🔬 condensed matter

Progress toward a better BOCS: Systematic coarse-graining with local density potentials

This paper introduces BOCS version 5.0 and the accompanying PKG-BOCS LAMMPS package, which utilize force-matching to parameterize coarse-grained models with local density and square gradient potentials, demonstrating significant improvements in the structural fidelity, thermodynamic properties, and transferability of water models.

Original authors: Maria C. Lesniewski, Michael R. DeLyser, W. G. Noid

Published 2026-06-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Maria C. Lesniewski, Michael R. DeLyser, W. G. Noid

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to understand a massive, bustling city. You could try to track every single person, car, and streetlight individually. This is like an "all-atom" computer simulation: incredibly detailed, but so slow that you can only watch a few minutes of city life before your computer runs out of memory.

To solve this, scientists use Coarse-Grained (CG) models. Instead of tracking every person, you group them into "neighborhoods." You treat each neighborhood as a single point. This is like looking at the city from a helicopter; you lose the details of individual faces, but you can see the traffic patterns and city growth over days or years.

The problem is: How do you decide how these neighborhoods interact? If you guess wrong, the city might collapse or fly apart in the simulation.

This paper introduces BOCS version 5.0, a new software tool that helps scientists build these "neighborhood" models much more accurately. Here is how it works, using simple analogies:

1. The Old Way: Just Looking at Neighbors

Previously, these tools mostly looked at pairwise interactions. Think of it like saying, "If two people are standing 5 feet apart, they push each other with this much force." It's a simple rule: Distance = Force.

But in real life, people don't just react to the person standing right next to them. They react to the crowd. If you are in a packed concert, you feel different than if you are in an empty park, even if the person next to you is the same distance away.

2. The New Feature: The "Local Density" Potentials

The big upgrade in BOCS 5.0 is the ability to account for Local Density (LD).

The Analogy: Imagine you are a person in a crowd.
- Old Model: You only care about the person touching your shoulder.
- New Model (LD): You care about how crowded the whole area around you is. Are you in a dense mosh pit? Or a sparse gathering?
How it works: The software calculates a "density score" for every particle based on how many neighbors are nearby. It then adjusts the force between particles based on this score.
- If the area is very crowded, the particles might push harder to avoid crushing each other.
- If the area is sparse, they might pull together to stay connected.

This allows the simulation to understand that a molecule in a solid block of ice behaves differently than the same molecule in a thin gas, even if they are just looking at their immediate neighbors.

3. The "Square Gradient" (SG): The Edge Effect

The paper also introduces a second new feature called Square Gradient (SG) potentials.

The Analogy: Think of a beach where the sand meets the ocean.
- In the middle of the sand, the density is constant.
- In the middle of the water, the density is constant.
- But at the edge? The density changes rapidly. This is a "gradient."
How it works: The SG potential specifically looks at how quickly the crowd density changes from one spot to the next. It helps the simulation accurately model interfaces (like the surface of a water droplet or the boundary between oil and water). Without this, simulations often get the "edge" of a liquid blob wrong, making it too fuzzy or too sharp.

4. The "Force-Matching" Magic

How does BOCS figure out the right rules for these crowded or edge-heavy situations? It uses a method called Force-Matching.

The Analogy: Imagine you have a high-definition video of a dance troupe (the detailed "all-atom" simulation). You want to teach a simplified robot (the coarse-grained model) to mimic the dance.
The Process: BOCS watches the high-definition video and measures the exact forces acting on every dancer at every moment. It then asks the robot: "If you move your simplified limbs this way, does it create the same push and pull as the real dancers?"
The Result: The software tweaks the robot's rules over and over until the robot's movements perfectly match the forces seen in the high-definition video. In this new version, the robot learns not just how to move with its neighbors, but how to move based on the crowd density and the edges of the group.

5. Why This Matters (According to the Paper)

The authors tested this new software on water.

The Problem: Water is tricky. It forms droplets, it has a surface, and it behaves differently in bulk liquid vs. vapor. Old "neighborhood" models often failed to get the density or the shape of the water droplet right.
The Result: When the scientists added the "Local Density" and "Square Gradient" rules to their water model:
- The simulated water had the correct density.
- The simulated water droplet had the correct shape and surface tension.
- The model could move from a liquid simulation to a vapor simulation without breaking.

Summary

The paper describes a software update (BOCS 5.0) and a new simulation tool (PKG-BOCS) that lets scientists build better, faster models of complex materials.

Instead of just asking, "How far apart are you?" the new software asks, "How crowded is it around you, and how fast is that crowding changing?" By answering these questions, the software can create simulations of liquids, interfaces, and soft materials that are far more accurate than ever before, specifically showing great promise for modeling water.

