Markov Chain Monte Carlo sampler for Bayesian parameter estimation. More...

#include <MCMC/MCMC.h>

Inheritance diagram for CMCMC< T >:

Collaboration diagram for CMCMC< T >:

Public Member Functions
	CMCMC (void)
	Default constructor.

	CMCMC (int nn, int nn_chains)
	Constructor with chain configuration.

	CMCMC (T *system)
	Constructor with model pointer.

bool	SetProperty (const string &varname, const string &value)
	Set MCMC properties from string key-value pairs.

	~CMCMC (void)
	Destructor.

void	initialize (CMBTimeSeriesSet *results, bool random=false)
	Initialize MCMC chains with starting parameter values.

void	initialize (vector< double > par)
	Initialize all chains with specific parameter values.

bool	step (int k, int chain_counter)
	Perform single MCMC step for one chain.

bool	step (int k, int nsamps, string filename, CMBTimeSeriesSet results=nullptr, ProgressWindow _rtw=0)
	Perform multiple MCMC steps with progress tracking and output.

vector< double >	purturb (int k)
	Generate proposed parameter values by perturbing current state.

void	writeoutput (string filename)
	Write MCMC results to output file.

double	posterior (vector< double > par, int chain_counter)
	Calculate log-posterior probability for given parameters.

void	model (T *Model1, vector< double > par)
	Evaluate model with given parameters.

int	getparamno (int i, int ts) const
	Get parameter index for a specific variable and time series.

int	get_act_paramno (int i)
	Get active parameter number (legacy, may be unused)

int	get_time_series (int i)
	Get time series index for parameter (legacy, may be unused)

Parameter *	parameter (int i)
	Get pointer to specific parameter.

Observation *	observation (int i)
	Get pointer to specific observation.

CVector	sensitivity (double d, vector< double > par)
	Calculate sensitivity of model output to parameters.

CVector	sensitivity_ln (double d, vector< double > par)
	Calculate log-space sensitivity (for lognormal parameters)

CMatrix	sensitivity_mat_lumped (double d, vector< double > par)
	Calculate lumped sensitivity matrix.

TimeSeriesSet< double >	prior_distribution (int n_bins)
	Generate histogram of prior distributions.

int	readfromfile (string filename)
	Read MCMC state from file to continue previous run.

TimeSeriesSet< double >	model (vector< double > par)
	Evaluate model for given parameters and return predictions.

void	ProduceRealizations (TimeSeriesSet< double > &MCMCout)
	Generate posterior predictive realizations.

void	get_outputpercentiles (TimeSeriesSet< double > &MCMCout)
	Calculate percentiles of model outputs from MCMC samples.

void	SetRunTimeWindow (ProgressWindow *_rtw)
	Set progress window for GUI updates.

void	Perform ()
	Main entry point to run complete MCMC analysis.

Public Attributes
T *	Model
	Pointer to the model object being calibrated.

T	Model_out
	Output model state after MCMC completion.

_MCMC_settings	MCMC_Settings
	MCMC algorithm configuration settings.

vector< vector< double > >	Params
	Parameter values for all samples and chains.

vector< double >	pertcoeff
	Perturbation coefficients for each parameter.

vector< double >	logp
	Log-posterior values for current state of each chain.

vector< double >	logp1
	Log-posterior values for proposed states.

vector< double >	u
	Uniform random values for Metropolis-Hastings accept/reject.

vector< CVector >	temp_predicted
	Temporary storage for predicted model outputs.

CMBTimeSeriesSet	predicted
	Model predictions as time series set.

vector< T >	CopiedModels
	Copies of model object for parallel chain evaluation.

vector< int >	params
	Parameter indices for model evaluation.

TimeSeriesSet< double >	MData
	Measured data (observations) for likelihood calculation.

_MCMC_file_names	FileInformation
	File paths for MCMC input/output.

vector< bool >	apply_to_all
	Flags indicating if parameter applies to all entities.

vector< Parameter > *	parameters = nullptr
	Pointer to vector of Parameter objects.

vector< Observation > *	observations = nullptr
	Pointer to vector of Observation objects.

