Replace the InitialState components by circuits #1374
Conversation
Cryoris
requested review from
levbishop,
manoelmarques,
pbark,
stefan-woerner and
woodsp-ibm
as code owners
Cryoris
added
Changelog: Deprecation
There was a problem hiding this comment.
Since InitialStates are now circuits wasn't the intent to reflect the circuit library and have a qiskit.chemistry.circuit.library where presumably we would have an initial_states folder in there containing these. Leaving just stubs where they are now, the stub extending the now moved one so it can be still be used from where it was, but which emit deprecation messages to change to use them from the new location.
| return np.zeros(self._num_parameters, dtype=np.float) | ||
| else: | ||
| return None | ||
| return np.zeros(self._num_parameters, dtype=np.float) |
There was a problem hiding this comment.
So this basically ends up suggesting to VQE that all zeros is the right start point no matter what initial state is supplied (when one is supplied). I guess this should be ok since for now this is normally used with HF init state. It was anticipated that for some other starting point like MP2 we would be indicating other values though clearly to date we have not had code that does this.
| Custom(state_vector) | qc = QuantumCircuit(n) | ||
| | qc.initialize(state_vector, qc.qubits) # normalize vector! | ||
| Custom('random') | use the initialize instruction with random amplitudes | ||
| VarFormBased | use the circuit of the variational form directly |
There was a problem hiding this comment.
We might need to provide some guidance somewhere on this due to the parameter sorting that is done by VQE? As long as you get the optimal dictionary I guess one is good - get the optimal params and you need to take care over sorting. I'm thinking of the use case where we take the var form that had been evolved somewhat by VQE and use that as the starting state for QPE - i.e. the test_vqe2iqpe that I see has been updated to do the above. Maybe we can show this in some tutorial as I am not sure there is a good place in the docs unless we did something in VQE
- move the initial state circuit to chemistry.circuit.library - globally replace num_particles by a tuple instead of list of integers
|
I moved the circuits to |
| var_form = TwoLocal(num_qubits, ['ry', 'rz'], 'cz', initial_state=init_state) | ||
| var_form = TwoLocal(num_qubits, ['ry', 'rz'], 'cz') | ||
|
|
||
| # add the initial state | ||
| var_form.compose(init_state, front=True) |
There was a problem hiding this comment.
So this is a required change in that initial_state will not accommodate a QuantumCircuit? If it is now done this way it seems the var form will not have visibility to the initial state as it had when it was passed directly and it would prepend it itself.
There was a problem hiding this comment.
We had a workaround in place where the NLocal circuits expected a InitialState. I'll add support for passing a QuantumCircuit and deprecate the InitialStates once this is merged. We can update the readme sample to use the kwarg again.
qiskit/chemistry/core/hamiltonian.py
Outdated
| two_qubit_reduction=self._two_qubit_reduction, | ||
| num_particles=self._molecule_info[self.INFO_NUM_PARTICLES]) | ||
| z2_symmetries = Hamiltonian._pick_sector(z2_symmetries, hf_state.bitstr) | ||
| from ..circuit.library.initial_states.hartree_fock import _build_bitstr |
There was a problem hiding this comment.
To me this does not look so good running off a private member. Maybe the better thing to do is to have some util somewhere that can build the bitstr so that the HF state or anything else like this can use it. If nothing it should be at least a public method of HFState if its preferable to keep it there for now.
There was a problem hiding this comment.
I thought about these two options too, but the problem with these are, that
- to extract the bitstring from the circuit we'd have to simulate the circuit and retrieve the bitstring, which is exponentially expensive
- having the property for the bitstring on it is a bit weird because initial states don't generally have to have a bitstring
- also we're creating a circuit but only care about the bitstring it is built from, this seems like an inverted workflow
How about changing this from a private function to a public method? E.g. hartree_fock_bitstring?
There was a problem hiding this comment.
A public method seems fine - the main suggestion was some sort of utility method/module somewhere outside of the circuit that the circuit could use. We have no util folder at present, hence other things, like MP2Info are in the root. Adding another file there for one method seems too much. Hence having it there with the circuit as a public method seems ok for now. hartree_fock_bitstring or hf_bitstring seems fine to me.
…kit-aqua#1374) * make chemistry initial states circuits * put VSCF circuit in new file * update several tests * fix type change to tuple * merge hf initial state and circuit * merge initstate+circuit * update EOH * update hhl and qaia * update chc and uvcc * deprecate InitialState components * update varformbased test * fix running tests & mypy * add reno * fix iqpe test * fix variable renaming from merge * move to circuit.lib and use tuple - move the initial state circuit to chemistry.circuit.library - globally replace num_particles by a tuple instead of list of integers * make _build_bitstr public functions
…kit-aqua#1374) * make chemistry initial states circuits * put VSCF circuit in new file * update several tests * fix type change to tuple * merge hf initial state and circuit * merge initstate+circuit * update EOH * update hhl and qaia * update chc and uvcc * deprecate InitialState components * update varformbased test * fix running tests & mypy * add reno * fix iqpe test * fix variable renaming from merge * move to circuit.lib and use tuple - move the initial state circuit to chemistry.circuit.library - globally replace num_particles by a tuple instead of list of integers * make _build_bitstr public functions
…kit-aqua#1374) * make chemistry initial states circuits * put VSCF circuit in new file * update several tests * fix type change to tuple * merge hf initial state and circuit * merge initstate+circuit * update EOH * update hhl and qaia * update chc and uvcc * deprecate InitialState components * update varformbased test * fix running tests & mypy * add reno * fix iqpe test * fix variable renaming from merge * move to circuit.lib and use tuple - move the initial state circuit to chemistry.circuit.library - globally replace num_particles by a tuple instead of list of integers * make _build_bitstr public functions
Summary
Convert
HartreeFockandVSCFto circuits and deprecated the other initial state components. This fits the overall picture of using circuits as algorithm building blocks instead of specialized Aqua types.Details and comments
Most of the functionality of the
InitialStatecomponents is already available in theQuantumCircuit(essentially by using theinitializemethod). Therefore, these components are no longer necessary. The specialized initial states have been converted toQuantumCircuittypes.The compatibility of all algorithms/components has been updated to support initial states given as circuits.
The replacement of
HartreeFockandVSCFis done in-place (i.e. they implement bothQuantumCircuitandInitialState) until we decide where circuits in the applications module should be kept.