Optimization or flow control problems have the usual three
ingredients. First, one has an objective, a reason why one wants to control the
flow. Mathematically such an objective functional is expressed as a cost, or
performance functional. Next, one has controls or design parameters are expessed
in terms of unknown data in the mathematical specification of the problem.
Finally, one has contraints that determine what type of flow one is interested
in and that place direct or indirect limited on candidate optimizers.
Mathematically, the type of flow is in terms of a specific set of partial
differential equations.
Actually, In this talk, we concern the
Stokes equations as the type of folw. In our case, the optimization problem we
consider is to find an optimal state $(\mathbf{u}, p)$ and an optimal control
$\mathbf{f}$ which minimize the $L^2$-norm distance between $\mathbf{u}$ and
target velocity $\hat{\mathbf{u}}$ subject to $(\mathbf{u},p,\mathbf{f})$
satisfying the Stokes system. Such constraint optimization problem can be
converted to the unconstrained optimization problems by the Lagrange multiplier
rule. It is a standard approach for solving finite-dimensional constrained
optimization problem([3,4]). As a result, we have a coupled optimality system
related to two Stokes type equations associate with state variables and adjoint
variables. In this system, we use only state variables and adjoint variables but
controls. This optimality system can be dealt with mixed finite element
approaches. But, in our case, we use first-order system least-squares(FOSLS)
technique by using the state flux $'\mathbf{U} = \nabla{\mathbf{u}}^t$ of the
state velocity $\mathbf{u}$ and the adjoint flux $\mathbf{V} =
\nabla{\mathbf{v}}^t$ of the adjoint velocity $\mathbf{v}$. We reformulate the
coupled second-order optimality system as the coupled first-order optimality
system. Then, we define the least-squares functionals based on the uncontrained
optimal system in terms of appropriate weights and norms for the system. In
particular, we impose a weight $\frac{1}{\sigma^2}$ on adjoint system to get a
good control result. We show that the least-squares functional is equivalent to
$H^1$ product norms by proving the $H^2$-regularity of that optimality
system[2]. This principles result in symmetric and positive definite algebraic
system. We show that a least-squares principle based on $L^2$ norms applied to
this system yields optimal discretization error estimates $H^1$ norm in each
variable.
For the numerical test, we use multigrid V-cycle
method using Gauss-Seidel smoothing iteration on the unit square domain. And we
also use the single approximating space of continuous piecewise linear
polynomials for the approximations of all unknowns. Note that we use a simple
target velocity that is a solution of the Stokes system. We examine the effects
of changes in the parameters \sigma and $\nu$. Generally, effecting good control
requires small values of penalty parameter $\sigma$. So, we observe the $L^2$
norm of between controlled flow and target flow. That is, we show the controlled
flow behavior when $\sigma =1,0.1,0.01$, and $0.001$ for fixed kinematic
viscosity $\nu=1$. Then, we see that the controlled flow matches the target flow
very well for small values of penalty parameter $\sigma$. Now, we consider the
controlled flow for different kinematic viscosity $\nu =1,0.1,0.01$ and $0.001$
when $\sigma=1$ is fixed. In this case, we see the norms of |u-hat{u}| and |f|
are very similar to the ones of above varying $\sigma$ case. These computational
results mean that if $\nu$ is small, then the flow is easily controlled by small
control flow $\mathbf{f}$. Finally, we consider the dependence of mesh size
$h=1/4,1/8,1/16,1/32$ for fixed $\nu=1$ and $\sigma=0.001$. We wee here that the
norm of $\|\mathbf{u}-\hat{\mathbf{u}}\|$ and the value of objective functional
are decreased.
As a conclusion, we applied on optimal control
problem with Stokes equations as contraints by using the earler FOSLS work[1] on
Stokes system. Moreover, we imposed weights $\frac{1}{\sigma^2}$ and $\nu^2$ for
good control effecting good control of small penalty parameter and viscosity. In
this sense, if we find the best weights then, a least-squares principles seems
to be proper for optimal control problems.
References
[1] Z. Cai, R.
Lazarov, T. A. Manteufel, and S. F. Mccormick, First-order system least squares
for the Stokes equation, with application to linear elasticity, SIAM J. Numer.
Anal. Vol. 34 (1997), pp. 1727-1741.
[2] S. Agmon, A. Douglis, and L.
Nirenberg, Estimates near the boundary for solutions of elliptic partial
differential equations satisfying general boundary conditions II, Comm. Pure
Appl. Math. 17 (1964), 35-92.
[3] P. Bochev and M. Gunzburger.
Least-squares finite element methods for optimality systems arising in
optimization and cotrol problems, SIAM J. Numer. Anal. Vol. 43 (2006), pp.
2517-2543.
[4] M. Gunzburger, Perspectives in Flow Control and
Optimization, SIAM, Philadelphia, 2003.
(This work was supported
by KRF-2005-070-C00017 and BK21 MCDI)