search

PennyLaneAI / pennylane

PennyLane is a cross-platform Python library for quantum computing, quantum machine learning, and quantum chemistry. Train a quantum computer the same way as a neural network.
https://pennylane.ai
Apache License 2.0
2.34k stars 602 forks source link

[BUG] `single_qubit_fusion` returns wrong angles for specific case #3023

Closed dwierichs closed 2 years ago

dwierichs commented 2 years ago

Expected behavior

The tape transform single_qubit_fusion always performs valid transforms, not changing the action of the tape.

Actual behavior

For the following specific case, the transform reports wrong angles. Having tracked this down to optimization_utils.fuse_rot_angles already, I just report the minimal example here:

>>> angles1 = [np.pi, np.pi/2, 0]
>>> angles2 = [0, -np.pi/2, 0]
>>> fuse_rot_angles(angles1, angles2)
array([0., 0., 0.])

As we can see, although only the RY rotation angles should cancel out, the RZ rotation caused by the first angle of angles1 also disappeared. As far as my experiments went, this bug seems to appear for a whole range of values, but I was not able to track down for which exactly. Tested examples include:

#works
angles1 = [np.pi, np.pi/2, 0.]
angles2 = [0., -np.pi/2, 0.23]

#doesn't work
angles1 = [np.pi, np.pi/2, 0.]
angles2 = [0., -np.pi/2, 0.]

#works 
angles1 = [np.pi, np.pi/2, 0.]
angles2 = [1., -np.pi/2, 0.3]

#doesn't work at all, produces `nan` angle
angles1 = [0.9, np.pi/2, 0.]
angles2 = [0., -np.pi/2, 0.]

#works
angles1 = [0.9, np.pi/2, 1.]
angles2 = [0., -np.pi/2, 0.]

Additional information

No response

Source code

No response

Tracebacks

No response

System information

Name: PennyLane
Version: 0.26.0.dev0
Summary: PennyLane is a Python quantum machine learning library by Xanadu Inc.
Home-page: https://github.com/XanaduAI/pennylane
Author: 
Author-email: 
License: Apache License 2.0
Location: /home/david/venvs/xanadu3.10/lib/python3.10/site-packages/PennyLane-0.26.0.dev0-py3.10.egg
Requires: appdirs, autograd, autoray, cachetools, networkx, numpy, pennylane-lightning, retworkx, scipy, semantic-version, toml
Required-by: amazon-braket-pennylane-plugin, PennyLane-Cirq, PennyLane-Lightning, PennyLane-PQ, PennyLane-qiskit, pennylane-qulacs

Platform info:           Linux-5.4.0-125-generic-x86_64-with-glibc2.31
Python version:          3.10.6
Numpy version:           1.23.1
Scipy version:           1.9.0rc2
Installed devices:
- braket.aws.qubit (amazon-braket-pennylane-plugin-1.6.9)
- braket.local.qubit (amazon-braket-pennylane-plugin-1.6.9)
- lightning.qubit (PennyLane-Lightning-0.25.0)
- projectq.classical (PennyLane-PQ-0.17.0)
- projectq.ibm (PennyLane-PQ-0.17.0)
- projectq.simulator (PennyLane-PQ-0.17.0)
- qulacs.simulator (pennylane-qulacs-0.13.0.dev0)
- cirq.mixedsimulator (PennyLane-Cirq-0.25.0.dev0)
- cirq.pasqal (PennyLane-Cirq-0.25.0.dev0)
- cirq.qsim (PennyLane-Cirq-0.25.0.dev0)
- cirq.qsimh (PennyLane-Cirq-0.25.0.dev0)
- cirq.simulator (PennyLane-Cirq-0.25.0.dev0)
- qiskit.aer (PennyLane-qiskit-0.25.0.dev0)
- qiskit.basicaer (PennyLane-qiskit-0.25.0.dev0)
- qiskit.ibmq (PennyLane-qiskit-0.25.0.dev0)
- qiskit.ibmq.circuit_runner (PennyLane-qiskit-0.25.0.dev0)
- qiskit.ibmq.sampler (PennyLane-qiskit-0.25.0.dev0)
- default.gaussian (PennyLane-0.26.0.dev0)
- default.mixed (PennyLane-0.26.0.dev0)
- default.qubit (PennyLane-0.26.0.dev0)
- default.qubit.autograd (PennyLane-0.26.0.dev0)
- default.qubit.jax (PennyLane-0.26.0.dev0)
- default.qubit.tf (PennyLane-0.26.0.dev0)
- default.qubit.torch (PennyLane-0.26.0.dev0)
- default.qutrit (PennyLane-0.26.0.dev0)

Existing GitHub issues

josh146 commented 2 years ago

This seems to happen at singular points of arctan2

Oof, very nice catch!