Giters

PennyLaneAI / pennylane-aqt

The PennyLane-AQT plugin integrates Alpine Quantum Technologies' ion-trap quantum computing hardware with with PennyLane's quantum machine learning capabilities.

Home Page:https://docs.pennylane.ai/projects/aqt

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

gate CNOT not supported

sassan72 opened this issue · comments

commented

i am using Pennylane_aqt for my simulation. i got this error:
File "/home/anaconda3/lib/python3.7/site-packages/pennylane/_device.py", line 673, in check_validity
"Gate {} not supported on device {}".format(operation_name, self.short_name)
pennylane._device.DeviceError: Gate CNOT not supported on device pennylane_aqt.Simulator

commented

Thanks for bringing this to our attention. We will start looking into a fix.

commented

thanks. i realized the problem. The ion trap computers supports two-qubit gate operation Mølmer-Sørenson not CNOT directly. each CNOT must be decomposed into Mølmer-Sørenson and some single qubit rotation gate. This paper https://doi.org/10.1088/1367-2630/ab3372 has mentioned the decomposition of CNOT into Mølmer-Sørenson.

commented

@sassan72 nice catch! It sounds like we should add a decomposition to the CNOT gate down into Mølmer-Sørenson gates :)

commented

yes. Thanks for your quick response.

commented

@sassan72 nice catch! It sounds like we should add a decomposition to the CNOT gate down into Mølmer-Sørenson gates :)

Indeed, this seems like the best solution to prevent this issue in future.
@josh146 is this device-specific decompositoin something that would be in the plugin, or in PL-core?

commented

@josh146 is this device-specific decompositoin something that would be in the plugin, or in PL-core?

It's a good question. We have two options:

  1. Add a decomposition to qml.CNOT directly in core. Possible because CNOT has no existing decompositions defined, and likely to work well since (as far as I know) no other qubit devices do not support CNOT. Currently, ops can only have one decomposition defined, although this might change with the operation refactor.

  2. Alternatively, once the expansion is moved to the device, we could have a custom CNOT->Mølmer-Sørenson decomposition occurring here in the AQT device.

I think I favour (1); only complication is if we expect there to be other devices that do not support CNOT nor Mølmer-Sørenson.

commented

as far as i know, IONQ and Honeywell can't also support CNOT directly. because their system are ION TRAP based.

commented

Honeywell should support CNOTs, they do a compilation step for submitted programs. But I don't believe AQT does this.

@josh146 option number 2 will be the best option, since the Molmer-Sorenson gate is only native to the AQT device.

@sassan72 it should be possible to auto-decompose CNOTs into Molmer-Sorenson gates hopefully by the next release of PennyLane

commented

Hi @sassan72, @rmoyard added a fix for this to the master branch. Until we release the plugin, the feature is accessible by cloning the repository and installing the project via pip install -e .. I'll be closing this issue, but let us know if you have any more questions/updates! 🙂