NWChem Home | Security & Privacy | PNNL

NWChem Known Bugs in version 4.6

Below is a list of the known bugs in NWChem 4.6. If you believe that you have found bugs that are not listed here, please send your bug report using the correct channel listed in the Reporting Problems with NWChem page.

1. Direct MP2 optimizations
2. Runtime problems with the BAND module using PGI compilers
3. Runtime problems with the PAW module using PGI compilers
4. Problem with RHF GIAO calculations on TMS molecules using basis sets that contain d functions
5. Problem in TI/PME evaulations
6. Possible compile problems with True64 Compaq Fortran
7. Medium and large size calculation under Linux produce NaNs
8. FPU corruption on IA64 Linux
9. Problem with rabuck option in DFT open shell runs
10. Problem with geometries the first time forces are called in QMD runs
11. Analytical gradients not available for metaGGA functionals
12. Numerical precision of the DFT code
13. Z-matrix failure

Direct MP2 optimizations:

MP2 optimizations that are performed with the direct option can have problems with CPHF convergence. In this case, the user is advised to use the default options which is a semi-direct calculation.

Runtime problems with the BAND module using Portland Group compilers on AMD Opteron and Intel x86:

the BAND module compiled using PGI compilers generates incorrect results, therefore the BAND module of NWChem 4.6 is not active when compiling with PGI compilers.

Runtime problems with the PAW module using Portland Group compilers on AMD Opteron and Intel x86:

the PAW module compiled using PGI compilers generates incorrect results, therefore the PAW module of NWChem 4.6 is not active when compiling with PGI compilers.

Problem with RHF GIAO calculations on TMS molecules using basis sets that contain d functions

A fix will be available in the next release.

Problem in TI/PME evaulations

A fix will be available in the next release.

Possible compile problems with True64 Compaq Fortran

Some versions of True64 Compaq Fortran default f77 to point to f90 and the compile will break because of problems with cpp and f90. This has been reported for V5.3-915. There are three known solutions:

1. Change the f77 link to point to f77 instead of f90,
2. Add the flag -old_f77 as the first flag in the FC definition of the DECOSF section of the $NWCHEM_TOP/src/config/makefile.h file, or
3. Upgrade to version X5.3-1155 plus the patches at http://www6.compaq.com/fortran/dfau.html#updates.

Medium and large size calculation under Linux produce NaNs

2.2 Linux kernels are known to produce random wrong floating point arithmetic (eventually causing NaNs), see for example:

http://www.ccl.net/cgi-bin/ccl/message.cgi?2000+06+06+002
http://www.ccl.net/cgi-bin/ccl/message.cgi?2001+02+23+012
BUG: Global FPU corruption in 2.2
Re: BUG: Global FPU corruption in 2.2

We have experienced the same problems running NWChem with a 2.2.x kernel. This is likely due to FPU problems that has been fixed in kernel 2.2.20 and in the 2.4 series.

http://www.kernel.org/pub/linux/kernel/v2.4/

or to upgrade your 2.2 kernel to version 2.2.20, whose source is available at

http://www.kernel.org/pub/linux/kernel/v2.2/linux-2.2.20.tar.gz

FPU corruption on IA64 Linux:

A kernel bug in the FPU management can cause wrong NWChem result on IA64 Linux boxes. More information at: http://www.gelato.unsw.edu.au/linux-ia64/0305/5681.html http://lia64.bkbits.net:8080/linux-ia64-2.4/cset@1.930.112.42

Problem with rabuck option in DFT open shell runs

A bug in the rabuck convergence option can cause DFT open shell runs to fail as described in

A fix has been put in the NWChem 4.6 source, but not in the NWChem 4.6 binaries available for download.

Problem with geometries the first time forces are called in QMD runs

A bug in the Quantum Molecular Dynamics code is described in

A fix is available on request and is currently available in the latest NWChem Source code distribution. The fix is not available in the NWChem 4.6 binaries.

Analytical gradients not available for metaGGA functionals

Analytical gradients for metaGGA functionals (xpkzb99, cpkzb99, xtpss03 and ctpss03) have not been coded yet. Therefore, numerical gradients need to specified in the input file for metaGGA functionals, e.g.

task dft optimize numerical

More details in

A fix will be available in the next release.

Numerical precision of the DFT code

A bug in the grid-based integration of the XC potential can cause inaccurate results. A patch is available

Patch grid_signf.patch

Install script grid_signf.install

Manual installation

1) cd $NWCHEM_TOP

2) cd ..

3) wget http://www.emsl.pnl.gov/docs/nwchem/support/patches/4.6/grid_signf.patch

4) patch -p0 < grid_signf.patch

5) cd nwchem-4.6/src/nwdft/grid

6) make

7) cd ../..

8) make link

Z-matrix failure

A bug was present in the z-matrix routines as described in http://www.emsl.pnl.gov/docs/nwchem/nwchem-support/2004/09/0034.z-matrix_oddity

A patch is now available

Patch geom_hnd.patch

Installation

1) cd $NWCHEM_TOP

2) cd ..

3) wget http://www.emsl.pnl.gov/docs/nwchem/support/patches/4.6/geom_hnd.patch

4) patch -p0 < geom_hnd.patch

5) cd nwchem-4.6/src/geom

6) make

7) cd ../..

8) make link