In some cases an assembly strategy might be needed where first a basis is assembled (and orthogonalized) on a few selected points and only after that assembled using a greedy strategy.
One use case is e.g. Hamiltonians with conserved quantities that show a block-diagonal structure such that a greedy assembly with a initial snapshot within a certain conserved plateau might get stuck immediately. In these cases one could apply latin hypercube sampling.
As an error estimate just provide the magnitude of the sum of neglected eigenvalues or sum of squared singular values, respectively (see Certified Reduced Basis Methods for Parametrized PDEs).
In some cases an assembly strategy might be needed where first a basis is assembled (and orthogonalized) on a few selected points and only after that assembled using a greedy strategy.
One use case is e.g. Hamiltonians with conserved quantities that show a block-diagonal structure such that a greedy assembly with a initial snapshot within a certain conserved plateau might get stuck immediately. In these cases one could apply latin hypercube sampling.