{
 "artifacts": null,
 "category": "regularization",
 "description": "For a neural network predicting A coupled concentration or density fields, decompose Fourier-space fluctuations into a charge direction and its charge-neutral composition subspace. Hard-project the predicted fields to eliminate the global charge mode, and regularize their low-wavenumber covariance so that neutral modes retain finite susceptibility while the charge structure factor follows the Coulombic suppression S_ZZ(k) proportional to k squared. This should improve long-range physical consistency without suppressing legitimate neutral composition fluctuations.",
 "formulas_latex": [
  "$$C_{ab}(k)=\\left\\langle\\delta c_a(k)\\delta c_b(-k)\\right\\rangle,\\qquad z(k)=q^T\\delta c(k),\\qquad S_{ZZ}(k)=\\left\\langle\\lvert z(k)\\rvert^2\\right\\rangle=q^T C(k)q.$$",
  "$$P_q=I-\\frac{qq^T}{q^Tq},\\qquad \\delta c_{\\mathrm{neutral}}(k)=P_q\\delta c(k),\\qquad q^T\\delta c_{\\mathrm{neutral}}(k)=0.$$",
  "$$S_{ZZ}(k)=\\alpha k^2+o(k^2)\\quad\\text{as }k\\to0,\\qquad C_{\\mathrm{neutral}}(k)=P_qC(k)P_q\\to C_0,$$",
  "$$\\mathcal{L}_{\\mathrm{SF}}=\\sum_{k\\in\\mathcal{K}_{\\mathrm{low}}}w_k\\left[\\log\\left(S_{ZZ}^{\\mathrm{pred}}(k)+\\varepsilon\\right)-\\log\\left(\\alpha k^2+\\varepsilon\\right)\\right]^2+\\lambda_N\\sum_{k\\in\\mathcal{K}_{\\mathrm{low}}}\\left\\|P_q\\widehat C^{\\mathrm{pred}}(k)P_q-P_q\\widehat C^{\\mathrm{data}}(k)P_q\\right\\|_F^2.$$"
 ],
 "id": 2862,
 "implementation": "Use a CNN, U-Net, or MLP that maps a conditioning variable x to A real-space fields c_a(r) on a periodic grid. At every forward pass, subtract the spatial mean, Fourier transform each channel, and apply the charge projection to the k = 0 coefficient: delta c(0) is replaced by P_q delta c(0). If the application requires local electroneutrality, apply the same projection to a chosen set of low-wavenumber modes. For each minibatch, estimate the Fourier covariance as C_hat_ab(k) = B^{-1} times the sum over examples b of delta c_a^(b)(k) delta c_b^(b)(-k). Compute S_ZZ(k) = q^T C_hat(k) q and add the displayed spectral loss to the task loss. Estimate alpha by fitting the empirical training spectra rather than assuming it is known. The projection is exact; covariance estimates and alpha are empirical. First test on a 64 by 64 synthetic two-species electrolyte-field dataset with known Gaussian Fourier covariance. Compare an unconstrained CNN, a hard-neutrality-only model, and the proposed model. The predicted signature is that S_ZZ(k) divided by k squared becomes approximately flat at low k, the k = 0 charge variance falls to numerical precision, and the neutral covariance approaches a nonzero plateau. Fit log-log slopes and require a charge slope of 2 plus or minus 0.2; the unconstrained baseline should show a nonzero zero-mode plateau or an incorrect slope.",
 "math_summary": "Let delta c_a(k) be the Fourier coefficient of the fluctuation of species a, with a ranging from 1 to A, and let q_a be its charge. Define the covariance matrix C_ab(k) as the expectation of delta c_a(k) delta c_b(-k), the charge vector q as (q_1,...,q_A)^T, and the charge amplitude z(k) as q^T delta c(k). The charge structure factor is S_ZZ(k) = q^T C(k) q. The neutral subspace is selected by P_q = I - q q^T divided by q^T q, which satisfies q^T P_q = 0. For a conducting Coulomb mixture, the long-wavelength signature is S_ZZ(k) = alpha k^2 + o(k^2), where alpha is positive, while the projected neutral covariance P_q C(k) P_q can approach a finite matrix as k approaches zero. Enforce exact global neutrality by replacing the zero Fourier mode with P_q delta c(0), and use a spectral loss matching the empirical neutral covariance and the quadratic low-wavenumber charge law.",
 "math_tags": [
  "statistical-mechanics",
  "linear-algebra",
  "harmonic-analysis",
  "probability"
 ],
 "ml_areas": [
  "cnn",
  "mlp",
  "regularization",
  "world-model"
 ],
 "paper": {
  "arxiv_id": "2608.30060",
  "arxiv_url": "https://arxiv.org/abs/2608.30060",
  "summary_what_math_gives_to_ml": "The paper provides a constructive separation of long-wavelength composition fluctuations from forbidden macroscopic charge fluctuations in charged mixtures. Its transferable asset is a covariance and structure-factor representation in which neutral composition modes have finite small-wavenumber susceptibility, while the charge mode is suppressed by electroneutrality and Coulomb correlations, typically with a charge structure factor proportional to k squared. This can become a physics-informed regularizer and hard projection for neural fields or generative models that predict multicomponent concentration fields, preventing unphysical global charge drift while preserving chemically meaningful neutral fluctuations.",
  "title": "Chemical potentials from structure factors: II. Charged multi-component mixtures",
  "year": "2026"
 },
 "ratings": {
  "difficulty": 5,
  "novelty": 7,
  "usefulness": 6
 },
 "solves": [
  "stability",
  "accuracy",
  "generalization"
 ],
 "title": "Neutral-Mode Structure-Factor Regularization",
 "url": "https://synthcore.org/idea/2862/neutral-mode-structure-factor-regularization",
 "verification": {
  "peer_reviewed": false,
  "status": "unverified",
  "status_label": "Unverified",
  "verdict_source": "deterministic test code (paired-seed permutation statistics)"
 }
}