07.02.2013 Views

Optimization and Computational Fluid Dynamics - Department of ...

Optimization and Computational Fluid Dynamics - Department of ...

Optimization and Computational Fluid Dynamics - Department of ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

3 Mathematical Aspects <strong>of</strong> CFD-based <strong>Optimization</strong> 65<br />

If g is symmetric <strong>and</strong> positive definite on the null space <strong>of</strong> A,thisisequivalent<br />

to the linear quadratic problem<br />

min 1/2∆z ⊤ G∆z+ � ∇f k − (A − cz(z k )) ⊤ λ k�⊤ ∆z (3.14)<br />

s.t. A∆z + c(z k ) = 0 (3.15)<br />

where the adjoint variable for Eq. (3.15) is λ k+1 . In this way, accurate evaluations<br />

<strong>of</strong> the matrices cz or H can be avoided. Nevertheless, the matrix cz<br />

appears in the objective <strong>of</strong> the quadratic program (QP), however, only in the<br />

form <strong>of</strong> a matrix-vector product with λ k , which can be efficiently realized<br />

such as in the adjoint mode <strong>of</strong> automatic differentiation [17].<br />

This formulation is generalizable to inequality constraints, cf. [6, 7], e.g.,<br />

<strong>of</strong> the form c(z) ≥ 0 in (3.10), which leads to linear-quadratic sub-problems<br />

<strong>of</strong> the form<br />

min 1/2∆z ⊤ G∆z+ � ∇f k − (A − cz(z k )) ⊤ λ k�⊤ ∆z (3.16)<br />

s.t. A∆z + c(z k ) ≥ 0 . (3.17)<br />

The adjoint variables <strong>of</strong> Eq. (3.17) converge to the adjoint variables <strong>of</strong> the inequality<br />

constraints at the solution. Therefore, they provide a proper decision<br />

criterion within an active-set strategy. This formulation <strong>of</strong> SQP methods allowing<br />

for approximations <strong>of</strong> derivatives, forms also the basis for the real-time<br />

investigations presented later (see also [14]).<br />

3.2.2 Modular SQP Methods<br />

Here, we discuss again the separability framework, i.e., problems <strong>of</strong> the type<br />

(3.4-3.6) – first without inequality constraints<br />

min f(y,p) (3.18)<br />

y,p<br />

s.t. c(y,p)=0. (3.19)<br />

The essential feature is the nonsingularity <strong>of</strong> cy, which means that we can<br />

look at the problem as an unconstrained problem <strong>of</strong> the form<br />

min<br />

p f(y(p),p) . (3.20)<br />

When applied to the necessary optimality condition ∇pf(y(p),p)=0,Newton’s<br />

method, or its variants, yield good local convergence properties. Every<br />

iteration consists <strong>of</strong> two steps<br />

(1) solve B∆p = −∇pf(y(p k ),p k )=0,whereB ≈∇p 2 f(y(p k ),p k )<br />

(2) update p k+1 = p k + τ · ∆p, whereτ is an appropriate step length

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!