UCCSD operator pool correction and integration in adapt simulator

[kvmto]

UCCSD operator pool correction and integration in adapt simulator (Issue #28)

This PR fixes #28 .

Problem

Adapt VQE did not provide correct results. The problem could be the creation of the UCCSD operators pool.

Changes

Changes so far have been applied to:

• The UCCSD operator pool 'generate' function.
• The Testing of the UCCSD operator pool, cpp and py files.
• Tested adap_vqe for final results.

Additional notes

Further testing is needed for making sure that adapt VQE works correctly, especially in terms of molecules.

[amccaskey]

Let's add a test to test_adapt.cpp that mimics the checkSimpleAdapt but uses the uccsd operator instead of the spin_complement_gsd one. Note that you'll need to update the coefficients on the operators to make them imaginary, since we assume that in adapt_simulator.cpp (line 152).

[bmhowe23]

I would recommend adding the phrase "fixes #28" or "resolves #28" to the description in order to automatically link this GitHub PR with the issue. If you do this, then when the PR is merged, the issue will automatically be closed.

Ref: https://docs.github.com/en/issues/tracking-your-work-with-issues/using-issues/linking-a-pull-request-to-an-issue#linking-a-pull-request-to-an-issue-using-a-keyword

[bmhowe23]

Let's add a test to test_adapt.cpp that mimics the checkSimpleAdapt but uses the uccsd operator instead of the spin_complement_gsd one. Note that you'll need to update the coefficients on the operators to make them imaginary, since we assume that in adapt_simulator.cpp (line 152).

I think I'm missing something ... why we are updating the generic UCCSD operator pools from real to imaginary to accommodate a single assumption in adapt_simulator.cpp? Are any docs updates needed for this change?

[marwafar]

The uccsd operator pool is correct. However, it might not be a good idea to multiply the imaginary number "i" by the uccsd operator pool. It will not make it generic. We might use this uccsd operator pool with other algorithm such as GQE and we will need to modify this again. I would suggest multiply the imaginary number "i" somewhere within the adapt-vqe algorithm.

In addition, I would suggest to run the adapt-vqe for these two geometry with sto3g basis.

H 0.0 0.0 0.0
H 0.0 0.0 0.7474

Li  0.0  0.0  0.0
H   0.0  0.0  1.5 

We need to make sure that it work for something beyond the simple H2 molecule with 4 qubits and 3 operator pools.

[amccaskey]

instead of making this copy, could you not retain the code that was there and just update the commutator creation to

std::complex<double> j = 1.; 
if (isImaginary) j = std::complex<double>{0.0, 1.0};
commutators.push_back(H * j * pool[i] - j * pool[i] * H); 

[kvmto]

In line 159 (auto op = pool[maxOpIdx];), the code selects the best operator and manipulates it. We could assume that each operator in the pool contains either only real or only imaginary coefficients, in which case we would only need to perform the check on the first item. Alternatively, if we can’t make that assumption, we would need to store the result of the check for each operator and provide it in line 159. The copy is just a proposal to accommodate different assumptions. I would be happy with either.

[amccaskey]

aren't these changes from Anna's PR? do you need her changes duplicated here?

[kvmto]

I don't need the duplication.
I just fetched and rebased, not sure how it ended up in here... I am not sure how changes happened in between.
I can reset the commit if needed.

[bmhowe23]

It looks like the formatter went wild on this one. Did you accidentally run clang-format on this file?