ProgressWindow *	rtw = nullptr
	Pointer to progress window for GUI updates.

vector< vector< TimeSeriesSet< double > > >	BTCout_obs
	Posterior predictive samples for observations.

vector< vector< TimeSeriesSet< double > > >	BTCout_obs_noise
	Posterior predictive samples with simulated measurement noise.

vector< vector< TimeSeriesSet< double > > >	BTCout_obs_prcntle
	Percentiles of posterior predictive distribution.

vector< vector< TimeSeriesSet< double > > >	BTCout_obs_prcntle_noise
	Percentiles of posterior predictive distribution with noise.

vector< CMatrix >	global_sens_lumped
	Global sensitivity analysis results (lumped)

TimeSeriesSet< double >	paramsList
	List of all parameter samples from MCMC.

TimeSeriesSet< double >	realized_paramsList
	Parameter samples for posterior predictive realizations.

vector< double >	calc_output_percentiles
	Percentiles to calculate for output distributions.

double	accepted_count = 0
	Count of accepted proposals.

double	total_count = 0
	Count of total proposals (accepted + rejected)

string	last_error
	Last error message from MCMC execution.

Private Member Functions
int_value_pair	Min (const vector< double > &vec, int current_counter, int n_chains)
	Find minimum value and its chain index.

int_value_pair	Max (const vector< double > &vec, int current_counter, int n_chains)
	Find maximum value and its chain index.

Detailed Description

template<class T>
class CMCMC< T >

Markov Chain Monte Carlo sampler for Bayesian parameter estimation.

The CMCMC class implements a sophisticated Metropolis-Hastings MCMC algorithm with advanced features for Bayesian inference in sediment source fingerprinting.

Algorithm Overview

MCMC generates samples from the posterior distribution p(θ|data) ∝ p(data|θ) × p(θ) using a random walk with accept/reject steps:

Propose: Generate candidate θ* from proposal distribution q(θ*|θ_current)
Evaluate: Calculate posterior ratio R = p(θ*|data) / p(θ_current|data)
Accept/Reject: Accept θ* with probability min(1, R), else keep θ_current
Repeat: Continue until desired number of samples obtained

Key Features

Multiple parallel chains for convergence diagnostics
Adaptive proposal tuning to achieve target acceptance rate
Sensitivity-based proposals for efficient exploration
Posterior predictive sampling for uncertainty quantification
Progress tracking with GUI integration
Checkpoint/resume capability for long runs

Usage in SedSat3

MCMC provides fully Bayesian estimates of source contributions with uncertainty:

Mean contributions from posterior means
Credible intervals (e.g., 95% intervals) from percentiles
Correlation between source contributions
Posterior predictive distributions for tracer concentrations

Mathematical Details

Log-posterior calculation:

log p(θ|data) = log p(data|θ) + log p(θ) + constant

= Σ log N(y_i | model(θ)_i, σ_i²) + Σ log p(θ_j)

CMCMC::model

void model(T *Model1, vector< double > par)

Evaluate model with given parameters.

Definition MCMC.hpp:182

AxisScale::log

@ log

Where:

y_i are observed tracer concentrations
model(θ)_i are predicted concentrations from mixing model
σ_i are measurement uncertainties
p(θ_j) are parameter prior distributions

Template Parameters

T	Model class type (e.g., SourceSinkData) that provides: Parameter evaluation interface Likelihood calculation Model predictions

Note: Thread-safe: Multiple chains can run in parallel

Warning: Requires careful tuning of proposal distribution for efficiency

See also: Parameter; Observation; SourceSinkData

Example usage:

// Create MCMC sampler for source apportionment
CMCMC<SourceSinkData> mcmc(&sourceData);
 
// Configure settings
mcmc.MCMC_Settings.total_number_of_samples = 50000;
mcmc.MCMC_Settings.number_of_chains = 4;
mcmc.MCMC_Settings.burnout_samples = 10000;
mcmc.MCMC_Settings.acceptance_rate = 0.234;
 
// Set parameters and observations
mcmc.parameters = &sourceParameters;
mcmc.observations = &tracerObservations;
 
// Run MCMC
mcmc.Perform();
 
// Extract results
mcmc.writeoutput("mcmc_results.txt");
mcmc.get_outputpercentiles(results);  // Get credible intervals

Definition at line 325 of file MCMC.h.

Constructor & Destructor Documentation

◆ CMCMC() [1/3]

template<class T >

CMCMC< T >::CMCMC ( void )

Default constructor.

Creates an uninitialized MCMC object. Must call other initialization methods before running.

Definition at line 19 of file MCMC.hpp.

