Matrix Riccati equations and matrix Sturm-Liouville problems. (Q1424488)

Let \(H_k\) denotes the set of all Hermitian \(k\times k\)-matrices. Let \(P, Q, R: [a, b] \to H_k\) be integrable functions and let \(\alpha \in \mathbb R\). Consider the matrix Sturm-Liouville eigenvalue problem \[ u' = Rv, \quad v' = -(\lambda P + Q) u, \tag{1} \] \[ u(a) = 0, u(b) \cos (\alpha /2) = v(b) \sin (\alpha /2). \tag{2} \] Let \(\lambda_0 \leq \lambda_1 \leq \lambda_2 \leq \cdots\) be the list of nonnegative repeated (according to multiplicity) eigenvalues of (1), (2). In [\textit{F. V. Atkinson}, Discrete and continuous boundary problems (1964; Zbl 0117.05806), Chapter 10], for continuous \(P, Q, R\) and positive definite \(P(t)\), \(R(t)\), an estimate was given: \(B_1^{-1} \leq \liminf_{r \to \infty} r^{-1/2} n(r) \leq \limsup_{r \to \infty} r^{-1/2} n(r) \leq A_1^{-1}\) for \(n\) large, where \(n(r)\) denotes the number of repeated eigenvalues within \([0, r]\) and \(A_1 > 0\), \(B_1 > 0\) could be computable from the data \(P, Q, R\). In this paper, the author improves the estimates. In fact, not only it is proved that \(r^{-1/2} n(r)\) has a limit as \(r \to \infty\) but also a corresponding formula is given. The method used in this article, following Atkinson, is a generalization from the Prüfer transformation for the case \(k = 1\). It proceeds as follows. Let \(U(t, \lambda), V(t, \lambda)\) denote the unique \(k\times k\)-matrix solution satisfying \[ U' = RV, \quad V' = -(\lambda P + Q)U \] with the initial conditions \(U(a, \lambda) = 0, V(a, \lambda) = E\), where \(E\) is the identity matrix. Let \(\Theta (t, \lambda) = [V(t, \lambda) + i U(t, \lambda)] [V(t, \lambda) - i U(t, \lambda)]^{-1}\). Then \(\Theta (t, \lambda)\) is well-defined and unitary for every \(t \in [a, b]\), \(\lambda \in \mathbb R\). Let \(\mu = \lambda^{1/2}\), \(U_k\) be the set of all unitary \(k\times k\)-matrices, then \(\Theta (\cdot , \mu) = \Theta : [a, b] \to U_k\) will satisfy the matrix Riccati equation \[ \Theta' = (i \mu)/2 [ F - G + \Theta (F - G) \Theta + \Theta (F + G) \Theta + (F + G) \Theta], \] where \(F, G: [a, b] \to H_k\) are obtainable from \(P, Q\) and are integrable as well. Let \(e^{i \phi_j (t, \lambda)}, j = 1, \cdots, k\), with \(\phi_j:[a, b] \times \mathbb R \to \mathbb R \text{ being continuous, } \phi_j (a, \lambda) = 0\), as well as \(\phi_1 (t, \lambda) \leq \phi_2 (t, \lambda) \leq \cdots \leq \phi_k (t, \lambda) \leq \phi_1 (t, \lambda) + 2 \pi\), be the repeated eigenvalues of \(\Theta (t, \lambda)\), then they are related to \(\lambda\) as follows: \(\lambda\) is a real eigenvalue of (1), (2), if and only if \(\phi_j (b, \lambda) \in \alpha + 2 \pi Z\). So the number of the multiplicity of \(j\)'s that satisfy \(\phi_j (b, \lambda) \in \alpha + 2 \pi Z\) is equal to the number of the multiplicity of \(\lambda\). Owing to this interaction, the author investigates the above-mentioned matrix Riccati equation. By considering the coefficients \(F, G\) in the Banach space \(L^1([a, b], H_k)\), through an approximation procedure under the norm in \( L^1([a, b], H_k) \times L^1([a, b], H_k)\), they get the following main result: \(\lim_{\mu \to \infty} \mu^{-1} (\arg \det \Theta (b, \mu) - \arg \det \Theta (a, \mu)) = \int_a^b \nu (F(t), G(t)) \, dt\), where \(\det \Theta (\cdot, \mu)\) is a continuous function \(\phi:[a, b] \to \mathbb R\) such that \(\det \Theta (t, \mu) = e^{i \phi (t)}\) for all \(t \in [a, b]\) and \(\nu (F, G) = \lim_{t \to \infty} t^{-1} \arg \det \Gamma (t)\); in the definition of \(\nu\), \(F, G\) are specified to be constant matrices, and \(\Gamma (t)\) is the corresponding solution of the above-mentioned Riccati equation with \(F(t) \equiv F, G(t) \equiv G\). The result above leads to the desired conclusion for (1), (2): If \(P(t)\) is positive semidefinite for \(t \in [a, b]\) a.e., then either every complex number is an eigenvalue or \(\lim_{r \to \infty} r^{-1/2} n(r) = ( | L |) /(2 \pi)\), where \(L = \int_a^b \nu (P(t), R(t)) \, dt\). Specifically, if \(P(t)\) is positive a.e. and \(R(t)\) is regular on a set of positive measure, only the latter case can occur. Furthermore, with suitable weaker assumptions on \(P, Q, R\), there are also some related results on the existence and distribution of large positive eigenvalues. In the last section, an example of computing \(L\) is given.