζ j ) ∣ . (29) \left|H_1(\Sigma_p(K);\mathbb{Z})\right| = \left|\prod_{j=1}^{p-1} \Delta_K(\zeta^j)\right|. \tag*{(29)} ∣ H 1 ( Σ p ( K ) ; Z ) ∣ = j = 1 ∏ p − 1 Δ K ( ζ j ) . (29)
For prime p p p , Lemma 2.5 guarantees finiteness. The product on the right is an algebraic field norm.
Theorem 7.1 (Simultaneous trace and norm recovery). Suppose p ≥ 2 g ( K ) + 1 p \ge2g(K)+1 p ≥ 2 g ( K ) + 1 is an odd prime. Set α K = ( ζ − 1 ) τ K s y m ( ζ ) \alpha_K=(\zeta-1)\tau_K^{\mathrm{sym}}(\zeta) α K = ( ζ − 1 ) τ K sym ( ζ ) . Then the single exact element α K ∈ Z [ ζ ] \alpha_K\in\mathbb{Z}[\zeta] α K ∈ Z [ ζ ] determines both
∣ H 1 ( Σ p ( K ) ; Z ) ∣ = ∣ N E p / Q ( α K ) ∣ (30) \left|H_1(\Sigma_p(K);\mathbb{Z})\right|=\left|N_{E_p/\mathbb{Q}}(\alpha_K)\right| \tag*{(30)} ∣ H 1 ( Σ p ( K ) ; Z ) ∣ = N E p / Q ( α K ) (30) and, for every Alexander grading a a a ,
χ ( H F K ^ ( K , a ) ) = 1 + Tr E p / Q ( ζ − a α K ) p . (31) \chi\bigl(\widehat{\mathrm{HFK}}(K,a)\bigr)=\frac{1+\operatorname{Tr}_{E_p/\mathbb{Q}}(\zeta^{-a}\alpha_K)}{p}. \tag*{(31)} χ ( HFK ( K , a ) ) = p 1 + Tr E p / Q ( ζ − a α K ) . (31) For odd p p p , the norm in (30) is positive.
Proof. The trace identity is Theorem 1.2 . The norm identity follows from (29) and the fact that Δ K ( ζ j ) = σ j ( α K ) \Delta_K(\zeta^j)=\sigma_j(\alpha_K) Δ K ( ζ j ) = σ j ( α K ) . Since the embeddings indexed by j j j and p − j p-j p − j take the same real nonzero value on the symmetric Alexander polynomial, the norm is a product of positive squares. □
Remark 7.2. The trace and norm yield different information from the same arithmetic datum. The norm alone is far from injective on Alexander polynomials, and neither the trace nor the norm determines the full bigraded knot Floer group. Theorem 7.1 is a package of two known types of topological information, not a recovery of Floer differentials or torsion-sensitive chain complexes.
Connected-sum gluing and Floer convolution
Let K 1 K_1 K 1 and K 2 K_2 K 2 be knots. Their connected-sum exterior decomposes along a properly embedded meridional annulus A ≃ S 1 × [ 0 , 1 ] A\simeq S^1\times[0,1] A ≃ S 1 × [ 0 , 1 ] into the two exteriors, with the same meridional holonomy t t t on all three pieces. The corresponding local system on A A A is acyclic when t ≠ 1 t\ne1 t = 1 , and its torsion in our convention is τ ( A ; t ) = ( t − 1 ) − 1 \tau(A;t)=(t-1)^{-1} τ ( A ; t ) = ( t − 1 ) − 1 .
Relative gluing is the natural sutured multiplication
The factor t − 1 t-1 t − 1 appearing in absolute gluing vanishes when we formulate the invariant in the meridional relative boundary condition. This is not an ad hoc renormalization: the annulus is already part of the sutured boundary data.
Proposition 8.1 (Relative multiplicativity). For knots K 1 , K 2 K_1,K_2 K 1 , K 2 and K = K 1 # K 2 K=K_1\#K_2 K = K 1 # K 2 , choose compatible meridional relative boundary data. Their symmetrized relative B F BF B F characters satisfy the exact identity
B K 1 # K 2 ( t ) = B K 1 ( t ) B K 2 ( t ) . (32) \mathcal{B}_{K_1\#K_2}(t)=\mathcal{B}_{K_1}(t)\mathcal{B}_{K_2}(t). \tag*{(32)} B K 1 # K 2 ( t ) = B K 1 ( t ) B K 2 ( t ) . (32) Before symmetrization the identity is an equality of torsion classes modulo ± t Z \pm t^{\mathbb{Z}} ± t Z , compatible with the gluing of determinant lines in the cellular B F BF B F system.
Proof. Decompose the exterior along the meridional connected-sum annulus A A A . For nontrivial holonomy the Mayer–Vietoris cellular sequence
0 → C ∙ ( A ; L z ) → C ∙ ( X K 1 ; L z ) ⊕ C ∙ ( X K 2 ; L z ) → C ∙ ( X K ; L z ) → 0 0\to C_\bullet(A;L_z)\to C_\bullet(X_{K_1};L_z)\oplus C_\bullet(X_{K_2};L_z)\to C_\bullet(X_K;L_z)\to0 0 → C ∙ ( A ; L z ) → C ∙ ( X K 1 ; L z ) ⊕ C ∙ ( X K 2 ; L z ) → C ∙ ( X K ; L z ) → 0
is a short exact sequence of acyclic complexes. Its determinant lines multiply with an inverse factor τ ( A ; z ) − 1 = z − 1 \tau(A;z)^{-1}=z-1 τ ( A ; z ) − 1 = z − 1 , which gives Proposition 8.2 below. Dividing each absolute torsion by the torsion of its meridional boundary annulus now gives
τ ( X K ; z ) τ ( R − ; z ) ≐ τ ( X K 1 ; z ) τ ( X K 2 ; z ) τ ( R − , 1 ; z ) τ ( R − , 2 ; z ) \frac{\tau(X_K;z)}{\tau(R_-;z)} \doteq\frac{\tau(X_{K_1};z)\tau(X_{K_2};z)}{\tau(R_{-,1};z)\tau(R_{-,2};z)} τ ( R − ; z ) τ ( X K ; z ) ≐ τ ( R − , 1 ; z ) τ ( R − , 2 ; z ) τ ( X K 1 ; z ) τ ( X K 2 ; z )
The same identity follows from the relative Mayer–Vietoris sequences after replacing the boundary annuli by compatible subcomplexes. Finally, symmetric representatives have constant value 1 1 1 at t = 1 t=1 t = 1 , so the residual sign and monomial ambiguity is removed. This proves (32) . □
Proposition 8.2 (Scalar torsion gluing). With the symmetric normalization (3) ,
τ K 1 # K 2 s y m ( t ) = ( t − 1 ) τ K 1 s y m ( t ) τ K 2 s y m ( t ) . (33) \tau^{\mathrm{sym}}_{K_1\#K_2}(t)=(t-1)\tau^{\mathrm{sym}}_{K_1}(t)\tau^{\mathrm{sym}}_{K_2}(t). \tag*{(33)} τ K 1 # K 2 sym ( t ) = ( t − 1 ) τ K 1 sym ( t ) τ K 2 sym ( t ) . (33) This agrees with the multiplicativity of acyclic torsion for the Mayer–Vietoris short exact sequence, after correcting by the annulus torsion.
Proof. The standard connected-sum identity is Δ K 1 # K 2 = Δ K 1 Δ K 2 \Delta_{K_1\#K_2}=\Delta_{K_1}\Delta_{K_2} Δ K 1 # K 2 = Δ K 1 Δ K 2 . Dividing this identity by t − 1 t-1 t − 1 gives (33) exactly, with all symmetric normalizations fixed. Independently, cellular Mayer–Vietoris gives a short exact sequence of acyclic twisted complexes 0 → C ∗ ( A ) → C ∗ ( X K 1 ) ⊕ C ∗ ( X K 2 ) → C ∗ ( X K 1 # K 2 ) → 0 0\to C_*(A)\to C_*(X_{K_1})\oplus C_*(X_{K_2})\to C_*(X_{K_1\#K_2})\to0 0 → C ∗ ( A ) → C ∗ ( X K 1 ) ⊕ C ∗ ( X K 2 ) → C ∗ ( X K 1 # K 2 ) → 0 . Multiplicativity of torsion gives τ ( X K 1 # K 2 ) ≐ τ ( X K 1 ) τ ( X K 2 ) / τ ( A ) \tau(X_{K_1\#K_2})\doteq\tau(X_{K_1})\tau(X_{K_2})/\tau(A) τ ( X K 1 # K 2 ) ≐ τ ( X K 1 ) τ ( X K 2 ) / τ ( A ) , whence the factor t − 1 t-1 t − 1 ; symmetric normalization removes the units. □
Theorem 8.3 (Trace–gluing compatibility). Let g i ≥ g ( K i ) g_i\ge g(K_i) g i ≥ g ( K i ) and take an odd prime p ≥ 2 ( g 1 + g 2 ) + 1 p\ge2(g_1+g_2)+1 p ≥ 2 ( g 1 + g 2 ) + 1 . For ζ = e 2 π i / p \zeta=e^{2\pi i/p} ζ = e 2 π i / p , set α i = ( ζ − 1 ) τ K i s y m ( ζ ) = Δ K i ( ζ ) \alpha_i=(\zeta-1)\tau^{\mathrm{sym}}_{K_i}(\zeta)=\Delta_{K_i}(\zeta) α i = ( ζ − 1 ) τ K i sym ( ζ ) = Δ K i ( ζ ) . Then
1 + Tr Q ( ζ ) / Q ( ζ − a α 1 α 2 ) p = ∑ b ∈ Z c b ( K 1 ) c a − b ( K 2 ) = c a ( K 1 # K 2 ) . (34) \frac{1+\operatorname{Tr}_{\mathbb{Q}(\zeta)/\mathbb{Q}}(\zeta^{-a}\alpha_1\alpha_2)}{p}=\sum_{b\in\mathbb{Z}}c_b(K_1)c_{a-b}(K_2)=c_a(K_1\#K_2). \tag*{(34)} p 1 + Tr Q ( ζ ) / Q ( ζ − a α 1 α 2 ) = b ∈ Z ∑ c b ( K 1 ) c a − b ( K 2 ) = c a ( K 1 # K 2 ) . (34) Equivalently, the torsion-gluing product corresponds under cyclotomic coefficient extraction to the convolution of knot Floer Euler characteristics.
Proof. Multiplicativity gives α 1 α 2 = Δ K 1 # K 2 ( ζ ) \alpha_1\alpha_2=\Delta_{K_1\#K_2}(\zeta) α 1 α 2 = Δ K 1 # K 2 ( ζ ) ; the degree support is contained in [ − ( g 1 + g 2 ) , g 1 + g 2 ] [-(g_1+g_2),g_1+g_2] [ − ( g 1 + g 2 ) , g 1 + g 2 ] . Apply Proposition 6.2 to the product. The coefficient of t a t^a t a in Δ K 1 ( t ) Δ K 2 ( t ) \Delta_{K_1}(t)\Delta_{K_2}(t) Δ K 1 ( t ) Δ K 2 ( t ) is the displayed convolution; it equals the Euler characteristic of the knot Floer connected-sum tensor product by the connected-sum formula of [18 ] . □
Remark 8.4 (Scope of gluing). Proposition 8.2 is a scalar torsion statement, which is compatible with the expected determinant-line behavior of Abelian BF theory. It is not a construction of a BV–BFV gluing equivalence with the bordered Floer pairing theorem. Such a categorified comparison would require additional chain-level and boundary-polarization data, and is not proved here.
Examples
The trefoil and the figure-eight knot
Both the right-handed trefoil 3 1 3_1 3 1 and the figure-eight knot 4 1 4_1 4 1 have genus 1 1 1 , so p = 3 p=3 p = 3 is admissible. Let ζ = e 2 π i / 3 \zeta=e^{2\pi i/3} ζ = e 2 π i /3 , so ζ + ζ − 1 = − 1 \zeta+\zeta^{-1}=-1 ζ + ζ − 1 = − 1 and E 3 = Q ( − 3 ) E_3=\mathbb{Q}(\sqrt{-3}) E 3 = Q ( − 3 ) .
For the trefoil,
Δ 3 1 ( t ) = t − 1 − 1 + t , α 3 1 = Δ 3 1 ( ζ ) = − 2. (35) \Delta_{3_1}(t)=t^{-1}-1+t,\qquad\alpha_{3_1}=\Delta_{3_1}(\zeta)=-2. \tag*{(35)} Δ 3 1 ( t ) = t − 1 − 1 + t , α 3 1 = Δ 3 1 ( ζ ) = − 2. (35)
The trace of the rational element − 2 -2 − 2 is − 4 -4 − 4 , and
Tr ( ζ − 1 α 3 1 ) = − 2 ( ζ + ζ − 1 ) = 2. \operatorname{Tr}(\zeta^{-1}\alpha_{3_1})=-2(\zeta+\zeta^{-1})=2. Tr ( ζ − 1 α 3 1 ) = − 2 ( ζ + ζ − 1 ) = 2.
Consequently c 0 = ( 1 − 4 ) / 3 = − 1 c_0=(1-4)/3=-1 c 0 = ( 1 − 4 ) /3 = − 1 and c ± 1 = ( 1 + 2 ) / 3 = 1 c_{\pm1}=(1+2)/3=1 c ± 1 = ( 1 + 2 ) /3 = 1 . The norm is N ( − 2 ) = 4 N(-2)=4 N ( − 2 ) = 4 , recovering ∣ H 1 ( Σ 3 ( 3 1 ) ) ∣ = 4 |H_1(\Sigma_3(3_1))|=4 ∣ H 1 ( Σ 3 ( 3 1 )) ∣ = 4 .
For the figure-eight,
Δ 4 1 ( t ) = 3 − t − t − 1 , α 4 1 = 4. (36) \Delta_{4_1}(t)=3-t-t^{-1},\qquad\alpha_{4_1}=4. \tag*{(36)} Δ 4 1 ( t ) = 3 − t − t − 1 , α 4 1 = 4. (36)
Now Tr ( 4 ) = 8 \operatorname{Tr}(4)=8 Tr ( 4 ) = 8 and Tr ( 4 ζ − 1 ) = − 4 \operatorname{Tr}(4\zeta^{-1})=-4 Tr ( 4 ζ − 1 ) = − 4 , so c 0 = 3 c_0=3 c 0 = 3 and c ± 1 = − 1 c_{\pm1}=-1 c ± 1 = − 1 . The norm gives ∣ H 1 ( Σ 3 ( 4 1 ) ) ∣ = 16 \lvert H_1(\Sigma_3(4_1))\rvert=16 ∣ H 1 ( Σ 3 ( 4 1 ))∣ = 16 .
The cinquefoil and the sharpness threshold
The ( 2 , 5 ) (2,5) ( 2 , 5 ) torus knot has genus 2 and
Δ T ( 2 , 5 ) ( t ) = t − 2 − t − 1 + 1 − t + t 2 . (37) \Delta_{T(2,5)}(t)=t^{-2}-t^{-1}+1-t+t^2. \tag*{(37)} Δ T ( 2 , 5 ) ( t ) = t − 2 − t − 1 + 1 − t + t 2 . (37)
At a primitive cube root ζ 3 \zeta_3 ζ 3 , one has t − 2 + t 2 = t + t − 1 t^{-2}+t^2=t+t^{-1} t − 2 + t 2 = t + t − 1 , hence Δ T ( 2 , 5 ) ( ζ 3 ) = 1 = Δ U ( ζ 3 ) \Delta_{T(2,5)}(\zeta_3)=1=\Delta_U(\zeta_3) Δ T ( 2 , 5 ) ( ζ 3 ) = 1 = Δ U ( ζ 3 ) . Therefore a single order-three torsion sector cannot distinguish even the Euler-characteristic polynomials of these two knots. The obstruction polynomial (28) for p = 3 p=3 p = 3 is precisely Δ T ( 2 , 5 ) \Delta_{T(2,5)} Δ T ( 2 , 5 ) .
At a primitive fifth root ζ 5 \zeta_5 ζ 5 , by contrast, Theorem 1.2 applies. Set s = ζ 5 + ζ 5 − 1 s=\zeta_5+\zeta_5^{-1} s = ζ 5 + ζ 5 − 1 . Since
1 + s + ( ζ 5 2 + ζ 5 − 2 ) = 0 , 1+s+(\zeta_5^2+\zeta_5^{-2})=0, 1 + s + ( ζ 5 2 + ζ 5 − 2 ) = 0 ,
α T ( 2 , 5 ) = 1 − s + ( ζ 5 2 + ζ 5 − 2 ) = − 2 s . \alpha_{T(2,5)}=1-s+(\zeta_5^2+\zeta_5^{-2})=-2s. α T ( 2 , 5 ) = 1 − s + ( ζ 5 2 + ζ 5 − 2 ) = − 2 s .
The trace identity gives
Tr ( α ) = 4 , Tr ( ζ 5 − 1 α ) = − 6 , Tr ( ζ 5 − 2 α ) = 4 , \operatorname{Tr}(\alpha)=4,\qquad\operatorname{Tr}(\zeta_5^{-1}\alpha)=-6,\qquad\operatorname{Tr}(\zeta_5^{-2}\alpha)=4, Tr ( α ) = 4 , Tr ( ζ 5 − 1 α ) = − 6 , Tr ( ζ 5 − 2 α ) = 4 ,
therefore c 0 = 1 c_0=1 c 0 = 1 , c ± 1 = − 1 c_{\pm1}=-1 c ± 1 = − 1 , and c ± 2 = 1 c_{\pm2}=1 c ± 2 = 1 , as expected. The field norm is N E 5 / Q ( − 2 s ) = 16 N_{E_5/\mathbb{Q}}(-2s)=16 N E 5 / Q ( − 2 s ) = 16 .