◆ CMCMC() [2/3]

template<class T >

CMCMC< T >::CMCMC	(	int	nn,
		int	nn_chains
	)

Constructor with chain configuration.

Parameters

nn	Number of parameters
nn_chains	Number of parallel chains

Initializes MCMC with specified dimensions and chain count.

◆ CMCMC() [3/3]

template<class T >

CMCMC< T >::CMCMC ( T * system )

Constructor with model pointer.

Parameters

system Pointer to model object to be calibrated

Preferred constructor that links MCMC to a specific model instance.

Definition at line 536 of file MCMC.hpp.

◆ ~CMCMC()

template<class T >

CMCMC< T >::~CMCMC ( void )

Destructor.

Cleans up allocated memory and closes output files.

Definition at line 24 of file MCMC.hpp.

Member Function Documentation

◆ get_act_paramno()

template<class T >

int CMCMC< T >::get_act_paramno ( int i )

Get active parameter number (legacy, may be unused)

Parameters

i	Parameter index

Returns: Active parameter number

◆ get_outputpercentiles()

template<class T >

void CMCMC< T >::get_outputpercentiles ( TimeSeriesSet< double > & MCMCout )

Calculate percentiles of model outputs from MCMC samples.

Parameters

MCMCout TimeSeriesSet containing MCMC samples

Computes credible intervals (percentiles specified in calc_output_percentiles) for all model outputs, providing uncertainty bounds on predictions.

Definition at line 696 of file MCMC.hpp.

◆ get_time_series()

template<class T >

int CMCMC< T >::get_time_series ( int i )

Get time series index for parameter (legacy, may be unused)

Parameters

i	Parameter index

Returns: Time series index

◆ getparamno()

template<class T >

int CMCMC< T >::getparamno	(	int	i,
		int	ts
	)		const

Get parameter index for a specific variable and time series.

Parameters

i	Variable index
ts	Time series index

Returns: Parameter index in MCMC parameter vector

◆ initialize() [1/2]

template<class T >

void CMCMC< T >::initialize	(	CMBTimeSeriesSet *	results,
		bool	random = `false`
	)

Initialize MCMC chains with starting parameter values.

Parameters

results	Pointer to results structure to store output (optional)
random	If true, randomize initial values within parameter ranges

Sets up chains at initial parameter values. If random=true and noinipurt=false, adds random perturbations to explore different starting regions.

Precondition: parameters and observations must be set

Postcondition: Chains are positioned at starting values

Definition at line 200 of file MCMC.hpp.

References high, log, and low.

Referenced by Conductor::ExecuteTestCMBBayesian(), and SourceSinkData::MCMC().

◆ initialize() [2/2]

template<class T >

void CMCMC< T >::initialize ( vector< double > par )

Initialize all chains with specific parameter values.

Parameters

par	Vector of parameter values [p1, p2, ..., pn]

All chains start at the same parameter values (may be perturbed if noinipurt=false).

Definition at line 278 of file MCMC.hpp.

References high, log, and low.

◆ Max()

template<class T >

int_value_pair CMCMC< T >::Max	(	const vector< double > &	vec,
		int	current_counter,
		int	n_chains
	)

private

Find maximum value and its chain index.

Parameters

vec	Vector of values (one per chain)
current_counter	Current sample number
n_chains	Number of chains

Returns: Pair containing chain index and maximum value

Utility for tracking worst (lowest likelihood) chain.

Definition at line 787 of file MCMC.hpp.

◆ Min()

template<class T >

int_value_pair CMCMC< T >::Min	(	const vector< double > &	vec,
		int	current_counter,
		int	n_chains
	)

private

Find minimum value and its chain index.

Parameters

vec	Vector of values (one per chain)
current_counter	Current sample number
n_chains	Number of chains

Returns: Pair containing chain index and minimum value

Utility for tracking best (highest likelihood) chain.

Definition at line 771 of file MCMC.hpp.

References int_value_pair::value.

◆ model() [1/2]

template<class T >

void CMCMC< T >::model	(	T *	Model1,
		vector< double >	par
	)

Evaluate model with given parameters.

Parameters

Model1	Pointer to model object to evaluate
par	Parameter values to use

Sets parameters in model and computes predictions. Used within posterior() for likelihood calculation.

Definition at line 182 of file MCMC.hpp.

