comp-chem/NWChem
1
0
Fork 0
mirror of https://github.com/nwchemgit/nwchem.git synced 2026-07-28 06:05:44 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions 1
NWChem/web/support/faq/guess.html
Edoardo Apra 2a32bca10e initial checkin
2006-01-12 18:52:52 +00:00

286 lines
11 KiB
HTML
Raw Permalink Blame History

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
"http://www.w3.org/TR/html4/loose.dtd">
<HTML>
<HEAD>
<TITLE>NWChem Frequently Asked Questions</TITLE>
<link rel="stylesheet" type="text/css" href="../../shared/nwchem_basic.css">
</HEAD>
<body text="#000000" bgcolor="#FFFFFF" link="blue" alink="blue" vlink="blue">
<table width="650"><tr><td>
<br>
<table>
<tbody>
<tr>
<td align="left" width="325"><img src="../../images/nwchem_logo_dark.gif" border=0 width=200 alt="NWChem - computational chemistry on parallel computers"></td>
<td align="right" width="325" valign="bottom">
<a href="../../nwchem_main.html">NWChem Home</a> |
<a href="../../disclaimer.html">Security & Privacy</a> |
<a href="http://www.pnl.gov" target="_blank">PNNL</a>
</td>
</tr>
</tbody>
</table>
<hr>
<table width="650">
<tbody>
<tr>
<td width="350" align="left"></td>
<td width="300" align="right"><a href="basis.html">Next FAQ Catagory</a> | <a href="NWChem_FAQ.html">Return to Main FAQ</a></td>
</tr>
</tbody>
</table>
<br><br>
<h1>NWChem FAQ</h1>
<p><h2>Initial Guess and Convergence Problems</h2></p>
<!--DO NOT REMOVE Begin Question and End Question Comments.-->
<ul>
<!--Begin Question-->
<li> <a href="#0">I am having problems getting my initial guess in the correct order. How I do use the swap directive?</a></li>
<li> <a href="#1">I am having problems getting an open shell sytem to converge. What strategies will help me?</a></li>
<li> <a href="#2">I am having problems converging an ionic, extended system. What can I do?</a></li>
<li> <a href="#3">I am having convergence problems with the final iterations of a calculation using a large basis set. What can I do?</a></li>
<li> <a href="#4">Do you have an example of a complex fragment guess?</a></li>
<!--End Question-->
</ul>
<hr>
<p>
<a name="0"></a>
<font color="purple">I am having problems getting my initial guess in the correct order. How I do use the swap directive?</font>
<p>
Remember that you can <b>(and should!)</b> look at your initial orbitals
by putting the line<br>
<pre>
print "initial vectors"
</pre>
<p>
into the SCF block of your input deck. If the order isn't correct, you can use
the "swap" directive to change the order of the orbitals.
<p>
For example:
<pre>
vectors atomic swap 173 175 174 176 output end.movecs
</pre>
will cause the orbitals 173 - 175 to be swapped to 175, 176, 173 and 174.
Note that the swaps are pairwise: first 173 and 175 are swapped, then the
the current orbitals 174 and 176 are swapped. More information on initial orbitals
can be obtained in <a href="http://www.emsl.pnl.gov/docs/nwchem/doc/user/node12.html#SECTION001250000000000000000" target="_blank">
Section 10.5</a> of the <a href="../doc/user/index.html" target="_blank">User's Manual</a>.
</p>
<hr>
<p>
<a name="1"></a>
<font color="purple"> I am having problems getting an open shell system to converge. What strategies will help me?</font>
<p>
Open shell systems, especially those containing transition metals or heavy
elements with many unoccupied d and f orbitals, can be especially hard
to converge. Running a spin unrestricted calculation can exacerbate the
problem since all other possible spin states can mix into the wavefunction.
It may also be true that your particular calculation has multireference
character and, by using SCF or DFT, you are forcing it into a single
determinate wavefunction.
