chromatographic peak shape analysis and modeling
TRANSCRIPT
Louisiana State UniversityLSU Digital Commons
LSU Historical Dissertations and Theses Graduate School
1990
Chromatographic Peak Shape Analysis andModeling.Mark Stephen JeansonneLouisiana State University and Agricultural & Mechanical College
Follow this and additional works at: https://digitalcommons.lsu.edu/gradschool_disstheses
This Dissertation is brought to you for free and open access by the Graduate School at LSU Digital Commons. It has been accepted for inclusion inLSU Historical Dissertations and Theses by an authorized administrator of LSU Digital Commons. For more information, please [email protected].
Recommended CitationJeansonne, Mark Stephen, "Chromatographic Peak Shape Analysis and Modeling." (1990). LSU Historical Dissertations and Theses.5061.https://digitalcommons.lsu.edu/gradschool_disstheses/5061
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer.
The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UM I a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion.
Oversize m aterials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand corner and continuing from left to right in equal sections with small overlaps. Each original is also photographed in one exposure and is included in reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced xerographically in this copy. Higher quality 6" x 9" black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact UMI directly to order.
University Microfilms In te rn a t io n a l A Bell & Howell In fo rm at ion C o m p a n y
3 0 0 N or th Z e e b R o a d . A nn Arbor. Ml 4 8 1 0 6 - 1 3 4 6 USA 3 1 3 /7 6 1 - 4 7 0 0 8 0 0 5 2 1 - 0 6 0 0
Order N um ber 9123204
Chrom atographic peak shape analysis and m odeling
Jeansonne, Mark Stephen, Ph.D.
The Louisiana State University and Agricultural and Mechanical Col., 1990
U M I300 N. Zeeb Rd.Ann Arbor, MI 48106
CHROM ATOGRAPHIC PEAK SHAPE ANALYSIS AND
MODELING
A Dissertation
Submitted to the Graduate Faculty of the Louisiana State University and
Agricultural and Mechanical College in partial fulfillment of the
requirements for the degree of Doctor of Philosophy
in
The Department of Chemistry
byMark Stephen Jeansonne
B.S., Louisiana State University, 1985 December 1990
D edication
In everyone's life there is a person or persons who have played a special role,
enabling one to aspire to and attain life's highest goals. In my life, these special people
have been my family, the total, unquestioning supporters of my goal of obtaining a Ph.
D. in Chemistry. Without their support and encouragement I certainly would not have
dared dream of spending the countless hours necessary for completion of this degree.
I know that they understood when I needed to apply myself to my work, even
when it meant that I might not spend as much time with them as I or they would have
liked. It is because of their unselfish dedication to my work that this dissertation is
dedicated to them, the members of my family. In this light, I consider them to be as
much an author of this dissertation as I am.
ACKNOW LEDGEM ENT
I wish to express my gratitude to Professor Joe P. Foley. His tireless dedication
his work and students has greatly inspired my work.
CONTENTS
List of Tables, ix
List of Figures, xii
Abstract, xviii
Chapter 1.
INTRODUCTION
References, 7
Chapter 2.
REVIEW OF THE EXPONENTIALLY MODIFIED GAUSSIAN
(EMG) CHROMATOGRAPHIC PEAK MODEL SINCE 1983
Introduction, 9
Development of Equations Utilizing the EMG Model, 12
Development of Equations For Single Peak Quantitation, 13
Development of Equations For Overlapping Peak Quantitation, 15
Use of the EMG Model for Study of Expected Errors in
Chromatographic Measurement, 18
Errors in the Measurement of Single Tailed Peaks, 16
Analysis of Errors in Overlapping Peaks Via the EMG Model, 21
Direct Chromatographic Application of the EMG Model, 22
Use of the EMG Model in Flow Injection Analysis, 22
Conclusion, 24
References, 31
Chapter 3. 38
IMPROVED EQUATIONS FOR THE CALCULATION OF
CHROMATOGRAPHIC FIGURES OF MERIT FOR IDEAL AND
SKEWED CHROMATOGRAPHIC PEAKS
Introduction, 39
Experimental, 40
EMG Peak Generation, 49
Method of Derivation, 41
Derivation of Equations For a, 42
Derivation of Equations for M2 ,42
Derivation of Area Equations, 43
Results and Discussion, 43
Accuracy of Equations, 43
Precision of Equations, 44
Accuracy and Preci sion of Area Equations, 44
Conclusion, 45
Acknowledgment, 45
Literature Cited, 59
Chapter 4. 60
MEASUREMENT OF STATISTICAL MOMENTS OF
RESOLVED AND OVERLAPPING CHROMATOGRAPHIC
PEAKS
v
Introduction, 62
Experimental, 64
EMG Peak Generation, 65
Real Chromatographic Peak Generation, 66
Peak Parameter Measurement, 67
Results and Discussion, 68
Simulated Peaks Without Noise, 68
Results for Real Chromatographic Peaks, 71
Modified Width-Asymmetry Method for True Peak Deconvolution, 73
Computational Time, 74
Acknowledgment, 75
Credits, 75
References, 93
Chaper 5. 94
THE Q TRANSFORMATION: A NOVEL METHOD OF
PEAK SHAPE ANALYSIS
Introduction, 97
Theory, 98
Definition of Q Transformation, 98
Properties of the Q Transformation, 99
Peak Shape Comparison Using the Q Transformation, 100
Experimental, 101
Peak Parameter Measurement, 103
Results and Discussion, 105
Measurement of Q and ZQ from real data, 105
Effect of Smoothing on Q Transformation, 105
Effect of Noise, 106
Effect of the Number of Points in a Peak, 107
Comparison of Methods Using Noiseless Data, 107
Comparison of Methods Using Noisy Data, 108
Potential Applications of the Q Transform, 110
Conclusion, 111
Credit, 112
Literature Cited, 129
Chapter 6.
SINGLE CHANNEL PEAK IMPURITY DETECTION
USING THE Q TRANSFORMATION
Introduction, 134
Experimental, 137
Peak Parameter Measurement, 139
Results and Discussion, 140
Peak Shape Comparison, 140
Simulated Impure Peaks, 145
Noisy Simulated Peaks, 141
Real Impure Peaks, 142
Experimental verification of paired-t option, 143
Potential For Qualitative Analysis, 143
Conclusion, 144
Literature Cited, 156
Chapter 7 157
SUMMARY
APPENDICES
A. Q Transformation Program Listing and Documentation 160
B . Tables for Interconversion Between Various Chromatographic
Separation Measures 188
Vita 243
v iii
LIST OF TABLES
Page
C hapter 2.
Table 2-1 25
Chromatographic Applications of the Exponentially Modified Gaussian (EMG)
Model Since 1983.
Chapter 3.
Table 3-1 46
Coefficients, Accuracy, and Precision of Equations for Calculating Gq, and M2
of Gaussian and Exponentially Modified Gaussian (EMG) Peaks.
Table 3-2 49
Error and Uncertainty of Chromatographic Peak Parameters calculated from Gq
and M2 using the Equations in Table I.
Table 3-3 53
Peak Area Equations for EMG and Gaussian Peaks.
Table 3-4 54
Comparison of maximum errors for the CFOMs reported here and those reported
previously.
Chapter 4.
Table 4-1 76
Maximum peak overlap (percent valleys) that can be tolerated by the width-
asymmetry and summation methods for a given accuracy (<5% error)
Table 4-2 77
Peak Parameters Measured for an Isolated, Real Chromatographic Peak.
Table 4-3 78
Comparison of Summation and Width-Asymmetry Methods for Two Sets of
Overlapping, Real Peaks.
Table 4-4 80
Peak Parameters for Deconvolved Peaks of Overlapping Peak Pairs With 40%
and 67% Valleys.
Chapter 5.
Table 5-1 113
Minimum Detectable Change in T/CT For Various Methods of Peak Shape
Analysis.
Table 5-2 114
% Relative Error in ZQ Due to Baseline Errors in EMG Peaks.
x
Chapter 6.
Table 6-1 146
Minimum Resolution Necessary for the Detection of Two Overlapping EMG
Peaks.
Table 6-2 147
Simulated Impure peaks detectable by the paired t-test and the Q profile.
Table 6-3 148
Impure experimental HPLC peaks detected via the paired t-test using the Q
profile.
Table 6-4 149
Comparison of some pure peaks via the paired t-test.
Appendix B.
Table B-l 191
Table B-2 204
Table B-3 222
LIST OF FIGURES
Page
C hapter 2.
Figure 2-1 29
Illustration of the effect of %/g q on EMG peak tailing.
Figure 2-2 30
Measurement of graphical peak parameters for use in asymmetry-based
equations.
Chapter 3.
Figure 3-1 56
Graphical parameters necessary for calculation of various CFOMs.
Figure 3-2 57
Plot construction for determination of f(b/a) for calculation of Gq at r=0.25.
Note curvature at lower asymmetries.
Figure 3-3 58
Plot construction for determination of f(b/a) for calculation of M2 at r = 0.25.
Due to curvature over entire asymmetry range, two least-squares quadratics were
fit.
Chapter 4.
Figure 4-1 81
Equations used for calculation of statistical moments and other peak parameters
by a) the summation method and by b) the width-asymmetry method.
Figure 4-2 82
Effect of the data sampling rate on the measurement of peak width, asymmetry,
and peak area.
Figure 4-3 83
Measurement of graphic parameters for an overlapping pair of chromatographic
peaks. tr and hp are the retention time and peak height of respective peaks, and
hv is the height of the valley. Peak width at the desired peak height fraction is
given by ta-tb-
Figure 4-4 84
Comparison of the errors in peak area and variance occurring in the summation
and width-asymmetry methods as a function of peak overlap (percent valley) for:
(a) x/ g q =1; (b) z / g q = 2; and (c) t /o q = 4. Labels in the plot refer to: a) the
parameter, b) the peak (first [L] or second [R]) for which a parameter was
obtained; and c) the relative peak height at which the width and asymmetry were
measured (width-asymmetry only). For example, "M2, L; 75%" refers to the
variance measured for the first peak of the overlapped pair at 75% of the peak
height, while "M2, R" refers to the variance of the second peak measured by the
summation method.
Figure 4-5 88
Comparison of the errors in peak area and variance occurring in the summation
and width-asymmetry methods as a function of peak overlap for a highly skewed
pair of peaks {x / g q = 4) with a less than ideal baseline level (0.1%). Conditions
as in Fig. 4-4.
Figure 4-6 89
Comparison of the errors in peak parameters other than peak area and variance
for overlapped peak pairs with x / g q = 4 occurring in: a) summation method; and
b) width-asymmetry method. Conditions as in Fig. 4-4.
Figure 4-7 92
Visual interpretation of the width-asymmetry/deconvolution method. The solid
black line indicates the real overlapping chromatographic peaks, while the lighter
lines show the individual peaks that are predicted by the width-
asymmetry/deconvolution method.
Chapter 5.
Figure 5-1 115
Measurement of a and b used for the calculation of Q in eq 1.
Figure 5-2 116
The Q transform can be used to visualize peak shape differences without regard
to peak height or width. Peak identity: (1) Exponentially Modified Gaussian
xiv
(EMG), x/o=2.6, 0=1.0 (2) EMG, 1/0=1, 0=0.26 (3) EMG, xl 0=2.2, o=0.26
(4) EMG, t /o =2 .8, o=0.5 (5) Log-Normal, asymmetry=1.6
Figure 5-3 117
Comparison of Q profiles for various peak shapes, (a) EMG peak, ref. 13 (b)
Gamma peak, eq 4 (c) Log-Normal peak, eq 5 (d) Overlay of Q profiles from
(a) through (c). Note that peak shapes can be differentiated on the basis of XQ or
on direct point-by-point comparison of individual Q profiles.
Figure 5-4 118
Differentiation of peak shapes using Q transform between (a) a symmetric peak
(Gaussian) and a slightly asymmetric peak (EMG, t/s = 0.3) and (b) two
symmetric peaks by a redefinition of Q as discussed in the text Note: Curves in
(a) appear noisy relative to curves in (b) because of scale difference.
Figure 5-5 119
Non-superimposability of Q profiles of fundamentally different type of peaks
with identical values of XQ. Peak identities: EMG with xto = 2.8 and log
normal with an asymmetry (eq 5) of 1.6.
Figure 5-6 120
Effect of Savitsky-Golay smoothing factor (eq 6) on (a) XQ ; and (b) Q vs r for
an EMG peak with x/a =1.
Figure 5-7 121
Effect of signal to noise (S/N) ratio on the Q transformation.
xv
Figure 5-8 122
Effect of the relative sampling rate (number of points) on the measurement of
XQ. Peak shape: EMG peak, x/o=l.
Figure 5-9 123
Effect of the relative sampling rate (number of points) on the accuracy of Q.
Number of points in the peak were measured from half-height to half-height.
Figure 9e shows true Q vs r profile (search algorithm) while Figure 9f shows the
average error associated with the measurement of Q at any r.
Figure 5-10 127
Relative abilities of statistical moments, excess, skew, second derivative, XQ and
DFM to detect changes in peak shape for noiseless peaks.
Figure 5-11 128
Automatic detection by the Q transform of instrumental problems in
chromatography, as illustrated with (a) a fronted peak (indicative of sample
overload); and (b) its Q profile.
Chapter 6.
Figure 6-1 150
Effect of co-elution on overall peak profile. Impurity elutes on the tail of the
parent peak. Conditions: EMG peaks (x/c =1); area ratio = 16:1; resolution =
xv i
0.25. Note that no shoulder or other visible evidence is apparent on the overall
profile.
Figure 6-2 151
Effect of co-elution on the Q transformation in which the impurity elutes on the
tail of the parent peak. Conditions: EMG peaks ('t/a =1); area ratio =16:1;
resolution = 0.25.
Figure 6-3 152
Effect of co-elution on (a) overall peak profile; and (b) Q transformation in which
the impurity elutes on the front of the parent peak. Other conditions as in Figure
6- 2 .
Figure 6-4 153
Impure and pure n-propyl benzene peaks. Visual inspection of original peaks
fails to indicate presence of impurity.
Figure 6-5 154
Example of clustering as a function of peak identity for a series of ketones.
Figure 6-6 155
Illustration, using simulated data, of Q transformation approach to qualitative
analysis.
x v ii
ABSTRACT
Various aspects of chromatographic peak quantitation and shape characterization
are investigated in detail for single and overlapping chromatographic peaks. From the
viewpoint of providing better quantitation of real chromatographic data while minimizing
computational complexity, the results presented should be easily incorporated into
existing routine chromatographic data analysis regimes. Three topics applicable to
modem chromatographic data analysis are considered.
First, progress in the application of the exponentially modified Gaussian (EMG)
function to chromatography is reviewed. The review covers the following areas: (1)
equations derived from the model, (2) studies of inherent errors in the quantitation of
chromatographic peaks via use of the EMG model, (3) chromatographic applications
since 1983 and (4) applications to flow injection analysis. The information discussed
and the references included in this review should provide a valuable resource for those
researchers considering or already using the EMG model in their studies.
Second, improved empirical equations based on the EMG model are presented. The
equations utilize measurements of the graphical parameters peak height (hp), width (W),
and asymmetry for the calculation of the following chromatographic figures of merit
(CFOMs): t, c , area, variance, third and fourth statistical moments, excess, and skew.
The equations are shown to be more accurate than traditional numerical methods when
applied to single and overlapped chromatographic peaks that fit the EMG model.
Application to simulated and real chromatographic peaks is discussed.
Finally, a computationally simple, normalized data transformation that can be
used for peak shape analysis and comparison is introduced. Compared to slope analysis,
moment analysis, and the distribution function, the "Q transformation" performs much
better over a wide range of experimental conditions, including signal to noise (S/N)
ratios of less than 100 in which the other methods often fail. This method is shown to be
computationally simple, easy to automate, and very intuitive in nature. Application of the
Q transformation for impure chromatographic peak detection using single-channel
channel data is discussed. Its advantages over the skew, excess, slope analysis and
distribution function methods are reported using both simulated and real data.
Chapter 1
INTRODUCTION
1
2
The field of chromatography has evolved substantially since its beginning. From
large particle silica stationary phases to today's microparticulate, tended phase packed
columms and capillary columns, chromatography has become a mainstay in modem
analytical chemistry. Our understanding of the chromatographic process has progressed
such that the prediction of retention behavior for many compounds is now possible. In
fact, expert computer programs now exist that allow a chromatographer to predict with
fair accuracy the retention behavior of the compounds in a complex mixture, with only a
minimum understanding of the actual chromatographic process. By predicting retention
behavior, one should be able to predict the best separation conditions for a sample with a
minimum of work.
However, the degree of separation (and other important chromatographic
parameters) also depends on the shape of each peak. We need to know the shape and
width of the peak in order to know the minimum difference in retention times between
adjacent peaks so that they are sufficiently separated. Even with the level of
understanding available today, accurate peak shape prediction for all compounds under
given conditions is still not possible. Some general theories do exist, though, for
explaining many aspects of chromatographic peak shape.
As a single type of solute molecule emerges from a chromatographic column, a
peak indicating the distribution of the molecules with time is detected. Because this
distribution is not infinitely narrow, or not as narrow as the band when it starts at the
beginning of the column, there must be processes occuring within the column acting to
broaden the peak. According to rate theory (1,2), the three processes that contribute to
broadening a chromatographic peak symmetrically are (1) resistance to mass transfer, (2)
eddy diffusion and (3) longitudinal diffusion.
3
Broadening due to the resistance to mass transfer occurs because an instantaneous
equilibrium is not established between solute molecules in the stationary phase and those
in the mobile phase. Therefore, the solute concentration profile in the stationary phase is
always displaced slightly behind the equilibrium position and the mobile phase profile is
displaced slightly ahead of the equilibrium position. The resulting peak profile observed
at the column outlet is broadened about its center, which is located where it would have
been for an instantaneous equilibrium. Resistance to mass transfer is the dominant cause
of symmetric band broadening in chromatography.
In packed columns there are multiple possible flow paths available for solute
molecules as they flow through the column. Because each path may be of different
length, solute molecules will not emerge from the column at the same time thereby
broadening the peak.
Longitudinal diffusion is due to simple molecular diffusion in the axial direction
along the length of a column. Its contribution to symmetric band broadening will
increase with the amount of time that a solute band is inside the column.
As stated before, the aforementioned processes tend to broaden a chromatographic
peak symmetrically. However, most real chromatographic peaks are not symmetric. The
processes that act to broaden a peak asymmetrically can be divided into intracolumn
(within column) and extracolumn (outside the column) processes. Important
intracolumn contributions to peak asymmetry include incomplete resolution, slow kinetic
processes, chemical reactions, column voids, and non-linear distribution isotherms.
Important extracolumn asymmetric band broadening processes include dead-volumes or
spaces and a slow detector time constant. Although all of the intracolumn asymmetric
broadening processes do occur, for most well behaved, linear chromatographic systems,
their effects will usually be slight or even negligible. However, the extracolumn
4
contributions to peak asymmetry will affect most chromatographic peaks and tend to add
an exponential tail to them.
Most basic chromatographic theory assumes a Gaussian peak shape. However,
most practicing chromatographers realize that this assumption is normally not valid due
to the possible asymmetric band broadening processes. Therefore, application of the
theory developed for symmetric peaks to experimental peaks can result in errors. For
example, when two chromatographic peaks overlap, integration of their separate areas is
impossible. If the assumption is made that the two peaks are symmetric, simply dividing
their areas by drawing a perpendicular at the valley between them can often give
sufficient accuracy. Quantitating the areas of overlapping asymmetric peaks in this
manner will typically cause large errors, depending on the degree of asymmetry (3).
However, this error can be reduced greatly by assuming a peak shape function that is
closer to the peak shape of the experimental peaks, even if the true experimental peak
shape is not known. In this manner, a tool or model can be developed for more accurate
quantitation of the peaks. The better model can also be used to simulate and therefore
understand the errors inherent in using the less accurate method. Because
chromatographic peaks cannot always be completely separated, the development of
techniques to better quantitate them when overlap occurs has played a very important role
in the development of chromatography as a routine analysis technique.
Better models can also be developed for the quantitation of other important
chromatographic parameters when peaks are asymmetric. Perhaps the most important
descriptor of chromatographic column performance is the efficiency (N), defined as
tR2n = — —— rVariance
5
where tR is the retention time of the peak. The variance of a peak can be evaluated
statistically using numerical method techniques. But, numerical method evaluation of
peak variance has proven to be highly error prone (4). Therefore, Equation 1 is often
simplified assuming a Gaussian peak shape to give
where W is the width of the peak at the baseline. Now the variance of the peak is
of the peak. This provides a very convenient method for measuring N. However, if the
experimental peak is not Gaussian in shape, error in the value of N will occur. Again, if
one were to assume a peak shape function closer to the true experimental peak shape, a
more accurate value of N could be obtained. Of course, to be routinely useful, in this
instance, the value of the variance needed for Equation 1 should be as easy to measure as
if one had assumed a Gaussian peak shape. That is, any new models should be
sufficiently easy to use so that those not familiar with the model or its underlying
concepts can still use it in routine analyses.
The development of easy to use models and methods for more accurate quantitation
of chromatographic peaks, overlapped or no t, is the general theme of the following
chapters. This area of research has become very important in recent years with advent of
computerized data analysis. Instrument manufacturers are always striving to find easier
and more accurate methods for describing chromatographic separations, thus allowing
greater automation and subsequently greater productivity by their customers. Because
computerized data analysis is now routinely available in most laboratories, computerized
data acquisition and its implications are an integral part of the work presented. We have
strived to keep computational complexity to a minimum, however.
[2]
assumed to be equal to the variance of a Gaussian function, which is related to the width
6
To a large extent, the following chapters utilize the exponentially modified
Gaussian (EMG) function (5,6) both as a model for real chromatographic peaks and as a
means for their simulation. This function results from the convolution of a Gaussian
function with an exponential decay function. As mentioned previously, if intracolumn
processes leading to peak asymmetry are negligible, then the chromatographic column
tends to broaden a peak symmetrically, with the resulting peak having a Gaussian shape.
However, the extracolumn effects usually tend to add some degree of exponential tailing
to a chromatographic peak. Therefore, the EMG model should give a more accurate
description of real chromatographic peaks than the Gaussian model. This peak model and
its applicability to experimental peaks are described in detail in Chapter 2.
Chapter 3 presents equations for the better quantitation of real chromatographic
peaks using the EMG. These equations can be used for both single and overlapping
chromatographic peaks and allow the user to apply the EMG model without resorting to
complicated curve fitting techniques. As mentioned before, simplicity is an important
aspect for computerized chromatographic quantitation.
Chapter 4 reports the results of applying the equations from Chapter 3 to both
overlapping and single peaks. Simulated and real chromatographic peaks were
examined. It was found that accurate statistical moment measurements may be made on
overlapped peaks that fit the EMG model. A very simple procedure for peak
deconvolution is also described.
Finally, Chapters 5 and 6 introduce a new method (the Q transformation) for peak
shape analysis and the detection of severely overlapped chromatographic peaks using
only single-channel data. This work is important in that it provides a computationally
simple method for comparison of peak shapes that is more sensitive to peak shape
differences than other simple methods of peak shape comparison. It was found that the Q
transformation is more sensitive to peak shape changes than the methods of slope
7
analysis, moment analysis, and the distribution function method. It is this sensitivity to
peak shape differences that allows for the detection of overlapping peaks, even when no
visible evidence of overlap occurs. The methods introduced in these chapters also have
important implications for the field of laboratoiy automation.
REFERENCES
1) Giddings, J.C. "Dynamics of Chromatography", Part 1; Chromatogr. Science
Series, Vol. 1; Dekker, New York, 1965
2) Poole, C.F.; Schuette, S.A. "Contemporary Practice of Chromatography";
Elsevier, New York, 1984
3) Foley, J.P. J. Chromatogr. 1987, 384, 301-303
4) Anderson, D. J.; Walters, R. R J. Chromatogr. Sci. 1984, 22, 353-359
5) Sternberg, J.C. "Advances in Chromatography"; Giddings, J.C.; Keller, R.A.,
Eds.; New York: Marcel Dekker: 1966; Vol. 2, pp 205-270
6) Foley, J. P.; Dorsey, J. G. J. Chromatogr. Sci. 1984, 22, 40-46
C h a p t e r 2
REVIEW OF THE EXPONENTIALLY MODIFIED GAUSSIAN (EMG) FUNCTION SINCE 1983
8
9
INTRO DUCTIO N
Assumption of a valid peak shape model has played an important role in the
development of chromatography to date. For example, the derivations of many
frequently used fundamental equations, such as the fundamental resolution (Rs) equation
(Equation 1), depend on the assumption of Gaussian peak shape. N is the column
efficiency, k' is the capacity factor, and a is the separation factor, equal to k'2 /k'i.
Another equation often used, the equation for calculation of efficiency (Equation 2), is
also based on this assumption. In Equation 2, tR refers to the retention time of the peak
and a refers to the standard deviation of the peak. However, as many practicing
chromatographers know, the Gaussian function rarely provides an accurate model for
real chromatographic peaks. Therefore, a search for a better function to model real
chromatographic peaks has been pursued for many years.
One model that has received much attention is the exponentially modified Gaussian
(EMG) function. This function results from the convolution of a Gaussian function with
an exponential decay function, and thus, can represent both symmetrical and tailed
peaks. Its evaluation is only somewhat more difficult than that of the simple Gaussian
function, while being a much better model for real chromatographic peaks. As with the
Gaussian function, the variables of the EMG can be related to physical parameters and
[1]
[2]
1 0
the function has been justified theoretically as a good model for real chromatographic
peaks (1, 2).
These attributes have enabled the EMG model for chromatographic peaks to receive
much attention in recent years. A 1984 review (2) of the EMG for use in
chromatography provided further impetus for its increased use. In that review, the
authors described a simpler method for evaluation of the EMG function, given in
Equation 3. The function is defined by three parameters, the retention time (to), the
standard
deviation of the Gaussian component (G q ) and the time constant of the exponential decay
component (z), where z = (t - tg) / Oq - Gq/z. A is the peak area. As the ratio z/Gq
increases, the amount of tailing in the EMG peak increases (see Figure 2-1). As shown
in Equation 3, three main terms exist in this form of the equation: (PE) the pre
exponential term, (E) the exponential term and (I) the integral term. Via a change of
variable, Equation 3 can be written as
hEMG(0 = ~ exP [ \ (^)2 - (7 - ^ ) ] Jexp (-x2) dx [3]* x x -00
z /a/2
J LPE E I
z
hEMG(l)
J L JLPE E I
The integral term (I) from Equation 4 can be approximated by the product of a
polynomial and an exponential term. That is, I (z<0) = NF(z)P(q) and I (z^O) = 1-5
l(z<0), where NF(z) = exp(-z^/2) / V27t), and P(q) = ^ bnqn, q= (1+pz)'*. Then=l
parameters p ,b j , ..., b5 are constants given in Table ID of reference 2. The error
reported for this method of EMG evaluation was 1% or less for x J g q > 0.2. A short
program in BASIC using this method was listed in that report (2).
Another method for evaluating the integral from Equation 3 utilizes the error
function (3,4). By writing the integral term as
I = erf i ( - ^ + ^ ) + e r f - ^ ( ^ - 5 a ) [5]\ 2 oq x \ 2 v ctq x
and using the evaluation algorithms for the error function available in many commercial
software packages or numerical methods textbooks (5), one can evaluate it easily.
Because the first term is nearly unity when tQ/CQ > 5, the integral can also be
approximated as
I . l + e r f - U H a . S l ) . [6]V 2 x G q X
Since the publication of these evaluation methods, many studies have indicated the
validity of the EMG model in chromatography (4 - 9). Furthermore, papers by Foley (8 )
and Wu (10) have introduced simple methods for validating the EMG model for peaks of
interest.
To summarize, the attributes of the EMG model discussed above - its ability to
model tailing peaks - its relative ease of evaluation - its demonstrated validity as a model
for real peaks - have provided further incentive for studies using the EMG function since
1 2
the review of 1984. Therefore, it is prudent now to review the progress made since
then.
Although we review the progress of the EMG in only chromatography and flow
injection analysis (FLA) in this report, the results presented are applicable to other fields
in which this model is used. Because we have already briefly discussed verification of
this model for chromatography, this aspect will not be discussed further. We leave it the
interested reader if more details are desired by using the articles referenced above. Also,
we are excluding from our discussion non-linear chromatography or other forms of
chromatography where non-EMG or non-Gaussian peaks are obtained. Our intentions
are to focus mainly on the applications of the EMG model to chromatography, with a
brief review of its use in flow injection analysis (FLA). More specifically, the
organization of this review will cover four main areas into which most studies utilizing
the EMG fall: (1) use of the EMG for development of equations that quantitate
chromatographic peaks and/or their mathematical separation, (2 ) investigations of
potential errors in chromatographic measurements by using the EMG for simulation of
real peaks, (3) direct application of the EMG model to chromatography, and (4) the
application of the EMG model to FIA. Because only three reports concerning evaluation
of the EMG function have occurred since 1983 (3,11,12), this topic is not discussed
further in this report.
DEVELOPM ENT OF EQUATIONS UTILIZING THE EMG
MODEL
Due to the attributes of the EMG model listed above, many studies have
concentrated on using this model to develop other equations that can be easily and
directly applied to real chromatographic data. These studies can be broken down into
1 3
those that apply strictly to single peaks and those that are applicable to the measurement
of resolution or deconvolution of two partially resolved peaks.
Development of equations for single peak quantitation . Within this
section we begin by presenting work specifically designed to extract the fundamental
parameters, x, Gq and t^ for an EMG fit to a real chromatographic peak. Some of these
studies also presented equations for the calculation of the first (Mi), second (M2 ) and
higher statistical moments. In 1983 Foley and Dorsey (13) reported equations for the
calculation of chromatographic figures of merit (CFOM) for symmetric and asymmetric
peaks based on the EMG model. The equations enabled one to calculate the basic
parameters (x, Oq and t^) and statistical moments of an experimental peak from the
simple graphical parameters a, b and peak height, illustrated in Figure 2-2. A particularly
important equation resulting from this study is shown in Equation 7, for the
measurement of system efficiency, where Wq.i is the width of the peak at the 10% of its
height. Following that report, this equation has been referred to as one of the best
methods for calculation of efficiency (14,15).
Although the Foley-Dorsey study was noted (although only briefly) in the previous
review (2 ), we decided that a brief re-introduction of these equations was necessary due
to their rather extensive use since then by other researchers (see Table 2-1). Also, these
equations appropriately set the tone for this section.
[7]
Soon after the publication of these equations, a report by Jung et al.(16) presented
a method for the extraction of the EMG parameters via the use of normal and derivative
peak heights measured at four or five time points. Their method gave good
measurements of the fundamental EMG variables and did not require the iterative
computer searches needed to find the graphical parameters a and b. However, their
approach was more complex intuitively than that reported by Foley and Dorsey and
required solution of cubic and quartic equations after measurement of the various normal
and derivative peaks. The Foley-Dorsey equations require only the "plugging in" of the
necessary graphical peak parameters. Later, a report by Anderson and Walters (17)
presented new equations for calculation of Oq , M j , and M2 that were valid to a higher
peak asymmetry than those of Foley and Dorsey. These equations were also based on
the parameters a, b, and peak height. Based on the measurement of the same graphical
peak parameters b and a, Foley (8 ) subsequently reported accurate equations for the
measurement of peak areas for chromatographic peaks fitting the EMG model.
f t a -0.133A = 0.586 hp W0 1 ( j?) [8]
A = 0.753 hD W0.25 [9]rf t ,\+0.235
A = 1.07 hpWo.5 ^ [1 0]/]b\+0.717
A = 1.64 hp W0_75 i ' [11]
Although these equations do serve as an alternative to the numerical integration methods
for the measurement of peak areas, their greatest utility lies in their ability to predict
whether a given peak accurately fits the EMG model and for the area deconvolution of
overlapping peaks (discussed later). As an attractive alternative to least-squares curve
fitting, one can compare the area values obtained by each equation, and based on the
suggested criteria (the spread in values from Equations 8-11 should be less than ten
1 5
percent for a peak to be considered EMG), one can decide on the applicability of the
EMG model for his data.
Development of equations for overlapping peak quantitation.
Previous equations for the calculation of resolution or deconvolution of areas for
overlapping peaks have been based mainly on the Gaussian peak model. However, due
to tailing of most real peaks, these equations have limited applicability. Because the
EMG model may better represent actual chromatographic peaks, equations using this
model should offer more accuracy in the quantitation of the overlap or provide a better
description of each single peak in the pair.
An article by Frans et al. (18) in 1985 presented a method for the reiterative least
squares resolution of overlapped chromatographic peaks using multi-wavelength
detection. For a six component mixture in which resolution ranged from 0.21 to 0.36,
the retention times of all components were accurately determined with less than 1.0 %
relative error. The average relative error in Oq for the six component mixture was less
than 15%.
Because multi-channel detection is not generally available for many analyses,
methods for quantitating overlapping peaks using single-channel detection have also been
reported. Foley (19) reported in 1987 that the area equation based on graphical peak
parameters a and b measured at 75% of the peak height (see Equation 11) could be used
to deconvolute the areas of an overlapped pair of chromatographic peaks. Because the
first peak in an overlapped pair is only slightly distorted at 75% of its height, an accurate
measurement of b and a could be obtained for that peak. The author reported an accuracy
of ± 4% in the area of of the first peak in the pair by this method when the valley height
between the peaks, relative to the first peak, was less than or equal to 45%.
Furthermore, by accurately measuring the total area of the peak pair by integration, and
1 6
then subtracting the area obtained for the first peak from the total area, an accurate value
for the area of the second peak can be determined. This method has been applied
successfully to real overlapping peaks that fit the EMG model (6 ).
Soon after this report appeared, Binsheng et al.(20) reported a table of correction
factors for quantitative area determinations of overlapping peak pairs. Their method
utilized values of the errors in area obtained when the perpendicular drop method was
used for overlapping EMG peak pairs. By calculating the area of the first peak via the
perpendicular drop method and utilizing Equation 13, a more accurate estimation of the
true area, Aj ,of the first peak can be ascertained. Av is the area of the first peak
measured by the perpendicular drop method and the value of P (correction factor) needed
for Equation 13 can be obtained from tables they provided.
Although the method of Binsheng gives greater accuracy for the quantitation of
overlapping peaks, the asymmetry of both peaks in the pair is assumed to be the same.
Because Foley's method (19) does not assume equal peak asymmetries in the pair, it may
be more practical for real overlapped peaks.
Along similar lines, Haddad and Sekulic (21) reported fourth-order polynomial
equations for the calculation of percentage area overlap of a peak pair. Their approach
was to determine the value of x/Oq for the first peak in a pair (using methods such as
those described above),to measure resolution of the peak pair, and from a fourth order
polynomial equation applicable at that t/Cq and resolution,to compute the value of
percent area overlap for the peaks. In this and related papers (7,22), the authors applied
this approach for the optimization of mobile phase conditions in high-performance liquid
chromatography (HPLC). Although developed strictly for optimization purposes, this
1+ P/1001 ) [13]
approach may prove useful in other areas where quantitation of overlapping peaks is
desired.
Schoenmakers et al.(23) have recently reported a corrected resolution function for
tailed chromatographic peaks (see Equation 14). A consequence of this function is that
two different resolution values are calculated, one for each peak.
measured at the 13.5% peak height fraction of peak i, lAgj is the asymmetry of peak i
measured at 13.5% of the peak height of j, *Nj is the efficiency of peak i as given in
the same parameters for peak i measured at 13.5% of peak height relative to peak j. As
stated by Schoenmakers, if the two peaks differ in height by no more than a factor of ten,
iRsji refers to the resolution measured for the first peak, i, tj is the retention time of
peak i, tj is the retention time of peak j (second peak), lAgj is the asymmetry of peak i
Equation 15 and xNj is the efficiency of peak i as given in Equation 16.
[15]
[16]
The parameters *ai and ^ are the same as the parameters a and b shown in Figure 2-1,
for peak i measured at 13.5% of the peak height of i, while the parameters Jaj and *bj are
if JNj = xNj = N, and if ̂ j = *AS j = As, then simplified expressions for the corrected
resolution can be given by Equations 17 and 18.
1 8
4 A stj + 4tj V l + l/21n(hj/h |)(M - tj) ( l + a s)Vn
[17]
4 A sti V I + l / 2 1n(hj/hj) + 4tj
(t, - tj) (1 + A S)VN[18]
USE O F THE EMG MODEL FOR STUDY OF EXPECTED
ERRORS IN CHROM ATOGRAPHIC MEASUREM ENT
The effective study of the measurement errors expected for tailed peaks via the
EMG function stems from its accuracy and practicality as a real peak model. Again, we
can subdivide the expected errors into those for a single tailed peak and those for
overlapping tailed peaks. We suggest reading references 24,25 and 26 for good general
reviews covering many general aspects of expected errors in chromatographic
measurements.
E rro rs in the measurement of single tailed peaks. Most of the studies
appearing since 1983 utilizing the EMG for quantitation of single peaks have dealt
directly or indirectly with statistical moment measurement. For example, two of the
more important chromatographic parameters, peak area and column efficiency, can be
defmed as statistical moments or functions of statistical moments (see Equations 19 and
20).
00 stopArea = zeroth moment = Mo = J h(t) dt = £ h(t) At [19]
-oo start
1 9
M i^ tp2Efficiency = N = - j^ - - ^ [20]
Thus those studies that specifically consider errors in either parameter are in reality
considering errors in the moments. Sources of error in the measurement of statistical
moments include inappropriate choice of model (e.g., when a Gaussian derived
efficiency equation is used for tailing peaks), insufficient sampling rate,, and most
importantly, noise and associated baseline errors.
Perhaps the greatest possibility for quantitation error occurs in the measurement
of the statistical moments for tailed peaks. As shown by Anderson et al.(17), large
errors in the first and second statistical moments (calculated via summation) can occur
when a portion of a tailed peak is truncated by an improperly constructed baseline.
These large errors result from the sensitivity of moments to the data in the extreme
regions of the peak. These errors normally occur when the integrator or data system
cannot detect the proper limits for the integration because noise is obscuring the correct
limits. If the integration bounds are improperly detected, the inaccurately constructed
baseline that results causes part of the peak to be lost after baseline subtraction. Of
course, improper setting of the slope sensitivity in the integrator can also cause
improperly detected integration limits. However, errors due to improper integrator
settings are avoidable, whereas errors resulting from noise are generally not. Anderson
et al. demonstrated errors in the first moment of greater than 30% and approximately
70% in the second moment when severe baseline construction errors occurred. They
showed that the use of asymmetry based equations for the first and second moments will
greatly reduce these errors.
As mentioned above, a significant source of error when measuring efficiency (or
other parameters) occurs when an inappropriate model is chosen. Because the efficiency
2 0
is related to the distribution of a sample peak over time, a symmetrical model, such as the
Gaussian model, cannot accurately represent the distribution of a tailed peak and thus
cannot accurately measure the efficiency. Therefore, a more representative model, such
as the EMG, should enable more accurate measurement of efficiency when the actual
moments (via summation) of a peak cannot be measured accurately. Most studies agree
(7, 14, 15,24,27,28) that in the absence of large amounts of noise, column efficiency
and variance (second moment, M2 ) are best calculated via Equations 20 and 21 using
numerical computer methods. However, the asymmetry based
stopM2 = Variance = J (t-Mj)2«h(t) dt + Area = ^ (t-Mi)2»h(t) At Area [21]
-00 start
methods for measurement of efficiency and variance will provide very good accuracy
when noise is severe or numerical integration is not feasible. Because the asymmetry
based methods seem to be simpler for routine use, they may be the best overall. For
higher moments (M3 and M4 ), integration methods are normally very inaccurate and
therefore the asymmetry based methods should be used for their measurement whenever
the peak shapes are appropriate. A dissenting view was presented by Colmsjo and
Ericsson (29) who suggested that measurement of the height equivalent to a theoretical
plate (HETP) for use in van Deemter plots can be made more reproducibly by using the
Gaussian model.
Sampling rate affects the accuracy of all possible measurements concerning
digitized chromatographic peaks. Two recent papers (25,26) and a book (24) have
thoroughly evaluated the effects of sampling rate on various chromatographic
measurements. Only one paper since 1983 has addressed sampling rate errors via the
21
EMG model. In that study, Rossi (30) presented a simple program for the calculation of
expected errors in the measurement of peak area for tailing peaks.
Analysis of errors in overlapping peaks via the EMG model. The
weak link in most chromatographic analyses continues to be the accurate area quantitation
of overlapping chromatographic peaks. The popular perpendicular drop and tangent
skim methods are still used almost exclusively due to their simplicity, although large
errors are possible with tailing peaks. Therefore, the EMG, being a good model for
tailing peaks, has been used in many studies since 1983 to demonstrate the severity of
these errors and present possible alternate methods for more realistic interpretation of the
individual areas of the peaks. However, most chromatographers agree that baseline
separation of important peaks is the best way to avoid these errors.
A widely cited study of the possible area quantitation errors for two overlapping
peaks was published in 1987 (19). There, the author demonstrated that errors in the area
when using the perpendicular drop method for quantitation of two severely overlapped
EMG peaks, with large differences in size, could be as high as 200%. He also reported
that errors in the measurement of heights could be as high as 80%. Papas et al. in 1987
(26) also showed the possibility of large errors in overlapping peak area measurement by
some integrator systems. Later, Jeansonne and Foley (6 ) investigated the errors
resulting from the measurement of statistical moments, excess, and skew on overlapping
peaks separated by the perpendicular drop method. They also demonstrated the
feasibility of accurate statistical moment measurement for overlapping peaks by using
asymmetry based methods.
A more recent paper by Papas et al.(31) discussed overlapping, tailed peak area
measurement errors when the perpendicular drop and tangent skim methods of area
deconvolution were used. Equations obtained via multiple linear regression were used to
determine whether the tangent skim or perpendicular drop method was better for
2 2
separating a particular set of overlapping peaks. In their opinion, these two
deconvolution methods are the most reliable and simple methods available for routine use
and therefore the decision of which method to use is important Other reports (7,20)
have demonstrated indirectly the potentially large errors resulting from the use of the
perpendicular drop method.
