PLUMED 2 Documentation

PLUMED includes a set of standalone command-line tools that can be used independently of a molecular dynamics code. These tools are invoked using the plumed executable followed by the tool name and its specific flags.

Common use cases for these tools include:

• Postprocessing trajectories.
• Benchmarking PLUMED calculations.
• Running molecular dynamics on Lennard Jones particle systems.

Example invocation: plumed <toolname> <list of input flags for that tool>

The generic module provides actions for extracting information from molecular dynamics trajectories. These can be used during an MD run or via post-processing with the driver.

Key capabilities include:

• Printing data: Use PRINT or DUMPVECTOR to output calculated values to files.
• Conditional output: Use UPDATE_IF to wrap printing commands, ensuring data is only output when collective variables (colvars) fall within specific ranges.
• Trajectory manipulation: Use WHOLEMOLECULES or WRAPAROUND to adjust atom positions passed from the MD engine.
• Constant storage: Use CONSTANT or PDB2CONSTANT to store fixed values.
• Stopping criteria: Use COMMITTOR to stop a calculation once a criterion is met or to track how often a criterion is satisfied.

The ttsketch module implements Tensor Train Metadynamics (TT-MetaD). Unlike standard metadynamics that represents bias potentials as a sum of Gaussian hills, TT-MetaD represents the bias potential as a Tensor Train (TT).

It uses the TT-Sketch algorithm to periodically compress accumulated Gaussian hills into a low-rank TT approximation. This approach ensures that both memory usage and evaluation costs remain bounded, even during very long simulations.

The metatomic module provides an interface between PLUMED and the [metatomic] library. It allows users to use arbitrary machine learning models as collective variables (CVs).

Workflow:

1. Define custom machine learning models using Python code following the metatomic interface.
2. Export these models to TorchScript.
3. Load the exported TorchScript model into PLUMED to be executed during simulations.

The LOGMFD module implements methods for enhanced sampling in a Collective Variable (CV) space and for on-the-fly free energy reconstruction along those CVs.

Key capabilities include:

• LogMFD/LogPD methods: For enhanced sampling and free energy reconstruction.
• Multiple-replica algorithm: Implements the Crooks-Jarzynski non-equilibrium work relation.
• Single-replica algorithm.
• TAMD/d-AFED: Implementation of Temperature Accelerated Molecular Dynamics (TAMD) and d-AFED.

EDS is a module providing methods for adaptively determining linear bias parameters. Its primary goal is to ensure that each biased Collective Variable (CV) samples a new target mean value.

Key features include:

• Implementation of the stochastic gradient descent algorithm as described in the original EDS literature.
• Additional minimization algorithms designed for Coarse-Grained Directed Simulation.

Note: If you require different optimization algorithms or features, consider using the MAXENT method available in PLUMED.

If you are migrating from PLUMED 2.0 to 2.1, be aware that several action labels have changed to be more consistent with steered MD. You may need to adjust your input files to use the new label formats:

• BIASVALUE: Changed from label.bias.argname to label.argname_bias.
• ABMD: Changed from label.min_argname to label.argname_min.
• PIECEWISE: Changed from label.argnumber to label.argname_pfunc.
• CONTACTMAP: Components are now labelled label.contact-$n$.
• Multicolvars (LESS_THAN/MORE_THAN): Components are now labelled lessthan and morethan (previously used underscores which are now reserved for special usage).

The PYCV module provides two primary actions for defining Collective Variables (CVs) or functions in Python:

1. PYCVINTERFACE: Used to implement single- or multi-component CVs. It supports a full lifecycle including PREPARE and UPDATE steps and can handle atom selection (via ATOMS or neighbor list keywords).
2. PYFUNCTION: Used to implement arbitrary functions. It is simpler and only supports the INIT and CALCULATE steps.

Plumed embeds a Python interpreter and imports a module specified by the IMPORT= keyword. This module must contain an init object (function or dictionary) and a calculate function.