Bode_GAMESS-intro - Quantum Chemistry with GAMESS Brett M....

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Quantum Chemistry with GAMESS Brett M. Bode Scalable Computing Laboratory Department of Electrical and Computer Engineering Iowa State University Outline Introduction to GAMESS GAMESS history GAMESS capabilities Novel capabilities Running GAMESS 2 GAMESS General Atomic and Molecular Electronic Structure System General purpose electronic structure code Primary focus is on ab initio quantum chemistry calculations Also can do Density functional calculations Other semi-empirical calculations (AM1, PM3) QM/MM calculations Its free and in wide use on everything from laptops to supercomputers. 3 Obtaining GAMESS Its free, but not “Open source” in the normal sense. Group license: You get the source and can do anything your want with it, except distribute it. See http://wwwmsg.fi.ameslab.gov/GAMESS/ for more information and the registration page link. Distribution is source code, with pre-built binaries also available for Macintosh, Linux and Windows. Full manual also on web site. See section 2 for complete keyword description for the input file, section 4 for references for all of the methods. 4 GAMESS People GAMESS is a product of Dr. Mark Gordon’s research group at Iowa State University. Dr. Mike Schmidt coordinates the development efforts and is the gatekeeper for the code. 5 GAMESS History The code base was began in 1980 from parts of other codes. Some code still goes back to that version! Currently stands at about 750000 lines of mostly Fortran 77 compatible code. Pretty much runs on any system with a working Fortran compiler. 6 GAMESS Parallelization Began in 1991 with the parallelization of the SCF Energy and Gradient computations (almost trivially parallel). Initial parallel work done on the TouchStone Delta. In 1996 the Distributed Data Interface (DDI) was developed to support the new parallel MP2 energy and gradient code. 7 GAMESS Parallelization In 2004 DDI was rewritten and optimizations for SMP using SystemV shared memory were added. Focus remains distributed memory systems! Also added was subgroup support to enable the Fragment Molecular Orbital method. 8 memory is the memory reserved by all the remaining parallel processes for their portions of the distributed data. Every process in a parallel job is allowed to access/modify any element in the distributed memory segment (regardless of its physical location); however, access to local distributed-memory is assumed to be faster than access to remote distributed-memory. Thus the DDI programming strategy aims to maximize the use of local distributed data while minimizing remote data requests. Note that the performance penalty for accessing distributed-memory (local or remote) is completely dependent on the underlying machine with high performance and potentially intelligent interconnect networks like Gigabit Ethernet [9], Myrinet [10], SCI [11], or Infiniband [12]. A similar trend is also evident in dedicated supercomputers, where, for example large scale IBM SP and HP SC systems now use SMP nodes. Indeed, very large shared memory computers, like the SGI Origin 3000 or HP GS, usually have Non-Uniform Memory Access (NUMA) architectures that can be viewed as a cluster of uniform memory SMPs linked via a network, albeit a very good network. DDI With this move away from single processor toward multi-processor based clusters we are confronted with a considerably more complicated memory model than that which was present when either DDI or GA were originally conceived. Now small groups of processes have equally fast access to chunks of memory, while accessing memory between groups of processes is slower. Recognizing this plus the success and popularity of these programming models, it is pertinent to consider how these models might be extended to better exploit SMP clusters. The aim of this paper is to begin to address this issue, presenting an enhanced version of DDI that includes new functionality specifically targeting SMP clusters. Using both the new and original versions of DDI, performance results are presented and discussed for a typical GAMESS computation run on a variety of MPP systems. First, however, we begin with a brief discussion of the existing DDI data server model used in GAMESS. Figure 1: The virtual shared-memory model. Each large box (grey) represents the memory available to a given CPU. The inner boxes represent the memory used by the parallel processes (rank in lower right). The gold region depicts the memory reserved for the storage of distributed data. The arrows indicate memory access (through any means) for the distributed operations: get, put and accumulate. Modeled on the Global Array Framework. The Distributed Data Interface provides a pseudo global shared memory interface for a portion of a nodes memory. 