Empower 3

Data Acquisition and Processing Theory

Guide
Revision A

Copyright © Waters Corporation 2010

All rights reserved
Copyright notice
© 2010 WATERS CORPORATION. PRINTED IN THE UNITED STATES OF
AMERICA AND IN IRELAND. ALL RIGHTS RESERVED. THIS
DOCUMENT OR PARTS THEREOF MAY NOT BE REPRODUCED IN ANY
FORM WITHOUT THE WRITTEN PERMISSION OF THE PUBLISHER.
The information in this document is subject to change without notice and
should not be construed as a commitment by Waters Corporation. Waters
Corporation assumes no responsibility for any errors that may appear in this
document. This document is believed to be complete and accurate at the time
of publication. In no event shall Waters Corporation be liable for incidental or
consequential damages in connection with, or arising from, its use.

Trademarks
Waters and Millennium are registered trademarks of Waters Corporation,
and ApexTrack, Empower, e-SAT/IN, and SAT/IN and “THE SCIENCE OF
WHAT’S POSSIBLE.” are trademarks of Waters Corporation.
Other registered trademarks or trademarks are the sole property of their
owners.

Customer comments
Waters’ Technical Communications department invites you to tell us of any
errors you encounter in this document or to suggest ideas for otherwise
improving it. Please help us better understand what you expect from our
documentation so that we can continuously improve its accuracy and
usability.
We seriously consider every customer comment we receive. You can reach us
at tech_comm@waters.com.
Contacting Waters
®
Contact Waters with enhancement requests or technical questions regarding
the use, transportation, removal, or disposal of any Waters product. You can
reach us via the Internet, telephone, or conventional mail.

Waters contact information

Contacting medium Information

Internet The Waters Web site includes contact
information for Waters locations worldwide.
Visit www.waters.com.
Telephone and fax From the USA or Canada, phone 800
252-HPLC, or fax 508 872 1990.
For other locations worldwide, phone and fax
numbers appear in the Waters Web site.
Conventional mail Waters Corporation
34 Maple Street
Milford, MA 01757
USA

Empower 3 software

Intended use
Use the Waters Empower 3 software for acquiring, processing, reporting, and
managing your chromatographic information.

Safety information
See the operator’s guides of the instruments or devices associated with this
software product for information on how to safely operate and maintain them.

iii
iv
Table of Contents
1 Data Acquisition ..................................................................................... 1-1
Overview ............................................................................................................. 1-2
What is data acquisition? ................................................................................ 1-2
What is processing? ......................................................................................... 1-2
Integration methods ........................................................................................ 1-2
Commonalities between ApexTrack and traditional integration.................. 1-3

Analog-to-digital conversion .......................................................................... 1-3

Data conversion ............................................................................................... 1-4
Data transfer and storage ............................................................................... 1-5

Detection sampling rates ................................................................................ 1-6

Determining the optimum sampling rate....................................................... 1-6
Displaying the data points .............................................................................. 1-7
Points Across Peak field .................................................................................. 1-7

Effects of data acquisition rate on disk space ........................................... 1-7

Reference ............................................................................................................ 1-8

2 ApexTrack Integration ......................................................................... 2-1

Features and capabilities ................................................................................ 2-2
ApexTrack features.......................................................................................... 2-2
How ApexTrack integrates peaks ................................................................... 2-3
Summary of processing method parameters ............................................. 2-3
Summary of timed events................................................................................ 2-6
Integration peak labels in ApexTrack ............................................................ 2-8

Apex detection ................................................................................................... 2-9

Detecting apices ............................................................................................... 2-9
Apex detection parameters............................................................................ 2-10
Auto-Peak Width ....................................................................................... 2-10
AutoThreshold ........................................................................................... 2-11
Obtaining the second derivative plot ............................................................ 2-11
Detecting the peak ......................................................................................... 2-12

Table of Contents v
 Resolved peaks and shoulder peaks ......................................................... 2-13
Round peaks .............................................................................................. 2-14
Second derivative apex and inflection point data in the Peaks table ......... 2-15
Baseline location ............................................................................................ 2-15
How ApexTrack determines the slope difference threshold ................... 2-16
How ApexTrack locates the baseline for an isolated peak ...................... 2-18
How ApexTrack locates the preliminary baseline for a cluster .............. 2-19
How ApexTrack determines the final cluster baseline ........................... 2-21
Effect of Liftoff % and Touchdown % on baseline location ..................... 2-21
Effect of changing Liftoff % and Touchdown % on cluster peaks ........... 2-23
Determination of peak boundaries ............................................................... 2-24
Sequence of operations .............................................................................. 2-24
Computing integration results...................................................................... 2-25
Peak area ................................................................................................... 2-25
Peak height ................................................................................................ 2-25
Retention time ........................................................................................... 2-26
Rules that determine which retention time method is used ................... 2-26
Retention time and height values for a manually adjusted peak ........... 2-28
Peak width parameter ................................................................................... 2-28
Auto-Peak Width ....................................................................................... 2-30
Using Auto-Peak Width ............................................................................ 2-30
Effect of varying the peak width parameter ............................................ 2-31
Detection threshold parameter ..................................................................... 2-31
AutoThreshold ........................................................................................... 2-33
Using AutoThreshold ................................................................................ 2-34
Using timed events ........................................................................................ 2-35
Peak detection events ............................................................................... 2-35
Peak integration events ............................................................................ 2-36
When timed events are active .................................................................. 2-36

References ........................................................................................................ 2-37

vi Table of Contents
3 Traditional Integration ........................................................................ 3-1
Features and capabilities ................................................................................ 3-2
Peak detection .................................................................................................. 3-2
Performing data bunching .......................................................................... 3-3
Determining peak start .............................................................................. 3-4
Determining preliminary peak apex .......................................................... 3-5
Determining peak end ................................................................................ 3-6
Determining peak width and threshold values ......................................... 3-7
Peak width parameter ................................................................................ 3-8
Threshold values ......................................................................................... 3-9
Peak Width and Threshold fields ............................................................. 3-11
Peak integration ............................................................................................ 3-11
Determining fused peaks .......................................................................... 3-12
Constructing the baseline ......................................................................... 3-14
Baseline adjustment ................................................................................. 3-16
Calculating peak retention time, height, and area ................................. 3-16
Retention time and height ........................................................................ 3-17
Area ............................................................................................................ 3-18
Peak rejection criteria ............................................................................... 3-19
Minimum area ........................................................................................... 3-19
Minimum height ........................................................................................ 3-19
5-point peak rejection ................................................................................ 3-19
Using timed events ........................................................................................ 3-20
Peak detection events ............................................................................... 3-20
Peak integration events ............................................................................ 3-20
Integration peak labels.................................................................................. 3-21

References ........................................................................................................ 3-22

4 Peak Identification and Quantitation of Sample Components ... 4-1

Features and capabilities ................................................................................ 4-2
Peak matching ................................................................................................... 4-2
Matching hierarchy ......................................................................................... 4-2
Calculating the match difference .................................................................... 4-3

Table of Contents vii

 Choosing the optimal peak match................................................................... 4-3
Single peak with a single component ......................................................... 4-3
Multiple peaks with a single component ................................................... 4-4
Single peak with multiple components ...................................................... 4-4
Shifting RT and RT windows .......................................................................... 4-4
RT Reference ............................................................................................... 4-4
Update RT ................................................................................................... 4-5

Quantitation ....................................................................................................... 4-7

Quantitation by calibration............................................................................. 4-7
Quantitation without calibration.................................................................... 4-8
Quantitation using sample weight and dilution ............................................ 4-9
Sample weight ............................................................................................. 4-9
Dilution ...................................................................................................... 4-10
Quantitation using injection volume ............................................................ 4-11
Quantitation using responses other than peak area and height ................ 4-12
External and internal standard quantitation .............................................. 4-13
External standard quantitation .................................................................... 4-13
Internal standard quantitation with separate standard and unknown
samples..................................................................................................... 4-18
Internal standard quantitation without separate standard and unknown
samples (RF internal standard) .............................................................. 4-23

Calibration curve fit types ............................................................................ 4-27

Single-level calibration curve........................................................................ 4-27
Linear through zero .................................................................................. 4-28
Response factor ......................................................................................... 4-29
Multilevel calibration matrix operations...................................................... 4-30
Matrix operations ...................................................................................... 4-31
Matrix operations example ....................................................................... 4-31
Multilevel calibration curves......................................................................... 4-33
Point-to-point fit ........................................................................................ 4-34
Determining component amount and/or concentration .......................... 4-35
Cubic spline fit .......................................................................................... 4-35
Linear fits .................................................................................................. 4-37
Quadratic fit .............................................................................................. 4-39

viii Table of Contents

 Cubic fit ..................................................................................................... 4-41
Fourth-order fit ......................................................................................... 4-42
Fifth-order fit ............................................................................................. 4-43
Multilevel forced-through-zero calibration curves....................................... 4-45
Weighting ....................................................................................................... 4-46
Statistics......................................................................................................... 4-48
Coefficient of determination ..................................................................... 4-49
Correlation coefficient ............................................................................... 4-49
Standard error of estimate of Y on X ....................................................... 4-49
Standard variance ..................................................................................... 4-50
Residual sum of squares ........................................................................... 4-50
Standard error of calibration .................................................................... 4-51
Calculated value and percent deviation of calibration points ................ 4-51
Percent relative standard deviation ......................................................... 4-52

References ........................................................................................................ 4-53

Index ..................................................................................................... Index-1

Table of Contents ix
x Table of Contents
1 Data Acquisition

This chapter describes how data are acquired and how analog data are
converted to digital data.
Contents:

Topic Page
Overview 1-2
Analog-to-digital conversion 1-3
Detection sampling rates 1-6
Effects of data acquisition rate on disk space 1-7
Reference 1-8

1-1
Overview

What is data acquisition?

A chromatogram is a series of detector responses, sampled uniformly across a
length of time. The elution of a compound results in a characteristic
chromatographic peak profile.
Integration is the process of calculating an area that is bounded in part or in
whole by a curved line. The goal of chromatographic peak integration is to
obtain retention times, heights, and areas of these peaks.
Peak integration uses two key algorithms: one that detects peaks and one that
determines their baselines. Once the peak apex and baseline are known, the
retention time (RT), height, and area can be calculated.

What is processing?
Processing is the manipulation of data to determine the identities and/or
amounts of separated components. It most often involves integrating
chromatographic peaks to calibrate standards and generate a calibration
curve, and to quantitate the source components.
Processing methods define how Empower software detects, integrates,
calibrates, and quantitates unprocessed, raw data from a 2D channel or a
2D-derived channel. The Processing Method wizard can help you create a
processing method, or you can interactively develop the processing method in
Review.
In Review, you start with unprocessed data acquired from a known standard
(Channels). You then create a multipoint calibration curve by using a range of
standard concentrations. Adding a processing method to a method set allows
the software to process raw data while it is being acquired.

Integration methods
Empower software offers two methods of peak detection and baseline
determination:
• ApexTrack™ integration – Detects a peak at its apex using the second
derivative of the chromatogram. Peak detection parameters are
independent of baseline location parameters. The baseline of each peak
is determined using the Liftoff % and Touchdown % parameters. A

1-2 Data Acquisition

 search algorithm starts from each peak’s apex and works downward and
outward to draw a baseline. Clusters are identified when the expanding
baselines meet and fuse.
To select the use of ApexTrack integration in a processing method, you
must first enable system and project policies.
• Traditional integration – Compares a slope against a fixed threshold to
identify the liftoff point of the peak (see Chapter 3, “Traditional
Integration”). Peak touchdown is determined by algorithms that proceed
down the chromatogram, finding features, valleys, apices, and so on,
until a suitable slope threshold is met.

Commonalities between ApexTrack and traditional integration

Although the algorithms that ApexTrack uses to detect peaks and to
determine baselines differ from those used by traditional processing,
significant functionality is the same for both traditional and ApexTrack
integration:
• Auto-Peak Width and AutoThreshold are supported.
• Baselines start and end on real, sampled data points.
• Timed events such as Inhibit Integration, Tangent Skim, Set Peak
Width, Set Minimum Height, Set Maximum Height, Set Minimum Area,
and Set Maximum Peak Width are supported.
• Peaks can be manually added or deleted.
• Peak start and stop markers can be manually changed.
• Allow Negative Peaks and Valley to Valley are supported.

Analog-to-digital conversion
The detector analog output signal must be converted to a digital
representation before Empower software can acquire and process data. This
section describes the sequential processes of data conversion and data transfer
and storage.

Analog-to-digital conversion 1-3

Data conversion
Analog-to-digital (A/D) conversion of detector data is performed in either of
two ways:
• A detector controlled over Ethernet, serial, or the IEEE-488 bus itself
performs the conversion.
• A detector not controlled by Empower software transmits an analog
output signal to a chromatographic interface (Waters e-SAT/IN™ or
busSAT/IN™ module). The magnitude of the transmitted signal (in
microvolts) corresponds to the amount of sample detected at a constant
rate.
The voltage range over which the incoming analog signal can vary is –0.25 to
+2.25 V. Each millivolt of signal represents 1,000 height counts (where
1 height count is equal to 1 µV). For example, with a detector set so that 1 AU
is equal to 1 V, a 1 AU peak is equal to a peak height of 1,000,000 height
counts (from the baseline, at 0 V).
The SAT/IN modules convert analog signals to digital signals at a specified
number of times per second (sampling rate).

Data acquisition process:

Digital
e-SAT/IN or signal
busSAT/IN (file)
Detector
Analog module Digital
(not under
signal signal
Empower Empower
control) (acquired busLAC/E
database
data) card
Detector under
Empower
control Empower software

1-4 Data Acquisition

Data transfer and storage
This sequence of events describes how the software transfers and stores data:
1. The converted digital signal is transmitted to one of the following
communication devices:
• BusLAC/E™ card
• Equinox serial card
• Serial hub
• COM port on the Empower computer
• Ethernet port on the Empower computer
Tip: You can use more than one type of communication device in an
Empower workstation, LAC/E32 Acquisition Server, or acquisition client.
2. The collected data are transmitted from the communication device to the
computer’s hard drive.
3. The digital voltage values are stored as acquired, unprocessed data. The
stored digital values are the raw data points of the chromatogram. Raw
data can be viewed in the Run Samples window as it is being acquired.
The Sample Sets, Injections, and Channels views of the Project window
represent the raw data in the current project.

Plot of acquired data points:

Raw data points

Detector output
signal (mV)

Time

Analog-to-digital conversion 1-5

Detection sampling rates
Empower software sets data collection frequency to the sampling rate you
specify in the associated instrument tab of the Instrument Method Editor. The
sampling rate needs to be high enough to provide a good representation of the
chromatogram, but not so high that you are collecting more data than you
need.
In liquid, gas, and ion chromatography, the best sampling rate produces a
minimum of 15 data points from peak start to peak end for the narrowest peak
of interest detected. The value of 15, as an optimum number of data points, is
determined by typical signal-to-noise ratios and the frequency content of an
exponentially modified Gaussian peak.
The amount of hard disk space required during data acquisition depends on
the sampling rate and run time. (For additional information on the theory of
data acquisition, see “Reference” on page 1-8.)

Determining the optimum sampling rate

You can use the following equation to determine the optimum sampling rate:
15
SR = ------
W
where:
SR = Sampling rate (points/second)
15 = Optimum number of data points from peak start to peak end
W = Measured width (in seconds) of the narrowest peak you want to detect
For example, for a measured peak width of 3 seconds, a sampling rate of 5
ensures data collection of 15 raw data points (where 15/3 = 5).
Tip: If the number of data points across the narrowest peak of interest is less
than 15, specify a faster sampling rate. Faster sampling rates produce more
data points and require a greater amount of disk space for data storage (see
“Effects of data acquisition rate on disk space” on page 1-7). If the calculated
sampling rate (as outlined above) is not available, select the next available
higher rate.

1-6 Data Acquisition

Displaying the data points
The Peaks tab in Review displays the Start Time, End Time, and Points
Across Peak for each integrated peak in the chromatogram. These are
reportable fields that you can display in any report group.

Points Across Peak field

Empower software calculates a Points Across Peak value for each integrated
peak in the chromatogram. Empower software builds later than 1154
32
calculate the value differently than does 1154 or Millennium software. The
calculation that the later builds use is the index of the data point closest to the
end time minus the index of the data point closest to the start time. This
improved calculation is accurate for peaks that have non-uniform data rates.

Effects of data acquisition rate on disk space

The amount of hard disk space required during data acquisition depends on
the sampling rate and run time. The table below illustrates the amount of
hard disk space needed to store a single channel of data collected at different
sampling rates and run times.

Effects of sampling rate and run time on hard disk space:

Sampling Data points Kilobytes per Approx.

Run Time
rate acquired per minute run time space used
(min)
(points/sec) minute run time (1024 bytes) (kilobytes)
1 60 0.23 10 2.3
5 300 1.17 10 12.0
20 1200 4.69 10 47.0

When you start data acquisition in Run Samples, the software determines the
current amount of disk space available. If disk space is insufficient, the
software warns you and does not start acquisition. If space becomes limited
during Run and Process or Run and Report modes, processing stops and
acquisition continues until all remaining disk space is used.

Effects of data acquisition rate on disk space 1-7

Reference
For further information on the theory of data acquisition, see the following
reference guide.
Dyson, Norman, Chromatographic Integration Methods, The Royal
Society of Chemistry, Thomas Graham House, Cambridge, 1998.

1-8 Data Acquisition

2 ApexTrack Integration

This chapter describes ApexTrack peak detection and integration

theory.
Contents:

Topic Page
Features and capabilities 2-2
Apex detection 2-9
References 2-37

2-1
Features and capabilities
ApexTrack peak detection and integration by Empower software includes the
following functions:
• Automatically determines appropriate peak width and detection
threshold values for the chromatogram unless already set in the
processing method.
• Detects peak apices in the chromatogram to determine the location of
peaks and shoulders.
• Integrates peaks to determine their retention times, areas, and heights.
The processing method defines the parameters that the software uses to
detect and integrate the peaks within the raw data file (channel).
Functionality in traditional processing that is identical to that in ApexTrack
processing is discussed in “Commonalities between ApexTrack and traditional
integration”, on page 1-3.

