Hi Anders,
The mpirun command you see is not something that is inserted by Amber. It’s handled by ORCA via the ORCA input file where you requested 10 processes (%pal nprocs 10 end).
The abort you observe is caused by executing an MPI program (ORCA) by system call from an MPI program (sander.MPI). It’s possible but requires careful setup like clearing environment variables that the MPI runtime uses and is hard to debug.
The simplest solution is to use serial sander (ORCA will still run in parallel). In general, the effect on performance will be negligible because the ab initio QM portion of the calculation is much more expensive and the classical force field.
Can you please try to use sander instead of “srun sander.MPI” and report back whether that works? Thanks.
All the best,
Andy
—
Dr. Andreas W. Goetz
Associate Research Scientist
San Diego Supercomputer Center
Tel: +1-858-822-4771
Email: agoetz.sdsc.edu
Web: www.awgoetz.de
On May 26, 2026, at 5:13 PM, Anders Frederiksen via AMBER <amber.ambermd.org> wrote:
Hi Clorice,
thank you very much for this input. I have tried to implement your path definitions in my slurm script, however still to no avail. I have below pasted the relevant part of my slurm script to start the simulation
ml PDC/24.11
ml amber/24-cpeGNU-24.11-ambertools-24-amdgpu
ml openmpi/5.0.8
ml orca/6.1.1
ORCA_PATH=/pdc/software/24.11/other/orca/6.1.1/
ORCARUN=/pdc/software/24.11/other/orca/6.1.1/orca
export PATH="/pdc/software/24.11/other/orca/6.1.1/orca:$PATH"
srun $AMBERHOME/bin/sander.MPI -O -i colvar.mdin \
-o orc_test.mdout \
-p neutral.prmtop \
-c rewrapped.rst7 \
-r cont.rst7 \
-x orc_test.nc \
similarly, my orca template contains
%pal nprocs 10 end
which ORCA correctly catches
************************************************************
* Program running with 10 parallel MPI-processes *
* working on a common directory *
************************************************************
Unfortunately, the simulation still crashes
Calling Command: mpirun -np 10 /pdc/software/24.11/other/orca/6.1.0/orca_startup_mpi orc_job.int.tmp orc_job
[file orca_tools/qcmsg.cpp, line 394]:
.... aborting the run
The mpirun call is not something I have inserted anywhere, it seems to me to happen on ambers part, though I am not sure if it is an issue that stems from the way the program was compiled on the cluster or if there is another underlying issue.
I will try and mail this thread to Andreas in the hopes that he has a solution.
Best,
Anders
________________________________
From: Clorice Reinhardt <Clorice.Reinhardt.uvm.edu<mailto:Clorice.Reinhardt.uvm.edu>>
Sent: Monday, May 25, 2026 11:54:21 PM
To: amber.ambermd.org<mailto:amber.ambermd.org>; Anders Frederiksen
Subject: Re: [AMBER] bug when running QM/MM with orca as external program
Hi Anders,
I have not any problems with Amber/ORCA interface. I'm not sure why you are calling ORCA in this way... Amber calls ORCA when it needs the forces on the atoms at each time step, or to perform the minimization.
Therefore, you only write the command to call AMBER not ORCA in your job submission file. The original QM/MM interface paper has a nice flow chart if your confused on the flow of information. Amber writes the input files for orca at each minimization or time step, you do not need to do this for it.
