Quantum Chemistry and Reaction Dynamics

busyicicleΜηχανική

22 Φεβ 2014 (πριν από 3 χρόνια και 3 μήνες)

69 εμφανίσεις

Basic Energy Sciences

Quantum Chemistry and Reaction
Dynamics

Jeff Nichols, Deputy Director

WR Wiley Environmental Molecular Sciences Laboratory

Pacific Northwest National Laboratory

Richland, WA 99352

Dr. Robert J. Harrison, Pacific Northwest National Laboratory

Dr. Martin Head
-
Gordon, Lawrence Berkeley National Laboratory

Dr. Mark S. Gordon, Ames Laboratory

Professor Piotr Piecuch, Michigan State University

Professor Peter R. Taylor, University of California, San Diego

Professor Henry Frederick Schaefer III, University of Georgia


Professor Gustavo E. Scuseria, Rice University



Professor Russ Pitzer, The Ohio State University

Dr. Walter C. Ermler, Program Director for Research, EHR/REC



Dr. Albert F. Wagner, Argonne National Laboratory

Professor Donald L. Thompson, Oklahoma State University

QC

RD

Basic Energy Sciences

Some Collaborations Have Already Been
Established

Ames, MSU

PNNL, LBNL,

UGA, UCSD

OkSU

ANL

UMd

SNL

Rice

OSU, U of Mem

ISICs

Colaboratory Pilot (CMCS)

Basic Energy Sciences

Advanced Methods for Electronic Structure

Robert J. Harrison and Jeff Nichols, PI


Collaborators


GI Fann, E Apra, PNNL


G Beylkin, U Colorado


HF Schaefer, III, UGA


PR Taylor, UCSD/U Warwick


M Head
-
Gordon, LBNL

One of three complementary projects:


Sandia National Laboratory,
“A Computational Facility for Reacting Flow Science,”

and


Argonne National Laboratory,
“Advanced Software for the Calculation of Thermochemistry,
Kinetics, and Dynamics.”


to address electronic structure, chemical kinetics, and fluid mechanics issues necessary to create a software
revolution in the simulation of chemically reacting turbulent flows. To the degree that funding allows, our
intention is to initiate a collaboration that will grow over time into a coordinated software development
program.


Basic Energy Sciences

A Multiresolution Approach to Quantum Chemistry in

Multiwavelet Bases


Complete elimination of the basis error



One
-
electron models (HF, DFT), pair models (e.g., MP2, CCSD)


Correct scaling of cost with system size


General approach


readily accessible by students, small computer code


Potential impact similar to the FFT on physics


Realizable numerical algebraic regularization


Faster algorithms with separable representation of kernels for integro
-
differential
equations


Applications to other domains including atmospheric dynamics and materials

Robert J. Harrison and G.I. Fann



Gregory Beylkin

Pacific Northwest National Laboratory


University of Colorado

Basic Energy Sciences

A Multiresolution Approach to Quantum Chemistry in

Multiwavelet Bases


Harrison, Fann and Beylkin


Distinguishing features


Multiresolution analysis in multiwavelet bases


Disjoint

intervals efficiently adapt to singularities


Non
-
standard representation of functions / operators


Integral operators scaling as
O(Nk
4
)

or better


Density Functional Theory


Integral formulation


nominally no derivatives


Initial applications


Benchmark calculations on polyatomic systems


Connections to ISICs


MRA in 3
-
D and 6
-
D; visualization of 3
-
D and 6
-
D numerical functions; NUMA
programming tools; sparse matrix BLAS and linear algebra,
CCA and
performance analysis and tuning


Basic Energy Sciences


Grid used to represent the nuclear
potential for H
2

using k=7 to a
precision of 10
-
5
.



Automatically adapts


it does not
know a priori where the nuclei are.



Nuclei at dyadic points on level 5


refinement stops at level 8



If were at non
-
dyadic points
refinement continues (to level ~12) but
the precision is still guaranteed.



In future will dyadically refine with
various lengths to force nuclei to box
corners.

