# Hodge-Selective Edge Dynamics

- ID: 1957
- Canonical URL: https://synthcore.org/idea/1957/hodge-selective-edge-dynamics
- API JSON: https://synthcore.org/api/idea/1957.json
- API Markdown: https://synthcore.org/api/idea/1957.md
- Verification status: mech_ok_no_baseline
- Source: [arXiv:2608.07768](https://arxiv.org/abs/2608.07768)
- Category: architecture
- Solves: stability, accuracy, sample-efficiency
- ML areas: graph-nn, ssm, training-dynamics
- Math tags: graph-theory, linear-algebra, dynamical-systems, control-theory
- Ratings: usefulness 7/10; difficulty 5/10; novelty 7/10

## Idea description

Replace an unconstrained edge-feature residual update in a graph neural network with separate cut-space and harmonic-space updates. The cut branch carries transfer information visible at nodes, while the harmonic branch models cycle circulation and can be given an independently chosen contraction rate, preventing persistent or unstable circulation features from contaminating node predictions.

## Mathematical statement

For a connected oriented graph with m edges and n nodes, let B in R^{n x m} be its signed node-edge incidence matrix and let f in R^m be an edge-feature vector. The paper uses the orthogonal Hodge decomposition R^m = im(B^T) direct-sum ker(B), where im(B^T) is the cut space and ker(B) is the harmonic cycle space. With the Euclidean inner product, the projections are P_C = B^T (B B^T)^dagger B and P_H = I_m - P_C, where (.)^dagger is the Moore-Penrose pseudoinverse, P_C f is the divergence-carrying component, and P_H f is the cycle-flow component. The ideal selective controller is dot(f) = -alpha P_C(f - f_*) - kappa P_H f, with alpha,kappa > 0 and f_* in im(B^T). Its exact solution is f_C(t) = f_* + exp(-alpha t)(f_C(0)-f_*) and f_H(t) = exp(-kappa t)f_H(0), so the two subspaces have independently controlled decay rates. A graph layer can transplant this as a discrete residual map f_{l+1} = f_* + (1-alpha_l)P_C(f_l-f_*) + (1-kappa_l)P_H f_l, with 0 <= alpha_l,kappa_l <= 1 for non-expansive one-step behavior. The key assumption is a fixed connected graph; for disconnected graphs, compute the pseudoinverse componentwise.

## Key formulas

- $$\mathbb{R}^{m}=\operatorname{im}B^{\top}\oplus\operatorname{ker}B$$
- $$P_C=B^{\top}(BB^{\top})^{\dagger}B,\qquad P_H=I_m-P_C$$
- $$\dot{f}=-\alpha P_C(f-f_*)-\kappa P_Hf$$
- $$f_{l+1}=f_*+(1-\alpha_l)P_C(f_l-f_*)+(1-\kappa_l)P_Hf_l$$

## Implementation notes

Integrate this into an edge-state graph neural network immediately before or after the ordinary edge message/update MLP. Inputs are an incidence matrix B, edge features F in R^{m x d}, and optionally a learned or data-derived cut target F_*; apply the same projection independently to each feature channel. First construct the sparse graph Laplacian L = B B^T. For small graphs, compute P_C explicitly using a pseudoinverse of L; for larger graphs, avoid materializing it and compute P_C X by solving L Y = B X with a pinned node or zero-mean constraint, then returning B^T Y. Compute P_H X = X - P_C X. A layer should execute: (1) compute ordinary edge message Z = MLP([F, edge_attributes, node_messages]); (2) split Z_C=P_C Z and Z_H=P_H Z; (3) update F_new = F_* + (1-alpha) Z_C + (1-kappa) Z_H, or use learned residual branches F_new = F + g_C(Z_C)+g_H(Z_H) followed by scalar gates constrained as alpha=sigmoid(a), kappa=sigmoid(b); (4) optionally feed B F_new to the node update. The paper's exact contribution is the orthogonal decomposition and independent contraction law; the MLP transformations, gates, and target F_* are engineering choices. Estimate projection correctness with ||P_C^2-P_C|| and ||P_C P_H||, and monitor separate norms ||P_C F|| and ||P_H F||. The first experiment should use synthetic ring, grid, and random connected graphs on a graph signal forecasting task, comparing an ordinary edge-GNN residual block, a single projected block, and the proposed two-rate block at matched parameters. Measure long-horizon rollout error, gradient norm, and node prediction accuracy after injecting a large cycle-flow perturbation. Success means lower rollout error and faster recovery from harmonic perturbations without degrading cut-space accuracy.

## Verification

- Status: mech_ok_no_baseline
- Mechanism evidence: yes
- Mechanism confirmed: yes
- Practical verdict: harms
- Verdict: Built a reproducible Hodge-selective edge-dynamics MVP with explicit incidence-based cut and harmonic projectors. Projector identities and divergence annihilation were verified at approximately 1e-15 error; independent-rate decay matched |1-alpha|^k and |1-kappa|^k predictions within 2e-15, and the stability boundary at rate 2 was confirmed. In the perturbation toy task, selective damping removed harmonic error rapidly while preserving a slower cut rate, but a shared fast baseline achieved lower total recovery error, so the mechanism is validated without evidence of a general accuracy win.

### Mechanism check

- Verdict: Built a reproducible Hodge-selective edge-dynamics MVP with explicit incidence-based cut and harmonic projectors. Projector identities and divergence annihilation were verified at approximately 1e-15 error; independent-rate decay matched |1-alpha|^k and |1-kappa|^k predictions within 2e-15, and the stability boundary at rate 2 was confirmed. In the perturbation toy task, selective damping removed harmonic error rapidly while preserving a slower cut rate, but a shared fast baseline achieved lower total recovery error, so the mechanism is validated without evidence of a general accuracy win.
- Confidence: 9/10
- Limitations: Only a small 8-node ring and synthetic linear recovery were tested; no learned MLP edge-GNN, grids or random graphs, disconnected-graph handling, training comparison, long-horizon forecasting benchmark, FLOP matching, or multi-seed statistical evaluation was performed.

### Practical benchmark

- Paired seeds: 8
- Baseline mean: 0
- Idea mean: 0
- p-value: 0.0444
- Paired wins: 2/8
- Benchmark verdict: idea worse (significant)

## Artifacts

- [bench_report.json](https://synthcore.org/code/748/bench_report.json)
- [hodge_experiment.py](https://synthcore.org/code/748/hodge_experiment.py)
- [report.md](https://synthcore.org/code/748/report.md)
- [report_bench_2026-09-01T022033.md](https://synthcore.org/code/748/report_bench_2026-09-01T022033.md)
- [results.json](https://synthcore.org/code/748/results.json)
- [stage2_hodge_bench.py](https://synthcore.org/code/748/stage2_hodge_bench.py)
- [Download all files as ZIP](https://synthcore.org/download/748)

## Disclaimer

AI-generated research hypothesis, automatically tested. Not peer-reviewed.