Adaptive circuits for quantum chemistry

Demos
1

Hi,

In the tutorial the final code is:

gates_select = doubles_select + singles_select
opt_state = opt.init(params)

for n in range(10):
    t1 = time.time()
    gradient = jax.grad(cost_fn, argnums=0)(params, excitations=doubles_select)
    updates, opt_state = opt.update(gradient, opt_state)
    params = optax.apply_updates(params, updates)
    energy = cost_fn(params, doubles_select)
    t2 = time.time()
    print("n = {:},  E = {:.8f} H, t = {:.2f} s".format(n, energy, t2 - t1))

However, I believe the gradient should use “excitations=gates_select” in order to use all the selected gates:

gradient = jax.grad(cost_fn, argnums=0)(params, excitations=gates_select)

Nonetheless, I tested both and I don’t get a significant difference in the results. Have anyone any idea why?

Thanks!

2

Hi @gnusci ,

I think you’re right and in fact it should be gates_select instead of doubles_select. Thanks for pointing this out!

Would you like to open an issue in the PennyLane qml repo? This is where we host the code for the demos. You can also open a PR to make the fix if you want!

If you prefer that someone else does this there’s no problem, just let us know.

In this particular case the single excitations don’t contribute to lowering the ground state. I used qml.draw_mpl() to draw the circuit when using the full set of gates (gates_select). As you can see the single excitations all have 0.0 as the parameter, which is the same as not applying the gate at all. So in practice only a few gates contribute to the circuit.

So just to clarify, we should indeed use gates_select instead of doubles_select, but the reason why the energy doesn’t really change in this case is because for this particular case those single excitations end up having 0.0 as the parameter.

image
image1420×1120 29.9 KB

I hope this helps!

And let us know if you want to open the issue and the PR :smiley: .

3

Awesome answer! Thanks a lot for the clarification.

I will try to open the issue and the PR! First time so I hope everything goes well :slight_smile:

Thanks again!

4

That’s great! Feel free to share the link to the issue and Pull Request here once you have them. And feel free to ask any questions you have along the way.