NQCD / surfgen

Generates analytical coupled PESs using quasi-diabatic Hamiltonians

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

surfgen

surfgen use ab initio data to generate analytical coupled [potential energy surfaces(PES)] (http://en.wikipedia.org/wiki/Potential_energy_surface) the quasi-diabatic Hamiltonian(Hd) approach.
The primary purpose of this project is to provide a way to accelerate dynamics simulations of non-adiabatic processes without compromising the accuracy of the representation of electronic states, which requires extremely expensive correlated multi-reference electronic structure methods such as MRCI.

The form of expansion and symmetry treatment is general and versatile, making it essentially capable of fitting almost any system as long as the underlying ab initio data can be supplied. The method has been used to generate coupled PESs for systems as large as phenol (J. Chem. Phys. 140, 024112 (2014)), where the coupled PESs accurately describe the full 33 dimensionalities in their entire dynamically relevant region, to a very high energy range of 50,000cm-1, including the 4 lowest singlet states.

The program uses energy, energy gradient and derivative coupling data from ab initio data to generate the fit. The use of energy gradient drastically reduce the number of data points needed to construct a converged fit, especially in higher dimensional cases, and at the same time significantly reduces oscilation in the fit surfaces. The program use derivative coupling data to automatically generate the most diabatic representation for the system, in a least-squares sense. Gradients and couplings are not required for all data points, but are crucial for the regions with strong couplings. We have found that obtaining high level ab initio data that contains gradient and couplings data and consistent in the entire domain is the most difficult step in the procedure.

Molecular symmetry, or subgroups of Complete Permutation Inversion Symmetry (CNPI) group is fully implemented in a general manner through a projection operators approach. CNPI group has an almost infinite number of group structures and irreducible representations, and it is in many aspect very different from point group symmetry. The user is advised to carefully study the concept of CNPI symmetry and the problem at hand to correctly choose the right symmetry for your system. The program can in principle treat any symmetry, but the user will have to supply the representation matrices.

Fortran 90 and 70 interfaces are supplied to facilitate the use of fit PESs in other programs.

This project is coded with Fortran 90 and contains both fitting programs used to generate coupled PESs and evaluation libraries that can be conveniently used to utilize the fit surface in any simulation program.

Features

Fitting Ab Initio Data

The program is capable of fitting energy, energy gradient and derivative coupling (AKA nonadiabatic coupling or vibronic coupling) data obtained from ab initio calculations. A weighed least squares procedure is used to generate the fit to simultaneously optimize the reproduction of adiabatic energies and energy gradients and the diabaticity of the quasi-diabaticrepresentation, defined by the residual coupling between diabatic states. Lagrange multipliers can be used to enable the exact reproduction of arbitrary selected set of data, such as energy and gradients at critical points on the potential or energy difference and derivative couplings at points of conical intersections.

Geometries, energies, gradients and couplings should all be prepared in COLUMBUS format. The program does not require all data to be present at all data points, and can operate even when large amount of data are missing. However, missing data makes the fitting procedure less efficient and less reliable. The availability of energy gradients and couplings will drastically reduce the number of data points need to construct a fit that describe the entire desired region. Gradients and couplings are also crucial in areas where states are strongly coupled.

The algorithm, through the use of gradients and couplings, is highly tolerant of discontinuities and numeric noises in the underlying ab initio data, and is found to be able to automatically smooth out these unphysical effects. However, if the discontinuities are too large the program can have trouble identifying and matching the electronic states.

Nonadiabatic couplings and Seams of Conical Intersections

The Hd approach is capable of extremely accurate description of nonadiabatic interactions. It has been used to successfully describe large portions of the seam of conical intersections, spanning completely different geometrically structures. This is enabled by the application of intersection adapted representation. The adiabatic representation changes drastically near avoided crossings, reaching singularity at the seam of conical intersections, which complicates the fitting procedure. In order to facilitate a stable fit, we use gradients and couplings to construct a stable intersection adapted representation (which is effectively the joint diagonalization of the matrix representation of vector operator ∇H) that is continuous near the seam. Ab initio data and the fit Hamiltonian are both rotated to this representation to facilitate the fit.

