Artin groups are the "braid-like" groups built from a reflection group (a Coxeter group) by forgetting that each reflection squares to the identity. A central way to study them is through their twisted homology: homology where each loop is weighted by a formal variable, so the answer is a module over a polynomial ring rather than a list of integers. The finite-order parts of these modules (torsion) carry most of the interesting information. Paolini and Salvetti found an efficient way to compute this homology when every generator is weighted by the same single variable $q$. Their method, discrete Morse theory with "precise matchings," cancels cells of a combinatorial model in pairs whose weights differ by exactly one cyclotomic factor (Weighted sheaves and homology of Artin groups). The method needs a principal ideal domain, meaning a ring like $\mathbb{Q}[q^{\pm1}]$ where every module splits into cyclic pieces. It therefore stops working for reflection groups with two generator lengths ($B_n$, $F_4$, $G_2$, and affine $\tilde B_n$, $\tilde C_n$, $\tilde F_4$, $\tilde G_2$). Those groups naturally carry two weights, $q$ and $t$, and the coefficient ring $\mathbb{Q}[q^{\pm1},t^{\pm1}]$ is no longer a principal ideal domain.
This note proposes a computational and theoretical program for that two-variable case. It reduces the problem to a finite family of one-variable problems, taken along "slopes" $t=q^{r}$ where Paolini–Salvetti matchings apply. It then isolates the one kind of torsion this reduction cannot see. Because Paolini and Salvetti proved the $K(\pi,1)$ conjecture for all affine Artin groups (Proof of the K(pi, 1) conjecture for affine Artin groups), the combinatorial model (the Salvetti complex) computes the actual group homology for the affine cases. The results are invariants of the groups themselves, not only of a model space.
Core Claim
Let $M$ be a two-variable twisted homology module of a non-simply-laced Artin group over $R=\mathbb{Q}[q^{\pm1},t^{\pm1}]$. Commutative algebra splits the torsion of $M$ into two parts. The first is supported on curves, which are height-one primes such as $\varphi_d(q^a t^b)$, where $\varphi_d$ is the $d$-th cyclotomic polynomial. The second is "pseudo-null" and supported on finitely many points. Localizing at a height-one prime gives a ring in which every module splits into cyclic pieces, so curve-supported torsion is fully described by a finite set of one-variable specializations.
We conjecture three things. First, every curve in the support has the cyclotomic form $\varphi_d(q^a t^b)$. Second, along each such slope the Salvetti complex admits a precise matching, so the one-variable machinery computes the local structure exactly. Third, for the spherical and affine types listed above, pseudo-null torsion either vanishes or appears only at a small, explicitly computable set of roots of unity.
If this holds, two-variable twisted homology becomes a finite "torsion profile over slopes" plus a small point-supported correction. That is a closed-form answer in a setting where the only general current tool is brute-force module computation.
Where the Other Directions Enter
The four directions are not given equal weight. The core mechanism is algebraic-topological reduction, used to settle questions in the $K(\pi,1)$ setting.
Rate-dependent error scaling supplies a methodological stance rather than a technique. Hassani, Mori, Tanaka and Urbanke replaced the rate-independent polar-code error bound with a profile indexed by the transmission rate $R$, with the second-order term governed by $Q^{-1}(R/I(W))$ (Rate-Dependent Analysis of the Asymptotic Behavior of Channel Polarization). In the same way, we replace the binary question "is $M$ torsion-free?" with a profile over the slope parameter $r$. We also borrow the local-to-global proof structure of General Strong Polarization, where a global property reduces to checkable local conditions.
Quark-hadron duality enters only as an analogy. A generic slope recovers the smooth, free part of $M$, much as averaging over resonances recovers the smooth deep-inelastic curve. Torsion concentrates at isolated "resonant" slopes. Determining exactly where those resonances sit is the analogue of locating the kinematic boundary of duality in parity-violating scattering (Measurements of Parity-Violating Asymmetries in Electron-Deuteron Scattering in the Nucleon Resonance Region).
We considered and rejected the more literal physics bridges. Gap 2 of the research file requires a channel-representation theorem for hadronization that no cited paper supplies. The only electroweak duality dataset in the file has seven kinematic points with no reported asymmetry values on Lacuna. Either project would be speculative in a way this one is not.
Research Program
Step 1: matching search. We build the Salvetti complex for each target Coxeter system with its bigraded cell weights. A cell's weight is a product of cyclotomic factors in $q$ and $t$ inherited from the two conjugacy classes of generators. We then search for acyclic Morse matchings that are precise with respect to a chosen slope. "Precise" means matched cells differ by exactly one irreducible factor of the form $\varphi_d(q^a t^b)$. The search reuses the cell structure and poset orderings from On the classifying space of Artin monoids and the weighted-sheaf formalism of Paolini and Salvetti.
Step 2: candidate slopes. We compute the Morse-reduced complex over the fraction field $\mathbb{Q}(q,t)$ to get generic ranks. We then read off candidate slopes from the irreducible factors of the reduced boundary determinants. Only finitely many slopes can matter.
Step 3: local structure along each slope. For each candidate slope we specialize to one variable and apply precise matchings. The spectral sequence collapses at $E^2$, and the local torsion structure is read off from incidence-matrix ranks, as in the Paolini–Salvetti $\tilde C_n$ computation.
Step 4: pseudo-null correction. We compute the remaining point-supported part by restricting a direct module computation to the small Morse-reduced complex, localized at the finitely many intersection points of candidate curves.
