{
 "artifacts": null,
 "category": "architecture",
 "description": "Replace an unconstrained Cartesian-product router over heterogeneous branches with a router whose joint expert or state assignments obey a finite-group conservation rule. Branch i emits a distribution over labels in its own subgroup H_i of a common finite abelian group G; only tuples whose group sum is zero are retained. This gives an exact, differentiable structural prior for modular arithmetic, multi-relational graphs, multi-view fusion, or any setting where latent labels compose by a conservation law.",
 "formulas_latex": [
  "$$\\boxed{\\mathcal{A}(G;H_1,\\dots,H_l)=\\{(g_1,\\dots,g_l):g_i\\in H_i,\\ \\textstyle\\sum_{i=1}^l g_i=0\\}}$$",
  "$$q(g_1,\\ldots,g_l)=\\frac{\\mathbf{1}[g_i\\in H_i\\ \\forall i]\\\\,\\mathbf{1}[\\sum_i g_i=0]\\\\,\\prod_{i=1}^l p_i(g_i)}{Z},\\qquad Z=\\sum_{(h_1,\\ldots,h_l)\\in\\mathcal{A}(G;H_1,\\ldots,H_l)}\\prod_{i=1}^l p_i(h_i)$$",
  "$$y=\\sum_{(g_1,\\ldots,g_l)\\in\\mathcal{A}(G;H_1,\\ldots,H_l)}q(g_1,\\ldots,g_l)\\,e_{g_1,\\ldots,g_l}$$",
  "$$\\mathcal{A}(\\mathbb{Z}_m;H_1,\\ldots,H_l)=\\{(g_1,\\ldots,g_l):g_i\\in H_i,\\ \\sum_i g_i\\equiv0\\pmod m\\}$$"
 ],
 "id": 3073,
 "implementation": "Integrate the layer immediately before MoE expert selection or multimodal or relational feature fusion. Assume l input branch tensors x_i with shape [batch, hidden]. Choose a small finite abelian group, initially G=Z_4 or Z_8, and define subgroup masks H_i. For Z_m, use the subgroup of size d_i dividing m, represented by {0,m/d_i,2m/d_i,...,(d_i-1)m/d_i}. Each branch has a linear classifier producing logits a_i over H_i, followed by p_i=softmax(a_i). Enumerate valid tuples A once at initialization. For each batch item, compute log weights s(g_1,...,g_l)=sum_i log p_i(g_i), set s=-infinity outside A, and obtain q=softmax(s over A). Associate every valid tuple with either an expert index or a learned code e_tuple. Compute y=sum_tuple q_tuple e_tuple, then use y to gate experts or add it to the fused hidden representation. Pseudocode: logits_i=Linear_i(x_i); p_i=softmax(logits_i); scores=stacked_sum(log(p_i) indexed by A); q=softmax(scores); y=q @ E; output=MoE(x, gate=y) or LayerNorm(concat(x_i)+Proj(y)). For large l, avoid full Cartesian enumeration by dynamically programming the modular sum: repeatedly convolve branch distributions on G, then recover tuple samples with a constrained beam or compute exact marginal probabilities using the probability that the remaining branches cancel the current label. The paper supplies the exact support constraint; all neural logits, code vectors, and expert weights are learned. First test on a synthetic compositional classification task where three branches contain Z_4 labels and the target is their zero-sum class, comparing an unconstrained Cartesian router, a masked router with H_i=G, and mixed subgroups such as sizes (4,2,2). Then test a small relational GNN on modular addition or colored-edge composition. Measure validation accuracy at fixed samples, number of active joint states, routing entropy, and wall-clock cost. Success means faster learning and higher accuracy when the conservation rule is correct, with no degradation on a control task using randomized labels.",
 "math_summary": "The paper defines the mixed-order configuration $\\mathcal{A}(G;H_1,\\ldots,H_l)=\\{(g_1,\\ldots,g_l):g_i\\in H_i,\\ \\sum_{i=1}^l g_i=0\\}$, where G is a finite abelian group written additively, H_i is the subgroup available to branch i, l is the number of branches, and equality is in G. The ordinary group-based model is the special case H_i=G for every i; heterogeneous cone orders correspond to proper subgroups on different leaves. Implement the same configuration as a masked product distribution: branch i produces probabilities p_i(g_i) for g_i in H_i, and the joint probability is proportional to the product of branch probabilities on valid tuples and zero elsewhere. The conservation constraint is exact, while gradients remain available through normalization. For G=Z_m, membership and the constraint are integer modular arithmetic: g_i is in {0,...,m-1} and the sum is congruent to zero modulo m.",
 "math_tags": [
  "algebra",
  "combinatorics",
  "representation-theory"
 ],
 "ml_areas": [
  "moe-routing",
  "embedding",
  "graph-nn",
  "transformer"
 ],
 "paper": {
  "arxiv_id": "2609.02630",
  "arxiv_url": "https://arxiv.org/abs/2609.02630",
  "summary_what_math_gives_to_ml": "The paper identifies orbifold boundary-condition classes with affine semigroups of finite-group label configurations satisfying a local conservation law: each branch label lies in a prescribed subgroup and the labels sum to zero. The transferable asset is not the orbifold application itself, but a principled mixed-domain constraint that couples heterogeneous categorical variables while retaining an explicitly enumerable configuration set. This can be turned into a differentiable constrained fusion or MoE-routing layer whose support contains only valid tuples, reducing the joint state space and enforcing an exact symmetry or conservation invariant. The main risk is inductive-bias mismatch, so the construction should be tested on tasks with genuine compositional or modular structure rather than imposed universally.",
  "title": "An Affine Semigroup from Orbifold Boundary Conditions: cut, phylogenetic and hierarchical models in the unit-weight sector, and weighted configurations beyond them",
  "year": "2026"
 },
 "ratings": {
  "difficulty": 5,
  "novelty": 7,
  "usefulness": 5
 },
 "solves": [
  "accuracy",
  "sample-efficiency",
  "scalability"
 ],
 "title": "Mixed-Subgroup Conservation Router",
 "url": "https://synthcore.org/idea/3073/mixed-subgroup-conservation-router",
 "verification": {
  "peer_reviewed": false,
  "status": "unverified",
  "status_label": "Unverified",
  "verdict_source": "deterministic test code (paired-seed permutation statistics)",
  "verification_axes": {
   "benchmark_mechanism": {
    "confirmed": null,
    "tested": false
   },
   "practical_benchmark": {
    "beats_baseline": null,
    "tested": false
   },
   "toy_mechanism_gate": {
    "confirmed": null,
    "tested": false
   }
  }
 }
}