Iterative improvement of free energy landscape reconstructions with optimal protocols derived from differentiable simulations
Oliver Cheng 1,†, Zofia Adamska 2,†, Michael P. Brenner 1, Megan C. Engel 1,3 · 2025
Read Iterative improvement of free energy landscape reconstructions with optimal protocols derived from differentiable simulations by Oliver Cheng 1,†, Zofia Adamska 2,†, Michael P. Brenner 1, Megan C. Engel 1,3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-17831.txt.
Introduction
Oliver Cheng, Zosia Adamska, Michael Brenner, and Megan Engel improve reconstructions of free energy landscapes, maps that show which molecular shapes are favored and how a molecule moves between them. Finding the best way to drive a molecule during measurement normally requires knowing the landscape in advance, which is the thing being sought.
Their loop starts from data, builds an approximate landscape, and uses fast differentiable simulations to design a better driving plan for the next run. Repeating the loop recovers known test cases and cuts error sharply versus simple linear plans. The method needs no prior knowledge of the true landscape.
Catalog record
| CREATOR | Oliver Cheng 1,†, Zofia Adamska 2,†, Michael P. Brenner 1, Megan C. Engel 1,3 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-17831
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Iterative improvement of free energy landscape reconstructions with optimal protocols derived from differentiable simulations Oliver Cheng 1,†, Zofia Adamska 2,†, Michael P. Brenner 1, Megan C. Engel 1,3 Abstract Free energy landscapes encode the kinetics, intermediates, and transition states that govern molecular processes and are thus a key target of single biomolecule research. Typical approaches to deriving optimal, error-minimizing, non-equilibrium driving protocols for estimating these landscapes require a priori knowledge of the landscape. Here, we present an alternative: an iterative algorithm for optimizing full free energy landscape reconstructions which can be used alongside experiments on unknown landscapes. Our approach (i) takes experimental or simulated trajectory data; (ii) reconstructs an ‘approximate’ energy landscape; (iii) derives optimal control protocols from low-dimensional differentiable Brownian dynamics simulations on the candidate landscape using automatic differentiation; (iv) re-runs the experiment or simulation using the updated protocol; and (v) iterates until convergence. Using this approach, we recover known benchmarks from the literature and probe far-from-equilibrium regimes for symmetric, asymmetric, and triple-well energy landscapes under both 1- and 2-dimensional control. Our control protocols – derived with no a priori knowledge of the energy landscape – yield substantially reduced variance and bias in free energy landscape reconstructions compared to naive linear protocols. I Introduction Calculating free energy differences is a key goal in diverse scientific contexts, from identifying and ranking drug candidates 45 ; 7 ; 35 ; 42 , to mapping the folding landscapes of disease-causing biomolecules 29 ; 13 , to elucidating enzyme catalysis mechanisms 30 . Performing free energy calculations at equilibrium, without enhanced sampling in some form, is intractable for many systems that exhibit slow dynamics 35 ; 59 . Advances in nonequilibrium statistical physics – namely, the Crooks theorem 17 and the Jarzynski equality 32 – provide a roadmap for computing equilibrium free energies from an ensemble of nonequilibrium experiments and simulations. While these tools have been successfully leveraged to calculate free energies in a range of contexts 52 ; 30 ; 39 ; 44 ; 43 ; 1 , a crucial limitation lies in the poor convergence behavior of the unidirectional estimator: often an impractically high number of trajectories are required to produce a reliable free energy estimate 3 ; 27 ; 33 ; 61 . There has been a recent surge of interest in approaches that improve the accuracy of Jarzynski-derived free energy calculations 37 ; 64 ; 57 ; 56 ; 8 . A closely related problem is that of devising external protocols that minimize the dissipated work in a nonequilibrium process. This is crucial because the number of trajectories required for the unidrectional Jarzynski free energy estimate to converge is exponential in the dissipated work 27 ; 61 as samples in the left tail of the work distribution dominate the estimator 33 ; 61 . This optimal control problem has been extensively studied and remains a topic of interest 11 ; 24 . In the near-equilibrium