upgrade qml.qsvt #6520
upgrade qml.qsvt #6520
Conversation
KetpuntoG
added
the
do not merge ⚠️
|
Hello. You may have forgotten to update the changelog!
|
| This circuit applies a polynomial transformation (:math:`Poly^{SV}`) to the singular values of | ||
| the block encoded matrix: | ||
| - ``"prepselprep"``: Embeds the Hamiltonian ``A`` using :class:`~pennylane.PrepSelPrep`. Default encoding for Hamiltonians. | ||
| - ``"qubitization"``: Embeds the Hamiltonian ``A`` using :class:`~pennylane.Qubitization`. |
There was a problem hiding this comment.
Isn't this just prep sel prep again? What value does adding the reflection have for block encoding ..?
There was a problem hiding this comment.
It is a fair point. In terms of efficiency it always makes more sense to use prepselprep. The reason I include it is:
-
to emphasize the idea that qubitization is a block encoding (which is not super widespread).
-
enable all the block encodings we have available in PennyLane so that the user has the power to do what he wants. Maybe you want to verify that QSVT gives the same results with one method and with another.
If you think it is still better to remove it, we can discuss it :)
| poly (tensor_like): Polynomial coefficients defining the transformation, represented in increasing order of degree. | ||
| This means the first coefficient corresponds to the constant term, the second to the linear term, and so on. |
There was a problem hiding this comment.
What happens if poly isn't a fixed parity?
There was a problem hiding this comment.
An error will be raise on poly_to_angles
| if block_encoding not in ["embedding", "fable", None]: | ||
| raise ValueError("block_encoding should take the value 'embedding' or 'fable'") | ||
|
|
There was a problem hiding this comment.
Can't we use Pauli decompose on the matrix and then use the PrepSelPrep or Qubitization methods?
There was a problem hiding this comment.
answered in previous comment:
"""
By definition, one is methods for block encoding arrays and the other for block encoding hamiltonians. I think that to avoid confusion, if a user wants to mix things up, they can do it outside the function
"""
Another problem that comes to mind is that we cannot predict the number of auxiliary qubit we will need since we do not know in advance the number of pauli terms we will have given a matrix
| if global_phase: | ||
| with qml.QueuingManager.stop_recording(): | ||
| global_phase_op = qml.GlobalPhase(-0.5 * np.pi * (4 - global_phase), wires=wires) | ||
| if block_encoding not in ["prepselprep", "qubitization", None]: |
There was a problem hiding this comment.
Why can't we use the other two methods here?
There was a problem hiding this comment.
By definition, one is methods for block encoding arrays and the other for block encoding hamiltonians. I think that to avoid confusion, if a user wants to mix things up, they can do it outside the function
A = qml.matrix(hamiltonian)
or
hamiltonian = qml.pauli_decompose(A)
| def qsvt(A, angles, wires, convention=None): | ||
| r"""Implements the | ||
| `quantum singular value transformation <https://arxiv.org/abs/1806.01838>`__ (QSVT) circuit. | ||
| def qsvt(A, poly, encoding_wires, block_encoding=None): |
There was a problem hiding this comment.
might be outside scope but you could probably make A also take a PauliSentence directly.
There was a problem hiding this comment.
It could be an option :)
In that case it would be enough to update qml.PrepSelPrep to make qml.Qubitization and qsvt work. Let's leave it out of this PR
| s = int(np.ceil(np.log2(max(len(A), len(A[0]))))) | ||
| encoding = qml.FABLE(2**s * A, wires=encoding_wires) |
There was a problem hiding this comment.
It's not clear what this code is doing even with the comment. Maybe re-name the variable s and add more to the comment?
There was a problem hiding this comment.
I extracted from the last cell in :
https://docs.pennylane.ai/en/stable/code/api/pennylane.FABLE.html
Basically FABLE does block encoding of A/2^n, which is not very intuitive since what we want is to do block encoding of A. Therefore, I am doing FABLE(2^n*A), so that the final result is block encoding of A.
Co-authored-by: Jay Soni <[email protected]>
…lane into upgrade_qsvt_2
|
Thanks for the review @Jaybsoni 🙌 |
This PR upgrades
qml.qsvtbased on this PRD . To check how the documentation is displayed -> [here]The main changes are:
qml.qsvtonly worked withqml.BlockEncode. With the newqml.qsvtthe user can choose the specific block encoding.qml.qsvtthe user had to pass the appropriate angles to apply a polynomial. Now the user passes the polynomial directlyThese changes aim to make
qml.qsvt, the friendly (but less customizable) version to use QSVT.[sc-72651] [sc-72650]