Nothing but 100 percent positive experiences from start to finish

The National eScience Infrastructure project provides early-stage researchers with world-class high performance computing facilities, as a PhD student discovered when she was able to complete more than five times as many calculations as others researching

Melting gallium nanoclusters to model their intriguing properties is the work of a Victoria University of Wellington science researcher. Nanoclusters are tiny clusters of particles that behave as a single unit. Oddly, nanoclusters of gallium comprising only tens of atoms melt at a much higher temperature their larger counterparts.

According to the principles of science, small quantities of matter are expected to melt at lower temperatures than larger quantities, due to an increase in the ratio of surface to interior atoms. Surface atoms are less strongly bound than interior atoms, so should become unbound and melt more easily. However, experiments show that the opposite appears to be true for gallium nanoparticles.

Krista Grace Steenbergen works at Victoria University of Wellington with her PhD supervisor, senior lecturer Dr Nicola Gaston, and also with Shaun Hendy, Professor of Physics, both at the School of Chemical and Physical Sciences. Steenbergen recently completed her PhD. Although there’s a lot of interest in the application of nanoclusters, only a handful of researchers globally are working on the specific properties of melting clusters of gallium.

“In everyday life we don’t really think about how things melt,” says Steenbergen. “If a small and a large piece of ice fell on a hot surface you would expect the smaller piece to melt more quickly, but most people probably wouldn’t even notice.”

Research into the thermodynamic properties of gallium nanoclusters, meanwhile, has led to the discovery that differences in cluster size of just one atom make a disproportionate difference to the melting temperatures. “If you take a 36-atom cluster, it melts at something like 250 degrees Celsius. If you add one atom there’s a 100-degree increase in the melting temperature.”

Applications for a potent element

A key limitation of working with nanoparticles is their typically low melting temperature, says Steenbergen. “In developing a better understanding of why small gallium clusters melt at such high temperatures we gain a better understanding of how these small systems behave, which could lead to new applications for nanoclusters in the future.”

The thermodynamic properties of gallium nanoclusters are of considerable interest to the scientific community, as gallium is such a potent element in society: it’s used in transistors, solar cells and LEDs, and 98 percent of its applications are in semiconductors. This suggests there might be future commercial applications to harness the properties of gallium nanoclusters. “The fact it has such unusual properties makes it of great interest because it’s part of our daily lives, even though it’s usually in compound form,” says Steenbergen. “Scientists are interested in it because it’s such a strange phenomenon. If we could figure out what’s going on, it might be harnessed for something we can’t even dream of at this stage.”

Computational performance boost

After discovering Dr Shaun Hendy’s computational nanotechnology site, US-born Steenbergen contacted him to ask what projects he had underway and was subsequently persuaded to do her research on this Marsden funded project by Dr Nicola Gaston, who would later become her PhD supervisor.

Before using NeSI’s HPC facilities Steenbergen’s research was restricted to a local computing cluster with a small number of nodes, limited disk space and long queue times. “When I look back on it, it was painful. The way the computer was configured made it very slow reading and writing to disk. At the time I started there were five researchers trying to use it and each of the researchers needed a good number of its nodes.”

Because of the need to timeshare with other researchers, Steenbergen’s models couldn’t run continuously and she had to wait her turn in the queue. “I would submit a cycle and it would run but I’d be stuck in line behind other people for the cycle to restart. So it was continuously trying to run over a six-month period.”

Although the files Steenbergen’s research generates are not particularly large, it produces many of them and she needs to access them concurrently. “I was having to zip some so that others could take up the space and then unzip the files to analyse them.”

Steenbergen was later granted access to BlueFern, the University of Canterbury’s supercomputing unit. BlueFern operates a number of supercomputers, accessible to researchers across the country via the REANNZ network and NeSI's access process.

For BlueFerns assistance in setting her up on the HPC facilities, Steenbergen reserves special praise for Tony Dale, supercomputing services and systems consultant, and Vladimir Mencl, e-research services and systems consultant, both at BlueFern; and Dan Sun, the BlueFern service delivery manager at the University of Canterbury. As a result of their help, if she were to start her project again with the knowledge she now has, she’d trust more in the access to support. “I’d been used to an impersonal approach from system administrators and such. It took me a while to get used to how great those guys are. They truly wanted to help with everything I needed: implementation of my code and how best to do it on HPC.”

Steenbergen used the BlueFern HPC facilities to run the Vienna Ab Initio Software Package (VASP) to carry out Density Functional Theory-based thermodynamic calculations on the small gallium clusters. “I run a wrapper code around it with a double layer of parallelisation. The way I implemented my algorithm was hugely affected by the help and the expertise of Tony and Vladimir and the crew at Canterbury.”

Her research wouldn’t be possible without high-performance computing, she says. “Accurate modelling at this scale required the extensive use of highly parallel quantum chemistry and physics codes for performing first-principles, density-functional molecular dynamics over hundreds of thousands of time steps.” As an example, one of her models required 19 parallel runs of 128 processors over 500 hours on Canterbury’s BlueGene/L system, totalling more than 1,000,000 CPU-hours for a single size of gallium cluster.

The performance gains achieved as a result of the processing power of NeSI were dramatic, says Steenbergen. “Because of access to the NeSI resources we were able to run our modelling on an unprecedented 11 gallium clusters. Due to the computational limitations, other researchers have only managed a maximum of two clusters. Being able to model the size ranges the way we have, we can really investigate the nature of how one atom changes things so much.”

Peer-reviewed success

When she initially signed up Steenbergen didn’t know what HPC resources would be available to her. “I knew that in order to do the work we were looking at we were going to have to have some pretty serious computational power, but didn’t realise how incredible it was until I got down here. It’s just unbelievable that the scientific community at large here in New Zealand has access to a machine like that.”

Training on the new system was not a significant barrier; she points out that the University of Canterbury provides free HPC courses to students in a postgraduate programme or their final year of an undergraduate degree, as well as HPC workshops and seminars. “They offer classes for people who may not have as much computational experience as I do on how to code and how to parallelise code,” says Steenbergen, “and that’s invaluable.”

Publications are an important measure of success for Steenbergen’s work. She and her fellow researchers have so far published one positively reviewed paper, with another paper in the review process and one more in draft stages. She recommends NeSI to others doing their PhDs or research masters.

“I’ve had nothing but 100 percent positive experiences from start to finish,” says Steenbergen. “I can’t say enough about how great these resources are.”

Further Material

If you would like to watch the recording of a seminar that Krista delivered on her work, visit and click on the green icon.

Next Case Study

High performance computing’s role in biosecurity

NeSI's high performance computing facilities are helping a Lincoln University researcher combat biosecurity breaches by speeding up genomic analysis and the ability to identify invasive pests.