First of all, I would like to express my sincere gratitude for developing such a powerful and versatile tool as GULP. Your work has significantly contributed to the materials community, and I deeply respect your contributions.
I am a beginner with GULP and currently working on fitting a Buckingham potential for Si-doped beta-Gallium Oxide. Since no existing potentials are available for this system, I have relied on initial guesses sourced from Scott Woodley’s interatomic potential library for the Ga-O, Si-O, and O-O pairs.
I am using energy data calculated from DFT (using VASP) for a structure where one Ga atom is substituted by Si at an octahedral position in a 6-unit cell supercell. I would like to note that the structure is not charge neutral due to this substitution. My goal is to derive a potential that can at least reproduce the lattice parameters accurately. Initially, I attempted to fit only the parameter
A, and then tried varying both A and ρ. However, in each case, the final sum of squares and gradient norms remained significantly high.
As an example, I have attached my input file along with two output files for your reference. I have experimented with several initial guesses and fitting combinations, but the results have not been satisfactory so far.
I understand that you have a busy schedule, but if you have any time to take a look and could provide me with some guidance or suggestions, it would be greatly appreciated.
Thank you very much for your time and for all the valuable resources you have provided to the community.
Best Regards
Shaon
Here is the google drive link with all the input and output:
Dear Shaon,
The sum of squares is likely to stay high since a force field model might not track the DFT energy values for everything exactly (if you want exact energy reproduction, rather than understanding, then machine learning would be the way to go). It would be better to fit a model to Ga2O3 first using the structure, elastic properties etc & then just fit the Si-O potential to the defect energy using the “reaction” option. You could even fit the Si-O potential first to stishovite to get something for the octahedral environment and then just see what the defect energy is.
Regards,
Julian
Thank you for the reply and suggestion. I am thinking to try the following steps:
Step 1: Fit the Ga₂O₃ Potential
a. Generate Ga₂O₃ Structures:
Begin with the undoped Ga₂O₃ structure, generating different configurations if necessary to explore various bonding and lattice arrangements.
b. Optimize the Ga₂O₃ Structure with DFT:
Run DFT calculations to obtain key properties such as lattice parameters, elastic constants, and cohesive energy for these pristine Ga₂O₃ structures.
c. Fit Ga-O and O-O Interactions Using GULP:
Use the DFT results to fit Ga-O and O-O potentials (such as Buckingham or Morse). Ensure the potential accurately reproduces the bulk properties of Ga₂O₃ before proceeding.
Step 2: Fit the Si-O Potential Using Defect Energy
a. Generate Si-Doped Ga₂O₃ Structure:
Use the existing DFT-optimized Si-doped Ga₂O₃ structure, where one Ga atom is substituted by Si, preferably in the octahedral site.
b. Calculate Defect Energy Using DFT:
Determine the energy difference between the pristine Ga₂O₃ structure and the Si-doped Ga₂O₃ structure, which represents the defect formation energy.
c. Fit the Si-O Potential:
Using GULP’s “reaction” option, adjust the Si-O potential so that the defect formation energy matches the DFT-calculated values, while keeping Ga-O and O-O potentials fixed.
Step 3: Fine-Tune and Validate the Full Potential
a. Validate the Combined Potential:
Combine the fitted Si-O, Ga-O, and O-O potentials and test how well they reproduce key properties of Si-doped Ga₂O₃, including lattice parameters and defect formation energies.
Please let me know if my understanding and approach are correct, or if there are any adjustments I should consider.
Thank you again for your valuable time and guidance.
Hi Shaon,
It would be worth running this by whoever is the academic lead on the project to make sure they are happy with the approach, since it’s not my role to define the strategy for your work. Broadly everything sounds OK, except that you need to think carefully about how you define the defect energy. If you just add/remove Si in an ionic model then your reference state would be Si4+, which is not necessarily sensible. Better to think in terms of reacting neutral solids together for the defect chemistry.
Regards,
Julian