INFO
Team
Moscito is written and maintained by the Moscito Team at the Phys. Chem. Department at TU Dortmund under the guidance of Dr. Dietmar Paschek. We would like to thank the following people who have contributed to Moscito in any possible way: Andreas Appelhagen, Ralf Baumert, Oliver Biermann, Robert Bieshaar, Frank Eikelschulte, Ingo Köper, Sascha Nonn, Amelie Rehtanz, Frank Schmauder and Ralf Schmelter. Ralf Schmelter has to be thanked extensively for contributing the Intel Assembler coding of force_ew_fast and some tuning of PME- and neighborlist routines. Sascha Nonn has contributed a X11-based crd-file trajectory viewer.
Forcefield
The freely available AMBER/Cornell et al. forcefield (parm94/parm96) is ready to be used together with the Moscito forcefield setup tool ffparse. Both are delivered with the current Moscito distribution. Moscito supports the AMBER open forcefield file format (OFF). We suggest Leap as a rather convinient tool for constructing molecular forcefields. In addition, a nice tool to build and handle molecular units is MOLDEN written by G. Schaftenaar.
MD-Algorithm
The time-reversible Verlet leapfrog algorithm is employed for propagation. Intramolecular distance constraints are handled by SHAKE. The pressure tensor is rigorously defined within a molecular center of mass framework, thus avoiding intramolecular correction terms due to the presence of constraints. A weak coupling scheme can be used to equilibrate a system close to a given (P,T) state point. The box lengths can be allowed to relax individually.
Ewald Summation
The heart of the matter are the force routines which are all based on the Ewald sum approach. Two different techniques have been implemented. One can either choose between a conventional reciprocal lattice sum and a FFT-based smooth particle mesh Ewald (SPME) method using cardinal B-spline interpolation. The highly efficient PME-routines were kindly provided by T. Darden (darden@t-rex.niehs.nih.gov). We employ the reasonably fast FFTW-routines for 3D-Fast-Fourier transformation. For system sizes below 50000 atoms the CPU-cost scales almost linearly with system size. The real-space part of the Coulomb-interaction can be interpolated from lookup tables.
Force Shifting
The Moscito MD code is atomistic, thus generally each "normal" center of interaction has to exhibit a non-vanishing mass. However, additional massless "virtual" sites can be introduced which have to be defined by a rigid framework of normal sites. A force shifting procedure that conserves total force and torque redistributes the forces. So, for example popular 4- and 5-center water models like TIP4P or TIP5P can be used.
Parallel MD
A parallel version of the program (moscito-net) based on the replicated data model is part of the distribution. The implementation uses the message passing interface (MPI) for communication. Using the LAM implementation we achieve a reasonable performance. Unfortunately, at the moment the PME is the most critical bottlneck for better parallelization.
Performance and Testing
The program has been thoroughly tested on Linux (i486/Pentium/P2/P3/P4 based), IBM RS/6000, DEC Alpha Server/Station and Sun machines. It performs well on PowerPC 604 based workstations and rather excellent on DEC/Alphas. In addition, Intel-Assembler coded nonbonded force routines make Moscito highly efficient on Intel/Linux-systems. (e.g. compare AMBER 5.0 4096wat with Moscito BIG-SPCE timings).