Debian Codegraph @

doc/Latex/apps/moldft/manual.tex

raw
1 \documentclass[letterpaper]{book}
2 \usepackage{amssymb,amsmath,latexsym,hyperref,graphics,color}
3
4 \setlength{\parindent}{0pt}
5 \setlength{\parskip}{2ex plus 0.5ex minus 0.2ex}
6
7 \begin{document}
8
9 % Title Page
10 \title{MADNESS Molecular electronic structure calculations}
11 \date{Last Modification: 7/7/2016}
12 \maketitle
13
14 % Copyright Page
15 \pagestyle{empty}
16 \null\vfill
17 \noindent
18 This file is part of MADNESS.
19
20
21 Copyright (C) 2007, 2010 Oak Ridge National Laboratory
22
23 This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public
24 License as published by the Free Software Foundation; either version 2 of the License, or(at your option) any later
25 version.
26
27 This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied
28 warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
29
30 You should have received a copy of the GNU General Public License along with this program; if not, write to the Free
31 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
32
33 For more information please contact:
34 \begin{quote}
35 Robert J. Harrison \\
36 Oak Ridge National Laboratory \\
37 One Bethel Valley Road \\
38 P.O. Box 2008, MS-6367 \\
39 Oak Ridge, TN 37831 \\
40 \\
41 email: [email protected] \\
42 tel: 865-241-3937 \\
43 fax: 865-572-0680
44 \end{quote}
45 \newpage
46
47
48 % Table of Contents Pages
49 \clearpage
50 \setcounter{page}{1}
51 \pagenumbering{roman}
52
53 \setcounter{tocdepth}{10}
54 \renewcommand\contentsname{Table of Contents}
55 \tableofcontents
56
57
58 \clearpage
59 \setcounter{page}{1}
60 \pagenumbering{arabic}
61
62 \chapter{Overview}
63
64 \section{Capabilities}
65
66
67
68 \section{Current status}
69
70 The software is still in what might best described as pre-production
71 quality. Most things mostly work, but there are enough quirks that
72 unfortunately it is far from being a black box like most Gaussian
73 codes.
74
75 Using the code to do actual science is the most effective mechanism
76 for us to find and fix issues, as well as to prioritize implementation
77 effort. Please report problems, issues and functionality requests to the
78 MADNESS issue tracker at XXXXXXXXXXXXXXX.
79
80 \chapter{Configuring and building}
81
82 This is just for LI-red for the purposes of the summer school.
83 MADNESS should build out of the box on most Linux and Apple Macintosh
84 boxes. Intel MKL is now free so you should alway use that (download
85 from Intel), and there is an open source version of Intel TBB (Linux
86 distributions should have that in their package managers). If you are
87 building to run on a virtual machine you should configure with the
88 never spin option.
89
90 \begin{enumerate}
91 \item Copy the modules from section \ref{sec:pbslired} below into your
92 \verb+.bashrc+ file. You will be using the Intel compiler, MPI, MKL
93 and TBB which is the recommended stack when running on a cluster.
94
95 \item Clone MADNESS from
96 \url{https://github.com/m-a-d-n-e-s-s/madness}. I (RJH) used
97 revision {\tt a1b4bae8acf57ae363f24cb8a7ea9e7229a0d95f} for testing.
98
99 \item {\tt cd madness}
100
101 \item \verb+sh autogen.sh+ --- only need to do this after initial checkout
102
103 \item \verb+autoreconf+ --- only need to do this if a configure script has changed
104
105 \item Configure with this command
106 \begin{verbatim}
107 ./configure --disable-shared \
108 CC=mpiicc CXX=mpiicpc +\verb+ MPICC=mpiicc MPICXX=mpiicpc
109 \end{verbatim}
110 If you have installed LIBXC you can specify it with option
111 \begin{verbatim}
112 --with-libxc=full-path-to-libxc+.
113 \end{verbatim}
114
115 \item \verb+make -j 10 libraries+ --- using just 10 processes on
116 LI-red is to be social. Also, if building on your own machine note
117 that some of the files can use over 2GB of memory to compile, so if
118 you have limited memory you may need to restrict the number of
119 processes. If compiling on {\tt cn-mem} I always use
120 \verb+make -j libraries+ which uses maximum parallelism (since {\tt
121 cn-mem} has 72 cores and 3 TB of memory.