ApexTrack features
Empower software supports both traditional integration and ApexTrack
integration. The term “traditional integration” refers to how data is processed
(detecting peaks and locating baselines) by Millennium32 software. ApexTrack
processes data differently from traditional integration:
• ApexTrack detects a peak at its apex rather than at its liftoff point,
detecting the apex by its curvature (second derivative). In contrast,
traditional integration detects a peak at its liftoff by its slope (first
derivative).
• Because ApexTrack uses a curvature criterion, ApexTrack can reliably
detect shouldered peaks.
• The ApexTrack algorithm finds baselines by starting at each peak’s
apex, expanding a trial baseline downward and outward.
• ApexTrack determines the end-points of peak and cluster baselines by
internal slope comparisons. As a result, the location of the baseline is
independent of detector drift, and ApexTrack can reliably integrate
small peaks on sloped baselines.

2-2 ApexTrack Integration

 ApexTrack offers the following major features:
• Shoulder detection – Detects shoulders and round peak pairs.
• Gaussian skims – Skims multiple peaks within a cluster with Gaussian
profiles.
• Tangential skims – Skims multiple peaks within a a cluster by drawing
a line to tangentially skim the rider peak(s) from the parent peak.
• Negative peak detection and integration – Integrates negative peaks
and clusters containing negative (and positive) peaks. Supports shoulder
detection and Gaussian skimming of negative peaks.

How ApexTrack integrates peaks

ApexTrack integration consists of three major processes:
• Detects peaks – Detects a peak at its second derivative apex. The
baseline slope does not affect peak detection. All apices above the
detection threshold are detected. (Traditional integration detects a peak
at its liftoff point.) (See “Apex detection” on page 2-9).
• Determines baselines – Determines the baseline of each peak using the
Liftoff % and Touchdown % parameters (see “Baseline location” on
page 2-15 and “Determination of peak boundaries” on page 2-24).
Peak detection and baseline determination are independent of each
other.
• Calculates the peak area, height, and retention time (RT) – Integrates
peaks and determines height, and retention time by the quadratic fit
method (5-point or 3-point fit), or by the time of the second derivative
apex or the time of the highest point. (see “Computing integration
results” on page 2-25).

Summary of processing method parameters

The default processing method for ApexTrack integration differs from that for
traditional integration.

Features and capabilities 2-3

 ApexTrack default processing method:

The processing parameters available in ApexTrack include these:

• Integration Algorithm – Determines whether ApexTrack or traditional
integration is used. The use of ApexTrack is enabled by the system and
the project policies. Once enabled, you can switch a processing method
between traditional and ApexTrack.
• Start (min) and End (min) – ApexTrack can detect only apices between
the start and end times. The effect of start and end is similar to that of
Inhibit Integration. You can specify start and end values manually, or
leave them blank (default). A blank entry in start means start of data,
and a blank entry in end means end of data. The range is 0 to 655 min. If
both start and end values are not blank, the value in start must be less
than the value in end.
Two parameters control peak detection:
• Peak Width (sec) – The peak width, in seconds, at 5% of the peak height.
You can specify a value from 0.01 to 9999.99 seconds. Alternatively, you
can leave the field blank (default), in which case the software applies the
Auto-Peak Width algorithm to the region of the chromatogram between
the start and end time values specified in the processing method. If an

2-4 ApexTrack Integration

 Inhibit Integration event occurs at the beginning and/or end of the
chromatogram, the software applies the Auto-Peak Width algorithm to
data between the first and last data points outside these Inhibit
Integration events. The peak width value controls only the smoothing of
the data, the effect of which is to set the minimum spacing allowable
between peaks. Reducing the peak width value generally increases the
number of detectable peaks (see “Peak width parameter” on page 2-28).
• Detection Threshold – The peak-to-peak baseline noise, in millivolts, in
response units scaled to the same units as peak height (microvolts). You
can specify a value from 0.000 to 1.000e+090 μV. Alternatively, you can
leave the field blank (default), in which case the software applies the
AutoThreshold algorithm to the region of the chromatogram between
the start and end time values specified in the processing method. If an
Inhibit Integration event occurs at the beginning and/or end of the
chromatogram, the software applies the AutoThreshold algorithm to
data between the first and last data points outside these Inhibit
Integration events. Reducing the detection threshold value increases the
number of detectable peaks (see “Detection threshold parameter” on
page 2-31).
Tip: Generally, it is good practice to ensure that an Inhibit Integration
event, and Start and End times occur within the baseline region of a
chromatogram.
Two parameters control baseline location:
• Liftoff % and Touchdown % – These define the values ApexTrack uses
when it determines the slope difference threshold to identify the start
and end of peaks and peak clusters. Liftoff % is used for peak start and
Touchdown % for peak end. The default value is 0.000 for Liftoff % and
0.500 for Touchdown %, and the values can range from 0.000 to
100.000%. Increasing a value raises the point on the peak at which liftoff
or touchdown occurs (see “How ApexTrack determines the slope
difference threshold” on page 2-16).
Restriction: The maximum value of Liftoff % and Touchdown % allowed
in a GPC processing method is 5.000. The default value for Liftoff % and
Touchdown %, for GPC, is 0.000.
Two parameters reject peaks that fall below specified values:
• Minimum Area and Minimum Height. Both reject peaks based on
integration results.

Features and capabilities 2-5

Summary of timed events
ApexTrack peak detection and integration events are available in the Type
column of the Integration table. ApexTrack events are available with all types
of processing methods. The Merge Peaks event is also available with GPC
processing methods.
Tip: Refer to the Empower Online Information System for an overview of all
timed events and a description of each.

ApexTrack timed events:

2-6 ApexTrack Integration

You can enable the following ApexTrack timed events within the time range
specified in the table entry:
• Detect Shoulders – Enables the detection of shoulder and round peaks.
The peak boundary and the integration results reflect the detection of
the shoulder and/or round peak.
• Valley-to-Valley – Enables the replacement of cluster baselines with a
separate baseline for each peak.
• Gaussian skim – Enables the replacement of vertical drop lines with
Gaussian skims.
• Allow Negative Peaks – Enables negative peak detection.
• Tangential Skim – Enables the replacement of vertical drop lines by
drawing a line tangentially to skim (front or rear) the rider peak(s) from
the parent peak.
These events are disabled by default. You can enable them in any combination
and over any time range. However, these events cannot overlap themselves.
You can also enable the following events to modify the values already entered
in the processing method:
• Inhibit Integration – Further delimits the time range within which
peaks can be detected.
• Set Events - Modify the corresponding method values.
– Set Peak Width (sec)
– Set Detection Threshold
– Set Liftoff %
– Set Touchdown %
– Set Minimum Area
– Set Minimum Height
– Set Maximum Height
– Set Maximum Width (sec)
• Merge Peaks Event – Available for GPC type processing methods.
All events require a start time; the default is 0.000 minutes. For events that
have a stop time, you leave the field blank to indicate that the event is enabled
until the end of the run, or you can specify a time in minutes. By default, the
start and stop times have a precision of three significant figures, and the valid

Features and capabilities 2-7

 range of each parameter is 0.000 through 655.000 minutes. The start time
must be less than the stop time, unless the stop time is blank.

Integration peak labels in ApexTrack

Each identified peak in a chromatogram is given a two-letter label that
describes the peak’s start and end boundary. The boundaries of a peak can be
described by any pair of letters. These letters appear in the Int Type column of
the Peaks tab, in the Results and Main windows of Review.
If no integration events are enabled, each peak starts or ends on the baseline
(B) or in a valley (V) above the baseline. Each peak is labeled as follows:
• BB indicates a baseline-resolved peak.
• BV indicates a peak that starts a cluster.
• VB indicates a peak that ends a cluster.
• VV indicates a peak within a cluster.
ApexTrack integration supports four Int Type letters specific to ApexTrack:
Shoulder, Round, Gaussian Skim, and Crossover.

Integration peak labels on a chromatogram:

Name of peak start

Letter Description
or end
Baseline B The boundary of the peak is a baseline.
Valley V The boundary of the peak is a valley.
Shoulder
a S The boundary of the peak is a shoulder (a
mathematically derived valley point).
Rounda R The boundary of the peak is a round (see
““Round peaks”” on page 2-14).
Gaussian Skima G The boundary of the peak is a Gaussian skim.

Crossovera X The boundary of the peak occurs where the

signal intersects the baseline. The peaks on
either side of this point have opposite signs
(one is positive, one is negative).
Tangential Skim
a T The boundary of the peak is a tangent skim.
a. You must enable the appropriate timed event.

2-8 ApexTrack Integration

 Capitalization of each letter indicates ApexTrack performed the integration
automatically. Lowercase letters indicate manual integration.
For instance, a baseline label of Bb indicates that, while the peak start and
peak end are both baseline-resolved, the peak start was automatically
integrated by the software and you manually adjusted the peak end.
Tip: Refer to the Empower Online Information System for information on
manual integration.

Apex detection
When ApexTrack processing is invoked, the first process applied to the data is
apex detection. The apex of a peak is the point of maximum curvature. Apex
detection is based on measuring the curvature (the rate of change of the slope,
or second derivative) of the peak. ApexTrack uses the curvature at the peak
apex to detect peaks, and the algorithm associates a peak with each detected
apex. After detecting peak apices, ApexTrack locates the baselines (see
“Baseline location” on page 2-15).
Tip: In describing or plotting the curvature of a chromatogram, this guide
adopts a negative curvature convention. The second derivative chromatogram
measures the chromatogram’s curvature at each point, and it is scaled
(multiplied) by –1 before plotting. Given this convention, the apex of a positive
peak has positive curvature, and the apex of a negative peak has negative
curvature.

Detecting apices
ApexTrack software detects peaks as follows:
1. Obtains the peak width parameter.
2. Uses the peak width to obtain the second derivative smoothing filter.
3. Uses the second derivative filter to obtain the chromatogram’s second
derivative (curvature) plot.
4. Within the second derivative plot, locates the times of each maximum
(for positive peaks) or minimum (for negative peaks, when Allow
Negative Peaks is enabled). ApexTrack also records the values of the
second derivative at each maximum (for positive peaks) or minimum (for
negative peaks).

Apex detection 2-9

 5. Obtains the detection threshold parameter.
6. Applies the second derivative threshold to the maximum (for positive
peaks) or minimum (for negative peaks), and retains only the apices
whose curvatures are above the threshold (for positive peaks) or below
the threshold (for negative peaks). For negative peaks, the threshold
value is -1 × (the value specified as the detection threshold parameter).

Apex detection parameters

Two parameters control apex detection:
• Peak width
• Detection threshold
ApexTrack requires values for both parameters. You can manually specify
peak width and detection threshold values in the processing method, or
Auto-Peak Width and AutoThreshold can automatically determine the values.
The operation of Auto-Peak Width is summarized in “Peak width parameter”
on page 2-28. The operation of AutoThreshold is summarized in “Detection
threshold parameter” on page 2-31.

Auto-Peak Width
Auto-Peak Width is the automatic determination of peak width. If the peak
width is unspecified in the processing method, the software applies the
Auto-Peak Width algorithm to the region of the chromatogram between the
start and end time values specified in the processing method.
If an Inhibit Integration event occurs at the beginning and/or end of the
chromatogram, the software applies the Auto-Peak Width algorithm to data
between the first and last data points outside these Inhibit Integration events.
If an Inhibit Integration event occurs elsewhere, the software uses the data
within the Inhibit Integration event to calculate the peak width.
Auto-Peak Width measures the peak width, in seconds, at 5% of the height of
the largest peak in the second derivative. ApexTrack uses this value to
integrate the chromatogram and includes it in the integration result. You can
specify this value in the processing method, and save it for subsequent
processing.

2-10 ApexTrack Integration

 AutoThreshold
AutoThreshold is the automatic determination of the threshold. If the
detection threshold is blank in the processing method, the software applies
the AutoThreshold algorithm to the region of the chromatogram between the
start and end time values specified in the processing method.
If an Inhibit Integration event occurs at the beginning and/or end of the
chromatogram, the software applies the AutoThreshold algorithm to data
between the first and last data points outside these Inhibit Integration events.
If an Inhibit Integration event occurs elsewhere, the software uses the data
within the Inhibit Integration event to calculate the peak threshold.
AutoThreshold measures the peak-to-peak noise in the baseline segments
between peaks. AutoThreshold reports its value in microvolts. ApexTrack uses
this value to integrate the chromatogram and includes it in the integration
result. You can enter this value into the method and save it for subsequent
processing.

Obtaining the second derivative plot

ApexTrack detects peaks by calculating the second derivative of the
chromatogram. The top plot shows an ideal Gaussian peak. The bottom plot
shows its second derivative profile.

Apex detection 2-11

 Gaussian peak and its second derivative:

1 Key
1.The maximum
of the second
derivative is the
2 2 highest point in
both plots.
3 3 4 2. In the bottom
plot, the inflection
Gaussian peak points are where
the second
derivative crosses
1 0. The times of
these points are
carried up to the
top plot.
3.The upslope
4 points in the top
0 2 2
plot are curvature
3 3 minima in the
Second derivative plot of Gaussian peak bottom plot.
4.The second
derivative of the
chromatogram’s
baseline is 0.

Tip: All second derivative plots in this guide are multiplied by –1, so the apex
of a positive peak appears as a positive second derivative.

Detecting the peak

A positive peak has a single maximum point of curvature. The time of that
maximum identifies the peak’s apex.
Below the apex are the inflection points, which straddle the apex and have
zero curvature (pass through the zero line on the second derivative plot).
Continuing down the peak are the upslope points, which have a minimum of
curvature.
Finally, ApexTrack reaches the baseline, which has zero curvature. Even if
the baseline has significant drift, it still has zero curvature because the
curvature of a straight line is zero.

2-12 ApexTrack Integration

Resolved peaks and shoulder peaks
The following figure shows the second derivative of a simulated
chromatogram with an isolated peak on the left and three pairs of fused
peaks. ApexTrack detects peaks by taking the second derivative of the
chromatographic signal and locating the maxima. This process identifies all
seven peaks, including the shoulder peak and round peaks.

Baseline resolved peak, valley boundary, shoulder boundary, and round

boundary:
Baseline Fused peaks Fused peaks Fused peaks
resolved peak (valley) (shoulder) (round)

Unprocessed
AU

chromatogram

Second
AU

derivative

Integrated
AU

chromatogram

Minutes

Tip: All second derivative plots in this guide are multiplied by –1, so the apex
of a positive peak appears as a positive second derivative.
The arrows in the second derivative plot point to the local maxima, or second
derivative apices.”All fused peaks are detected by their second derivative
apex.
A valley drop line is added between fused peaks when there is a minimum in
the chromatographic signal between the two second derivative apices.
A shoulder drop line is added between fused peaks when there is no minimum
in the chromatographic signal between the two second derivative apices.

Apex detection 2-13

 A round drop line is added between fused peaks where there is no minimum in
the chromatographic signal between the two second derivative apices and the
minimum in the second derivative plot (between the apices) is greater than
zero.

Round peaks
A pair of round peaks occurs when peaks of nearly equal height fuse at low
(but not zero) resolution. When two peaks are not totally resolved, they could
have a valley between them. At lower resolution, that boundary will be a
shoulder (for two peaks of differing heights) or a round (for two peaks of
similar height).
Tip: When the Detect Shoulders timed event is enabled, both shoulder peaks
and round pairs are detected. Shoulder boundaries are labeled by (S) and
round boundaries are labeled by (R).
The following figure shows two unresolved pairs of peaks illustrating the
formation of a pair of round peaks. The pair on the left results in a valley
boundary. The pair on the right results in a pair of round peaks. A pair of
round peaks occurs when, at lower resolution, the valley disappears and the
apex appears to be rounded or flattened. For valley (and shoulder) boundaries,
each apex is straddled by a pair of inflection points (indicated by diamonds).
The defining characteristic of the round boundary is that the two apices share
the same single pair of inflection points.

Fused peaks with valley (left) and fused round peaks (right):

Simulated
chromatogram
Valley Round

Second
derivative

Negative curvature Positive curvature

2-14 ApexTrack Integration

Second derivative apex and inflection point data in the Peaks
table
The time of the second derivative apex for each peak appears in the Peaks
table in the column labeled 2nd Derivative Apex. This time is generally not
the same as the peak retention time. If the time of the second derivative apex
is used for the retention time, a peak code of I20 appears. For tailed peaks, the
second derivative apex time generally precedes the retention time.
The displayed value for the second derivative apex of a particular peak does
not change with changes to the processing method unless the Peak Width
parameter is changed. The only exception is when there are shoulder and/or
round boundaries that have been removed from the peak because the Detect
Shoulders event is not enabled. If a peak has “hidden” round or shoulder
boundaries, the second derivative apex displayed for it may change with
changes to the processing method.
Inflection points straddle the apex and have zero curvature (pass through the
zero line on the second derivative plot). The time between a peak’s inflection
points appears in the Peaks table, in the column labeled Inflection Point
Width (sec).

Baseline location
After valid apices are found, ApexTrack determines the baselines associated
with them.

To determine the baseline for positive peaks, ApexTrack completes these

steps:
1. Initially draws a baseline between the inflection points of each peak.
2. Draws lines tangent to the inflection points.
3. Determines the slope differences between the inflection point baseline
and a tangent line at each inflection point (upslope and downslope).
4. Determines the slope difference thresholds. The peak start slope
difference threshold is defined as (Liftoff % × slope difference)/ 100. The
peak end slope difference threshold is defined as (Touchdown % × slope
difference)/100.
– A Liftoff % or Touchdown % of 100% is at the inflection point.
– A Liftoff % or Touchdown % of 0% merges with baseline noise.

Apex detection 2-15

 5. Expands the baselines until the slope difference threshold criteria are
met for peak start and peak end.

How ApexTrack determines the slope difference threshold

The processing method requires values for the following peak integration
parameters:
• Liftoff % (default value is 0.000)
• Touchdown % (default value is 0.500, except GPC default value is 0.000)

Baseline parameters:

The algorithm then computes two slope difference thresholds for each peak
based on the Liftoff % and Touchdown %.

To calculate the slope difference thresholds, ApexTrack completes these

steps:
1. Identifies the inflection points that straddle a peak apex.
2. Draws a tangent at each inflection point and draws a baseline that
connects the inflection points.

