# Schur torsion-filter feature layer

- ID: 2511
- Canonical URL: https://synthcore.org/idea/2511/schur-torsion-filter-feature-layer
- API JSON: https://synthcore.org/api/idea/2511.json
- API Markdown: https://synthcore.org/api/idea/2511.md
- Verification status: unverified
- Source: [arXiv:2608.18302](https://arxiv.org/abs/2608.18302)
- Category: architecture
- Solves: accuracy, sample-efficiency, stability
- ML areas: embedding, graph-nn, attention
- Math tags: representation-theory, algebra, linear-algebra, harmonic-analysis
- Ratings: usefulness 5/10; difficulty 6/10; novelty 6/10

## Idea description

Add a deterministic feature layer that evaluates symmetric Schur-type features on a fixed cyclic orbit and learned reciprocal latent pairs, then projects the resulting channels onto selected residue classes with an exact roots-of-unity filter. The reciprocal construction makes the layer invariant under replacing each latent scalar by its inverse, while the torsion projector prevents leakage between cyclic frequency sectors.

## Mathematical statement

The paper defines N=t+2r, the root-of-unity orbit mu_t={1,zeta,...,zeta^(t-1)} with zeta=exp(2 pi i/t), and reciprocal pairs z_a,z_a^(-1) for a=1,...,r. For a strictly decreasing integer vector beta=(beta_1,...,beta_N) normalized by beta_N=0, let delta=(N-1,N-2,...,0) and lambda=beta-delta; then lambda is a partition and the Schur feature is Phi_{t,r}(beta)=s_lambda(mu_t,z_1,z_1^(-1),...,z_r,z_r^(-1)). The determinantal definition is Phi_{t,r}(beta)=det(x_i^(beta_j))/det(x_i^(delta_j)), where x=(1,zeta,...,zeta^(t-1),z_1,z_1^(-1),...,z_r,z_r^(-1)). The adaptation uses the exact cyclic projector P_q[f]=(1/t) sum_{k=0}^{t-1} zeta^(-qk) f(zeta^k), where q is the retained residue class and f is a polynomial feature; if f(u)=sum_m a_m u^m, then P_q[f]=sum_{m congruent to q modulo t} a_m. Reciprocal pairs are parameterized as z_a=exp(u_a) and z_a^(-1)=exp(-u_a), with real outputs obtained from conjugate-symmetric combinations.

## Key formulas

- $$\Phi_{t,r}(\beta)=\frac{\det\bigl(x_i^{\beta_j}\bigr)_{i,j=1}^{N}}{\det\bigl(x_i^{\delta_j}\bigr)_{i,j=1}^{N}},\qquad x=(1,\zeta,\dots,\zeta^{t-1},z_1,z_1^{-1},\dots,z_r,z_r^{-1}),\quad \zeta=e^{2\pi i/t}.$$
- $$P_q[f]=\frac{1}{t}\sum_{k=0}^{t-1}\zeta^{-qk}f(\zeta^k),\qquad f(u)=\sum_m a_m u^m\ \Longrightarrow\ P_q[f]=\sum_{m\equiv q\ (\mathrm{mod}\ t)}a_m.$$
- $$z_a=e^{u_a},\qquad z_a^{-1}=e^{-u_a},\qquad \widetilde{\Phi}_{q,\beta}(u)=\operatorname{Re}\left[\frac{1}{t}\sum_{k=0}^{t-1}\zeta^{-qk}\,s_\lambda(\zeta^k,e^{u_1},e^{-u_1},\ldots,e^{u_r},e^{-u_r})\right].$$

## Implementation notes

(1) Integration point: place the layer after an embedding or graph/time-series encoder, before the MLP head or attention Q/K projection. For each token or node, map the real hidden vector h to r scalars u_a using a learned linear projection W. Use z_a=exp(alpha tanh(u_a)) with a small fixed alpha, so reciprocal values remain numerically bounded. Choose a small torsion order, initially t=4 or t=8, and retain one or several sectors q. (2) Pseudocode: `u = alpha*tanh(W*h); x = [roots_of_unity(t), exp(u), exp(-u)]; for beta in beta_set: numerator=det(x[i]**beta[j]); denominator=det(x[i]**delta[j]); phi[beta]=numerator/(denominator+eps); for q in sectors: y[q,beta]=real(sum_k exp(-2*pi*i*q*k/t)*eval_schur_at_root_k[k,beta])/t; output=concat(h,y)`. In practice compute Schur polynomials with a stable symmetric-polynomial recurrence or batched complex LU rather than explicit powers for large beta. (3) The roots-of-unity projector and reciprocal parameterization are exact consequences of the construction; W, the scalars u, selected partitions lambda, and sector weights are learned. Monitor the Vandermonde denominator and add a penalty or clipping rule when its magnitude is below a threshold. (4) First experiment: train a small transformer on synthetic length-64 sequences generated by mixtures of periodic components with periods 4, 8, and 16, comparing a baseline embedding/MLP against the same model plus this layer at equal parameter count. Evaluate active-period classification and extrapolation to unseen phase shifts. Success means lower validation error, better phase-shift generalization, and exact suppression of deliberately irrelevant residue sectors. A second test uses node classification on cycle graphs, measuring accuracy and output variance under cyclic relabeling.

## Disclaimer

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