From: Dave Rusin Subject: Re: STABILITY ANALYSIS : Help Please Date: Thu, 6 May 1999 07:36:24 -0500 (CDT) Newsgroups: [missing] To: lsouriau@club-internet.fr Keywords: boundedness of terms of a linear recurrence >Here is the PROBLEM : > > (1) A(k) Wn+1(k) - B(k) Wn(k) + C(k) Wn-1(k) = D(k) > n coresponds to the current time step : t=n Dt. >Initial conditions : W0(k) = W1(k) = 0 > >I'd like to know what is the condition on factors A, B, C , D >(function of k, real number) so that Wn(k) is stable versus time ? >(No divergence of Wn(k) torward + infinity ) Your problem may depend on k, but the question concerns one k at a time. So the question (for each k) seems to be, If W0=W1=0 and A Wn+1 - B Wn + C Wn-1 = D for all n >=1, is {Wn} bounded? That's a simple question of linear recurrence relations. If, for example, A is nonzero [in your problem there may be particular values of k which make A(k)=0 even if A(k) is not identically zero] then we let F = sum( Wn * X^n ) be the formal power series; the given conditions show F = D / ( (1-X)(A X^2 - B X + C) ). Now use partial fractions to write F = a/(1-X) + b/(1-rX) + c/(1-sX) [special cases if A-B+C=0 or B^2=4AC] to see Wn = a + b r^n + c s^n, which stays bounded iff |r|<=1 and |s|<=1 [special cases if b=0 or c=0, as well as if r and s are purely imaginary] So apart from a few special cases the conclusion is, "W stays bounded iff both roots of the polynomial A X^2 - B X + C have complex magnitude at most equal to 1". This in turn simply means that the ratios (B/A, C/A) lie in a certain compact subset of the plane. [deletia -- djr]