|
|
sci.math |
d^2u/dx^2 - P.du/dx + (P^2-Q^2)/4.u = 0 where the matrix coefficients - with infinitesimal grid-spacings (dx) - a = exp(-(P-Q)/2.dx) ; b = exp(+(P+Q)/2.dx) And the general solution is of the form (with A,B arbitrary): u(x) = A.exp((P-Q)/2.x) + B.exp((P+Q)/2.x) The key reference is: http://hdebruijn.soo.dto.tudelft.nl/hdb_spul/calculus.pdf Han de Bruijn Jules wrote: Special cases of such infinite systems of linear equations have been
> Suppose we have a doubly-indexed sequence {b_(i, j)} of real numbers,
> with i, j positive integers. Suppose also that we have a sequence
> {c_i} of reals. We wish to find a sequence {a_j} of reals so that the
> sum as j goes from 1 to infinity of b_(i, j) * a_j = c_i for each i.
> This is, in some sense, a collection of countably-many linear equations
> in countably-many variables. Are there any conditions on the
> coefficients b_(i, j) that would guarantee existence and/or uniqueness
> of a solution {a_j}? If there were only finitely-many equations and
> variables (the same number of each), then one could simply check that
> the determinant of the coefficient matrix is non-zero. Is there any
> analog of determinant for an "omega-by-omega" matrix?
investigated. In "Multigrid Calculus" it is proved that the infinite
tri-diagonal system of linear equations:
........
-b 1 -a
-b 1 -a
-b 1 -a
.........
is equivalent with a second order ordinary differential equation:
are given by: