The Operator module¶
The Operator module defines the common interfaces used to interact with quantum operators and super-operators, as well as several concrete implementations of different operators such as netket.hilbert.LocalOperator, netket.hilbert.Ising and others.
NetKet’s operators are all sub-classes of the abstract class netket.hilbert.AbstractOperator, which defines a small set of API respected by all implementations. The inheritance diagram for the class hierarchy of the Operators included with NetKet is shown below (you can click on the nodes in the graph to go to their API documentation page). Dashed nodes represent abstract classes that cannot be instantiated, while the others are concrete and they can be instantiated.
Similarly to Hilbert spaces, there are two large classes of operators: DiscreteOperator and ContinuousOperator. Evidently the formers will only work with Discrete Hilbert spaces, while the latters will only work with continuous Hilbert spaces.
The main function of operators in NetKet is to define the logic and some values used to compute expectation values over a variational state. Functions implemented by operators are either needed to compute expectation values, or are nice utilities useful to manipulate or inspect the operators but are not needed by the Monte-Carlo logic interacting with the variational states.
All AbstractOperators act on a well defined hilbert space that can be accessed through the
It is also possible to check if the operator is hermitian through the boolean property
There are only two other operations defined on all operator types: it is possible to take the conjugate or conjugate-transpose of an operator by accessing the methods
Those will usually return lazy wrappers.
Finally, it is also possible to call
collect() to get rid of any possible lazy wrapper.
The bare-minimum requirement when defining a custom operator is to define it’s hilbert space.
Most likely you will also want to define the
expect and/or the
expect_and_grad method to compute the expectation value of such operator over a certain Variational State.
Contrary to more standard Pythonic code, those methods are not defined as class-functions in your custom operator class, but you have to use multiple dispatch (netket.utils.dispatch.dispatch) to define those methods on a specific signature such as
expect(vstate: MCState, O: MyCustomOperator).
This is needed because the way you compute expectation values and/or gradients might not only change depending on the exact operator, but depends also on the type of variational state that you are working with.
To learn more about multiple dispatch, check this section
An explanation of how to define the
expect method for custom operators is given in the Custom Operator documentation.
The DiscreteOperator interface¶
Operators defined on Discrete Spaces must implement the DiscreteOperator interface if they wish to work ou