Fitting Algorithms

Available in MaCh3 fitting algorithms. More...

Classes
class	DelayedMR2T2
	Implementation of delayed rejection for MR2T2. More...
class	MinuitFit
	Implementation of Minuit fitting algorithm [22]. More...
class	MR2T2
	MCMC algorithm implementing the Metropolis–Rosenbluth–Rosenbluth–Teller–Teller (MR \(^2\)T \(^2\)) method. More...
class	PSO
	Class PSO, consist of a vector of object Class Particle and global best Takes in the size (number of particle) and number of iteration functions includes: finding global best, updating velocity, actual minimisation function. More...

Functions
std::unique_ptr< FitterBase >	MaCh3FitterFactory (Manager *FitManager)
	MaCh3 Factory initiates one of implemented fitting algorithms. More...

Detailed Description

Available in MaCh3 fitting algorithms.

Introduction

MaCh3 provides multiple fitting strategies that fall into two main categories: Markov Chain Monte Carlo samplers and likelihood minimisation algorithms.

---

Markov Chain Monte Carlo (MCMC)

These algorithms explore parameter space by stochastic sampling of the posterior. They are primarily used for full Bayesian inference and high-dimensional problems.

Implemented methods:

• MR2T2 — Metropolis–Rosenbluth–Rosenbluth–Teller–Teller sampler.
• DelayedMR2T2 — Extension including delayed rejection to improve sampling efficiency and acceptance in difficult likelihood surfaces.

Implemented Variants: There are multiple ways to modify algorithm by using so called variants. This modifies only single aspect of algorithm and are enable or disabled on on config level. This list includes

• Adaptive MCMC (AdaptiveMCMCHandler) - modifies proposal matrix resulting in better acceptance and smaller autocorrelations
• Multi-Canonical or Umbrella Sampling (MulticanonicalMCMCHandler) - modifies the sampling distribution to improve exploration of difficult or multimodal posterior regions.

  General:
    MCMC:
      Multicanonical:
        Enabled: true

• Temperature annealing (MCMCBase::AnnealTemp) - modifies the effective sampling temperature, allowing the sampler to explore the parameter space more broadly.

  General:
    MCMC:
      AnnealTemp: 0.5

---

Likelihood Minimizers

These algorithms deterministically minimise the negative log-likelihood to obtain a best-fit point. They are typically much faster than MCMC and are useful for diagnostics, validation, and systematic scans.

Implemented methods:

• MinuitFit — Gradient-based minimiser.
• PSO (Particle Swarm Optimisation) — Global optimisation using a swarm of walkers to locate the likelihood minimum.

Factory Method

A factory method is implemented to select the algorithm from the configuration file instead of hard-coding it.

Example configuration:

General:
  FittingAlgorithm: "MR2T2"

or

General:
  FittingAlgorithm: "PSO"

This allows switching algorithms without recompilation.

Function Documentation

MaCh3FitterFactory()

std::unique_ptr<FitterBase> MaCh3FitterFactory

(

Manager *

FitManager

)

MaCh3 Factory initiates one of implemented fitting algorithms.

Parameters

FitManager

pointer to Manager class

Note

Example YAML configuration:

General:
  FittingAlgorithm: ["MCMC"]

Definition at line 5 of file MaCh3Factory.cpp.

                                                              {
// ********************************************
  std::unique_ptr<FitterBase> MaCh3Fitter = nullptr;
 
  auto Algorithm = GetFromManager<std::string>(fitMan->raw()["General"]["FittingAlgorithm"], "MCMC", __FILE__ , __LINE__);
 
  if(Algorithm == "MCMC" || Algorithm == "MR2T2") 
  {
    MaCh3Fitter = std::make_unique<MR2T2>(fitMan);
  }
  else if (Algorithm == "DelayedMCMC"  || Algorithm == "DelayedMR2T2")
  {
    MaCh3Fitter = std::make_unique<DelayedMR2T2>(fitMan);
  }
  else if (Algorithm == "PSO")
  {
    MaCh3Fitter = std::make_unique<PSO>(fitMan);
  }
  else if (Algorithm == "Minuit2")
  {
    #ifdef MaCh3_MINUIT2
    MaCh3Fitter = std::make_unique<MinuitFit>(fitMan);
    #else
    MACH3LOG_ERROR("Trying to use Minuit2 however MaCh3 was compiled without Minuit2 support");
    throw MaCh3Exception(__FILE__ , __LINE__ );
    #endif
  }
  else
  {
    MACH3LOG_ERROR("You want to use algorithm {}, I don't recognize it, sry", Algorithm);
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  return MaCh3Fitter;
}