Hindawi

Journal of Spectroscopy
Volume 2018, Article ID 5837214, 13 pages
https://doi.org/10.1155/2018/5837214

Research Article
Statistically Coherent Calibration of X-Ray Fluorescence
Spectrometry for Major Elements in Rocks and Minerals

Surendra P. Verma ,1 Sanjeet K. Verma ,2 M. Abdelaly Rivera-Gómez,3

Darı́o Torres-Sánchez ,4 Lorena Dı́az-González,5 Alejandra Amezcua-Valdez,6
Beatriz Adriana Rivera-Escoto,7 Mauricio Rosales-Rivera ,8 John S. Armstrong-Altrin,3
Héctor López-Loera,2 Fernando Velasco-Tapia,9 and Kailasa Pandarinath1
1
Instituto de Energı́as Renovables, Universidad Nacional Autónoma de México, Temixco, Mor 62580, Mexico
2
División de Geociencias, Instituto Potosino de Investigación en Ciencia y Tecnologı́a, Camino a la Presa San José # 2055,
Col. Lomas 4a Sec., San Luis Potosı́, SLP 78216, Mexico
3
Instituto de Ciencias del Mar y Limnologı́a, Unidad de Procesos Oceánicos y Costeros,
Universidad Nacional Autónoma de México, Circuito Exterior s/n, 04510 CDMX, Mexico
4
Posgrado en Geociencias Aplicadas, Instituto Potosino de Investigación en Ciencia y Tecnologı́a,
Camino a la Presa San José # 2055, Col. Lomas 4a Sec., San Luis Potosı́, SLP 78216, Mexico
5
Centro de Investigación en Ciencias, Instituto de Investigación en Ciencias Básicas y Aplicadas,
Universidad Autónoma del Estado de Morelos, Cuernavaca, Mor 62209, Mexico
6
Posgrado en Ingenierı́a, Instituto de Energı́as Renovables, Universidad Nacional Autónoma de México, Temixco,
Mor 62580, Mexico
7
División de Materiales Avanzados, Instituto Potosino de Investigación en Ciencia y Tecnologı́a,
Camino a la Presa San José # 2055, Col. Lomas 4a Sec., San Luis Potosı́, SLP 78216, Mexico
8
Doctorado en Ciencias, Instituto de Investigación en Ciencias Básicas y Aplicadas,
Universidad Autónoma del Estado de Morelos, Cuernavaca, Mor 62209, Mexico
9
Universidad Autónoma de Nuevo León, Facultad de Ciencias de la Tierra, Ex–Hacienda de Guadalupe,
Carretera Linares–Cerro Prieto km 8, Linares, N.L. 67700, Mexico

Correspondence should be addressed to Surendra P. Verma; spv@ier.unam.mx

Received 8 August 2018; Revised 8 October 2018; Accepted 19 October 2018; Published 11 December 2018

Academic Editor: Rafal Sitko

Copyright © 2018 Surendra P. Verma et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

We applied both the ordinary linear regression (OLR) and the new uncertainty weighted linear regression (UWLR) models for
the calibration and comparison of a XRF machine through 59 geochemical reference materials (GRMs) and a procedure blank
sample. The mean concentration and uncertainty data for the GRMs used for the calibrations (Supplementary Materials)
(available here) filewere achieved from an up-to-date compilation of chemical data and their processing from well-known
discordancy and significance tests. The drift-corrected XRF intensity and its uncertainty were determined from mostly
duplicate pressed powder pellets. The comparison of the OLR (linear correlation coefficient r∼0.9523–0.9964 and 0.9771–
0.9999, respectively, for before and after matrix correction) and UWLR models (r∼0.9772–0.9976 and 0.9970–0.9999, re-
spectively) clearly showed that the latter with generally higher values of r is preferable for routine calibrations of analytical
procedures. Both calibrations were successfully applied to rock matrices, and the results were generally consistent with those
obtained in other laboratories although the UWLR model showed mostly narrower confidence limits of the mean (slope and
intercept) or lower uncertainties than the OLR. Similar sensitivity (∼2.69–46.17 kc·s1·%1 for the OLR and ∼2.78–59.69 kc·s1·%1
for the UWLR) also indicated that the UWLR could advantageously replace the OLR model. Another novel aspect is that the
total uncertainty can be reported for individual chemical data. If the analytical instruments were routinely calibrated from the
UWLR model, this action would make the science of geochemistry more quantitative than at present.
2 Journal of Spectroscopy

1. Introduction 2. Evaluation of Major Element Data for GRMs

All modern analytical instruments require some kind of A total of 59 GRMs (listed in alphabetical order in Table S1;
calibration of the instrumental response (y-variable) as a this and four other tables are provided in Supplementary
function of the concentration (x-variable) [1–3]. This Materials), along with a procedure blank, were used in this
calibration is generally achieved through an ordinary least- study. The procedure blank was a pellet prepared in du-
squares linear regression (OLR) model. However, such plicate with only pure N,N′-Ethylene bis(stearamide) beads
a procedure is not strictly valid because all requirements without any sample (Section 3). The individual data reported
for the statistical validity of the OLR model are not ful- in earlier compilations [31–47] were first compiled in new
filled. Usually, the assumptions “independent concentra- databases.
tion variable x is error-free or less than one-tenth of the The statistical parameters obtained in these early com-
error in the dependent response variable y” and “error in y pilations could not be directly used for instrumental cali-
is homoscedastic” (i.e., equal errors for all y values) are not brations due to the following reasons: (i) the statistical
satisfied and, therefore, more sophisticated and statistically methods used to achieve the statistical estimates were
coherent regression procedures, such as weighted least- outdated (see [17, 18, 48, 49] for possible reasons), and the
squares linear regression (WLR) models, should be used inferred statistical values were of low quality (high values of
[4–18]. dispersion); (ii) there are still determinations reported
X-ray fluorescence (XRF) spectrometry is among the during about 30 or more years (postcompilation years) that
most popular analytical techniques for the determination of were not obviously available to those compilers; (iii) the
all major and some trace elements in rocks [4, 19–27]. precision of more recent determinations is likely to have
Natural geochemical reference materials (GRMs) are com- improved due to the availability of online computers on
monly used for XRF calibrations and posterior character- most modern instruments; (iv) newer more reliable statis-
ization of those and other GRMs as well as of similar rock tical techniques are now available for improving both
and mineral matrices [4, 19, 28–30]. As for most other precision and accuracy of the statistical inferences, e.g., the
analytical instruments, XRF spectrometers are also cali- use of discordancy tests with the highest power and lowest
brated under the statistically incoherent OLR model. swamping and masking effects [18, 48, 50–52]; and (v)
To apply the WLR and compare it with the OLR, both importantly, new computer programs have been developed
central tendency (e.g., mean) and dispersion (e.g., confidence by our group [52–54], available at http://tlaloc.ier.unam.mx
limits of the mean) estimates on both x-axis (concentration, for download or online processing of data (after previous
generally expressed in the unit of % m/m, i.e., mass/mass unit registration onto our server), which can be advantageously
expressed in percent) and y-axis (response, in this case used for efficient processing of experimental databases.
XRF intensity, generally reported in the unit of kc·s−1, The same kinds of objections are applicable even today
i.e., kilo counts per second) variables are required. More for the originator’s websites, such as https://gbank.gsj.jp/
precise (and accurate) estimates of the central tendency geostandards/welcome.html for Japanese GRMs or https://
will also be useful for both types of regressions. Therefore, crustal.usgs.gov/geochemical_reference_standards for United
precise concentrations of GRMs with the respective lowest States GRMs. The statistical information at these websites is
possible “confidence limits of the mean” (referred here- based on early compilations (around 30 or more years ago).
after as the “uncertainty” of the measured variable) Furthermore, we were unable to use the recent work [55]
[2, 17, 18] are required to apply the regression procedure. because this paper reported significantly larger uncertainty
Sometimes, we had to use also the term “error” (instead of values as compared to those achievable from our new val-
the uncertainty) because the use of the error is widespread idated statistical procedure [51–54]; besides, updated sta-
in the literature. tistical information on the mean and its uncertainty was not
We report the following five aspects: (a) evaluation of 59 available in [55] for many GRMs used in our work.
GRMs to achieve the least possible uncertainties in the mean The initial databases were complemented by individual
concentrations of all major elements (SiO2 to P2O5); (b) the data from a large number of posterior publications (∼480;
comparison of regression models (OLR and WLR) applied to Table S1), whose complete listing is available at our server
net drift-corrected XRF intensities before the correction of http://tlaloc.ier.unam.mx under the heading of “Quality
matrix effects; (c) the second (or final) comparison of both Control.” These major element data were classified according
models after achieving the matrix correction as well as for to the analytical method groupings [56]. Data from each
the estimation of sensitivities of the regression models; (d) method group were considered as a univariate statistical
application of the entire procedure to four GRMs treated as sample. Appropriate discordancy and significance tests were
“unknown” samples and their comparison with the previous applied from thoroughly automatized software UDASys2
literature compilations; and (e) development of a computer [52] and UDASys3 (unpublished), which, in their “recom-
program to achieve the abovementioned objectives. Thus, mended procedure,” apply the most powerful five (two new
the regression equations (intercept and its uncertainty, slope and three conventional) recursive tests with prior applica-
and its uncertainty, and linear correlation coefficient values) tion of respective single-outlier tests having nil swamping
for each constituent from SiO2 to P2O5 and their application and low masking effects [48, 57–60]. Although the appli-
to similarly complex rock matrices are presented in this cation of discordancy tests is identical for both UDASys2
work. and UDASys3, the difference lies in that the latter applies the
Journal of Spectroscopy 3

