Thanks for the answer.
I speak Russian but do not have a Russian keyboard with me right now.
However: The problem is NOT in storing the results, but in the way optimal control is defined.
I have control u=f(x,y,t) with x and y state and costate variables respectively. I need to evaluate u as f(t) at every step to avoid computational burden.
For example I have u(x,y,t)=x^2+y^2-ln(t) and x=f(t), y=f(t), but defined through numerical procedures as proc(t)... proc.
I need maple to evaluate u(t) given this. However with procedure-defined functions this seems to be difficult...
This control function is supposed to be used then for the next step of the procedure, namely
0. Take initials;
1. Compute x,y as functions from initials and t
2. Use x, y to define u(t) - control function
3. Update x,y from 1.
And I have the problem at step 2.