This document discusses exploring new 2D materials through computational modeling and simulations using high-performance computing (HPC). It describes the University of Bath's research agenda and need for flexible HPC resources to efficiently model experimental data from cutting-edge materials. The document outlines the HPC resources used, including a new cloud-based Janus environment, and how modeling scales up based on factors like system size and vacuum levels in 2D materials simulations. The goal is extracting maximum understanding from large experimental datasets through timely modeling and calculations.

1. Exploring new
high potential 2D
materials
Cloud flexibility for time-
optimised experiments

2. 2
Exploring new high-potential 2D materials
Daniel Wolverson
Centre for Nanoscience and Nanotechnology
University of Bath, UK
Stefano Angioni
Research Computing
University of Bath, UK

3. 3
University of Bath
Public research university with
a strong focus on technology
and science
One of the top ten universities
in the UK
20,000 students from 147
nationalities

4. 4
On-premise HPC challenges
Replacement every 6-7 years
Fixed and aging system
Long queues slowing down
research
Research needs evolve at pace

5. 5
Bath’s ambitious research agenda
Cutting-edge computational
facilities
Flexible and scalable resources
for changing needs
Grow and diverse user base
Increase capabilities and
productivity

6. 6
Turbocharging research with Cloud
Highly flexible and on-demand
computing
Computational environments
which can evolve
Modern and cutting-edge
Cost-optimised
Increased cost transparency

7. 7
Research Computing Portfolio
Cloud based
Teaching HPC environment
Research HPC – phase 1 Janus
Research HPC – phase 2 Nimbus
On-premise
Restricted Software HTC – Anatra
GW4 Tier 2 HPC – Isambard

8. 8
Cloud solution
CycleClou
d

9. 9
Research HPC environments
Phase 1
Janus
Phase 2
Nimbus
Phase 3
Phase 1 Lightweight Cloud environment for selected user groups
Phase 2 Functional and resilient platform mirroring Balena Cost
control and full software stack
Phase 3 Non-traditional HPC groups, optimising and tuning the
environment

10. 10
Janus environment
Our phase 1 HPC system in Azure.
Nodes are not physical, but virtual
machines. Nearly 800 compute nodes, but
we are only charged if they are used ☺
Cloud resources are orchestrated using
Azure CycleCloud.
Users can access Janus using SSH.
The Job scheduling software is Slurm.
Only one login / scheduler node 

11. 11
Nimbus environment
Development guided from learnings from
Janus
Improved resilience
New lifecycle model for max availability
New cluster monitoring with TIG stack
New cost management tools and cost
controls
New storage options

12. 12
Rods
All nanoscience is
contained in a bag
of liquorice allsorts!
Nanoscience and 2D materials A boy and his atom
Little men
https://www.youtube.com/watch?v=oSCX78-8-q0
Tubes
https://en.wikipedia.org/wiki/Car
bon_nanotube#/media/File:Multi
-walled_Carbon_Nanotube.png
Quantum wells;
Layered 2D materials
https://www.nature.com/articles/nature12385
Core-shell spheres
https://www.photonics.com/Articles/Quantum_Dots
_Small_Structures_Poised_to_Break/a22350

13. 13
https://en.wikipedia.org/wiki/Phosphorene#/medi
a/File:Orthorhombic_bulk_black_phosphorus.png
https://en.wikipedia.org/wiki/Graphene#/media/File:Grap
hen.jpg
Ruitao Lv et al., Nano Today (2015) 10 559-592
Graphene is unique, but:
40 other natural layered MX2 compounds;
Thousands more we could make if we wish.
Isn’t nanoscience all about graphene?

