Lifting the fog—a case for non-reversible “lifted” Markov chains
Gabriele Tartero, Sora Shiratani, Werner Krauth · 2026
Read Lifting the fog—a case for non-reversible “lifted” Markov chains by Gabriele Tartero, Sora Shiratani, Werner Krauth (2026) 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-2603-16855.txt.
Introduction
Tartero, Shiratani, and Krauth show how non-reversible Monte Carlo speeds up phase coarsening. They use fog as the everyday image: many tiny droplets slowly merge into fewer large drops. The classic Metropolis algorithm obeys reversibility, which keeps this coarsening slow. Their lifted variants break that rule at the microscopic level. Droplets then move relative to each other through a density-displacement coupling. For a phase-separated Lennard-Jones system, the authors report a larger coarsening growth exponent. The outcome matches Metropolis, but large systems finish far faster.
Catalog record
| CREATOR | Gabriele Tartero, Sora Shiratani, Werner Krauth |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37157 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.16855
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Lifting the fog—a case for non-reversible “lifted” Markov chains Gabriele Tartero, Sora Shiratani, Werner Krauth Abstract Phase transitions appear all over science, and are familiar from everyday life, as water boiling, sugar melting into caramel or as nematic molecules turning smectic in liquid-crystal displays. The dynamics of phase transitions can be extremely slow, as for example when fog in winter does not lift, that is when the coarsening takes much time from many tiny water droplets to fewer but larger rain drops that feel the pull of gravity. The dynamics of phase transitions is relevant also for the performance of computer algorithms, whose time evolution is often patterned after physical processes. In the ubiquitous Metropolis Monte Carlo algorithm, the mixing dynamics towards equilibrium leads towards the solution of a sampling problem. It is governed by the same reversibility and detailed-balance principles as the overdamped physical dynamics of fog. In the example of the phase-separated Lennard-Jones system, we describe here how the coarsening dynamics of non-reversible “lifted” variants of the Metropolis algorithm proceeds on much faster time scales, with the microscopic non-reversibility translating into large-scale relative motion of droplets that is impossible under the Ostwald-ripening condition of reversibility. A density–displacement coupling moves droplets relative to each other through a lensing effect. Efficient implementations of the long-range Metropolis algorithm and its non-reversible lifting (event-chain Monte Carlo) allow us to show that, in consequence, the coarsening growth exponent is larger under lifting. For large system sizes, the computing problem is thus solved infinitely faster than before, with the outcome (the equilibrium state) strictly unchanged with respect to the Metropolis algorithm. We also discuss the larger setting of our findings, namely that “lifted” non-reversible algorithms can be set up for generic reversible sampling methods, with applications going much beyond our example of lifting fog. \dates This manuscript was compiled on August 10, 2026 www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXX † † lead-author: Tartero † † corresponding-author: 1 To whom correspondence should be addressed. E-mail: [email protected] † † firstpage: [ 9]2 \dropcap For a range of temperatures and volumes, the equilibrium state of a system of N particles may be composed of two coexisting phases, following their separation at a first-order transition. During a cool night, for example, part of the water vapor contained in air may nucleate into tiny liquid droplets, creating fog. In spite of its persistence in time, familiar to every reader of 19th century English novels, fog is far from being an equilibrium state: It is unstable at large times. The explanation for this inescapable fact is that Nature strives to minimize free energy and that a single drop has smaller surface and lower surface free energy than two droplets of same combined volume. Eventually, droplets must thus coalesce until (in a finite volume) they form a single approximately spherical drop or until they precipitate as rain under the action of gravity (see Fig. 1 ). The persistent and apparently paradoxical Dickensian fog of Bleak House Dickens2008 , sticking around for days, illustrates that the coarsening dynamics of droplets increasing in size and decreasing in number is excruciatingly slow. Fog practically never precipitates into rain, and it generally lifts through a change of the external parameters: with an increase in temperature or with wind that starts to blow. After a while, a new equilibrium state is attained at a different temperature, without the original phase separation. Computational science patterns many of its techniques after physical processes. Thus, molecular dynamics and gradient descent mimic Newtonian time evolution in applications from physics to machine learning, and the vast field of Monte Carlo sampling, pioneered in physics by Metropolis et al. in 1953, now permeates all of science. The Metropolis algorithm, one of the most cited works in science Metropolis1953 , is a black box for computing and is perceived as a panacea for arbitrary sampling problems. The Metropolis algorithm implements time-reversible dynamics that converges towards equilibrium. The time reversibility, in other words the detailed-balance condition, built into the algorithm in or out of physics, is identical to the governing principle of the fog dynamics of tiny droplets. In very general applications, the Metropolis algorithm will, so to speak, eventually turn small droplets into rain, but the mixing time Levin2008 for this to come to fruition can be enormous, just as for the “fog everywhere” in “implacable November weather” ( Dickens2008 , p.12) . Two aspects of the analogy with London fog are crucial: First, the equilibrium state encoded in the Metropolis algorithm represents the solution of a given computational problem, and we require a speed-up of the dynamics without changing the steady state (without having the sun dissipate the fog). Second, in the case of a two-phase mixture, it has been understood since Ostwald in the late 1800s that a configuration with several droplets cannot be stationary, thus providing a clear-cut criterion for stationarity. In general, however, it is very difficult to ascertain that the steady state has been reached. This risks flawed conclusions, as for example in Fig. 1 b (if we lacked Ostwald’s insight), which is “almost” stationary under local reversible dynamics, but which is not at all in equilibrium (represented by Fig. 1 d). Figure 1: Coarsening of the two-dimensional Lennard-Jones system ( N=10^{4} particles at density \rho=0.1 ) from super saturation (a) into a single liquid drop surrounded by vapor (d). After nucleating several droplets (b), the system goes through a coarsening process, that gradually reduces the number of liquid clusters while increasing their average size (c). In the absence of gravity, the equilibrium state contains a single floating drop. The Metropolis algorithm, that mimics the physics of Ostwald ripening, needs about 10^{12} steps to complete this process, while its lifted counterpart, event-chain Monte Carlo, reaches equilibrium after around 10^{9} events. Non-reversible (thus, effectively, non-equilibrium) Markov chains violate detailed balance, and they feature finite steady-state flows even if the steady state is the equilibrium Boltzmann distribution. The perceived paradox of non-equilibrium in equilibrium has been clarified in exactly solved models KapferKrauth2017 ; Essler2024 . The practical construction of non-reversible Markov chains with an imposed steady state has long lacked clear design principles, but the concept of “lifting”, which copies every configuration of a parent chain into multiple configurations, has allowed definite progress Diaconis2000 ; Chen1999 . An example of non-reversible lifted Markov chains is given by the event-chain Monte Carlo algorithms for particle systems whose Boltzmann distributions then appears as a stationary state with finite flows Krauth2021eventchain ; Peters2026 . The characteristic convergence times for lifted Markov chains obey mathematical constraints Chen1999 . First, in order to speed up a reversible parent Markov chain, its lifting must be non-reversible. This validates our intuition that the slow evaporation of liquid particles from a small droplet into the vapor and the condensation of vapor particle to a larger liquid droplet (Ostwald ripening Lifshitz1961 ; Wagner1961 ) explains the persistence of fog, and more generally the slowness of the local Metropolis algorithm. Second, the convergence times of a lifted Markov chain are lower bounded by the square root of the times of its reversible parent algorithm. This validates the concept of ballistic vs. diffusive motion for general Markov chains, much beyond the realm of particle systems. However,