2-16 ApexTrack Integration

Inflection point baseline:

Upslope Downslope
tangent tangent

Inflection point
baseline

3. Computes two slope differences:

– Δm1, between the slope of the tangent at the upslope inflection point
and the inflection point baseline.
– Δm2, between the tangent at the downslope inflection point and the
inflection point baseline.

Computing slope differences:

Δm1 Δm2

ApexTrack computes two slope difference thresholds (Tstart and Tend) using
Baseline % Thresholds from the method:
• Tstart = (Δm1 × Liftoff %)/100
• Tend = (Δm2 × Touchdown %)/100

Apex detection 2-17

 How ApexTrack locates the baseline for an isolated peak
For each peak, ApexTrack expands the baselines downward and outward until
the slope difference threshold criteria are met. At each point in the expansion,
the slope difference threshold criteria are tested.

Computing the final baseline:

Δm1 Δm2

Tstart = Angle between Tstop = Angle between

arrow lines arrow lines

This figure shows the simulation of a tailed peak and uses it to illustrate how
ApexTrack locates the baseline of a baseline-resolved peak. The initial
baseline is the inflection point baseline. The baseline expands as it moves
down the peak and the slope difference thresholds are tested. With each step,
the ends of the baseline become more tangent to the peak. The expansion
stops when the slope difference thresholds are met at both ends.

2-18 ApexTrack Integration

Example of search for baseline in a tailed peak:

Inflection point
baseline

Final baseline

How ApexTrack locates the preliminary baseline for a cluster

ApexTrack uses a slightly different method to determine the baseline for
cluster peaks.

To determine the preliminary baseline for cluster peaks:

1. ApexTrack expands each peak’s baseline until its ends meet the slope
difference threshold criteria. If peaks are not resolved as the baselines
expand, they overlap. The following figure shows the preliminary
overlapped baselines for a two-peak cluster.

Apex detection 2-19

 Preliminary baselines in cluster peaks:

Preliminary
baseline Preliminary baseline

2. Identifies the valleys by the overlap of the expanded baselines.

3. Replaces the two overlapped preliminary baselines with one fused
baseline that starts at the beginning of the first preliminary baseline
and ends at the last preliminary baseline on the cluster.

Cluster baselines:

Valley drop
line

Preliminary fused
baseline

Final baseline after

expansion

2-20 ApexTrack Integration

How ApexTrack determines the final cluster baseline
After a baseline is fused, the slope difference thresholds are tested at the
beginning and end of the cluster baseline.

If the slope difference thresholds have not been met, ApexTrack behaves as
follows:
1. Expands the cluster baseline as before.
2. Stops the expansion when the slope difference thresholds are met:
– Slope difference threshold Tstart = (Δm1 × Liftoff %)/100
– Slope difference threshold Tend = (Δm2 × Touchdown %)/100
3. Positions valley drop lines at the point of minimum height above the
final baseline.

Effect of Liftoff % and Touchdown % on baseline location

The location of the baseline is controlled by Liftoff % and Touchdown %. If
both are set to 0.000%, the resulting baselines are tangent to the detector
baseline. If both are set to 1.000%, then the slope difference thresholds are
1.000% of the inflection points’ slope differences (Δm1 and Δm2). The resulting
baselines terminate at about 1.000% of the peak’s height. If both are set to
100%, the baseline used for each peak is its inflection point baseline.
Because ApexTrack uses a percentage to calculate the slope difference
threshold, the threshold computed for a series of peaks is proportional to peak
height. Thus, big peaks have big slope difference thresholds and small peaks
have small slope difference thresholds, which allows a single method to
successfully integrate peaks of varying sizes using the same Liftoff % and
Touchdown % values.
The following figure shows an example of peaks whose height ratios – 1, 1/10,
and 1/100 – use the default values Liftoff % = 0.000 and Touchdown % =
0.5000.
• Liftoff is the same for each peak.
• Touchdown is the same for each peak.
• Touchdown is well positioned for each peak.

Apex detection 2-21

 Touchdown of largest peak:

Touchdown of
largest peak

Initial peak 1

1/10 peak

1/100 peak

Zooming in to focus on the middle (1/10) peak, touchdown is well positioned,

although the slopes are different. The end point for the middle peak occurs at
the same relative point to the peak tail.

Touchdown of 1/10 peak:

1/10 peak

1/100 peak

Zooming in to focus on the smallest (1/100) peak, touchdown is well positioned,

although the slopes are different. The end point for the third peak occurs at
the same relative point to the peak tail. The slope difference threshold for the
largest peak is 100 times that of the smallest peak.

2-22 ApexTrack Integration

Touchdown of 1/100 peak:

1/10 peak

1/100 peak

Effect of changing Liftoff % and Touchdown % on cluster peaks

Changing the Liftoff % and Touchdown % changes the baseline location, and
can cause the time of a valley drop line to change. This is because the drop line
is set at the lowest point relative to the current baseline.
Changing Liftoff % and Touchdown % can cause shoulder drop lines to become
valley drop lines, and vice versa. Determining whether a drop line is a
shoulder or valley depends upon the slope of the baseline. If there is a
minimum between the two adjoining peak apices with respect to the current
baseline, the drop line is a valley boundary (V). If there is no minimum, the
drop line is a shoulder boundary (S). Shoulder drop lines can become valley
drop lines (and vice versa) when changes in Liftoff % and Touchdown %
change the slope of the baseline.
If Detect Shoulders is not enabled, peaks can be added or deleted as you
change Liftoff % and Touchdown %. The determination of whether a peak is a
shoulder is made using the current baseline. If Detect Shoulders is not
enabled, a shoulder peak will not appear. However, if you change Liftoff % and
Touchdown %, the slope of the baseline changes. The boundary of that peak
may appear as a valley with respect to the new baseline and the peak will
appear. In general, the disappearance of a shoulder boundary has the effect of
combining the two adjoining peaks into one.

Apex detection 2-23

Determination of peak boundaries
After detecting the apices and locating the baselines, ApexTrack identifies the
start and stop of each peak. The default boundaries are baseline and valley. If
a peak is baseline resolved, the start and stop are the baseline’s end points,
and are labeled by a B. If a peak is in a cluster, the boundaries between peaks
are vertical drop lines placed at valley points and are labeled “V “(see
“Integration peak labels in ApexTrack” on page 2-8).
Timed events enable the following additional types of boundaries:
Detect Shoulders – In regions where shoulder detection is enabled, the
boundaries for shoulder and round peaks are vertical drops, labeled “S”and
“R.”
Gaussian Skim – In regions where Gaussian skimming is enabled, a Gaussian
profile replaces vertical drop lines and the new peak boundary is labeled “G.”
Tangential Skim – In regions where tangential skimming is enabled, a line
drawn tangentially to skim (front or rear) the rider peak(s) from the parent
peak, replaces vertical drop lines and the new peak boundary is labeled “T.”
Negative Peaks – If a peak cluster contains only negative peaks, and Allow
Negative Peaks is enabled, peak start and end boundaries are labeled “B” and
“V”. If Detect Shoulders or Gaussian Skim is enabled in these regions, the S,
R, and G boundaries are also allowed.
Crossover – If a cluster contains positive and negative peaks, the
chromatographic signal will intersect the baseline between these adjoining
peaks. In this case, the boundary is a crossing point and is labeled “X.”

Sequence of operations
The determination of peaks, baselines, and boundaries involves three
processes (apex detection, baseline location, and boundary determination)
with the processing of timed events (Allow Negative Peaks, Detect Shoulders,
Valley-to-Valley, Gaussian Skim, and Tangential Skim).

The order of operation of these processes is as follows:

1. Detect apices.
2. Process Allow Negative Peak events.
3. Locate baselines.
4. Locate peak boundaries.

2-24 ApexTrack Integration

 5. Process Valley-to-Valley events.
6. Process Detect Shoulders events.
7. Process Gaussian Skim events.
8. Process Tangential Skim events.

Computing integration results

Once the apices are detected, the baselines are placed, and the boundaries are
identified, ApexTrack obtains the integration results for each peak. Peak area,
retention time, and height are all computed using the baseline-corrected
signal.
Tip: ApexTrack integration uses the baseline-corrected signal to determine
retention time. In contrast, Traditional integration uses the uncorrected
signal to determine retention time. For relatively flat baselines, the retention
time values match or differ by negligible amounts. For peaks on highly sloped
baselines, the ApexTrack calculation, based on the baseline-corrected signal,
is more accurate.
If one or both of a peak’s boundaries is a Gaussian skim, then a portion of the
peak profile and/or baseline is replaced by the skim, prior to baseline
correction. For example, if a peak generates a skim, the skim profile replaces
the responses for the larger (parent) peak between the start and stop time of
the skim. This same profile becomes the baseline of the adjoining, smaller
(child) peak that is being skimmed.

Peak area
Peak area is obtained by applying Simpson’s rule. The contribution to peak
area from each adjoining pair of sample points is the average of the
baseline-corrected responses at those sample points, multiplied by the sample
period (the time between the adjoining sample points).

Peak height
Peak height is the value of the baseline-corrected response at the retention
time.

Apex detection 2-25

 Retention time
ApexTrack determines the retention time and height in one of four ways,
depending on the peak boundaries and the properties of the peak shape:
• A 5-point fit of a quadratic curve to the points at the peak apex.
• 3-point fit of a quadratic curve to the points at the peak apex.
• The time of the second derivative apex.
• The time of the highest point.
Tips:
• Under most conditions, the 5-point quadratic fit is used to
determine a peak’s height and retention time and no processing code
is reported.
• The 3-point fit and the time of the second derivative are unique to
ApexTrack and are not implemented in traditional integration.
• In ApexTrack integration, the 3-point and 5-point fit is to the
baseline-corrected signal. In traditional integration, only the 5-point
fit is implemented, and that fit is to the uncorrected signal.

Rules that determine which retention time method is used

For each peak, ApexTrack carries out a hierarchy of tests to determine the
retention time method.

The tests and the order in which they are done is as follows:
1. The retention time of the second derivative apex is used:
– When either boundary of the peak is a round (R).
– When the highest point on the baseline-corrected signal is at a peak
boundary. In general this ensures that the retention time of a
shoulder peak is obtained from the second derivative apex.
The processing code I20 is included in the integration result of the
peak whenever the retention time and height reported in the result
are calculated at the second derivative apex.
If the retention time of the second derivative apex cannot be used
because it falls outside the peak boundary, the retention time of the
highest point is used instead. The processing code I23 is included in
the integration result, signifying that the attempt at using the
second derivative apex retention time failed.

2-26 ApexTrack Integration

2. If the peak does not fit the criteria for the first test, the 3-point fit is
used:
– When there are fewer than four sample points within the inflection
point width of the peak.
The processing code I19 is included in the integration result of the
peak whenever the retention time and height reported in the result
are calculated from a 3-point fit.
If the retention time falls outside the 3 points used for the fit, the
second derivative apex value is used, and the processing code I22 is
included in the integration result to indicate that a 3-point fit was
attempted but failed. The processing code I20 is included in the
integration result of the peak whenever the retention time and
height reported in the result are calculated at the second derivative
apex.
If the retention time of the second derivative apex falls outside the
peak boundary, the retention time of the highest point is used
instead. The processing code I23 is included in the integration
result, signifying that the attempt at using the second derivative
retention time failed.
3. If the peak does not fit the criteria for either test, the 5-point fit is used.
– No processing code is reported.
The 5-point fit may fail if these conditions apply:
• The first or last point of the 5 points used for the fit lies outside the
start and stop times of the peak.
• The retention time from the fit falls outside the 5 points used for the
fit.
The processing code I21 is included in the integration result
signifying that a 5-point fit was attempted but failed. In either case
a 3-point fit is then attempted. The processing code I19 is included
in the integration result of the peak whenever the retention time
and height reported in the result are calculated from a 3-point fit.
If the retention time from the 3-point fit falls outside the 3 points
used for the fit, then the processing code I22 is included in the
integration result, signifying that a 3-point fit was attempted but
failed. In this case the second derivative value is attempted. The
processing code I20 is included in the integration result of the peak

Apex detection 2-27

 whenever the retention time and height reported in the result are
calculated at the second derivative apex.
If the retention time of the second derivative apex falls outside the
peak boundary, the retention time of the highest point is used
instead. The processing code I23 is included in the integration
result, signifying that the second derivative retention time was
attempted but failed.

Retention time and height values for a manually adjusted peak

Baseline location and peak boundaries are needed to determine a peak’s
retention time, height, and area. When a manually determined baseline
location and peak boundaries coincide with the automatically determined
values, the manually determined retention time, height, and area values will
be identical to the automatically determined values.
Tip: If retention time and height are computed using the 2nd derivative apex,
an I20 integration event is reported. If an I20 peak is manually adjusted, the
second derivative value is unavailable, so the software uses a different set of
rules to determine how to calculate retention time and height. As a result, the
retention time and height will be determined by a 5-point fit, a 3-point fit, or
by time of the highest point.
If Detect Shoulders is enabled, an I20 event can occur if a peak is determined
to be a shoulder or round. If Detect Shoulders is not enabled, an I20 event can
occur for a narrow, low level peak.

Peak width parameter

Different separations can produce peaks whose widths vary over a wide range,
from seconds to minutes. ApexTrack requires the peak width as input into the
processing method. The peak width sets the widths of the digital filters, which
are used internally to obtain the smoothed, first and second derivative
chromatograms. In ApexTrack, the role of the peak width is solely to
determine these filter widths.
Tip: The number of points in the filtered chromatograms is the same as in the
original chromatogram. ApexTrack, unlike traditional integration, does not
bunch data points. ApexTrack processing uses either the original or the
filtered chromatograms.
The width of a filter determines how much smoothing is included. Wider
filters produce increased smoothing in the smoothed, first or second derivative

2-28 ApexTrack Integration

chromatogram. Smoothing removes high frequency components (noise),
leaving frequencies that correspond to actual chromatographic features
(peaks).
Different conventions can be employed to express the width of a
chromatographic peak. The ApexTrack processing method expects as input
the width of a peak measured at 5% of its height, expressed in seconds.
Tip: The most universal measure of the width of a distribution is its standard
deviation (SD). For a Gaussian peak, the chromatographic peak width is
defined as 4.0 × SD.
In ApexTrack, you can measure the peak width at 5% height by visual
inspection and specify this value in the method, or you can use Auto-Peak
Width to determine the peak width automatically.

Peak width definitions in a Gaussian peak:

Inflection width,
2.0 × SD

Half height, 2.4 × SD

Chromatographic
peak width, 4.0 × SD

5% height,
1% height,
4.9 × SD
6.1 × SD

After a peak is integrated, Empower 3 software recalculates its inflection

points and tangent lines at the inflection points. To do so, it applies the data
between the peak’s start and end times after those data are re-smoothed
based on an internally determined peak width (not the peak width specified in
the peak width parameter). Improving the placement of the inflection points
and tangent lines improves the accuracy of the System Suitability Width @
Tangent value and that of the fields that depend on its value (USP Resolution,
USP Plate Count, and Relative Resolution).

Apex detection 2-29

 Auto-Peak Width
Auto-Peak Width automatically measures the peak width in a region of a
chromatogram containing one or more peaks. Auto-Peak Width selects the
peak having the largest magnitude second derivative in that region and
determines its peak width. It does this by accurately measuring the time
interval between the inflection points. The time interval is then multiplied by
a factor of 4.89549/2, which gives the width at 5% for a Gaussian peak.
You can select the region used to calculate Auto-Peak Width two ways:

• Zoom in on a region of the chromatogram, and click (Set Processing

Method Peak Width). The peak width is entered into the processing
method. When you process the data, ApexTrack reports and applies this
value.
• Leave the Peak Width field in the processing method blank. When you
process the data, Auto-Peak Width uses the regions between the Start
and End times, and the region between any Inhibit Integration events at
the start and end of the chromatogram, to determine the peak width.
ApexTrack reports and uses this value.

Using Auto-Peak Width

If the Peak Width field is blank, Auto-Peak Width uses the data between start
and end, and the region between any Inhibit Integration events at the start
and end of the chromatogram, to determine the peak width. Choose the start
and end times to exclude injection artifacts and artifacts associated with the
return-to-initial conditions. For example, if the void volume is included,
ApexTrack might select an injection artifact as the highest peak, which
generally gives a width that is too small.
Even if start and end times are properly chosen, Auto-Peak Width can give
inaccurate results under the following circumstances:
• If the largest peak is saturated, the peak width value can be too large.
• If the largest peak is coeluting with another peak, the peak width value
can be too large.
• If the largest peak is noisy, Auto-Peak Width can measure the width of a
noise artifact and produce a width that is too small.
To address these problems, zoom in on a region of the chromatogram that
contains a valid peak with a valid width, then click (Set Processing Method
Peak Width). The peak width specified in the processing method. When you

2-30 ApexTrack Integration

 process the data with this method, ApexTrack reports and uses the specified
value. Generally, the peak width obtained from a reference separation is
relevant for subsequent separations.
When you enable the Inhibit Integration event, Auto-Peak Width uses the
data between the first and last good data points outside the Inhibit
Integration event. For example, if you enter an Inhibit Integration start time
at 0 minutes and end time at 1 minute, and then another Inhibit Integration
event starting at 5 minutes to end, Auto-Peak Width is calculated using the
first good data point after 1 minute and the last good data point before
5 minutes.

Effect of varying the peak width parameter

On baseline-resolved peaks, the variation of width about the Auto-Peak Width
value by a factor of up to 1.5 should have little effect on peak detection or
baseline placement.
For complex chromatograms with coeluted peaks that span a range of peak
heights, changing the width by a factor of 1.5 (and redetermining the
threshold with AutoThreshold) can change which low-level peaks are
detected.
Increasing the width by about a factor of 2 (and redetermining the threshold
with AutoThreshold) should result in detection of peaks near the baseline that
might otherwise be missed. However, this level of increase also reduces the
number of detected shoulders.

Detection threshold parameter

To distinguish chromatogram peaks from noise peaks, ApexTrack requires a
detection threshold as input into the processing method. The software
interprets this value as the baseline’s peak-to-peak noise, expressed in
microvolts.
Two sources of noise can add fluctuations to the chromatographic signal:
• The irreducible statistical fluctuations inherent in any detection
process.
• The detector’s response to the solvent stream.
In ApexTrack, the baseline noise in the second derivative chromatogram is
relevant in distinguishing peaks from baseline artifacts. The following figure

Apex detection 2-31

 shows the second derivative of a chromatogram with two chromatographic
peaks and the baseline noise.