DIRECT CHROMATOGRAPHIC APPLICATION OF THE
EMG MODEL
No review on the EMG model would be complete without reporting the practical
applications of this model. These studies have used either the EMG model directly, as
described in the introduction, or equations derived from the EMG model, such as those
presented in the previous sections. Because the number of these reports is large, they are
not discussed here in detail. Instead, we give a chronological listing in Table 2-1 of
those studies that have appeared since 1983. It is hoped that this table will facilitate easy
reference for those who wish to consider the use of the EMG model for their own
investigations. However, a note of caution is needed here. Although almost any tailed
chromatographic peak is better modeled by the EMG than the Gaussian function, EMG
validity should be checked first by the methods discussed earlier.
USE OF THE EMG MODEL IN FLOW INJECTION
A N A L Y SIS
Although not strictly a chromatographic technique, flow injection analysis (HA)
consists of a flow system similar to that found in chromatography. In fact, the
dispersion processes in FIA can be considered analogous to the extracolumn band
2 3
broadening phenomena found in chromatography. The total band broadening in terms of
variance can be described by Equation 22 (32), and is equal to that for extracolumn band
broadening in chromatography.
°^peak “ °^injection + ^ transport+ ^detection [2 2 ]
Also, because several studies (32 - 37) have demonstrated the suitability of the EMG
model for the FIA peaks, a short discussion of some results from this field is
appropriate.
Brooks et al. (32) reported several advantages for using the variance instead of
the dispersion coefficient in an FIA system. (1) The peak width can be obtained from the
variance; (2) various FIA manifolds can be compared by using variance values; (3)
variance is a more direct measure of sample throughput; and (4) the individual
contributions to total variance (including the contribution from chemical reaction) are
additive and therefore easily obtained. The EMG character of the FIA peaks was verified
in this study by using Equations 8-11. The authors considered a peak to be EMG if the
largest outlier of the four measurements was within 2 0 % of the mean area calculated by
the equations. They found that 73% of the peaks fit the EMG model to within 15% at
flow rates up to 1 .2 ml/min for flow manifolds consisting of 1 0 0 cm of straight tubing,
approximately 14 mm in diameter. At flow rates of less than 0.6 ml/min they found that
92% of the peaks fit the EMG model to within 20%. They also reported that when coiled
tubing and a flow rate of 1.04 ml/min was used, all peaks fit the EMG to within 10%.
After verification of the EMG character of their peaks, the authors found a low (6.5%)
relative standard deviation for the measurement of the total variance, without chemical
reaction. The authors concluded that moment analysis using the EMG model was a valid
2 4
method for the examination of FIA peaks. Subsequent work (33, 34) using the EMG
model for FIA peaks has also yielded excellent results under a variety of conditions.
C O N C LU SIO N
The authors of the previous EMG review (2) hoped that their simple, clear
approach for evaluation of the EMG function would contribute to its greater use by
scientists when appropriate. Based on the large number of studies published since then
for which the EMG model has been directly applied or used for development of peak
analysis methods in chromatography, that review was a success. It is likewise hoped
that this compilation will contribute to the overall use of the EMG model by providing a
valuable reference resource for those researchers interested in applying the model to their
own studies.
2 5
Table 2-1. Chrom atographic Applications of the Exponentially
M odified Gaussian (EMG) M odel Since 1983
EMG APPLICATION METHOD CHROMATOGRAPHIC R EF.
Measurement of N a
MODE
Micellar HPLC 38
Measurement of G q c HPLC 39
Measurement of N a, c Affinity 40
Generation of test peaks N/A General 41
Measurement of x and Oq c HPLC 42
Measurement of x /G q c HPLC 43
Measurement of M2, x and G q b HPLC 44
EMG curve fit N/A HPLC 45
Measurement of N,M2 a HPLC/MS 46
Measurement of CFOM a HPLC 47
Generation of test peaks N/A General 48
Measurement of N b, c HPLC 49
Measurement of M j ,M2 d HPLC/MS 50
Measurement of N a Micellar HPLC 51
Measurement of Mj a HPLC 52
Measurement of M2 a HPLC 53
Measurement of x and G q a,b OT HPLC/MS 54
Measurement of x and G q b OTHPLC 55
EMG curve fit N/A GC/MS 56
Generation of test peaks N/A General 57
Measurement of N a HPLC 58
EMG curve fit NA GC 59
Measurement of M2 a HPLC 60
Measurement of M2 a HPLC 61
Measurement of N,x, 0 (3,M^,M2 c HPLC 62
Measurement of M2 a HPLC 63
Measurement of N,x, Gq a HPLC 64
Measurement of x and Gq a, d HPLC 65
Measurement of M2 a HPLC 6 6
EMG curve fit N/A GC/MS 67
Measurement of N a Micellar HPLC 6 8
Measurement of N a Micellar HPLC 69
Measurement of N,x, Cq a HPLC 70
Measurement of x/Gq b GC 71
Measurement of M2 a HPLC 72
Measurement of M2 b tubular reactors 73
Measurement of N a Micellar HPLC 74
Measurement of M2 a HPLC 75
Measurement of M2 a HPLC 76
Measurement of M2 a, d HPLC 77
Measurement of M2 a HPLC 78
Measurement of M2 a HPLC 79
Measurement of N a HPLC 80
Measurement of M2 a HPLC/MS 81
Measurement of N a Micellar HPLC 82
Measurement of M2 a HPLC 83
Measurement of N a HPLC 84
2 7
Measurement of N a Micellar HPLC 85
Measurement of N a Chiral HPLC 8 6
Measurement of CFOM a HPLC 87
Measurement of N a HPLC 88
Measurement of CFOM a HPLC 89
Measurement of N a RP-HPLC 90
Measurement of N a, c HPLC 91
Generation of test peaks N/A General 92
EMG curve fit N/A General 93
Measurement of M2 a HPLC 94
Measurement of CFOM a GC 95
Measurement of N,t, Oq a HPLC 96
Measurement of N,t, Oq a HPLC 97
EMG curve fit N/A General 98
Measurement of % and Oq a GLC 99
Measurement of CFOM a Micellar HPLC 1 0 0
Measurement of N a Chiral HPLC 101
Measurement of N a GLC 102
EMG curve fit N/A GC-MS 103
EMG curve fit N/A HPLC 104
Measurement of N a CSP HPLC 105
Measurement of M2 a HPLC 106
Measurement of N a Micellar HPLC 107
Measurement of CFOM a HPLC 108
Simplex Optimization, curve fit N/A HPLC 109
Measurement of M2 a HPLC 110
2 8
EMG curve fit N/A GC/MS 111
Measurement of x, Oq b GAC 112
Measurement of N a Micellar HPLC 113
Measurement of CFOM a HPLC 114
Measurement of N a Micellar HPLC 115
Measurement of M2 a, b HPLC 116
Measurement of N a SFC 117
Measurement of M2 a HPLC 118
Measurement of N a GC 119
Measurement of M 1 a HPLC 1 2 0
Measurement of N a Cap. electrophoresis 121
Measurement of M2 a HPLC 122
Measurement of N a HPLC 123
a Used method of ref 13. b Used method of ref 124. c Used method of ref 125. d Used
method of ref 17.
2 9
t / o = 0 (Gaussian)
x/ o = 2
Figure 2-1. Illustration of the effect of x/ g q on EMG peak tailing.
3 0
Peak height = hp
Peak width = W = a + b
Peak asymmetry = (b/a)
Figure 2-2. Measurement of graphical peak parameters for use in asymmetry-based
equations.
31
R eferences
1) Sternberg, J.C. "Advances in Chromatography"; Giddings, J .C .; Keller, R.A.,
Eds.; New York: Marcel Dekker: 1966; Vol. 2, pp 205-270
2) Foley, J. P.; Dorsey, J. G. J. Chromatogr. Sci. 1984, 22, 40-46
3) Hanggi ,D.; Carr, P. W. Anal. Chem. 1985, 57, 2394-2395
4) Naish, P. J.; Hartwell, S. Chromatographia 1988, 26, 285-296
5) Press, W H .; Flannery, B.P.; Teukolsky, S.A.; Vetterling, W.A. Numerical
Recipes : The A n o f Scientific Computing , 1st ed.; Cambridge University Press:
Cambridge, 1986, pp. 163-165
6 ) Jeansonne, M. S.; Foley, J. P. J. Chromatogr. 1989, 461, 149-163
7) Sekulic, S.; Haddad, P.R. J. Chromatogr. 1988, 459, 65-77
8 ) Foley, J. P. Anal. Chem. 1987, 59, 1984-1987
9) Niessen, W.M.A.; van Vliet, H.P.M .; Poppe, H. Chromatographia 1985, 20,
357-363
10) Wu, N.S.; Hu, M. Chromatographia 1989, 28, 415-416
11) Delley, R. Anal. Chem. 1986, 58 2344-2346
12) Delley, R. Anal. Chem. 1985, 57,388
13) Foley, J. P.; Dorsey, J. G. Anal. Chem. 1983, 55, 730-737
14) Bidlingmyer, B. A.; Warren, F. V ., Jr. Anal. Chem. 1984, 56, 1583A-1596A
15) Dezaro,D.A.; Floyd, T. R.; Raglione, T. V.; Hartwick, R. A. Chromatogr.
Forum. 1986, May-June, 34-37
16) Jung, K. H.; Yun , S. J.; Kang, S. H. Anal. Chem. 1984, 56, 457-462
17) Anderson, D. J.; Walters, R. R. J. Chromatogr. Sci. 1984, 22, 353-359
18) Frans, S.D.; McConnell , M.L.; Harris, J.M. Anal. Chem. 1985, 57, 1552-
1559
19) Foley, J.P. J. Chromatogr. 1987, 384, 301-303
20) Binsheng, L.; Peichang, L. HRC & CC. 1987, 10, 449-454
21) Haddad, P.R.; Sekulic, S. J. Chromatogr. 1988, 459: 79-90
22) Haddad, P.R.; Sekulic, S. J. Chromatogr. 1989, 485: 501-515
23) Schoenmakers, P. J.; Strasters , J. K.; Bartha, A. J. Chromatogr. 1988, 458:
355-370
24) Dyson, N. Chromatographic Integration Methods., R.M. Smith, ed. Royal
Society of Chemistry: Cambridge, 1990; pp. 17-67
25) Papas, A. N. CRC Crit. Rev. Anal. Chem. 1989, 20: 359-404
26) Papas, A. N.; Delaney, M. F. Anal. Chem. 1987, 59, 54A-59A
27) Schudel, J. V.H.; Guiochon, G. J. Chromatogr. 1988, 457, 1-12
28) Doehl, J.; Greibrokk, T. J. Chromatogr. Sci. 1987, 25, 99-103
29) Colmsjo, A.L.; Ericsson, M. W. J. Chromatogr. 1987, 398, 63-71
30) Rossi, D. T. J. Chromatogr. Sci. 1988, 26, 101-105
31) Papas, A. N.; Tougas, T. P. Anal. Chem. 1990, 62, 234-239
32) Brooks, S. H.; Leff, D. V.; Torres, M. A. H.; Dorsey, J. G. Anal. Chem.
1988, 60, 2737-2744
33) Brooks, S. H.; Dorsey, J. G. Anal. Chim. Acta. 1990, 229, 35-46
34) Brooks, S. H.; Williams, R. N. ; Dorsey, J. G. Anal. Lett. 1988, 21, 583-598
35) Vanderslice, J.T.; Rosenfeld, A.G.; Beecher, G.R. Anal. Chim Acta 1986,
179, 119
36) Hemandez-Torres, M. A.; Khaledi, M. G.; Dorsey, J. G. Anal. Chim. Acta.
1987, 201, 67-76
37) Locascio-Brown, L.; P lan t, A. L.; Durst, R. A. Anal. Chem. 1988, 60, 792-
3 3
38) Dorsey, J. G.; DeEchegaray, M. T.; Landy, J. S. Anal. Chem. 1983, 55, 924-
928
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Hearn, M. T.W.; Grego, B. J. Chromatogr. 1983, 255, 125-136
Walters, R. R. Anal. Chem. 1983, 55, 1395-99
Wegscheider, W.; Lankmayr, E. P.; Otto, M. Anal. Chim. Acta 1983, 150, 87-
103
Haddad, P.R.; Low , G.K.C.; Heckenberg, A. L. Chromatographia 1984, 18,
417-423
Nondek, L. Anal. Chem. 1984,56, 1192-1194
van den Berg, J.H.M.; Horsels, H.W.M.; Deelder, R.S. J. Liq. Chromatogr.
1984, 7, 2351-2365
Peichang, L. and Xiaoming, L. J. Chromatogr. 1984, 292, 169-188
Hayes, M.J.; Schwartz, H.E.; Vouros, P.; Karger, B .L .; Thruston, A.D., Jr.;
McGuire, J.M. Anal. Chem. 1984, 56, 1229
Lansden, J. A. J AO AC. 1984, 67, 728-731
Davis, J. M.; Giddings, J. C. J. Chromatogr. 1984, 289, 277-298
Marshall, D. B.; Stutler, K. A. ; Lochmuller, C. H. J. Chromatogr. Sci. 1984,
22, 217-220
Karger, B. L. and Vouros, P. J. Chromatogr. 1985, 323, 13-32
Landy, J. S.; Dorsey, J. G. Anal. Chim. Acta. 1985, 178, 179-188
Nondek, L.; Minarik, M. J. Chromatogr. 1985, 324, 261-268
Maris, F.A.; Van Der Vliet, A.; Geerdink, R.B.; U.A. Th. Brinkman. J.
Chromatogr. 1985, 347, 75-81
Niessen, W.M.A.; H. Poppe. J. Chromatogr. 1985, 323, 37 - 46
van Vliet, H.P.M.; Poppe, H. J. Chromatogr. 1985, 346, 149-160
Garland, W.A.; Barbalas, M.P. J. Clin. Pharmacol. 1986, 26, 412-418
3 4
57) Voigtman, E.; Winefordner, J. D. Anal. Inst. 1986, 15, 309-328
58) Sands, B.W.; Kim, Y.S.; Bass, J.L. J. Chromatogr. 1986, 360, 353-369
59) Jung, K.H.; Shin, Y.H. B. Kor. Chem. 1986, 7, 403-405
60) Krull, I. S.; Bushee, D.S.; Schleicher, R.G.; Smith, S.B., Jr. Analyst. 1986,
111,345-349
61) Marshall, D. B.; Bums, J. W.; D. E. Connolly. J. Chromatogr. 1986, 360, 13-
24
62) Bacaloglu, R.; Co tare a, L.; Isfan, Isfan, A.; C. Rev RO Chim. 1986, 31, 403-
409
63) Gluckman, J.C.; Barcelo, D.; De Jong, G .J .; Frei, R .W .; Maris, F.A.;
Brinkman, U.A. Th. J. Chromatogr. 1986, 367, 35-44
64) Minkler, P. E.; Erdos, E. A.; Ingalls, S. T.; Griffin , R. L.; Hoppel, C. L. J.
Chromatogr. 1986, 380, 285-299
65) van Vliet, H.P.M.; Bruin, G.J.M.; Kraak, J.C.;Poppe, H. J. Chromatogr.
1986, 363, 187-198
6 6 ) Tomellini, S. A.; Hsu , S.-H.; Hartwick, R. A. Anal. Chem. 1986, 58, 904-
906
67) Garland, W.A.; Hess, J.; Barbalas, M.P. Trends in Analytic. Chemistry 1986,
5, 132-138
6 8 ) Hemandez-Torres, M. H. ; Landy , J. S.; Dorsey, J. G. Anal. Chem. 1986,
58, 744-747
69) Mullins, F.G.P.; Kirkbright, G. F. Analyst. 1986, 111, 1273-1279
70) Pauls , R.E.; McCoy, R.W. J. Chromatogr. Sci. 1986, 24, 66-69
71) Li, H.; Lin, B.; Luo, C ; Lu, P. Sci. Sin. B. 1986, 29, 1248-1260
72) Marshall, D. B.; Cole, C.L.; Connolly, D. E. J. Chromatogr. 1986,361,71-82
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
3 5
Leclerc, D. F.; Bloxham, P. A.; Toren, E. C., Jr. Anal. Chim. Acta. 1986, 184,
173-185
Dorsey, J. G.; Hinze, W. L.; Armstrong, D. W. , ed., ACS Symp. Ser. 1987,
342, 105-114
Selavka, C. M.; Jiao, K.-S.; Krull, I. S. Anal. Chem. 1987, 59, 2221-2224
Shelley , D. C.; Colkins, T. J. J. Chromatogr. 1987, 411, 185-199
De Ruiter, C.; Wolf, J.H.; Th. Brinkman, U.A.; Frei, R.W. Anal. Chim. Acta
1987, 192, 267-275
Bushee, D.S.; Krull, I. S. Anal. Chim. Acta 1987, 194, 235-245
Barcelo, D.; Maris, F.A.; Frei, R.W.; de Jong, G.J.; Th. Brinkman, U.A.
Intern. J. Environ. Anal. Chem. 1987, 30, 95-104
Inman, E. L . . J. Chromatogr. 1987, 410, 363-372
Kresbach, G. M.; Baker, T. R.; Nelson, R. J.; Wronka, J.; Karger, B. L.;
Vouros, P. J. Chromatogr. 1987, 394, 89-100
Mullins, F. G.P.; Hinze, W. L. ; Armstrong, D. W. , Eds., ACS Symp. Ser.
1987, 342, 115-129
de Jong, G.J.; Lammers, N.; Spruit, F.J.; C. Dewaele; M. Verzele. Anal. Chem.
1987, 59, 1458-1461
Berthod, A.; Li, K. P.; Yu, T.; Winefordner, J. D. Anal. Chem. 1987, 59,
1485-1488
Dorsey, J. G. Advances in Chromatography, Vol. 27. Giddings, J.C. , Grushka,
E. and Brown, P.R., Eds. Marcel Dekker, New York, 1987, pp. 167-214
Tambute, A.;Begos, A.; Lienne, M.; Caude , M.; Rosset, R. J. Chromatogr.
1987, 396, 65-81
Karcher, B.D.; Krull, I .S ., Schleicher, R.G.; Smith, S.B. Chromatogr. 1987,
24, 705-712
3 6
8 8 ) Foley , J. P.; May, W. E. Anal. Chem. 1987, 59, 110-115
89) Minkler, P. E.; Ingalls, S. T.; Griffin , R. L.; Hoppel, C. L. J. Chromatogr.
1987, 413, 33-42
90) Gorse, J.; Burke, M.F. J. Liq. Chromatogr. 1987, 10, 1049-1064
91) Marshall, D. B.; Cole , C.L.; Norman, A. D. J. Chromatogr. Sci. 1987, 25,
262
92) Strasters, J.K.; Bartha, A.; B illiet, H.A.H.; de Galan, L. J. Liq. Chromatogr.
1988, 11, 1827-1854
93) Felinger, A.; Pap, T.; Inzdedy, J. Magy KemFo 1988,94,227-231
94) Colgan, S. T.; Krull, I. S.; Dorschel, C ; Bidlingmeyer, B. A. J. Chromatogr.
Sci. 1988, 26, 501-512
95) Roberts , D.; Bertsch, W. HRC & CC 1988, 11, 783-787
96) Chai, B.; Minkler, P. E.; Hoppel, C. L. J. Chromatogr. 1988, 430, 93-101
97) Minkler, P. E.; Hoppel, C. L. J. Chromatogr. 1988, 428, 388-394
98) Peichang, L.; Hongxin, H. J. Chromatogr. 1988, 452, 175-189
99) Petrishchev, S.D.; Sakodynskii, K.I. Russ. J. Phys. Chem. 1988, 62, 1101
100) Berthod, A.; Dorsey, J. G. Analusis. 1988, 16, 75-89
101) Lienne, M.; Caude, M.; Rosset, R.; Tambute, A. J. Chromatogr. 1988, 448,
55-72
102) Lekova, K.; Bursukova, M.; Ivanichkova, S.; Bojkova, N.; Petkova, R. J.
Chromatogr. 1988,446,31-38
103) Barbalas, M.P.; Hess, J.; Sandor, D.; Garland, W.A. J. Pharm. Sci. 1988, 77,
679
104) Bauer, J. E.; Kristensen, E. W.; Wightman, R. M. Anal. Chem. 1988, 60,
2334-2338
105) Wilson, T. D. J. Chromatogr. 1988, 448, 31-39
3 7
106) Karlsson, K.-E.; Novotny, M. . Anal. Chem. 1988, 60, 1662-1665
107) Berthod , A.; Roussel, A. J. Chromatogr. 1988, 449, 349-360
108) Floyd, T.R. Chromatographia 1988, 25, 791-796
109) Peichang , L.; Hongxin, H. J. Chromatogr. Sci. 1989, 27, 690-697
110) Chou, T.-Y.; Gao, C.-X.; Grinberg , N.; Krull, I. S. Anal. Chem. 1989, 61,
1548-1558
111) Garland, W. A.; Crews, T.; Brown, S. Y.; Fukuda, E. K. J. Chromatogr.
1989, 472, 250-255
112) Baker, R.J.; Suffet, I. H.; Yohe, T. L. Adv. Chem. Ser. 1989, 219, 533-548
113) Borgerding, M. F.; Hinze, W. L.; Stafford, L. D.; Fulp, G. W., Jr.; Hamlin, W.
C . , Jr. Anal. Chem. 1989, 61, 1353-1358
114) Baur , J. E.; Wightman, R. M. J. Chromatogr. 1989, 482, 65-73
115) Borgerding, M. F.; Williams, R. L., Jr.; Hinze, W. L.; Quina, F. H. J. Liq.
Chromatogr. 1989,12, 1367-1406
116) Kientz, Ch. E.; Verweij, A.; Boter, H.L.; Poppema, A.; Frei, R .W .; De Jong,
G.J.; Th. Brinkman, V.A. J. Chromatogr. 1989, 467, 385-394
117) Crow , J.A. ; Foley, J.P. HRC & CC. 1989, 12, 467-470
118) Gao, C.-X.; Chou , T.-Y.; Krull, I. S. Anal. Chem. 1989, 61, 1538-1548
119) Villalobos, R.; Annino, R. HRC & CC 1989, 12, 149-160
120) de Bokx, P.K.; Boots, H.M.J. J. Phys. Chem. 1989, 93, 8243-8248
121) Guttman, A.; Cohen, A.S.; Heiger, D.N.; Karger, B.L. Anal. Chem. 1990,
62,137-141
122) Gao, C.-X. ; Krull, I. S.; Trainor, T. J. Chromatogr. Sci. 1990, 28, 102-108
123) Berthod, A. J. Liq. Chromatogr. 1989,12,1187-1201
124) Yau, W.W. Anal. Chem. 1977, 49, 395-398
125) Barber, W.E.; Carr, P.W. Anal. Chem. 1981, 53, 1939-1942
Chapter 3
IM PROVED EQUATIONS FOR THE CALCULATION OF
CHROM ATOGRAPHIC FIGURES OF MERIT (CFOMs) FOR
IDEAL AND SKEW ED CHROMATOGRAPHIC PEAKS
3 8
INTRO DUCTIO N
3 9
In a recent publication we presented an equation-based approach for the accurate
measurement of chromatographic peak statistical moments, excess, and skew (1). These
empirical equations, based on the exponentially modified Gaussian (EMG) function,
utilize the peak width, asymmetry, and in some cases, the peak height. The EMG
function, resulting from the convolution of Gaussian and exponential decay functions,
can model any tailed peak more accurately than a Gaussian function. Due to various intra
and extracolumn band broadening processes, real chromatographic peaks are seldom
symmetric, so that use of a tailed peak model should be more accurate. Hie EMG
function has been justified both theoretically (2,3,4) and experimentally (5,6) and has
been thoroughly reviewed (7,8). Because a convenient procedure for verification of the
EMG character of chromatographic peaks exists (5), these equations can be confidently
applied once EMG peak shape has been established.
A variety of approaches has been used to make this model more practical for
routine use.(5,9-15) One particularly popular approach has been to relate the graphical
parameters a and b, (Figure 3-1), measured at a particular fraction of the peak height, to
the fundamental EMG parameters, T and Gq . Relationships between a and b and the
EMG parameters have been reported in the form of both graphical curves and empirical
equations. Both techniques circumvent tedious, computer intensive curve fitting.
Equations reported by Foley and Dorsey (13) in 1983 have perhaps shown the greatest
routine applicability, based on the number of citations to that article. In that work,
equations were derived for the calculation of many important chromatographic peak
parameters (i.e., T, Gq , statistical moments, skew, excess and efficiency) that were
collectively termed chromatographic figures of merit (CFOMs). Equations based on the
measurement of b and a at 10%, 30%, and 50% of the peak height fraction were
4 0
reported, although only the equations at 1 0% were recommended for calculation of the
CFOMs. The equations at 30% and 50% were intended solely for peak modeling.
Although the CFOM equations reported by Foley and Dorsey earlier were very
accurate and precise, they were meant to be used with manually measured b and a values
where the accuracy of these measurements was the limiting factor in the accuracy of the
equations. Due to the greater accuracy available with modem electronic integrators and
data systems, the accuracy of CFOM measurement may be unnecessarily limited by the
equations themselves. Two further limitations are: (i) the somewhat narrow asymmetry
(b/a) range of the equations; and (ii) the absence of equations based on peak
measurements higher than 50% of the peak height
With regard to (ii), the ability to use b and a from higher peak height fractions is
desirable when peaks are overlapped because distortion from the adjacent peak is less at
higher peak height fractions. We have shown that accurate measurement of CFOMs
utilizing b and a measurements at 75% of the peak height is possible for both overlapping
and resolved peaks using improved equations we have not reported before.(l) In this
paper we report those improved equations (from the standpoint of the previous
limitations) for the measurement of peak area, variance (M2 ), third (M3 ) and fourth (M4 )
statistical moments, excess, and skew.
EXPERIM ENTAL
An Apple Macintosh Plus microcomputer programmed in Microsoft Basic was
used for EMG peak generation. The EMG function and universal data were calculated as
described before (7) via a search algorithm. All polynomial curve fitting was done using
commercially available software.
41
EMG Peak Generation. Equation 1 shows the form of the EMG function
used.z
hEMG(t)-A — V2 "ex p [I (S i)2 . ( i l i f i ) ]X X X
A is the peak amplitude, X is the exponential modifier, G q is the standard deviation of the
unmodified Gaussian , tQ is the retention time of the unmodified Gaussian and z = (t -
tc ) /O G - G q / t . The ratio X/Oq is used to describe the overall shape of an EMG peak; at
X/<J<3 close to zero, the peak approaches Gaussian shape, while higher T /G q values give
greater tailing. Values of A = 1, Iq = 100 and G q - 5 were used for evaluation of the
function. The times for tR (EMG peak apex), ta, and t^ (Figure 3-1) at r (peak height
fraction) = 0.10, 0.25, 0.5, 0.71, 0.73, 0.75, 0.77, and 0.79 for V G q values from 0 to
4.5 in increments of 0.05 were obtained to within 0.001 using a search algorithm. The
peak height (hp) at each V G q was determined concurrently with tR using the same search
algorithm. From these values, a, b, W (a +b), and asymmetry (b/a) could be calculated.
The values of asymmetry (b/a), W /G q , O r - 1q ) / G q and hp, termed universal data, were
used in the following derivations.
Our approach was to derive width and asymmetry based equations for G q and
M2 , to use equation 2 to determine X, and to use the equations (13) listed in column 1 of
Table 3-2 for the computation of the remaining moments, excess, and skew.
METHOD OF DERIVATION
12]
4 2
We found this approach slightly more accurate than developing asymmetry based
equations for Gq and T/Gq, and then calculating 1 from their product Due to the
inferiority of the latter approach, it is not discussed further. Equations at several peak
heights were derived, with those for 0.71 < r < 0.79 developed for purposes of
averaging to reduce uncertainty and bias.
Derivation of Equations for G q . For each r, construction of a plot of
Wj/Gq v s asymmetry (corresponding to the T/Gq range given above) gave results similar
to Figure 3-2 for r = 0.25. As shown, the plot is curved at the lower b/a values and
linear at the higher values. Therefore, a quadratic least-squares fit was used at lower
asymmetries and a linear least-squares fit was used for higher asymmetries resulting in
two equations at each r. G q could then be related to width and asymmetry as shown in
equation 3.
f(b/a) [3]
where f(b/a) is the resulting fitted equation of the form f(b/a) = Co + Cj(b/a) + C2 (b/a)“
(The coefficient C2 is 0 for f(b/a) resulting from the linear fit). Table 3-1 lists the
coefficients obtained at each r for both linear and quadratic fits and their valid asymmetry
ranges. The cut point between curved and linear fits was determined so that the relative
error in G q over the entire b/a range was minimized.
Derivation of Equations for M2 . Analogous to the derivation of G q , plots
of M2/W r 2 v s asymmetry were constructed at each r. However, as shown in Figure 3-3
for r=0.25, curvature was evident over the entire b/a range. Because an accurate
quadratic fit over the entire b/a range was not possible, two least-squares quadratic fits
were used, again choosing the cut point to minimize relative error over the entire
4 3
asymmetry range. Equation 4 shows the resulting relationship between M2 , Wf2 and
f(b/a).
M2 = Wr2f(b/a) [4]
The values of the coefficients for calculation of f(b/a) are given in Table 3-1. At r values
of 0.71 to 0.79, the plots were similar to those obtained for the G q derivations, thereby
allowing linear and quadratic fits for the two sections of curves, as before.
Derivation of Area Equations. Area equations based on width, asymmetry,
and peak height measurements at r = 0.10,0.25, 0.50, and 0.75 have been previously
reported (5) along with the method of their derivation. Additional area equations at r =
0.71,0.73,0.77, and 0.79, were derived in this report for purposes of comparison.
The area equations for these r values are reported in Table 3-3 along with the area
equations at r = 0.10, 0.25, 0.50, and 0.75 previously reported (5).
RESULTS AND DISCUSSIO N
Accuracy of CFOM Equations. Tables 3-1 and 3-2 show the maximum
errors over the T /G q range 0.3 to 4.3 for the equations derived for the various CFOMs
considered in this study. In general, the equations for G q are accurate to within 1%
while those for M2 are accurate to within 2%. Table 3-4 compares the results of the
maximum errors reported for the Foley-Dorsey (13) equations to those reported in Tables
3-1 and 3-2. Because Anderson and Walters (12) also derived equations for M j, M2,
Gq , and T, the relative errors for their equations are reported in Table 3-4 for purposes of
comparison. Only CFOMs at r = 0.10 are compared, because the Foley-Dorsey
equations at other r values are useful only for peak modeling and should not be used for
4 4
CFOM calculation. For all CFOMs considered, the accuracy of the present set of
equations is better than those derived by the previous methods. Although the Anderson -
Walters equations provide a greater valid asymmetry range, most experimental peaks will
have asymmetries within the valid range of the present equations. Because our equations
give better accuracy than the other methods while providing for the calculation of the
higher moments (whereas the Anderson - Walters equations do not), they should be the
method of choice in most instances. Also, our equations for higher r (not available
previously) allow the calculation of CFOMs when the lower peak height fractions are not
accessible due to peak overlap.
Precision of CFOM Equations. Also reported in Tables 3-1 and 3-2 are the
relative uncertainties of the CFOMs based on error propagation. Percent relative
standard deviations (%RSDs) of 0.25% and 0.5% for W and asymmetry, respectively,
were propagated through the various equations. Note that the actual precision of the
width and asymmetry measurements may be better for some electronic integration
systems. As seen in Table 3-1, the uncertainties for Oq and M2 are in general less than
2%, and less than 1% at some r values. From Table 3-2 it is evident that the values of
the higher moments, excess, and skew are slightly more imprecise. Where a range for
the uncertainty is reported, the larger number refers to the least asymmetric peak, while
the smaller number refers to most asymmetric peak. In general, the uncertainty decreases
exponentially from the larger value to the lower value.
Accuracy and Precision of Area Equations. The maximum errors and
uncertainties for the area equations derived here, along with those reported previously,
are shown in Table 3-3. The errors at all r are well within 1.5%, and are not more than
1.0% in most cases. Assuming the same %RSD values for width and asymmetry as
before, and a %RSD in peak height (hp) of 0.25%, the resulting uncertainties for all of
the area equations are less than 0 .6 %.
4 5
C O N C LU SIO N
It is hoped that the more accurate equations presented here will facilitate greater
use of the EMG peak model within electronic integrators and data systems. Because
widths and asymmetries are now commonly measured by many chromatographic data
systems, integration of these equations into existing software should be easy, while
providing better accuracy in the desired CFOMs. Also, as shown before (1), these
equations provide for the direct CFOM measurement of the first eluted peak in an
overlapped pair, even when the valley height exceeds 50% of the first peak's height.
ACKNOW LEDGEM ENT
The authors gratefully acknowledge the partial support of this work by the Computer-
Aided Chemistry Division of the Perkin-Elmer Corporation.
Table 3-1. Coefficients, Accuracy, and Precision of Equations for Calculating Oq, and M2 of Gaussian and Exponentially Modified Gaussian (EMG) Peaks.3
peak height asymmetry asymmetry coefficients maximum relativefraction, r range,t/o range, (b/a)r c 2 Cl c 0 error b uncertainty c
0 .1 0 0.30-1.15 1.03-1.46 -1.2951 6.6162 -0.9516 -0 .2 ,+0 .6 ±0.52
1.20-4.30 1.46-3.67 0 3.3139 1.11470.135 0.30-0.65 1.02-1.14 -4.1347 12.7655 -4.5742 -0.2, +0.3 ±0.53
0.65-4.30 1.14-3.42 0 3.1665 1.00510.25 0.30-1.05 1.02-1.26 -2.4352 8.8715 -3.0485 ±0 .2 ±0.52
1.10-4.30 1.26-2.85 0 2.8504 0 .6 8 8 8
0.50 0.30-1.25 1 .0 1 - 1 .21 -4.2670 12.5178 -5.8561 -0.4, +0.8 ±0 .8
1.30-4.30 1.21-2.09 0 2.4685 0.09810.71 0.30-1.85 1.01-1.23 -2.7901 9.0992 -4.6165 -0.5, +0.2 ± 1 .0
1.85-4.30 1.23-1.64 0 2.2527 -0.39830.73 1.30-1.85 1 .0 1 - 1 .2 2 -2.9116 9.3374 -4.8028 -0.5, +0.2 ± 1 .0
1.85-4.30 1.22-1.60 0 2.2449 -0.46660.75 0.30-1.25 1 .0 1 - 1 .1 2 -7.4174 18.8306 -9.8710 ±0.7 ± 1 .2
1.30-4.30 1.12-1.56 0 2.3245 -0.6604
0.77 0.30-1.85 1.01-1.19 -3.3185 10.1672 -5.3703 -0.5, +0.2 ± 1.1
1.85-4.30 1.19-1.52 0 2.2326 -0.61110.79 0.30-1.90 1.01-1.19 -3.3724 10.2546 -5.4779 -0.5, +0.2 ± 1 .2
1.90-4.30 1.19-1.49 0 2.2268 -0.6859
Table 3-1. Cont'd
M2
peak height asymmetry asymmetryfractions range,t/c range, (b/a)r C2 x 10
0 .1 0 0.30-1.15 1.03-1.46 0.1270
1.15-4.30 1.46-3.67 -0.02990.135 0.30-2.85 1.03-2.50 -0.0378
2.85-4.30 2.50-3.42 0
0.25 0.30-0.70 1.02-1.13 2.34180.70-4.30 1.13-2.85 -0.1842
0.50 0.30-1.25 1 .0 1 -1 .21 7.91611.30-4.30 1.21-2.09 -1.3369
0.71 0.30-1.30 1.01-1.14 52.7771.30-4.30 1.14-1.64 0
0.73 0.30-1.35 1.01-1.14 63.5451.35-4.30 1.14-1.60 0
0.75 0.30-1.35 1.01-1.13 84.4481.40- 4.30 1.13-1.56 0
0.77 0.30-1.70 1.01-1.17 79.7711.70-4.30 1.17-1.52 0
0.79 0.30-1.70 1.01-1.16 103.9561.70-4.30 1.16-1.49 0
coefficients maximum relativeCj x 10 Co x 10 error b uncertainty c
-0.06458 0.4766 -0 .2 , +0 .6 ±0 .6
0.3569 0.19090.4993 0.1470 -1.4, +0.5 ±0 .6
0.2493 0.5394-4.1376 2.6930 -0.2, +0.3 ±0 .8
1.6032 -0.5715-12.8196 6.6750 -0.6, +0.4 ±1.5
9.3616 -6.6407-96.293 47.101 -0 .6 , +0 .8 ±2.425.039 -22.696
116.389 56.743 -1.0, +0.5 ±2 .6
29.935 -27.601157.205 77.0333 -1.3, +0.7 ±2 .8
35.9111 -33.5484142.903 67.779 -1.3, +0.7 ±2.744.119 -41.993
188.641 89.840 -1.3, +0.7 ±2.954.353 -52.398
Table 3-1. cont'd
a General form of the equations for each parameter: G q = Wr / f(b/a); M2 = Wr 2 x f(b/a), where Wr = width of peak at the peak
height fraction given by the subscript, b/a = an asymmetry factor measured at the same peak height fraction as the width, and f(b/a) = Co + Ci(b/a) + C2 (b/a)2.
b Maximum error of the equations over the asymmetry range.
c Percent relative standard deviation (%RSD) predicted from error propagation, assuming RSDs of 0.25% and 0.5% for Wr and
b/a. Note that the precision of W and b/a can be better than what we have assumed for many data acquisdon systems. Where a range is reported, the larger number refers to the least asymmetric peak (smallest b/a value) and the smaller number refers to the most asymmetric peak (largest b/a value). The uncertainty decreases exponentially from the larger value to the lower value as b/a increases.
00
Table 3-2. Error and Uncertainty of Chromatographic Peak Parameters calculated from Oq and M2 using the Equations in Table I.a
Parameter
1.T = (M2 -(7g2)1/2
2) l/G
Peak height asymmetry asymmetry maximum relativefraction ,r range, x/a range, (b/a)r error uncertainty
0 .1 0 0.5- 4.3 1.09-3.67 -0.4, +0.2 2.4-0.3
0.135 0.5-4.3 1.08-3.42 -0.3, +0.7 2.6-0.3
0.25 0.5-4.3 1.07-2.85 ±0.4 3.0-0.30.50 0.5-4.3 1.04-2.09 -0.4, +0.6 4.2-0.4
0.71 0.5-4.3 1.03-1.64 -0.4, +0.6 5.8-0.60.73 0.5-4.3 1.03-1.60 -0.7, +0.4 6 .1-0.70.75 0.5-4.3 1.03-1.56 ±0 .8 6.5-0.70.77 0.5-4.3 1.02-1.52 -0.5, +1.0 7.0-0 .8
0.79 0.5-4.3 1.02-1.49 -0 .8 , + 1 .0 7.3-0.8
0 .1 0 0.5-4.3 1.09-3.67 -0.8, +0.3 2.5-0.6
0.135 0.5-4.3 1.08-3.42 -0.3, +0.8 2.7-0.60.25 0.5-4.3 1.07-2.85 ±0.5 3.0-0.60.50 0.5-4.3 1.04-2.09 -1 .0 , +0 .6 4.3-0.70.71 0.5-4.3 1.03-1.64 -0.6, +0.9 5.9-0.90.73 0.5-4.3 1.03-1.60 -0.6, +0.9 6.2-0.90.75 0.5-4.3 1.03-1.56 -0.9, +1.4 6 .6 - 1 .0
0.77 0.5-4.3 1.02-1.52 -0.6, +0.9 7.1-1.00.79 0.5-4.3 1.02-1.49 ±0 .8 7.4-1.1
Table 3-2. Cont'd
Parameter Peak height asymmetry asymmetryfraction/ range, t/o range, (b/a)r
3) M3
4) M4
0 .1 0 0.5-4.3 1.09-3.67
0.135 0.5-4.3 1.08-3.420.25 0.5-4.3 1.07-2.850.50 0.5-4.3 1.04-2.090.71 0.5-4.3 1.03-1.640.73 0.5-4.3 1.03-1.600.75 0.5-4.3 1.03-1.560.77 0.5-4.3 1.02-1.520.79 0.5-4.3 1.02-1.49
0 .1 0 0.5-4.3 1.09-3.67
0.135 0.5-4.3 1.08-3.420.25 0.5-4.3 1.07-2.850.50 0.5-4.3 1.04-2.090.71 0.5-4.3 1.03-1.640.73 0.5-4.3 1.03-1.600.75 0.5-4.3 1.03-1.560.77 0.5-4.3 1.02-1.520.79 0.5-4.3 1.02-1.49
maximumerror
-1.0, +0.5
-0.9, +2.2 - 1.2, + 1.1 -1.2, +1.9 -1.1, +1.7 -2.0, +1.4 -2.4, +2.5 -1.5, +2.9 -2.5, +2.9
-0.6, +0.9
-0.5, +1.2 -0.5, +0.9 ±0.9 - 1.2, + 1.6 -2.3, +1.4 -2.8, +2.3 -1.7, +1.6 -2.9, +1.7
relativeuncertainty
7.3-0.9
7.8-0.98.8-0.9 12.6- 1.217.5-2.018.3-2.019.6-2.120.9-2.322.0-2.4
3.1-1.2
3.1-1.23.5-1.24.9-1.6 6.7-2.67.0-2.77.6-2.88.0-3.0 8.4-3.2
Table 3-2. Cont'd
Parameter
5) skew = M3 / M 2 ^
6 ) excess = M4 / W ifi - 3
Peak height asymmetry asymmetryfraction,r range, x/o range, (b/a)r
0 .1 0 0.5-4.3 1.09-3.67
0.135 0.5-4.3 1.08-3.420.25 0.5-4.3 1.07-2.850.50 0.5-4.3 1.04-2.090.71 0.5-4.3 1.03-1.640.73 0.5-4.3 1.03-1.600.75 0.5-4.3 1.03-1.560.77 0.5-4.3 1.02-1.520.79 0.5-4.3 1.02-1.49
0 .1 0 0.5-4.3 1.09-3.67
0.135 0.5-4.3 1.08-3.420.25 0.5-4.3 1.07-2.850.50 0.5-4.3 1.04-2.090.71 0.5-4.3 1.03-1.640.73 0.5-4.3 1.03-1.600.75 0.5-4.3 1.03-1.560.77 0.5-4.3 1.02-1.520.79 0.5-4.3 1.02-1.49
maximum relativeerror uncertainty
-1 .0 , +0 .6 7.3-1.2
-0.7, +1.5 7.9-1.3-1.2, +0.9 8.9-1.3-1.2, +1.3 12.7-1.6-1 .0 , +1.1 17.7-2,7-1.1, +0.7 18.5-2.8-2.3, +1.8 19.8-3.0-0.4, +2.1 21.2-3.1-0.9, +1.8 22.2-3.3
-1.3, +0.8 42.1-2.5
-0.9, +2.0 45.4-2.6-1 .6 , + 1 .2 51.1-2.7-1 .6 , + 1 .8 73.0-3.4±1.4 102-5.5-1.4, +0.9 106-5.8-3.0, +2.5 105-6.1-0 .6 , +2 .8 122-6.4-1.2, +2.5 128-6.8
Table 3-2. Cont'd
a Refer to Table 3-1 for an explanation of terms.