A Multiresolution Approach to Quantum Chemistry in

Multiwavelet Bases


Harrison, Fann and Beylkin

Basic Energy Sciences

Accurate Properties for Open
-
Shell States of Large Molecules

Peter R. Taylor, UCSD/ U. Warwick

Objectives


Extend the applicability of accurate, reliable chemical
predictions for closed
-
shell molecules to radicals, excited
states, and situations in which chemical bonds are broken


Develop response methods for calculating properties for
these systems


Improve convergence of results by exploiting two
-
particle
basis functions


Extend this approach to systems with dozens of open
-
shell
electrons

Basic Energy Sciences

Accurate Properties for Open
-
Shell States of Large Molecules

Peter R. Taylor, UCSD/ U. Warwick

Approach


Implement scalable CASPT2 energies and response
properties (new software)


Integrate existing work on CASPT2 using two
-
particle
basis functions (existing software)


Develop and implement group function approach for
weakly coupled open
-
shell sites such as polynuclear
transition
-
metal complexes or ferredoxins (new equations,
new algorithms, new software)

Basic Energy Sciences

Development of Next
-
Generation, Explicitly
-
Correlated
Electronic Structure Methods for Sub
-
Chemical Accuracy

Fritz Schaefer, University of Georgia


New closed
-
shell, integral
-
direct MP
n
-
R12 and CC
n
-
R12 codes are being created to
take advantage of our recently developed integrals package CINTS. Our MP2
-
R12
work has already provided the scientific community explicitly
-
correlated methodologies
for workstation computations with as many as 1500 basis functions and essentially no
angular momentum limits, at least through
k

functions.


The

full

promise

of

R
12

methods

is

being

pursued

by

further

development

of

evaluation

methods

for

many
-
electron

integrals
.

Through

demanding

computations

with

enormous

basis

sets,

deficiencies

in

the

standard

approximations

of

R
12

theory

are

being

identified

and

understood
.

Improvements,

such

as

dual

basis

set

schemes,

are

being

investigated

for

better

completeness

insertions,

particularly

ones

with

improved

consistency

over

broad

regions

of

potential

energy

hypersurfaces
.


For

combustion

and

atmospheric

chemistry

applications,

we

plan

critical

development

and

implementation

of

explicitly
-
correlated

(R
12
)

open
-
shell

perturbation

and

coupled
-
cluster

theories,

both

within

less

complicated

unrestricted

(UHF)

formalisms

and

more

difficult

spin
-
adapted

(ROHF)

approaches
.


Basic Energy Sciences

Local Correlation


Martin Head
-
Gordon, LBNL


Coupled cluster

(CC) methods are the
most accurate

quantum chemistry
methods in wide use.


They are accurate for reaction barriers where simpler density functional
theory (DFT) methods often fail.


But

CC methods are restricted to small molecules, and are
very

computationally expensive compared to DFT.


Application to large molecules is blocked because cost rises with the 6
th

and 7
th

powers of molecular size.


Thus immediate development of parallel algorithms is
not

going to
permit application to larger molecules.


Goal:

develop alternative formulations of CC theory which reduce the
cost scaling to quadratic or linear with molecular size, without
destroying accuracy.

Basic Energy Sciences

Local Correlation
-

Overview of Planned Work

Martin Head
-
Gordon, LBNL


Our reduced scaling CC methods must permit:


Continuous potential surfaces.


Viability with large basis sets.


Both energies and forces must be available.


We’ve defined a new ansatz that reduces the number of CC variables to
quadratic, and satisfies the above criteria.


The accuracy of this ansatz must be established


We’ll avoid the difficulty of very large basis sets by combining our CC
methods with a density functional style correlation functional.


The performance of this hybrid approach must be tested.


Serial code will be developed for testing and refinement.


When results justify it, parallel algorithms will be pursued in
collaboration with the NWChem team.

Basic Energy Sciences

Computational Chemistry


Mark Gordon, Ames Laboratory

Participants:


Mark Gordon


Mike Schmidt


Klaus Ruedenberg


James Evans


Collaborators:


Piotr Piecuch (Michigan
State)


Don Truhlar (Minnesota)


Scalable Computing Lab:
Ames Lab


PNNL

Basic Energy Sciences

Computational Chemistry


Objectives for Highly Accurate
Quantum Chemistry Codes 1

Mark Gordon, Ames Laboratory


Full Configuration Interaction (Full CI)


Exact wavefunction for a given atomic basis


Serves as benchmark for approximate correlated methods:


Currently limited to modest basis sets, atoms, very small molecules


Need scalable code to apply Full CI more broadly


Replicated data algorithm: limited molecular size


Distributed data algorithm: communication issues


Programming Models interface: Data compression (Kendall)


Scalable General CI


Full CI limited in scope


Most truncated CI based on “full space”
ansatz


Extend scope of CI by eliminating “deadwood”: only include
important configurations

