Thread-Safe C++ Trust Region Optimizers • RcppTrust

RcppTrust is a thread-safe C++ port of the trust-region optimizer in Charles J. Geyer’s CRAN package trust, built for use in nlmixr2est. It implements the exact same algorithm – same Newton / easy-easy / hard-easy / hard-hard trust-region subproblem, same termination criteria – and exposes it three ways:

trust(), a drop-in R replacement for trust::trust(), with the same arguments, defaults, and return value.
trust_solve_c(), a thread-safe C entry point (a C objective function pointer plus a small options struct, no R API calls) safe to call from parallel C++ code such as an OpenMP loop.
A header-only, positionally-indexed function-pointer table (inst/include/RcppTrust.h), so another package can call trust_solve_c() without linking against this package’s shared library – the same pattern rxode2/n1qn1/lbfgsb3c use across the nlmixr2 ecosystem.

It also includes three more trust-region optimizers built the same way, each with an R function and a thread-safe C entry point:

R function	Ported from	Needs	C entry point
`steihaug()`	basin (Rust)	gradient + Hessian or Hessian-vector products	`steihaug_solve_c()`
`bobyqa()`	Powell’s BOBYQA Fortran, via `minqa`	function values only; box bounds	`bobyqa_solve_c()`
`newuoa()`	Powell’s NEWUOA Fortran, via `minqa`	function values only	`newuoa_solve_c()`

bobyqa()/newuoa() are drop-in replacements for their minqa counterparts and give bitwise-identical results; steihaug() reproduces basin’s iterates bit for bit. See vignette("trust-region-methods").

See vignette("RcppTrust") for what’s the same as upstream trust, what’s different, and a worked example of the C interface and the registration pattern.

Note this package was generated with the help of AI (Claude/Gemini).

Installation

You can install the development version of RcppTrust from GitHub with:

# install.packages("pak")
pak::pak("nlmixr2/RcppTrust")

Example

trust() is a straight substitute for trust::trust():

library(RcppTrust)

# Rosenbrock's function, the example from ?trust::trust
objfun <- function(x) {
  f <- expression(100 * (x2 - x1^2)^2 + (1 - x1)^2)
  g1 <- D(f, "x1"); g2 <- D(f, "x2")
  h11 <- D(g1, "x1"); h12 <- D(g1, "x2"); h22 <- D(g2, "x2")
  x1 <- x[1]; x2 <- x[2]
  list(
    value = eval(f), gradient = c(eval(g1), eval(g2)),
    hessian = rbind(c(eval(h11), eval(h12)), c(eval(h12), eval(h22)))
  )
}

out <- trust(objfun, c(3, 1), 1, 5)
out[c("value", "argument", "converged", "iterations")]
#> $value
#> [1] 5.165437e-15
#> 
#> $argument
#> [1] 1 1
#> 
#> $converged
#> [1] TRUE
#> 
#> $iterations
#> [1] 21

Authors

Charles J. Geyer – original algorithm and R implementation (trust)
M. J. D. Powell – BOBYQA and NEWUOA algorithms and original Fortran 77 code
Douglas Bates, Katharine M. Mullen, John C. Nash, Ravi Varadhan – minqa, the R distribution of Powell’s code this port follows
Johan Larsson and basin contributors – basin’s Steihaug trust-region implementation
Matthew Fidler – thread-safe C++ ports (RcppTrust)