Hello! I’m working on replicating the Pennylane quantum chemistry tutorial: “Optimization of Molecular geometry”, but for water instead of the Trihydronium molecule they use. However, when I’m running the final part of the code, it seems that the gradient for the nuclear coordinates starts to diver…

Hi @ImranNasrullah and thanks for using PennyLane! Could you please run the code in this demo but just change the molecular symbols and Hamiltonian as provided in the following? Depending on your results, we go further and find the problem. symbols = ["O", "H", "H"] x = np.array([0.028, 0.054, 0.0…

Hello Soran! Thanks so much for the suggestion. I did as you said, and the code in the tutorial, this time using the smaller-active space that you suggested. The code was able to converge with a OH1 bond-radius of 0.98958 Angstroms, and an OH1-OH2 bond angle of 100 degrees. However, I’m curious fo…

Hi @ImranNasrullah . Great that the geometry converges for the small active space. Could you please increase the size of the activate space step-by-step, e.g., 2 electron and 1 orbital at a time, and see when the optimisation goes wrong? Also, please change the initial coordinates, e.g., by rotati…

Sounds good, I’ll give it a shot

Hello Soran, apologies for the delay. As you suggested, I ran my code starting from 4 electrons and 4-orbitals, increasing the # of electrons by 2 and the orbitals by 1 every time. 4 e- and 4 m.o., as well as 6 e- and 5 m.o. seem to converge just fine. But when I get to 8-electrons and 6-m.o, here a…

Hi @ImranNasrullah , could you please use the ‘pyscf’ backend and let me know the results? It will help us to find the actual reason for the diverging coordinates. def H(x): return qml.qchem.molecular_hamiltonian(symbols, x, method="pyscf")[0] Regarding the energy fluctuations, they are actuall…

Hello Soran, I tried using the ‘pyscf’ backend like you suggested, and the bond length becomes very large, while the angle becomes super small. Not sure why this is the case. I’ve attached this code for you to take a look over. Both files are pretty much the same thing, one has results with the d…

Hi @ImranNasrullah . The issue is very likely due to using the finite difference method which makes gradient computations problematic in cases where the sign of some of the Hamiltonian terms changes with small perturbations. There are a couple of possible ways to fix it. Here I mention one: You can …

Hello Soran, I tried out the code you sent me, this time with: 1). The full 8 electrons and 6 molecular orbitals. 2). All the selected double and single excitation gates that came from the ‘ADAPT-VQE’ algorithm. The great news is that the algorithm jumped immediately to the proper values, and wa…

Optimizing water's molecular geometry

PennyLane Help

ImranNasrullah February 27, 2024, 9:36pm 18

hello, actually would it have to do with the difference in calculation of the dhf method? Soran, I saw ur answer about how using the psi4 method would work, linked here:

Topic		Replies	Views
Using AutoGrad/Automatic Diffrentiation Along a Direction PennyLane Help	11	581	March 18, 2024
Geometry optimization on hydrogen with distinct spatial configurations and non-integer charge PennyLane Help	3	280	July 11, 2024
Unable to construct Hamiltonian PennyLane Help	36	3463	April 26, 2022
Gradient in differentiable Hartree-Fock demo PennyLane Help	13	150	September 9, 2024
Help in tutorial optimization of molecules PennyLane Help	3	248	June 11, 2024

Optimizing water's molecular geometry

Related topics