Differentiable Simulations for Enhanced Sampling of Rare Events
Authors: Martin Sipka, Johannes C. B. Dietschreit, Lukáš Grajciar, Rafael Gomez-Bombarelli
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The potential of Diff Sim is demonstrated in the successful discovery of transition paths for the Muller-Brown model potential as well as a benchmark chemical system alanine dipeptide. In Section 5, we demonstrate the usefulness of Diff Sims in the context of chemical reactions by training the bias function promoting barrier crossing for the well-studied Muller-Brown potential as well as the alanine-dipeptide molecule. |
| Researcher Affiliation | Academia | 1Department of Materials Science and Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA 2Mathematical Institute, Faculty of Mathematics and Physics, Charles University, Sokolovsk a 83, 186 75 Prague, Czech Republic 3Department of Physical and Macromolecular 853 Chemistry, Faculty of Sciences, Charles University, 128 43 854 Prague 2, Czech Republic. |
| Pseudocode | No | The paper describes steps in numbered list format in '4. Practical implementation' but does not include a structured pseudocode or algorithm block. |
| Open Source Code | Yes | The source code for all examples (2D and 5D Muller-Brown, Alanine Dipeptide) are available online on Github https: //github.com/martinsipka/rarediffsim |
| Open Datasets | Yes | We apply it to a commonly used two dimensional model PES, the Muller-Brown potential(M uller & Brown, 1979), where any linear combination of the Cartesian coordinates does not yield a good CV. Second, we investigate the benchmark system for enhanced sampling in molecular systems, alanine dipeptide (amino acid alanine capped at both ends). |
| Dataset Splits | No | The paper does not explicitly provide traditional training/validation/test dataset splits with percentages or sample counts, as the data is generated through simulations rather than being a static dataset. |
| Hardware Specification | No | No specific hardware details (e.g., GPU models, CPU models, or memory specifications) used for running the experiments are provided in the paper. |
| Software Dependencies | No | The paper mentions 'Torch MD library' and 'Amber ff19SB forcefield' and 'torchdiffeq python package' but does not provide specific version numbers for these software dependencies. |
| Experiment Setup | Yes | The equations we simulate are (3), discretized by the Leapfrog algorithm. The method is symplectic and conserves energy. The constants and parameters of the method were chosen as follows: case m [g/mol] γ [ps 1] T [K] dt [fs] timesteps epochs 2D 0.1 0.1 10 1 6 103 101 5D 0.01 1.0 300 1 2 104 66 Ala2 0.1 300 1 104 301. The setup of the bias function differed for every test case: ... 5d Muller-brown: We use a fully connected network with all five degrees of freedom used as five continuous input neurons. The network has four hidden layers, each 150 neurons with Si LU as activation functions. The final layer has a single output neuron the bias and no activation. We use Adam optimizer for all our cases. |