Published on June 4th, 2013 | by Stephanie
Talking exploding supernovas and MPE Australia with Dr David Bailey
Dr David Bailey will be talking about how to conquer numerical errors in his address on the opening day of MPE: The Conference. We had a hangout with him to get a little taste of what he will be discussing.
A problem plaguing scientists that use large data sets and even larger computers is the issue of reliability and reproducibility. ‘We don’t just have to worry about the how close, theoretically, the results fit to reality. But we have to ensure that that the numerical computations being done are both reliable numerically and reproducible,’ Dr Bailey said.
Reproducible for him means that there need to be standards in place that allow anyone to pick up a peer reviewed article and code the problem themselves to generate the same answers.
‘The entire scientific community must strive for this reproducibility,’ he said.
He went on to tell us that there are two main sources for errors in simulations. The first, he said, ‘keeps scientists up at night’. When we wish to simulate data simplifications and assumptions are necessary, we do not have the computing power at present to not simplify some of the mathematical equations that are needed for modelling. However, Dr Bailey pointed out that: ‘The more simplifications that we make, the less faithful to reality our models become.’
The second source is related to numerical errors that accumulate over the simulation itself. These errors are mainly due to rounding off. With the development of computers that can process more and more data this errors can be significantly reduced.
Something that I hadn’t realised was that the best models we have for making climate predictions are only able to take data every 25 km or so. Which leaves a great deal of terrain unmapped. If we could have data for every kilometre our predictions would be much closer to reality – this is what scientists refer to as resolution.
‘We need to have higher resolution in our models. This means having millions or billions of points. This is the job for large scale, very powerful, parallel computing,’ Dr Bailey said.
‘We need these computers to simulate data and to process experimental data,’ Dr Bailey continued.
He is optimistic that this sort of computing isn’t far off. ‘Up until only a few years ago our models couldn’t even simulate a supernova explosions but now we have programs that make them explode all by themselves,’ he said quite excitedly.
Dr Bailey is looking forward to MPE: The conference as it gives him the ‘opportunity to bump heads with a new crowd. And hopefully come home with some new ideas and perspectives’.[subscribe2]