Skip to content

lgsmith/timemachine

 
 

Repository files navigation

CircleCI

Time Machine

This package is designed with three goals in mind:

  1. Enable rapid prototyping of potential functions and automatic generation of gradients of arbitrary order.
  2. Generate performant code using a jax/xla backend that can be run on CPUs, GPUs, and TPUs.
  3. Meta-optimization of optimizers and integrators, allowing one to generate analytic derivatives of a trajectory.

The code is implemented against OpenMM for numerical accuracy.

Example Code:

import functools
import numpy as np
import jax

from timemachine.potentials import bonded

x0 = np.array([
    [1.0, 0.2, 3.3], # H 
    [-0.5,-1.1,-0.9], # C
    [3.4, 5.5, 0.2], # H 
], dtype=np.float64)

params = np.array([10.0, 3.0, 5.5], dtype=np.float64)

param_idxs = np.array([
    [0,1],
    [1,2],
], dtype=np.int32)

bond_idxs = np.array([
    [0,1],
    [1,2]
], dtype=np.int32)

# wrap the energy function for convenience
energy_fn = functools.partial(
    bonded.harmonic_bond,
    param_idxs=param_idxs,
    bond_idxs=bond_idxs)

# dE/dx, shape [N,3]:
dedx_fn = jax.grad(energy_fn, argnums=(0,))
dedx_fn(x0, params, box=None)

# dE/dparams, shape [3,]:
dedp_fn = jax.grad(energy_fn, argnums=(1,))
dedp_fn(x0, params, box=None)

# d^2E/dx^2, shape [N,3,N,3]:
d2edx2_fn = jax.hessian(energy_fn, argnums=(0,))
d2edx2_fn(x0, params, box=None)

# d^2E/dxde, shape [N,3,3]:
d2edxde_fn = jax.jacfwd(jax.jacrev(energy_fn, argnums=(0,)), argnums=(1,))
d2edxde_fn(x0, params, box=None)

Warning

This code is under heavy development. APIs for potential energies are fairly stable now.

Supported Potentials

We currently support the following functional forms:

  • (Periodic) Harmonic Bond
  • (Periodic) Harmonic Angle
  • (Periodic) Periodic Torsion
  • (Periodic) Electrostatic
  • (Periodic) Lennard-Jones
  • GBSA OBC

Requirements

See requirements.txt

License

Copyright 2019 Yutong Zhao

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

About

Differentiate all the things!

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 81.2%
  • Python 13.1%
  • Cuda 4.4%
  • C++ 1.1%
  • Other 0.2%