Welcome to my talk. Good afternoon, ladies and gentlemen. First, we have to move this.

Yeah, I'm presenting the work of my research group. My research group name is legacy research

group. I started this research group after my PhD at the IP. It belongs to two institutes,

which is very unusual. We are really an interdisciplinary research group. So we take two

views. The one view is the engineering view where people have problems. They don't care about the

methods as long as their problems are solved. And since I'm a mathematician, I also like to take the

method view. And my working group is mostly oriented on the method view. We have one method

and we look for problems which we can solve with them. And you will see in my talk, the

field of applications is very, very vast. My group consists of 10 PhD students. So all the work I'm

presenting here is done by my PhD students. A little bit of myself, but this is really old stuff

from my PhD time. These are the current PhD students, but also PhD students who finished.

So the talk is about facing challenges in computational dynamics. And we use three things,

one nice method called lattice Boltzmann method. We use OpenRB, this is open source

software. And we use high performance computers to solve these problems.

So the class of problems we are considering are threefold. The first class of problems is turbulence.

And here it is important to capture small scales because they really change the integral results.

The second one are suspensions. Small effects can have big influences. So for example, the shape of

a particle. And the third class of problems are optimal control and optimization problems where

we have to solve a complex problem again and again and again to converge to the optimal solution.

All three classes of problems were mentioned in the NASA report, challenges in CFD 2030,

published a few years ago. So it's not that I thought about these problems,

where you will find them in this paper and many other people are thinking about them.

These three problems have in common that they are, the models are known,

but they are computationally expensive to solve them.

And yeah, compute powers, where you probably all know Moore's law, the number of transistors

for microprocessor data about every three years. You also see this shift here towards

multi-core hardware in 2003. And here I've plotted the number of publications with

several keywords in the title in the F. You can see for example, I should do it here for the

ones which are online. You see that it's going in line well until this multi-core shift until 2003.

So it's a logarithmic scale that means the number of publications concerning numerical simulations,

at least in the title, also doubles every two, three years. But only until this shift.

And you'll see here that some methods like machine learning can still take advantage of it and others

cannot or can not as fast as Moore's method. So it depends on the method. So the message here is

that compute power is there and it's increasing. So we hope to solve these challenges there.

But you need a nice message. Lettuce Boltzmann is there now for 40 years. From the very beginning on,

people used that in parallel computing. There were many papers published on this topic, especially

here in Erlangen. And this method is there which can take advantage and is almost perfect

scalable. So the models are there. But it's not enough to have a nice model. You need a

software tool to solve complex problems. And we started with the Open AB project in 2006.

And the idea from the very beginning was difficult, go beyond one PhD cycle, not starting,

writing new code, doing well documentation, writing it extensible, that you can use different

algorithms of the same method, using modern C++ templates, having in mind always doing

software development, which is very, very important, the method view and the application view.

So approaching these challenges from two sides. Thinking as a mathematician and as a computer

scientist of methods and algorithms and also thinking from the application side to solve these

problems. So here we face these challenges. There's a pure compute power. We are using DNS,

which means direct numerical simulation or large eddy simulation instead of Reynolds,

Ivergen, and Restokes models. So as much as we can, we resolve the turbulent structures

of the particles. As much as we can, we resolve the properties of these particles, the shape,

the forces between acting between these particles. And for the optimization algorithms,