Connected sum of two trefoils
For K = 3 1 # 3 1 K=3_1\#3_1 K = 3 1 # 3 1 , the genus is 2 and
Δ K ( t ) = ( t − 1 − 1 + t ) 2 = t − 2 − 2 t − 1 + 3 − 2 t + t 2 . \Delta_K(t)=(t^{-1}-1+t)^2=t^{-2}-2t^{-1}+3-2t+t^2. Δ K ( t ) = ( t − 1 − 1 + t ) 2 = t − 2 − 2 t − 1 + 3 − 2 t + t 2 .
The first admissible prime is p = 5 p=5 p = 5 . In particular the Alexander-graded Euler characteristics are ( 1 , − 2 , 3 , − 2 , 1 ) (1,-2,3,-2,1) ( 1 , − 2 , 3 , − 2 , 1 ) in degrees − 2 , − 1 , 0 , 1 , 2 -2,-1,0,1,2 − 2 , − 1 , 0 , 1 , 2 , agreeing with the convolution of the trefoil coefficients. Torsion gluing gives τ K s y m = ( t − 1 ) ( τ 3 1 s y m ) 2 \tau_K^{\mathrm{sym}}=(t-1)(\tau_{3_1}^{\mathrm{sym}})^2 τ K sym = ( t − 1 ) ( τ 3 1 sym ) 2 , and Theorem 8.3 reconstructs exactly those coefficients from the fifth-root torsion value.
Limits of reconstruction from scalar torsion
A natural temptation is to regard the collection of all complex rank-one BF torsion sectors as a model of knot Floer homology itself, rather than of its Euler characteristic. The following elementary obstruction specifies why a further enhancement is necessary.
Proposition 10.1 (Scalar torsion does not determine knot Floer homology). There is no assignment that recovers the isomorphism type of the Alexander-graded groups H F K ^ ( K , a ) \widehat{\mathrm{HFK}}(K,a) HFK ( K , a ) from the family of scalar normalized rank-one torsions { τ K s y m ( z ) } z ≠ 1 \{\tau_K^{\mathrm{sym}}(z)\}_{z\ne1} { τ K sym ( z ) } z = 1 alone, even when all values, or equivalently the entire rational torsion function, are given.
Proof. Let U U U be the unknot and let K T KT K T be a nontrivial Kinoshita–Terasaka knot of genus 2. The latter satisfies Δ K T ( t ) = 1 = Δ U ( t ) \Delta_{KT}(t)=1=\Delta_U(t) Δ K T ( t ) = 1 = Δ U ( t ) , a standard example discussed in the knot Floer calculations of Ozsváth and Szabó [19 ] . Hence both knots have identical normalized rank-one torsions 1 / ( t − 1 ) 1/(t-1) 1/ ( t − 1 ) , and therefore identical evaluations at every z ≠ 1 z\ne1 z = 1 . Nevertheless H F K ^ ( U , a ) = 0 \widehat{\mathrm{HFK}}(U,a)=0 HFK ( U , a ) = 0 for all a ≠ 0 a\ne0 a = 0 , whereas genus detection for knot Floer homology yields H F K ^ ( K T , 2 ) ≠ 0 \widehat{\mathrm{HFK}}(KT,2)\ne0 HFK ( K T , 2 ) = 0 [20 ] . Their Alexander-graded knot Floer homologies cannot be isomorphic. □
Remark 10.2. The proposition is a limitation on reconstruction from scalar Abelian torsion . It does not rule out categorification using additional BV–BFV boundary structures, non-Abelian sectors, observables, or a Floer-theoretic chain complex. It also explains why even an exact knowledge of all the cyclotomic trace and norm data in the preceding sections stops at decategorification.
Linear BV–BFV gluing
The preceding comparison uses torsion, which discards the differential up to acyclic simple-homotopy information. We now retain the entire cellular cochain complexes long enough to implement cutting and regluing by an explicit chain homotopy. The construction concerns the linear BV–BFV model and its cellular restriction maps.
The boundary Q Q Q -complex and two explicit contractions
Let k k k be a field and z ∈ k × z \in k^{\times} z ∈ k × with z ≠ 1 z \ne1 z = 1 , and put u = z − 1 u=z-1 u = z − 1 . Fix a cellular decomposition of a boundary interface Y Y Y together with its twisted local system L z L_z L z . The linearized A \mathcal{A} A -part of the BFV field complex is C ∙ ( Y ; L z ) [ 1 ] C^\bullet(Y;L_z)[1] C ∙ ( Y ; L z ) [ 1 ] ; the B \mathcal{B} B -part is the dual-coefficient complex C ∙ ( Y ; L z ∨ ) [ 1 ] C^\bullet(Y;L_z^\vee)[1] C ∙ ( Y ; L z ∨ ) [ 1 ] on the dual cell structure. Their Q Q Q -operators are the cellular coboundaries. The finite-dimensional boundary symplectic pairing is induced by evaluation L z ⊗ L z ∨ → k L_z \otimes L_z^\vee\to k L z ⊗ L z ∨ → k and cellular Poincaré duality. These statements specify the linearized boundary fields ; a quantum BFV state complex also requires a polarization, a measure and possibly corner data.
Lemma 11.1 (Explicit meridional interface contractions). For z ≠ 1 z \ne1 z = 1 , the following two complexes admit the displayed contracting homotopies h h h satisfying d h + h d = 1 dh+hd=1 d h + h d = 1 and h 2 = 0 h^2=0 h 2 = 0 :
(i) On a meridional annulus A A A , deformation retracting onto its core circle, use
D A 0 = k → u D A 1 = k , h A : D A 1 → D A 0 , h A ( b ) = u − 1 b . D_A^0=k \xrightarrow{u} D_A^1=k,\qquad h_A:D_A^1\to D_A^0,\qquad h_A(b)=u^{-1}b. D A 0 = k u D A 1 = k , h A : D A 1 → D A 0 , h A ( b ) = u − 1 b . (ii) On a torus T 2 T^2 T 2 with meridional holonomy z z z and longitudinal holonomy 1 1 1 , use the complex
D T 0 = k → d 0 D T 1 = k 2 → d 1 D T 2 = k , d 0 ( c ) = ( u c , 0 ) , d 1 ( a , b ) = u b , D_T^0=k \xrightarrow{d^0} D_T^1=k^2 \xrightarrow{d^1} D_T^2=k,\qquad d^0(c)=(uc,0),\qquad d^1(a,b)=ub, D T 0 = k d 0 D T 1 = k 2 d 1 D T 2 = k , d 0 ( c ) = ( u c , 0 ) , d 1 ( a , b ) = u b , with
h T 1 ( a , b ) = u − 1 a , h T 2 ( c ) = ( 0 , u − 1 c ) , h T 0 = 0. h_T^1(a,b)=u^{-1}a,\qquad h_T^2(c)=(0,u^{-1}c),\qquad h_T^0=0. h T 1 ( a , b ) = u − 1 a , h T 2 ( c ) = ( 0 , u − 1 c ) , h T 0 = 0. The corresponding complexes with coefficients in L z ∨ L_z^\vee L z ∨ contract by the same formulas with u u u replaced by z − 1 − 1 z^{-1}-1 z − 1 − 1 .
Proof. For the annulus, d h dh d h and h d hd h d act as the identity in degrees 1 and 0, respectively. For the torus, one computes
h T 1 d 0 ( c ) = c , ( d 0 h T 1 + h T 2 d 1 ) ( a , b ) = ( a , 0 ) + ( 0 , b ) = ( a , b ) , d 1 h T 2 ( c ) = c . h_T^1d^0(c)=c,\qquad(d^0h_T^1+h_T^2d^1)(a,b)=(a,0)+(0,b)=(a,b),\qquad d^1h_T^2(c)=c. h T 1 d 0 ( c ) = c , ( d 0 h T 1 + h T 2 d 1 ) ( a , b ) = ( a , 0 ) + ( 0 , b ) = ( a , b ) , d 1 h T 2 ( c ) = c .
All compositions h 2 h^2 h 2 vanish by the displayed definitions. Dualizing the character simply replaces its meridional holonomy. □
Remark 11.2 (The resonant point). The contraction contains the coefficient ( z − 1 ) − 1 (z-1)^{-1} ( z − 1 ) − 1 and is therefore meromorphic in the holonomy. It has no extension to a contracting homotopy over k [ z , z − 1 ] k[z,z^{-1}] k [ z , z − 1 ] near z = 1 z=1 z = 1 : already for the meridional circle, u h = 1 uh=1 u h = 1 would require u u u to be a unit. At z = 1 z=1 z = 1 the cohomology of both the annulus and torus is nonzero. Thus the nonresonant chain-level gluing mechanism below cannot be continued unchanged through trivial holonomy. This pole is a property of the explicit linear complex , not an assertion about a singularity of any completed quantum BV–BFV theory.
A homotopy pullback with an explicit strong deformation retract
Consider a finite CW decomposition of a pair ( X , R ) (X,R) ( X , R ) obtained by gluing two CW pairs along a common subpair:
( X , R ) = ( X 1 , R 1 ) ∪ ( Y , S ) ( X 2 , R 2 ) , Y = X 1 ∩ X 2 , S = R ∩ Y , R i = R ∩ X i . (38) (X,R)=(X_1,R_1)\mathop{\cup}_{(Y,S)}(X_2,R_2),\qquad Y=X_1\cap X_2,\qquad S=R\cap Y,\qquad R_i=R\cap X_i. \tag*{(38)} ( X , R ) = ( X 1 , R 1 ) ∪ ( Y , S ) ( X 2 , R 2 ) , Y = X 1 ∩ X 2 , S = R ∩ Y , R i = R ∩ X i . (38)
Choose a flat rank-one local system L L L on X X X ; all restrictions are understood. Write
C i = C ∙ ( X i , R i ; L ) , D = C ∙ ( Y , S ; L ) , r = r 1 − r 2 : C 1 ⊕ C 2 ⟶ D . C_i=C^\bullet(X_i,R_i;L),\qquad D=C^\bullet(Y,S;L),\qquad r=r_1-r_2:C_1\oplus C_2\longrightarrow D. C i = C ∙ ( X i , R i ; L ) , D = C ∙ ( Y , S ; L ) , r = r 1 − r 2 : C 1 ⊕ C 2 ⟶ D .
Let P = holim ( C 1 → r 1 D ← r 2 C 2 ) P=\operatorname{holim}(C_1\xrightarrow{r_1}D\xleftarrow{r_2}C_2) P = holim ( C 1 r 1 D r 2 C 2 ) be represented by the shifted mapping cone:
P n = C 1 n ⊕ C 2 n ⊕ D n − 1 . (39) P^n=C_1^n\oplus C_2^n\oplus D^{n-1}. \tag*{(39)} P n = C 1 n ⊕ C 2 n ⊕ D n − 1 . (39)
d P ( x 1 , x 2 , e ) = ( d 1 x 1 , d 2 x 2 , r 1 x 1 − r 2 x 2 − d D e ) . (40) d_P(x_1,x_2,e)=(d_1x_1,d_2x_2,r_1x_1-r_2x_2-d_De). \tag*{(40)} d P ( x 1 , x 2 , e ) = ( d 1 x 1 , d 2 x 2 , r 1 x 1 − r 2 x 2 − d D e ) . (40)
The sign − d D -d_D − d D in the third component makes d P 2 = 0 d_P^2=0 d P 2 = 0 . The strict cellular matching complex C ∙ ( X , R ; L ) C^\bullet(X,R;L) C ∙ ( X , R ; L ) maps to P P P by restriction x ↦ ( x ∣ X 1 , x ∣ X 2 , 0 ) x\mapsto(x|_{X_1},x|_{X_2},0) x ↦ ( x ∣ X 1 , x ∣ X 2 , 0 ) .
Theorem 11.3 (Constructive acyclic-interface gluing). Suppose that the cellular interface complex D D D admits a contraction h : D n → D n − 1 h:D^n\to D^{n-1} h : D n → D n − 1 with d h + h d = 1 dh+hd=1 d h + h d = 1 and h 2 = 0 h^2=0 h 2 = 0 . Then:
(i) The cellular restriction C ∙ ( X , R ; L ) → P C^\bullet(X,R;L)\to P C ∙ ( X , R ; L ) → P is a quasi-isomorphism.
(ii) The maps
q : P ⟶ C 1 ⊕ C 2 , q ( x 1 , x 2 , e ) = ( x 1 , x 2 ) , s : C 1 ⊕ C 2 ⟶ P , s ( x 1 , x 2 ) = ( x 1 , x 2 , h ( r 1 x 1 − r 2 x 2 ) ) , H : P n ⟶ P n − 1 , H ( x 1 , x 2 , e ) = ( 0 , 0 , − h e ) \begin{aligned}
q &: P \longrightarrow C_1 \oplus C_2, & q(x_1,x_2,e)&=(x_1,x_2),\\
s &: C_1 \oplus C_2 \longrightarrow P, & s(x_1,x_2)&=(x_1,x_2,h(r_1x_1-r_2x_2)),\\
H &: P^n \longrightarrow P^{n-1}, & H(x_1,x_2,e)&=(0,0,-he)
\end{aligned} q s H : P ⟶ C 1 ⊕ C 2 , : C 1 ⊕ C 2 ⟶ P , : P n ⟶ P n − 1 , q ( x 1 , x 2 , e ) s ( x 1 , x 2 ) H ( x 1 , x 2 , e ) = ( x 1 , x 2 ) , = ( x 1 , x 2 , h ( r 1 x 1 − r 2 x 2 )) , = ( 0 , 0 , − h e ) are chain maps in the case of q , s q,s q , s and a homotopy in the case of H H H . They give a strong deformation retract:
q s = 1 , d P H + H d P = 1 − s q , H s = q H = H 2 = 0. (41) qs=1,\qquad d_P H+Hd_P=1-sq,\qquad Hs=qH=H^2=0. \tag*{(41)} q s = 1 , d P H + H d P = 1 − s q , H s = q H = H 2 = 0. (41) In particular, the glued cochain complex is related by an explicit zigzag of quasi-isomorphisms to C 1 ⊕ C 2 C_1\oplus C_2 C 1 ⊕ C 2 .
(iii) If all three complexes C 1 , C 2 , D C_1,C_2,D C 1 , C 2 , D are acyclic and their cellular bases and lifts are compatible, their Milnor torsions obey
τ ( X , R ; L ) ≔ τ ( X 1 , R 1 ; L ) τ ( X 2 , R 2 ; L ) τ ( Y , S ; L ) . (42) \tau(X,R;L)\coloneqq\frac{\tau(X_1,R_1;L)\tau(X_2,R_2;L)}{\tau(Y,S;L)}. \tag*{(42)} τ ( X , R ; L ) : = τ ( Y , S ; L ) τ ( X 1 , R 1 ; L ) τ ( X 2 , R 2 ; L ) . (42) The equality is in the appropriate determinant line before choosing compatible bases; the displayed scalar formula retains the usual deck-translation and sign ambiguity. This theorem concerns the linearized cellular BV–BFV Q Q Q -complex and its torsion, not a quasi-isomorphism of interacting quantum boundary state complexes.
Proof. The cochain-level cellular Mayer–Vietoris sequence for (38) is degreewise exact:
0 ⟶ C ∙ ( X , R ; L ) ⟶ C 1 ⊕ C 2 → r D ⟶ 0. (43) 0\longrightarrow C^\bullet(X,R;L)\longrightarrow C_1\oplus C_2\xrightarrow{r}D\longrightarrow0. \tag*{(43)} 0 ⟶ C ∙ ( X , R ; L ) ⟶ C 1 ⊕ C 2 r D ⟶ 0. (43) The map r r r is degreewise surjective: any cellular cochain on a cell of Y ∖ S Y\setminus S Y ∖ S extends by zero to a cochain on X 1 X_1 X 1 . The associated homotopy-pullback comparison of the strict kernel with Cone ( r ) [ − 1 ] \operatorname{Cone}(r)[-1] Cone ( r ) [ − 1 ] is a quasi-isomorphism. More explicitly, the quotient of the cone by the strict kernel is an acyclic cone of the surjective map after choosing a linear splitting of r r r in each degree. This proves (i).
For (ii), abbreviate x = ( x 1 , x 2 ) x=(x_1,x_2) x = ( x 1 , x 2 ) and d C = d 1 ⊕ d 2 d_C=d_1\oplus d_2 d C = d 1 ⊕ d 2 . Since r r r is a cochain map,
d P s ( x ) = ( d C x , r x − d D h r x ) = ( d C x , h d D r x ) = s ( d C x ) . d_Ps(x)=(d_Cx,rx-d_Dhrx)=(d_Cx,hd_Drx)=s(d_Cx). d P s ( x ) = ( d C x , r x − d D h r x ) = ( d C x , h d D r x ) = s ( d C x ) . Thus s s s is a chain map, as is the evident projection q q q . Furthermore,
( d P H + H d P ) ( x , e ) = ( 0 , d D h e + h d D e − h r x ) = ( 0 , e − h r x ) = ( 1 − s q ) ( x , e ) . (d_PH+Hd_P)(x,e)=(0,d_Dhe+hd_De-hrx)=(0,e-hrx)=(1-sq)(x,e). ( d P H + H d P ) ( x , e ) = ( 0 , d D h e + h d D e − h r x ) = ( 0 , e − h r x ) = ( 1 − s q ) ( x , e ) . The side conditions follow from h 2 = 0 h^2=0 h 2 = 0 and the formulas. This also exhibits a concrete chain homotopy inverse to the projection P ⟶ C 1 ⊕ C 2 P\longrightarrow C_1\oplus C_2 P ⟶ C 1 ⊕ C 2 .
