Fixed-Point SCF Models

Fixed-point SCF models solve a self-consistent electrostatic density problem. They build a local density contribution, compute long-range field features from a trial density, update the density, and repeat until convergence or until the configured SCF step limit is reached.

Pages

Its reccomended to read these in order:

• Fixed-Point SCF Model
  • Core Idea
  • Why A Fixed Point?
  • Charge Control And Fermi Level
  • Architecture Overview
  • Electrostatic Features
  • Energy Terms
  • Training Data Requirements
  • Training And Inference Mismatch
  • Use Cases
  • Main Outputs
• Using Fixed-Point SCF Models
  • The SCF Loop
  • Convergence
  • SCF Settings
  • Practical Starting Options
  • Diagnosing A Difficult SCF Run
  • Restarts
• Training Fixed-Point SCF Models
  • Architecture Hyperparameters
  • Training Method
  • Training Modes
  • Losses
• Implicit Differentiation For Fixed-Point SCF
  • Implicit Differentiation
  • Constant Total Charge
  • Linearize and Solve
• Evaluating Fixed-Point SCF Models
  • SCF Evaluation
  • Direct Fixed-Point Evaluation
  • Boundary Conditions
  • Output Keys
  • Notes
• Fixed-Point SCF ASE Calculator
  • Class
  • Minimal Example
  • Constructor Arguments
  • SCF Options
  • Input Keys
  • Arguments You Usually Do Not Need
  • Constant Charge And Constant Fermi Level
  • Results
  • Developer Restart Options