# Charged Diffusive-Precession State Space

- ID: 1210
- Canonical URL: https://synthcore.org/idea/1210/charged-diffusive-precession-state-space
- API JSON: https://synthcore.org/api/idea/1210.json
- API Markdown: https://synthcore.org/api/idea/1210.md
- Verification status: unverified
- Source: [arXiv:2607.20991](https://arxiv.org/abs/2607.20991)
- Category: dynamics
- Solves: stability, accuracy, sample-efficiency
- ML areas: ssm, rnn, world-model, training-dynamics
- Math tags: dynamical-systems, pde, spectral-theory, linear-algebra
- Ratings: usefulness 5/10; difficulty 5/10; novelty 5/10

## Idea description

Replace part of a sequence or spatiotemporal model's unconstrained recurrence with a bank of stable second-order filters whose poles are a frequency-shifted precession pole and a diffusion pole. The chemical-potential parameter produces oscillatory memory, while the diffusion parameter produces scale-dependent decay; a learned residual branch preserves expressivity when the prior is imperfect.

## Mathematical statement

The paper's charged longitudinal correlator is \(\Pi^{\parallel,\pm}(\omega,\vec{k})=-\frac{i\Sigma\,\omega((\omega\pm\mu_{3})^{2}-\vec{k}^{2})}{(\omega\pm\mu_{3}+iD\vec{k}^{2})(\omega\pm\mu_{3})}\), with \(\mu_3\) the isospin chemical potential, \(D\ge 0\) the diffusion constant, \(\Sigma\) the flavor conductivity, \(\omega\) temporal frequency, and \(\vec{k}\) spatial wavevector. The denominator has poles at \(\omega=-\mu_3\) and \(\omega=-\mu_3-iD\lvert\vec{k}\rvert^2\); with the Fourier convention \(e^{-i\omega t}\), these correspond to an oscillatory mode at angular frequency \(\mu_3\) and a mode with decay rate \(D\lvert\vec{k}\rvert^2\). The transverse response \(\Pi^{\perp,\pm}(\omega,\vec{k})=-i\Sigma\omega\) supplies a direct high-frequency/residual channel. We use only the pole structure, enforce \(D\ge0\) by softplus, and discretize the two continuous-time modes exactly over step \(\Delta t\): \(z_1=e^{(-i\mu_3)\Delta t}\) and \(z_2(\vec{k})=e^{(-i\mu_3-D\lvert\vec{k}\rvert^2)\Delta t}\).

## Key formulas

- $$\Pi^{\parallel,\pm}(\omega,\vec{k})=-\frac{i\Sigma\,\omega((\omega\pm\mu_{3})^{2}-\vec{k}^{2})}{(\omega\pm\mu_{3}+iD\vec{k}^{2})(\omega\pm\mu_{3})}$$
- $$\Pi^{\perp,\pm}(\omega,\vec{k})=-i\Sigma\,\omega$$
- $$z_1=e^{-i\mu_3\Delta t},\qquad z_2(\vec{k})=e^{(-i\mu_3-D\lVert\vec{k}\rVert^2)\Delta t},\qquad D=\operatorname{softplus}(\theta_D)$$
- $$h_{t+1}^{(1)}(\vec{k})=z_1h_t^{(1)}(\vec{k})+b_1x_t(\vec{k}),\qquad h_{t+1}^{(2)}(\vec{k})=z_2(\vec{k})h_t^{(2)}(\vec{k})+b_2h_t^{(1)}(\vec{k}),\qquad y_t(\vec{k})=W_1h_t^{(1)}(\vec{k})+W_2h_t^{(2)}(\vec{k})+W_rx_t(\vec{k})$$

## Implementation notes

Integrate this as a replaceable mixer inside a small SSM, RNN, or spatiotemporal Transformer block. For an image, video, or grid sequence, FFT the input along spatial axes; for a plain 1D sequence, use a bank of learned frequencies \(\mu_{3,j}\) and omit \(\lVert\vec{k}\rVert^2\), or assign each bank element a fixed scale \(q_j\). For each channel group and frequency, maintain complex states \(h^{(1)},h^{(2)}\). Initialize \(\mu_{3,j}=\operatorname{softplus}(\theta_{\mu,j})\), \(D_j=\operatorname{softplus}(\theta_{D,j})\), compute \(z_1=\exp(-i\mu_{3,j}\Delta t)\) and \(z_2=\exp((-i\mu_{3,j}-D_jq^2)\Delta t)\), then execute the two recurrences in the displayed formula. Apply a learned real projection after inverse FFT and add a gated residual from the original block. Use both signs of \(\mu_3\) by pairing conjugate channels, guaranteeing real outputs. The paper supplies the pole locations and the stability condition \(D\ge0\); the network learns \(\Sigma\)-like gains, input/output matrices, and the mixture with the residual branch. Estimate \(q^2\) exactly from FFT frequencies, or use fixed logarithmically spaced scales for 1D data. First test on Copy, Adding, and Long Range Arena tasks using a 100k-token context, comparing equal-width S4/Mamba-style diagonal SSMs and a GRU. Track validation loss versus context length, gradient norm, and recurrent rollout stability. A successful signal is lower long-context loss or fewer divergence events at equal parameters and FLOPs; on video prediction, measure multistep error and whether the learned \(D\) orders low spatial frequencies as slowly decaying modes.

## Disclaimer

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