# Delay-Aware Frequency-Preserving Recurrent Coupling

- ID: 1128
- Canonical URL: https://synthcore.org/idea/1128/delay-aware-frequency-preserving-recurrent-coupling
- API JSON: https://synthcore.org/api/idea/1128.json
- API Markdown: https://synthcore.org/api/idea/1128.md
- Verification status: mech_ok_no_baseline
- Source: [arXiv:2607.18913](https://arxiv.org/abs/2607.18913)
- Category: architecture
- Solves: stability, accuracy
- ML areas: rnn, ssm, training-dynamics, world-model
- Math tags: dynamical-systems, control-theory, spectral-theory, linear-algebra, harmonic-analysis
- Ratings: usefulness 7/10; difficulty 6/10; novelty 7/10

## Idea description

For coupled recurrent or state-space modules that represent oscillatory or periodic signals, explicitly account for communication or attention delay in the characteristic equation. Tune the coupling gain or add a phase-lead compensator so that the desired latent frequency remains a closed-loop mode instead of being shifted by small delays.

## Mathematical statement

Let \(x_i(t)\in\mathbb{R}^d\) be the state of module \(i\), with local dynamics \(A\), coupling matrices \(B,K\), graph Laplacian \(L\), and transmission delay \(\tau\). The delayed coupled system is \(\dot x_i(t)=Ax_i(t)-BK\sum_jL_{ij}x_j(t-\tau)\). After diagonalizing \(L\), each graph mode with eigenvalue \(\lambda_j\) has characteristic equation \(\det[sI-A+\lambda_jBK e^{-s\tau}]=0\). The consensus mode \(\lambda_1=0\) is delay-independent, but nonzero modes acquire a frequency-dependent phase factor. At a target oscillation \(s=\mathrm{i}\omega_0\), delay changes the coupling phase by \(-\omega_0\tau\), so periodic synchronization is generally lost. A phase-lead compensator \(C(s)\) should satisfy \(C(\mathrm{i}\omega_0)\approx e^{\mathrm{i}\omega_0\tau}\).

## Key formulas

- $$\dot x_i(t)=Ax_i(t)-BK\sum_{j=1}^N L_{ij}x_j(t-\tau),\qquad x_i\in\mathbb{R}^d.$$
- $$\det\!\left[sI-A+\lambda_jBK e^{-s\tau}\right]=0,\qquad j=1,\ldots,N.$$
- $$e^{-\mathrm{i}\omega_0\tau}=\cos(\omega_0\tau)-\mathrm{i}\sin(\omega_0\tau).$$
- $$\det\!\left[\mathrm{i}\omega_0I-A+\lambda_jBK C(\mathrm{i}\omega_0)e^{-\mathrm{i}\omega_0\tau}\right]=0,\qquad C(\mathrm{i}\omega_0)\approx e^{\mathrm{i}\omega_0\tau}.$$

## Implementation notes

Integrate this into a recurrent or state-space architecture whose hidden state is updated in several coupled branches, for example four diagonal SSM blocks or four small GRU-like modules connected by a graph. Use a continuous-time-inspired Euler update `x_i[t+1] = x_i[t] + h*(A*x_i[t] - B*K*sum_j L[i,j]*x_j[t-d])`, where \(h\) is the integration step and \(d\) is an integer delay. First choose \(A\) with a lightly damped complex pair at target frequency \(\omega_0\), and train only a readout on synthetic mixtures of sinusoids so the effect of coupling is isolated. Implement two variants: ordinary delayed coupling and phase-compensated coupling using a two-tap predictor \(\widehat{x}_j[t+d]=x_j[t]+d(x_j[t]-x_j[t-1])\), or a learnable FIR filter constrained to have frequency response near \(e^{\mathrm{i}\omega_0\tau}\). The paper contributes the modal delay characteristic equation and the distinction between delay-invariant zero-frequency consensus and delay-sensitive periodic modes. Estimate \(A,B,K\) from module weights and estimate dominant roots by scanning complex \(s=\alpha+\mathrm{i}\omega\) for minima of the characteristic determinant; measure the output spectrum with an FFT. Sweep \(\tau\), coupling gain \(\gamma\), and target frequency. The falsifiable prediction is a loss of phase locking or a dominant-frequency shift once \(\omega_0\tau\) becomes appreciable, with instability beginning when a characteristic root reaches \(\operatorname{Re}s=0\). The compensated model should preserve peak frequency within 5 percent up to delays for which the uncompensated model exhibits a measurable shift, while constant inputs should remain nearly delay-invariant.

## Verification

- Status: mech_ok_no_baseline
- Mechanism evidence: yes
- Mechanism confirmed: yes
- Practical verdict: no_effect
- Verdict: Built and ran a delayed modal recurrent-coupling MVP with ordinary delayed feedback and a two-tap phase-lead predictor. The three mechanism predictions were confirmed: delay phase slope was -1.000 versus -1.000 predicted, the lambda=0 mode was exactly delay-invariant while a nonzero mode changed by relative RMS 0.825, and the small-gain frequency-shift slope differed from perturbation theory by 3.9%. In the four-branch ring toy experiment, compensation reduced mean nonzero-mode frequency error from 0.03425 rad/time to 0.000563 rad/time, a 98.4% reduction; this is a real signal for the stated mechanism, though not evidence of a general ML-training win.

### Mechanism check

- Verdict: Built and ran a delayed modal recurrent-coupling MVP with ordinary delayed feedback and a two-tap phase-lead predictor. The three mechanism predictions were confirmed: delay phase slope was -1.000 versus -1.000 predicted, the lambda=0 mode was exactly delay-invariant while a nonzero mode changed by relative RMS 0.825, and the small-gain frequency-shift slope differed from perturbation theory by 3.9%. In the four-branch ring toy experiment, compensation reduced mean nonzero-mode frequency error from 0.03425 rad/time to 0.000563 rad/time, a 98.4% reduction; this is a real signal for the stated mechanism, though not evidence of a general ML-training win.
- Confidence: 9/10
- Limitations: Only a deterministic complex single-mode Euler DDE and a four-branch ring modal equivalent were tested. No learned GRU/SSM, synthetic readout training, multi-frequency mixture, broad delay/gain stability-boundary sweep, wall-clock comparison, or GPU experiment was included; FFT output-spectrum evaluation was replaced by a continuous phase-slope estimator.

### Practical benchmark

- Paired seeds: 8
- Baseline mean: 0
- Idea mean: 0
- p-value: 1
- Paired wins: 0/8
- Benchmark verdict: no measurable effect

## Artifacts

- [bench_report.json](https://synthcore.org/code/450/bench_report.json)
- [delay_bench.py](https://synthcore.org/code/450/delay_bench.py)
- [delay_coupling_experiment.py](https://synthcore.org/code/450/delay_coupling_experiment.py)
- [report.md](https://synthcore.org/code/450/report.md)
- [report_bench_2026-08-31T123143.md](https://synthcore.org/code/450/report_bench_2026-08-31T123143.md)
- [results.json](https://synthcore.org/code/450/results.json)
- [Download all files as ZIP](https://synthcore.org/download/450)

## Disclaimer

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