5 3
Table 3-3. Peak Area Equations for EMG and Gaussian Peaks a
r equation %RE b %RSDC
0 .1 A = 0.586 hp W0.i (b/a)-0 1 3 3 ±0.50 0.36
0.135 A = 0.631 hp Wo. 135 (b/a)-° -106 -0.3, +0.6 0.36
0.25 A = 0.753 hp W0.25 -1 .0 , +0 .6 0.35
0.50 A = 1.07 hp W0 .5 (b/a)* 0 -236 -1 .2 , + 1 .0 0.37
0.71 A = 1.514 hp W0 .7 i (b/a)+0-591 - 1 .1 , +0 .6 0.46
0.73 A = 1.58 hp W0.73 (b/a)* 0 -661 -0.7, +1.1 0.48
0.75 A= 1.64 hp W0.75 (b/a)*0-71? -1 .1, +0 .6 0.50
0.77 A = 1.73 hp W0.77 (b/a)* 0 -763 -0.9, +0.6 0.52
0.79 A = 1.82 hp W0.79 (b/a)* 0 -836 -1.0, +0.4 0.55
a A = Area of peak, hp = peak height, W = width of peak at designated peak height
fraction, and b/a is the asymmetry factor measured at the same peak height fraction as the width. See Figure 1.
b Largest relative error in Area over the interval 0 < x/ct < 4.3, expressed as a percentage.
c Percent relative standard deviation (%RSD) of the area measurement predicted from error propagation assuming RSDs of 0.25%, 0.25%, and 0.5% for hp, W, and b/a. Note that for many data systems the precision of hp, W, and b/a can be better than
what we have assumed.
Table 3-4. Comparison of maximum errors for the CFOMs reported here and those reported previously.
Parameter3 This Report Foley and Dorsey b Anderson and Waltersc
Valid b/a range
maximumerror
Valid b/a range
maximumerror
Valid b/a range
maximumerror
<*G 1.03-3.67 -0 .2 ,+0 .6 1.09-2.76 -1.0, +0.5 1.0-5.21 -1.3, +1.6
T 1.09-3.67 -0.4, +0.2 1.09-2.76 -1.0, +3.5 1.0-5.21 -4.0, +0.4
t /o G 1.09-3.67 -0.8, +0.3 1.09-2.76 -1.0, +4.5 d d
M! d d 1.00-2.76 ± 1 .0 1.0-5.21 -0 .2 , +0 .1
m 2 1.03-3.67 -0 .2 , +0 .6 1.00-2.76 -1.5, +0.5 1.0-5.21 -0.7, +1.0
m 3 1.09-3.67 -1.0, +0.5 1.09-2.76 -2.5, +10.5 d d
M4 1.09-3.67 -0.6, +0.9 1.09-2.76 -3.0, +1.5 d d
skew 1.09-3.67 -1.0 , +0 .6 1.09-2.76 -1.0 , + 1 0 d d
excess 1.09-3.67 -1.3, +0.8 1.09-2.76 -1.5, +14 d d
014k
Table 3-4. Cont'd
a Parameters based on r = 0.10 measurements,
b From Foley, J.P.; Dorsey, J.G. Anal. Chem. 1983, 55, 730-737
c From Anderson, D J.; Walters, R.R. / . Chromatogr. Sci. 1984,22, 353-359
d Equations not derived for this parameter.
5 6
Peak height = h_pPeak width = W = a + b
Peak asymmetry = (b/a)
Figure 3-1. Graphical parameters necessary for calculation of various CFOMs.
5 7
9.0 n
8.0
7.00.25
G
6.0
5.0
4.0-
3.01.0 1.5 2.0 3.02.5
x a '0.25
Figure 3-2. Plot construction for determination of f(b/a) for calculation of <Tg at r=0.25.
Note curvature at lower asymmetries.
5 8
0.25
'0.25
0.15
0.053.02.52.01.51.0
a 7 0.25
Figure 3-3. Plot construction for determination of f(b/a) for calculation of M2 at r =
0.25. Due to curvature over entire asymmetry range, two least-squares
quadratics were fit.
5 9
LITERATURE CITED
1) Jeansonne, M.S.; Foley, J.P. J. Chromatogr. 1989, 461, 149-163
2) Grushka, E. Anal. Chem. 1972, 44, 1733-1738
3) Giddings, J.C. "Dynamics in Chromatography", Part 1; Chromatogr. Science
Series, Vol. 1; Dekker, 1965
4) Sternberg, J.C. "Advances in Chromatography”; Giddings, J.C., Keller, R.A.,
Eds.; New York: Marcel Dekker: 1966; Vol. 2, pp. 205-270
5) Foley, J.P. Anal. Chem. 1987, 59, 1984-1987
6 ) Naish, P J; Hartwell, S. Chromatographia 1988, 26, 285-296
7) Foley, J.P.; Dorsey, J.G. J. Chromatogr. Sci. 1984, 22, 40-46
8 ) Jeansonne, M.S.; Foley, J.P. 1990, Submitted to / . Chromatogr. Sci.
9) Yau, W.W. Anal. Chem. 1977, 49, 395-398
10) Barber, W.E.; Carr, P.W. Anal. Chem.. 1981, 53, 1939-1942
11) Jung, K.H.; Yun, S.J.; Kang, S.H. Anal. Chem. 1984, 56, 457-462
12) Anderson, D.J.; Walters, R.R. J. Chromatogr. Sci. 1984, 22, 353-359
13) Foley, J.P.; Dorsey, J.G. Anal. Chem. 1983, 55, 730-737
14) Haddad, P.R.; Sekulic, S. J. Chromatogr. 1988, 459, 79-90
15) Wu, N.S.; Hu, M. Chromatographia 1989,28,415-416
Chapter 4
M EASUREM ENT OF STATISTICAL MOMENTS OF
RESOLVED AND OVERLAPPING CHROMATOGRAPHIC
PEA K S
Reprinted with permission from the Journal o f Chromatography, Vol. 461, pp. 240-253,
1989.
6 0
61
Department o f C hem is try
L o u i s i a n a S t a t e U n i v e r s i t y and agricultural and mlch apical colli-clBATON ROUGE • LOUISIANA • 70803-1804 , , , ,
Mrs. M. Verhaar Elsevier Science Publishers P.O. Box 330 1000 AH Amsterdam The Netherlands
Dear Mrs. Verhaar:
I am writing to you in reference to the article entitled," Measurement of S tatis t ical Moments of Resolved and Overlapping Chromatographic Peaks", published in the Journal o f Chromatography, Vol. 461, pp. 240-253, 1989, for which I am the f i rs t author. I would like to use the manuscript as part of my Ph.D. dissertation. The completed dissertation will be submitted to University Microfilms, Incorporated. Please forward permission for reprint of the manuscript.
I appreciate your prompt reply.
Permission granted subject to permission from the author(s) and to full acknowledgement of the source.Elsevier Scienoe Publishers Physical Sciences & Engineering Div.
504/388-3361 FAX 5041388-3458
September 13, 1990
Sincerely,
Mark S. Jeansonne
6 2
INTRO DUCTIO N
The importance of statistical moment analysis to the chromatographer cannot be
overemphasized because a large amount of information can be derived from such an
analysis. Statistical moment analysis can be used not only to measure directly parameters
such as area (zeroth moment), peak centroid (first statistical moment), and variance
(second statistical moment), but also to calculate other important parameters indirectly as
well. For example, column efficiency can be calculated from N=Mj^ / M2 , where N is
the column efficiency, M j is the first statistical moment, and M2 is the variance. Other
parameters, such as the third and fourth statistical moments give information on peak
asymmetry and peak flattening, respectively. Peak skew and excess are parameters
related to statistical moments and provide a measure of the deviation of the
chromatographic peak from a Gaussian peak profile.
Traditionally, statistical moments for digitally represented chromatographic peaks
have been approximated by the simple summation of the magnitude of the peak signal at
each data point between the peak start and stop limits, as shown in Fig. 4 - l a .
However, this approach to the calculation of statistical moments has several
shortcomings when applied to real chromatographic data.
First, it has been shown that the accuracy and precision of the calculation of
statistical moments is directly affected by the amount of noise present in the
chromatogram. The noise level has been shown to affect peak start / stop
assignments, and this affects the limits of summation and, consequently, the value of the
statistical moments calculated. 3
Secondly, the accuracy of the summation method (equations shown in Fig. 4-la)
deteriorates rapidly as the peaks begin to overlap. We have recently shown 4 that errors
6 3
in peak area can exceed 1 0 0 % when the summation approach (perpendicular drop
algorithm) is applied to overlapping peaks. As we shall show later in this report, errors
in the higher moments calculated via the summation method are usually much larger
under the same circumstances.
A final drawback of the summation method is that it is computationally intensive,
requiring numerous calculations (e.g. Fig. 4-la) for every data point in the peak of
interest This is particularly true for the higher moments and related parameters.
Although this problem has been alleviated somewhat by the advances in computer
technology (faster computations), the summation method remains noticeably time-
comsuming on many commercial chromatographs with microcomputer-based data
systems.
Most, if not all, of the problems associated with the measurement of statistical
moments can be reduced or eliminated if one has an accurate model for the
chromatographic peaks of interest. A model that has been reported to be accurate for
most chromatographic peaks 5-7 is the Exponentially Modified Gaussian (EMG)
function, which is the convolution of a Gaussian and an exponential decay function.
Recently, we introduced 8 a convenient procedure for determining whether or not the use
o f the EMG model is appropriate. This procedure utilizes empirical equations for
calculating peak area based on peak width, asymmetry and peak height measurements.
Once the validity of the EMG model for a given set of peaks has been confirmed, these
equations can also be used for the accurate measurement of peak areas of overlapping
chromatographic peaks. 4. Note that this method relies on the measurement of peak
width and asymmetry for the less distorted peak of the overlapped pair (the first peak) at
a point above the valley where distortion from the second peak is low.
6 4
Although some of the problems associated with the traditional measurement of
statistical moments can be reduced or eliminated via the use of a variety of sophisticated,
curve-fitting/deconvolution procedures, these procedures also have numerous
drawbacks. First, they are nearly always even more time-consuming than the traditional
summation approach. In many cases a final summation step is required after the
preliminary curve-fitting/deconvolution procedures. Secondly, some of the procedures
require multi-channel detection which is not always available. Thirdly and most
importantly, for a variety of reasons the curve-fitting/deconvolution approaches have not
yet proven to be sufficiently reliable. For example, with iterative procedures, lack of
convergence is frequently observed. In general, these and other disadvantages have
dissuaded most, if not all, commercial manufacturers from implementing the curve-
fitting/deconvolution approaches into their chromatographic data systems.
The purpose of this paper is to report an alternative to both the traditional and
least squares/deconvolution methods for the measurement of statistical moments. Our
present approach utilizes empirical equations (Fig. 4-lb) similar to those we already
reported for peak area 4, but also includes a veiy simple deconvolution procedure for a
pair of overlapping peaks. The derivation of these equations will not be included here,
as this topic will constitute a separate paper. 9 . For the remainder of this report, we will
refer to our method of statistical moment measurement as the width-asymmetry method.
EXPERIM ENTAL
Both an Apple Macintosh and IBM PC-AT were utilized for simulated peak
generation and other calculations. All programs were written in either Microsoft BASIC
or TRUE BASIC.
6 5
EMG Peak generation. All peaks generated were based on the EMG
function^ expressed as
hEMG(0 = A — V2- exp (2S ) 2 - < ^ ) J X X X
-y 2 exp ( ~ 2 ~ ^
V27Cdy [1]
where A is the peak area, tQ is the retention time, Gq is the standard deviation of the
Gaussian function, x is the time constant of the exponential decay function convoluted
with the Gaussian function, and Z = ( t-tQ ) / o q - g q / x . As the x / g q ratio increases,
the peak in question will become more skewed, and as it decreases, the peak approaches
a Gaussian shape.
Single chromatographic peaks at x / g q ratios of 0 ,0 .5 ,1 ,2 , 3, and 4, with g q a
constant at 0 .1 min, were generated for this study, using a sampling rate of three points
per second. As shown in Fig. 4-2 (x/ g q = 2), about 30 points per peak measured from
1 0 % peak height to 1 0 % peak height were needed for < 2 % error.
Overlapped chromatographic peaks at x / g q ratios of 0 .5 ,1 ,2 , 3, and 4 were
generated at resolution values of 0.625, 0.75, 0.875, 1, 1.125, 1.25, 1.375 and 1.5 by
using the same sampling rate as the single peaks. Resolution was defined as AtQ /
4(variance)l/2, where A t<3=tQ 2 - kj,l> anc* variance was defined as g q 2 + x% for an
exponentially modified Gaussian peak. The peaks were overlapped by adding two
individual, simulated peaks of equal area and x/ g q value. However, the degree of peak
overlap will be reported here as the percent valley due to the inadequacy of the resolution
parameter for fully describing tailed overlapped peaks. 4 Percent valley was defined as
hv / hp X 100%, where hv and hp are shown in Fig. 4-3.
6 6
Real Chrom atographic Peak Generation. Real single and overlapped pairs
of peaks were generated on a Series 400 liquid chromatograph (Perkin-Elmer, Norwalk,
CT, U.S.A.), using pyrene as the analyte. The mobile phase composition was 75%
aqueous acetonitrile at a flowrate of 1.5 ml/min. The column used was a Vydac pH-
stable C-8 column. A Model variable wavelength ultraviolet absoiption detector
(Isco, Lincoln, NB), set at 330 nm was used to detect pyrene. An Omega-2 data system
(Perkin-Elmer) utilizing an IBM AT computer, was used for storage of the
chromatograms.
Overlapped peaks were obtained from precise, rapid duplicate injections of a
standard solution of pyrene. This single-standard, rapid, duplicate-injection approach has
many advantages over a two-component standard solution method which would require
changing conditions to obtain different degrees of overlap. First, it allows the degree of
peak overlap to be easily controlled by simply varying the length of time between
injections. Second, this method avoids any relative change in the molar absorptivities of
two analytes in the mixture as mobile-phase conditions are changed to obtain different
degrees of overlap. Third, it permits a single peak to be obtained under the same
conditions as the overlapped peaks, thus allowing the statistical moments measured by
the summation and width-asymmetry methods for the isolated peak to be compared with
those measured for the overlapped peak pair without any concern about changes in the
peak shape and/or concentration. Finally, one can be confident that the true area ratio of
the overlapped peaks is unity, since equal amounts of the same compound are being
injected.
Two pairs of tailed, overlapping peaks with percent valleys of 40% and 67%
were generated. A single control peak with the same amount of peak tailing as the
6 7
overlapping peaks was also generated. The amount of peak tailing was adjusted by
adding or removing dead-volume ahead of the column.
Peak Param eter Measurement The equations given in Fig. 4-la, for
determining the zeroth through fourth statistical moments by summation, were applied to
both simulated and real peaks. For the simulated peaks, start/stop assignments (limits of
integration) for isolated peaks were taken as the point where the peak was determined to
be "on baseline", which depended on the " baseline level" being used for the peak.
Baseline levels of l x l 0 ' 9 and 3x10"^ were used, which corresponded to approximately
0.00 % and 0.1% of the peak height, respectively. The lower baseline was used as an
ideal baseline in order to obtains maximum level of accuracy for purposes of
comparison. The ideal baseline was chosen to be slightly above zero, since the value of Z
in Eq. 1 necessary to give a zero baseline would result in an overflow condition. This
ideal baseline was used only for peaks in which no noise was present, since noise affects
peak/start assignments in real chromatograms.
For overlapped peak pairs, the starting point for the first peak and stopping point
for the second peak were chosen as for isolated peaks. The peak stop for the first peak
and the peak start for the second peak were taken as the intersection of the baseline being
used and a perpendicular line, drawn to the minimum of the valley between the peaks
(see Fig. 4-3 ). This method for dealing with overlapped peaks is commonly referred to
as the perpendicular drop algorithm.
For the real chromatographic peaks, the baseline level and the start/stop
assignments were determined by the data system, the perpendicular drop method being
employed for overlapped peaks. The peak detection algorithms in the data system were
optimized for the types of real peaks that were generated.
68
The widths of the peaks at 10%, 25%, 50%, and 75% relative peak height were
determined by utilizing a four point least squares fit where four points on each side of the
peak, symmetric about the particular height, were used and the difference in time
between the two points ( ta - t^w as taken as the peak width (see Fig. 4-3). Four points
were used, since the accuracy of the value obtained for peak width did not increase when
more points were fit.
Peak height was obtained by subtracting the baseline value being used from the
peak maximum obtained via a quadratic least-squares curve fit of the seven highest points
in the peak. The seven-point group was selected so that the middle point had the highest
value. The time at which the maximum was calculated from the quadratic fit was used as
the retention time of the peak. Seven points were used, since this number represented a
compromise between the optimum number of points for a peak with a t/sQ ratio of 1
(mildly skewed) and a peak with a t /o q ratio of 4 (heavily skewed) for the data sampling
rate used. This compromise was selected so that the quadratic fit could be used for peaks
for which the value of t/Oq was not known, as in real chromatograms. The asymmetry
of the peak was taken as the value of b/a, where a and b were determined as shown in
Fig. 4-3 at the appropriate peak heights.
RESULTS AND DISCUSSIO N
Simulated peaks without noise. The values for the area and the variance
obtained by the summation and width-asymmetry methods for simulated peaks and an
ideal baseline are compared in Fig. 4-4 as a function of the percent valley between the
peaks. For only slightly tailed peaks (x / g q = 1, Fig. 4-4 a), it appears that the
summation method for peak area and variance is fairly accurate for the noiseless peaks
69
used in this study. However, for moderately tailed peaks (x/ g q = 2, Fig. 4-4 b), the
area of the second peak and the variance of the first peak have become much less accurate
relative to the same parameters measured by using the width-asymmetry method. This
trend continues in Fig. 4-4 c for x / g q = 4. Thus, the width-asymmetry method can be
used to measure accurately both the area and variance for the left peak of a highly skewed
and overlapped pair of peaks, while the same parameters cannot be measured as
accurately for either peak when the summation method is used.
Fig. 4-5 illustrates the same results for peak area and variance as Fig. 4-4 c, but
with a less ideal baseline. Here, the variance for the first peak, as measured by the
summation method, is very inaccurate. This is due to a significant portion of the tail of
the first peak being truncated by the higher baseline, and shows the sensitivity of the
variance to baseline eirors when measured by the summation method. However, the
variance measured by our width-asymmetry method does not show this sensitivity.
The errors in the higher moments, skew, and excess are compared in Fig. 4-6 for
an overlapped pair of highly skewed peaks ( x/ g q = 4) and an ideal baseline. Since an
ideal baseline was simulated, the moments measured for the right peak by the summation
method (see Fig. 4-6 a) show a fair accuracy up to a high percent valley for these
noiseless peaks. However, results this accurate cannot be expected for real
chromatographic peaks, due to the problems outlined in the introduction for the
summation method. As expected, the higher moments, skew, and excess, measured by
the summation method for the left peak, show greater sensitivity to the truncation of the
peak tail than do the area and variance for the left peak, measured by the summation
method.
However, the higher moments, skew, and excess measured by the width-
asymmetry method for the left peak do not show this sensitivity (see Fig. 4-6 b). In fact,
7 0
the accuracy for these parameters, measured by the width-asymmetry method, is much
better than that obtained by the summation method for the right peak. This shows that the
width-asymmetry method is tetter overall for measuring the higher statistical moments,
excess, and skew for at least one peak of an overlapped pair of peaks.
Table 4-1 shows the maximum percent valleys (maximum overlap) for which the
two methods described here are in error by less than 5%. As seen there, for most of the
moments calculated by the width-asymmetry method for the left peak, the maximum
overlap that can be tolerated is higher. However, for the right peak, some of the
moments calculated by the summation method are more accurate. These results were
expected, since the tail for the right peak is fully included in the limits of integration.
Peak overlap of the peaks prevents the tail of the left peak from being included in the
summation method. This confirms the well-known results that the tail of a skewed peak is
especially important in calculating the higher moments by the summation method. Also,
the results show that the degree of distortion in the left peak is low for two overlapped
EMG peaks. Overall, these results indicate that all the statistical moments, including
excess and skew can be measured accurately for peaks that are moderately overlapped,
the more skewed peaks giving the test results. This latter trend is due primarily to the
percent valley parameter, which tends to underestimate peak overlap for symmetrical
peaks and to overestimate peak overlap for skewed peaks. (However, this measure of
peak overlap is no worse than any other parameter, and is more practical than most other
measures of peak overlap for skewed peaks 4.) Of course, Fig. 4-6 also show that for
those overlapped peaks which are not highly skewed the summation method may
occasionally give tetter results.
Although the summation method appears to be fairly accurate for the higher
moments of the second peak of a moderately to highly skewed overlapping peak pair, it
71
will generally be very inaccurate with real chromatographic data (peaks with noise). In
modem chromatographic integrators and data systems, much of the tail of even a mildly
skewed peak is often not included in the summation, due to baseline errors occurring
when the algorithm used detects a peak stop before the actual end of the peak is reached.
Many data systems rely on the first derivative, second derivative, or similar tests to detect
peak end with a slope sensitivity setting which depends on the degree of noise in the
chromatogram. ^ The slope sensitivity is set at a level higher than what might be
expected for the baseline drift. However, this setting may frequently also be higher than
the slope on the tail of a skewed chromatographic peak. On overlapped pairs of peaks,
the premature peak end would affect the second peak almost exclusively, therefore
disallowing the use of the second peak in accurate computation by the summation method
of the higher statistical moments for that peak, and often of the area and the second
moment as well. In contrast, the width-asymmetry method is relatively unaffected by this
type of truncation error. 11 Furthermore, in this example, the width-asymmetry method,
which is applied to the first peak of an overlapped pair, would be entirely unaffected by
the premature peak stop on the trailing edge of the second peak.
Results for real chrom atographic peaks. Table 4-2 shows the results for
a real isolated chromatographic peak, obtained under ideal conditions (high signal-to-
noise (S/N) ratios, no overlap, baseline resolution, etc.). As seen in Table 4-2, the zeroth
through fourth statistical moments, along with peak excess and skew for the single peak,
were found to be similar for the width-asymmetry and summation methods. Under less
ideal conditions, with a much smaller S/N ratio, the summation method would probably
give results very different from the width-asymmetry method, due to the limitations of
the summation method mentioned in the introduction.
7 2
The appropriateness of the EMG model for this real chromatographic peak is
demonstrated by the agreement obtained for the various peak parameters at different
relative peak heights. Although the agreement is not exact (there is a slight spread), the
spread in peak parameters is small compared to the error encountered when using the
summation method on most real peaks, again due to the problems outlined in the
Introduction.
The advantage of the width-asymmetry method over the summation method for
real peaks becomes apparent when overlapping peaks are examined. Table 4-3 gives the
results for the summation and width-asymmetry methods for two pairs of peaks that
overlap by different amounts. As seen for the 40% valley case, the summation method
gave relative areas of 39.5% for the left peak and 60.5% for the right peak peak. Since
the true relative areas of the peaks are 50%, the error in peak area for each peak is 10.5%
when using the summation method and perpendicular drop algorithm. In contrast, the
width-asymmetry method gave relative areas of 50.0% for each peak, i.e., exactly the
correct result.
For the more heavily overlapped peak pair (67% valley), the errors associated
with the summation method increased, whereas the width-asymmetry method again gave
results very close to the correct result (areas measured for the single peak shown in Table
4-2). The relative areas for the left and right peaks, determined by the summation
method, were in error by 17% for each peak, whereas the relative areas measured by the
width-asymmetry method were in error by only 1%. The other statistical moments,
including skew and excess for the left peak, show a large difference between those
calculated by the width-asymmetry and summation methods. Also, a comparison of these
parameters measured for the left peak by using the width-asymmetry method to those for
7 3
the single peak (see Table 4-2) shows that the width-asymmetry method gave very good
results.
Modified width-asym m etry method for true peak deconvolution. In
order to determine relative areas, the present width-asymmetry method requires
knowledge of the total area of the two overlapping peaks, as measured by the
summation method. The total area cannot be calculated by the width-asymmetry method,
because the distortion of the right peak by the tail of the left peak causes an erroneous
contribution to the total area. Thus, the width-asymmetry method, as presently
employed, is not a true peak deconvolution method, since parameters other than the peak
area are not additive and therefore cannot be determined for the second peak by
subtraction. However, the width-asymmetry method can be used to deconvolve an
overlapping pair of peaks as follows: First, the first peak of the overlapping pair is
calculated point by point via Eqn. 1 over an appropriate time interval from values of o q ,
x, and A that are estimated from peak width, asymmetry, and peak-height (area only)
measurements by using equations described elsewhere. 5 Next, tQ is estimated from Mi
- x and then adjusted so that the maxima of the calculated peak coincides exactly with the
maxima of the first peak in the overlapping pair of real peaks. Finally, the lone second
peak of the overlapped pair is obtained by subtracting, point by point, the values of the
calculated peak from the total chromatographic signal.
The accuracy of the width-asymmetry deconvolution method is evident from
Table 4-4 for real chromatographic peaks. As Tables 4-3 and 4-4 show, both the original
width-asymmetry method and the modified width-asymmetry/deconvolution method give
accurate results for all parameters for the first peak of the overlapped pair. Note,
however, that the original width-asymmetry method uses values of peak height, width,
and asymmetry obtained directly from the actual chromatogram (at 15% relative peak
7 4
height), whereas the width-asymmetry/deconvolution method uses the same values from
an artificially constructed peak.
Fig. 4-7 illustrates the results of the width-asymmetry deconvolution method,
applied to the chromatogram of overlapped peaks with the 40% valley (relative to the left
peak); superimposed on the real chromatogram are the two simulated peaks, obtained
from the width-asymmetry deconvolution method. As seen, the first simulated peak
falls directly on the actual first peak of the overlapped pair. This was expected, since
there is little distortion of the first peak in the overlapped pair from the second peak. The
distortion of the second peak in the overlapped pair, caused by the tail of the first peak, is
readily apparent, though, as the difference in both height and area of the second peak and
its corresponding simulated peak is large. This distortion is the reason why the
petpendicular drop method underestimates the area of the first peak and overestimates the
area of the second peak in an overlapped pair of peaks. It is also why, as mentioned
above and shown here, the width-asymmetry method cannot be applied directly to the
second peak of an overlapped pair of peaks.
Computational Time. As stated in the Introduction, one problem with the
summation method is that eveiy point in chromatographic peak must be involved in
moment calculations. However, most of the points in a peak of interest do not have any
calculations performed on them when the width-asymmetry method is used. When the
two methods were timed against each other, the width-asymmetiy method was found to
be about twice as fast for single-peak chromatograms and up to ten times faster for
multiple-peak chromatograms.
7 5
ACKNOW LEDGEM ENT
We gratefully acknowledge the support of this work by the Computer-Aided
Chemistry Department of the Perkin-Elmer Corporation and the National Oceanographic
and Atmospheric Administration (NOAA). Mark S. Jeansonne also thanks Jeffrey Crow
for his helpful discussion on this topic.
CREDITS
This work was presented in part at the 1988 Pittsburgh Conference and Exposition in
New Orleans, LA , February 1988, paper # 088.
Table 4-1
Maximum peak overlap (percent valleys) that can be tolerated by the width-asymmetry and summation methods for a given accuracy (<5% error)
PEAK AREA m2 m 3 m4 SKEW EXCESS
Width-asymmetry First 47 34 16 24 17 15
Method Second 4 17 8 8 8 7
Summation method/ First 9 < 2 < 2 < 2 < 2 < 2
ideal baseline* Second 8 34 5 11 5 2
Summation method/ First 8 < 2 < 2 < 2 < 2 < 2
non-ideal baseline** Second 9 24 9 < 2 9 < 2
* Peak start/stop corresponds to points where signal is 5 x 10"8 % of maximum.
** Peak start/stop corresponds to points where signal is 0.1% of maximum.
TABLE 4-2
Peak parameters measured for an isolated, real chromatographic peak (pyrene).
Height(%)
Peakarea(mV • s)
Mj M2 m3 M4 SKEW EXCESS
Summationmethod
NotApplicable
5449.6 1.718 0.0424 0.0186 0.0216 2.146 8.974
Width- 10 5370.0 1.722 0.0410 0.0149 0.0138 1.794 5.191asymmetry 25 5530.7 0.0440 0.0167 0.0159 1.807 5.242method 50 5670.8 1.892 0.0468 0.0184 0.0181 1.816 5.278
75 5802.7 0.0500 0.0205 0.0208 1.827 5.321
TABLE 4-3
COMPARISON OF SUMMATION AND WIDTH-ASYMMETRY METHODS FOR TWO SETS OF OVERLAPPING, REAL PEAKS
Overlapping peaks generated from rapid, duplicate injections of pyrene (see Experimental section). Therefore, values for all the parameters except Mj should be the same for all peaks within experimental error.
Summation method40% Valley*
Left Peak Right Peak
67% Valley*Left peak Right peak
Width-asymmetry method 40% Valley*
Left peak
Right peak
67% Valley* Left peak
* *
Right peak* *
Height (%) Peak area (mV«s)
Mj m2 m3 m4 SKEW EXCESS
Not applicable Not applicable
4481.46 8 6 6 .8
1.63292 .1 0 1 0
0.00930.0501
3.7X10' 50.0319
1.9X10-40.0469
-0.7952.839
0.04215.669
Not applicable Not applicable
3733.67347.5
1.60901.9974
0.00610.0416
-6.13X10-50.0175
5.1X10-50.0179
-1.6262.067
-0.1307.340
50%75%50%75%
5672.75676.86731.06435.0
0.04050.04190.03710.0354
0.01400.01500.01180.0113
0.01300.01410.01050.0098
1.7201.7441.6541.690
4.9075.0004.6564.794
1.9902
50%
75%50%75%
5694.27470.2 6594.9
0.04540.02180.0258
0.01700.00240.0062
0.01660 .0 0 2 0
0.0047
1.7610.7461.483
5.0631.6124.025
1.8933
TABLE 4-3 cont'd
* Valley height relative to left (first) peak.** Areas calculated as A r = (A l + AR)summation * other parameters calculated using the width-asymmetry
equations in Fig. 4-lb. Width-asymmetry equations should normally not be employed for the more distorted peak of an overlapping pair of peaks. Results are shown here for purposes of comparison only.
TABLE 4-4. PEAK PARAMETERS FOR DECONVOLVED PEAKS OF OVERLAPPING PEAK PAIRS WITH 40% AND 67% VALLEYS a
40% VALLEY** AREA0 M i m 2 m 3 M4 SKEW EXCESS
SIMULATED LEFT PEAK
5654.4 1.798 0.0415 0.0147 0.0138 1.738 4.976
SIMULATED RIGHT PEAK
5598.1 2.197 0.0411 0.0150 0.0139 1.804 5.229
67% VALLEY**
SIMULATED LEFT PEAK
5676.4 1.827 0.0450 0.0168 0.0164 1.755 5.042
SIMULATED RIGHT PEAK
5272.2 2.090 0.0356 0 .0 1 2 0 0.0104 1.780 5.138
a Values for each parameter were calculated from width-asymmetry equations atlO, 25,50 and 75% relative peak height The values obtained were then averaged for this table. The spread of values for each parameter never exceeded 5% for the left peak and 10% for the right peak.
b Valley height relative to left peak.
c Peak area given in units of mV-sec.
ooo
81
(a)oo Stop
Mq «= AREA «= Jh(t) dt - X h ( t ) At•oo start
oo StopM i= Jt*h(t) dt / Area - X l*h(t)At + Area
•oo start
oo StopM2 = J (t-Mi )2 »h(t) dt / Area - £ (t-M-j )2*h(t) At + Area
-oo start
oo stopM3 = f (t-M-j )3 »h(t) dt / Area * X (t_M1 )3 ®h(t) At + Area
start
00 stopM4 = f (t*Mi )4 *h(t) dt / Area « X (t_M1 )4*h(l) At + Area
-ed start
YS = SKEW - M3/M23/2 YE = EXCESS * M4/M22 - 3
(b)
CG - f f lg s j M2 = W 2 x f2(b/a)
M i = tR + W x f 3 (b/a) T = ' V m 2 - 0 G 2
M3 = 2 x3 M4 = 3 GG4 + 6 OG2 x2 + 9 T4
Ys = M3/M2 3/ 2 Y e = M4 /M2 2 - 3
Figure 4-1. Equations used for calculation of statistical moments and other peak
parameters by a) the summation method and by b) the width-asymmetry
method.
OBS
/ TR
UE
1.02i
1.0 0 -WIDTH
0.98-
AREA0.96-
0.94-ASYMMETR
0.920.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4 .0
LOG( # PTS,10%)
Figure 4-2. Effect of the data sampling rate on the measurement of peak width,
asymmetry, and peak area.
8 3
1 a
P,2
time
Figure 4-3. Measurement of graphic parameters for an overlapping pair of
chromatographic peaks. tr and hp are the retention time and peak height of
respective peaks, and hv is the height of the valley. Peak width at die
desired peak height fraction is given by ta-tfc.
8 4
Figure 4-4. Comparison of the errors in peak area and variance occurring in the
summation and width-asymmetry methods as a function of peak overlap
(percent valley) for: (a) x / g q = 1; (b) x /g q = 2; and (c) x /o q = 4. Labels
in the plot refer to: a) the parameter, b) the peak (first [L] or second [R]) for
which a parameter was obtained; and c) the relative peak height at which the
width and asymmetry were measured (width-asymmetry only). For
example, "M2, L; 75%" refers to the variance measured for the first peak of
the overlapped pair at 75% of the peak height, while "M2, R" refers to the
variance of the second peak measured by the summation method.
|OB
S|/T
RU
E
85
2 . 0 1
M2, L; 75%
1.5AREA, L; 75%
AREA, R: AREA, R; 75%
M2,R1.0
AREA,L
M2, R; 75%
M2,L0.5
0.00 20 40 60 80 100
% VALLEY
|OB
S| /
TRU
E
8 6
2.01
1 .5 - (b) M2, L; 75%
AREA, R; 75%AREA, R
AREA, L; 75%
1 .0 - M2, R
AREA, L
M2, R; 75%0 .5 -
M2, L
0.010060 804020
% VALLEY
anui/lsaol
87
2.0 n
AREA, R
1.5AREA, R;75%
M2, R : M2, L;75% : AREA, L;75%1.0
AREA, L0.5 M2, R; 75%
M2, L
0.00 20 40 60 80 100
% VALLEY
|OB
S| /
TRU
E
88
2.0
1.5 AREA, RAREA, R; 75%
M2, L; 75%: AREA, L; 75%1.0
M2, R
AREA, L
M2, R; 75%0.5
M2, L
0.00 20 40 60 80 100
% VALLEY
Figure 4-5. Comparison of the errors in peak area and variance occurring in the
summation and width-asymmetry methods as a function of peak overlap for
a highly skewed pair of peaks (x/ g q = 4) with a less than ideal baseline
level (0.1%). Conditions as in Fig. 4.
89
Figure 4-6. Comparison of the errors in peak parameters other than peak area and
variance for overlapped peak pairs with x/ g q = 4 occurring in: a)
summation method; and b) width-asymmetry method. Conditions as in
Fig. 4-4.
|OB
S|/T
RU
E
9 0
2 .0 1
1.5 -
M3, R ; M4, R ; SKEW, R ; EXCESS, R
1.0 -
M3, L ; M4, L0 .5 -
SKEW , L
EXCESS, L
0.00 20 40 60 80 100
% VALLEY
|OBS
| /
TRU
E
91
2.0 1
(b)
M3, L; 75% : M4, L; 75% : SKEW, L; 75% : EXCESS, L; 75%
SKEW, R; 75%
EXCESS, R; 75%0 .5 -
M3, R; 75%
M4, R; 75%
0.00 20 40 60 80 100
% VALLEY
SIG
NA
L (m
V)
9 2
400
REAL CHROMATOGRAM
SIMULATED LEFT PEAK250
SIMULATED RIGHT PEAK
100
-50
0 1 2 3TIME (MIN)
Figure 4-7. Visual interpretation of the width-asymmetry/deconvolution method. The
solid black line indicates the real overlapping chromatographic peaks, while
the lighter lines show the individual peaks that are predicted by the width-
asymmetry/deconvolution method.
9 3
REFERENCES
1. Chesler, S. N.; Cram, S. P Anal. Chem. 1971, 43, 1922 - 1933
2. Petitclerc, T . ; Guiochon, G. J. Chromatogr. Sci. 1976, 14, 531-535
3. Grushka, E.; Myers, M.N.; Schettler, P. D. Anal. Chem. 1969, 41, 889 - 892
4. Foley, J.P. J. Chromatogr. 1987, 384, 301 - 303
5. Foley, J.P.; Dorsey, J.G. J. Chromatogr. Sci. 1984, 22, 40 - 46
6 . Pauls , R.E.; Rogers, L. B. Anal. Chem. 1977, 49, 625 - 628
7. Grushka, E. Anal. Chem. 1972, 44 , 1733 - 1738
8 . Foley, J.P. Anal. Chem. 1987, 59, 1984 - 1987
9. Jeansonne, M.S.; Foley, J.P. This Dissertation, Chapter 3.
10. Colmsjo, A. L. Chromatographia 1987,23,257 - 260
11. Anderson, D.P.; Walters, R.R. / . Chromatogr. Sci. 1984,22,353-359
Chapter 5
THE Q TRANSFORM ATION: A NOVEL METHOD OF PEAK SH APE ANALYSIS.
Reprinted with permission from Burgess, S.W., Jeansonne, M.S. and Foley, J.P.;
submitted for publication in Anal. Chem. Unpublished work Copyright 1990 American
Chemical Society.
9 4
9 5
Department o f Chemistr\/
L o u i s i a n a S t a t e u n i v e r s i t y and agricultural and mechanical colleceBATON ROUGE • LOUISIANA • 70803-1804
Barbara Polanski Publications Division American Chemical Society 1 155 16th St. Northwest Washington, DC 20036
Dear Ms. Polanski,
I am writing to you in reference to the article en t i t led ," The Q Transformation: A Novel Method of Peak Shape Analysis", AC9006886, received by A nalytical Chem istry on August 24, 1990, for which I am the f i rs t author. I would like to use the manuscript as part of my Ph. D. dissertation. The completed dissertation will be submitted to University Microfilms, Incorporated. Please forward permission to reproduce the manuscript.
I appreciate your prompt reply.
5041388-3361 FAX 504/388-3458
September 13, 1990
Sincerely,
Mark S. Jeansonne
American Chemical Society 9
PUBLICATIONS DIVISION 1155 SIXTEENTH STREET, N.W. W ASHINGTON, D.C. 20036 P h o n e (202) 872-4600 Fax (202) 872-6060
Septem ber 28, 1990
Mr. M ark S. Jeansonne D epartm ent o f Chem istry Louisiana S ta te University B aton Rouge, Louisiana 70803-1804
D ear Mr. Jeansonne:
Thank you for your le tter o f Septem ber 13, in which you requested permission to include in your thesis, two papers that you subm itted to Analytical Chemistry :"SingIe C hannel Peak Im purity D etection Using the Q Transform ation," and "The Q Transform ation: A Novel M ethod o f Peak Shape Analysis."
I would be happy to grant you this permission free o f charge provided that you print the required ACS copyright credit line on the first page o f your article and that your paper is published by ACS first before you submit it to UMI. Also, please inform UM I that permission is granted subject to the terms o f the ACS/UMI Agreem ent dated O ctober 1, 1984.
T he credit line we require is, "Reprinted with permission from FU LL R E FE R E N C E CITA TIO N Copyright Y EA R American Chemical Society." Please be sure to insert the appropria te inform ation in place o f the words in capital letters.
If your paper is no t accepted before you include it in your thesis, please use the following credit line: "Reprinted with permission from Burgess, S.W., (A UTHO RS NAM ES); subm itted for publication in Anal. Chem. U npublished work Copyright 1990 Am erican Chemical Society."
Thank you for your request. Please let me know if I can be o f any further help.
Sincerely,
Barbara F. Polansky Copyright A dm inistrator
9 7
INTRO DUCTIO N
In many forms of analytical chemistry the analyst must rely on a detection system
that gives a peak shaped signal as output. Most often the peak's height or area is
measured to calculate properties of the chemical or physical system being measured.
Sometimes, however, the shape of the peak holds special importance in supplying added
clues to the behavior of a particular system. For example, in chromatography the analyst
is usually interested in height, area, retention time and resolution between peaks to
characterize a sample. While the measurement of first three parameters may depend very
little on the shape of the signal peak, the accurate calculation of resolution or overlap
between adjacent peaks certainly does. Individual peak shapes in chromatography can
also give insight into many intra-column and extra-column effects that may or may not be
desirable. That is why much research has been done on peak shapes and peak shape
modelling in chromatography. In many other areas of analytical chemistry such as
nuclear magnetic resonance spectroscopy and infrared spectroscopy, chemists may also
be interested in peak shapes (1). In Mossbauer spectroscopy, for example, peak shapes
are especially important (2 ).
