The Chemical Variability of some Common

Igneous Rocks
by R. W. LE MAITRE

Department of Geology, University of Melbourne,

Parkville, 3052, Victoria, Australia
(Received 23 June 1935; in revised form 25 February 1976)

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
ABSTRACT
The chemical variability of 39 groups of igneous rocks has been investigated using a data
base of 26,373 published analyses. Each group was extracted from the data base using the
name given in the reference from which the analysis was collected. The numerical data for
each group includes wt. per cent and molecular per cent means, standard deviations, C.I.P.W.
norms, Niggli values, the first three eigenvalues and eigenvectors together with a breakdown
of the geographical distribution of the analyses used. Nine variation diagrams are also
given for each group.

INTRODUCTION
I T is over twenty years since Nockolds (1954) published calculated average
compositions of a wide variety of igneous rocks. Prior to that the standard
reference for average compositions of igneous rocks had been Daly (1933).
Both of these authors were primarily concerned with delineating specific rock
types in terms of their chemistry but paid no attention to the variability within
each type. Most of their averages were derived from fewer than thirty individual
analyses and in many cases, from less than ten. Since then other averages pub-
lished have been more specific. For example, Green & Poldervaart (1955)
examined the composition of basaltic rocks from a geographical point of view,
while Manson (1967) and Chayes (1969) concerned themselves with the variation
in composition within basalts and andesites, respectively. Unfortunately many
users of averages seem to attach great importance to the individual values with-
out any particular concern for the amount of variability that such rock types
might be expected to show. It is hoped that this paper will make people more
aware of the variability within rock types.
The data used in this paper were extracted from a master file of 26,373 pub-
lished analyses of igneous rocks which forms part of the CLAIR data system
in use at the University of Melbourne and briefly described by Le Maitre (1973).
All the data reductions and computerized plots used are part of the CLAIR
system.

PRESENTATION OF DATA
All the diagrams and numerical data are presented in the same manner for
each rock type, in order to aid comparisons.
[Journal of Petrology, Vol. 17, Part 4, pp. 589-637, 1976]
590 R. W. LE MAITRE

DIAGRAMS
These have been produced on a computer line-printer by the CLAIR system
as density plots which simulate half-tone printing. As a number of points falling
within one cell of the plot increases so the intensity of the characters printed
increases. Up to three over-printings were necessary to produce the five scales
of intensity used. Although the program used has options for four types of
scaling (linear, sine, square root, logarithmic) the square root scale was found
in most cases to give the best visual expression of the distribution of points.

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
The linear scale under-emphasizes the peaks of the distribution while the
logarithmic scale over-emphasizes the background. Analyses were only plotted
in the triangular diagrams if the sum of the components was greater than or
equal to 10 per cent or 20 per cent as indicated by the 2 sign associated with
the diagrams. In each diagram the mean position of all those points included
in the diagram is plotted with a heavy circular symbol. It should be noted that
in many diagrams the position of the mean is not the same as the position of
the greatest density of points (the mode of statistics), thus emphasizing the
dangers of using averages in an indiscriminate manner.

Top-left (PCA diagram)

This is a two-dimensional representation of a three-dimensional projection
from nine-dimensional oxide space as described and used previously by Le
Maitre (1968) and Cundari & Le Maitre (1970). Although this diagram was
originally named PLVVD (principal latent vector variation diagram) the name
no longer appears to be appropriate because the term latent vector is now rarely
used compared to the alternative term eigenvector. It is, therefore, proposed
to rename the P L W D the PCA diagram after the statistical procedure, principal
component analysis, used to extract the eigenvectors. To aid comparison
between rock types, the eigenvectors used in each diagram are the set deter-
mined for 'igneous rocks' (No. 1). The scale is in oxide units i.e. 10 units corres-
ponds to a change of 10 per cent along an oxide axis.

Centre-left
This is a total alkali-silica diagram extended to cover the range of all rock
types. The short inclined line is the line of demarcation between the alkali
basalt and tholeiitic basalt fields of Macdonald & Katsura (1964). Although
originally popularized by Tilley (1950) to distinguish between alkali and tholei-
itic basalts the diagram has been used by other authors to cover a wide range
of rocks (Le Maitre, 1962; Baker, 1969).

Bottom-left
This is the QLM diagram as described by Burri (1964). In this diagram M
represents the olivine and ore, P the pyroxene, L the feldspathoid, F the feldspar
 CHEMICAL VARIABILITY OF COMMON IGNEOUS ROCKS 591
and Q the quartz molecules, respectively. Points plotting above the line PF are
oversaturated with respect to silica.