Second derivative showing baseline noise:

Second peak Detection threshold

value
First peak
Baseline noise

Zero curvature
line

The noise in the baseline of the second derivative chromatogram is

proportional to the noise in the baseline of the original chromatogram. Thus,
the software can obtain a second derivative threshold from the peak-to-peak
noise in the original chromatogram’s baseline.
Internally, the threshold entered into the processing method is converted to a
value of curvature (microvolts/sec/sec). This converted value is applied to the
second derivative chromatogram. Only peaks whose second derivative rises
above the curvature threshold are accepted as valid detections of peak apices.
A properly chosen threshold rejects all artifacts due to detector noise and
accepts only valid peaks.
In ApexTrack, you can measure the baseline’s peak-to-peak noise by visual
inspection and specify the value in the method, or you can use AutoThreshold
to measure the peak-to-peak noise automatically.

2-32 ApexTrack Integration

Example of peak-to-peak baseline noise:

1 × SD 4 × SD =
peak-to-peak noise

AutoThreshold
AutoThreshold automatically determines the baseline noise amplitude.
AutoThreshold requires a region of a chromatogram and a value for peak
width as input. AutoThreshold identifies the regions within the
chromatogram that are free of peaks, and estimates the peak-to-peak noise in
those regions. The selected region might contain one or more peaks, or no
peaks. For an accurate measurement, the selected region must contain a
segment of the chromatogram that is free of peaks and is at least one peak
width wide.
As with Auto-Peak Width, you can select the region that is input to
AutoThreshold in two ways:

• Zoom in on a region of the chromatogram, and click (Set Processing

Method Threshold) to set a detection threshold value in the processing
method. (This button is active only if a peak width is specified in the
processing method.) When you process the data, ApexTrack reports and
uses this value.
• Leave the Threshold field in the processing method blank. When you
process the data, AutoThreshold uses the regions between the start and
end times and between Inhibit Integration events at the beginning and
end, to determine the detection threshold. ApexTrack reports and uses
this value.
AutoThreshold determines the threshold by examining the noise regions in
the second derivative chromatogram. It converts the noise in the second
derivative chromatogram to the equivalent peak-to-peak noise threshold that
would be seen in the original baseline, and reports this value.

Apex detection 2-33

 The following figure shows the integration of a sample chromatogram whose
baseline is shown in the following figure. AutoThreshold reports the threshold
as 23.00 μV.

Automatic measurement of baseline noise:

AU

Minutes

You can also manually zoom in on a baseline and estimate the peak-to-peak
noise visually. A straight line with a positive or negative slope indicates drift,
and you can estimate the peak-to-peak height. The magnitude of peak-to-peak
noise in this example is approximately 25 μV. You can specify this value in the
ApexTrack method.

Manual measurement of baseline noise:

AU

Minutes
1 Minutes, -0.0005753 AU

Using AutoThreshold
If the Detection Threshold field is blank, AutoThreshold uses the data
between start and end times and between Inhibit Integration events to
determine the detection threshold. Choose the start and end times to exclude

2-34 ApexTrack Integration

 regions that can have baseline noise that differs from the noise in the
separation region.
When you enable the Inhibit Integration event at the beginning and/or end of
the chromatogram, AutoThreshold uses the data between the first and last
good data points outside the Inhibit Integration event. For example, if you
enter an Inhibit Integration start time at 0 minute and end time at 1 minute,
then another Inhibit Integration event starting at 5 minutes to end,
AutoThreshold is calculated using the first good data point after 1 minute and
the last good data point before 5 minutes.
Even if the processing method parameters are properly chosen, there is a
circumstance under which AutoThreshold can give inaccurate results. In a
chromatogram containing many components, there may be no region that is
free of peaks. In this case, AutoThreshold may give a value that is too high, so
that valid peaks are not detected. To address this problem, zoom in on the
baseline region and set the Detection Threshold to the height of the smallest
peak. It can be useful to temporarily set the Detection Threshold to 0.0, so you
can see all the peaks that previously were not detected.

Using timed events

Empower software supports timed events for peak detection and integration.
Tip: Refer to the Empower Online Information System for an overview of all
timed events, and a description of each timed event.

Peak detection events

In ApexTrack, timed events can modify the detection of peaks. You select
ApexTrack peak detection events from the Type column of the Integration
table.
ApexTrack events that affect peak detection are as follows:
• Inhibit Integration Event
• Detect Shoulders Event
• Allow Negative Peaks Event
• Set Events, as follows:
– Set Peak Width (sec)
– Set Detection Threshold

Apex detection 2-35

 Peak integration events
Events can change the integration results associated with detected peaks.
These ApexTrack peak integration events are available in the Type column of
the Integration table. The Merge Peaks event is also available for GPC
processing methods.
All events require a start time, the default for which is 0.000 minutes. For
events that have a stop time, you can either leave the stop time in the Stop
(min) column blank, to indicate the event is enabled until the end of the run,
or specify a time in minutes. Both the start and stop times have a precision of
three significant figures by default, and the valid range of Start and Stop
times is 0 to 655 minutes. The start time must be less than the stop time
unless the stop time is blank.
These ApexTrack events affect peak integration:
• Valley-to-Valley Event
• Gaussian Skim Event
• Tangential Skim Event
• Merge Peaks Event (GPC, GPCV, GPC-LS, GPCV-LS only)
• Set Liftoff % Event
• Set Touchdown % Event
• Set Minimum Area
• Set Minimum Height
• Set Maximum Height
• Set Maximum Width (sec)

When timed events are active

These events affect peaks whose second derivative apex occurs during the
time of the event:
• Inhibit Integration
• Allow Negative Peaks
• Set Detection Threshold
• Set Liftoff %
• Set Touchdown %

2-36 ApexTrack Integration

 These events affect peaks whose retention time occurs during the time of the
event:
• Set Minimum Area
• Set Minimum Height
These events affect vertical drop lines that occur during the time of the event:
• Detect Shoulders
• Gaussian Skim
• Valley-to-Valley
• Merge Peaks (for GPC only)
This event affects the data point immediately after the timed event is enabled:
Set Peak Width (sec).

References
For further information on the theory of ApexTrack peak detection and
integration, see:
ApexTrack Integration: Theory and Application, Waters Corp., Milford, MA,
2007. Posted on www.waters.com.

References 2-37
2-38 ApexTrack Integration
3 Traditional Integration

This chapter describes traditional peak detection and integration theory.

Contents:

Topic Page
Features and capabilities 3-2
References 3-22

3-1
Features and capabilities
Traditional peak detection and integration by Empower software includes
these functions:
• Automatically determining appropriate peak width and threshold values
for the chromatogram, unless already set in the processing method
• Detecting peaks in the chromatogram to determine their location
• Integrating peaks to determine their retention times, areas, and heights
The processing method defines the parameters (including detection and
integration events) that the software uses to detect and integrate the peaks
within the raw data file (channel).

Peak detection
The peak detection processes include:
1. Performing data bunching
2. Determining peak start
3. Determining preliminary peak apex
4. Determining peak end
5. Determining peak width and threshold values in the processing method
6. Inhibiting integration
The detection algorithm first determines the presence of peaks by comparing
the rate of change of the signal to specific acceptance criteria, determining
where peaks in the acquired raw data file start and end. The software must
perform these peak detection tests before it can integrate the peaks.
You determine the peak detection test criteria several ways:
• Peak width and threshold selections in the Integration tab of the
Processing Method window in Review
• Integration toolbar of the Review Main window or the Processing
Method Editor
• Processing Method wizard in Review
See also: For additional information on peak detection theory, see
“References” on page 3-22.

3-2 Traditional Integration

Performing data bunching
As the detection algorithm tests the data for a peak, the software averages
individual raw data points into discrete groups, or bunches, to produce a
single point. The number of data points in a bunch is set by the peak width
parameter.
In most instances, each bunch of data contains one point when the sampling
rate is optimized. Data bunching has no effect on the acquired raw data. It is
an internal calculation used to enhance the process of determining peak start
and peak end. When the peaks contain more data points than necessary, the
bunched data points are used only to detect peaks; all raw data points are
used for integration.
The following figure illustrates the effects of data bunching on a noisy signal.
In this example, with the peak width set to 60 and sampling rate set to 1, the
detection algorithm produces a bunched data point for each set of four raw
data points. This optimizes the number of bunches to 15 (across a 60-second
peak containing 60 data points) and effectively smooths the data.

Data bunching example:

Raw data

Four data
points
averaged into
one
Bunched data

During detection, the software calculates the number of points in a bunch

using the equation:

PB = (-------------------------
PW × SR )-
15

Features and capabilities 3-3

 where:
PB = Points in a bunch
PW = Peak width (in seconds)
SR = Sampling rate (data points/second as specified in the instrument
method used for acquisition)
Tip: The peak detection algorithm functions most effectively with 15 data
points across each peak. For this reason, the software organizes the raw data
into 15 discrete bunches when setting the peak-width value. When peak width
is 15 and the sampling rate is 1, no data point bunching occurs; all data points
are used to detect peak start and peak end.

Determining peak start

The liftoff threshold specified in the processing method defines the minimum
slope of the signal in µV/sec, at or above which the start of a peak is detected.
Tip: By default, negative peaks are not detected. To activate the Allow
Negative Peaks event, see “Using timed events” on page 3-20.

To determine peak start, the software’s detection algorithm behaves as

follows:
1. It performs the threshold test on the signal:
• The software averages the signal slope across two data bunch
intervals and then compares it to the liftoff threshold.
• When the averaged slope of the signal between bunches B1 and B3
is greater than, or equal to, the liftoff threshold value, the software
flags B1 as the possible peak start.
2. It examines the individual points in the B1 bunch to determine the
actual start point. For positive peaks, this is the data point with the
minimum Y-value. For negative peaks, this is the data point with the
maximum Y-value.
The preliminary start point of a detected peak can be affected by an Inhibit
Integration event that ends near the start of the peak. Because the bunched
point that would have been selected as the preliminary start point of the peak
is inside the Inhibit Integration event, a different bunched point is selected as
the preliminary start point, even when the “bunch” contains only one data
point.

3-4 Traditional Integration

 Determining preliminary peak apex:

B2 - B1
Slope 1 =
Averaged t2 - t1
B3
slope
Slope 2
Possible B3 - B2
B2 Slope 2 =
peak start t3 - t2
B1
Slope 1 + Slope 2
Averaged =
t1 t2 t3 2
slope

Slope 1 Theoretical baseline

Slope of the (subject to change by
liftoff threshold integration)

Determining preliminary peak apex

To determine the preliminary peak apex (peak maximum) after the peak start
is confirmed, the software performs as follows:
1. It monitors the signal until the slope changes sign from positive to
negative. For a negative peak, the slope changes sign from negative to
positive.
2. It analyzes the bunch where the slope change occurs (bunch B12 in the
next figure) and assigns a tentative peak apex to the data point within
the bunch that is farthest away from the theoretical baseline.
Tip: This peak apex is preliminary because the software does not
determine the actual peak apex until integration occurs and baselines
are assigned.

Features and capabilities 3-5

 Determining the preliminary peak apex:
B12
Peak apex point

B13
Negative slope
B10
B14

Positive slope

B17

B5
Peak start point
Theoretical baseline
B20

B1
B24

The preliminary apex of a detected peak can be affected by an Inhibit

Integration event placed too close to the peak apex. Because the bunched point
that would have been selected as the preliminary peak apex is inside the
Inhibit Integration event, the peak can go undetected.

Determining peak end

To determine the peak end, the software performs as follows:

1. It compares the slope of the signal to the touchdown threshold. When
two consecutive slopes are less than the threshold value, the algorithm
flags the last data point in the last bunch as the possible peak end.
2. It examines the individual data points in the current and next bunch to
determine the actual peak end. For positive peaks, this is the data point
with the minimum Y-value. For negative peaks, this is the data point
with the maximum Y-value.
3. During the peak-end test, it checks for a change in the sign of the slope.
A change in sign before the touchdown indicates a preliminary peak

3-6 Traditional Integration

 valley (the end point of the current peak and the start point of the next
peak).
Tip: This peak end point and start point is preliminary because the
software does not determine the actual data point for the end of a peak
until the integration process occurs and baselines are assigned.
4. It proceeds from this peak start point to the peak apex test and
continues until it successfully determines peak touchdown.

Determining peak end:

Slope of
B20 touchdown
threshold

B21
B23 B24
B22

Peak end within

this region

The preliminary end point of a detected peak can be affected by an Inhibit

Integration event that starts near the end of the peak. Because the bunched
point that would have been selected as the preliminary end point of the peak
is inside the Inhibit Integration event, a different bunched point is selected as
the preliminary end point, even when the “bunch” contains only one data
point.

Determining peak width and threshold values

The software uses a second derivative to set the peak width and threshold
values in the processing method automatically.
Tip: You can use the Peak Width and Threshold determination method that
was used in all Millennium32 software, by clicking Configuration Manager >
View > System Policies, and then clicking Use v3.0X Style Peak Width and
Threshold Determination in the Data Processing tab (see the Empower Online
Information System). When this system policy is active, the Peak Width and

Features and capabilities 3-7

 Threshold buttons, Processing Method wizard, processing methods, and
32
results all function as they do in all Millennium software.

Peak width parameter

The software automatically determines the peak-width value (Auto-Peak
Width) using the inflection points of the second derivative of the peak that
exhibits the highest second derivative within a chromatographic region.
Since the software uses the peak-width value to determine a bunching factor
during peak detection (see “Performing data bunching” on page 3-3), this
value affects the sensitivity of peak detection. The guideline is to use a peak
width value within a range of plus-or-minus two times the
software-determined peak-width value.
If the signal-to-noise (S/N) ratio is acceptable, the peak-width value at the
high end of this range can increase sensitivity and allow relatively small
peaks to be properly integrated. However, shoulders on larger peaks, if
present, might no longer be detected. Increasing the peak-width value above
this range results in a decrease in sensitivity.
The valid range of the peak width setting is 0.01 through 9999.99. The default
peak-width setting is blank.
There are several ways to set a peak-width value in Review:
• When using the Processing Method wizard in Review to create a new
processing method or edit an existing one, the software automatically
determines an appropriate peak width using the data contained within
the zoomed region (in the Integration - Integration Region wizard page).
• When viewing data in the Review Main window, clicking the Peak Width
button in the Integration toolbar automatically sets the peak-width
value to the peak with the highest second derivative within the current
zoom region (which may be the entire chromatographic region).
• With no peak width set in the active processing method, you can
integrate the data by clicking Process > Integrate (or clicking the
Integrate button in the toolbar). The peak width is automatically set
according to the data in the entire chromatogram (unless there is an
Inhibit Integration event at the start and/or the end of the
chromatogram).
The region of the chromatogram used to set peak width starts at the
beginning of the chromatogram or the stop time of an Inhibit Integration
event that starts at the beginning of the chromatogram. The region of the

3-8 Traditional Integration

chromatogram used to set peak width ends at the end of the chromatogram or
the start time of the Inhibit Integration event that stops at the end of the
chromatogram.
Inhibit Integration events that do not overlap the beginning or end of the
chromatogram are ignored when setting peak width.
Tip: When using this method, the peak width is placed in the Result Peak
Width field only. The Processing Method Peak Width field remains blank. To
copy the Result Peak Width value to the Method Peak Width field, click Copy
to Processing Method from the right-click menu.
• Set the peak width value manually by entering a value in the
Integration toolbar of the Main window of Review or in the Integration
tab of the Processing Method window.
Tip: If the peak with the highest second derivative is fused, the peak
width value might not be optimal. in such cases, zoom in on peaks other
than the fused peak when setting the peak width parameter.

Threshold values
The software automatically determines the threshold value (Auto-Threshold)
by first applying a median filter to the second derivative of the
chromatographic data to determine the noise. The software then derives the
threshold value by multiplying the second derivative noise by the current
peak width value.
The threshold value is a slope measurement that the software uses to
determine peak start and end points during peak detection (as described in
“Determining peak start” on page 3-4, and “Determining peak end” on
page 3-6). A relatively low threshold value increases sensitivity and may allow
relatively small peaks to be properly integrated. If too many small, baseline
noise peaks are being integrated, increasing the threshold value can prevent
these small peaks from being integrated.
The software normally uses the global threshold value, in the processing
method, to determine both peak start (liftoff) and peak end (touchdown). If
you need to use a different threshold value for peak starts or ends, due to a
tailing or a sloping baseline, use the Set Liftoff or Set Touchdown event.
The valid range of the threshold setting is 0.0 or greater. The default
threshold setting is blank.

Features and capabilities 3-9

 There are several ways to set a threshold value in Review:
• When using the Processing Method wizard to create a new processing
method or edit an existing one, the software automatically determines
an appropriate threshold using the data in the zoomed region in the
Integration – Integration Region wizard page.
• When viewing data in the Review Main window, clicking the Threshold
button in the Integration toolbar automatically sets the threshold value
using data in the current zoom region (which can be the entire
chromatographic region).
Tip: The Set Processing Method Threshold button is disabled when the
Processing Method Peak Width field is blank.
• With no threshold set in the active processing method, you can integrate
the data by clicking Process > Integrate (or clicking the Integrate button
in the toolbar). The software automatically sets the threshold according
to the data in the entire chromatogram (unless there is an Inhibit
Integration event at the start and/or the end of the chromatogram).
The region of the chromatogram used to set the threshold starts at the
beginning of the chromatogram or the stop time of an Inhibit Integration
event that starts at the beginning of the chromatogram. The region of
the chromatogram that is used to set the threshold ends at either the
end of the chromatogram or the start time of the Inhibit Integration
event that stops at the end of the chromatogram.
Inhibit Integration events that do not overlap the beginning or end of
the chromatogram are ignored when setting threshold.
Tip: When using this method, the determined threshold is placed in the
Result Threshold field only. The Processing Method Threshold field
remains blank. To copy this value to the Method Threshold field, click
Copy to Processing Method from the right-click menu.
A peak-width value is required before determining a threshold value. If
the processing method does not include a peak-width value, the
threshold button is not available. If you use this method to perform
integration without a peak width value, the software first determines
the peak width value and then the threshold value. Both values are
automatically placed in their respective toolbar fields.
• You can set the threshold value manually by entering a value in the
Integration toolbar of the Main window of Review or in the Integration
tab of the Processing Method window.

3-10 Traditional Integration

 Peak Width and Threshold fields
The peak width and threshold values are reported as both method and result
fields. These fields are in the Integration toolbar of the Review Main window,
and the result fields are available for reports. The method fields report the
peak width and threshold values from the processing method. The result fields
report the peak width and threshold values used when the raw data was
processed.
During processing, the software uses the values in the Processing Method
Peak Width and the Processing Method Threshold fields. It then stores these
values in the Result Peak Width and Result Threshold fields. In this case, the
Result Peak Width and the Result Threshold fields are the same as the
Processing Method Peak Width and the Processing Method Threshold fields.
If the Processing Method Peak Width and/or the Processing Method Threshold
are blank, then the software determines the Result Peak Width and/or Result
Threshold fields during data processing.
When data is processed using a processing method that contains a blank
Processing Method Peak Width and/or Processing Method Threshold, each
result may be produced using a different Result Peak Width and Result
Threshold.
Tip: You can disable the Auto-Peak Width and Auto-Threshold
determinations by clicking Configuration Manager > View > System Policies,
and then clicking Use v3.0X Style Peak Width and Threshold Determination,
in the Data Processing tab (see the Empower Online Information System).
When this system policy is active, the Peak Width and Threshold buttons, the
Processing Method wizard, processing methods, and results all function as
32
they do in all Millennium software.

Peak integration
This section describes the following peak integration processes:
• Determining fused peaks
• Constructing the baseline
• Calculating peak retention time, height, and area
Integration uses the peak start and peak end values identified during peak
detection to determine baselines and integrate isolated and fused (clustered)
peaks.

Features and capabilities 3-11

 If your chromatograms are complicated, you can enable time-based
integration events to refine peak integration.
See also: For in-depth information on peak integration theory, see
“References” on page 3-22.

Determining fused peaks

The first process in integration is distinguishing any fused and isolated peaks
in the chromatogram. The software checks the distance between adjacent
peaks.

Adjacent peak width comparison:

Start End Start End

W3

W1 W2

When determining fused peaks, the traditional integration algorithm

operates as follows:
1. It compares the width of the space between the detected start and end
points of adjacent peaks (W3) to the width of the wider adjacent peak
(either W1 or W2).
2. It locates the wider adjacent peak (W2 > W1).
3. It computes the ratio of the wider adjacent peak to the space between
the two peaks (W3) using the equation W2/W3. If the ratio is greater
than or equal to 3.0, the peaks are considered fused. If the ratio is less
than 3.0, the peaks are considered resolved.
Tip: The software uses the ratio of 3.0 to increase the chances of
detecting peak overlap.

3-12 Traditional Integration

To set the valley point between fused peaks, the software operates as
follows:
1. It draws a projected baseline from the start point of the first peak in the
cluster to the end point of the last peak in the cluster.
2. It searches for the valley point between each pair of adjacent fused
peaks, and chooses the raw data point closest to the projected baseline
as the valley point. The software adjusts the end point of the peak
preceding the valley to the time of the valley point. Similarly, it adjusts
the start point of the peak following the valley to the time of the valley
point.
3. It draws a vertical line from the valley point to the projected baseline,
thereby separating the peaks.
In the next figure, for example, the integration algorithm locates two fused
peak groups and a total of six peaks within the chromatogram.

Features and capabilities 3-13

 Determination of resolved and fused peaks:
Valley point

Peak 1 Peak 2
Vertical drop line

Projected baseline

End 1, End 2, End 1,

start 2 start 3 start 2

Start 1 End 1 Start 1 End 2

Peak 1 Peak 2 Peak 3

Peak 1 Fused peak Fused peak

group 1 group 2

Projected baseline Start 1 End 3

Constructing the baseline

Once resolved and fused peaks are identified within the chromatogram, the
integration algorithm draws a baseline from the start to the end of each peak
or fused peak group.
The fields Start Time and End Time display the start and end times of the
peak calculated during peak integration. The fields Baseline Start and
Baseline End display the start and end times of the baseline used to integrate
a peak. The Baseline Start and Baseline End values are as follows:
• The same as the Start Time and End Time values when integration is
for a baseline-resolved peak (baseline-to-baseline).
• Different from the Start Time and End Time values when integration is
for a fused peak (peaks not baseline-resolved).

3-14 Traditional Integration

Baseline construction:

1 2 3

B B B B
BB = Baseline-to-baseline Peak 1 = BV (baseline-to-valley)
Peak 2 = VV (valley-to-valley)
Peak 3 = VB (valley-to-baseline)

When you use the default integration setting, each identified peak is given a
two-character label that indicates whether the peak starts or ends at a point
on the baseline (B) or in a valley (V) above the baseline. A peak can have four
types of baseline construction. The label appears in the Int Type column of the
Peaks tab of the Results and Main windows of Review.

Default integration peak labels:

Peak start and end point Label

Baseline-to-Baseline BB
Baseline-to-Valley BV
Valley-to-Baseline VB
Valley-to-Valley BB

Tip: When you use the Exponential Skim or Tangential Skim integration
events, additional types of baseline construction can appear (see “Integration
peak labels” on page 3-21).
Capitalization of the label indicates the following conditions:
• Capital letters – The integration was performed automatically by the
software.
• Lowercase letters – The integration was performed manually.

Features and capabilities 3-15

 For instance, a baseline label of Bb indicates that while the peak start and
peak end are both baseline-resolved, the peak start was automatically
integrated by the software, and the peak end was manually adjusted.

Baseline adjustment
If the projected baseline intersects the signal in the chromatogram, the
software adjusts the baseline to the lowest point within the fused peak group,
separating the peak group into individual and/or fused peaks, as appropriate.
The software then rechecks the new baselines to make sure they do not
intersect the chromatographic signal except at peak start or end points, and it
readjusts the baseline as necessary.

Baseline adjustment:

Baseline intersects signal

Initial baseline Baseline after adjustment

Calculating peak retention time, height, and area

Once actual baselines are constructed, the integration algorithm operates

as follows:
1. It calculates the retention time, height, and area for each peak.
2. It compares each integrated peak to the minimum area and minimum
height rejection criteria you specify in the processing method.

3-16 Traditional Integration

Retention time and height

To determine retention time and height, the software operates as follows:

1. Locates the retention time of the data point in the peak that is farthest
from the constructed baseline.
2. Fits a quadratic curve to the five points at the top of the peak (the
highest data point and the two data points on either side of this point).
3. Sets the peak apex point to the inflection point of the fitted curve. The X-
value of the peak apex is the retention time of the peak.
4. Calculates the peak height as the distance (in µV) from the constructed
baseline to the Y-value of the calculated peak apex.
Tip: If the software fails to fit a curve to the top of the peak, it uses the apex
point (the data point farthest from the baseline) to calculate retention time
and height as in Millennium software version 2.15 and earlier. The software
adds a processing code (I05, I06, I07, or I08) to the Codes column in the Peaks
tab in Review to explain why the curve fitting to the top of the peak failed.
The following figure illustrates the peak retention time and peak height
calculation.

Peak retention time and peak height calculation:

Calculated
peak apex
Highest point
Calculated peak apex
(not on data point)

P3
P4
P2 P5 Peak
Constructed height
Start of baseline
P1
chromatogram Peak Peak
start end

Retention time

Tip: You can disable fitting a quadratic curve to the top of the peak by clicking
Configuration Manager > View > System Policies and implementing a system

Features and capabilities 3-17

 policy named Use v2.XX Style Retention Time Calculations (see the Empower
Online Information System) in the Data Processing tab. When results are
integrated with this system policy active, processing code I09 is added to the
Codes field in the result. This field is visible in the chromatogram Result
table, in the Result window of Review or in the Project window’s Results tab.

Area
The algorithm calculates the total peak area by adding the areas for each raw
data point interval between peak start and peak end. The region below the
constructed baseline (Ab) is subtracted from the total area of the peak (At).
This yields the peak area above the constructed baseline (Ap).

Peak area calculation:

Peak area At
Peak
start
Constructed
baseline Peak
end

A/D minimum
Total peak area (At) signal

Peak area Ap

Peak
Ap = At - Ab
start
Constructed
baseline Peak
end

Baseline
area (Ab)
A/D minimum
Subtracting area below baseline (Ab) signal

3-18 Traditional Integration

Peak rejection criteria
When the software integrates a peak, the integration algorithm compares the
peak with the integration rejection criteria you specify. You can set the
rejection criteria in the Integration tab of the Processing Method window in
Review, by using the Minimum Area and Minimum Height buttons in the
Review Main window or using the Processing Method wizard. Based on this
comparison, the algorithm accepts or rejects the peak. Integration rejection
criteria can include:
• Minimum area
• Minimum height
• Minimum of five points across a peak

Minimum area
The minimum area criterion determines the minimum area (in μV • sec)
required for an integrated peak to be included in the peak list. If the area of
the integrated peak falls below the set value, the peak is removed from the
peak list. If the area is equal to or greater than the set value, the peak is
accepted.

Minimum height
The minimum height criterion determines the minimum height (in μV)
required for an integrated peak to be included in the peak list. If the height of
the integrated peak falls below the set value, the peak is removed from the
peak list. If the absolute value of the height is greater than or equal to the set
value, the peak is accepted.
Tip: Minimum Area and Minimum Height parameters are useful for removing
small integrated peaks from the result. A high value may cause integrated
peaks to be rejected as noise; conversely, a low value may cause baseline noise
to be integrated as peaks.

5-point peak rejection

The 5-point peak rejection criterion instructs the software to remove from the
peak list any peak that contains fewer than five points.
Tip: The 5-point peak rejection criterion is built into the software and occurs
automatically. It is not a parameter found in the processing method.

Features and capabilities 3-19

Using timed events
Empower software supports timed events for peak detection and integration.
Tip: Refer to the Empower Online Information System for an overview of all
timed events, and a description of each timed event.

Peak detection events

The software supports the following time-based detection events to further
refine peak detection:
• Allow Negative Peaks
• Set Liftoff
• Set Touchdown
• Set Peak Width
Tip: The Set Liftoff, Set Touchdown, and Set Peak Width events take effect
only in the baseline regions outside detected single peaks or fused peak
groups. If the event starts within an isolated peak or fused peak group, the
event takes effect at the end of the isolated or fused peak group.

Peak integration events

The software supports the following time-based integration events to further
refine peak integration:
• Force Baseline by Time
• Force Baseline by Peak
• Forward Horizontal by Time
• Forward Horizontal by Peak
• Reverse Horizontal by Time
• Reverse Horizontal by Peak
• Valley-to-Valley
• Force Drop Line
• Force Peak
• Exponential Skim
• Tangential Skim
• Set Minimum Height

3-20 Traditional Integration

 • Set Minimum Area
• Set Maximum Height
• Set Maximum Width (sec)

Integration peak labels

When you use the default integration setting, each identified peak in a
chromatogram is given a two-character label that indicates whether the peak
starts or ends at a point on the baseline (B) or in a valley (V) above the
baseline. The label appears in the Int Type column of the Peaks tab of the
Results and Main windows of Review.

Integration Peak Labels on a Chromatogram:

Peak Start and End Point Label

Baseline-to-Baseline BB
Baseline-to-Valley BV
Exponential-to-Exponential EE
Exponential-to-Valley EV
Tangential-to-Tangential TT
Tangential-to-Valley TV
Valley-to-Baseline VB
Valley-to-Exponential VE
Valley-to-Tangential VT
Valley-to-Valley VV

Capitalization of the label indicates the integration was performed

automatically by the software. Lowercase letters indicate the integration was
performed manually.

Features and capabilities 3-21

References
For further information on the theory of peak detection and integration, see:
• Dyson, Norman, Chromatographic Integration Methods, The Royal
Society of Chemistry, Thomas Graham House, Cambridge, 1990.
• Massart, D.L., et al., Chemometrics: A Textbook, Elsevier Science
Publishers, Amsterdam, 1988.
• Papoulis, Athanasios, Signal Analysis, McGraw-Hill, New York, 1977.
• Snyder, L.R. and J.J. Kirkland, Introduction to Modern Liquid
Chromatography, second ed., Wiley-Interscience, New York, 1979.

3-22 Traditional Integration

4 Peak Identification and
Quantitation of Sample
Components
This chapter describes how peaks are quantitated.
Contents:

Topic Page
Features and capabilities 4-2
Peak matching 4-2
Quantitation 4-7
Calibration curve fit types 4-27
References 4-53

4-1
Features and capabilities
Empower software identifies and quantifies unknown components using peak
matching and quantitation:
• Peak matching – The process of matching unknown peak retention
times (RT) against the RT of known standard peaks.
• Quantitation – The process of calculating the amounts of unknown
peaks using the integration results of each peak and a calibration curve
based on the amounts and integration results of known peaks
(standards).

Peak matching
When performing peak matching, the software chooses the integrated peaks
in the chromatogram that most closely match the components in the
Components table from the processing method. To accomplish this, the
software operates as follows:
1. It uses the time region defined by the RT of the component’s calibration
curve, plus or minus the component’s RT window, together with the
Peak Match type.
2. It matches peaks inside the RT windows of the components by
calculating the difference between each unknown peak and component
RT defined in the processing method.
3. It uses the differences to choose the unknown peak that most closely
matches the component peak.

Matching hierarchy
The software uses a hierarchy of peak match types when matching unknown
peaks to components. The software matches each component to the unknown
peaks in its RT window. If a peak matches multiple components, the software
determines the most appropriate component for that peak – first by position,
then by size, and finally by RT.

4-2 Peak Identification and Quantitation of Sample Components

Calculating the match difference
When the peak match type is Closest or Closest Negative, the software
calculates the difference as the absolute value of the component’s RT minus
the unknown peak’s RT. For the other match types, the match difference is
either 0 (a perfect match) or not matched.
A match is considered perfect if at least one of the following conditions exists:
• A peak is in first, second, third, fourth, fifth, or last position in the RT
window that corresponds to its match type.
• The size of a peak, relative to other peaks in the RT window, conforms to
its match type:
– Greatest area or height
– Least area or height
– Greatest width (GPCV data only)
• There is a 0.0 difference between the RT of the peak and the component.

Choosing the optimal peak match

The next step in the matching process is to determine whether any
components match multiple unknown peaks or whether any unknown peaks
match multiple components. These are the three possible outcomes of the
initial component matching process:
• Single peak matching a single component
• Multiple peaks matching a single component
• Single peak matching multiple components

Single peak with a single component

The peak matching process is straightforward when the RT windows of the
components do not overlap and, at most, one unknown peak is found in each
window. In such a case, matching type and difference are not necessary.
Tip: Never use the matching types Second, Third, Fourth, or Fifth when there
are always fewer than two, three, four, or five peaks, respectively, in the RT
window.

Peak matching 4-3

 Multiple peaks with a single component
If there are multiple unknown peaks in the RT window of a component, the
software uses the match difference to choose the peak that most closely fits
the matching type criteria.
• If two peaks match a single component, the software chooses the peak
with the smallest match difference. The remaining peak is then matched
to its next closest component.
• If two peaks equally match a single component (the match differences
are equal), the software chooses the first peak. The second is matched to
its next closest component.

Single peak with multiple components

If the same peak is matched to two or more components, the software then
picks the component with the smallest difference from among the possible
matches. If two components have equal differences for a peak, a choice cannot
be made, and the unknown peak remains unmatched. In the Peaks tab of the
Main and Results windows of Review, the software lists the components’ Peak
Types as Missing and a Q04 code is copied into the Processing Code field for
the unmatched peak, indicating the reason the components are missing.

Shifting RT and RT windows

If the software cannot identify peaks because they are shifting so much that
they are outside the RT windows, you can increase the size of the RT windows.
If that is not possible, or if it causes peaks to be misidentified, you can use the
RT reference peak or the update RT parameter.

RT Reference
The RT Reference field allows you to temporarily adjust the RT of a
component based on where in the chromatogram the defined RT Reference
peak is found in the chromatogram. The RT of the component temporarily
shifts by the same percentage, and in the same direction, as the shift of the RT
Reference peak. (The RT Reference peak is determined by comparing the RT
for the reference peak listed in its calibration curve to the actual RT of the
reference peak in the chromatogram.) The software uses the adjusted RT to

4-4 Peak Identification and Quantitation of Sample Components

match the unknown peak in the chromatogram and calculates the adjusted
RT of a component as follows:
RT ref peak found
RT adjusted = RT x ---------------------------------------------------------
RT ref peak in cal curve

Use a RT reference peak that is always found in your chromatogram, that is

well separated from other peaks, and whose RT shifts with your other
components. The peak characteristics usually compensate for RT shifts that
can affect a particular chromatogram.

Update RT
The Update RT field adjusts the RT of calibration curves, thus affecting the
RT that the software uses to match unknown peaks. Adjusting is done to more
accurately reflect the actual RT of the peaks in the chromatogram when RT
shifting or drifting is problematic.
Normally during peak matching, the software compares the RT of integrated
peaks to the RT of the calibration curves and also to the RT windows listed for
the components in the Components tab of the processing method. When
Update RT is selected, the software uses the RT where the components were
actually found during the processing of a previous chromatogram. Each time a
chromatogram is processed, the software may store a new RT to use for peak
matching and processing the subsequent chromatogram (depending on the
Update RT selection). The updated RT is stored in the calibration curve for the
component and is displayed in the Time field of the Calibration Curve window.
The Update RT functionality does not affect the retention times listed in the
Components tab of the Review window.
These are the Update RT choices:
• Never – The RT of the calibration curve is not updated.
• Replace – The RT of the calibration curve is updated every time a
chromatogram is calibrated or quantitated, regardless of sample type.
When a chromatogram is calibrated or quantitated, if any peak is
identified in the RT window, the software replaces the RT in the
calibration curve (not in the Components table) with the newly found
RT.
• Replace Standards – The RT of the calibration curve is updated only
when standards are calibrated. When a chromatogram is calibrated, if a
standard peak is identified in the RT window, the software replaces the

Peak matching 4-5

 RT in the calibration curve (not in the Components table) with the newly
found standard RT.
• Average – The RT of the calibration curve is updated every time a
chromatogram is calibrated or quantitated, regardless of sample type.
When a chromatogram is calibrated or quantitated, if any peak is
identified in the RT window, the software averages the RT in the
calibration curve (not in the Components table) with the newly found
RT.
• Average Standards – The RT of the calibration curve is updated only
when standards are calibrated. When a chromatogram is calibrated, if a
standard peak is identified in the RT window, the software averages the
RT in the calibration curve (not in the Components table) with the newly
found standard RT.
The software updates the RT using any averaging choice as follows:

RT c = (------------------------------------------------------------------------------------------------------------------------------------------------------------------
Average Time from Calibration Curve × n + New Retention Time -)
n+1
where:
RTc = The retention time of the calibration curve
n = The number of times the value was previously averaged
Update RT is a coarse adjustment that should only be used when these
conditions apply:
• Peak retention times are shifting in one direction.
• The overall shift cannot be offset by either increasing the RT window of
the component peaks or using the RT reference peak.
Tip: The Replace and Average choices should be used only when the unknown
samples have no peaks that can be misidentified.

