Initial age deviation penalties — get_init_devs

Population and region specific penalties on the initial age deviations (ln_InitDevs), plus the tie holding each later sex's curve near the first sex's. Called from get_recruitment_penalty.

Usage

get_init_devs_penalty(
  n_pop,
  n_regions,
  n_ages,
  rec_region_prop_spec,
  rec_region_prop,
  equil_init_age_strc,
  ln_InitDevs,
  init_age_devs_shared,
  ln_sigmaR,
  bias_ramp,
  InitDevs_pen_center = 0,
  init_devs_pen_use = NULL,
  Use_init_sex_pen = 0,
  ln_sigma_init_sex = 0,
  init_bias_ramp = NULL,
  map_ln_InitDevs = NULL,
  init_sigmaR_dsem = NULL,
  init_sigmaR_dsem_use = NULL,
  RecDevs_model = 1,
  RecDevs_rho = NULL
)

Arguments

n_pop, n_regions, n_ages: Dimension sizes.
rec_region_prop_spec: Integer switch; when 1, populations and regions with a fixed zero recruitment proportion are skipped.
rec_region_prop: Array [pop, region] of recruitment regional apportionment.
equil_init_age_strc: Integer switch naming which initial age deviations are penalized (0: none, 1: all but the plus group, 2: all, 3: the shared subset).
ln_InitDevs: Array [pop, region, age, sex] of initial age deviations. A 3-D [pop, region, age] array, the layout before the sex dim existed, is accepted and treated as one shared curve.
init_age_devs_shared: Integer vector of shared initial age deviation indices, read when equil_init_age_strc == 3.
ln_sigmaR: Array [early/late, pop, region] of log sigma. Initial ages read the early one.
bias_ramp: Numeric vector [year] of bias ramp adjustment factors.
InitDevs_pen_center: Integer. 1 centers on the deviations' own weighted mean, 0 on the bias-corrected mean.
init_devs_pen_use: Array of 0/1 matching ln_InitDevs, naming which cells are penalized. Sexes sharing one parameter keep only the first sex's copy flagged so the shared parameter is not penalized twice; sex-specific deviations flag every sex. NULL penalizes only the first sex's slice, which is the pre-sex-dim behavior.
Use_init_sex_pen: Integer (0/1). Whether each later sex's deviations are tied to the first sex's through a Gaussian on their difference at every penalized age. Only meaningful when the sexes have their own curves.
ln_sigma_init_sex: Log standard deviation of that tie.
init_bias_ramp: Numeric vector of length n_ages - 1, the bias ramp read at the year each initial age was born (deviation index 1 - age). NULL reads the first model year's ramp value at every age.
map_ln_InitDevs: Numeric array matching ln_InitDevs of map levels (NA where fixed). Cells sharing a level hold one parameter and split one penalty between them. NULL penalizes every cell in full.
init_sigmaR_dsem, init_sigmaR_dsem_use: Array [pop, region] of the settled marginal sd of the recruitment series under the arrows, and 1 where a cell reads it in place of ln_sigmaR. NULL without a dsem.
RecDevs_model, RecDevs_rho: Recruitment process error form and its unconstrained AR1 correlation. Under 3 the initial ages read the stationary sd \(\sigma_R / \sqrt{1 - \rho^2}\).

Value

List with Init_Rec_nLL (array [pop, region, age, sex]) and Init_Sex_nLL (the same layout, the between-sex tie, zero for the first sex and whenever the tie is off), each holding negative log-likelihood penalties and zero where nothing is penalized.

Details

Each penalized deviation is Gaussian on the log scale with the early recruitment sigma, \(-\log \phi(d_{a,s} \mid \mu, \sigma_{R,1})\), where

\(d_{a,s}\) is the deviation at age \(a\) and sex \(s\), log scale, estimated.
\(\sigma_{R,1} = \exp(\code{ln_sigmaR[1,p,r]})\) is the early recruitment sigma, log scale.
\(\mu\) is the center, either the bias-corrected mean \(-\sigma_{R,1}^2 b_a / 2\) with \(b_a\) the bias ramp read at the year age \(a\) was born, or the deviations' own weighted mean pooled over ages and sexes.