2 Normal MPI version uses 2 processes per processor, 1 compute, 1 data server. Sockets are used for interrupts on data servers because MPI often polls in receive. SHMEM and LAPI versions also available... Also provides processor subgroup support. 9 Program Capabilities Types of wavefunctions Hartree-Fock (RHF, ROHF, UHF, GVB) CASSCF CI, MRCI Coupled cluster methods (closed shells) Second order perturbation theory MP2 (closed shells) ROMP2 (spin-correct open shells) UMP2 (unrestricted open shells MCQDPT(CASSCF - MRMP2) Localized orbitals (SCF, MCSCF) 10 Program Capabilities Energy-related properties Total energy as function of nuclear coordinates (PES): All wavefunction types Analytic energy gradient RHF, ROHF, UHF, MCSCF, CI, MP2, UMP2, DFT ROMP2 in progress Analytic hessian RHF, ROHF, TCSCF/GVB MCSCF just completed 11 Program Capabilities Energy-related properties (cont’d) Numerical hessians from finite differences of analytic gradients Fully numerical derivatives for all methods Saddle point (TS) search (requires hessian) Minimum energy path=Intrinsic Reaction Coordinate Several IRC options - GS2 is most effective Requires frequency input, gradients along path Follow reaction path from reactants through TS to products Build reaction path Hamiltonian (RPH): dynamics 12 Program Capabilities Energy-related properties (cont’d) Dynamic reaction coordinate (DRC) Add kinetic energy to system at any geometry Add photon(s) to any vibrational mode Classical trajectory using QM-derived energies Requires gradients Monte Carlo sampling: find global minimum Molecular dynamics (in progress) 13 Program Capabilities Other functionalities Spin-orbit coupling Any spin states, any number of states Full two-electron Breit-Pauli Partial two-electron (P2e)-very efficient, accurate Semi-empirical one-electron Zeff RESC Averaging over vibrational states Derivative (vibronic) coupling: planned 14 Program Capabilities Interpretive tools Localized molecular orbitals (LMO) Localized charge distributions (LCD) Nuclear and spectroscopic properties Spin densities at nucleus (ESR) NMR spin-spin couplings (in progress) NMR chemical shifts Polarizabilities, hyperpolarizabilities IR and Raman intensities Transition probabilities, Franck-Condon overlaps 15 Program Capabilities QM/MM Methods Effective fragment potential (EFP) method for Cluster studies of liquids Cluster studies of solvent effects Interfaced with continuum methods for study of liquids and solvation in bulk Covalent link for study of enzymes, proteins, materials SIMOMM: QM/MM method for surface chemistry QM part can be any method in GAMESS MM part from Tinker (Jay Ponder) 16 Current Capabilities Run Type Energy Analytic Gradient Numerical Hessian Analytic Hessian MP2 energy MP2 gradient CC Energy EOMCC CI energy CI gradient DFT energy DFT gradient MOPAC Energy MOPAC gradient RHF CDFP CDFP CDP CDP CDFP CDFP CDF CD CDP CD CDFP CDFP yes yes ROHF CDP CDP CDP CDP CDP DP CDP CDP CDP yes yes SCF Type UHF GVB CDP CDP CDP CDP CDP CDP CDP CDP CDP CDP CDP yes yes CDP yes - C= conventional storage of AO integrals on disk D= direct evaluation of AO integrals F= Fragment Molecular Orbital enabled P= parallel execution 17 MCSCF CDFP CDFP CDP CDP CP CDP - Solvation Solvation Methods Explicit vs. implicit methods Explicit Methods TIP3P, TIP4P SPC, SPC/E EFP Method for Solvation Summary of EFP1 method for water Generalized EFP Method (EFP2) 18 General Effective Fragment Potential Discrete solvation method Fragment potential is one electron contribution to the ab initio Hamiltonian Potentials are obtained by separate ab initio calculations depend on properties of isolated molecules can be systematically improved 19 General Effective Fragment Potential Effective Fragment Potential System is divided into System is divided into an ab initio region for the “solute” and an ab initio region for the “solute” and a ragment region he olvent molecules. a ffragment region for tforsthe solvent molecules. E = Eab initio + Einteraction 10 20 Hartree Fock based EFP Hartree Fock based EFP Interaction energy consists of : electrostatic, polarization and exchange repulsion/charge transfer term Einteraction= Ecoulomb + Epolarization + Eexchange repulsion/charge transfer K Einteraction = ! V k =1 Distributed Multipolar expansion Elec k L M R ( µ, s) + ! Vl (µ , s) + ! Vm e p (µ, s) Pol l =1 LMO polarizability expansion 21 m =1 Fit to Functional Form 12 EFP results gOH(r): EFP1/HF, EFP1/DFT, SPC/E 62 waters 1.8 1.6 1.4 1.2 gOH(r) 1 EFP1/HF EFP1/DFT SPC/E Exp (ND) 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 r (Angstroms) 6 7 22 Exp (ND): Neutron Diffraction; Soper et. al. 22 Generalized EFP2 Method Generalized EFP2 Method Interaction energy consists of : electrostatic, polarization and exchange repulsion term Einteraction= Eelectostatic + Epolarization + Eexchange repulsion Distributed Multipolar expansion LMO polarizability expansion 23 From first principles using LMO overlaps 26 EFP Performance Energy +EFPadient Calculation Gr Performance: Energy + Gradient Calculation Method1 20 water molecules 62 water molecules 122 water molecules 512 water molecules Ab initio2 3.19 hrs --- --- ~157 yrs3 EFP2 3.3 sec 26.1 sec 95.3 sec 26.8 min EFP1/HF 0.2 sec 2.6 sec 5.1 sec 97.8 sec SPC/E4 0.02 sec 0.02 sec 0.1 sec 0.7 sec 1Run 2Ab on 1200 MHz Athlon/Linux machine initio: DZP basis set, 3Assuming N4 scaling, 4SPC/E = Simple Extended Point Charge model 24 30 Fragment Molecular Orbital Divide up the system into fragments Basic idea Ignore exchange and self-consistency due to other fragments Do ab initio calculations of fragments in the Coulomb field due to the whole system. Dividebyhe system into fragments. work t Kitaura, Ishida and Likewise, compute pairs each fragment,atgnore exchange and s For Federov i AIST and triples of fragments. consistency due to other fragments but retain Coulomb field. Otherwise, do ab initio calcula 25 fragments and their n-mers. FMO Features No hydrogen caps. All n-mer calculations are ab initio. Interfragment charge transfer, dispersion and exchange are included. Systematic many-body effects. Total properties closely reproduce ab initio values. No fitted parameters. 26 FMO Can also add in electron correlation. MP2 Coupled Cluster DFT MCSCF Can be multilayer - ie MCSCF for active site, RHF everywhere else. 27 ! FMO-MCSCF !"#$"%&%!' "#$!%&#&%$'!()!*$+(#$*!,)!-./.01! ! "23$'!%&#&%$')!,'$!4501! "%' &%! ()!' ! 6(%$')!(#789*(#:!23$!-./.0! ! %&#&%$'!,'$!-./.0!;'$*!8(#$)<1! ! "23$'!*(%$')!,'$!450!;=89$!8(#$)<1! 28 ()! Applications of FMO !""#$%&'$()*+(,+-./+ ! 01(21'34+("'$2$*&'$()!"!#$%%%!&'()*+,! ! !"#$%$&$'!-./.-!*$)5#1+"($)'+1)135$1*!"!#%$%%%!&'()*+,! ! 6&$3+$)'13&%'$()+&)&#4*$*!"!#%$%%%!&'()*+0! ! 1234!1.*546$! ! -54&61!1(78564$! ! 9(-:).2!7;.)5*'2:$! ! )(-.73-&2!7-3*'.2*,! !! 29 FMO results !"#$%"&'()*&+,%-.'!"#$%$&'"$(")$)*+' ! ! ! /01234/562789:;'<=>?')@"-,'@$'6AA'1B%*-@$,' =A;?C8'#%@&,;'<<;<?D'*E*(%-@$,;'86D;DD='F#,+,'G"$(%+@$,>' ! "#!$%&'()*!!"#$%&+!,-.&/01).-2*3'!4556!78&92!2&0:3*0(;!.(.&/!(<(/=>#! 30 Running GAMESS GAMESS runs on Any Unix-based system available in the U.S. Any Linux based system Any Macintosh Windows based system using WinGAMESS or PCGAMESS Obtained from www.msg.ameslab.gov GAMESS GAMESS is a back-end program, ie no GUI. T ypically it is run via a script Input is taken from a file (usually .inp) Output appears in .log file (stdout) This is intended to be human readable MO Vectors, coordinates, hessians, etc appear in .dat file. Can be used for restarts. IRC and DRP data and numerical hessian restart information appear in .irc file. These are all ASCII text files. GAMESS Input file Input files are modular, arranged in $groups Most common input groups $SYSTEM: specifies memory, time limit $CONTRL: specifies basics of calculation $BASIS: specifies basis set if standard $DATA: specifies nuclear coordinates, basis set if nonstandard Other important groups: $GUESS, $SCF, $FORCE, $HESS, $VEC, $IRC, $VIB GAMESS Input file The input file is mostly free-format (ie flexible spacing) except: ‘$’ sign specifying group must be in column 2! All groups must terminate with a $END (this ‘$’ can be anywhere except column 1). anything in column 1 indicates a comment line Some key groups $SYSTEM group: TIMLIM=(default=525600 min = 1 yr) MWORDS=(default=1=8MB) MEMDDI= relevant for parallel run T otal required memory (divide by number of processors to get memory requested/node) Some key groups $CONTRL ICHARG= (specifies charge on system) MUL (specifies spin multiplcity) T= 1 for singlet, 2 for doublet, ... EXETYP= Check: checks input for errors Run: actual run UNITS= angs (default) bohr Some key groups $CONTRL Runtyp= (type of run) Energy (single point energy run) Gradient (energy 1st derivative wrt coordinates) Optimize (optimize geometry) Hessian (energy second derivative, vibrational frequencies, thermodynamic properties): generates $HESS group in .dat file) Sadpoint (saddle point search:requires hessian in $HESS group) IRC (performs IRC calculation: usually requires $IRC group, $HESS group) Some key groups $CONTRL scf typ= (type of wavefunction) RHF ROHF UHF MCSCF GVB mplevl= 0 (default, no perturbation theory) 2 (MP2: valid for RHF, ROHF, MCSCF) Some key groups $CONTRL Cctyp= NONE (no coupled cluster, default) LCCD (linearized doubles CC) CCD (doubles CC) CCSD (singles+doubles) CCSD(T) adds perturbative triples to CCSD Most popular method T riples essential for accurate calculations R-CC, CR-CC Specialized methods to approximate bond-breaking Some key groups $BASIS - Used to select among the built-in basis sets GBASIS= STO N21 N31 TZV... NGAUSS=(# gaussians for STO, N21, N31) NDFUNC=(# sets of d’s on heavy atoms NPFUNC=(# sets of p’s on hydrogens) NFFUNC=(# sets of f ’s on TM’s) DIFFSP=.T. (diffuse sp functions on heavy atoms) DIFFS=.T. (diffuse s functions on hydrogens) Some key groups $Data - Gives the molecular geometry Title line (will be printed in output) Symmetry group C1 CS CNV 2 (C2V), ... Blank line except C1 Symbol Z xcoord ycoord zcoord Symbol = atomic symbol Z = atomic number xcoord,ycoord, zcoord = Cartesian coordinates Internal coordinates is another option Some key groups $Data - continued Repeat this line for each symmetry unique atom (see below) Need to specify basis set after each coordinate line if $BASIS is not present symmetry unique atoms H2O: O and 1 H NH3: N and 1 H saves CPU time (e.g., numerical hessians only displace symmetry unique atoms) Need to follow conventions in GAMESS manual Cs, Cnh: plane is XY Cnv: axis is Z For Cinfv, use C4v For Dinfh, use D4h Some key groups $GUESS - Initial MO guess Built-in guess (default) works much of the time $GUESS=MOREAD NORB=xx $END Requires $VEC group (usually from .dat file) NORB=# MO’s to be read in Useful when SCF convergence is difficult Necessary for MCSCF, CI GAMESS output The log file output is intended to be human readable: -------------------------RHF SCF CALCULATION -------------------------- NUCLEAR ENERGY = 8.9064898741 MAXIT = 30 NPUNCH= 2 EXTRAP=T DAMP=F SHIFT=F RSTRCT=F DIIS=F DEM=F SOSCF=F DENSITY MATRIX CONV= 1.00E-05 MEMORY REQUIRED FOR RHF STEP= 30441 WORDS. ITER EX DEM TOTAL ENERGY E CHANGE DENSITY CHANGE DIIS ERROR 1 0 0 -74.7936151096 -74.7936151096 .595010038 .000000000 2 1 0 -74.9519661838 -.1583510742 .180249713 .000000000 ... 11 6 0 -74.9659012167 -.0000000014 .000018538 .000000000 12 7 0 -74.9659012170 -.0000000003 .000008228 .000000000 13 8 0 -74.9659012171 -.0000000001 .000003650 .000000000 ----------------DENSITY CONVERGED ----------------TIME TO FORM FOCK OPERATORS= .0 SECONDS ( .0 SEC/ITER) TIME TO SOLVE SCF EQUATIONS= .0 SECONDS ( .0 SEC/ITER) FINAL RHF ENERGY IS -74.9659012171 AFTER 13 ITERATIONS The Dat file The dat file contains formatted numerical data. Useful, sometimes required for restarts. Contains items such as: MO Vectors ($VEC) Gradient ($GRAD) and Hessian ($HESS) When copying a group make sure you copy everything from the beginning $ sign through the corresponding $END. GAMESS output You will need to look at the log file to verify the results. Did the run finish correctly? Was the input specified correctly? Were there errors in the computation? Running GAMESS You frequently need the results from one run as input to another run. restarting incomplete runs Multi step problems A Saddle point search might take several optimization and hessian computations followed by IRC computations. Multi-reference computations often multiple runs to get the orbital guess correct. Visualization A number of programs can visualize GAMESS results to varying degrees. MacMolPlt is one such program that has been specifically designed for visualizing GAMESS output. Demo This af ternoon I will present a demo of running GAMESS and using MacMolPlt. Acknowledgments Mark Gordon Dmitri Federov the rest of the Gordon group in Ames Financial Support Air Force Office of Scientific Research National Science Foundation DoD CHSSI Software Development DOE SciDAC Program Ames Laboratory DoD HPC Grand Challenge Program 51 ...
View Full Document

This note was uploaded on 12/07/2011 for the course CHEM 350 taught by Professor Duanejohnson during the Summer '06 term at University of Illinois, Urbana Champaign.

Ask a homework question - tutors are online