In this paper we wish to introduce a new method by which individual peak
shaped signals may be analyzed and/or compared. The new approach that we have
developed has become known by us as the Q transformation. In technical terms, it can
be thought of as a normalized, horizontal differentiation of a peak, in distinct contrast to
conventional "vertical" differentiation, i.e., differentiation of the signal with respect to
the abscissa.
Other peak shape analysis methods of similar computational complexity have
been introduced previously, including slope analysis (3,4), statistical moment analysis
(5), and the distribution function method (DFM) (6 -8 ). These methods have notable
98
deficiencies, however. Both the second derivative and moment methods are extremely
sensitive to noise and also, in the case of moments, baseline errors, thereby limiting their
usefulness when applied to real data (9-13). Finally, although the DFM handles noise
well, we will later show that it is not as good as the Q transformation at detecting small
changes in peak shape.
Definition of Q transformation. The Q transformation is defined and
illustrated in eq 1 and Figure 5-1,
where r is the peak height fraction (in percent) at which the measurement of Q is initiated,
b and a are graphical parameters defined in Fig.5-la, and Ab is calculated as shown in eq
2 with Aa measured similarly. By definition, the value of Q at any value of r will be less
than, equal to, or greater than unity for a fronted, symmetric, or tailed peak, respectively.
Illustrated in Figure 5-lb is the resulting Q profile (plot of Q vs r) for the peak of
Figure 5-la. As shown, the Q profile decreases monotonically toward a value of 1 as r
increases from 2 to 98%, consistent with the fact that all continuous, asymmetric peak
profiles become more symmetric as their apex is approached. Since the Q profile can
give visual indication of peak shape as a function of r, it can be used to visually
determine relative peak shape independent of time scale or sample concentration.
THEORY
[1]
[Ab]r = bj- - bj-_i [2]
9 9
As illustrated in Figure 5-2, it is very difficult to determine visually which peaks
are the same shape and which are fundamentally different. By using the Q transform,
however, it is easy to determine that peaks 1 and 2 are the same.
An important parameter, XQ, is simply the summation of Q over all r for a given
peak where r can run from 2 to 98%. As we will show, the XQ can give valuable,
compact peak shape information by compressing the Q profile data (Q vs r) into a single
number.
Properties of the Q transformation. In general, we expected the Q
transformation to be independent of both the time and concentration scales while being
very sensitive to peak shape. The results shown in Figure 5-3 for a wide range of
fundamentally different types of asymmetric peaks confirm these expectations. Shown
in Figure 5-3a are the results of applying eq 1 to single exponentially modified Gaussian
(EMG) peaks with X/(J between 0.5 and 4. The value of Q at any fraction of the peak
height depends only on the ratio X/O and not on the individual values of these width-
related parameters. Figure 5-3b shows the results for Gamma type peaks (eq 4) with a
ranging from 4 to 10. The value of Q here does not depend on the width-related
parameter (3 — only on a. The results for log-normal type peaks (Figure 5-3c, eq 5)
show that the value of Q at any height depends only on the asymmetry and not on the
value of width used in eq 5. An overlay of Figures 5-3a, 5-3b, and 5-3c, shown in
Figure 5-3d, illustrates that the shape of the Q profile also depends on the particular peak
family or class to which the Q transform is applied. These properties indicate that the Q
transformation can be used for both inter- and intra-family peak shape comparison.
A minor limitation of the Q transformation (eq 1) is its inability to distinguish
between different types of symmetric peaks (e.g., Gaussian vs Lorentzian). By
definition, any symmetric peak shape will give a value of 1 for Q at every value of r.
Note, however, that two or more perfectly symmetric peaks are seldom, if ever,
100
observed in chromatography or spectroscopy, even though many experimental peaks
may appear to be symmetric to the naked eye. Moreover the Q transform can distinguish
between a perfectly symmetric peak and a slighdy asymmetric peak. This situation is
much more probable when comparing the shapes of two peaks that appear to be
symmetric.
If the unlikely problem of comparing two symmetric peaks ever did arise, one
could simply redefine Q as Ab/Abref, and perform the comparison as shown in Figure 5-
4, where the Gaussian peak with a = 1.2 was arbitrarily used in the numerator, and the
Gaussian peak with ct= 1.0 was arbitrarily employed in the denominator (reference
peak). In this case the shapes of the two peaks are the same when one ignores scale, but
by defining Q differently, we are able to distinguish between similar symmetric peaks of
different widths. If two peaks were symmetric but of different shape (Gaussian and
Lorentzian), they could also be distinguished using this modified definition of Q. In
general, we believe such a modification of Q is unnecessary since both types of
symmetric peak shape comparisons (different width or different shape) can be performed
by simple (visual) inspection. The modification of Q is also undesirable because of the
reduction in sensitivity to peak shape differences and the lack of invariance to scale that
results. Therefore, the remainder of this paper focuses on the definition of Q given in
Eq. 1 and the comparison of peak shapes for more realistic scenarios (slightly
asymmetric to very asymmetric peaks).
Peak shape comparison using the Q transform.When comparing the
shapes of peaks, real or simulated, there can be two approaches when using the Q
transformation. Assuming that the individual peaks being compared belong to the same
family of peak shapes, the simpler approach is to compare values for XQ, the area under
the Q profile curve (Q vs. r), for the different peaks. As illustrated in Figure 5-3a, any
101
set of peaks belonging to the same class or family will have different values for XQ if
their shapes are different
Alternatively, a more sophisticated approach is to compare the values ofQ at each
r, either visually or statistically, using tests such as the paired t-test or the Wilcoxon
signed rank test (14). Such as approach would be necessary when two peaks have the
same value of XQ (within experimental error). Figure 5-5 illustrates this point for an
EMG peak and a log-normal peak. Despite their identical values for XQ (not shown),
the non-superimposability of the two Q vs r curves clearly illustrates that these peak
shapes are fundamentally different. The statistics tests mentioned above are helpful in
detecting these differences when the O profiles are "noisy" (e.g., Figure 5-9), although
they were not necessary for our comparison of the Q approach with slope analysis,
moment analysis, and the distribution function method.
EXPERIM ENTAL
An Apple Macintosh Plus personal computer was used for simulated peak
generation and other calculations. All programs were written in either Microsoft Basic or
True Basic and were copyrighted after debugging. Single, noiseless exponentially
modified Gaussian (EMG) peaks were generated with tau/sigma (t/a) ratios between 1
and 4 as described earlier (13). As x/a increases, an EMG peak becomes more tailed; as
x/a decreases, the EMG peak approaches a Gaussian shape. Eleven single EMG peaks
with x/a = 1 were generated at each signal-to-noise (S/N) ratio of 1000, 500, and 100
where S/N was defined as the signal at peak maximum (peak height) divided by the root
mean square (RMS) of the noise. We assumed that the major noise component was white
noise, and that the distribution of the amplitudes was normal. Since the peak height was
known for noiseless EMG peaks, the required RMS noise could then be calculated.
From the RMS noise calculated, a random noise factor (random number of appropriate
102
amplitude) was calculated and added to the signal as shown in eq 3, such that the RMS
of many peak data points would be the RMS needed for the correct S/N.
EMG(t) with noise = EMG(t) ± Noise factor [3]
The S/N was then checked by calculating the RMS from a large number of peak baseline
points after the noise factor had been added.
Single noiseless Gamma type peaks were generated using the following equation,
(15)
Signal(t)= t(a_1) expf-I p a ( a - l ) !
[4]
For the present study, a was varied between 3 and 10 in increments of 2 while P was
held constant at 0.2. The actual value of p did not influence the values measured for Q.
Single log-normal (16) peaks were generated using
Signal(t) = hp exp In rln2Ac [5]
JJ
where hp is the maximum height of the peak, r is from Figure 5-1, t is time, ^ is the
retention time of the peak, and W and As are the width and asymmetry of the peak
measured at r. For this study, the value of r was 0.5; the peak height was 5, the width
was 1 minute and the asymmetry was varied between 1.05 and 1.6. The value of the
width in eq 5 has no effect on the value of Q.
1 0 3
Peak param eter measurement. Values of Q for noiseless, single peaks
were measured to 5 decimal point accuracy using a search algorithm. For simulated
peaks that contained noise (EMG only), start/stop assignments (limits of integration)
were taken as the point where the signal of the peak (before noise was added) crossed the
baseline threshold level, defined as 0 .0 0 1 % of the peak height.
Peak parameters a and b were measured for the digitized peaks for r from 2 to
98% in increments of 1% of the peak height. A 4 to 10 point least-squares fit, symmetric
about the particular height, was used on each side of the peak to find the time
corresponding to the exact peak height fraction. The parameters a and b were then
calculated as shown in Figure 5-1. The number of points used in the fit was the nearest
even integer of
Nfront / 14, where Nfront is the number of points on the front of the peak (from 1% peak
height to the peak maximum). Thus, the same range of peak height was fit for all peaks,
despite any differences in Nfront. We used 1% increments for Ar, although it could have
been larger or smaller. Larger values for Ar would affect the value obtained for XQ,
depending on Ar, since XQ is simply a summation. However, smaller values of Ar
would probably have a smaller impact on the calculated value of XQ because one would
simply obtain a larger number of a and b measurements using the same line fits as
described above. Therefore, the value of Ar should only impact the XQ value as would
the increment in any summation under a curve. Of course, a more accurate method of
measuring the time corresponding to a particular height should give an even more
accurate measurement of Q at any r, but the computational complexity may be increased
also. Peak height was measured by subtracting the baseline value being used from the
peak maximum obtained via a quadratic least-squares curve fit of the seven highest points
in the peak. For peaks with finite signal to noise ratio (noise added as in eq 3), a
Savitsky-Golay polynomial smooth (17-19) was applied to the raw data prior to the
1 0 4
measurement of a and b. A smoothing factor defined in eq 6 was used to determine the
number of points employed in the smooth.
c _ # pts in Savitsky-Golay Smooth , , ,smoothing factor = # p‘, s j„ top half of pH it ~ [6)
We found that a smoothing factor of 0.2 offered a good compromise between S/N
enhancement and the peak distortion that is caused by polynomial smoothing (2 0 ).
The statistical moments, skew, and excess of the noiseless EMG peaks were
calculated from fundamental relationships (21) while the same parameters for EMG
peaks with noise were calculated after a Savitsky-Golay smooth (smoothing factor = 0.2)
using the equations in Figure la of ref. 13, except that MJ-M4 were normalized by peak
area.
The second derivative of all EMG peaks also was calculated by using the method
of Savitsky and Golay (17-19). This was done without prior smoothing because the
Savitsky-Golay method smooths the data and calculates the second derivative in a single
operation. The same number of points was used to calculate the Savitsky-Golay second
derivative as the pure smoothing performed prior to the calculation of other parameters
(Q, etc.) so that no differences in peak shape would be introduced among the different
methods of peak shape analysis. The second derivative of most peaks has two maxima
and one minimum. Grushka et al. (4) used the ratio of each maximum to the minimum
and plotted these values versus each other for peak shape comparison. Because we
wanted a single value that was related to peak shape, we decided to use the ratio of the
two maxima, and found that this was a more sensitive function of peak shape.
The distribution function was calculated for all peaks using equations published
elsewhere (6 -8 ). For the peaks containing noise, a Savitsky-Golay smooth was applied
using the same smoothing factor as above before the DFM calculation.
1 0 5
RESULTS AND DISCUSSIO N
M easurement of Q and XQ from real data. Application of the Q
transformation on real (noisy, digitized) data requires the consideration of several data-
handling factors, such as (i) noise, (ii) smoothing, and (iii) the number of points in a
peak. Since each of these factors may potentially affect the measurement of Q and XQ, it
is appropriate to discuss them here.
Effect of smoothing on Q transformation. Depending on the degree of
noise, one may need to smooth the data prior to applying eq 1. However, smoothing of
peak data usually results in distortion of the peak shapes being measured. Therefore,
one should smooth the data in a manner that maximizes the accuracy of the measurements
being made while minimizing any peak shape distortion due to smoothing (2 0 ).
Fortunately, this is fairly easy to accomplish via judicious selection of the smoothing
parameters. Furthermore, when the objective is peak shape comparison of similar peaks,
any slight errors caused by smoothing process itself will be virtually eliminated
(cancelled) so long as the smoothing is performed consistently for all peaks.
Based on the above considerations and our prior experience with
chromatographic peak characterization (13,21,22), we selected the efficient smoothing
algorithm of Savitsky and Golay (17-19). As with all smoothing methods, some
distortion of the peak shape, however slight, is inevitable. Figure 5-6 shows the
resulting errors in XQ introduced into the data for noiseless EMG peaks with T/(7=l.
Since the peaks contain no noise, the errors are assumed to be due solely to the distortion
of the peak's shape. Because the errors are so small (< 0.25%), we believe that if one
uses the same smoothing factor (eq 6) for aR the peaks being compared (unknowns and
standards), no error between peaks due to distortion caused by smoothing should be
1 0 6
observed. In addition to errors in XQ, one can expect some error in the individual
points within the Q profile. In general, however, the errors in Q at any value of r are so
slight that Q profiles of the same peak calculated with different smoothing factors are
virtually superimposable. Therefore to illustrate just how small this effect of smoothing98
is, we calculated the normalized residual sum, ^
r= 2
the smoothing factor. The results, shown in Figure 5-6b, demonstrate that similarly to
the effect of smoothing on XQ, the effect of smoothing on Q itself is entirely negligible if
an appropriate degree of smoothing is employed.
Effect of noise. Whether or not the real (noisy) data are smoothed, they will
still contain some noise which may-have an effect on the measurement of Q and XQ.
Note that since Q represents a data transformation in which both the signal and the noise
are transformed, the effect of noise on Q cannot be predicted intuitively based on
conventional S/N wisdom. Figure 5-7 shows the relative error of Q (averaged over r = 2
to 98%) for various signal to noise (S/N) ratios. A smoothing factor of 0.2 was applied
to the data before measuring Q and calculating the errors. This smoothing factor was
chosen because it offers a good compromise between peak distortion (see Figure 5-6)
and S/N enhancement (17), while also requiring only a small number of convolute points
for the smooth, thereby reducing computation time. Although the relative error for each
point occasionally exceeds 6 % as shown in Figure 5-7, the errors appear to be randomly
distributed. Therefore the value of XQ will be only slightly affected by the magnitude of
the noise. For a S/N of 100 (same data as before), the relative error in XQ is only 0.4%;
larger S/N ratios resulted in even smaller relative errors.
Because a least squares line fit is used to determine a and b accurately at the
desired height, some additional smoothing may be introduced by this operation.
However, the effect will be slight, as the peak has already been smoothed via the
* as a function of
1 0 7
Savitsky-Golay method. As shown before (20), the fraction of noise remaining after
multiple smooths varies approximately with the inverse eighth root of the number of
smooths.
Effect of the num ber of points in a peak. The final consideration when
measuring Q is the number of points within the peak(s) being investigated. Because we
are dealing with digitized data, we can expect increased accuracy in the measurement of
the parameters a and b, and subsequently greater accuracy in the values of Q, as the
number of points in a peak increases. Figure 5-8 shows the accuracy of XQ as a
function of the number of points in the top half of a peak. Instead of a monotonically
changing curve as one would expect, the data seems to be discontinuous, but with all
values of relative error less than 0.2%. We believe this scatter is due to the complex
relationship between the number of points in the peak, the shape of the peak, and the
changing number of points used in the line fit for measurement of the parameters b and a
(see Experimental Section). As seen in Figure 5-8, very few points are needed for
accurate measurement of XQ, in sharp contrast to the moment method which needs at
least 4 to 5 times as many points for accurate quantitation (12).
Although the error in XQ remains small as the number of points in a peak
decreases from 250 to 20 in Figure 5-8, the error in Q at all relative peak heights (r)
increases (Figure 5-9). As illustrated in Figure 5-9f, the error in Q (averaged over all r
values) increases almost exponentially from 0.1 to 3.5% as the number of points in a
peak decreases. This error is due to the errors in the measurement of the parameters a
and b as discussed in ref. 13. Again, as shown in Figure 5-8, these errors tend to cancel
when XQ is evaluated.
Comparison of methods using noiseless data. To compare the abilities
of the four methods (XQ, DFM, moments and second derivative) for detecting peak
shape changes or differences, we first calculated the sensitivity of each method for
1 0 8
noiseless EMG peaks. We defined this noiseless sensitivity as the percent change in
each peak shape parameter (ZQ, etc.) relative to its value at %Jo=\ as the asymmetry of
the peak is increased from at x/a=l. The basis of peak shape measurement for the DFM
was simply the area under the curve while the ratio of the peak height maxima was used
for the second derivative. As mentioned in the Experimental Section, we found that the
percent change in the second derivative maxima ratio was greater than either the first
maximum to minimum or second maximum to minimum ratio.
Shown in Figure 5-10 are the sensitivities of the methods for detecting peak
shape changes in noiseless data. Apparently, the third and fourth statistical moments are
the most sensitive in terms of percent change per change in peak shape. The sensitivities
of the other methods (in descending order) are excess, skew, second derivative, ZQ and
DFM for noiseless EMG peaks.
Comparison of methods using noisy data. When noise is introduced
into the EMG peaks, the relative abilities of the methods to detect differences in peak
shape change dramatically. The relative ability of each method, defined in terms of the
minimum detectable change in peak shape, relative to the shape at x/a= 1, was calculated
using a t-test (14). Rearranging equation 7, for calculation of t for the comparison of
two means
we obtain equation 8 . The parameter S is the pooled standard deviation for the two sets
of data, n corresponds to the number of points in each data set, and x is equal to mean of
parameter being considered (i.e. ZQ).
[7]
1 0 9
x j - x2 = tSni + n2
n l n 2[8]
Equation 8 allows one to calculate the minimum difference in means between two data
sets needed for the means to be considered significantly different at the desired
confidence interval. For this work we chose the 99% confidence interval, nj and n2
were equal to eleven, as that was the number of peaks simulated for each S/N (see
Experimental section). Equation 8 can then be reduced to
parameter of interest (ZQ, DFM, etc.), we can determine the corresponding peak shape
change ('t/a increased) from a reference value (t/a=l) that would give this difference
between means. For this work we assumed that the necessary significant difference in
means was equal to the difference in the true values, thereby significantly reducing the
number of simulated peaks needed. S was calculated from the eleven peaks at the desired
S/N, assuming that the standard deviations of the reference value and the difference value
were equal. Shown in Table 5-1 are the results for ZQ, DFM, second derivative, and
all the moment-based parameters at three different S/N ratios. Obviously, the Q
transform is superior to the other methods when noise is included in the data, especially
at the lower S/N ratios, while the second moment and DFM methods compare favorably
to the ZQ method at higher S/N ratios. We arbitrarily chose 't/a =1 as the reference peak
shape. However, the relative abilities of the various methods to detect peak shape
differences should not change with a different reference peak shape. Note that because
we used ideal baseline levels in the calculations for all the methods, the adverse effects of
low S/N ratios are not completely reflected in the results for the statistical moments in
Ax = 1.2131 S [9]
Therefore, by knowing the necessary significant difference between means for the
110
Table 5-1, i.e.,the performance o f all the moments, including the excess and skew,
would have been much poorer had more realistic baseline levels been used. This is due
to the extreme sensitivity of the statistical moments to baseline errors (9-11,23). Shown
in Table 5-2 is the performance of the Q method when baseline errors are introduced;
these results are significantly better than those obtained for the moments of EMG peaks
described elsewhere (23).
Although the performance of the DFM and XQ methods are comparable at
moderate to high S/N ratios, the XQ method is significantly better for low S/N ratios. In
addition, the Q transform approach inherently provides well established graphical
parameters (a, b) that can also be used to calculate other important chromatographic
figures of merit such as plate count (2 1 ), peak area (2 2 ), and statistical moments (13,
14). Given the similar moderate computational complexity of the XQ and DFM
methods, the Q data transformation approach to peak shape analysis and comparison
arguably represents a significant improvement over the DFM method, and certainly over
the slope and moment-based single-channel methods.
Potential applications of Q the transform . The Q transformation may be
employed in other ways in chromatography and analytical chemistry. For example,
based on preliminaty unreported gas chromatographic data, we have found that the XQ
can be very useful for solute identification when retention times alone are insufficient.
Because each compound interacts with the stationary phase differently, their peak shapes
may be different enough to identify them based on their XQ value even when their
retention times are similar. We have been able to distinguish between solutes with
identical retention times whose XQ values are different. In the paper immediately
following this issue, we examine the ability of the Q transform to detect the presence of
impure chromatographic peaks when the degree of overlap between two overlapping
111
peaks is so severe that no shoulder or other visible evidence exists to indicate this
coelution.
Because many times peak shapes can give indications of instrumental problems,
the Q transformation may also be useful for automatic instrument troubleshooting. For
example, consistently negatives values of Q vs r would signal the presence of a fronted
peak in chromatography, which in turn is often indicative of sample overload. As
illustrated in Figure 5-11, a fronted chromatographic peak will give a Q profile {small
positive values of Q, Q < 1) and/or EQ value that is different from normal peaks. Also,
peaks with excessive tailing {large positive values of Q) may indicate that (i) the detector
time constant is excessive for the present conditions, (ii) retention is due to two or more
mechanisms, or (iii) active sites in the chromatographic column are present.
C O N C LU SIO N
For asymmetric peaks, such as the exponentially modified Gaussian peaks
employed primarily in this study, the Q tranformation approach introduced here appears
to be superior to slope analysis, moments, and the distribution function method for the
measurement and characterization of peak shape, including the detection of peak shape
changes, especially at low S/N ratios. At higher S/N ratios the differences are somewhat
less significant, particularly between EQ and DFM. The relative insensitivity of the latter
methods to baseline errors and their computational simplicity would appear to make both
suitable for peak shape difference measurement As we will show in the article
immediately following this one, however, the Q transform is more sensitive than the
DFM to peak shape changes caused by the overlap of two peaks and is therefore better
for the single-channel detection of impure chromatographic peaks. Given this advantage,
the lower sensitivity of the EQ method to low S/N ratios, and the fact that it inherently
112
provides values for graphical parameters (a and b) that can be used to calculate other
important chromatographic figures of merit, however, we believe that the XQ approach
is moderately superior to the DFM. The Q transform may also provide additional
advantages for other applications that we have not yet explored.
CREDIT
This work was supported in part by the National Oceanographic and Atmospheric
Administration, NOAA-50-ABNC-7-00100.
1 1 3
TABLE 5-1. MINIMUM DETECTABLE CHANGE IN (x/c) FOR VARIOUS
METHODS OF PEAK SHAPE ANALYSIS a
S/N
1 0 0 500 1 0 0 0
XQ 0.047 0.018 0.016
Slope Analysis 0 .2 0 0.16 0.067
m 2 0.074 0.026 0.018
m 3 0.36 0 .1 0 0.091
M4 0.56 0.17 0.096
EXCESS b 1 .0 2 0.48
SKEW 2.32 0.23 0 .1 2
DFM 0 .2 2 0 .0 2 0 0.018
a t/ct for the reference peak is 1 .0 .
b No change in peak shape detectable in the range 1.0 < 't/a < 4.0 at this signal-to-noise
by the specified method.
1 1 4
TABLE 5-2. % RELATIVE ERROR IN I Q DUE TO BASELINE ERRORS IN EMG
PEAKS.
HORIZONTAL BASELINE a
BASELINE HEIGHT x/a=l i /o=3
2% 1.74 2.86
4% 3.16 5.30
6 % 4.39 7.52
8 % 5.50 9.57
10% 6.51 11.51
SLOPING BASELINE b
BASELINE HEIGHT x/o=l x/a=3
2% 6.79 8.80
4% 11.59 15.20
6 % 15.40 20.34
8 % 18.60 24.69
10% 21.38 28.48
a Baseline was a line drawn horizontally at indicated fraction of peak height. It was
then subtracted from the peak.
b Baseline was drawn from beginning of peak(approximately 0.001 % of peak height)
to a point on the tail of the peak nearest indicated baseline height (fraction of original
peak height) and then subtracted from peak.
1 1 5
Q TRANSFORMATION
r-1r-1
imi i i i i | i i i i im i | in i i i i i | i i i i i i i i i |80 100
Figure 5-1. Measurement of a and b used for the calculation of Q in eq 1.
1 1 6
WHICH TWO PEAKS HAVE THE SAME SHAPE?
ANSWER : PEAKS 1 AND 2
1 EMG ( T / a =2.6)2 EMG ( T / 0 = 2.6)3 EMG ( X / a = 2.2)4 EMG ( T / a = 2.8)5 LOG-NORMAL ( Asymmetry = 1.6)
Figure 5-2. The Q transform can be used to visualize peak shape differences without
regard to peak height or width. Peak identity: (1) Exponentially Modified
Gaussian (EMG), T/0 =2.6, 0=1.0 (2) EMG, T/0=1, 0=0.26 (3) EMG,
T/0=2.2, 0=0.26 (4) EMG, T/0=2.8, 0=0.5 (5) Log-Normal,
asymmetry=1 .6
1 1 7
1311
9
7
Q
5
3
0.51
1006040206040200 '0,0/
EMG
GAMMAASYMMETRYLOG-NORMAL
1.4
10040
Figure 5-3. Comparison of Q profiles for various peak shapes, (a) EMG peak, ref. 13
(b) Gamma peak, eq 4 (c) Log-Normal peak, eq 5 (d) Overlay of Q
profiles from (a) through (c). Note that peak shapes can be differentiated
on the basis of LQ or on direct point-by-point comparison of individual Q
profiles.
A b
/ (A
b,a=
1)
1 1 8
1 .5 -
GAUSSIAN, <7 = 1.2
1.0 .
GAUSSIAN, 0 = 1.0
0 - 5 ■ ' • ■'' ‘ r — ■ ■ ■ ■ I • ■ ■ ""> I ■' » " "» ............ ........ ' I- - » """l
20 40 60 80 100f fOA
Figure 5-4. Differentiation of peak shapes using Q transform between two symmetric
peaks by a redefinition of Q as discussed in the text.
1 1 9
1 1 1LOG-NORMAL PEAK SHAPE ’Asym = 1.6
Q
EMG (tau/sigma = 2.8)
80 1000 20 40 60
Figure 5-5. Non-superimposability of Q profiles of fundamentally different type of
peaks with identical values of XQ. Peak identities: EMG with x/o = 2.8
and log-normal with an asymmetry (eq 5) of 1.6.
120
0 .22-1
0 .2 0 -
owulCC 0 .1 8 -
0 .16
0.14*1 "i—r -T —r "'"| - t i i i | i i i i " » i i | i i i i | i i i i |
0. 0 0. 1 0. 2 0. 3 0. 4 0. 5 0. 6
SMOOTHING FACTOR
O
<=>q(/)h iCCoUJNZj<sCCoz
0 .04
0 .0 3
0.02
0.01
0.000.2 0 .3 0 .5 0.60.0 0.1 0 .4
SMOOTHING FACTOR
Figure 5-6. Effect of Savitsky-Golay smoothing factor (eq 6 ) on (a) Z Q ; and (b) Q vs
r for an EMG peak with T/G =1.
AVER
AGE
% RE
, Q
121
8
7
6
5
4
3
2
1
0200015001000500
S/N
Figure 5-7. Effect of signal to noise (S/N) ratio on the Q transformation.
122
0 .20 -l
0.15-
0.05-
0.002.51.5 2.01.0
LOG (#PTS, 50%)
Figure 5-8. Effect of the relative sampling rate (number of points) on the measurement
of IQ . Peak shape: EMG peak, T/G=l.
1 2 3
Figure 5-9. Effect of the relative sampling rate (number of points) on the accuracy of
Q. Number of points in the peak measured from half-height to half-height.
Figure 5-9e shows true Q vs r profile (obtained via search algorithm) while
Figure 5-9f shows the average error associated with the measurement of Q
at any r.
1 2 4
3.01
2.5-
2.0
1008040 600 20
3.0 ~|
2.5
2.0
1.5
60 10020 40 600
1 2 5
Q
3.0-I
2.5
2.0
80 10040 60f fo,
3.0 n
2.5
2 .0 -
80 10040 60I* / o>
AVERAGE % RE, Q
Us- 0H0 )
01o
rooo
cn
roo
roUl
o
o cno cn oo
AO
o
00o
126
LOG
(% C
HA
NG
E)
1 2 7
SLOPE ANALYSIS
EXCESS
SKEW
IQ
DFM
3
t/ ct
Figure 5-10. Relative abilities of statistical moments, excess, skew, second derivative,
I Q and DFM to detect changes in peak shape for noiseless peaks.
SIG
NA
L
128
0.5-1
0.4
0.3
0.2
0.1
0.066420
TIME (MIN)
1.0i
0.8
0.6
Q0.4
0.2 :
0.0100806020 400
r (%)
Figure 5-11. Automatic detection by the Q transform of instrumental problems in
chromatography, as illustrated with (a) a fronted peak (indicative of sample
overload); and (b) its Q profile.
1 2 9
LITERATURE CITED
1) Brown, Steven D.; Barker, Todd Q.; Larivee, Robert J.; Monfre, Stephen L.;
Wilk, Harlan R. Anal. Chem, 1988, 60, 252R-273R.
2) Stevens, J. G.; Bowen, L. H.; Whatley, K. M. Anal. Chem, 1988, 60, 90R-
106R.
3) Grushka, Eli. Anal. Chem., 1972, 44, 1733-1738.
4) Grushka, Eli.; Monacelli, G. C. Anal. Chem., 1972,44,484-489.
5) Grushka, Eli.; Myers, M. N.; Giddings, J. C. Anal. Chem, 1970, 42, 21-26.
6 ) Rix, H. Anal. Chim. Acta, 1986, 191, 467-472.
7) Rix, H. J. Chromatogr., 1981, 204, 163-165.
8 ) Excoffler, J.-L.; Jaulmes, A.; Vidal-Madjar, C.; Guiochon, G. Anal. Chem.,
1982, 54, 1941-1947.
9) Chesler, S. N.; Cram, S. P. Anal. Chem, 1971, 43, 1922-1933.
10) Petitclerc, T.; Guiochon, G. J. Chromatogr. Sci., 1976, 14, 531-535.
11) Grushka, Eli.; Myers, M. N.; Schettler, P. D. Anal. Chem., 1969, 41, 889-892.
12) Eikens, David I.; Carr, Peter W. Anal Chem., 1989, 61, 1058-1062.
13) Jeansonne, Mark S.; Foley, Joe P. J. Chromatogr., 1989, 461, 149-163.
14) Miller, J.C.; Miller, J.N., Statistics For Analytical Chemistry, 2nd ed.; Horwood:
1988; Chapters 3 and 6 .
15) Mendenhall, William; Scheaffer, Richard L ., Mathematical Statistics with
Applications, 1st ed.; Duxbury Press: California, 1973; Chapter 4.
16) Grimalt, J.; Iturriaga, H.; Thomas, X. Anal. Chim. Acta, 1982, 139, 155-166.
17) Savitsky, A.; Golay, M. J. E. Anal. Chem., 1964, 36, 1627-1639.
18) Steinier, J.; Termonia, Y.; Deltour, J. Anal. Chem., 1972, 44, 1906-1909.
19) Madden, H. H. Anal. Chem., 1978, 50, 1383-1386.
1 3 0
20) Enke, C. G.; Nieman, T. A. Anal. Chem., 1976, 48, 705A-709A.
21) Foley, Joe P.; Dorsey, John G. Anal. Chem., 1983,55,730-737.
23) Anderson, David.J.; Walters, Rodney.R.; J. Chromatogr. Sci., 1984, 22, 353-359
22) Foley, Joe P. Anal. Chem., 1987, 59, 1984-1987.
Chapter 6
SINGLE CHANNEL PEAK IM PURITY DETECTION USING
THE Q TRANSFORMATION
Reprinted with permission from Burgess, S.W., Jeansonne, M.S. and Foley, J.P.;
submitted for publication in Anal. Chem. Unpublished work Copyright 1990 American
Chemical Society.
13 1
1 3 2
Department o f Chem is try
L o u i s i a n a S t a t e U n i v e r s i t y and agricultural and mechanical collegeBATON ROUGE • LOUISIANA ■ 70803-1804 , « 7
Barbara Polanski Publications Division American Chemical Society 1 155 16th St. Northwest Washington, DC 20036
Dear Ms. Polanski,
1 am writing to you in reference to the article en t i t led ," Single Channel Peak Impurity Detection Using the Q Transformation ", AC900689Y, received by A nalytical Chem istry on August 24, 1990, for which J am the f irs t author. I would like to use the manuscript as part of my Ph. D. dissertation. The completed d isserta t ion will be submitted to University Microfilms, Incorporated. Please forward permission to reproduce the manuscript.
I appreciate your prompt reply.
S04/388-3361 FAX 5041388-3458
September 13, 1990
Sincerely,
'Prof. Joe P. FMeyT toA jtw d i
Mark S. Jeansonne
American Chemical Society 133
PUBLICATIONS DIVISION 1155 SIXTEENTH STREET, N.W. WASHINGTON, D.C. 20036 P h o n e (202) 872-4600 Fax (202) 872-6060
Septem ber 28, 1990
Mr. M ark S. Jeansonne D epartm ent o f Chemistry Louisiana S ta te University B aton Rouge, Louisiana 70803-1804
D ear Mr. Jeansonne:
Thank you for your le tter o f September 13, in which you requested permission to include in your thesis, two papers that you subm itted to Analytical Chemistry :"Single Channel Peak Im purity D etection Using the Q Transform ation," and "The Q Transform ation: A Novel M ethod o f Peak Shape Analysis."
I would be happy to grant you this permission free o f charge provided that you print th e required ACS copyright credit line on the first page o f your article and that your paper is published by ACS first before you submit it to UMI. Also, please inform UM I that perm ission is granted subject to the terms o f the ACS/UM I Agreem ent dated O ctober 1, 1984.
The credit line we require is, "Reprinted with permission from FU LL R E FE R E N C E CITA TIO N Copyright Y EA R American Chemical Society." Please be sure to insert the appropria te inform ation in place o f the words in capital letters.
If your paper is not accepted before you include it in your thesis, please use the following credit line: "Reprinted with permission from Burgess, S.W., (A U TH O RS NAM ES); subm itted for publication in Anal Chem. U npublished work Copyright 1990 Am erican Chem ical Society."
Thank you for your request. Please let me know if I can be o f any further help.
Sincerely,
Barbara F. Polansky Copyright A dm inistrator
1 3 4
INTRO DUCTIO N
The problem of overlapping peaks is .frequently encountered in chromatographic
analysis. In many instances, the peaks are sufficiently resolved that the overlap is
obvious from visual inspection and detection of the overlapping peaks can be easily
accomplished visually and/or automatically. In other situations, however, the resolution
of peaks may be low enough that no shoulder or other easily detectable distinguishing
characteristics appear, i.e., the composite peak is indistinguishable from a single peak as
illustrated in Figure 6-1.
Modem multichannel, hyphenated techniques such as GC/MS, GC/FTIR,
HPLC/photodiode array and HPLC/ICP can be used to assess peak purity, but are not
always available or directly amenable for use with present general-puipose laboratory
single-channel data systems (1). Also, these multichannel techniques usually assume
that the characteristics of the compounds forming the composite peak (mass spectrum,
UV spectrum, e tc .,) are different. As pointed out before, the co-elution of two
compounds is frequently due to their very similar structures, and therefore their spectra
could also be too similar for the multichannel techniques to be successful (2). Hence,
various single-channel methods for peak purity determination have been pursued,
including slope analysis (3,4), moment analysis (5) and the distribution function method
(DFM) (6-8).
Slope analysis utilizes the second derivative of a chromatographic peak to
determine peak purity by comparing properties of the second derivative for pure and
impure peaks. By comparing the ratios of the peak's second derivative maxima and
minimum to those for a known pure peak, a determination of purity can be made.
However, there are two major drawbacks to this approach. First, in the method
described by Grushka and co-workers (3,4), a very accurate peak model must be used
1 3 5
or assumed. Unfortunately, real chromatographic peaks are not always described with
sufficient accuracy by any given model. Secondly, the second derivative is very
sensitive to noise. Because accurate values of the peak's second derivative minimum and
maxima must be measured, noise can cause a large error.
Moment analysis is a mathematical peak purity determination method in which the
skew and excess of pure peaks are compared to the skew and excess of impure peaks.
Here, the excess is plotted versus the skew for pure modeled or reference peaks in order
to obtain a pure reference curve. When the plotted excess versus skew point for other
peaks do not fall on the curve, the peaks can be suspected of being impure. As with
slope analysis, this method must assume a peak shape model or use internal standards
with exactly the same peak shape as those in the chromatogram to be examined.
Assuming that an accurate model or reference peak(s) is/are available for comparison
with the real chromatographic peaks, yet another challenge awaits. The skew and excess
for the real peaks must now be measured very accurately. As seen in eqs 1 and 2, the
skew and excess depend on the second, third, and/or fourth central moments.
Skew = Ys = j^S/2 I‘]
Excess = 7e - M ^2 *_ j PJ
The inability to measure accurately the third and fourth statistical moments for a real
chromatographic peaks due to noise (9-11) and baseline errors (12,13) results in a
significant propagation of errors to the skew and excess. Thus, the comparison of the
skew and excess of real peaks to reference values obtained either via mathematical
models or internal standards can be quite error prone, and represents a major limitation of
moment analysis for peak purity determinations.
1 3 6
In the distribution function method (DFM), the distribution of area under a peak is
plotted as a function of time and the resulting distribution is compared to a standard or a
reference. Although this method is sensitive to peak shape changes (6-8,14), we will
show later in this paper that it is not as sensitive to peak impurities as the method we are
introducing.
Due to the shortcomings of the above methods, we believe that another method for
the single-channel detection of co-eluting peaks is needed. In this current project, our
goal was to develop such an alternative that circumvents the problems associated with the
second derivative, moment analysis, and DFM while retaining their computational
simplicity.
In the preceding chapter, we reported a mathematical transformation for data
typically associated with digitized chromatograms or spectra. The signal vs. time data
are transformed into a shape factor measured from 2 to 98% of the peak height in
increments of 1%. We term this the "Q transformation", defined in eq 1 of the preceding
chapter (14). Also introduced were the "Q vs r" profile and XQ, the latter simply the
summation of Q over all r for a given peak.
Whereas the previous chapter focused on the benefits of this new data
transformation for the analysis and comparison of pure peaks, the purpose of the present
article is to demonstrate that this same approach can be used for the single-channel
detection of severely overlapped peaks (no valley, shoulder, or other visible artifact).
The advantages of this approach over multi-channel approaches are: 1) the universal
availability of single-channel chromatographic detection in any laboratory; and 2) the
ability to work even when the impurity and major peak have very similar or even
identical absorption or fluorescence spectra. Potential advantages of this approach over
the second derivative, moment analysis, and DFM methods are as follows: 1) no prior
assumption of a particular peak shape model; 2) less vulnerability to noise than the
1 3 7
second derivative; 3) and more sensitivity than the DFM to peak shape changes caused
by co-eluting peak impurities.
EXPERIM ENTAL
An Apple Macintosh Plus personal computer was used for simulated peak
generation and other calculations. All programs were written in either Microsoft Basic or
True Basic. An Omega-4 data system (Perkin-Elmer, Norwalk, CT, USA) utilizing an
IBM PC-AT computer was used for chromatogram collection and storage. Liquid
chromatographic (LC) data was collected from a Series 4 liquid chromatograph (Perkin
Elmer, Norwalk, CT, USA) using a reversed phase C8 column. Detection was achieved
using a Model variable wavelength ultraviolet absorption detector (Isco, Lincoln,
NB, U.S.A.). Injections were made manually using a Model 7125 standard six-port
injection valve (Rheodyne, Cotati, CA) with 20 |iL loop.
Individual liquid chromatographic peaks of toluene, ethylbenzene, isopropyl
benzene, n-propyl benzene, n-butyl benzene, 4-methyl cyclohexanone, 3-heptanone, 4-
heptanone and 5-nonanone were generated using a mobile phase consisting of 60%
HPLC grade acetonitrile (ACN) and 40% water. Individual peaks of toluene,
ethylbenzene, n-propyl benzene, and n-butylbenze were generated using a mobile phase
of 70% HPLC grade methanol (MeOH) and 30% water. Individual liquid
chromatographic peaks of valerophenone and acetophenone were generated using a
mobile phase of 50% HPLC grade ACN and 50% water. Individual peaks of n-propyl
benzene and toluene were obtained with a gradient of 5% ACN/min while individual n-
butyl benzene and ethyl benzene peaks were generated using a gradient of 5%
MeOH/min. All peaks were generated with a flow rate of 2.0 ml/min with detection of
the alkyl benzenes at 254 nm and the ketones at 272 nm. The water used was double
1 3 8
distilled and deionized. All of the compounds used were at least 99% pure and the
sample solvent was identical to the mobile phase in all cases. This was done to minimize
any effects on peak shape caused by the delayed-injection method of generating co
eluting peaks (vide infra), whereby the solvent or void peak passes through a previously
injected analyte peak(s). Except when using the delayed-injection method, an approach
not applicable to real samples, our experience and the results of others have shown that
composition of sample solvent has little effect on peak shape provided that its solvent
strength is less than or equal to that of the mobile phase (15).
Overlapping real chromatographic peak pairs consisting of n-propyl benzene :
toluene, isopropyl benzene : ethylbenzene, and valerophenone : acetophenone were
generated using a delayed injection technique. That is, in each case the compound with
the greater retention time was injected, a time approximately equal to the difference in
retention time between the two was allowed to pass, and the second compound injected.
In this way, overlapping peaks could be generated with various resolutions, depending
on the delay time used. However, the actual difference in retention times could only be
estimated to within 0.5 second due to the inherent imprecision of manual injections.
Due to experimental difficulties, we were not able to generate any impure peaks by
the double-injection method when employing a gradient. Therefore, we artificially
contaminated a real peak by mathematically adding to it another peak obtained under the
same conditions, after the size of the impurity peak had been mathematically reduced.