For (iii), use the chain version of (43) , 0 ⟶ C ∙ ( Y , S ; L ) ⟶ C ∙ ( X 1 , R 1 ; L ) ⊕ C ∙ ( X 2 , R 2 ; L ) ⟶ C ∙ ( X , R ; L ) ⟶ 0 0\longrightarrow C^\bullet(Y,S;L)\longrightarrow C^\bullet(X_1,R_1;L)\oplus C^\bullet(X_2,R_2;L)\longrightarrow C^\bullet(X,R;L)\longrightarrow0 0 ⟶ C ∙ ( Y , S ; L ) ⟶ C ∙ ( X 1 , R 1 ; L ) ⊕ C ∙ ( X 2 , R 2 ; L ) ⟶ C ∙ ( X , R ; L ) ⟶ 0 . The multiplicativity theorem for based acyclic complexes gives τ ( C 1 ) τ ( C 2 ) = τ ( D ) τ ( C X ) \tau(C_1)\tau(C_2)=\tau(D)\tau(C_X) τ ( C 1 ) τ ( C 2 ) = τ ( D ) τ ( C X ) . Compatibility of the cell lifts identifies these with the displayed torsions, proving (42) . □
Corollary 11.4 (Annular and toroidal gluing). For a meridional interface Y = A ≃ S 1 × [ 0 , 1 ] Y=A\simeq S^1\times[0,1] Y = A ≃ S 1 × [ 0 , 1 ] with S = ∅ S=\varnothing S = ∅ , the relative BF cochain complex admits the explicit retract (41) for z ≠ 1 z\ne1 z = 1 , and
τ ( X , R ; L z ) ≐ ( z − 1 ) τ ( X 1 , R 1 ; L z ) τ ( X 2 , R 2 ; L z ) (44) \tau(X,R;L_z)\doteq(z-1)\tau(X_1,R_1;L_z)\tau(X_2,R_2;L_z) \tag*{(44)} τ ( X , R ; L z ) ≐ ( z − 1 ) τ ( X 1 , R 1 ; L z ) τ ( X 2 , R 2 ; L z ) (44) whenever the complexes are acyclic. For a torus interface Y = T 2 Y=T^2 Y = T 2 with meridional holonomy z z z and longitudinal holonomy 1 1 1 , the same chain-level retract exists and
τ ( X , R ; L z ) ≐ τ ( X 1 , R 1 ; L z ) τ ( X 2 , R 2 ; L z ) , (45) \tau(X,R;L_z)\doteq\tau(X_1,R_1;L_z)\tau(X_2,R_2;L_z), \tag*{(45)} τ ( X , R ; L z ) ≐ τ ( X 1 , R 1 ; L z ) τ ( X 2 , R 2 ; L z ) , (45) since τ ( T 2 ; L z ) ≐ 1 \tau(T^2;L_z)\doteq1 τ ( T 2 ; L z ) ≐ 1 . The statements hold for the L z ∨ L_z^\vee L z ∨ part of the linear BV–BFV fields as well.
Proof. Apply Lemma 11.1 to Theorem 11.3. The Milnor torsion of a meridional circle (hence annulus) is ( z − 1 ) − 1 (z-1)^{-1} ( z − 1 ) − 1 , while the torus complex splits as the sum of a two-term complex with boundary u u u and a shifted two-term complex with the same boundary. Their torsions are inverse, giving torsion 1 1 1 up to sign. □
Remark 11.5 (What the deformation retract does not do). The maps s s s , q q q , H H H depend on a cellular decomposition, the local system and the contracting homotopy. In particular, h h h uses ( z − 1 ) − 1 (z-1)^{-1} ( z − 1 ) − 1 . We do not claim these maps implement the quantum BFV integral kernel, respect a chosen boundary polarization symplectically, or are compatible with a bordered Floer A ∞ A_\infty A ∞ module structure. Such claims require a separate quantization and comparison construction. What is established is a chain-level homological gluing theorem for the linearized BFV complexes, not merely an equality of their scalar torsions.
Floer gluing and decategorification
Two different gluing operations
Two different gluing operations. The natural gluing operations of the two theories are different. The linear cellular BFV interface is represented by a homotopy pullback of cochain complexes; in bordered Floer theory the boundary carries a differential graded algebra A ( F ) \mathcal{A}(F) A ( F ) , and a bordered three-manifold has a type D D D structure or an A ∞ A_\infty A ∞ type A A A module. The pairing theorem of Lipshitz–Ozsváth–Thurston identifies the chain complex of a glued closed manifold with a suitable box tensor product [14 ] . The categorification theorem of Hom–Lidman–Watson identifies Donaldson’s Alexander-module TQFT as a decategorification of a bordered Floer construction [16 ] .
For orientation, if A \mathcal{A} A is a differential graded algebra over F 2 \mathbb{F}_2 F 2 , a type D D D structure is a graded vector space N N N with a map δ 1 : N → A ⊗ N \delta^1:N\to\mathcal{A}\otimes N δ 1 : N → A ⊗ N satisfying
( d A ⊗ 1 ) δ 1 + ( μ ⊗ 1 ) ( 1 ⊗ δ 1 ) δ 1 = 0. (d_{\mathcal{A}}\otimes1)\delta^1+(\mu\otimes1)(1\otimes\delta^1)\delta^1=0. ( d A ⊗ 1 ) δ 1 + ( μ ⊗ 1 ) ( 1 ⊗ δ 1 ) δ 1 = 0.
An A ∞ A_\infty A ∞ right module M M M has structure maps m j + 1 : M ⊗ A ⊗ j → M m_{j+1}:M\otimes\mathcal{A}^{\otimes j}\to M m j + 1 : M ⊗ A ⊗ j → M obeying the A ∞ A_\infty A ∞ relations. Writing δ j : N → A ⊗ j ⊗ N \delta^j:N\to\mathcal{A}^{\otimes j}\otimes N δ j : N → A ⊗ j ⊗ N for the iterates of δ 1 \delta^1 δ 1 and δ 0 = 1 \delta^0=1 δ 0 = 1 , the box tensor differential on M ⊗ N M\otimes N M ⊗ N is schematically
∂ ⊠ ( m ⊗ n ) = ∑ j ≥ 0 ( m j + 1 ⊗ 1 ) ( m ⊗ δ j ( n ) ) . \partial_{\boxtimes}(m\otimes n)=\sum_{j\geq0}(m_{j+1}\otimes1)(m\otimes\delta^j(n)). ∂ ⊠ ( m ⊗ n ) = j ≥ 0 ∑ ( m j + 1 ⊗ 1 ) ( m ⊗ δ j ( n )) .
Under the usual boundedness hypothesis the sum is finite, and the defining relations imply ∂ ⊠ 2 = 0 \partial_{\boxtimes}^2=0 ∂ ⊠ 2 = 0 . This operation differs from the homotopy pullback (39) : the latter uses only linear restriction maps of cellular cochains, rather than the higher operations of a bordered algebra.
These theorems do not give a functor from the cellular BFV boundary Q Q Q -complex to the bordered algebra A ( F ) \mathcal{A}(F) A ( F ) , nor a natural quasi-isomorphism between the associated state spaces. For knot connected sums, however, the knot Floer Künneth formula is an especially transparent chain-level pairing in its own right, and it can be compared exactly with the BF torsion of an annular gluing.
For a knot K K K , let F K = gr A C F K ^ ( K ; F 2 ) \mathcal{F}_K=\operatorname{gr}_A\widehat{CFK}(K;\mathbb{F}_2) F K = gr A C F K ( K ; F 2 ) be the associated graded complex of the Alexander-filtered hat knot Floer chain complex. It is a finite bigraded chain complex; its differential preserves Alexander grading and decreases Maslov grading by one. Write
Eu ( F K ; t ) = ∑ a ∈ Z ( ∑ m ( − 1 ) m dim F 2 H m ( F K ) a ) t a . \operatorname{Eu}(\mathcal{F}_K;t)=\sum_{a\in\mathbb{Z}}\left(\sum_m(-1)^m\dim_{\mathbb{F}_2}H_m(\mathcal{F}_K)_a\right)t^a. Eu ( F K ; t ) = a ∈ Z ∑ ( m ∑ ( − 1 ) m dim F 2 H m ( F K ) a ) t a .
The Euler characteristic can equivalently be calculated from the chain groups, as long as the differential preserves Alexander degree.
Theorem 12.1 (Compatibility of torsion and Euler characteristic under gluing). Let K 1 , K 2 K_1,K_2 K 1 , K 2 be oriented knots and let K = K 1 # K 2 K=K_1\#K_2 K = K 1 # K 2 be their connected sum. Work with a nontrivial meridional character z z z for which all relevant twisted complexes are acyclic, with the normalization and lift conventions of § 3 –§ 4 .
(i) On the BF side , the cellular homotopy pullback for the meridional connected-sum annulus admits the explicit strong deformation retract of Theorem 11.3 . Its torsion is the annular gluing product
τ ( X K ; L z ) = ( z − 1 ) τ ( X K 1 ; L z ) τ ( X K 2 ; L z ) . \tau(X_K;L_z)=(z-1)\tau(X_{K_1};L_z)\tau(X_{K_2};L_z). τ ( X K ; L z ) = ( z − 1 ) τ ( X K 1 ; L z ) τ ( X K 2 ; L z ) .
(ii) On the Floer side , the knot Floer connected-sum formula gives an Alexander- and Maslov-compatible quasi-isomorphism (over F 2 \mathbb{F}_2 F 2 )
F K 1 # K 2 ≃ F K 1 ⊗ F 2 F K 2 , (46) \mathcal{F}_{K_1\#K_2} \simeq\mathcal{F}_{K_1} \otimes_{\mathbb{F}_2} \mathcal{F}_{K_2}, \tag*{(46)} F K 1 # K 2 ≃ F K 1 ⊗ F 2 F K 2 , (46)
where the tensor differential is the usual sum and the Alexander and Maslov degrees are added.
(iii) After applying the respective decategorification maps, namely symmetric relative torsion for the BF side and graded Euler characteristic for the Floer side, these chain-level gluing operations give the same identity
B K 1 # K 2 ( t ) = B K 1 ( t ) B K 2 ( t ) = Eu ( F K 1 ; t ) Eu ( F K 2 ; t ) . (47) \mathcal{B}_{K_1\#K_2}(t) = \mathcal{B}_{K_1}(t)\mathcal{B}_{K_2}(t) = \operatorname{Eu}(\mathcal{F}_{K_1};t)\operatorname{Eu}(\mathcal{F}_{K_2};t). \tag*{(47)} B K 1 # K 2 ( t ) = B K 1 ( t ) B K 2 ( t ) = Eu ( F K 1 ; t ) Eu ( F K 2 ; t ) . (47)
The result is a pairing-compatible comparison of the Euler/torsion shadows of two chain-level gluing procedures; it is not a natural transformation between the two procedures.
Proof. Part (i) is Corollary 11.4 applied to the connected-sum decomposition along a meridional annulus. For (ii), use the connected-sum Künneth theorem for the knot Floer complex [18 ] . The tensor differential preserves the sum of Alexander gradings, so its Euler characteristic is multiplicative:
Eu ( F K 1 ⊗ F K 2 ; t ) = Eu ( F K 1 ; t ) Eu ( F K 2 ; t ) . \operatorname{Eu}(\mathcal{F}_{K_1} \otimes\mathcal{F}_{K_2};t) = \operatorname{Eu}(\mathcal{F}_{K_1};t)\operatorname{Eu}(\mathcal{F}_{K_2};t). Eu ( F K 1 ⊗ F K 2 ; t ) = Eu ( F K 1 ; t ) Eu ( F K 2 ; t ) .
For (iii), the relative torsion comparison gives B K i ( t ) = Δ K i ( t ) \mathcal{B}_{K_i}(t) = \Delta_{K_i}(t) B K i ( t ) = Δ K i ( t ) for each knot and the connected sum. The Ozsváth–Szabó graded Euler characteristic formula identifies Eu ( F K i ; t ) = Δ K i ( t ) \operatorname{Eu}(\mathcal{F}_{K_i};t) = \Delta_{K_i}(t) Eu ( F K i ; t ) = Δ K i ( t ) . Together with the multiplicativity proved on both sides, these give (47) .
An obstruction to an ordinary chain quasi-isomorphism
There is a simple obstruction to strengthening Theorem 12.1 by identifying the standard complexes themselves.
Theorem 12.2 (Acyclicity obstruction to a direct BF–Floer quasi-isomorphism). Let k = F 2 ( t ) k = \mathbb{F}_2(t) k = F 2 ( t ) , *and endow a knot exterior and its meridional relative pair with the rank-one local system defined by the abelianization π 1 ( X K ) → Z \pi_1(X_K) \to\mathbb{Z} π 1 ( X K ) → Z followed by 1 ↦ t 1 \mapsto t 1 ↦ t . For every knot K ⊂ S 3 K \subset S^3 K ⊂ S 3 ,
H ∙ ( X K , R − ; k ϕ ) = 0 , H ∙ ( F K ⊗ F 2 k ) = H F K ^ ( K ; F 2 ) ⊗ F 2 k ≠ 0. (48) H^\bullet(X_K,R_-;k_\phi) = 0,\qquad H_\bullet(\mathcal{F}_K \otimes_{\mathbb{F}_2} k) = \widehat{HF\!K}(K;\mathbb{F}_2)\otimes_{\mathbb{F}_2} k \ne0. \tag*{(48)} H ∙ ( X K , R − ; k ϕ ) = 0 , H ∙ ( F K ⊗ F 2 k ) = H F K ( K ; F 2 ) ⊗ F 2 k = 0. (48) Consequently there is no quasi-isomorphism in the ordinary derived category D ( k ) D(k) D ( k ) between the standard twisted relative cellular BF complex and the scalar extension of the associated graded knot Floer complex. The obstruction remains if one allows an overall cohomological shift. Any genuinely categorified BV–BFV/Floer comparison must change its objects or introduce additional non-acyclic boundary/categorical data.
Proof. The Alexander polynomial satisfies Δ K ( 1 ) = 1 \Delta_K(1) = 1 Δ K ( 1 ) = 1 , so its reduction modulo 2 is nonzero in F 2 [ t , t − 1 ] \mathbb{F}_2[t,t^{-1}] F 2 [ t , t − 1 ] . The Fox complex over the fraction field F 2 ( t ) \mathbb{F}_2(t) F 2 ( t ) is acyclic by invertibility of an Alexander–Fox minor and of t − 1 t-1 t − 1 . The meridional annulus is also acyclic because t − 1 t-1 t − 1 is a unit in k k k , hence the long exact sequence of the pair yields H ∙ ( X K , R − ; k ϕ ) = 0 H^\bullet(X_K,R_-;k_\phi) = 0 H ∙ ( X K , R − ; k ϕ ) = 0 . On the other hand, H F K ^ ( K ; F 2 ) \widehat{HF\!K}(K;\mathbb{F}_2) H F K ( K ; F 2 ) is nonzero: its graded Euler characteristic is Δ K ( t ) \Delta_K(t) Δ K ( t ) , which is not the zero polynomial (and has value 1 at t = 1 t=1 t = 1 ). Flat scalar extension from F 2 \mathbb{F}_2 F 2 to k k k preserves nonzero homology. Two complexes with different cohomology cannot be quasi-isomorphic, irrespective of a common shift.
Remark 12.3 (Scope of the chain comparison). The contractions in Theorem 11.3 use the standard mapping-cone and Mayer–Vietoris constructions. The Floer tensor pairing and its Euler-characteristic identity are established results. Theorem 12.2 concerns the ordinary complexes specified there; it does not exclude comparisons using additional boundary data.
A concrete comparison: two trefoils
For the right-handed trefoil, write Eu ( F 31 ; t ) = t − 1 − 1 + t \operatorname{Eu}(\mathcal{F}_{31};t)=t^{-1}-1+t Eu ( F 31 ; t ) = t − 1 − 1 + t . Its connected sum with itself gives
Eu ( F 31 # 31 ; t ) = ( t − 1 − 1 + t ) 2 = t − 2 − 2 t − 1 + 3 − 2 t + t 2 . \operatorname{Eu}(\mathcal{F}_{31\#31};t)=(t^{-1}-1+t)^2=t^{-2}-2t^{-1}+3-2t+t^2. Eu ( F 31#31 ; t ) = ( t − 1 − 1 + t ) 2 = t − 2 − 2 t − 1 + 3 − 2 t + t 2 .
The relative torsion has the same normalized Laurent expression. On the absolute cellular BF side, if τ F 31 s y m ( t ) = ( t − 1 − 1 + t ) / ( t − 1 ) \tau^{\mathrm{sym}}_{\mathcal{F}_{31}}(t)=(t^{-1}-1+t)/(t-1) τ F 31 sym ( t ) = ( t − 1 − 1 + t ) / ( t − 1 ) , then annular gluing produces
( t − 1 ) ( τ F 31 s y m ( t ) ) 2 = ( t − 1 − 1 + t ) 2 t − 1 , (t-1)(\tau^{\mathrm{sym}}_{\mathcal{F}_{31}}(t))^2=\frac{(t^{-1}-1+t)^2}{t-1}, ( t − 1 ) ( τ F 31 sym ( t ) ) 2 = t − 1 ( t − 1 − 1 + t ) 2 ,
exactly the absolute torsion of the connected-sum complement. The annular contracting homotopy h A = ( t − 1 ) − 1 h_A=(t-1)^{-1} h A = ( t − 1 ) − 1 displays at chain level the same correction factor that is often inserted only after forming scalar torsions. The example illustrates a precise common multiplicative shadow while leaving the Maslov-graded Floer differential out of the BF complex.
L-space knots
The acyclicity obstruction of Theorem 12.2 rules out a direct quasi-isomorphism of the standard BF and Floer complexes. It does not prohibit using torsion as input to a reconstruction algorithm for a Floer complex when a classification theorem makes this possible. Positive L-space knots provide a nontrivial class where this stronger kind of relationship can be made completely explicit.