significance (ANOVA, F and t) tests in order to provide the 8 to 10 times each day. First, the expected monitor intensity
final results automatically. was established as an average value of the first two days when
The resulting statistical information after the application the intensities were fairly stable and reproducible. Then, the
of well-known discordancy tests at the strict 99% confidence average drift correction factors were calculated for each
level (mean and uncertainty values rounded according to the chemical element from the two monitors run in the XRF
flexible rules [18]) is listed in Table S2. These GRM com- instrument periodically before and after a set of GRMs used
positional data showed by far the lowest 99% uncertainty for the calibration. These correction factors were then ap-
(Table S2), much lower than any existing compilation plied to the bracketed GRMs for the entire period of cali-
[31–47, 55]. We may also stress once again that this was bration, including the first two days and analysis of
achieved through an objective combination of discordancy “unknown” samples.
tests having the highest performance and lowest swamping Now, although the X-ray counts may obey a Poisson
and masking effects [17, 18, 48, 53], i.e., from the meth- distribution, we are dealing with average values of count
odology having the lowest type I and type II errors and the rates, which are likely to follow a normal distribution be-
highest power. cause of the central limit theorem. A normal distribution of
Therefore, the population mean of these GRMs is now measured intensities was also assured for each pellet from
known within the narrowest possible 99% confidence limits the application of discordancy tests as explained above for
of the mean to best represent the concentration (x) axis in GRM concentrations. The intensity results for all pellets
the instrumental calibrations as suggested [2, 5, 7, from a given GRM were then combined, the tests applied
10, 17, 18, 53]. These data (in units of % m/m; Table S2) will again to the combined data, and new mean and 99% un-
also be useful for those who wish to achieve instrumental certainty values were calculated for X-ray intensity of each
calibrations or simply use them for quality control of their GRM. This was done to take into account the variance of the
results for rock and mineral matrices. sample preparation method, which was significantly higher
than the instrumental variance of intensity measurements
3. XRF Instrumentation and for individual pellets. The drift-corrected intensity values
Intensity Measurements and their 99% uncertainties (kc·s−1) for all GRMs, along with
the concentration data and their 99% uncertainties (% m/m),
A wavelength dispersive X-ray fluorescence (WDXRF) are listed in Table S2.
spectrometer Rigaku ZSX Primus II model (rhodium X-ray
tube; 4 kW maximum power) was used for this work. We 4. Regression Models
made the effort to best represent the response (y) axis (x-ray
intensity in the units of kilo counts per second, kc·s−1) for the Two different regression models (OLR and UWLR) were used
calibrations. For each GRM, duplicate (41 samples) or even and compared in this work. The OLR model most frequently
triplicate (8 samples) pressed powder pellets were prepared. used for instrumental calibrations (x-axis concentration and
First, an appropriate amount of each GRM was dried y-axis response; GRM concentration and X-ray intensity,
overnight in an oven at about 105°C. For each pellet, ac- respectively, in XRF spectrometry) requires the following
curately weighed 3.5 g of moisture-free GRM was thor- assumptions to be fulfilled [4, 7, 10, 12–18]: (i) all errors are in
oughly mixed with accurately weighed 3.0 g pure N,N′- the y-axis; (ii) x-axis is either error-free or has at most 10%
ethylene bis(stearamide) beads, <840 μm as wax (Sigma- error of the y-axis errors; (iii) errors in both axes are normally
Aldrich), and stored in a desiccator. Pressed powder pel- distributed; and (iv) errors in the y-axis are homoscedastic.
lets were prepared at 20 tons·inch−2 pressure (about Some or all of these assumptions are violated in most in-
310 MPa). However, for 10 GRMs, sufficient material was strumental calibrations through the OLR model.
not available; therefore, only a single pellet could be pre- Thus, from the literature on the GRMs, it has been
pared but the measured intensity uncertainty (u99 ) at the demonstrated that the concentration axis is not error-free
99% confidence level was increased by a factor of 2 to take (see non-zero uncertainties for all GRM concentrations in
into account the sample preparation variance. Similarly, Table S2) [31–47, 51–53]. One can also clearly see that the
accurately weighed 6.5 g of pure N,N′-ethylene bis(stear- errors in the intensity axis are not homoscedastic (see un-
amide) beads, <840 μm as wax, was pressed to prepare a equal, i.e., heteroscedastic uncertainties for any element in
procedure blank sample. This was done in duplicate. different GRMs in Table S2). For a heteroscedastic linear
For the intensity measurements, the optimum in- regression system, even if each error or noise term is still
strumental conditions were first established through pre- Gaussian, the OLR model is no longer the maximum
liminary experiments prior to the routine measurements likelihood estimate and consequently, it is no longer efficient
(Table S2). Each pellet was run at least 8 to 10 times in a [10]. The main advantage that the WLR has over the OLR is
random sequence, along with two drift monitors prepared the ability to handle regression situations in which the data
from two volcanic rocks (basalt and rhyolite) from the San points are of varying quality as is the case with most in-
Luis Potosı́ Volcanic Field, San Luis Potosı́ (central Mexico). struments including the XRF spectrometers.
The peak and background measuring conditions and However, the major disadvantage of the WLR is that the
time periods are also listed in Table S3. Appropriate mean approach is based on the assumption that the weights are
drift corrections from two monitors were applied to all known exactly. They can be estimated using several different
intensity measurements. Both monitors were run randomly equations or algorithms, but when the weights are produced
4 Journal of Spectroscopy

from small numbers of replicated observations, the re- 􏽐ni�1 􏼈 xi − x􏼁 × yi − y􏼁􏼉

