Definition (Free endpoint problem)
We wish to find Admissible Controls that defined on [t0,t1], starting at x(t0)=x0 s.t. we minimize the cost J(u)=t0∫t1(xu⊤(t)L(t)xu(t)+u⊤(t)u(t))dt+xu⊤(t1)Qxu(t1)where the index u in xu is used to emphasize that the trajectory depends on the choice of control, L:[t0,t1]→Mn(R) is a Continuous function, and Q∈Mn(R). Note that in this problem we are not concerned with where the trajectory is going to at time t1, but only w.r.t. minimizing the cost.
Lemma (Free endpoint lemma)
Consider the LTVC system x˙=A(t)x(t)+B(t)u(t)and suppose that K(⋅) is a differentiable matrix function on [t0,t1] s.t. for K(t)=K⊤(t),∀t∈[t0,t1]. Then x⊤(t1)K(t1)x(t1)−x⊤(t0)K(t0)x(t0)=t0∫t1[u⊤(t)x⊤(t)][0K(t)B(t)B⊤(t)K(t)K˙(t)+A⊤(t)K(t)+K(t)A(t)]dt
Theorem (Solution to free endpoint problem)
Consider a LTVC system x˙(t)=A(t)x(t)+B(t)u(t)where A,B are Continuous functions of time and x(t0)=x0. If ∃K:[t0,t1]→Mn(R) with K(t)=K⊤(t),∀t∈[t0,t1] which satisfies the Riccati Differential Equation: K˙(t)K(t1)=−A⊤(t)K(t)−K(t)A(t)−L(t)+K(t)B(t)B⊤(t)K(t)=Qthe solution to which we denote by t↦Π(t,Q,t1), then there exists an optimal control u∗:[t0,t1]→Rm for the Free endpoint problem with the cost J(u)=t0∫t1(xu⊤(t)L(t)xu(t)+u⊤(t)u(t))dt+xu⊤(t1)Qxu(t1)In particular, the optimal control is u∗(t)=−B⊤(t)Π(t,Q,t1)x(t),t∈[t0,t1] a continuous function, and the optimal cost is Jmin=x0⊤Π(t0,Q,t1)x0