Wujie Wang's website

Bio:

I am Wujie Wang, a 4th year graduate student at MIT in Materials Sc. & Eng. I hold a B.A. in Physics from Wesleyan and B.S in Engineering and Applied Sciences from Caltech. I am from Nanjing, China and attended Nanjing Foreign Language School. 

I am maintaining this website to showcase my work. 

Research Interests/Expertise:

  • Differentiable simulations for control and learning tasks
  • Develop computational methods to learn interpretable and symmetry preserving representation of molecular systems 
  • Emergent information processing capacity in out-of-equilibrium molecular systems 
  • High throughput screening of molecules/polymers, particularly, for electrolytes 
  • Energy-Based Models(EBM) for variational inference 

Notes:

  • basic thermodynamics

   https://github.com/wwang2/thermo-notes/blob/main/thermodynamics/StatMechNotes.pdf

  • Apply Automatic Differentiation to compute the Virial Stress Tensor

  https://github.com/wwang2/thermo-notes/blob/main/AutoDiff/On_computing_Virials_with_AD.pdf

Selected Projects:

NeuralForceField:

Graph convolution models for molecular dynamics 

https://github.com/learningmatter-mit/NeuralForceField

Torchmd:

A package for differentiable molecular simulations in PyTorch  

https://github.com/wwang2/torchmd

Coarse-Graining Auto-Encoders

Code for learning for coarser represenation from Molecular Dynamics data

 https://github.com/wwang2/Coarse-Graining-Auto-encoders

Selected Demos:

    

controllable polymer fold with Graph Neural Networks (Ref. [1])

      

Differentiable simulations: learning Hamiltonains to reporduce water pair correlations (Ref. [1])

   

ML-exploration of chemical space for ether-based Solvate Ionic Liquids (Ref. [2])

Unsupervised learning for Coarse-Grained Modeling (Ref. [3])

Papers/Preprints:

[1] Wang, W., Axelrod, S., & Gómez-Bombarelli, R. (2020). Differentiable Molecular Simulations for Control and Learning. ArXiv. Retrieved from https://arxiv.org/abs/2003.00868

[2] Wang WYang THarris WGómez-Bombarelli RActive learning and neural network potentials accelerate molecular screening of ether-based solvate ionic liquids. Chem. Commun., 2020,56, 8920-8923 

[3] Ang, S. J., Wang, W., Schwalbe-Koda, D., Axelrod, S., & Gomez-Bombarelli, R. (2020). Active Learning Accelerates Ab Initio Molecular Dynamics on Pericyclic Reactive Energy Surfaces. ChemRxiv. Preprint. https://doi.org/10.26434/chemrxiv.11910948.v1

[4] Wang, W., & Gómez-Bombarelli, R. (2019). Coarse-graining auto-encoders for molecular dynamics. Npj Computational Materials5(1). https://doi.org/10.1038/s41524-019-0261-5

[5] Wang, W., Nocka, L. M., Wiemann, B. Z., Hinckley, D. M., Mukerji, I., & Starr, F. W. (2016). Holliday Junction Thermodynamics and Structure: Coarse-Grained Simulations and Experiments. Scientific Reports6(3), 3–4. https://doi.org/10.1038/srep22863