Cryoris
commented
2 years ago The bound parameter values may not show up in your circuit diagram as being bound, but they are 🙂 Here's your example with 3 repetitions (odd and larger than 2):
┌───────────┐┌───────────┐┌───────────┐┌───────────┐┌───────────┐┌───────────┐┌───────────┐┌───────────┐┌───────────┐┌───────────┐
q: ┤ exp(it X) ├┤ exp(it Y) ├┤ exp(it X) ├┤ exp(it Y) ├┤ exp(it X) ├┤ exp(it Y) ├┤ exp(it X) ├┤ exp(it Y) ├┤ exp(it X) ├┤ exp(it Y) ├
└───────────┘└───────────┘└───────────┘└───────────┘└───────────┘└───────────┘└───────────┘└───────────┘└───────────┘└───────────┘but if we call .decompose() on the circuit and have a look inside we see the values are actually bound:
┌───────┐┌───────┐┌───────┐┌───────┐┌───────┐┌───────┐┌───────┐┌───────┐┌───────┐┌───────┐
q: ┤ Rx(2) ├┤ Ry(2) ├┤ Rx(2) ├┤ Ry(2) ├┤ Rx(2) ├┤ Ry(2) ├┤ Rx(2) ├┤ Ry(2) ├┤ Rx(2) ├┤ Ry(2) ├
└───────┘└───────┘└───────┘└───────┘└───────┘└───────┘└───────┘└───────┘└───────┘└───────┘By the way, you might want to consider using the qiskit.circuit.library.PauliEvolutionGate for these evolutions as it's the most up to date implementation 🙂
fazlekarim-anthem
Environment
What is happening?
When running a
PauliTrotterEvolutionwith odd order greater than 1 on anEvolvedOpoperator that has aPauliOporPauliSumOpprimitive, the code ignores the parameters binded to it.How can we reproduce the issue?
When running the following example:
where the order is 2 (or an even number) we get the following circuit:
If we repeat the above code with an order of 3 (or an odd number greater than 2) we will get the following circuit:
As you notice, the parameters are not being binded when the order is an odd number greater than 2.
What should happen?
The expected outcome is when there should be a parameter binded to the circuit for odd orders
Any suggestions?
(As a disclaimer, I am still familiarizing myself with the subject and codebase).
I came to the realization that regardless of what the
trotter_modeis inPauliTrotterEvolution, if the operator is anEvolvedOPand if its primitive is aPauliOporPauliSumOp, we will use the predefined trotter in the class where itsSuzukiTrotterwhen the order is greater than 1 andLieTrotterwhen the order is 1 (regardless of what the user defines it to be). ThisSuzukiTrotterused in the class has a different implementation thanqiskit.opflow.evolutions.trotterizations.suzuki.py.When we look into the implementation of
SuzukiTrotter: https://github.com/Qiskit/qiskit-terra/blob/e148f97b9afe869511fe82dc6a0e83b7b33dd217/qiskit/synthesis/evolution/suzuki_trotter.py#L129-L146we notice when
order==1we return apauli_listwithout multiplying it with time. This begs the question, what do we mean by order here? Is an order of n the same as 2k of the implemented paper or an order of n the same as k?If the answer is the prior, then we need to have an assertion such that the order, n, can not be an odd number as k is always an integer greater than 1.
If its the later, then we must return the following
[(op, coeff * time) for op, coeff in pauli_list]when order==1.I am looking at order == 1 because we recursively will go back to it when the order is an odd number greater than 2: https://github.com/Qiskit/qiskit-terra/blob/e148f97b9afe869511fe82dc6a0e83b7b33dd217/qiskit/synthesis/evolution/suzuki_trotter.py#L139-L146.
One more thing to do is compare these two codes (
SuzukiTrotterandqiskit.opflow.evolutions.trotterizations.suzuki.py). In one implementation time is taken in consideration and in the other time is not.