gression parameters can be unpredictably affected [10]. In mO � 2 , (2)
􏽐ni�1 xi − x􏼁
the example of the XRF calibration that we are presenting,
the numbers of observations were relatively large for both where x and y are, respectively, the mean values of the x and
the x and y axes (concentration and X-ray intensity pa- y variables:
rameters). Besides, instead of the sample variance, we used 􏽶�����������������
the uncertainty values (that take into account the number of ⎪􏽴
⎧
⎪ 2 ⎪
⎫
⎪
⎨ 􏽐ni�1 yi − y􏽢i 􏼁 ⎬ CL
observations in the formula for uncertainty or confidence umO � ⎪ n 2 ⎪ × t(n−2) , (3)
⎪
⎩ (n − 2)􏽐i�1 xi − x􏼁 ⎪ ⎭
limits of the mean calculations) [2, 18] for estimating the
weight factors. The problem of the sensitivity to outliers in
the regression equations [10] was also appropriately handled where y􏽢i is the value of yO for xi in equation (1) and t is the
by discordancy tests programmed in the UDASys and Student’s t test value for (n − 2) degrees of freedom, and the
BiDASys software [53, 54, 61]. superscript CL is the confidence level, generally 95% or 99%:
Therefore, although frequently used, the OLR model is bO � y − mO × x􏼁,
not statistically correct or coherent. The statistically coherent
WLR, especially the uncertainty-based WLR (UWLR [17]) 􏽶�������������������
⎪􏽴 n
⎧
⎪ ⎪
⎫ (4)
model, should be used. The confidence level, such as 95% or ⎨ 􏽐i�1 yi − y􏽢i 􏼁2 × 􏽐ni�1 x2i ⎪
⎬
ubO �⎪ × tCL
(n−2) .
99% (significance level of 5% or 1%, respectively, or α of 0.05 ⎪ n(n − 2)􏽐i�1 xi − x􏼁 ⎪
⎩
n 2 ⎪
⎭
and 0.01, respectively), can be explicitly expressed in the
confidence limits of the mean or uncertainty used in the
UWLR model as well as to estimate the weight factors [17]. It is a general practice in most instrumental calibrations
We will deal with the 99% uncertainty to have the type I to ignore all uncertainties in equation (1) and use an OLR
error small (about 1%). Unfortunately, software of most equation without any error (or uncertainty) as follows:
analytical instruments, including XRF spectrometers, allows yO � bO + 􏼂mO × x􏼃. (5)
only the OLR calibration. Therefore, any sophisticated re-
gression model, such as the UWLR, will have to be applied The resulting standard deviation values of repeat mea-
outside the instrumental software. Thus, the probability surements of unknown samples are reported as the final
concept (99% confidence level) can be explicitly used in the errors. However, these are only partial errors because the
UWLR model for weight factors based on the inverse of the errors in the calibration equation (1) are not taken into
squared 99% uncertainty of the mean. account. In this work, we will use equation (1) to report total
We now present a synthesis of the regression equations errors (in fact, 99% uncertainties) for the OLR model.
for instrumental calibrations [2, 10, 17, 18, 61].
4.2. Uncertainty Weighted Least-Squares Linear Regression
4.1. Ordinary Least-Squares Linear Regression (OLR) Model. (UWLR) Model. For the UWLR model, the n pairs (xi , yi )
Let us assume that we have a series of n reference materials of calibrators as well as the respective uncertainties (uxi , uyi )
or standard calibrators having individual mean concentra- are taken into account in order to achieve the best least-
tions xi with respective uncertainties uxi where i varies from squares linear fit.
1 to n. In order to calibrate an instrument, each of these n The uncertainties uxi in the x-axis are first propagated to
calibrators were run several times, obtaining individual the y-axis, combined with the uyi , and the total uncertainty
mean responses yi with respective uncertainties uyi where i ui values on the y-axis are used for the weighting factors
varies from 1 to n. Thus, we have n bivariate concentration- [2, 10, 17, 18, 61]: 􏽲����������������
response data pairs or calibrators (xi , yi ) with the respective 2
ui � 􏼐mO × uxi 􏼑 + 􏼐uyi 􏼑 .
2
(6)
uncertainties (uxi , uyi ).
We can apply the OLR model to these data for obtaining
The weights are calculated from ui as follows:
a calibration equation. The OLR fits a least-squares linear
−2
equation to the n pairs (xi , yi ) but does not take into ac- n × ui 􏼁
count the respective uncertainties (uxi , uyi ). wi � −2 , (7)
􏽐ni�1 ui 􏼁
The general regression equation for the OLR is as follows
(the subscript O is for the OLR model): where wi values have the following property:
n
yO 􏼐 ± uyO 􏼑 � bO 􏼐 ± ubO 􏼑 + 􏽨mO 􏼐 ± umO 􏼑 × x􏽩, (1)
􏽘 wi � n. (8)
i�1
where m is the slope, um is the resulting uncertainty in the
slope, b is the intercept, ub is the resulting uncertainty in the Thus, the UWLR fits a linear equation to the n pairs
intercept, x is the independent variable, yO is the dependent (xi , yi ) with the respective weighting factors wi as follows
variable from the OLR model, and uyO is the resulting (the subscript UW is for the UWLR model):
uncertainty in y. The following equations allow the calcu-
lations of these parameters: yUW 􏼐 ± uyUW 􏼑 � bUW 􏼐 ± ubUW 􏼑 + 􏽨mUW 􏼐 ± umUW 􏼑 × x􏽩. (9)
Journal of Spectroscopy 5

Note that this regression line will pass closer to the data (Table S4) can be similarly understood. The OLR regression
with lesser uncertainty ui . The intercept and slope variables equation from the first row of statistical information in
and their uncertainties are calculated from the following Table S4 is as follows (after the element SiO2 , subscript O is
equations: for the OLR and p is for provisional concentration; note
􏼈􏽐ni�1 wi × xi × yi 􏼁􏼉 − 􏼈n × xUW × yUW 􏼉 many decimal places are used for the regression variables in
mUW � 2 , (10) such equations, because these values are not final results, and
􏼈􏽐ni�1 wi × x2i 􏼁􏼉 − 􏽮n × xUW 􏼁 􏽯 we should not introduce rounding errors during the cal-
culation stage):
where xUW and yUW are, respectively, the weighted mean
values of the x and y variables: CSiO2Op 􏼒 ± uCSiO 􏼓 � 11.47071( ± 4.90411)
􏽶������������������� 2Op

⎪􏽴
⎧
⎪ 2 ⎪
⎫
⎪
⎨ 􏽐ni�1 yi − y􏽢iUW􏼁 ⎬ CL + 􏼔0.14325( ± 0.01605) (13)
umUW � ⎪ ×t , (11)
⎪ (n − 2)􏽐i�1 xi − xUW 􏼁 ⎪
⎩
n 2 ⎪ (n−2)
⎭
× ISiO2 􏼒 ± uISiO 􏼓􏼕.
2
where y􏽢iUW is the value of yUW for xi in equation (9):
Similarly, the UWLR equation from the second row of
bUW � yUW − mUW × xUW 􏼁, statistical information in Table S4 is as follows:
􏽶���������������������
⎪
⎧ 􏽴 ⎪
⎫ CSiO2UWp 􏼒 ± uCSiO 􏼓 � −0.01316( ± 2.83181)
⎨ 􏽐ni�1 yi − y􏽢iUW􏼁2 × 􏽐ni�1 x2i ⎪
⎪ ⎬ CL (12) 2UWp
ubUW �⎪ × t(n−2) .
⎪ n(n − 2)􏽐i�1 xi − xUW 􏼁 ⎪
⎩
n 2 ⎪
⎭ + 􏼔0.18539( ± 0.00927) (14)

The best regression equation for a calibration curve × ISiO2 􏼒 ± uISiO 􏼓􏼕.
2
should have the following characteristics (without dis-
tinguishing the subscripts O and UW): (i) intercept b small The implications of these regression equations can be
approaching to zero; (ii) slope m large; and (iii) both ub and understood from the comparison of the uncertainties of the
um small. Further, the quality of the regression, whether a intercept and slope, which are lower for the UWLR (equation
calibration curve or any other bivariate relationship, is also (14)) than for the OLR (equation (13)). This means that the
expressed as the linear regression coefficient (r; rO and rUW , uncertainty of the calculated concentration will be lower for
respectively, for the OLR and UWLR), which is ideally the UWLR than for the OLR. Correspondingly, the r value for
+1.00000 for a calibration curve [5, 18, 61]. the UWLR (0.99004, n � 60; R2 � 0.98017) is much higher
than that for the OLR (0.95229, n � 60; R2 � 0.90687;
Table S4). Similar trend in the r (and R2 ) values was obtained
5. Application of Regression Models for for all other elements except MnO (Table S4).
XRF Calibration
5.1. Original Drift-Corrected Net Intensities and GRM Con- 5.2. Matrix-Effect-Corrected Intensities and GRM Concen-
centrations: The First Set of Two Regression Equations for Each trations: The Second Set of Two Regression Equations for Each
Element. The evaluations for both regression types on the Element. Matrix correction is certainly required because the
drift-corrected net intensity-concentration (Int-Conc) re- abovementioned least-squares linear regression fits are far
lationships (Table S2) for all major elements from SiO2 to from “perfect” (r ≠ +1.00000; in fact, r < 1; n � 60; r �
P2O5 were performed (Table S4), for which the new online 0.95229–0.99638 for the OLR and r � 0.97715–0.99760 for
software BiDASys was used [61] at http: the UWLR; Table S4). There is a vast literature on the subject
//tlaloc.ier.unam.mx. BiDASys allows the application of of matrix effects in XRF and their correction procedures
the conventional OLR as well as the newly proposed UWLR [63–75]. In this study, the Lachance-Traill algorithm [73]
model [17] and provides the output of all regression pa- was used for the matrix effect correction [63, 71]. This was
®
rameters in an Excel file. Contrary to the common practice,
we will refrain from showing the numerous x-y (variable x is
done outside the XRF instrument software. In a review of the
existing algorithms, Rousseau [63] showed that the
drift-corrected net intensity “Int” and variable y is the GRM Lachance-Traill algorithm could be considered as one of the
concentration “Conc”) plots. This is because Table S4 sta- most appropriate procedures for the matrix effect correction
tistically quantifies the visual interpretation of such dia- because other algorithms have limited application range or
grams. The quality parameters (standard errors seb and sem , lack of accuracy. Thus, for each element from SiO2 to P2O5, a
uncertainty ub and um , and linear correlation coefficient r system of overdetermined equations was solved and the
and its squared value R2 parameters) are reported in resulting alpha coefficients were used to correct all in-
Table S4. Because we are using these several different quality tensities for matrix effects.
parameters, the concern against the use of solely R2 pa- From the alpha coefficients, matrix-corrected intensities
rameter [62] is not important for comparison purposes. and improved concentration values for the GRMs and their
We will explain the implications of the statistical results uncertainties were calculated iteratively under the condition
for the first element SiO2; the statistics for other elements that the convergence parameter (absolute relative difference
6 Journal of Spectroscopy

of the GRM calculated and input concentrations) for each 1.00

compositional constituent (SiO2 to P2O5) be minimized.
New regression equations for achieving the corrected

