comp-chem/NWChem
1
0
Fork 0
mirror of https://github.com/nwchemgit/nwchem.git synced 2026-07-29 06:35:39 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

new chapter on use of Python

Browse source
This commit is contained in:
Robert Harrison 1999-05-10 18:31:59 +00:00
parent 0994b651e5
commit 8500b9ab89
3 changed files with 524 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

518
doc/user/python.tex Normal file
View file

@ -0,0 +1,518 @@
\label{sec:python}
Python (version 1.5.1) programs may be embedded into the NWChem input
and used to control the execution of NWChem. Python is a very
powerful and widely used scripting language that provides useful
things such as variables, conditional branches and loops, and is also
readily extended. Example applications include scanning potential
energy surfaces, computing properties in a variety of basis sets,
optimizing the energy w.r.t. parameters in the basis set, computing
polarizabilities, and simple molecular dynamics.
Visit the Python web-site \verb+http://www.python.org+ for a full manual
and lots of useful code and resources.
\section{How to input and run a Python program inside NWChem}
A Python program in input into NWChem inside a Python compound directive.
\begin{verbatim}
python [print|noprint]
...
end
\end{verbatim}
The \verb+END+ directive must be flush against the left
margin (see the Troubleshooting section for the reason why).
The program is by default printed to standard output when read, but
this may be disabled with the \verb+noprint+ keyword. Python uses
indentation to indicate scope (and the initial level of indentation
must be zero), whereas NWChem uses optional indentation only to make
the input more readable. For example, in Python, the contents of a
loop, or conditionally-executed block of code must be indented further
than the surrounding code. Also, Python attaches special meaning to
several symbols also used by NWChem. For these reasons, the input
inside a \verb+PYTHON+ compound directive is read verbatim except that
if the first line of the Python program is indented, the same amount
of indentation is removed from all subsequent lines. This is so that
a program may be indented inside the \verb+PYTHON+ input block for
improved readability of the NWChem input, while satisfying the
constraint that when given to Python the first line has zero
indentation.
E.g., the following two sets of input specify the same Python program.
\begin{verbatim}
python
print 'Hello'
print 'Goodbye'
end
python
print 'Hello'
print 'Goodbye'
end
\end{verbatim}
whereas this program is in error since the indentation of the second
line is less than that of the first.
\begin{verbatim}
python
print 'Hello'
print 'Goodbye'
end
\end{verbatim}
The Python program is not executed until the following directive
is encountered
\begin{verbatim}
task python
\end{verbatim}
which is to maintain consistency with the behavior of NWChem in general.
{\em The program is executed by all nodes.} This enables the full functionality and speed of NWChem to be accessible from Python, but there are some gotchas
\begin{itemize}
\item Print statements and other output will be executed by all nodes
so you will get a lot more output than probably desired unless the
output is restricted to just one node (by convention node zero).
\item The calls to NWChem functions are all collective (i.e., all
nodes must execute them). If these calls are not made collectively
your program may deadlock (i.e., cease to make progress).
\item When writing to the database (\verb+rtdb_put()+) it is the data
from node zero that is written.
\end{itemize}
\section{NWChem extensions}
Since we have little experience using Python, the NWChem-Python
interface might change in a non-backwardly compatible fashion as we
discover better ways of providing useful functionality. We would
appreciate suggestions about useful things that can be added to the
NWChem-Python interface. In principle, nearly any Fortran or C
routine within NWChem can be extended to Python, but we are also
interested in ideas that will enable users to build completely new
things. For instance, how about being able to define your own energy
functions that can be used with the existing optimizers or dynamics
package?
Python has been extended with the following NWChem-specific commands.
They all handle errors by throwing exceptions which may be handled in
the standard Python manner. The first four commands will be of the
widest interest.
\begin{itemize}
\item \verb+input_parse(string)+ --- invokes the standard NWChem input
parser with the data in \verb+string+ as input. Note that the usual
behavior of NWChem will apply --- the parser only reads input up to
either end of input or until a \verb+TASK+ directive is encountered
(the task directive is {\em not} executed by the parser).
