Geometry Optimizations

YASS provides geometry optimization capabilities through Optim.jl. This section explains how to optimize molecular structures and crystal cells.

Basic Usage

The simplest way to optimize a molecular structure is:

using Optim
using YetAnotherSimulationSuite

# Read initial structure
molecule = readSystem("water.xyz")

# Run geometry optimization
optimized = opt(TIP4Pf(), LBFGS(), molecule)

# Save optimized structure
write("optimized.xyz", optimized)

Optimization Algorithms

YASS optimizations can be done with any optimizer in Optim.jl

using Optim
using YetAnotherSimulationSuite

# Using algorithms directly from Optim
opt(calc, LBFGS(), molecule)
opt(calc, ConjugateGradient(), molecule)
opt(calc, OACCEL(), molecule)

Configuring Optimizations

The opt function accepts several keyword arguments to control the optimization:

optimized = opt(
    TIP4Pf(),           # Calculator
    LBFGS(),            # Algorithm
    molecule;           # Structure
    f_abstol=1e-8,      # Function value tolerance
    g_abstol=1e-5,      # Gradient tolerance  
    iterations=100_000, # Maximum iterations
    show_trace=true     # Show progress
)

Common optimization parameters:

• f_abstol: Tolerance for changes in energy (default: 0.0)
• g_abstol: Tolerance for forces/gradients (default: 1e-8)
• iterations: Maximum optimization steps (default: 1000)
• show_trace: Display optimization progress (default: false)

Periodic Systems

For periodic systems, use a Cell object:

# Read periodic structure
crystal = readSystem("crystal.xyz")

# Optimize atomic positions only
optimized = opt(calc, LBFGS(), crystal)

# Optimize cell parameters (lattice vectors)
optimized = optCell(calc, LBFGS(), crystal)

Constraints

You can apply constraints during optimization:

# Fix specific atoms (by index)
fixed = FixedAtoms([1,2,3])
calc = Calculator(TIP4Pf(); constraints=[fixed])

# Run constrained optimization
optimized = opt(calc, LBFGS(), molecule)

Convergence Monitoring

To monitor optimization progress:

optimized = opt(
    calc, LBFGS(), molecule;
    show_trace=true,
    extended_trace=true,  # Show detailed info
    trace_simplex=true    # For Nelder-Mead
)

Advanced Usage

For more control over the optimization:

# Custom convergence criteria
optimized = opt(
    calc, LBFGS(), molecule;
    x_tol=1e-6,        # Position tolerance
    f_calls_limit=1000, # Max energy evaluations
    g_calls_limit=1000  # Max gradient evaluations
)

# Use different line search method
optimized = opt(
    calc, LBFGS(), molecule;
    linesearch=LineSearches.BackTracking()
)

For additional options and algorithms, refer to the Optim.jl documentation.