Linear correction coefficient (r)

0.99
concentrations were established from the relationship of
the calculated GRM concentrations (ConcCalc) and the
original GRM concentrations (Conc) given in Table S2, 0.98
for which the online BiDASys software [61] was used at
http://tlaloc.ier.unam.mx. These equations can be formu- 0.97
lated from the regression coefficient values given in
Table S4 (see ConcCalc-Conc rows corresponding to the Before After
OLR and UWLR). Again, we will highlight their signifi- 0.96 matrix matrix
cance for SiO2 only. correction correction
The OLR regression equation from the third row of 0.95
statistical information in Table S4 is as follows: rOLR1 rUWLR2 rOLR3 rUWLR4
Regression model
CSiO2O 􏼒 ± uCSiO 􏼓 � 1.87016( ± 4.2048)
2O SiO2 MnO Na2O
TiO2 MgO K2O
+ 􏼔0.96759( ± 0.07369) (15) Al2O3 CaO P2O5
Fe2O3t
× CSiO2Oc 􏼒 ± uCSiO 􏼓􏼕,
2Oc Figure 1: Linear correlation coefficient (r) values for the ordinary
least-squares linear regression (OLR) and uncertainty-based
where the subscripts O and c stand for the OLR model and weighted least-squares linear regression (UWLR) models for the
calculated concentration (ConcCalc), respectively. XRF calibration of major elements (SiO2 to P2O5) in rocks and
Similarly, the UWLR equation from the fourth row of minerals. OLR1: OLR model 1 for Int-Conc before matrix cor-
statistical information in Table S4 is as follows: rection; UWLR2: UWLR model 2 for Int-Conc before matrix
correction; OLR: OLR model 3 for ConcCalc-Conc after matrix
CSiO2UW 􏼒 ± uCSiO 􏼓 � −0.02185( ± 1.47356) correction; and UWLR4: UWLR model 4 for ConcCalc-Conc after
2UW
matrix correction. Symbols are shown as inset. The horizontal line
+ 􏼔1.00325( ± 0.02701) (16) at the r value of 1 represents the “ideal” or “perfect” linear fit.

× CSiO2UWc 􏼒 ± uCSiO 􏼓􏼕, Before the matrix correction, the intercepts of the Int-
2UWc
Conc regression lines were closer to zero for the UWLR
where the subscripts UW and c stand for the UWLR model (range ∼−0.013 to +0.011) than for the OLR (range ∼−2.098
and calculated concentration (ConcCalc), respectively. to +11.47) model (Table S4; Figure 2). The same is true for
Equations (15) and (16) show that the concentration the intercept values (ConcCalc-Conc relationship) after the
values from the UWLR would be more reliable (lesser un- matrix correction (∼−0.025 to +0.021 for the UWLR and
certainty values in both intercept and slope) than the OLR ∼−0.110 to +1.87 for the OLR).
model. The r value is higher for the UWLR (0.99704, n � 60; Finally, the uncertainties on both intercept and slope
R2 � 0.99408; Table S4) than the OLR (0.97710, n � 60; parameters were mostly lower for the UWLR than the OLR
R2 � 0.95472). (Table S4). We highlight these differences (lower un-
After the matrix correction, in fact most regression certainties for the UWLR) from dimensionless (free of the
equations are better because all r and R2 values are higher for measurement units) parameters δub and δum defined as
both OLR and UWLR than without the correction (Table S4; follows:
Figure 1 for r only). For the OLR, the matrix correction ub − ubUW
δub � O × 100,
increased the r values (n � 60) from 0.95229–0.99638 ubUW
(R2 � 0.90687 − 0.99277) to 0.97710–0.99992 (R2 � 0.95472 (17)
− 0.99994). Similarly, for the UWLR, this increase was from umO − umUW
0.97715–0.99760 (R2 � 0.95472 − 0.99521) to 0.99704– δum � × 100.
umUW
0.99993 (R2 � 0.99408 − 0.99986). Thus, after matrix cor-
rection, all r values increased for both OLR and UWLR. For Plots of these two parameters are presented in Figure 3. If
the UWLR, the r values approached the ideal value of ubO > ubUW , the δub will be positive, otherwise it will be
+1.00000 (Figure 1). One has to keep in mind that when the r negative. The same is true for δum . For the comparison of
values are closer to the maximum possible value of 1 (the two models OLR and UWLR before the matrix correction,
“ideal” fit), the improvement expressed by the actual (ab- the uncertainty for the UWLR were lower than the OLR for 7
solute) value of r will apparently be small. However, as long elements (positive δub and δum ), whereas for after the matrix
as the r value increases for the UWLR as compared to the correction, it was so for 8 elements (out of 10; Figure 3). The
OLR (Figure 1; Table S4), we can objectively infer that the exceptions were for 3 elements Mno, CaO, and P2O5
UWLR is a better regression model than the OLR. (negative δub and δum ) for the uncertainties before matrix
Journal of Spectroscopy 7

0.2 correction and for 2 elements MnO and MgO for those after
Before the matrix correction (Figure 3). Even for the exceptions of
matrix
the elements MnO and MgO, the UWLR values should be
correction
0.1 usable (Table S4), i.e., it is not actually necessary to resort to
Regression intercept (b)

the OLR model for these two exceptions (2 out of 10 cases).

Thus, we can use the UWLR model for all purposes.
0.0
6. Sensitivities of Major Elements
–0.1 We calculated the sensitivities as the slope of the Conc-
After IntCorr (GRM concentrations of Table S2 and matrix-
matrix
correction corrected intensities of Table S5; see Supplementary Mate-
–0.2
rials at http://tlaloc.ier.unam.mx) from the regression curve
bOLR1 bUWLR2 bOLR3 bUWLR4 (line) for all 10 elements and for both models (Table 1).
Regression model Because the r values are significantly high (all >0.961, n � 60;
Table 1) and the residuals are randomly distributed (graphs
SiO2 MnO Na2O
not shown), the straight line is the most likely, statistically
TiO2 MgO K2O
Al2O3 CaO P2O5
valid fit for the concentration-matrix-corrected intensity
Fe2O3t data [5, 17, 18]. Therefore, the slope of the regression line
represents an average sensitivity value for a given element
Figure 2: Intercept (b) values for the ordinary least-squares linear under the chosen working conditions (Table S3).
regression (OLR) and uncertainty-based weighted least-squares The intercept values were closer to zero (zero being the
linear regression (UWLR) models for the XRF calibration of major theoretically ideal intercept) for the UWLR regression
elements (SiO2 to P2O5) in rocks and minerals. Symbols are shown
(∼−0.113 to +0.104; Table 1) as compared to the OLR
as inset. For abbreviations, see Figure 1. Note some intercept values
plotted outside the graph; this is indicated by arrows next to the
(∼−47.8 to +12.3; Table 1). The sensitivity values represented
data point. The horizontal line at the intercept value of zero by the slopes of the regression lines (Table 1) were generally
represents the “ideal” intercept. similar for both models (∼2.69–46.17 kc·s−1·%−1 for the OLR
and ∼2.78–59.69 kc·s−1·%−1 for the UWLR). The sensitivity
actually depends on the measuring conditions (Table S3),
100 which were the same for both models.
For the matrix-corrected intensity-concentration
(IntCorr-Conc) regressions, the parameters are listed in
Before After Table 2. All intercepts for the UWLR model, without ex-
50 matrix matrix ception, were closer to zero as compared to the OLR model.
correction correction This confirms the superiority of the UWLR model.
δub or δum

7. Application to Rock Matrices

0
The calibrations achieved in this work (Table S4) were ap-
plied to the analysis of four GRMs (attapulgite or Fuller’s
earth clay ATT1; bentonite clay CSB1; granite GH; and
–50 tonalite TLM1) taken as “unknown” samples. These GRMs,
bOUW12 mOUW12 bOUW34 mOUW34 having similarly complex matrices as the calibration sam-
b and m ratios for regression model ples, were not included in the calibrations because their
SiO2 MnO Na2O mean values were available only from early description or
TiO2 MgO K2O compilations (for ATT1 and CSB1 [76]; for GH [77]; and for
Al2O3 CaO P2O5 TLM1 [78]). We were unsuccessful in complementing these
Fe2O3t “old” concentration values with newer ones for these GRMs.
Therefore, these GRMs were used as unknown samples. They
Figure 3: New parameters δub (for the intercept) and δum (for the were analysed in exactly the same manner as the calibration
slope) for the evaluation of intercept (b) and slope (m) of two samples.
regression models (OLR: ordinary least-squares linear regression
All calculations for the unknown samples were done
and UWLR: uncertainty-based weighted least-squares linear re-
gression) before (OUW12) and after (OUW34) the matrix cor- outside the instrumental software. The drift-corrected net
rection. The horizontal solid line at the y value of zero represents intensities and the corresponding uncertainties were pro-
the line with no difference in the uncertainties of the two models. cessed from the first set of two regression equations
The arrows indicate that these data plotted above the scale are used (Int-Conc OLR and UWLR models; Table S4) to obtain
for the diagram. provisional concentration and uncertainty values. The
8 Journal of Spectroscopy