122
123 \item \verb+cd src/apps/moldft+
124
125 \item \verb+make moldft+
126
127 \item \verb+MAD_NUM_THREADS=10 ./moldft+ --- run a quick test (the {\tt input} file by default does LDA water which is quick).
128
129 \end{enumerate}
130
131
132 \chapter{Execution}
133
134 To run {\tt moldft} with a single, multi-threaded process simply
135 invoke the name of the executable with the name of the input file on
136 the command line. If the input file is ommited, it defaults to a file
137 named {\tt input}. E.g.,
138
139 \begin{verbatim}
140 ./moldft inputfilename
141 \end{verbatim}
142
143
144 Recommendations:
145 \begin{itemize}
146 \item Use Intel TBB if it is available since it is more scalable than the old task pool.
147 \item By default, MADNESS will create one thread for every core in the
148 computer, but even with TBB it may be more effective to use fewer threads
149 (see the environment variable \verb+MAD_NUM_THREADS+). E.g.,
150 \begin{verbatim}
151 MAD_NUM_THREADS=10 ./moldft inputfile
152 \end{verbatim}
153 \item Oversubscribing the cores can lead to very poor performance due to
154 cache contention and locks/mutexes being held by de-scheduled
155 threads. Oversubscription can happen due to other processes (of yourself or other users)
156 running at the same time, or from use of hyperthreading (which on Intel processors
157 makes one phyiscal core appear to the system as two cores).
158 \item Many threads seem to interact badly with how we are using MPI.
159 Thus, it can help both robustness and performance to use multiple
160 MPI processes per node with fewer threads per processes. As a side benefit you
161 can bind the threads associated with each process to execute on the same
162 socket, which improves thread scaling. See the SeaWulf batch script below
163 for an example.
164 \end{itemize}
165
166 To run in parallel using MPI to create multiple, multi-threaded
167 processes use the command {\tt mpirun} (or on some systems {\tt
168 mpiexec}) --- this is very system dependent so you may need
169 additional options to specify where and how many processes are created
170 (see below for use on SeaWulf and LI-red). E.g.,
171
172 \begin{verbatim}
173 export MAD_NUM_THREADS=10
174 mpirun -np 12 ./moldft inputfilename
175 \end{verbatim}
176
177 \subsection{Environment variables}
178
179 \verb+MAD_NUM_THREADS+ --- Sets the total number of threads to be used
180 by MADNESS. When running with just one process all threads are
181 devoted to computation (1 main thread with the remainder in the thread
182 pool). When running with multiple MPI processes, one of the threads
183 is devoted to communication. If you have 8 or more cores per node, it
184 is recommended to leave at least one free for use by the O/S and MPI.
185
186 \verb+MRA_DATA_DIR+ --- Full path to the directory containing the MRA
187 data files (twoscale and autocorrelation coefficients; Gauss-Legendre
188 quadrature weights). This is usually not needed unless you have moved
189 the build/installation directory.
190
191 \verb+MAD_BUFFER_SIZE+ --- Sets the buffer size (in bytes) used by the
192 active messages (default is 1.5MBytes). Never needed by moldft?
193
194 \verb+MAD_RECV_BUFFERS+ --- Sets the number of receive buffers used by
195 the communication thread (default is 128 and a minimum of 32 is
196 enforced). If you are experiencing hangs when running with MPI,
197 making this number a bit larger (e.g., 256) can sometimes help. Too
198 many buffers can cause performance problems.
199
200 \verb+MAD_SEND_BUFFERS+ --- Sets the number of outstanding
201 asynchronous send buffers (default 128 and a minimum of 32 is
202 enforced). If you are experiencing hangs when running with MPI try
203 making this smaller (to throttle the volume communication).
204
205 \verb+MAD_NSSEND+ --- Every \verb+MAD_NSSEND+ messages, MADNESS
206 requests a receipt acknowledgment from the receiver in an attempt to
207 throttle the volume of communication. The default is the number of
208 send buffers. Set to the value \verb+1+ to force receipt every
209 message to be acknowledged. This makes things a bit slower but can
210 workaround MPI hangs.
211
212 \subsection{PBS script for LI-red}
213 \label{sec:pbslired}
214
215 This example makes a 2 hour job running in the short queue on 8 nodes
216 with one process per node and job name {\tt moldft}. Each process
217 will use 20 threads.
