Chapter 12 X-Ray fluorescence of obsidian: approaches to ...
Transcript of Chapter 12 X-Ray fluorescence of obsidian: approaches to ...
Chapter 12
X-Ray fluorescence of obsidian: approaches to calibration and the
analysis of small samples Jeffrey R. Ferguson
Introduction
For decades, archaeologists have used a number of techniques to dClcnnine the
compositions of geological sources of chemically homogenous materials and then
attempted to match artifacts to sources to understand trade and exchange of material
objects. Such provenance research is common with glassy rhyolites (obsidian) that
were lIsed wherever available as a source of raw material for flaked slone artifacts
such as projectile points and cutting tools. Handheld X-ray fluorescence. of all of
the analytical techniques available for obsidian compositional analysis, has the
potential to make the greatest impact by combining non-destructive analysis with
rapid results, relatively low equipment and analysis cost, and the option of in-field
analysis. While handheld XRF is a powerful research tool, its successful use in
obsidian provenance research requires an understanding of X-ray physics. igneous
petrology, the calibration process, and the ability 10 test a sufficient variety of
homogeneous and well-characterized reference materials suitable for developing
a valid calibration curve. The potential for "point and shoot" handheld XRF is
hampered in pan by the all of lhese factors.
This chapter addresses precision requirements of a matrix-specific calibration,
the choice and analysis of reference standards, spectral normalization to account
ror differences in sample geometry and size, and the use of secondary targets to
fine tune the spectra to maximize sensitivity to the elements of interest. While the
focus here is on obsidian provenance research, many of the same basic principles apply to almost all applications of XRF.
Compositional studies of anifacts have been an integral pan or archaeological
"cientific investigations for decades, providing insight into many aspects of past
human behavior, and obsidian sourcing studies are often presented as the "poster
child" of adherence to the provenance postulate (Weigand, et al. 1977; Neff 1998).
The requirements for materials to permit successful provenance studies as stated in the provenance postulate are that measurable intersource variability must exceed
intrasource variability. Obsidian is a nearly ideal material for provenance studies
402 Jeffrey R. Ferguson
because it occurs in a relatively limited number of geologic contexts and typically
exhibits uniform chemistry within particular outcrops. There are numerous
problematic cases of sources with complex and even overlapping chemistry
(Braswell and Glascock 1998; Glascock et al. 1999: Ambroz et al. 200 1), but in
general obsidian studies are extremely successful at linking artifacts with geologic
sources wherever it was used in the past.
Obsidian is a glassy rhyolitic rock, meaning that it has high silica content and
a disordered atomic structure (Shackley 2005). The disordered atomic structure
allows obsidian to fracture predictably and with extremely sharp edges. Obsidians
arc formed on ly in specific volcank events when magma of the right chemistry
is cooled under specific conditions making it a remarkably rare material. The
fracturing ability and limited natural outcrops made obsidian a valuable resource
that in many cases was carried or exchanged across great distances.
Provenance studies are particularly appealing to archaeologists because they
can document migrations. logistical movements, exchange relationships, and
tool production , use. and discard in a spatial context (Duff 1999; Shackley 2002:
Carter et al. 2007: Eerkens el al. 2007; Ferguson et al. 2010) . Obsidian has the
additional benefit of potentially providing direct temporal informalion through the
measurement of water absorption; a process known as obsidian hydration (Friedman
and Trernbour 1978, ] 983; Origer 1989). When coupled with temporal information
acquired either through obsidian hydration analysis or by other means . such
temporally-sensitive artifact types , stratigraphic sequences, or direct association
with dated materials, it is possible to examine changes in obsidian procurement
patterns over time spans ranging from decades to hundreds of thousands of years.
No doubt the reasonable cost of obsidian sourcing relative to the valuable cultuml
information obtained has entrenched obsidian provenance in the world of contract
archaeology, driving further research in the development of analytical techniques.
geologic characterization, and the cultural interpretation of sourcing results.
This chapter is aimed at providing a technical background for the analysiS
of obsidian using XRF, with a particular focus on proper calibration techniques
and the analysis of small artifacts. This chapter is included in a volume focusing
on handheld XRF because handheld XRF users tend to be most likely to
underestimate the complexity of proper analysis. The analytical issues presented
here are common to both handheld and lab-based XRF. This paper is not intended
to discuss the geologic context of obsidian in detail , proper methods of source
sampling and characterization, or the anthropological implications of obsidian
provenance studies. There are a number of excellent publications on these topics.
starting with volumes by Shackley (l998a, 2005) and edited volumes by Glascock
and others (Taylor 1976; Hughes 1984, 1989; Glascock 2002; Glascock el al.
X-ray fluorescence of obsidian 403
2(07) and these volumes contain references to many of the hundreds of relevant
publications.
Analytical techniques for obsidian
Unlike many other geologically-derived materials typically recovered from
archaeological sites, obsidian sources lend to follow the requirements of the
provenance postulate for major, minor, and trace elements, permitting the
successful application of a variety of analytical techniques. Specific details of
the analytical techniques are readily found in existing publications (i.e. Shackley
1998b) so the discussion here is limited to the differences between the techniques
and their advantages and limitations as they relate to obsidian analysis.
Visual sourcing is occasionally put forth as a more cost-effective method for
examining large assemblages in contexts where all geologic sources are well
known (Weisler and Clague 1998), however the reliability of visual sourcing is
highly questionable (Bettinger et al. 1984; Braswell et al. 2(00). On more than
one occasion 1 have seen archaeologists working in the American Southwest
get excited after finding an artifact with a green translucent color, a feature
characteristic of the Pachuca source in Central Mexico. In one case it turned out
to be from Pachuca, comprising one of the handful of Mexican obsidian artifacts
recovered from north of the Mexican border (except for a small cluster in the
southern tip of Texas [Hester et al. 1991: Ferguson and Skinner 2006]). In the
second , the artifact was from the Antelope Wells source in southwestern New
Mexico that is somewhat similar to the Pachuca source as it also very high in
zirconium and has a green translucent color. While there may be a few cases of
material so visually unique as to aUow some correct visual source assignments ,
there is an error rate most researchers are not comfortable accepting.
Neutron activation analysis (NAA) was one of the first techniques used for the
chemical characterization of archaeological obsidian (Gordus et al. 1968; Glascock
et al. 1994; 1998). NAA allows for the precise determination of a wide range
of clements that can vary depending on the nuclear properties of the elements ,
irradiation conditions, and timing of the gamma count(s), but it is an expensive
technique and typically requires the destructive analysis of at least some portion
of the artifacts (although, as described later, there are options for non-destructive
NAA of small samples). NAA also requires considerable sample preparation time
and cost in the fonn of irradiation and the use of high-purity sample containers.
Another common technique is inductively coupled plasma mass spectrometry
(ICP-MS), often incorporating a laser (LA-ICP-MS) to ablate, or vaporize, an
404 Jeffrey R. Ferguson
extremely small ponion of the anifact for analysis in the ICP-MS. LA-ICP-MS
can provide data on a wide range of elemental compositions, comparable to NAA,
but can suffer from short and long term difficulties with analytical accuracy due
to difficulties with consistent calibration (Speakman el al. 2007). LA-ICP-MS is
often louted as a non-destructive analytical method, but almost every instrument
in use today requires the removal of a portion of a sample in order to fit inside the
laser ablation chamber.
Techniques with limited application to the compositional analysis of obsidian
include proton induced X-ray emission (PIXE: Pollard el al. 2007; Bellot-Gurlct
el al. 2008) and electron microprobe analysis (EPMA; Merrick and Brown 1984;
Pollard el al. 2007). These techniques often involve the same basic method, .,
XRF, but they vary in the means of removing the electron from the inner shell of
the atom in order to create the emitted X-ray as a result of outer shell electron~
dropping down to fill the void. Many of the same issues discussed here for XRF
apply to these techniques as well. Additional analytical techniques include, but are
not limited to: natural radioactivity (Leach el al. 1978). obsidian density (Reeves
and Armitage 1973), fission track analysis (Bigazzi el al. 1998), and numerous
chemical methods (Tykot 2004).
XRF offers the potential for totally non-destructive analysis, ponable and
handheld operation , minimal or no sample preparation. rapid results. widely
available analytical instrumentation, low cost. and quantitative analysi, of some
of the most important elements used for discriminating obsidian sources and
providing provenance data. All of the other techniques described above requ ire
large stationary equipment - ranging from the size of a chest freezer to a nuclear
reactor - and require the sample to fil the space requirements of an instrument
sample chamber rather than introducing a smal l instrument to the object of interest
as is possible with handheld XRF. Most of the commercial obsidian provenance
laboratories employ non-ponable energy-dispersive (ED-XRF) as their main
instruments , yet portable instruments can provide high quality data in the lab while
also aJlowing in-field analysis including obsidian artifacts in countries or curation
facilities that will not allow their removal of artifacts. While the decreasing co ... 1.
increasing availability, and portability have all contributed to making XRF a more
attractive analytical technique to a broader spectrum of researchers across man)
disciplines. the fundamental physical principles behind the technique and the need
to understand them have not changed. The following section describe' some of
the most imponant issues in handheld XRF analysis of obsidian, with panicular
attention to some of the issues encountered in the research at the Archaeometr)
Laboratory at the University of Missouri Research Reactor (MURR).
X-ray fluorescence of obsidian 405
XRF methodology
One of the firM considerations in XRF analysis is whether or not quantitative
analysis is necessary, or even possible, with the sample set under investigation.
While qualitative comparative spectra] overlay can be used to assign individual
obsidian artifacts to particular sources, it is not the most efficient means of
processing large numbers of samples and can lead to difficulties in discriminating
geologic sources with similar elemental ratios (i.e. the shape of the spectra are
similar but the magnitudes differ). Without the ability to extract quantitative
concentrations for certain clements. it is not possible to compare new analyses with
previously published compositional data, limiting comparisons to those samples
previously analyzed using the same equipment. Thus, quantitative analysis is
necessary in obsidian studies. In contrast, qualitative analysis can be highly useful
in. for example. attempting to determine if some large pit features were used to
smelt iron or to fire limestone to produce lime. In this case qualitative spectra]
overlay comparing areas inside the features to areas of the surrounding natural soil
;howed significantly higher calcium and strontium levels in deposits at the bottom
orthe features compared to outside the features and no difference in the amount of
iron. Additional tests may be warranted. but it seems likely thatlhe features were
used as lime ki lns strict ly based on spectral comparisons (Ferguson 20 10).
The main reason for not perfonnjng quantitative analysis on the pit features
described above is I have not yet developed a comprehensive calibration for the
analysis of soils. and the difficulties in properly quantifying such a heterogeneous
material in the field make the process difficult at best. The "point and shoot" claims
by some instrument manufacturers have led to misleading data being published
without a proper matrix-specific calibration (i.e. Morgenstein and Redmount 2005;
Goren el 01. 20 I I). A single calibration is simply incapable of such a broad range
of different matrices, and matrix-specific calibrations are required for accurate
quantitative analysis due to a number of factors, including absorption propenies of
the matrix, secondary fluorescence, and different concentration ranges for specific elements.
Developing accurate quantitative approaches for obsidian analysis allows
the rapid (although not fool-proof) assignment of geologic sources to artifacts
Using a variety of multivariate statistical methods, something not possible with
qualitative spectral overlay. Proper calibration is the key to efficient and accurate
SOurce assignment. MURR has a long history of obsidian analysis by NAA and
consequently a massive source library including almost every known obsidian
SOurce in the world. The MURR handheld XRF generally uses a calibration based
...
II ~ III I" '
1'1'
II II
I
I
I, ".:
406 Jeffrey R. Fergu,on
on compatibi lity with NAA data in order to maximize the potential to find sources
not anticipated in a panicular study, and thus not yet fully characterized by XRF.
The most common deviations between M URR XRF data and that produced at other
labs using only XRF are in the iron and zirconium concentrations. Unfortunately
NAA is not as capable of producing rughly accurate zirconium concentration, due
to the production of both "Zr and "Zr during "'u fission. It is possible to correct
for the uranium interference, but the correction has a large uncertainty (Glascock
el al. 1986).
The differences between the iron concentrations reported by the Berkeley
XRF Laboratory and MURR are more complicated to explain, but Mike Glascock
(pers. com. 201 I) provides a possible explanation. The published values for iron
in many USGS standards used by some XRF labs for calibration arc generally
listed as total iron. USGS standards, as a historical convention resulting from the
use of gravometric techniques, report concentrations for multiple iron oxides. It is
possible to convert the weight percent into ppm concentrations, but there is often
the false assumption that all of the iron exists as the single oxide reported. Thi,
can result in inaccurate iron concentrations entered into the calibration standards.
NAA iron concentrations are based on a single 1ST standard (fly ash) and the
concentration in the standard is not reported as an oxide, but as total iron. AA
calibration is linear and only requires a single standard for iron. The MURR
XRF calibration is based on the data derived from AA measurements of the
40 calibration standards and thus is not susceptible to the oxide weight percent
conversion problem.
Obsidian XRF calibration
Proper calibration is not a simple process. It is important to have a broad range
of concentrations in one's standards that cover the full range of concentrations
anticipated in the unknown samples. This was made clear with the initial analysis
of obsidians from the Central Rift Valley of Kenya. The initial calibration included
samples from 40 well-known obsidian sources in the Americas]. For these
obsidians, iron rarely exceeds 6-7 percent. zirconium is rarely above 1,000 ppm,
and niobium is rarely over 2oo ppm. It was clear from the spectral overlay that
many of the Kenyan samples had abnormally high concentrations of these three
elements , yet the concentrations calculated by the calibration produced lower
concentrations as the peak areas increased. Bruce Kaiser (personal communication
2oo9) has termed this phenomenon the "wet noodle effect" in which peak areas for
panicular elements that exceed the range of the standards used in the calibration
X-ray fluorescence of obsidian 407
can trail off in extreme directions. sometimes leading to inverse relationships
between peak areas and concentrations.
The data from the 2009 Kenya source samples (n:14I , collected by Stanley
Ambrose) serve as an example of the need to include the complete range of
variability for each element. Figure 12.1 is a plot of the NAA-determined
concentrations for iron (X -axis) and the XRF-detcrmined concentrations for
iron (Y-axis) for both the original and extended calibrations. As noted above, the
original calibration produced decreasing iron concentration for larger peaks once
the concentrations exceeded the range in the calibration. The problem was corrected
by including a couple Kenya obsidian samples with high iron , zirconium, and
niobium concentrations. previously analyzed by NAA. The corrected calibration
produces XRF results comparable to NAA .
15000 +--~-__ -~-~-_--~ lSOOO 25000 lSOOO 4SOOO SSOOO 65000 75000
Iron (ppm) from NM
Figure 12.1. Comparison of the initial and extended MURR XRF calibrations for iron. The lower curve (triangles) are the inItial concentrations reported for a group of 141 source samples from Kenya. The upper linear spread ("x' symbols) are the same analyses recalculated using the extended calibration. The Y-axis values are the XRF concentratIons for Iron and the X-axis values are concentrations determined by NAA.
The current MURR calibration uses either 40 or 50 different obsidian secondary
standards previously characterized by NAA at MURR (the main calibration uses
only 40 standards, and the calibration for high iron , zirconium, and/or niobium
samples uses an additional 10 standards, this is explained in greater detail below).
The calibration produces concentrations for 19 elements (Na. AI, K, Ba, Ti, Mn.
I'
408 Jeffrey R. Ferguson
Fe, Co, Ni, Cu, Zn, Ga, Pb, Th, Rb, Sf. Y, Zr, and Nb). Although not a ll these
elements are relevant to the overall composition of obsidian, they are included to
improve the overa ll calibration . Only six elements are routinely used in obsidian
provenance studies. These six elements and their concentration ranges are listed in
Table 12.1. It is necessary to not only bracket the expected range of concentrations,
but to have uncorrelated variation between the elements in the suite of standards.
This multi-elemental variability helps the ca~bration algorithms account for peak
interferences and secondary fluorescence (especially for the elements Rb- 'b
where the spacing of the peaks allows for the K~ peaks from a particular clement
to interfere with the Ka peaks of the element two units greater in atomic number).
It is also important to test calibrations using samples of known composition that
are not included in the calibralion (Mauser 2006).
Mmimum (ppm) Maximum (ppm)
Fe 3554 69319
Rb II 48 1
Sf 0 488
Y 12 662
Zr 61 3455
Nb 4 372
Table 12.1: Minimum and maximum concentrations for standards used in extended MURR XRF calibration.
Unfortunately, concentrations for yttrium and niobium are not avai lable using
NAA. and AA has a relatively high detection limit (approximately 50ppm) for
strontium. Standard values for yttrium, niobium. and low strontium concentrations
were averaged from the ED-XRF analysis of the same standards at twO other
obsidian studies laboratories: Northwest Research Obsidian Studies LaboralOry
in Corvallis Oregon , and the Berkeley Archaeological XRF Lab althe Unive"ity
of California, Berkeley. Methods for analysis and calibration used by these
laboratories arc available through their respective web sites (Northwest Rc!)carch
Obsidian Studies Laboratory 20 II ; Shackley 20 II a).
Shackley (2010) recently commented on the need to use certified international
standards in calibration. This is certainly an ideal goal that would help lO increase
inter-laboratory compatibility, but there are a number of problems with this
approach. First of all, most certified standards are available only as powders.
While the powders can be pressed into pellets there is the potential for analytical
II X-ray Ouorescence of obsidian 409 I'!
~I differences when analyzing solid glass samples using a calibration based on !.
pressed powders. Empirical tests of this powder versus solid sample difference
have shown minimal , if any. effecl. Shackley (201Ib) has analyzed both solid
and pressed pellets from the single boulder used by USGS to create the RGM-
I standard and found very little difference. A recent test at MURR examined a
polished solid piece of obsidian from the Whitewater Ridge source in Oregon and
compared the spectra to loose powder and a pressed pellet from the same rock.
There was a reduction in the backscatter for the powder and pellet samples , but
there was no apparent difference in the peaks of the main elements of interest (Rb-
b). Others have observed morc significant differences as a function of grain size
of powdered samples. Liritzis and Zacharias (20 II) observed differences between
solid and powdered obsidians ranging from up to 5% for K. Ca, Sr, Zr, and Rb,
to 20 and 25% for Fe and Ti respectively. The different matrix effects of different
types of geologic materials (i.e. basalt versus rhyolite) may be more problematic
but are not yet fully studied.
Second, there are very few certifieq,&Iandards for obsidian or compositionally
similar materials . The inter-elemental interactions in XRF are incredibly complex,
and it is desirable to have a large number of calibration standards that not only
bracket the range of concentrations in the samples, but aJso include a large mix
of these elements. The goal of the MURR calibration is to use a large number of
secondary standards in the actual working range for obsidian rather than include a
number of certified standards with concentrations irrelevant for obsidian analysis.
It is critical to create a matrix-specific calibration. Finally. it is unlikely given
the compositional variations of obsidians that si mply using a series of common
standards will allow one to match source and artifact data collected at different
laboratories and using different instruments. Proper calibration using appropriate
standards can increase the probability of identifying possible sources for an
unassigned artifact in published data using other laboratories or techniques. but
assigning an artifact to a particular source unless the sample and source material
are analyzed on the same instruments using the same settings and calibration is
unwise. Some other analytical techniques. such as NAA. are much morc stable
due. in part , to linear standardizations and allow for calibration using a small
number of standards. These techniques are not susceptible to matrix effects,
and thus specific procedures to allow for inter-laboratory exchange of data are
warranted (Graham ef al. 1982; Glascock ef al. 1998). but XRF is based on
different excitation and absorption phenomena and requires a more conservative
approach. A recent inter-laboratory comparison of quantitative XRF of copper
alloys found that the accuracy of the data was a function of variables such as the
type of calibration used (fundamental parameters with and without standards and
II •
III
I"
Ii: I::
410 Jeffrey R. Ferguson
empirical calibrations). the number of standards used in the calibralion, bUl other
variables such as delector type and valid count rates, had Iiule apparent impact on
the accuracy (Heginbotham 20 I I) .
'"
12C
'0<
~
E Q. eo Q. ~
.D Z ou
Mule Moun"lns
,,>U 200 jW J~"
Rb (ppm'
Figure 12.2: Bivariate plot of rubidium and niobium for source samples from Mule Creek. The open symbols are source samples analyzed at MURR and the solid symbols are source samples analyzec at the Berkeley XRF Lab. The specific samples are not the same for each lab. Ellipses represent 90 percent confidence Intervals for membership in the group.
The goal of archaeological obsidian provenance stud ies is to match obsidian
artifacts to specific geologic sources of obsidian . While every effort should be
made to detcrmjne accurate elemental concentrations by using appropriate
standards in a robust calibration , consistent separation of sources is the ultimate
goal. Shackley (2005:7) correctly described the balance between precision and the
ultimate goals of obsidian provenance research: "[sjome physicists and chemists in
the past have made what 1 consider the mjstake of focusing on precision versus the
archaeological accuracy we seek in source provenance studies, and archaeologistS
have been guilty of the opposite - ignoring precision and accuracy in provenance
studies by wholly trusting the analyst because they don ' t understand the process."
In the case of XRF at MURR, the calibration is more valuable the closer it matches
the NAA data to facilitate the identification of artifacts from sources not yet
analyzed by XRF. The differences between the MURR calibration and other major
X·ray fluorescence of obsidian 411
labs are relatively minor. For example, Table 12.2 provides a comparison belween
the Be rke le y Lab and M U RR fo r som e o f the m ajor source s in N e w M ex ico. While
the re a re s uffi c ie nt differences to complic ate inte r- labo ratory s haring o f data , both
labs produce quantitalive elemental concentration data capable of discriminating
most sources, including two compositionally-similar sources from Mule Creek
( Figure 12.2).
Fe Rb Sr y Zr Nb n mean s.d mean s.d. mean s.d. mean s.d. mean !..d. mean
EI Rechuclos Bcrkelcy 15 6676 296 152 6 9 3 23 I 77 3 47 Rh)olile I MURR It 3803 114 138 7 I I I 22 2 7 1 4 39
rObSidian Ridge Berkeley 20' 9735 666 207 II 5 3 63 3 183 7 98
MURR 10 7289 185 186 5 6 I 54 4 168 7 89
f(-cITO del Medio Berkeley 2' 103 15 18 17 160 9 10 I 43 I 172 4 54
MURR 22 7053 396 143.6 7 10 I 4 1 3 167 9 50
Bcar Springs Berkeley 24 ' 6593 43 1 116 5 43 4 2 1 2 108 4 53 Peak
M URR 7 4877 206 110 4 51 5 20 I 108 5 47
~ran" Ridge Berkeley 10 8253 474 564 34 4 I 76 2 11 8 4 196
M UR R 10 5016 256 47 1 22 II I 67 5 10 1 5 171
lHomce ~ l e,. Berkeley 12 9568 321 530 17 4 I 87 3 142 4 236
M URR 13 5945 4 12 422 21 13 I 73 5 11 7 5 205
~I ulc Creek Berkeley 21 9619 385 246 15 16 3 42 4 122 7 27 i(AnlelOpe Creek)
I- MURR 54 6441 380 206 10 2 1 3 37 4 106 5 23
IMuie Creck (N.Sawmill )
Berkelcy 34 8423 564 41 1 15 6 2 7 1 2 118 7 120
L MURR 26 5007 300 358 13 12 2 63 4 98 7 101
Mule Creek Berkeley 15 ~M u le
7936 297 184 7 10 2 25 2 120 5 32
Mounta in<;)
L MURR 13 4861 22 1 163 6 13 2 24 2 11 6 5 28
Gwynn Canyon Berkeley 20 85 12 539 226 \3 19 2 3 1 2 156 7 22
I---- MURR 12 6079 266 205 5 21 2 30 2 156 12 22
melope Well ... Berkeley 25 22521 19 14 348 22 7 I 128 6 1255 47 97
M URR 3 18 178 900 271 8 8 I 115 6 1069 50 82
Fewer samples with Iron concentrations.
Table 12.2:
Mean and standard deviations (in ppm by XRF) for some of the major sources In New Mexico for both the Berkeley Lab and MURR. Berkeley Lab concentrations taken from the laboratory
webSite (http://www.swxrfiab.netlswobsrcs.htm).
!I.d.
2
4
4
3
I
3
5
3
7
10
7
12
3
2
4
5
2
2
3
2
5
9
41 2 Jeffrey R. Ferguson
Maximizing the potential of XRF for obsidian
The goal of XRF analysis is to maximize the nuorcscence of, and thus sensitivity
lO, clements best able to discriminate geologic sources of obsidian. The mOst
useful group of elements that occur in sufficient concentrations for accurate
quantification by EO-XRF are the elements Rb through Nb (atomic numbe"
37-41). These clements generally occur in concentrations of tens to hundreds of
parts per million and the K01
nuorescence energies for each of these elements
are roughly in the middle of the spectrum of energies produced by most handheld
XRF instruments operating at 40kV.
The probability of nuorescing a particular atom is related to the cxcitarion energy of the atom relative to the energy of the incident X-ray. The probability of
fluorescence is greatest when using incident X-rays with energies just above the
exc itation energy of the atom in the sample. For example. iron has a Ka , of 6.4
keY. and there is no chance of K-shell electron nuorescence from incoming X-rays
with energies less than or equal to 6.4 keY, while X-rays lightly higher than 6.4
have the highest probability of fluorescence. The probability of fluorescence
decreases exponentially with higher energy. Thus, the goal should be to produce an
abundance of X-rays with energies just above the absorption edge of the elements
of greatest interest. The elements Rb-Nb have Ka, energies from 13.40 - 16.62
keY. An ideal incoming array would contain X-rays primarily above 17 keY and
tapering off at much higher energies.
Bruce Kaiser (with Bruker Scientific) and R. Jeffrey Speakman (wi th the
Museum Conservation Institute at the Smithsonian) have developed a filter that
acts like a secondary target to absorb the incoming X-rays below about 18 keY
in order to both reduce the background of scattering X-rays at the same energy as
Rb-Nb and to remove thc X-rays of low energy that do not significantly contribute
to the fluorescence of Rb-Nb. The filter consists of thin foils of copper ( 150 Iill')'
aluminum (50 1lJ11) and titanium (300 11m). The copper absorbs the X-ray'
below 18 keY, but the fluorescence of copper produces a large spike of X-ray'
at approximately 8-9 keY. The titanium layer absorbs the 8 .05 keY X-rays. but
produces another peak between 4 .5 and 5 keY as a result of titanium fluorescence.
The aluminum layer absorbs the titanium peak, and the resulting aluminum
fluorescence does not have sufficient energy to impact sample analysis. Figure
12.3 shows the difference in background and increased sensitivity for the same
obsidian sample analyzed under the same conditions with and without the filter.
-
X-ray fluorescence of obsidian 413
16000
14000
4j12000 c c 210000 v
~8000 tl § 6000 0
~ 4000 15 ~ 2000
0
0 S 10 IS 20 2S 30 3S 40 KeV
Figure 12.3: S~tra of the same large flake of obsidian showing the difference in background when using the coppeHitanium-aluminum filter (solid line) versus without the filter (dashed line). The filter reduces the total number ofX:7aYs hitting the sample so the filtered data have been multiplied eight times. Note the minimal background under the peaks between 13 and 17 keV that correspond to the elements rubidium to niobium.
Normalization and small/thin sample analysis
Samples selected for XRF analysis should ideally be infinitely thick fortheelements
of interest. Infinite thickness is defined as the thickness at which additional sample
thickness does not result in additional fluorescent X-rays . lnfinite thickness is
different for each element in direct correlation with the excitation energy and
varies between matrices. although sample matrix differences are not addressed
here because obsidians are relatively similar in their overall high-silica matrix.
For example, 99 percent of the iron fluorescent X-rays (6.4 keY) are from up to
0.1 mm from the surface of a pure silica matrix (obsidian is close to a pure silica
matrix). The X-rays with the highest probability of exciting iron have relatively
low energy and thus arc not capable of penetrating deep into a sample, and even if
they did fluoresce an iron atom deep in the sample, it is unlikely for the low energy
fluorescent X-ray to make it back through a thick sample and into the detector.
While iron peaks in a spectrum result primarily from nuorescence at the surface
of the sample. under the same conditions, zirconium Ka X-rays (15.8 keY) escape
from up to 2.2 mm deep, and barium Ka X -rays (32 .2 keY) from up to 17 mOl
deep.
414 Jeffrey R. Ferguson
The differences in infinite thicknesses are critical 10 an understanding of the
difficulties in analyzing thin samples (I will expand on the analysis of thin samples
in the following seclion), but normalization can help account for some of the issues
related to sample thickness variability. Normalization was initia lly developed as
a means to neutralize the effects of small fluctuations in the output of the X-ray
sources. The method involves selecting a region of the spectrum that does not have
peaks derived from elements of interest in the sample and that directly reneel the
amount and energy of X-rays hitting the sample. The Bruker Tracer lll-V PXRF
in use at MURR uses a rhodium source and thus produces a characteristic peak of
X-rays for Rh Ka at approximately 20.2 keY due to Raleigh scatter of the source
X-rays on the sample. This interaction of Rh X-rays with the sample produces
both elastic and inelastic scatter resulting in both the Raleigh (elastic scalter at -
20.2 keY as mentioned above) and Compton (inelastic scalter at approx imately
18.5-19.5 keY) peaks being produced in the spectrum based on infinitely thick
samples. In order to gel quantitative concentrations for a batch of sarnpl e~ it is
necessary to normalize the peak areas in the rhodium Compton peak and apply the
same correct ion to the other peaks in the spectra. This minimizes the differences
in the incoming X-ray profile.
While the Compton peak serves well for normalization in most obsidians. there
are certain sources with elevated niobium concentrations that have a sufficiently
large niobium K~ peak to interfere with the rhodium Compton peak area . In
these cases, the calibration interprets the additional peak area from niobium as
increased X-ray output requiring comparable reduction in all peaks in the spectra.
Obsidians with such elevated niobium concentrations are rare worldwide. but they
are common for the sources in the Rift Valley of Kenya. For Kenyan obsidian
a slightly modified version of the calibration that normalizes to a portion of the
backscatter in a region above obvious elemental interferences is used. This avoidS
the niobium effects on normalization. but it moves the normalization region
even farther from the peaks for the main elements of interest (Rb-Nb) making
correction for sample size even more difficult. XRF calibration requ ires a series
of compromises to arrive at the best possible data. There is no obvious one-size-
fits-all answer.
The same normalization procedure used to account for tube fluctuations can
also be used , with somewhat less success, to account for variabi lity in sample size.
particularly thickness. Many applications of XRF involve thin samples. such as
the analysis of thin films and filters. This type of thin sample is relatively easy to
correct for if the samples have consistent thicknesses and it is possible lO creat~ comparable calibration standards (Markowicz 2008; Piorek 2008). Although It
is possible to create thin slices of obsidian. artifacts have complex morphology
4
X·ray fluorescence of obsidian 415
and thicknesses are not only inconsistent between samples, but across a single
,ample. By understanding the physics behind the fluorescence of thin samples as
well as the potential errors inherent in normalization it is possible to confidently
determine the source of obsid ian artifacts that are on ly a couple millimeters in
diameter and less than 05mm thick. Figure 12.4 shows two overlaid spectra of a
largelthick flake and a smallithin flake fro m the same obsidian cobble.
700
600
~ 500 c ~
-'= ~ 400 [ ~300
'\' .. ~ 0 ~ 200 , g • , >-'- 100
,
•
. ! ~j
NI ., ~::i
0
0 5 10 15 20 25 30 35 40 KeV
~ igure 12.4: Comparison of large (dashed line) and small (solid line) flakes from the same cobble. The large flake covered the entire beam with a thickness of 7mm. and the small flake measured 4 by 3mm and O.4mm thick.
The current solution to small sample analysis at MURR involves a number of strategies. First of all , we use an instrument with a small beam cross-section .
Figure 125 is a briefly exposed dental X-ray of the Bruker Tracer III-V PXRF
compared to that of the benchtop ElvaX. The Bruker beam focu ses on an area of
approxi mately 2 x 3 mm . Although a narrow beam is more susceptible to var iat ion
in heterogenous samples. obsidian is usually homogenous enough to not create any
problems. The beam in the Tracer III-V is large compared to that of a micro-XRF
and small beam use has great potential for the analysis of obsidian microdebitage.
The small beam improves the analysis of samples with a small diameter by simply
Covering more of the beam. If samples were infirtitely thick, the small sample
diameter would easily be accounted for by the typical normalization process and
no data shift should occur.
416 Jeffrey R Ferguson
I I Smm
Figure 12.5: Dental X-ray film brieOy exposed to the Elva-X (left) and Tracer III-V (right) beams. Note the much smaller beam area of the Tracer.
The real problems arise with the analysis of samples of less than infinite thickness.
For the purposes of this discussion, consider an example in which we are only
concerned with obtaining concentration data for iron and niobium. The niobium
Ka, peak is between 16.3 and 16.9 keY, and this is relatively close to the rhodium
Compton peak used for normalization that is between 18.5 and 19.5 keY. The
closer the two peaks are in energy, the closer they are to having the same infinite
thickness value, thus the two peaks are absorbed at simi lar rates with increases in
sample thickness. When the calibration normalizes to the rhodium Compton peak,
and the same correction factor is applied to the niobium peak the quantitative
results are reasonable. But , iron has an infinite thickness in obsidian measured in
microns , and thus the size of the iron peak does not change as a function of sample
thickness until the samples are extremely thin. This may not seem like a problem,
but the normalization applies the same correction to all peak areas. whether it
is needed or not. Therefore, in a sample thin enough to generale only half the
rhodium Compton peak area. the niobium peak area (and thus the calcu lated
concentration) will be correctly increased by a factor of two, but the iron peak arca
will be significantly over-corrected. The primary elements of interest in obsidian
characterization, Rb-Nb, are all relatively close to the rhodium Compton peak. and
the normalization does a good job of correcting for sample size although there is an
increasing overcorrection moving down in energy. This overcorrection is extreme
for low-energy elements like iron. When analyzing assemblages that include small
samples, it is best to discount data from all elements with fluorescence energie~
below that of rubidium.
It should now be clear why the calibration discussed previously that account,
for the elevated niobium levels in some East African obsidians is an example of
X~ray fluorescence of obsidian 41 7
a difficult compromise. The East African calibration sacrifices accuracy of small
sample analysis because of the higher energy of the normalization region in order
to avoid the negative effects of high niobium concentrations. In other areas of the
world. namely the western United States, the niobium levels are low enough that
sample size is more important and one should use the calibration that normalizes
to the rhodium Compton region.
The combination of the variably erroneous normalization correction in thin
samples and the overall reduced counts for small/ thin samples causes a slight
shift in the data the correlates with sample size. It can make source assignment
by whatever means (i.e. bivariate plots, cluster analysis, Mahalanobis distance
projection) more difficult. Bivariate plots can show the correlation in the scattered
data that should co-vary with the correct source data across all plots for clements
Rb-Nb. It is very important to confirm the source assignment with direct spectral
overlay, keeping in mind the expected shifts resulting from small samples.
A number of recenr-projects on obsidian from a number of si tes in central and
southern California has recently been conducted at MURR to demonmate that
XRF is effective for small sample source assignment. Of the more than 1,000
artifacts analyzed by XRF, 40 of the smallest and most difficult to assign samples
were selected for additional analysis by short-irradiation NAA . The advantages of
the short NAA are that the samples become only sl ightly radioactive and can be
safely handled after on ly a brief decay period , the technique is a bulk technique that
is not subjected to variations in sample morphology because the samplc rotated ,
and it can produce precise data for about six elements on samples of only a few
milligrams. Of the forty samples, XRF source ass ignment was confirmed for 38
of them . One sample was not assigned to any source by XRF but was assigned by
NAA, and one sample was incorrectly assigned by XRF to a source with similar
composition. This error rate of five percent is higher than one would like, but it
occurs on the smallest and most challenging samples. The error rate for the entire
study, including the larger artifacts, would likely be much lower. Similar tests have
been done with smaller assemblages from the American Southwcst with similarly
encouraging resu lts.
The current approach to small sample analysis by XRF is not yet an ideal
solution. Attempts have been made to create an element-specific correction
factor that uses the rhodium Compton peak as an indicator of sample mass. One
possibil ity is to lise a two-element standard behind the artifact that provides a
good approximation of the thickness of the sample in the beam. The standard
would contain sufficient concentrations of two elements, one with a peak below
the elements of interest and one above. In some preliminary tests using a pressed
pellet containing gallium and indium in which the lower energy element (in this
418 Jeffrey R. Ferguson
case gallium) would have a peak area reduced in relation to the higher energy
e lement (indium) in a ratio reOecting sample mass. This method has some promise.
but it is time consuming to keep the standard in proper position and this would
also greatly increase the time required to interpret the final data. It is important to
verify source assignments. particularly with small samples. using other analytical
techniques such as NAA or LA-ICP-MS, but the ultimate solution to small artifact
analysis may involve micro-XRF.
Conclusions
XRF has held a prominent position in the provenance study of archaeological
obsidian for decades, and all indications are that it wil l continue to be a major
analytical technique. Handheld XRF instruments provide some logistical
advantages over lab-based instruments, but aside from a few, relatively minor,
limjtations the two groups of instruments pose the same analytical potential as
well as limitations. Handheld XRF has been the unfair subject of recent skepticism
in the literature (i.e. Shackley 20 I 0). The problems arise not from the portability of
a particular instrument. but from the lower cost of most portable instruments that
places the technology in the hands of researchers with insufficient understanding
of the physics, calibration methods , and analytical limitations. This chapter begi ns
to address some of these problems .
Notes
I . A completely new set of calibration Mandards is in preparation that will cover the kno\\n concentration ranges in obsidian for the elements iron , rubidium , strontium , yttrium. zirconium, niobium and barium. This new calibration set wi ll include 40 obsidian samples from around the world. The concentrations wi ll be derived from a combination NAA, microwave digestion ICP-MS. and XRF analysis of the individual rocks. A publication detailing the concentrations of the originaJ set of caJibrat ion standards in currently in preparation , and considering it will soon be obsolete upon the completion of the new set of calibration standards the values are not reported here.
