Effective bandstructure that goes beyond quantuamatk limitation

[sukhito teh]

Dear developers and users,
I am trying to plot the effective bandstructure of a MoS2 Moire Superlattice. Unfortunately the number of atoms in a "unit cell" of Moire Superlattice is not always equal to that of primitive cell, thus I am unable to use the built in EffectiveBandstructure analysis. I humbly ask if any of you can answer the following questions:
1. Is there any workaround to plot the effective bandstructure using quantumatk?
2. Is this limitation inherent from the atomic basis? Is it possible for me to extract internal variable of quantumatk and write my own code (for example using method in this paper https://iopscience.iop.org/article/10.1088/0953-8984/25/34/345501)to plot the effective bandstructure. I suppose I need the overlap matrix and Hamiltonian (which can be assessed through function calculateHamiltonianAndOverlap() )to plot the effective bandstructure. I also hope developers can shed some insights if there's some inherent difficulty in implementing such method, so that I can assess the feasibility of writing the codes. 
Thank you for your attention.
 
Update: the issue was resolved when I set the processes_per_kpoint to 2 increase the threads per process. So I think the error happens due to insufficient memory, contrary to my initial assumption.     

[Anders Blom]

Point 2 is probably not a way forward, you'd just end up reimplementing the feature we have.

What did you try so far? The whole point of the effective band structure is that the supercell has a lot more atoms than the original primitive cell, so that is not a fundamental limitation.

Do you have any reference publication for the plot you want to make? I am not sure this is a clearcut application of the effective band structure method. What information will you gain from the effective band structure which is not seen from just doing the corresponding calculation of the supercell (which you need to do anyway)?

[sukhito teh]

Point 2 is probably not a way forward, you'd just end up reimplementing the feature we have.

What did you try so far? The whole point of the effective band structure is that the supercell has a lot more atoms than the original primitive cell, so that is not a fundamental limitation.

Do you have any reference publication for the plot you want to make? I am not sure this is a clearcut application of the effective band structure method. What information will you gain from the effective band structure which is not seen from just doing the corresponding calculation of the supercell (which you need to do anyway)?

Thank you for your reply. In general, I plot effective bandstructure to see the change in curvature of bandstructure, direct/indirect bandgap.

The initial error message I receive when performing the calculation are as below:
===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   PID 274903 RUNNING AT node05
=   EXIT CODE: 9
=   CLEANING UP REMAINING PROCESSES
=   YOU CAN IGNORE THE BELOW CLEANUP MESSAGES
===================================================================================
YOUR APPLICATION TERMINATED WITH THE EXIT STRING: Terminated (signal 15)
This typically refers to a problem with your application.
Please see the FAQ page for debugging suggestions

The above issue was resolved by setting processes_per_kpoint to 2 increasing the threads per process. This leads me to the following questions:
1. Memory Issues? Does this observation suggest that the original error was due to insufficient memory allocation?
2. Optimizing Parallelization: What are the best practices for determining the optimal processes_per_kpoint value and the corresponding number of nodes to use for maximum computational efficiency? Are these parameters related to the k-points of the bulk configuration, the k-points used for the effective band structure calculation, or both?
3. Effective Band Structure Limitations: The effectivebandstructure module has a specific requirement: the 'configuration' must be divisible into an integer number of smaller unit cells, each identical in size and number of basis functions, as defined by the 'primitive_configuration'. This limitation is not present when plotting effective band structures derived from standard Plane-Wave DFT band structure calculations (e.g., VASP). Could you briefly explain the reason for this limitation within the effectivebandstructure module?"

[Anders Blom]

1. Yes, memory, and you found the right solution - more threads rather than more MPIs per node reduces memory usage but keep the speed roughly the same, in many cases.
2. There are two guides in the manual around performance/parallelization and memory usage:
https://docs.quantumatk.com/manual/technicalnotes/parallelization/parallelization.html
https://docs.quantumatk.com/manual/technicalnotes/advanced_performance/advanced_performance.html
It's hard to give a single advice that works for all systems, but QuantumATK speeds up very well to high count of cores, and the more memory you need, the more you should rely on threading rather than MPI.
3. To me this is a very natural part of the projection algorithm, not least considering the way QuantumATK uses atomic orbitals as basis. It also provides a logical way to interpret the results. If your larger system is not really an extension of something with a smaller periodicity, the effective band structure may not be the best tool to analyze the results. The band gap for any system can always be inferred from the density of states, and the curvature is the same as the effective mass basically, which can also be computed directly for the supercell system.

[sukhito teh]

1. Yes, memory, and you found the right solution - more threads rather than more MPIs per node reduces memory usage but keep the speed roughly the same, in many cases.
2. There are two guides in the manual around performance/parallelization and memory usage:
https://docs.quantumatk.com/manual/technicalnotes/parallelization/parallelization.html
https://docs.quantumatk.com/manual/technicalnotes/advanced_performance/advanced_performance.html
It's hard to give a single advice that works for all systems, but QuantumATK speeds up very well to high count of cores, and the more memory you need, the more you should rely on threading rather than MPI.
3. To me this is a very natural part of the projection algorithm, not least considering the way QuantumATK uses atomic orbitals as basis. It also provides a logical way to interpret the results. If your larger system is not really an extension of something with a smaller periodicity, the effective band structure may not be the best tool to analyze the results. The band gap for any system can always be inferred from the density of states, and the curvature is the same as the effective mass basically, which can also be computed directly for the supercell system.

Thank you for your reply.
1. You are right, the effective bandstructure that I had obtained is indeed very disordered due to poor symmetry.
2. Thanks, I would look into the effective mass module.