Optimal control problem governed by an infinite dimensional two-nilpotent bilinear system (Q6107954): Difference between revisions

The authors consider the bilinear system: \(\overset{.}{x}(t)=Ax(t)+u(t)Bx(t)\) , \(x(0)=x_{0}\), where \(A\) and \(B\) are bounded linear operators on a Hilbert space \(E\), \(x(t)\in E\) is the trajectory associated to the control \(u(t)\in \mathbb{R}\) and \([0,T]\) is a fixed time interval. For all \(t\in \lbrack 0,T]\) , the optimal control problem consists to find the control \(u(t)\) which minimizes the functional \(J(u)=\int_{0}^{T}u^{2}dt+h\left\langle x(T),Fx(T)\right\rangle \), under the constraint associated to the preceding system, where \(F\) is a coercive, symmetric and bounded operator. The authors first observe that the space \(\mathcal{L}(E)\) of bounded operators in \(E\) has a structure of Lie algebra for the bracket \([A,B]=AB-BA\) and they define the endomorphism \(Ad\) on \(\mathcal{L}(E)\) through \(Ad_{A}(B)=[A,B]\). For any parts \(M\) and \(N\) of \(\mathcal{L}(E)\), \(Ad_{M}(N)\) is the space generated by the elements \(Ad_{A}(B)\)), with \(A\in M\) and \(B\in N\). \((Ad_{M})^{k}(N)\) is then defined by recurrence on \(k\). \(M\) is a nilpotent Lie sub-algebra of degree \(k\) if there exists an integer \(l\) such that \((Ad_{M})^{l}(M)=\{0\}\) and \(k\) is the smallest integer \(l\) which has this property. The purpose of the paper is to construct two matrices \(A\) and \(B\) which generate a nilpotent Lie algebra of degree 2 such that: \( (Ad_{A})^{2}(B)=(Ad_{B})^{2}(A)=0\) and\ \([A,B]\neq 0\) in finite and infinite dimensions. In the finite dimension (\(n\)) case, the main result proves that for any nonzero matrix \(C\) of \(M_{n}(\mathbb{C})\) which is defined by its elementary divisors \(\{\underset{p}{\underbrace{\lambda ^{2},\ldots ,\lambda ^{2}},}\underset{q}{\underbrace{\lambda ,\ldots ,\lambda }}\}\), with \(p\neq 0 \), \(q\neq 0\) and \(n=2p+q\), then for all \(n>3\), there are two matrices \(A\) and \(B\) of \(M_{n}(\mathbb{C})\) such that \([A,B]=C\), \([A,C]=0\), and \([B,C]=0\) . The authors first consider the cases \(n=1,\ldots ,6\).\ Considering the general case, they use the Jordan normal form of the matrix \(C\) and of the matrices which commute with \(C\). In the infinite dimension case, the main result proves that for any operator \(C\) of \(\mathcal{L}(E)\) such that and \( C^{2}=0\) and \(C\neq 0\), there are two compact operators \(A,B\) such that: \( [A,B]=C\), \([A,C]=0\), and \([B,C]=0\). The authors decompose \(E\) as \( E=E_{0}\oplus _{i\in \mathbb{N}^{+}}E_{i}\), where \(E_{0}\) is a subspace of \( Ker(C)\) and \(E_{i}=span\{e_{i},C(ei)\}\).\ They consider the spaces \( E_{n}=E_{0}\oplus _{i=1,\ldots ,n}E_{i}\) and they build a chain of finite rank operators \(C_{i}\), through \(C_{n}=\oplus _{i=1,\ldots ,n}C_{i}\oplus C_{0}\) and prove that the limit operator \(C\) is a compact operator which satisfies the desired properties. The paper ends with considerations on an optimal control of the error in the problem: \(\overset{.}{x}(t)=\Delta x(t)+Au(t)+B(x(t),u(t))\), \(x(0)=x_{0}\), \(y(t)=Cx(t)\), involving an observer.