This is the version with improved speed.
The emphasis shifted toward maximum efficiency when working on the graphics card (GPU).
Monte Carlo Look Ahead Hamiltonian Monte Carlo (LAHMC) algorithm of Sohl-Dickstein et al increases the probability of acceptance of HMC steps. In practice, it is in 2-3 times more efficient than the classical HMC.
In Abalone algorithm was implemented as the Extra Chance Generalized Hybrid Monte Carlo (xHMC) version of Campos and Sanz-Serna. As an integrator of molecular dynamics steps a two-step algorithm of Tuckerman et al (r-RESPA) was used.
xHMC is enabled by default on the MC panel. The algorithm is implemented effectively on the GPU (CUDA) and in many cases preferable than the molecular dynamics.
Molecular dynamics The default parameters of molecular dynamics have been changed in order to work more efficiently on the GPU. So the default RESPA algorithm was transferred from 2x3x1 to 2x1x3 mode. This means that, instead of the two levels of the not valence forces, the one level of the not valence forces and the one level of the valence forces are used. This improves GPU acceleration about twice. Thermodynamic properties of some suffering. However, one should keep in mind that examination of thermodynamic properties preferable in the HMC in any case.
Quantum chemistry Quantum chemical methods using NWChem now can be run in parallel not only from a command line, but also from a graphical environment.
Also, a number of minor improvements were made, such as the calculation of electrostatic moments until octupole etc.