modified on 23 February 2016 at 11:20 ••• 18,461 views

FAQ

From NWChem

(Difference between revisions)
Jump to: navigation, search
(I get the message: ! warning: processed input with no task. What is wrong?)
(AUTOZ fails to generate valid internal coordinates. Now what?)
Line 40: Line 40:
Once the obvious has been checked, there are several possible modes of failure, some of which may be worked around in the input.
Once the obvious has been checked, there are several possible modes of failure, some of which may be worked around in the input.
-
Strictly linear molecules with 3 or more atoms. AUTOZ does not generate linear bend coordinates, but, just as in a real Z-matrix, you can specify a dummy center that is not co-linear. There are two relevant tips:
+
1. Strictly linear molecules with 3 or more atoms. AUTOZ does not generate linear bend coordinates, but, just as in a real Z-matrix, you can specify a dummy center that is not co-linear. There are two relevant tips:
i) constrain the dummy center to be not co-linear otherwise the center could become co-linear. Also, the inevitable small forces on the dummy center can confuse the optimizer.
i) constrain the dummy center to be not co-linear otherwise the center could become co-linear. Also, the inevitable small forces on the dummy center can confuse the optimizer.
ii) put the dummy center far enough away so that only one connection is generated.  
ii) put the dummy center far enough away so that only one connection is generated.  
Line 69: Line 69:
</pre>
</pre>
-
Larger molecules that contain a strictly linear chain of four or more atoms (that ends in a free atom). For these molecules the autoz will fail and the code can currently not recover by using cartesians. One has to explicitly define noautoz in the geometry input to make it work. If internal coordinates are required one can fix it in the same manner as described above. However, you can also force a connection to a real nearby atom.
+
2. Larger molecules that contain a strictly linear chain of four or more atoms (that ends in a free atom). For these molecules the autoz will fail and the code can currently not recover by using cartesians. One has to explicitly define noautoz in the geometry input to make it work. If internal coordinates are required one can fix it in the same manner as described above. However, you can also force a connection to a real nearby atom.
-
Very highly connected systems generate too many internal coordinates which can make optimization in redundant internals less efficient than in Cartesians. For systems such as clusters of atoms or small molecules, try using a smaller value of the scaling factor for covalent radii
+
 
 +
3. Very highly connected systems generate too many internal coordinates which can make optimization in redundant internals less efficient than in Cartesians. For systems such as clusters of atoms or small molecules, try using a smaller value of the scaling factor for covalent radii
<tt>    zcoord; cvr_scaling 0.9; end </tt>
<tt>    zcoord; cvr_scaling 0.9; end </tt>

Revision as of 12:53, 29 January 2016

NWChem Frequently Asked Questions

Contents

General information about NWChem

Where is the User's Manual?

NWChem User's Manual

Where do I go for help with a Global Arrays problem?

If you have problems with compiling the tools directory, please visit the Global Arrays Google group at http://groups.google.com/group/hpctools or visit the Global Arrays website at http://hpc.pnl.gov/globalarrays/

Where do I go for help with NWChem problems?

Please post your NWChem issue to the NWChem Community forums at http://www.nwchem-sw.org//index.php/Special:AWCforum

Where do I find the instructions for installing NWChem?

For updated instructions for compiling NWChem please visit the following URL http://www.nwchem-sw.org/index.php/Compiling_NWChem

Installation Problems

How to fix configure: error: could not compile simple C MPI program

When compiling the tools directory, you might see the compilation stopping with the message configure: error: could not compile simple C MPI program This is most likely due to incorrect settings for the MPI_LIB, MPI_INCLUDE and LIBMPI environment variables. The suggested course of action is a) to use NWChem 6.6, b) unset all of the three variables above and c) point your PATH env. variable to the location of mpif90

What's this business with ARMCI and ARMCI_NETWORK?

ARMCI is a library used by Global Arrays (both ARMCI and GA source code is located in NWChem's tools directory). More information can be found at the following URL http://hpc.pnl.gov/armci If your installation uses a fast network and you are aiming to get optimal communication performance, you might want to assign a non-default value to ARMCI_NETWORK. The following links contained useful information about ARMCI_NETWORK: Choosing the ARMCI library and Setting up the proper environment_variables when compiling NWChem

Input Problems

I get the message: ! warning: processed input with no task. What is wrong?

Have you used emacs to create your input file? Emacs usually does not put and an end-of-line as a last character of the file, therefore the NWChem input parser ignores the last line of your input (the one containing the task directive). To fix the problem, add one more blank line after the task line and your task directive will be executed.

AUTOZ fails to generate valid internal coordinates. Now what?

If AUTOZ fails, NWChem will default to using Cartesian coordinates (and ignore any zcoord data) so you don't have to do anything unless you really need to use internal coordinates. An exception are certain cases where we have a molecule that contains a linear chain of 4 or more atoms, in which case the code will fail (see item 2. for work arounds). For small systems you can easily construct a Z-matrix, but for larger systems this can be quite hard.

First check your input. Are you using the correct units? The default is Angstroms. If you input atomic units but did not tell NWChem, then it's no wonder things are breaking. Also, is the geometry physically sensible? If atoms are too close to each other you'll get many unphysical bonds, whereas if they are too far apart AUTOZ will not be able to figure out how to connect things.

Once the obvious has been checked, there are several possible modes of failure, some of which may be worked around in the input.

1. Strictly linear molecules with 3 or more atoms. AUTOZ does not generate linear bend coordinates, but, just as in a real Z-matrix, you can specify a dummy center that is not co-linear. There are two relevant tips: i) constrain the dummy center to be not co-linear otherwise the center could become co-linear. Also, the inevitable small forces on the dummy center can confuse the optimizer. ii) put the dummy center far enough away so that only one connection is generated.

E.g., this input for acetylene will not use internals

           geometry
             h  0  0  0
             c  0  0  1
             c  0  0  2.2
             h  0  0  3.2
           end

           but this one will

           geometry
             zcoord
               bond    2 3  3.0  cx constant
               angle 1 2 3 90.0 hcx constant
             end
             h  0  0  0
             c  0  0  1
             x  3  0  1
             c  0  0  2.2
             h  0  0  3.2
           end

2. Larger molecules that contain a strictly linear chain of four or more atoms (that ends in a free atom). For these molecules the autoz will fail and the code can currently not recover by using cartesians. One has to explicitly define noautoz in the geometry input to make it work. If internal coordinates are required one can fix it in the same manner as described above. However, you can also force a connection to a real nearby atom.

3. Very highly connected systems generate too many internal coordinates which can make optimization in redundant internals less efficient than in Cartesians. For systems such as clusters of atoms or small molecules, try using a smaller value of the scaling factor for covalent radii

zcoord; cvr_scaling 0.9; end

In addition to this you can also try specifying a minimal set of bonds to connect the fragments.

If these together don't work, then you're out of luck. Use Cartesians or construct a Z-matrix.