TEBDEvolver
class TEBDEvolver(evolution_state, *args, order=2, **kwargs)
A TEBD algorithm for evolving an internal MPO.
As discussed in more detail in quimb_tebd
, this extension of quimb
’s existing quimb.tensor.TEBD
implementation time-evolves an internal matrix product operator (MPO) rather than the conventional matrix product state (MPS).
More concretely, the internal object is expected to be an MPOState
.
The API of this class is actually much larger than shown here, because it inherits additional methods from the quimb.tensor.TEBD
base class. However, we do not duplicate that API here.
Initialize a TEBDEvolver
instance.
Parameters
-
evolution_state (MPOState) –
a reference to the time-evolution state. This overwrites the
p0
argument of the underlyingquimb.tensor.TEBD
class.WarningIn contrast to the default behavior, this state will NOT be canonicalized. Instead, it is taken as is and is kept by reference (i.e. no copy is created). This ensures that the same object can be shared between two instances of this class, as required by the
DynamicMPF
algorithm. -
args – any further positional arguments will be forwarded to the
quimb.tensor.TEBD
constructor. -
order (int) – the order of the builtin Suzuki-Trotter formula to use during time evolution. This will be the value forwarded to the
quimb.tensor.TEBD.step()
method. -
kwargs – any further keyword arguments will be forwarded to the
quimb.tensor.TEBD
constructor.
Attributes
conjugate
Returns whether this time-evolver instance acts on the right-hand side.
evolved_time
Returns the current evolution time.
Methods
step
step()
Perform a single time step of TEBD.
This essentially calls quimb.tensor.TEBD.step()
and forwards the value of the order
attribute that was provided upon construction.
Return type
None
sweep
sweep(direction, dt_frac, dt=None, queue=False)
Perform a single sweep of the TEBD algorithm [1].
The TEBD algorithm updates the even and odd bonds of the underlying tensor network in alternating fashion. In the implementation of the quimb.tensor.TEBD
base class, this is realized in the form of alternating “sweeps” in left and right directions over the internal state.
We are overwriting the behavior of this method in this subclass, in order to call the specialized gate_split()
method.
Parameters
- direction (Literal['left', 'right']) – the direction of the sweep. This must be either of the literal strings,
"left"
or"right"
. - dt_frac (float) – what fraction of the internal time step (
dt
) to time-evolve for. This is how any builtin Suzuki-Trotter formula specifies its splitting. - dt (float | None) – an optional value to overwrite the internal time step.
- queue (bool) – setting this to
True
will raise aNotImplementedError
.
Raises
- NotImplementedError – if
queue=True
. - NotImplementedError – if
cyclic
isTrue
. - NotImplementedError – if
imag
isTrue
. - RuntimeError – if an invalid
direction
is provided.
Return type
None
References
[1]: https://en.wikipedia.org/wiki/Time-evolving_block_decimation