A knot K ⊂ S 3 K\subset S^3 K ⊂ S 3 is called a positive L-space knot if some positive integral Dehn surgery on K K K is an L-space, i.e. a rational homology sphere Y Y Y satisfying rk H F ^ ( Y ) = ∣ H 1 ( Y ; Z ) ∣ \operatorname{rk}\widehat{HF}(Y)=|H_1(Y;\mathbb{Z})| rk H F ( Y ) = ∣ H 1 ( Y ; Z ) ∣ . The Ozsváth–Szabó lens-space-surgery restriction and the resulting staircase normal form for its knot Floer complex imply that a positive L-space knot has a symmetrized Alexander polynomial of the form
Δ K ( t ) = ∑ i = − k k ( − 1 ) k − i t n i , n − k < ⋯ < n k , n − i = − n i , (49) \Delta_K(t)=\sum_{i=-k}^{k}(-1)^{k-i}t^{n_i},\qquad n_{-k}<\cdots<n_k,\qquad n_{-i}=-n_i, \tag*{(49)} Δ K ( t ) = i = − k ∑ k ( − 1 ) k − i t n i , n − k < ⋯ < n k , n − i = − n i , (49)
with integral exponents and n k = g ( K ) n_k=g(K) n k = g ( K ) . The staircase normal form is recalled below; for the classification input see [12 , 13 ] .
Theorem 13.1 (Cyclotomic reconstruction of an L-space knot Floer chain complex ). Fix g ≥ 0 g\ge0 g ≥ 0 and an odd prime p ≥ 2 g + 1 p\ge2g+1 p ≥ 2 g + 1 , and let ζ = exp ( 2 π i / p ) \zeta=\exp(2\pi i/p) ζ = exp ( 2 π i / p ) . Suppose K K K is a positive L-space knot with g ( K ) ≤ g g(K)\le g g ( K ) ≤ g . Knowing the single exact algebraic number
α = B K ( ζ ) = τ s y m ( X K , R − ; L ζ ) = Δ K ( ζ ) , \alpha=\mathcal{B}_K(\zeta)=\tau^{\mathrm{sym}}(X_K,R_-;L_\zeta)=\Delta_K(\zeta), α = B K ( ζ ) = τ sym ( X K , R − ; L ζ ) = Δ K ( ζ ) , with the explicitly stipulated symmetric normalization , determines the filtered chain homotopy type of C F K − ( K ; F 2 ) CFK^-(K;\mathbb{F}_2) C F K − ( K ; F 2 ) over F 2 [ U ] \mathbb{F}_2[U] F 2 [ U ] . More precisely, the following algorithm produces a filtered free chain complex representing it:
For − g ≤ a ≤ g -g\le a\le g − g ≤ a ≤ g , compute the integer coefficients
c a = 1 + Tr Q ( ζ ) / Q ( ζ − a α ) p . (50) c_a=\frac{1+\operatorname{Tr}_{\mathbb{Q}(\zeta)/\mathbb{Q}}(\zeta^{-a}\alpha)}{p}. \tag*{(50)} c a = p 1 + Tr Q ( ζ ) / Q ( ζ − a α ) . (50) List the exponents for which c a ≠ 0 c_a\ne0 c a = 0 in increasing order, { n − k , … , n k } \{n_{-k},\ldots,n_k\} { n − k , … , n k } . Under the positive L-space-knot hypothesis, they satisfy (49) .
Let S ( α ) S(\alpha) S ( α ) be free over F 2 [ U ] \mathbb{F}_2[U] F 2 [ U ] on generators x − k , … , x k x_{-k},\ldots,x_k x − k , … , x k with Alexander filtration A ( x i ) = n i A(x_i)=n_i A ( x i ) = n i and A ( U x i ) = n i − 1 A(Ux_i)=n_i-1 A ( U x i ) = n i − 1 . Define its differential by
∂ x i = { 0 , i ≡ k ( m o d 2 ) , x i − 1 + U n i + 1 − n i x i + 1 , i ≡ k + 1 ( m o d 2 ) . (51) \partial x_i=
\begin{cases}
0, & i\equiv k\pmod2,\\
x_{i-1}+U^{n_{i+1}-n_i}x_{i+1}, & i\equiv k+1\pmod2.
\end{cases} \tag*{(51)} ∂ x i = { 0 , x i − 1 + U n i + 1 − n i x i + 1 , i ≡ k ( mod 2 ) , i ≡ k + 1 ( mod 2 ) . (51) The second line only occurs at interior indices. The usual absolute Maslov normalization for S 3 S^3 S 3 determines the remaining grading shift, with M ( U x ) = M ( x ) − 2 M(Ux)=M(x)-2 M ( U x ) = M ( x ) − 2 .
Then S ( α ) S(\alpha) S ( α ) is filtered chain homotopy equivalent to C F K − ( K ; F 2 ) CFK^{-}(K;\mathbb{F}_2) C F K − ( K ; F 2 ) . Thus, for this class of knots, one exact cyclotomic value of the normalized meridional BV–BFV torsion character reconstructs more than the Euler characteristic: it recovers the filtered Floer chain homotopy type and hence its standard filtered-chain invariants.
Proof. By Theorem 1.2 and Proposition 6.2, the exact cyclotomic value α = Δ K ( ζ ) \alpha=\Delta_K(\zeta) α = Δ K ( ζ ) together with the degree bound and Δ K ( 1 ) = 1 \Delta_K(1)=1 Δ K ( 1 ) = 1 uniquely determines all coefficients via (50) . The L-space-knot condition now supplies the alternating nonzero coefficients and ordered exponents in (49) by the Ozsváth–Szabó classification theorem [12 ] .
Equation (51) defines a differential. Indeed, if i ≡ k ( m o d 2 ) i\equiv k\pmod{2} i ≡ k ( mod 2 ) then ∂ x i = 0 \partial x_i=0 ∂ x i = 0 . If i ≡ k + 1 ( m o d 2 ) i\equiv k+1\pmod{2} i ≡ k + 1 ( mod 2 ) , its two neighbors have parity k k k , so ∂ 2 x i = 0 \partial^2x_i=0 ∂ 2 x i = 0 . Moreover, the two terms in the nonzero differential have Alexander filtrations
A ( x i − 1 ) = n i − 1 < n i , A ( U n i + 1 − n i x i + 1 ) = n i A(x_{i-1})=n_{i-1}<n_i,\qquad A(U^{n_{i+1}-n_i}x_{i+1})=n_i A ( x i − 1 ) = n i − 1 < n i , A ( U n i + 1 − n i x i + 1 ) = n i
so ∂ \partial ∂ respects the increasing Alexander filtration. The polynomial (49) specifies all the gaps n i + 1 − n i n_{i+1}-n_i n i + 1 − n i and hence all the nonzero differential entries. The staircase normal-form theorem for positive L-space knots (e.g. the basis-and-differential formulation in [13 Proposition 1.1] ) identifies this complex, with its U U U -action and Alexander filtration, with C F K − ( K ; F 2 ) CFK^{-}(K;\mathbb{F}_2) C F K − ( K ; F 2 ) up to filtered chain homotopy equivalence. The usual Maslov grading conventions fix the overall shift. □
Example 13.2 (The trefoil from one cubic holonomy value). For the right-handed trefoil, g = 1 g=1 g = 1 , and the minimal permitted prime is p = 3 p=3 p = 3 . Since Δ 31 ( t ) = t − 1 − 1 + t \Delta_{31}(t)=t^{-1}-1+t Δ 31 ( t ) = t − 1 − 1 + t , a primitive cube root ζ 3 \zeta_3 ζ 3 gives α = Δ 31 ( ζ 3 ) = − 2 \alpha=\Delta_{31}(\zeta_3)=-2 α = Δ 31 ( ζ 3 ) = − 2 . Formula (50) returns c − 1 = 1 c_{-1}=1 c − 1 = 1 , c 0 = − 1 c_0=-1 c 0 = − 1 , c 1 = 1 c_1=1 c 1 = 1 . Therefore k = 1 k=1 k = 1 and ( n − 1 , n 0 , n 1 ) = ( − 1 , 0 , 1 ) (n_{-1},n_0,n_1)=(-1,0,1) ( n − 1 , n 0 , n 1 ) = ( − 1 , 0 , 1 ) . The reconstructed F 2 [ U ] \mathbb{F}_2[U] F 2 [ U ] -complex has generators x − 1 , x 0 , x 1 x_{-1},x_0,x_1 x − 1 , x 0 , x 1 and
∂ x − 1 = ∂ x 1 = 0 , ∂ x 0 = x − 1 + U x 1 . \partial x_{-1}=\partial x_1=0,\qquad\partial x_0=x_{-1}+Ux_1. ∂ x − 1 = ∂ x 1 = 0 , ∂ x 0 = x − 1 + U x 1 .
This is the positive-trefoil staircase model, not just its Euler-characteristic polynomial.
Example 13.3 (The cinquefoil). For the positive ( 2 , 5 ) (2,5) ( 2 , 5 ) torus knot, g = 2 g=2 g = 2 ; a primitive fifth root ζ 5 \zeta_5 ζ 5 is sufficient. The trace recovers the five nonzero exponents ( − 2 , − 1 , 0 , 1 , 2 ) (-2,-1,0,1,2) ( − 2 , − 1 , 0 , 1 , 2 ) , and the resulting model has
∂ x − 1 = x − 2 + U x 0 , ∂ x 1 = x 0 + U x 2 , \partial x_{-1}=x_{-2}+Ux_0,\qquad\partial x_1=x_0+Ux_2, ∂ x − 1 = x − 2 + U x 0 , ∂ x 1 = x 0 + U x 2 ,
with all other generator differentials zero. In particular, the two step lengths of the staircase are recovered from the one specialized BV–BFV torsion character.
Remark 13.4 (The L-space hypothesis). The Alexander polynomial does not determine whether a knot is an L-space knot. Theorem 13.1 assumes this property and uses the established staircase classification; it does not construct a map from the twisted BF complex. For general knots, the Kinoshita–Terasaka example in Proposition 10.1 obstructs reconstruction from the Alexander polynomial.
Scalar extension and pairing obstructions
The preceding theorems compare torsion and Euler characteristic, but a functor between boundary theories would have to intertwine their gluing operations before taking decategorification. We now make the simplest possible such attempt fully precise, prove that it cannot work, and isolate which additional algebraic operations are indispensable. The result is an algebraic theorem about the standard torus algebra; it does not assume a hypothetical equivalence of BV–BFV and Floer field theories.
Derived complexes and torsion
For a knot exterior X K X_K X K and nonresonant rank-one character z z z , the finite-dimensional linear BF model produces a twisted relative complex C ∙ ( X K , R ; κ z ) \mathcal{C}^{\bullet}(X_K,\mathbb{R};\kappa_z) C ∙ ( X K , R ; κ z ) together with cellular bases, and hence a Milnor-torsion element. There are at least three distinguishable constructions:
(b) the based complex with a chosen determinant-line trivialization and its torsion; (c) a quantum BV–BFV state with boundary observable algebra, polarization, residual fields and higher composition operations.
Our earlier torsion comparison concerns (b). It gives no identification of the full structure in (c) with a bordered Floer module. Even (a) and (b) cannot be conflated:
Lemma 14.1 (Acyclicity loses torsion). Let k k k be a field and u , v ∈ k × u,v\in k^{\times} u , v ∈ k × . The two based cochain complexes
E u : 0 ⟶ k → u k ⟶ 0 , E v : 0 ⟶ k → v k ⟶ 0 E_u: 0 \longrightarrow k \xrightarrow{u} k \longrightarrow0,\qquad E_v: 0 \longrightarrow k \xrightarrow{v} k \longrightarrow0 E u : 0 ⟶ k u k ⟶ 0 , E v : 0 ⟶ k v k ⟶ 0 are isomorphic to the zero object in D ( k ) D(k) D ( k ) , but their based torsions can have different values. With the chain-torsion convention of Definition 2.1 , their torsions differ by u / v u/v u / v or its inverse according to the cochain-to-chain convention. In particular there is no function on the isomorphism classes of D ( k ) D(k) D ( k ) that recovers based Reidemeister torsion for all such complexes.
Proof. Both differentials are invertible, so both complexes are contractible. Their images in D ( k ) D(k) D ( k ) are zero. In the fixed standard bases the only nontrivial determinant entering the acyclic torsion is the determinant of the one differential; changing u u u to v v v changes this factor by u / v u/v u / v (possibly inverted). Taking u ≠ v u\ne v u = v establishes the claim. □
This small example rules out an exact functor defined only on the ordinary twisted derived BF complex as a mechanism for remembering Alexander torsion. Simple-homotopy data, cellular lifts or a suitable determinant-line enhancement are not optional decorations. We next show that retaining only a scalar determinant line is still insufficient for bordered Floer gluing.
The torus algebra and its chord ideal
We work initially over F 2 \mathbb{F}_2 F 2 and then extend scalars to an arbitrary field k k k of characteristic two, for instance k = F 2 ( t ) k=\mathbb{F}_2(t) k = F 2 ( t ) . We use the ordinary hat bordered torus algebra of Lipshitz–Ozsváth–Thurston [14 Section 10.1] ; the generators and relations are also recorded in [15 equation (1.18)] . Its underlying differential is zero and its eight-dimensional vector-space basis is
A = F 2 ⟨ ι 0 , ι 1 , ρ 1 , ρ 2 , ρ 3 , ρ 12 , ρ 23 , ρ 123 ⟩ . (52) \mathcal{A}=\mathbb{F}_2\langle\iota_0,\iota_1,\rho_1,\rho_2,\rho_3,\rho_{12},\rho_{23},\rho_{123}\rangle. \tag*{(52)} A = F 2 ⟨ ι 0 , ι 1 , ρ 1 , ρ 2 , ρ 3 , ρ 12 , ρ 23 , ρ 123 ⟩ . (52)
The idempotents are orthogonal, ι i ι j = δ i j ι i \iota_i\iota_j=\delta_{ij}\iota_i ι i ι j = δ ij ι i and ι 0 + ι 1 = 1 \iota_0+\iota_1=1 ι 0 + ι 1 = 1 . The Reeb-chord elements have endpoints
ι 0 A ι 0 ι 0 , ρ 12 ι 0 A ι 1 ρ 1 , ρ 3 , ρ 123 ι 1 A ι 0 ρ 2 ι 1 A ι 1 ι 1 , ρ 23 \begin{array}{c|c}
\iota_0\mathcal{A}\iota_0 & \iota_0,\ \rho_{12}\\
\iota_0\mathcal{A}\iota_1 & \rho_1,\ \rho_3,\ \rho_{123}\\
\iota_1\mathcal{A}\iota_0 & \rho_2\\
\iota_1\mathcal{A}\iota_1 & \iota_1,\ \rho_{23}
\end{array} ι 0 A ι 0 ι 0 A ι 1 ι 1 A ι 0 ι 1 A ι 1 ι 0 , ρ 12 ρ 1 , ρ 3 , ρ 123 ρ 2 ι 1 , ρ 23
The nonzero products of positive-length chords are determined by
ρ 1 ρ 2 = ρ 12 , ρ 2 ρ 3 = ρ 23 , ρ 1 ρ 23 = ρ 12 ρ 3 = ρ 123 , (53) \rho_1\rho_2=\rho_{12},\qquad\rho_2\rho_3=\rho_{23},\qquad\rho_1\rho_{23}=\rho_{12}\rho_3=\rho_{123}, \tag*{(53)} ρ 1 ρ 2 = ρ 12 , ρ 2 ρ 3 = ρ 23 , ρ 1 ρ 23 = ρ 12 ρ 3 = ρ 123 , (53)
while ρ 2 ρ 1 = ρ 3 ρ 2 = 0 \rho_2\rho_1=\rho_3\rho_2=0 ρ 2 ρ 1 = ρ 3 ρ 2 = 0 ; remaining incompatible products vanish. We suppress the refined group grading in the calculations below, regarding A k : = A ⊗ F 2 k \mathcal{A}_k:=\mathcal{A}\otimes_{\mathbb{F}_2}k A k := A ⊗ F 2 k as a differential-graded algebra concentrated in cohomological degree zero. No claim about an equivalence preserving the original bordered grading is hidden in this simplification.
Let J ⊂ A k J\subset\mathcal{A}_k J ⊂ A k be the two-sided ideal spanned by the six Reeb-chord elements. Equation (53) gives
J 4 = 0 , A k / J ≅ k ι 0 ⊕ k ι 1 ≅ k × k . (54) J^4=0,\qquad\mathcal{A}_k/J\cong k\iota_0\oplus k\iota_1\cong k\times k. \tag*{(54)} J 4 = 0 , A k / J ≅ k ι 0 ⊕ k ι 1 ≅ k × k . (54)
In particular, J J J is the Jacobson radical of this finite-dimensional algebra. The two quotient projections define strict unital augmentations
ϵ i : A k ⟶ k , ϵ i ( ι j ) = δ i j , ϵ i ( J ) = 0 ( i = 0 , 1 ) . (55) \epsilon_i:\mathcal{A}_k\longrightarrow k,\qquad\epsilon_i(\iota_j)=\delta_{ij},\qquad\epsilon_i(J)=0\qquad(i=0,1). \tag*{(55)} ϵ i : A k ⟶ k , ϵ i ( ι j ) = δ ij , ϵ i ( J ) = 0 ( i = 0 , 1 ) . (55)
Each augmentation makes k k k into a left or right A k \mathcal{A}_k A k -module, denoted k ϵ i k_{\epsilon_i} k ϵ i .
Proposition 14.2 (Universality of rank-one strict scalarization). For every field k k k of characteristic two, the only unital k k k -algebra homomorphisms A k → k \mathcal{A}_k \to k A k → k are ϵ 0 \epsilon_0 ϵ 0 and ϵ 1 \epsilon_1 ϵ 1 . When both algebras are regarded as dg algebras concentrated in cohomological degree zero, the same statement holds for unital A ∞ A_\infty A ∞ morphisms: all higher Taylor maps f n f_n f n for n ≥ 2 n \ge2 n ≥ 2 vanish for degree reasons, and f 1 f_1 f 1 is one of the two augmentations.
