# Minimal-Channel Harmonic Controller

- ID: 1958
- Canonical URL: https://synthcore.org/idea/1958/minimal-channel-harmonic-controller
- API JSON: https://synthcore.org/api/idea/1958.json
- API Markdown: https://synthcore.org/api/idea/1958.md
- Verification status: unverified
- Source: [arXiv:2608.07768](https://arxiv.org/abs/2608.07768)
- Category: memory
- Solves: memory, scalability, stability
- ML areas: graph-nn, ssm, memory, inference-speedup
- Math tags: graph-theory, linear-algebra, topology, control-theory
- Ratings: usefulness 6/10; difficulty 6/10; novelty 8/10

## Idea description

Use a low-rank controller that observes and actuates only the graph's harmonic coordinates rather than all edge features. For a graph with first Betti number beta_1 = dim ker(B), a beta_1-dimensional cycle basis is sufficient to represent the entire harmonic sector, yielding a compact recurrent memory or adapter for circulation-dependent graph dynamics.

## Mathematical statement

Let H in R^{m x beta_1} have orthonormal columns spanning ker(B), where beta_1 = dim ker(B) = m - rank(B) for the connected graph considered here. Then P_H = H H^T and every harmonic edge state is f_H = H z with coordinates z = H^T f. The ideal harmonic damping term -kappa P_H f becomes -kappa H z, and the harmonic coordinate dynamics are dot(z) = -kappa z. A finite channel implementation uses an actuator/sensor matrix A in R^{m x r}; exact harmonic compression requires rank(H^T A)=beta_1, hence r >= beta_1. The paper emphasizes that this rank condition alone does not guarantee local decoupling: if A is not aligned with H, channel feedback can produce cut-harmonic spillover. A practical learned controller is therefore f_{t+1}=f_t + A u_t, u_t=-eta H^T A^T H H^T f_t, with the normalized harmonic Gram matrix G=A^T H H^T A used to precondition channels. For exact aligned control A=H, this simplifies to z_{t+1}=(1-eta)z_t and leaves the cut component unchanged.

## Key formulas

- $$\beta_1=\dim\ker(B),\qquad P_H=HH^{\top},\qquad f_H=HH^{\top}f=Hz,\quad z=H^{\top}f$$
- $$\dot{f}=-\kappa P_Hf\quad\Longleftrightarrow\quad\dot{z}=-\kappa z$$
- $$\operatorname{rank}(H^{\top}A)=\beta_1\quad\text{and therefore}\quad r\geq\beta_1$$
- $$f_{t+1}=f_t+A u_t,\qquad u_t=-\eta H^{\top}A^{\top}H H^{\top}f_t$$

## Implementation notes

Add a compact harmonic-state controller to a recurrent graph neural network, graph state-space model, or edge-feature adapter. Preprocessing takes the fixed incidence matrix B and computes an orthonormal basis H for ker(B): use sparse QR/SVD for small graphs, or obtain beta_1 cycle vectors with a spanning-tree fundamental-cycle construction followed by QR orthogonalization. At each time step, read the edge state F and compute harmonic coordinates Z=H^T F. Pass Z through a small recurrent cell or linear controller, then inject the result through A: U=Controller(Z), F_new=F+A U. For the simplest falsifiable version, set A=H and U=-eta Z, giving F_new=F-eta H H^T F; constrain eta to (0,1) with eta=sigmoid(raw_eta) so harmonic energy contracts monotonically. For a localized-device version, choose A as r selected physical edge columns, with r=beta_1 and optimize the selected edges or a learned sparse A. Before training, calculate the rank and condition number of H^T A; reject channel sets with rank deficiency. During training, log harmonic energy E_H=||H^T F||_F^2, cut energy E_C=||F-HH^T F||_F^2, and spillover ratio ||P_C A U||_F/||A U||_F. The mathematics determines the minimum rank and ideal contraction; the recurrent cell and channel selection are empirical additions. First test on temporal node/edge forecasting over ring, mesh, and IEEE-style graphs, comparing full edge-state recurrence, ordinary low-rank adapters, and the beta_1-channel controller at equal hidden size. Inject unobserved cycle perturbations at inference. Success is lower memory and parameter count with stable long-horizon predictions, harmonic energy decay at the prescribed rate, and no significant increase in cut-space forecasting error.

## Disclaimer

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