Web Throwing in OpenPathSampling

Authors: Sander Roet, Anders Lervik, Enrico Riccardi

This module implements the web throwing method in OpenPathSampling

Purpose of Module

Web throwing is a Monte Carlo move in path space designed to improve the efficiency of transition interface sampling (TIS). It consists of a smart selection of shooting points and shooting moves that respects super detailed balance. The web throwing algorithm generates paths that have a much lower correlation with their original paths, even if it is more computational expensive than standard shooting. The strategy thus can significantly reduce the computational time required to study transition events and to quantify their rates. The web throwing algorithm is described by Riccardi, Dahlen, and van Erp (http://dx.doi.org/10.1021/acs.jpclett.7b01617)

The web throwing method is an shooting method for transition interface sampling that decorrelates the trajectory between an interface \lambda and an associated surface of unlikely return \lambda_{\text{SOUR}}. It does this by doing n_cycles of Transition Path Sampling (TPS) from \lambda_{\text{SOUR}} to \lambda. The first frame in this volume is selected for a forwards shot and the last frame in this volume is selected for a backwards shot. After the n_cycles the new sub-trajectory is extended in both ways to satisfy the TIS ensemble. The sampling efficiency is increased significantly if \lambda_{\text{SOUR}} and \lambda are positioned before and after the barrier in the potential, respectively.

The implementation introduces the following new classes:

  • SecondFrameSelector inherits from ShootingPointSelector and selects the second frame (index = 1) from a trajectory.
  • SecondToLastFrameSelector inherits from ShootingPointSelector and selects the second to last frame of a trajectory (index = len(trajectory)-2).
  • WebEnsemble inherits from SequentialEnsemble and implements the sampling ensemble between the Volume \lambda_{\text{SOUR}} and the Volume \lambda in which the web throwing occurs.
  • WebBackwardExtendEnsemble inherits from SequentialEnsemble and implements the ensemble the partial path has to fit in after the backwards extend. This ensemble consists of one frame in the initial state followed by frames outside of any states. This should only be used in backward moves, as this only has an acceptable stop in this direction.
  • WebShootingMover inherits from RandomChoiceMover and implements the TPS shooting algorithm in the WebEnsemble using the SecondFrameSelector for the ForwardShootMover and the SecondToLastFrameSelector for the BackwardShootMover.
  • WebCycleMover inherits from SequentialMover and implements n_cycles sequential WebShootingMover in the WebEnsmeble.
  • WebExtendMover inherits from ConditionalSequentialMover and extend the sub trajectory after web throwing. It first simulates backwards to reach the initial TIS state. If this reaches any other state, the whole move is rejected. If this indeed finds the correct state the sub trajectory is extended in the forward direction to complete the TIS ensemble.
  • WebThrowingMover inherits from SubPathMover and is the canonical mover for the web throwing algorithm. From a TIS ensemble it first selects a random sub-trajectory in the web_ensemble. Then it calls the WebCycleMover with n_cycles. Finally it calls the WebExtendMover to extend the sub-trajectory back into the TIS ensemble.
  • WebThrowingStrategy inherits from MoveStrategy and it builds the WebThrowingMover for every {interface: lambda_sour} in the sours dictionary. Every interface should have only 1 lambda_sour associated with it.
  • WebThrowingScheme inherits from MoveScheme. This is a MoveScheme that only does web throwing. This is mostly a convenience class used in examples and testing.

Background Information

This module builds on OpenPathSampling, a Python package for path sampling simulations. To learn more about OpenPathSampling, you might be interested in reading:


The tests for this module can be run by installing OpenPathSampling, downloading source code for the module (see the Source Code section below), and installing it by running by running python setup.py install from the root directory of the package. Test this module with the nose package, by running the command nosetests from the root directory of the repository.


All examples can be found in the examples directory (https://gitlab.e-cam2020.eu/sroet/web_throwing/tree/master/examples). Open the ipython notebook (files with the extension .ipynb) of interest by running jupyter notebook file_of_interest.ipynb in this directory, replacing file_of_interest.ipynb by the file you want to open. (see Jupyter notebook documentation at http://jupyter.org/ for more details)

The examples are:

  • An example on how to set up a multiple interface set TIS (MISTIS) simulation with only web throwing and the analysis of the web throwing moves can be found in mistis_using_only_webthrowing.ipynb.
  • An example on how to add the web throwing to a default MISTIS simulation can be found in adding_webthrowing_to_mistis.ipynb. This example also shows how to change the selection weight of this move and some analysis.

Source Code

The source code for this module can be found in: https://gitlab.e-cam2020.eu/sroet/web_throwing/tree/master.