Averaging method in the asymptotic integration problem for systems with oscillatory-decreasing coefficients (Q2480483)

The author presents a change of variables that, combined with the Levinson asymptotic integration theorem, allows for an accurate asymptotic investigation of the solutions to \[ \frac{d^2y}{dt^2}+\left(1+a\frac{\sin\varphi(t)}{\sqrt{t}}\right)y=0, \] where \(\varphi(t)=t+\alpha t^{\beta}\) and \(\varphi(t)=t+\alpha\log t\). These are cases that escape the powerful asymptotic development results in the literature. His main theorem reads as follows: Consider the differential system \[ \begin{multlined} \frac{dx}{dt}=[A_0+\sum_{i=1}^n A_i(t)v_i(t)+\sum_{1\leq i_1\leq i_2\leq n}A_{i_1i_2}(t)v_{i_1}(t)v_{i_2}(t)+\cdots+\\ \sum_{1\leq i_1\leq\cdots\leq i_k\leq n}A_{i_1\cdots i_k}(t)v_{i_1}(t)\cdots v_{i_k}(t)+R(t)]x,\end{multlined} \] where (1) \(A_0\) is a constant matrix with real eigenvalues; (2) \(v_i(t)\rightarrow 0\) as \(t\rightarrow+\infty\); (3) \(v_{i}^{\prime}\in L^1\) and \(v_{i_1}\times\dots\times v_{i_k}\in L^1\) for all subsets of \(\{1,\dots, n\}\); (4) the entries of the matrices \(A_{i_1\dots i_l}(t)\) are trigonometric polynomials, that is \(A_{i_1\dots i_l}(t)=\sum_{j}\Delta_{j}^{(i_1,\dots,i_l)}\exp({\mathbf i}\;t\lambda_jt)\), where the \(\Delta\)'s are complex matrices; (6) \(R\in L^1\). Then, the change of variables \[ \begin{multlined} x=[I+\sum_{i=1}^n Y_i(t)v_i(t)+\sum_{1\leq i_1\leq i_2\leq n}Y_{i_1i_2}(t)v_{i_1}(t)v_{i_2}(t)+\cdots+\\ \sum_{1\leq i_1\leq\cdots\leq i_k\leq n}Y_{i_1\cdots i_k}(t)v_{i_1}(t)\cdots v_{i_k}(t)]y,\end{multlined} \] where \(I\) is the identity and the entries of the matrices \(Y\) are trigonometric polynomials with zero mean value, reduces the differential system for sufficiently large \(t\) to the form \[ \begin{multlined} \frac{dx}{dt}=[A_0+\sum_{i=1}^n A_iv_i(t)+\sum_{1\leq i_1\leq i_2\leq n}A_{i_1i_2}v_{i_1}(t)v_{i_2}(t)+\cdots+\\ \sum_{1\leq i_1\leq\cdots\leq i_k\leq n}A_{i_1\cdots i_k}v_{i_1}(t)\cdots v_{i_k}(t)+R_1(t)]y,\end{multlined} \] where the matrices \(A\) are constant and \(R_1\in L^1\). Several ways of by-passing the restriction to a real spectrum imposed to \(A_0\) are dicussed in Remarks 1,2 -- for instance, in the important case when the matrices \(A\) are periodic of period \(T\), the matrix \(A_0\) should not possess eigenvalues such that \(\lambda_r-\lambda_p=\frac{2\pi q\;{\mathbf i}}{T}\) for \(r\neq p\). For the first case of \(\varphi\), it is established that \(y(t)=C_1\sin(t-(a^2/12)\log t+C_2)+o(1)\) as \(t\rightarrow+\infty\), where \(C_1,C_2\in\mathbb{R}\). This estimate holds for \(\beta\in(0,1)\). Here, the ''\(\log\)'' has \(e\) as base. For the second case of \(\varphi\), it is shown that \(-5a^2/24\leq\alpha\leq a^2/24\) is a parametric resonance domain, that is, here the equation exhibits a solution verifying \(y(t)=C_1\log t\sin(t+\alpha\log t-\nu)+O(1)\) as \(t\rightarrow+\infty\), where \(\nu=0\) for \(\alpha=-5a^2/24\) and \(\alpha=\pi/2\) when \(\alpha=a^2/24\). After performing in both cases changes of variables following from this theorem, the authors presents in detail a complicated Lyapunov (type of) change of variables that allows for recasting the (transformed) first order differential system into another first order system that is suitable for applying the Levinson theorem. At some point the author incorporates in his analysis the recent Bodine-Lutz improvement of the fundamental Harris-Lutz approach in the Levinson asymptotic integration theory. This is a very interesting paper and a must for people interested in the adiabatic oscillator problems/quantum mechanics.