Re: Matrix is close to singular or badly scaled.

[Jawad Gull]

kavitha <kavitha...@yahoo.co.in> wrote in message <24238363.1213399712...@nitrogen.mathforum.org>...
> Please suggest.
I also had this problem with the stiffness matrix of beam elements. What I find is sometimes there are uncoupled and unused degree of freedoms. Like in 3DOF beam element we might not use the axial degree of freedom. I mean we don't apply any load in that direction and don't constraint it in that direction. And this cause problem. When I constraint my unused degree of freedom (No matter there is any load in that direction or not) this problem of "Matrix is close to singular or badly scaled" goes away

[Karsten Reuß]

kavitha <kavitha...@yahoo.co.in> wrote in message <24238363.1213399712...@nitrogen.mathforum.org>...
> Please suggest.

Just a practical tip from my experience:
It usually helps to look at the correlation matrix of your matrix "x".
That is corr(x).
I visualize it by commands like imagesc first and then browse through the columns for strange correlations. Things that can happen are: Correlations close to 1 or -1 (apart from the diagonal where correlations are always 1). Then you have to look at those highly or perfectly correlated variables and try to figure out the reason, e.g. if it's a coding error, if some variable is screwed up or if you have to drop one or several of the variables.