Geometry Optimizations

JMD uses Optim.jl for optimizations. You can optimize the geometry of molecular systems just using JMD.

using JMD

# Read in bdys
bdys = readSystem("./myFile.xyz")

# Run a geometry optimization
new  = opt(TIP4Pf(), JMD.LBFGS(), bdys)

The optimization algorithms from Optim.jl are not re-exported by JMD.jl, for this reason you must either use JMD.LBFGS() (as shown above) or load Optim.jl within your script (as shown below). Note, for this to work you need to have already installed Optim.jl.

using JMD
using Optim

# Read in bdys
bdys = readSystem("./myFile.xyz")

# Run a geometry optimization
new  = opt(TIP4Pf(), LBFGS(), bdys)

Configuerable options for Optim.jl optimizations can be passed as keyword arguments to the opt function. For a full list of algorithms and configuerable options please see the Optim.jl documentation. Here we only give a few key optional arguments that are typically used in geometry optimizations.

• f_abstol: Absolute tolerance in changes of the objective value. Defaults to 0.0.
• g_abstol: Absolute tolerance in the gradient, in infinity norm. Defaults to 1e-8. For gradient free methods, this will control the main convergence tolerance, which is solver specific.
• iterations: How many iterations will run before the algorithm gives up? Defaults to 1_000.
• show_trace: Should a trace of the optimization algorithm's state be shown on stdout? Defaults to false.

An example of passing these optional arguments to opt is given below.

# You can also replace the ; with , but Julia convention is ;
new  = opt(
  TIP4Pf(), LBFGS(), bdys; f_abstol=1e-8, g_abstol=1e-5, 
  iterations=1000000, show_trace=true 
)