Misc

[PhilipVinc]

expect(op::AbstractOperator{B,B}, state::Ket{B}) where B<:Basis = state.data' * (op * state).data
expect(op::AbstractOperator{B1,B2}, state::AbstractOperator{B2,B2}) where {B1<:Basis,B2<:Basis} = tr(op*state)

we could add a specialization for data operators
which would make expect non allocating.

expect(op::DataOperator{B,B}, state::Ket{B}) where B<:Basis = dot(state.data', op.data, state.data)
expect(op:: DataOperator{B1,B2}, state::AbstractOperator{B2,B2}) where {B1<:Basis,B2<:Basis} = dot(op.data', state.data)

[david-pl]

There already are specialized methods implemented: https://github.com/qojulia/QuantumOpticsBase.jl/blob/46b7693e7742095ad60a092ae23e17ffe7a56b9d/src/operators_dense.jl#L148

Or do you mean adding methods for when state::AbstractOperator ?

[PhilipVinc]

yeah, I mean for a mixed state (state::AbstractOperator), currently (if I am not mistaken), the definition is
expect(op::AbstractOperator{B1,B2}, state::AbstractOperator{B2,B2}) where {B1<:Basis,B2<:Basis} = tr(op*state)

which will allocate because of the matmul.

but by being smart with linear algebra, it is in general true that tr(op*state) = dot(op.data', state.data).

[david-pl]

Getting back to this now and I'm still a bit confused.

The methods for ::AbstractOperator types use the matmul definition tr(op*state) because they don't necessarily have a .data field (they can be lazy). So you can't just use dot (you'd need to implement a custom version).

For operators that do have .data fields (<:DataOperator), there are already specialized methods that don't allocate e.g. here https://github.com/qojulia/QuantumOpticsBase.jl/blob/1a432191a43490814e946803296d602c59fda37c/src/operators_dense.jl#L158 (and another one for sparse data).

Funny enough, the method for state::Ket actually wasn't using dot 🤷 Here's the fix: qojulia/QuantumOpticsBase.jl#34

Can we close this issue?

[PhilipVinc]

i think so, yes. probably i had missed the specialized method