NAMD Wiki: NamdKnownBugs
Resolved or unresolved issues, by version. New issues and versions added to top. Please let us know if you come across an easily reproducible example of any of the unresolved issues below. Also feel free to add workarounds or other insights below.
Please note that when a bug is "fixed" that means it should not appear in newer versions, not that we have rebuilt all of the binaries on the download page. If you need the fix for a bug in the most recent version you will need to download the nightly build or compile from the source code in CVS.
2.11b1, 2.11b2, 2.11
Beginning with 2.11b1 (days before the release) DCD trajectory files written in parallel by memory optimized builds are corrupted (only contain data from a single writer). Fixed in CVS.
Minimizer may appear to hang while stuck in an infinite loop. Fixed in CVS.
Targetted MD was applying the right forces to the wrong atoms for parallel runs. Fixed in CVS.
If you have switching off and don't specify a cutoff distance you will get a random value for cutoff (typically zero but possibly any other floating point number). Always specify cutoff distances. Fixed in CVS.
When using the coorfile command to read a trajectory containing periodic cell coordinates (e.g., when using the "NAMD Energy" plugin of VMD) the alpha and gamma angles are swapped. This is only a problem for non-orthogonal (e.g., hexagonal) periodic cells. Fixed in CVS.
Parallel runs will often crash (segment fault) during startup phase 1 when CMAP crossterms are present in the psf file. Fixed.
NAMD will report zero energies on pair interaction runs when reading a DCD trajectory file that contains periodic cell information if the config file does not specify periodic boundaries. This is easy to spot if the "PATCH GRID IS" line on startup doesn't say "(PERIODIC)" in all three dimensions, but you see volume and pressure in the energy output. Fixed.
NAMD will hang during the first timestep if any of the "global" force methods such as TclForces, IMD, SMD, or conformational free energy are used and more than 64 processors are being used and the simulation has more patches (see the patch grid dimension output and multiply) than processors. Fixed.
Net forces from pair interactions calculations (pairInteraction on) are incorrect. Forces are accumulated with essentially random signs that are consistent between runs. For a single pair of atoms, this may produce a sign error, while for larger numbers of atoms the net forces are effectively random. The calculated interaction energies are correct, but the forces are meaningless. Fixed.
There is a report of "BUG ALERT: Stray PME grid charges detected!" after 40,000 steps when running Linux-i686 on an Opteron cluster, but not for Linux-i686-TCP.
There is a similar report of errors from Linux-i686 but not from Linux-i686-TCP. I was able to reproduce this locally. Adding +netpoll to the namd2 command line for Linux-i686 results in stable operation. PPL is looking into it.
Eventually resolved as an icc issue, and the following text added to release notes:
If you are building a non-smp, non-tcp version of net-linux with the Intel icc compiler you will need to disable optimization for some files to avoid crashes in the communication interrupt handler. The smp and tcp builds use polling instead of interrupts and therefore are not affected. Adding +netpoll to the namd2 command line also avoids the bug, but this option reduces performance in many cases. These commands recompile the necessary files without optmization:
cd charm/net-linux-icc /bin/rm tmp/sockRoutines.o /bin/rm lib/libconv-cplus-* ( cd tmp; make charm++ OPTS="-O0 -DCMK_OPTIMIZE=1" )
There are reports of floating point exceptions, infinite loops, or NaN (not a number) in output when very small systems are minimized for 10,000 steps or so. Not sure which platforms are affected, but I can get a divide by zero due to floating point underflow on Alpha. Fixed by comparing divisor to zero first.
These meaningless warnings are printed, once per processor:
Warning: FOUND 0 ATOMS WITH ZERO OR NEGATIVE MASSES! CHANGED TO 0.001 Warning: FOUND 0 ATOMS WITH ZERO OR NEGATIVE MASSES! CHANGED TO 0.001 Warning: FOUND 0 ATOMS WITH ZERO OR NEGATIVE MASSES! CHANGED TO 0.001 Warning: FOUND 0 ATOMS WITH ZERO OR NEGATIVE MASSES! CHANGED TO 0.001 ...
Please ignore these. Simple missing logic, fixed.
Dies during startup for Amber rigid waters with:
FATAL ERROR: CAN'T FIND BOND PARAMETERS FOR BOND HW - HW IN PARAMETER FILES
- In psfgen the readpsf command may introduce spurious data in the beta field of the read atoms. Fixed for future versions.
- The MOLLY method is broken. Fixed for future versions.
- NAMD will fail to write final output files when minimization is on. This has been present since version 2.2b2. The workaround is to use the Tcl scripting capability. Rather than specifying the parameters "minimization on" and "numsteps 1000", for example, make the last line in the file "minimize 1000". Fixed for future versions.
- At startup, on workstation networks, conv-host will occasionally report "rsh phase failed". This is fixed in newer versions of Converse which will be used in future versions of NAMD.
- Binary file (DCD trajectory and binary coordinate and velocity) reading and writing are broken on Windows only because files are opened in text mode. Fixed for future versions.
- NAMD will report "BUG ALERT: Bad global exclusion count!" and exit when nonbondedFreq is not one. Fixed for future versions.
- Very old CHARMM parameters that utilize an effective electron count rather than a well depth to specify nonbonded interactions will be misread and cause floating point exceptions (on some architectures) or grossly incorrect energies (if atoms of the affected types, which are usually noble gasses, are present in the system). These parameters are distinguished by the second number in the nonbonded specification being large and positive rather than small and negative (CHARMM format only). The next release will ignore these parameters and issue warnings.
- We have discovered a sporadic transmission bug in Converse which is exposed by compiling Charm++/Converse with recent GNU compilers on Intel platforms (Linux and Solaris-x86). This problem is fixed in the new 2.1b4 binaries.
- The Alpha binaries compiled with -fp_reorder, which were up for download until 9/23/99, incorrectly subtracted the center of mass velocity at the start of the simulation. This seems to be a compiler bug so a new binary was put up for download.
- When running minimization the maximumMove parameter is not set by default. You should manually set it to 1.0 or less. Fixed.
- Skips some nonbonded interactions if only last atom on patch is not fixed. Only happens in very unusual systems with fixed atoms. Fixed.
- On the T3E anything using random numbers explodes. Fixed.
- XSC and XST files written with vectors instead of floats. Fixed.
- Output files are not written when firstTimeStep != 0. Fixed.
- SMD, Tcl, and free energy don't deal with periodic boundaries. Fixed.
- Pressure incorrect when using MOLLY. Fixed.
- FullDirect leaks memory and behaves oddly. Fixed.
- FreeEnergy chokes on whitespace and long input files. Fixed.
NAMD 2.0 beta 3
- Issues BUG ALERT and exits with fixed atoms. Fixed.
NAMD 2.0 beta 2
- Fails during startup on more that 28 processors under Linux. Fixed.
- Velocity rescaling prints messages but does nothing. Fixed.