4-6 Peak Identification and Quantitation of Sample Components

Quantitation
You can perform quantitation using these approaches:
• Calibration
• No calibration
• Sample weight and dilution
• Injection volume
• Responses other than peak area and height
See also: For additional information on the processes the software uses
during quantitation, see “References” on page 4-53.

Quantitation by calibration
Empower software performs calibration on a set of processed standards
acquired by the chromatographic system. When you run the standards, the
software requires that you specify this information:
• Injection volumes in the Samples table of the Run Samples window,
Sample Set Method Editor, or Alter Sample window
• Component names and amounts, or concentrations in the Default
Amount tab of the Processing Method window, the Component Editor of
the Run Samples window, or the Alter Sample window.
During processing of chromatograms, the software calculates a response
based on the detector signal for each peak. This response can be one of these:
• Peak area
• Peak height
• Another peak value (including a custom peak value).
Once calibration standards are processed, the software generates a calibration
curve for each standard component listed in the Components table. The
calibration curve displays this information:
• Response (Y-Value field) versus Amount or Concentration (X-Value
field) for external standard calibration
• Response ratio multiplied by the internal standard amount, or
concentration, versus Amount or Concentration (X-Value field) for
internal standard calibration.

Quantitation 4-7
 Calibration curve shape is based on a fit type you select (as described in
“Calibration curve fit types” on page 4-27). There are three calibration curve
fit types:
• Linear– Always results in a linear fit.
• Non-Linear– Allows you to select different fits to a multilevel calibration
curve.
• Forced-Through-Zero – Allows you to force the calibration line through
zero.
The software calculates and updates calibration curves using individual or
averaged points based on an Average By value you specify in the Components
tab of the Processing Method window.
During sample processing, the software performs the following steps in
sequence:
• Matches the RT of the integrated peaks found in the unknown
chromatogram with the RT of the components in the calibration curve.
• Applies the response of each matched unknown peak to the
corresponding component calibration curve.
During quantitation, the software calculates the amount, or concentration, of
the unknown sample from the calibration curve. It uses the response of the
sample to find the X-value that corresponds to the amount or concentration. It
then displays the final component amount in the Peaks tab of the Main and
Results windows of Review.

Quantitation without calibration

If you want to quantitate without performing calibration, the software
calculates the relative amount of each unknown peak in the sample as both
percent area and percent height. Peak area and height percent are calculated
as the percent of each integrated peak relative to the total area or height of all
integrated peaks.

4-8 Peak Identification and Quantitation of Sample Components

Quantitation using sample weight and dilution
Sample weight and dilution values are used to adjust the amounts and
concentrations of standard and unknown components. These two values are
optional and can be used to compensate for differences due to factors such as
these:
• Varying dilutions
• Different initial sample mass or volume
You enter sample weight and dilution on a per-standard or per-sample basis
in the Samples table of the Run Samples window, Sample Set Method Editor,
or Alter Sample window. Typically, sample weights and/or dilutions are used
for either the standard samples or the unknown samples, but not for both.

Sample weight
Sample weight is typically used in samples to calculate the ratio of quantity of
component injected into the system to total quantity of original sample.
During calibration, the software multiplies the entered amount(s) or
concentration(s) of the standard component(s) by the sample weight to
calculate amounts and concentrations for the standard sample.
During quantitation, the software divides the amount or concentration
(X-Value field) determined from the calibration curve by the sample weight to
calculate amounts and concentrations for the unknown sample.
For example, if the mass of sample that you weighed is 0.5 mg, and you want
to report the amount determined by the software as the amount of the
component as compared to the mass of the total sample, enter a sample weight
of 0.5 for your unknown sample. The software quantitates the component
amount from the calibration curve and then divides that value by the sample
weight to obtain the final ratio of component amount to total sample amount.
The amount can then be converted into a percentage by multiplying it by 100,
either by using a dilution value of 100 or by creating a custom field and
specifying a formula of Amount*100.
When using sample weight, be sure that this value is equivalent to the units
for the component amount or concentration that you are reporting. For
example, if you weigh 1.44 mg of sample, and the units of your standard
amounts are in μg, use a sample weight of 1440 (μg).

Quantitation 4-9
 Dilution
The Dilution field is typically used when you dilute a sample (a standard, an
unknown, or a control) prior to injection and want to report the quantity of
analyte in the original, undiluted sample. This could occur when an undiluted
sample, injected directly onto the column, would fall above the range of the
calibration curve. The sample dilution should be entered into the Samples
table of Run Samples window, and the sample should be injected at the usual
injection volume.
During calibration, the software divides the entered amounts or
concentrations of the standard component(s) by the dilution value to calculate
amounts and concentrations for the standard component(s).
During quantitation, the software multiplies the amount or concentration
(X value) determined from the calibration curve by the dilution value to
calculate amounts and concentrations for the unknown sample.
For example, if a 1:10 dilution was performed on a standard sample
containing one component at an amount of 100 μg, enter the amount of the
standard component into the Component Editor or the Default Amounts tab of
the Processing Method window as 100 μg (the original, undiluted quantity)
and the dilution as 10. When the software calibrates this standard, it takes
the specified amount of 100 μg and divides by the specified dilution of 10. The
software reports the resulting amount as 10 μg (the amount injected on the
column). This value is also plotted on the calibration curve.
If that same sample were an unknown sample, you would not enter a quantity
for the component; nevertheless, the dilution would still be 10. When the
software quantitates the unknown, it reads an amount of 10 μg directly from
the calibration curve and multiplies that value by 10 (the dilution value) for a
resulting amount of 100 μg (the prediluted amount).
When you are working with dilutions of standards, specify the dilutions and
the original, undiluted quantities of the standard components.When you are
working with dilutions of unknown samples, if you specify the dilutions, the
amounts and concentrations reported by the software will be those of the
original, undiluted samples. The use of the dilution field eliminates the need
to correct for a dilution by adjusting the injection volume.
Tip: You can correct for the dilution of samples by adjusting the injection
volume. If, by mistake, you dilute a sample or standard by a factor of 10, you
could inject 10 times the usual injection volume instead of specifying a value
of 10 in the Dilution field. If the sample is a standard, you also need to specify
the undiluted amount(s) for the standard component(s). If the sample is an

4-10 Peak Identification and Quantitation of Sample Components

 unknown, do not specify the dilution but adjust the injection volume 10-fold.
The software determines the undiluted quantity of component. This quantity,
given the higher injection volume, is 10 times higher than it would have been
had the normal injection volume been used. For both standard and unknown
samples, if a dilution is corrected for by the injection volume, do not adjust the
value in the Dilution field.

Quantitation using injection volume

The software calculates both amounts and concentrations for standard and
unknown samples. This allows you to print the value that is meaningful to you
on your reports.
The software determines whether you are entering your standard component
quantities in units of amount or concentration by the Sample Value Type list
in the Components tab of the Processing Method window. Selecting amount or
concentration will cause the software to interpret component quantities as
follows:
• Amount – The software interprets the component quantities that you
enter (in the Component Editor of the Alter Sample window, the
Component Editor of the Run Samples window, or the Default Amount
tab of the Processing Method window) as amounts. The software
calculates the corresponding concentration by dividing the specified
amount by the injection volume, in μL.
• Concentration – The software interprets the component quantities that
you enter (in the Component Editor of the Alter Sample window, the
Component Editor of the Run Samples window, or the Default Amount
tab of the Processing Method window) as concentrations. The software
calculates the corresponding amount by multiplying the specified
concentration by the injection volume, in μL.
Regardless of whether you define your standard component quantities in
amounts or concentrations, you can create a calibration curve that uses
standard amounts or concentrations. The X-value for the component as
entered in the Components tab of the Processing Method window determines
whether the calibration curve is a plot of Response versus Amount or
Response versus Concentration.
If the calibration curve is a plot of Response versus Amount (which occurs
when the X-Value field is set to Amount), the software quantitates unknown
samples by using a component’s response to determine its amount directly
from the calibration curve. The software then determines the component’s

Quantitation 4-11
 corresponding concentration value by dividing the calculated amount by the
injection volume, in μL.
Tip: If the X-Value field is set to Amount, the sample injection volume affects
the calculated concentrations for unknown samples but not the calculated
amounts.
Likewise, if the calibration curve is a plot of Response versus Concentration
(which occurs when the X-Value field is set to Concentration), the software
quantitates unknown samples by using a component’s response to determine
its concentration directly from the calibration curve. The software then
determines the component’s corresponding amount value by multiplying the
calculated concentration by the injection volume, in μL.
Tip: If the X-Value field is set to Concentration, the sample injection volume
affects the calculated amounts for unknown samples but not the calculated
concentrations.
Be careful when you specify a component’s unit label (μg, μg/μL, etc.) because
the software reports the label exactly as you specify it. In cases where the
component’s quantity is affected by the injection volume, be sure that the unit
label is appropriate. The software always uses microliters for injection volume
units.

Quantitation using responses other than peak area and height

The software allows you to use the responses of area, height, % Area,
% Height, for response. Additionally, any peak type custom field using a data
type of real can be used for response, except for custom fields using these:
• Time fields
• Baseline fields
• Response
• Amount
• Concentration
• % Amount
Make the appropriate selection in the Y-Value field, in the Components tab of
the Processing Method window.

4-12 Peak Identification and Quantitation of Sample Components

External and internal standard quantitation
The component amount calculations use one of these quantitation methods:
• External standard
• Internal standard with separate standard and unknown samples
• Internal standard without separate standard and unknown samples
(typically used with gas chromatography)
This section briefly describes these methods.
See also: For assistance reproducing amounts and/or concentrations
calculated by the software, see “Calibration curve fit types” on page 4-27.

External standard quantitation

The external standard method of quantitation determines component
amounts and/or concentrations by applying the detector response of a
component peak to a calibration curve. The calibration curve is generated
from a separately acquired and processed set of standards.
Tip: The standard set may contain only one standard (referred to as
single-level calibration).
The following criteria are also required:
• You must define standard samples as Standards, either by using the
Inject Standard function during sample loading in Run Samples or (after
the sample is acquired) by defining a Sample Type of Standard in Alter
Sample.
• You must define unknown samples as Unknowns, either by using the
Inject Unknown function during sample loading in Run Samples or
(after the sample is acquired) by defining a Sample Type of Unknown in
Alter Sample.
• You must define component names and amounts or concentrations of
each standard component in the Default Amount tab of the Processing
Method window, the Component Editor of the Run Samples window, or
the Alter Sample window.
• The response is the Y-value of the calibration curve. You choose the
parameter to use as the y-axis by selecting the Y-Value field in the
Components tab of the Processing Method window.

Quantitation 4-13
 • The X value of the calibration curve is either amount, concentration, or a
custom field. You choose the x-axis in the X-Value field in the
Components tab of the Processing Method window.
You can use any peak type custom field using a data type of real for the
X-value, except for custom fields using:
– Time fields
– Baseline fields
– Response
– % Amount
• External standard quantitation generates each calibration curve by
plotting the detector response of a standard component versus the
amount or concentration of the standard component.
• You define the fit type used for the calibration curve in the Components
table of the Processing Method window.

To perform single-level external standard quantitation, the software

operates as follows:
1. It identifies the component peak(s) in the standard injection using peak
matching.
2. It determines the response and amount or concentration for each
standard peak, then plots the two values as a calibration point on the
calibration curve for the component with the same name.
Given a chromatogram such as that in the following figure, the values
used to determine the calibration points are the concentration and
response values from the table that follows the figure.

4-14 Peak Identification and Quantitation of Sample Components

 External standard chromatogram:

Peak B

Peak A
Peak C

Tip: In the following table, the X-values are set to concentration.

Standard peak values, external standard calibration:

Amount or
Quantitation
concentration
basis
Component in standard Component Component
(user- Response
name (user- area height
specified
specified
Y-value)
X-value)
A Area 20 μg/μL 10000 900 10000
B Height 100 μg/μL 12000 1100 1100
D Height 5 μg/μL 8000 700 700

3. Calculates a calibration curve (response versus concentration) for each

component listed in the Name column of the Components table.
4. Quantitates unknown samples comparing them against the generated
calibration curve, completing the following steps in sequence:
• Identifies each unknown peak by matching its retention time with a
component from the Components table.
• Calculates amount and concentration for each unknown peak from
the component calibration curve using sample peak response and
injection volume.
• Adjusts the amount and concentration by the sample weight and
dilution fields as entered during sample loading. The final

Quantitation 4-15
 calculated amount and concentration appear in the Peaks tab of the
Main and Results windows of Review.
The following figure illustrates the quantitation for peak components A,
B, and D. Each calibration curve uses the single-level fit type (linear
through zero).

4-16 Peak Identification and Quantitation of Sample Components

External standard component calibration curves (single-level,
concentration):
Response

10000 μV*sec Std

Sample response

0
Sample 20 μg/μl Std
concentration
Peak A

Response

1100 μV*sec Std

Sample response

0
Sample 100 μg/μl Std
concentration
Peak B

Response

700 μV*sec Std

Sample response

0
Sample 5 μg/μl Std
concentration
Peak D

Tip: If you are using a multilevel calibration curve, the software

performs an equivalent process.

Quantitation 4-17
Internal standard quantitation with separate standard and
unknown samples
This technique uses an internal standard added to both the standard and
unknown samples as a recovery standard. It is commonly used to correct for
losses during sample preparation.
This technique determines component amounts and concentrations by
applying a response to a calibration curve generated first by calculating the
response for the set of standards containing the internal standard. The
response is calculated from the responses of the component peak and its
internal standard peak. You select the type of response by using the Y-Value
field in the Components table of the Processing Method window.
The classic internal standard quantitation method plots the response ratio of
the standard component to the internal standard versus the amount or
concentration ratio of the standard component to the internal standard to
generate a calibration curve. The software uses an equivalent calibration
curve produced by plotting the response times the internal standard X-value
versus the component X-value, where the X-Value field is set to amount or
concentration in the Components table of the Processing Method window.

Classic response versus amount (or concentration) plot:

ResponseStd

ResponseIstd

0
AmountStd

AmountIstd

4-18 Peak Identification and Quantitation of Sample Components

Multiplying response by internal standard amount (or concentration):

ResponseStd
l AmountIstd
ResponseIstd

0
AmountStd

The following criteria are also required:

• You must define standard samples as standards, either by using the
Inject Standards function during sample loading in Run Samples
window or (after the sample is acquired) by defining a Sample Type of
Standard in Alter Sample.
• You must define unknown samples as unknowns, either by using the
Inject Unknowns function during sample loading in Run Samples
window or (after the sample is acquired) by defining a Sample Type of
Unknown in Alter Sample.
• You must enter component names and amounts or concentrations of
each standard component in the Default Amount tab of the Processing
Method window, the Component Editor of the Run Samples window, or
the Alter Sample window.
• The X-value of the calibration curve is either amount or concentration.
You choose amount or concentration as the x-axis in the X-Value field in
the Components tab of the Processing Method window.
• You define the fit type used for the calibration curve in the Components
table of the Processing Method window.

To perform internal standard quantitation (with separate standard and

unknown samples), the software operates as follows:
1. It identifies the component peaks in the chromatogram using peak
matching.
2. It determines the responses and amounts, or concentrations, for
standard peaks and internal standard(s). The software calculates a

Quantitation 4-19
 response for each standard peak and multiplies that value by the
amount, or concentration, of the internal standard component. The
resulting response value is plotted against the amount, or concentration,
value of the standard peak on the calibration curve for the component
with the same name.
Given a standard chromatogram like the one in the figure below, the
values used to determine the calibration points are the concentration
and response values from the table that follows.

Internal standard chromatogram:

Peak B

Peak A
Peak D
Peak C

Internal standard peak

Tip: In the following table, the X-values are set to amount.

Standard peak values, internal standard calibration with separate standard

and unknown samples:

Amount or
Quantitation
concentration
basis
Component in standard Component
(user- Response
name (user- area
specified
specified
Y-value)
X-value)
A Area 20 µg 10000 AreaA- = 10000 --------------- × 10 = 16.67
-----------------
AreaC 6000

B Area 100 µg 12000 AreaB 12000

------------------ = --------------- × 10 = 20.0
AreaC 6000

4-20 Peak Identification and Quantitation of Sample Components

 Standard peak values, internal standard calibration with separate standard
and unknown samples: (Continued)

Amount or
Quantitation
concentration
basis
Component in standard Component
(user- Response
name (user- area
specified
specified
Y-value)
X-value)
C (Int Std) Area 10 µg 6000 N/A
D Area 5 µg 8000 AreaD 8000
------------------- = ------------ × 10 = 13.33
AreaC 6000

3. Calculates a calibration curve (response times internal standard amount

versus Amount or Concentration) for each component listed in the Name
field of the Components table of the Processing Method window.
4. Quantitates unknown samples comparing them against the generated
calibration curve by performing the following steps in sequence:
• Identifies each unknown peak by matching its retention time with a
component from the Components table.
• Calculates a response for each matched peak by dividing the peak’s
response by its internal standard’s response.
• Generates a response by multiplying the response by the amount or
concentration of the internal standard for each matched peak.
• Calculates amount and concentration for each sample peak from the
component calibration curve using the sample peak response and
the injection volume.
• Adjusts amount and concentration by the Sample Weight and
Dilution fields as specified in the Samples tab of Run Samples or in
Alter Sample. The final calculated amount and concentration
appear in the Peaks tab of the Main and Results windows of Review.
The following figure illustrates quantitation for peak components A, B,
and D (internal standard C is not shown). Note that each calibration
curve uses the single-level fit type.

