| Operational Modes |
Parameter estimation
Predictive analysis
Regularization
Pareto |
| Parameter Estimation |
Gauss Marquardt Levenberg with Broyden Jacobian updating
Shuffled Complex Evolution
Covariance Matrix Adaptation |
| Parameters |
No upper limit on number of parameters
Log-transformed
Untransformed
Fixed
Tied to other parameters
Parameter scaling and offsetting
Arbitrary parameter transformation and mathematical manipulation
Self-regularizing parameter bounds enforcement through temporary, sequential fixing
|
| Observations |
Individually weighted
Covariance matrix can be supplied to grouped observations
Objective function contributions calculated for individual observation groups
Adjustment of observation group weight factors for equalization of contribution to objective function
Computation of orthogonal "super observations"
No limit to number of observations |
| Prior Information |
Applied to individual parameters or to linear relationships between parameters
Individually weighted
Covariance matrix can be supplied to grouped prior information equations
Objective function contributions calculated for individual prior information groups
No limit to number of prior information equations |
| Regularization |
Truncated singular value decomposition
Tikhonov
Pareto-adjustable Tikhonov
Subspace-enhanced Tikhonov
LSQR
"SVD-Assist"
All combinations of the above
"Automatic user intervention" |
| Model Interaction |
Writes model input files using templates of those files
Reads model output files using instruction sets
Runs model (or sequence of models) through system call
Option to receive model-computed Jacobian matrix from model
Parallelization of model runs across different nodes or machines |
| Derivatives |
Two, three or five point finite difference stencil
Parabolic, outside-points or best-fit options for three point stencil
Maximum precision or minimum error variance options for five point stencil
User-selectable relative or absolute parameter increments and/or combination of these
Analysis of finite-difference derivatives integrity
Partial detection and elimination of numerically-corrupted derivatives
Computation of composite sensitivities
Display, analysis and manipulation (including singular value decomposition) of (weighted) Jacobian matrix |
| Uncertainty Analysis |
Linear over-determined
Nonlinear over-determined through calibration-constrained predictive maximization/minimization
Linear highly-parameterized parameter/predictive error variance assessment
Linear highly-parameterized parameter/predictive uncertainty assessment
Nonlinear, regularized, calibration-constrained parameter/predictive maximization/minimization
Random parameter generation
Monte Carlo analysis
Post-calibration null-space Monte Carlo analysis
"Predictive calibration" analysis
Pareto-adjustable hypothesis testing |
| Uncertainty-Related |
Parameter identifiability
Relative parameter error variance reduction and uncertainty reduction
Resolution matrix
Parameter (group) contributions to predictive error variance
Parameter (group) contributions to predictive uncertainty
Worth of different new or existing measurements in reducing parameter/predictive error variance
Worth of different new or existing measurements in reducing parameter/predictive uncertainty
Solution and null space contributions to parameter/predictive error variance
Optimality of singular value truncation for minimization of predictive error variance
Inference statistics such as observation leverage, Cook's D, DFBETAS
Kullback-Leibler (K-L) Information Loss Statistics |
| Other |
Restart without loss of data
User-intervention
Formulation of equivalent linear model
Formulation of input dataset for scaled parameters
General manipulation of matrices and vectors
Global Jacobian matrix assembly from submatrices
Comprehensive checking of PEST input dataset for correctness and consistency
Automatic generation of PEST input dataset |
| Operating System |
PC or UNIX (source code provided for compilation under UNIX)
32 bit and 64 bit executables available for PC
|
