M.DynaMix - a scalable portable parallel MD simulation package for arbitrary molecular mixtures. A.P. Lyubartsev, A. Laaksonen.

PROGRAM SUMMARY
Title of program: M.DynaMix
Catalogue identifier: ADLW
Ref. in CPC: 128(2000)565
Distribution format: tar gzip file
Operating system: UNIX, Windows 98/NT
High speed store required: 64MK words
Number of lines in distributed program, including test data, etc: 24216
Keywords: Parallel algorithms, Molecular dynamics, Computer simulations, Solid state physics, Other.
Programming language used: Fortran, C
Computer: IBM RISC 600 , DEC Alpha , Pentium PC , Cray T3E , IBM SP2 , Linux SMP , Linux network cluster , Fujitsu VX .

Nature of physical problem:
Many-body problem with interacting particles. Structural, thermodynamical and dynamical properties of molecular liquids and liquid mixtures, including organic molecules or biomacromolecules as solutes.

Method of solution
Numerical integration of classical (Newtonian) equations of motion, optionally modified for constant temperature or/and constant pressure simulations. Forces are calculated from standard molecular-mechanical force fields.

Restrictions on the complexity of the problem
The principal limitation is the size of the non-bonded neighbour lists, but it rarely reaches the limits of RAM memory available. In practice, typical systems consist of about 10**4 atoms, covered during several nanoseconds on modern parallel computers (of Cray T3E, IBM SP2 type).

Typical running time
Varies greatly depending on the complexity of the problem. Typically for a system of 10000 atoms, a simulation of 1 nano-second would take several days of cpu time using 16-32 processors.

Unusual features of the program
The memory required to execute with typical data depends on the size of the simulated system, simulation parameters and architecture in hand (some data structures use distributed memory). As an example for a 10000 atoms system on a single-processor computer, 64MB is suggested. The number of processors used is arbitrary, but effective speedup depends on the communication bandwidth. For parallel computers with distributed memory (IBM SP2, Cray T3E) very good scaling is observed up to 32 processors. In general, the scaling properties increase with the number of particles.