Table 1: Instrumental sensitivities (x-y: concentration-matrix-corrected intensity (Conc-IntCorr) regression model; mean and 99%
uncertainty) for major elements.
Quality of
Regression Regression equation parameters regression
Number of data pairs equation
Element
(calibrators)
Variables Intercept Slope
Model seb ub sem um r R2
x-y (b) (m)
Conc-
OLR 60 −47.8441 13.1332 34.9772 6.3592 0.2401 0.6395 0.96105 0.92362
IntCorr
SiO2
Conc-
UWLR 60 0.1035 5.3870 14.3471 5.3476 0.0985 0.2623 0.99157 0.98321
IntCorr
Conc-
OLR 60 −0.2298 0.3000 0.7991 13.2255 0.2889 0.7694 0.98644 0.97307
IntCorr
TiO2
Conc-
UWLR 60 −0.0188 0.2637 0.7024 12.3181 0.2539 0.6763 0.99113 0.98235
IntCorr
Conc-
OLR 60 8.1398 4.1673 11.0987 6.1191 0.2269 0.6043 0.96237 0.92615
IntCorr
Al2O3
Conc-
UWLR 60 0.0594 3.0552 8.1369 7.0715 0.1663 0.4430 0.98557 0.97135
IntCorr
Conc-
OLR 60 12.2628 2.4739 6.5886 4.3943 0.1337 0.3561 0.97419 0.94904
IntCorr
Fe2O3t
Conc-
UWLR 60 0.0715 2.7723 7.3835 5.4403 0.1498 0.3991 0.98112 0.96259
IntCorr
Conc-
OLR 60 2.0367 0.2963 0.7891 46.1686 0.6958 1.8531 0.99348 0.98700
IntCorr
MnO
Conc-
UWLR 60 0.0147 0.7535 2.0069 59.6875 1.7695 4.7127 0.99160 0.98328
IntCorr
Conc-
OLR 60 1.1050 1.0231 2.7246 3.6928 0.0901 0.2400 0.98317 0.96662
IntCorr
MgO
Conc-
UWLR 60 0.0108 0.9641 2.5676 3.9863 0.0849 0.2262 0.98745 0.97505
IntCorr
Conc-
OLR 60 −14.3055 3.5661 9.4974 15.4215 0.3371 0.8977 0.98643 0.97304
IntCorr
CaO
Conc-
UWLR 60 −0.0121 5.6252 14.9814 11.5666 0.5317 1.4160 0.98657 0.97332
IntCorr
Conc-
OLR 60 0.0166 0.1553 0.4137 2.6888 0.0571 0.1521 0.98716 0.97449
IntCorr
Na2O
Conc-
UWLR 60 −0.0111 0.1154 0.3073 2.7843 0.0424 0.1130 0.99344 0.98692
IntCorr
Conc-
OLR 60 1.2200 0.5896 1.5702 13.8296 0.1431 0.3811 0.99691 0.99383
IntCorr
K2O
Conc-
UWLR 60 −0.1130 0.5491 1.4623 14.4829 0.1333 0.3549 0.99793 0.99587
IntCorr
Conc-
OLR 60 −1.7376 0.5433 1.4469 39.1055 1.2932 3.4442 0.96972 0.94035
IntCorr
P2O5
Conc-
UWLR 60 0.0626 0.7039 1.8747 29.4822 1.6756 4.4624 0.96926 0.93946
IntCorr
b, intercept; se, standard error; u, uncertainty at 99%; m, slope; r, linear correlation coefficient; R2 , squared linear correlation coefficient.

provisional concentrations were then used to obtain matrix The results are listed in Table 3 and compared with the
corrections for each sample. The method was iteratively literature compilations [75–77]. On the other hand, because
applied with the newer concentrations to obtain the final 99% uncertainties were not reported in the original com-
calculated concentration values (Table 3). These calculated pilations, they were computed for the comparison from the
concentration values were used to compute the final mean standard deviation, number of determinations, and ap-
concentrations (x) and 99% uncertainties of the mean (u99 ) propriate two-sided t values at 99% confidence level [2, 18].
for each sample from the second sets of regression equa- Firstly, although the mean concentration values de-
tions (ConcCalc-Conc, OLR and UWLR models; Table S4). termined by the OLR and UWLR models showed a general
The loss on ignition (LOI) was required to optimise the agreement, the 99% uncertainty values (u99 ; Table 3) were
final results. generally lower for the UWLR models, which clearly
Journal of Spectroscopy 9

Table 2: Regression (x-y): matrix-corrected intensity-concentration (IntCorr-Conc) parameters.

Quality of
Regression Regression equation parameters regression
Number of data pairs equation
Element
(calibrators)
Variables Intercept Slope
Model seb ub sem um r R2
x-y (b) (m)
IntCorr-
OLR 60 10.8541 1.6762 4.4641 0.1452 0.0055 0.0146 0.96105 0.92362
Conc
SiO2
IntCorr-
UWLR 60 −0.0148 0.9765 2.6006 0.1848 0.0032 0.0085 0.99157 0.98321
Conc
IntCorr-
OLR 60 0.0344 0.0220 0.0587 0.0736 0.0016 0.0043 0.98644 0.97307
Conc
TiO2
IntCorr-
UWLR 60 0.0020 0.0204 0.0542 0.0801 0.0015 0.0040 0.99114 0.98235
Conc
IntCorr-
OLR 60 −0.1675 0.6763 1.8011 0.1514 0.0056 0.0150 0.96237 0.92615
Conc
Al2O3
IntCorr-
UWLR 60 −0.0070 0.4419 1.1769 0.1397 0.0037 0.0098 0.98557 0.97135
Conc
IntCorr-
OLR 60 −2.0576 0.5960 1.5873 0.2160 0.0066 0.0175 0.97419 0.94904
Conc
Fe2O3t
IntCorr-
UWLR 60 −0.0092 0.5458 1.4536 0.1772 0.0060 0.0160 0.98112 0.96259
Conc
IntCorr-
OLR 60 −0.0412 0.0067 0.0178 0.0214 0.0003 0.0009 0.99348 0.98700
Conc
MnO
IntCorr-
UWLR 60 0.0001 0.0134 0.0356 0.0165 0.0007 0.0017 0.99160 0.98328
Conc
IntCorr-
OLR 60 −0.0977 0.2748 0.7319 0.2618 0.0064 0.0170 0.98317 0.96662
Conc
MgO
IntCorr-
UWLR 60 −0.0014 0.2692 0.7168 0.2376 0.0063 0.0167 0.98745 0.97505
Conc
IntCorr-
OLR 60 1.0561 0.2173 0.5789 0.0631 0.0014 0.0037 0.98643 0.97304
Conc
CaO
IntCorr-
UWLR 60 0.0064 0.4616 1.2293 0.0853 0.0029 0.0078 0.98657 0.97331
Conc
IntCorr-
OLR 60 0.0447 0.0567 0.1511 0.3624 0.0077 0.0205 0.98716 0.97449
Conc
Na2O
IntCorr-
UWLR 60 0.0054 0.0421 0.1122 0.3558 0.0057 0.0152 0.99345 0.98693
Conc
IntCorr-
OLR 60 −0.0720 0.0430 0.1146 0.0719 0.0007 0.0020 0.99691 0.99383
Conc
K2O
IntCorr-
UWLR 60 0.0082 0.0391 0.1042 0.0687 0.0007 0.0018 0.99793 0.99587
Conc
IntCorr-
OLR 60 0.0537 0.0128 0.0341 0.0241 0.0008 0.0021 0.96972 0.94035
Conc
P2O5
IntCorr-
UWLR 60 0.0010 0.0202 0.0538 0.0320 0.0013 0.0033 0.96929 0.93952
Conc
b, intercept; se, standard error; u, uncertainty at 99%; m, slope; r, linear correlation coefficient; R2 , squared linear correlation coefficient.

