# # A linear-quadratic control problem as suggested by # # K. Schittkowski, # ``Numerical solution of a time-optimal parabolic boundary-value # control problem'', J. Optim. Theory Appl., vol. 27, pp. 271-290, 1979, # # H. Goldberg and F. Troeltzsch, # ``On a Lagrange-Newton method for a nonlinear parabolic boundary control # problem'', Optim. Meth. Software, vol. 8, pp. 225-247, 1998. # # and # # H. D. Mittelmann, # ``Sufficient Optimality for Discretized Parabolic and # Elliptic Control Problems'', to appear in # Proc. Workshop Fast solution of discretized optimization problems, # WIAS Berlin, 5/2000, Birkhaeuser-Verlag, Basel # See also Hans Mittelmann's WWW article # # http://plato.la.asu.edu/papers/paper91/paper.html # # where the problem is example 5.2-I. # # this example was also coded in SIF by Nick Gould and added to CUTE # as CONT5-QP, classification QLR2-MN-V-V # param n default 200; param m default n; param n1 := n-1; param m1 := m-1; param dx := 1/n; param T default 1.58; param dt := T/m; param h2 := dx^2; param a := .001; param yt{j in 0..n} := .5*(1 - (j*dx)^2); var y{0..m, 0..n}; var u{i in 1..m}; minimize f: .25*dx*( (y[m,0] - yt[0])^2 + 2* sum{j in 1..n1} (y[m,j] - yt[j])^2 + (y[m,n] - yt[n])^2) + .25*a*dt*( 2* sum{i in 1..m1} u[i]^2 + u[m]^2); s.t. pde{i in 0..m1, j in 1..n1}: (y[i+1,j] - y[i,j])/dt = .5*(y[i,j-1] - 2*y[i,j] + y[i,j+1] + y[i+1,j-1] - 2*y[i+1,j] + y[i+1,j+1])/h2; s.t. ic {j in 0..n}: y[0,j] = 0; s.t. bc1 {i in 1..m}: y[i,2] - 4*y[i,1] + 3*y[i,0] = 0; s.t. bc2 {i in 1..m}: (y[i,n-2] - 4*y[i,n1] + 3*y[i,n])/(2*dx) = u[i]-y[i,n]; s.t. cc{i in 1..m}: -1 <= u[i] <= 1; s.t. sc{i in 0..m,j in 0..n}: 0 <= y[i,j] <= 1;