Release Data: Dec. 11, 2014
Git tag "6.2" (https://github.com/pandegroup/openmm/releases/tag/6.2)

Improved Performance
====================

The main focus of this release is on performance.  It contains dozens
of different optimizations throughout the code.  The CUDA and OpenCL
platforms both have large, across the board speed improvements.  The
CPU platform now has a just-in-time compiler for custom expressions,
so most custom forces are dramatically faster.  Multi-GPU performance
on the CUDA platform is significantly improved.  And so on.

In addition to the improved computation speed, there are a few other
significant new features in this release:

Membrane Barostat
=================

There is a new barostat called MonteCarloMembraneBarostat that
implements features specifically needed for membrane simulations.  For
example, it can apply a surface tension in the plane of the membrane,
and supports new types of Monte Carlo moves, such as semi-isotropic
moves and volume preserving moves.

CustomManyParticleForce
=======================

There is a new custom force called CustomManyParticleForce.  It can
implement a wide range of non-pairwise nonbonded forces, where the
interaction energy depends explicitly on more than two particles at a
time.

PDBx/mmCIF Support
==================

There is a reader for PDBx/mmCIF files.  This is a file format that is
being pushed by wwPDB as a replacement for the old PDB format.  To use
the new reader, just specify the PDBxFile class in place of PDBFile.

64 Bit Only
===========

One important change to be aware of: in this release, the precompiled
binaries we provide for Linux and Mac OS X are 64 bit only.  This
change was largely forced on us by the fact that NVIDIA is dropping
support for 32 bit binaries on those platforms.  Using CUDA 6.5, it is
impossible to compile a 32 bit version of the CUDA platform on them. 
In any case, 32 bit Linux is becoming increasingly rare, and all
versions of OS X for quite a while have supported 64 bit binaries.  If
you compile from source, you can still build 32 bit binaries as long
as you don't need CUDA (or build against an older version of CUDA). 
At least for the moment, we are continuing to provide 32 bit binaries
for Windows.

Bugs Fixed in This Release
==========================

The following bugs that were present in 6.1 are fixed in this release
(see https://simtk.org/tracker/?atid=435&group_id=161&func=browse):

2031: getInducedDipoles() is incorrect on CUDA platform
2034: Uninitialized memory using PME with multiple GPUs
2057: Crash in GBSA on CPU platform
2058: Inconsistent cutoffs in CHARMM polarizable force field
2059: Drude particles sometimes put at wrong location

Known Issues
============

The following known issues that could potentially affect results are
still present in 6.2 (see
https://simtk.org/tracker/?atid=435&group_id=161&func=browse):

1948: Degrees incorrectly converted to radians when using
AmoebaAngleForce with constraints.
2060: Uninitialized memory in variable integrators
2061: Unexpected behavior of arrays with units