Technical Summary: Progress toward a better BOCS: Systematic coarse-graining with local density potentials

Problem Statement
Bottom-up coarse-grained (CG) models offer computational efficiency for simulating length and time scales inaccessible to atomically detailed (AA) models. While the force-matching (FM) variational principle allows for the parameterization of CG potentials to optimally approximate the exact CG potential, standard implementations often rely on conventional pair potentials. These pair potentials struggle to capture the thermodynamic and structural properties of inhomogeneous systems, such as liquid-vapor interfaces, and often lack transferability between bulk and interfacial environments. Furthermore, volume potentials, while effective for reproducing pressure-density equations of state in bulk, cannot model interfaces. There is a need for a systematic framework that incorporates potentials dependent on the local density (LD) and the square gradient (SG) of that density to improve the fidelity and transferability of CG models for complex soft materials.

Methodology
The authors describe version 5.0 of the Bottom-up Open-source Coarse-graining Software (BOCS) package and a corresponding extension to the PKG-BOCS package for LAMMPS.

Theoretical Framework: The software parameterizes CG potentials using the force-matching variational principle, which minimizes the mean-squared difference between the forces in the AA model and the CG model over the mapped ensemble. The total non-bonded potential is expanded to include three components:
1. Conventional central pair potentials ( $U_{pair}$ ).
2. Local density potentials ( $U_{LD}$ ), which depend on the instantaneous local density $\rho_I$ around each particle $I$ .
3. Square gradient potentials ( $U_{SG}$ ), which depend on the square of the gradient of the local density, $|\nabla_I \rho_I|^2$ .
  The local density is defined as a sum of contributions from neighboring particles weighted by a function $w(r)$ , for which five different functional forms (e.g., DPD, Lucy, Shell, smooth, sphere) are implemented.
Software Implementation:
- BOCS v5.0: This version introduces tools (ldcalc) to calculate local densities and their gradients from AA trajectories. It employs a generalized-Yvon-Born-Green (g-YBG) framework to perform force-matching without explicit force information for bonded, pair, and LD terms. The software uses spline-based basis functions to represent force functions and coefficients, allowing for flexible functional forms. It solves the resulting normal equations via singular value decomposition to determine optimal parameters.
- PKG-BOCS (LAMMPS): A new pair style (ldd) was implemented in LAMMPS to simulate these potentials. This implementation supports complex molecular topologies, multiple particle types, and various weighting functions. It properly accounts for intramolecular contributions and exclusions.
Validation and Application: The authors validated the implementation through three benchmark tests:
1. A system of 400 particles with linear LD potentials.
2. A system of 2 particles with quadratic LD and linear SG potentials.
3. A complex system of 125 branched molecules with 4 site types, involving bonded, pair, LD, and SG interactions.
  Additionally, the software was applied to parameterize 1-site CG water models based on the SPC/E AA model, comparing models parameterized from bulk simulations, slab simulations, and bulk simulations augmented with LD potentials.

Key Results

Implementation Accuracy: In all benchmark tests, BOCS v5.0 quantitatively recovered the known "ground truth" potentials and forces used in the LAMMPS simulations. The force-matching calculations yielded negligible errors ( $\chi^2 \approx 0$ ), demonstrating that the software correctly implements the variational principle for LD and SG potentials.
Physical Insight: The study revealed that the effective pair interactions generated by LD and SG potentials are highly sensitive to the choice of the weighting function $w(r)$ . For instance, different weighting functions can lead to significantly different effective pair potentials and force profiles, even when the underlying LD or SG functional form is identical.
Water Model Performance:
- A standard pair-only model parameterized from bulk water (PAIR-BULK) failed to stabilize the liquid phase at 1 bar, resulting in vaporization.
- A pair-only model parameterized from a slab simulation (PAIR-SLAB) stabilized the liquid but significantly overestimated density and compressibility and poorly reproduced the liquid-vapor interface width.
- The LD-BULK model (pair + LD potentials parameterized from bulk) significantly improved performance. It reproduced the SPC/E equation of state with high accuracy (underestimating density by only 0.1%) and reasonably reproduced the liquid-vapor interface width and density profile, despite being parameterized solely from a bulk simulation.
- Analysis showed that the LD potential modulates pair forces based on the local environment, effectively deepening attractive interactions in low-density regions (mimicking interface behavior) while remaining repulsive in high-density regions.

Significance and Claims
The authors claim that BOCS version 5.0 and the PKG-BOCS extension provide a robust and accurate tool for systematically parameterizing CG models with local density and square gradient potentials. The paper asserts that these potentials offer a significant improvement in the structural fidelity, thermodynamic properties, and transferability of bottom-up CG models, particularly for inhomogeneous systems like liquid-vapor interfaces. The work demonstrates that LD potentials can effectively capture cooperative many-body interactions and improve the description of density fluctuations without requiring explicit many-body force terms in the simulation, as the environmental dependence is captured through the local density. The authors position this software as a useful complement to existing packages for developing CG models of complex soft materials.