Peering into the future - with simulations

It’s a complicated world, that much is certain. One great aim of science is not only to understand it but also to get behind the interrelationships - in fact, to make them predictable. The challenge is: the incalculable. In this regard, a third avenue to understanding has opened up in the last few decades as a support for experimentation and theory: simulation.

The Stuttgart Research Centre for simulation Technology (SRC) networks next-generation scientists from quite unrelated disciplines at the University of Stuttgart and the SimTech Excellence Cluster. The goal: to support gifted young scientists with a broad-based offering of courses and seminars, intensive mentoring, research and study abroad, and interdisciplinary stimulation of ideas. 

Just to give an example: SimTech provides interesting perspectives that include a graduate school with about 150 doctoral candidates from nine of the University of Stuttgart’s ten faculties, an elite course of studies leading to Bachelor’s and Master’s degrees, postdoc positions, and junior professorships. The simulation models developed here with state-ofthe- art technology and the know-how of far more than 100 researchers aim to develop answers to complex issues and scientific questions.

Simulations have one purpose above all: to reduce the number of incalculables. Not only do they reproduce processes, they also predict them. This scientific technique for peering into the future has established itself today in all areas of life: from financial crises to environmental catastrophes that we can see coming with the help of simulations before they occur. This sometimes involves supercomputers whose immense calculatory abilities can take the user virtually through global scenarios from the future. Many simulation tasks require such great calculation capacities that they challenge even the underlying software and models. Computer pioneer Seymour Cray summed the issue up this way: ‘If you’re going to plow a field, which would you prefer to use: two strong oxen, or 1024 chickens?’. We naturally tend to think that it’s easier to use one big, fast processor than many slow ones.

The fact is, however, that computers have sped up in the last 10 years or so not because individual processors have become faster, but because there are more of them: more than 180,000, for example, in ‘Hazel Hen’, Germany’s speediest computer in Stuttgart-Vaihingen. That makes it a challenge to use such computers. Just for comparison: if automobiles since 1970 had become faster at the same rate as computers, they would be traveling today at several times the speed of light. The result: today’s scientists in this field travel between virtual and real worlds to gain new, sometimes epochal insights into the forces that move the world and will continue to do so.

This in turn requires that more precise geological information be procured, which in turn will require a few highly expensive drilling operations in order to procure samples. ‘Unfortunately, we do not have an X-ray machine that can make the entire planet transparent. That would, of course, make our simulation work much easier,’ is how Dirk Pflüger illustrates the challenge in dealing with incalculables when creating simulation models. ‘What we’re looking for is to turn ‘What don’t I know?’ into ‘At least I know what I don’t know,’ or better: ‘At least I know how much I don’t know’, adds Sergey Oladyshkin.

Dirk Pflüger is a specialist in informatics, Sergey Oladyshkin in applied mathematics; but even though their disciplines are different, their research has some interesting interfaces: Pflüger works out the research methods for working with a simulation created by Oladyshkin. The parameters to be considered in CO₂ sequestration are manifold: where can CO₂ be pumped into the ground, for example, in what quantities and under what pressure? Will CO₂ leak uncontrollably out of drill holes or due to weak points in the rock layers? Conventional scientific methods quickly reach their limits when the attempt is made to take all these dimensions into account in a single model. The description used by scientists in this context is, ‘the curse of dimensionality’. But now computer simulations offer new ways of inputting existing data as fruitfully as possible.

To this end, Dirk Pflüger and his group are working out methods for the approximation of high-dimensional functions on the basis of ‘sparse grids’. He compares this principle to that of sinking warships: ‘The first thing is to sink the battleship with as little loss of time and effort as possible; then come the lesser ships, and then the smallest one.’ The same holds true for simulation: the first thing is to present the key interrelationships as quickly as possible and with as little time and effort as possible.

The aim is to make the results more reliable on the basis of data-driven approaches and to make factual incalculables visible. Why is it important to quantify incalculables? That makes it possible to develop methods for reducing the complexity of computer-intensive simulation concerning CO₂ sequestration, thus making it computable within an acceptable period of time. The development of better methods of computer simulation puts the researchers on the one hand in the field of basic research. But at the same time, their work remains application-oriented inasmuch as they are dealing with real data. The results are of interesting not only to them but also for international research institutions: For example, Dirk Pflüger is also collaborating with the Sandia National Laboratories in the U.S.A. to pursue and publish the results. ‘We are working above all on the question of how to master high-dimensional relations of dependency, not only with regard to the mathematics on which they are based but also regarding efficient implementation of all this and how to turn it into useful tools and software. What’s nice about this project is that we constantly move back and forth between basic research and its application,’ says Dirk Pflüger.

‘Our vision is for ongoing measurements to take place within the auto itself and for calculations to be carried out at every moment in order to ensure the individual safety of the driver at moments of danger,’ is how Jörg Fehr, Junior Professor for Multibody Dynamics at Sim-Tech, outlines the overall aim. After becoming a simulation engineer for multiphysical crash simulations at TRW (now ZF TRW), he worked with a team on the development of new safety systems for automobiles, was one of the first graduates of the Simulation Technology Cluster of Excellence, and leaped at the chance in his third year of working in industry to come back to the academic world. Now, as a next-generation researcher, his current work lets him combine research questions from industry and science in order to improve the safety of vehicles. There is a manifold range of new approaches, like considerations involving the driver’s muscle activity, or longer simulation horizons, or the uncertainty of parameters which must be studied in order to better simulate human behavior during accidents.

The question is: How can we reduce the order of the equation system while still retaining the system’s essential information? After all: the moment required for the accident itself to occur ranges between only 80 and 120 milliseconds, but the process as a whole has already begun some seconds before that, namely at the time when the active safety system is triggered. A simulation, however, must take the entire phase into account, with the result that the computer’s calculations need more time. ‘The speed is breathtaking; ten million equations must be solved 500,000 times throughout a single night in order to completely calculate a simulation of 120 milliseconds’ duration. But a look at the entire elapsed time required by active vehicle safety systems shows that it quickly takes ten times as long,’ explains Fehr. The solution is to make calculations faster by reducing the model to the most important factors, meaning that orders and dimensions must be made as small as possible, but as big. as necessary. The results make it possible to arrive at precise statements about whether the safety system in question is good or bad - and for whom. These results are directly related to the development of driving safety systems and help to make them even safer. ‘It’s very important to us not to do our research in a vacuum,’ says Fehr determinedly; ‘we want to create added value and work out solutions that have concrete applications.’