Extending Hamiltonian-Adaptive Resolution Simulation to Interfaces: An Updated LAMMPS Implementation and Application to Porous Solids

This paper presents an updated LAMMPS implementation of the Hamiltonian-Adaptive Resolution Simulation (H-AdResS) method that supports diverse potentials and fluctuating densities, demonstrating its efficiency and accuracy in modeling gas adsorption and transport in porous metal-organic frameworks while preserving atomistic-level properties.

The Gist

Imagine you are trying to watch a movie, but you only have a very expensive, high-definition camera and a limited battery. You want to film a specific scene where a complex action happens (like a car crash or a chemical reaction), but the scene is surrounded by a huge crowd of people just standing around.

If you try to film the entire crowd with your high-definition camera, your battery will die instantly, and the computer will crash because there's too much data to process. But if you zoom out and use a blurry, low-resolution camera for the crowd, you lose the details you need for the main action.

This is the problem scientists face when simulating materials. They need to see the tiny, atomic details of a specific spot (like gas entering a porous rock) but also need to account for the massive environment surrounding it.

This paper introduces a clever solution called H-AdResS (Hamiltonian-Adaptive Resolution Simulation), and the authors have built a new, upgraded "camera" for it inside a popular software called LAMMPS.

Here is the breakdown of what they did, using simple analogies:

1. The "Smart Zoom" Camera

Think of the simulation box as a room.

• The High-Res Zone (All-Atom): In the center of the room, where the interesting action happens, every single atom is visible. It's like looking through a microscope. This is accurate but slow.
• The Low-Res Zone (Coarse-Grained): In the corners of the room, the atoms are grouped together into "beads" or clusters. It's like looking at the crowd from a helicopter; you see the movement, but not individual faces. This is fast but less detailed.
• The Hybrid Zone (The Transition): Between the microscope and the helicopter view, there is a smooth transition area. As a molecule moves from the center to the corner, it slowly morphs from a detailed atom into a simple bead, and back again.

The Innovation: The authors updated the software (LAMMPS) to make this "morphing" process much smoother and more flexible. Before, this software was like an old, clunky camera that only worked with specific types of film. Now, it works with any "film" (force fields) and is much easier to set up.

2. The "Density Balancer" (The Tricky Part)

Here is the biggest challenge they solved: Porous Materials.

Imagine a sponge (a Metal-Organic Framework, or MOF). It has holes and solid parts.

• In a normal liquid (like water), the density is the same everywhere.
• In a sponge, some parts are solid, and some parts are empty air.

The old version of the software got confused when it tried to calculate the "transition" in a sponge. It would see an empty hole, get a math error, and the simulation would crash. It was like trying to blend a photo of a solid wall with a photo of empty space, and the software didn't know how to handle the gap.

The Fix: The authors rewrote the math to handle these "empty" spots. They added a "balancing routine" that acts like a smart thermostat. If the density gets too high in one spot or too low in another, the software gently pushes or pulls the particles to keep everything stable, even in a sponge-like structure.

3. Why This Matters: The Gas in the Sponge

To prove their new camera works, they simulated ZIF-8, a type of sponge-like material used to capture carbon dioxide ($CO_2$) from the air.

• The Test: They let $CO_2$ gas float into the sponge.
• The Result: They found that the gas moved and behaved exactly the same way in their "Smart Zoom" simulation as it did in a full, high-definition simulation.
• The Benefit: Because they didn't have to calculate every single atom in the empty air pockets, the simulation ran 20% faster.

The Big Picture

Think of this new software update as upgrading from a manual transmission car to an automatic one with a turbocharger.

• Easier to use: You don't need to be a coding wizard to set it up anymore.
• Faster: You get to your destination (the scientific result) quicker.
• More capable: You can now drive off-road (simulate porous solids and gas interfaces) without getting stuck in the mud.

Why should you care?
This technology helps scientists design better materials for:

• Energy Storage: Making better batteries.
• Climate Change: Creating filters that suck $CO_2$ out of the air more efficiently.
• Medicine: Designing better drug delivery systems.

In short, the authors took a powerful but difficult tool, fixed its bugs, made it easier to use, and proved it works perfectly for studying how gases interact with porous materials—a key step toward solving real-world energy and environmental problems.

Technical Summary

1. Problem Statement

Multiscale phenomena often occur across varying length and time scales, where a region of interest (e.g., a chemical reaction or interface) is influenced by its surrounding environment. Traditional high-resolution atomistic (AA) simulations are computationally expensive for large systems, while coarse-grained (CG) models lack the necessary molecular detail.

• Limitations of Previous Methods: The Adaptive Resolution Simulation (AdResS) and its Hamiltonian variant (H-AdResS) allow for dual-resolution simulations but were previously limited to bulk liquids and solutions.
• Specific Challenges:
  • Software Obsolescence: The existing H-AdResS implementation in LAMMPS (from 2016) was incompatible with modern LAMMPS versions (post-2016 C++11 upgrades), making it impractical to use.
  • Inhomogeneous Systems: Previous implementations struggled with systems having non-uniform densities (e.g., porous solids, gases, interfaces). Standard compensation routines failed when simulation bins were empty, leading to unphysical forces or simulation crashes.
  • Force Field Rigidity: Existing implementations were often restricted to specific interaction potentials, limiting their applicability to diverse force fields (e.g., MARTINI, tabular potentials).

