Reduced Density Matrices about tensornetwork HOT 7 CLOSED

Thanks for trying out TensorNetwork! As you may have noticed, we don't currently build any semantics into TensorNetwork itself - we make no assumptions in the core code about the meaning of the tensor that the network represents. This is a clean way to implement basic tensor network features.

That said, it would be interesting to explore how best to add such quantum functionality! One way would be to create a "wavefunction" or "quantum_state" class that inherits from our Network class. This could then provide features like the ones you suggest on top of what we have already. Alternatively, this could also be done via a library of functions that take tensor networks as inputs, returning e.g. a new network representing the reduced state.

Any thoughts @Thenerdstation and @mganahl ?

from tensornetwork.

I agree with Ash, this seems like something for for an extension of TensorNetwork.

Though I also wonder if it's even really needed. Doing it with the current features of the library should be fairly straightforward.

For example:

net1, qubit_edges1 = make_wavefunction(...)
net2, qubit_edges2 = make_wavefunction(...)
net1.add_subnetwork(net2)
for qubit in [3, 5, 7]: # or whatever you want.
  net.connect(qubit_edges1[qubit], qubit_edges2[qubit])

tensornetwork.contractors.naive(net1)

net1.get_final_node().tensor # Here is your RDM.

Pretty short and sweet. The only other reason I could see for adding this is to optimize the compute needed to calculate the RDM. This would relate back to our "inner shape vs outer shape" issue from earlier.

from tensornetwork.

Yes, this stuff is not hard to do (you're quite the quantum physicist these days, @Thenerdstation!). However, including it either in TensorNetwork, or in a package that extends it, would surely be useful for some people. This could form a new "example", or be an extension of the "wavefunctions" example we currently have, or it could be a separate project, or it could become a core part of TensorNetwork.

My inclination would be not to add it to the core since I think it is better to focus on the basics in TensorNetwork. So, insofar as this goes beyond an "example" or "experiment", my feeling is it would be better off as a separate, dependent project. How about "QuantumTensorNetwork"? ;)

from tensornetwork.

Thanks! I've been taking classes ^-^.

from tensornetwork.

Related question: What is our long-term plan for the "experiments"? Do they eventually get upgraded to fully-fledged parts of the library, or split out into their own separate projects?

from tensornetwork.

Thanks! If you do make a QuantumTensorNetwork dependent project, I would use it!

from tensornetwork.

I think we'll keep it under this TensorNetwork package.

from tensornetwork.