218
219 \begin{verbatim}
220 #!/bin/bash
221 #PBS -l nodes=8:ppn=1,walltime=02:00:00
222 #PBS -q short
223 #PBS -N moldft
224
225 export EXE=/home/rjh/madness/src/apps/moldft/moldft
226 export INPUT=benzenehf.in
227 export OUTPUT=benzenehf.out
228
229 export I_MPI_FABRICS=shm:ofa
230 export MAD_NUM_THREADS=20
231
232 cd $PBS_O_WORKDIR
233
234 mpirun -ppn 1 $EXE $INPUT >& $OUTPUT
235 \end{verbatim}
236
237 The modules I (RJH) used for testing were (put the following in your \verb+.bashrc+)
238 \begin{verbatim}
239 module load shared
240 module load torque/5.1.0
241 module load maui/3.3.1
242 module load gcc/4.9.2
243 module load intel/compiler/64/16.0.2/2016.2.181
244 module load intel/mkl/64/11.3.2/2016.2.181
245 module load intel-mpi/64/5.1.2/5.1.2.150.XXX
246 module load intel/tbb/64/4.4.3/2016.2.181
247 module load intel/vtune/2016/2016.1.1.434111
248 \end{verbatim}
249
250 \subsection{PBS script for SeaWulf}
251
252 On SeaWulf the PBS option {\tt ppn} is presently not consistent with
253 that on LI-red --- this will change soon. In the meantime the script
254 is a little more complicated in order to get just two MADNESS
255 processes per node with threads able to use all cores.
256
257 ASIDE: SeaWulf has 28 cores per node and a 40 Gbit/s InfiniBand
258 network whereas LI-red has 24 cores per node and a 56 Gbit/s network.
259 Somehow these factors (or others?) combine to make MADNESS execution
260 on SeaWulf with 1 MPI process that uses 26 MADNESS threads/node not
261 reliable. Hence, the job below uses two MPI processes per node each
262 with 11 MADNESS threads.
263
264 \begin{verbatim}
265 #!/bin/bash
266 #PBS -l nodes=12:ppn=28,walltime=10:00:00
267 #PBS -q default
268 #PBS -N geomopt
269
270 # Probably best to put these in your .bashrc so don't need here
271 module load intel/compiler/64/16.0/2016.1.056.XXX
272 module load intel/mkl/64/11.3/2016.1.056.XXX
273 module load intel-mpi/64/5.1.2/5.1.2.150.XXX
274 module load intel/tbb/64/4.4.3/2016.2.181
275 export [email protected]
276
277 # Intel MPI should use shared-memory and OFED for communication
278 export I_MPI_FABRICS=shm:ofa
279
280 # We will create two processes per node --- bind to separate sockets
281 export I_MPI_PIN_DOMAIN=socket
282
283 # Should be set elsewhere but may not be
284 export I_MPI_HYDRA_BOOTSTRAP=rsh
285
286 # 14 cores per socket = Linux+MPI+unknown+madness
287 export MAD_NUM_THREADS=11
288
289 # Change to the directory from which the job was submitted
290 cd $PBS_O_WORKDIR
291
292 # Presently SeaWulf is configured so that to get a dedicated node we
293 # need to set ppn=28 above which gives us 28 processes per node, but we
294 # want two. So need to make our own host file.
295 sort < $PBS_NODEFILE | uniq > $$.hosts
296 let NPROC=$PBS_NUM_NODES*2
297
298 # Paths to files
299 MOLDFT=/gpfs/home/rharrison/madness/src/apps/moldft/moldft
300 INPUTFILE=input
301 OUTPUTFILE=output
302
303 # Redirect output to a file so that can see it while job is running
304
305 mpirun -f $$.hosts -ppn 2 -n $NPROC $MOLDFT $INPUTFILE >& $OUTPUTFILE
306
307 \end{verbatim}
308
309
310 \section{Input structure}
311
312 Presently the input {\em must} contain both a \verb+dft+ and a
313 \verb+geometry+ input block. The file is scanned for the first such
314 block (so you can have multiple blocks in a file --- just the first is
315 used). Within these blocks, lines beginning with \verb+#+ are read as comments.
