Browsing by Author "Enriquez, Marco"

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    A C++ Class Supporting Adjoint-State Methods
    (2009-09) Enriquez, Marco
    The adjoint-state method is widely used for computing gradients in simulation- driven optimization problems. The adjoint-state evolution equation requires access to the entire history of the system states. There are instances, however, where the required state for the adjoint-state evolution is not readily accessible; consider large- scale problems, for example, where the entire simulation history is not saved to con- serve memory. This thesis introduces a C++ state-access class, StateHistory, to support a myriad of solutions to this problem. Derived StateHistory classes im- plement a (simulation) time-altering function and data-access functions, which can be used in tandem to access the entire state history. This thesis also presents a derived StateHistory class, GriewankStateHistory, which uses Griewank's opti- mal checkpointing scheme. While only storing a small fraction of simulation states, GriewankStateHistory objects can reconstitute unsaved states for a small computa- tional cost. These ideas were implemented in the context of TSOpt, a time-stepping library for simulation-driven optimization algorithms.
  • Thumbnail Image
    Item
    The Effects of Coupling Adaptive Time-Stepping and Adjoint-State Methods for Optimal Control Problems
    (2011) Enriquez, Marco; Symes, William W.
    This thesis presents the implications of using adaptive time-stepping schemes with the adjoint-state method, a widely used algorithm for computing derivatives in optimal-control problems. Though we gain control over the accuracy of the timestepping scheme, the forward and adjoint time grids become mismatched. Despite this fact, I claim using adaptive time-stepping for optimal control problems is advantageous for two reasons. First, taking variable time-steps potentially reduces the computational cost and improves accuracy of the forward and adjoint equations' numerical solution. Second, by appropriately adjusting the tolerances of the timestepping scheme, convergence of the optimal control problem can be theoretically guaranteed via inexact Newton theory. I present proofs and computational results to support this claim.
  • Thumbnail Image
    Item
    Time-Stepping Classes for Optimization (TSOpt)
    (2009-09) Enriquez, Marco; Symes, William W.
    This report introduces the "Time Stepping Package for Optimization", or TSOpt, which is an interface for time-stepping simulation written in C++. It packages a simulator together with its derivatives (\sensitivities") and adjoint derivatives with respect to simulation parameters in a single object called a Jet, which can be used in conjunction with an optimization algorithm to solve a simulation-driven optimization problem. Further, TSOpt interfaces with the Rice Vector Library (RVL), allowing Jet objects to define a Operator subclass.