Top-centre
This is a plot similar in concept to the QLM diagram to show the relationship
of normative feldspar and ferromagnesian molecules with silica saturation. As
many compositions have neither normative Q nor normative Ne, Lc, or Kp
many rocks plot on the horizontal line between the left and right apices of the
diagram—a disadvantage that the QLM diagram does not have.

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
Top-right
This is a normative approximation to Streckeisen's (1967) modal classification
of igneous rocks. One of the difficulties of equating the normative and modal
approaches is in allotting normative Ab between alkali feldspar and plagioclase
(Le Maitre, 1976). In this case the ratio of alkali feldspar (Or') to plagioclase
(An') is taken as the ratio of normative Or to An. The normative Ab value is only
used to maintain the correct ratio between total normative feldspar and norm-
ative Q (or Ne+Lc+Kp). Again a disadvantage of this diagram is that many of
the analyses fall on the horizontal line between 'alkali feldspar' and 'plagioclase',
having neither normative Q nor normative Ne, Lc, or Kp.

Centre and centre-right

These diagrams show the normative feldspar compositions for those rocks
in each group which have normative Q (centre, oversaturated) and those that
have no normative Q (centre-right, undersaturated).

Bottom-centre (KCN diagram) and Bottom-right (FMA diagram)

These two diagrams are in common use and popularized by Nockolds & Allen
(1953, 1954, 1956).

NUMERICAL DATA
All the numerical data has been processed by computer and the values then
rounded-off to one, two, or three decimal places. The individual values quoted
may not, therefore, exactly add up to the total given or implied due to these
rounding-off errors.
The left-hand column gives the mean wt. per cent of all those analyses in the
CLAIR file containing separate FeO, Fe2O3, H2O+ and H2O- determinations.
The total number of analyses found for the particular rock name is indicated
against the asterisk sign in the bottom right-hand of the page. The next two
columns give the C.I.P.W. norm and Niggli values of the mean analysis in the
left-hand column. The differentiation index (Thornton & Tuttle, 1960) and
crystallization index (Poldervaart & Parker, 1964) of this mean are given below
the Niggli values.
592 R. W. LE MAITRE
Under 'Data recalculated to 100 per cent' is given the mean wt. per cent and
standard deviation of all analyses containing separate FeO and Fe2O3 deter-
minations (total number against hash sign in bottom right of page) after the
nine oxides listed have been recalculated to 100 per cent. The mean wt. per cent
analysis has also been recalculated into molecular per cent. The first three
eigenvectors and eigenvalues as per cent and accumulative per cent are also given
for all these recalculated values. The meaning of eigenvectors and eigenvalues
can be found in Le Maitre (1968) or in many texts on statistics, e.g. Davis (1973).
Briefly the first eigenvector is the direction of 'the best straight-line fit' through

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
the analyses in n-space and the first eigenvalue is a measure of the spread of
points along that direction. If a single function is required that will optimize the
variability within a particular rock group then one cannot do better than to
choose the first eigenvector. For example, to show best the variability among
a group of andesites, evaluate the expression:
—0-888 SiO 2 +0050 TiO 2 + . . . - 0 0 3 4 Na 2 O-0065 K2O
for each andesite analysis and the values obtained will give the relative position
of the andesites along the first eigenvector.
The eigenvalues are actually the sums of squares of the data points on to
hyperplanes normal to the corresponding eigenvector and passing through the
mean. They are thus indicators of the shape of the distribution of the data
points in space, e.g. spherical, planar, elongated, which in turn is a function of
the factors controlling the variability.
The geographical distribution of the analyses are given for all analyses in
the CLAIR file and the total number is given against the dagger sign in the
bottom right-hand corner of this page. For the purposes of this paper the world
is divided as shown in Fig. 1.
The number of references given in the bottom left-hand corner of the page
refers to the total number of individual references from which the analyses were
obtained.

