TL;DR; anyone? Originally I thought this lets one calculate eigenvectors directly from matrix and its submatrices's eigenvalues, but looking at the paper, it is not exactly that good.
Element wise:
v(i,j)=√∏δ(k,i)(λi(A) − λk(A))/δ(k,n)(λi(A) − λk(Mj))
δ(x,y)(f):= ±1 if x=y else f
1. Roots of characteristic equation det(A-λI)=0 give the eigenvalues.
2. Solve (A-λI)x = 0 for each eigenvalue to find each eigenvector x.
It seems very plausible that step 2 can be reduced to an equation.