Features in brief:

• based on finite elements (suitable for non-cuboidal structures)

• problem description in Python, therefore high degree of flexibility

• inbuilt mesh post processing tools

• efficient data storage (binary compressed) and extraction into vtk files (of course the raw data can be extracted)

• arbitrary crystal anisotropy

• pseudo) periodic boundary conditions (“macro geometry approach”)

• Spin torque transfer (Zhang-Li model for uniform current density)

• Supports use of matrix compression library (HLib) for BEM

• extensive documentation in html and pdf, including Tutorial

Nmag’s home page is at http://nmag.soton.ac.uk/nmag/.

Nmag is supported on clusters Atlas4, Atlas5, Atlas6 and Atlas7. Nmag can run well on multiple CPU-cores within one compute server. The version installed on the HPC clusters is 0.20.

To run Nmag software, you need to run command “nmag_install” once on any cluster:

> nmag_install

After that you can enter command “nsim” to start Nmag software interface.

> nsim

To process large simulations that needs hours, days or even weeks, Nmag simulation can be submitted to batch LSF queue and executed in batch mode.

For a small model that can be processed on one CPU core, you can submit the simulation as:

> bsub -q linux64 -o stderr.o nsim case1.py

For a large model that needs to be processed on multiple CPU cores, you can submit the simulation as:

> bmpionenode -q atlas4_parallel -o stderr.o -n 4 nsim case1.py

> bmpionenode -q atlas5_parallel -o stderr.o -n 8 nsim case1.py

> bmpionenode -q atlas6_parallel -o stderr.o -n 6 nsim case1.py

In the above commands, “cas1.py” is the input file to Nmag; “stderr.o” is the file to store the job information and error messages if any. The suggested number of CPU cores for queue atlas4_parallel is (-n 4 or -n 8), atlas5_parallel is (-n 8), and atlas6_parallel is (-n 6 or -n 12).