Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
72 H.G. Bock <strong>and</strong> V. Schulz<br />
Itisshownin[12]thatthereforethecoarse grid correction is indeed a secant<br />
direction for the optimization problems on the higher refinement levels.<br />
A typical area <strong>of</strong> application for this multigrid strategy is the inverse problems<br />
for distributed parameters <strong>and</strong> control or shape optimization problems,<br />
where the optimization variable is <strong>of</strong> function type.<br />
3.3 Unsteady Problems<br />
In principle, unsteady optimization problems <strong>of</strong> the form (3.1, 3.2, 3.3) can<br />
be treated by the methods discussed above after a discretization in space<br />
<strong>and</strong> time. However, the adjoint variables to the CFD equation always have<br />
the same dimensionality as the solution <strong>of</strong> the CFD equation, i.e., if y is a<br />
function <strong>of</strong> space <strong>and</strong> time, then so is λ. When performing simulations <strong>of</strong><br />
CFD problems, one typically provides only a space discretization <strong>of</strong> the flow<br />
variables <strong>and</strong> marches forward in time so that during the whole simulation,<br />
the main memory is only populated by one (or two, but at most six, depending<br />
on the time marching scheme) spatially discretized state variable pertaining<br />
to the solution at the current time-step. Unfortunately, the time direction is<br />
reversed for the corresponding adjoint problems <strong>and</strong>, in nonlinear problems,<br />
the solution <strong>of</strong> the (discretized) adjoint problem needs information from the<br />
(discretized) primal CFD solution y. That means, a direct application <strong>of</strong> the<br />
methodology sketched above to time-dependent problems requires storage for<br />
at least the size <strong>of</strong> the whole space-time history <strong>of</strong> the solution y <strong>of</strong> the CFD<br />
problem. If this amount <strong>of</strong> storage easily fits into the available memory, one<br />
can just skip the following comments <strong>and</strong> continue with the next section,<br />
where we discuss special features <strong>of</strong> the resulting SQP methods. Otherwise,<br />
there are mainly four options:<br />
• It is possible to counteract the lack <strong>of</strong> storage by investing more computing<br />
time. In the situation <strong>of</strong> time-dependent optimization problems, this can<br />
be facilitated by the use <strong>of</strong> so-called check-pointing strategies as described<br />
in [18]. The algorithmic complexity then grows only logarithmically in the<br />
number <strong>of</strong> time-steps to be thus virtually stored <strong>and</strong> so does the actual<br />
storage space required. Note that the multiple shooting approach outlined<br />
below provides a natural kind <strong>of</strong> checkpoints.<br />
• Another interesting option is model reduction. In a trivial form this could<br />
just mean choosing a coarser space-grid, which is however not feasible in<br />
many cases. But the same line <strong>of</strong> thought leads to proper orthogonal decomposition<br />
(POD), where low dimensional ansatz spaces are constructed<br />
by the use <strong>of</strong> snapshots, which mirror already the major properties <strong>of</strong> the<br />
solution <strong>of</strong> the CFD problem. The result is a low dimensional system <strong>of</strong><br />
Ordinary Differential Equations (ODE) or Differential Algebraic Equations<br />
(DAE), for which the optimization problems can be solved in the<br />
fashion <strong>of</strong> the methods discussed below. A description <strong>of</strong> the use <strong>of</strong> POD