Basic Energy Sciences

Computational Chemistry


Objectives for Highly Accurate
Quantum Chemistry Codes 2

Mark Gordon, Ames Laboratory


Scalable MCSCF


Frequently require orbitals optimized in CI space


More demanding than CI


Most common MCSCF implementation: FORS/CAS


Expand Accessible MCSCF Active Space


MCSCF based on General CI: eliminating deadwood


Q
-
CAS based on localized orbitals


Efficient Scalable Coupled Cluster Methods: Piecuch


Open Shell Perturbation Theory Gradients


Basic Energy Sciences

Computational Chemistry


Objectives for Spanning Multiple
Time and Length Scales

Mark Gordon, Ames Laboratory


Scalable Kinetic Monte Carlo Codes with Focus on
Surface Phenomena


Develop New Integrated Atomistic & Mesoscale
Descriptions of Surface Phenomena


Pattern formation in catalytic surface reactions


Nanostructure evolution during surface processes


Chemical vapor deposition (CVD)


Etching of semiconductor surfaces


Heterogeneous catalysis


Interface with continuum mesoscale modeling

Basic Energy Sciences

The “holy grail” of the
ab initio

electronic structure theory:

The development of

simple
,

black
-
box
, and

affordable

methods that can provide

highly
accurate

(~spectroscopic) description of
ground
-

and excited
-
state potential energy
surfaces

New Coupled
-
Cluster Methods For Molecular Potential
Energy Surfaces

Piotr Piecuch, Michigan State University



Examples of applications:


dynamics of reactive collisions


highly excited and metastable ro
-
vibrational states of molecules


rate constant calculations


collisional quenching of electronically
excited molecular species

Motivation:


elementary processes that occur in
combustion (e.g., reactions involving OH
and N
x
O
y
)


collisional quenching of the OH and
other radical species

Basic Energy Sciences

New Coupled
-
Cluster Methods For Molecular Potential
Energy Surfaces
-

Specific Goals

Piotr Piecuch, Michigan State University


New CC methods for ground
-
state potential energy
surfaces:


method of moments of CC equations


renormalized CC approaches



New CC methods for excited
-
state potential energy
surfaces:


method of moments of CC equations


active
-
space EOMCC approaches


MBPT(2)
-
like choices of


lead to the renormalized and completely
renormalized CCSD(T), CCSD(TQ), CCSDT(Q), etc. approaches


Basic Energy Sciences

New Coupled
-
Cluster Methods For Molecular Potential
Energy Surfaces


Future Work

Piotr Piecuch, Michigan State University


Methods and algorithms
-

ground
-
state problem


Incorporation of the renormalized CCSD(T), CCSD(TQ), and CCSDT(Q) methods in GAMESS


Development of the MMCC schemes with the non
-
perturbative choices of



Extensions of the MMCC and renormalized CC methods to open
-
shell states and reference configurations of the
ROHF type


Work with Professor Mark S. Gordon and coworkers on parallelizing the MMCC and renormalized CC methods
within GAMESS


Methods and algorithms
-

excited
-
state problem


Extension of the MMCC theory to the MMCC(2,4) case and extension of the active
-
space EOMCC theory to the
EOMCCSDtq case


Development of efficient computer codes for the MMCC and active
-
space EOMCC methods and incorporation of
these codes in GAMESS


Development of the MMCC and active
-
space EOMCC methods for non
-
singlet states and formulation of the EA and
IP extensions of the active
-
space EOMCC approaches


Extension of the active
-
space EOMCC approaches to properties other than energy


Development of the MMCC schemes with the perturbative choices of


(renormalized EOMCCSD(T) method ?)


Work with Professor Mark S. Gordon and coworkers on parallelizing the excited
-
state MMCC codes within
GAMESS

Basic Energy Sciences

Quantum Chemistry for Periodic Systems

Gustavo E. Scuseria, Rice University


Objectives


To develop methods and computational programs for the
accurate

prediction of
electronic

and
optical

properties of polymers, surfaces, and solids with large unit
cells


Applications to
polymers

and
catalysis


Approach


Use Gaussian orbitals for across the Periodic Table
reliability


Use Fast
-
Multipole Methods for achieving
linear scaling

in the Coulomb problem


Fast quadratures and alternatives to diagonalization for O(N) performance


Develop
better

DFT functionals for increased
accuracy


Payoffs


Increase capabilities for
modeling

materials and processes


Big
impact

on areas where quantum molecular modeling is routinely used:
chemical, pharmaceutical, and defense industries


Incorporation to
Gaussian

package


Order (N) tools for Periodic Systems

Basic Energy Sciences

Relativistic Multireference Quantum Chemistry


R. M. Pitzer, B. E. Bursten, I. Shavitt


Ohio State University


Strengths


Includes relativistic effects


Describes complicated electron coupling


Includes electron correlation


Needs


Extensive testing and tuning of recent parallel version


More flexible core potentials (W. C. Ermler)


Parallel versions of ancillary programs


Refinement of diagonalization algorithm


Applications


Actinide complexes in solution and on surfaces


Lanthanide intensities in crystals

Basic Energy Sciences



Reliable Electronic Structure Calculations for Heavy Element
Chemistry: Molecules Containing Actinides, Lanthanides,
and Transition Metals

W. C. Ermler and M. M. Marino, Department of Chemistry, The
University of Memphis

Relativistic Pseudopotentionals (RPPs)


RPPs are based on extending the usual two
-
space representation of atomic electrons
(core and valence) to three spaces (core/outer core/valence).