316 Outside these blocks, data is ignored.
317
318 \section{Reproducibility}
319
320 For small molecules, the calculations should usually be fully
321 reprodicible from run to run or with different numbers of threads or
322 processes. However, for larger molecules, the different rounding
323 error arising from different order of parallel execution can cause
324 different execution paths. So results may vary in the last digit or
325 so, and sometimes more especially if additional/fewer iterations are
326 made. However, results at convergence should {\em always} agree to
327 the that expected from the truncation and convergence thresholds, and
328 the number of iterations between runs should only differ by about one.
329 If you are seeing worse behavior than this, then please send the input
330 and output to the MADNESS issue tracker.
331
332 \chapter{Geometry specification}
333
334 \begin{verbatim}
335 geometry
336 units atomic or angstrom (default atomic)
337 eprec precision (default 1e-4)
338 tag x y z
339 pstag x y z charge
340 end
341 \end{verbatim}
342
343 The units and energy precision ({\tt eprec}) must presently be specified {\em before} coordinates of atoms are specified.
344
345 {\tt units} --- E.g., {\tt units angstrom} --- The user input units which can be {\tt atomic} or {\tt angstrom} (default is {\tt atomic}). Beware --- presently, the output file always contains atomic units.
346
347 {\tt eprec} --- E.g., {\tt eprec 1e-5} --- The energy error per atom
348 in atomic units due to smoothing the nuclear potential (default is
349 {\tt 1e-4}). The default is adequate for most chemical purposes
350 (giving energy differences to at least {\tt 1e-5} atomic units per atom
351 and geometries accurate to about {\tt 1e-3} atomic units) unless you
352 are doing very high accuracy benchmark studies. Gradients may get
353 noisy if you make {\tt eprec} too small, though we don't have much
354 experience with this yet.
355
356 {\tt tag x y z} --- E.g., {\tt Be 0.3 -0.1 3.1} --- Specifies an atom using its atomic symbol (case insensitive) and Cartesian coordinates.
357
358 {\tt bqtag x y z charge} --- E.g., {\tt Bq9 20.0 20.0 20.0 -10.0} --- The tag must begin with {\tt Bq} (case insensitive). It is used to place a charge at an arbitrary location (e.g., to simulate an external field).
359
360 {\tt pstag x y z charge} --- E.g., {\tt psBe 0.3 -0.1 3.1} --- The tag must begin with {\tt ps} (case insensitive). It is used to put a pseudopotential on selected atoms ({\em I think ???????}).
361
362
363 \chapter{Ground state molecular DFT and HF}
364
365 \begin{verbatim}
366 dft
367 directives/keywords
368 end
369 \end{verbatim}
370
371 {\tt nopen value} --- E.g., {\tt nopen 3} --- The number of unpaired spin orbitals, $n_\alpha - n_\beta$ (default {\tt 0})
372
373 {\tt unrestricted} --- Selects a spin-unrestricted calculation (default is spin restricted)
374
375 {\tt xc value} --- E.g., {\tt xc HF} --- Selects the exchange correlation potential (default is {\tt LDA}). See XC section for more details.
376
377 {\tt aobasis value} --- E.g., {\tt aobasis sto-3g} --- Sets the atomic orbital basis used for the initial guess. Options are {\tt sto-3g} (down to Iodine) or {\tt 6-31g} (down to Zinc, default).
378
379 {\tt charge value} --- E.g., {\tt charge -1.0} --- Total charge (default {\tt 0}) on the molecule. Atomic units.
380
381 {\tt nvalpha value} --- E.g., {\tt nvalpha 2} --- The number of alpha spin virtual orbitals to solve for (default {\tt 0}) --- is this working now?
382
383 {\tt nvbeta value} --- E.g., {\tt nvbeta 2} --- The number of beta spin virtual orbitals to solve for (default {\tt 0}) --- is this working now?
384
385 {\tt no\_orient} --- Do not reorient/translate the molecule to orientation/center.
386
387 {\tt core\_type value} --- E.g., {\tt What is available?} Selects the pseudopotential to be used on all atoms (can also do mixed all-electron/pseudopotential calculation). Not heavily tested and unoptimized. (default is all electron).
