Chemical potential calculations in dense liquids using metadynamics

Claudio Perego, Federico Giberti, Michele ParrinelloThe European Physical Journal Special Topics 2016 · DOI · arXiv

Abstract

Calculating the chemical potential in dense systems is a long-standing challenge in computational chemistry. Standard "test-particle" methods often fail when atoms are packed tightly, as the probability of finding a suitable insertion cavity becomes vanishingly small. This paper introduces a robust alternative using Well-Tempered Metadynamics. By defining a collective variable that biases the system toward rare, "insertion-friendly" configurations, the authors enable efficient and accurate chemical potential calculations even in supercooled, high-density fluids like the Kob-Andersen binary Lennard-Jones mixture.

Introduction

The chemical potential ($\mu$) is a fundamental thermodynamic quantity governing phase transitions, chemical equilibria, and solubility. In molecular simulations, Widom’s test-particle insertion method has been the "gold standard." It involves periodically attempting to insert a "ghost" particle and measuring the resulting energy change. While effective for sparse gases, this method fails in dense liquids because the likelihood of a random insertion not overlapping with existing atoms is nearly zero, leading to poor convergence and noisy data.

Methodology: Biasing for Cavities

To overcome the sampling bottleneck in dense systems, the authors employ Well-Tempered Metadynamics (WT-MetaD). The core of the approach is a novel Collective Variable (CV), $s$, which represents the average Boltzmann factor of insertion energies across a grid of $M$ fixed points within the simulation box.

Free energy profiles F(s) for the collective variables s_B (a) and s_A (b) in the Kob-Andersen fluid.

As shown in Figure 1, the metadynamics bias pushes the system to explore states where at least one grid point becomes "hollowed out" enough to accept a particle. This "insertion-centric" approach forces the system's atoms to move cooperatively to create space, rather than relying on random chance. A regularized potential is used to handle short-range overlaps, and standard reweighting techniques are applied to recover the unbiased chemical potential from the biased trajectory.

Results: Success in the "Torture Test"

The method was validated using the Kob-Andersen binary Lennard-Jones fluid, a classic benchmark for supercooled liquids. This system contains large (Type A) and small (Type B) particles. Finding a cavity for the larger Type A particles in a dense liquid is statistically nearly impossible for standard random sampling.

Convergence of the excess chemical potential for Type B (a) and Type A (b) particles over simulation time.

The results in Figure 3 demonstrate the method's robustness. For the smaller Type B particles, both Widom’s method and Metadynamics converge. However, for the challenging Type A particles, Widom’s method (blue line in 3b) remains unstable and fails to converge even after extensive sampling. In contrast, the Metadynamics approach (red line) achieves stable convergence with an accuracy within 1%. The method proved consistent regardless of the number of insertion points used, though grid density affects the speed of convergence.

Conclusion and Future Outlook

By leveraging metadynamics to sample the extreme tails of the energy distribution—where rare cavity-opening fluctuations occur—the authors have provided a powerful tool for studying dense fluids. While the current implementation assumes a homogeneous fluid, future work aims to extend this framework to non-homogeneous systems, such as interfaces and complex molecular environments. This could significantly impact fields like drug discovery and chemical engineering, where calculating solvation energies for large molecules is critical.

Summary updated

BibTeX
@article{perego2016chemical,
  author = {Claudio Perego and Federico Giberti and Michele Parrinello},
  title = {{Chemical potential calculations in dense liquids using metadynamics}},
  journal = {The European Physical Journal Special Topics},
  year = {2016},
  doi = {10.1140/epjst/e2016-60094-x},
}
Figure