Adaptive Precision Optimization via Output Error Estimates

1
Adaptive Precision Optimization via Output Error
Estimates

• James Lu Advisor David Darmofal
• Motivation flow optimization traditionally
  performed on grids of fixed refinement and
  residual convergence.
• Objective significantly reduce compute time for
  flow optimization with use of adaptively
  controlled grid refinement and iterative solution
  while ensuring convergence to true optimum.
• Approach use of a posteriori discretization/itera
  tion output error estimates to drive adaptive
  process

2
Efficient concurrent flow-adjoint solver and
iterative output error estimate

• Concurrent solver capability line-implicit,
  p-multigrid.
• Residual, output convergence and CPU cost for
  concurrent flow-adjoint line-implicit solution

3
Adjoint consistency/proof-of-concept

• Need for discrete/PDE adjoint consistency

• Quasi-1D flow pressure matching test case
• Comparison of adaptive/fixed precision approaches