<p>
That being said, to converge your system, you need to guide the code
to the particular state that you want. This may be a very difficult
task and different starting guesses can converge to different solutions,
especially if there is symmetry breaking. Here are four different
suggestions to help you with your problem:
<ol>
<li> Use Auf bau construction. Start by removing most or all of the
open-shell electrons and converge the resulting closed-shell wavefunction.
Incrementally add the missing electrons, maintaining a closed-shell form for
the wavefunction until the last calculation. Once you have an ROHF
wavefunction converged, then do the UHF calculation if desired (note that
UHF is the only option with the DFT code). When adding electrons, pay
attention to how well separated the virtual orbitals are in energy.</li>
<li> Converge a very high multiplicity state and then incrementally
pair electrons until you get the state you want.</li>
<li> Converge in a minimal basis set where there is very little variational
freedom and it is not as expensive to run many iterations. Then
<a href="http://www.emsl.pnl.gov/docs/nwchem/doc/user/node12.html#SECTION001250000000000000000">project</a>
to a larger basis set (you can use this
<a href="http://www.emsl.pnl.gov/docs/nwchem/support/proj.nw">example</a>
as a model).
It is also useful to use spherical functions since this
tends to decrease the amount of linear dependence. This method isn't
as reliable as the previous methods.</li>
<li> Use the <a href="http://www.emsl.pnl.gov/docs/nwchem/doc/user/node12.html#SECTION001251000000000000000">fragment</a>
guess. This is especially useful if you know the configuration of the atoms.
Prepare the atoms in the appropriate state and assemble the molecule from the fragments.
If you want an antiferromagnetic state, this is the only way to do it.</li>
</ol>
</p>
<hr>
<p>
<a name="2"></a>
<font color="purple"> I am having problems converging an ionic, extended system. What can I do?</font>
<p>
This problem usually shows up when a minimum basis set is used and where a
lot of charge has to be moved a long way but the minimal basis does not
providea lot of coupling between the orbitals. The initial guess may be
modified by
<ol>
<li><a href="http://www.emsl.pnl.gov/docs/nwchem/doc/user/node12.html#SECTION001252000000000000000">putting the charge on the atoms</a>
where it is believed to be,</li>
<li> look at the orbitals and
<a href="http://www.emsl.pnl.gov/docs/nwchem/doc/user/node12.html#SECTION001250000000000000000">swap the orbitals</a>
to obtain the correct occupation, or</li>
<li> ue the <a href="http://www.emsl.pnl.gov/docs/nwchem/doc/user/node12.h
tml#SECTION001251000000000000000">fragment</a> guess.</li>
</ol>
Better yet, use a better basis set which will allow for better coupling of
the orbitals!
</p>
<hr>
<p>
<a name="3"></a>
<font color="purple"> I am having convergence problems with the final iterations of a calculation using a large basis set. What can I do?</font>
<p>
This problem generally occurs when high precision is necessary because of
near linear dependencies in the basis set and there is insufficient precision
in the computed gradient or hessian vector product to converge to the
requested threshold. The first item to check is to make sure that the convergence
<a href="http://www.emsl.pnl.gov/docs/nwchem/doc/user/node12.html#SECTION001270000000000000000">threshold</a> is realistic (not below 1e<sup>-7</sup> or
perhaps 1e<sup>-8</sup>). If doing an MP2 gradient with the
<a href="http://www.emsl.pnl.gov/docs/nwchem/doc/user/node17.html#SECTION001712000000000000000">TIGHT</a> directive, make sure the SCF input block is
after the MP2 block and then in the SCF block put
<pre>
SCF
thresh 1e-7
tol2e 1e-10
END
</pre>
This may still not fix the problem if the calculation is using the semi-direct
algorithm. In this case:
<ol>
<li>Run the calculation on enough nodes to run fully disk resident,</li>
<li>Run as before, but switch off the density screening with<p>
set fock:densityscreen f<p></li>
<li>Run the calculation fully direct.</li>
</ol>
<hr>
<p>
<a name="4"></a>
<font color="purple"> Do you have an example of a complex fragment guess?</font>
<p>
The following is an advanced fragment guess (NOT for beginners!). It consists
of two Fe<sup>3+</sup> ions in a simulated lattice and the objective is
to compute the energy difference between the ferro and anti-ferromagnetic
states. This is accomplished by first using the atomic guess to do an ROHF
calculation on an Fe<sup>3+</sup> ion with d<sup>5</sup> occupation. The
next calculation is on the high spin system and it merely has to start from
the two ions ... the open shell orbitals will automatically be contiguous and
so nothing else is needed.