The RPP has embedded within it the standard small
-
core RECP that relegates the outer
core electrons to the valence space.


The RPP is ultimately calculated in its entirety at runtime and is specific to the
molecular geometry and electronic state.


Only the smallest numbers of molecular valence electrons need to be treated explicitly.


The RPP can be used in any treatment of the electronic structure [DFT, CI, CCSD(T),
etc.]

Small
-
core RECPs embedded within very
-
large
-
core RPPs, in conjunction with advanced computing platforms, permit the
highly accurate ab initio treatment (including correlation, outer
-
core/valence polarization, and spin
-
orbit coupling) of
systems possessing orders of magnitude more electrons than are tractable using current codes and platforms.

Basic Energy Sciences

Advanced Software for the Calculation of Thermochemistry,
Kinetics and Dynamics
-

Parallelization of Cumulative
Reaction Probabilities (CRP)

Al Wagner, ANL


computationally intensive core of reaction rate constants


mathematical kernel

(all matrices are sparse with some structure):


-

method 1:

-

iterative eigensolve (imbedded iterative linearsolves)





-

clever preconditioning important




-

portability based on ANL PETSc library of kernels


-

method 2:

-

Chebyschev propagation (=> matrix vector multiplies)





-

novel finite difference representation (helps parallelize)


programing issues
:


-

parallelization


-

exploiting data structure (i.e., preconditioning)

Basic Energy Sciences

Advanced Software for the Calculation of Thermochemistry,
Kinetics and Dynamics
-

Parallel Implementation of Subspace
Projection Approximate Matrix (SPAM) method

Al Wagner, ANL


novel iterative method to solve general matrix equations

-
eigensolve


-

linear solve


-

nonlinear solve


applications are widespread


in chemistry: CRP, electronic structure (SCF, MRSDCI,…)


mathematical kernel
:


related to Davidson, multigrid, and conjugate gradient methods


subspace reduction (requiring usual matrix vector multiplies)


projection operator decomposition of matrix vector product


substitution of user
-
supplied approximate matrix in computationally
intensive part of decomposition


sequence of approx. matrices => multilevel method


programing issues
:


generalization of approach (only done for eigensolve)


incorporation into libraries (connected to TOPS project at ANL)


test of efficacy in realistic applications (e.g., CRP)

Basic Energy Sciences

Theoretical Chemical Dynamics of Elementary Combustion
Reactions

Don Thompson, Oklahoma State University

1. Potential Energy Surfaces

We are developing methods for direct use of ab initio energies and forces in dynamics
simulations based on interpolating moving least
-
squares. The method can be used to
generate global representations for dynamics calculations. It can also be integrated into
direct (on
-
the
-
fly) dynamics simulations, providing a means of reducing the
computation expense by using interpolation. Furthermore, we will attempt to develop
methods for "on the fly scaling" of values so that lower
-
level quantum methods can be
used.

Note: The interpolating moving least
-
squares method requires the determination of
optimum polynomial expansion coefficients. The extension of this approach to large
numbers of atoms will be made practical by the using parallel computing.

Project goal: Development of methods and software for more accurate and efficient
simulations and rate calculations for complex chemical reactions involving large
polyatomic molecules and radicals.

Basic Energy Sciences

Theoretical Chemical Dynamics of Elementary Combustion
Reactions

Don Thompson, Oklahoma State University

2. Molecular Dynamics Simulations and Rate Calculations

The treatment of the dynamics of reactions involving polyatomic
molecules presents a challenging problem since they often involve
quantum effects, yet quantum mechanical methods are not capable of
treating more than a few atoms.
We are developing quasiclassical and
semiclassical methods for practical dynamics simulations of complex
reactions in large molecules.

We are developing a general computer code
that implements these methods and that incorporates various ways of
representing the potential, including the direct use of ab initio forces. The
size of problems that can be treated can be greatly increased by taking
advantage of many processors to compute the large ensembles of
trajectories needed to determine rates and other dynamical properties of a
system.

Basic Energy Sciences

Basic Energy Sciences

Basic Energy Sciences