GLODS_vs_GLODSSI

GLODS vs GLODS-SI (pairwise comparison)

This folder contains the complete results of the head-to-head comparison between GLODS (original) and GLODS-SI (scale-invariant). It is the source of the GLODS and GLODS-SI columns of Table 3 in the accompanying paper:

J. F. A. Madeira, GLODS-SI: Scale-Invariant Global-Local Direct Search for Engineering Design Optimization, Journal of Computational Design and Engineering, 2026. Manuscript ID JCDE-2026-065.

Data profiles

For each of the eight scaling strategies, a PDF data profile compares the two algorithms over a budget of 20,000 function evaluations:

• Profile_GLODS_vs_GLODSSI_baseline_BEST.pdf (κ = 1)
• Profile_GLODS_vs_GLODSSI_moderate_BEST.pdf (κ = 10⁴)
• Profile_GLODS_vs_GLODSSI_progressive_BEST.pdf (κ = 10⁶)
• Profile_GLODS_vs_GLODSSI_extreme_BEST.pdf (κ = 10⁸)
• Profile_GLODS_vs_GLODSSI_sobol_oscillatory_BEST.pdf (κ = 10⁶)
• Profile_GLODS_vs_GLODSSI_sobol_digit_oscillatory_BEST.pdf (κ = 10⁸)
• Profile_GLODS_vs_GLODSSI_halton_oscillatory_BEST.pdf (κ = 10⁶)
• Profile_GLODS_vs_GLODSSI_spatial_thermal_BEST.pdf (κ ≈ 9 × 10⁴)

Summary tables

ASCII summary file per strategy, reporting success rates at selected evaluation budgets:

• Summary_GLODS_vs_GLODSSI_<strategy>_BEST.txt (eight files)

A problem p is considered solved at evaluation k for algorithm a when

H(k, p, a) <= f_min(p) + tau * (f_0(p) - f_min(p)),     tau = 1e-5,

with f_min(p) taken as the minimum across the two algorithms in this pairwise comparison.

LaTeX table

• latex_table_GLODS_vs_GLODSSI.tex

Auto-generated LaTeX source containing two tables: final success rates across all eight strategies, and success rates at three reference budgets (500, 2000, 20000 function evaluations). The first of these tables is the source of the GLODS and GLODS-SI columns of Table 3 in the paper.

Reproducibility

All files were generated by the script profile_GLODS_vs_GLODS_SI_all_strategies.m (located in the local working directory used to produce the paper) from the raw optimization runs in results_GLODS_*_BEST/ and results_GLODS_SI_*_BEST/. The pipeline is deterministic: rerunning with the same benchmark suite and the default parameters listed in parameters_glods_si.m reproduces these files exactly.