Remove a site from an MPS/MPO about itensor HOT 3 CLOSED

Hi, the short answer is: at the moment, no. A loop like the one in your example would be the way to do it. We probably won't add such a method unless there is a common use case. Typically, removing a tensor from an MPO (of ITensors or IQTensors) would lead to it being ill-defined because there would now be a pair of virtual/link indices that no longer match up. Or, if you removed it from the end there would be a "dangling" virtual/link index. Does your use case not run in to either of these issues?

from itensor.

Hi, thanks for the quick reply. It runs fine, it's just a bit slow. I am not running a normal DMRG-like algorithm (http://arxiv.org/abs/1401.4874) so I have been careful in re-defining the spin and link index vectors. I am only really using the MPO as a convenient way of accessing the operators rather than the more complex operations that it allows. I suppose that some vector of pointers to ITensors would work, but I'll have to think a bit more about that.

from itensor.

Oh, ok I see. Nice paper - I had been wondering what this type of RG would look like as a tensor network. I wrote a tree tensor network code for a very different use case and for that I just created a new container class instead of using MPO. One tip (you may have already realized) is to define a new "Model" class at each layer. (Another person who was trying something like this was using the primeLevel feature to use the same Site indices at each level of the tree but with more and more primes---not a good way to go.) Shoot me an email [email protected] if you'd like to discuss any more code details. E.g. I wrote a special Model class that can automatically create operators at each level of the tree based only on the quantum numbers of the site indices.

from itensor.