388
389 {\tt psp\_calc} --- Perform pseusopotential calculation on all atoms. Not heavily tested and unoptimized. (default is all-electron)
390
391 {\tt L value} --- E.g., {\tt L 50} --- Sets the computational box size to $[-L,L]^3$ atomic units (mostly for testing). Default is to find cube that contains all nuclei and to add 50 atomic units.
392
393 \section{XC --- DFT exchange correlation}
394
395 Without LIBXC, the code just provides either Hartree-Fock ({\tt xc HF}) or local (spin) density approximation ({\tt xc LDA}, the default).
396
397 With LIBXC, in addition to HF and LDA (default) there are wide varity of GGA and hybrid functionals available --- the ones that have been tested (to some extent) have been provided with simple input formats
398 \begin{itemize}
399 \item Becke-Perdew (91?) --- \verb+xc bp+
400 \item Becke-Perdew 86 --- \verb+xc bp86+
401 \item PBE --- \verb+xc pbe+
402 \item PBE-0 --- \verb+xc pbe0+
403 \item B3LYP --- \verb+xc b3lyp+
404 \end{itemize}
405
406 There is also a more general input format in which you can provide a list of functional names (using LIBXC's naming scheme) and the coefficient by which to scale it. For instance, to get PBE-0 you could specify
407 \verb+xc GGA_X_PBE .75 GGA_C_PBE 1. HF_X .25+.
408
409 We have not yet implemented the near linear-scaling algorithm for HF
410 exchange, which as a consequence is fairly slow and may require a lot
411 of memory (there is low-memory algorithm in the code but I don't think
412 it is yet available as an input option).
413
414 \section{Restarting}
415
416 At completion of an HF or DFT calculation, the molecular orbitals are
417 saved in the files \verb+restartdata.*+ (with one file per I/O server
418 process). The projection of the orbtials onto the {\tt sto-3G} AO basis
419 set is saved into the file {\tt restartaodata}.
420
421 {\tt restart} --- Restart from numerical orbitals from a previous calculation (default is no)
422
423 {\tt restartao} --- Restart from projection of orbitals onto AO basis
424 set from a previous calculation (default is no unless doing geometry
425 optimization). If a restart file is not found, or the file contains
426 incompatible data then the default atomic guess is used.
427
428 {\tt save value} --- E.g., {\tt save false} --- Boolean flag to save (or not) orbitals at completion (default is true).
429
430 \section{Controlling convergence and accuracy}
431
432 The default convergence test is on both the 2-norm of change in density per atom (separately for each spin) between iterations and the residual error in each wave function.
433
434 \begin{verbatim}
435 converged = (da < dconv * molecule.natom()) &&
436 (db < dconv * molecule.natom()) &&
437 (conv_only_dens || (max_residual < 5.0 * dconv))
438 \end{verbatim}
439
440 {\tt dconv value} --- E.g., {\tt dconv 1e-5} --- SCF convergence criterion (default 1e-4 atomic units). Suggest decreasing this to {\tt 1e-5} for geometry optimization or property calculations.
441
442 {\tt canon} --- Solves for canonical orbitals or eigenfunctions (default is localized orbitals except for atoms and diatomics).
443
444 {\tt pm} --- Selects use of the Pipek-Mezy localized orbitals (default).
445
446 {\tt boys} --- Selects use of the Boys localized orbitals.
447
448 {\tt maxrotn value} --- E.g., {\tt maxrotn 0.1} --- Used to restrict maximum rotation of orbitals (default {\tt 0.25})
449
450 {\tt maxiter value} --- E.g., {\tt maxiter 20} --- The maximum number of iterations (default is {\tt 20})
451
452 {\tt maxsub value} --- E.g., {\tt maxsub 5} --- The size of the iterative subspace (default is {\tt 5}). Sometimes it helps to make this larger.
453
454 {\tt protocol valuelist} --- E.g., {\tt protocol 1e-4 1e-6 1e-8} --- Sets the solution protocol. The default is `{\tt 1e-4 1e-6} which means solve first using a truncation threshold of {\tt 1e-4} (using $k=6$) and with a threshold of {\tt 1e-6} (using $k=8$).
455
456 {\tt orbitalshift} --- E.g., {\tt orbitalshift 0.1} --- Shifts the occupied orbitals down in energy by the given amount (default {\tt 0}). Is this working?