Single, noiseless exponentially modified Gaussian (EMG) peaks were generated
with tau/sigma (x /a) ratios between 1 and 4 as described previously (12). The EMG
function results from the convolution of a Gaussian (standard deviation of O) and an
exponential decay function (standard deviation of X). As the T/O ratio increases, the peak
in question will become more skewed, and as it decreases, the peak approaches a
1 3 9
Gaussian shape. Eleven single EMG peaks of x /a =1 were generated at Signal-to-Noise
(S/N) ratios of 100, 500, and 1000 (14).
Simulated overlapping EMG peaks were generated by simple addition of the major
and impurity peaks; both elution orders were considered, i.e., the impurity could elute on
either the front or tail of the major peak. The area ratio of major peak to impurity was
16/1, with both peaks having a T/O ratio of unity. The resolution was varied from
0.03125 to 0.5 in increments of 0.0625. Resolution was defined as AtR /4 VM2 , where
M2 , the variance, is equal to Oq2 + x2 for an EMG peak.
Peak param eter measurement.For simulated peaks, start/stop assignments
(limits of integration) were taken as the point where the signal height of the peak (before
noise was added) crossed the baseline threshold level, arbitrarily defined as r = 0.001%
(see Figure 5-la of preceding article for definition of r). The parameters a and b for
calculation of Q, skew, excess, and second derivative were measured by methods
explained before (14). The distribution function was calculated using equations
published elsewhere (6-8).
The second derivative of most peaks has two maxima and one minimum. Grushka
et al. (4) used the ratio of each maxima to the minimum and plotted these values versus
each other for simulated peaks using a model they thought appropriate. They then
compared this plot to those points determined for sample peaks in order to detect peak
impurities. We used either the ratio of maxi or max2 to the minimum, whichever was
more sensitive. In general, the maxl/min ratio will more sensitive to impurities eluting in
front of a peak while the max2/min ratio will be more sensitive to impurities eluting in the
tail of a peak.
1 4 0
RESULTS AND DISCUSSIO N
Peak shape comparison. When comparing the shapes of peaks, real or
simulated, there can be two approaches when using the Q transformation. First, one can
compare values for XQ, which corresponds to the area under the Q profile (plot of Q vs
r). If the assumption is made that the individual peaks being compared belong to the
same family or class of peak shapes (for example, EMG peak shapes), then any
difference in peak shape will translate into a difference in area under the Q profile or XQ
(14). One can make this assumption for either of the following scenarios: (1) a sample
peak is compared to a reference peak; or (2 ) the compounds are known to be pure and
elute with the same retention mechanism, such as compounds of a homologous series,
where any differences in peak shapes may be attributed only to differences in
intracolumn band broadening.
The second approach is to compare point by point the shapes of the individual Q
profiles either visually or statistically using a test such as the paired t-test (16,17). This
approach is very useful, as shown later, for the detection of impure peaks with only one
reference peak and one sample peak.
Simulated im pure peaks. In general, the deviation of the Q profile measured
for an impure peak will be influenced by location of the impurity within the peak as
shown in Figures 6-2 and 6-3. In these two examples, the areas under the Q profiles for
simulated, noiseless standard and sample peaks were compared for peak purity
determination. Figure 6-1 shows a simulated impure peak composed of two overlapping
EMG peaks, each with X/G = 1 at a resolution of 0.25 while Figure 6-2 shows the
corresponding Q profiles for the pure and impure peaks from Figure 6-1. The important
feature to note in Figure 6-2 is the positive deviation of the Q profile for the impure peak
from that of the pure peak for r = 15 to 98% without any compensating negative
141
deviation from r = 2 to 14%. This difference enables an impure peak to be detected by
comparing XQ for both peaks.
Figure 6-3a again illustrates a simulated impure peak composed of two overlapping
EMG peaks, each with T/G = 1 at a resolution of 0.25, except that the impurity elutes on
the front of the major peak. In this case the resulting Q profile for the impure peak
deviates negatively from that for the pure peak, again allowing one to use XQ for
impurity detection while also allowing one to judge whether the impurity is eluting on the
front or the back of the major peak.
Noisy simulated peaks. When random noise is present in the peaks (as in real
data), one must treat the data statistically in a way similar to that reported in the preceding
paper, except that the resolution between co-eluting peaks is changed instead of the
shape of the single pure peak. The ability of each method to detect co-eluting peaks was
evaluated as follows: For a given signal to noise (S/N) ratio, the necessary change in the
parameter of interest (XQ, XDFM, etc.) was calculated depending on the standard
deviation of eleven single noisy peaks, as described in the previous chapter. Another
smaller peak was then superimposed on the original, and the resolution was then
increased until the value of the parameter of interest (XQ, DFM, etc.) exceeded the
necessary change for the 99% confidence level. The resolution at which this occur ed is
reported in Table 6-1 for the various methods as a function of S/N and elution order
(major and minor (impurity) peak).
Importantly, Table 6-1 shows that the ZQ method is better at detecting peak
impurities than any o f the other methods. Note also that the skew and excess methods
failed to detect any of the impurities co-eluting with the major peaks over the resolution
range shown here, even for large S/N values. Finally, although in the preceding paper
(14) we found the DFM to be almost as good as the Q transform approach for the
142
detection of overall changes in peak shape, as Table 6-1 shows the DFM is not nearly as
sensitive as the XQ approach to peak shape differences caused by impurities.
Real impure peaks. In order to determine peak purity in a noisy,
multicomponent chromatogram, one can use a procedure similar to that described above,
provided that the identity of the analytes in question are known or can be assumed. Once
their identity has been determined, replicate injections of both the sample and standard
peaks will provide two sets of XQ points. A t-test can then be used to determine whether
the two sets of XQ points are significantly different at the 99% confidence interval.
Because the t-test for comparison of means performs better as the number of injections is
increased, one may need many XQ points for both the standard and sample. For routine
use this multi-injection approach is somewhat laborious. Another possible statistical test
exists that requires only one injection of the standard and one injection of the sample. By
comparing the individual Q profiles via a paired t-test (16,17) one can determine whether
the two profiles are significantly different. The results of applying this procedure to
simulated peaks are shown in Table 6-2. As seen there, this test is not quite as sensitive
as the multiple injection method, although the results are comparable. Because only
discreet resolutions were used, the results may be better than those presented had smaller
resolution increments been considered. Nevertheless, the paired t-test using Q profiles
allows better detection of impure peaks than the multiple injection method used in
conjunction with the DFM,slope analysis, excess, and skew methods . When the paired
t-test was used with DFM and slope analysis, no impurites could be detected over the
resolution range used. In fact, the DFM always found that two simulated pure peaks
with the same x/a and same S/N were significantly different! The paired t-test cannot be
used with the excess and skew methods because those methods only give a single value
pertaining to peak shape.
1 4 3
To summarize, the analyst has two options for comparing standard peaks and
sample peaks. First, by injecting the the standard many times and the sample several
times and comparing the resulting two sets of XQ points via a t-test. If time and labor
are not large concerns, this option will give superior results. In fact, by using a larger
number of peaks than were used for determination of the values in Table 6-1, one may
achieve even better results than those presented there. Second, one can inject the
standard and sample peaks each once and utilize a paired t-test to compare the individual
Q profiles. Although this option is slightly less sensitive to impurities than the first
option, it may be better in terms of time and labor.
Experim ental verification of paired-t option. Because the t-test for
comparison of means discussed above is not amenable for most routine analyses, we
present experimental verification for only the paired t-test in this report Shown in Table
6-3 are the results of applying the paired t-test to detect impure peaks under various
mobile phase conditions in high performance liquid chromatography (HPLC). Note that
the paired t-test worked over a range of mobile phase conditions and S/N ratios. Both
the second derivative and DFM methods were unable to detect the impurities. Figure 6-4
illustrates the appearance of the pure and impure n-propyl benzene peaks that were used
for the data in row four of Table 6-3. How does the paired t-test treat identical pure
peaks obtained under the same chromatographic conditions? Table 6-4 shows that groups
of pure peaks, all obtained under the same experimental conditions, are not detected as
impure by the paired t-test.
Potential for qualitative analysis. Another possible application of the Q transform
is qualitative analysis. Presently, most qualitative analysis in chromatography is
performed using retention time windows. As illustrated in Figure 6-5, XQ data obtained
for a given compound during different runs (but otherwise identical chromatographic
conditions) will cluster together when its XQ values are plotted vs log t^ (Figure 6-6).
1 4 4
From preliminary data not presented in this article, we have observed that ZQ data for
pure compounds with similar or identical retention times (same set of chromatographic
conditions) will often cluster in different regions because their peak shapes are different.
As seen in Figure 6-5, peaks of 3-heptanone and 4-heptanone cluster slightly apart from
each other even though their retention times are almost identical. These peak shape
differences could be the result of differences in the sites at which retention occurs
(differences in solute-stationary phase interactions) or some other phenomenon.
Provided that differences in peak shape among compounds exist, it should be possible,
in instances where two or more compounds of interest have the same retention time but it
is known that only one of them is present, to identify the unknown analyte on the basis
of its peak shape.
C O N C LU SIO N
Our goal in this project was to develop a method for the single-channel detection of
peak impurities. We also wanted an approach that was computationally simple, but
could detect impurities at a lower S/N ratios, lower resolution, and lower contamination
levels than other methods of similar complexity, i.e., slope analysis, moments, and the
distribution function. There are other methods for the detection of impure
chromatographic peaks using single-channel data (18,19), but these methods appear to
be much too complex for general use.
The Q transformation approach that we reported here appears to fulfill all of these
objectives, although a minor drawback of all of these methods including the Q approach
is the need for some type of peak shape reference. Note, however, that our approach to
the detection of co-eluting peaks requires only one injection of a reference
multicomponent solution. In most routine labs where quantitative analysis is done, a
1 4 5
standard or mixture of standards is injected at least once per day, and in many instances,
before every sample. This standardization could easily include a peak shape calibration
as well.
Although the Q transformation approach to peak shape analysis is still in its infancy
at this time, we feel that further investigation into its properties and applications is
warranted. In fact, improvements in the measurement of the graphical parameters of a
and b or in the statistical analyses of Q transform data may result in substantial
improvements in the ability to detect peak impurities. As of this writing we know of no
other peak shape analysis techniques that can characterize peaks in the way that the Q
transform does. This coupled with its computational simplicity, and its easy automation
should provide ample incentive for further investigation.
1 4 6
TABLE 6-1. Minimum Resolution Necessary for the Detection of Two Overlapping
EMG Peaks. a
First Peak 16 Times Larger Than Second Peak.
S/N
1 0 0 500 1 0 0 0
IQ .23 .13 .1 1
Slope Analysis b .36 .24
XDFM b .42 .375
Skew b b b
Excess b b b
Second Peak 16 Times Larger than First Peak.
S/N
1 0 0 500 1 0 0 0
IQ 0.14 0.07 0.04
Slope Analysis b 0.24 0.175
XDFM 0.36 0 .2 0 0.18
Skew b b b
Excess b b b
a t / a of EMG peaks was 1.0.
b Impurity not detectable by this method at specified Signal-to-Noise before a shoulder
appears on the peak.
1 4 7
Table 6-2. Simulated Impure peaks detectable by the paired t-test and the Q profile.
Rs Ara 10 0
S/N
500 1 0 0 0
0.125 1/16 no no yes
0.1875 16/1 no no yes
0.1875 1/16 no yes yes
0.25 16/1 no yes yes
0.25 1/16 yes yes yes
0.3125 16/1 yes yes yes
a Area ratio of first peak to the second peak
1 4 8
Table 6-3. Impure experimental HPLC peaks detected via the paired t-test using the Q
profile.
Peak combination (Pure peak/impurity)
Rs Ar S/N Mobile phase
Valerophenone/Acetophenone 0 .1 16/1 2 0 0 0 Isocratic, 50% ACN
Valerophenone/Acetophenone 0 .2 16/1 2 0 0 0 Isocratic, 50% ACN
n-propyl benzene/ toluene 0 .1 2 0 /1 2 0 0 0 Isocratic, 70% MeOH
n-propyl benzene/ toluene 0 .1 2 0 /1 1 0 0 Isocratic, 60% ACN
isopropyl benzene/ethylbenzene 0.15 14/1 1 0 0 Isocratic, 60% MeOH
n-propyl benzene/n-propyl benzene a 0 .2 2 0 /1 1 0 0 Gradient, 5% ACN/min
n-butyl benzene/ethylbenzene a 0 .1 16/1 2 0 0 0 Gradient, 5%
MeOH/min
a Peaks obtained via mathematical addition of single peaks as described in the
Experimental Section.
1 4 9
Table 6-4. Comparison of some pure peaks via the paired t-test.
peaks compared mobile phase t value calculated S/N
1 & 2 Isocratic, 60% ACN 0.981 1 0 0
1&3 Isocratic, 60% ACN 1.488 1 0 0
2&3 Isocratic, 60% ACN 0.923 1 0 0
1 & 2 Isocratic, 70% MeOH 0.082 2 0 0 0
1&3 Isocratic, 70% MeOH 0 .1 0 2 0 0 0
2&3 Isocratic, 70% MeOH 0 .2 2 0 0 0
a Ethyl benzene peaks injected sequentially under given mobile phase conditions,
b Tabulated t-value for 99% confidence was 2.576. Curves are not significantly
different if calculated t is less than the tabulated value.
1 5 0
0.4-I
0.3
0.2
0.0
PURE PEAK-
IMPURITY
IMPURE PEAK (PURE PEAK + IMPURITY)
7.0 9.0 11.0 13.0 15.0 17.0
TIME (MINUTES)
Figure 6-1. Effect of co-elution on overall peak profile. Impurity elutes on the tail of the
parent peak. Conditions: EMG peaks (x/cr = 1); area ratio = 16:1; resolution
= 025. Note that no shoulder or other visible evidence is apparent on the
overall profile.
151
2.5
Q 2.0
IMPURE PEAK
1.5
PURE PEAK
1.080 10020 40 600
r (%)
Figure 6-2. Effect of co-elution on the Q transformation in which the impurity elutes on
the tail of the parent peak. Conditions: EMG peaks (x/o =1); area ratio =
16:1; resolution = 0.25.
SIG
NA
L
1 5 2
0 .41
0.3
IMPURE PEAK PURE PEAK + IMPURITY0.2
IMPURITYPURE PEAK
0.07.0 9.0 11.0 13.0 15.0 17.0
TIME (MINUTES)
2.5
2.0
PURE PEAK
IMPURE PEAK
1.00 20 40 60 100
r (%)
Figure 6-3. Effect of co-elution on (a) overall peak profile; and (b) Q transformation in
which the impurity elutes on the front of the parent peak. Other conditions as
in Figure 6-2.
1 5 3
IMPURE n-PROPYL BENZENE
PURE n-PROPYL BENZENE
Figure 6-4. Impure and pure n-propyl benzene peaks. Visual inspection of original
peaks fails to indicate presence of impurity.
1 5 4
170 i
160A
1
150
IQ140
130
2 .
8
(1) 4-Methyl Cyclohexanone(2) 3-Heptanone(3) 4-Heptanone(4) 5-Nonanone
120
110
1000.0 0.1
> I " " '
0.2' i '0.3
"r ~ T ~ r
0.4> i ■0.5 0.6
LOG t:R
Figure 6-5. Example of clustering as a function of peak identity for a series of ketones.
1 5 5
HOMOLOGOUS SERIES
DIFFERENT COMPOUNDS, SAME t'R
LOG t'R
Figure 6 -6 . Illustration, using simulated data, of Q transformation approach to
qualitative analysis.
LITERATURE CITED
1 5 6
1) Papas, A. N. CRC Crit. Rev. Anal. Chem., 1989, 20, 359-404.
2) Schieffer, G.W. J. Chromatogr., 1985, 319, 387-391.
3) Grushka, Eli; Monacelli, G. C. Anal. Chem., 1972, 44, 484-489.
4) Grushka, Eli. Anal. Chem., 1972, 44, 1733-1738.
5) Grushka, Eli.; Myers, M. N.; Giddings, J. C. Anal. Chem., 1970, 42, 21-26.
6) Rix, H. Anal. Chim. Acta, 1986, 191, 467-472
7) Rix, H. J. Chrom., 1981, 204, 163-165
8 ) Excoffler, J.-L.; Jaulmes, A.; Vidal-Madjar, C.; Guiochon, G. Anal. Chem.,
1982, 54, 1941-1947
9) Chesler, S. N.; Cram, S. P. Anal. Chem., 1971, 43, 1922-1933.
10) Petitclerc, T.; Guiochon, G. J. Chromatogr. Sci., 1976, 14, 531-535.
11) Eikens, David I.; Carr, Peter W. Anal Chem., 1989, 61, 1058-1062.
12) Jeansonne, Mark S.; Foley, Joe P. J. Chromatogr., 1989, 461, 149-163.
13) E. Grushka, M.N. Myers, and P.D. Schettler, Anal. Chem., 1969, 41, 889-892.
14) Jeansonne, Mark S.; Foley, Joe P. Manuscript submitted to Anal. Chem.
15) Snyder, L.R.; Kirkland, J.J., Introduction to Modern Liquid Chromatography,
2nd ed.; Wiley: 1979; Chapter 7
16) Miller, J.C.; Miller, J.N., Statistics For Analytical Chemistry, 2nd ed.; Horwood:
1988; Chapters 3 and 6 .
17) Harris, D.C., Quantitative Chemical Analysis; W.H. Freeman: 1982; Chapter 4
18) Sybrandt, L.B. ; Perone, S.P., Anal Chem, 1972, 44, 2331-2339
19) Pichler, M.A.; Perone, S.P., Anal. Chem., 1974, 46, 1790-1798
Chapter 7
SUM M ARY
1 5 7
1 5 8
As mentioned in the Introduction, the field of chromatography has advanced greatly
since its inception. Of particularly fast growth has been the use of computerized data
analysis. In fact, computer data analysis has now become a routine part of
chromatography, due, in part, to the computer’s ability to process large amounts of data
quickly and efficiently. Quicker and more efficient processing of data thus allows more
information to be obtained than would have been feasible without the computer. The
availabilty of more information should result in more productivity, because trial and error
techniques become less necessary. Enhancing productivity is a major goal of computer
data analysis.
Recently, the trend in chromatographic automation has teen the development of
expert systems. It is believed that these systems will allow those with only limited
chromatographic background to do more difficult and complex types of analyses.
However, the expert system needs large amounts of information in order to make its
decisions. In the opinion of this author, chromatographic data analysis techniques like
those presented in the preceding chapters are necessary to supply the needed information.
Of particular importance will be peak shape information. Because chromatographic
peak shapes in chromatography are somewhat unpredictable, yet reveal much about the
processes occurring within the column, efficient methods of peak shape measurement
and comparison are necessary. Therefore, the information presented in Chapters 2 - 4
should enable more accurate measurement of important chromatographic parameters via
use of the EMG model because it describes real chromatographic peaks tetter than the
frequently used Gaussian model. Although methods more mathematically rigorous than
those presented in chapters 2-4 may be available, their use for routine data analysis is
often limited by their complexity and time consuming calculations. The easier and more
1 5 9
intuitive approaches presented in those chapters will encourage greater use of the more
accurate EMG model for routine chromatographic analyses.
Similarly, the Q transformation, discussed in Chapters 5 and 6 should encourage
more routine use of peak shape comparison with single-channel data in order to extract
more information during routine analyses. Again, more sophisticated techniques are
available, but their complexity often discourages their routine use. Also, because the Q
transformation is better than other methods of similar complexity for comparing peak
shapes, it should be the technique of choice when peak shape comparison is necessary.
Increasing the information available and shortening analysis times through the use
of the techniques presented here will play a major role in the future development of
chromatography. The importance of these techniques cannot be overemphasized.
Appendix A
Q TRANSFORM ATION PROGRAM DOCUM ENTATION AND
LISTIN G
1 6 0
161
Q T r a n s f o r m — O V E R V I E W
Q TRANSFORM is a program that can be used for peak shape analysis. Q is measured at fractions of the peak height (r), between 2 and 98% in increments of 1%. The total or £Q over an entire peak gives a single number that can indicate peak shape. This program will take a user specified data file and topographical file and output a Q file in ASCII format. The Q file output contains (i) the value of Q measured at peak height fractions between 2 and 98%; and (ii) XQ, the sum of these Q values that indicates peak shape in a single number. The interpretation of the Q output data which permits the detection of co-eluting peaks is described in the accompanying manuscript preprint "Single Channel Peak Impurity Detection...".
When Q TRANSFORM is started, the user will be presented with a choice of whether to "CONFIGURE", "RUN Q" or "STOP". Choosing "CONFIGURE" enables the user to set up the RUNDATA.TXT file (described below), while choosing "RUN Q" will start the calculations for the chromatogram files that have been listed in RUNFILES.TXT (see INPUT section below). To stop the program after choosing "RUN Q", simply press uppercase Q on the keyboard. It may take a few seconds, but the program will eventually quit
The file SAMPLEDATA. 1 included on your disk is a simulated chromatogram that is used for the sample chromatogram and topographical files discussed below. You can run Q TRANSFORM on this data in order to verify that the program is working. The resulting data will be printed at the end of the Q output file already present on the disk. Any time Q TRANSFORM is run on a data file for which a Q output file already exists, the ouput data will simply be added to the end of the existing Q ouput file.
I N P U T
The Q TRANSFORM program is designed to allow users to input their own chromatograms or other peak shaped data. Two ASCII (text) files are required for each chromatogram:1. User-named raw data file, with the extension ".txt". Example: "Sampledata.l.txt"2. User-named topographical file, with extension ".top". Example:
"Sampledata. l.top"
1 6 2
Two other ASCII files are required each time Q Transform is run.3. "Runfiles.txt"4. "Rundata.txt."Note that for a given run of Q Transform, all files must have the same filepath (be in the same folder).
The first three files should be set up by the user before running Q Transform. The last file is set up automatically while running Q Transform by using the CONFIGURE routine. The purpose and format of each of these files is given below. Two sets of raw data and topographical files are included on the demonstration disk: Sampledata.l.txt, Sampledata. 1.top, Sampledata.2.txt, and Sampledata.2.top
"Runfiles.txt" is an ASCII (text) file that contains the name(s) of the data file(s) on which the Q tranform will be run. Open this file using any word processing program (Word, MacWrite, etc.) to verify the file name(s) BEFORE running Q Transform.
"Rundata.txt" is an ASCII (text) file that contains user-specified parameters for the subsequent running of Q Transform. This file is generated from within Q Transform by running the CONFIGURE routine. The next section describes this in detail.
R U N D A TA .TX T FILE
This file is read by the Q TRANSFORM program when it is run. When run, a menu is presented that allows the user the option of running Q TRANSFORM or configuring the Rundata.txt file. Rundata.txt is in ASCII format so that the user can modify the contents using his own programs instead of configuring it from the CONFIGURE option in Q TRANSFORM. Each line in the file gives a parameter that Q TRANSFORM uses. There can be no extra lines or blank lines in the file.
Line 1 in this file consists of the path to the data files to be used. For example, to examine a file on the Q TRANSFORM disk not in any subdirectories or folders the path would simply be Q TRANSFORM: for Macintosh computers or Q TRANSFORM\ for IBM personal computers.
Line 2 gives the smoothing factor used by the program. There must be a number here. There cannot be a blank line or non-numerical character. A smoothing factor of 0 will tell Q TRANSFORM not to smooth the data.
1 6 3
Unless a smoothing factor of 0 is input by the used, the Q TRANSFORM will perform a weighted moving average quadratic smooth on the data as described by Savitsky and Golay L The smoothing factor is defined as the number of points in the moving window divided by the number of points in the peak from half-height on the front of the peak to half-height on the back of the peak. For example, a peak that has 25 points from half-height to half-height would have a 25 point moving average window applied to it if the smoothing factor was 1 , while a peak with 11 points from half-height to half-height would have an 11 point moving average window applied to it with a smoothing factor of 1. Therefore, peaks that do not have the same number of points can still be smoothed equally. Moving windows of between 5 and 25 points are available with the program and if a smoothing factor is used that indicates more than 25 points should be used then 25 points will be used. Likewise, if the smoothing factor indicates that less than 5 points should be used in the moving window, then 5 will be used, unless the smoothing factor is 0 .
Line 3 is a "Y" or "N" depending on whether the user would like each individual peak to be plotted on the screen.
Line 4 is again a "Y" or "N" depending on whether the user would like a bell to ring after each chromatogram is finshed.
Note: When using the CONFIGURE option from Q TRANSFORM, the answer in square brackets given after each question is the default answer. Simply pressing the return key will enter this answer.
DATA FILE FORMAT
The name of a chromatogram data file to be used with this program is always followed by the ".TXT" extension. Case does not matter.
The file must be in ASCII format. These files consist of one number per line with no leading or trailing spaces on each line. The numbers are the signal heights for the chromatographic data. The sampling interval or times at which each point occurs is included within the topographical file, described below. Due to the type of smoothing
1 A. Savitsky and M. J. E. Golay, "Smoothing and Differentiation of Data by Simplified Least Squares Procedures", Analytical Chemistry 36,1627 (1964)
1 6 4
employed by this program, a minimum of 16 data points must be present in this file before the first peak's data actually starts, while 16 points must also be included after the data for the last peak in the chromatogram ends. These "extra" points may be added as dummy points, set to 0 , if needed.
TOPOGRAPHICAL FILE FORMAT
The name for the topographical file corresponding to a data file is the same as the root name for the data file with the extension ".TOP" instead of ".TXT". Again, case does not matter.
The topographical files to be used are in ASCII format Each line contains 3 comma delimited numbers. The first number in each line is a code number that tells Q TRANSFORM what the second and third numbers refer to.
Sam ple Topographical File
o, 0 .0000, 3.000006 , 2 .0 000 , 1.000001, 2.0050, 3.000002 , 2.7310, 3.076595, 3.1150, 14.83301, 5.0350, 3.001087, 54.000, 7.000002 , 5.6320, 3.085055, 6.1440, 8.640751, 8 .0210 , 3.000827, 56.000, 4.000002 , 8.4480, 3.066515, 9.1310, 12.37851, 11.264, 3.000607, 6 6 .0 00 , 6.00000
CODE 0
All topographical files begin this way. The second number in the line refers to the chromatogram start time in minutes, e.g., the time at which the first point in the chromatogram occurs. Most of the time this number is 0. The third number in this line is
1 6 5
a code referring to the sample interval for the chromatogram. Note: the sample interval must be constant for the entire chromatogram in order for this version of Q TRANSFORM to run correctly. The following chart explains the sample interval codes.
Sam pling Code Sampling interval(S econ d s)
3 2.564 1.285 0.646 0.327 0.168 0.089 0.0410 0 .0211 0.0112 0.313 0.214 0.115 0.05
CODE 6
This line should always be the same as shown in the sample topographical file.
CODE 1
This line tells Q TRANSFORM where a basepoint occurs. The program draws straight lines between basepoints, so that the baseline level can be subtracted from each peak before Q is calculated at each value of r. The second number is the time (in minutes) at which the signal level is "on baseline" and the third number is the signal height at that time. The sample chromatogram in Fig. 1 below shows the 4 basepoints that are given in the sample topographical file above.
1 6 6
4.0
<zo
3.0(NJ n
Q.LUcn
CL L1J in
CL LU tn
LU00
mCD
8.0 10.0 12.00*0 4.0 6.02*0
T I M E ( M I N )
F ig . 1
CODE 2
Indicates a peak start. The second number in the line indicates time of peak start while the third number indicates the height of the signal at the peak start Can also indicate a peak end(peak start) if a valley occurrs between two peaks and the signal level is not yet at the baseline level. Normally, a code of 1 will indicate a peak end if two peaks are well separated.
CODE 5
Indicates peak apex. The second number is the retention time of the peak while the third number is the signal height at that retention time.
CODE 7
1 6 7
Second number indicates the number of points in the peak and the third number indicates the area of the peak. This code always comes after the peak end that this line applies to whether the peak end code is indicated by a 1 or 2 .
OTHER CONSIDERATIONS
1) There must not be any extra blank lines in the topographical file, either before the first line or between lines.
2) There must not be a line feed and/or carriage return after the last number in the last line.
O U T P U T
Q O UTPUT FILE FORM AT
The name for the Q file that is output is derived from the name of the data file used by the program. For example, the file Sampledata.l.txt results in an output file named Sampledata.l.Q.
A sample Q output file is included on the Q TRANSFORM disk. You can open this file with any text editing program. The first line gives information about whether the data was smoothed prior to calculating Q .what the smoothing factor was and the date and time the Q analysis was done. The time given is in ANS standard format "hh:mm:ss", where hours are measured by a 24 hour clock (midnight is 00:00:00). This format allows one to do multiple Q TRANSFORM runs on the same data file and store all runs in the same .Q file with each subsequent run stored after prior runs. More about the smoothing factor is given later. The second line in the file consists of the column names. In the sample file (Sampledata. l.Q), you will see a column for r, Q l, Q2, and Q3. These columns correspond to columns for the r values used, and the Q value at each r for each peak in the chromatogram. In this case there are 3 peaks, so 3 Q columns are seen, one for each peak. The last line in the file shows the XQ value for each peak. Columns are tab delimited for this file making it easy for the user to open the file directly into most popular spreadsheet and graphing programs.
1 6 8
Note: If a valley between two peaks is high enough then Q TRANSFORM will not be able to calculate Q for every r. If there are zero's for Q for a peak at the lower r values this indicates that two peaks were not resolved enough to calculate Q at these lower r values. The file Sampledata.2.Q illustrates this for two partially overlapping peaks.
USE OF Q OUTPUT
The preprint of an article describing the use of the Q TRANSFORM data for the detection of impure peaks in chromatography is included with the demo disk. The data from the ”.Q” file and the ".TOP" file are needed for the methods given in the article.
TIPS ON USING Q TRANSFORM
1) The same smoothing factor should be used for both the reference peak(s) and sample peak(s).
2) For best results, there should be at least 30 points from half-height to half-height in the peaks used. The signal-to-noise (defined as the height of peak divided by the root-mean-square noise in the baseline) should be greater than 2 0 0 for best results. We have used this program down to signal-to-noise values of less than 1 0 0 although the ability to detect impurities decreases as the signal-to-noise decreases.
3) When running Q TRANSFORM for the first time, set the "plot peak on screen" option in RUNDATA.TXT to "Y" in order to make sure that the program is reading in peaks correcdy. Be aware that 16 extra points will be plotted before and after each peak. These extra points are used in the smoothing routine but not in the actual calculations of Q.
4) Sample peak(s) and reference peak(s) must be obtained under the same chromatographic conditions because different conditions can introduce differences in peak shape.
1 6 9
RUNNING Q TRANSFORM
Two versions of Q TRANSFORM have been written—one for the Macintosh Personal Computer and one for the IBM Personal Computer. Please refer to individual owner guides for information on how to start an application program.
