Given a matrix A that is potentially close to a TL matrix, we would like to generate a TL matrix by
T = TLMat(A)
so that an efficient representation of A is obtained. This is currently achieved by computing its displacement D(A), and computing a SVD of D(A), which is expensive (cubic operation).
A more economic choice would be to use Adaptive Cross Approximation on D(A) to recover a low rank approximation (essentially d * n^2, where d is the rank). We then should also support syntax for controlling the rank, i.e.
T = TLMat(A, d) % prescribed drank d
T = TLMat(A, tol) % prescribed tolerance pivot_k / pivot_1 that indicates to stop
Given a matrix A that is potentially close to a TL matrix, we would like to generate a TL matrix by
T = TLMat(A)
so that an efficient representation of A is obtained. This is currently achieved by computing its displacement D(A), and computing a SVD of D(A), which is expensive (cubic operation).
A more economic choice would be to use Adaptive Cross Approximation on D(A) to recover a low rank approximation (essentially d * n^2, where d is the rank). We then should also support syntax for controlling the rank, i.e.
T = TLMat(A, d) % prescribed drank d T = TLMat(A, tol) % prescribed tolerance pivot_k / pivot_1 that indicates to stop