There are some reasons that people chose to train on ab-inition MD, I will list three that I think are really important:
Ab-initio MD provides a high level of accuracy by directly solving the electronic Schrödinger equation using DFT. This allows for precise calculations of energies, forces, and stress tensors. By doing that, you can create potentials that has DFT or close to DFT precision, and you can get the potentials to use larger cells in your calculations that can be hard to handle by DFT.
Flexibility is also important. Unlike classical force fields, which are typically restricted to specific material classes, ab-initio MD can be applied to all types of materials individually and in any combination.
Reactivity is also important, that is why DFT is famous in chemistry community. Ab-initio MD potentials are reactive force fields, capturing chemical reactions and bond-breaking/bond-forming events.
About the temperature, when generating potentials using ab-initio MD, you need to simulate the system at a specific temperature, as you mentioned. The choice of temperature depends on the material and the physical conditions you want to model. Commonly used temperatures include room temperature (around 300 K) or higher temperatures relevant to specific applications. For example, if you’re interested in studying phase transitions, melting behavior, or thermal properties, you might choose temperatures above room temperature. Usually if you want to study a specific phase of a material, you need to do the simulations in the temperature that the material is stable.
Remember that ab-initio MD simulations involve solving the equations of motion for atoms using quantum mechanical forces obtained from DFT calculations.
Using ab initio molecular dynamics (AIMD) simulations based on density functional theory (DFT) for generating forces and final atomic coordinates is a common approach in the development of Embedded Atom Method (EAM) potentials for several reasons:
Accuracy: AIMD simulations provide a highly accurate description of the electronic structure and atomic interactions based on quantum mechanics. DFT, in particular, is known for its balance of accuracy and computational efficiency, making it suitable for simulating the electronic properties of materials.
Transferability: EAM potentials aim to capture the underlying atomic interactions in a material by considering the effective interactions between atoms. AIMD simulations can provide valuable insights into the nature of these interactions, helping to ensure that the resulting EAM potential is transferable and applicable across different material configurations and conditions.
Data Generation: AIMD simulations produce large amounts of data, including atomic positions, forces, energies, and other properties, which can be used as input for training the EAM potential. This data allows for the fitting of the potential to accurately reproduce the structural and energetic properties observed in the AIMD simulations.
Validation: By comparing the results of AIMD simulations with experimental data and theoretical predictions, researchers can validate the accuracy of the DFT calculations and gain confidence in the reliability of the generated dataset. This validation step is crucial for ensuring that the resulting EAM potential accurately reproduces the properties of interest.
Flexibility: AIMD simulations can capture a wide range of phenomena, including structural changes, phase transitions, and dynamic processes, providing comprehensive insights into the behavior of materials under different conditions. This flexibility allows researchers to explore various aspects of material behavior and tailor the EAM potential accordingly.
Overall, leveraging AIMD simulations in DFT for generating forces and final atomic coordinates provides a robust and reliable approach for developing EAM potentials that accurately capture the complex atomic interactions in materials.