2. Methodology

The authors developed a new, optimized implementation of the Hamiltonian-Adaptive Resolution Simulation (H-AdResS) method within LAMMPS (2023 version).

• Theoretical Framework:

  • The system is divided into three regions: Atomistic (AA), Coarse-Grained (CG), and a Hybrid transition region.
  • A transition function, $\lambda(x)$, varies from 1 (AA) to 0 (CG) based on the center of mass of molecules.
  • The global Hamiltonian interpolates the non-bonded potential energy: $H = \kappa + U_{bonded} + \{\lambda_a U^{AA}_a + (1-\lambda_a) U^{CG}_a\}$.
  • A drift force arises from the resolution change, which is counteracted by compensation routines (pressure and density compensation) to maintain thermodynamic equilibrium.

• Implementation Improvements:

  • New Atom Style: Introduced atom_style hars (derived from full) to store per-atom variables: center of mass ($X$), leader atom flag, CG bead type, resolution ($\lambda$), and gradient ($\nabla\lambda$).
  • Enhanced Pair Styles: Developed new pair styles (lj/gromacs/hars, table/hars) to support diverse force fields, including analytical potentials (LJ), tabular potentials (IBI, MS-CG), and generic CG force fields (MARTINI).
  • Input Simplification: Added set type commands to assign leader atoms and CG bead types directly within the LAMMPS input script, eliminating the need for external preprocessing scripts.
  • Robust Compensation Routines:
    • Empty Bin Handling: Modified algorithms to interpolate energy for empty bins in the hybrid region, preventing discontinuities in systems with voids (porous solids).
    • Force Capping: Introduced a mechanism to cap forces/energies at region boundaries to prevent simulation crashes caused by sudden re-introduction of AA degrees of freedom in large molecules.

3. Key Contributions

1. Modern LAMMPS Port: Successfully migrated H-AdResS to LAMMPS 2023, ensuring compatibility with modern C++ standards and new fix/compute styles.
2. Generalization of Force Fields: The new implementation supports a wide range of interaction potentials, making the method accessible for complex CG models like MARTINI and tabular potentials.
3. First Application to Porous Solids: This is the first application of H-AdResS to Metal-Organic Frameworks (MOFs) and porous solid/gas interfaces, a significant expansion beyond bulk liquids.
4. User Accessibility: Streamlined the setup process via dedicated LAMMPS commands, reducing the barrier to entry for new users.

4. Results and Validation

The authors validated the implementation using two primary systems: Bulk Liquid Water and ZIF-8 (a Zeolitic Imidazolate Framework) with CO2 adsorption.

• Validation (Water):
  • Radial Distribution Functions (RDFs) and density profiles of the AA region in H-AdResS simulations matched perfectly with fully atomistic reference simulations and the legacy 2016 implementation.
• Application (ZIF-8/CO2):
  • Structural Integrity: RDFs for Zn-Zn, Zn-C, and C-C pairs in the AA region of the H-AdResS simulation were indistinguishable from fully atomistic results, confirming that the crystalline structure of the MOF and gas interactions are preserved.
  • Dynamic Properties:
    • Mean Residence Time ($\tau_R$): Calculated as 47 ± 7 ps for H-AdResS (with both compensations), matching the fully atomistic value (47 ± 4 ps) and literature values.
    • Diffusion Coefficient: The Mean Squared Displacement (MSD) and self-diffusion coefficient ($1.5 \times 10^{-9} m^2/s$) in the AA region matched the reference AA simulation.
  • Compensation Necessity: The study demonstrated that both pressure and density compensation are critical. Using only one led to density imbalances and incorrect dynamic properties (e.g., artificially high residence times due to gas crowding).
• Performance:
  • The H-AdResS implementation achieved a ~18–25% performance increase (ns/day) compared to fully atomistic simulations for the ZIF-8/CO2 system.
  • Efficiency scales linearly with the number of processes.
  • Efficiency decreases as the size of the AA region increases, highlighting the trade-off between resolution detail and computational cost.

5. Significance

This work represents a major step forward in multiscale molecular dynamics:

• Bridging Scales in Porous Materials: It enables the simulation of complex interfaces (solid/gas) in porous materials like MOFs, allowing researchers to study defect concentrations and composite compositions without the prohibitive cost of full atomistic simulations.
• Future Applications: The method paves the way for studying energy storage, electrocatalysis, and membrane technologies where interfaces are critical.
• Community Tool: By integrating these features into the widely used LAMMPS package with improved flexibility and ease of use, the authors provide a robust tool for the broader scientific community to simulate heterogeneous systems with high accuracy and efficiency.