TIME-X

• Traditional way of achieving parallelism
  • Spatial decomposition of data
  • Parallel execution within each timesteps
  • Successive timesteps executed sequentially
  • This limits scalability

• New algorithmic approach: parallel-in-time integration
  • (Additionally) parallelise across time dimension
  • Requires research and development of new algorithms
  • Convergence and validation of results are concerns

Objectives

• Advance parallel-in-time integration from academic research topic into usable algorithms
• Demonstrate potential for delivering Exascale performance for a range of scientific and industrial applications”

Technical approach

• HPC software development considering
  • Load balancing
  • Adaptivity
  • Inexactness and robustness (communication)
• Algorithm development including
  • Optimisation and optimal control
  • Uncertainty quantification and data assimilation
  • Multiscale propagators
  • Molecular dynamics simulations
• Application showcases in
  • Medicine, molecular dynamics, electromagnetics, climate/weather