COMMENTS ON THE DATA

The order in which the data are presented is shown in Table 1. As can be
seen the rock types are roughly grouped according to their chemical associations.
Originally only rock types with more than 100 analyses in the CLAIR master
file were included, but this limit was reduced to seventy to include a few more
rock types of current interest. Rock types with field names, e.g. porphyry,
porphyrite were excluded even though they had well in excess of seventy
analyses. It must be stressed that the name associated with each analysis is that
given by the author of the paper from which the analysis was taken. Therefore,
as no attempt has been made to impose a uniform nomenclature on the rocks,
the data represent what a wide variety of people, of different nationalities, often
using slightly different classification criteria, mean by a particular rock name.
Pacific
m http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014

Ocean

Fio. 1. Location of boundaries used to describe the geographical distribution of analyses in each set of averages.
594 R. W. LE MAITRE
TABLE 1
Order ofpresentation of averages

No. Name No. Name

1. Igneous rocks 23. Gabbro
24. Norite
2. Nepheline syenite 25. Dolerite
3. Tinguaite 26. Diabase
4. Phonolite 27. Basalt
5. Lujavrite 28. Tholeiite

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
6. Syenite 29. Basanite
7. Trachyte 30. Nepheline leucite basalt
31. Nephelinite
8. Granite 32. Tephrite
9. Rhyolite
10. Adamellite 33. Anorthosite
11. Granodiorite 34. Websterite
12. Rhyodacite 35. Pyroxenite
13. Dacite 36. Lherzolite
14. Tonalite 37. peridotite
15. Diorite 38. Harzburgite
16. Andesite 39. Dunite
17. Monzonite
18. Latite
19. Trachyandesite
20. Mugearite
21. Trachybasalt
22. Hawaiite

In one sense then, this presentation must have a minimum number of dis-
satisfied customers!

IGNEOUS ROCKS (No. 1)

Before looking at the individual groups it is worthwhile studying the igneous
rocks as a whole. From the PCA diagram it is clear that the majority of analysed
igneous rocks lie in a slightly curved distribution in 9-oxide space between
. 'basaltic' and 'granitic' compositions (from left to right). The major departure
from this trend are the 'ultrabasic' compositions which hang down as a large
lobe in the bottom part of the diagram. This marked change in direction means
that the processes that relate the 'ultrabasic' compositions to the 'basaltic'
compositions cannot be the same as those relating the 'basaltic' to the 'granitic'
compositions. Furthermore in 9-oxide space the distance between 'ultrabasic'
and 'basaltic' compositions is virtually the same as the distance between the
'basaltic' and 'granitic' compositions, which makes the petrologist's pre-
occupations with subdividing this latter series rather puzzling.
In the commonly used FMA triangle, and to a lesser extent in the KCN tri-
angle, there is a very marked concentration of analyses following a well-defined
trend. This means that a series of unrelated analyses taken at random would
 CHEMICAL VARIABILITY OF COMMON IGNEOUS ROCKS 595
tend to fall on the trend—USER BEWARE! In the normative feldspar triangles
there is a marked concentration of points along a slightly curved path from an
intermediate plagioclase to an intermediate alkali feldspar composition. It is
also interesting to note that in the two double triangular plots the distribution
of analysis is much more uniform for the undersaturated compositions than
for the oversaturated compositions. The reason for this is not immediately
obvious.
It is also clear from the diagrams that any classification of igneous rocks will

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
be largely arbitrary as there appear to be no genuine minima in any of the
distributions.

UNDERSATURATED SALIC ROCKS (NOS. 2-5)

These form a reasonably coherent chemical group as can be seen in the PCA
diagrams. Both the phonolites and tinguaites show a tendency to spread towards
trachytic and oversaturated compositions, possibly a reflection of the difficulty
of estimating the amount of nepheline in this type of rock. Because of the rarity
of occurrence the lujavrite data (No. 5) are heavily weighted towards the ex-
amples from the U.S.S.R. The main difference between this type and the other
three is its lower CaO and higher Na2O and Fe2O3 contents as can be seen in the
KCN and normative feldspar triangles and in the abundance of normative Ac.

SATURATE SALIC ROCKS (Nos. 6 and 7)

These have a wider spread of composition than the undersaturated salic rocks
and obviously include types that many people would have classified as phono-
lites or nepheline syenites. There is a slight majority of oversaturated analyses
in both rock types but somewhat surprisingly the greatest spread is among the
syenites (No. 6). This probably reflects the fact that quartz syenites and syenites
with small amounts of nepheline are still called syenites, but in volcanic types
the presence of quartz or nepheline would, to many people, indicate rhyolite or
phonolite. Note the uselessness of the KCN diagram in categorizing these types.

