TensorNetwork

TensorNetwork copied to clipboard

mganahl

We should add an implementation of the CGM linear solver, which is provably optimal for symmetric systems. NumPy and Jax both have implementations of this already, although I don't think...

alewis

We should add an implementation of the LGMRES sparse linear solver, which outperforms GMRES in most cases, especially for nearly symmetric operators.

alewis

It seems to me that we should support integer and perhaps boolean operations as much as possible due to their ubiquity in machine learning applications. Currently I think most of...

alewis

PEPS

2

The major remaining v1.0 roadblock is a functioning PEPS example that runs acceptably on CPU NumPy and Jax. Let us get this done and discuss the issue here. Some time...

alewis

zeros_like, ones_like, empty_like for BlockSparseTensor and symmetric backend

4

This is useful for initializing new BlockSparseTensor objects from existing ones with compatible shapes, charges and flows

mganahl

Once we have batch support for `ncon` we need to have that as well

mganahl

Environments

1

We should develop a way to easily/cheaply calculate all of the environments of a tensor network without any dangling edges.

chaserileyroberts

Have FiniteMPS.{left, right}_envs use BaseMPS.apply_transfer_operator

1

I might be missing something, but I believe the contractions in `FiniteMPS.left_envs` are identical to those in `BaseMPS.left_transfer_operator`. I suggest the alteration to make use of the latter's JITting. I...

AidanGG

Array class

3

It seems there is widespread agreement that, in the longer term, `Node` is eventually to be replaced by or supplemented with some sort of "I can't believe it's not an...

alewis