Using MoleCuilder in a docker container

Since the release of version v1.6.0, MoleCuilder is available as a debian package (installable at least under Ubuntu 14.04).

Since that it is pretty easy to build a docker image such that MoleCuilder can be run inside a docker container. A Dockerfile is provided in the same blog post.

To make it even easier, I created a public repository at ​Docker, where the image can be pulled from directly: frederikheber:molecuilder/latest

In other words and assuming you have a working docker installation at hand, all you need to do to try the latest version is the following.

docker pull frederikheber/molecuilder:latest
docker run -it frederikheber/molecuilder:latest

This will open a root shell in the container. Type

molecuilder --version

And you'll see the version installed.

For computations you need both molecuilder_server and at least one instance of molecuilder_poolworker running. The Poolworker is the number cruncher. It's a modified version of ​MPQC. The Server accepts computation jobs from the MoleCuilder program and pushes them on to an idle worker. We start either one in the background, logging to a file:

molecuilder_server --controllerport 10024 --workerport 10025 2&>server.log &
molecuilder_poolworker --server 127.0.0.1:10025 --listen 10026 --hostname 127.0.0.1 2&>worker.log &

The server listens to MoleCuilder on port 10024, the Worker enrolls with the Server on port 10025, and finally the Worker accepts jobs on 10026. Simply choose any ports available. You can also add more than one worker at a time using GNU's parallel, which is not installed on the container,

apt-get install parallel
parallel --gnu molecuilder_poolworker --server 127.0.0.1:10025 --hostname 127.0.0.1 --listen ::: 10027 10028 &

Check the server.log file to see that indeed 3 workers are in the queue now.

Well, that's all. Would you like to calculate the energy of a methane molecule?

molecuilder \
  --input methane.data  \
  --set-output tremolo \
  --set-tremolo-atomdata "id type x=3 F=3 neighbors=4" --reset 1 \
  --set-parser-parameters mpqc --parser-parameters "basis=STO-3G;theory=CLHF;" \
  --add-atom C --domain-position "10,10,10" \
  --select-atom-by-element C --saturate-atoms \
  --select-all-atoms --optimize-structure --steps 10 --deltat 1. --server-address 127.0.0.1 --server-port 10024

This should bring the largest remaining force components to about 1e-5 a.u..

Don't be overwhelmed by the lot of typing and let's explain step by step:

• First, we specify an input file (which is empty but will contain the methane on exit).
• Next, we set the output format, namely tremolo.
• And afterwards, we need to specify the tremolo format to contain the id of the atom, its element, three spatial and three gradient coordinates and at most four (covalent) neighbor ids.
• The ab-initio calculations will be done at Closed-Shell Hartree Fock (CLHF) with a simple STO-3G basis set. As we use MPQC as number cruncher, we have to give these information to the parser mpqc.
• Then, finally, we create the methane molecule by adding a carbon atom and saturating it.
• Last, all atoms are selected and the structural optimization is performed for 10 steps with an initial step length of 1 a.u.. Notice that we give the port 10024, which we told the Server to use for listening to molecuilder's jobs.

Wanna take it from here? Increase the optimization steps. Choose another basis set, e.g. 6-31G (any of MPQC's sets is fine), or Moeller-Plesset Perturbation theory 2nd order (MBPT2), ...

Some pro tips:

• have one executable running per container, i.e. start you server as

  docker run --name "MoleCuilderServer" -p 10024:10024 -p 10025:10025 frederikheber/molecuilder:latest \
  /usr/bin/molecuilder_server --signal 2 15 --controllerport 10024 --workerport 10025
  

• similarly, start you workers each in a single container, too, where we automatically extract the servers IP address

  SERVERIP=`docker inspect MolecuilderServer | grep IPAddress | tail -n 1 | awk -F":" '{print $2}' | tr -d \"\ ,`
  docker run --name "MoleCuilderWorker" -p 10026:10026 frederikheber/molecuilder:latest \
  mpirun -np 1 /usr/bin/molecuilder_poolworker --signal 2 15 --server ${SERVERIP}:10025 --listen 10026
  

• use GNU's parallel to start e.g. 8 workers at once, assume the above is contained in script start_worker.sh

  parallel --gnu --delay 1 --ctrlc ./start_worker.sh ::: `seq 1 8`
  

• you can also set up docker to start containers automatically at startup, see ​here. But this seems to be changing in the docker API.
• install molecuilder on you local machine and use molecuildergui for the methane optimization example above.