Quantitation 4-21
 Internal standard component calibration curves (single-level, amount):

Response

16.67 Std
RespSample
l AmtIstd
RespIstd
(from
Component
Loading Table) 0
Sample 20 μg Std

Peak A

Response

20.0 Std
RespSample
l AmtIstd
RespIstd
(from
Component 0
Loading Table)
Sample 100 μg Std

Peak B

Response

13.33 Std
RespSample
l AmtIstd
RespIstd
(from
Component
0
Loading Table)
Sample 5 μg Std

Peak D

Tip: If you are using a multilevel calibration curve, the software

performs an equivalent process.

4-22 Peak Identification and Quantitation of Sample Components

Internal standard quantitation without separate standard and
unknown samples (RF internal standard)
This technique is often used in gas chromatography. All samples are spiked
with one or more standard compounds that elute at a different time than the
unknown component or components you want to quantitate. When each
chromatogram is processed, it is treated as an unknown sample type. The
software calculates a response factor (RF) for the standard component(s) and
then quantitates the unknown components using the RF of the standard
component(s) rather than using coefficients of a calibration curve.
The following criteria are also required:
• You must define all samples as an RF internal standard, either by using
the Inject RF Internal Standards function during sample loading in Run
Samples window or (after the sample is acquired) by defining a Sample
Type of RF Internal Standard in Alter Sample.
• The X-value of the calibration curve is either amount or concentration
(defined in the X-Value field, in the Components tab of the processing
method).
• You must specify the names of the standard components in the
Components tab of the processing method.
• You must specify amounts or concentrations of each standard
component in the Default Amount tab of the processing method in the
Component Editor of the Run Samples window, or the Alter Sample
window.
• You may enter component names of the unknown components in the
Components tab of the processing method (optional).
• Unknown components whose names are defined in the Components tab
of the processing method are typically quantitated using the RF of a
standard component defined by using a curve reference peak.
Nevertheless, they may alternatively be quantitated using a default
peak.
• Unknown components whose names are not defined in the Components
tab of the processing method are quantitated using a default peak.
Tip: Use the Default Pk field to define which standard component’s
response factor to use during quantitation of named or unnamed
components. To use a standard peak as a default peak, in the
Components tab of the processing method, on the row containing the
default peak, check the Default Pk field. Define the region of the

Quantitation 4-23
 chromatogram in which this default peak should be used in the Default
Pk Start and Default Pk End fields. These fields allow you to use a
different default peak for different regions of your chromatogram, if
necessary. During quantitation, any peak that is detected within the
default peak start and end range will use the RF of that default peak in
determining its amount or concentration.
Tip: Use the Curve Reference field to define which standard component’s
RF to use during quantitation of named components. To use a Curve
Reference, in the Components tab of the processing method, on the row
for the named unknown component, specify the name of the appropriate
standard component in the Curve Reference field.
• The RF is calculated using the Y-value and the X-value (amount or
concentration) defined in the Components tab of the Processing Method
window.

To perform internal standard quantitation (without separate standards and

unknown samples), the software operates as follows:
1. It identifies the component peak(s) in the standard injection using peak
matching.
2. It determines the RF for each standard component using the following
formula:
Y Value-
RF = ------------------
X Value

where:
RF = Response factor
Y-Value = Response of the standard component calculated by the
software
X-Value = Component amount or concentration of the standard
component
Given an RF internal standard chromatogram such as the one in the
next figure, the values used to determine the RF are the amount and
response values from the table that follows.
Tip: When this type of internal standard method is used, no calibration
curves are generated.

4-24 Peak Identification and Quantitation of Sample Components

 RF Internal Standard Chromatogram:

Peak B

Peak A
Peak D
Peak C

Unknown Standard
component component
Unknown Unknown
component component

Tip: In the following table, the X-values are set to amount.

Standard peak values, RF internal standard calibration without separate

standard and unknown samples:

Amount or
concentration Component
Component Quantitation in standard area
Response factor
name basis (user- (response
specified or Y-value)
X-value)
C (standard Area 10 µg 6000 6000
------------ = 600
component) 10

3. Uses the response of each unknown component and the appropriate RF

to determine unknown component amounts (or concentrations) as
follows:

X Value = Y Value
------------------
RF

where:
RF = Response factor value calculated for the standard peak

Quantitation 4-25
 Y-Value = Response of the unknown component calculated by the
software
X-Value = Component amount or concentration
The values used to determine the amounts of the unknown components
in the following table are the RFs determined in the previous table and
the unknown component values in the following table.
Tip: In the following table, the software-determined X-values are
amounts.

Unknown component values, RF internal standard calibration without

separate standard and unknown samples:

Amount or
Component
concentration in
Component area
Quantitation basis unknown component
name (response or
(software-determined
Y-value)
X- value)
A (unknown Area 10000 10000
--------------- = 16.667
component) 600 µg
B (unknown Area 12000 12000
--------------- = 20
component) 600 µg
D (unknown Area 8000 8000
------------ = 13.333
component) 600 µg

4-26 Peak Identification and Quantitation of Sample Components

Calibration curve fit types
A variety of calibration curve fits are available for external and internal
standard calibration with multiple levels of standards. The calibration curve
fit types are divided into three groups of increasing complexity:
• Single-level calibration (linear through zero and response factor)
• Multilevel calibration matrix operations:
– Multilevel calibration (linear, inverse linear, log-log linear,
quadratic, cubic, fourth-order, and fifth-order, power curve)
– Multilevel forced through zero (linear, quadratic, cubic,
fourth-order, fifth-order, and response factor)
• Multilevel calibration (point-to-point and cubic spline)
Tip: You can apply weighting only to the linear, quadratic, cubic, fourth-order,
and fifth-order fit types.
When you refer to the calibration curve fit types in this section keep in mind
this information:
• The software uses matrix operations to perform multilevel calibration.
• For background material on the processes used by the software to create
calibration curves, see “References” on page 4-53.
• The equations shown in the following examples are not adjusted for
sample weight and dilution (see “Quantitation” on page 4-7).

Single-level calibration curve

For single-level calibration, the curve fit is linear; the intercept is zero. The
software supports the following single-level calibration curve fits:
• Linear through zero
• Response factor
The following figure illustrates a single-level calibration curve.

Calibration curve fit types 4-27

 Single-level calibration curve:

Linear through zero

The linear-through-zero calibration curve is represented by the equation:
y = Bx

where:
y = Response of the standard component calculated by the software
B = Slope of the calibration curve
x = Component amount or concentration

4-28 Peak Identification and Quantitation of Sample Components

The component amount or concentration for a quantitated sample can be
determined by the equation:
y-
x = ---
B

where:
x = Component amount or concentration
y = Response of the sample peak calculated by the software
B = Slope of the calibration curve

Response factor
The response factor (RF) fit type eliminates the need to create an RF custom
field. When using an RF fit type, you should specify the appropriate X-Value
and Y-Value in the Components tab of the processing method as when using a
linear-through-zero fit.
The software plots the standard component’s response versus its amount (or
concentration) on the calibration curve. The RF is the slope of the curve. If
multiple data points are plotted on the calibration curve, the RF for each point
is determined, then the average RF is used as the slope of the curve.
The RF is represented by the equation:

Y Value-
RF = ------------------
X Value

where:
RF = Response factor (slope of the calibration curve)
y = Response of the standard component calculated by the software
x = Component amount or concentration of the standard component
A linear-through-zero fit to the average RF point results in the equation of the
curve.
The following figure illustrates an RF calibration curve.

Calibration curve fit types 4-29

 Response factor calibration curve:

Multilevel calibration matrix operations

For the multilevel polynomial curve fits (linear, inverse linear, log-log linear,
quadratic, cubic, fourth-order, fifth-order, and through zero fits), the software
uses matrix operations to obtain the required coefficients.
For all polynomial fits, the software uses an unweighted or weighted
least-squares fitting technique to a set of x-y points or x, y, and weight points.
This technique is the numerical routine called LU (lower and upper triangular
matrix) Decomposition for weighted and unweighted fits.
• When weighting is disabled in the Components table, the software uses
an unweighted least-squares fitting technique to a set of x-y points.
• When weighting is enabled in the Components table, the software uses a
weighted least-squares fitting technique to a set of x-y weight points.

4-30 Peak Identification and Quantitation of Sample Components

Matrix operations
The objective is to use the least-squares fit to calculate the coefficients for the
curve:
[Y] = [A] × [C]
where:
[Y] = Response vector
[A] = Design matrix
[C] = Coefficient vector
To accomplish this, the least-squares fit solves the following linear normal
equations:
([A]T × [W] × [A]) × [C] = [A]T × [W] × [Y]
where [W] is a diagonal weight matrix, which becomes unity (all diagonal
elements equal to 1) for an unweighted fit. The solution is:
[C] = ([A]T × [W] × [A]) -1 × ([A]T × [W] × [Y])
The matrix inversion is done by using LU Decomposition. The least-squares
fit is described in Section 15.4 General Linear Least Squares in Numerical
Recipes in C William H. Press, et al., (2nd Edition).

Matrix operations example

The software uses the following matrices to calculate the coefficients based on
the number of standards that are run. These operations illustrate an
unweighted multilevel fifth-order fit.
As an example, assume that seven standards are run, one at each level. The
software attempts to apply a fifth-order fit to the calibration points.

To find the coefficients of the calibration equation:

1. The seven standards produce the following amount, response sets (x, y
plotted points on the calibration curve):
(x1, y1), (x2, y2), (x3, y3), (x4, y4), (x5, y5), (x6, y6), (x7, y7)
2. The following seven equations, using the data in step 1, contain the six
unknown coefficients (c5 through c0) and the seven sets of points:
5 4 3 2 1 0
y1 = c5(x1) + c4(x1) + c3(x1) +c2(x1) + c1(x1) + c0(x1)

Calibration curve fit types 4-31

 5 4 3 2 1 0
y2 = c5(x2) + c4(x2) + c3(x2) + c2(x2) + c1(x2) + c0(x2)
5 4 3 2 1 0
y3 = c5(x3) + c4(x3) + c3(x3) +c2(x3) + c1(x3) + c0(x3)
5 4 3 2 1 0
y4 = c5(x4) + c4(x4) + c3(x4) + c2(x4) + c1(x4) + c0(x4)
5 4 3 2 1 0
y5 = c5(x5) + c4(x5) + c3(x5) + c2(x5) + c1(x5) + c0(x5)
5 4 3 2 1 0
y6 = c5(x6) + c4(x6) + c3(x6) + c2(x6) + c1(x6) + c0(x6)
5 4 3 2 1 0
y7 = c5(x7) + c4(x7) + c3(x7) + c2(x7) + c1(x7) + c0(x7)
3. The previous equations can now be written using matrix notation as:

x 15 x 14 x 13 x 12 x 11 x 10

x 25 x 24 x 23 x 22 x 21 x 20
y1
c5
y2 x 35 x 34 x 33 x 32 x 31 x 30 c4
y3
= • c3
y4 x 45 x 44 x 43 x 42 x 41 x 40 c2
y5
c1
y6 x 55 x 54 x 53 x 52 x 51 x 50
c0
y7
x 65 x 64 x 63 x 62 x 61 x 60

x 75 x 74 x 73 x 72 x 71 x 70

or
[Y] = [A] × [C]
where:
[Y] = Response vector
[A] = Design matrix
[C] = Vector of coefficient to be computed
Design Matrix A is constructed with n+1 columns and i rows (where n is
the order of the polynomial and i is the number of levels). The

4-32 Peak Identification and Quantitation of Sample Components

 construction of Design Matrix A for the fifth-order fit type is previously
illustrated.
4. The software then uses LU Decomposition to solve the normal equations
of the least-squares fit:
T T
([A] × [A]) × [C] = [A] × [Y]

Multilevel calibration curves

The software supports the following multilevel calibration curve fits:
• Point-to-point
• Cubic spline
• Linear
• Inverse linear
• Log-log linear
• Quadratic
• Cubic
• Fourth-order
• Fifth-order
• Power curve
The equations used to calculate the goodness-of-fit statistics are described in
“Statistics” on page 4-48, with the following results:
• For all fit types, the software reports only positive X-value amounts or
concentrations.
• For linear fit types, the software reports X-values within the range of
the calibration curve (from 0 to the highest X-value) as well as X-values
greater than the highest X-value because it extrapolates values above
the highest X-value of the standard data point.
• For all nonlinear fit types, the software reports X-values from 0 to the
highest X-value of the standard points.

Calibration curve fit types 4-33

 Point-to-point fit
To calculate a point-to-point calibration curve, the software performs a linear
fit between the different levels. The first and last segments of the curve are
extrapolated linearly so that they can be used to calculate X-values that fall
outside the range of the lowest to highest X-value.
Because the point-to-point calibration curve is fit through every point, the
correlation coefficient equals 1, and the standard error equals 0. No curve
coefficients are calculated or stored for this fit type. The following figure
illustrates a point-to-point calibration curve.

Point-to-point calibration curve:

Each point-to-point segment of the calibration curve is represented by the

equation:
y = Ai + Bi x

4-34 Peak Identification and Quantitation of Sample Components

where:
y = Response of the standard peak calculated by the software
Ai = y-intercept of the ith curve segment
Bi = Slope of the ith curve segment
x = Component amount or concentration

Determining component amount and/or concentration

Component amount and/or concentration for a quantitated sample peak can
be determined by the equation:
y–A
x = ---------------i
Bi

where:
x = Component amount and/or concentration
y = Response of the sample peak, calculated by the software
Ai = y-intercept of the ith curve segment
Bi = Slope of the ith segment

Cubic spline fit

To generate a cubic spline calibration curve, the software performs a cubic
polynomial fit between every two successive levels, matching the slope and
curvature at every point boundary. The cubic spline fit adjusts the shape of
the calibration curve on a point-by-point basis.
Because the cubic spline calibration curve is fit through every point, the
correlation coefficient = 1, and the standard error = 0. No curve coefficients
are calculated or stored when the cubic spline fit type is used.
The following figure illustrates a cubic spline calibration curve.

Calibration curve fit types 4-35

 Cubic spline calibration curve:

Each cubic spline segment of the calibration curve is represented by the

equation:
y = Ai + Bi + Ci x 2 + Di x 3

where Ai, Bi, Ci, and Di are the polynomial coefficients of the segment.
Tip: The software uses an iterative method to solve for x when given y.

4-36 Peak Identification and Quantitation of Sample Components

Linear fits
Linear fits include linear, inverse linear, log-log linear, and power curve.
Linear fit
To calculate a linear calibration curve, the software calculates the line that
best fits the amounts or concentrations and responses of the calibration
points. The Y-value of each point is the response of the standard peak and the
X-value of each point is the amount or concentration of the standard peak. The
following figure illustrates a linear least-squares fit calibration curve.
Inverse linear fit
To calculate an inverse linear calibration curve, the software performs a
linear fit to the X- and Y-values of the calibration points. The Y-value of the
point is the response of the standard peak and the X-value is the 1/X value
(either amount or concentration) of the standard peak.
Log-log linear fit
To calculate a log-log linear calibration curve, the software performs a linear
fit to the X- and Y-values of the calibration points. The Y-value of each point is
the log of the response of the standard peak and the X-value is the log of the
X-value (either amount or concentration) of the standard peak.
Tip: Inverse linear and log-log linear use the linear fit equation.
Power curve fit
To calculate a power curve, the software performs a linear fit of log y versus
log x, where y is the response and x is the X-value. The plot and equation of
the curve are reported in terms of the X-value and response.
The power curve fit generates a calibration curve represented by the equation:
B
y = Ax

where:
y = Response of the standard peak calculated by the software
A = Multiplier in the calibration curve formula
B = Exponent in the calibration curve formula
x = Component amount or concentration

Calibration curve fit types 4-37

 Linear least-squares fit calibration curve:

The linear fit generates a calibration curve represented by the equation:

y = A + Bx

where:
y = Response of the standard peak calculated by the software
A = y-intercept of the calibration curve
B = Slope of the calibration curve
x = Component amount or concentration

4-38 Peak Identification and Quantitation of Sample Components

 Determining component amount or concentrations
Component amount or concentration for a quantitated sample peak can be
determined by the equation:

x = y------------
– A-
B

where:
x = Component amount or concentration
y = Response of the sample peak calculated by the software
A = y-intercept of the calibration curve
B = Slope of the calibration curve

Quadratic fit
To calculate a quadratic calibration curve, the software performs a
least-squares fit of a quadratic polynomial to the calibration points. The fit
cannot be performed with fewer than three calibration points, and a minimum
of five points is strongly recommended.
The following figure illustrates a quadratic fit calibration curve.

Calibration curve fit types 4-39

 Quadratic fit calibration curve:

The quadratic fit generates a calibration curve that is represented by the

equation:
y = A + Bx + Cx 2

where:
y = Response of the standard peak calculated by the software
x = Component amount or concentration
A, B, and C = Polynomial coefficients of the curve
Determining component amount or concentration
Component amount or concentration for a quantitated sample peak can be
determined by solving for x:

– B ± B 2 – 4C ( A – y )
x = ----------------------------------------------------------
2C

4-40 Peak Identification and Quantitation of Sample Components

where:
y = Response of the sample peak calculated by the software
x = Component amount or concentration
A, B, and C = Polynomial coefficients of the curve
The software reports only positive values of x that are within the range of the
calibration curve.

Cubic fit
To calculate a cubic fit calibration curve, the software performs a
least-squares fit of a cubic polynomial to the calibration points. The fit cannot
be performed with fewer than four calibration points, and a minimum of six
points is strongly recommended.
The following figure illustrates a cubic fit calibration curve.

Cubic fit calibration curve:

Calibration curve fit types 4-41

 The cubic fit generates a calibration curve that is represented by the equation:

y = A + Bx + Cx 2 + Dx 3

where:
y = Response of the standard peak calculated by the software
x = Component amount or concentration
A, B, C, and D = Polynomial coefficients of the curve

Fourth-order fit
To calculate a fourth-order fit calibration curve, the software performs a
least-squares fit of a fourth-order polynomial to the calibration points. The fit
cannot be performed with fewer than five calibration points, and a minimum
of seven points is strongly recommended.
The following figure illustrates a fourth-order fit calibration curve.

