GALAHAD: Geometry-Adaptive Lyapunov-Assured Hybrid Optimizer with
Softplus Reparameterization and Trust-Region Control
Implements the GALAHAD algorithm (Geometry-Adaptive
Lyapunov-Assured Hybrid Optimizer), updated in version 2 to replace the
hard-clamp positivity constraint of v1 with a numerically smooth softplus
reparameterization, add rho-based trust-region adaptation (actual vs.
predicted objective reduction), extend convergence detection to include
both absolute and relative function-stall criteria, and enrich the
per-iteration history with Armijo backtrack counts and trust-region
quality ratios. Parameters constrained to be positive (rates,
concentrations, scale parameters) are handled in a transformed z-space
via the softplus map so that gradients remain well-defined at the
constraint boundary. A two-partition API (positive / euclidean) replaces
the three-way T/P/E partition of v1; the legacy form is still accepted
for backwards compatibility. Designed for biological modeling problems
(germination, dose-response, prion RT-QuIC, survival) where rates,
concentrations, and unconstrained coefficients coexist. Developed at
the Minnesota Center for Prion Research and Outreach (MNPRO), University
of Minnesota. Based on Conn et al. (2000)
<doi:10.1137/1.9780898719857>, Barzilai and Borwein (1988)
<doi:10.1093/imanum/8.1.141>, Xu and An (2024)
<doi:10.48550/arXiv.2409.14383>, Polyak (1969)
<doi:10.1016/0041-5553(69)90035-4>, Nocedal and Wright (2006,
ISBN:978-0-387-30303-1), and Dugas et al. (2009)
<https://www.jmlr.org/papers/v10/dugas09a.html>.
Documentation:
Downloads:
Reverse dependencies:
Linking:
Please use the canonical form
https://CRAN.R-project.org/package=GALAHAD
to link to this page.
These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
[aut, cre]