Van Vleck's theorem on continued fractions (Q2467668)

The general theory given in the paper concerns continued composition of mappings in \({\mathbb R}^{N}\). The \(N \times N\) matrix \(a(j)\) is quasi-orthogonal in the sense that \(a(j) \operatorname{transp} \{ a(j) \}\) is a diagonal matrix whose diagonal elements are all equal to \(\|a(j)\|^2\) (\(\|a(j)\|\) positive) (\(j \geq 1\)). \(\Sigma\) is a half space consisting of those \(x\) in \({\mathbb R}^N\) for which \((x,e) > 0\), \(e\) being a specified unit vector in \({\mathbb R}^N\) where \((.,.)\) denotes the scalar product. The inverse \(inv(x)\) of nonzero \(x\) in \({\mathbb R}^N\) is taken to be \(x / (x,x)\). \(t(j|x)\) is \(a(j) \operatorname{inv}(b(j)+x)\), \(b(j)\) being so chosen that \(t(j|\Sigma) \subset \Sigma\) (\(j \geq 1\)). \(T(a,b\mid |n| x)\) is the continued composition \(t(1|t(2|\dots t(n-1|t(n|x))\dots))\) . (\(T(a,b\mid |0|x)=e\)). The notation \(T(1,b\mid|n|x) \) is used in the special case in which all \( a(j) \) are the unit matrix. \( \tau (x) \) being \( x - 2(x,e)e\), \(k(n)\) is \(\tau\{ T(a,b\mid|n-1|0)\}\) (\(n\geq 1\)). \(|x|\) denotes the Euclidean norm of \(x\in {\mathbb R}^N\). \(\lim T(a,b \| n | \Sigma) \) refers to the increase \( n \to \infty \); it is either a hyper-sphere of nonzero radius or a point. Four results are derived. [0] \( \lim T(1,b || n | \Sigma) \) is a point if and only if the series \( \{ \sum (b(j),e) / (k(j),e) | j \geq 1 \}\) diverges. [1] Suppose the \( a(j) \) to be strictly orthogonal (i.e. \( \| a(j) \| = 1 \). The series \( \{ \sum (b(j),e) / (k(j),e) | j \geq 1 \} \) converges if and only if (i) \( \sum | b(j) | j \geq 1 \} \) converges and (ii) \( \alpha > 0 \) exists such that \( | k(2j) | \to \alpha \) and \( | k(2j+1) | \to 1 / \alpha \) as \( j \to \infty \). [2] If the series \[ \{ \sum \sqrt \{(b(j),e) . (b(j+1),e) / \| a(j) \| | j \geq 1 \} \] diverges, \( \lim T(1,b \| n | \Sigma) \) is a point. [3] Positive real \( M \) being fixed, let \( \Omega \) be the set of all \( x \) for which \( | x | \leq M . (x,e) \) and let \( b(j) \) be in \( \Omega \) (\( j \geq 1 \)). The following assertions are equivalent. (a) \( \lim T(a,b \| n | \Sigma) \) is a point. (b) \( x \) being in \( \Omega \), \( \lim T(a,b \| n | x) \) is in the closure of \( \Sigma \). (c) \( \{ \sum b(j) | j \geq 1 \}\) diverges. Setting \( N = 2 \), the above results may be transformed in terms of complex numbers. Now \(inv_x\) is simply \(1/x\). The limit hyper-sphere case becomes a limit disc case. In [0] above, \( e \) becomes \( 1 + 0i \) and the inner product ratio \( (b(j),e) / (k(j),e) \) becomes \(\text{Re}{ b(j) } /\text{Re}{k(j) } \). In [1], reference to \(a(j)\) may be discarded and the modulus signs removed from \(|k(2j)|\) , \(|k(2j+1)|\). In [2] the inner products may be replaced by real parts, as for [0]. In the reformulated version of [3], \( K(0) \) is taken to be the right half plane, and \( \zeta (0) \) to be \( 1 \). Discs \( K(j) \) (\( j \geq 1 \)) for which \( K(0) \supset K(1) \supset K(0) \supset \ldots \), \( K(j+1) \) being internally tangent to \( K(j) \) at \( \zeta (j) \), the \( \zeta (j) \) being distinct are postulated. \( b(j) \) with \(\text{Re}{b(j) } \geq 0 \) (\( j \geq 1 \)) exist for which \( K(j) = T(1,b \| j | K(0)) \) and \( \zeta (j) = T(1,b \| j | 0) \) (\(j \geq 1\)). In the paper the complex variable results precede the more general theory. In particular the authors construct a continued fraction of limit disc type which nevertheless converges. Vector continued fractions of the form \(b(0)+a(1) \text{inv}[b(1)+a(2) \text{inv}[b(2)+\ldots ]]\) in which the \(a(j)\) are scalars and the \(b(j)\) vectors were introduced by the reviewer [Lin. Alg. Appl. 1, 357--395 (1968; Zbl 0164.18503)]. Convergence results for such continued fractions have been obtained [\textit{H. Zhao} and \textit{P. Xiao}, Numer. Math., Nanjing 26, No. 1, 18--24 (2004; Zbl 1054.40003)]. In the Zbl review of that paper the present reviewer pointed out that the results hold in a far more general setting -- in particular they apply to the case in which the \( a(j) \) are scalar multiples of rotation matrices. Precisely results of this sort are the concern of the paper considered in this review.