◆ model() [2/2]

template<class T >

TimeSeriesSet< double > CMCMC< T >::model ( vector< double > par )

Evaluate model for given parameters and return predictions.

Parameters

par	Parameter values

Returns: TimeSeriesSet of model predictions

Wrapper for model evaluation that returns predictions in convenient format.

Definition at line 46 of file MCMC.hpp.

◆ observation()

template<class T >

Observation * CMCMC< T >::observation ( int i )

Get pointer to specific observation.

Parameters

i	Observation index

Returns: Pointer to Observation object

Definition at line 40 of file MCMC.hpp.

◆ parameter()

template<class T >

Parameter * CMCMC< T >::parameter ( int i )

Get pointer to specific parameter.

Parameters

i	Parameter index

Returns: Pointer to Parameter object

Definition at line 33 of file MCMC.hpp.

◆ Perform()

template<class T >

void CMCMC< T >::Perform ( )

Main entry point to run complete MCMC analysis.

Executes full MCMC workflow:

Initialize chains
Run burn-in samples
Run production samples
Calculate statistics and diagnostics
Generate posterior predictive distributions
Save results to files

Precondition: parameters and observations must be set; MCMC_Settings must be configured

Postcondition: Results written to output files; paramsList contains posterior samples

See also: MCMC_Settings

Definition at line 725 of file MCMC.hpp.

◆ posterior()

template<class T >

double CMCMC< T >::posterior	(	vector< double >	par,
		int	chain_counter
	)

Calculate log-posterior probability for given parameters.

Parameters

par	Parameter vector [θ1, θ2, ..., θn]
chain_counter	Chain index for thread-safe model evaluation

Returns: ln(p(θ|data)) = ln(p(data|θ)) + ln(p(θ))

Computes:

Log-likelihood: Σ ln N(y_observed | y_predicted(θ), σ²)
Log-prior: Σ ln p(θ_i) from Parameter::CalcLogPriorProbability()
Log-posterior: Sum of above

Note: Uses log-space to prevent numerical underflow

See also: Parameter::CalcLogPriorProbability()

Definition at line 164 of file MCMC.hpp.

◆ prior_distribution()

template<class T >

TimeSeriesSet< double > CMCMC< T >::prior_distribution ( int n_bins )

Generate histogram of prior distributions.

Parameters

n_bins Number of bins for histogram

Returns: TimeSeriesSet containing prior PDFs for each parameter

Creates visual representation of priors for plotting and comparison with posteriors.

Definition at line 610 of file MCMC.hpp.

◆ ProduceRealizations()

template<class T >

void CMCMC< T >::ProduceRealizations ( TimeSeriesSet< double > & MCMCout )

Generate posterior predictive realizations.

Parameters

MCMCout TimeSeriesSet containing MCMC samples

Samples parameters from posterior and evaluates model to create posterior predictive distribution for uncertainty quantification.

Number of realizations set by MCMC_Settings.number_of_post_estimate_realizations.

Definition at line 652 of file MCMC.hpp.

◆ purturb()

template<class T >

vector< double > CMCMC< T >::purturb ( int k )

Generate proposed parameter values by perturbing current state.

Parameters

k	Chain index to perturb

Returns: Vector of proposed parameter values

Proposal distribution: θ* ~ N(θ_current, diag(pertcoeff²)) Respects parameter bounds by reflecting or rejecting out-of-bounds proposals.

Definition at line 377 of file MCMC.hpp.

◆ readfromfile()

template<class T >

int CMCMC< T >::readfromfile ( string filename )

Read MCMC state from file to continue previous run.

Parameters

filename Path to file containing saved MCMC state

Returns: Number of samples read, or negative value on error

Restores:

Current parameter values for each chain
Perturbation coefficients
Acceptance statistics

Allows resuming long MCMC runs without starting over.

Definition at line 584 of file MCMC.hpp.

◆ sensitivity()

template<class T >

CVector CMCMC< T >::sensitivity	(	double	d,
		vector< double >	par
	)

Calculate sensitivity of model output to parameters.

Parameters

d	Finite difference step size (fractional)
par	Current parameter values

Returns: Vector of sensitivities dOutput/dθ for each parameter

Computes numerical derivatives using forward finite differences: S_i = (model(θ + δθ_i) - model(θ)) / δθ_i

Used for sensitivity-based proposal adaptation.

Definition at line 554 of file MCMC.hpp.