Arbitrary energy difference range and arbitrary number of states can be treated, provided that all gradient and coupling data are present. (see J. Chem. Phys. 141, 174109 (2014))

Flexibility

The program allows the user to define the blocks of the matrix Hd with a large set of customizable basis functions, enabling the flexibility to describe complex features on the surface to a high level of accuracy. The function basis are constructed from symmetric projections of the monomials of single-coordinate functions.
A number of forms of single coordinate function forms are provided, including internuclear distances, bond angles and different types of out-of-plane bending motions. Many scaling options are availble for each type of coordiantes. The user can control the number, parameters and types of the coordinates used as well as their combination to generate the monomials through coords.in file.

This in principle allows the user to achieve any precision. However, in practice the memory use scales with n2, the fitting time scales with n3 and the evaluation time when using the fit Hd scales with n, where n is the size of funciton basis used. This can quickly render the fit impractical if the basis is too large.
A basis size below 10,000 is prefered, but a size around 20,000 is usually still managable provided that sufficient memory is available. We recommend fine tuning the parameters and type of coordinates instead of simply increase the order of monomial. Adding different types and scaling of coordinates and coordinates between different group of interacting atoms is also found to be much more efficient than increasing the order. It is recommended that Morse functions for most interatomic distances be included, as well as sufficient out-of-plane coordinates, to ensure that the coordinates is capable of describing all motions. Hyperbolic tangent, gaussian functions and cosines of bond angles are found to be very efficient in improving quality of the fit when many different regions are involved, since they behave like smooth window functions.

Global Symmetry Treatment

Standard group-theoretic projection operator method is used by the program to facilitate an arbitrary subgroup of the Complete Nuclear Permutation Inversion(CNPI) group to construct symmetry adapted basis for the fitting procedure.
With the help of such feature, the program can correctly treat the symmetry in problems that involve large amplitude motions, as well as vibrational problems.

CNPI group is a global symmetry group, which provides symmetry relation between different parts of the surfaces. This is to be compared to point group symmetry, which relates different vibrational component of the surface at the same data point. In fact, with few exceptions, CNPI group symmetry automatically generates the corresponding point group symmetry when the geometry carries any specific point group. The point group is homomorphic to a subgroup of the CNPI group induced by the subset of CNPI operations and/or rotations that keeps the geometry invariant. The only exception being the C axis of linear molecules, which is a pure rotation that cannot be achieved by any permutation/inversion.

Relative signs between different parts of the potential can be very difficult to access, making the assignment of CNPI symmetry difficult. One should therefore take advantage of the induced subgroup at high symmetry points to determine the symmetry of diabatic states.

The projection operator approach requires representation matrices. Due to the large amount of possible combination of groups and irreducible representations, we currently do not supply these representation matrices. The user needs to work out the effect of symmetry operations and representation matrices and supply them in the irrep.in file.

A problem that very often come up is the change of symmetry of states in the process of a reaction. Since only a limited number of states are treated, states with different symmetry are often found to enter and exit the set of treated states in the course of a reaction, changing the symmetry. This means whatever symmetry that we set the diabatic states to be, they are likely to be incorrect in some regions where states with other symmetries intrude into the problem. One can still converge a fit in this case, but the couplings and the topography of conical intersections will be incorrect. Usually this can be overcome by including 1 or 2 more states than the problem physically require, and allowing these states to have incorrect symmetry, so that the problematic crossings happen above the energy limit of the problem.

Efficiency

With the fully analytical form of Hd, the evaluation time for a single point is usually within 50ms. Future update is planned to allow vectorized evaluation of large number of data points. The fitting program achieve high efficiency through the extensive use of optimized and threaded LAPACK and BLAS libraries. Other functionalities such as automatic local internal coordinate construction, automatic null space removal and GDIIS extrapolations provide tools to enhance performance.

Documentation

Chapter 1 manual pages are provided for surfgen and several input files are provided in man/man1. To view these manuals, add the man directory to $MANPATH or copy the manual files to /usr/shared/man/man1.

$ cp man/man1/* /usr/shared/man/man1

Then simply use man to view the pages.

$ man surfgen

You can also use -M flag to explicitly specify the manual directory or use -t flag to generate PostScript version of the manual page. For example, use the following command to view PDF versions on Mac OS X

$ man -M man -t surfgen | open -f -a /Applications/Preview

PDF versions of the documents can also be found in pdf directory. However, they may not be as up-to-date as the man files.

Installation

Please be advised that we have only tested installation on Mac OS X Mountain Lion, RHEL 4.x and CentOS 6.x, with Intel Fortran compiler or gfortran. Other set up should also work but might require modifications of code or Makefile. A proper Fortran 90 compiler, GNU make and implementations of BLAS and LAPACK are also required. They can all be obtained free of charge (gfortran, GNU make, ATLAS)

A number of precompiled binaries fitting programs can be found in bin directory.

