Hyperparameters

Hyperparameters are used to facilitate structure optimization. The values do not affect the outcome but control the runtime and accelerate convergence to the target values. Here is a complete list of hyperparameters. The usage can be found in example 10, 11, 12, 13. Their definition and suggested value are listed below:

Metropolis temperature

Tmetro, Tmetro_out: Greater than 0, usually smaller than 300 K.

Control the acceptance rate for inner and outer optimization loop respectively. The higher temperature, more likely a Monte Carlo move will be accepted.

Max number of iterations in each loop

i_max, i_max_out, i_max_ring: Usually ~1000 -10000

Limit the number of MC attempts and runtime of the simulations in each optimization loop.

Expected polymer size

expected_size: Greater than 0, usually smaller than 100.

We consider the polymer size from a population following a normal distribution This value sets the estimated mean size for the polymer molecule in the inner loop.

Maximum polymer size

max_size: Greater than 0, usually smaller than 100.

Set a threshold for the polymer molecule size in the inner loop.

Size distribution scaling factor

distribution_scaling: Between 0 to 1.

Controls the width of this distribution and affects the weighted average molecular weight. The larger the k, the wider the range of polymer sizes is; the weighted average molecular weight would be greater than the number average molecular weight, indicating the polymer population is polydisperse.

Branching propensity

branching_propensity: Between 0 to 1.

Determine the branching probability for newly added monomers. The value should be similar to the experimental branching coefficient.

Hyperparameters can be passed into optimization.Simulation class. Code example to define all the hyperparameters:

# set Metropolis temperatures
Tmetro = 10
Tmetro_out = 10

# set the max iterations
i_max = 1000
i_max_out = 1000
i_max_ring = 500

# set the expected and max sizes
expected_size = 7
max_size = 100
# size distribution scaling factor
distribution_scaling = 0.1

# whether the size is in Molecular weight or monomer count
size_in_MW = False

branching_propensity = 0