Once Q TRANSFORM has started, the user will be presented with a choice of whether to "CONFIGURE", "RUN Q" or "STOP". Choosing "CONFIGURE" enables the user to set up the RUNDATA.TXT file, while choosing "RUN Q" will start the calculations for the chromatogram files that have been listed in RUNFILES.TXT. If the user wishes to stop the program after choosing "RUN Q" simply press uppercase Q on the keyboard. It may take a few seconds, but the program will eventually quit
Q TRANSFORM ATION PROGRAM4c 4c * * 4c * * * * * * * 4c 4c >|c * * * * 4c * * * * * * * 4c 4c * # 4c * * 4c * * * * * * * * 4> * * * * * * * * * * * * * * * *
! Copyright 1990! by! Mark S. Jeansonne and Joe P. Foley! Department of Chemistry, 232 Choppin Hall! Louisiana State UniversityI Baton Rouge, LA 70803i
LIBRARY "Pictlib*"DIM M$(3)LET COL$='"'OPEN #2:SCREEN 0,.01,0,.01 DOCALL STARTUP CLOSE #9 RESTORE LOOPSUB STARTUP
CLEARCALL set_Frame(.l,.8,.3,.8)
CALL draw_pictfile("Copyright notice",0)CALL MENU_SET("TOP",COL$,3,10,MENU$,#9)MAT READ M$DATA CONFIGURE,RUN Q.STOPCALL MENU_ALL(M$,3,""ANS,MENU$,#2,#9)
LETT1=TIME SELECT CASE ANS CASE 1
CLOSE #2OPEN #2:SCREEN 0,1,0,1 CALL CONFIG CLOSE #9 RESTORE CALL STARTUP
CASE 2OPEN #l:Name "runfiles.txt",access input,organization text DO while more #1
INPUT #l:file$LET a$=E$&file$CALL Qtransform3Mult(a$)
LOOP CASE 3
STOP CASE ELSE
STOP END SELECT
END SUB LETT2=TIME
171
END
EXTERNALSUB Qtransform3Mult(FILE$)
LIBRARY "sgsmooth*"CLEAR
DIM TIM(3000), TEMPTIM(3000), RT(50), RTHEIGHT(50), PSTART(50), PEND(50),TPEND(50)
DIM A(50,100), B 1(50,100), PEAK(3000), TEMPPEAK(3000), temp(50)DIM NUMSAMP(3000), TEMP_HALF_HEIGHT(50), START(50),PTEND(50),
point(50)DIM X(50,2), Y(50,l), Xt(2,50), XtX(2,2),XtY(2,l), XtXi(3,3), B(2,l)DIM Xraw(4,l), Yraw(4,l),TA(100),TB(100)DIM AREA 1(1),AREA2( 1 )rAREA(8 )JHEIGHT( 100)DIM PERCENTHT(8 )DIM TPSTART(50),TBSPOINT(50),BSPOINT(50),BSLNSLP(50)DIM XC(60,3), YC(60,1), XTC(3,60), XTXC(3,3), XTYC(3,1), XTXIC(3,3),
BC(3,1)DIM XRAWC(60,1 ),YRAWC(60,1 ),H 1 $(20),H2(50),H3(50),H4(50)DIM H5(50),H6(50),H7(50),H8(50)DIM changeina(100),changeinb(100)DIM Q(50,100),SUMQ(50)OPEN #4: Name "rundata.txt".access input,organization text LINE INPUT #4 :5 LET f$=trim$(f$)LET INFO$=HLE$OPEN #1 : screen 0, 1, .8 , 1IINPUT prompt "Name of data file ? ":file$LET file$=f$&file$! INPUT #4:answer$INPUT #4:smoothfact IINPUT #4:Threshold 1INPUT #4:Start_Q IINPUT #4 :End_Q IINPUT #4:a6$INPUT #4:a7$INPUT #4:a8$INPUT #4:all$CALL CONTINU SUB CONTINU
IF KEY INPUT THEN GET KEY KEY IF KEY=81 THEN STOP
END IF END SUB
OPEN #7:NAME FILE$&".Q",CREATE NEWOLD, ACCESS OUTPUT,ORGANIZATION TEXT !******
SET #7:POINTER END SET #7:MARGIN 2000j****************************************************************
CLEAROPEN #2:name file$&".txt",access INput,ORGANIZATION textOPEN #3 : screen 0, 1,0, .8WINDOW #3OPEN #5:PRINTERSET MARGIN 70PRINT "NOW READING IN TOPOGRAPHICAL DATA FROM ";INFO$ PRINTOPEN #6 :NAME file$&".TOP",ORGANIZATION TEXT ACCESS INPUT LET NUMPEAK=0 LET K=1DO WHILE MORE # 6
WHEN error inINPUT #6:H4(k),H6(k),H8(K)
USELET h4(K)=20
END WHEN LETK=K+1
LOOPFORL=l TO K-l
IF H4(L)=6 THEN LET BSCODE=H6 (L) -
END IF
IF H4(L)=0 THEN LET RUNTIME=H6 (L)SELECT CASE H8 (L)CASE 3
LET DT=2.56/60 CASE 4
LET DT=1.28/60 CASE 5
LET DT=.64/60 CASE 6
LET DT=.32/60 CASE 7
LET DT=.l 6/60 CASE 8
LET DT=.08/60 CASE 9
LET DT=.04/60 CASE 10
LET DT=.02/60 CASE 11
LET DT=.01/60 CASE 12
LET DT=0.3/60 CASE 13
LET DT=0.2/60 CASE 14
LET DT=0.1/60 CASE 15
1 7 3
LET DT=0.05/60 CASE 16
LET DT=0.001/60 CASE 17
LET DT= 15/60 CASE 18
LET DT=4/60 CASE 19
LET DT=8/60 CASE 20
LET DT= 12/60 CASE 21
LET DT=6/60 CASE 22
LET DT= 10/60 CASE 23
LET DT=18/60 CASE 24
LET DT=20/60 CASE 25
LET DT=9/60 CASE 26
LET DT=11/60 END SELECT
END IFIF H4(L)=1 THEN
LET NUMBSPT=NUMBSPT +1 LET TB SPOINT(NUMB SPT)=H6 (L)LET BSPOINT(NUMBSPT)=H8 (L)
END IF NEXT LFOR L=1 TO K-l
IF H4(L)=5 THEN IF h8 (L)>=threshold then
LET NUMPE AK=NUMPE AK+1 LET PEAKNUM=NUMPEAK LET RTHEIGHT(PEAKNUM)=H8 (L) LET RT(PEAKNUM)=H6 (L)IF H4(L-1)=2 OR H4(L-1)=1 THEN
LET TPSTART(PEAKNUM)=H6 (L-1) LET PS TART (PE AKNU M )=H 8 (L-1)
ELSEIF H4(L-2)=2 OR H4(L-2)=1 THEN LET TPSTART(PEAKNUM)=H6(L-2) LET PSTART(PEAKNUM)=H8(L-2)
END IFIF H4(L+1)=7 THEN
LET NUM S AMP(PE AKNUM)=H6 (L+1) ELSEIF H4(L+2)=7 THEN
LET NUMSAMP(PEAKNUM)=H6(L+2) END IF
END IF END IF
174
NEXT LCALL CONTINU LET J=1LET T=RUNTIME LET K=1 LET FLAG=0
LETT1=TIME FOR PEAKNUM=1 TO NUMPEAK
LET START(PEAKNUM)=round((TPSTART(PEAKNUM)-T-16*DT)/DT) LET
PTEND(PEAKNUM)=round(START(PEAKNUM)+NUMSAMP(PEAKNUM)+32) NEXT PEAKNUMPRINT "NOW READING IN PEAK DATA FROM ";INFO$PRINTDO WHILE MORE #2
LINE INPUT #2: RAW$FOR kk=l to numbspt
IF J>round((tbspoint(kk)-T)/dt)-3 and J<round((tbspoint(kk)-T)/dt)+3 then LET temp(kk)=temp(kk)+val(raw$)! PRINT val(raw$)
END IF NEXT kk LET J=J+1
LOOP
FOR kk=l to numbspt LET bspoint(kk)=temp(kk)/5
NEXT kk
LET HOLDJ=l RESET #2:BEGIN
FOR PEAKNUM=1 TO NUMPEAKMATTA=0MATTB=0
MAT PEAK=TEMPPEAK MAT TIM=TEMPTIM LET J=HOLDJ DOLINE INPUT #2: RAW$
IF J > START(PEAKNUM) then IF J < PTEND(PEAKNUM) THEN
LET JJ=J-START(PEAKNUM)+ADDON LET PEAK( J J)=V AL(RA W$)LET tim(jj)=(jj-l)*dt+tpstart(peaknum)-16*dt
END IF END IFIF J>START(PEAKNUM+1) AND PEAKNUMoNUMPEAK THEN
LET II=J-START(PEAKNUM+1)LET TEMPPEAK(II)=V AL(RA W $)
1 7 5
LET TEMPTTM(II)=(II-1 )*DT+TPSTART(PEAKNUM+1)-16*DT END IFCALL CONTINU LET J=J+1 LET HOLDJ=J
LOOP until J> PTEND(PEAKNUM)LET T2=TTME
CALL CONTINU CALL initialize CLEARCALL CONTINU CALL smooth CALL CONTINU CALL BSLN.SUBTRACT CALL CONTINU CALLMAXFIT CALL CONTINU CALL QCALC CALL CONTINU CALL CONTINU
NEXT PEAKNUM CALL PRINTQ
IF al 1$="Y" then CALL ring_bell
SUBMAXFIT LET XMEAN=0 LET YMEAN=0 LET XSUM=0 LET YSUM=0 MAT XRAWC=0 MAT YRAWC=0 MAT XC=0 MAT YC=0 MAT XTC=0 MAT XTXC=0 MATXTYC=0 MAT XTXIC=0 MATBC=0 LET N=7LET JJ=(RT(PEAKNUM)-tpstart(peaknum))/DT+1+16 FOR 1=1 TO N
LET XRAWC(l,l)=tim((JJ-ROUND(N/2)+I))LET YRAWC(1,1 )=PEAK((J J-ROUND(N/2)+I))1PRINT USING .#########": XRAWC(I,1),YRAWC(I,1)
NEXT I FOR 1=1 TO N
LET XSUM=XSUM+XRAWC(I,1)LET YSUM=YSUM+YRAWC(I,1)
NEXT ILET XMEAN=XSUM/N LET YMEAN=Y SUM/N
1 7 6
FOR 1=1 TO N LET XC(I,1)=1LET XC(I,2)=XRA W C(1,1 )-XMEAN LET XC(I,3)=XC(I»2)A2
LET Y C(1,1 )=YRA WC(1,1)-YMEAN NEXT IMAT XTC=TRN(XC)! MAT PRINT XTC;MAT XTXC=XTC*XC MAT XTXIC=INV(XTXC)1PRINT "DET(XTXC)=";DETMAT XTYC=XTC*YC!MAT PRINT XTYCMAT BC=XTXIC*XTYCLET TEMPXMAX=-BC(2,1 )/(2*B C(3,1))LET TEMPYMAX=BC( 1,1 )+BC(2,1 )*TEMPXMAX+BC(3,1 )*TEMPXMAXA2 LET XMAX=TEMPXMAX+XMEAN LET YMAX=TEMPYMAX+YMEAN LET RTHEIGHT(PEAKNUM)=YMAX LET RT(PEAKNUM)=XMAX
END SUB
SUB initialize FOR k=l to (numbspt-1)
IF rt(peaknum)>tbspoint(k) and rt(peaknum)<tbspoint(k+l) then LET point(peaknum)=kLET bslnslp(peaknum)=(bspoint(k+1 )-bspoint(k))/(tbspoint(k+1 )-tbspoint(k))
END IF NEXT kLETTPEND(PEAKNUM)=TPSTART(PEAKNUM)+(NUMSAMP(PEAKNUM)-
1)*DTLETBSLN=BSLNSLP(peaknum)*(RT(PEAKNUM)-
tbspoint(point(peaknum)))+bspoint(point(peaknum))LETTEMP_HALF_HEIGHT(PEAKNUM)=.5*(RTHEIGHT(PEAKNUM)-
BSLN)END SUB
SUB smooth LET N=17FOR I=N to (N+Numsamp(peaknum))
LET bsln=bslnslp(peaknum)*(tim(i)- tbspoint(point(peaknum)))+bspoint(point(peaknum))
LET temp_pointl=peak(I)-bsln LET bsln=bslnslp(peaknum)*(tim(i-1 )-
tbspoint(point(peaknum)))+bspoint(point(peaknum))LET temp_point2=peak(I-1 )-bslnIF tim(I)<Rt(peaknum) and temp_pointl >TEMP_HALF_HEIGHT(PEAKNUM)
and temp_point2<TEMP_HALF_HEIGHT(PEAKNUM) then LET tl=tim(i)
END IFIF tim(I)>rt(peaknum) and temp_point 1 <TEMP_HALF_HEIGHT(PEAKNUM)
and temp_point2>TEMP_HALF_HEIGHT (PEAKNUM) then
177
LET t2=tim(i)END IF
NEXT iLET ptshalfHt=round((t2-tl)/dt)LET hold=int(SMOOTHFACT*ptshalfht)IF mod(hold,2)oO then
LET smoothwidth=int(hold)ELSE
LET smoothwidth=int(hold-l)END IFIF SMOOTHWIDTH<=0 THEN
LET SMOOTHWIDTH=0 ELSEIF S MOOTHWIDTH>0 AND SMOOTHWIDTH<5 THEN
LET SMOOTHWIDTH=5 ELSEIF SMOOTHWIDTH>25 THEN
LET SMOOTHWIDTH=25 END IF PRINTIF SMOOTHWIDTHoO THENPRINT "SMOOTHING PEAK # ";STR$(PEAKNUM)PRINTPRINT "SMOOTHING FACTOR = ";SMOOTHFACT PRINTPRINT "NUMBER OF POINTS IN SMOOTH = ";SMOOTHWIDTH
SELECT CASE smooth width CASE 5
CALL sgsmooth5(PEAK,(NUMSAMP(PEAKNUM)+32)) CASE 7
CALL sgsmooth7(PEAK,(NUMSAMP(PEAKNUM)+32)) CASE 9
CALL sgsmooth9(PEAK,(NUMSAMP(PEAKNUM)+32)) CASE 11
CALL sgsmooth 11 (PEAK,(NUMSAMP(PEAKNUM)+32)) CASH 13
CALL sgsmooth 13(PEAK,(NUMS AMP(PEAKNUM)+32)) CASE 15
CALL sgsmooth 15 (PEAK,(NUMS AMP(PEAKNUM)+32)) CASE 17
CALL sgsmooth 17(PEAK,(NUMSAMP(PEAKNUM)+32)) CASE 19
CALL sgsmooth 19(PEAK,(NUMSAMP(PEAKNUM)+32)) CASE 21
CALL sgsmooth21(PEAK,(NUMSAMP(PEAKNUM)+32)) CASE 23
CALL sgsmooth23(PEAK,(NUMSAMP(PEAKNUM)+32)) CASE 25
CALL sgsmooth25(PEAK,(NUMSAMP(PEAKNUM)+32)) END SELECT ELSEPRINT "NO SMOOTH APPLIED TO THIS DATA”
END IFFOR 1=1 TO NUMSAMP(PEAKNUM)+32
1 7 8
IF TIM(I)=TBSPOINT(POINT(PEAKNUM)) THEN LET BSPOINT(POINT(PEAKNUM))=PEAK(I)
END IFIF TIM(I)=TBSP0INT(P0INT(PEAKNUM)+1) THEN
LET B SPOINT (POINT (PEAKNUM)+1 )=PE AK(I)END IF
NEXT I END SUB
SUB BSLN_SUBTRACT LET MAX=0LET BSLNSLP(PEAKNUM)=(BSPOINT(POINT(PEAKNUM)+l)-
B S POINT (POINT (PE AKNUM)) )/(TB S POINT (POINT (PEAKNUM)+1 )- TBSPOINT(POINT(PEAKNUM)))
LET N=17FOR I=N TO (NUMSAMP(PEAKNUM)+N-1)
LET BSLN=(BSLNSLP(peaknum)*((tim(i))- tbspoint(point(peaknum)))+bspoint(point(peaknum)))
LET PEAK(I)=PEAK(I)-BSLN !PRINT #5:I,PEAK(I,PEAKNUM)IF PEAK(I)>MAX THEN
LET MAX=PEAK(I)- LET RT(PEAKNUM)=tim(i)
END IF NEXT ILET PEND(PEAKNUM)=peak((16+numsamp(peaknum)))LET PSTART(PEAKNUM)=peak(17)IF a7$="Y" then
CLEAR SET WINDOW
TB SPOINT (POINT(PEAKNUM)),TB SPOINT (POINT (peaknum)+1 1 *max,MAX FOR 1=17 TO (NUMS AMP(PEAKNUM)+16)
PLOT LINES: tim(i),PEAK(I);tim(i-l), Peak(I-l)NEXT I PLOT
LINES:TBSPOINT(POINT(PEAKNUM)),0;TBSPOINT(POINT(PEAKNUM)+1),0 PAUSE 5
END IF END SUB
SUB QCALC CLEARSET WINDOW 0,1,0,1PRINT "CALCULATING Q FOR PEAK # ”;STR$(peaknum)PRINTFOR KK=1 TO 99
!***DETERMINE SIGNAL HEIGHT AT WHICH PEAK FRACTIONS OCCUR***
LET HEIGHT(KK)=(KK/100)*(RTHEIGHT(PEAKNUM))NEXT KKLET complete=0LET increment=.5102041
1 7 9
PLOT TEXT, AT .27,.8:"% complete"LET place=98 LET COUNT2=98LET HOLD=RTHEIGHT(PEAKNUM)LET N=17 LET tempO l=le200 LET temp02=le200
LET PLACE=1 LET PLACE2=98
FOR I=N TO (N+NUMSAMP(PEAKNUM))IF RTHEIGHT(PEAKNUM)>THRESHOLD THEN
IF TIM(I)<RT(PEAKNUM) THEN FOR COUNT=PLACE TO 98
IF PEAK(I)>HEIGHT (COUNT) THEN IF PEAK((I-l))<HEIGHT(COUNT) THEN caU CONTINU
LET complete=complete+increment BOX CLEAR .2,.25,.7,.87
PLOT TEXT, AT .2,.8:str$(int(complete))IF COUNT=l THEN
CALL FITFACTCALC CALL Linefit
END IFIF COUNT<91 THEN
IF PEAK(I)oTEMP01 AND PEAK((I-l))oTEM P02 THEN CALL LINEFIT
END IFLETTA(COUNT)=(HEIGHT(COUNT)-B(l,l))/B(2,l)
ELSELET M=(PEAK(I)-PEAK((I- 1)))/(TIM(I)-TIM((I-1)))LET TA(COUNT)=(HEIGHT(COUNT)-PEAK((I-l))+M*TIM((I-
1)))/MEND IF
LET TEMPO 1=PEAK(I)LET TEMP02=PEAK((I-1))LET PLACE=PLACE+1•PRINT COUNT;" ";FITFACT;" ";HEIGHT(COUNT);" ”;TIM(I);"
”;TIM((I-1));" ’VTA^'iTACCOUNT)END IF
END IF NEXT COUNT
END IFIF TIM(I)>RT(PEAKNUM) THEN
FOR COUNT2=PLACE2 TO 1 STEP -1 IF PEAK(I)<HEIGHT (COUNT2) THEN
IF PEAK((I-1 ))>HEIGHT(COUNT2) THEN CALL CONTINU
LET complete=complete+increment BOX CLEAR .2,.25,.7,.87
PLOT TEXT, AT .2,.8:str$(int(complete))IF COUNT2<91 THEN
1 8 0
IF count2=90 then CALL linefitIF TEM P03oPEAK(I) AND TEM P04oPEAK((I-1)) THEN
CALL LINEFIT END IFLET TB (COUNT2)=(HEIGHT (COUNT2)-B (1,1 ))/B(2,1)
ELSELET M=(PEAK(I)-PEAK((I-1 )))/(TIM(I)-TIM(a-1)))LET TB (COUNT2)=(HEIGHT(COUNT2)-PE AK((I-
1 ))+M*TIM((I-1 )))/MEND IF
LET TEMP03=PEAK(I)LET TEMP04=PEAK((I-1))!PRINT COUNT2;" ";FITFACT;" ”;HEIGHT(COUNT2);" ,,;TIM(I);,,
”;TIM((I-1));" ";"TB=";TB(COUNT2)LET place2=place2-l
END IF END IF
NEXT COUNT2 END IF
END IF NEXT I
!DETERMINE PEAK WIDTH AND ASYMMETRY AT DESIRED PEAK HEIGHT FRACTIONS
FOR KK =1 TO 98 IF TA(KK)>0 AND TB(KK)>0 THEN
LET A(PEAKNUM,KK)=RT(PEAKNUM)-TA(KK)LET B1 (PEAKNUM,KK)=TB(KK)-RT(PEAKNUM)
ELSELET A(PEAKNUM,KK)=0 LET B 1 (PEAKNUM,KK)=0 END IF NEXT KK FOR JJ=2 TO 98IF A(PE AKNUM, J J)>0 AND A(PEAKNUM,JJ-1)>0 AND
B 1 (PEAKNUM,JJ)>0 AND B1(PEAKNUM,JJ-1)>0 THENLET changeina(ij)=a(PEAKNUM jj- l)-a(PEAKNUM jj)LET changeinb(jj)=bl (PEAKNUM ,jj-l )-bl (PEAKNUM jj)IF CHANGEDMA(J3)>0 AND CHANGEINB(JJ)>0 THEN
LET Q(PEAKNUM,JJ)=changeinB(JJ)/changeinA(jj)ELSE
LET Q(PE AKNUM, J J)=0 END IF
END IFLET sumQ(peaknum)=SumQ(peaknum)+Q(peaknumjj)
NEXT JJ
END SUB
SUB FITFACTCALC
! ** NMPTSHALFPEAK=# PTS FROM PEAK START TO PEAK APEX
LET NMPTSHALFPE AK=(RT(PEAKNUM)-TIM(I))/DT +1 LET FITFACT=ROUND (NMPTS HALFPEAK/14)IF MOD(FITFACT,2)>0 THEN LET FITFACT=FITFACT-1 IF FITFACT>10 THEN LET FITFACT-10 IF FTTFACT<2 THEN LET FITFACT=2 LET FITFACTHOLD=FITFACT
END SUB
SUB LINEFIT MAT x=0 MAT y=0 MAT xt=0 MAT xtx=0 MAT xtxi=0 MAT b=0 LET COUNTER=0FOR L=I-(FITFACT/2) TO I+(FITFACT/2-l)
LET COUNTER=COUNTER+1 LET X(COUNTER, 1 )=1 LET X(COUNTER,2)=TIM(L)LET Y (COUNTER, 1 )=PEAK(L)
NEXT LMAT XT=TRN(X)MAT XTX=XT*X MAT XTY=XT*Y MAT XTXi=inv(xtx)MAT B = XtXi*XtY
END SUB
SUB PRINTQ CLEARPRINT "PRINTING Q TO FILE ";INFO$&".Q"PRINTLET MONTH$=DATE$[5:6]LET YEAR$=DATE$[ 1:4]LET DAY$=DATE$[7:8]
PRINT #7:"SMOOTHFACT=";SMOOTHFACT;" DATE ";MONTH$;"/”;DAY$;'‘/u;YEAR$;1’ TIME: ";TIME$
PRINT #7:"r";Chr$(9);If a8 $="y" or a8 $="Y" then FOR Peaknum=l to Numpeak
PRINT #7:"Q”; STR$(peaknum); chr$(9);"b";str$(peaknum); chi$(9);"a";str$(peaknum); chr$(9);
NEXT peaknumelse
FOR Peaknum=l to Numpeak PRINT #7:"Q";STR$(peaknum);chr$(9);
NEXT peaknumend if
PRINT #7
1 8 2
If a8$="y" or a8 $="Y" then FOR jj=l to 98
PRINT #7:str$(jj);FOR peaknum=l to numpeak
PRINT #7:CHR$(9); Str$(Q(peaknumjj)); chr$(9); str$(bl(peaknumjj)); chr$(9);str$(a(peaknumjj));
NEXT peaknum PRINT #7
NEXTjj
FOR jj=2 to 98 PRINT #7:str$(jj);FOR peaknum=l to numpeak
PRINT #7:CHR$(9);Str$(Q(peaknumjj));NEXT peaknum PRINT #7
NEXTjjend if
PRINT #7 PRINT #7:"XQ";FOR peaknum=l to numpeak
PRINT #7:chr$(9);str$(sumQ(peaknum));NEXT peaknum PRINT #7
END SUB
SUB ring_bell FOR RING=1 TO 2
FOR 1=1 TO 15 SOUND 600, .03 SOUND 2000, .02
NEXT I PAUSE 1
NEXT RING END SUB
END SUB
SUB CONFIGOPEN #l:Name "Rundata.txt",create newold,access outin,organization text WHEN error in
INPUT #l:a$USEEND WHEN PRINT A$LINE INPUT prompt "Change path to data?[N]":a0$LET a0$=trim$(a0$)IF a0$="y" or a0$="Y” then
INPUT prompt "New Path:":al2$IF a 12$="" then LET al2$=a$
ELSE LET a!2$=a$
END IFLINE INPUT prompt "Input smoothing factor:[0.3] ":a2$LET a2$=trim$(a2$)IF a2$="" then LET a2$="0.3"LINE INPUT prompt "Plot individual peaks on the screen?[N] ":a7$
LET a7$=trim$(a7$)IF a7$="" then LET a7$="N"Line Input prompt "Print a and b to data file?[N] ":a8 $ let a8 $=trim$(a8$)If a8 $="" then let a8 $="N"LINE INPUT prompt "Ring bell when program is finished?[Y] ":al 1$ LET all$=trim $(all$)IF al 1$="" then LET al 1$="Y" f r a s f #1PRINT #l:Ucase$(al2$)PRINT #l:Ucase$(a2$)PRINT #l:ucase$(a7$)Print #1 :ucase$(a8 $)PRINT #l:ucase$(all$)
END SUB
Menu routines
a True BASIC(tm), Inc. product
ABSTRACT Library of portable menu routines for use in user programs.
Copyright (c) 1985 by True BASIC, Inc.
SUB menu_set(where$,c$,maxent,maxlen,menu$,#9)
! Where$: top, bottom, left, or right ! C$ : color choice, ignored in MAC version ! maxent: maximum items in menu ! Maxlen: maximum length of menu item ! Menu$: for internal use ! #9 window
! Compute parameters
LET w$ = lcase$(where$) ! Screen positionIF w$="left" or w$="right" then LET vert = 1 ! Vertical windowLET m = min(maxent,10) ! Max entriesLET ml = min(maxlen,10) ! Max lengthLET fill = 3 ! Button, spacesLET zone = ml + fill ! ZonewidthLET chars = 70 ! Chars per line
1 8 4
IF vert = 0 then ! Horizontal windowIF m>5 then LET lines=3 else LET lines=l
END IF
! Pack them
LET menu$ = ""CALL packb(menu$, 1,1, large)CALL packb(menu$,2,1,vert)CALL packb(menu$,3,1,extra)CALL packb(menu$,4,1,inverse)CALL packb(menu$,5,4,m)CALL packb(menu$,9,4,ml)CALL packb(menu$,13,4,perline)
! Open window
IF vert = 1 then ! Vertical windowLET lines = 2*m LET yl = (20-lines)/20 LET y2 = 1LET width = max(zone,12)/chars IF where$ = "left" then
LETxl = 0 LET x2 = width
FT SFLET xl = 1 - width LET x2 = 1
END IFOPEN #9: screen xl,x2,yl,y2 SET WINDOW 0,5,20*yl,20
ELSE ! Horizontal windowLETxl = 0 LET x2 = 1IF m<=5 then LET width = .1 else LET width = .2 IF where$ = "top" then
LET yl = .95 - width LET y2 = .95
ELSE LET yl - 0 LET y2 = width
END IFOPEN #9: screen xl,x2,yl,y2 IF m>5 then LET ym=0 else LET ym=1.75 SET WINDOW 0,25,ym,3
END IF
SET COLOR "black/white"CLEAR
END SUB
SUB menu_show(M$(),ml,menu$,#9)
WINDOW #9LET large = unpackb(menu$, 1,1)LET vert = unpackb(menu$,2,l)LET extra = unpackb(menu$,3,l)LET inverse = unpackb(menu$,4,l)LET m = unpackb(menu$,5,4)LET m = min(ml,m)LET ml = unpackb(menu$,9,4)LET perline = unpackb(menu$,13,4)
IF vert=0 then FOR i = 1 to m
LET x$ = using$("############",m$(i))IF i<=5 then
DRAW button(x$) with shift(5*i-4.75,2) ELSE
DRAW button(x$) with shift(5*i-29.75,.25) END IF
NEXT i ELSE
FOR i = 1 te mLET x$ = using$("############",m$(i)) DRAW button(x$) with shift(.25,20.5-2*i)
NEXT i END IF
END SUB
PICTURE button(s$)PLOT .25,1;4.25,1;4.5,.75;4.5,.25;PLOT 4.25,0;.25,0;0,.25;0,.75;.25,1
PLOT TEXT, AT .25,.25: s$END PICTURE
SUB menu_ask(ml,z,menu$,#9)
LET ml = unpackb(menu$,5,4)LET m = min(ml,ml)LET vert = unpackb(menu$,2,l)WINDOW #9
DO ! Force right answerDO
GET MOUSE x,y,s LOOP until s=2 IF vert=0 then
LET v = int(x/5)+l IF y>=1.75 and y<=3 then
LET z = v ELSEEF y<1.75 then
LET z = v+5 ELSE
LET z = 0 END IF
ELSE LET v = int(19.5 - y)IF mod(v,2)=l then
LET z=0 ELSE
LET z = v/2 + 1 END IF
END IFIF z<l or z>m then LET z = 0
LOOP until z>0 CALL blink(z,vert)
SUB blink(i,v)IF v=0 then
IF i<=5 then L E T xl =5*i-4.5 LET x2 = 5*i-.5 LET yl =2.1 LET y2 = 2.9
ELSE L E T xl =5*i-29.5 LET x2 = 5*i-25.5 L E T yl = .35 LET y2= 1.15
END IF ELSE
LE T xl = .5 LET x2 = 4.5 LET yl = 20.5 - 2*i + .1 LET y2 = yl + .8
END IFBOX KEEP xl,x2,yl,y2 in k$BOX CLEAR xl,x2,yl,y2P A T T C F y
BOX SHOW k$ at x l,y l END SUB
END SUB
SUB menu(M$(),m,a,menu$,#9) ! Show menu, get answer
CALL menu_show(M$,m,menu$,#9)CALL menu_ask(m,a,menu$,#9)CLEAR
END SUB
SUB menu_all(M$(),m,prompt$,ans,menu$,#l ,#9)
1 8 7
WINDOW #1 PRINT prompt$;M?CALL menu(M$,m,ans,menu$,#9) WINDOW #1 PRINT MS(ans)
END SUB
Appendix B
TABLES FOR INTERCONVERSION BETW EEN VARIOUS
CHROM ATOGRAPHIC SEPARATION MEASURES
1 8 8
1 8 9
The following tables are a compilation of various separation measures employed in
chromatography. They were developed by systematic variation of resolution between
two overlapped EMG peaks. As the true resolution (RS)True> Equation 1), was varied
from 0.5 to 1.5 in increments of 0.1, the height ratio between the two peaks was varied
from 1 to 8 in increments of 1. Also, x/a for the two peaks was varied from 0 to 3 in
increments of 1, both peaks having the same x/a. Thus, many possible combinations of
overlapping peaks were considered.
R s= AtG [1 ]4 V variance
The discrimination factor for each peak was defined as
h n j ■ h y j
[ 2 ]
where hp is the height of the peak, hv is the valley height, and the subscript i refers to
peak in question.
The measured resolution, Rs m, based on the Gaussian peak model, was calculated
via Equations 3 and 4 for width (W) measured at 50% or 75% of the peak height,
respectively, where the subscripts 1 and 2 refer to the first or second peak in the pair.
( lR .2 - tR ,l)Rs,m,50% - 1-16 (W lj50% + W2>50%) 13 3
(*R,2 ~ tR .l) RS,m,7 5 % = 0.759 (W l j 0 7 5 + W2 0 .75) [ 4 ]
1 9 0
The percent relative area overlap (%RO) is defined as
%ROj = a [ X 100% [ 5 ]
where i refers to the peak in question, Ajj refers to the common area of the two
overlapped peaks, and Aj refers to the total area of the ith peak, before overlap.
The overlap integral (£2, ref. 1) was defined as
O O
( JcAcB dt)2W — ^ -------= -------- [6]
|C A2 d t JC B 2 dt
where Qa and Cb are their signal heights at time L
The parameters b, a, and width (W), d0 >j, do 2 , and tR were measured using a
search algorithm (See Figure 2-1), while %RQ and £2 were measured from digitized
overlapping peaks with a sampling rate of 50 points/GQ.
To use the tables one must first measure dG j , dQ 2 and the parameters a and b
(for both peaks), at 25% (Table B-l), 50% (Table B-2), or 75% (Table B-3) of the peak
height The values of a and b are measured with respect to each peak individually and
therefore, in some situations one may be able to obtain b and a for one peak at the desired
peak height fraction, but not for the other peak. Width (W) for each peak is is the sum of
a and b for the respective peak. Secondly, by matching as close as possible the measured values of ^ ( ^ ) 2 , do l and do 2 , one can arrive at an
approximate value for %RO or Rs,true- Alternatively, one can arrive at Rs,true by using
Rs,m . b, and a from both peaks. The tables are arranged in acending order of the values
of dQ j. Therefore, one should match the values of dGj and do2 first
Table B -l. W, a and b measured at the 25% peak height fraction.
do.l do ,2 ^s.true %ROi %R0 2 n
0.751 0.878 0 .8 8.948 4.474 3.395E-03
0.752 0.876 1.1 3.895 1.947 6.252E-05
0.759 0.883 0.7 10.792 5.396 6.724E-03
0.770 0.986 1.3 3.321 0.208 1.344E-06
0.770 0.885 1 .0 5.475 2.738 4.910E-04
0.771 0.986 0.9 8.438 0.527 1.388E-03
0.778 0.791 0.7 9.853 9.853 6.724E-03
0.779 0.787 0 .8 7.898 7.898 3.395E-03
0.779 0.972 1.1 5.210 0.651 1.589E-04
0.787 0.947 1 .2 3.173 0.793 9.930E-06
0.791 0.974 0.9 7.648 0.956 1.388E-03
0.798 0.987 0 .8 9.309 0.582 2.893E-03
0.811 0.976 0 .8 8.685 1.086 2.893E-03
0.811 0.953 0.9 6.892 1.723 1.388E-03
0.812 0.814 1 .0 4.480 4.480 4.910E-04
*>1 alW i W j
0.6471 0.3529
0.5047 0.4953
0.6932 0.3068
0.5000 0.5000
0.5673 0.4327
0.6340 0.3660
0.6899 0.3101
0.6349 0.3651
0.5618 0.4382
0.5007 0.4993
0.6317 0.3683
0.6880 0.3120
0.6879 0.3121
0.6308 0.3692
0.5588 0.4412
b2 a 2w2 w2
0.6145 0.3855
0.4634 0.5366
0.6722 0.3278
0.4825 0.5175
0.5436 0.4564
0.6289 0.3711
0.6465 0.3535
0.5888 0.4112
0.5522 0.4478
0.4851 0.5149
0.6278 0.3722
0 .6 8 6 6 0.3134
0.6855 0.3145
0.6256 0.3744
0.5299 0.4701 191
Table B-l. Cont’d
d0,l do ,2 Rs,true %ROi %R0 2 Q
0.815 0.954 1 .1 4.371 1.093 1.589E-04
0.822 0.822 1 .1 2.781 2.781 6.252E-05
0.822 0.822 1 .1 2.781 2.781 6.252E-05
0.823 0.956 0 .8 8.067 2.017 2.893E-03
0.825 0.978 1.3 2.470 0.309 1.344E-06
0.829 0.916 0.9 6.171 3.085 1.388E-03
0.830 0.989 1 .2 4.005 0.250 5.132E-05
0.835 0.919 0 .8 7.452 3.726 2.893E-03
0.843 0.922 1 .2 2.299 1.150 9.930E-06
0.845 0.990 1 .0 5.734 0.358 5.675E-04
0.846 0.923 1 .1 3.639 1.819 1.589E-04
0.847 0.853 0 .8 6.838 6.838 2.893E-03
0.847 0.851 0.9 5.480 5.480 1.388E-03
0.856 0.982 1 .2 3.400 0.425 5.132E-05
0.858 0.982 1 .0 5.213 0.652 5.675E-04
b l alWi Wj
0.5570 0.4430
0.5105 0.4895
0.5105 0.4895
0.6878 0.3122
0.5001 0.4999
0.6303 0.3697
0.5551 0.4449
0.6878 0.3122
0.5014 0.4986
0.6300 0.3700
0.5551 0.4449
0.6877 0.3123
0.6301 0.3699
0.5542 0.4458
0.6300 0.3700
^ 2 a 2w2 w2
0.5509 0.4491
0.4895 0.5105
0.4895 0.5105
0.6832 0.3168
0.4934 0.5066
0.6207 0.3793
0.5531 0.4469
0.6782 0.3218
0.4939 0.5061
0.6292 0.3708
0.5482 0.4518
0.6659 0.3341
0.6090 0.3910
0.5527 0.4473
0.6286 0.3714
19
2
Table B-l. Cont'd
do,l do ,2 &s,true %ROi %R02 Q
0.862 0.991 1.4 1.845 0.115 1.550E-07
0.863 0.991 0.9 6.357 0.397 1.245E-03
0.869 0.967 1.3 1.811 0.453 1.344E-06
0.871 0.984 0.9 5.945 0.743 1.245E-03
0.871 0.968 1 .0 4.716 1.179 5.675E-04
0.873 0.874 1.1 3.001 3.001 1.589E-04
0.876 0.752 1 .1 1.947 3.895 6.252E-05
0.878 0.970 1 .2 2 .8 6 8 0.717 5.132E-05
0.879 0.970 0.9 5.538 1.384 1.245E-03
0.883 0.942 1 .0 4.241 2 .1 2 0 5.675E-04
0.887 0.944 0.9 5.133 2.567 1.245E-03
0 .8 8 8 0 .8 8 8 1 .2 1.639 1.639 9.930E-06
0 .8 8 8 0 .8 8 8 1 .2 1.639 1.639 9.930E-06
0.890 0.993 1.3 2.589 0.162 1.656E-05
0.895 0.898 0.9 4.731 4.731 1.245E-03
*>1 a lW j Wi
0.5000 0.5000
0.6877 0.3123
0.5002 0.4998
0.6877 0.3123
0.6299 0.3701
0.5542 0.4458
0.5047 0.4953
0.5538 0.4462
0.6877 0.3123
0.6299 0.3701
0.6877 0.3123
0.5028 0.4972
0.5028 0.4972
0.5536 0.4464
0.6877 0.3123
t>2 a 2w2 w2
0.4973 0.5027
0.6870 0.3130
0.4970 0.5030
0.6863 0.3137
0.6272 0.3728
0.5420 0.4580
0.4634 0.5366
0.5520 0.4480
0.6848 0.3152
0.6243 0.3757
0.6817 0.3183
0.4972 0.5028
0.4972 0.5028
0.5532 0.4468
0.6747 0.3253
19
3
Table B-l. Cont’d
do,l do,2 Rs,true %ROi %R02 Q
0.895 0.897 1.0 3.787 3.787 5.675E-04
0.895 0.993 1.1 3.889 0.243 2.320E-04
0.896 0.987 1.4 1.365 0.171 1.550E-07
0.897 0.797 1.1 2.449 4.899 1.589E-04
0.898 0.949 1.2 2.403 1.201 5.132E-05
0.903 0.816 0.9 4.328 8.655 1.245E-03
0.904 0.988 1.1 3.547 0.443 2.320E-04
0.905 0.952 1.3 1.309 0.655 1.344E-06
0.906 0.988 1.3 2.209 0.276 1.656E-05
0.906 0.819 1.0 3.352 6.704 5.675E-04
0.907 0.994 1.0 4.336 0.271 5.357E-04
0.912 0.989 1.0 4.064 0.508 5.357E-04
0.912 0.978 1.1 3.219 0.805 2.320E-04
0.915 0.916 1.2 1.996 1.996 5.132E-05
0.917 0.979 1.0 3.795 0.949 5.357E-04
*>1 a lW! W j
0.6299 0.3701
0.6299 0.3701
0.5000 0.5000
0.5538 0.4462
0.5536 0.4464
0.6877 0.3123
0.6299 0.3701
0.5004 0.4996
0.5535 0.4465
0.6299 0.3701
0.6877 0.3123
0.6877 0.3123
0.6299 0.3701
0.5535 0.4465
0.6877 0.3123
*>2 a2W2 w 2
0.6178 0.3822
0.6295 0.3705
0.4987 0.5013
0.5247 0.4753
0.5505 0.4495
0.6557 0.3443
0.6291 0.3709
0.4985 0.5015
0.5530 0.4470
0.6005 0.3995
0.6872 0.3128
0.6868 0.3132
0.6282 0.3718
0.5474 0.4526
0.6858 0.3142
19
4
Table B-l. Cont'd
d0>l do ,2 Rs,trae
0.920 0.960 1 .1
0.920 0.980 1.3
0.921 0.995 1.5
0.922 0.843 1 .2
0.923 0.962 1 .0
0.923 0.981 1.4
0.928 0.929 1 .0
0.928 0.929 1 .1
0.929 0.996 1.4
0.929 0.996 1 .2
0.931 0.862 1 .2
0.932 0.932 1.3
0.932 0.932 1.3
0.933 0.967 1.3
0.933 0.871 1 .0
%ROi %R0 2 Q
2.906 1.453 2.320E-04
1.873 0.468 1.656E-05
0.986 0.062 1.523E-08
1.150 2.299 9.930E-06
3.529 1.764 5.357E-04
0.996 0.249 1.550E-07
3.264 3.264 5.357E-04
2.607 2.607 2.320E-04
1 .6 6 8 0.104 5.342E-06
2.634 0.165 9.486E-05
1.643 3.287 5.132E-05
0.932 0.932 1.344E-06
0.932 0.932 1.344E-06
1.578 0.789 1.656E-05
3.000 6 .0 0 0 5.357E-04
*>1 alW j Wj
0.6299 0.3701
0.5534 0.4466
0.5000 0.5000
0.5014 0.4986
0.6877 0.3123
0.5000 0.5000
0.6877 0.3123
0.6299 0.3701
0.5534 0.4466
0.6299 0.3701
0.5535 0.4465
0.5007 0.4993
0.5007 0.4993
0.5534 0.4466
0.6877 0.3123
b2 a 2w2 w2
0.6264 0.3736
0.5526 0.4474
0.4995 0.5005
0.4939 0.5061
0.6838 0.3162
0.4994 0.5006
0.6796 0.3204
0.6225 0.3775
0.5533 0.4467
0.6296 0.3704
0.5402 0.4598
0.4993 0.5007
0.4993 0.5007
0.5518 0.4482
0.6697 0.3303 19
5
Table B-l. Cont'd
do.l do ,2 Rs,true %ROi %R02 n
0.935 0.992 1 .2 2.408 0.301 9.486E-05
0.935 0.874 1.1 2.321 4.642 2.320E-04
0.937 0.996 1 .1 2.954 0.185 2.305E-04
0.938 0.770 1 .0 2.735 10.938 5.357E-04
0.939 0.992 1.4 1.429 0.179 5.342E-06
0.940 0.993 1.1 2.774 0.347 2.305E-04
0.940 0.985 1 .2 2.192 0.548 9.486E-05
0.941 0.993 1.5 0.726 0.091 1.523E-08
0.943 0.780 1.1 2.047 8.189 2.320E-04
0.944 0.986 1.1 2.596 0.649 2.305E-04
0.944 0.778 1 .2 1.339 5.355 5.132E-05
0.944 0.944 1.3 1.320 1.320 1.656E-05
0.944 0.972 1.4 0.718 0.359 1.550E-07
0.946 0.973 1 .2 1.986 0.993 9.486E-05
0.947 0.787 1 .2 0.793 3.173 9.930E-06
*>1 alW i Wj