<p>
The final calculation is a UHF for which the 5 open shell electrons on one
atom must be spin up and on the other spin down. The initial guess orbitals
start off from the previous high spin solution.
However, the d orbitals
on the second atom must have the beta orbitals in the right place, hence
the swap directive. The net result is a completely localized guess with the
desired spin coupling.
<pre>
start fe2
geometry noautoz autosym
Fe 1.45 0.071 0.84
Fe -1.45 0.071 -0.84
Bq 3.02 0.84 0.03 charge -1.
Bq -0.00 1.04 0.03 charge -1.
Bq 1.33 -1.67 0.03 charge -1.
Bq 2.78 -0.69 2.33 charge -1.
Bq 1.45 1.61 2.33 charge -1.
Bq 0.11 -0.69 2.33 charge -1.
Bq -1.33 -1.67 0.03 charge -1.
Bq -3.02 0.84 0.03 charge -1.
Bq -0.12 -0.69 -2.25 charge -1.
Bq -1.45 1.61 -2.25 charge -1.
Bq -2.78 -0.69 -2.25 charge -1.
end
geometry Fe
Fe 0 0 0
symmetry c2v
end
basis
Fe library sto-3g
end
set atomscf:tags_z Fe
set atomscf:z 3.
title "Initial calculation on Fe3+"
set geometry Fe
charge 3
scf; rohf ; nopen 5; vectors atomic output fe.mos; end
task scf
title "Ferro-magnetic state"
unset geometry
charge -5
scf; print "initial vector analysis"; rohf; nopen 10
vectors fragment fe.mos fe.mos output feferr.mos;
maxiter 0
print mulliken
end
task scf ignore
title "Anti-ferro-magnetic state"
scf; uhf; nopen 0;
vectors input feferr.mos output feantif.mos\
swap beta 19 24 20 25 21 26 22 27 23 28
maxiter 99
end
task scf
</pre>
For DFT calculation you can either the same method just shown or a
different approach that is illustrated in this
<a href="http://www.emsl.pnl.gov/docs/nwchem/support/fe2dft.nw">example</a>.
</p>
<p>
<hr>
<font size="0">
<center>
<a href="../../nwchem_main.html">NWChem</a> |
<a href="../../capabilities/nwchem_capab.html">Capabilities</a> |
<a href="../../platforms/platforms.html">Platforms</a> |
<a href="../../download.html">Download</a> |
<a href="../../doc/user/index.html">User's Manual</a> |
<a href="../../doc/prog/index.html">Programmer's Manual</a> |
<a href="../../release-notes/index.html">Release Notes</a> |
<a href="NWChem_FAQ.html">FAQ</a>
</center>
<br>
<center>
<a href="../known_bugs.html">Known Bugs</a> |
<a href="../support.html">Support</a> |
<a href="../../training/training.html">Tutorial</a> |
<a href="../../contributors.html">Contributors</a> |
<a href="../../benchmarks/index.html">Benchmarks</a> |
<a href="http://www.emsl.pnl.gov/forms/search_nwchem.html">Search</a> |
<a href="http://www.emsl.pnl.gov/docs/mssg/index.html" target="_blank">Mol Sci. Soft. Group</a> |
<a href="../../citation.html">Citation</a>
</center>
</font>
<hr>Contact: <a href="mailto:ms3distribution@emsl.pnl.gov">NWChem Support</a><br>
Updated: February 22, 2005
</td></tr>
</table>
</body>
</html>
Reference in a new issue View git blame Copy permalink