indicates that this model should be used routinely, instead of results for established GRMs and their comparison to other
the conventional OLR model. Secondly, there is also a laboratories.
general agreement among all mean values, especially for
granite GH and tonalite TLM1. The two clay samples (ATT1 8. Computer Program XRFCalcUnknown
and CSB1) showed some differences with the preliminary
values obtained by the originators of these GRMs [75]. These An online computer program JSpectrom_XRFCalcUnknown
values for comparison were obtained in only one laboratory. will be available at our server https://tlaloc.ier.unam.mx for
The errors (uncertainties) reported in the literature were use for unknown samples, which will guide other users to
underestimated, because they did not include those resulting achieve the UWLR calibration outside of the instrumental
from the calibrations. Furthermore, the accuracy data of the software and its routine application to unknown samples. This
originator’s laboratory were not reported [75], such as the program incorporates the iteration process to achieve reliable
 10

Table 3: Application of the XRF calibrations for the determination of major elements (mean concentration in % m/m; mass/mass unit expressed in percentage, habitually called wt.% and
99% uncertainty u99 or simply u) in four rock samples (data reported as rounded value following flexible criteria proposed by [18]) and their comparison with the literature data.
SiO2 TiO2 Al2O3 Fe2O3t MnO MgO CaO Na2O K2O P2O5
Rock Method (ref.)
x u99 x u99 x u99 x u99 x u99 x u99 x u99 x u99 x u99 x u99
OLR 61.7 6.1 0.4961 0.0173 10.00 0.49 4.246 0.292 0.02394 0.00188 5.373 0.377 2.139 0.086 0.083 0.094 0.7777 0.0378 0.8074 0.0134
ATT1 UWLR 62.30 2.23 0.4951 0.0143 9.666 0.321 4.402 0.235 0.02416 0.00269 6.17 0.52 2.144 0.076 0.058 0.065 0.7781 0.0366 0.7828 0.0125
[76] 59.6 0.8 0.49 0.015 9.50 0.29 3.31 0.06 0.025 0.018 9.14 0.06 1.87 0.03 0.1 0.03 0.86 0.06 0.76 0.01
OLR 52.7 5.5 0.1448 0.0157 24.60 0.71 3.493 0.289 0.01408 0.00188 1.523 0.270 1.117 0.085 2.126 0.119 0.4169 0.0374 0.0383 0.0060
CSB1 UWLR 52.76 2.04 0.1343 0.0130 24.40 0.48 3.564 0.233 0.01398 0.00268 1.604 0.398 1.105 0.075 2.123 0.083 0.4337 0.0361 0.0357 0.0057
[76] 55.3 0.8 0.16 0.03 21.75 0.18 3.88 0.03 0.020 0.003 1.94 0.08 1.28 0.06 2.07 0.20 0.54 0.03 — —
OLR 73.3 6.5 0.0968 0.0157 13.98 0.53 1.336 0.285 0.05084 0.00190 0.074 0.262 0.843 0.084 3.618 0.151 5.302 0.059 0.0119 0.0059
GH UWLR 73.62 2.38 0.0861 0.0129 13.714 0.349 1.466 0.229 0.05064 0.00271 0.038 0.390 0.828 0.074 3.625 0.105 5.216 0.057 0.0091 0.0057
[77] 75.8 0.39 0.08 0.07 12.50 0.16 1.34 0.2 0.05 0.009 0.03 0.18 0.69 0.13 3.85 0.11 4.76 0.05 0.08 0.03
OLR 57.7 5.7 0.8497 0.0200 17.38 0.58 7.637 0.306 0.11614 0.00199 3.972 0.319 6.807 0.100 2.675 0.130 1.7585 0.0403 0.1378 0.0063
TLM1 UWLR 57.67 2.11 0.8402 0.0165 17.046 0.386 7.682 0.247 0.11592 0.00281 4.281 0.447 6.821 0.088 2.667 0.090 1.742 0.0390 0.1336 0.0060
[78] 58.85 0.35 0.820 0.013 17.48 0.25 7.67 0.15 0.105 0.013 3.3 0.08 6.67 0.08 2.98 0.07 1.67 0.08 0.15 0.013
Journal of Spectroscopy
Journal of Spectroscopy 11

concentrations as demonstrated in this work. Example input and Nanotechnology National Research Laboratory
data files and a ReadMe document are provided to facilitate (LINAN), Carbon Nanostructures and Two-Dimensional
the application of JSpectrom_XRFCalcUnknown. One im- Systems Laboratory at IPICYT, and Dr. Emilio Muñoz-
portant aspect of the program is that for a sample to be Sandoval for providing access to the required facilities. M.
identified as an “unknown” sample, the value of LOI (loss on Abdelaly Rivera-Gómez is grateful to CTIC and DGAPA for
ignition in percent) should be input in the first sheet of the a postdoctoral fellowship at the ICML-UNAM. Darı́o
measured intensity file. Torres-Sánchez and Mauricio Rosales-Rivera thank CON-
A novel aspect of the present work is that total 99% ACYT for the doctoral fellowship. The GRM compilation
uncertainty can be calculated for individual datum in a given was initiated long ago in our group by the participation of R.
sample (treated as unknown; Table 3). This innovation if put González-Ramı́rez although the bulk of the work was carried
into practice can entirely change the geochemical literature, out by the present authors. We are grateful to the IER-
and in fact make geochemistry a more quantitative science. UNAM library personnel for efficiently providing some of
Further, if an appropriate GRM is analysed as unknown and the literature materials for compilation and to Alfredo
the analytical data (both mean and total uncertainly) are Quiroz-Ruiz for the maintenance of the computing facility at
reported along with the field samples, the data accuracy can IER-UNAM. Diego Villanueva-López helped us during the
be statistically judged from such reports. pressed powder pellet preparation and for checking the
correctness of the information in GRM databases.
9. Conclusions
The XRF spectrometer calibrated under both the OLR and Supplementary Materials
UWLR models clearly showed that the UWLR provides Five tables (Tables S1–S5) are provided. (Supplementary
more reliable results (lower uncertainty estimates) than the Materials)
OLR model commonly practiced for most XRF in-
struments. The sensitivity and LOD values presented for
both models also supported the use of the UWLR model. References
The UWLR model should therefore be used routinely in [1] H. R. Rollinson, Using Geochemical Data: Evaluation, Pre-
such calibrations. The use of a large number of well- sentation, Interpretation, Longman Scientific Technical, Essex,
characterized GRMs is also recommended for this pur- UK, 1993.
pose as illustrated in the present work. The application of [2] J. N. Miller and J. C. Miller, Statistics and Chemometrics for
our procedure was well documented for the analysis of Analytical Chemistry, Pearson Prentice Hall, Essex, UK, 2010.
similarly complex rock matrices. The reporting of total [3] R. G. Brereton, “Chemometrics in analytical chemistry. A
uncertainty values for individual datum is highly recom- review,” Analyst, vol. 112, pp. 1635–1657, 1987.
mended for all future geochemical research. This work for [4] M. Guevara, S. P. Verma, F. Velasco-Tapia, R. Lozano-Santa
the XRF shows that such a practice is easy to achieve in any Cruz, and P. Girón, “Comparison of linear regression models
for quantitative geochemical analysis: an example using x-ray
other analytical calibration procedures. As the major
fluorescence spectrometry,” Geostandards and Geoanalytical
conclusion, we can confirm that the statistically coherent Research, vol. 29, no. 3, pp. 271–284, 2005.
WLR model was shown to perform better than the fre- [5] P. R. Bevington, Data Reduction and Error Analysis for the
quently used conventional statistically incoherent OLR Physical Sciences, Mc-Graw Hill Book Company, New York,
model. NY, USA, 1969.
[6] D. York, “Least squares fitting of a straight line with correlated
Data Availability errors,” Earth and Planetary Science Letters, vol. 5, pp. 320–
324, 1969.
The list of all compiled references (Table S1) will be available at [7] A. H. Kalantar, “Weighted least squares evaluation of slope
http://tlaloc.ier.unam.mx. These references are not included from data having errors in both axes,” Trends in Analytical
with the manuscript because they are too many (∼480). Chemistry, vol. 9, no. 5, pp. 149–151, 1990.
Similarly, as stated, the online program JSpec- [8] K. L. Mahon, “The New “York” regression: application of an
trom_XRFCalcUnknown will also be added onto this web portal improved statistical method to geochemistry,” International
http://tlaloc.ier.unam.mx. This program needs to be available Geology Review, vol. 38, no. 4, pp. 293–303, 1996.
[9] M. E. Zorn, R. D. Gibbons, and W. C. Sonzogni, “Weighted
online for future use; it cannot be submitted to the journal.
least-squares approach to calculating limits of detection and
quantification by modeling variability as a function of con-
Conflicts of Interest centration,” Analytical Chemistry, vol. 69, no. 15, pp. 3069–
3075, 1997.
The authors declare that they have no conflicts of interest. [10] N. R. Draper and H. Smith, Applied Regression Analysis, John
Wiley & Sons, New York, NY, USA, 1998.
Acknowledgments [11] A. Schick, “Improving weighted least-squares estimates in
heteroscedastic linear regression when the variance is a
This work was supported through the Newton Advanced function of the mean response,” Journal of Statistical Planning
Fellowship Award (grant NA160116) of the Royal Society, and Inference, vol. 76, no. 1-2, pp. 127–144, 1999.
U.K., to the second author (SKV) and from the sabbatical [12] A. Sayago, M. Boccio, and A. G. Asuero, “Fitting straight lines
stay of SPV at IPICYT. We are grateful to the Nanoscience with replicated observations by linear regression: the least
12 Journal of Spectroscopy

