Glossary of Terms
adaptive Gaussian process (AGP)
A Gaussian process method that adaptively adds new sample points based on the location of the maximum error estimation of the most recent Gaussian process model.
adaptive sparse polynomial chaos expansion (ASPCE)
A sparse polynomial chaos expansion surrogate model that adaptively adds new sample points from a
Latin hypercube sampling (LHS)
if error estimation is larger than the relative tolerance.
bivariate correlation (Pearson’s correlation)
A sample-based sensitivity analysis method that computes the linear relationship between a
QoI (quantity of interest)
and an input parameter.
calibration parameters
Global parameters with unknown probability distributions. The distributions of these are determined (calibrated) in an inverse uncertainty quantification study.
compute action
An action type that defines how to generate and use sampled input parameter data and model evaluations to perform UQ analysis.
covariance
A covariance function encodes the assumption of the function you want to learn with a GP model, for instance, its smoothness and length scale.
cumulative distribution function (CDF)
The probability that an uncertain variable is small or equal to a certain value.
efficient global reliability analysis (EGRA)
A global reliability analysis method that balances the exploitation of the surrogate model and exploration of the unobserved region to maximize the accuracy of the surrogate model near the location where the QoIs are close to the threshold, thereby efficiently increasing the accuracy of the reliability analysis.
expected feasibility function (EFF)
The error estimation used to find an adaptation point for the AGP used in reliability analysis. The EFF defines the expectation of the sample lying in the vicinity around the limit state where the QoIs are equal to the thresholds.
experimental parameters
Global parameters with known probability distributions. These are the parameters both used to create the surrogate model and to produce experimental data in the inverse uncertainty quantification study.
Gaussian process (GP)
A popular probabilistic surrogate model that provides both the prediction and the variance of the prediction at every point sampled from the input parameter space.
importance sampling
A method that, through sampling, forms a distribution that overweights an importance region. The method is used for reliability analysis.
input parameters
Global parameters with uncertainties. These need to be specified before the model can be used to conduct uncertainty analysis.
inverse uncertainty quantification
A UQ analysis that computes the posterior distribution of the calibration parameters based on experimental data and prior distributions of the calibration parameters.
kernel density estimation (KDE)
A numerical technique for estimating the probability density function based on sample data. The method is formulating the probability density functions by using a known kernel function.
Latin hypercube sampling (LHS)
A method used to generate sample data with good space filling in the input-parameter space.
Markov chain Monte Carlo (MCMC)
A method that forms a chain of samples from the posterior distribution for the calibration parameters in the inverse uncertainty quantification study.
model evaluation
A COMSOL Multiphysics computation for the sampled input parameter values and the QoIs evaluated based on the corresponding solution.
Morris one-at-a-time method (MOAT)
The sample-based global screening method used to rank the importance of the input parameters.
partial correlation
A sample-based sensitivity analysis method that computes the monotonic relationship between a QoI and an input parameter.
probability density function (PDF)
The probability weight function for an uncertain variable.
quantity of interest (QoI)
The output from the COMSOL Multiphysics model evaluation that provides data for the uncertainty analysis.
rank bivariate correlation (Spearman’s correlation)
A sample-based sensitivity analysis method that computes the linear relationship between the ranking of a QoI and the ranking of an input parameter.
rank partial bivariate correlation
A sample-based sensitivity analysis method that computes the monotonic relationship between the ranking of a QoI and the ranking of an input parameter.
reliability analysis
A UQ analysis that computes the probability that the QoIs satisfy a condition defined by thresholds.
response surface
A postprocessing method that displays the relationship between each QoI and all the input parameters. Two parameters at a time can be visualized in a Table Surface Plot.
screening
An UQ analysis that qualitatively ranks the importance of all input parameters for each of the QoIs separately.
sensitivity analysis
A UQ analysis that quantitatively computes the influence of all input parameters for each of the QoIs separately.
sparse polynomial chaos expansion (SPCE)
A polynomial chaos expansion surrogate model that finds the sparse representation of a multivariate orthonormal polynomial basis.
Sobol method
A sensitivity analysis method that decomposes the variance of each QoI into a sum of contributions from the input parameters and their interactions.
surrogate model
A class of inexpensive-to-evaluate models used instead of a COMSOL Multiphysics model for Monte Carlo-type UQ analysis.
uncertainty propagation
A UQ analysis that propagates the uncertainty in inputs to the uncertainty in the outputs (QoIs).
verify action
An action type that defines how to generate and use sampled data and to verify the accuracy of a surrogate model.