◆ sensitivity_ln()

template<class T >

CVector CMCMC< T >::sensitivity_ln	(	double	d,
		vector< double >	par
	)

Calculate log-space sensitivity (for lognormal parameters)

Parameters

d	Finite difference step size (fractional)
par	Current parameter values in log-space

Returns: Vector of log-space sensitivities

Definition at line 572 of file MCMC.hpp.

◆ sensitivity_mat_lumped()

template<class T >

CMatrix CMCMC< T >::sensitivity_mat_lumped	(	double	d,
		vector< double >	par
	)

Calculate lumped sensitivity matrix.

Parameters

d	Finite difference step size
par	Current parameter values

Returns: Matrix of sensitivities [outputs × parameters]

Computes full Jacobian matrix of model outputs with respect to parameters. Used for global sensitivity analysis and identifiability assessment.

◆ SetProperty()

template<class T >

bool CMCMC< T >::SetProperty	(	const string &	varname,
		const string &	value
	)

Set MCMC properties from string key-value pairs.

Parameters

varname	Property name (e.g., "number_of_samples", "acceptance_rate")
value	Property value as string

Returns: true if property was recognized and set successfully

Allows configuration from text files or user input.

Definition at line 68 of file MCMC.hpp.

Referenced by Conductor::ExecuteTestCMBBayesian(), and SourceSinkData::MCMC().

◆ SetRunTimeWindow()

template<class T >

void CMCMC< T >::SetRunTimeWindow ( ProgressWindow * _rtw )

Set progress window for GUI updates.

Parameters

_rtw	Pointer to ProgressWindow object

Links MCMC to a progress dialog for real-time status updates.

Definition at line 530 of file MCMC.hpp.

◆ step() [1/2]

template<class T >

bool CMCMC< T >::step	(	int	k,
		int	chain_counter
	)

Perform single MCMC step for one chain.

Parameters

k	Sample iteration number
chain_counter	Index of chain to advance

Returns: true if proposed state was accepted, false if rejected

Executes one Metropolis-Hastings update:

Propose new parameters
Calculate log-posterior
Accept/reject based on Metropolis ratio
Update acceptance statistics

Definition at line 337 of file MCMC.hpp.

References log.

Referenced by Conductor::ExecuteTestCMBBayesian(), and SourceSinkData::MCMC().

◆ step() [2/2]

template<class T >

bool CMCMC< T >::step	(	int	k,
		int	nsamps,
		string	filename,
		CMBTimeSeriesSet *	results = `nullptr`,
		ProgressWindow *	_rtw = `0`
	)

Perform multiple MCMC steps with progress tracking and output.

Parameters

k	Starting sample number
nsamps	Number of samples to generate
filename	Output file for saving samples
results	Pointer to results structure (optional)
_rtw	Pointer to progress window for GUI updates (optional)

Returns: true if completed successfully

Runs the main MCMC loop with:

Periodic output saving
Adaptive proposal tuning
Progress bar updates
Convergence monitoring

See also: ProgressWindow

Definition at line 393 of file MCMC.hpp.

◆ writeoutput()

template<class T >

void CMCMC< T >::writeoutput ( string filename )

Write MCMC results to output file.

Parameters

filename Path to output file

Saves:

Parameter samples (after burn-in)
Acceptance rates
Posterior statistics (mean, std dev, percentiles)
Convergence diagnostics

Format is typically tab-delimited text for easy import into analysis software.

Member Data Documentation

◆ accepted_count

template<class T >

double CMCMC< T >::accepted_count = 0

Count of accepted proposals.

Running total across all chains and samples. Used to calculate overall acceptance rate.

Definition at line 812 of file MCMC.h.

◆ apply_to_all

template<class T >

vector<bool> CMCMC< T >::apply_to_all

Flags indicating if parameter applies to all entities.

Used for shared parameters across multiple sources or observations.

Definition at line 612 of file MCMC.h.

◆ BTCout_obs

template<class T >

vector<vector<TimeSeriesSet<double> > > CMCMC< T >::BTCout_obs

Posterior predictive samples for observations.

Model predictions at sampled parameter values. Structure: BTCout_obs[observation][realization]

Definition at line 725 of file MCMC.h.

◆ BTCout_obs_noise

template<class T >

vector<vector<TimeSeriesSet<double> > > CMCMC< T >::BTCout_obs_noise

Posterior predictive samples with simulated measurement noise.