A library is built with ifort with static link to LAPACK/BLAS on CentOS 6. It can be found in lib directory.

If you cannot file a working binary or library, it will only take you a few minutes to build them from source code, located in source directory. See the following sections for compilation guides.

  • Stacksize

Please set stacksize to unlimited to allow execution for larger systems.

Compiling the Fitting Program

If you cloned the repository in Xcode on Mac OS X, you should be able to use Xcode to run or build for any type immediately after download if you have gfortran installed in /usr/bin.

To compile the fitting program, you have to set up environmental variables $FC to point to your fortran compiler and $LIBS(or $BLAS_LIB) to be your LAPACK/BLAS link line flags.
If you included MKL path in $LD_LIBRARY_PATH or if you have included LAPACK/BLAS link options in LDFLAGS, then you can skip setting up the $LIBS(or $BLAS_LIB) variable.

After the variables are set, simply do

$ make 

The compiled program can be found in bin directory with the name surfgen-{version}-{OS}-{OS version}-{compiler}

You can also pass any unset variables through arguments of make

$ make FC=ifort BLAS_LIB="-Wl,--start-group  $MKLROOT/lib/intel64/libmkl_intel_ilp64.a $MKLROOT/lib/intel64/libmkl_intel_thread.a $MKLROOT/lib/intel64/libmkl_core.a -Wl,--end-group -lpthread -lm"

If for any reason you suspect a bug, you can include debug symbols or check flags by setting $DEBUGFLAG variable or set DEBUGGING_SYMBOLS to YES. Note that if you click the Run button in Xcode or compile for Debugging, this option is automatically enabled.

Extra compilation flags can be added through $CPOPT or $FFLAGS. Extra linking flags can be added through $LKOPT or $LDFLAGS.

Compiling the Evaluation Library

To compile the library, you also need to supply environmental variable $FC, either in the shell or through arguments.

Example:

$ make lib FC=gfortran

Extra compilation flags can be added through $CPOPT or $FFLAGS. Extra linking flags can be added through $LKOPT or $LDFLAGS. You can also change the default archiver by setting $AR.

For example, if you want to enable ipo:

$ make lib FC=ifort FFLAGS="-parallel -O3 -xHost -i8 -ip -ipo" AR="xiar -rv"

Library can be found in lib directory.

Help, Problem Reports and Wiki

Please email Xiaolei Zhu if you need help to compile or use the program. Please use the Issues page on github to report bugs and other problems.

We are planning on implementing an wiki site but have yet to realize that thought. Anyone is invited to help to construct the wiki.

Authors

surfgen is created by Xiaolei Zhu and other awesome members from Yarkony Group, Department of Chemistry, Johns Hopkins University

This program is based on, and would not have been possible without the quadratic Hd fitting program for vibrational spectroscopy simulation, developed by Michael Schuurman, then also a member of Yarkony group.

Xiaolei Zhu's Research Profile: Follow me on ResearchGate My ORCID profile

References

Xiaolei Zhu's Ph.D. dissertation, The Quasi-Diabatic Hamiltonian Approach to Accurate and Efficient Nonadiabatic Dynamics with Correct Treatment of Conical Intersection Seams(2014)

Xiaolei Zhu and David Yarkony, Quasi-diabatic representations of adiabatic potential energy surfaces coupled by conical intersections including bond breaking: A more general construction procedure and an analysis of the diabatic representation , J. Chem. Phys., 137, 22A511 (2012)

Xiaolei Zhu and David Yarkony, Toward eliminating the electronic structure bottleneck in nonadiabatic dynamics on the fly: An algorithm to fit nonlocal, quasidiabatic, coupled electronic state Hamiltonians based on ab initio electronic structure data , J. Chem. Phys., 132, 104101 (2010)

Copyright Notice

Copyright 2011-2015 Yarkony Group, The Johns Hopkins University.

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program.  If not, see <http://www.gnu.org/licenses/>.

Creative Commons License
The texts in Surfgen Program Suite by Yarkony Group, Johns Hopkins University is licensed under a Creative Commons Attribution 3.0 Unported License.

About

Generates analytical coupled PESs using quasi-diabatic Hamiltonians

License:GNU General Public License v3.0


Languages

Language:SourcePawn 79.1%Language:Fortran 20.2%Language:Makefile 0.6%Language:C 0.1%Language:C++ 0.1%Language:Shell 0.1%