Abstract:We show that the minimum effort control of colloidal self-assembly can be naturally formulated in the order-parameter space as a generalized Schrödinger bridge problem -- a class of fixed-horizon stochastic optimal control problems that originated in the works of Erwin Schrödinger in the early 1930s. In recent years, this class of problems has seen a resurgence of research activities in the control and machine learning communities. Different from the existing literature on the theory and computation for such problems, the controlled drift and diffusion coefficients for colloidal self-assembly are typically nonaffine in control, and are difficult to obtain from physics-based modeling. We deduce the conditions of optimality for such generalized problems, and show that the resulting system of equations is structurally very different from the existing results in a way that standard computational approaches no longer apply. Thus motivated, we propose a data-driven learning and control framework, named `neural Schrödinger bridge', to solve such generalized Schrödinger bridge problems by innovating on recent advances in neural networks. We illustrate the effectiveness of the proposed framework using a numerical case study of colloidal self-assembly. We learn the controlled drift and diffusion coefficients as two neural networks using molecular dynamics simulation data, and then use these two to train a third network with Sinkhorn losses designed for distributional endpoint constraints, specific for this class of control problems.

What problem does this paper attempt to address?

The paper primarily aims to address the following issues: ### Research Background and Objectives - **Problem Definition**: The study investigates how to guide the colloidal self-assembly (SA) process from an initial disordered state to a desired terminal ordered state through minimum effort control. - **Mathematical Formulation**: This problem is naturally formulated as a generalized Schrödinger bridge problem (GSBP), a class of stochastic optimal control problems over a fixed time interval. This type of problem was initially proposed by Erwin Schrödinger in the 1930s and has recently garnered renewed attention in the control theory and machine learning communities. ### Key Challenges - **Nonlinear Dynamics**: The drift coefficient \(f\) and diffusion coefficient \(g\) associated with colloidal self-assembly are typically nonlinear and non-affine in the control input \(u\). This makes the problem difficult to solve directly using physics-based modeling methods. - **Data-Driven Approach**: Due to the difficulty in establishing precise physical models, researchers have adopted a data-driven learning and control framework called the "neural Schrödinger bridge." This framework uses molecular dynamics simulation data to train neural networks to approximate these nonlinear coefficients. ### Main Contributions - **Theoretical Contribution**: A new perspective is proposed to formulate the controlled colloidal self-assembly problem as a distribution control problem. This is the first work to derive and numerically solve the control non-affine generalized Schrödinger bridge problem in a multi-dimensional state. - **Computational Framework**: A computational framework named "neural Schrödinger bridge" is proposed, which includes two aspects: first, using neural networks (\(ND_{\text{drift}}\) and \(ND_{\text{diffusion}}\)) to approximate the drift coefficient \(f\) and diffusion coefficient \(g\); second, using a physics-informed neural network (PINN) combined with the Sinkhorn loss function to solve the optimal control problem. - **Optimality Conditions**: The optimality conditions for the control non-affine generalized Schrödinger bridge problem are derived, involving a set of coupled partial differential equations (PDEs). These equations are structurally different from the results in the control affine case, making traditional numerical methods inapplicable. ### Methodological Distinctions - **Differences from Other Works**: The framework proposed in this paper differs from previous works, such as learning uncontrolled dynamical systems through SBP or solving classical SBP problems between images. The colloidal self-assembly problem in this paper requires learning the controlled dynamical system first and then synthesizing optimal control based on it. ### Numerical Case Study - **Specific Implementation**: The effectiveness of the proposed framework is demonstrated through a numerical case study of a specific colloidal self-assembly system. The drift and diffusion coefficients are learned from molecular dynamics simulation data, and these learned results are used to train a third network to solve the generalized Schrödinger bridge problem. In summary, this paper aims to solve the optimal control problem in the colloidal self-assembly process by introducing a novel data-driven method—the "neural Schrödinger bridge," effectively addressing the challenges posed by nonlinear and non-affine control.

Neural Schrödinger Bridge with Sinkhorn Losses: Application to Data-driven Minimum Effort Control of Colloidal Self-assembly

Soft-constrained Schrodinger Bridge: a Stochastic Control Approach

Schrodinger Bridge over Averaged Systems

Schrödinger Bridge with Quadratic State Cost is Exactly Solvable

Control, Transport and Sampling: Towards Better Loss Design

Aligned Diffusion Schrödinger Bridges

Transport with Support: Data-Conditional Diffusion Bridges

Weyl Calculus and Exactly Solvable Schrödinger Bridges with Quadratic State Cost

On the Contraction Coefficient of the Schrödinger Bridge for Stochastic Linear Systems

Nonlinear Optimal Control of Electron Dynamics within Hartree-Fock Theory

Unbalanced Diffusion Schrödinger Bridge

Plug-in estimation of Schrödinger bridges

Schrödinger Bridge Flow for Unpaired Data Translation

Diffusion Bridge Mixture Transports, Schrödinger Bridge Problems and Generative Modeling

Time-Reversible Bridges of Data with Machine Learning

How to Learn and Generalize From Three Minutes of Data: Physics-Constrained and Uncertainty-Aware Neural Stochastic Differential Equations

Schrödinger's control and estimation paradigm with spatio-temporal distributions on graphs

Generalized Schrödinger Bridge Matching

Stochastic Optimal Control for Collective Variable Free Sampling of Molecular Transition Paths

Stochastic Optimal Control for Diffusion Bridges in Function Spaces