squares postulates,” Critical Review of Analytical Chemistry, [29] S. P. Verma, T. Besch, M. Guevara, and B. Schulz-Dobrich,
vol. 34, no. 1, pp. 39–50, 2004. “Determination of twelve trace elements in twenty-seven and
[13] A. Sayago and A. G. Asuero, “Fitting straight lines with ten major elements in twenty-three geochemical reference
replicated observations by linear regression: Part II. Testing samples by X-Ray fluorescence spectrometry,” Geostandards
for homogeneity of variances,” Critical Review of Analytical Newsletter, vol. 16, no. 2, pp. 301–309, 1992.
Chemistry, vol. 34, no. 3-4, pp. 133–146, 2004. [30] X. Wang, G. Li, Q. Zhang, and Y. Wang, “Determination of
[14] A. G. Asuero, A. Sayago, and A. G. González, “The correlation major/minor and trace elements in seamount phosphorite by
coefficient: an overview,” Critical Review of Analytical XRF spectrometry,” Geostandards and Geoanalytical Re-
Chemistry, vol. 36, no. 1, pp. 41–59, 2006. search, vol. 28, no. 1, pp. 81–88, 2004.
[15] J. Tellinghuisen, “Weighted least-squares in calibration: what [31] K. Govindaraju, “1987 compilation report on Ailsa Craig
difference does it make?,” Analyst, vol. 132, no. 6, pp. 536–543, Granite AC-E with the participation of 128 GIT-IWG labo-
2007. ratories,” Geostandards Newsletter, vol. 11, no. 2, pp. 203–255,
[16] S. P. Verma, L. Dı́az-González, and R. González-Ramı́rez, 1987.
“Relative efficiency of single-outlier discordancy tests for [32] K. Govindaraju, “Report (1980) on three GIT-IWG rock
processing geochemical data on reference materials and ap- reference samples: anorthosite from Greenland, AN-G;
plication to instrumental calibration by a weighted least- basalte d’ Essey-la-Côte, BE-N; granite de Beauvoir, MA-
squares linear regression model,” Geostandards and Geo- N,” Geostandards Newsletter, vol. 4, no. 1, pp. 49–138, 1980.
analytical Research, vol. 33, no. 1, pp. 29–49, 2009. [33] E. S. Gladney and I. Roelandts, “1987 compilation of ele-
[17] S. P. Verma, “Geochemometrics,” Revista Mexicana de mental concentration data for USGS BHVO-1, MAG-1, QLO-
Ciencias Geológicas, vol. 29, no. 1, pp. 276–298, 2012. 1, RGM-1, SCo-1, SDC-1, SGR-1, and STM-1,” Geostandards
[18] S. P. Verma, Análisis Estadı́stico de Datos Composicionales, Newsletter, vol. 12, no. 2, pp. 253–262, 1988.
Universidad Nacional Autónoma de México, CDMX, Mexico, [34] K. Govindaraju, “Report (1967-1981) on four ANRT rock
2016. reference samples: diorite DR-N, serpentine UB-N, bauxite
[19] P. J. Potts, A Handbook of Silicate Rock Analysis, Blackie, BX-N, disthene DT-N,” Geostandards Newsletter, vol. 6, no. 1,
Glasgow, UK, 1987. pp. 91–159, 1982.
[20] P. J. Potts and P. C. Webb, “X-ray fluorescence spectrometry,” [35] E. S. Gladney and I. Roelandts, “1987 compilation of elemental
Journal of Geochemical Exploration, vol. 44, no. 1–3, concentration data for USGS BIR-1, DNC-1 and W-2,” Geo-
pp. 251–296, 1992. standards Newsletter, vol. 12, no. 1, pp. 63–118, 1988.
[21] M. El Maghraoui, J.-L. Joron, J. Etoubleau, P. Cambon, and [36] S. Abbey, C. R. McLeod, and W. Liang-Guo, “FeR-1, FeR-2,
M. Treuil, “Determination of forty four major and trace el- FeR-3 and FeR-4 Four Canadian iron-formation samples
ements in GPMA magmatic rock reference materials using prepared for use as reference materials,” Geological Survey of
x-ray fluorescence spectrometry (XRF) and instrumental Canada Report, Geological Survey of Canada, Ottawa, ON,
neutron activation analysis (INAA),” Geostandards Newslet- Canada, 1983.
ter: Journal of Geostandards and Geoanalysis, vol. 23, no. 1, [37] K. Govindaraju, “Report (1973-1984) in two ANRTgeochemical
pp. 59–68, 1999. reference samples: granite GS-N and Potash Feldspar FK-N,”
[22] K. Tani, H. Kawabata, Q. Chang, K. Sato, and Y. Tatsumi, Geostandards Newsletter, vol. 8, no. 2, pp. 173–206, 1984.
“Quantitative analyses of silicate rock major and trace ele- [38] E. S. Gladney, C. E. Burns, and I. Roelandts, “1982 compi-
ments by X-ray fluorescence spectrometer: evaluation of lation of elemental concentration data for the United States
analytical precision and sample preparation,” Frontiers in Geological Survey’s geochemical exploration reference sam-
Research of Earth Evolution, vol. 2, pp. 1–8, 2004. ples GXR-1 to GXR-6,” Geostandards Newsletter, vol. 8, no. 2,
[23] J. Enzweiler and M. A. Vendemiatto, “Analysis of sediments pp. 119–154, 1984.
and soils by x-ray fluorescence spectrometry using matrix [39] E. S. Gladney and I. Roelandts, “1988 compilation of ele-
corrections based on fundamental parameters,” Geo- mental concentration data for USGS geochemical exploration
standards and Geoanalytical Research, vol. 28, no. 1, reference materials GXR-1 to GXR-6,” Geostandards News-
pp. 103–112, 2005. letter, vol. 14, no. 1, pp. 21–118, 1990.
[24] L. P. Bédard, “Neutron activation analysis, atomic absorption [40] K. Govindaraju, “Report (1984) on two GIT-IWG geo-
and x-ray fluorescence spectrometry review for 2004-2005,” chemical reference samples: albite from Italy, AL-I and Iron
Geostandards and Geoanalytical Research, vol. 30, no. 3, Formation sample from Greenland, IF-G,” Geostandards
pp. 183–186, 2006. Newsletter, vol. 8, no. 1, pp. 63–113, 1984.
[25] K. Nakayama, Y. Shibata, and T. Nakamura, “Glass beads/x- [41] N. Imai, S. Terashima, S. Itoh, and A. Ando, “1994 compi-
ray fluorescence analysis of 42 components in felsic rocks,” X- lation of analytical data for minor and trace elements in
Ray Spectrometry, vol. 36, no. 2, pp. 130–140, 2007. seventeen GSJ geochemical reference samples, “igneous rock
[26] W. Wu, T. Xu, Q. Hao, Q. Wang, S. Zhang, and C. Zhao, series”,” Geostandards and Geoanalytical Research, vol. 19,
“Applications of x-ray fluorescence analysis of rare earths no. 2, pp. 135–213, 1995.
in China,” Journal of the Rare Earths, vol. 28, pp. 30–36, [42] S. Terashima, “Elemental concentrations in nine new GSJ rock
2010. reference samples,” Geostandards Newsletter, vol. 14, no. 1,
[27] H. Mashima, “XRF analyses of major and trace elements in pp. 1–5, 1990.
silicate rocks calibrated with synthetic standard samples,” [43] S. Terashima, S. Itoh, M. Ujiie, H. Kamioka, T. Tanaka, and
Natural Resources of Environment and Humans, vol. 6, H. Hattori, “Three new GSJ rock reference samples: rhyolite
pp. 39–50, 2016. JR-3, gabbro JGb-2 and hornblendite JH-1,” Geostandards
[28] D. Robinson and M. C. Bennett, “XRF determination of 19 Newsletter, vol. 17, no. 1, pp. 1–4, 1993.
trace elements in international geochemical reference [44] S. Abbey, “Reference materials: rock samples SY-2, SY-3,
samples,” Geostandards Newsletter, vol. 5, no. 2, pp. 175–181, MRG-1,” Energy, Mines and Resources Canada Report,
1981. Natural Resources Canada, Ottawa, ON, Canada, 1979.
Journal of Spectroscopy 13