OVERSATURATED ACID TO SATURATED INTERMEDIATE ROCKS

(Nos. 8-16)
In these types no attempt has been made to distinguish the alkaline from calc-
alkaline varieties and inspection of any of the plots will show that there is no
apparent natural division into the two types. Most of these groups show con-
siderable spread with the exception of adamellite (No. 10), rhyodacite (No. 12),
and tonalite (No. 14) which seem to be more clearly defined. However there is
still considerable overlap between adjacent types in the series, e.g. between
dacite (No. 13) and rhyodacite (No. 12). This is particularly evident in the
simulated Streckeisen diagram and the normative feldspar triangle. The only
way to efficiently distinguish between these sets of data is to use multivariate
discriminant analysis which will be the subject of a forthcoming paper. Another
596 R. W. LEJMAITRE
point of interest is the consistently higher average Fe/Me+Mg ratio in the
C.I.P.W. normative silicate minerals in the plutonic rocks compared with the
chemically equivalent volcanic rocks, presumably caused by the higher Fe2O3/
FeO ratio of the volcanic rocks.
The average andesite (No. 16) is slightly lower in SiO2 than the average given
by Chayes (1969) due no doubt to the inclusion of some Ne normative types
which Chayes automatically excluded. Although few people would tolerate a
nepheline-bearing andesite, it seems presumptuous to automatically exclude Ne

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
normative types until it has been conclusively shown that the ferromagnesian
minerals in andesites are never Ne normative.

INTERMEDIATE ALKALINE ROCKS (Nos. 17-22)

Four of the volcanic types, latite (No. 18), mugearite (No. 20), trachybasalt
(No. 21), and hawaiite (No. 22) show reasonable chemical coherence in their
distributions. Again it is a plutonic rock that seems to show the greatest spread
in its distribution, i.e. monzonite (No. 17). Both monzonite and the chemically
equivalent latite are peculiar in the magnitude of the first eigenvalue (82 to 83
per cent). This indicates that the variability shown by these two rock types is
caused by variation between two closely defined 'end members' or to use
popular statistical jargon is caused by a single 'factor'. Some of these types
have a rather biased geographical distribution, e.g. a large majority of the
monzonite and latite analyses come from N. America and most of the hawaiite
analyses come from the Pacific.

BASIC ROCKS (Nos. 23-28)

All of these types show considerable scatter in their distribution, with the
exception of tholeiite (No. 28) which not surprisingly forms a well-defined
group as half the analyses come from the Pacific. As would be expected they
nearly all fall below the line of demarcation between the alkali and tholeiitic
basalt fields on the alkali-silica diagram. In the other groups no attempt has
been made to distinguish alkali from tholeiitic types. Using the alkali-silica
diagram and the state of silica saturation as criteria the basalt analyses (No. 27)
are evenly split between alkali and tholeiitic types while the gabbro (No. 23),
dolerite (No. 25), and diabase (No. 26) have a slight majority of tholeiitic types
confirming field and petrological observations. As might be expected there is no
significant difference in chemistry between dolerite (No. 25) and diabase (No.
26). Note the remarkable coincidence of the equal values of 51-58 for the per-
centages of SiO2 in the two averages of the recalculated analyses!
All these types have a fairly small first eigenvalue (37 per cent-47 per cent)
and appreciable third eigenvalues (13 per cent to 17 per cent) indicating at least
two major 'factors' causing the variation—these are almost certainly the feldspar
and ferromagnesian components.
The average basalt and dolerite are very similar to those given by Manson
 CHEMICAL VARIABILITY OF COMMON IGNEOUS ROCKS ', 597
(1967) but the gabbro is somewhat different possibly due to the small number
of analyses in Manson's data (127) compared with 1317 in this compilation.
A further point of interest is the tendency for the basalt analyses to show a
correlation between the amount of undersaturation and O l + D i + H y which is
completely absent in the gabbro and dolerite. This is also reflected in the direc-
tion of the first eigenvectors.

UNDERSATURATED BASIC ROCKS (NOS. 29-32)

For their small sample size these types show considerable scatter, especially

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
nephelinite (No. 31) particularly in the FMA triangle. Although the tephrite
(No. 32) and basanite (No. 29) analyses somewhat overlap, their normative
feldspar compositions do generally behave as they ought to, with the basanites
being somewhat more plagioclase-rich and the tephrites more alkali feldspar-
rich. However, there seems to be very little difference between the nephelinite
(No. 31) and nepheline, leucite basalt (No. 30) analyses probably due to the
confusion that has existed in the nomenclature of these rocks (see Streckeisen
(1967), p. 187-8).

ULTRABASIC ROCKS (NOS. 33-39)

All these types show considerable scatter especially in the PCA diagram.
Although there is also considerable overlap their mineralogy is reasonably well
reflected in the C.I.P.W. norms. The low totals for the average of the websterite
(No. 34), lherzolite (No. 36), and harzburgite (No. 38) are due to the fact that
most of the data for these types were compiled from a single Russian source of
rocks containing small amounts of chromite and nickel sulphides which were
not included in the averages. Some of the rocks were also slightly altered as can
be seen from the average H2O+ contents.

ACKNOWLEDGEMENTS
This study was made possible by a grant from the Australian Research Grants
Committee. Many thanks are due to Mrs. S. Monahan, Miss B. Edwards, Mrs.
K. Mehre, and Miss H. Johns who patiently accumulated the CLAIR master
file; to Dr. A. Cundari and A. K. Ferguson for discussions involving the format
of the presentation; to Professor J. F. Lovering and Professor K. Yagi for
critically reviewing the manuscript; to Mr. L. J. Course, Department of Fine
Arts, University of Melbourne for discussions involving the theory of half-tone
printing and photographic techniques for the production of the diagrams; to
Mrs. Gwen Waterman for patiently typing the tables and to Mr. Don Campbell
for producing the diagrams in their final form.
598 R. W. LE MAITRE

REFERENCES
BAKER, I., 1969. Petrology of the volcanic rocks of St. Helena, S. Atlantic. Bull. geol. Soc. Am. 80,
1283-1310.
BURRI, C, 1964. Petrochemical calculations based on equivalents. Jerusalem: Israel Program for
Scientific Translations.
CHAYES, F., 1969. The chemical composition of Cenozoic andesite. Proc. Andesite Conference. Oreg.
Dept. Geol. Min. Ind. Bull. 65, 1-11.
CUNDARI, A., & LE MAITRE, R. W., 1970. On the petrogeny of theleucite-bearing rocks of the Roman
and Birunga regions. / . Petrology, 11, 33-47.
DALY, R. A., 1933. Igneous Rocks and the Depths of the Earth. New York: McGraw-Hill.

Downloaded from http://petrology.oxfordjournals.org/ at National Chung Hsing University Library on March 31, 2014
DAVIS, J. C , 1973. Statistics and data analysis in geology. New York: Wiley.
GREEN, J., & POLDERVAART, A., 1955. Some basaltic provinces. Geochim. cosmochim. Ada, 7,177-88.
LE MAITRE, R. W., 1962. Petrology of volcanic rocks, Gough Island, S. Atlantic. Bull. geol. Soc. Am.
73, 1309-40.
1968. Chemical variation within and between volcanic rock series—A statistical approach. / .
Petrology, 9, 220-52.
1973. Experiences with CLAIR: a computerized library of analysed igneous rocks. Chem.
Geol. 12, 301-8.
1976. Some problems of the projection of chemical data into mineralogical classifications.
Contr. Miner. Petrol. 56, 181-9.
MACDONALD, G. A., & KATSURA, T., 1964. Chemical composition of Hawaiian lavas. / . Petrology,
5, 82-133.
MANSON, V., 1967. Geochemistry of Basaltic rocks; major elements. In HESS, H. H., & POLDERVAART,
A. (Eds), Basalts, I, 215-69. New York: Interscience.
NOCKOLDS, S. R., 1954. Average chemical composition of some igneous rocks. Bull. geol. Soc. Am.
65, 1007-32.
• & ALLEN, R., 1953. The geochemistry of some igneous rock series. Geochim. cosmochim. Ada,
4, 105-42.
1954. The geochemistry of some igneous rock series: Part II. Ibid., 5, 245-85.
1956. The geochemistry of some igneous rock series: Part m . Ibid. 9, 34-77.
POLDERVAART, A., & PARKER, A. B., 1964. The crystallization index as a parameter of igneous dif-
ferentiation in binary variation diagrams. Am. J. Sci. 262, 281-9.
STRECKEISEN, A. L., 1967. Classification and nomenclature of igneous rocks. Neues. Jb. Miner. Abh.
107, 144-240.
THORNTON, C. P., & TUTTLE, O. F., 1960. Chemistry of igneous rocks: pt. I, Differentiation index.
Am. J. Sci. 258, 664-84.
TILLEY, C. E., 1950. Some aspects of magmatic evolution. Q. J. geol. Soc. Lond. 106, 37-61.