0.6299 0.3701
0.6299 0.3701
0.6877 0.3123
0.6877 0.3123
0.5534 0.4466
0.6877 0.3123
0.6299 0.3701
0.5000 0.5000
0.6299 0.3701
0.6877 0.3123
0.5534 0.4466
0.5534 0.4466
0.5001 0.4999
0.6299 0.3701
0.5007 0.4993
*>2 a 2w2 w2
0.6294 0.3706
0.6137 0.3863
0.6874 0.3126
0.6344 0.3656
0.5532 0.4468
0.6871 0.3129
0.6288 0.3712
0.4998 0.5002
0.5850 0.4150
0.6865 0.3135
0.5175 0.4825
0.5501 0.4499
0.4997 0.5003
0.6277 0.3723
0.4851 0.5149 19
6
Table B-l. Cont’d
d0,l ^o ,2 fts.true %ROi %R0 2 Q
0.947 0.974 1.1 2.421 1 .2 1 0 2.305E-04
0.948 0.987 1.4 1.218 0.304 5.342E-06
0.951 0.951 1 .1 2.247 2.247 2.305E-04
0.951 0.951 1 .2 1.789 1.789 9.486E-05
0.952 0.997 1.3 1.781 0 .1 1 1 3.878E-05
0.952 0.905 1.3 0.655 1.309 1.344E-06
0.954 0.908 1.3 1.095 2.189 1.656E-05
0.954 0.910 1 .1 2.073 4.147 2.305E-04
0.955 0.997 1.5 1.071 0.067 1.723E-06
0.956 0.913 1 .2 1.601 3.202 9.486E-05
0.956 0.995 1.3 1.632 0.204 3.878E-05
0.956 0.978 1.4 1.031 0.516 5.342E-06
0.957 0.989 1.5 0.528 0.132 1.523E-08
0.957 0.997 1 .2 2 .0 1 1 0.126 9.920E-05
0.957 0.837 1.1 1.900 7.600 2.305E-04
bl alWi W i
0.6877 0.3123
0.5534 0.4466
0.6877 0.3123
0.6299 0.3701
0.6299 0.3701
0.5004 0.4996
0.5534 0.4466
0.6877 0.3123
0.5534 0.4466
0.6299 0.3701
0.6299 0.3701
0.5534 0.4466
0.5000 0.5000
0.6877 0.3123
0.6877 0.3123
*>2 a 2w2 w2
0.6852 0.3148
0.5530 0.4470
0.6826 0.3174
0.6253 0.3747
0.6297 0.3703
0.4985 0.5015
0.5465 0.4535
0.6768 0.3232
0.5534 0.4466
0.6203 0.3797
0.6296 0.3704
0.5525 0.4475
0.4999 0.5001
0.6875 0.3125
0.6620 0.3380 19
7
Table B-l. Cont'd
do.l ^o,2 Rs.true %RO\ %R0 2 Q
0.959 0.995 1.2 1.892 0.236 9.920E-05
0.960 0.990 1.3 1.490 0.373 3.878E-05
0.960 0.960 1.4 0.511 0.511 1.550E-07
0.960 0.960 1.4 0.511 0.511 1.550E-07
0.960 0.846 1.2 1.421 5.685 9.486E-05
0.961 0.995 1.5 0.921 0.115 1.723E-06
0.962 0.990 1.2 1.774 0.444 9.920E-05
0.962 0.849 1.3 0.900 3.598 1.656E-05
0.963 0.963 1.4 0 .8 6 8 0 .8 6 8 5.342E-06
0.963 0.982 1.3 1.354 0.677 3.878E-05
0.964 0.982 1.2 1.658 0.829 9.920E-05
0.966 0.966 1.2 1.543 1.543 9.920E-05
0.966 0.992 1.5 0.788 0.197 1.723E-06
0.967 0.967 1.3 1.225 1.225 3.878E-05
0.967 0.869 1.3 0.453 1.811 1.344E-06
b l alWi Wi
0.6877 0.3123
0.6299 0.3701
0.5002 0.4998
0.5002 0.4998
0.6299 0.3701
0.5534 0.4466
0.6877 0.3123
0.5534 0.4466
0.5534 0.4466
0.6299 0.3701
0.6877 0.3123
0.6877 0.3123
0.5534 0.4466
0.6299 0.3701
0.5002 0.4998
*>2 a 2w 2 w 2
0.6873 0.3127
0.6292 0.3708
0.4998 0.5002
0.4998 0.5002
0.6077 0.3923
0.5533 0.4467
0.6869 0.3131
0.5379 0.4621
0.5516 0.4484
0.6285 0.3715
0.6861 0.3139
0.6844 0.3156
0.5532 0.4468
0.6270 0.3730
0.4970 0.5030 19
8
Table B-l. Cont'd
d0 ,l do ,2 Rs,true %ROi %R0 2 Q
0.968 0.998 1.4 1.202 0.075 1.586E-05
0.968 0.938 1.2 1.429 2.858 9.920E-05
0.969 0.984 1.5 0.380 0.190 1.523E-08
0.969 0.939 1.4 0.725 1.449 5.342E-06
0.969 0.759 1.3 0.731 5.849 1.656E-05
0.970 0.940 1.3 1.101 2.202 3.878E-05
0.970 0.996 1.4 1.105 0.138 1.586E-05
0.971 0.886 1.2 1.315 5.262 9.920E-05
0.971 0.998 1.3 1.367 0.085 4.268E-05
0.971 0.986 1.5 0.671 0.336 1.723E-06
0.972 0.944 1.4 0.359 0.718 1.550E-07
0.972 0.997 1.3 1.288 0.161 4.268E-05
0.973 0.993 1.4 1.011 0.253 1.586E-05
0.973 0.893 1.3 0.982 3.930 3.878E-05
0.973 0.796 1.2 1.202 9.612 9.920E-05
*>1 alW i Wj
0.6299 0.3701
0.6877 0.3123
0.5000 0.5000
0.5534 0.4466
0.5534 0.4466
0.6299 0.3701
0.6299 0.3701
0.6877 0.3123
0.6877 0.3123
0.5534 0.4466
0.5001 0.4999
0.6877 0.3123
0.6299 0.3701
0.6299 0.3701
0.6877 0.3123
t>2 a2w 2 w 2
0.6298 0.3702
0.6808 0.3192
0.4999 0.5001
0.5496 0.4504
0.5045 0.4955
0.6240 0.3760
0.6297 0.3703
0.6727 0.3273
0.6876 0.3124
0.5529 0.4471
0.4997 0.5003
0.6875 0.3125
0.6295 0.3705
0.6171 0.3829
0.6486 0.3514 19
9
Table B-l. Cont'd
do,l d0 ,2 Rs.true %ROi %R0 2 Q
0.974 0.994 1.3 1 .2 1 0 0.303 4.268E-05
0.975 0.899 1.4 0.600 2.401 5.342E-06
0.975 0.988 1.4 0.922 0.461 1.586E-05
0.975 0.988 1.3 1.134 0.567 4.268E-05
0.976 0.812 1.3 0.869 6.953 3.878E-05
0.976 0.976 1.5 0.568 0.568 1.723E-06
0.977 0.977 1.3 1.058 1.058 4.268E-05
0.977 0.977 1.4 0.836 0.836 1.586E-05
0.978 0.978 1.5 0.270 0.270 1.523E-08
0.978 0.978 1.5 0.270 0.270 1.523E-08
0.978 0.825 1.3 0.309 2.470 1.344E-06
0.978 0.999 1.5 0.811 0.051 6.483E-06
0.978 0.957 1.3 0.983 1.966 4.268E-05
0.979 0.836 1.4 0.492 3.938 5.342E-06
0.979 0.959 1.4 0.755 1.509 1.586E-05
*>1 alW i W j
0.6877 0.3123
0.5534 0.4466
0.6299 0.3701
0.6877 0.3123
0.6299 0.3701
0.5534 0.4466
0.6877 0.3123
0.6299 0.3701
0.5000 0.5000
0.5000 0.5000
0.5001 0.4999
0.6299 0.3701
0.6877 0.3123
0.5534 0.4466
0.6299 0.3701
^ 2 a 2w2 w2
0.6872 0.3128
0.5454 0.4546
0.6290 0.3710
0.6867 0.3133
0.5984 0.4016
0.5524 0.4476
0.6856 0.3144
0.6281 0.3719
0.5000 0.5000
0.5000 0.5000
0.4934 0.5066
0.6299 0.3701
0.6833 0.3167
0.5351 0.4649
0.6262 0.3738 200
Table B-l. Cont'd
<*0,1 do,2 Rs.true %ROj %R02 n
0.980 0.960 1.5 0.477 0.954 1.723E-06
0.980 0.921 1.3 0.908 3.633 4.268E-05
0.980 0.998 1.5 0.746 0.093 6.483E-06
0.980 0.999 1.4 0.929 0.058 1.837E-05
0.981 0.923 1.4 0.249 0.996 1.550E-07
0.981 0.998 1.4 0.876 0.110 1.837E-05
0.981 0.926 1.4 0.677 2.707 1.586E-05
0.981 0.857 1.3 0.834 6.669 4.268E-05
0.982 0.995 1.5 0.685 0.171 6.483E-06
0.982 0.996 1.4 0.825 0.206 1.837E-05
0.983 0.992 1.5 0.626 0.313 6.483E-06
0.983 0.933 1.5 0.398 1.592 1.723E-06
0.983 0.869 1.4 0.602 4.817 1.586E-05
0.983 0.992 1.4 0.774 0.387 1.837E-05
0.984 0.984 1.4 0.724 0.724 1.837E-05
*>1 alWi Wj
0.5534 0.4466
0.6877 0.3123
0.6299 0.3701
0.6877 0.3123
0.5000 0.5000
0.6877 0.3123
0.6299 0.3701
0.6877 0.3123
0.6299 0.3701
0.6877 0.3123
0.6299 0.3701
0.5534 0.4466
0.6299 0.3701
0.6877 0.3123
0.6877 0.3123
t>2 a2w2 w2
0.5513 0.4487
0.6785 0.3215
0.6298 0.3702
0.6876 0.3124
0.4994 0.5006
0.6876 0.3124
0.6222 0.3778
0.6667 0.3333
0.6296 0.3704
0.6874 0.3126
0.6293 0.3707
0.5491 0.4509
0.6127 0.3873
0.6870 0.3130
0.6863 0.3137 201
Table B-l. Cont'd
d0,l do,2 Rs,true
0.984 0.969 1.5
0.985 0.985 1.5
0.985 0.772 1.4
0.985 0.971 1.4
0.986 0.770 1.3
0.986 0.972 1.5
0.986 0.889 1.5
0.986 0.946 1.4
0.987 0.999 1.5
0.987 0.896 1.4
0.987 0.900 1.4
0.987 0.949 1.5
0.987 0.998 1.5
0.988 0.997 1.5
0.988 0.820 1.4
%ROi %R02 Q
0.190 0.380 1.523E-08
0.570 0.570 6.483E-06
0.531 8.491 1.586E-05
0.675 1.349 1.837E-05
0.208 3.321 1.344E-06
0.516 1.032 6.483E-06
0.329 2.631 1.723E-06
0.626 2.502 1.837E-05
0.630 0.039 7.903E-06
0.171 1.365 1.550E-07
0.577 4.613 1.837E-05
0.465 1.858 6.483E-06
0.596 0.075 7.903E-06
0.562 0.140 7.903E-06
0.528 8.443 1.837E-05
*>1 alW i
0.5000 0.5000
0.6299 0.3701
0.6299 0.3701
0.6877 0.3123
0.5000 0.5000
0.6299 0.3701
0.5534 0.4466
0.6877 0.3123
0.6877 0.3123
0.5000 0.5000
0.6877 0.3123
0.6299 0.3701
0.6877 0.3123
0.6877 0.3123
0.6877 0.3123
*>2 a2w2 w2
0.4999 0.5001
0.6288 0.3712
0.5804 0.4196
0.6849 0.3151
0.4825 0.5175
0.6276 0.3724
0.5442 0.4558
0.6819 0.3181
0.6877 0.3123
0.4987 0.5013
0.6751 0.3249
0.6251 0.3749
0.6876 0.3124
0.6875 0.3125
0.6571 0.3429 202
Table B-l. Cont'd
d0,l <*0 ,2 ^s,true %ROi %R0 2 Q
0.989 0.909 1.5 0.416 3.324 6.483E-06
0.989 0.994 1.5 0.528 0.264 7.903E-06
0.989 0.820 1.5 0.269 4.308 1.723E-06
0.989 0.957 1.5 0.132 0.528 1.523E-08
0.989 0.989 1.5 0.495 0.495 7.903E-06
0.990 0.840 1.5 0.369 5.897 6.483E-06
0.990 0.980 1.5 0.462 0.925 7.903E-06
0.991 0.963 1.5 0.430 1.720 7.903E-06
0.991 0.931 1.5 0.398 3.182 7.903E-06
0.991 0.862 1.4 0.115 1.845 1.550E-07
0.992 0.874 1.5 0.366 5.851 7.903E-06
0.993 0.941 1.5 0.091 0.726 1.523E-08
0.995 0.921 1.5 0.062 0.986 1.523E-08
bi alW i Wi
0.6299 0.3701
0.6877 0.3123
0.5534 0.4466
0.5000 0.5000
0.6877 0.3123
0.6299 0.3701
0.6877 0.3123
0.6877 0.3123
0.6877 0.3123
0.5000 0.5000
0.6877 0.3123
0.5000 0.5000
0.5000 0.5000
b2 a 2w2 w2
0.6198 0.3802
0.6873 0.3127
0.5317 0.4683
0.4999 0.5001
0 .6 8 6 8 0.3132
0.6063 0.3937
0.6859 0.3141
0.6840 0.3160
0.6799 0.3201
0.4973 0.5027
0.6703 0.3297
0.4998 0.5002
0.4995 0.5005
20
3
Table B-2. W, a and b measured at the 50% peak height fraction.*>1Wj
alW i
*>2w2
a 2w2do,l do,2 Rs,trae Rs.meas %ROi %R0 2 Q
0.501 0.875 1 .0 0.921 8.703 2.176 3.355E-04 0.5016 0.4984 0.4425 0.5575
0.504 0.969 0.7 0.982 18.161 1.135 8.304E-03 0.6077 0.3923 0.5835 0.4165
0.508 0.939 0.9 0.981 1 2 .0 1 2 1.502 1.507E-03 0.5672 0.4328 0.5327 0.4673
0.508 0.757 0 .8 0.849 1 2 .1 0 1 6.051 4.550E-03 0.5752 0.4248 0.5215 0.4785
0.536 0.598 0.5 0.788 20.203 20.203 3.632E-02 0.6381 0.3619 0.5806 0.4194
0.543 0.585 0 .6 0.794 16.151 16.151 2.030E-02 0.5988 0.4012 0.5365 0.4635
0.550 0.944 0.7 1 .0 0 0 16.338 2.042 8.304E-03 0.5919 0.4081 0.5822 0.4178
0.564 0.945 1 .1 1.059 7.215 0.902 6.252E-05 0.5003 0.4997 0.4788 0.5212
0.566 0.973 0 .6 1.019 19.870 1.242 1.563E-02 0.6313 0.3687 0.6281 0.3719
0.587 0.897 0.9 1.004 9.965 2.491 1.507E-03 0.5448 0.4552 0.5310 0.4690
0.594 0.950 0 .6 1.018 18.437 2.305 1.563E-02 0.6303 0.3697 0.6267 0.3733
0.595 0.900 0.7 1 .0 0 0 14.595 3.649 8.304E-03 0.5879 0.4121 0.5795 0.4205
0.604 0.615 0 .8 0.854 9.721 9.721 4.550E-03 0.5471 0.4529 0.5029 0.4971
0.605 0.605 0.9 0.841 7.186 7.186 1.534E-03 0.5227 0.4773 0.4773 0.5227
0.605 0.605 0.9 0.841 7.186 7.186 1.534E-03 0.5227 0.4773 0.4773 0.5227 20
4
Table B -2. Cont’d
d0,l do,2 Rs.true
0.607 0.976 1 .0
0.622 0.908 0 .6
0.625 0.812 1 .0
0.633 0.977 1 .2
0.639 0.825 0.7
0.650 0.833 0 .6
0.660 0.831 0.9
0.662 0.979 0 .8
0 .6 6 6 0.917 1 .1
0.667 0.958 1 .0
0.678 0.707 0 .6
0.681 0.700 0.7
0.693 0.962 0 .8
0.704 0.982 0.7
0.707 0.513 0 .6
Rs.meas %RO] %R02
1.127 9.442 0.590
1.015 17.011 4.253
0.965 6.363 3.182
1.171 5.743 0.359
0.993 12.927 6.463
1.006 15.583 7.791
1.005 8.179 4.089
1.155 12.393 0.775
1.073 5.353 1.338
1.130 7.941 0.993
0.986 14.142 14.142
0.974 11.327 11.327
1.154 11.192 1.399
1.193 13.611 0.851
0.916 12.670 25.340
n blWi
alWi
b2w2
a2w2
4.910E-04 0.5403 0.4597 0.5338 0.4662
1.563E-02 0.6299 0.3701 0.6238 0.3762
3.355E-04 0.5032 0.4968 0.4847 0.5153
9.930E-06 0.5000 0.5000 0.4908 0.5092
8.304E-03 0.5863 0.4137 0.5737 0.4263
1.563E-02 0.6297 0.3703 0.6175 0.3825
1.507E-03 0.5390 0.4610 0.5275 0.4725
3.395E-03 0.5851 0.4149 0.5839 0.4161
6.252E-05 0.5005 0.4995 0.4913 0.5087
4.910E-04 0.5371 0.4629 0.5334 0.4666
1.563E-02 0.6296 0.3704 0.6026 0.3974
8.304E-03 0.5855 0.4145 0.5596 0.4404
3.395E-03 0.5849 0.4151 0.5831 0.4169
6.724E-03 0.6295 0.3705 0.6286 0.3714
1.563E-02 0.6295 0.3705 0.5404 0.4596 20
5
Table B -2. Cont'd
<*o,l do, 2 Rs,true
0.717 0.965 1 .2
0.721 0.931 1 .0
0.722 0.508 0.7
0.722 0.966 0.7
0.723 0.931 0 .8
0.725 0.729 0.9
0.729 0.729 1 .0
0.729 0.729 1 .0
0.740 0.984 1 .1
0.741 0.936 0.7
0.751 0.878 0 .8
0.752 0.876 1 .1
0.759 0.883 0.7
0.770 0.986 1.3
0.770 0.885 1 .0
Rs.meas %ROi %R02
1.177 4.304 0.538
1.130 6.624 1.656
0.899 9.786 19.572
1.191 12.665 1.583
1.151 10.046 2.511
0.996 6.632 6.632
0.970 4.550 4.550
0.970 4.550 4.550
1.247 6.167 0.385
1.188 11.727 2.932
1.145 8.948 4.474
1.078 3.895 1.947
1.182 10.792 5.396
1.278 3.321 0.208
1.128 5.475 2.738
Q
9.930E-06
4.910E-04
8.304E-03
6.724E-03
3.395E-03
1.507E-03
3.355E-04
3.355E-04
1.589E-04
6.724E-03
3.395E-03
6.252E-05
6.724E-03
1.344E-06
4.910E-04
bl alWj W i
0.5001 0.4999
0.5356 0.4644
0.5851 0.4149
0.6295 0.3705
0.5848 0.4152
0.5365 0.4635
0.5068 0.4932
0.5068 0.4932
0.5349 0.4651
0.6295 0.3705
0.5848 0.4152
0.5010 0.4990
0.6295 0.3705
0.5000 0.5000
0.5349 0.4651
*>2 a 2w 2 w 2
0.4958 0.5042
0.5325 0.4675
0.4970 0.5030
0.6277 0.3723
0.5815 0.4185
0.5194 0.4806
0.4932 0.5068
0.4932 0.5068
0.5340 0.4660
0.6258 0.3742
0.5779 0.4221
0.4960 0.5040
0.6219 0.3781
0.4982 0.5018
0.5306 0.4694 206
Table B-2. Cont'd
do ,l do ,2 Rs.true Rs.meas %ROi %R02
0.771 0.986 0.9 1.301 8.438 0.527
0.778 0.791 0.7 1.168 9.853 9.853
0.779 0.787 0 .8 1.132 7.898 7.898
0.779 0.972 1 .1 1.247 5.210 0.651
0.782 0.578 0.9 0.965 5.303 10.606
0.787 0.947 1 .2 1.180 3.173 0.793
0.791 0.974 0.9 1.300 7.648 0.956
0.796 0.640 0.7 1.135 8.902 17.805
0.798 0.987 0 .8 1.364 9.309 0.582
0.805 0.640 0 .8 1 .1 0 0 6.889 13.778
0.811 0.976 0 .8 1.363 8.685 1.086
0.811 0.953 0.9 1.298 6.892 1.723
0.812 0.814 1 .0 1 .1 2 1 4.480 4.480
0.812 0.625 1 .0 0.965 3.182 6.363
0.815 0.954 1.1 1.246 4.371 1.093
a blWi
alWi
b2w2
a2w2
1.388E-03 0.5847 0.4153 0.5842 0.4158
6.724E-03 0.6295 0.3705 0.6133 0.3867
3.395E-03 0.5848 0.4152 0.5701 0.4299
1.589E-04 0.5346 0.4654 0.5338 0.4662
1.507E-03 0.5354 0.4646 0.4955 0.5045
9.930E-06 0.5001 0.4999 0.4980 0.5020
1.388E-03 0.5847 0.4153 0.5837 0.4163
6.724E-03 0.6295 0.3705 0.5907 0.4093
2.893E-03 0.6295 0.3705 0.6289 0.3711
3.395E-03 0.5848 0.4152 0.5496 0.4504
2.893E-03 0.6295 0.3705 0.6283 0.3717
1.388E-03 0.5847 0.4153 0.5826 0.4174
4.910E-04 0.5346 0.4654 0.5265 0.4735
3.355E-04 0.5032 0.4968 0.4847 0.5153
1.589E-04 0.5344 0.4656 0.5333 0.4667 20
7
Table B -2. Cont'd
do,l do,2 Rs,true
0.822 0.822 1 .1
0.822 0.822 1 .1
0.823 0.956 0 .8
0.825 0.978 1.3
0.829 0.916 0.9
0.830 0.989 1 .2
0.835 0.919 0 .8
0.843 0.922 1 .2
0.845 0.990 1 .0
0.846 0.923 1 .1
0.847 0.853 0 .8
0.847 0.851 0.9
0.849 0.704 1 .0
0.856 0.982 1 .2
0.858 0.982 1 .0
Rs.meas %ROi %R02
1.079 2.781 2.781
1.079 2.781 2.781
1.360 8.067 2.017
1.280 2.470 0.309
1.294 6.171 3.085
1.362 4.005 0.250
1.356 7.452 3.726
1.181 2.299 1.150
1.446 5.734 0.358
1.244 3.639 1.819
1.346 6.838 6.838
1.285 5.480 5.480
1.106 3.621 7.242
1.361 3.400 0.425
1.445 5.213 0.652
n
6.252E-05
6.252E-05
2.893E-03
1.344E-06
1.388E-03
5.132E-05
2.893E-03
9.930E-06
5.675E-04
1.589E-04
2.893E-03
1.388E-03
4.910E-04
5.132E-05
5.675E-04
bl alW! W i
0.5019 0.4981
0.5019 0.4981
0.6295 0.3705
0.5000 0.5000
0.5847 0.4153
0.5343 0.4657
0.6295 0.3705
0.5003 0.4997
0.5847 0.4153
0.5343 0.4657
0.6295 0.3705
0.5847 0.4153
0.5344 0.4656
0.5343 0.4657
0.5847 0.4153
b2 a 2w2 w2
0.4981 0.5019
0.4981 0.5019
0.6271 0.3729
0.4992 0.5008
0.5805 0.4195
0.5341 0.4659
0.6246 0.3754
0.4990 0.5010
0.5844 0.4156
0.5322 0.4678
0.6193 0.3807
0.5759 0.4241
0.5168 0.4832
0.5340 0.4660
0.5841 0.4159 208
Table B -2. Cont'd
d0,l do,2 Rs,true
0.859 0.740 0 .8
0.862 0.991 1.4
0.863 0.991 0.9
0.865 0.743 0.9
0.869 0.967 1.3
0.871 0.984 0.9
0.871 0.968 1 .0
0.871 0.561 0 .8
0.873 0.874 1.1
0.875 0.501 1 .0
0.876 0.752 1 .1
0.878 0.970 1 .2
0.879 0.970 0.9
0.882 0.571 0.9
0.882 0.544 1 .0
Rs.meas %ROi %R0 2
1.326 6 .2 2 1 12.441
1.379 1.845 0.115
1.535 6.357 0.397
1.266 4.818 9.636
1.280 1.811 0.453
1.534 5.945 0.743
1.444 4.716 1.179
1.267 5.591 22.364
1.240 3.001 3.001
0.921 2.176 8.703
1.078 1.947 3.895
1.361 2 .8 6 8 0.717
1.532 5.538 1.384
1.213 4.182 16.727
1.059 2 .8 8 6 11.545
Q
2.893E-03
1.550E-07
1.245E-03
1.388E-03
1.344E-06
1.245E-03
5.675E-04
2.893E-03
1.589E-04
3.355E-04
6.252E-05
5.132E-05
1.245E-03
1.388E-03
4.910E-04
bl alW i W i
0.6295 0.3705
0.5000 0.5000
0.6295 0.3705
0.5847 0.4153
0.5000 0.5000
0.6295 0.3705
0.5847 0.4153
0.6295 0.3705
0.5343 0.4657
0.5016 0.4984
0.5010 0.4990
0.5343 0.4657
0.6295 0.3705
0.5847 0.4153
0.5343 0.4657
t>2 a 2w2 w2
0.6071 0.3929
0.4997 0.5003
0.6291 0.3709
0.5652 0.4348
0.4996 0.5004
0.6287 0.3713
0.5834 0.4166
0.5687 0.4313
0.5300 0.4700
0.4425 0.5575
0.4960 0.5040
0.5337 0.4663
0.6279 0.3721
0.5324 0.4676
0.4846 0.5154 20
9
Table B •2. Cont'd
do,l do ,2 ^s.true
0.883 0.942 1 .0
0.887 0.944 0.9
0 .8 8 8 0 .8 8 8 1 .2
0 .8 8 8 0 .8 8 8 1 .2
0.890 0.993 1.3
0.895 0.898 0.9
0.895 0.897 1 .0
0.895 0.993 1 .1
0.896 0.987 1.4
0.897 0.797 1 .1
0.898 0.949 1 .2
0.903 0.816 0.9
0.904 0.988 1.1
0.905 0.952 1.3
0.906 0.988 1.3
^s.meas %ROj %R0 2
1.441 4.241 2 .1 2 0
1.529 5.133 2.567
1.181 1.639 1.639
1.181 1.639 1.639
1.475 2.589 0.162
1.523 4.731 4.731
1.435 3.787 3.787
1.591 3.889 0.243
1.379 1.365 0.171
1.232 2.449 4.899
1.360 2.403 1 .2 0 1
1.509 4.328 8.655
1.590 3.547 0.443
1.280 1.309 0.655
1.475 2.209 0.276
Q
5.675E-04
1.245E-03
9.930E-06
9.930E-06
1.656E-05
1.245E-03
5.675E-04
2.320E-04
1.550E-07
1.589E-04
5.132E-05
1.245E-03
2.320E-04
1.344E-06
1.656E-05
*>1 alWi W i
0.5847 0.4153
0.6295 0.3705
0.5005 0.4995
0.5005 0.4995
0.5343 0.4657
0.6295 0.3705
0.5847 0.4153
0.5847 0.4153
0.5000 0.5000
0.5343 0.4657
0.5343 0.4657
0.6295 0.3705
0.5847 0.4153
0.5001 0.4999
0.5343 0.4657
^ 2 a 2W2 w 2
0.5821 0.4179
0.6263 0.3737
0.4995 0.5005
0.4995 0.5005
0.5342 0.4658
0.6230 0.3770
0.5792 0.4208
0.5845 0.4155
0.4998 0.5002
0.5253 0.4747
0.5331 0.4669
0.6158 0.3842
0.5843 0.4157
0.4998 0.5002
0.5341 0.4659 210
Table B -2. Cont'd
^0 ,1 do ,2 Rs,true
0.906 0.819 1 .0
0.907 0.994 1 .0
0.911 0.679 0.9
0.912 0.989 1 .0
0.912 0.978 1.1
0.915 0.916 1 .2
0.917 0 .6 6 6 1.1
0.917 0.690 1 .0
0.917 0.979 1 .0
0.918 0.679 1.1
0.920 0.960 1.1
0.920 0.980 1.3
0.921 0.995 1.5
0.922 0.843 1 .2
0.923 0.962 1 .0
Rs.meas %ROi %R0 2
1.423 3.352 6.704
1.706 4.336 0.271
1.476 3.920 15.680
1.705 4.064 0.508
1.589 3.219 0.805
1.357 1.996 1.996
1.073 1.338 5.353
1.394 2.935 11.739
1.704 3.795 0.949
1 .2 1 2 1.975 7.901
1.587 2.906 1.453
1.475 1.873 0.468
1.478 0.986 0.062
1.181 1.150 2.299
1.702 3.529 1.764
n *>1Wi
alWi
*>2w2
a2w2
5.675E-04 0.5847 0.4153 0.5731 0.4269
5.357E-04 0.6295 0.3705 0.6292 0.3708
1.245E-03 0.6295 0.3705 0.5980 0.4020
5.357E-04 0.6295 0.3705 0.6290 0.3710
2.320E-04 0.5847 0.4153 0.5839 0.4161
5.132E-05 0.5343 0.4657 0.5319 0.4681
6.252E-05 0.5005 0.4995 0.4913 0.5087
5.675E-04 0.5847 0.4153 0.5581 0.4419
5.357E-04 0.6295 0.3705 0.6285 0.3715
1.589E-04 0.5343 0.4657 0.5138 0.4862
2.320E-04 0.5847 0.4153 0.5830 0.4170
1.656E-05 0.5343 0.4657 0.5339 0.4661
1.523E-08 0.5000 0.5000 0.4999 0.5001
9.930E-06 0.5003 0.4997 0.4990 0.5010
5.357E-04 0.6295 0.3705 0.6274 0.3726 Ki
Table B ■2. Cont'd
<*0 ,1 do ,2 Rs.true
0.923 0.981 1.4
0.928 0.929 1 .0
0.928 0.929 1 .1
0.929 0.996 1.4
0.929 0.996 1 .2
0.931 0.862 1 .2
0.932 0.932 1.3
0.932 0.932 1.3
0.933 0.967 1.3
0.933 0.871 1 .0
0.935 0.992 1 .2
0.935 0.874 1 .1
0.936 0.508 1 .1
0.937 0.996 1 .1
0.938 0.770 1 .0
Rs.meas %ROi %R0 2
1,379 0.996 0.249
1.697 3.264 3.264
1.583 2.607 2.607
1.589 1 .6 6 8 0.104
1.736 2.634 0.165
1.352 1.643 3.287
1.281 0.932 0.932
1.281 0.932 0.932
1.474 1.578 0.789
1.687 3.000 6 .0 0 0
1.735 2.408 0.301
1.575 2.321 4.642
1.135 1.570 12.557
1.876 2.954 0.185
1.667 2.735 10.938
a
1.550E-07
5.357E-04
2.320E-04
5.342E-06
9.486E-05
5.132E-05
1.344E-06
1.344E-06
1.656E-05
5.357E-04
9.486E-05
2.320E-04
1.589E-04
2.305E-04
5.357E-04
*>1 alWj W]
0.5000 0.5000
0.6295 0.3705
0.5847 0.4153
0.5343 0.4657
0.5847 0.4153
0.5343 0.4657
0.5001 0.4999
0.5001 0.4999
0.5343 0.4657
0.6295 0.3705
0.5847 0.4153
0.5847 0.4153
0.5343 0.4657
0.6295 0.3705
0.6295 0.3705
*>2 a2w2 w2
0.4999 0.5001
0.6253 0.3747
0.5813 0.4187
0.5342 0.4658
0.5846 0.4154
0.5294 0.4706
0.4999 0.5001
0.4999 0.5001
0.5336 0.4664
0.6209 0.3791
0.5845 0.4155
0.5776 0.4224
0.4629 0.5371
0.6293 0.3707
0.6108 0.3892 212
Table B -2. Cont'd
^0 ,1 do, 2 Rs.true
0.939 0.992 1.4
0.940 0.993 1 .1
0.940 0.985 1 .2
0.941 0.993 1.5
0.943 0.780 1 .1
0.943 0.606 1 .0
0.944 0.986 1.1
0.944 0.778 1 .2
0.944 0.944 1.3
0.944 0.972 1.4
0.945 0.564 1 .1
0.946 0.973 1 .2
0.947 0.787 1 .2
0.947 0.974 1 .1
0.948 0.987 1.4
Rs.meas %ROi %R0 2
1.589 1.429 0.179
1.876 2.774 0.347
1.734 2.192 0.548
1.478 0.726 0.091
1.557 2.047 8.189
1.613 2.465 19.721
1.875 2.596 0.649
1.341 1.339 5.355
1.473 1.320 1.320
1.379 0.718 0.359
1.059 0.902 7.215
1.733 1.986 0.993
1.180 0.793 3.173
1.874 2.421 1 .2 1 0
1.589 1.218 0.304
Q. t>lW j
alW i
^ 2W2
a 2w 2
5.342E-06 0.5343 0.4657 0.5342 0.4658
2.305E-04 0.6295 0.3705 0.6291 0.3709
9.486E-05 0.5847 0.4153 0.5842 0.4158
1.523E-08 . 0.5000 0.5000 0.5000 0.5000
2.320E-04 0.5847 0.4153 0.5693 0.4307
5.357E-04 0.6295 0.3705 0.5828 0.4172
2.305E-04 0.6295 0.3705 0.6288 0.3712
5.132E-05 0.5343 0.4657 0.5239 0.4761
1.656E-05 0.5343 0.4657 0.5329 0.4671
1.550E-07 0.5000 0.5000 0.5000 0.5000
6.252E-05 0.5003 0.4997 0.4788 0.5212
9.486E-05 0.5847 0.4153 0.5837 0.4163
9.930E-06 0.5001 0.4999 0.4980 0.5020
2.305E-04 0.6295 0.3705 0.6281 0.3719
5.342E-06 0.5343 0.4657 0.5341 0.4659 213
Table B •2. Cont'd
do,l do,2 Rs.true
0.950 0.628 1 .1
0.951 0.951 1 .1
0.951 0.951 1 .2
0.952 0.997 1.3
0.952 0.905 1.3
0.954 0.908 1.3
0.954 0.910 1.1
0.955 0.997 1.5
0.955 0.651 1 .2
0.956 0.913 1 .2
0.956 0.995 1.3
0.956 0.978 1.4
0.957 0.989 1.5
0.957 0.997 1 .2
0.957 0.837 1.1
^s.meas %ROj %R02
1.513 1.784 14.274
1.870 2.247 2.247.
1.730 1.789 1.789
1.880 1.781 0 .1 1 1
1.280 0.655 1.309
1.470 1.095 2.189
1.864 2.073 4.147
1.702 1.071 0.067
1.316 1.077 8.613
1.725 1.601 3.202
1.880 1.632 0.204
1.588 1.031 0.516
1.478 0.528 0.132
2.047 2 .0 1 1 0.126
1.850 1.900 7.600
Q
2.320E-04
2.305E-04
9.486E-05
3.878E-05
1.344E-06
1.656E-05
2.305E-04
1.723E-06
5.132E-05
9.486E-05
3.878E-05
5.342E-06
1.523E-08
9.920E-05
2.305E-04
*>1 alWi Wj
0.5847 0.4153
0.6295 0.3705
0.5847 0.4153
0.5847 0.4153
0.5001 0.4999
0.5343 0.4657
0.6295 0.3705
0.5343 0.4657
0.5343 0.4657
0.5847 0.4153
0.5847 0.4153
0.5343 0.4657
0.5000 0.5000
0.6295 0.3705
0.6295 0.3705
t>2 a 2w2 w2
0.5472 0.4528
0.6268 0.3732
0.5825 0.4175
0.5847 0.4153
0.4998 0.5002
0.5316 0.4684
0.6240 0.3760
0.5343 0.4657
0.5100 0.4900
0.5803 0.4197
0.5846 0.4154
0.5339 0.4661
0.5000 0.5000
0.6294 0.3706
0.6179 0.3821 214
Table B ■2. Cont'd
do ,l do,2 Rs,true
0.959 0.995 1 .2
0.960 0.990 1.3
0.960 0.960 1.4
0.960 0.960 1.4
0.960 0.846 1 .2
0.961 0.714 1.1
0.961 0.995 1.5
0.962 0.990 1 .2
0.962 0.849 1.3
0.963 0.963 1.4
0.963 0.982 1.3
0.964 0.982 1 .2
0.964 0.522 1.1
0.965 0.717 1 .2
0.965 0.734 1 .2
Rs,meas %ROi %RC>2
2.047 1.892 0.236
1.879 1.490 0.373
1.379 0.511 0.511
1.379 0.511 0.511
1.714 1.421 5.685
1.818 1.725 13.800
1.702 0.921 0.115
2.046 1.774 0.444
1.463 0.900 3.598
1.587 0 .8 6 8 0 .8 6 8
1.879 1.354 0.677
2.045 1.658 0.829
1.711 1.546 24.743
1.177 0.538 4.304
1 .6 8 8 1.249 9.989
a
9.920E-05
3.878E-05
1.550E-07
1.550E-07
9.486E-05
2.305E-04
1.723E-06
9.920E-05
1.656E-05
5.342E-06
3.878E-05
9.920E-05
2.305E-04
9.930E-06
9.486E-05
*>1 alW i Wi
0.6295 0.3705
0.5847 0.4153
0.5000 0.5000
0.5000 0.5000
0.5847 0.4153
0.6295 0.3705
0.5343 0.4657
0.6295 0.3705
0.5343 0.4657
0.5343 0.4657
0.5847 0.4153
0.6295 0.3705
0.6295 0.3705
0.5001 0.4999
0.5847 0.4153
*>2 a 2w2 w2
0.6293 0.3707
0.5844 0.4156
0.5000 0.5000
0.5000 0.5000
0.5754 0.4246
0.6035 0.3965
0.5342 0.4658
0.6291 0.3709
0.5287 0.4713
0.5335 0.4665
0.5840 0.4160
0.6286 0.3714
0.5486 0.4514
0.4958 0.5042
0.5641 0.4359 215
Table B-2. Cont'd
do,l do,2 Rs.true
0.966 0.966 1 .2
0.966 0.992 1.5
0.967 0.967 1.3
0.967 0.869 1.3
0.968 0.998 1.4
0.968 0.938 1 .2
0.969 0.984 1.5
0.969 0.939 1.4
0.969 0.558 1 .2
0.969 0.759 1.3
0.970 0.940 1.3
0.970 0.996 1.4
0.971 0 .8 8 6 1 .2
0.971 0.998 1.3
0.971 0.986 1.5
Rs.meas %ROi %RC>2
2.043 1.543 1.543
1.702 0.788 0.197
1.877 1.225 1.225
1.280 0.453 1.811
2.025 1 .2 0 2 0.075
2.038 1.429 2.858
1.478 0.380 0.190
1.585 0.725 1.449
1.613 1.083 17.322
1.450 0.731 5.849
1.873 1 .1 0 1 2 .2 0 2
2.025 1.105 0.138
2.029 1.315 5.262
2.218 1.367 0.085
1.702 0.671 0.336
f i
9.920E-05
1.723E-06
3.878E-05
1.344E-06
1.586E-05
9.920E-05
1.523E-08
5.342E-06
9.486E-05
1.656E-05
3.878E-05
1.586E-05
9.920E-05
4.268E-05
1.723E-06
*>1 alWj Wi
0.6295 0.3705
0.5343 0.4657
0.5847 0.4153
0.5000 0.5000
0.5847 0.4153
0.6295 0.3705
0.5000 0.5000
0.5343 0.4657
0.5847 0.4153
0.5343 0.4657
0.5847 0.4153
0.5847 0.4153
0.6295 0.3705
0.6295 0.3705
0.5343 0.4657
b2 a 2w2 w2
0.6277 0.3723
0.5342 0.4658
0.5833 0.4167
0.4996 0.5004
0.5847 0.4153
0.6259 0.3741
0.5000 0.5000
0.5328 0.4672
0.5277 0.4723
0.5223 0.4777
0.5820 0.4180
0.5846 0.4154
0.6221 0.3779
0.6294 0.3706
0.5341 0.4659
Table B-2. Cont’d
do,l d0>2 Rs,true
0.972 0.944 1.4
0.972 0.997 1.3
0.973 0.993 1.4
0.973 0.893 1.3
0.973 0.796 1 .2
0.974 0.994 1.3
0.975 0.899 1.4
0.975 0.988 1.4
0.975 0.647 1 .2
0.975 0.988 1.3
0.976 0.812 1.3
0.976 0.622 1.3
0.976 0.976 1.5
0.977 0.977 1.3
0.977 0.633 1 .2
Rs.meas %ROi %R0 2
1.379 0.359 0.718
2.218 1.288 0.161
2.024 1 .0 1 1 0.253
1 .8 6 6 0.982 3.930
2.009 1 .2 0 2 9.612
2.217 1 .2 1 0 0.303
1.582 0.600 2.401
2.024 0.922 0.461
1.959 1.086 17.377
2.216 1.134 0.567
1.849 0.869 6.953
1.417 0.587 9.385
1.701 0.568 0.568
2.215 1.058 1.058
1.171 0.359 5.743
n
1.550E-07
4.268E-05
1.586E-05
3.878E-05
9.920E-05
4.268E-05
5.342E-06
1.586E-05
9.920E-05
4.268E-05
3.878E-05
1.656E-05
1.723E-06
4.268E-05
9.930E-06
1>1 alW i Wi
0.5000 0.5000
0.6295 0.3705
0.5847 0.4153
0.5847 0.4153
0.6295 0.3705
0.6295 0.3705
0.5343 0.4657
0.5847 0.4153
0.6295 0.3705
0.6295 0.3705
0.5847 0.4153
0.5343 0.4657
0.5343 0.4657
0.6295 0.3705
0.5000 0.5000
b2 a 2w 2 w 2
0.5000 0.5000
0.6293 0.3707
0.5845 0.4155
0.5790 0.4210
0.6138 0.3862
0.6292 0.3708
0.5312 0.4688
0.5843 0.4157
0.5922 0.4078
0.6289 0.3711
0.5725 0.4275
0.5052 0.4948
0.5338 0.4662
0.6283 0.3717
0.4908 0.5092
Table B •2. Cont'd
<*0,1 <*o,2 ^s.true
0.977 0.977 1.4
0.978 0.978 1.5
0.978 0.978 1.5
0.978 0.825 1.3
0.978 0.999 1.5
0.978 0.957 1.3
0.979 0.679 1.3
0.979 0.836 1.4
0.979 0.959 1.4
0.980 0.960 1.5
0.980 0.921 1.3
0.980 0.998 1.5
0.980 0.999 1.4
0.981 0.923 1.4
0.981 0.998 1.4
Rs.meas %ROj %RC>2
2.023 0.836 0.836
1.478 0.270 0.270
1.478 0.270 0.270
1.280 0.309 2.470
2.170 0.811 0.051
2.212 0.983 1.966
1.811 0.760 12.165
1.574 0.492 3.938
2.020 0.755 1.509
1.700 0.477 0.954
2.205 0.908 3.633
2.170 0.746 0.093
2.388 0.929 0.058
1.379 0.249 0.996
2.388 0.876 0.110
Q
1.586E-05
1.523E-08
1.523E-08
1.344E-06
6.483E-06
4.268E-05
3.878E-05
5.342E-06
1.586E-05
1.723E-06
4.268E-05
6.483E-06
1.837E-05
1.550E-07
1.837E-05
bl alWi Wi
0.5847 0.4153
0.5000 0.5000
0.5000 0.5000
0.5000 0.5000
0.5847 0.4153
0.6295 0.3705
0.5847 0.4153
0.5343 0.4657
0.5847 0.4153
0.5343 0.4657
0.6295 0.3705
0.5847 0.4153
0.6295 0.3705
0.5000 0.5000
0.6295 0.3705
b2 a 2w 2 w 2
0.5839 0.4161
0.5000 0.5000
0.5000 0.5000
0.4992 0.5008
0.5847 0.4153
0.6272 0.3728
0.5565 0.4435
0.5279 0.4721
0.5830 0.4170
0.5334 0.4666
0.6248 0.3752
0.5847 0.4153
0.6294 0.3706
0.4999 0.5001
0.6294 0.3706 218
Table B -2. Cont'd
do,l do,2 Rs.true
0.981 0.926 1.4
0.981 0.857 1.3
0.982 0.995 1.5
0.982 0.996 1.4
0.983 0.746 1.3
0.983 0.992 1.5
0.983 0.933 1.5
0.983 0.869 1.4
0.983 0.992 1.4
0.983 0.737 1.4
0.984 0.984 1.4
0.984 0.969 1.5
0.985 0.985 1.5
0.985 0.772 1.4
0.985 0.971 1.4
Rs.meas %ROj %R02
2.015 0.677 2.707
2.192 0.834 6.669
2.169 0.685 0.171
2.388 0.825 0.206
2.162 0.759 12.139
2.169 0.626 0.313
1.698 0.398 1.592
2.005 0.602 4.817
2.388 0.774 0.387
1.558 0.399 6.387
2.386 0.724 0.724
1.478 0.190 0.380
2.168 0.570 0.570
1.982 0.531 8.491
2.384 0.675 1.349
Q *>1 al *>2 a2W i W j w 2 w 2
1.586E-05 0.5847 0.4153 0.5811 0.4189
4.268E-05 0.6295 0.3705 0.6196 0.3804
6.483E-06 0.5847 0.4153 0.5846 0.4154
1.837E-05 0.6295 0.3705 0.6293 0.3707
4.268E-05 0.6295 0.3705 0.6079 0.3921
6.483E-06 0.5847 0.4153 0.5845 0.4155
1.723E-06 0.5343 0.4657 0.5326 0.4674
1.586E-05 0.5847 0.4153 0.5772 0.4228
1.837E-05 0.6295 0.3705 0.6291 0.3709
5.342E-06 0.5343 0.4657 0.5203 0.4797
1.837E-05 0.6295 0.3705 0.6287 0.3713
1.523E-08 0.5000 0.5000 0.5000 0.5000
6.483E-06 0.5847 0.4153 0.5842 0.4158
1.586E-05 0.5847 0.4153 0.5685 0.4315
1.837E-05 0.6295 0.3705 0.6280 0.3720
I!