Proof. Let f : A k → k f:\mathcal{A}_k \to k f : A k → k be a unital algebra homomorphism. For every x ∈ J x \in J x ∈ J , equation (54) gives x 4 = 0 x^4=0 x 4 = 0 , so f ( x ) 4 = 0 f(x)^4=0 f ( x ) 4 = 0 . Since k k k is a field, f ( x ) = 0 f(x)=0 f ( x ) = 0 . Thus f f f factors through A k / J ≅ k × k \mathcal{A}_k/J \cong k \times k A k / J ≅ k × k . The orthogonal idempotents f ( ι 0 ) f(\iota_0) f ( ι 0 ) and f ( ι 1 ) f(\iota_1) f ( ι 1 ) are idempotents in a field whose sum is one. Exactly one is one and the other zero, proving the first assertion. For an uncurved A ∞ A_\infty A ∞ morphism of dg algebras concentrated in degree zero, f n f_n f n has cohomological degree 1 − n 1-n 1 − n , so f n = 0 f_n=0 f n = 0 for n ≥ 2 n \ge2 n ≥ 2 . The n = 2 n=2 n = 2 A ∞ A_\infty A ∞ identity then says that f 1 f_1 f 1 respects products, and strict unitality says f 1 ( 1 ) = 1 f_1(1)=1 f 1 ( 1 ) = 1 . □
Derived scalar extension
The augmentations are not merely formal maps. Each gives a well-defined exact derived functor on the category of right modules,
R i : Perf ( A k ) ⟶ D ( k ) , R i ( M ) = M ⊗ A k L k ϵ i . (56) R_i:\operatorname{Perf}(\mathcal{A}_k) \longrightarrow D(k), \qquad R_i(M)=M\otimes_{\mathcal{A}_k}^{\mathsf{L}} k_{\epsilon_i}. \tag*{(56)} R i : Perf ( A k ) ⟶ D ( k ) , R i ( M ) = M ⊗ A k L k ϵ i . (56)
Here Perf ( A k ) \operatorname{Perf}(\mathcal{A}_k) Perf ( A k ) is the category of perfect dg modules. An A ∞ A_\infty A ∞ right module may be replaced by a bar-type dg model before applying the derived tensor product. Consequently (56) is homotopy invariant and respects all A ∞ A_\infty A ∞ identities internal to this extension-of-scalars construction . Its construction uses the torus algebra and does not assign a BV–BFV state to a knot exterior.
For a left module we similarly write L i ( N ) = k ϵ i ⊗ A k L N L_i(N)=k_{\epsilon_i}\otimes_{\mathcal{A}_k}^{\mathsf{L}}N L i ( N ) = k ϵ i ⊗ A k L N . A gluing-preserving rank-one reduction would require, for all appropriate module pairs, a natural quasi-isomorphism between the original derived pairing and the pairing of the reductions:
M ⊗ A k L N ≃ R i ( M ) ⊗ k L L i ( N ) . (57) M\otimes_{\mathcal{A}_k}^{\mathsf{L}}N \simeq R_i(M)\otimes_k^{\mathsf{L}}L_i(N). \tag*{(57)} M ⊗ A k L N ≃ R i ( M ) ⊗ k L L i ( N ) . (57)
The following theorem shows that equation (57) cannot hold on the entire bordered module category, even after forgetting all gradings.
Theorem 14.4 (Explicit Reeb-chord obstruction to scalar gluing). Fix i ∈ { 0 , 1 } i \in\{0,1\} i ∈ { 0 , 1 } and let j = 1 − i j=1-i j = 1 − i . In the degree-zero model of the bordered torus algebra over k k k , take the projective right module P i = ι i A k P_i=\iota_i\mathcal{A}_k P i = ι i A k and the projective left module Q j = A k ι j Q_j=\mathcal{A}_k\iota_j Q j = A k ι j . Then
H 0 ( P i ⊗ A k L Q j ) ≅ ι i A k ι j ≠ 0 , R i ( P i ) ⊗ k L L i ( Q j ) ≃ 0. (58) H^0(P_i\otimes_{\mathcal{A}_k}^{\mathsf{L}}Q_j)\cong\iota_i\mathcal{A}_k\iota_j\ne0,\qquad R_i(P_i)\otimes_k^{\mathsf{L}}L_i(Q_j)\simeq0. \tag*{(58)} H 0 ( P i ⊗ A k L Q j ) ≅ ι i A k ι j = 0 , R i ( P i ) ⊗ k L L i ( Q j ) ≃ 0. (58) For i = 0 i=0 i = 0 the first space has dimension three; for i = 1 i=1 i = 1 it has dimension one. Hence no natural quasi-isomorphisms (57) exist for all perfect modules. In particular, scalar character base change does not preserve the bordered gluing operation.
Proof. Since P i P_i P i and Q j Q_j Q j are summands of free right and left A k \mathcal{A}_k A k -modules, respectively, they are projective. The derived tensor product is therefore represented by the ordinary tensor product, and multiplication induces the isomorphism
( ι i A k ) ⊗ A k ( A k ι j ) → ≅ ι i A k ι j . (\iota_i\mathcal{A}_k)\otimes_{\mathcal{A}_k}(\mathcal{A}_k\iota_j)\xrightarrow{\cong}\iota_i\mathcal{A}_k\iota_j. ( ι i A k ) ⊗ A k ( A k ι j ) ≅ ι i A k ι j .
By the table in § 14.2 , ι 0 A k ι 1 \iota_0\mathcal{A}_k\iota_1 ι 0 A k ι 1 has basis ρ 1 , ρ 3 , ρ 123 \rho_1,\rho_3,\rho_{123} ρ 1 , ρ 3 , ρ 123 , while ι 1 A k ι 0 \iota_1\mathcal{A}_k\iota_0 ι 1 A k ι 0 has basis ρ 2 \rho_2 ρ 2 . On the other hand, projectivity and the augmentation identities give
R i ( P i ) ≅ ι i k ϵ i ≅ k , L i ( Q j ) ≅ k ϵ i ι j = 0 ( j ≠ i ) . (59) R_i(P_i)\cong\iota_i k_{\epsilon_i}\cong k,\qquad L_i(Q_j)\cong k_{\epsilon_i}\iota_j=0\qquad(j\ne i). \tag*{(59)} R i ( P i ) ≅ ι i k ϵ i ≅ k , L i ( Q j ) ≅ k ϵ i ι j = 0 ( j = i ) . (59)
The right-hand side of (58) is therefore zero. A nonzero complex concentrated in degree zero cannot be quasi-isomorphic to the zero complex. □
Corollary 14.5. (Two idempotents without Reeb chords do not suffice). Replacing A k A_k A k by its semisimple quotient A k / J ≅ k × k A_k/J \cong k \times k A k / J ≅ k × k also destroys the off-diagonal pairings of Theorem 14.4 . Thus even a boundary model carrying both scalar idempotent sectors cannot reproduce all bordered tensor products if it discards the Reeb-chord ideal.
Proof. The quotient maps each ι i A k ι j \iota_i A_k \iota_j ι i A k ι j for i ≠ j i \ne j i = j to ι i ( A k / J ) ι j = 0 \iota_i(A_k/J)\iota_j = 0 ι i ( A k / J ) ι j = 0 , although the original space is nonzero. □
Remark 14.6. (Interpretation and precise domain of the theorem). The modules P i P_i P i and Q j Q_j Q j are elementary algebraic projective modules; we do not assert that each is the bordered Floer invariant of a particular knot exterior. The theorem rules out a rank-one, gluing-preserving scalar reduction on the whole perfect module category ; it does not rule out specialized comparisons for particular geometric objects. It also does not identify the derived functor R i R_i R i with the BFV pushforward or with a physical choice of polarization. Establishing such an identification would require a separate construction of a boundary observable algebra and a functor from it to A k A_k A k .
The projective pairing matrix
The preceding counterexample uses a specified augmentation. There is a stronger obstruction that does not assume that the scalar reductions come from algebra maps, or even that the reductions are functorial. Write P a = ι a A k P_a = \iota_a A_k P a = ι a A k and Q b = A k ι b Q_b = A_k\iota_b Q b = A k ι b for a , b ∈ { 0 , 1 } a,b \in\{0,1\} a , b ∈ { 0 , 1 } . Their derived pairing is concentrated in degree zero, with dimension matrix
C a b = dim k ( ι a A k ι b ) , C = ( 2 3 1 2 ) , det C = 1. (60) C_{ab} = \dim_k(\iota_a A_k \iota_b), \qquad C = \begin{pmatrix} 2 & 3 \\ 1 & 2 \end{pmatrix}, \qquad\det C = 1. \tag*{(60)} C ab = dim k ( ι a A k ι b ) , C = ( 2 1 3 2 ) , det C = 1. (60)
The determinant in this equation is a determinant of integer dimension counts ; it is not a computation in the characteristic-two coefficient field.
Theorem 14.7. (Obstruction to independent scalar tensor factorization). There do not exist four perfect complexes V 0 , V 1 , W 0 , W 1 V_0,V_1,W_0,W_1 V 0 , V 1 , W 0 , W 1 over k k k such that, for every a , b ∈ { 0 , 1 } a,b \in\{0,1\} a , b ∈ { 0 , 1 } ,
P a ⊗ A k L Q b ≃ V a ⊗ k L W b . (61) P_a \otimes^{\mathrm{L}}_{A_k} Q_b \simeq V_a \otimes^{\mathrm{L}}_k W_b. \tag*{(61)} P a ⊗ A k L Q b ≃ V a ⊗ k L W b . (61) Consequently no pair of assignments from all perfect right and left A k A_k A k -modules to perfect scalar complexes can preserve every derived pairing by ordinary tensor product over k k k . The obstruction already occurs on the four projective pairs and requires neither naturality nor an augmentation hypothesis.
Proof. For a perfect scalar complex V V V , put b ( V ) = ∑ n dim k H n ( V ) b(V) = \sum_n \dim_k H^n(V) b ( V ) = ∑ n dim k H n ( V ) , its finite total homology dimension. The Künneth formula over a field gives b ( V ⊗ k L W ) = b ( V ) b ( W ) b(V \otimes^{\mathrm{L}}_k W) = b(V)b(W) b ( V ⊗ k L W ) = b ( V ) b ( W ) . If (61) held, then
C a b = b ( V a ) b ( W b ) . C_{ab} = b(V_a)b(W_b). C ab = b ( V a ) b ( W b ) .
The resulting 2 × 2 2 \times2 2 × 2 matrix is an outer product, so it has determinant zero over the integers. This contradicts (14.8). □
Remark 14.8. (Scope of the rank obstruction). This theorem concerns independent scalar complexes with the ordinary tensor product as pairing. It does not exclude a target with an additional boundary kernel, a non-scalar observable algebra, or a pairing other than ordinary tensor product. It also uses the same degree-zero algebra model as Theorem 14.4 ; no extension to every grading or every geometric subcategory is asserted. The rank computation is stronger than the augmentation example, but it is not a general nonexistence theorem for BV–BFV/Floer functors.
Semisimple reduction and its pairing defect
Let S = A k / J = k ι 0 ⊕ k ι 1 S = A_k/J = k\iota_0 \oplus k\iota_1 S = A k / J = k ι 0 ⊕ k ι 1 , and denote the quotient map by π : A k → S \pi: A_k \to S π : A k → S . For perfect right and left modules define
R S ( M ) = M ⊗ A k L S , L S ( N ) = S ⊗ A k L N . R_S(M) = M \otimes^{\mathsf L}_{A_k} S, \qquad L_S(N) = S \otimes^{\mathsf L}_{A_k} N. R S ( M ) = M ⊗ A k L S , L S ( N ) = S ⊗ A k L N .
These are perfect S S S -complexes. Since S S S is a product of two fields, each consists of the two corresponding scalar complexes. Keeping both sectors has a useful positive property which must be distinguished from preservation of the pairing.
Proposition 14.9 (The two-sector reduction is conservative). The functors R S R_S R S and L S L_S L S on perfect modules reflect quasi-isomorphisms. Equivalently, R S ( M ) ≃ 0 R_S(M) \simeq0 R S ( M ) ≃ 0 implies M ≃ 0 M \simeq0 M ≃ 0 , and the analogous assertion holds on the left. Nevertheless, R S R_S R S is not faithful on morphisms, and neither reduction preserves all derived pairings.
Proof. Each bimodule J r / J r + 1 J^r/J^{r+1} J r / J r + 1 , for 1 ≤ r ≤ 3 1 \le r \le3 1 ≤ r ≤ 3 , is annihilated by J J J on both sides, and hence is an S S S -bimodule. Associativity of derived tensor product gives
M ⊗ A k L ( J r / J r + 1 ) ≃ R S ( M ) ⊗ S L ( J r / J r + 1 ) , M \otimes^{\mathsf L}_{A_k} (J^r/J^{r+1}) \simeq R_S(M) \otimes^{\mathsf L}_S (J^r/J^{r+1}), M ⊗ A k L ( J r / J r + 1 ) ≃ R S ( M ) ⊗ S L ( J r / J r + 1 ) , where the right side is viewed as a right A k A_k A k -complex by restriction along π \pi π . If R S ( M ) ≃ 0 R_S(M) \simeq0 R S ( M ) ≃ 0 , all these complexes vanish. Starting with J 4 = 0 J^4 = 0 J 4 = 0 and using the triangles associated to 0 → J r + 1 → J r → J r / J r + 1 → 0 0 \to J^{r+1} \to J^r \to J^r/J^{r+1} \to0 0 → J r + 1 → J r → J r / J r + 1 → 0 , downward induction gives M ⊗ A k L J ≃ 0 M \otimes^{\mathsf L}_{A_k} J \simeq0 M ⊗ A k L J ≃ 0 . The triangle obtained from 0 → J → A k → S → 0 0 \to J \to A_k \to S \to0 0 → J → A k → S → 0 then gives M ≃ 0 M \simeq0 M ≃ 0 . Applying this argument to the cone of a morphism proves reflection of quasi-isomorphisms. The left-module proof is identical.
For nonfaithfulness, consider the nonzero right-module morphism
f : P 1 ⟶ P 0 , f ( ι 1 a ) = ρ 1 a . f : P_1 \longrightarrow P_0, \qquad f(\iota_1a) = \rho_1a. f : P 1 ⟶ P 0 , f ( ι 1 a ) = ρ 1 a . It is nonzero in the derived category because both modules are projective and concentrated in degree zero. Its reduction is zero since π ( ρ 1 ) = 0 \pi(\rho_1) = 0 π ( ρ 1 ) = 0 . Failure to preserve pairings follows already from Corollary 14.5, and is quantified below.
Theorem 14.10 (Canonical defect and its three chord layers). For perfect right and left A k A_k A k -modules M M M , N N N , the quotient π \pi π induces a natural pairing map
θ M , N : M ⊗ A k L N ⟶ R S ( M ) ⊗ S L L S ( N ) . (62) \theta_{M,N} : M \otimes^{\mathsf L}_{A_k} N \longrightarrow R_S(M) \otimes^{\mathsf L}_S L_S(N). \tag*{(62)} θ M , N : M ⊗ A k L N ⟶ R S ( M ) ⊗ S L L S ( N ) . (62) Its homotopy fiber is
D ( M , N ) = M ⊗ A k L J ⊗ A k L N . (63) \mathcal{D}(M,N) = M \otimes^{\mathsf L}_{A_k} J \otimes^{\mathsf L}_{A_k} N. \tag*{(63)} D ( M , N ) = M ⊗ A k L J ⊗ A k L N . (63) In particular there is a natural distinguished triangle
D ( M , N ) ⟶ M ⊗ A k L N → θ M , N R S ( M ) ⊗ S L L S ( N ) ⟶ D ( M , N ) [ 1 ] . (64) \mathcal{D}(M,N) \longrightarrow M \otimes^{\mathsf L}_{A_k} N \xrightarrow{\theta_{M,N}} R_S(M) \otimes^{\mathsf L}_S L_S(N) \longrightarrow\mathcal{D}(M,N)[1]. \tag*{(64)} D ( M , N ) ⟶ M ⊗ A k L N θ M , N R S ( M ) ⊗ S L L S ( N ) ⟶ D ( M , N ) [ 1 ] . (64) The defect admits a finite filtration with three quotients
gr r D ( M , N ) ≃ R S ( M ) ⊗ S L V r ⊗ S L L S ( N ) , V r = J r / J r + 1 , 1 ≤ r ≤ 3. (65) \operatorname{gr}^{r}\mathcal{D}(M,N) \simeq R_S(M) \otimes^{\mathsf L}_S V_r \otimes^{\mathsf L}_S L_S(N), \qquad V_r = J^r/J^{r+1}, \quad1 \le r \le3. \tag*{(65)} gr r D ( M , N ) ≃ R S ( M ) ⊗ S L V r ⊗ S L L S ( N ) , V r = J r / J r + 1 , 1 ≤ r ≤ 3. (65) Representatives for the three S S S -bimodules and their endpoint dimension matrices G a b ( r ) = dim k ( ι a V r ι b ) G^{(r)}_{ab} = \dim_k(\iota_a V_r \iota_b) G ab ( r ) = dim k ( ι a V r ι b ) are
Table 1.
Thus the associated finite spectral sequence computes the homology of the pairing defect from these layers and their extension maps. For projective pairs,
D ( P a , Q b ) ≃ ι a J ι b , ( dim k D ( P a , Q b ) ) a b = C − I = ( 1 3 1 1 ) . (66) \mathcal{D}(P_a,Q_b) \simeq\iota_a J \iota_b, \qquad\bigl(\dim_k \mathcal{D}(P_a,Q_b)\bigr)_{ab} = C-I = \begin{pmatrix}1 & 3 \\ 1 & 1\end{pmatrix}. \tag*{(66)} D ( P a , Q b ) ≃ ι a J ι b , ( dim k D ( P a , Q b ) ) ab = C − I = ( 1 1 3 1 ) . (66)
The quotient map (62) is a quasi-isomorphism for a given pair exactly when its defect is acyclic.
