Adress:

Technische Universität Darmstadt
FB Mathematik, Fachgebiet Optimierung
Dolivostr. 15, 64293 Darmstadt

Office: S4|10 109
Phone: +49 (0)6151 16-70801
Fax: +49 (0)6151 16-3185

Mail: rroth(at)mathematik.tu-darmstadt.de

• Flow Control
• Optimal Control Theory
• Multilevel Optimization Methods
• Automatic Differentiation
• Proper Orthogonal Decomposition (POD)
• Parallel Methods and Efficient Implementation
• Feasible Sequential Quadratic Programming (SQP) Methods

Effiziente numerische Multilevel-Verfahren zur Optimierung von Gasturbinenbrennkammern

Collaborative Research Centre 568: Flow and Combustion in Future Gas Turbine Combustion Chambers (Strömung und Verbrennung in zukünftigen Gasturbinenbrennkammern)

Multilevel Optimization of Turbulent Flows by Discrete Adjoint Techniques, Dissertation, Technische Universität Darmstadt, 2012, Verlag Dr. Hut, PDF (8.7 MB)

• Parallel, Multigrid Discrete Adjoint Solver for Flow Solver FASTEST (Unsteady, LES, RANS, Temperature, Shape Optimization, ...) (Fortran)

• Parallel Krylov-Subspace Solvers along with Preconditioners (Fortran)

• General Purpose Implementation of the Recursive Multilevel Trust-Region Algorithm (MATLAB)

12.09.11 - "Multilevel optimization for flow control with discrete adjoints", 25th IFIP TC07 Conference, Berlin

17.05.11 - "Multilevel optimization for flow control with discrete adjoints", SIAM Conference on Optimization 2011, Darmstadt

06.04.10 - "Discrete adjoint techniques for flow optimization", 11th Copper Mountain Conference On Iterative Methods, Copper Mountain

29.07.10 - "Discrete adjoint techniques for flow optimization and POD supported shape optimization", The International Conference on Continuous Optimization 2010, Santiago de Chile

28.08.09 - Discrete adjoint techniques for flow optimization based on parallel Large Eddy Simulation, 20th International Symposium of Mathematical Programming, Chicago

Germano LES of Re 5000 flow (left) and corresponding discrete adjoint (right, drag as target), 1.1 mio grid points, whole calculation with 1000 time steps took 68h on 22 cores, gradient w.r.t. cylinder rotation vs. divided differences yields a rel. error of 2%

Control of Weakly Conductive Fluids by Near Wall Lorentz-Forces (Re=300), reduction of the wake

Shape Optimization of Windtunnel (LES, Smagorinksy)

Shape Optimization: Reduction of Drag (Above), Attain Lift (Below)