--- title: 'Marginal Flow: Optimal Transport & Density Estimation' url: https://www.emergentmind.com/topics/marginal-flow type: topic --- # Marginal Flow: Optimal Transport & Density Estimation “Marginal Flow” is not a single standardized object but a family of constructions in which the central mathematical role is played by marginal distributions: preserving them, matching them at multiple times, transporting between them, or parameterizing densities through marginalization. In optimal transport, the term denotes a cost-specific, marginal-preserving neural-ODE update that moves inside the feasible set of couplings [2209.14577]. In density estimation, it denotes a model of the form \[ q_\theta(x)=\int q(x\mid w)\,q_\theta(w)\,dw, \] where latent parameters are marginalized rather than optimized as fixed mixture components [2509.26221]. Related usages appear in multi-marginal Schrödinger bridges, continuum-marginal optimal transport, bridge-aware samplers, and several field-theoretic settings involving marginal or marginally irrelevant operators [1902.08319] [2604.24226] [2102.01829]. This suggests that the term is best understood through recurring structural motifs rather than a single canonical definition. ## 1. Marginals as constraints, invariants, and dynamical observables A recurring formal idea is that a flow acts on couplings or path measures while the observable constraints are marginal laws. In the optimal-transport usage of “Marginal Flow,” one starts from a differentiable stochastic process \(\{X_t\}_{t\in[0,1]}\) with velocity field \[ v_t^{\mathbf X}(z)=\mathbb{E}[\dot X_t \mid X_t=z], \] whose marginals \(\rho_t=\law(X_t)\) satisfy \[ \partial_t \rho_t + \nabla\!\cdot\!\big(v_t^{\mathbf X}\rho_t\big)=0. \] A vector field \(r_t\) is \(\mathbf X\)-marginal-preserving if \[ \int_0^t \mathbb{E}\big[\nabla h(X_s)\cdot r_s(X_s)\big]\,ds = 0 \quad \forall h\in \mathcal C_c, \] equivalently, \[ \nabla\cdot\big(r_t\rho_t\big)=0. \] The key theorem states that two processes with the same initial law have identical marginals for all \(t\) iff the difference of their velocity fields is marginal-preserving [2209.14577]. In continuum-marginal optimal transport, marginals are not merely endpoint constraints but a full time-continuous input \(\mu=\{\mu_t\}_{0\le t\le T}\). The admissible velocity field \(u(t,x)\) must reproduce every marginal through the weak continuity equation \[ \int_{\mathbb{R}^d}\varphi(t,x)\,p(t,x)\,dx - \int_{\mathbb{R}^d}\varphi(0,x)\,p(0,x)\,dx = \int_0^t\!\int_{\mathbb{R}^d} \bigl(\partial_s\varphi + u(s,x)^{\mathsf T}\nabla_x\varphi\bigr)p(s,x)\,dx\,ds, \] and the optimization problem is \[ V_0(\mu) := \inf_{u\in\mathcal{U}_0} \int_{[0,T]\times\mathbb{R}^d} |u(t,x)|^2\,p(t,x)\,dx\,dt. \] Here the marginal flow itself is the object being recovered [2604.24226]. A related continuum viewpoint appears in the Sinkhorn/IPFP limit. As \(\varepsilon\downarrow 0\) and \(k\approx t/\varepsilon\), the one-marginals produced by the discrete entropic OT iterations converge to an absolutely continuous curve \((\rho_t)_{t\ge 0}\) in \( \mathcal P_2(\mathbb R^d)\), called the Sinkhorn flow [2307.16421]. Across these formulations, “marginal flow” refers less to a particular architecture than to a dynamical law constrained or characterized by marginals. ## 2. Marginal-preserving optimal transport and interior updates The most explicit algorithmic use of the name is the cost-specific extension of rectified flow in “Rectified Flow: A Marginal Preserving Approach to Optimal Transport” [2209.14577]. The problem is the Monge–Kantorovich program \[ \min_{(X_0,X_1)} \mathbb{E}[c(X_1-X_0)] \quad \text{s.t.} \quad \law(X_0)=\pi_0,\;\law(X_1)=\pi_1, \] with \(c\) convex. The method starts from any valid coupling, often \(\pi_0\times\pi_1\), forms the linear interpolation \[ X_t=(1-t)X_0+tX_1, \] and learns a new velocity restricted to the structured class \[ g_t^{\mathbf X,c}(z)=\nabla c^*\!\big(\nabla f_t^{\mathbf X,c}(z)\big), \] where \(f^{\mathbf X,c}\) minimizes \[ L_{\mathbf X,c}(f) = \int_0^1 \mathbb{E}\!