Imperial College London

MrLarryKostorz

Faculty of EngineeringDepartment of Earth Science & Engineering

Research Postgraduate
 
 
 
//

Contact

 

w.kostorz16

 
 
//

Location

 

ACE ExtensionSouth Kensington Campus

//

Summary

 

Summary

Background:
Polish, a mathmo by training. BA (2012-15, I) and MMath (2015-16, D) from Peterhouse, CB. Both degrees focused strongly on applied mathematics, all pen & paper with the exception of separate programming projects

Main study areas:
mathematical fluid dynamics, linear and nonlinear wave theory, numerical analysis, linear algebra (the "practical" aspect in all applications from QM to stability theory), asymptotic/perturbation methods

Imperial:
Joined the 4-yr Fluids CDT programme in 2016; Obtained MRes (D) in 2016-17 (1/4 years) and started the PhD straight away. The project is industry-funded by TOTAL S.A. (1 supervisor from the company included)

PhD:
Typically, the trade-off between simulation time and accuracy is dictated by mesh resolution; indeed, lower-resolution models run faster but are less accurate due to discretization errors. This project investigates another trade-off, namely simulation time vs "simulator's accuracy" while keeping the resolutions high. Recently proposed ideas enable one to construct those less accurate but super fast "simulators" automatically, by running a fixed piece of code on any suitable set of data. Practically, a single "builder program" can automatically generate super fast approximate models for a huge range of dynamical systems. The essence of the project is extending some and implementing all of those ideas so that the method can be applied to petroleum engineering problems

Current working areas:
mathematical modelling, multidimensional interpolation methods, surrogate-based and/or derivative-free optimisation, linear algebra, data processing

Additional work:
- Top-down redesign and reimplementation of a high-order IMPES porous media flow simulator "MISTRESS" (Miscible-Immiscible Simulation Tool for RESearch Studies) together with a dedicated MATLAB interface and data preprocessor

- Efficient implementations of derivative-free optimization routines DIRECT (MATLAB, Cpp), MCS (Cpp), ACiD (MATLAB) and EGO (MATLAB)

- Design and implementation of a non-intrusive modelling MATLAB toolbox based on Radial Basis Function and Polyharmonic Spline interpolation methods

- Small amount of experience with GPU programming (Nvidia CUDA), Python and Fortran77/90