PhD project: Developing the direct numerical simulation of turbulence for GPU based computing with application to fundamental problems
Project description
Direct numerical simulation (DNS) is one of the most important tools
today for studying fluid turbulence. This project has a strong
computational focus with a combination of
programming and computational work alongside application to
fluid turbulence. There is an opportunity to
study fundamental questions about turbulence using DNS and analytic methods.
The limiting factor for large DNS simulations in the
future would be the speed and management of the memory.
This project starts with an existing and well tested CPU based DNS code
The goal is to convert it to a GPU based code as well as to extend
functionality. GPUs are an increasingly important resource in scientific
computing and a portable code with effective performance is essential.
Other possible areas of interest related to GPUs are: the use of
low-precision floating point arithmetic (often used in machine learning)
for DNS - the fidelity of the results would have to be assessed; the
model used to program multiple GPUs (with both fast and less fast memory
transfers available) is also of interest.
The end result will be an open source versatile DNS code.
Alongside development of the DNS code, the project will involve
running simulations to study fundamental questions about
turbulence. This will include examining chaotic and spectral properties
of turbulence. In additional, a unique functionality of the DNS
code is studying turbulence in higher spatial dimensions beyond
three, which has fundamental interest motivated by quantum
field theory, to search for simplified behaviour that
could then reveal new insights relevant for turbulence in
three spatial dimensions.
The student needs to be a competent programmer in C or C++, and have a
strong interest in parallel programming and HPC architecture.
A desirable skill would be experience with numerical simulation in C++
but it is not a necessary requirement. The student should have a background in physics and/or mathematics.
This will be a project running in partnership between the School of Physics and Astronomy and Edinburgh Parallel Computing Centre (EPCC)
Project supervisor
- Professor Arjun Berera (School of Physics & Astronomy, University of Edinburgh)
The project supervisor welcomes informal enquiries about this project.
Find out more about this research area
The links below summarise our research in the area(s) relevant to this project:
What next?
- Find out how to apply for our PhD degrees.
- Find out about fees and funding and studentship opportunities.
- View and complete the application form (on the main University website).
- Find out how to contact us for more information.