\left[ c(\dot X_t)- \dot X_t\cdot \nabla f_t(X_t) + c^*(\nabla f_t(X_t)) \right] dt. \] The updated coupling is obtained from the neural ODE \[ dZ_t = g_t^{\mathbf X,c}(Z_t)\,dt,\qquad Z_0=X_0. \] The defining feature is the “interior approach”: the iterate remains feasible by construction. The update changes the coupling while preserving all time marginals, rather than enforcing constraints through dual variables or penalties. The velocity admits a Bregman/Helmholtz-like decomposition \[ v_t^{\mathbf X} = \nabla c^*(\nabla f_t^{\mathbf X,c}) + r_t^{\mathbf X,c}, \] where the residual is exactly the marginal-preserving component removable without changing the marginals. The cost-improvement theorem states \[ \mathbb{E}[c(X_1-X_0)]-\mathbb{E}[c(Z_1-Z_0)] = S_c(\mathbf Z)+L_{\mathbf X,c}(f^{\mathbf X,c}) \ge 0, \] and the fixed-point criterion is \[ (X_0,X_1)=\crectify(X_0,X_1) \quad \Longleftrightarrow \quad \ell^*_{\mathbf X,c}=0 \quad \Longleftrightarrow \quad (X_0,X_1)\ \text{is \(c\)-optimal}. \] For quadratic cost, \[ c(x)=\frac12\|x\|^2,\qquad c^*(y)=\frac12\|y\|^2,\qquad \nabla c^*(y)=y, \] the construction reduces to standard rectified flow. A common misconception is that “marginal-preserving” means the process is unchanged. In fact, the entire purpose is to alter the coupling while keeping the prescribed marginals fixed [2209.14577]. ## 3. Multi-marginal, bridge, and continuum-time formulations Several works generalize marginal flow from two endpoint constraints to multi-time or all-time settings. In the multi-marginal Schrödinger bridge problem for inertial particles, the state evolves by \[ dx = v\,dt, \qquad dv = dw, \] but only positional marginals \[ \int \mu_{t_i}(x,v)\,dv = \rho_i(x), \qquad i=0,\dots,N, \] are observed. The solution is the path-space law \(\mathbb Q\) minimizing relative entropy with respect to the inertial Brownian prior, equivalently the optimal control problem \[ \min \left\{\int_0^1 \|a(t)\|^2\,dt\right\} \] subject to \[ dx = v\,dt,\qquad dv = a\,dt + dw,\qquad x(t_i)\sim \rho_i. \] In density form, the formulation becomes a Benamou–Brenier-like variational problem, and after the substitution \(\hat a = a - \frac12 \nabla_v \log \mu\), the cost acquires a Fisher information term in the velocity variable [1902.08319]. For high-dimensional snapshot data at irregular time points, Multi-Marginal Stochastic Flow Matching (MMSFM) models a stochastic flow \[ dX(t) = u_t(X(t))\,dt + g(t)\,dW(t), \] with the decomposition \[ u_t(x) = u_t^\circ(x) + \frac{g^2(t)}{2}\nabla \log p_t(x). \] The method constructs overlapping windows of consecutive marginals, uses transport splines to build Gaussian conditional paths \[ p_t(x|z)=\mathcal N(x\mid \mu_t,\sigma_t^2), \] and trains flow and score networks jointly by a simulation-free objective. The paper emphasizes spline reparameterization over actual time intervals and stratified time sampling as the mechanisms that handle irregular snapshot timing [2508.04351]. ALI-CFM introduces adversarially learnt interpolants \[ G_\phi(x_0,x_1,t)=(1-t)x_0 + tx_1 + t(1-t)f_\phi(x_0,x_1,t), \] and matches intermediate snapshot distributions through a GAN-style objective so that \[ (G_\phi(\cdot,\cdot,t_i))_\#\pi = q_{t_i} \] at observed times. The learned interpolants are then marginalized by conditional flow matching to obtain a time-dependent vector field [2510.01159]. A further continuum limit is the Sinkhorn flow, which is both a limit of entropic OT iterations and an example of Wasserstein mirror gradient flow. Its continuity equation is \[ \partial_t \rho_t + \operatorname{div}(\rho_t v_t)=0, \] with mirror-gradient velocity \[ v_t(x)= -\left(\nabla^2 u_t(x)\right)^{-1}\nabla\bigl(\log \rho_t(x)+f(x)\bigr), \] and potential evolution governed by the parabolic Monge–Ampère equation \[ \partial_t u_t(x)=f(x)-g(x^{u_t})+\log\det\!\left(\frac{\partial x^{u_t}}{\partial x}\right). \] The paper also proves exponential convergence under a logarithmic Sobolev inequality and a uniform lower bound on mirror curvature [2307.16421]. ## 4. Density estimation, tail control, and parameter marginalization In density modeling, “Marginal Flow” denotes a different construction: a conditional density with marginalized latent parameters. The framework defines \[ q_\theta(x)=\int q(x\mid w)\, q_\theta(w)\, dw = \mathbb{E}_{w\sim q_\theta(w)}[q(x\mid w)], \] implemented in practice as \[ q_\theta(x)\coloneqq \frac{1}{N_c}\sum_{i=1}^{N_c} q(x\mid w_{\theta,i}), \qquad w_{\theta,i}\sim q_\theta(w). \] The latent parameters are sampled by \[ w_{\theta,i} \coloneqq f_\theta(z_i), \qquad z_i\sim p_{\text{base}}(z), \] so \(f_\theta\) need not be invertible and the method imposes no bijectivity constraint. The paper emphasizes exact density evaluation, efficient sampling, forward-KL or reverse-KL training, architectural freedom, and the ability to model lower-dimensional manifolds by choosing \(p_{\text{base}}\) in \(\mathbb R^m\) with \(m