2. Tutorials

The following tutorials show how to set up complete optimization workflows with the package. All example scripts and input files are located in the package’s examples/ folder and are thoroughly commented.

Most tutorials illustrate each step for both the ParAMS GUI and using python scripts.

Other AMS tutorials (not using ParAMS) can be found in the Tutorials section of the main AMS documentation.

• 2.1. Introduction to parametrization
  • 2.1.1. Training data
    • 2.1.1.1. Where does training data come from?
    • 2.1.1.2. What items can I include in my training set?
    • 2.1.1.3. What types of jobs should I run?
    • 2.1.1.4. What should I consider when designing my training set?
  • 2.1.2. Loss function
    • 2.1.2.1. How do I balance my training set?
    • 2.1.2.2. How do I minimize the loss function?
  • 2.1.3. Model parameters
  • 2.1.4. Where to now?
• 2.2. Getting Started: Lennard-Jones Potential for Argon
  • 2.2.1. Lennard-Jones Parameters, Engine, and Interface
  • 2.2.2. Files
  • 2.2.3. ParAMS input
    • 2.2.3.1. Parameter interface (parameter_interface.yaml)
    • 2.2.3.2. Job Collection (job_collection.yaml)
    • 2.2.3.3. Training Set (training_set.yaml)
    • 2.2.3.4. ParAMS settings (params.conf.py)
  • 2.2.4. Run the example
  • 2.2.5. Parametrization results
    • 2.2.5.1. The best parameter values
    • 2.2.5.2. Correlation plots
    • 2.2.5.3. Error plots
    • 2.2.5.4. Parameter plots
    • 2.2.5.5. Editing and Saving Plots
    • 2.2.5.6. Predicted values
    • 2.2.5.7. Loss contributions
    • 2.2.5.8. Summary statistics
    • 2.2.5.9. All output files
    • 2.2.5.10. summary.txt
  • 2.2.6. Close the ParAMS GUI
  • 2.2.7. Appendix: Creation of the input files
  • 2.2.8. Next steps
• 2.3. Import training data (GUI)
  • 2.3.1. Choose preferred units
  • 2.3.2. Reference calculation #1: Geometry optimization of a water molecule
  • 2.3.3. Charges and forces
  • 2.3.4. Bonds and angles
  • 2.3.5. PES scans: bond scan, angle scan, or volume scan
  • 2.3.6. Energies
  • 2.3.7. Import a molecular dynamics trajectory
    • 2.3.7.1. Reference calculation #2: MD simulation of liquid Ar
    • 2.3.7.2. MD with reference method
    • 2.3.7.3. MD with fast method followed by Replay with reference method
  • 2.3.8. Import a structure and settings from AMSinput
  • 2.3.9. Import VASP or Quantum ESPRESSO calculations
• 2.4. Import training data (Python)
  • 2.4.1. Run a reference job
  • 2.4.2. Water molecule reference data
  • 2.4.3. Initialize the ResultsImporter
  • 2.4.4. Add a singlepoint calculation on the optimized geometry
  • 2.4.5. Add a geometry optimization job extracting the bond length and bond angle
  • 2.4.6. Add a trajectory
  • 2.4.7. Add a reaction energy
  • 2.4.8. Training set, validation set, and other data sets
  • 2.4.9. Exit PLAMS
  • 2.4.10. Save to disk
  • 2.4.11. More ResultsImporters
• 2.5. ReaxFF: Gaseous H₂O
  • 2.5.1. Calculate the reference data
  • 2.5.2. Import the reference data into ParAMS
  • 2.5.3. Set the parameters to optimize
  • 2.5.4. Run the ReaxFF parametrization
  • 2.5.5. Visualize the results
  • 2.5.6. ffield.ff: The force field file
  • 2.5.7. Summary of the 3 different ways
  • 2.5.8. Next steps with ReaxFF parametrization
• 2.6. ReaxFF: Adsorption on ZnS(110)
  • 2.6.1. Mix force fields from the AMS ReaxFF library
  • 2.6.2. Choose the parameters to optimize
  • 2.6.3. ZnS training and validation set
  • 2.6.4. Run the ZnS ReaxFF parametrization
  • 2.6.5. ZnS parametrization results
    • 2.6.5.1. ZnS energies
    • 2.6.5.2. ZnS volume scans
    • 2.6.5.3. H₂S bond and angle scans
    • 2.6.5.4. ZnS optimized distances
    • 2.6.5.5. ZnS forces on distorted structures
• 2.7. ReaxFF: Convert old training sets to ParAMS format
  • 2.7.1. Convert geo (and control) to job_collection.yaml
  • 2.7.2. Convert trainset.in to training_set.yaml
  • 2.7.3. Convert ffield (and params) to parameter_interface.yaml
  • 2.7.4. Using convert.py
• 2.8. ReaxFF: Training set for cobalt
  • 2.8.1. Weighting of individual entries
  • 2.8.2. General energies, Cluster models and the Co₂ dimer
  • 2.8.3. Description of crystalline phases
  • 2.8.4. Description of Co-surfaces
  • 2.8.5. Adatoms
  • 2.8.6. Vacancies and defects
  • 2.8.7. Elastic strain moduli
• 2.9. GFN1-xTB: Lithium fluoride
  • 2.9.1. Energy-volume scan reference calculation for LiF
  • 2.9.2. Import the energy-volume scan into ParAMS
  • 2.9.3. Import experimental formation enthalpy into ParAMS
  • 2.9.4. Import experimental F₂ bond length
  • 2.9.5. Job-dependent engine settings (k-space)
  • 2.9.6. Set parameters to optimize and their ranges
  • 2.9.7. Set the optimizer settings
  • 2.9.8. Run the xTB parametrization
  • 2.9.9. Results of the xTB parametrization
• 2.10. Training and validation sets
  • 2.10.1. Explicit validation set (validation_set.yaml)
    • 2.10.1.1. Validation set settings
    • 2.10.1.2. Run the optimization
    • 2.10.1.3. Training and validation set results
  • 2.10.2. Scripting: Random split of a dataset into training and validation sets
• 2.11. Restarting (continuing) an optimization
  • 2.11.1. Restarting with the CMAOptimizer
    • 2.11.1.1. Run the first 100 evaluations
    • 2.11.1.2. Continue for another 100 iterations with a restart file
    • 2.11.1.3. Compare to an uninterrupted run of 200 iterations
  • 2.11.2. Restart with other optimizers (e.g. Nelder-Mead from Scipy)
• 2.12. Calculate reference values with ParAMS
  • 2.12.1. Prerequisites
  • 2.12.2. The input files
    • 2.12.2.1. Training set without reference values (training_set.yaml)
    • 2.12.2.2. Jobs and Engines (job_collection.yaml, job_collection_engines.yaml)
  • 2.12.3. Calculate the reference values
  • 2.12.4. Output files for reference calculations and data
    • 2.12.4.1. The reference.cache folder
    • 2.12.4.2. The training_set.ref.yaml file
• 2.13. The params Python library
  • 2.13.1. Run an optimization with the Python library
  • 2.13.2. Calculate reference values
  • 2.13.3. Training and validation sets with the Python library
• 2.14. DFTB repulsive potential
  • 2.14.1. job_collection.yaml and training_set.yaml
  • 2.14.2. Background: znorg-0-1
  • 2.14.3. Parameter interface
  • 2.14.4. Run the parametrization
  • 2.14.5. Results
  • 2.14.6. Modify the analytical repulsive potential