For example, my slurm file looks like the following for QM/M I do (with some information redacted 🙂 )
module load amber/24
echo $AMBERHOME
module load gcc/13.3.0
module load openmpi/5.0.5
module load orca/6.0.1
## amber can't find orca unless you explicitly tell it where the paths are. Probably could remedy this by passing the path while compiling, but I didn't compile the version on the cluster and it works fine
ORCA_PATH=YOURPATH
ORCARUN=YOURPATHTOEXECUTEABLE
export PATH="/gpfs1/sw/x86_64/rh7/pkgs/orca/6.0.0/orca:$PATH"
top=my_parmtop
$sander -O -i qmmm_md.in -o qmmm_md.out -p ${top}.prmtop -c YOUR_RESTART.rst -r qmmm_md.rst -x qmmm_md.netcdf
I read in an orca template file with into amber (contain these among others):
%pal nprocs 24 end
%maxcore 3600
Confirming it worked (looking at orc_job.dat which gets produced each time orca is called and is the default name of the output file):
" ************************************************************
* Program running with 24 parallel MPI-processes *
* working on a common directory *
************************************************************
------------------------------------------------------------------------------
ORCA STARTUP CALCULATIONS
-- RI-GTO INTEGRALS CHOSEN --
------------------------------------------------------------------------------
"
Your happy to reach out to me if you have questions. I run a ton of QM/MM and have dug into the source code and used the interface with Qchem, Gaussian and ORCA without a hitch, but Andreas Goetz is who'd I reach out if you have questions. He wrote the original paper on the interface, and he's super nice.
https://urldefense.com/v3/__https://scholar.google.com/citations?user=PJusscIAAAAJ&hl=en__;!!Mih3wA!EKWJmUkOC3zyw0BXi5O1toowNp6rDZFQ0X-0ypz9ULQkuwQUb8spo46Fq-W47mW2CQp6sKF2_IaSMw$
Clorice
________________________________
From: Anders Frederiksen via AMBER <amber.ambermd.org<mailto:amber.ambermd.org>>
Sent: Monday, May 25, 2026 9:52 AM
To: amber.ambermd.org<mailto:amber.ambermd.org> <amber.ambermd.org<mailto:amber.ambermd.org>>
Subject: [AMBER] bug when running QM/MM with orca as external program
Dear amber forum,
I have in the last week been working with the QM/MM interface of the
amber programme and especially focussing with performing a QM/MM
simulation with orca as the external program.
I have had great success running with orca as external program on small
test systems for short simulations with sander running on multiple
processors through MPI and orca running on single core through num
threads = 1.
However, if I want to run any kind of multi processing the QM part of
the simulation crashes with the error
ORCA finished by error termination in orca_util
Calling Command: mpirun -np 2
/pdc/software/24.11/other/orca/6.1.0/orca_util_mpi orc_job orca_hostfile
orc_job
[file orca_tools/qcmsg.cpp, line 394]:
.... aborting the run
This in principle is not surprising if we take a quick look at the orca
documentation where it states:
Do not start the ORCA driver with mpirun!
I have tested and if I use ORCA the intended way, that is calling ORCA
with its full path followed by the input file it runs just fine on
multiple processors and therefore the fix for the amber call should be
quite simple: Remove the "mpirun -np #MPIprocessors" call when orca is
initiated.
I should not that this should not be a version dependent thing for orca
as I have found documentation all the way back in ORCA 3 (i.e.
documentation from 2015), stating that starting ORCA with mpirun will
make it crash.
I have tried to work around the issue by using a template file
(use_template =1) however sadly AMBER seems to read the nprocs flag and
will make ORCA run through mpirun, leading to the same error as if
initiated with AMBER input.
If anyone have found another workaround I am all ears, though I hope
this mail reaches a developer as the fix on paper should be rather
simple, though given my inexperience in development it could not be as
simple as I want it to be.
Best,
Anders Frederiksen
Postdoctoral researcher
KTH, department of Chemistry, Stockholm, Sweden.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org<mailto:AMBER.ambermd.org>
https://urldefense.com/v3/__http://lists.ambermd.org/mailman/listinfo/amber__;!!Mih3wA!EKWJmUkOC3zyw0BXi5O1toowNp6rDZFQ0X-0ypz9ULQkuwQUb8spo46Fq-W47mW2CQp6sKGuaycqdQ$
_______________________________________________
AMBER mailing list
AMBER.ambermd.org<mailto:AMBER.ambermd.org>
https://urldefense.com/v3/__http://lists.ambermd.org/mailman/listinfo/amber__;!!Mih3wA!EKWJmUkOC3zyw0BXi5O1toowNp6rDZFQ0X-0ypz9ULQkuwQUb8spo46Fq-W47mW2CQp6sKGuaycqdQ$
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue May 26 2026 - 21:30:05 PDT