As BTCout_obs but with added observation errors sampled from measurement uncertainty distributions.

Definition at line 733 of file MCMC.h.

◆ BTCout_obs_prcntle

template<class T >

vector<vector<TimeSeriesSet<double> > > CMCMC< T >::BTCout_obs_prcntle

Percentiles of posterior predictive distribution.

Credible intervals for model predictions (e.g., 2.5%, 50%, 97.5% percentiles).

Definition at line 740 of file MCMC.h.

◆ BTCout_obs_prcntle_noise

template<class T >

vector<vector<TimeSeriesSet<double> > > CMCMC< T >::BTCout_obs_prcntle_noise

Percentiles of posterior predictive distribution with noise.

Credible intervals including measurement uncertainty.

Definition at line 747 of file MCMC.h.

◆ calc_output_percentiles

template<class T >

vector<double> CMCMC< T >::calc_output_percentiles

Percentiles to calculate for output distributions.

Vector of percentile values (0-1 scale) to compute. Typical: {0.025, 0.16, 0.5, 0.84, 0.975} for 95% and 68% intervals.

Definition at line 796 of file MCMC.h.

◆ CopiedModels

template<class T >

vector<T> CMCMC< T >::CopiedModels

Copies of model object for parallel chain evaluation.

Each chain gets its own model copy to allow thread-safe parallel execution. Dimensions: [number_of_chains]

Definition at line 520 of file MCMC.h.

◆ FileInformation

template<class T >

_MCMC_file_names CMCMC< T >::FileInformation

File paths for MCMC input/output.

Configuration structure containing output directory and filenames.

Definition at line 557 of file MCMC.h.

◆ global_sens_lumped

template<class T >

vector<CMatrix> CMCMC< T >::global_sens_lumped

Global sensitivity analysis results (lumped)

Sensitivity matrices calculated across posterior samples.

Definition at line 754 of file MCMC.h.

◆ last_error

template<class T >

string CMCMC< T >::last_error

Last error message from MCMC execution.

If MCMC fails or encounters issues, error description is stored here.

Definition at line 826 of file MCMC.h.

◆ logp

template<class T >

vector<double> CMCMC< T >::logp

Log-posterior values for current state of each chain.

Dimensions: [number_of_chains] Updated after each MCMC step.

Definition at line 415 of file MCMC.h.

◆ logp1

template<class T >

vector<double> CMCMC< T >::logp1

Log-posterior values for proposed states.

Temporary storage for evaluating proposed parameter values. Dimensions: [number_of_chains]

Definition at line 423 of file MCMC.h.

◆ MCMC_Settings

template<class T >

_MCMC_settings CMCMC< T >::MCMC_Settings

MCMC algorithm configuration settings.

All tuning parameters and flags controlling MCMC behavior. Modify before calling Perform() or initialize().

Definition at line 391 of file MCMC.h.

◆ MData

template<class T >

TimeSeriesSet<double> CMCMC< T >::MData

Measured data (observations) for likelihood calculation.

Contains observed tracer concentrations and their uncertainties. Used to calculate likelihood: p(data|θ).

Definition at line 550 of file MCMC.h.

◆ Model

template<class T >

T* CMCMC< T >::Model

Pointer to the model object being calibrated.

The model provides likelihood evaluation and predictions. For SedSat3, this is typically a SourceSinkData object.

Definition at line 334 of file MCMC.h.

Referenced by Conductor::ExecuteTestCMBBayesian(), and SourceSinkData::MCMC().

◆ Model_out

template<class T >

T CMCMC< T >::Model_out

Output model state after MCMC completion.

Contains model configured with best-fit or posterior mean parameters.

Definition at line 341 of file MCMC.h.

◆ observations

template<class T >

vector<Observation>* CMCMC< T >::observations = nullptr

Pointer to vector of Observation objects.

Contains observed tracer concentrations and measurement uncertainties. Must be set before running MCMC.

Definition at line 628 of file MCMC.h.

◆ parameters

template<class T >

vector<Parameter>* CMCMC< T >::parameters = nullptr

Pointer to vector of Parameter objects.

Contains prior distributions, ranges, and names for all parameters. Must be set before running MCMC.

Definition at line 620 of file MCMC.h.

◆ Params

template<class T >

vector<vector<double> > CMCMC< T >::Params

Parameter values for all samples and chains.

