Reduced Order Methods in HPC

Primary tabs

Model order reduction aims at attaining the result of many simulations with a single, larger, high-dimensional simulation whose results can be readily stored for future use. The benefit is double: enabling real-time access to an arbitrary scenario and speeding up multiparametric simulations.


Actual computer science understands the geophysical simulations as complex HPC algorithmics trying to find efficient solutions to very challenging parameter-dependent problems. These parameters are either estimated physical properties (e.g. direct modelling), or large observation datasets that must be reproduced from the numerical simulations (e.g. inverse modelling). In both cases, thousands of offline solutions to 3D complex geophysical problems are constantly required. This research line works towards constructing a generalized (high-dimensional) solution capable of providing fast online simulations for all the parameters at once. Thus, it enables the possibility of drastically accelerate the modelling processes associated with geophysical applications. For this purpose, reduced order model (ROM) methods are worked out to find an optimal HPC-driven approximation of the targeted generalized solution.


  • Develop ROM techiques for direct modelling applications in geophysics including seismic and electromagnetic fields.
  • Analize the potential use of different parameterizations of the proposed high-dimensional models and the capabilities of actual ROM methods to provide feasible solutions.
  • Analize the HPC potential of the current ROM algorithms and explore novel strategies to maximize their efficiency with our HPC technology.
  • Generate efficient HPC-driven ROM algorithms to accelerate the most challenging inverse modelling problems in geophysics.
    • DAVID MODESTO's picture
    • Contact
    • Postdoc Junior
    • Tel: +34 934137992