Proof. The short exact sequence 0 → J → A k → π S → 0 0 \to J \to\mathcal{A}_k \xrightarrow{\pi} S \to0 0 → J → A k π S → 0 is a short exact sequence of A k \mathcal{A}_k A k -bimodules. Derived tensoring on the left with M M M and on the right with N N N gives a distinguished triangle. Its middle term is M ⊗ A k L N M \otimes_{\mathcal{A}_k}^{\mathrm{L}} N M ⊗ A k L N . Associativity identifies its last term as
M ⊗ A k L S ⊗ S L N ≃ ( M ⊗ A k L S ) ⊗ S L ( S ⊗ A k L N ) . M \otimes_{\mathcal{A}_k}^{\mathrm{L}} S \otimes_{S}^{\mathrm{L}} N \simeq(M \otimes_{\mathcal{A}_k}^{\mathrm{L}} S) \otimes_{S}^{\mathrm{L}} (S \otimes_{\mathcal{A}_k}^{\mathrm{L}} N). M ⊗ A k L S ⊗ S L N ≃ ( M ⊗ A k L S ) ⊗ S L ( S ⊗ A k L N ) .
This proves (64) and the fiber assertion.
The multiplication table gives
J 2 = ⟨ ρ 12 , ρ 23 , ρ 123 ⟩ k , J 3 = ⟨ ρ 123 ⟩ k , J 4 = 0. J^2 = \langle\rho_{12},\rho_{23},\rho_{123}\rangle_k,\qquad J^3 = \langle\rho_{123}\rangle_k,\qquad J^4=0. J 2 = ⟨ ρ 12 , ρ 23 , ρ 123 ⟩ k , J 3 = ⟨ ρ 123 ⟩ k , J 4 = 0.
Choose bounded complexes of finitely generated projective modules representing M M M and N N N . Tensoring them with the filtration J ⊃ J 2 ⊃ J 3 ⊃ 0 J \supset J^2 \supset J^3 \supset0 J ⊃ J 2 ⊃ J 3 ⊃ 0 gives a filtration of a representative of (63) ; projectivity preserves the short exact sequences in each degree. Its quotients are M ⊗ A k L V r ⊗ A k L N M \otimes_{\mathcal{A}_k}^{\mathrm{L}} V_r \otimes_{\mathcal{A}_k}^{\mathrm{L}} N M ⊗ A k L V r ⊗ A k L N . Since the two actions on V r V_r V r factor through S S S , associativity identifies these with (65) . The endpoint table for the chords proves (14.14). The filtration is finite and the representative complexes are bounded and finite-dimensional over k k k , so its spectral sequence converges to H ∙ ( D ( M , N ) ) H^\bullet(\mathcal{D}(M,N)) H ∙ ( D ( M , N )) .
For M = P a M=P_a M = P a , N = Q b N=Q_b N = Q b , no resolutions are needed. Multiplication identifies P a ⊗ A k J ⊗ A k Q b P_a \otimes_{\mathcal{A}_k} J \otimes_{\mathcal{A}_k} Q_b P a ⊗ A k J ⊗ A k Q b with ι a J ι b \iota_a J \iota_b ι a J ι b , yielding (66) . The last assertion follows directly from the distinguished triangle. □
Corollary 14.11 (Euler pairing and its exact scalar defect). Let m a = χ ( R S ( M ) ι a ) m_a=\chi(R_S(M)\iota_a) m a = χ ( R S ( M ) ι a ) , n b = χ ( ι b L S ( N ) ) n_b=\chi(\iota_bL_S(N)) n b = χ ( ι b L S ( N )) , and regard m , n m,n m , n as two-component columns of integers. Then
χ ( M ⊗ A k L N ) = m T C n , (67) \chi(M \otimes_{\mathcal{A}_k}^{\mathrm{L}} N)=m^{\mathrm{T}}Cn, \tag*{(67)} χ ( M ⊗ A k L N ) = m T C n , (67) χ ( R S ( M ) ⊗ S L L S ( N ) ) = m T n , (68) \chi(R_S(M)\otimes_S^{\mathrm{L}}L_S(N))=m^{\mathrm{T}}n, \tag*{(68)} χ ( R S ( M ) ⊗ S L L S ( N )) = m T n , (68) χ ( D ( M , N ) ) = m T ( C − I ) n . (69) \chi(\mathcal{D}(M,N))=m^{\mathrm{T}}(C-I)n. \tag*{(69)} χ ( D ( M , N )) = m T ( C − I ) n . (69) Here the Euler characteristics refer to the ordinary cohomological grading of the specified degree-zero algebra model.
Proof. Euler characteristic is additive under the finite filtration and under (64) , and multiplicative for tensor products of finite scalar complexes. The decomposition S = k ι 0 ⊕ k ι 1 S=k\iota_0\oplus k\iota_1 S = k ι 0 ⊕ k ι 1 therefore gives
χ ( gr r D ( M , N ) ) = m T G ( r ) n . \chi(\operatorname{gr}^{r}\mathcal{D}(M,N))=m^{\mathrm{T}}G^{(r)}n. χ ( gr r D ( M , N )) = m T G ( r ) n . The matrices satisfy I + G ( 1 ) + G ( 2 ) + G ( 3 ) = C I+G^{(1)}+G^{(2)}+G^{(3)}=C I + G ( 1 ) + G ( 2 ) + G ( 3 ) = C . Adding the three layer contributions proves the formulas. □
Remark 14.12 (What the filtration does and does not reconstruct). Conservativity detects whether a perfect object or a morphism cone is acyclic; it is weaker than faithfulness or equivalence of categories. Likewise, the matrices (14.14) determine the Euler defect, but do not determine the extension maps in the filtered defect complex. No direct-sum splitting of (65) is asserted for general M , N M,N M , N . In particular, the dimensions of the scalar sectors alone do not reconstruct the original pairing. The nontrivial products ρ 1 ρ 2 = ρ 12 \rho_1\rho_2=\rho_{12} ρ 1 ρ 2 = ρ 12 , ρ 2 ρ 3 = ρ 23 \rho_2\rho_3=\rho_{23} ρ 2 ρ 3 = ρ 23 and ρ 1 ρ 2 ρ 3 = ρ 123 \rho_1\rho_2\rho_3=\rho_{123} ρ 1 ρ 2 ρ 3 = ρ 123 are precisely the algebraic information omitted by the semisimple quotient.
This calculation supplies a concrete condition on a proposed enriched boundary comparison: if it uses π \pi π as its reduction, then its additional gluing data must represent the bimodule J J J and the derived defect (63) , rather than simply retaining two independent torsion lines. It is an algebraic requirement; a realization by quantum BV–BFV states, in characteristic zero and with the original bordered grading, would still require a separate construction.
Torsion and object reconstruction
Theorem 14.4 is an obstruction to preserving composition . There is a separate obstruction to recovering objects from the scalar torsion function, even if all nonresonant characters are retained. Define the normalized meridional BF torsion function
T K ( t ) = τ ( X K , R − ; Q ( t ) ϕ ) s y m = Δ K ( t ) . \mathcal{T}_K(t)=\tau\left(X_K,R_-;\mathbb{Q}(t)_\phi\right)^{\mathrm{sym}}=\Delta_K(t). T K ( t ) = τ ( X K , R − ; Q ( t ) ϕ ) sym = Δ K ( t ) .
Here the equality uses the chosen symmetric normalization and means the particular relative Milnor-torsion representative used throughout this paper.
Theorem 14.13 (Failure of object reconstruction from all scalar twists). There are knots K K K and K ′ K' K ′ with T K ( t ) = T K ′ ( t ) \mathcal{T}_K(t)=\mathcal{T}_{K'}(t) T K ( t ) = T K ′ ( t ) as rational functions, and hence with equal nonresonant specializations at every character, but whose Alexander-graded knot Floer homologies are not isomorphic. Consequently no rule assigning a Floer object solely as a function of the normalized scalar meridional BF torsion can recover knot Floer homology for all knots.
Proof. Let K K K be the unknot and K ′ K' K ′ the Kinoshita–Terasaka knot. Both have Alexander polynomial 1, so their relative torsion functions and every specialization coincide. The unknot has Seifert genus zero, whereas the Kinoshita–Terasaka knot has genus two. Knot Floer homology detects Seifert genus [20 , 19 ] ; hence H F K ^ ( K ) \widehat{HFK}(K) H F K ( K ) and H F K ^ ( K ′ ) \widehat{HFK}(K') H F K ( K ′ ) have different Alexander support and are nonisomorphic.
The resulting comparison diagram
The results above give the following diagram of known operations and obstructions. The upper row is a shadow identity; there is no implied upper-to-lower equivalence of categories:
based relative BF complexes → torsion Q ( t ) / { ± t n } ← χ A knot Floer complexes \text{based relative BF complexes}\xrightarrow{\text{torsion}}\mathbb{Q}(t)/\{\pm t^n\}\xleftarrow{\chi_A}\text{knot Floer complexes} based relative BF complexes torsion Q ( t ) / { ± t n } χ A knot Floer complexes
Perf ( A k ) → − ⊗ A k L k ϵ D ( k ) \operatorname{Perf}(\mathcal{A}_k)\xrightarrow{-\otimes_{\mathcal{A}_k}^{\mathbb{L}} k_{\epsilon}}D(k) Perf ( A k ) − ⊗ A k L k ϵ D ( k )
The top horizontal comparison holds on the geometrically specified knot data with the normalizations of this paper. The lower arrow is an actual derived, A ∞ A_\infty A ∞ -compatible base-change functor. Theorem 14.4 shows why it is not a tensor/gluing bridge to a scalar boundary theory. Lemma 14.1 says why its ordinary derived output cannot even retain based torsion without additional structure. Theorem 14.13 shows that scalar torsion is insufficient to reconstruct all knot Floer objects. The distinction between these statements is important: they rule out three particular modes of comparison, not every potential enriched BV–BFV/Floer correspondence.
Requirements for an enriched comparison
The obstruction suggests a more sharply formulated research problem. Rather than seeking a functor from one-dimensional BFV torsion lines to bordered modules, one would need an enhancement B ∂ \mathcal{B}_\partial B ∂ of the boundary theory, containing at least nonzero off-diagonal morphisms that can model the Reeb-chord sector, together with a homotopy-coherent comparison with the bordered torus algebra. Such an enhancement should respect the knot’s peripheral parametrization, derived pairings, gradings and the relative determinant-line normalization. The existence of a geometrically natural BV–BFV construction with all these properties remains open within the present work; we do not claim to have constructed it.
A dg resolution of the torus algebra
The scalar obstructions do not prevent a comparison that retains the chord algebra. We now construct a small dg resolution of that algebra and prove a pairing-preserving equivalence of perfect-module categories. We also realize the differential on its framed representation space as a classical BFV Hamiltonian and as the target of a one-dimensional BV–BFV cotangent model. These are two explicit constructions. They must not be conflated with a derivation of the resolved algebra from the three-dimensional BF theory on a knot exterior: that identification is not supplied by choosing a resolution of the Floer algebra.
A six-generator resolution
In this subsection k k k is any field. Let A k A_k A k be the algebra with the same generators, idempotents and unsigned multiplication table as the degree-zero torus algebra. In characteristic two it is the A k A_k A k of Section 14. In other characteristics it denotes an algebraic lift of that table, not an assertion about sign-refined bordered Floer invariants. Set a = ρ 1 a=\rho_1 a = ρ 1 , b = ρ 2 b=\rho_2 b = ρ 2 , c = ρ 3 c=\rho_3 c = ρ 3 ; thus
a , c ∈ ι 0 A k ι 1 , b ∈ ι 1 A k ι 0 , b a = c b = 0. a,c\in\iota_0A_k\iota_1,\qquad b\in\iota_1A_k\iota_0,\qquad ba=cb=0. a , c ∈ ι 0 A k ι 1 , b ∈ ι 1 A k ι 0 , ba = c b = 0.
The multiplication convention reads paths from left to right, as in [15 equation (1.18)] . Define the free graded path algebra B k B_k B k over S = k ι 0 ⊕ k ι 1 S=k\iota_0\oplus k\iota_1 S = k ι 0 ⊕ k ι 1 by six generators:
Table 15.1.
There are no relations besides the endpoint rules. In particular B k B_k B k has finitely many generators but is infinite-dimensional. Its differential is the graded derivation determined by
d a = d b = d c = 0 , d u = b a , d v = c b , d w = c u − v a . (70) da=db=dc=0,\qquad du=ba,\qquad dv=cb,\qquad dw=cu-va. \tag*{(70)} d a = d b = d c = 0 , d u = ba , d v = c b , d w = c u − v a . (70)
The generators u , v u,v u , v resolve the two relations, and w w w resolves their overlap. There is an augmented dg-algebra map
q : B k ⟶ A k , q ( a ) = ρ 1 , q ( b ) = ρ 2 , q ( c ) = ρ 3 , q ( u ) = q ( v ) = q ( w ) = 0. (71) q:B_k\longrightarrow A_k,\qquad q(a)=\rho_1,\qquad q(b)=\rho_2,\qquad q(c)=\rho_3,\qquad q(u)=q(v)=q(w)=0. \tag*{(71)} q : B k ⟶ A k , q ( a ) = ρ 1 , q ( b ) = ρ 2 , q ( c ) = ρ 3 , q ( u ) = q ( v ) = q ( w ) = 0. (71)
Theorem 15.1 (Explicit dg resolution of the torus algebra). The differential (70) squares to zero, and q q q is a quasi-isomorphism. More precisely,
H n ( B k ) = 0 ( n ≠ 0 ) , H 0 ( B k ) = k ⟨ ι 0 , ι 1 , a , b , c , a b , b c , a b c ⟩ ≅ A k . H^n(B_k)=0\quad(n\ne0),\qquad H^0(B_k)=k\langle\iota_0,\iota_1,a,b,c,ab,bc,abc\rangle\cong A_k. H n ( B k ) = 0 ( n = 0 ) , H 0 ( B k ) = k ⟨ ι 0 , ι 1 , a , b , c , ab , b c , ab c ⟩ ≅ A k . Every expanded path containing a forbidden subword admits an explicit contracting homotopy. Thus the assertion follows from a direct word-by-word contraction, without assuming Koszul formality.
Proof. The only generator with a potentially nonzero second differential is w w w , and
d 2 w = c d u − ( d v ) a = c b a − c b a = 0. d^2w=c\,du-(dv)a=cba-cba=0. d 2 w = c d u − ( d v ) a = c ba − c ba = 0.
Since d d d is a graded derivation, this proves d 2 = 0 d^2=0 d 2 = 0 .
For a monomial in B k B_k B k , form its expanded path by replacing u u u by b a ba ba , v v v by c b cb c b , and w w w by $cba. The differential preserves the expanded path. Hence the underlying complex of B k B_k B k is the direct sum of the complexes associated to individual expanded paths in the alphabet a , b , c a,b,c a , b , c , together with the two idempotents.
Mark each occurrence of b a ba ba or c b cb c b in an expanded path. Two marked occurrences overlap only as the pair c b , b a cb,ba c b , ba inside c b a cba c ba . There is no chain of three overlapping occurrences, since the letters that finish b a ba ba and begin c b cb c b are incompatible with extending such a chain. The marked occurrences therefore split into disjoint clusters of two types: a single relation word b a ba ba or c b cb c b , or the overlapping word c b a cba c ba . The remaining intervals consist of ordinary arrows.
The complex for a single relation cluster is k u → d k b a ku\xrightarrow{d}kba k u d k ba , or k v → d k c b kv\xrightarrow{d}kcb k v d k c b , with the first term in degree − 1 -1 − 1 . It contracts by h ( b a ) = u h(ba)=u h ( ba ) = u , respectively h ( c b ) = v h(cb)=v h ( c b ) = v , and h ( u ) = h ( v ) = 0 h(u)=h(v)=0 h ( u ) = h ( v ) = 0 . For an overlap cluster the complex is
k w → d k ( c u ) ⊕ k ( v a ) → d k ( c b a ) , kw\xrightarrow{d}k(cu)\oplus k(va)\xrightarrow{d}k(cba), k w d k ( c u ) ⊕ k ( v a ) d k ( c ba ) ,
in degrees − 2 , − 1 , 0 -2,-1,0 − 2 , − 1 , 0 , with d w = c u − v a dw=cu-va d w = c u − v a and d ( c u ) = d ( v a ) = c b a d(cu)=d(va)=cba d ( c u ) = d ( v a ) = c ba . An explicit contraction is
h ( c b a ) = c u , h ( c u ) = 0 , h ( v a ) = − w , h ( w ) = 0. (72) h(cba)=cu,\qquad h(cu)=0,\qquad h(va)=-w,\qquad h(w)=0. \tag*{(72)} h ( c ba ) = c u , h ( c u ) = 0 , h ( v a ) = − w , h ( w ) = 0. (72)
Direct substitution gives d h + h d = 1 dh+hd=1 d h + h d = 1 and h 2 = 0 h^2=0 h 2 = 0 . The formulas remain valid in characteristic two, where the minus sign becomes a plus sign.
A basis monomial with a fixed expanded path is exactly a tiling of that path by its arrows and the generators u , v , w u,v,w u , v , w . Such tilings factor over the disjoint clusters just described. The corresponding complex is therefore a tensor product of the cluster complexes, with the ordinary Koszul signs, and one-dimensional degree-zero factors for the remaining arrow intervals. If a relation cluster is present, apply its contraction to the first such factor and the identity to the others. This contracts the whole expanded-path complex.