Storage structure: Params[chain_index][parameter_index] Dimensions: [number_of_chains][number_of_parameters]

Definition at line 399 of file MCMC.h.

◆ params

template<class T >

vector<int> CMCMC< T >::params

Parameter indices for model evaluation.

Maps MCMC parameter space to model parameter space. May be a subset if some parameters are fixed.

Definition at line 542 of file MCMC.h.

◆ paramsList

template<class T >

TimeSeriesSet<double> CMCMC< T >::paramsList

List of all parameter samples from MCMC.

Complete chain history after burn-in for posterior analysis.

Definition at line 761 of file MCMC.h.

◆ pertcoeff

template<class T >

vector<double> CMCMC< T >::pertcoeff

Perturbation coefficients for each parameter.

Adaptive proposal standard deviations, one per parameter. Updated during sampling to achieve target acceptance rate.

Definition at line 407 of file MCMC.h.

◆ predicted

template<class T >

CMBTimeSeriesSet CMCMC< T >::predicted

Model predictions as time series set.

Contains predicted tracer concentrations for plotting and comparison with observations.

Definition at line 512 of file MCMC.h.

Referenced by SourceSinkData::MCMC().

◆ realized_paramsList

template<class T >

TimeSeriesSet<double> CMCMC< T >::realized_paramsList

Parameter samples for posterior predictive realizations.

Subset of posterior samples used to generate model predictions.

Definition at line 768 of file MCMC.h.

◆ rtw

template<class T >

ProgressWindow* CMCMC< T >::rtw = nullptr

Pointer to progress window for GUI updates.

If not nullptr, MCMC will update progress bar and display current acceptance rate, parameter values, etc.

Definition at line 672 of file MCMC.h.

◆ temp_predicted

template<class T >

vector<CVector> CMCMC< T >::temp_predicted

Temporary storage for predicted model outputs.

Used during likelihood evaluation to store model predictions for each chain without modifying the main model object.

Definition at line 504 of file MCMC.h.

◆ total_count

template<class T >

double CMCMC< T >::total_count = 0

Count of total proposals (accepted + rejected)

Running total of all MCMC steps attempted.

Definition at line 819 of file MCMC.h.

◆ u

template<class T >

vector<double> CMCMC< T >::u

Uniform random values for Metropolis-Hastings accept/reject.

Pre-generated uniform random numbers U(0,1) for comparing with acceptance probability. Dimensions: [number_of_chains]

Definition at line 432 of file MCMC.h.

The documentation for this class was generated from the following files:

MCMC/MCMC.h
MCMC/MCMC.hpp

Public Member Functions

Public Attributes

Private Member Functions

Detailed Description

Algorithm Overview

Key Features

Usage in SedSat3

Mathematical Details

Constructor & Destructor Documentation

◆ CMCMC() [1/3]

◆ CMCMC() [2/3]

◆ CMCMC() [3/3]

◆ ~CMCMC()

Member Function Documentation

◆ get_act_paramno()

◆ get_outputpercentiles()

◆ get_time_series()

◆ getparamno()

◆ initialize() [1/2]

◆ initialize() [2/2]

◆ Max()

◆ Min()

◆ model() [1/2]

◆ model() [2/2]

◆ observation()

◆ parameter()

◆ Perform()

◆ posterior()

◆ prior_distribution()

◆ ProduceRealizations()

◆ purturb()

◆ readfromfile()

◆ sensitivity()

◆ sensitivity_ln()

◆ sensitivity_mat_lumped()

◆ SetProperty()

◆ SetRunTimeWindow()

◆ step() [1/2]

◆ step() [2/2]

◆ writeoutput()

Member Data Documentation

◆ accepted_count

◆ apply_to_all

◆ BTCout_obs

◆ BTCout_obs_noise

◆ BTCout_obs_prcntle

◆ BTCout_obs_prcntle_noise

◆ calc_output_percentiles

◆ CopiedModels

◆ FileInformation

◆ global_sens_lumped

◆ last_error

◆ logp

◆ logp1

◆ MCMC_Settings

◆ MData

◆ Model

◆ Model_out

◆ observations

◆ parameters

◆ Params

◆ params

◆ paramsList

◆ pertcoeff

◆ predicted

◆ realized_paramsList

◆ rtw

◆ temp_predicted

◆ total_count

◆ u