The Pawsey Supercomputing Centre


Simulating the Epoch of Reionisation

A team of researchers led by Dr Alan Duffy from Swinburne University of Technology’s Centre for Astrophysics and Supercomputing is using the petascale power of the Pawsey Supercomputing Centre to uncover the secrets behind the birth of the Universe’s first galaxies. The project is simulating the little-understood formation of these galaxies at an unrivalled level of detail, which is not only groundbreaking research in its own right, but will also help unlock the potential of next-generation radio telescopes.


Tenuous filaments of gas connect a central galaxy more widely to the large structures around it in the early universe. Image: Alan Duffy, Stuart Wythe and the DRAGONS team

Tenuous filaments of gas connect a central galaxy more widely to the large structures around it in the early universe. Image: Alan Duffy, Stuart Wyithe and the DRAGONS team

“We’re now trying to understand the earliest moments of our universe’s history,” says Dr Duffy.

“It’s challenging because we don’t understand the physics of the early universe particularly well; we don’t really know exactly how to ‘build’ a galaxy in detail.”

The processes behind the birth of these galaxies can be studied both through simulation using supercomputing, as in this project, or directly observed using telescopes such as the Australian Square Kilometre Array (SKA) precursor projects.

“However, we are limited in those observations, quite drastically at this stage. So, we have to infer from these tiny bits of information we have from these vast distances, essentially what the galaxies are doing,” says Dr Duffy.

As these telescopes produce data sets in the form of raw statistical outputs, simulations like Dr Duffy’s are essential to allow researchers to properly interpret the results of their projects – acting as a “numerical laboratory” to make “…solid predictions for these first galaxies, and then to try to guide the interpretation of the observations as well.”

However, this process is practically impossible to perform using traditional computing methods, according to Dr Duffy.


“You have to run the same simulation over and over again, tweaking the physics each time – it’s extremely time-consuming, computationally speaking, to run one of these simulations even once. To do it over and over again, systematically varying your experiment, is only possible thanks to the extraordinary facility the Petascale Pioneers Program offered.”


Using highly optimised code running on the petascale ‘Magnus’ supercomputer, Dr Duffy and his team were able to simulate the formation of these galaxies at a scale previously unseen in Australia.

“The code we were using had been systematically improved over the years working on Pawsey’s previous machine Epic in particular, and then having the confidence to scale this up to the billion particle scale we are ultimately able to run on Magnus.”

Dr Duffy and his team are also able to take advantage of Pawsey’s world-class data storage resources and expertise to help archive and access the vast amounts of information produced by a project like this.

“It’s one thing to crunch the numbers but an entirely different thing to actually store it and to analyse it,” says Dr Duffy.

“What is also crucial is knowing the Petabyte storage facility had the capacity and the fast disk to take the data.”


Dr Duffy and his team are able to simulate the formation of some of the earliest galaxies in the universe at an “unsurpassed” resolution – over a billion particles.

“The results from this project are going to be used in interpreting the results from the Murchison Widefield Array telescope facilities in Western Australia. With the telescopes, you basically get a statistical description of the night sky, and to actually turn that into ‘where’s the galaxies, what are they doing?’ that has to come through these simulations.”

The vast amounts of data produced by this project are also spurring the development of next-generation software and infrastructure to transfer and analyse this data and allow practical sharing of the results of this project and others like it.

“This data set ended up being a test case for that,” says Dr Duffy.

“That protocol has been widely used since – it’s stupendously fast, orders of magnitude faster than anything else that was available. All in all it’s been a challenge, but that challenge has inspired some amazing development.”

Comments are closed.