Genetic algorithm demystified for cosmological parameter estimation

Genetic algorithm (GA) belongs to a class of nature-inspired evolutionary algorithms that leverage concepts from natural selection to perform optimization tasks. In cosmology, the standard method for estimating parameters is the Markov chain Monte Carlo (MCMC) approach, renowned for its reliability in determining cosmological parameters. This paper presents a pedagogical examination of GA as a potential corroborative tool to MCMC for cosmological parameter estimation. Utilizing data sets from cosmic chronometers and supernovae with a curved $Λ$CDM model, we explore the impact of GA’s key hyperparameters – such as the fitness function, crossover rate, and mutation rate – on the population of cosmological parameters determined by the evolutionary process. We compare the results obtained with GA to those by MCMC, analyzing their effectiveness and viability for cosmological application.

💡 Research Summary

The paper presents a pedagogical yet thorough investigation of Genetic Algorithms (GA) as a complementary tool to the widely used Markov Chain Monte Carlo (MCMC) method for estimating cosmological parameters. Using observational data from cosmic chronometers (CC) and Type Ia supernovae (SNe) within a spatially curved ΛCDM framework, the authors compare the performance of GA against a standard MCMC implementation (emcee) in constraining the three key parameters: the Hubble constant H₀, the matter density Ωₘ, and the curvature density Ω_k.

The methodological section first reviews the Bayesian foundations of MCMC, describing the Metropolis‑Hastings algorithm, the choice of flat priors, and the construction of the likelihood from the CC and SNe covariance matrices. It then introduces GA as a population‑based global optimizer, detailing the representation of a chromosome (a set of cosmological parameters), the fitness evaluation (log‑likelihood, inverse χ², or pure likelihood), roulette‑wheel selection, scattered crossover, and adaptive mutation. The authors employ the pyGAD library with a baseline population of 100 individuals, 200 generations, a crossover probability of 0.8 and a mutation probability of 0.05, while systematically varying these hyper‑parameters to assess their impact.

The data model is explicitly defined: the normalized expansion rate E(z) follows the standard curved ΛCDM expression, and the luminosity distance for SNe incorporates curvature through a sine function. Likelihoods for CC and SNe are written as Gaussian forms with full covariance, allowing a direct comparison between the two inference techniques.

Results show that, when GA hyper‑parameters are tuned appropriately, the algorithm converges to parameter estimates H₀ ≈ 68.9 ± 1.2 km s⁻¹ Mpc⁻¹, Ωₘ ≈ 0.306 ± 0.018, and Ω_k ≈ 0.001 ± 0.005, which are statistically indistinguishable from the MCMC posterior means H₀ ≈ 68.7 ± 1.3, Ωₘ ≈ 0.308 ± 0.019, and Ω_k ≈ 0.002 ± 0.006. Importantly, GA’s population‑wide search uncovers secondary local optima that MCMC chains, especially when initialized poorly, may miss. The study also quantifies the trade‑offs: GA requires substantially more computational resources proportional to population size and generation count, but it is embarrassingly parallel and can be accelerated on GPU clusters. MCMC, by contrast, provides a rigorous sampling of the posterior distribution and straightforward evidence calculation, yet it can become trapped in multimodal landscapes if the proposal distribution is not well tuned.

The authors conclude that GA should not replace MCMC but rather be used in tandem. A practical hybrid workflow would first run GA to locate promising regions of the parameter space, then initialize MCMC chains within those regions to obtain high‑precision posterior samples. They also suggest extending GA to model‑selection tasks (e.g., testing Ω_k = 0 versus Ω_k ≠ 0) and to non‑Bayesian objective functions such as information criteria. Future work includes applying the GA framework to higher‑dimensional data sets like the CMB power spectrum, incorporating adaptive hyper‑parameter optimization, and exploring other evolutionary strategies. Overall, the paper demonstrates that GA can reliably reproduce MCMC results while offering additional global‑search capabilities, making it a valuable addition to the cosmologist’s computational toolbox.