Optimal control problem in non-standard form with many integral constraints

118 Views Asked by At

I am trying to solve the following optimization program and I am hoping for some help as I am stuck:

$\min_{R_{t}\in[0,1]}\,\,\int_{0}^{\tau}e^{-t}R_{t}\,dt$

s.t. $e^{-t}R_{t}-\int_{t}^{\tau}e^{-s}R_{s}ds\geq c$ for all $t$ (with equality at $t=\tau$)

for some fixed $\tau >0$ and $c>0$.

I conjecture that the optimal solution has $R_t = c e^\tau$ for all t. But I cannot prove it.

This problem looks somewhat like an optimal control problem, but it's in an odd form. For example, I tried converting it into an optimal control problem by replacing the constraint with $e^{-t}R_{t}-\int_{t}^{\tau}e^{-s}R_{s}ds + u_t = c$, where $u_t \geq 0$ is the control, and $u_\tau = 0$, but I don't know how to deal with the integral.

If I assume that $\dot{R}_t$ exists (which I would rather not), I can use integration by parts, and rewrite the constraint as $\int_{t}^{\tau}e^{-s}\dot{R}_{s}ds\leq0$ for all t and $R_\tau = c e^\tau$. It's still unclear, however, how I might go about solving this problem.

Even if I consider the relaxed problem where I consider only the constraint corresponding to t=0, so that the problem can be rewritten as

$\min_{R_{t}\in[0,1]}\,\,\int_{0}^{\tau}e^{-t}R_{t}\,dt$

s.t. $\int_{0}^{\tau}e^{-t}\dot{R}_{t}dt\leq0$,

I still don't know how to solve this program.

If I can show that $\dot{R}_t = 0$ for all $t$ in this relaxed problem, because the constraints for all $t \in (0,\tau)$ are satisfied by this solution, I'll be done.

Any help would be very much appreciated! Thank you!

Happy Holidays!

1

There are 1 best solutions below

0
On

Figured out how to solve the problem. First, define the state variable $X_{t}=\int_{\tau}^{t}e^{-s}R_{s}ds$. Now the problem can be rewritten as

$\max_{u_{t}\geq0}X_{0}$

s.t. $\dot{X}_{t}=u_{t}+c-X_{t}\\X_{\tau}=0$

This problem is non-standard, because the terminal value of $X$ is fixed, while the initial value is free. By changing variables $r = \tau - t$, the problem can be transformed into a canonical optimal control problem with a free endpoint and solved using standard techniques.

The optimal solution has $R_t = ce^{\tau}$.