[45] K. Govindaraju, “Report (1968-1978) on two mica reference [62] J. B. Willet and J. D. Singer, “Another cautionary note about
samples: biotite Mica-Fe and phlogopite Mica-Mg,” Geo- R2: its use in weighted least-squares regression analysis,”
standards Newsletter, vol. 3, no. 1, pp. 3–24, 1979. American Statistician, vol. 42, no. 3, pp. 236–238, 1988.
[46] T. W. Steele and R. G. Hansen, “Major element data (1966- [63] R. M. Rousseau, “Corrections for matrix effects in X-ray
1978) for the six “Nimroc” reference samples,” Geostandards fluorescence analysis—a tutorial,” Spectrochimica Acta Part B:
Newsletter, vol. 3, no. 2, pp. 135–172, 1979. Atomic Spectroscopy, vol. 61, no. 7, pp. 759–777, 2006.
[47] E. S. Gladney, E. A. Jones, E. J. Nickell, and I. Roelandts, “1988 [64] T. Shiraiwa and N. Fujino, “Theoretical calculation of fluo-
compilation of elemental concentration data for USGS DTS-1, rescent x-ray intensities in fluorescent x-ray spectrochemical
G-1, PCC-1, and W-1,” Geostandards Newsletter, vol. 15, analysis,” Japanese Journal of Applied Physics, vol. 5, no. 10,
no. 2, pp. 199–396, 1991. pp. 886–899, 1966.
[48] V. Barnett and T. Lewis, Outliers in Statistical Data, John [65] R. Rousseau and F. Claisse, “Theoretical alpha coefficients for
Wiley & Sons, Chichester, UK, 1994. the Claisse-Quintin relation for x-ray spectrochemical anal-
[49] K. Hayes, A. Kinsella, and N. Coffey, “A note on the use of ysis,” X-Ray Spectrometry, vol. 3, no. 1, pp. 31–36, 1974.
outlier criteria in Ontario laboratory quality control schemes,” [66] R. M. Rousseau, “Fundamental algorithm between concen-
Clinical Biochemistry, vol. 40, no. 3-4, pp. 147–152, 2007. tration and intensity in XRF analysis 2-Practical application,”
[50] S. P. Verma, L. Dı́az-González, J. A. Pérez-Garza, and X-Ray Spectrometry, vol. 13, no. 3, pp. 121–125, 1984.
M. Rosales-Rivera, “Quality control in geochemistry from a [67] R. M. Rousseau, J. P. Willis, and A. R. Duncan, “Practical XRF
comparison of four central tendency and five dispersion es- calibration procedures for major and trace elements,” X-Ray
timators and example of a geochemical reference material,” Spectrometry, vol. 25, no. 4, pp. 179–189, 1996.
Arabian Journal of Geosciences, vol. 9, p. 740, 2016. [68] A. J. Klimasara, “XRF analysis—theory, experiment, and
[51] S. P. Verma, L. Dı́az-González, J. A. Pérez-Garza, and regression,” in Proceedings of X-Ray Conference (DXC) on
M. Rosales-Rivera, “Erratum to: quality control in geo- Applications of X-Ray Analysis, Denver, CO, USA, August
chemistry from a comparison of four central tendency and 1997.
five dispersion estimators and example of a geochemical [69] B. Tan and W. Sun, “Correction method for the matrix effect
reference material,” Arabian Journal of Geosciences, vol. 10, in x-ray fluorescence spectrometric analysis,” X-Ray Spec-
p. 24, 2017. trometry, vol. 27, no. 2, pp. 95–104, 1998.
[52] S. P. Verma, M. Rosales-Rivera, L. Dı́az-González, and [70] B. I. Kitov, “Calculation features of the fundamental pa-
A. Quiroz-Ruiz, “Improved composition of Hawaiian basalt rameter method in XRF,” X-Ray Spectrometry, vol. 29, no. 4,
BHVO-1 from the application of two new and three con- pp. 285–290, 2000.
ventional recursive discordancy tests,” Turkish Journal of [71] A. J. Klimasara, “Logical steps in the automated Lachamp–
Earth Science, vol. 26, no. 5, pp. 331–353, 2017. Triall XRF matrix correction method utilizing an electronic
[53] S. P. Verma, R. Cruz-Huicochea, and L. Dı́az-González, spreadsheet,” in Advances in XRF Analysis, vol. 42, pp. 53–83,
“Univariate data analysis system: deciphering mean compo- JCPDS-International Centre for Diffraction Data, Newtown
sitions of island and continental arc magmas, and influence of Square, PA, USA, 2000.
underlying crust,” International Geology Review, vol. 55, [72] R. M. Rousseau, “Correction for long-term instrumental
no. 15, pp. 1922–1940, 2013. drift,” X-Ray Spectrom, vol. 31, no. 6, pp. 401–407, 2002.
[54] M. Rosales Rivera, Desarrollo de herramientas estadı́sticas [73] G. R. Lachance and R. J. Traill, “A practical solution to the
computacionales con nuevos valores crı́ticos generados por matrix problem in X-ray analysis,” Canadian Journal of
simulación computacional, Universidad Autónoma del Estado Spectroscopy, vol. 11, pp. 43–48, 1966.
de Morelos, Cuernavaca, Morelos, Mexico, 2018. [74] J. P. Willis and G. R. Lachance, “Comparison between some
[55] K. P. Jochum, U. Weis, B. Schwager et al., “Reference values common influence coefficient algorithms,” X-Ray Spectrom-
following ISO guidelines for frequently requested rock ref- etry, vol. 33, no. 3, pp. 181–188, 2004.
erence materials,” Geostandards and Geoanalytical Research, [75] J. P. Willis and G. R. Lachance, “A new approach to correcting
vol. 40, no. 3, pp. 333–350, 2016. theoretical emitted intensities for absorption and enhance-
[56] F. Velasco-Tapia, M. Guevara, and S. P. Verma, “Evaluation of ment effects,” X-Ray Spectrometry, vol. 33, no. 3, pp. 204–211,
concentration data in geochemical reference materials,” 2004.
[76] J. W. Hosterman and F. J. Flanagan, “USGS reference samples
Chemie der Erde, vol. 61, no. 1, pp. 69–91, 2001.
Attapulgite ATT-1 and Bentonite CSB-1,” Geostandards and
[57] B. Rosner, “On the detection of many outliers,” Technometrics,
Geoanalytical Research, vol. 11, no. 1, pp. 1–9, 1987.
vol. 17, no. 2, pp. 221–227, 1975.
[77] K. Govindaraju, “1995 working values with confidence limits
[58] F. E. Grubbs and G. Beck, “Extension of sample sizes and
for twenty-six CRPG, ANRT and IWG-GIT geostandards,”
percentage points for significance tests of outlying observa-
Geostandards and Geoanalytical Research, vol. 19, pp. 1–32,
tions,” Technometrics, vol. 14, no. 4, pp. 847–854, 1972.
1995.
[59] R. B. Jain and L. A. Pingel, “On the robustness of recursive
[78] F. J. Flanagan, “Rock reference samples, san marcos gabbro,
outlier detection procedures to nonnormality,” Communi-
GSM-1 and lakeview mountain tonalite, TLM-1,” Geo-
cations in Statistics - Theory and Methods, vol. 10, no. 13,
standards and Geoanalytical Research, vol. 10, no. 2,
pp. 1323–1334, 1981.
pp. 111–119, 1986.
[60] R. B. Jain, “Detecting outliers: power and some other con-
siderations,” Communications in Statistics - Theory and
Methods, vol. 10, no. 22, pp. 2299–2314, 1981.
[61] M. Rosales-Rivera, L. Dı́az-González, and S. P. Verma, “A new
online computer program (BiDASys) for ordinary and un-
certainty weighted least-squares linear regressions: case
studies from food chemistry,” Revista Mexicana de Ingenierı́a
Quı́mica, vol. 17, no. 2, pp. 507–522, 2018.
 Nanomaterial
Nanomaterials
Journal of

Journal of
The Scientific
Photoenergy
International Journal of
Analytical Methods Journal of

Hindawi
in Chemistry
Hindawi
World Journal
Hindawi Publishing Corporation
Applied Chemistry
Hindawi Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 http://www.hindawi.com
www.hindawi.com Volume 2018
2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Advances in International Journal of

Physical Chemistry
Hindawi
Medicinal Chemistry
Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Submit your manuscripts at

www.hindawi.com

Bioinorganic Chemistry Journal of

and Applications
Hindawi
Materials
Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Advances in Journal of BioMed International Journal of International Journal of

Tribology
Hindawi
Chemistry
Hindawi
Research International
Hindawi
Spectroscopy
Hindawi
Electrochemistry
Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

International Journal of Journal of Journal of Enzyme Biochemistry

Analytical Chemistry
Hindawi
Spectroscopy
Hindawi
Nanotechnology
Hindawi
Research
Hindawi
Research International
Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018