An inverse spectral theorem for a Hill's matrix (Q788216)

The author is concerned with a second order linear difference equation analogous to Hill's differential equation, namely \[ -b_{n-1}x_{n- 1}+(\lambda -a_ n)x_ n-b_ nx_{n+1}=0\quad(n=0,\pm 1,...), \] where \(a_ n,b_ n\) are real, \(b_ n>0\) and \(a_{n+N}=a_ n\), \(b_{n+N}=b_ n\) for some ''period'' N. The behaviour of the solutions is governed by a certain polynomial \(\Delta^ 2(t)-4\) of degree 2N, whose roots can be arranged in the order \(\lambda_ 0>\lambda_ 1\geq \lambda_ 2>\lambda_ 3\geq \lambda_ 4>...>\lambda_{2N-1}\). If \(\lambda\) lies in one of the ''instability intervals'' \((\lambda_ 2,\lambda_ 1),(\lambda_ 4,\lambda_ 3),...\), then the fundamental solutions with \(x_{n+N}/x_ n\) constant are unbounded. When do the coefficients \(a_ n,b_ n\) in fact have a smaller period N/d, where d is a divisor of N? The author's main result is that this is so if, and only if, \(\lambda_{2k}=\lambda_{2k-1}\) whenever k is not a multiple of d. The proof rests on a uniqueness theorem which states that the difference equation is completely specified by certain information about it.