An expanded path without a forbidden subword contributes precisely one degree-zero basis element. The only such nonempty paths are a , b , c , a b , b c , a b c a,b,c,ab,bc,abc a , b , c , ab , b c , ab c : a compatible path of length four necessarily contains b a ba ba or c b cb c b . Including the two idempotents gives the eight listed cohomology classes. The map q q q identifies their products with the torus multiplication table and annihilates all contracted blocks, proving the quasi-isomorphism. □
Remark 15.2 (Why the coherence generator is indispensable). If one adds only u , v u,v u , v with d u = b a , d v = c b du=ba,dv=cb d u = ba , d v = c b , the element c u − v a cu-va c u − v a is a degree-− 1 -1 − 1 cycle. Its expanded-path block has no degree-− 2 -2 − 2 generator, so it is not a boundary. Such a two-relation construction is not a dg resolution of A k A_k A k . Adding w w w eliminates exactly this overlap class; Theorem 15.1 proves that no further generators are needed. This is the concrete coherence that is lost when the chords are replaced by scalar idempotents.
Derived equivalence and tensor pairings
Theorem 15.3 (Resolved boundary-category equivalence). Extension and restriction of scalars along q q q give inverse equivalences
F = − ⊗ B k L A k : Perf ( B k ) ⇄ Perf ( A k ) : G = q ∗ . (73) F=-\otimes_{B_k}^{L} A_k:\operatorname{Perf}(B_k)\rightleftarrows\operatorname{Perf}(A_k):G=q^*. \tag*{(73)} F = − ⊗ B k L A k : Perf ( B k ) ⇄ Perf ( A k ) : G = q ∗ . (73) For perfect right A k A_k A k -modules M M M and perfect left A k A_k A k -modules N N N there is a natural quasi-isomorphism
q ∗ M ⊗ B k L q ∗ N ≃ M ⊗ A k L N . (74) q^*M\otimes_{B_k}^{L}q^*N\simeq M\otimes_{A_k}^{L}N. \tag*{(74)} q ∗ M ⊗ B k L q ∗ N ≃ M ⊗ A k L N . (74) The same scalar-extension comparison respects derived composition of perfect bimodules. Thus the resolved boundary category retains the entire algebraic gluing operation, rather than only its Euler shadow.
Proof. The quasi-isomorphism q : B k → q ∗ A k q:B_k\to q^*A_k q : B k → q ∗ A k identifies the restricted free A k A_k A k -module with the free B k B_k B k -module in the derived category. The unit and counit of scalar extension and restriction are therefore quasi-isomorphisms on free generators. They remain quasi-isomorphisms under shifts, finite cones and direct summands. These operations generate the perfect categories, which proves (73) . This is a concrete instance of standard dg derived Morita theory [2 ] .
Likewise, the multiplication map A k ⊗ B k L A k → A k A_k\otimes_{B_k}^{L}A_k\to A_k A k ⊗ B k L A k → A k is a quasi-isomorphism: tensor the quasi-isomorphism B k → A k B_k\to A_k B k → A k with a semifree replacement and use the derived unit. Associativity then gives
q ∗ M ⊗ B k L q ∗ N ≃ M ⊗ A k L ( A k ⊗ B k L A k ) ⊗ A k L N ≃ M ⊗ A k L N . q^*M\otimes_{B_k}^{L}q^*N\simeq M\otimes_{A_k}^{L}(A_k\otimes_{B_k}^{L}A_k)\otimes_{A_k}^{L}N\simeq M\otimes_{A_k}^{L}N. q ∗ M ⊗ B k L q ∗ N ≃ M ⊗ A k L ( A k ⊗ B k L A k ) ⊗ A k L N ≃ M ⊗ A k L N .
Applying the same argument to bimodules gives compatibility with composition; the associativity is the usual associativity of derived tensor products, as realized by semifree or bar models. □
For the four projective pairs, equation (74) recovers all entries of the matrix C C C in (14.8), including its off-diagonal entries. There is no conflict with Theorem 14.7 : the target category uses the non-scalar dg algebra B k B_k B k and keeps the relation and overlap homotopies. Theorem 15.3 is an equivalence with the specified algebraic torus-module model . We have not changed the original grading qualifications or proved that these module categories are the quantum state categories of Abelian BF theory.
A classical BFV Hamiltonian for the resolution
Here work over a characteristic-zero field, for instance Q \mathbb{Q} Q , with ordinary graded symplectic geometry. Fix finite vector spaces V 0 V_0 V 0 , V 1 V_1 V 1 and consider the graded affine space R \mathcal{R} R of framed matrix coordinates for the six generators:
Table 2.
Their degrees are those in (15.1). The entries of the odd coordinates u , v u,v u , v are graded-commuting odd variables. The degree-one vector field Q R Q_{\mathcal{R}} Q R is
Q R a = Q R b = Q R c = 0 , Q R u = b a , Q R v = c b , Q R w = c u − v a . (75) Q_{\mathcal{R}}a=Q_{\mathcal{R}}b=Q_{\mathcal{R}}c=0,\qquad Q_{\mathcal{R}}u=ba,\qquad Q_{\mathcal{R}}v=cb,\qquad Q_{\mathcal{R}}w=cu-va. \tag*{(75)} Q R a = Q R b = Q R c = 0 , Q R u = ba , Q R v = c b , Q R w = c u − v a . (75)
The same overlap cancellation proves Q R 2 = 0 Q_{\mathcal{R}}^2=0 Q R 2 = 0 . This is the commutative coordinate dg algebra of the framed representation construction for B Q B_{\mathbb{Q}} B Q . At an ordinary degree-zero point, u , v , w u,v,w u , v , w vanish and the Q Q Q -fixed-point equations are b a = c b = 0 ba=cb=0 ba = c b = 0 . It is a framed representation model; no quotient by changes of basis and no category of quantized branes is implicit in this statement.
Let P = T ∗ R \mathcal{P}=T^*\mathcal{R} P = T ∗ R , with its degree-zero canonical symplectic form and Liouville one-form λ \lambda λ . For a coordinate x x x , its momentum p x p_x p x has degree − ∣ x ∣ -|x| − ∣ x ∣ . Using the canonical graded evaluation convention, define
Θ = ⟨ p , Q R x ⟩ = ⟨ p u , b a ⟩ + ⟨ p v , c b ⟩ + ⟨ p w , c u − v a ⟩ . (76) \Theta=\langle p,Q_{\mathcal{R}}x\rangle=\langle p_u,ba\rangle+\langle p_v,cb\rangle+\langle p_w,cu-va\rangle. \tag*{(76)} Θ = ⟨ p , Q R x ⟩ = ⟨ p u , ba ⟩ + ⟨ p v , c b ⟩ + ⟨ p w , c u − v a ⟩ . (76)
Each term has degree one. The pairings are the trace pairings between rectangular matrix spaces and their duals, with the Koszul convention fixed by the canonical cotangent primitive.
Proposition 15.4 (Classical cotangent BV–BFV realization). The Hamiltonian vector field of Θ \Theta Θ is the cotangent lift of Q R Q_{\mathcal{R}} Q R and is cohomological. In particular, { Θ , Θ } = 0 \{\Theta,\Theta\}=0 { Θ , Θ } = 0 . For an oriented interval I I I , the fields
F I = Map ( T [ 1 ] I , P ) \mathcal{F}_I=\operatorname{Map}(T[1]I,\mathcal{P}) F I = Map ( T [ 1 ] I , P ) carry the classical one-dimensional AKSZ BV–BFV structure with bulk symplectic form and action
ω I = ∫ T [ 1 ] I δ p δ x , S I = ∫ T [ 1 ] I ( ⟨ p , D x ⟩ + Θ ( x , p ) ) , (77) \omega_I=\int_{T[1]I}\delta p\,\delta x,\qquad S_I=\int_{T[1]I}\left(\langle p,Dx\rangle+\Theta(x,p)\right), \tag*{(77)} ω I = ∫ T [ 1 ] I δ p δ x , S I = ∫ T [ 1 ] I ( ⟨ p , D x ⟩ + Θ ( x , p ) ) , (77) where D D D is the source de Rham vector field. The degrees of ω I \omega_I ω I and S I S_I S I are − 1 -1 − 1 and 0 0 0 . The boundary data are P \mathcal{P} P at each endpoint, with outward orientation signs:
α ∂ I = λ 1 − λ 0 , ω ∂ I = δ α ∂ I , S ∂ I = Θ 1 − Θ 0 . \alpha_{\partial I}=\lambda_1-\lambda_0,\qquad\omega_{\partial I}=\delta\alpha_{\partial I},\qquad S_{\partial I}=\Theta_1-\Theta_0. α ∂ I = λ 1 − λ 0 , ω ∂ I = δ α ∂ I , S ∂ I = Θ 1 − Θ 0 . With the corresponding canonical sign conventions the bulk and boundary structures satisfy the BV–BFV compatibility identity.
Proof. For any vector field Q Q Q on a graded manifold, the fiber-linear Hamiltonian ⟨ p , Q x ⟩ \langle p,Qx\rangle ⟨ p , Q x ⟩ generates its cotangent lift. The cotangent lift respects graded Lie brackets; hence its square vanishes because Q R 2 = 0 Q_{\mathcal{R}}^2=0 Q R 2 = 0 . Equivalently, the Poisson self-bracket of this Hamiltonian is the fiber-linear Hamiltonian for [ Q R , Q R ] [Q_{\mathcal{R}},Q_{\mathcal{R}}] [ Q R , Q R ] , which is zero. This proves the target master equation.
Transgression over T [ 1 ] I T[1]I T [ 1 ] I lowers degree by one. Both the kinetic term and Θ \Theta Θ have target total degree one, giving the displayed bulk degrees. Varying the kinetic term and integrating the source differential by parts yields the endpoint Liouville forms with opposite orientation signs. The bulk cohomological vector field is the sum of D D D and the lifted target field; its square is zero by the source differential, the target master equation and their graded commutation. The integration-by-parts identity is precisely the classical BV–BFV bulk–boundary compatibility relation. This is the standard AKSZ-with-boundary construction applied to the explicitly specified target [6 ] . ▫
Geometric interpretation
The construction now supplies a concrete dg algebra and a classical BV–BFV realization, as well as an unconditional algebraic equivalence (73) . A full BV–BFV/bordered-Floer equivalence still requires the following independent geometric work:
(i) Construct a quantum boundary category from the chosen three-dimensional field theory and compute a generator’s derived endomorphism algebra. Prove that it is quasi-isomorphic to B k B_k B k (or derived Morita equivalent to it), using that field theory’s observables and gauge fixing rather than inserting A k A_k A k as input.
(ii) For each parametrized knot exterior or cobordism, construct its boundary module from the same theory and identify it with the appropriate bordered Floer invariant. The abstract equivalence of module categories does not identify which module a manifold supplies.
(iii) Identify the field-theoretic sewing map with derived tensor composition, preserving peripheral parametrizations, grading data, coefficients, and the determinant-line normalization of the meridional relative model.
If these identifications are established, the equivalence and its algebraic pairing compatibility follow from Theorem 15.3 . They are not consequences of that theorem alone.
There are relevant existing geometric constructions, which also clarify the limits of a shortcut. Auroux identifies the bordered algebra with endomorphisms of generating arc-product Lagrangians in a partially wrapped Fukaya category [3 Theorems 1.2–1.3] . The same paper explicitly qualifies the scope of the full geometric module and gluing arguments after its Theorems 1.4–1.5; we do not invoke them as an unqualified BV–BFV equivalence. Chekeres–Losev–Mnev–Youmans obtain enhanced Fukaya–Morse compositions from a gauge-fixed BF effective action [4 ] . Their construction involves gradient-tree compositions for generating functions on a closed manifold, rather than an automatic identification of the knot-exterior BFV category with the stopped Heegaard-Floer boundary category. Matching these geometric settings, including stops and boundary conditions, is an additional research problem.
Dg and Koszul techniques also occur in bordered Floer theory [5 ] . The resolution and cotangent model above address the algebraic part of the comparison; the geometric identifications remain to be established. For the particular polynomial quantum sector of nonresonant three-dimensional Abelian BF theory computed in the next section, the proposed dg Morita identification is obstructed. Thus an abstract choice of B k B_k B k cannot be regarded as a computation of that field theory’s boundary observable algebra.
Quantum BFV reduction on the torus
The preceding algebraic constructions leave open whether the original three-dimensional linear BF theory supplies the torus algebra. In a precisely specified quantum sector, this question can be answered directly. We compute the polarized BFV differential, its polynomial state cohomology, and the polynomial Weyl observable algebra. The calculation uses the boundary differential of the three-dimensional Abelian theory, rather than an algebra imported from Floer theory.
The polarized state complex
Fix a characteristic-zero field k k k , a holonomy z ∈ k × ∖ { 1 } z \in k^\times\setminus\{1\} z ∈ k × ∖ { 1 } and u = z − 1 u = z - 1 u = z − 1 . Set K = k ( ( ℏ ) ) K = k((\hbar)) K = k (( ℏ )) , so ℏ \hbar ℏ is invertible. Use the twisted cellular cochain complex of the closed boundary torus with meridional holonomy z z z and longitudinal holonomy 1 1 1 :
D 0 = k → d 0 D 1 = k 2 → d 1 D 2 = k , d 0 ( t ) = ( u t , 0 ) , d 1 ( r , s ) = u s . D^0 = k \xrightarrow{d^0} D^1 = k^2 \xrightarrow{d^1} D^2 = k, \qquad d^0(t) = (ut,0),\quad d^1(r,s) = us. D 0 = k d 0 D 1 = k 2 d 1 D 2 = k , d 0 ( t ) = ( u t , 0 ) , d 1 ( r , s ) = u s .
In the A A A -polarization of the three-dimensional Abelian BF theory, take polynomial functions on D [ 1 ] D[1] D [ 1 ] as the finite cellular state model, after a fixed half-density trivialization. This is the polynomial version of the standard cellular A A A -polarized state construction and quantum BFV operator [10 Section 7, especially equations (71)–(74)] . The coordinates and ghost degrees are
ξ ( 1 ) , x , y ( 0 ) , η ( − 1 ) . \xi\ (1),\qquad x,y\ (0),\qquad\eta\ (-1). ξ ( 1 ) , x , y ( 0 ) , η ( − 1 ) .
The odd coordinates ξ , η \xi,\eta ξ , η are exterior generators; the even coordinates x , y x,y x , y are polynomial generators. The normalized degree-one BFV differential on
S = K [ x , y ] ⊗ K Λ K ( ξ , η ) (78) S = K[x,y]\otimes_K\Lambda_K(\xi,\eta) \tag*{(78)} S = K [ x , y ] ⊗ K Λ K ( ξ , η ) (78)
is
δ ξ = δ y = 0 , δ x = u ξ , δ η = u y , δ = u ξ ∂ x + u y ∂ η . (79) \delta\xi= \delta y = 0,\qquad\delta x = u\xi,\qquad\delta\eta= uy,\qquad\delta= u\xi\partial_x + uy\partial_\eta. \tag*{(79)} δ ξ = δ y = 0 , δ x = u ξ , δ η = u y , δ = u ξ ∂ x + u y ∂ η . (79)
These formulas are the dual coordinate differential of the displayed cellular complex, with ghost degree 1 − r 1-r 1 − r for an r r r -cell.
Let p f p_f p f be the canonical momentum conjugate to a coordinate f f f , with ∣ p f ∣ = − ∣ f ∣ |p_f|=-|f| ∣ p f ∣ = − ∣ f ∣ . Choose the algebraic quantization convention p f = ℏ ∂ f p_f=\hbar\partial_f p f = ℏ ∂ f and graded commutators [ p f , f ] g r = ℏ [p_f,f]_{\mathrm{gr}}=\hbar [ p f , f ] gr = ℏ . The normal-ordered quantum charge is
Θ ^ = u ( ξ p x + y p η ) = ℏ δ on S . (80) \widehat{\Theta}=u(\xi p_x+yp_\eta)=\hbar\delta\quad\text{on }S. \tag*{(80)} Θ = u ( ξ p x + y p η ) = ℏ δ on S . (80)
Conventional factors of i i i multiply this charge by a nonzero scalar and do not change the cohomology over K K K . The operators in its two summands anticommute and square to zero, so Θ ^ 2 = 0 \widehat{\Theta}^2=0 Θ 2 = 0 ; no higher quantum correction is needed for this linear boundary charge. The coordinate vector field δ \delta δ has zero divergence, so the constant half-density trivialization introduces no additional divergence term in this operator.
Vacuum cohomology
Theorem 16.1 (Nonresonant polarized vacuum cohomology). The polynomial state complex ( S , δ ) (S,\delta) ( S , δ ) has a strong deformation retract onto the vacuum line K ⋅ 1 K\cdot1 K ⋅ 1 in ghost degree zero. Writing ε \varepsilon ε for evaluation at zero and ι : K → S \iota:K\to S ι : K → S for the inclusion of constants, there is a homotopy H H H with
δ H + H δ = 1 − ι ε , H 2 = H ι = ε H = 0. \delta H+H\delta=1-\iota\varepsilon,\qquad H^2=H\iota=\varepsilon H=0. δ H + H δ = 1 − ι ε , H 2 = H ι = ε H = 0. In particular H ∙ ( S , δ ) = K H^\bullet(S,\delta)=K H ∙ ( S , δ ) = K . The same formulas extend continuously to completion in total polynomial degree. They also contract the complex with differential Θ ^ \widehat{\Theta} Θ over K K K , after replacing H H H by ℏ − 1 H \hbar^{-1}H ℏ − 1 H .
