modified __add__ for Observable and Hamiltonian to support adding 0

[Qi-Hu]

Context:

We would like to be able to sum over a list of observables as follows:

H = sum([qml.PauliX(i) for i in range(10)])

Since sum starts at integer 0 by default, we need to define addition between observables and 0.

Description of the Change:

We modified the __add__ method for Observable and Hamiltonian classes, so that integer 0 can be added to those objects, either from the left side or the right side. Unit tests were added for this feature.

Benefits:

Now we can directly sum a list of observables in an elegant way.

Possible Drawbacks:

None.

Related GitHub Issues:

Closes #2423.

[codecov]

Codecov Report

Merging #2603 (8d9d442) into master (2b07d22) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #2603   +/-   ##
=======================================
  Coverage   99.58%   99.58%           
=======================================
  Files         248      248           
  Lines       19975    19985   +10     
=======================================
+ Hits        19892    19902   +10     
  Misses         83       83           

Impacted Files	Coverage Δ
pennylane/operation.py	96.59% <100.00%> (+0.02%)	⬆️
pennylane/ops/qubit/hamiltonian.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2b07d22...8d9d442. Read the comment docs.

[co9olguy]

Hi @Qi-Hu, let us know when the PR is ready for review, or if you have any questions 🙂

[Qi-Hu]

Hi @Qi-Hu, let us know when the PR is ready for review, or if you have any questions 🙂

Hi @co9olguy, this PR is ready for review. Thanks!

[antalszava]

Hi @Qi-Hu, the core solution is looking great overall! 🎉 🙂 Great to have support for adding 0 to observables both from the left side and the right side.

There's one subtlety that would be great to adjust and a couple of additional things to test.

1. The in-place summation (__iadd__ method) of the qml.Hamiltonian class has a different implementation to the regular addition (__add__ method).

Although not explicitly mentioned in the original issue, attempting to add 0 to a Hamiltonian object using in-place addition raises an error:

H = sum([qml.PauliX(i) for i in range(10)])
H += 0

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-14-2918b18b0578> in <module>
----> 1 H += 0

~/pennylane/pennylane/ops/qubit/hamiltonian.py in __iadd__(self, H)
    620             return self
    621 
--> 622         raise ValueError(f"Cannot add Hamiltonian and {type(H)}")
    623 
    624     def __imul__(self, a):

ValueError: Cannot add Hamiltonian and <class 'int'>

2. As for testing, it would be great to add a couple of further cases:

A. Checking two remaining qubit observables:

• SparseHamiltonian
• Projector

B. Testing the in-place addition support via the += operator for a few general observables.

C. Testing that the feature works for continuous-variable (CV) observables too. These observables would subclass CVObservable.

[Qi-Hu]

Hi @antalszava, thanks a lot for your detailed comments and useful suggestions! I have made a few changes according to your suggestions.

I have two questions:

1. The __iadd__ method is only explicitly defined for the Hamiltonian class. Do we need to test it for general observables?
2. I added more test cases for the __add__ method of Observable class, including SparseHamiltonian, Projector and CVObservable. However, there were a few test failures. They seemed to be caused by the same issue, which is related to the attribute 'tobytes'. I haven't figured out why this happened, but I will continue working on it.

pennylane/operation.py:1635: in compare
    return other._obs_data() == self._obs_data()
pennylane/operation.py:1598: in _obs_data
    parameters = tuple(param.tobytes() for param in ob.parameters)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

.0 = <list_iterator object at 0x7fc90b25a850>

>   parameters = tuple(param.tobytes() for param in ob.parameters)
E   AttributeError: 'float' object has no attribute 'tobytes'

pennylane/operation.py:1598: AttributeError

[Qi-Hu]

Hi @antalszava, I realized the compare method is not supported by a few classes: Projector, SparseHamiltonian, QuadOperator and FockStateProjector. To be more specific, their parameters do not have tobytes attribute. I do not know if this is an issue that needs to be fixed, so I just didn't include these test cases in my most recent commit.

[antalszava]

Hi @Qi-Hu, this looks good to me! 👍 It will be very useful to have this in. 🙂

Apologies for the delay here, wasn't feeling completely well and then it has been a bit busy.

Thank you for addressing the suggestions.

I realized the compare method is not supported by a few classes: Projector, SparseHamiltonian, QuadOperator and FockStateProjector. To be more specific, their parameters do not have tobytes attribute. I do not know if this is an issue that needs to be fixed, so I just didn't include these test cases in my most recent commit.

Good catch! This should be good as such indeed. 👍

[Qi-Hu]

Thanks @antalszava!, please let me know if there is anything else I could do.

[antalszava]

Sure 🙂 Merged it, thank you for this contribution! 🎉

[Qi-Hu]

Sure 🙂 Merged it, thank you for this contribution! 🎉

Glad to hear that! Thank you for your help!

[co9olguy]

🎉