{
 "artifacts": null,
 "category": "architecture",
 "description": "Replace an unconstrained latent-to-data map in a small generative model with a learned Brenier map T equal to the gradient of a convex potential, transporting a Gaussian latent distribution toward a density proportional to exp(-B) times the Gaussian density. Constrain B to be globally L-Lipschitz and use the resulting dimension-free Jacobian ceiling as an adaptive stability target. This should reduce pathological local expansion without requiring B itself to be convex or its Hessian to be bounded.",
 "formulas_latex": [
  "$$d\\gamma_d(x)=(2\\pi)^{-d/2}e^{-|x|^2/2}\\,dx$$",
  "$$d\\mu_B(x)=Z_B^{-1}e^{-B(x)}d\\gamma_d(x),\\qquad Z_B=\\int_{\\mathbb{R}^d}e^{-B(x)}d\\gamma_d(x),\\qquad |B(x)-B(y)|\\leq L|x-y|$$",
  "$$T=\\nabla\\phi,\\qquad T_{\\#}\\gamma_d=\\mu_B,\\qquad \\operatorname{Lip}(T)\\leq\\mathfrak{C}(0,L),\\qquad \\log\\mathfrak{C}(0,L)=4L^2+\\log L+\\mathcal{O}(1)$$",
  "$$0\\preceq D^2\\phi(x)\\preceq\\mathfrak{C}(0,L)\\,\\mathrm{Id}\\quad\\text{a.e.},\\qquad \\widehat{L}=\\max_{x\\in\\mathcal{B}}\\|\\nabla B(x)\\|_2,\\qquad \\widehat{C}=\\exp(4\\widehat{L}^{\\,2})\\max(1,\\widehat{L})$$"
 ],
 "id": 3161,
 "implementation": "(1) Integration point: use this as the final latent adapter in a VAE or as an initialization map for a diffusion model. Sample z from N(0,I_d), compute x=T_theta(z)=grad_z phi_theta(z), and pass x to the decoder or diffusion network. Parameterize phi_theta as an input-convex neural network: use nonnegative hidden-to-hidden weights and convex activations, then obtain T_theta by automatic differentiation. Parameterize B_psi with a spectrally normalized MLP so that its gradient norm is controllable. (2) Pseudocode: sample z; compute x=grad_z(phi_theta(z)); evaluate the generative loss; compute L_hat as the maximum minibatch value of norm(grad_x B_psi(x)); set log_C_hat=min(4*L_hat^2+log(max(1,L_hat)), log(C_max)); estimate the largest singular value s_hat of the Jacobian of T_theta using one or two power iterations with Jacobian-vector products; add lambda times relu(log(s_hat)-log_C_hat)^2 to the objective; update theta, psi, and decoder parameters. (3) The paper supplies the dimension-independent implication and the quadratic-in-L logarithmic scaling. Estimate L_hat, s_hat, and the multiplicative calibration factor empirically; do not treat the asymptotic expression as an exact finite-L constant. Clip log_C_hat to prevent overflow. (4) First experiment: train 2D and 16D VAEs on mixtures of Gaussians, comparing an unconstrained MLP decoder, an unconstrained convex-gradient adapter, and the proposed adapter. Measure held-out likelihood, reconstruction error, maximum and 99th-percentile Jacobian singular values, gradient explosions, and failed runs as dimension increases from 2 to 64. The expected signal is fewer unstable runs and smaller Jacobian tails at matched likelihood, with weaker degradation as dimension grows. A secondary experiment should initialize a diffusion sampler with T(z) and compare FID or sliced Wasserstein distance at equal sampling steps.",
 "math_summary": "Let gamma_d be the standard Gaussian probability measure on R^d, with density d gamma_d(x) = (2 pi)^(-d/2) exp(-|x|^2/2) d x. For a globally L-Lipschitz function B:R^d to R, define mu_B by d mu_B(x) = Z_B^(-1) exp(-B(x)) d gamma_d(x), where Z_B = integral of exp(-B) against gamma_d and |B(x)-B(y)| \u003c= L |x-y|. The quadratic-cost Brenier map T = grad phi satisfies T#gamma_d = mu_B and has a globally Lipschitz representative with Lip(T) \u003c= C(0,L), where log C(0,L) = 4 L^2 + log L + O(1) as L tends to infinity. Since phi belongs to C^(1,1), its Hessian exists almost everywhere and obeys 0 \u003c= D^2 phi(x) \u003c= C(0,L) Id in Loewner order. The transferable asset is that first-order control of B implies global differential control of the transport map, without assumptions that D^2 B is positive or small.",
 "math_tags": [
  "optimal-transport",
  "convex-analysis",
  "measure-theory",
  "geometry"
 ],
 "ml_areas": [
  "vae",
  "diffusion-sampling",
  "training"
 ],
 "paper": {
  "arxiv_id": "2609.04052",
  "arxiv_url": "https://arxiv.org/abs/2609.04052",
  "summary_what_math_gives_to_ml": "The paper establishes a dimension-free stability principle for quadratic optimal transport: if a Gaussian target density is perturbed by a globally L-Lipschitz log-density term, its Brenier map remains globally Lipschitz even without convexity or semiconvexity of the perturbation. This can be transferred to latent-variable generators by parameterizing the generator as the gradient of a convex potential and controlling the Lipschitz constant of the target log-density perturbation. The resulting bound is conservative and grows exponentially in L squared, so its most practical use is as a stability certificate or adaptive Jacobian regularizer rather than as a hard architectural constraint.",
  "title": "Caffarelli Estimates under Lipschitz Perturbations",
  "year": "2026"
 },
 "ratings": {
  "difficulty": 7,
  "novelty": 6,
  "usefulness": 6
 },
 "solves": [
  "stability",
  "accuracy",
  "generalization"
 ],
 "title": "Lipschitz-Perturbation Brenier Latent Adapter",
 "url": "https://synthcore.org/idea/3161/lipschitz-perturbation-brenier-latent-adapter",
 "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
   }
  }
 }
}