Simulate a stochastic or conditional-Gaussian model

Generates truth-twin realisations by Euler-Maruyama integration, optionally retaining the hidden path and the driving Wiener increments so that a later ensemble smoother can reuse them.

Usage

# S3 method for class 'stochastic_model'
simulate(
  object,
  nsim = 1,
  seed = NULL,
  t_end,
  dt,
  ic = NULL,
  burn_in = 0,
  keep_hidden = TRUE,
  keep_noise = TRUE,
  ...
)

Arguments

object

A stochastic_model or cgns_model object.

nsim

Positive whole number of realisations to generate.

seed

Optional non-negative whole number seeding the generator. Seeding is contained: the caller's .Random.seed is restored when the call returns, so a reproducible path leaves the caller's stream where it was.

t_end

Positive 1-length numeric total simulated time, excluding burn-in. Called T before 0.1.0; that name is accepted with a warning until 0.2.0.

dt

Positive 1-length numeric integration step.

ic

Optional list with elements x0 and y0 giving the initial state; NULL uses the model's default initial condition.

burn_in

Non-negative 1-length numeric time discarded before recording.

keep_hidden

TRUE to retain the hidden path.

keep_noise

TRUE to retain the driving Wiener increments.

...

Must be empty; unused arguments are an error.

Value

An object of class aci_sim when nsim is one, otherwise a list of such objects. Its observed trajectory carries the model's own channel names when meta$vars$observed supplies one unique non-empty name per observed channel, as the built-in constructors do, so that aci_conditional()'s given and target can be named straight off a simulation. The names are labels: no numeric result depends on them.

See also

aci_model(), aci_conditional()