\item \verb+task_energy(theory)+ --- returns the energy as if computed
with the NWChem directive \verb+TASK ENERGY <THEORY>+.
\item \verb+task_gradient(theory)+ --- returns a tuple
\verb+(energy,gradient)+ as if computed with the NWChem
directive \verb+TASK GRADIENT <THEORY>+.
\item \verb+ga_nodeid()+ --- returns the number of the parallel
process.
\item \verb+rtdb_print(print_values)+ --- prints the contents of the
RTDB. If \verb+print_values+ is 0, only the keys are printed, if it
is 1 then the values are also printed.
\item \verb+rtdb_put(name, values)+ or
\verb+rtdb_put(name, values, type)+ --- puts the values into the
database with the given name. In the first form, the type is inferred
from the first value, and in the second form the type is specified
using the last argument as one of \verb+INT+, \verb+DBL+,
\verb+LOGICAL+, or \verb+CHAR+.
\item \verb+rtdb_get(name) + --- returns the data from the database
associated with the given name.
\end{itemize}
\section{Examples}
Several examples will provide the best explanation of how the extensions
are used, and how Python might prove useful.
\subsection{Hello world}
\begin{verbatim}
python
print 'Hello world from process ', ga_nodeid()
end
task python
\end{verbatim}
This input prints the traditional greeting from each parallel process.
\subsection{Scanning a basis exponent}
\begin{verbatim}
geometry units au noprint
O 0 0 0
H 0 1.430 -1.107
H 0 -1.430 -1.107
end
python noprint
exponent = 0.1
while (exponent <= 2.01):
input_parse('''
basis noprint
H library 3-21g
O library 3-21g
O d; %f 1.0
end
''' % (exponent))
print ' exponent = ', exponent, ' energy = ', task_energy('scf')
exponent = exponent + 0.1
end
print none
task python
\end{verbatim}
This program augments a 3-21g basis for water with a d-function on
oxygen and varies the exponent from 0.1 to 2.0 in steps of 0.1,
printing the exponent and energy at each step.
The geometry is input as usual, but the basis set input is embedded
inside a call to \verb+input_parse()+ in the Python program. The
standard Python string substitution is used to put the current value of
the exponent into the basis set (replacing the \verb+%f+) before being
parsed by NWChem. The energy is returned by \verb+task_energy('scf')+
and printed out. The \verb+print none+ in the NWChem input switches
off all NWChem output so all you will see is the output from your
Python program.
Note that execution in parallel may produce unwanted output since
all process execute the print statement inside the Python program.
\subsection{Scanning a basis exponent revisited.}
\begin{verbatim}
geometry units au
O 0 0 0
H 0 1.430 -1.107
H 0 -1.430 -1.107
end
print none
python
if (ga_nodeid() == 0): plotdata = open("plotdata",'w')
def energy_at_exponent(exponent):
input_parse('''
basis noprint
H library 3-21g
O library 3-21g
O d; %f 1.0
end
''' % (exponent))
return task_energy('scf')
exponent = 0.1
while exponent <= 2.0:
energy = energy_at_exponent(exponent)
if (ga_nodeid() == 0):
print ' exponent = ', exponent, ' energy = ', energy
plotdata.write('%f %f\n' % (exponent , energy))
exponent = exponent + 0.1
if (ga_nodeid() == 0): plotdata.close()
end
task python
\end{verbatim}
This input performs exactly the same calculation as the previous one,
but uses a slightly more sophisticated Python program, also writes
the data out to a file for easy visualization with a package such as
\verb+gnuplot+, and protects write statements to prevent
duplicate output in a parallel job. The only significant differences
are in the Python program. A file called \verb+"plotdata"+ is opened,
and then a procedure is defined which given an exponent returns the
energy. Next comes the main loop that scans the exponent through the
desired range and prints the results to standard output and to the
file. When the loop is finished the additional output file is closed.