14. 14
• Digital economy
• Energy and decarbonisation
• Engineering
• Healthcare technologies
• Information and communication technologies (ICT)
• Manufacturing the future
• Mathematical sciences
• Physical sciences
• Quantum technologies
• Research infrastructure.
low-power electronics and spintronics
energy harvesting through motion
energy harvesting through photocatalysis
energy harvesting as photovoltaics
healthcare; wearable diagnostics
low-power electronics and spintronics
flexible, wearable electronics
quantum devices
single photon light emitters for quantum
2D materials promising for:
But we are at TRL1 to TRL4 in most cases
(proof of concept, lab validation)
Why are 2D materials important?
UK funding priorities in engineering and physical science

15. 15
How are these materials studied?
Elettra synchrotron, Italy
Energy (eV)
X-ray photoemission beamline:
Diamond synchrotron,
UK
https://www.diamond.ac.uk/Home/About.html, https://www.synchrotron-soleil.fr/en

16. 16
Electronic structure?
t
r
c
Energy (eV)
Energy (eV)
-28 -26 -24 -22
0
20
40
60
80
100
Intensity
after
baseline
subtraction
/
1000
Binding energy / eV
tip
ring
Synchrotron X-ray photoemission spectra:

17. 17
o We generate large and expensive
datasets on highly topical
materials.
o These deserve modelling to the
best of our ability to extract
maximum information and
understanding.
o We need to work quickly.
o It’s appropriate to do some of this
within an experimental group;
postgrads can easily be involved in
both theory and experiment.
o We can even involve
undergraduates at an appropriate
level; HPC is not necessary for
training them but is needed for real
problems.
ReSe2
What’s the role of HPC computing?

18. 18
We have also been
able to contribute
some new ideas
about the
methodology that
we are using.
New tricks

19. 19
What physical things do we calculate?
1. Approximation to the ground state
electronic wavefunction and energy for
some fixed configuration of some atoms;
2. From this, can obtain the forces between
the atoms, can move them and obtain an
equilibrium model of the crystal;
3. From the forces near equilibrium, can
obtain the vibrational frequencies of the
crystal (which we also measure);
1.
3.

20. 20
4. From the equilibrium structure, can calculate
the energy bands (is the material a metal,
semiconductor, etc?);
5. We can ask how the vibrations affect the
energy bands (a hot topic experimentally);
6. Can do these also for structures with defects
(e.g, missing atoms), multi-layered crystals,
mixtures, etc…
7. We are not limited to real crystals and, often,
we will know how to make any interesting
crystals we “discover”.
arXiv preprint
arXiv:2206.00126
What physical things do we calculate?

21. 21
What tools do we need?
Density functional theory (DFT) codes (Walter Kohn, Nobel Prize in Chemistry 1998)
• CASTEP, VASP, QUANTUM ESPRESSO and many others.
• Plus post-processing, extensions and visualisation codes (Wannier90, yambo)
• We mainly use QE: GNU licence, open source, and we do modify source code.
• Libraries typically: BLAS, LAPACK, FFTW, OpenMPI, MPICH, HDF5.
• Qauntum Espresso code is very easy to install on desktops / small servers for
training, so is well-suited for the early stages of undergraduate and Masters
projects. Cloud resources allow one to escalate the calculations to a “production”
level fast and flexibly.

22. 22
How do the resource requirements scale?
1. Approximation to the ground state
electronic wavefunction and energy for
some fixed configuration of some atoms;
Simple answer: scaling goes as 𝑁3where 𝑁 is
related to the system size – this includes:
• Number of atoms
• Number of bands (number of electrons)
• Number of plane waves (the “basis set”)
The basic computational task is:

23. 23
How do the resource requirements scale?
For large volumes of vacuum either side of a 2D slab (below) the
basis set becomes very large, independent of the other factors.
This is too complex a topic to give more
detail here:
the main point is that publication-quality
results can require a significant escalation
of the resources required.

24. 24
What’s the role of HPC computing?
We generate large and
expensive experimental
datasets on highly topical
materials.
These deserve modelling to
the best of our ability to
extract maximum information
and understanding.
We need to work quickly.
ReSe2