Hi,
I am wondering if OMPS alows to add two generic MPOs and truncate/approximate the result to a new MPO using a rounding procedure (based for example of SVD decompositions of the cores) in order to reduce the bond dimensions.
thanks
Raffaele
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
no, at the moment both the addition and the reduction are steps which are
not implemented. The addition seems to be not that useful as we could
define the MPO directly from both rule sets. As an additional insight: the
addition of MPS is implemented variationally as this approach is more
efficient, where the target MPS has a fixed bond dimension and we can avoid
the SVD, which would reduce the bond dimension after a "true" addition. The
SVD cannot be copied from MPS or MPDO for the MPO; the MPO is designed as a
sparse matrix. Before thinking about the implementation of such
sophisticated algorithms for MPOs, I would try a lot to reduce the bond
dimension of the single MPO and maybe define a tailored rule for my case.
What bond dimension of the MPO are you expecting for your research?
Hi,
I am wondering if OMPS alows to add two generic MPOs and
truncate/approximate the result to a new MPO using a rounding procedure
(based for example of SVD decompositions of the cores) in order to reduce
the bond dimensions.
Hi,
I am wondering if OMPS alows to add two generic MPOs and truncate/approximate the result to a new MPO using a rounding procedure (based for example of SVD decompositions of the cores) in order to reduce the bond dimensions.
thanks
Raffaele
Hello Raffaele,
no, at the moment both the addition and the reduction are steps which are
not implemented. The addition seems to be not that useful as we could
define the MPO directly from both rule sets. As an additional insight: the
addition of MPS is implemented variationally as this approach is more
efficient, where the target MPS has a fixed bond dimension and we can avoid
the SVD, which would reduce the bond dimension after a "true" addition. The
SVD cannot be copied from MPS or MPDO for the MPO; the MPO is designed as a
sparse matrix. Before thinking about the implementation of such
sophisticated algorithms for MPOs, I would try a lot to reduce the bond
dimension of the single MPO and maybe define a tailored rule for my case.
What bond dimension of the MPO are you expecting for your research?
Best regards,
Daniel
On Wed, Sep 25, 2019 at 3:02 PM lello lelloborrelli@users.sourceforge.net
wrote: