v>EPA
United States
Environmental Protection
Agency
Health Effects Research
Laboratory
Cincinnati OH 45268
EPA-600/1-78-058
September 1978
Research and Development
The Multielemental
Analysis of
Drinking Water
Using Proton-
Induced X-Ray
Emission (PIXE)
EP 600/1
78-058
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U S Environmental
Protection Agency, have been grouped into nine series These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields
The nine series are
1 Environmental Health Effects Research
2 Environmental Protection Technology
3 Ecological Research
4 Environmental Monitoring
5 Socioeconomic Environmental Studies
6. Scientific and Technical Assessment Reports (STAR)
7 Interagency Energy-Environment Research and Development
8 "Special" Reports
9 Miscellaneous Reports
This report has been assigned to the ENVIRONMENTAL HEALTH EFFECTS RE-
SEARCH series This series describes projects and studies relating to the toler-
ances of man for unhealthful substances or conditions This work is generally
assessed from a medical viewpoint, including physiological or psychological
studies In addition to toxicology and other medical specialities, study areas in-
clude biomedical instrumentation and health research techniques utilizing ani-
mals — but always with intended application to human health measures
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
EPA-600/1-78-058
September 1978
THE MULTIELEMENTAL ANALYSIS OF DRINKING WATER
USING PROTON-INDUCED X-RAY EMISSION (PIXE)
by
P. C. Simms and F. A. Rickey
Purdue Research Foundation
Division of Sponsored Programs
West Lafayette, Indiana 47907
Contract No. 68-03-2178
Project Officer
Nancy S. Ulmer
Field Studies Division
Health Effects Research Laboratory
Cincinnati, Ohio 45268
HEALTH EFFECTS RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U. S. ENVIRONMENTAL PROTECTION AGENCY
CINCINNATI, OHIO 45268
-------
DISCLAIMER
This report has been reviewed by the Health Effects Research Laboratory,
U. S. Environmental Protection Agency, and approved for publication.
Approval does not signify that the contents necessarily reflect the views
and policies of the U. S. Environmental Protection Agency, nor does mention
of trade names or commercial products constitute endorsement or recommenda-
tion for use.
11
-------
FOREWORD
The U.S. Environmental Protection Agency was created because of in-
creasing public and government concern about the dangers of pollution to
the health and welfare of the American people. Noxious air, foul water,
and spoiled land are tragic testimony to the deterioration of our natural
environment. The complexity of that environment and the interplay be-
tween its components require a concentrated and integrated attack on
the problem.
Research and development is that necessary first step in problem
solution and it involves defining the problem, measuring its impact, and
searching for solutions. The primary mission of the Health Effects Re-
search Laboratory in Cincinnati (HERL) is to provide a sound health effects
data base in support of the regulatory activities of the EPA. To this end,
HERL conducts a research program to identify, charcterize, and quantitate
harmful effects of pollutants that may result from exposure to chemical,
physical, or biological agents found in the environment. In addition to
valuable health information generated by these activities, new research
techniques and methods are being developed that contribute to a better
understanding of human biochemical and physiological functions, and how
these functions are altered by low level insults.
This report describes a rapid and economical method for the simul-
taneous determination of 76 elements in aqueous samples. The occurrence
data, provided by this multielemental approach, will afford a sound basis
for future epidemiological and toxicologir&al studies.
Garner
Director
Health Effects Research Laboratory
111
-------
ABSTRACT
Proton Induced X-ray Emission (PIXE) was used in this research program
to provide analysis of drinking water samples for 76 elements heavier than
aluminum. Atomic absorption was used for sodium analysis. A new technique
called "vapor filtration" was used to prepare suitable targets for PIXE
analysis from aqueous samples. Excellent detection limits (0.1 to 100 ppb)
were obtained for most elements heavier than silicon. Energetic protons
were provided by a nuclear accelerator, the x rays were observed with an
energy sensitive semiconductor detector, and the data were recorded and
processed with an on-line computer. Advanced electronic systems and a high
degree of automation were used to provide high quality, low cost analysis
for 6200 samples. The experimental equipment and procedures are described.
The results of the analysis of drinking water samples will be available from
the Health Effects Research Laboratory of the U. S. Environmental Protection
Agency.
This report was submitted in fulfillment of Contract No. 68-03-2178 by
Purdue University under the sponsorship of the U. S. Environmental
Protection Agency. This report covers the period February 24, 1975 to
April 29, 1977, and work was completed as of April 26, 1977.
-------
CONTENTS
Foreword iii
Abstract iv
Figures vi
Tables vii
List of Abbreviations and Symbols viii
Acknowledgments ix
1. Introduction 1
General principles of PIXE analysis 1
Techniques developed for PIXE analysis of aqueous
samples 4
2. Conclusions and Recommendations 5
3. General Description of Techniques and Procedures 6
Sample preparation 6
Bombardment chamber and proton beam controls 9
Detector and electronics 10
Computer processing 13
System calibration 16
4. Initial Evaluation of PIXE Analysis for Aqueous Samples ... 21
5. Precision and Detection Limits 27
Definitions 27
Evaluation and discussion 29
6. Procedures and Quality Control Used for PIXE Analysis of
Drinking Water Samples 37
7. Analysis for Sodium Using Atomic Absorption 46
References 48
-------
FIGURES
Number Page
1 X-ray yield for PIXE analysis 2
2 Typical background seen in PIXE analysis 3
3 Vapor filtration apparatus 8
4 Schematic drawing of bombardment and detection apparatus 11
5 X-ray spectrum for a drinking water sample which has typical
concentrations of most elements and a relatively large
concentration of Cd 14
6 Comparison of evaporated calibrators and liquid single-
element targets prepared by vapor filtration 19
7 Sketch of probability distributions used to define
detection limits 29
VI
-------
TABLES
Number Page
1 Elements Used for X-ray Yield Calibration 18
2 Recovery Test for Targets Prepared by Vapor Filtration
for the Elements of Primary Interest 22
3 Percent Recoveries of Trace Elements in Presence of Various
Concentrations of Calcium and Sodium 24
4 Evaluation of Analytical Precision 30
5 Detection Limits for PIXE Analysis of Drinking Water 3^
6 Trace Elements in Ultra-pure Water 42
7 Acidified Ultra-pure Water Analysis Used to Evaluate
Contamination 43
8 Evaluation of Cleaning Procedures for Vapor Filtration
Devices 44
9 Precision of Atomic Absorption Measurements of Sodium
in Drinking Water Samples 47
VI1
-------
ABBREVIATIONS AND SYMBOLS
ABBREVIATIONS
C -- centigrade
Cv -- coefficient of variation or relative standard deviation
DL -- detection limit of an element
eV -- electron volt
keV -- thousand electron volts
1 -- liter
MeV -- million electron volts
ml -- milliliter
PIXE -- proton-induced x-ray emission
ppb -- part per billion
ppm -- part per million
s -- estimate of the standard deviation of replicate observations
p,a -- microampere
U,C -- microcoulomb
M,g -- microgram
Z -- atomic number
SYMBOLS
Ac actinium
Ag silver
Al aluminum
As arsenic
At astatine
Au gold
Ba barium
Bi bismuth
Br bromine
C carbon
Ca calcium
Cd cadmium
Ce cerium
Cl chlorine
Co cobalt
Cr chromium
Cs cesium
Cu copper
Dy dysprosium
Er erbium
Eu europium
F fluorine
Fe iron
Fr francium
Ga gallium
Gd gadolinium
Ge germanium
H hydrogen
Hf hafnium
Hg mercury
Ho hoImium
I iodine
In indium
Ir iridium
K potassium
La lanthanum
Li lithium
Lu lutecium
Mn manganese
Mo molybdenum
N nitrogen
Na sodium
Nb niobium
Nd neodymium
Ni nickel
Np neptunium
0 oxygen
Os osmium
P phosphorus
Pa protactinium
Pb lead
Pd palladium
Pm promethium
Po polonium
Pr praseodymium
Pt platinum
Pu plutonium
Ra radium
Rb rubidium
Re rhenium
Rh rhodium
Ru ruthenium
S sulfur
Sb antimony
Sc scandium
Se selenium
Si silicon
Sm samarium
Sn tin
Sr strontium
Ta tantalum
Tb terbium
Tc technetium
Te tellurium
Th thorium
Ti titanium
Tl thallium
Tin thulium
W tungsten
U uranium
V vanadium
Y yttrium
Yb ytterbium
Zn zinc
Zr zirconium
viii
-------
ACKNOWLEDGMENTS
The entire staff of the Purdue University Accelerator Laboratory have
viewed this project as a challenge to their creative talents. The success
of the program has depended on the willingness of many people to continue
beyond their normal work hours to find solutions to nagging problems.
K. A. Mueller has been responsible for the direct supervision of the
target preparation and analysis staff during this project. He has also made
many creative contributions to the development of PIXE analysis for aqueous
samples. The operation and maintenance of the accelerator has been super-
vised by B. D. Michael. A. M. Sills has made valuable contributions to the
accelerator operations and target preparation. H. E. Haldman has been
responsible for the operation of the electronic systems and the on-line
computer. K. E. Melson and D. L. Coppage have exhibited an unusual degree
of scientific skill and perseverance during thousands of hours of computer
assisted analysis.
The atomic absorption measurements were conducted under the supervision
of I. E. Smiley and K. J. Yost of the Purdue Bionucleonics Department.
N. S. Ulmer, Project Officer, has played an active role in this research
project. She has understood the scientific and technical problems which
are inevitable in a new development of this type and has made many construc-
tive suggestions for improvements.
IX
-------
SECTION I
INTRODUCTION
GENERAL PRINCIPLES OF PIXE ANALYSIS
The potential for using energetic protons from a nuclear accelerator to
perform quantitative, multielemental analysis was first discussed by
Johansson, Akselsson,and Johansson (1). An excellent review of PIXE
analysis has been written by Johansson and Johansson (2). When a sample is
bombarded with protons, characteristic x rays emitted from each element
heavier than magnesium can be used in uniquely identifying the element in
the sample. The x rays are detected by an energy sensitive semiconductor
device and the data are processed by an on-line computer. Proton excitation
provides much better sensitivity than electron excitation because the yield
of characteristic x rays compared to background radiation is much higher.
Proton excitation has an advantage over x-ray excitation since good sensi-
tivity can be obtained for a much larger group of elements in a single
measurement.
A thorough discussion of the probability for characteristic x-ray
production and the various types of background radiation that are produced
has been given by Folkmann, Gaarde, Huss, and Kemp (3). The most important
factors will be summarized here to show why PIXE provides excellent sensi-
tivity for a very broad range of elements. The probability for characteris-
tic x-ray production is very large in the energy region where the background
radiation is maximum. This relationship is illustrated in Figures 1 and 2.
Figure 1 shows the number of K and L x rays detected in our experimental
arrangement as a function of the atomic number (Z) of the target atom.
(The data were obtained using an absorber in front of the x-ray detector.
Absorber selection is discussed in section 6.) Elements lighter than cerium
(Z = 58) are detected by K x rays while L x rays are used for the heavier
elements. Figure 2 shows a simplified representation for the intensity of
background radiation as a function of x-ray energy. Energetic protons
passing through the target knock out showers of electrons. When these
electrons stop in the target, they produce a continuous x-ray spectrum
called Bremsstrahlung or braking radiation. The x-ray detector is also
slightly sensitive to nuclear \ rays which produce a small, continuous
background over the total energy range of interest. The positions at which
the characteristic x rays would appear in the spectrum are shown for a few
typical elements by the vertical arrows. For example, chromium x rays occur
near the peak of the Bremsstrahlung, but the yield for chromium x rays is
very large as shown in Figure 1. On the other hand, the x-ray yield for
cadmium is small, but the x rays occur at much higher energy than the
-------
X-RAY YIELD /yuC/>g/ cm2
5000 -
2000 -
1000
500 -
200
100
4 MeV PROTONS
10 mm DIAMETER DETECTOR
6.5cm FROM TARGET
0.71 mm MYLAR FILTER
20
30
40 50 60 70
Z, ATOMIC NUMBER
Figure 1. X-ray yield for PIXE analysis,
80
90
-------
c
N
O
O
u_
o
cr
UJ
GO
2
z
o
o
o
o
o
o
o
o
o
o
o
o
o
\
o
_o
e>
cr
LU
z
UJ
TO
•H
CO
e
w
x
C
Q)
0)
CO
•o
o
VJ
Ml
-------
Bremsstrahlung where there is only a very small y~ray background. The
sensitivity for elements in the region of selenium and for heavy metals
like lead is good because the x-ray yield is large and the background small.
It is useful to compare PIXE to optical emission techniques which are
widely used for elemental analysis. The silicon-lithium semiconductor x-ray
detector has many advantages. First, the response of the detector is
essentially constant for months or even years. This is generally not true
for photomultipliers used to detect light. Furthermore, PIXE uses only one
detector while multielemental techniques employing optical emission have a
separate detector for each element. In optical systems, most of the
available light is not used, and the detectors must be preset for selected
lines. This is unnecessary with PIXE since all x rays which reach the
detector are stored in the x-ray spectrum. Unexpected elements can be
identified and the characteristic x rays and background can be measured
simultaneously. The intensity of the proton beam is measured directly during
PIXE analysis; hence the excitation probability is a known quantity.
Consequently, the background is measured directly, the detector is very
stable, and the analytical calibration of the system is usually constant for
long periods. Frequent comparisons to standards and reference samples,
which are common for optical systems, are not needed in PIXE analysis. The
characteristics of PIXE analysis will be discussed in more detail in
sections 3, 4, 5, and 6.
TECHNIQUES DEVELOPED FOR PIXE ANALYSIS OF AQUEOUS SAMPLES
Although PIXE analysis has been used for a wide variety of samples (2),
there were three areas that required extensive development before the
technique could be used with aqueous samples:
1. A method was needed for depositing the elements from the aqueous
sample on a thin backing suitable for proton bombardment. The
procedure must recover extremely small quantities of material, and
it must minimize the danger of introducing outside contamination.
2. A large quantity of data was needed to obtain extremely low detection
limits for a wide range of elements. Since the time required for
bombardment was the major factor in the cost of analysis, a high data
rate was very desirable. We developed an "on-demand" pulsed proton
beam and an advanced x-ray "pile-up" rejector that permits much higher
x-ray counting rates than are generally available for PIXE analysis.
3. A more complicated computer program than that usually used for PIXE
analysis was needed to provide the low detection limits and broad
elemental range required. The new computer program is essentially
automatic and requires minimum operator intervention; however, the
operator is able to check the analysis quickly and make corrections
if they are required.
A more complete description of the general principles of PIXE analysis
is given in Section 3, and the procedures used for the analysis of drinking
water samples under this contract are discussed in Section 6.
-------
SECTION 2
CONCLUSIONS AND RECOMMENDATIONS
The project has demonstrated that PIXE is a powerful tool for the
raultielemental analysis of drinking water samples. PIXE analysis provides
excellent detection limits for a wide range of trace elements at relatively
modest cost ($42.60 per sample in this project). PIXE is not the most
sensitive method available nor the least expensive method if information is
required for only a few elements.
It would be expensive to install PIXE analysis systems at laboratories
that do not currently have a nuclear accelerator and a relatively large
on-line computer. However, there are a large number of university and
government laboratories which have most of the facilities needed for PIXE
analysis. With some improvements in the Purdue University Accelerator
Laboratory we could provide the quality of analysis demonstrated in this
project for 20,000 samples per year at a cost-per-sample reduction of
approximately 50 percent.
It would be easy to add the light elements, lithium, boron, fluorine,
sodium, and magnesium, to the analysis by observing nuclear y rays with a
germanium-lithium detector at the same time that the x rays are detected.
The sensitivity of PIXE analysis could be improved and the cost reduced by
depositing larger quantities of the trace elements on the targets. The
volume of water that can be processed by vapor filtration is limited by the
quantity of calcium and sodium in the sample, so it would be desirable to
study methods of removing at least part of these elements from the sample
before vapor filtration.
In summary, it has been demonstrated that a nuclear accelerator can be
used for large scale, multielemental analysis of drinking water. The full
potential of the Purdue University accelerator for analyzing large numbers
of samples was not utilized in the present work. The many suitable acceler-
ators in the United States should be considered as a valuable national
resource for large projects involving multielemental analysis in the areas
of health effects, environmental control, mineral resource evaluation, and
energy utilization.
-------
SECTION 3
GENERAL DESCRIPTION OF TECHNIQUES AND PROCEDURES
SAMPLE PREPARATION
*
The process described here was developed by the Purdue University
Accelerator Laboratory staff prior to the award of U. S. Environmental
Protection Agency (USEPA) Contract No. 68-03-2178. The process was included
in the Purdue proposal submitted to the USEPA for "The Multielemental
Analysis of Drinking Water Using Proton Induced X-ray Emission" and dated
October 1, 1974.
A great improvement in sensitivity for trace element analysis of aqueous
samples by PIXE can be obtained by eliminating the water before bombardment.
In order to reach sensitivity levels of one part per billion (ppb) or better,
approximately 30 ml of the sample must be reduced to dryness without signifi-
cant addition or loss of trace elements. There are several problem areas
in this process. First, the effects of kinetics must be minimized. For
example, in a simple boiling process large amounts of material can be lost
when drops of solution leave the surface of the sample. A similar effect
occurs in freeze-drying processes as water vapor leaves the ice surface.
Second, a closed system must be used to minimize contamination from outside
sources. Third, one must quantitatively transfer the solids from the concen-
tration unit to a suitable support. The apparatus must be easy to clean to
prevent contamination from other samples. Finally, the trace elements in
the sample may exist in various chemical forms, and the preparation technique
must be independent of the chemistry of the sample.
We have developed a simple and novel solution to these problems (4).
Consider a container whose bottom surface is a membrane permeable to water
vapor but not readily permeable to water. The bottom surface of the membrane
is exposed to a vacuum system. Water which approaches the vacuum side of
the membrane immediately changes to the vapor phase, effectively depositing
all solids on the membrane. The water vapor passing through the membrane is
pumped away by the vacuum system. The membrane itself is then used as a
target backing in PIXE analysis.
This concept, which we call "vapor filtration", has many attractive
features. The water is not agitated, so there is no danger of water droplets
spattering with subsequent loss of material. The kinetic motion of the water
*
Purdue Research Foundation, Lafayette, Indiana, has filed a patent applica-
tion (652,381) for the vapor filtration process.
-------
moves the trace elements toward the desired deposit location. The system can
be closed at the top so that airborne contaminants cannot enter. There is no
final transfer of solids involved, so the process can be quantitative. Since
no chemical processing is required, the process is not affected by the
chemical composition of the dissolved or suspended solids.
There are five basic requirements for a practical vapor filtration
device. 1) the membrane must pass water vapor, be reasonably strong, and
contain minimal impurities. The device must provide 2) a good liquid seal
between the membrane and the water container, 3) a good vacuum seal between
the membrane and the vacuum system, and 4) a porous support so that the wet
membrane will not break under atmospheric pressure. 5) The apparatus should
be easy to clean.
Our solution for these requirements is shown in Figure 3. The water
container is a cylindrical tube glued into a collar. The tube, which has an
inside diameter of 1.6 cm, defines the area on the membrane where material
from the sample will be deposited. The base has a recess to hold a porous
polyethylene disk, a n-butyl 0-ring seal to a vacuum port, and a lip on the
outside top for easy alignment of the circular membrane and the water
container collar. (The 15-mm, porous polyethylene column support disc,
No. 275-264-0015, was purchased from Fisher and Porter Co., 1531 County Line
Rd., Warminster, PA.) These pieces are made of General Electric Lexan
polycarbonate plastic which is resistant to nitric acid and easy to clean.
The collar and base are held together by a spring loaded aluminum clamp.
The bottom of the collar and the top of the base must be carefully machined
and polished to prevent water leaks on top of the membrane and vacuum leaks
on the bottom. All parts of the vapor filtration apparatus were made in
our laboratory machine shop. Medicine cups are used for the top cover.
(General Medical Company, No. 721.)
A wide variety of materials were investigated to find a suitable
membrane. The best material that is readily available commercially is DuPont
PD 215 cellophane (2.3 X 10~2 mm thick). After the sample is dried, the
membrane is mounted in a 35-mm, plastic photographic slide frame for bombard-
ment. (The frames are manufactured by GEPE Garanti.)
The vacuum system has a number of requirements. The pressure on the
vacuum side of the membrane must be kept below 4 Torr, the vapor pressure of
ice at 0°C. Below this pressure it is impossible for water to exist in the
liquid or solid phase. The disk that supports the membrane is sufficiently
porous for the pressure at the membrane surface to be approximately 1 Torr.
A mechanical pump will easily maintain this pressure, but it will not
continually pump water vapor. Our pumping system consists of a large
mechanical pump (Kinney Model KDH-130) whose outlet pressure is maintained
at approximately 100 Torr by a water-ring pump (Kinney Model K-32-3). The
water vapor is continuously pumped from the warm pump oil by the ring pump
which is very efficient in the 100-Torr range.
-------
COLLAR
ALIGNMENT
LIP
•COVER
WATER
TUBE
•CLAMP
.SPRING
THIN-FILM
VAPOR-FILTER
POROUS /[
SUPPORT
BASE
0-RING SEAL
TO
VACUUM
PUMP
Figure 3. Vapor filtration apparatus.
-------
Since approximately three days are required to pass each 30-ml water
sample through the vapor filter, the vacuum manifold was designed to process
120 samples at the same time. The exact time required for drying each sample
depends on the total dissolved solids in the sample. Prolonged pumping after
the sample is dry weakens the membrane, so each pumping port has a separate
vacuum shut-off valve. A less expensive pumping system could have been made
with a small mechanical pump and a cold trap to condense water vapor.
However, a large capacity pump is desirable so that good vacuum can be
maintained even if a few membranes break during the drying cycle. The
medicine cup which covers the top of the water container also aids in
maintaining good vacuum. If a membrane breaks, the medicine cup forms a
partial seal and limits the flow of air into the vacuum system.
Approximately five percent of the targets are not usable because of
either a water leak, a membrane break during pumping, or a membrane tear
during removal from the system. The membranes do not appear to be adversely
affected by the normal nitric acid used to preserve samples (1.5 ml concen-
trated nitric acid per liter of sample), i.e., the breakage rate for acid
preserved samples is essentially the same as that for pure water. However,
some standard solutions prepared with other acids are difficult to process.
For example, samples that contain aqua regia are very likely to damage the
membrane. Also we have found that a small sample with a large concentration
of acid is more likely to damage the membrane than a larger sample with the
same quantity of acid.
The uniformity of the target depends on the type and quantity of
material in the sample. There is a tendency for the deposit to be heavier
around the edges of the target circle due to the meniscus at the end of the
drying process. Target nonuniformity is not a major limitation as long as
the proton bombardment is essentially uniform across the surface of the
target.
A summary of the tests that have been performed to evaluate the quality
of targets prepared by vapor filtration is given in Section 4.
BOMBARDMENT CHAMBER AND PROTON BEAM CONTROLS
Energetic protons are obtained from a High Voltage Engineering Model
FN Tandem Van de Graaff Accelerator. The target is normally bombarded with
a one-microamp (|j,a) beam of 4-MeV protons (approximately 6.3 x W protons
per second). The beam passes through a 2.5 x 10 -mm aluminum foil located
3 meters in front of the target to diffuse the beam and ensure that the
proton bombardment is uniform over the target area. The beam is restricted
to a 1.9-cm diameter circle by a series of tantalum collimators to ensure
that the protons can strike only the sample and the thin cellophane support
film. The beam spot is 3 mm larger than the target and 6 mm smaller than
the target frame. Lead absorbers are used to prevent tantalum x rays
produced in the collimators from reaching the x-ray detector.
-------
A sketch of the target chamber is shown in Figure 4. Eighty targets
are held in the'circular slide tray of a modified 35-mm slide projector
(Keystone, Model 1100). The target to be bombarded is dropped into the
proton beam by the automatic mechanism of the projector. The protons pass
through the target and are stopped on a carbon block located one meter behind
the target so that Bremsstrahlung from the stopper cannot reach the x-ray
detector. A 10-cm diameter plastic tube (not shown) extends from the back
of the target down the exit beam pipe. This tube prevents scattered protons
from striking metal components of the chamber and producing characteristic
x-ray background.
X rays from the target go through a 0.025-mm Mylar window in the
bombardment chamber and through a 0.025-mm beryllium window in the detector
housing. These windows absorb most x rays emitted from elements lighter than
aluminum. Several filters that can be changed from the computer are included
to absorb high intensity x rays which would otherwise interfere with the
detection of trace elements. The types of filters used for the measurements
will be described in Section 6.
The detector is located 6.5 cm from the target. The axis of the
circular detector is perpendicular to the surface of the target. In this
geometry, the difference in the probability of detecting an x ray from the
center and from the edge of the target is only 1.5 percent, so nonuniformity
of the target cannot have a significant effect on the x-ray detection
probability. In many other PIXE analysis systems the detector is located
perpendicular to the beam axis at a 45° angle to the target. In that
geometry the change in detection probability across the target would be
38 percent for the same target size and detector distance.
DETECTOR AND ELECTRONICS
The silicon-lithium detector (Kevex Model 3010 AR) is 1 cm in diameter
and 3 mm thick. A 6-mm diameter mask is used in front of the detector to
prevent x rays from interacting near the detector edges. This precaution is
necessary to minimize the possibility that part of the x-ray energy will be
lost from the detector by escaping electrons. The detector provides 170-eV
energy resolution for 5.9-keV x rays. The system can respond to thousands
of x rays per second, but even a few scattered protons per second will jam
the detector-electronics system because the protons have approximately 1000
times as much energy as the average x ray. An absorber of at least
9 x 10 mm of Mylar is needed in the filter position to prevent scattered
protons from reaching the x-ray detector.
Normally when an x ray is absorbed in the detector, all of the x-ray
energy is converted into an electrical pulse. However, it is also possible
for a silicon x ray, created in the detection process, to escape from the
detector. These "silicon escape" events create extraneous peaks in the
x-ray spectrum. For example, if a zinc K x ray strikes the detector and a
silicon K x ray escapes, the event will appear to have almost exactly the
same energy as the cobalt K x ray. It is possible for electrons released
in the detection process to leave the surface of the detector. These events
10
-------
to
4-1
CO
S-i
CO
C
O
•H
4-1
O
CLJ
e
u
a
T3
00
c
•1-1
S-i
O
4-1
CO
>*
01
3
bO
•r-l
11
-------
produce a small tail on the low energy side of peaks in the x-ray spectrum.
The computer must be able to correct for these two features of the detector
response in order to get accurate analytical results.
The voltage pulse from the detector preamplifier is amplified and
shaped by a Kevex Model 4532P Pulse Processor. Electronic noise from the
detector is reduced in the 4532P processor by an active filter which has a
4-microsecond time constant. Approximately 50 microseconds are required to
process each x-ray pulse. If a second x ray is detected before the energy
of the first x ray has been stored in the computer processing circuit, both
x rays must be discarded. If the second x ray is detected in the latter part
of the first x-ray pulse, only the second x ray is lost. This effect imposes
a serious limitation on conventional data taking systems. For example, if
ICr" x rays were hitting the detector per second, approximately 60 percent
of the x rays would have to be discarded.
The problem of pile-up is greatly reduced in our system by switching the
proton beam off the target approximately 1.5 microseconds after the first
x ray is detected. As soon as the pulse processor has finished with an x-ray
event, the beam is automatically returned to the target to produce a new
x ray. With this "on-demand" beam pulsing system (5) there is typically
only a 3 percent loss of data at an average rate of 10^ x rays per second.
Even with the beam switching system there is still a small probability
that a second x ray will be produced before the beam is removed from the
target. When two x-ray pulses pile up in the noise filter, a single pulse
is produced corresponding in energy to the summed energies of the two x rays.
We have developed a circuit (6) that recognizes and rejects pile-up events
unless the second x ray is emitted within Tr seconds of the first x ray.
The resolving time of the pile-up detector (Tr) is approximately 0.1 micro-
second for 2-keV x rays. Tr decreases as the x-ray energy increases. Tr is
approximately 0.05 microsecond when the x-ray energy is greater than 6 keV.
The undetected pile-up events are rare so they usually do not present a
significant limit for the analysis.
The task of separating events that come from two x rays of similar
energy is greatly simplified if the energy calibration of the electronic
system is stable. The energy calibration is always a function of the
counting rate in the detector. The Kevex Model 4532P pulse processor used
in our system has correction circuits that limit the calibration variation
to less than 0.03 percent for counting rates up to 10 x rays per second.
The computer uses a large x-ray peak in the spectrum to make a fine adjust-
ment of the energy calibration so that the energy uncertainty is less than
0.01 percent.
The proton beam is collimated so that all protons entering the target
chamber must pass through the target. The total charge that is accumulated
in the chamber during a measurement is recorded by a digital current inte-
grator (Ortec Model 439) and stored in the computer. The quantity of charge
on each proton is well known, so the computer converts the total charge into
the number of protons passing through the sample. The computer also corrects
for the number of events lost due to pile-up.
12
-------
COMPUTER PROCESSING
On-line data accumulation and processing are performed by a Digital
Equipment Corporation PDP-15/40 computer. The soft-ware currently used for
FIXE analysis has required three years of development by Professor
Frank A. Rickey. The analysis process is monitored and the computer is
controlled with an interactive video display.
The first step in the analysis process is to subtract background
radiation. Many experimental background spectra, each similar to the sketch
shown in Figure 2, are stored on a magnetic disk. The operator selects a
background spectrum which represents the host material of the target being
analyzed. The computer adjusts the background to match the target including
the effect of the peak-tails described above. The background is then
displayed on the video monitor as an overlay on the experimental data.
Usually, the computer generated background accurately describes the actual
background from the sample, but the operator can instruct the computer to
refine the background by contacting the display with a "light pen".
The computer also has stored experimental data that describe the shape
and the position of the characteristic x-ray peaks in the spectrum. Even
though the imperfections in the detector response are small, they can be
important when there are intense x rays in the spectrum. The computer
locates large peaks and identifies their tails and the peaks that are
produced by silicon-escape x rays. Very large peaks can also produce
simultaneous pile-up peaks. The computer performs a least-squares fit of
the data to determine the number of x rays at each energy of interest. The
stability of the energy calibration, the advanced knowledge of the shape of
the characteristic x-ray peaks, and the careful treatment of imperfections
and extra peaks are all essential factors for accurate analytical results.
The result of the least-squares fit is displayed on the video monitor
as an overlay on the data. Most fitting errors are evident from a visual
examination of the display, so the operator can instruct the computer with
the light pen to refine the fit to the data. The most important part of a
typical spectrum obtained for a water sample is shown in Figure 5. The
vertical axis gives the number of x rays on a log-^Q scale and the horizontal
axis gives the x-ray energy. Upon request the computer will expand any part
of the spectrum so that the operator can examine details of the fit to the
data.
After the visual examination, the computer makes additional checks on
the analysis. The relative intensities of the major x rays emitted by each
element in the unknown sample are compared to standard values which were
determined from single element targets. Inconsistencies are reported to the
operator. The computer also warns the operator of possible sources of error
caused by two x rays with approximately the same energy.
13
-------
J-Ll I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
o
ro
in
CVJ
o tr
c\j
LJ
or
I
y2 x
to
10
irl
o
in
•
in
ro
o
ro
m
o
, O
4-1
C3
co O
O 4J
•H C
fi CU
is o
c
CU O
•-I O
ex
6 a)
cd t>o
co Vj
cd
cu
Cd r-4
M -H
C -U
•
d a)
•H S-)
S-i
4-4 CO
4-J
i g
^ E
4J S
a <-t
cu cu
a,
CO 4J
CO
>> O
2 B
I <4-l
X! O
to
•H
14
-------
Usually it is not necessary for the operator to intervene in the
analysis; however, the following example will illustrate the steps that the
operator can take should a question arise. Suppose that the computer had
identified a large bromine and a small uranium component in the sample. The
bromine Kg 2 intensity is reported as too small compared to the other bromine
x rays. In addition, the uranium L^ x ray is reported as present, but the
other uranium x rays are not clearly identified. This information indicates
that the analysis is not correct because the bromine K, 2 (13465 eV) x rays
and the uranium L^ (13596 eV) x rays have similar energy. The operator
would instruct the computer to delete uranium and refit those portions of the
spectrum that contain any of the appropriate x rays. The computer now
reports an intensity for bromine Kg 2 consistent with the other bromine x rays
and a good fit to the data in all regions where uranium x rays would appear.
The operator can be sure that the original fit had falsely assigned part of
the bromine K_ 2 intensity to the uranium Li x ray.
When the operator has approved the analysis of a sample, the computer
uses the observed number of x rays and calculates the concentration of each
element in the sample. The number of protons that have passed through the
sample is recorded as units of charge, microcoulombs (JJ.C), by the beam
current integrator (Ortec Model 439). In a continuous beam system, a large
correction would be required for the number of protons which passed through
the target while the pulse processing system was busy. This correction is
not necessary with our pulsed-beam system. A small correction (typically 3
percent) is made for the quantity of charge that was lost due to pile-up that
occurred while the beam was being deflected from the target. Data for this
correction is recorded automatically by an electronic sealer which counts the
number of events that are rejected by the pile-up detector circuit. The
x-ray yield per \j,C of protons for a given filter and proton energy are stored
in the computer. (The procedure for obtaining this calibration is given in
the following subsection.) The concentration, C, of each element in an
individual sample is determined from the following equation:
C = N/YV
where
N is the number of characteristic x rays/|j,C
Y is the x-ray yield (number of characteristic x rays/(j,C/|j,g element)
V is the volume (liters) of sample analyzed
The final analytical results for the sample, including confidence
limits, are printed on a single sheet of paper and stored on a magnetic tape
for subsequent interpretation. The original data, the computer generated
background, and the fit to the data are stored on a separate magnetic tape
so the entire analysis can be checked should a question arise at a later
time.
15
-------
SYSTEM CALIBRATION
Calibration stability is a very important advantage of PIXE analysis.
We will first summarize the most important points and then describe the
details of the calibration.
1. It is not necessary to have a calibration standard for each element
that is included in the analysis.
2. The source of excitation, i.e., the proton beam current, is measured
directly.
3. In contrast to multielemental techniques which use optical emission,
only one detector is used for PIXE analysis.
4. The detector is a semiconductor device whose characteristics usually
are constant for several years.
The x-ray yield (Yx) is defined as the number of x rays which are
observed from a particular element during the bombardment of the target:
Y = N. N_ P_, P^
x APED
where :
N is the number of atoms of the element on the target.
A
Np is the number of protons that pass through the target.
P£ is the probability of emitting an x ray of energy EX.
PTJ is the probability of detecting an x ray of energy E .
N. , which is used to determine the weight of an element on the target,
can be calculated from YX when the other factors are known. Np is measured
for each sample so it is a known quantity in the analysis.
Pg is a constant of nature. Statement (1) above is true because the
change in Pg from element to element follows a smooth curve. The curve can
be accurately defined by measuring YX for selected elements. The calibration
for elements which are not included as standards can be obtained from the
smooth curve. P has three components:
PD
where:
P is the probability that an x ray with energy Ex will pass through the
absorber in front of the x-ray detector.
fL is the solid angle of the x-ray detector.
e_ is the probability that an x ray with energy EX will deposit all of its
energy in the x-ray detector.
16
-------
For the type of absorbers used in this analysis, P^ is a smooth function
of EX. However, P^ can change during the course of the experiment if the
surface of the absorber becomes coated with small particles and oil from the
vacuum pumps. The factor, Q^, is constant because the detector is rigidly
mounted, and e^ is constant because it depends only on the volume of the
x-ray detector.
2
Thin films with approximately 100 jj,g or less material per cm were used
to measure the x-ray yield curves so that it would not be necessary to
correct for slowing down of the protons or x-ray absorption in the calibra-
tion material. Two types of calibration targets were used to provide a
consistency check on calibrator preparation. Thirty-three calibration
targets were obtained from Micromatter, Inc., 197 34th Avenue East, Seattle,
WA, in the form of single elements or stable compounds vacuum-evaporated on
Mylar backings. These targets are listed in column 3. of Table 1. Thirty
calibrators were made in our laboratory from atomic absorption liquid
standards. One ml of solution containing 100 ^g of the calibration element
was processed by vapor filtration to make each calibrator. These standards
are listed in column 4 of Table 1. (The atomic absorption liquid solutions
are more completely described in Table 2 in Section 4.)
The targets cover the range of elements from aluminum (Z = 13) to
thorium (Z = 90). Forty-five elements are included in the calibration.
Eighteen elements are present in both types of targets. Most elements in the
atomic number range from 13 to 32 were included because the x-ray absorbers
produce a relatively large curvature in the plot of the x-ray yield in this
region.
The targets which are used for a particular calibration depend on the
absorber which is present in front of the x-ray detector. (Absorber selec-
tion is discussed in Section 6.) For example, when an absorber is used to
reduce the intensity of x rays from calcium, there is no point in analyzing
x rays from elements lighter than calcium. These x rays would not pass
through the absorber. Calibrators for calcium and heavier elements are used.
The results of the x-ray yield measurements were fit with an algebraic
function to obtain the smooth curve shown in Figure 1. (This curve is for
Ki and L^ x rays. Similar curves are obtained for other less intense
x rays for use in the consistency check of the x-ray intensity measurements
described previously.) The results from individual calibrators are not shown
in Figure 1 because most of the points would be so close to the line that it
would be very difficult to evaluate the consistency of the calibration.
Instead, Figure 6 shows in a much clearer form the data used to plot Figure 1.
Each x-ray yield has been normalized by the corresponding value obtained from
the best-fit function. The certified values of 10 of the calibrators were 8
or 9 percent lower than the values from the best-fit function. Two of the
evaporated calibrators were obviously not correct. Since 46 other calibra-
tors were very consistent, the 12 low calibrators were dropped from the best
fit determination. The standard deviation of the 46 calibrators from the
best fit function was 3 percent.
17
-------
TABLE 1. ELEMENTS USED FOR X-RAY YIELD CALIBRATION
E lement
Aluminum
Silicon
Sulfur
Chlorine
Potassium
Calcium
Scandium
Titanium
Vanadium
Chromium
Manganese
Iron
Cobalt
Nickel
Copper
Zinc
Gallium
Germanium
Selenium
Rubidium
Strontium
Yttrium
Niobium
Molybdenum
Palladium
Silver
Cadmium
Tin
Antimony
Tellurium
Iodine
Cesium
Barium
Lanthanum
Neodymium
Gadolinium
Erbium
Tantalum
Tungsten
Platinum
Gold
Thallium
Lead
Bismuth
Thorium
Atomic
number
13
14
16
17
19
20
21
22
23
24
25
26
27
28
29
30
31
32
34
37
38
39
41
42
46
47
48
50
51
52
53
55
56
57
60
64
68
73
74
78
79
81
82
83
90
Material used for
evaporated film
calibrators
Al
SiO
CuS
NaCl
KC1
CaFs
Sc
Ti
Cr
Fe
Ni
Cu
Zn
GaP
Ge
Se
SrF2
Y
Nb203
Pd
Cd
Sn
Te
Csl
Csl
LaF3
NdF3
Gd
Er
Ta
Au
Pb
ThF4
Material used for
vapor filtration
calibrators
KC1
CaCOs
ScCls
Ti(in HC1)
VgCfe
KgCrgO,
Mn(in HN03)
FeCl3
Co (in HN03)
Ni(in HN03)
CuO
Zn(in HN03)
Ga(N03)3
(NH4)2GeF6
RbCl
Sr(N03)2
Y(N03)3
Mo (in HN03 + HC1)
PdCl2
AgN03
Cd(in HN03)
Sn(in HC1)
SbCl3
Na2Te04
BaCl2
Na2W04
H2PtCl6
T13S04
Pb(in HN03)
Bi(in HN03)
18
-------
x
X C
X
X
X
°X 0
0 <
0
*
*i
&
i
- o c
o:
0
«5
o:
GO
o
z
o
H
o:
5
i
? v O
f X Q_
I ^T
1 ^A
1 >
O
1
1
1
1
1
1
1
1 1 1 1 1
D 00 N CD ID
CO
I .
CO
_J
o
_ J
O P>,
QJ 0) ^
)-( 01
H-l CO X
O O. J-1
CU
C M ^
o a.,0
CO
•>-l CO T3
M 4-1 CU
CO 0) T3
PL, bO -i-l
0 P >
O CO -H
VD
W)
•r-l
19
-------
It is essential to understand the basic difference in the calibrations
for PIXE analysis and optical emission analysis. If one uses an incorrect
standard for optical emission analysis, the results for that element is
affected directly. The calibration of a particular element in PIXE analysis,
however, does not depend directly on the reliability of the target which
contains that element. Each point on a calibration curve is determined by
many elements.
The expressions given above for the x-ray yield reveals another
important point. The calibration is reliable as long as the proton charge-
measuring system is working correctly and the absorption probability is
constant. These factors can be checked easily by using two single-element
standard targets. A high energy x ray that is not effected by the absorber
is used to check the accuracy of the charge-measuring system. A low energy
x ray is used to ensure that the absorption probability is unchanged. We
require that these tests agree with the calibration curve to within 3 percent.
A variety of targets are alternated in these tests to minimize the danger
that a mistake will be made if a particular target deteriorates with use.
20
-------
SECTION 4
INITIAL EVALUATION OF PIXE ANALYSIS FOR AQUEOUS SAMPLES
Before the PIXE technique could be used for the analysis of drinking
water, it was necessary to develop and test a method for preparing suitable
targets from aqueous samples. Techniques such as freeze drying and various
forms of evaporation were investigated and judged unsatisfactory because of
the danger of trace material loss and contamination. The concept of vapor
filtration that developed from these studies seemed very promising, but
extensive development was required to refine the process. The tests
described in this section were completed before the proposal was submitted
for this contract.
It was demonstrated that water vapor would go through cellophane and
other films at a reasonable rate. Even though cellophane is strong when it
is dry, good mechanical support is required when it is wet.
The next step was to prove that the vapor filtration process was
quantitative. The conditions and results of recovery measurements for the
35 elements of primary interest are summarized in Table 2. Most of the test
solutions were prepared from commercial atomic absorption standards purchased
from Fisher Scientific Co. and Alfa Division of Ventron Corp., as indicated
in column 2 of the table. It was not possible to mix all of the standards
in a single solution because of the various properties of the chemicals
present. The 35 elements were divided into four groups as shown in column 3.
A solution of each group of standards was prepared and diluted with ultra-
pure water. The final elemental concentrations are shown in column 4. Nine
aliquots, each 30 ml, were obtained from each of the four original standard
mixtures. Targets were prepared and analyzed over a six-week period. Each
recovery value in column 5 is the average of nine analyses. The standard
deviation in column 6 should be interpreted in the usual way: the deviation
of approximately two thirds of the measurements from the average was equal
to or less than the specified value. The standard solutions were not held
for longer than three months, so no attempt was made to evaluate their
long term stability. During the period of the contract, analyses of
fresh solutions of the same four groups of elements yielded results similar
to those in Table 2.
The vapor filtration process is not sensitive to the chemical composi-
tion of the sample (except for volatile components and excessive acid in the
sample). Therefore, the tests summarized above provide strong evidence that
the vapor filtration process is an excellent method for preparing targets
from aqueous solutions. Nevertheless, a number of additional tests were
performed to evaluate problems that could occur due to large concentrations
21
-------
TABLE 2. . RECOVERY TEST FOR TARGETS PREPARED BY VAPOR FILTRATION
FOR THE ELEMENTS OF PRIMARY INTEREST
Element
Potassium
Calcium
Scandium
Titanium
Vanadium
Chromium
Manganese
Iron
Cobalt
Nickel
Copper
Zinc
Gallium
Germanium
Bromine
Rubidium
Strontium
Yttrium
Molybdenum
Palladium
Silver
Cadmium
Indium
Tin
Antimony
Tellurium
Iodine
Cesium
Barium
Lanthanum
Tungsten
Platinum
Thallium
Lead
Bismuth
*
Source
F
F
A
F
F
F
F
F
F
F
F
F
A
A
P
A
A
A
F
A
F
F
A
F
F
A
P
A
F
A
A
A
A
F
F
Group
1
3
1
2
2
4
3
2
3
2
3
3
2
4
4
1
2
3
2
1
3
3
1
1
2
4
4
1
2
1
4
2
1
3
3
Concentration
p/g/l
1000
1000
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
30
30
30
10
10
10
10
10
Average
percent
recovery*
96
97
101
96
99
94
102
103
94
99
91
99
93
91
89
103
93
100
99
98
92
97
95
93
102
92
56
97
96
93
92
98
101
99
88
Standard
deviationt
6
5
8
5
4
7
5
6
8
3
7
5
6
7
14
5
4
3
5
7
8
5
6
6
7
9
16
4
8
8
5
6
8
4
9
*
The standard solutions were obtained from Fisher Scientific Co. (F) and
Alfa Division of Ventron Corp. (A). The test solutions for I and Br
labeled (P) were prepared at Purdue University.
Refer to the text for a description of the measurements and interpretation.
t
22
-------
of sodium and calcium in drinking water samples. The four groups of standard
solutions were added to drinking water samples which were collected in the
Lafayette area. The recovery factors observed in these tests were similar
to that shown in Table 2. Of course, the solids observed in the local water
samples did not cover the elemental range that would occur in water samples
from the proposed study, so the next step was to prepare test solutions which
would simulate drinking water samples with large variations in calcium and
sodium concentrations.
One obvious concern was the possibility of deposits being formed on the
inner walls of the cylindrical tube during the vapor filtration process.
Replicate 30-ml aliquots of samples containing 40, 80, 120, and 200 ppm of
calcium were processed and analyzed. The average recoveries were 95, 101,
97 and 96 percent, respectively. Although the observed calcium recoveries
were satisfactory, there was still the possibility that even a small calcium
or sodium deposit might somehow trap a larger fraction of a trace component.
Solutions were prepared that contained from 0 to 200 ppm calcium or
sodium and ppb concentrations of manganese, cobalt, nickel, copper, zinc,
yttrium, silver, cadmium, lead, and bismuth. The results of the analyses of
these solutions are presented in Table 3. The slightly lower recoveries for
manganese, cobalt, copper, and nickel are probably the result of x-ray
absorption rather than loss in the vapor filtration device. No significant
loss can be attributed to the presence of sodium. Since the required
accuracy of the analysis is 100 + 10 percent, and very few drinking water
samples contain 200 ppm concentrations of calcium these tests demonstrate
that calcium and sodium deposits do not present a significant limitation on
the analysis of these 10 elements.
While investigating the possible effects of calcium and sodium on the
other 25 trace elements, solubility problems occurred when calcium concen-
trations of 80 ppm or greater were added to the solutions of trace elements.
Typically only 80 percent of each element, including calcium, was recoverable.
To determine whether the elements were retained on the wall of the cylin-
drical tube, a 30-ml aliquot of ultra-pure water was processed through each
tube before it was cleaned. No significant concentrations of calcium or
trace elements were observed in the analyses of these "rinse" samples. The
missing material probably precipitated from solution during the process of
combining the component solutions.
X-ray absorption is not expected to be significant for the analysis
of elements heavier than calcium. A simple consideration of solubility
suggests that the major components would precipitate first. The trace
components would be located near the surface of the deposit, so x-ray
absorption would be small. The data in Table 3 indicate this is the case.
For example, if all the manganese in a 30-ml sample with 200 ppm of calcium
is deposited under the calcium, 70 percent of the manganese K x rays would
be absorbed. Since approximately 90 percent of the expected manganese was
observed, it is evident that most of it was deposited on top of the calcium.
The x-ray absorption probability of sodium is much smaller than for calcium,
so absorption effects are expected to be smaller for sodium than for calcium.
This trend is evident in Table 3.
23
-------
TABLE 3. PERCENT RECOVERIES OF TRACE ELEMENTS IN PRESENCE
OF VARIOUS CONCENTRATIONS OF CALCIUM AND SODIUM*
Ca(ppm)
0
40
80
120
200
50 ppb concentrations
Mn
Co
Ni
Cu
Zn
Y
Ag
Cd
Pb
Bi
97(5)
95(5)
97(5)
93(5)
97(5)
100(4)
99(5)
96(4)
98(5)
97(3)
97(5)
96(4)
97(3)
94(4)
101(3)
104(5)
97(4)
95(5)
100(4)
96(3)
91(4)
91(3)
93(3)
92(3)
94(3)
97(3)
101(5)
93(4)
94(4)
91(5)
92(4)
93(2)
94(5)
92(3)
97(3)
100(4)
99(3)
96(4)
96(3)
94(5)
89(5)
90(5)
92(5)
89(4)
96(4)
99(4)
98(4)
92(5)
96(3)
97(3)
5 ppb concentrations
Mn
Co
Ni
Cu
Zn
Y
AS
Cd
Pb
Bi
Na(ppm)
105(5)
96(5)
90(5)
92(5)
102(5)
98(4)
92(5)
93(4)
102(4)
97(3)
0
87(5)
89(5)
88(5)
91(5)
99(5)
92(5)
91(5)
93(4)
95(5)
101(4)
40
95(3)
99(5)
101(5)
94(4)
100(5)
102(5)
98(5)
99(3)
100(3)
95(3)
80
91(5)
96(3)
96(4)
91(5)
105(5)
98(5)
96(5)
94(3)
99(4)
98(4)
120
88(5)
91(4)
93(5)
90(5)
97(3)
97(4)
95(5)
96(4)
97(5)
98(4)
200
5 ppb concentrations
Mn
Co
Ni
Cu
Zn
Y
Ag
Cd
Pb
Bi
100(3)
96(5)
100(4)
92(5)
101(4)
99(3)
98(3)
96(4)
98(4)
94(4)
97(4)
99(4)
97(3)
93(5)
98(3)
96(4)
101(3)
91(5)
95(5)
97(3)
96(4)
92(5)
95(3)
89(4)
97(4)
96(3)
98(4)
98(3)
98(3)
91(3)
95(3)
94(3)
98(4)
92(5)
104(5)
102(4)
92(5)
94(3)
99(3)
95(5)
98(3)
96(4)
100(4)
93(3)
98(3)
97(4)
96(5)
95(3)
97(4)
96(3)
Each number in parenthesis represents the standard deviation of the
observations of percent recovery.
24
-------
Since the probability of x-ray absorption increases as the x-ray energy
decreases, additional tests were performed to determine the recovery of
elements lighter than manganese. A series of solutions containing 200 ppm
calcium and 50 ppb of a single element were analyzed. The recoveries of
vanadium, titanium, and chromium were 91, 89, and 95 percent, respectively.
The recovery of even lighter elements was much poorer; e.g., for sulfur it
was as low as 30 percent.
Analyses of drinking water samples containing 30 to 80 ppm calcium and
sodium concentrations were also performed to evaluate their effects on the
recovery of light elements. Targets were made from one-mi and 30-ml samples
to provide a large difference in the total material on the target. The
observed intensity of x rays from potassium and sulfur, normalized for
sample volume, was much smaller from the 30-ml samples than from the one-mi
sample. This confirmed the conclusion that x-ray absorption is a serious
problem in the analysis of light elements.
In principle it would be possible to correct for x-ray absorption if
one knew how the material was deposited on the target. Since absorption can
be large for light elements and there can be large variations in the forma-
tion of target deposits, we concluded that one-mi samples should be used for
the analysis of the following light elements: phosphorus, sulfur, chlorine,
potassium, and calcium. The advantage of the one-mi target is simple; even
in the extreme case of phosphorus totally covered by the deposit from 200 ppm
of calcium, the x-ray absorption would only be 13 percent.
The dissolved solids in the sample also influence target reliability.
Since material might fall from the backing when the target deposit was heavy,
drinking water samples and samples made from known solutions with high
concentrations of calcium and sodium were evaluated. Three aliquots of each
sample were prepared and analyzed. One target was coated with a dear
plastic film from an aerosol spray can (Borden Number 1302). One uncoated
target was handled very carefully and another uncoated target was shaken
with the sample surface down. Targets with blank cellophane backing were
also sprayed with plastic and analyzed to correct for the components in the
plastic film. The results of these tests showed that there was no detectable
loss of material as long as the total material on the target backing was less
than 5 mg. (A 30-ml sample with a 167-ppm calcium concentration would
produce a 5-mg deposit.) All of the analyses reported here and furnished in
the drinking water study were performed on uncoated targets, and there is no
evidence that the vapor filtration targets require more than normal care in
handling.
Since FIXE analysis depends on the comparison of x-ray yields from
relatively thick targets to yields from thin calibrators, it is necessary to
evaluate the effect of slowing down the incident protons in the target. The
average energy of a proton going through a target prepared from a 30-ml
sample containing 50 ppm of calcium and 50 ppm of sodium would be approxi-
mately 3.95 MeV rather than 4 MeV. The variation in x-ray yield as a
function of proton energy is not the same for all elements, but it is
typically only 30 percent per MeV. If the deposit on the target is uniform,
the change in yield as the proton slows down would be only 1.5 percent.
25
-------
Unless the targets are extremely nonunifortn, this correction would be
insignificant. The recovery evaluation data, given in Table 3, were not
corrected for a decrease in proton velocity, so it does not appear that the
nonuniformity of the targets is large enough for this effect to be signifi-
cant.
26
-------
r 2 i
n (C. - C)
v- J
i, ~
._
1/2
SECTION 5
PRECISION AND DETECTION LIMITS
DEFINITIONS
The statistical interpretation of multielemental data requires an
evaluation of the magnitude and sources of variability in the elemental
observations. Since replicate observations of-each element will be normally
distributed about some mean with a variance, a , the magnitude of the vari-
ability of the replicate determinations of each elemental concentration can
be expressed in terms of a, the standard deviation or square root of variance.
In this research the variability is actually presented in terms of s, the
estimate of the standard deviation, and C , the coefficient of variation or
relative standard deviation. These terms are derived as follows:
s =
and C = 100 s/C
where:
C. = individual observed elemental concentration Qug/1)
C = mean of the individual observed elemental concentrations (ug/1)
n = number of replicate observations of the elemental concentration
Attention is now directed to the sources of variability in the elemental
observations. Two of the most significant sources of variability are the
counting and instrumental effects. Interelemental effects may also occur and
introduce additional variabilities.
2
The counting variance, s , is a measure of the variability in the
emission of x rays from a sample. Since the emission of x rays follows a
Poisson distribution, the best estimate of the numberical value of the count-
i-j .--—
ing variance, expressed in (pg/1) , is derived from N, the mean number of
characteristic x rays/^C; Y, the x-ray yield or number of characteristic
x rays/^iC/ug element; and V, the volume of sample expressed in liters. Thus,
s = N/(YV)2
If the total number of characteristic x rays emitted from an element is large,
however, the variance may be estimated from one count without performing a
series of counts and calculating a mean count.
27
-------
2
The instrumental variance, s , is a measure of the variability of the
function of the components of the instruments, such as the proton beam and
the detector. Since the function of the instrumental components follows an
unknown distribution, the best estimate of the instrumental variance, ex-
pressed in (pg/1) , is derived from SN , the variance of replicate observa-
tions of N, the number of observed characteristic x rays/uC; Y, the x-ray
yield; and V, the volume of sample, expressed in liters. Thus,
s.2 = S//CYV)2
The counting and instrumental effects are independent of^one another. If the
interelemental effects are considered negligible, then s , the estimate of
the variance of replicate observations of the elemental concentration, may be
expressed as follows:
2 2,2
s = s + s.
C 1
Since s, the standard deviation expressed in pg/1, may be estimated by deriv-
ing the square root of the variance, then
t 2 j. 2^/2
s = (s + s. )
C 1
When the number of individual elemental observations is large, the pro-
portion of total observations lying within any given range about the mean of
the observations is related to the standard deviation. _Thus, 68 percent of
the observations lie within C + 1 s, 95 percent within C +_ 2 s, and 99 percent
within C +_ 3 s. In this research the confidence limits, C +1 s, were employed.
The detection limit, DL, is generally interpreted as the smallest concen-
tration of an element that can be observed with reliability. However, there
seems to be no general agreement on the definition of "reliability." The
procedure used in this work can be described in terms of the probability
distributions shown in Figure 7. The detection limit was set equal to a,
the theoretical standard deviation. When the actual concentration of an
element is zero or equal to DL, the concentrations observed in repeated mea-
surements would follow distribution A or B, respectively. From Figure 7 it
is evident that the probability of obtaining an observed concentration between
zero and DL is the same when the actual concentration is zero or equal to DL.
Thus observed concentrations in this range are unreliable and are not re-
ported. On the other hand, only 16 percent of distribution A. exceeds DL.
Any observed concentration, Co, which is larger than DL is considered reliable
because there is at least an 84 percent probability that the element is
present in the sample. Given the fact that 50 percent of distribution B
exceeds DL, the detection limit is the smallest concentration of an element
that is needed to have a 50 percent probability of obtaining a reliable
value for C . The sample standard deviation, s, will be used as the best
estimate for o, that is the detection limit.
28
-------
Probability of Observation
0 DL
Concentration
Figure 7. Sketch of probability distributions used to define detection limits,
EVALUATION AND DISCUSSION
Now that the terms for statistical interpretation have been defined,
the procedures used to determine the magnitude of the variability of the
observations of elemental concentration will be described. A number of
drinking water samples, collected from local public water supplies and home
wells, were analyzed. Calculations were then performed to determine the
standard deviation of the observations of each of the 10 elements present in
sufficient concentration for 1) s , the estimate of the counting standard
deviation, to be less than 3 percent of C, the mean of the elemental
observations, and 2) the interelemental effects to be negligible. (Use of
the 3 percent value assured that the counting variance was less than 10
percent of the instrumental variance as shown below.) The results of the
replicate analyses of three typical samples are given in Table 4. Examina-
tion of the data reveals that the average percent coefficient of variation
or relative standard deviation was about 10. Consequently, the magni-
tude of the confidence limits assigned to each elemental concentration
observed thereafter was 10 percent of the observation, i.e., C* + 1 s or
C^ + 0.1 C. These limits can only be considered approximations, however,
since they were derived from observations of a limited number of elements in
a few samples from one geographical area and then applied to a single
observation of each element in each sample.
29
-------
TABLE 4. EVALUATION OF ANALYTICAL PRECISION
E lament
K
Ca
Ti
Fe
Cu
Zn
Sr
Mo
Ba
Pb
Sample 1
*
Average C
V
(ppb) (Percent)
2,100
24,000
14
76
43
36
84
56
12
8
9
8
7
11
10
8
Sample 2 Sample 3
* *
Average C Average C
(ppb) (Percent) (ppb) (Percent)
1,900 9 1,100
7,600 6 53,000
104 10 14
29 9
19
39 9 168
17
107
9 10
10
9
9
9
9
11
12
Five targets were analyzed for each sample.
The relative significance and magnitude of the counting and instrumental
variances may be derived after dividing the second of the previously cited
equations for s , the variance of the replicate observations of elemental
concentrations, by C^, the square of the mean of the observed elemental
concentrations. Thus,
o _?
2-9
= s
When the percent coefficient of variation is 10 (i.e. s= .1 C) and the
counting standard deviation is 3 percent of the mean elemental concentra-
tion, the equation becomes
2 ,-2
(0.1) = (0.03)'
+ s.
i
7c'
then
2 -2
s . /C
= (0.095) ~ (0.1)'
The magnitude of the instrumental variance is approximately the variance of
the elemental observations, i.e., 10 percent of the mean concentration. The
instrumental variance is, therefore, considered the primary factor in the
variance of replicate observations of elemental concentration above the
detection limit. As the mean of the observed elemental observations
approaches the elemental detection limit, however, the value of the expres-
sion, s^/C , approaches unity since the detection limit is equal to s.
30
-------
Therefore' s 2/C2 + s.2/C) *1
01 2-2
Substituting the previously derived approximate value for s. /C , the re-
lationship now becomes,
s 2/C2 + 0.01) ^ 1
c
The counting variance appears to become the most significant factor in the
variance of the elemental observations as the latter approach the detection
limits.
The magnitude of the counting variance may be estimated more specifi-
cally after considering the x-ray spectrum. As shown in Figure 5, the spec-
trum consists of peaks produced by monoenergetic, characteristic x rays and
continuous background from Bremsstrahlung and nuclear gamma rays. In any
particular peak, N , the total number of observed x rays/uC; N , the number
of observed characteristic x rays/jiC; and N^, the sum of the numbers of
observed background x rays and gamma rays/^C, are related as follows:
N = N - N,
p t b
When the total number of observed x rays is large, the numerical values for
s , sh > an tne corresponding variances of N , N, , and N , each ex-
pressed in (ug/l)2, may be estimated as follows:
st2 = Nt/(YV)2 = (Np + Nfe)/(YV)2
sb2 = Nb/(YV)2
sp2 = (Np + 2 Nb)/(YV)2
(In these expressions, Y and V are defined as before.) The numerical value
of s , the counting variance expressed in (ug/1) , then becomes
s2 = s2=(N +2 N, )/(YV)2
c p p b
When the elemental observations are near the detection limits and N is less
than N, , the numerical value of the counting variance expressed in (ug/1)2
then approaches the following:
sc2 -v 2 Nb/(YV)2
In actual practice a least-squares program is utilized to calculate the
contribution of the counting variance to the variance of the observations
of elemental concentration.
For the purpose of evaluating the detection limits of the various
elements in targets prepared from 30-ml aliquots of sample containing 50 ppm
calcium, a test target containing 1.5 mg calcium was bombarded under the
same conditions used in the multielemental analyses of drinking water
31
-------
samples. Then N^, the sum of the observed numbers of background x rays and
gamma rays per ^C,was determined at the energy of each characteristic x ray
that would be counted in the analyses. Since s , the variance of the repli-
cate observations of the elemental concentration, was previously shown to
approach s^-, the counting variance, then s, the standard deviation of the
elemental observations expressed in p.g/1 can be estimated from the following:
s ~ s ~ (2N/Y2V2)1/2
c b
Since DL, the detection limit, is set equal to s in this research, then the
approximate value of DL can be derived as follows:
9 9 119
DL = s ~ (2N, /Y V ) '
b
Attention is now directed to other factors which can effect detection
limits and analytical precision. Although the material in the target deposit
does not have a major effect on the Bremsstrahlung background, large
quantities of sodium in the sample can have a dominant effect on the high-
energy background. The probability of exciting emission from sodium nuclei
during proton bombardment is relatively large. Most of the 440-keV rays
from sodium do not interact with the x-ray detector, but a few will leave a
small part of their energy in the detector, thus adding to the continuous
high-energy background. Although the number of these events is small, the
K x-ray yield for elements with atomic numbers between 46 and 57 is also
small, so sodium can increase the standard deviation for these elements.
Large quantities of elements heavier than silicon can affect s in
another way. As noted in the previous description of the detector and
electronics, the x-ray counting rate is limited by the noise filter which
is essential for good energy resolution. Thus the number of x rays that can
be observed from trace elements in a fixed time period will be limited by
intense radiation from an abundant element. Since analysis costs are
directly related to bombardment time, the most practical solution is to
reduce the intensity of the dominant radiation with an absorber. When any
type of absorber is used, there will be some loss of x rays needed in the
analysis. Hence a number of compromises must be made. The factors
influencing the selection of the best absorber for drinking water analysis
will be described in Section 6.
The finite energy resolution of the x-ray detector can also affect s
for some elements. An intense x ray from an abundant element may have
approximately the same energy as a low intensity x ray from a trace element.
The standard deviation for that trace element will be limited to a fraction
of the abundant element. This limitation is most common for rare earth
elements. As shown in Figure 1, elements heavier than lanthanum (Z = 57)
are detected by L x rays. The L x rays from rare earth elements occur in the
same energy region as the K x rays from elements with atomic numbers from
24 to 30. The latter are frequently present in relatively large concen-
trations.
32
-------
If the sum of the energy of simultaneous pile-up x rays is close to the
energy of an x ray of interest, the standard deviation for that element can
be increased. Our advanced pile-up detector minimizes the number of pile-up
events in the x-ray spectrum, so this effect is rarely significant.
The problems caused by x rays of similar energy from different elements
can be reduced by selecting the particular x rays used in the analysis.
Elements heavier than calcium emit at least two x rays which are distinctly
different in energy. Typically the K_ intensity is 20 percent of the Ka
intensity. For many elements, the LQ, and La x rays have approximately the
same intensity. If one of these x rays is obscured by a stronger x ray,
the other can be used for analysis.
The detection limits given in Table 5 were calculated using the back-
ground observed from a typical target as described previously. A restric-
tion is placed on the detection limits for samples that have large concen-
trations of calcium. When the calcium concentration was greater than 50
parts per million (pptn), it was necessary to reduce the proton beam intensity
to reduce the calcium x-ray counting rate. The detection limits for other
elements were increased because fewer x rays were observed in the normal
counting period.
The detection limits for elements lighter than scandium are relatively
large because targets were made from one-mi samples. This procedure provided
the detection limits required for the investigation and reduced the correc-
tions necessary for the absorption of low-energy x rays in the sample
material. This problem has been discussed in Section 4.
The variations in detection limits for adjacent elements are caused by
a variety of effects discussed in previous sections of this report. Some
examples of these effects will now be described. The detection limit for
scandium is limited to 0.1 percent of the calcium concentration because the
energy of the Kg x ray from calcium is close to the energy of the Ka x ray
of scandium. The detection limits for titanium and vanadium are larger than
that of chromium because the Mylar absorber used to reduce the calcium x rays
also absorbs significant numbers of titanium and vanadium x rays. In addi-
tion, L x rays from barium, which are relatively abundant in some samples,
occur in the same energy region as the x rays from titanium and vanadium.
The analysis of arsenic and lead present a good example of the use of
alternate x rays to minimize the variability caused by x rays of similar
energy. The difference in energy between the K x ray of arsenic and the
LQ, x ray of lead is only 2 eV, so it is impossible to distinguish between
these two x rays. Fortunately, the L,i x ray from lead is relatively
intense and does not interfere with any common x rays, so it is used to
detect lead. The intensity ratio of the lead L and LSJ ^ x rays has been
measured, so the lead contribution to the combined peak from lead and arsenic
can be subtracted to determine the arsenic concentration. In principle, the
Kj x ray from arsenic could also be used, but it is frequently obscured by
relatively intense x rays from bromine in drinking water samples.
33
-------
TABLE 5. DETECTION LIMITS* FOR PIXE ANALYSIS OF DRINKING WATER
Element
Si
P
S
Cl
K
Ca
Sc
Ti
V
Cr
Mn
Fe
Co
Ni
Cu
Zn
Ga
Ge
As
Se
Br
Rb
Sr
Y*
Zr
Nb
Detection
Limit"!"
(ppb)
5000
500
100
100
100
100
3 or 0.17, Ca
1
1
0.1
1
1
0.1
0.1
1
1
0.1
0.1
0.5
1
0.1
0.1
0.1
0.1
1
0.1
Element
Mo
Tc
Ru
Rh
Pd
Ag
Cd
In
Sn
Sb
Te
I
Cs
Ba
La
Ce
Pr
Nd
Pm
Sm
Eu
Gd
Tb
Dy
Ho
Er
Detection
Limit'1'
(ppb)
1
1
1
1
0.1
1
0.1
0.2
0.3
0.4
0.7
1
2
2.5
2.5
20
18
16
13
12
11
9
8
7
6
5
Element
Tm
Yb
Lu
Hf
Ta
W
Re
Os
Ir
Pt
Au
Hg
Tl
Pb
Bi
Po
At
Fr
Ra
Ac
Th
Pa
U
Np
Pu
Detection
Limit"1"
(ppb)
3
3
3
3
3
0.3
3
3
3
0.3
3
3
0.3
0.1
0.3
3
3
3
3
3
3
3
3
3
3
*The term detection limit is defined and discussed in the text.
t,
For samples with greater than 50 ppm of calcium, the cited detection limit
is multiplied by the factor (calcium concentration in ppm)/50 ppm. This
modification does not apply to scandium.
Since yttrium is used as an internal standard, it is not determined in the
sample.
34
-------
The effect of imperfections in the detector response can be illustrated
with manganese. If the concentration of iron in the sample is very large,
there will be a slight tail on the low energy side of the iron Ka x-ray peak.
Since this makes it somewhat more difficult to determine the background
under the manganese K^ x ray, the detection limit for manganese is not as
good as for chromium.
The increase in detection limit for the elements between cadmium and
lanthanum is caused by the steady decrease in the x-ray yield for these
elements. The detection limits are larger for the rare earth elements
because the analysis uses L x rays occurring in the energy region where
the Bremsstrahlung is large.
Many of the other variations result from the economical use of computer
time during the analysis. The time required to perform and evaluate the
least-squares fit of the x-ray spectrum would be much longer if all elements
were automatically included. If the computer time exceeds the bombardment
time, the cost of analysis would be increased. As a practical compromise,
the elements of primary interest listed in Table 2 were always included in
the least-squares fit. The computer was programmed to recognize other peaks
and add them to the analysis. Of course it is very difficult for the program
to distinguish very small peaks from random variations in the background.
To minimize computer time a relatively high detection limit was used for each
secondary element.
It was necessary to bombard some samples for long periods to obtain the
detection limits given in Table 5. For example, many samples contained
larger amounts of copper than we had anticipated. It was not possible to
reduce the intensity of copper x rays and maintain the required detection
limits for lighter elements. The only solution was to reduce the proton beam
intensity, thus preventing the counting rate of copper x rays from exceeding
the practical limits of the electronic system. We also did not fully
appreciate the significance of y rays from sodium. Again, long runs were
required to reduce the statistical uncertainty in the Y~ray background.
The characteristics of the sample have a major effect on detection
limits in PIXE analysis. Limits for typical samples of a particular type
can be given in the relatively simple form shown in Table 5. Simple state-
ments of performance are appealing, but they are unrealistic when the
technique is sensitive to many elements and sources of interference modify
results. Since the goal of this program was to obtain the best possible
detection limits for a wide range of elements, we have maintained the detec-
tion limits listed in Table 5 for the vast majority of samples. As described
above, many samples required unexpectedly long bombardment periods; hence the
actual cost of those analyses exceeded the charge made to the program.
If PIXE analysis is to be provided at fixed cost, the detection limits
must be given in a more complex form than that used in Table 5. For example,
factors should be included that account for the presence of high y-ray
background from elements, such as sodium, and intense x rays from common
medium-weight elements, such as copper and zinc. Interference effects can
be stated in a relatively simple form as illustrated by the effect of calcium
35
-------
on scandium described in this work. However, if all conceivable possibil-
ities of interference were included, the table of detection limits would be
too long. A subjective judgement is needed to define the interferences
which are likely to be significant in the analysis.
With any practical statement of detection limits, it is very difficult
to obtain the desired results for some unusual samples. For example, in
this program approximately 50 samples contained several ppm of mercury, a
preservative, accidentally added when the samples were collected. A heavy
element, such as mercury, emits 15 x rays which cover a wide energy range.
When the concentration of mercury was several orders of magnitude larger
than anticipated, it was essentially impossible to avoid x-ray interferences
and an increase in the variability of the observations for other elements.
An important distinction should be made between PIXE and many other
"multielemental" techniques. Although the computer program may be designed
to look for selected x rays in the first phase of the analysis, all x rays
detected from the sample are maintained in the x-ray spectrum. All optical
emission techniques, claimed to be multielemental, retain only a limited
portion of the emitted spectrum. Therefore unusual elements can not be
added to the analysis after the sample has been processed. Not only is
potentially valuable information lost, but it is impossible to recognize the
presence of unusual elements which may interfer with the elements of
interest. This is not the case with PIXE analysis. Unusual elements, whose
concentrations exceed the detection limits, are added to the analysis and
any variability caused by interference is evaluated.
36
-------
SECTION 6
PROCEDURES AND QUALITY CONTROL USED IN THE
ANALYSIS OF DRINKING WATER SAMPLES
The analysis for all elements heavier than calcium was made on targets
prepared by vapor filtration from approximately 30-ml samples of drinking
water. Each sample, contained in a polyethlene cubitainer, was shaken
vigorously by hand before pouring an aliquot into the cylindrical tube of
the vapor filtration apparatus. This procedure eliminated the possibility
of contamination which could occur when graduated cylinders or pipets are
used.
Each vapor filtration cylinder was calibrated so that the volume of
sample in the cylinder could be determined from the height of the water
column. Thirty ml of water were pipetted into each of the vapor filtration
cylinders. The height of the water column was measured with a scale marked
in millimeters. The scale rested on the collar at the bottom of the
cylinder (see Figure 3). Each collar was machined to a tolerance of less
than 0.1 mm so that the position of the scale was reproducible and the
height of the column below the first mark on the scale was known. The
total height of the water column in each cylinder was measured. The average
height of a. 30-ml column in 120 cylinders was 150 mm with a standard de-
viation of 1.5 mm. It is evident that the inside diameter of the cylinders
(16 mm) is constant to approximately 1 percent.
A scratch mark was made on each cylinder at the 30-ml level as an aid
in pouring a sample. However, the actual volume of the sample was measured
by holding the millimeter scale at the side of the cylinder. The height of
the column could easily be measured to within 1 mm. When this variability
is combined with the standard deviation in the average height of a 30-ml
column, the expected standard deviation of the measurement is less than 1.8
mm or 1.2 percent. Each mm on the scale corresponds to 0.2 ml of water; the
measured height in mm was multiplied by 0.2 ml to obtain the sample volume.
In Section 4 we explained that one-mi samples were used for the analysis
of elements lighter than scandium (Z = 21) to reduce the corrections re-
quired for the absorption of low energy x rays in the sample material. The
water used for analysis of light elements was poured into a clean plastic
beaker. One ml of water was transferred to the vapor filtration device with
an Eppendorf pipet (Brinkmann Number 22-35-090-1). New pipet tips were
cleaned with our standard procedure, described later in this section. Each
tip was used one time, then discarded. Although the samples used in the analy-
sis of light elements required more handling than those used in the analysis
37
-------
heavy elements, the danger of contamination was negligible because the
desired detection limits were 100 times higher than for heavy elements.
To evaluate the accuracy of the complete analytical procedure, an
internal standard, 4 u,g yttrium in one ml ultra-pure water, was added to each
sample aliquot using an Eppendorf pipet. (The previously described precau-
tions for the pipet tips were used.) The same quantity of yttrium tracer
was added to all ultra-pure water samples processed as blanks. The role that
the yttrium played in the analysis will be considered at several points in
the subsequent discussion.
Occasionally a water leak would develop at the top of the cellophane film
shortly after the sample was poured into the cylinder. Some targets were also
lost when cellophane films broke on the pumping stations after the samples
had dried. Sample preparation was scheduled to enable close observation at
these times. A vacuum valve was included at each of the 120 pumping ports
so that faulty samples could be isolated from the vacuum system.
All sample preparation, handling, and target storage was done in a
pressurized clean room. The room had a single door and no windows. Air was
pulled into the room through a high efficiency electrostatic filter (Trion
Model TTM-II). The air intake was located underground in an air-conditioned
part of the building; hence the input air to the filter was already relative-
ly clean. The air pressure in the room was higher than in the outside hall
so that air left the room when the door was opened. The water vapor pumps
were located one floor below the clean room so there was no danger of the
pumps causing sample contamination. The pumps were connected to the vapor
filtration apparatus with a three-inch pipe passing through the floor of the
clean room.
When the sample was dry, the cellophane film was removed from the vapor
filtration device and mounted in a plastic frame. Since the film was precut
to fit clamping ridges in the frame, only a few seconds were required to
position the sample deposit in the center of the frame. The two halves of
the frame were then snapped together and the finished target was stored in a
covered plastic tray. When the targets were scheduled for bombardment they
were transferred to circular plastic trays which fit the modified photography
projector in the bombardment chamber (described in Section 3).
Standard solutions provided by the USEPA to check the quality of the
analyses were handled in essentially the same way. The solutions were
diluted with ultra-pure water according to the instructions provided with the
standards. Normally, 30 ml of the diluted solution were used in the vapor
filtration device. A smaller volume was used to prepare a few of the
standards to minimize the damage to the cellophane by the aqua regia present
in the solutions. At first 15 ml of the standard was prepared. If the
target broke, then 5 ml was used. Targets, prepared from these smaller
volumes, were bombarded for a longer period to obtain the desired detection
limits for the heavier elements.
38
-------
Several alternatives were tried before a satisfactory method was found
for analyzing the 1.5-ml aliquots of acid preservative furnished by the USEPA.
At first the 1.5-ml aliquot was diluted with one liter of ultra-pure water.
With this dilution, very few elements were detected. Next, aliquots were
diluted with 200 ml of ultra-pure water. Many of these targets broke because
of the relatively high concentration of acid. Finally, aliquots were diluted
with 500 ml of ultra-pure water to obtain durable targets having a two-fold
weight increase in trace elements.
Targets prepared from 30-ml and 1-ml drinking water samples were
bombarded with 4-MeV protons. The most intense x rays emitted were from
calcium. In the analysis of the 30-ml samples, one-microampere (u.a) beam
current and a 0.8-mm thick mylar absorber were used. The latter absorbed
the intense calcium x rays and provided optimum transmission for x rays from
elements heavier than calcium. A 0.09-mm Mylar absorber and a 0.05~(j,a beam
current were used for the analysis of elements lighter than scandium in one-mi
samples. This absorber prevented scattered protons from reaching the x-ray
detector, but it was sufficiently thin to permit silicon to be included in
the analysis. Although the 1-ml sample had much less material than the
30-ml sample, a lower beam current was used since the calcium x-ray intensity
was not reduced by the 0.09-mm Mylar filter.
The analysis of each 30-ml sample for elements heavier than calcium
required 30 minutes to obtain the detection limits given in Table 5. The
analysis time for elements lighter than scandium was only 5 minutes because
less data was required to achieve the relatively higher detection limits.
Analyses were normally performed from 4 pm until 8 am the following
morning. At the beginning of each 16-hour running period, several system
tests were performed. A target having small chips of titanium, zinc, molyb-
denum, and tin metals was bombarded to provide a series of clean x-ray
peaks covering the energy range of interest. The computer analyzed these
peaks to determine the energy calibration of the system and to check the
energy resolution of the detector. There was no significant change in the
response of the detector over the two-year period of the measurements. The
energy calibration of the system varied less than 0.01 percent.
As explained in Section 3, it was not necessary to remeasure a full
set of calibration targets to check the x-ray yield calibration. One
calibration target was analyzed at the beginning of each 16-hour analysis
period to ensure that the proton charge-measuring system was operating
correctly. The charge-measuring system proved to be very reliable and
required maintenance only once in the two-year period. Once each week,
calibration targets were used to ensure that absorption in the Mylar filters
was unchanged. This test did not need to be performed more frequently
because the buildup of material on the surface of the absorber was very slow.
When the absorption had increased by 3 percent, the absorber was cleaned.
39
-------
Although the system has proved to be extremely reliable, the possibility
of intermittent failures was considered. The yttrium internal standard
provided a running check on each analysis. A temporary failure in the
charge-measuring system or the detector would immediately produce an
abnormal yield of yttrium x rays. The computer checked the yttrium counting
rate several times a minute and would stop the measurement and inform the
operator if a problem developed. Almost all low yttrium yields could be
traced to a failure in the target; either some of the sample was lost in the
vapor filtration process, or the target was broken during bombardment. New
targets had to be made for approximately 10 percent of the samples. A
series of low yttrium yields and subsequent checks with calibration targets
identified the one occurrence of failure in the charge-collection system.
At the beginning of the project, the calcium concentration in each
sample was checked with an ion sensitive electrode so the volume of water
poured into the vapor filtration device could be adjusted. However, since
relatively few of the samples had sufficient calcium to require reduced
volume targets for heavy element analysis, this procedure was discontinued.
Instead, we adopted a standard procedure of analyzing approximately 30-ml
samples. When a sample with high calcium concentration reached the data
collection stage, the intensity of the proton beam was reduced to limit the
x-ray counting rate. Although a large volume sample was used, the calcium
still had the specified effect on detection limits because less data could
be collected in the normal running period. Partial loss of the sample
deposit from some of these targets during bombardment was easily recognized
by a low yttrium yield. When this occurred, a new target was made from a
20-ml sample. Since we were handling thousands of targets and some of them
had to be remade for other reasons, this was the most efficient method for
routine handling of samples with high calcium concentrations.
The performance of the beam-switching system and the pile-up rejector
was continuously monitored by the computer. The computer calculated the
amount of pile-up that should occur from the observed counting rates of
intense x rays. Any pile-up peaks statistically significant relative to the
background were included in the least square fit to prevent mistakes from
being made in the measurement of x-ray peaks. The computer compared the
observed pile-up to that expected and notified the operator if the two rates
did not agree. This procedure provided a running check on the pile-up
detector and on the beam-pulsing system. If the beam was not being switched
correctly, the probability of pile-up increased dramatically. The beam-
switching system has required maintenance twice but the pile-up rejector
has required no maintenance in the two-year running period.
The reproducibility of the analysis was evaluated periodically by analyz-
ing multiple targets from typical samples. Since one-liter drinking water
samples were provided for analysis, they could be used in the evaluation of
precision. The procedure for obtaining and analyzing the targets was varied.
For example, multiple targets were made at the same time for several samples.
Sometimes these targets were analyzed immediately to evaluate the instrumental
contribution, s^, to the standard deviation (discussed in Section 5). Elements
present in relatively high concentrations were used so that the variations
40
-------
due to counting statistics were not a significant factor in the standard
deviation. On other occasions, the multiple targets were held for four weeks
and then analyzed. This procedure evaluated the stability of the targets and
served as a continuing check on s . To evaluate the consistency of the
target-making procedure, aliquots of some special solutions were prepared on
a weekly basis. The analytical observations of these samples were averaged
and the coefficient of variation was calculated. Typical results for tests
of this type have been given in Table 3. Throughout the course of this
contract, the average percent coefficient of variation for repeated analyses
was approximately 10.
One of our most difficult tasks has been to establish and maintain
procedures which would minimize incidences of sample contamination. Preven-
tion of contamination was an important goal in the development of the vapor
filtration technique. It was also essential that the samples be prepared in
a very clean environment such as the clean room described previously in this
section. An ample supply of ultra-pure water was also necessary to clean
apparatus, to prepare standard solutions, and to evaluate cleaning procedures.
Ultra-pure water was provided by a system obtained from Culligan Corp.
The Physics Department distilled water supply was used as the input for the
ultra-pure water system. The total dissolved solids in the input water were
less than 10 ppm. The water passed through a carbon filter tank (Cullar-F,
1618-00), two mixed bed ion exchange tanks (Model 2603-26) and finally
through a submicron filter (Model 9095-76). The resistance of the water,
continuously monitored by a built-in meter, was always greater than 16
megohms. Since distilled water was used for the input of the ultra-pure
water system, component replacement was required only at three-month
intervals.
Several 30-ml samples of ultra-pure (non-acidified) water were analyzed
once a month. The number of targets, average elemental observations, and the
corresponding standard deviations for a typical three-month period are given
in Table 6. There was some variation in the quality of the ultra-pure water.
The concentrations of 70 of the 76 elements were less than the detection limits.
The most abundant element was iron while nickel, copper, zinc, cadmium, and
lead were sometimes present in significant concentrations. In this example,
the concentrations were higher during the middle of the period than at the
beginning or end. No general pattern was found for the variation in concen-
trations. This ultra-pure water was more than adequate for cleaning purposes
since only a few drops of water were left in the vapor filtration devices at
the end of the cleaning cycle.
It was difficult to evaluate the occurrence of contamination at very
low elemental concentrations. The ultra-pure water used to evaluate contam-
ination was acidified so the conditions during the vapor filtration process
were approximately the same as during the preparation of drinking water
samples. Some low-level impurity in the acid and ultra-pure water was inevit-
able, so it was impossible to measure contamination satisfactorily. The
best available technique was to make replicate targets from the same
acidified ultra-pure water. The trace elements in the acidified ultra-pure
water should have been consistent from target to target, but there could have
41
-------
TABLE 6. TRACE ELEMENTS IN ULTRA-PURE WATER
Sample 1
Element
Cr
Fe
Ni
Cu
Zn
Cd
Pb
*
Average
ppb
0.31
6.8
0.21
0.9
<0.5
0.22
0.31
Standard
deviation
ppb
0.11
1.5
0.10
0.5
0.12
0.22
Sample 2
, Standard
Average deviation
ppb ppb
0.12
8.1
0.36
1.3
2.5
0.50
0.40
0.10
2.0
0.15
0.8
1.0
0.32
0.33
Sample 3
^ Standard
Average deviation
ppb ppb
0.40
6.1
0.26
<0.5
1.0
<0.1
1.10
0.20
1.8
0.10
0.8
0.20
t
tt
Average of three target analyses.
Average of five target analyses.
been considerable variation in the contamination. If the observed variations
were consistent with the expected standard deviation, then the contamination
was either negligible or constant. We expect the most likely source of
contamination was airborne particles that would not produce a constant con-
tamination level.
Ultra-pure water was a better solution for contamination tests
than a standard solution with known elemental concentrations. The
instrumental variation was usually negligible for the ultra-pure water
analyses, but its significance would increase with the elemental concentra-
tion. A 2-ppb contamination in ultra-pure water would be much easier to
detect than the same contamination in a solution containing 20 ppb of the
element.
A set of six vapor filtration devices were assigned for use with
acidified ultra-pure water. No other type of sample was placed in these
devices. The devices were originally cleaned with a 10 percent nitric acid
solution. They were rinsed with ultra-pure water before each test sample
was processed. These devices were randomly placed on the pumping apparatus,
and 30 ml of acidified ultra-pure water were processed at the same time that
regular samples were being processed. Thus the probability of contaminating
these targets with airborne particles was the same as for regular samples.
The results for a typical evaluation of this type are given in Table 7.
42
-------
TABLE 7. ACIDIFIED ULTRA-PURE WATER ANALYSES USED TO EVALUATE CONTAMINATION
Observations (ppb) Standard
Element 12 3 45 6 Average deviation
ppb ppb
Cr
Fe
Ni
Cu
Zn
Cd
Pb
< 0.1
17.0
0.21
1.3
3.1
0.15
0.9
0.2
14.0
0.27
1.7
2.2
0.46
0.5
< 0.1
9.5
< 0.10
0.9
1.6
0.15
1.1
< 0.1
10.0
0.31
2.1
3.6
0.28
1.0
0.3
15.0
0.28
1.4
2.0
0.30
1.3
< 0.1
11.0
0.20
3.3
1.8
0.35
0.8
--
12.8
0.25
1.8
2.4
0.30
0.9
--
3.0
0.05
0.8
0.8
0.10
0.3
The observed standard deviation for nickel, copper, zinc, and cadmium
were smaller than the expected standard deviation (0.1, 1.0, 1.0, and 0.1,
respectively), so contaminations from these elements were either negligible
or constant. In four out of six targets the chromium concentration was less
than the detection limit. The analysis of the other two targets showed a
low level of chromium contamination.
A lead concentration of 1 ppb with a standard deviation of 0.2 ppb was
observed in the cellophane target backing material. The average lead
concentration shown in Table 7 was essentially the same as that in the
cellophane but the standard deviation here was slightly larger. Since
several ppb lead were observed in most drinking water samples, small varia-
tions should not have a major effect on the interpretation of lead analyses.
The vapor filtration devices had to be thoroughly cleaned to avoid
contamination. The first step in cleaning the devices was a 30-minute rinse
with flowing distilled water in a siphon-type pipette washer made of poly-
ethylene. (Made by Nalgene Company; Cole-Parmer stock numbers 6190 washer-
rinser and 6155 basket.) Next, the devices were soaked for 30 minutes in
10 percent nitric acid. The tank used for the nitric acid soak was located
under a hood to prevent acid vapor from attacking metal surfaces in the clean
room. The final rinse was for 30 minutes in flowing ultra-pure water.
Water droplets were shaken off and the devices stored in covered poly-
ethylene trays. The devices were not dried prior to storage. There was
probably much less contamination from having a few drops of ultra-pure water
left in the device than from leaving the devices open to air for a long
drying period. As stated previously, all handling and storage were confined
to the clean room.
43
-------
The cleaning procedure was thoroughly tested before drinking water
analysis began. The vapor filtration devices were contaminated by processing
solutions containing large concentrations of the elements of primary interest
(Table 2). The devices were cleaned with the procedure described above.
Then targets were made by processing 30 ml of acidified ultra-pure water
through the devices. No measurable contamination was observed in the vapor
filtration devices.
Since it was possible that the vapor filtration devices became more
difficult to clean as they were used, similar tests were continued during
the drinking water analyses. When a set of regular samples was being put on
the pumping manifold, six devices, selected randomly, were checked for
contamination. These devices previously had been used with regular samples
and cleaned by the standard procedure. Targets were made from 30-ml aliquots
of acidified ultra-pure water. The analyses for a typical set of these
targets are shown in Table 8 and they are statistically indistinguishable
from the results in Table 7. The same solution was used for the two tests,
TABLE 8. EVALUATION OF CLEANING PROCEDURES FOR VAPOR FILTRATION DEVICES
Observations (ppb)
Element
Cr
Fe
Ni
Cu
Zn
Cd
Pb
1
0.25
14
0.19
2.1
2.7
0.25
1.1
2
< 0.1
21
0.21
1.8
2.0
0.1
1.3
3
< 0.1
12
0.30
3.1
3.3
0.15
0.7
4
< 0.1
10
0.15
2.9
2.2
0.31
1.0
5
0.20
13
0.24
2.4
2.8
0.33
0.9
6
< 0.1
17
0.27
2.3
3.6
0.27
0.8
Average
ppb
—
14.5
0.23
2.4
2.8
0.24
1.0
Standard
deviation
ppb
—
4.0
0.05
0.5
0.6
0.1
0.2
but the vapor filtration devices used to obtain the results in Table 7 were
only used with ultra-pure water. Thus, any contamination left in the vapor
filtration devices during the washing process was too small to be measured.
One source of unexpected contamination was observed. Field personnel,
associated with the cardiovascular disease study, collected a number of
drinking water samples at each site for determination of a variety of
inorganic constituents by several laboratories. Unfortunately, a number of
samples were either incorrectly preserved or the samples were not sent to the
correct laboratory. As a result, the Purdue Accelerator Laboratory received
about 50 samples preserved with mercuric chloride (HgC^). The pH of the
incoming samples was measured to detect the absence of nitric acid, the
preservative for the trace element analyses. These measurements did not
44
-------
reveal, however, the accidental preservation of a sample with both mercuric
chloride and nitric acid. A target with large quantities of mercury would
contaminate other targets in the bombardment chamber at the same time. It
was necessary to make new targets for many samples to determine whether
mercury was really present in the original drinking water. Although a
determined effort was made to check cases that were suspicious, it is
possible that some of our reports for mercury are high due to contamination.
Since mercury was not one of the elements of primary interest in this
investigation, we did not evaluate the probability of mercury loss during
bombardment when mercury was present at trace levels.
Five copies of a computer printout of the results of each analysis were
forwarded to the USEPA project oficer. The results of all analyses were
also recorded on computer magnetic tape for statistical evaluation by the
USEPA staff.
The original data and analyses were stored on magnetic tape and will be
held at the Purdue Accelerator Laboratory until May 1, 1978. The analysis
of any sample can be re-examined by reading the data and analysis parameters
back into the computer. The remainder of each sample has been retained in
its original container and will be stored at the Purdue Accelerator Labor-
atory until May 1, 1978. The targets have not been retained.
45
-------
SECTION 7
ANALYSIS FOR SODIUM USING ATOMIC ABSORPTION
Plastic labware was used in the preparation and storage of all solutions
to avoid contamination with sodium, often observed with the use of glass.
The labware was washed using a tap water rinse, followed by one percent
nitric acid bath and then two washes in deionized water. Blanks were
prepared by filling some of the washed sample bottles with deionized water
and analyzing them in the same manner as regular water samples. The concen-
tration of sodium in the blanks was always less than 0.25 ppm, indicating
that the cleaning procedure was satisfactory.
Nalgene flasks and Eppendorf pipets were used in the preparation of
standard solutions. Appropriate aliquots of Fisher Scientific Company sodium
atomic absorption standard were diluted with deionized water. A 100-ml
aliquot of each drinking water sample was transferred from the cubitainer to
a small plastic bottle and stored at room temperature until the sodium
analysis was performed.
A Perkin Elmer Model 306 spectrophotometer with deuterium background
correction was employed for all measurements. The instrument was operated
in the absorption mode with an air-acetylene flame. Samples containing more
than 2.5 ppm sodium were analyzed using the 330.2-nm line, a three-slot
Baling burner, an air flow of 23.2 liters per minute, and an acetylene flow
of 7.6 liters per minute. A set of six sodium standards with concentrations
of 3.13, 6.25, 12.5, 25, 50, and 100 ppm were run before and after each set
of measurements (100 or less samples per set). A linear least-squares fit
of the observations was performed to obtain a calibration graph. The
standards were also run singly during a set of measurements for quality
control purposes. If the concentration of the sodium in a drinking water
sample exceeded 100 ppm, the sample was diluted by a factor of ten using an
Eppendorf pipet and a Nalgene flask and the analysis was repeated.
Samples containing less than 2.5 ppm sodium were analyzed using the
589.6-nm line, a nitrous oxide burner, an air flow of 9.2 liters per minute,
and an acetylene flow of 5.8 liters per minute. Five solutions with sodium
concentrations of 0.25, 0.5, 0.75, 1.0, and 2.0 ppm were used for standard-
ization and quality control in the manner described previously.
Twelve drinking water samples were selected to evaluate the reproduc-
ibility of the analyses. Six aliquots of each sample were analyzed. The
average of the six analyses of each sample and the standard deviation of the
measurements are presented in Table 9.
46
-------
TABLE 9. PRECISION OF ATOMIC ABSORPTION MEASUREMENTS
OF SODIUM IN DRINKING WATER SAMPLES
*
Average concentration
(ppm)
53.5
53.1
24.1
23.7
19.3
18.6
18.3
12.6
11.8
8.6
9.2
4.8
Standard
deviation
(ppm)
1.33
1.59
0.93
0.93
1.1
1.6
0.9
1.4
1.8
1.3
1.8
0.9
Average of six determinations,
47
-------
REFERENCES
1. Johansson, T. B., R. Akselsson, and S. A. E. Johansson. X-Ray Analysis:
Elemental Trace Analysis at the 10~^ g Level. Nuclear Instruments and
Methods 84, 141-143, 1970.
2. Johansson, A_ E., and T. B. Johansson. Analytical Applications of
Particle Induced X-Ray Emission. Nuclear Instruments and Methods 137.
473-516, 1976.
3. Folkmann, F., J. Borggreen, and A. Kjeldgaard. Sensitivity in Trace-
Element Analysis by p, a, and 1°0 Induced X-Rays. Nuclear Instruments
and Methods 119, 117-123, 1974.
4. Rickey, F. A., K. A. Mueller, P. C. Simms, and B. D. Michael. Sample
Preparation for Multielemental Analysis of Water. IN: X-Ray Fluores-
cence Analysis of Environmental Samples, T. B. Dzubay, Ed. Ann Arbor
Science Publishers, Ann Arbor, Mich., 1977. pp. 135-143.
5. Butler, N. S., and P. C. Simms. A Pulse Pile-Up Detector for Use with
with Si-Li X-Ray Detectors. To be published.
6. Simms, P. C., and N. S. Butler. On-Demand Pulsed Beam System for Proton
Induced X-Ray Analysis. To be published.
48
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EPA-600/1-78-058
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
The Multielemental Analysis of Drinking Water Using
Proton-Induced X-Ray Emission (PIXE)
5. REPORT DATE
September 1975
issuing date
6. PERFORMING ORGANIZATION CODE
7 AUTHOR(S)
P.C. Simms and F.A. Rickey
(See item 15)
8. PERFOPM'NG ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Purdue Research Foundation
Division of Sponsored Programs
Executive Building
West Lafayette, IN ^7907
10. PROGRAM ELEMENT NO.
ICC6lk
11. CONTRACT/GRANT NO.
68-03-2178
12. SPONSORING AGENCY NAME AND ADDRESS
Health Effects Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH ^5268
13. TYPE OF REPORT AND PERIOD COVERED
Final, 2/2U/75 - V26/77
14. SPONSORING AGENCY CODE
EPA/600/10
15 SUPPLEMENTARY NOTES
Dept. of Physics, Purdue University
West Lafayette, IN Vf907
16 ABSTRACT
A new, rapid, and economical method for the nmltielemental analysis of drinking
water samples is described. The concentrations of 76 elements heavier than
aluminum are determined using proton-induced x-ray emission (PIXE) technology.
The concentration of sodium is evaluated using an atomic absorption approach.
Targets for PIXE analysis are prepared from aqueous samples using a new "vapor
filtration" technique Each target is bombarded with protons, supplied by a
nuclear generator The resulting x rays are observed with an energy-sensitive
semiconductor detector and the data recorded and processed with an on-line com-
puter. Excellent detection limits (0.1 to 100 ppb) are obtained for most of
the elements heavier than silicon.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
Chemical analysis, elements, x-ray
fluorescence analysis, x-ray spectra,
emission spectroscopy, atomic spectro-
scopy, accuracy, precision, sensitivity,
water supply, drinking water, potable
water
b.IDENTIFIERS/OPEN ENDED TERMS
Multielemental analysis,
proton-induced x-ray
emission, atomic ab-
sorption
COSATI Field/Group
99A
68D
18. DISTRIBUTION STATEMENT
Release to public
19. SECURITY CLASS (ThisReport)
Unclassified
21. NO. OF PAGES
59
20 SECURITY CLASS (This page)
Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77)
PREVIOUS EDITION IS OBSOLETE
49
•&U S GOVERNMENT PRINTING OFFICE 1978-757-140/1461 Region No. 5-11
------- |