limit, solutions are geodesics in the space of thermodynamic geometry 24 ; 49 ; 9 ; 63 , but such general analytical solutions are intractable for all but the simplest systems 46 ; 11 . Outside of the linear response regime, numerical approaches are necessary to make progress 23 ; 62 ; 64 ; 54 ; 20 . For example, Whitelam et al. 54 used genetic algorithms to train neural-network parameterized optimal protocols for a variety of physical systems. Zhong et al. used a trajectory reweighting scheme to compute bidirectional optimal protocols, which vastly outperform unidirectional free energy estimates 64 . Engel et al. used automatic differentiation (AD) to compute minimum dissipation protocols via stochastic gradient descent 20 . Excepting the work of Whitelam et al. 54 and Zhong et al. 64 , techniques for computing optimal control protocols on bistable landscapes require the mathematical form of the energy landscape to be known a priori . Furthermore, existing approaches to maximizing the accuracy of nonequilibrium free energy estimates are typically tailored to calculating a single free energy difference \Delta F rather than full free energy landscapes F(x) ; focus on a one-dimensional control protocol (usually the position of a harmonic potential as a function of time); and treat bistable, symmetric energy landscapes. Here, we address each of these gaps. Inspired by single-biomolecule experiments that seek to reconstruct full free-energy landscapes as a function of a molecular coordinate of interest 40 ; 22 ; 28 ; 19 ; 59 –typically, molecular end-to-end extension–we extend the approach of Engel et al. 20 to the experimentally relevant situation wherein the molecular free energy landscape is a priori unknown and simulations for which the gradients of the work values cannot be easily obtained. This allows experimentalists and simulators to calibrate protocols to enhance sampling of their systems without requiring any new technology. In particular, we propose the following iterative scheme: (1) collect trajectories and measure the dissipated work for each, beginning with a linear initial protocol guess; (2) reconstruct a full free energy landscape from the resulting trajectories using the method developed by Hummer and Szabo (See Equation. 2 ) 31 for adapting the Jarzynski equality to internal molecular coordinates; (3) compute the minimum-dissipation protocol for this landscape using AD, and (4) iterate by repeating steps (1) - (3) with increasingly optimized protocols until the free energy estimate converges (see Figure 1 ). In this paper, we generate trajectories via Brownian dynamics in JAX MD 47 , but we emphasize that any experimental or MD source suffices; the only requirement is a recording of the time dependent work (Equation 7 ) used in free energy reconstructions. With our approach, we are able to successfully reconstruct a variety of free energy landscapes, including triple well and asymmetric landscapes, with a marked improvement over the performance of the naive linear protocol. This is relevant for the broad class of biomolecules that feature intermediate, metastable configurations on their energy landscapes 28 ; 12 . Our method is in principle applicable in any regime, from weak to strong protocol strength and slow to fast driving. Additionally, the computational cost of reverse-mode AD scales well with the number of input parameters 5 ; in particular, our method scales linearly in memory with the number of protocol dimensions. Here, we demonstrate the successful use of AD for optimizing 1- and 2-dimensional control protocols across a range of barrier heights. The remainder of this paper is organized as follows: in Section II we review the relevant background: free energy landscape reconstruction following Hummer and Szabo 31 , computing optimal protocols for barrier crossing problems, and using automatic differentiation on Brownian dynamics simulations in JAX MD. Section III outlines our iterative method, with simulation details supplied in Appendix A . Our main results are presented in Section IV , and we conclude with a discussion in Section V . Figure 1 : Schematic of our iterative scheme for reconstructing a priori unknown free energy landscapes. (a) Nonequilibrium barrier crossing trajectories are collected, either through simulations or experiments. Shown here is a hypothetical laser optical tweezers (LOT) setup measuring the unfolding of a DNA hairpin. Reproduced from Ref. 38 with permission. (b) For each barrier crossing trajectory, external work is measured as a function of time. The