X-ray fiuorescence of obsidian 41 9
References
Ambroz. Jessica A .. Michael D. Glascock, and Craig E. Skinner. 2001. "Chemical Differenliation of Obsidian within the Glass Buttes Complex, Oregon:' JOII",al of Archaeological Science Vol. 28 (7):741-746.
Bellot-Gurlet. Ludovic. Olivier Dorighel, and Gerard Poupeau. 2008. "Obsidian provemmce studies in Colombia and Ecuador: obsid ian sources revisited." Joumal of Archaeological Science Vol. 35 (2):272-289.
Bigazzi, G" G . Poupeau. Z. Yeg.ingil, and L. Bellot-Gurlct. 1998. "Provenance Studies of Obsidian Artefacts in Anatolia Using Fission-Track Dating Method: An Overview." In L'obsidien"e au Proche et Moyen Oriem. edited by M.-C. Cauvin. A. Gourgaud. B. Gratulc. N. Arnaud. G . Poupeau. J.-L. Poidevin and C. Chataigner, 70-89. Oxford: BAR International Series.
Braswell. Geoffrey E., John E. C lark, Kazuo Aoyama, Heather McKillop, and Michael D. Glascock. 2000. "Determining the Geochemica1 Provenance of Obsidian Anifacts from the Maya Region: A test of the Efficacy of Visual Sourcing." Latin American Allriqlliry Vol. 11 (3):269-282.
Braswell. Geoffrey E .. and Micfnle1 D. Glascock. 1998. " Interprcting Intrasource Variation in the Composition of Obsidian: The Geoarchaeology of San Manin Jilotepeque, Guatemala ." Latift American Antiquity Vol. 9 (4):353-369.
Carter, Tristan, G6rard Poupcau. C61ine Bressy, and Nicholas J. G . Pearce. 2006. "A New Programme of Obsidian Characterization al v31alhoyUk, Turkey ." JOIln/al oj Arclweologiclli Sciellce Vol. 33 (7):893-909.
Duff. Andrew Ian Lyman. 1999. RegiolJal lllleraction and the Transjormation oj Westen! Pueblo Identities, A.D. 1275-1400. PhD. Department of Anthropology, Arizona State Uni versity, Phoenix.
Eerkens, Jelmer W., Jeffrey R. Ferguson. Michael D. Glascock. Craig E. Skinner, and Sharon A. Waechter. 2007. "Reduction Strategies and Geochemical Characterization of Lithic Assemblages: A Comparison of Three Case Studies from Western North America." American Amiquity Vol. 72 (3):585-597.
Ferguson . Jeffrey R. 2010. In -Field Qualitative X-Ray Fluorescence of Kiln Features at LA 155669. Capitan Mountains , New Mexico. Report on file at MURR.
Ferguson, Jeffrey R .. Stephen H. Lckson, Margaret elson, Toni S. Laumbach. and Karl W. Laumbach. 20 10. Characterization of Carbon-Painted Ceramics from Southwestern and South Centra l New Mexico. Paper presented at the Society for American Archaeology Meeting, at SI. Louis, Missouri .
Ferguson. Jeffrey R .. and Craig E. Skinner. 2006. Geochemical Evidence of Mesoamerican Prismatic Blades in the United States. In Societ),jor American Arc/weolog)' Meeting . San Juan , Puerto Rico.
Friedman , I. and F. W. Trembour. 1978. "Obsidian : The Dating Stone." American Scietllisl Vol. 66 (1):44-45.
---. 1983. "Obsidian Hydration Dating Update." American Antiquity Vol. 48 (3):544-547 .
Glascock, M . 0 .. H. Neff. K. S. Stryker.nndT. N. Johnson. 1994. "Sourcing Archaeological Obsidian by an Abbreviated NAA Procedure." Journal oj RadiolUwl),lical and Nllclear Chemisrry Vol. 180 (1 ):29-35.
I I II
420 Jeffrey R. Ferguson
Glascock. Michael D. 2002. Ge()('hemical Evidellcefor umg Disumce £rc/ulIIge . Westport. Connecticut: Bergin and Garvey.
Glascock. Michael D., Geoffrey E. Braswell. and Robert H. Cobean. 1998. "A Systematic Approach to Obsidian Source Characterization," In Archaeological Obsidian Studies: Method mId Theory, edited by M . Steven Shackley. 15·65. New York: Plenum Press.
Glascock. Michael D .• Raymond Knuselrnan . and Daniel Wolfman . 1999. "Imrasourcc Chemical Differentiation of Obsidian in the Jemez Mountains and Taos Plateau. New Mexico:' JOllrnal of ArchaeologiclIl Science Vol. 26 (8):86 1·868.
Glascock. Michael D .. Robert J. Speakman. and Rachel S. Popelka-FiJcoff. 2007. Archaeological Chemistry: Analytical Techniques alld Archaeological lllferprewtioll. Vol. 968. Washington. DC: American Chem ica l Society.
Gordus. Adon A. , Gary A . Wright , and James B. Griffin . 1968. "Obsidian Characterized by Neutron·Activation Analysis ," Science Vol. 161 (3839):382-384.
Goren. Yuval. Hans Mommsen, and Jorg Klinge r. 20 11." on-Destructive Provenance Study of Cuneifonn Tablets Using Portable X-Ray F1uorescence," Joumal oj Archaeological Science Vol. 38 (3):684·696.
Graham , Christoper C .. Michael D. G lascock. James J . Carni . James R. Vogt. and Thomas G. Spalding. 1982. "Detemlination of Elements in National Bureau of Standards' Geological standard Reference Materia ls by eulron Activation Analysis," Allalytiwl Chemisrry Vol . 54 (9): 1623- 1627.
Heginbotham. Arlen , A. Bezur. M . Bouchard . J . M. Davis, K . Ere min. J. H. Frontz. L. Glinsman, L. Hayek, D. Hook. V. Kantare lou. A . G. Karydas. L. Lee. J . Mass, C. Matsen, B. McCarthy. M . McGath , A. Shugart Sirois J. D. Smith. and R. J. Spea!..man. 201 1. "An Evaluation of Inte r· L~lboratory Reproducibility for Quantitative XRF of Historic Copper Alloys." In Mewl 2010: IlIlernmiollal Conjerence 011 Melal Conservation, edited by P. Mardikian. C. Chemello. C . Watters and P. Hull. 244·255. C le mson. South Carolina: C le mson Univers ity.
Hester, T. R .. F. Asaro , F. Stross, H. Michae l. A. C. Kerr, and P. Headrick. 1991 . "An Overview of the Resu lts of the Texas Obsidian Project ." La TIerra Vol. 18 ( I ):4·7.
