How to specify tolerance using linsolve ?

Hi,

I have a linear system that I want to solve with Matlab using linsolve. My problem is that the coefficients' range of the system is huge. The minimum coefficient start at e-20 while the max reach 10e25. When I use linsolve, I get a warning telling me "Warning: Rank deficient, rank = 15, tol = 5.2711e+005.". When I use rank on my system, I effectively observe that my system is not full rank. But when I use rank(sys, tol) with a small tol value (~e-15), I obtain a full rank system.

So I suppose that linsolve choose a tolerance that is to high for my system and my question is if there exists a way to specify a tolerance value to linsolve in ordre to solve system with a high range of coefficients.

Thanks

Answer 1

[pietro]

It is strongly suggested to normalize your design variables, contraints and so on in order to have a similar variation of the gradient and lagragian functions. Please post your problem and I will help you

[Hugo]

I try to solve a system in ordrer to compute the coefficients of the solution of a Laplace problem in a domain that looks like a superposition of annulus. The linear system to solve is described here (from page 8) : http://onlinewww.jpier.org/PIER/pier92/01.09032301.pdf

In summary, I put the coefficients of all equations given in this paper in a matrix representing my system. Having infinite sum and indices this system is theoretically of infinite size but I truncate it to a certain order. My code is pretty long but is nothing else than the coefficients expressions listed in this paper (eq (38)–(49), (52)–(54), (56)–(60), (62) and (64)). I can put it here if needed.

[pietro]

well the design variables may be scaled in this way:

(x-lb)./(ub-lb)

where x is the array with the design variables, ub the array with upper bounds and lb is the array with lower bounds

[Hugo]

Thanks for your help, normalization indeed solve my problem.