Tm. Gordon et al., EXPLORATORY DATA-ANALYSIS IN THERMOBAROMETRY - AN EXAMPLE FROM THE KISSEYNEW SEDIMENTARY GNEISS BELT, MANITOBA, CANADA, The American mineralogist, 79(9-10), 1994, pp. 973-982
Generalized thermobarometry uses measured partial molar free-energy fu
nctions computed from standard state properties, activity models, and
measured compositions, along with the linear dependence of the solutio
ns to these functions at equilibrium to estimate pressure, temperature
, and partial molar free energies of individual mineral end-members. S
uch problems are often inconsistent, i.e., there are no values of pres
sure and temperature that simultaneously satisfy all the equations. Th
ese inconsistencies result from errors in the measured partial molar f
ree-energy functions and may be classified into random errors (those t
hat can be modeled as arising from a statistical population) and syste
matic errors (those that cannot). Because the systematic error may dom
inate, exploratory data analysis is an essential step in the interpret
ation of thermobarometry results. Several complementary techniques may
be used to provide insight into the sensitivity of estimated pressure
s and temperatures to systematic errors in the measured partial molar
free-energy functions for individual end-members. The number of equati
ons is small, and there are usually only one to three fewer unknowns t
han equations; hence, statistical methods designed to investigate inco
nsistencies in large data sets cannot be readily applied. However, sev
eral procedures can be used to guide interpretation of the results of
thermobarometric calculations. (1) Examination of residuals provides a
direct indication of relative misfit of the measured partial molar fr
ee-energy functions to the theoretical values of chemical potential at
equilibrium. The intrinsic correlation of residuals in small data set
s limits the utility of this approach. (2) Contours of the sum of squa
res of residuals illustrate the covariance and overall reliability of
the least-squares solution to the problem. Assigning actual confidence
levels to contours requires that the errors have statistical properti
es that may not exist in practice. (3) Case deletion studies require t
hat the equations be solved repeatedly, eliminating the equation for e
ach individual end-member in turn. Diagrams illustrating these results
serve to confirm or refute the existence of end-members that are high
ly influential in determining best-fit pressures and temperatures. (4)
Perturbation analysis also requires the repeated solution of the prob
lem, but, instead of deleting individual end-members, their measured f
ree-energy estimates are perturbed by fixed amounts, and a new pressur
e and temperature are determined. Displays of this kind provide quanti
tative estimates of the change in estimated pressures and temperatures
resulting from arbitrary perturbations in thermodynamic constants and
measured compositions.