Hughes. Richard E . 1984 . Obsidi(1II Studies ill 'he Grear Basin. Contributi on"i of the Unive rsity of Californ ia Archaeological Research Facility No. 45 . Berkeley' Archaeological Research Facility. University of Californ ia .
---. J989 . Current Direclimls iff California Obsidian SllIdies. Contribution~ of the University of California Archaeological Research Facility No. 48. Berkeley. Archaeologica l Research Facility. Universi ty of California.
Leach. B. Foss, Stanley Warre n. and Barry Fankhauser. 1978. "Obsidian from the Far North of New Zealand: A Method of Sourcing Based on Natural Radioactive Emission." New Zealand Journal oj Science Vol . 2 1: 123- 128.
Lirill.i s. Joannis. and Nikolaos Zacharias. 201 1. " Portable XRF aof ArchlJeological
Artifacts: Current Research, Potent ials and Limitations." In X~Rav Fillorescell('{' Speclrometry (XRF) if! Geoarc/Illeology, edi ted by M . Steven Sha~klcy, 109-142. New York: Springer.
Mauser, KarJ -Eugen. 2006. "Standardless Methods. In Handbook of Practical X-Ray F1uorescence Analysis ." In . edited by B. Beckhoff. B. KanngieBer, N. Langhoff. R Wedell and H. Wolff, 384-393. Berlin: Springer.
X-ray fluorescence of obsIdian 421
Merrick. Harry V .. and Francis H. Brown. 1984 . "Rapid Chemical Characterization of Obsidian Artifacts by Electron MicroprobeAnalysis." Archaeometry Vol. 26 (2):230-236.
Morgcll!llein. Maury, and Carol A. Redmount. 2005. "Using Portable Energy Dispersive X-Ray Fluoresccnce (EDXRF) Analysis for On-Site Study of Ceramic Sherds at EI Hibeh. Egypt:' JOllrnal of Archaeological Sciellce Vol. 32 (1 1): 161 3- 1623.
Neff. Hector. 1998. "Units in Chemistry-Based Ceramic Provenance In"estigations:' In Unit Issues in Arclweology. edi ted by A. F. Ramenofsky and A. Sleffan. 115-127. Sail Lake City: University of Utah Press.
Northwest Research Obsidian Studies Laboratory. 2011. Introduction 10 Obsidian CJlllrllcter;:.,atiorl Studies 2011 [cited September 2011]. Available from http://www. obsidianlab.com/ info_xrf.hunl.
Origer. Thomas M. 1989. "Hydration Analysis of Obsidian Flakes Produced by Io;;hi During the Histori c Period." In Cllrrent Directions ill California Obsidian Studies. ed ited by Richard E. Hughes. 69-77. Contributions of the University of California Archaeological Research Facility No. 48. Berkeley: University of California.
Pollard, Mark. Catherine Bau. Ben Stern, and SUlanllc M. M. Young. 2007. Analytical Chemistry in Archaeology. Cambridge Manuals in Archaeology. Cambridge: Cambridge Universi ty Press.
Reeves. R. D .. and Gaye C. Armitage. 1973. "Density Measurements and Chemical Analysis in the Identification of New Zealand Archaeological Obsidians:' New Zealand JOllrlwl of Science Vol. 16:561-572.
Shack.ley. M. Steven. 1 998a. Archaeological Obsidian Studies: Method and Theory. New York: Plenum Press.
- --. 1998b. "Current Issues and Future Directions in Archaeological Volcanic Glass Studies: An Introduction." In Obsidian Geology and Archaeology ill the Nonh American Southwest . edited by M. Steven Shackley, 1-14. Tucson: University of Arizona Pre!o;s.
- - -. 2002. "More than Exchange: Pre-Ceramic through Ceramic Period Obsidian Studies in the Greater North American Southwe~I." In Geochemical Evidence for Lollg-Disumce £rc/umge. edited by Michael D. Glascock, 53-87. Westport. Connecticu t: Bergin and Garvey.
--- .2005. Obsidian: Geology lIIul Archaeology ill the Nonh American SOll1liwest. Tucson: University of Arizona Pres!o;.
-. 20 10. " Is There Reliability and Validity in Portale X-Ray Fluorescence Spectrometry (PXRF)T SAA Archaeological Record Vol. 10 (5): 17-20,44.
- - - .20 I I a. ThermoFislter ScielJlific Quant'X Allalsyis alld IIIstrllmelJllllioll 20 II rcited September 20 I J I. Available from http://www.swxrnab.nct//anIY!lis .htm .
. 20 11b. "An IlHroduction to X-Ray Fluorescence (XRF) Analysis in Archaeology," In X-Ray Fluorescence Spectrometry (XRF) ill Geoarchaeology. edited by M. Steven Shack.ley. 7-44. New York: Springer.
Speakman. Robert J.Michael D. Glascock. Robert H. Tykot.Christophe De'iCantes.Jennifer J. Thatcher. Craig E. Skinner. and Kyra M. Lienhop. 2007. "Selected Applications of Laser Ablation Inductively Coupled Plasma-Mass Spectrometry to Archaeological Research:' In Archaeological Chemistry: Analytical Techniquel atlll Archaeological Interpretation, edj ted by Michael D. Glascock, Robert J. Speakman and Rachel S. Popelka-Filcoff. 275-296. Washington. DC: American Chemical Society.
422 Jeffrey R. Ferguson
Taylor, R. E. 1976. Advances;'1 Obsidian Glass Swdies: Archae%giClli and Geochemical Perspectives. Park Ridge, New Jersey: Noyes Press.
Tykot. R. H. 2004. "Scientific Methods and Applications to Archaeological Provenance Studies." In Proceedbrgs of the Imemolioltal School oj Physics "Enrico Fermi" Course CLlV, edited by M. Martini. M. Milazzo and M. Piacentini. 407-432. Amsterdam: lOS Press.
Weigand, Phil C., Garman Harbottle, and Edward V. Sayre. 1977 . "Turquoise Sources and Source Analysis: Mesoamerica and the Southwestern U.S.A." In Exchange Systems i" Prehistory, edited by T. K . Earle and J. E. Ericson, 15-34. New York: Academic Pre!ls.
Weisler. Marshall I., and David A. Clague. 1998. "Characlcril3lion of Archaeological Volcanic Glass from Oceania: Ihe UlililY of Three Techniqucs ." In Archaeological Obsidian 5ll1dies: Method alld Theory, edilcd by M. Steven Shackley. 123- 128 . Ne" York: Plenum Press.