{
 "artifacts": null,
 "category": "architecture",
 "description": "Replace a local smoothness penalty or local state transition along a sequence or depth coordinate by a marginal fractional quadratic energy with Fourier multiplier |k|. The sigma=1 kernel is nonlocal and scale-free, so it can preserve long-range correlations while suppressing high-frequency instability more selectively than an ordinary Laplacian penalty.",
 "formulas_latex": [
  "$$E_{\\sigma}(x)=\\frac{1}{2L}\\sum_k |k|^{\\sigma}\\|\\widehat{x}_k\\|_2^2,\\qquad \\nabla E_{\\sigma}(x)=\\mathcal{F}^{-1}\\left[|k|^{\\sigma}\\widehat{x}_k\\right],\\quad \\sigma=1.$$",
  "$$x^{+}=x-\\alpha\\mathcal{F}^{-1}\\left[|k|\\widehat{x}_k\\right],\\qquad 0\u003c\\alpha\u003c\\frac{2}{\\lambda_{\\max}},\\quad \\lambda_{\\max}=\\max_k |k|.$$",
  "$$C(r)=\\frac{1}{L-r}\\sum_{j=0}^{L-r-1}\\frac{\\langle x_j\\cdot x_{j+r}\\rangle}{\\langle\\|x_j\\|_2^2\\rangle},\\qquad \\eta=-\\lim_{r\\to\\infty}\\frac{\\log C(r)}{\\log r}.$$",
  "$$\\chi_{\\mathrm{tw}}(L)\\propto L^{-\\eta(T)}.$$"
 ],
 "id": 2991,
 "implementation": "1. Integration point: add the operator to a residual sequence model, SSM, or deep network whose states are indexed by token position or layer depth. For hidden states H with shape [batch, length, channels], compute an FFT over the length dimension, multiply each frequency by |k|, inverse FFT, and use either a residual layer H_new=H+g(MLP(H))-alpha q or a regularizer lambda_frac E_1(H) added to the task loss. Use a real FFT, set the zero-mode multiplier to zero, and normalize frequencies by the Nyquist frequency. 2. Pseudocode: h_hat=rfft(H); q_hat=abs(freq)*h_hat; q=irfft(q_hat); H_new=H+beta*MLP(H)-alpha*q; loss=task_loss+lambda_frac*mean(abs(freq)*abs(h_hat)^2). Choose alpha initially as 0.9*2/lambda_max and optionally estimate the realized Jacobian gain with power iteration. 3. Computed from the paper: the |k| multiplier, fractional energy, algebraic-correlation prediction, and twist-response scaling. Estimated empirically: eta from a linear regression of log C(r) on log r over the middle 50 percent of distances, and the largest effective Jacobian eigenvalue. 4. First cheap experiment: compare a small residual MLP or SSM on Copy, Adding, and Long Range Arena ListOps at lengths 128, 256, 512, and 1024 against no coupling and first- or second-difference penalties. The predicted signature is an approximately length-independent intermediate-distance power law C(r) proportional to r^{-eta}, rather than exponential decay. Sweep alpha and verify that divergence begins near alpha_c=2/lambda_max within 20 percent after accounting for nonlinear gain. Also apply a phase ramp H_j to exp(i theta j/L)H_j and fit the response versus L; the prediction is a power law L^{-eta}, not a size-independent stiffness.",
 "math_summary": "Let x_j in R^d be a feature sequence of length L, let x_hat_k be its discrete Fourier transform, and let lambda_k=|k|^sigma with sigma=1. Define the fractional energy E_sigma(x)=(1/(2L)) sum_k |k|^sigma ||x_hat_k||_2^2. Its gradient is the fractional Laplacian F^{-1}[|k|^sigma x_hat_k]. The paper's marginal mechanism predicts algebraic correlations C(r) proportional to r^{-eta(T)}, where eta(T)=-lim_{r to infinity} log C(r)/log r. For an explicit linear fractional update x_plus=x-alpha F^{-1}[|k| x_hat_k], stability requires 0\u003calpha\u003c2/lambda_max, where lambda_max=max_k |k|. This last bound is the implementation-specific spectral step-size ceiling.",
 "math_tags": [
  "statistical-mechanics",
  "dynamical-systems",
  "harmonic-analysis",
  "spectral-theory"
 ],
 "ml_areas": [
  "ssm",
  "rnn",
  "regularization",
  "training-dynamics"
 ],
 "paper": {
  "arxiv_id": "2609.00721",
  "arxiv_url": "https://arxiv.org/abs/2609.00721",
  "summary_what_math_gives_to_ml": "The paper identifies a nonlocal one-dimensional U(1) field whose marginal fractional dispersion |k| at sigma=1 produces logarithmic fluctuations and BKT-like algebraic correlations without a conventional helicity-modulus jump. The transferable mechanism is a fractional spectral coupling: a quadratic penalty weighted by |k| creates scale-free interactions, while twist responses provide a finite-size diagnostic of the resulting correlation exponent. A concrete neural implementation is a fractional spectral regularizer or residual state-space layer that couples positions, depth, or training-time iterates through the |k| multiplier. The experiment should test whether the predicted power-law correlation and size-scaling signatures appear, and whether they improve long-range sequence stability relative to local first- or second-difference penalties.",
  "title": "BKT-like Correlation Scaling and Twist Responses in a One-Dimensional Fractional $U(1)$ Ginzburg--Landau Model",
  "year": "2026"
 },
 "ratings": {
  "difficulty": 5,
  "novelty": 7,
  "usefulness": 6
 },
 "solves": [
  "stability",
  "accuracy",
  "generalization"
 ],
 "title": "Marginal Fractional Coupling Layer",
 "url": "https://synthcore.org/idea/2991/marginal-fractional-coupling-layer",
 "verification": {
  "peer_reviewed": false,
  "status": "unverified",
  "status_label": "Unverified",
  "verdict_source": "deterministic test code (paired-seed permutation statistics)",
  "verification_axes": {
   "benchmark_mechanism": {
    "confirmed": null,
    "tested": false
   },
   "practical_benchmark": {
    "beats_baseline": null,
    "tested": false
   },
   "toy_mechanism_gate": {
    "confirmed": null,
    "tested": false
   }
  }
 }
}