Table B•2. Cont'd
do,l do,2 Rs,true
0.986 0.770 1.3
0.986 0.972 1.5
0.986 0.889 1.5
0.986 0.946 1.4
0.987 0.999 1.5
0.987 0.896 1.4
0.987 0.900 1.4
0.987 0.949 1.5
0.987 0.998 1.5
0.988 0.997 1.5
0.988 0.820 1.4
0.989 0.909 1.5
0.989 0.994 1.5
0.989 0.820 1.5
0.989 0.957 1.5
Rs,meas %ROi %r o 2
1.278 0.208 3.321
2.166 0.516 1.032
1.694 0.329 2.631
2.380 0.626 2.502
2.559 0.630 0.039
1.379 0.171 1.365
2.371 0.577 4.613
2.163 0.465 1.858
2.559 0.596 0.075
2.559 0.562 0.140
2.351 0.528 8.443
2.156 0.416 3.324
2.558 0.528 0.264
1.685 0.269 4.308
1.478 0.132 0.528
Q blW i
alW j
b2w 2
a2w 2
1.344E-06 0.5000 0.5000 0.4982 0.5018
6.483E-06 0.5847 0.4153 0.5836 0.4164
1.723E-06 0.5343 0.4657 0.5308 0.4692
1.837E-05 0.6295 0.3705 0.6264 0.3736
7.903E-06 0.6295 0.3705 0.6295 0.3705
1.550E-07 0.5000 0.5000 0.4998 0.5002
1.837E-05 0.6295 0.3705 0.6232 0.3768
6.483E-06 0.5847 0.4153 0.5825 0.4175
7.903E-06 0.6295 0.3705 0.6294 0.3706
7.903E-06 0.6295 0.3705 0.6294 0.3706
1.837E-05 0.6295 0.3705 0.6162 0.3838
6.483E-06 0.5847 0.4153 0.5801 0.4199
7.903E-06 0.6295 0.3705 0.6293 0.3707
1.723E-06 0.5343 0.4657 0.5269 0.4731
1.523E-08 0.5000 0.5000 0.5000 0.5000 220
Table B ■2. Cont'd
d0,l do,2 Rs.true
0.989 0.989 1.5
0.990 0.840 1.5
0.990 0.980 1.5
0.991 0.963 1.5
0.991 0.931 1.5
0.991 0.862 1.4
0.992 0.874 1.5
0.993 0.941 1.5
0.995 0.921 1.5
^s,meas %ROi %R02
2.558 0.495 0.495
2.142 0.369 5.897
2.556 0.462 0.925
2.553 0.430 1.720
2.547 0.398 3.182
1.379 0.115 1.845
2.534 0.366 5.851
1.478 0.091 0.726
1.478 0.062 0.986
n *>1 al *>2 a2Wi w2 w2
7.903E-06 0.6295 0.3705 0.6290 0.3710
6.483E-06 0.5847 0.4153 0.5750 0.4250
7.903E-06 0.6295 0.3705 0.6285 0.3715
7.903E-06 0.6295 0.3705 0.6275 0.3725
7.903E-06 0.6295 0.3705 0.6254 0.3746
1.550E-07 0.5000 0.5000 0.4997 0.5003
7.903E-06 0.6295 0.3705 0.6211 0.3789
1.523E-08 0.5000 0.5000 0.5000 0.5000
1.523E-08 0.5000 0.5000 0.4999 0.5001
221
Table B-3. W, a and b measured at the 75% peak height fraction.
do.l do,2 Rs.true Rs,meas %ROi %R02 Q blW i
alWi
*>2w 2
a2w 2
0.253 0.623 0.8 0.689 15.255 7.628 5.9760E-03 0.5094 0.4906 0.4123 0.5877
0.258 0.655 0.5 0.694 26.592 13.296 4.9457E-02 0.6020 0.3980 0.5355 0.4645
0.265 0.265 0.7 0.527 16.151 16.151 1.9841E-02 0.5849 0.4151 0.4151 0.5849
0.265 0.265 0.7 0.527 16.151 16.151 1.9841E-02 0.5849 0.4151 0.4151 0.5849
0.280 0.955 0.6 0.899 26.539 1.659 2.0300E-02 0.5738 0.4262 0.5495 0.4505
0.294 0.822 0.9 0.840 13.630 3.408 1.5338E-03 0.5016 0.4984 0.4541 0.5459
0.297 0.912 0.8 0.897 17.969 2.246 4.5500E-03 0.5515 0.4485 0.5194 0.4806
0.309 0.662 0.7 0.759 17.688 8.844 1.3305E-02 0.5560 0.4440 0.5100 0.4900
0.347 0.919 0.6 0.912 23.774 2.972 2.0300E-Q2 0.5598 0.4402 0.5484 0.4516
0.352 0.441 0.5 0.689 22.864 22.864 4.9457E-02 0.5668 0.4332 0.5133 0.4867
0.363 0.920 1.0 0.969 11.639 1.455 3.3546E-04 0.5002 0.4998 0.4769 0.5231
0.365 0.961 0.5 0.943 28.960 1.810 3.6315E-02 0.5804 0.4196 0.5767 0.4233
0.404 0.852 0.8 0.916 14.844 3.711 4.5500E-03 0.5328 0.4672 0.5180 0.4820
0.407 0.927 0.5 0.941 26.782 3.348 3.6315E-02 0.5791 0.4209 0.5755 0.4245
0.413 0.857 0.6 0.912 21.129 5.282 2.0300E-02 0.5548 0.4452 0.5462 0.4538
222
Table B-3. Cont'd
do,l do,2 Rs.true Rs,meas %ROi %RC>2
0.422 0.964 0.9 1.043 14.336 0.896
0.443 0.470 0.7 0.755 14.076 14.076
0.447 0.447 0.8 0.743 10.961 10.961
0.447 0.447 0.8 0.743 10.961 10.961
0.448 0.965 1.1 1.085 9.544 0.597
0.450 0.867 0.5 0.936 24.609 6.152
0.457 0.728 0.9 0.874 10.028 5.014
0.479 0.752 0.6 0.903 18.591 9.296
0.493 0.764 0.5 0.925 22.423 11.212
0.501 0.875 1.0 0.986 8.703 2.176
0.504 0.969 0.7 1.079 18.161 1.135
0.508 0.939 0.9 1.047 12.012 1.502
0.508 0.757 0.8 0.916 12.101 6.051
0.536 0.598 0.5 0.900 20.203 20.203
0.543 0.585 0.6 0.880 16.151 16.151
o *>1 al *>2 a2W i W j w2 W2
1.5066E-03 0.5280 0.4720 0.5204 0.4796
1.3305E-02 0.5344 0.4656 0.4948 0.5052
5.9760E-03 0.5218 0.4782 0.4782 0.5218
5.9760E-03 0.5218 0.4782 0.4782 0.5218
6.2522E-05 0.5001 0.4999 0.4892 0.5108
3.6315E-02 0.5784 0.4216 0.5729 0.4271
1.5338E-03 0.5034 0.4966 0.4839 0.5161
2.0300E-02 0.5526 0.4474 0.5414 0.4586
3.6315E-02 0.5781 0.4219 0.5677 0.4323
3.3546E-04 0.5006 0.4994 0.4903 0.5097
8.3040E-03 0.5511 0.4489 0.5499 0.4501
1.5066E-03 0.5243 0.4757 0.5201 0.4799
4.5500E-03 0.5263 0.4737 0.5150 0.4850
3.6315E-02 0.5780 0.4220 0.5556 0.4444
2.0300E-02 0.5516 0.4484 0.5302 0.4698 223
Table B-3. Cont'd
<*o,l do,2 Rs.true
0.550 0.944 0.7
0.564 0.945 1.1
0.566 0.973 0.6
0.581 0.361 0.5
0.587 0.897 0.9
0.594 0.950 0.6
0.595 0.900 0.7
0.604 0.615 0.8
0.605 0.605 0.9
0.605 0.605 0.9
0.607 0.347 0.6
0.607 0.976 1.0
0.622 0.908 0.6
0.623 0.253 0.8
0.625 0.812 1.0
Rs.meas %RO\ %R02
1.078 16.338 2.042
1.093 7.215 0.902
1.136 19.870 1.242
0.839 17.904 35.807
1.047 9.965 2.491
1.134 18.437 2.305
1.074 14.595 3.649
0.905 9.721 9.721
0.881 7.186 7.186
0.881 7.186 7.186
0.818 13.790 27.580
1.169 9.442 0.590
1.130 17.011 4.253
0.689 7.628 15.255
0.993 6.363 3.182
Q 1>1W i
alWi
t>2w 2
nw 2
8.3040E-03 0.5508 0.4492 0.5492 0.4508
6.2522E-05 0.5001 0.4999 0.4950 0.5050
1.5627E-02 0.5778 0.4222 0.5770 0.4230
3.6315E-02 0.5779 0.4221 0.5193 0.4807
1.5066E-03 0.5225 0.4775 0.5193 0.4807
1.5627E-02 0.5778 0.4222 0.5763 0.4237
8.3040E-03 0.5507 0.4493 0.5478 0.4522
4.5500E-03 0.5235 0.4765 0.5081 0.4919
1.5338E-03 0.5072 0.4928 0.4928 0.5072
1.5338E-03 0.5072 0.4928 0.4928 0.5072
2.0300E-02 0.5510 0.4490 0.4941 0.5059
4.9097E-04 0.5215 0.4785 0.5206 0.4794
1.5627E-02 0.5778 0.4222 0.5747 0.4253
5.9760E-03 0.5094 0.4906 0.4123 0.5877
3.3546E-04 0.5011 0.4989 0.4954 0.5046 224
Table B -3. Cont'd
do,l do,2 Rs.true
0.633 0.977 1.2
0.639 0.825 0.7
0.650 0.833 0.6
0.660 0.831 0.9
0.662 0.979 0.8
0.666 0.917 1.1
0.667 0.958 1.0
0.678 0.707 0.6
0.681 0.700 0.7
0.691 0.416 0.8
0.693 0.962 0.8
0.704 0.982 0.7
0.707 0.513 0.6
0.717 0.965 1.2
0.721 0.931 1.0
Rs,meas %ROi %R02
1.198 5.743 0.359
1.066 12.927 6.463
1.121 15.583 7.791
1.044 8.179 4.089
1.235 12.393 0.775
1.097 5.353 1.338
1.169 7.941 0.993
1.104 14.142 14.142
1.050 11.327 11.327
0.870 7.678 15.356
1.234 11.192 1.399
1.326 13.611 0.851
1.065 12.670 25.340
1.199 4.304 0.538
1.168 6.624 1.656
o bi al t>2 a2W i W i w2 w2
9.9295E-06 0.5000 0.5000 0.4978 0.5022
8.3040E-03 0.5506 0.4494 0.5449 0.4551
1.5627E-02 0.5778 0.4222 0.5714 0.4286
1.5066E-03 0.5216 0.4784 0.5177 0.4823
3.3952E-03 0.5506 0.4494 0.5501 0.4499
6.2522E-05 0.5001 0.4999 0.4976 0.5024
4.9097E-04 0.5212 0.4788 0.5204 0.4796
1.5627E-02 0.5778 0.4222 0.5642 0.4358
8.3040E-03 0.5506 0.4494 0.5386 0.4614
4.5500E-03 0.5221 0.4779 0.4889 0.5111
3.3952E-03 0.5506 0.4494 0.5497 0.4503
6.724IE-03 0.5778 0.4222 0.5773 0.4227
1.5627E-02 0.5778 0.4222 0.5466 0.4534
9.9295E-06 0.5000 0.5000 0.4989 0.5011
4.9097E-04 0.5210 0.4790 0.5199 0.4801 22
5
Table B -3. Cont'd
<*o,l <*o,2 ^s,true
0.722 0.508 0.7
0.722 0.966 0.7
0.723 0.931 0.8
0.725 0.729 0.9
0.728 0.457 0.9
0.729 0.729 1.0
0.737 0.257 0.6
0.740 0.984 1.1
0.741 0.936 0.7
0.751 0.878 0.8
0.752 0.876 1.1
0.759 0.883 0.7
0.763 0.252 0.7
0.770 0.986 1.3
0.770 0.885 1.0
**s,meas %ROi %R02
1.014 9.786 19.572
1.325 12.665 1.583
1.231 10.046 2.511
1.036 6.632 6.632
0.874 5.014 10.028
0.994 4.550 4.550
0.928 11.128 44.511
1.287 6.167 0.385
1.322 11.727 2.932
1.226 8.948 4.474
1.099 3.895 1.947
1.316 10.792 5.396
0.870 8.291 33.162
1.300 3.321 0.208
1.165 5.475 2.738
Q b lWi
alWi
*>2w2
a2w2
8.3040E-03 0.5506 0.4494 0.5226 0.4774
6.724IE-03 0.5778 0.4222 0.5768 0.4232
3.3952E-03 0.5505 0.4495 0.5488 0.4512
1.5066E-03 0.5212 0.4788 0.5141 0.4859
1.5338E-03 0.5034 0.4966 0.4839 0.5161
3.3546E-04 0.5022 0.4978 0.4978 0.5022
1.5627E-02 0.5778 0.4222 0.4609 0.5391
1.5892E-04 0.5208 0.4792 0.5207 0.4793
6.7241E-03 0.5778 0.4222 0.5758 0.4242
3.3952E-03 0.5505 0.4495 0.5470 0.4530
6.2522E-05 0.5003 0.4997 0.4988 0.5012
6.7241E-03 0.5778 0.4222 0.5736 0.4264
8.3040E-03 0.5506 0.4494 0.4309 0.5691
1.3438E-06 0.5000 0.5000 0.4996 0.5004
4.9097E-04 0.5209 0.4791 0.5190 0.4810 226
Table B-3. Cont'd
d0,l ^o,2 ^s,tnie ^s.meas %ROi %R02 Cl b lW i
alW i
*>2w2
a2w2
0.771 0.986 0.9 1.390 8.438 0.527 1.3881E-03 0.5505 0.4495 0.5503 0.4497
0.778 0.791 0.7 1.303 9.853 9.853 6.7241E-03 0.5778 0.4222 0.5692 0.4308
0.779 0.787 0.8 1.215 7.898 7.898 3.3952E-03 0.5505 0.4495 0.5432 0.4568
0.779 0.972 1.1 1.286 5.210 0.651 1.5892E-04 0.5208 0.4792 0.5206 0.4794
0.782 0.578 0.9 1.017 5.303 10.606 1.5066E-03 0.5210 0.4790 0.5060 0.4940
0.787 0.947 1.2 1.200 3.173 0.793 9.9295E-06 0.5001 0.4999 0.4995 0.5005
0.791 0.974 0.9 1.389 7.648 0.956 1.3881E-03 0.5505 0.4495 0.5500 0.4500
0.796 0.640 0.7 1.277 8.902 17.805 6.724IE-03 0.5778 0.4222 0.5593 0.4407
0.798 0.987 0.8 1.516 9.309 0.582 2.8934E-03 0.5778 0.4222 0.5775 0.4225
0.805 0.640 0.8 1.191 6.889 13.778 3.3952E-03 0.5505 0.4495 0.5346 0.4654
0.811 0.976 0.8 1.515 8.685 1.086 2.8934E-03 0.5778 0.4222 0.5772 0.4228
0.811 0.953 0.9 1.387 6.892 1.723 1.3881E-03 0.5505 0.4495 0.5494 0.4506
0.812 0.814 1.0 1.160 4.480 4.480 4.9097E-04 0.5208 0.4792 0.5172 0.4828
0.812 0.625 1.0 0.993 3.182 6.363 3.3546E-04 0.5011 0.4989 0.4954 0.5046
0.815 0.954 1.1 1.286 4.371 1.093 1.5892E-04 0.5208 0.4792 0.5203 0.4797 22
7
Table B -3. Cont'd
do,l do,2 Rs.true
0.815 0.417 0.7
0.822 0.822 1.1
0.822 0.822 1.1
0.822 0.294 0.9
0.823 0.956 0.8
0.825 0.978 1.3
0.829 0.916 0.9
0.830 0.989 1.2
0.832 0.423 0.8
0.832 0.371 0.9
0.835 0.919 0.8
0.843 0.922 1.2
0.845 0.990 1.0
0.846 0.923 1.1
0.847 0.853 0.8
Rs.meas %ROi %R02
1.214 7.923 31.690
1.099 2.781 2.781
1.099 2.781 2.781
0.840 3.408 13.630
1.513 8.067 2.017
1.301 2.470 0.309
1.384 6.171 3.085
1.404 4.005 0.250
1.135 5.915 23.662
0.969 4.170 16.680
1.509 7.452 3.726
1.200 2.299 1.150
1.545 5.734 0.358
1.284 3.639 1.819
1.500 6.838 6.838
a t>lW i
alWi
b2w 2
a2w 2
6.7241E-03 0.5778 0.4222 0.5321 0.4679
6.2522E-05 0.5006 0.4994 0.4994 0.5006
6.2522E-05 0.5006 0.4994 0.4994 0.5006
1.5338E-03 0.5016 0.4984 0.4541 0.5459
2.8934E-03 0.5778 0.4222 0.5765 0.4235
1.3438E-06 0.5000 0.5000 0.4998 0.5002
1.388 IE-03 0.5505 0.4495 0.5484 0.4516
5.1316E-05 0.5208 0.4792 0.5208 0.4792
3.3952E-03 0.5505 0.4495 0.5106 0.4894
1.5066E-03 0.5209 0.4791 0.4813 0.5187
2.8934E-03 0.5778 0.4222 0.5751 0.4249
9.9295E-06 0.5001 0.4999 0.4997 0.5003
5.675 IE-04 0.5505 0.4495 0.5504 0.4496
1.5892E-04 0.5208 0.4792 0.5198 0.4802
2.8934E-03 0.5778 0.4222 0.5723 0.4277 228
Table B-3. Corn'd
d0,l do,2 Rs.true
0.847 0.851 0.9
0.849 0.704 1.0
0.856 0.982 1.2
0.858 0.982 1.0
0.859 0.740 0.8
0.862 0.991 1.4
0.863 0.991 0.9
0.865 0.743 0.9
0.869 0.967 1.3
0.871 0.984 0.9
0.871 0.968 1.0
0.871 0.561 0.8
0.873 0.874 1.1
0.875 0.501 1.0
0.876 0.752 1.1
^s,meas %RO\ %R02
1.376 5.480 5.480
1.149 3.621 7.242
1.404 3.400 0.425
1.544 5.213 0.652
1.481 6.221 12.441
1.401 1.845 0.115
1.706 6.357 0.397
1.361 4.818 9.636
1.301 1.811 0.453
1.706 5.945 0.743
1.543 4.716 1.179
1.441 5.591 22.364
1.281 3.001 3.001
0.986 2.176 8.703
1.099 1.947 3.895
f l *>1Wi
alWi
b2w2
a2w2
1.3881E-03 0.5505 0.4495 0.5460 0.4540
4.9097E-04 0.5208 0.4792 0.5131 0.4869
5.1316E-05 0.5208 0.4792 0.5206 0.4794
5.6751E-04 0.5505 0.4495 0.5502 0.4498
2.8934E-03 0.5778 0.4222 0.5663 0.4337
1.5498E-07 0.5000 0.5000 0.4999 0.5001
1.2450E-03 0.5778 0.4222 0.5776 0.4224
1.3881E-03 0.5505 0.4495 0.5410 0.4590
1.3438E-06 0.5000 0.5000 0.4999 0.5001
1.2450E-03 0.5778 0.4222 0.5774 0.4226
5.675 IE-04 0.5505 0.4495 0.5498 0.4502
2.8934E-03 0.5778 0.4222 0.5521 0.4479
1.5892E-04 0.5208 0.4792 0.5188 0.4812
3.3546E-04 0.5006 0.4994 0.4903 0.5097
6.2522E-05 0.5003 0.4997 0.4988 0.5012 229
Table B-3. Cont'd
<*0,1 do,2 Rs.true Rs.meas %RO\ %RC>2
0.878 0.970 1.2 1.403 2.868 0.717
0.879 0.970 0.9 1.704 5.538 1.384
0.882 0.571 0.9 1.326 4.182 16.727
0.882 0.544 1.0 1.125 2.886 11.545
0.883 0.942 1.0 1.540 4.241 2.120
0.884 0.315 0.8 1.330 4.936 39.488
0.887 0.944 0.9 1.701 5.133 2.567
0.888 0.888 1.2 1.200 1.639 1.639
0.888 0.888 1.2 1.200 1.639 1.639
0.890 0.993 1.3 1.521 2.589 0.162
0.895 0.898 0.9 1.695 4.731 4.731
0.895 0.897 1.0 1.535 3.787 3.787
0.895 0.993 1.1 1.700 3.889 0.243
0.896 0.987 1.4 1.401 1.365 0.171
0.897 0.797 1.1 1.275 2.449 4.899
QbiWi
alWj
*2w2
a2w2
5.1316E-05 0.5208 0.4792 0.5206 0.4794
1.2450E-03 0.5778 0.4222 0.5769 0.4231
1.3881E-03 0.5505 0.4495 0.5291 0.4709
4.9097E-04 0.5208 0.4792 0.5034 0.4966
5.675 IE-04 0.5505 0.4495 0.5491 0.4509
2.8934E-03 0.5778 0.4222 0.5037 0.4963
1.2450E-03 0.5778 0.4222 0.5761 0.4239
9.9295E-06 0.5002 0.4998 0.4998 0.5002
9.9295E-06 0.5002 0.4998 0.4998 0.5002
1.6558E-05 0.5208 0.4792 0.5208 0.4792
1.2450E-03 0.5778 0.4222 0.5743 0.4257
5.675 IE-04 0.5505 0.4495 0.5477 0.4523
2.3202E-04 0.5505 0.4495 0.5504 0.4496
1.5498E-07 0.5000 0.5000 0.5000 0.5000
1.5892E-04 0.5208 0.4792 0.5166 0.4834 23
0
Table B ■3. Cont'd
d0,l do,2 Rs.true
0.898 0.949 1.2
0.898 0.330 0.9
0.903 0.816 0.9
0.904 0.988 1.1
0.905 0.952 1.3
0.906 0.988 1.3
0.906 0.819 1.0
0.907 0.994 1.0
0.909 0.328 1.0
0.911 0.679 0.9
0.912 0.989 1.0
0.912 0.978 1.1
0.915 0.916 1.2
0.917 0.666 1.1
0.917 0.690 1.0
Rs.meas %ROi %RC>2
1.402 2.403 1 .2 0 1
1.232 3.566 28.526
1.682 4.328 8.655
1.699 3.547 0.443
1.301 1.309 0.655
1.521 2.209 0.276
1.525 3.352 6.704
1.896 4.336 0.271
1.059 2.262 18.092
1.655 3.920 15.680
1.896 4.064 0.508
1.698 3.219 0.805
1.401 1.996 1.996
1.097 1.338 5.353
1.503 2.935 11.739
Q bl al b2 a2W i Wl w2 w2
5.1316E-05 0.5208 0.4792 0.5202 0.4798
1.3881E-03 0.5505 0.4495 0.4888 0.5112
1.2450E-03 0.5778 0.4222 0.5705 0.4295
2.3202E-04 .0.5505 0.4495 0.5503 0.4497
1.3438E-06 0.5000 0.5000 0.4999 0.5001
1.6558E-05 0.5208 0.4792 0.5207 0.4793
5.675 IE-04 0.5505 0.4495 0.5447 0.4553
5.3573E-04 0.5778 0.4222 0.5777 0.4223
4.9097E-04 0.5208 0.4792 0.4705 0.5295
1.2450E-03 0.5778 0.4222 0.5622 0.4378
5.3573E-04 0.5778 0.4222 0.5775 0.4225
2.3202E-04 0.5505 0.4495 0.5501 0.4499
5.1316E-05 0.5208 0.4792 0.5197 0.4803
6.2522E-05 0.5001 0.4999 0.4976 0.5024
5.675 IE-04 0.5505 0.4495 0.5380 0.4620 231
Table B -3. Cont'd
do,l do,2 Rs.true
0.917 0.979 1.0
0.918 0.679 1.1
0.919 0.471 0.9
0.920 0.363 1.0
0.920 0.960 1.1
0.920 0.980 1.3
0.921 0.995 1.5
0.922 0.843 1.2
0.923 0.962 1.0
0.923 0.981 1.4
0.928 0.929 1.0
0.928 0.929 1.1
0.928 0.493 1.0
0.929 0.996 1.4
0.929 0.996 1.2
Rs.meas %RO| %R02
1.895 3.795 0.949
1.262 1.975 7.901
1.593 3.502 28.016
0.969 1.455 11.639
1.696 2.906 1.453
1.521 1.873 0.468
1.501 0.986 0.062
1.200 1.150 2.299
1.893 3.529 1.764
1.401 0.996 0.249
1.888 3.264 3.264
1.693 2.607 2.607
1.452 2.533 20.262
1.638 1.668 0.104
1.854 2.634 0.165
Q blWi
alWi &2
w2a2w2
5.3573E-04 0.5778 0.4222 0.5772 0.4228
1.5892E-04 0.5208 0.4792 0.5120 0.4880
1.2450E-03 0.5778 0.4222 0.5411 0.4589
3.3546E-04 0.5002 0.4998 0.4769 0.5231
2.3202E-04 0.5505 0.4495 0.5496 0.4504
1.6558E-05 0.5208 0.4792 0.5206 0.4794
1.5230E-08 0.5000 0.5000 0.5000 0.5000
9.9295E-06 0.5001 0.4999 0.4997 0.5003
5.3573E-04 0.5778 0.4222 0.5767 0.4233
1.5498E-07 0.5000 0.5000 0.5000 0.5000
5.3573E-04 0.5778 0.4222 0.5755 0.4245
2.3202E-04 0.5505 0.4495 0.5487 0.4513
5.6751E-04 0.5505 0.4495 0.5208 0.4792
5.3418E-06 0.5208 0.4792 0.5208 0.4792
9.4861E-05 0.5505 0.4495 0.5505 0.4495 232
Table B ■3. Cont'd
d0,l do,2 Rs.true
0.931 0.862 1.2
0.932 0.932 1.3
0.932 0.932 1.3
0.933 0.967 1.3
0.933 0.871 1.0
0.935 0.992 1.2
0.935 0.874 1.1
0.936 0.508 1.1
0.937 0.996 1.1
0.938 0.770 1.0
0.939 0.992 1.4
0.940 0.993 1.1
0.940 0.985 1.2
0.941 0.993 1.5
0.943 0.780 1.1
Rs.meas %ROi %RC>2
1.397 1.643 3.287
1.301 0.932 0.932
1.301 0.932 0.932
1.520 1.578 0.789
1.879 3.000 6.000
1.854 2.408 0.301
1.686 2.321 4.642
1.231 1.570 12.557
2.086 2.954 0.185
1.861 2.735 10.938
1.638 1.429 0.179
2.086 2.774 0.347
1.853 2.192 0.548
1.501 0.726 0.091
1.672 2.047 8.189
n blW i
alW!
b2w 2
a2W2
5.1316E-05 0.5208 0.4792 0.5185 0.4815
1.3438E-06 0.5000 0.5000 0.5000 0.5000
1.3438E-06 0.5000 0.5000 0.5000 0.5000
1.6558E-05 0.5208 0.4792 0.5205 0.4795
5.3573E-04 0.5778 0.4222 0.5731 0.4269
9.4861E-05 0.5505 0.4495 0.5504 0.4496
2.3202E-04 0.5505 0.4495 0.5469 0.4531
1.5892E-04 0.5208 0.4792 0.5003 0.4997
2.3053E-04 0.5778 0.4222 0.5777 0.4223
5.3573E-04 0.5778 0.4222 0.5680 0.4320
5.3418E-06 0.5208 0.4792 0.5208 0.4792
2.3053E-04 0.5778 0.4222 0.5776 0.4224
9.4861E-05 0.5505 0.4495 0.5503 0.4497
1.5230E-08 0.5000 0.5000 0.5000 0.5000
2.3202E-04 0.5505 0.4495 0.5429 0.4571 233
Table B •3. Cont'd
do.l do,2 Rs.true
0.943 0.606 1.0
0.944 0.986 1.1
0.944 0.778 1.2
0.944 0.944 1.3
0.944 0.972 1.4
0.945 0.564 1.1
0.946 0.973 1.2
0.947 0.787 1.2
0.947 0.974 1.1
0.948 0.987 1.4
0.949 0.372 1.0
0.950 0.628 1.1
0.951 0.951 1.1
0.951 0.951 1.2
0.951 0.283 1.1
Rs,meas %ROi %R02
1.821 2.465 19.721
2.085 2.596 0.649
1.389 1.339 5.355
1.519 1.320 1.320
1.401 0.718 0.359
1.093 0.902 7.215
1.852 1.986 0.993
1.200 0.793 3.173
2.083 2.421 1.210
1.638 1.218 0.304
1.719 2.186 34.983
1.640 1.784 14.274
2.080 2.247 2.247
1.850 1.789 1.789
1.133 1.226 19.609
n 1>1Wi
alW i
^2w 2
a2w 2
5.3573E-04 0.5778 0.4222 0.5564 0.4436
2.3053E-04 0.5778 0.4222 0.5774 0.4226
5.1316E-05 0.5208 0.4792 0.5160 0.4840
1.6558E-05 0.5208 0.4792 0.5202 0.4798
1.5498E-07 0.5000 0.5000 0.5000 0.5000
6.2522E-05 0.5001 0.4999 0.4950 0.5050
9.4861E-05 0.5505 0.4495 0.5499 0.4501
9.9295E-06 0.5001 0.4999 0.4995 0.5005
2.3053E-04 0.5778 0.4222 0.5771 0.4229
5.3418E-06 0.5208 0.4792 0.5207 0.4793
5.3573E-04 0.5778 0.4222 0.5222 0.4778
2.3202E-04 0.5505 0.4495 0.5338 0.4662
2.3053E-04 0.5778 0.4222 0.5763 0.4237
9.4861E-05 0.5505 0.4495 0.5494 0.4506
1.5892E-04 0.5208 0.4792 0.4524 0.5476 23
4
Table B -3. Cont'd
do.l do,2 ^s,true
0.952 0.997 1.3
0.952 0.905 1.3
0.954 0.908 1.3
0.954 0.910 1.1
0.955 0.997 1.5
0.955 0.651 1.2
0.956 0.913 1.2
0.956 0.995 1.3
0.956 0.978 1.4
0.956 0.406 1.1
0.957 0.989 1.5
0.957 0.997 1.2
0.957 0.837 1.1
0.959 0.995 1.2
0.960 0.990 1.3
Rs.meas %ROi %R02
2.009 1.781 0.111
1.301 0.655 1.309
1.517 1.095 2.189
2.074 2.073 4.147
1.755 1.071 0.067
1.373 1.077 8.613
1.845 1.601 3.202
2.009 1.632 0.204
1.638 1.031 0.516
1.562 1.530 24.484
1.501 0.528 0.132
2.276 2.011 0.126
2.061 1.900 7.600
2.276 1.892 0.236
2.008 1.490 0.373
n *>1W i
alW i
*>2w 2
nw 2
3.8783E-05 0.5505 0.4495 0.5505 0.4495
1.3438E-06 0.5000 0.5000 0.4999 0.5001
1.6558E-05 0.5208 0.4792 0.5195 0.4805
2.3053E-04 0.5778 0.4222 0.5747 0.4253
1.7232E-06 0.5208 0.4792 0.5208 0.4792
5.1316E-05 0.5208 0.4792 0.5106 0.4894
9.486IE-05 0.5505 0.4495 0.5483 0.4517
3.8783E-05 0.5505 0.4495 0.5505 0.4495
5.3418E-06 0.5208 0.4792 0.5206 0.4794
2.3202E-04 0.5505 0.4495 0.5077 0.4923
1.5230E-08 0.5000 0.5000 0.5000 0.5000
9.9195E-05 0.5778 0.4222 0.5778 0.4222
2.3053E-04 0.5778 0.4222 0.5716 0.4284
9.9195E-05 0.5778 0.4222 0.5777 0.4223
3.8783E-05 0.5505 0.4495 0.5504 0.4496 23
5
Table B -3. Cont'd
do,l do,2 Rs.true
0.960 0.960 1.4
0.960 0.960 1.4
0.960 0.846 1.2
0.961 0.714 1.1
0.961 0.995 1.5
0.962 0.990 1.2
0.962 0.849 1.3
0.963 0.963 1.4
0.963 0.982 1.3
0.964 0.982 1.2
0.964 0.522 1.1
0.965 0.717 1.2
0.965 0.734 1.2
0.965 0.470 1.2
0.965 0.448 1.1
Rs.meas %ROj %R02
1.401 0.511 0.511
1.401 0.511 0.511
1.835 1.421 5.685
2.035 1.725 13.800
1.755 0.921 0.115
2.275 1.774 0.444
1.512 0.900 3.598
1.637 0.868 0.868
2.007 1.354 0.677
2.274 1.658 0.829
1.975 1.546 24.743
1.199 0.538 4.304
1.815 1.249 9.989
1.334 0.853 13.649
1.085 0.597 9.544
n bl alWi
b2w2
a2w2
1.5498E-07 0.5000 0.5000 0.5000 0.5000
1.5498E-07 0.5000 0.5000 0.5000 0.5000
9.4861E-05 0.5505 0.4495 0.5458 0.4542
2.3053E-04 0.5778 0.4222 0.5646 0.4354
1.7232E-06 0.5208 0.4792 0.5207 0.4793
9.9195E-05 0.5778 0.4222 0.5776 0.4224
1.6558E-05 0.5208 0.4792 0.5182 0.4818
5.3418E-06 0.5208 0.4792 0.5204 0.4796
3.8783E-05 0.5505 0.4495 0.5502 0.4498
9.9195E-05 0.5778 0.4222 0.5773 0.4227
2.3053E-04 0.5778 0.4222 0.5478 0.4522
9.9295E-06 0.5000 0.5000 0.4989 0.5011
9.486IE-05 0.5505 0.4495 0.5405 0.4595
5.1316E-05 0.5208 0.4792 0.4964 0.5036
6.2522E-05 0.5001 0.4999 0.4892 0.5108 236
Table B-3. Cont'd
<*o,l do,2 Rs.true Rs.meas %ROj %R02 Q t>lW i
alW i
t>2w 2
a2W2
0.966 0.966 1 .2 2.272 1.543 1.543 9.9195E-05 0.5778 0.4222 0.5768 0.4232
0.966 0.992 1.5 1.755 0.788 0.197 1.7232E-06 0.5208 0.4792 0.5207 0.4793
0.967 0.967 1.3 2.006 1.225 1.225 3.8783E-05 0.5505 0.4495 0.5498 0.4502
0.967 0.869 1.3 1.301 0.453 1.811 1.3438E-06 0.5000 0.5000 0.4999 0.5001
0.968 0.998 1.4 2.164 1 .2 0 2 0.075 1.5856E-05 0.5505 0.4495 0.5505 0.4495
0.968 0.938 1 .2 2.267 1.429 2.858 9.9195E-05 0.5778 0.4222 0.5759 0.4241
0.969 0.984 1.5 1.501 0.380 0.190 1.5230E-08 0.5000 0.5000 0.5000 0.5000
0.969 0.939 1.4 1.636 0.725 1.449 5.3418E-06 0.5208 0.4792 0.5201 0.4799
0.969 0.558 1 .2 1.770 1.083 17.322 9.4861E-05 0.5505 0.4495 0.5278 0.4722
0.969 0.759 1.3 1.503 0.731 5.849 1.6558E-05 0.5208 0.4792 0.5153 0.4847
0.970 0.940 1.3 2.003 1 .1 0 1 2 .2 0 2 3.8783E-05 0.5505 0.4495 0.5491 0.4509
0.970 0.996 1.4 2.163 1.105 0.138 1.5856E-05 0.5505 0.4495 0.5505 0.4495
0.971 0.886 1 .2 2.259 1.315 5.262 9.9195E-05 0.5778 0.4222 0.5738 0.4262
0.971 0.998 1.3 2.466 1.367 0.085 4.2684E-05 0.5778 0.4222 0.5778 0.4222
0.971 0.986 1.5 1.755 0.671 0.336 1.7232E-06 0.5208 0.4792 0.5207 0.4793 237
Table B -3. Cont'd
do,l
c-i©•o
Rs.true
0.972 0.944 1.4
0.972 0.997 1.3
0.973 0.993 1.4
0.973 0.893 1.3
0.973 0.796 1.2
0.974 0.994 1.3
0.975 0.899 1.4
0.975 0.988 1.4
0.975 0.647 1.2
0.975 0.988 1.3
0.976 0.812 1.3
0.976 0.622 1.3
0.976 0.976 1.5
0.977 0.977 1.3
0.977 0.633 1.2
Rs,meas %ROi %RC>2
1.401 0.359 0.718
2.466 1.288 0.161
2.163 1.011 0.253
1.996 0.982 3.930
2.241 1.202 9.612
2.465 1.210 0.303
1.633 0.600 2.401
2.163 0.922 0.461
2.202 1.086 17.377
2.464 1.134 0.567
1.983 0.869 6.953
1.483 0.587 9.385
1.755 0.568 0.568
2.463 1.058 1.058
1.198 0.359 5.743
n blWi
alW i
*>2w 2
a2w 2
1.5498E-07 0.5000 0.5000 0.5000 0.5000
4.2684E-05 0.5778 0.4222 0.5777 0.4223
1.5856E-05 0.5505 0.4495 0.5504 0.4496
3.8783E-05 0.5505 0.4495 0.5475 0.4525
9.9195E-05 0.5778 0.4222 0.5695 0.4305
4.2684E-05 0.5778 0.4222 0.5776 0.4224
5.3418E-06 0.5208 0.4792 0.5193 0.4807
1.5856E-05 0.5505 0.4495 0.5503 0.4497
9.9195E-05 0.5778 0.4222 0.5599 0.4401
4.2684E-05 0.5778 0.4222 0.5775 0.4225
3.8783E-05 0.5505 0.4495 0.5444 0.4556
1.6558E-05 0.5208 0.4792 0.5089 0.4911
1.7232E-06 0.5208 0.4792 0.5206 0.4794
4.2684E-05 0.5778 0.4222 0.5772 0.4228
9.9295E-06 0.5000 0.5000 0.4978 0.5022 23
8
Table B •3. Cont'd
<*0,1 d0,2 **s,true
0.977 0.977 1.4
0.978 0.978 1.5
0.978 0.978 1.5
0.978 0.825 1.3
0.978 0.999 1.5
0.978 0.957 1.3
0.979 0.679 1.3
0.979 0.836 1.4
0.979 0.959 1.4
0.980 0.960 1.5
0.980 0.921 1.3
0.980 0.998 1.5
0.980 0.999 1.4
0.981 0.923 1.4
0.981 0.998 1.4
Rs,meas %ROi %R02
2.161 0.836 0.836
1.501 0.270 0.270
1.501 0.270 0.270
1.301 0.309 2.470
2.318 0.811 0.051
2.460 0.983 1.966
1.954 0.760 12.165
1.628 0.492 3.938
2.159 0.755 1.509
1.754 0.477 0.954
2.454 0.908 3.633
2.318 0.746 0.093
2.655 0.929 0.058
1.401 0.249 0.996
2.655 0.876 0.110
Q blWi
alWi
b2w2
a2w2
1.5856E-05 0.5505 0.4495 0.5501 0.4499
1.5230E-08 0.5000 0.5000 0.5000 0.5000
1.5230E-08 0.5000 0.5000 0.5000 0.5000
1.3438E-06 0.5000 0.5000 0.4998 0.5002
6.4826E-06 0.5505 0.4495 0.5505 0.4495
4.2684E-05 0.5778 0.4222 0.5765 0.4235
3.8783E-05 0.5505 0.4495 0.5373 0.4627
5.3418E-06 0.5208 0.4792 0.5178 0.4822
1.5856E-05 0.5505 0.4495 0.5496 0.4504
1.7232E-06 0.5208 0.4792 0.5204 0.4796
4.2684E-05 0.5778 0.4222 0.5752 0.4248
6.4826E-06 0.5505 0.4495 0.5505 0.4495
1.8367E-05 0.5778 0.4222 0.5778 0.4222
1.5498E-07 0.5000 0.5000 0.5000 0.5000
1.8367E-05 0.5778 0.4222 0.5777 0.4223 239
Table B •3. Cont'd
<*o,l <*o,2 **s,true
0.981 0.926 1.4
0.981 0.857 1.3
0.982 0.995 1.5
0.982 0.996 1.4
0.983 0.746 1.3
0.983 0.992 1.5
0.983 0.933 1.5
0.983 0.869 1.4
0.983 0.992 1.4
0.983 0.737 1.4
0.984 0.984 1.4
0.984 0.969 1.5
0.985 0.985 1.5
0.985 0.772 1.4
0.985 0.971 1.4
**s,meas %ROi %R02
2.155 0.677 2.707
2.441 0.834 6.669
2.318 0.685 0.171
2.655 0.825 0.206
2.415 0.759 12.139
2.317 0.626 0.313
1.752 0.398 1.592
2.146 0.602 4.817
2.655 0.774 0.387
1.616 0.399 6.387
2.653 0.724 0.724
1.501 0.190 0.380
2.317 0.570 0.570
2.128 0.531 8.491
2.651 0.675 1.349
Q blWi
alWi
*>2w2
a2w2
1.5856E-05 0.5505 0.4495 0.5487 0.4513
4.2684E-05 0.5778 0.4222 0.5725 0.4275
6.4826E-06 0.5505 0.4495 0.5505 0.4495
1.8367E-05 0.5778 0.4222 0.5777 0.4223
4.2684E-05 0.5778 0.4222 0.5666 0.4334
6.4826E-06 0.5505 0.4495 0.5504 0.4496
1.7232E-06 0.5208 0.4792 0.5200 0.4800
1.5856E-05 0.5505 0.4495 0.5467 0.4533
1.8367E-05 0.5778 0.4222 0.5776 0.4224
5.3418E-06 0.5208 0.4792 0.5146 0.4854
1.8367E-05 0.5778 0.4222 0.5774 0.4226
1.5230E-08 0.5000 0.5000 0.5000 0.5000
6.4826E-06 0.5505 0.4495 0.5503 0.4497
1.5856E-05 0.5505 0.4495 0.5425 0.4575
1.8367E-05 0.5778 0.4222 0.5770 0.4230 240
Table B-3. Cont’d
d0fl d0,2 Rs.true
0.986 0.770 1.3
0.986 0.972 1.5
0.986 0.889 1.5
0.986 0.946 1.4
0.987 0.999 1.5
0.987 0.896 1.4
0.987 0.900 1.4
0.987 0.949 1.5
0.987 0.998 1.5
0.988 0.997 1.5
0.988 0.820 1.4
0.989 0.909 1.5
0.989 0.994 1.5
0.989 0.820 1.5
0.989 0.957 1.5
Rs.meas %ROi %R02
1.300 0.208 3.321
2.315 0.516 1.032
1.749 0.329 2.631
2.647 0.626 2.502
2.845 0.630 0.039
1.401 0.171 1.365
2.639 0.577 4.613
2.312 0.465 1.858
2.845 0.596 0.075
2.845 0.562 0.140
2.621 0.528 8.443
2.307 0.416 3.324
2.845 0.528 0.264
1.742 0.269 4.308
1.501 0.132 0.528
n blW i
alWi
*>2W2
a2w 2
1.3438E-06 0.5000 0.5000 0.4996 0.5004
6.4826E-06 0.5505 0.4495 0.5499 0.4501
1.7232E-06 0.5208 0.4792 0.5191 0.4809
1.8367E-05 0.5778 0.4222 0.5761 0.4239
7.9032E-06 0.5778 0.4222 0.5778 0.4222
1.5498E-07 0.5000 0.5000 0.5000 0.5000
1.8367E-05 0.5778 0.4222 0.5744 0.4256
6.4826E-06 0.5505 0.4495 0.5494 0.4506
7.9032E-06 0.5778 0.4222 0.5778 0.4222
7.9032E-06 0.5778 0.4222 0.5777 0.4223
1.8367E-05 0.5778 0.4222 0.5707 0.4293
6.4826E-06 0.5505 0.4495 0.5481 0.4519
7.9032E-06 0.5778 0.4222 0.5777 0.4223
1.7232E-06 0.5208 0.4792 0.5174 0.4826
1.5230E-08 0.5000 0.5000 0.5000 0.5000
241
Table B -3. Cont’d
do,l do,2 ^s.true
0.989 0.989 1.5
0.990 0.840 1.5
0.990 0.980 1.5
0.991 0.963 1.5
0.991 0.931 1.5
0.991 0.862 1.4
0.992 0.874 1.5
0.993 0.941 1.5
0.995 0.921 1.5
Rs.meas %ROi %R02
2.844 0.495 0.495
2.294 0.369 5.897
2.842 0.462 0.925
2.839 0.430 1.720
2.834 0.398 3.182
1.401 0.115 1.845
2.822 0.366 5.851
1.501 0.091 0.726
1.501 0.062 0.986
n *>1Wi
alWi
*>2w2
a2w2
7.9032E-06 0.5778 0.4222 0.5776 0.4224
6.4826E-06 0.5505 0.4495 0.5456 0.4544
7.9032E-06 0.5778 0.4222 0.5773 0.4227
7.9032E-06 0.5778 0.4222 0.5767 0.4233
7.9032E-06 0.5778 0.4222 0.5756 0.4244
1.5498E-07 0.5000 0.5000 0.4999 0.5001
7.9032E-06 0.5778 0.4222 0.5732 0.4268
1.5230E-08 0.5000 0.5000 0.5000 0.5000
1.5230E-08 0.5000 0.5000 0.5000 0.5000
VITA
Name:
Bom:
Marital Status:
Children:
Formal Education:
Mark S. Jeansonne
May 12,1962 in Alexandria, Louisiana
Married
None
B.S., 1985, Louisiana State University
Major: Chemistry
Ph. D., 1990, Louisiana State University
Major: Chemistry (Chromatography)
Dissertation Title: " Chromatographic Peak Shape Analyis
and Modeling”, supervised by Professor Joe P. Foley.
2 4 3
DOCTORAL EXAMINATION AND DISSERTATION REPORT
Candidate: M ark S. J e a n s o n n e
Major Field: C h e m is try
Title of Dissertation: C h ro m a to g ra p h ic P eak S hape A n a ly s is and M o d e lin g
Approved:
M ajo r P ro fe sso r ai lha irm an
D ean of th e G ra d u a te School
EXAMINING COMMITTEE:
Date of Examination:
O c to b e r 15, 1990