\subsection{Scanning a geometric variable}
\begin{verbatim}
python
geometry = '''
geometry noprint
symmetry d2h
C 0 0 %f
H 0 0.916 1.224
end
'''
x = 0.6
while (x < 0.721):
input_parse(geometry % x)
energy = task_energy('scf')
print ' x = %5.2f energy = %10.6f' % (x, energy)
x = x + 0.01
end
basis
C library 6-31g*
H library 6-31g*
end
print none
task python
\end{verbatim}
This scans the bond length in ethene from 1.2 to 1.44 in steps
of 0.2 computing the energy at each geometry. Since it is using
$D_{2h}$ symmetry the program actually uses a variable (verb+x+) that is
half the bond length.
\subsection{Scan using the BSSE counterpoise corrected energy}
\begin{verbatim}
basis spherical noprint
Ne library cc-pvdz; He library cc-pvdz
BqNe library Ne cc-pvdz; BqHe library He cc-pvdz
end
mp2; tight; freeze core atomic; end
print none
python noprint
supermolecule = '''
geometry noprint
Ne 0 0 0
He 0 0 %f
end
'''
fragment1 = '''
geometry noprint
Ne 0 0 0
BqHe 0 0 %f
end
'''
fragment2 = '''
geometry noprint
BqNe 0 0 0
He 0 0 %f
end
'''
def geom_energy(geometry):
input_parse(geometry)
input_parse('scf; vectors atomic; end\n')
return task_energy('mp2')
def bsse_energy(z):
return geom_energy(supermolecule % z) - \
geom_energy(fragment1 % z) - \
geom_energy(fragment2 % z)
z = 3.3
while (z < 4.301):
energy = bsse_energy(z)
if (ga_nodeid() == 0):
print ' z = %5.2f energy = %10.7f ' % (z, energy)
z = z + 0.1
end
task python
\end{verbatim}
This example scans the He---Ne bond-length from 3.3 to 4.3 and prints out
the BSSE counterpoise corrected MP2 energy.
The basis set is specified as usual, noting that we will need
functions on ghost centers to do the counterpoise correction. The
Python program commences by defining strings containing the geometry
of the super-molecule and two fragments, each having one variable to be
substituted. Next, a function is defined to compute the energy given
a geometry, and then a function is defined to compute the counterpoise
corrected energy at a given bond length. Finally, the bond length is
scanned and the energy printed. When computing the energy, the atomic
guess has to be forced in the SCF since by default it will attempt to
use orbitals from the previous calculation which is not appropriate
here.
Since the counterpoise corrected energy is a linear combination of
other standard energies, it is possible to compute the analytic
derivatives term by term. Thus, combining this example and the next
could yield the foundation of a BSSE corrected geometry optimization
package.
\subsection{Scan the geometry and compute the energy and gradient}
\begin{verbatim}
basis noprint; H library sto-3g; O library sto-3g; end
python noprint
print ' y z energy gradient'
print ' ----- ----- ---------- ------------------------------------'
y = 1.2
elo = 0.0
while y <= 1.61:
z = 1.0
while z <= 1.21:
input_parse('''
geometry noprint units atomic
O 0 0 0
H 0 %f -%f
H 0 -%f -%f
end
''' % (y, z, y, z))
(energy,gradient) = task_gradient('scf')
if (energy < elo):
elo = energy
ylo = y
zlo = z
print ' %5.2f %5.2f %9.6f' % (y, z, energy),
i = 0
while (i < len(gradient)):
print '%5.2f' % gradient[i],
i = i + 1
print ''
z = z + 0.1
y = y + 0.1
print ''
print ' Lowest energy =',elo,' at y=',ylo,', z =',zlo
print ' '
end
print none
task python
\end{verbatim}
This program illustrates evaluating the energy and gradient
by calling \verb+task_gradient()+. A water molecule is scanned
through several $C_{2v}$ geometries by varying the y and z coordinates
of the two hydrogen atoms. At each geometry the coordinates, energy
and gradient are printed. The lowest energy geometry is recorded.
