[LCRC Accounts] Project Allocation Request
Hello, A change in allocation has been requested: Requester: jlow (John J. Low) Project: metadissolve Title: Ab Initio Molecular Dynamics Simulations of Aqueous Reactions of AluminoSilicates Description: The project will predict the rates of aqueous polymerization reactions of aluminosilicates with ab initio molecular dynamics. We will generate free energy surfaces for dimerization reactions of SiOH4 + SiOH4, Al(OH)3 + SiOH4 and B(OH3) + SiOH4 in water and vacuum. Each of these reactions will carried in pH neutral, acidic (1M HCl) and basic (1M NaOH) solutions to model the effect of pH on these reactions. We will use metadynamics to define the reaction coordinate and estimates of free energies. Umbrella sampling will be use to predict accurate free energies and estimates of rates. CPMD (http://www.cpmd.org), PLUMED (http://www.plumed-code.org) and WHAM (http://membrane.urmc.rochester.edu/content/wham) will be used in this project. CPMD is a plane-wave pseudopotential code which exploits Carr-Parrinello Molecular Dynamics. CPMD is very efficient method for ab initio molecular dynamics. PLUMED is an open source plugin, which works together with CPMD, for free energy calculations in molecular systems. Although CPMD can do metadynamics, PLUMED implements the reaction path collective variable (RPCV). The RPCV should increase the efficiency of metadynamics and umbrella sampling. WHAM is a program which performs a weighted histogram analysis of the trajectory from umbrella sampling to generate the free energy changes along the reaction coordinate. CPMD is a highly scalable parallel code. The hybrid MPI/openMP Version of CPMD 3.15.3 scales very well. CPMD can use 560 cores on either blues or fusion very efficiently for these models considered in this project. Additional scaling can be obtained with multiple walkers and task groups, which is available in CPMD. This will enable additional levels of parallelism could scale efficiently to thousands of cores. PLUMED implements a reaction path collective variable will allow us to simulate this reaction with one dimensional metadynamics. One dimensional metadynamics is at least two to three times more efficient than two or three dimensional sampling. PLUMED will increase the efficiency of our metadynamics and umbrella sampling by a factor of two to three. We expect that there will three members of this project. Current: undetermined amount Justification: CPMD is well known as a program with efficient parallel scaling. I have documented the scaling of CPMD in previous project reports for aimd_catalysis. The table below show the scaling of CPMD for Cellobiose (C12O11H22) dissolved in 1M HCL (6HCl + 256 H2O) for a total on 569 atoms. CPMD scaling on blues on cellobiose in 1M HCl. MPI OMP Task # of elasped time Core cores (seconds) 216 1 1 216 31138 216 4 1 864 9180 432 4 4 1728 5695 432 8 4 3456 4290 The new version of CPMD (3.17.1) is claimed to scale better then 3.15.3. Using the new version will likely increase the efficiency of the code. It is possible that performance profiling of the code will reveal opportunities for vectorization. We study the CPU and communications bottlenecks of the code to determine whether code modifications may improve performance. Requested: 500000 A specific reason has been given: A large fraction of the allocation was used in production runs for another project "SRF-materials". Now that cp2k has been modified to enable these calculations, the requested time will be used to complete original work. This needs to be approved and the final allocation amount decided upon. Thank You, The LCRC Accounts System
participants (1)
-
accounts@lcrc.anl.gov