4-42 Peak Identification and Quantitation of Sample Components

Fourth-order fit calibration curve:

The fourth-order fit generates a calibration curve that is represented by the

equation:

y = A + Bx + Cx 2 + Dx 3 + Ex 4

where:
y = Response of the standard peak calculated by the software
x = Component amount or concentration
A, B, C, D, and E = Polynomial coefficients of the curve

Fifth-order fit
To calculate a fifth-order fit calibration curve, the software performs a
least-squares fit of a fifth-order polynomial to the calibration points. The fit
cannot be performed with fewer than six calibration points, and a minimum of
eight points is strongly recommended.

Calibration curve fit types 4-43

 The following figure illustrates a fifth-order fit calibration curve.

Fifth-order fit calibration curve:

The fifth-order fit generates a calibration curve that is represented by the

equation:
y = A + Bx + Cx 2 + Dx 3 + Ex 4 + Fx 5

where:
y = Response of the standard peak calculated by the software
x = Component amount or concentration
A, B, C, D, E, and F = Polynomial coefficients of the curve

4-44 Peak Identification and Quantitation of Sample Components

Multilevel forced-through-zero calibration curves
The software supports forced-through-zero for the following multilevel curve
fits:
• Linear
• Quadratic
• Cubic
• Fourth-order
• Fifth-order
• Response factor
Each forced-through-zero fit is similar to the corresponding
nonforced-through-zero fit except that the curve is mathematically
constrained to pass through zero. Forcing the calibration curve through zero
results in different coefficients than those for nonforced calibration curves.
For forced-through-zero fits, the zeroth order coefficient (C0) is set to 0, and
the software computes the remaining coefficients.
5 4 3 2 1 0
y1 = c5(x1) + c4(x1) + c3(x1) + c2(x1) + c1(x1) + 0(x1)
5 4 3 2 1 0
y2 = c5(x2) + c4(x2) + c3(x2) + c2(x2) + c1(x2) + 0(x2)
5 4 3 2 1 0
y3 = c5(x3) + c4(x3) + c3(x3) + c2(x3) + c1(x3) + 0(x3)
5 4 3 2 1 0
y4 = c5(x4) + c4(x4) + c3(x4) + c2(x4) + c1(x4) + 0(x4)
5 4 3 2 1 0
y5 = c5(x5) + c4(x5) + c3(x5) + c2(x5) + c1(x5) + 0(x5)
5 4 3 2 1 0
y6 = c5(x6) + c4(x6) + c3(x6) + c2(x6) + c1(x6) + 0(x6)
5 4 3 2 1 0
y7 = c5(x7) + c4(x7) + c3(x7) + c2(x7) + c1(x7) + 0(x7)
The following table shows the differences between standard equations and
forced-through-zero equations.

Standard and forced-through-zero equation format comparison:

Forced-through-zero
Fit type Standard equation
equation
Linear y = A + Bx y = Bx

Calibration curve fit types 4-45

 Standard and forced-through-zero equation format comparison: (Continued)

Forced-through-zero
Fit type Standard equation
equation
Quadratic y = A + Bx + Cx 2 y = B + Cx 2

Cubic y = A + Bx + Cx 2 + Dx 3 y = Bx + Cx 2 + Dx 3

For information on forced-through-zero fits, see the relevant

nonforced-through-zero section for that type of fit. The software performs an
equivalent computation, but coefficient A is always zero.

Weighting
Weighting is applied while fitting curves to multilevel points to achieve the
following ends:
• Ensure that the points that have the most certainty (least error)
contribute most strongly to the determination of the coefficients.
• Adjust for the differences in precision of the Y-value (response or
response ratio) with respect to the X-value (amount or concentration).
To fit a curve to the calibration data, the software performs a least-squares fit
to select the coefficients that minimize the sum of the differences between the
individual points in the curve.
• Without weighting, all points contribute equally to that sum.
• With weighting, the contributions are adjusted to reflect the variance at
each calibration level.
The equation that is minimized is:
( ŷ i – y i ) 2 w i
∑ --------------------------------------------------------
DegreesOfFreedom
-
i=1

where:
yi = Observed data point
ŷ i = Calculated data point
wi = Weighting factor for each data point

4-46 Peak Identification and Quantitation of Sample Components

 Degrees Of Freedom = The number of points minus the number of
coefficients calculated
Unweighted data assumes equal precision at all levels (wi = 1).
To select the weighting type, plot the standard deviation for each level versus
the X-value. Then select the weighting type based on the observed variation of
the standard deviation by level.
Weighting can be applied to the standard fit types:
• Linear
• Quadratic
• Cubic
• Fourth-order
• Fifth-order
The types of weighting and the results of their application are described in the
following table.

Weighting application results:

Weighting type Weighting equation

x wi = xi: This results in a fit to the points at the high end
of the curve (amounts or concentrations).
1/x and 1/x2 2
wi = 1/xi or wi= 1/xi : This results in a fit to the points at
the low end of the curve (amounts or concentrations).
The weight for the point and the coefficients cannot be
calculated if x = 0. If there is a point where x = 0, the
coefficients of the curve are not calculated and a
processing code Q28 is copied into the curves code field.
1/y and 1/y2 wi = yi: This results in a fit to the points at the low end of
the curve (response). The weight for the point and the
coefficients cannot be calculated if y = 0. If there is a
point where y = 0, the coefficients of the curve are not
calculated and a processing code Q30 is copied into the
curves code field.
x2 wi = xi2: This results in a fit to the points at the high end
of the curve (amounts or concentrations).

Calibration curve fit types 4-47

 Weighting application results: (Continued)

Weighting type Weighting equation

log x wi = log xi: Produces a fit that weights the points on the
calibration curve by a factor of log base 10, resulting in a
logarithmic fit to the points on the high end of the
calibration curve. If xi < 0, the weight for the point and
the coefficients of the calibration curve are not
calculated and a processing code Q29 is copied into the
curves code field.
ln x wi = ln xi: Produces a fit that weights the points on the
calibration curve by a factor of the natural log of X,
resulting in a logarithmic fit to the points on the high
end of the calibration curve. If xi < 0, the weight for the
point and the coefficients of the calibration curve are not
calculated and a processing code Q29 is copied into the
curves code field.

where:
wi = Weighting factor for each data point
xi = X value of the point
yi = Y value of the point
Tip: If the software cannot calculate the weighted points, the coefficients for
the curve are not calculated and a processing code is generated, indicating the
reason it is not calculated.

Statistics
Statistics indicate goodness of fit. The software calculates the following
statistical criteria:
• Coefficient of determination
• Correlation coefficient
• Residual sum of squares
1
• Standard error of estimate of y on x (no report)
1
• Standard variance (no report)
• Standard error of calibration

4-48 Peak Identification and Quantitation of Sample Components

 • Percent Relative Standard Deviation
• Calculated value and percent deviation of the calibration points

Coefficient of determination
2
Coefficient of determination (R ) is a rough indicator of the goodness of fit and
is calculated by:
2
( Sy )
R 2 = 1 – --------------
2
σ y

where:
R2 = Coefficient of determination
R = Correlation coefficient
Sy = Standard error of estimate of y on x
σ2y = Standard variance

Correlation coefficient
The correlation coefficient (R) is an indicator of goodness of fit. It is the square
root of the coefficient of determination.

Standard error of estimate of Y on X

The standard error of estimate of y on x (Sy) determines R2 (the coefficient of
determination) and R (the correlation coefficient) and is calculated by:

n 2
1 ˆ
Sy = --- ∑ w i ( y i – y i )
ni = 1

where:
n = Number of points

1. The software calculates these two criteria that are not reported as
intermediate values.

Calibration curve fit types 4-49

 wi = Weighting factor (set to 1 for uniform weighting)

ˆ
yi = Responses as predicted using the calibration curve

yi = Response of a calibration point

Standard variance
2
The standard variance (σ y) is used to calculate the coefficient of
determination and correlation coefficient. It is computed as follows:
n
1 2
σ 2 y = --- ∑ w i ( y i – y )
ni = 1

where:
wi =Weighting factor (set to 1 for uniform weighting)

yi =Response of a calibration point

y =Weighted mean given by the equation:

∑ wi yi
y = i---------------------
=1
n

∑ wi
i=1

Residual sum of squares

Residual sum of squares (RSS) is an indicator of goodness of fit and precision
of data. It is used to calculate the standard error of estimate and the standard
error of calibration. It is calculated by:
n
ˆ 2
RSS = ∑ wi ( yi – yi )
i=1

where:
RSS = Residual sum of squares

4-50 Peak Identification and Quantitation of Sample Components

 n = Number of points

wi = Weighting factor (set to 1 for uniform weighting)

yˆi = Responses as predicted using the calibration curve

yi = Response of a calibration point

Standard error of calibration

The standard error of calibration (E) is the square root of the sum that is
minimized when fitting coefficients to the curve and is calculated by:

⎛ n ⎞
1⎜
--- ∑ w i ( yˆi – y i ) ⎟ = --1- RSS
2
E =
⎜ ⎟ d
d⎝ ⎠
i=1

where:
d = Degrees of Freedom = Number of points minus the number of
coefficients calculated
wi = Weighting factor (set to 1 for uniform weighting)
ˆ
yi = Responses as predicted using the calibration curve

yi = Response of a calibration point

RSS = Residual sum of squares

Calculated value and percent deviation of calibration points

The calculated value and percent deviation of calibration points can be used to
assess how well the points fit the curve by visual inspection or by plotting
against the X value.
Percent deviation is calculated by:
ˆ
⎛ x i – x i⎞
% Deviation = 100 ⎜ --------------
-⎟
⎝ xˆi ⎠

where:

Calibration curve fit types 4-51

 xˆi
= X value as predicted using the calibration curve (the calculated
value)
xi = X value of the calibration point

Plots of percent deviation versus amount should display random scatter if the
fit type is correct. Plots of calculated value versus amount or concentration
should be linear.

Percent relative standard deviation

The Percent RSD is an indication of goodness of fit and precision of the data.
Percent RSD is calculated by:
1
---
⎛ n 2 ⎞2
⎜ ⎟
⎜∑ [ w • y – YWM ] ⁄ [n-1]
i i ⎟
⎝i = 1 ⎠
%RSD = --------------------------------------------------------------------------------- • 100
YWM

where:
wi = Weighting factor (set to 1 for uniform weighting)
yi = Response of a calibration point
YWM = Weighted mean response of all calibration points, which is
expressed as:

∑ ( wi • yi )
i=1
YWM = -------------------------------
n -

n = The number of points

4-52 Peak Identification and Quantitation of Sample Components

References
For further information on the theory of quantitation, see:
• Press, William H., Teukolsky, Saul A., Vetterling, William T., and
Flannery, Brian P., Numerical Recipes, Cambridge University Press,
Cambridge, UK, 2007.
• Massart, D.L., et al., Chemometrics: A Textbook, Elsevier Science
Publishers, Amsterdam, 1988.
• Papoulis, Athanasios, Signal Analysis, McGraw-Hill, New York, 1977.
• Snyder, Lloyd R., Kirkland, Joseph J., and Dolan, John W., Introduction
to Modern Liquid Chromatography, third ed., Wiley-Interscience,
Hoboken, New Jersey, 2010.
• Strang, Gilbert, Linear Algebra and Its Applications, Harcourt Brace
Jovanovich, Inc., New York, 1988.

References 4-53
4-54 Peak Identification and Quantitation of Sample Components
Index
Numerics B
5-point peak rejection, Traditional 3-19 Baseline construction, Traditional 3-14
Baseline determination, ApexTrack
A 2-3, 2-15
A/D conversion 1-4 Baseline noise 2-5
Analog-to-digital conversion 1-4 estimating 2-33
Apex role in ApexTrack 2-31
defining in ApexTrack 2-12 Baseline, ApexTrack 2-12
detection in ApexTrack 2-3
ApexTrack C
apex determination 2-3 Calculated value and percent deviation
apex, defining 2-12 of calibration points 4-51
Auto-Peak Width 2-10, 2-30 Calibration curve
AutoThreshold 2-11, 2-33 calculated value and percent
baseline determination 2-3 deviation of calibration
crossover 2-24 points 4-51
Detection Threshold 2-5 coefficient of determination 4-49
Gaussian skim 2-7 correlation coefficient 4-49
inflection points 2-12, 2-15 residual sum of squares 4-50
integration 1-2 standard error of calibration 4-51
integration events 2-6, 2-35, 2-36 standard error of estimate of Y on X
integration peak labels 2-8 4-49
liftoff 2-21 standard variance 4-50
Liftoff % 2-5 Calibration curve fits
negative peaks 2-7 cubic 4-41
peak detection process 2-9 cubic spline 4-35
Peak Width 2-4 fifth-order 4-43
procedure for peak detection 2-13 forced-through-zero 4-45
processing parameters 2-4 inverse linear 4-37
round peak 2-14 linear 4-37
slope difference threshold 2-16 linear through zero 4-28
touchdown 2-21 log-log linear 4-37
Touchdown % 2-5 point-to-point 4-34
Area, Traditional integration 3-18 quadratic 4-39
Auto-Peak Width, ApexTrack 2-10, response factor 4-29
2-30 statistics 4-48
AutoThreshold, ApexTrack 2-11, 2-33 weighting 4-46

Index-1
Calibration during quantitation 4-7 GPC processing, Merge Peaks event
Coefficient of determination 4-49 2-6
Conversion, A/D 1-4
Correlation coefficient 4-49 H
Crossover Hard disk space 1-7
ApexTrack 2-24 I
Cubic fit 4-41 Inflection points, ApexTrack 2-12, 2-15
Cubic spline fit 4-35 Integration
Curvature, use in ApexTrack 2-9 ApexTrack 1-2
D ApexTrack algorithm 2-4
Data acquisition 1-2 Traditional 1-3
Data bunching, Traditional 3-3 Integration events
Data storage 1-5 ApexTrack 2-6, 2-36
Data transfer 1-5 ApexTrack Integration table 2-35
Detect Shoulders event, ApexTrack 2-7 Traditional integration, overview
Detection events 3-20
Detect Shoulders 2-7 Integration peak labels, ApexTrack 2-8
Detection sampling rate 1-6 Integration tab in ApexTrack
Detection Threshold, ApexTrack 2-5, processing method 2-4
2-10 Integration, definition 1-2
Dilution 4-10 Internal standard quantitation 4-18
Disk space 1-7 Inverse linear fit 4-37

E L
External standard quantitation 4-13 Liftoff %, ApexTrack 2-5, 2-16
Liftoff in ApexTrack 2-21
F Linear fit 4-37
Fifth-order fit 4-43 Linear through zero fit 4-28
Forced-through-zero 4-45 Log-log linear fit 4-37
Fused peaks
second derivative, ApexTrack 2-13 M
Traditional 3-12 Match difference 4-3
Matrix operations
G example 4-31
Gaussian skimming, ApexTrack 2-3, multilevel calibration 4-30
2-7 Maximum curvature of peaks,
GPC ApexTrack 2-9
Liftoff % and Touchdown % 2-5 Merge Peaks event
GPC processing 2-6

Index-2
Minimum Area retention time, Traditional 3-17
Traditional 3-19 Peak labels
Minimum Height ApexTrack 2-8
Traditional 3-19 Traditional integration 3-21 I
Peak matching 4-2
N match difference 4-3
Negative peak detection in ApexTrack retention times, shifting 4-4
2-3 Update Retention Time 4-5
P Peak rejection criteria, Traditional
Peak apex, determining, Traditional 3-19
3-5 Peak start, determining, Traditional
Peak area, determining, Traditional 3-4
3-18 Peak Width, ApexTrack 2-4, 2-9
Peak detection Peak, maximum curvature, ApexTrack
ApexTrack 2-3, 2-9 2-9
data bunching, Traditional 3-3 Peak-to-peak baseline noise 2-33
End (min), ApexTrack 2-4 Point-to-point fit type 4-34
peak apex, ApexTrack 2-3 Processing methods
peak apex, Traditional 3-5 default in ApexTrack 2-4
peak end, Traditional 3-6 definition 1-2
peak start, Traditional 3-4 Processing, definition 1-2
Start (min), ApexTrack 2-4 Q
Traditional 3-2 Quadratic fit 4-39
Peak end, determining, Traditional 3-6 Quantitation
Peak height, determining, Traditional by calibration 4-7
3-17 external standard 4-13
Peak integration fit type 4-8
ApexTrack peak labels 2-8 internal standard 4-18
baseline construction, Traditional RF Internal Standard 4-23
3-14 using dilution 4-10
determining fused peaks, using injection volume 4-11
Traditional 3-12 using sample weight 4-9
events, Traditional 3-20
overview, Traditional 3-11 R
peak area, Traditional 3-18 Raw data points 1-5
peak height, Traditional 3-17 Rejection
peak labels, Traditional 3-21 criteria, Traditional 3-19
peak rejection criteria, Traditional Residual sum of squares 4-50
3-19 Response factor fit 4-29

Index-3
Retention time Touchdown, in ApexTrack 2-21
determining 3-17 Traditional integration 1-3, 3-11
Retention time reference 4-4 peak detection 3-2
shifting 4-4 peak integration events 3-20
RF Internal Standard quantitation peak labels 3-21
4-23
Round peak, ApexTrack 2-3, 2-7, 2-13, U
2-14 Update RT 4-5
RT Reference 4-4 Updating retention time 4-5
Upslope points, ApexTrack 2-12
S
Sample weight 4-9 V
Sampling rate Valley-to-Valley event, ApexTrack 2-7
data collection frequency 1-6 W
formula 1-6 Weighting 4-46
optimum sampling rate 1-6
Second derivative
apex 2-15
example 2-13
filter, ApexTrack 2-10
peak profile 2-11
threshold, ApexTrack 2-10
Second derivative filter, ApexTrack 2-9
Shifting retention times 4-4
Shoulders
detection 2-3, 2-13
Skim events
ApexTrack 2-3
Slope difference threshold 2-17
ApexTrack 2-16
Slope differences, ApexTrack 2-17
Standard error of calibration 4-51
Standard error of estimate of Y on X
4-49
Standard variance 4-50
Statistics 4-48
T
Tangential Skim 2-7
Touchdown %, ApexTrack 2-5, 2-16

Index-4