`NaN` summary when fitting simulated `toa` models with `sigma.toa` trap-extended

[JosephReps]

Here we are simulating data with:

• intercept (sigma.toa) = 0.001
• trap_typesony(sigma.toa) = 5

If we simulate and try to fit:

# 10 detectors, arranged in 2 rows of 5, 100m spacing
detectors <- data.frame(x = rep(seq(-200, 200, length.out=5), 2), y = c(rep(100, 5), rep(-100, 5)))
# First 5 traps (first row), sony brand, second 5 traps (second row), panasonic
trap_cov <- data.frame(trap = 1:10, trap_type = c(rep("sony", 5), rep("panasonic", 5)))

set.seed(12345)
trap_extended_data <- sim.capt(
    detfn="hn",
    param = list(D = 2, sigma = 100, g0 = 1, sigma.toa = log(c(0.001, 5000))),
    traps = detectors,
    control.mask = list(buffer=500),
    trap.cov = trap_cov,
    loc.cov = loc_cov,
    model = list(sigma.toa = ~trap_type)
)

trap_ext.acre <- read.acre(
  trap_extended_data$capt,
  detectors,
  control.mask = list(buffer=500),
  trap.cov = trap_cov
)

trap_ext.fit <- fit.acre(
  trap_ext.acre,
  model = list(sigma.toa =~trap_type)
)

We get:

> summary(trap_ext.fit)
Detection function: Halfnormal 
Number of sessions: 1 
Information types: Times of arrival
Confidence interval method: Wald 

 
 Parameters: 
                          Estimate Std. Error 2.5 % 97.5 %
g0                        1.000000        NaN    NA     NA
sigma                   103.722394        NaN   NaN    NaN
sigma.toa.(Intercept)     0.406755        NaN   NaN    NaN
sigma.toa.trap_typesony   0.348817        NaN   NaN    NaN
D                         1.805549        NaN   NaN    NaN

 
 Extended parameters link functions: 
sigma.toa : log 

Really weird, sometimes we get NaN gradient, and infinity likelihood, other we get like below:

obj <- TMB::MakeADFun(data = data, parameters = parameters, map = map, silent = (!tracing), DLL="acre")
Browse[2]> obj$par
          g0        sigma    sigma_toa    sigma_toa            D 
 2.944438979  5.123466964  0.039068762  0.000000000 -0.058589736 
Browse[4]> obj$fn()
[1] 207.2983
Browse[4]> obj$gr()
outer mgc:  175.79967 
          [,1]     [,2]       [,3]       [,4]          [,5]
[1,] 1.0587695 41.75982 -175.79967 -166.77821 5.0492943e-13

[JosephReps]

The issue is with higher sigma.toa values not making sense in the context of this buffer / mask size I think.