Model configuration

Model configuration#

When initialising the fragmentmnp.FragmentMNP model class, a config dict must be provided. Examples are given in the fragmentmnp.examples module:

from fragmentmnp.examples import minimal_config, full_config
minimal_config

{'n_size_classes': 7, 'particle_size_range': [-9, -3], 'n_timesteps': 100}

full_config

{'n_size_classes': 7,
 'particle_size_range': [-9, -3],
 'n_timesteps': 100,
 'dt': 1,
 'solver_method': 'RK45',
 'solver_rtol': 0.001,
 'solver_atol': 1e-06,
 'solver_max_step': inf,
 'solver_t_eval': 'timesteps'}

The minimal_config contains only required variables, whilst full_config includes variables that have defaults. Here we take a look at the schema for this config dict.

n_size_classes: Required, int, less than or equal to 100.; The number of bins the model should use for particle size distributions.
particle_size_classes: Optional, list of int or floats of length n_size_classes, units: m.; The mean diameter of each particle size class. If particle_size_classes is not specified, then particle_size_range will be used to calculate this diameters instead. Note that whilst particle_size_range requires exponents of 10 to be provided, particle_size_classes requires absolute values.
particle_size_range: Optional, list/tuple of int or floats of length 2, units: m.; The particle size range that should be included in the model in the format [min, max]. The values given are used as the exponent of 10, so, for example, [-9, -3] results in a size range of 10e-9 m to 10e-3 m (1 nm to 1 mm). The size class bins themselves will be deduced by evenly splitting the range by n_size_classes on a log scale. If neither particle_size_range nor particle_size_classes is specified, you will get an error.
n_timesteps: Required, int.; The number of timesteps to run the model for.
dt: Optional, int, units: s, default: 1.; The length of each timestep. Defaults to 1 second. The model time grid uses the midpoint of these timesteps, and therefore is calculated as t_grid = np.arange(0.5*dt, n_timesteps*dt, 0.5*dt).

solver_method: Optional, str equal to a valid method, default: “RK45”.; The method that is used to numerically solve the model differential equation. This method is passed directly to scipy.intergrate.solve_ivp and must be a valid method available in SciPy (“RK45”, “RK23”, “DOP853”, “Radau”, “BDF” or “LSODA”). By default, LSODA is used. See the SciPy documentation for more details. If you are having numerical stability issues, try “Radau”, “BDF” or “LSODA”, and read Dealing with numerical instability.

solver_rtol, solver_atol: Optional, float or list of floats for atol, defaults: rtol=1e-3, atol=1e-6; Relative and absolute tolerances passed to scipy.integrate.solve_ivp. The solver keeps the local error estimates less than solver_atol + solver_rtol * abs(y). For solver_atol, a list of length n_size_classes can be given, such that a different absolute tolerance is used for each size class. Setting these tolerances might be particularly useful for stiff problems that cause numerical instability.

solver_max_step: Optional, float, default: np.inf.; Set a maximum step size for the ODE solver. This might be particularly useful for stiff problems that cause numerical instability.

solver_t_eval: Optional, str equal to “timesteps”, None, or list of integers or floats, default: “timesteps”.; Time points at which to store the computed solution, passed to scipy.integrate.solve_ivp. If the string “timesteps” is provided, the model stores the model timesteps deduced from the n_timestep and dt config options, i.e. solver_t_eval = np.arange(0, n_timesteps, dt). If None, the points selected by the solver are used (which might be controlled by the solver_max_step option). If a list of integers or floats is provided, these are used directly as the time points. An error will be thrown if these lie outside of the model time span.