Proof. Define the degree-− 1 -1 − 1 graded derivation
κ = u − 1 ( x ∂ ξ + η ∂ y ) . \kappa=u^{-1}(x\partial_\xi+\eta\partial_y). κ = u − 1 ( x ∂ ξ + η ∂ y ) . On each of the four generators, direct substitution gives ( δ κ + κ δ ) f = f (\delta\kappa+\kappa\delta)f=f ( δ κ + κ δ ) f = f . The graded commutator of the two odd derivations is a degree-zero derivation, so on all polynomials
δ κ + κ δ = N , \delta\kappa+\kappa\delta=N, δ κ + κ δ = N , where N N N counts total polynomial degree, including the exterior generators. Both derivations preserve this degree. On the degree-n n n homogeneous component define H = κ / n H=\kappa/n H = κ / n for n > 0 n>0 n > 0 , and H = 0 H=0 H = 0 on constants. All denominators are invertible in characteristic zero. This gives the retract identity. The identity κ 2 = 0 \kappa^2=0 κ 2 = 0 follows on the generators and hence on the graded polynomial algebra, proving the side conditions. For the completed algebra the formula is applied componentwise. Finally, Θ ^ = ℏ δ \widehat{\Theta}=\hbar\delta Θ = ℏ δ and ℏ \hbar ℏ is invertible in K K K . □
Remark 16.2 (The formal-parameter restriction). Over K [ [ ℏ ] ] K[[\hbar]] K [[ ℏ ]] , the unnormalized operator ℏ δ \hbar\delta ℏ δ can leave ℏ \hbar ℏ -torsion in cohomology; the theorem does not discard that torsion by an unstated division. It concerns either the normalized BRST differential δ \delta δ , or the quantum charge after passage to K ( ( ℏ ) ) K((\hbar)) K (( ℏ )) . It also concerns polynomial states and their stated completion, not arbitrary distributional sectors or added defects. The disappearance of residual boundary fields is the familiar cohomological reduction of cellular BF theory [10 Section 7.4] ; the displayed homotopy supplies it explicitly for this meridional torus.
The quantum observable dg algebra
Let W \mathcal{W} W be the graded polynomial Weyl algebra generated by the four coordinates and their four momenta, over K K K , with the canonical graded commutation relations above and all other pairs graded commuting. The algebra has the degree-one differential
d W = ℏ − 1 [ Θ ^ , − ] g r . d_{\mathcal{W}}=\hbar^{-1}[\widehat{\Theta},-]_{\mathrm{gr}}. d W = ℏ − 1 [ Θ , − ] gr .
It restricts to δ \delta δ on the coordinates. On momenta it is
d W p ξ = u p x , d W p y = − u p η , d W p x = d W p η = 0. (81) d_{\mathcal{W}}p_{\xi}=up_x,\qquad d_{\mathcal{W}}p_y=-up_{\eta},\qquad d_{\mathcal{W}}p_x=d_{\mathcal{W}}p_{\eta}=0. \tag*{(81)} d W p ξ = u p x , d W p y = − u p η , d W p x = d W p η = 0. (81)
The signs are fixed by the graded canonical relations; for instance the anticommutator of p ξ p_{\xi} p ξ with ξ \xi ξ is ℏ \hbar ℏ , while [ p y , y ] = ℏ [p_y,y]=\hbar [ p y , y ] = ℏ .
Theorem 16.3 (Vacuum reduction of the polynomial BFV observables). The unit inclusion
K ⟶ ( W , d W ) K\longrightarrow(\mathcal{W},d_{\mathcal{W}}) K ⟶ ( W , d W ) is a quasi-isomorphism of dg algebras. Consequently Perf ( W ) ≃ Perf ( K ) \operatorname{Perf}(\mathcal{W})\simeq\operatorname{Perf}(K) Perf ( W ) ≃ Perf ( K ) .
Proof. Normal ordering, with coordinates to the left and momenta to the right, gives the graded PBW vector-space isomorphism
σ : Sym K ( V ⊕ V ∨ ) ⟶ W , \sigma:\operatorname{Sym}_{K}(V\oplus V^{\vee})\longrightarrow\mathcal{W}, σ : Sym K ( V ⊕ V ∨ ) ⟶ W , where V V V is the four-dimensional coordinate complex and V ∨ V^{\vee} V ∨ denotes the conjugate momentum complex with the differential (81) . This is a vector-space isomorphism, not an algebra isomorphism. Since d W d_{\mathcal{W}} d W sends coordinates to coordinates and momenta to momenta, it preserves normal ordering and corresponds under σ \sigma σ to the linear derivation on this graded symmetric algebra. There are no lower-order reordering terms in this differential.
Both summands of V ⊕ V ∨ V\oplus V^{\vee} V ⊕ V ∨ are contractible. Extend the coordinate contraction from the previous theorem by
κ ( p x ) = u − 1 p ξ , κ ( p η ) = − u − 1 p y , κ ( p ξ ) = κ ( p y ) = 0. \kappa(p_x)=u^{-1}p_{\xi},\qquad\kappa(p_{\eta})=-u^{-1}p_y,\qquad\kappa(p_{\xi})=\kappa(p_y)=0. κ ( p x ) = u − 1 p ξ , κ ( p η ) = − u − 1 p y , κ ( p ξ ) = κ ( p y ) = 0. On symbols, d κ + κ d = N d\kappa+\kappa d=N d κ + κ d = N again counts total PBW degree. Dividing by this degree on the positive-degree components and transporting by σ \sigma σ contracts the observable complex onto its constants. This contraction need not be an algebra derivation. It proves that H ∙ ( W ) = K ⋅ [ 1 ] H^{\bullet}(\mathcal{W})=K\cdot[1] H ∙ ( W ) = K ⋅ [ 1 ] . The unit inclusion is itself a unital dg-algebra map and induces that cohomology isomorphism, proving the theorem. The perfect-module equivalence follows from dg scalar extension as in Theorem 15.3 .
Corollary 16.4 (Regular formal interactions do not add vacuum classes). Let λ \lambda λ be a formal parameter of degree zero and let d λ = d W + Δ d_{\lambda}=d_{\mathcal{W}}+\Delta d λ = d W + Δ be a unital K [ [ λ ] ] K[[\lambda]] K [[ λ ]] -linear dg-algebra differential on W [ [ λ ] ] \mathcal{W}[[\lambda]] W [[ λ ]] , where Δ \Delta Δ raises λ \lambda λ -adic order and d λ 2 = 0 d_{\lambda}^{2}=0 d λ 2 = 0 . Then the unit inclusion
K [ [ λ ] ] ⟶ ( W [ [ λ ] ] , d λ ) K[[\lambda]]\longrightarrow(\mathcal{W}[[\lambda]],d_{\lambda}) K [[ λ ]] ⟶ ( W [[ λ ]] , d λ ) is a quasi-isomorphism. Thus a regular unital formal perturbation with this same leading boundary complex still has only vacuum cohomology.
Proof. Extend the contraction of Theorem 16.3 λ \lambda λ -adically. The series in the homological perturbation lemma converge in this topology, since Δ \Delta Δ raises λ \lambda λ -adic order. Unitality gives Δ ( 1 ) = 0 \Delta(1)=0 Δ ( 1 ) = 0 . Hence the corrected inclusion remains the inclusion of constants and the transferred differential on K [ [ λ ] ] K[[\lambda]] K [[ λ ]] is zero. The lemma supplies a deformation retract onto this complex [10 Section 4] . The inclusion is a unital dg-algebra map, so the cohomology isomorphism proves the assertion.
A Hochschild obstruction
Let A K A_K A K be the characteristic-zero degree-zero algebraic lift of the torus table, as defined in § 15.1 . It is placed in ghost degree zero with zero differential.
Theorem 16.5 (No Morita identification in the computed quantum sector). There is no K K K -linear dg derived Morita equivalence between ( W , d W ) (\mathcal{W},d_{\mathcal{W}}) ( W , d W ) and A K A_K A K , or between W \mathcal{W} W and its six-generator resolution B K B_K B K . More precisely, the Morita invariants are
H H 0 ( W ) ≅ K , H H 0 ( A K ) ≅ K 2 . HH_0(\mathcal{W}) \cong K,\qquad HH_0(A_K) \cong K^2. H H 0 ( W ) ≅ K , H H 0 ( A K ) ≅ K 2 . Here the left side is the degree-zero homology of the derived dg Hochschild complex; it is not the trace quotient of the underlying Weyl algebra with its differential forgotten.
Proof. By Theorem 16.3 , the dg algebra W \mathcal{W} W is quasi-isomorphic to K K K , so its derived Hochschild homology is that of K K K . For the degree-zero algebra A K A_K A K , H H 0 ( A K ) = A K / [ A K , A K ] HH_0(A_K)=A_K/[A_K,A_K] H H 0 ( A K ) = A K / [ A K , A K ] . Every off-diagonal chord is a commutator with an idempotent. The remaining chord elements are
a b = [ a , b ] , b c = [ b , c ] . ab=[a,b],\qquad bc=[b,c]. ab = [ a , b ] , b c = [ b , c ] . because b a = c b = 0 ba=cb=0 ba = c b = 0 . Thus the whole chord ideal J J J lies in the commutator subspace. Conversely, the quotient A K / J = K × K A_K/J=K\times K A K / J = K × K is commutative, so every commutator lies in J J J . It follows that [ A K , A K ] = J [A_K,A_K]=J [ A K , A K ] = J and H H 0 ( A K ) = K ι 0 ⊕ K ι 1 HH_0(A_K)=K\iota_0\oplus K\iota_1 H H 0 ( A K ) = K ι 0 ⊕ K ι 1 .
Hochschild homology is invariant under K K K -linear dg derived Morita equivalence [1 ] . Its displayed dimensions rule out the claimed equivalence. Since B K B_K B K is quasi-isomorphic to A K A_K A K , the same obstruction applies to B K B_K B K . □
Remark 16.6 (Exact scope of the conclusion). This result computes a boundary observable algebra of the original linear BF model in its fixed nonresonant sector and obstructs its proposed identification with the specified torus module algebra. Adding only contractible BRST pairs, or replacing this observable algebra by a quasi-isomorphic resolution, cannot remove the obstruction. Corollary 16.4 also treats regular unital formal interactions with the same leading complex. The theorem does not exclude other interacting theories, resonant holonomy, an all-character family, new boundary defects, stopped or wrapped brane categories, or a different quantization sector. It does not assert a theorem about changing Floer coefficients: the torus algebra in this comparison is the explicitly stated characteristic-zero algebraic lift.
Reduced sewing
The reduced polarized state space in this sector is a line. Any BFV-compatible sewing pairing that descends to these reduced spaces is therefore a bilinear map K ⊗ K K → K K\otimes_K K\to K K ⊗ K K → K , determined by a scalar. With compatible vacuum normalizations its action is ordinary scalar multiplication; determinants, torsion factors and density conventions may change that scalar but cannot create additional reduced boundary states. The standard Gaussian cellular sewing construction and cohomological reduction are described in [10 Sections 7.3–7.4] .
When the bulk residual fields are also absent, the modified quantum master equation makes a bulk boundary state a BFV cocycle. Theorem 16.1 then reduces that state to a scalar multiple of the vacuum. Its scalar determinant-line information is the appropriate location for the Alexander torsion comparison already proved in this manuscript. The quantum calculation does not promote that scalar to a bordered Floer module with chord operations. Consequently a geometric equivalence cannot be obtained through the computed polynomial Weyl sector by gauge fixing or dg resolution alone. A proposed enriched theory must change the actual boundary category and compute its new algebra and manifold modules independently.
Discussion
The relative meridional torsion identifies the Alexander polynomial within a based cellular BF model. Exact prime-cyclotomic evaluation, combined with a genus bound, determines its coefficients through field traces; the threshold is sharp. Field norms recover branched-cover homology orders, and annular gluing recovers connected-sum multiplicativity. For positive L-space knots, the established staircase classification turns this arithmetic reconstruction into a filtered Floer model.
Scalar torsion does not determine knot Floer homology for general knots. At the boundary-algebra level, the degree-zero torus model also obstructs single-field tensor factorizations.
Retaining both idempotents makes reduction conservative on perfect modules, but does not preserve pairings: the chord ideal gives the precise derived defect and its three-layer filtration.
The dg resolution in § 15 retains the chord operations and their relation and overlap homotopies. It preserves perfect-module categories and derived pairings, and its representation differential has a classical one-dimensional BV–BFV realization. This construction starts from the torus algebra. It does not derive that algebra, or the modules associated to manifolds, from a three-dimensional field theory.
In the fixed nonresonant polynomial sector of § 16 , over characteristic zero with ℏ \hbar ℏ inverted, the quantum BFV states and Weyl observables contract to the coefficient field. Their Hochschild invariant obstructs a dg Morita equivalence with the degree-zero torus algebra. This conclusion is specific to the stated sector and coefficient conventions. A full geometric BV–BFV/Floer comparison would require an independently constructed boundary category, compatible manifold modules, and sewing maps respecting gradings and peripheral parametrizations.
Appendix A. Explicit torsion matrices for the trefoil
We verify the absolute/relative normalization by a direct Fox calculation. For the right-handed trefoil take the presentation π 1 ( X ) = ⟨ x , y ∣ x y x = y x y ⟩ \pi_1(X)=\langle x,y\mid xyx=yxy\rangle π 1 ( X ) = ⟨ x , y ∣ x y x = y x y ⟩ and relator r = x y x y − 1 x − 1 y − 1 r=xyxy^{-1}x^{-1}y^{-1} r = x y x y − 1 x − 1 y − 1 . Both generators map to t t t under abelianization. Fox’s rule ∂ ( u v ) / ∂ x = ∂ u / ∂ x + u ∂ v / ∂ x \partial(uv)/\partial x=\partial u/\partial x+u\,\partial v/\partial x ∂ ( uv ) / ∂ x = ∂ u / ∂ x + u ∂ v / ∂ x and ∂ x − 1 / ∂ x = − x − 1 \partial x^{-1}/\partial x=-x^{-1} ∂ x − 1 / ∂ x = − x − 1 gives
ϕ ( ∂ r ∂ x ) = 1 + t 2 − t = t 2 − t + 1 , ϕ ( ∂ r ∂ y ) = − ( t 2 − t + 1 ) . (82) \phi\left(\frac{\partial r}{\partial x}\right)=1+t^2-t=t^2-t+1,\qquad\phi\left(\frac{\partial r}{\partial y}\right)=-(t^2-t+1). \tag*{(82)} ϕ ( ∂ x ∂ r ) = 1 + t 2 − t = t 2 − t + 1 , ϕ ( ∂ y ∂ r ) = − ( t 2 − t + 1 ) . (82)
The cellular chain complex (with the preferred basis of generators) is
0 ⟶ F → ( q − q ) F 2 → ( t − 1 t − 1 ) F ⟶ 0 , q = t 2 − t + 1. 0\longrightarrow F \xrightarrow{\begin{pmatrix}q\\-q\end{pmatrix}} F^2 \xrightarrow{\begin{pmatrix}t-1&t-1\end{pmatrix}} F\longrightarrow0,\qquad q=t^2-t+1. 0 ⟶ F ( q − q ) F 2 ( t − 1 t − 1 ) F ⟶ 0 , q = t 2 − t + 1.
Its torsion is q / ( t − 1 ) q/(t-1) q / ( t − 1 ) , up to a conventional sign and monomial. The meridional annulus contributes 1 / ( t − 1 ) 1/(t-1) 1/ ( t − 1 ) , leaving the relative torsion class q q q . Multiplying by t − 1 t^{-1} t − 1 yields the unique symmetric normalized representative B 31 ( t ) = t − 1 − 1 + t \mathcal{B}_{31}(t)=t^{-1}-1+t B 31 ( t ) = t − 1 − 1 + t . There are no singular prime-order specializations, since q ( ζ p ) ≠ 0 q(\zeta_p)\ne0 q ( ζ p ) = 0 for every prime p p p .
Appendix B. Acyclic Gaussian determinants and the scope of BV pushforward
For completeness, let ( C ∙ , d ) (C^\bullet,d) ( C ∙ , d ) be a finite-dimensional acyclic complex with fixed bases and a decomposition
C k = B k ⊕ L k , B k = d ( C k − 1 ) , d : L k → ∼ B k + 1 . C^k=B^k\oplus L^k,\qquad B^k=d(C^{k-1}),\qquad d:L^k\xrightarrow{\sim}B^{k+1}. C k = B k ⊕ L k , B k = d ( C k − 1 ) , d : L k ∼ B k + 1 .
In a basis adapted to the direct sums, the torsion is an alternating product of determinants of the isomorphisms d ∣ L k : L k → B k + 1 d|_{L^k}:L^k\to B^{k+1} d ∣ L k : L k → B k + 1 , together with the determinant of the basis changes. In the Milnor convention used in this paper, this product is fixed by the requirement that the two-term circle complex ((16) ) has torsion ( z − 1 ) − 1 (z-1)^{-1} ( z − 1 ) − 1 . The mixed relative cochain complex carries an analogous splitting, and its determinant-line element is independent of the splitting as an algebraic torsion, though it still depends on cellular lifts.
Finite-dimensional Abelian BV pushforward is a Gaussian/Berezin integration after choosing a Lagrangian gauge-fixing subspace. Its determinant factors are precisely such alternating products; an actual state also involves half-densities and possibly universal phases. For a non-unimodular L ∞ L_\infty L ∞ , no invariant complex number is assigned without controlling Euler structures or bases. Our BF character is therefore explicitly the normalized Milnor-torsion branch of the finite-dimensional BF determinant-line output. None of the present proofs requires selecting a functional integral measure or claiming an equality of unnormalized path integrals.
Use of generative AI
The manuscript was prepared in LaTeX and compiled with pdf-LaTeX. Generative AI tools assisted with the formulation and editing of the prose, LaTeX typesetting, exploratory calculations, the development of mathematical arguments, and the identification of relevant references. The author takes responsibility for the content of the manuscript.
References
Institut für Mathematik, Universität Zürich, Winterthurerstrasse 190, CH-8057 Zürich, Switzerland
Email address: nima.moshayedi@math.uzh.ch