[Openmm-betatesters] OpenMM 7.0 beta

[Peter Eastman]

A beta of OpenMM 7.0 is now available.  This is a major update with lots of new features.  Here are some of the most important ones.

We've done a lot of work on optimization, and you should see significant across the board performance improvements.  We've especially worked hard on performance for AMOEBA, including switching to a new algorithm for computing interactions between multipoles.  We also added a new "extrapolated" polarization method, which performs just a few iterations of the induced dipoles, then tries to extrapolate them to their fully converged values.  This produces a large speedup with only a very small decrease in accuracy.

CustomIntegrator now supports complex flow control, including "if" and "while" blocks.  This greatly extends the range of algorithms you can implement with it.

You can now save states or trajectories to PDBx/mmCIF files.

There's a new class called CustomCentroidBondForce, which implements forces between the centers of clusters of atoms.  It's especially useful for implementing various types of restraints that operate between groups of atoms rather than individual atoms.

There's also a new CompoundIntegrator class.  It lets you have multiple integrators for one simulation, and switch between them freely.  That's useful for simulation protocols where you need to switch back and forth between different algorithms.

When running simulations in mixed precision mode, energy accumulation is now done in double precision.  That means you can now get much more accurate energies without having to sacrifice speed.

Starting in this release, we're recommending the conda package manager as the preferred way of installing OpenMM.  To use it, you first need to install either the Anaconda Python distribution (https://www.continuum.io/downloads) or Miniconda (http://conda.pydata.org/miniconda.html).  Then just type the following command:

conda install -c omnia openmm-beta

We do still provide the traditional installers, which are available on the download page.  (Click the "List all hidden releases" link to get to the beta of 7.0).  Eventually, though, we hope to move to conda as the exclusive means of distributing precompiled binaries, so please let us know how it works for you.

The binaries for this release are compiled against CUDA 7.5.  You can still compile from source if you need to use an earlier version, but the precompiled binaries only work with CUDA 7.5.

Give it a try, and please let us know about any problems you find.

Peter