The basis set (sto-3g) is input as usual. The two while loops vary
the y and z coordinates. These are then substituted into a geometry
which is parsed by NWChem using \verb+input_parse()+. The energy and
gradient are then evaluated by calling \verb+task_gradient()+ which
returns a tuple containing the energy (a scalar) and the gradient (a
vector or list). These are printed out exploiting the Python
convention that a print statement ending in a comma does not print
end-of-line.
\subsection{Reaction energies varying the basis set}
\begin{verbatim}
mp2; freeze atomic; end
print none
python
energies = {}
c2h4 = '''
geometry noprint
symmetry d2h
C 0 0 0.672
H 0 0.935 1.238
end
'''
ch4 = '''
geometry noprint
symmetry td
C 0 0 0
H 0.634 0.634 0.634
end
'''
h2 = '''
geometry noprint
H 0 0 0.378
H 0 0 -0.378
end
'''
def energy(basis, geometry):
input_parse('''
basis spherical noprint
c library %s ; h library %s
end
''' % (basis, basis))
input_parse(geometry)
return task_energy('mp2')
for basis in ('sto-3g', '6-31g', '6-31g*', 'cc-pvdz', 'cc-pvtz'):
energies[basis] = 2*energy(basis, ch4) - \
2*energy(basis, h2) - \
energy(basis, c2h4)
if (ga_nodeid() == 0): print basis, ' %8.6f' % energies[basis]
end
task python
\end{verbatim}
In this example the reaction energy for
$2H_2 + C_2H_4 \rightarrow 2CH_4$ is evaluated using MP2 in several
basis sets. The geometries are fixed, but could be re-optimized in
each basis. To illustrate the useful associative arrays in Python,
the reaction energies are put into the associative array
\verb+energies+ --- note its declaration at the top of the program.
\subsection{Using the database}
\begin{verbatim}
python
rtdb_put("test_int2", 22)
rtdb_put("test_int", [22, 10, 3], INT)
rtdb_put("test_dbl", [22.9, 12.4, 23.908], DBL)
rtdb_put("test_str", "hello", CHAR)
rtdb_put("test_logic", [0,1,0,1,0,1], LOGICAL)
rtdb_put("test_logic2", 0, LOGICAL)
rtdb_print(1)
print "test_str = ", rtdb_get("test_str")
print "test_int = ", rtdb_get("test_int")
print "test_in2 = ", rtdb_get("test_int2")
print "test_dbl = ", rtdb_get("test_dbl")
print "test_logic = ", rtdb_get("test_logic")
print "test_logic2 = ", rtdb_get("test_logic2")
end
task python
\end{verbatim}
This example illustrates how to access the database from Python.
\section{Troubleshooting}
Common problems with Python programs inside NWChem.
\begin{enumerate}
\item You get the message
\begin{verbatim}
0:python_input: indentation must be >= that of first line: 4
\end{verbatim}
This indicates that NWChem thinks that a line is less indented than
the first line. If this is not the case then perhaps there is a tab
in your input which NWChem treats as a single space character but
appears to you as more spaces. Try running \verb+untabify+ in Emacs.
It could also be the \verb+END+ directive that terminates the
\verb+PYTHON+ compound directive --- since Python also has an
\verb+end+ statement then to avoid confusion the \verb+END+ directive
for NWChem {\em must} be at the start of the line.
\item The last (or only) line of your input to \verb+input_parse()+
seems to be ignored --- An entire line of input must be provided,
including the end of line. Try terminating the string with
\verb+'\n'+.
\item Your program hangs or deadlocks --- most likely you have a piece
of code that is restricted to executing on a subset of the processors
(perhaps just node 0) but is calling (perhaps indirectly) a function
that must execute on all nodes.
\end{enumerate}