cnick1 / nlbalancing Goto Github PK

Construct the pseudoinverse rather than solving for it. N*N is a diagonal matrix, as is the Nk (1 \otimes Sigma^2) squared, but then the cross terms are also diagonal, so I have a block matrix with diagonal blocks.

1. Instead of computing those blocks with matrix multiplication, I can cosntruct
   (A*A.') directly/analytically.

In particular example 5 seems to be misbehaving potentially

Make each verbose function announce itself and then indent all the messages

need to substitute d-1 with d

The unsymmetrized coefficients are sparse. This could be used to reduce memory usage. Furthermore, certain operations may be faster with a small number of nonzero entries. On the other hand, certain operations are faster using symmetry.
Can we only symmetrize when necessary? Or unsymmetrize when possible to accelerate?

If rotary inertia is included, the mass matrix will be asymmetric and a standard LU solver is needed instead of cholesky

For example, 1 and 5 are the same if only up to quadratic, so just combine
Use syntax [A,B,C,N,f,g,h] = getSystem#()

Then use runExample to differentiate them