457
458 {\tt k value} --- E.g., {\tt k 8} --- Sets the wavelet order to a fixed value (mostly only used for testing)
459
460 {\tt convonlydens} --- Just test on the change in the density for convergence.
461
462
463 \section{Geometry optimization}
464
465 By default geometry optimization is performed using the BFGS Hessian update algorithm. The convergence test is on all of the 2-norm of the gradient, the change in the energy between iterations, and the maximum change in Cartesian coordinates (all in atomic units). The
466
467 For geometry optimization it is recommended to select {\tt dconv 1e-5} to obtain more accurate gradients.
468
469 {\tt gopt} --- Requests optimization of the geometry
470
471 {\tt gtol value} --- E.g., {\tt gtol 1e-4} --- Sets the convergence threshold for the 2-norm of the gradient (default {\tt 1e-3}).
472
473 {\tt gtest value} --- E.g., {\tt gtest 1e-4} --- Sets the convergence threshold for the maximum change in Cartesian coordinates (default {\tt 1e-3} atomic units).
474
475 {\tt gval value} --- E.g., {\tt gval 1e-6} --- Sets the available precision in the energy (default is {\tt 1e-5} atomic units).
476
477 {\tt gprec value} --- E.g., {\tt gtest 1e-6} --- Sets the available precision in the gradient (default is {\tt 1e-5} atomic units).
478
479 {\tt gmaxiter value} --- E.g., {\tt gmaxiter 100} --- Sets the maximum number of geometry optimization iterations (default is 20).
480
481 {\tt algopt value} --- E.g., {\tt algopt SR1} --- Selects the quasi-Newton update method (default is ). Options are {\tt BFGS} (default) or {\tt SR1} (not heavily tested). Case sensitive.
482
483 \section{Properties}
484
485 {\tt derivatives} --- Compute the derivates (default is false).
486
487 {\tt dipole} --- Compute the molecular dipole moment (default is false --- why?).
488
489 {\tt response} --- TBD
490
491 {\tt response\_freq} --- TBD
492
493 {\tt response\_axis} --- TBD
494
495 {\tt rconv} --- TBD
496
497 {\tt efield} --- TBD
498
499 {\tt efield\_axis x y z} --- TBD
500
501 {\tt print\_dipole\_matels} --- TBD
502
503 \section{Plotting}
504
505 Plots are generated to OpenDX files. In the {\tt moldft} source directory are two useful files
506
507 \begin{itemize}
508 \item {\tt vizit.net} --- An OpenDX visual program that displays a
509 molecule (from file {\tt molecule.dx}) along with positive+negative
510 isosurfaces (with adjustable value)for a scalar field read from a
511 file.
512
513 \item {\tt moldx.py} --- A Python program you can run with your {\em
514 output} file as standard input to produce a {\tt molecule.dx} file.
515 It is important to use your output file since {\tt moldft} will (by
516 default) translate and rotate the molecular coordinates.
517
518 \end{itemize}
519
520 {\tt plotmos lo hi} --- E.g., {\tt plotmos 10 12} --- Plots the molecular orbitals in the given inclusive range (default is none). Orbitals are numbered from zero. Seems like this needs extending to accomodate unrestricted calculations.
521
522 {\tt plotdens} --- Plots the total electronic charge density and, if spin unrestricted, the spin density (default is off).
523
524 {\tt plotcoul} --- Plots the total (electronic + nuclear) electrostatic potential (default is off).
525
526 {\tt plotnpt value } --- E.g., {\tt plotnpt 501} --- Sets the number of plots used per dimension in the cube of points (default 101).
527
528 {\tt plotcell xlo xhi ylo yhi zlo zhi} --- E.g., {\tt plotcell -10 10 -15 15 -10 5} --- Sets the cell (in atomic units) used for plotting (default is the entire simulation cell).
529
530 \section{Parallel execution}
531
532 {\tt loadbal vnucfac parts} --- E.g., {\tt loadbal 12 2} --- Adjusts data/loadbalance when running in parallel with MPI. {\tt vnucfac} (default 12) is extra weight associated with nuclear potential and {\tt parts} (default 2) is the number of partitions (or subtrees) per node. SCF
533
534 {\tt nio value} --- E.g., {\tt nio 10} --- The number of MPI processes to use as I/O servers (default is 1)
535
536
537 \end{document}