Decompositions and coalescing eigenvalues of symmetric definite pencils depending on parameters (Q2098811)

Let \(\Omega \) be an open connected region in \(\mathbb{R}^{2}\) and let \(A\) and \(B\) be smooth functions from \(\Omega \) into \(\mathbb{R}^{n\times n}\) such that \(A(x)\) is symmetric and \(B(x)\) is positive definite for all \(x\in \Omega \). Consider the generalized eigenvalue problem \[(A(x)-\lambda B(x))v=0. \tag{*}\] It can be shown that, with a suitable choice of ordering of the eigenvalues and eigenvectors, the eigenvalues and eigenvectors of (*) also depend smoothly on \(x\) except at values of \(x\) where there are equal eigenvalues (referred to as points of conical intersections or coalescences). The aim of this paper is to detect these conical intersections. Since \(B(x)\) is positive definite, its Cholessky decomposiiton can be used to reduce (*) to an equivalent ordinary eigenvalue problem. However, the authors point out that this route has several disadvantages; for example, sparseness or banding in the matrices \(A(x)\) and \(B(x)\) is usually lost. They therefore approach the problem of directly solving (*) using the following algorithm (called ``Algorithm 1.1'' in the paper) to detect coalescences. Step 1. Let \(\Gamma \) be a simple closed loop in \(\Omega \) parametrized by the smooth, \(1\)-periodic function \(\gamma (t)\) (\(0\leq t\leq 1\)) and assume that (*) has distinct eigenvalues \(\lambda _{1}(\gamma (t))>\lambda _{2}(\gamma (t))>\cdots >\lambda _{n}(\gamma (t))\) for all \( t\in \lbrack 0,1]\); Step 2. Obtain the smooth diagonalization of the pencil \( (A(x),B(x))\) along the loop \(\Gamma \): \(A(x)V(x)=V(x)B(x)\Lambda (x)\) with \( x:=\gamma (t)\) (where \(\Lambda (x):=\operatorname{diag}(\lambda _{1}(x),\dots,\lambda _{n}(x)) \) and \(V(\gamma (t))\) is nonsingular for all \(t\in \lbrack 0,1]\)); Step 3. Compare \(V(\gamma (0))\) and \(V(\gamma (1))\); determine if a pair of eigenvalues coalesced inside \(\Gamma \) and, if so, which ones. They use an implementation of their approach to provide a statistical study of the number of conical intersections.