Step 5: integral refinement. Following the criterion of Equivariant K-homology for hyperbolic reflection groups, we test whether the reduced boundary matrices are totally unimodular after a basis change. Total unimodularity means every square submatrix has determinant $-1$, $0$, or $1$. If it holds, we obtain integral rather than rational statements at each slope.
The theoretical deliverable is a proof of the cyclotomic-curve conjecture for the spherical types $B_n$, $F_4$, $G_2$ and for affine $\tilde B_n$, $\tilde C_n$ with three-class weights, $\tilde F_4$, and $\tilde G_2$, or an explicit counterexample.
Baselines
The naive reference point is rational, generic homology. It computes ranks of $M$ over the fraction field $\mathbb{Q}(q,t)$, or equivalently at a random numerical specialization. It is cheap and always correct about the free part, but by construction it reports no torsion. Beating it means exhibiting torsion it misses. The importance of that torsion is the lesson of Homotopy classification of 4–manifolds whosefundamental group is dihedral, where whether a classification holds is decided entirely by torsion in a derived module. We run it on identical complexes as a sanity check: our generic ranks must match it exactly.
The strongest existing method is the one-variable precise-matching computation of Weighted sheaves and homology of Artin groups, applied on the diagonal $t=q$. It produced the first $\tilde C_n$ twisted homology and recovered $A_n$, $B_n$, $\tilde A_n$. No reported result was found on Lacuna giving a quantitative figure (such as runtime or largest rank solved) for this method. The comparison is therefore qualitative and exact. On the diagonal, our outputs must coincide with the published one-variable results. Our method beats this baseline if it produces, for the same groups, torsion on off-diagonal slopes, or pseudo-null torsion, that the diagonal computation cannot express.
The ground-truth baseline is a direct two-variable module computation (Gröbner-basis presentation of homology) on the full, unreduced Salvetti complex. It runs on the same machine under the same wall-clock and memory budget. For small ranks it gives the complete answer. Agreement with it on every small case validates our decomposition. The crossover rank at which it exceeds the budget while our pipeline finishes measures practical gain.
The key ablation removes precise matchings. We keep the slope decomposition but compute each specialization by Smith normal form on the unreduced complex, or with an arbitrary non-precise acyclic matching. This isolates whether the precise matching, rather than the slope decomposition alone, is what makes the computation tractable and the $E^2$ collapse exact.
Why not just…?
Why not just run a Gröbner-basis computation on the whole two-variable complex? For small cases this is the right answer, and we use it as ground truth. It is a real threat if the Salvetti complexes of the target groups stay small enough for direct computation at every rank of interest. The crossover experiment settles this. If direct computation keeps up through $B_8$ and $\tilde C_6$, the computational claim weakens to a structural one: the cyclotomic-slope description and the pseudo-null analysis still say why the answer looks as it does, which a raw module presentation does not.
Why not just compute the one-variable specializations independently and stitch them together? This is nearly our method, minus the localization argument. Without that argument one does not know which finitely many slopes suffice. More importantly, it cannot see pseudo-null torsion, which no curve detects. The ablation and the pseudo-null step show directly whether this torsion exists. If it never appears in any target group, independent specialization is enough and we will say so.
Why not just work rationally, or use a classical spectral sequence? Rationalizing removes exactly the torsion that carries the content. Classical spectral sequences for these local systems involve differentials on many pages. Precise matchings were introduced to force collapse at $E^2$, and our ablation measures what is lost when that collapse is not forced.
Proposed Experiments
The benchmark is the family of known computations plus brute-force ground truth. No standard numeric leaderboard exists here, and no reported result was found on Lacuna for timing or size limits.
The first check is consistency. On the diagonal $t=q$, outputs for $B_n$ ($n\le 8$) and $\tilde C_n$ ($n\le 6$) must reproduce the published one-variable results. Generic ranks must match the fraction-field computation. Euler characteristics must match cell counts.
The second check is ground-truth agreement. For $G_2$, $F_4$, $\tilde G_2$, and small $B_n$ and $\tilde B_n$, the slope-plus-correction answer must equal the direct two-variable module computation exactly. We report wall-clock time and memory against the same budget to locate the crossover.
The decisive check is the conjecture itself. For every computed case we verify that all supporting curves are cyclotomic in a monomial $q^a t^b$ and that a precise matching exists along each one. We record whether pseudo-null torsion appears and at which roots of unity.
A single non-cyclotomic support curve would falsify the conjecture. So would a slope admitting no precise matching. Either outcome would be informative, since it would mark exactly where the single-variable theory's structure fails to extend. Conversely, consistent pseudo-null vanishing across the affine families would turn the computation into a uniform closed form. For the affine types, which are $K(\pi,1)$ by Paolini–Salvetti, this would be a statement about the group homology of Artin groups with two-parameter coefficients.
References
- Weighted sheaves and homology of Artin groups
- Proof of the $$K(\pi , 1)$$ conjecture for affine Artin groups
- Rate-Dependent Analysis of the Asymptotic Behavior of Channel Polarization
- General Strong Polarization
- Measurements of Parity-Violating Asymmetries in Electron-Deuteron Scattering in the Nucleon Resonance Region
- On the classifying space of Artin monoids
- Equivariant K-homology for hyperbolic reflection groups
- Homotopy classification of 4–manifolds whosefundamental group is dihedral