Contrib Mineral Petrol (1996) 123:92—108 ( Springer-Verlag 1996

Keith Putirka · Marie Johnson · Rosamond Kinzler ·

John Longhi · David Walker

Thermobarometry of mafic igneous rocks based on

clinopyroxene-liquid equilibria, 0—30 kbar

Received: 16 May 1994/Accepted: 15 June 1995

Abstract Models for estimating the pressure and tem- and diopside#hedenbergite (DiHd) components are
perature of igneous rocks from co-existing clino- calculated from a normative scheme which assigns the
pyroxene and liquid compositions are calibrated from lesser of Na or octahedral Al to form Jd; any excess AlVI
existing data and from new data obtained from experi- forms Calcium Tschermak’s component (CaTs;
ments performed on several mafic bulk compositions CaAlAlSiO ); Ca remaining after forming CaTs and
6
(from 8—30 kbar and 1100—1475° C). The resulting CaTiAl O is taken as DiHd. Experimental data not
2 6
geothermobarometers involve thermodynamic expres- included in the regressions were used to test models (i)
sions that relate temperature and pressure to equili- and (ii). Error on predictions of ¹ using model (i) is
brium constants. Specifically, the jadeite (Jd; $40 K. A pressure-dependent form of (i) reduces this
NaAlSi O )—diopside/hedenbergite (DiHd; Ca(Mg, Fe) error to $30 K. Using model (ii) to predict pressures,
2 6 the error on mean values of 10 isobaric data sets (0—25
Si O ) exchange equilibrium between clinopyroxene
2 6 kbar, 118 data) is $0.3 kbar. Calculating thermodyn-
and liquid is temperature sensitive. When composi-
tional corrections are made to the calibrated equili- amic properties from regression coefficients in (ii) gives
brium constant the resulting geothermometer is VJ$ of 23.4 $1.3 cm3/mol, close to the value anticip-
&
ated from bar molar volume data (23.5 cm3/mol). Ap-
plied to clinopyroxene phenocrysts from Mauna Kea,
C D
104 Jd19*Ca-*2*Fm-*2
(i) "6.73!0.26* ln Hawaii lavas, the expressions estimate equilibration
¹ DiHd19*Na-*2*Al-*2
depths as great as 40 km. This result indicates that
transport was sufficiently rapid that at least some
C D
Mg-*2
!0.86* ln #0.52*ln [Ca-*2] phenocrysts had insufficient time to re-equilibrate at
Mg-*2#Fe-*2
lower pressures.
an expression which estimates temperature to $27 K.
Compared to (i), the equilibrium constant for jadeite
formation is more sensitive to pressure resulting in
a thermobarometer Overview

C D
¹ ¹ Jd19 Thermobarometry applied to phenocrysts in volcanic
(ii) P"!54.3#299* #36.4* ln
104 104 [Si-*2]2*Na-*2*Al-*2 rocks provides a method for determining depths and
temperatures of magma storage. If magma stalls at the
#367*[Na-*2*Al-*2]
base of the crust prior to discharge, thermobarometry
which estimates pressure to $ 1.4 kbar. Pressure is in constrains crustal thickness, and perhaps also the
kbar, ¹ is in Kelvin. Quantities such as Na-*2 represent geothermal gradient at the time of eruption. Few means
the cation fraction of the given oxide (NaO ) in the of temperature estimation widely applicable to basaltic
0.5 rocks exist and even fewer means of determining the
liquid and Fm"MgO#FeO. The mole fractions of Jd
pressures of phenocryst equilibration for such rocks
K. Putirka ( ) · M. Johnson · R. Kinzler · J. Longhi · currently exist. The thermobarometers presented here
D. Walker are calibrated using new experimental data and are
Lamont-Doherty Earth Observatory of Columbia based on clinopyroxene-liquid equilibria. Equilibria in-
University Palisades, NY 10964, USA volving clinopyroxene were chosen for two reasons.
Editorial responsibility: T.L Grove (1) Clinopyroxene is a common phenocryst phase in
 93

volcanic rocks; clinopyroxene-based thermobaro- smaller; these exchange-equilibria are temperature-de-

meters will thus be of general applicability. (2) Forma- pendent and are calibrated as thermometers. Both ex-
tion of the clinopyroxene component jadeite is accom- change equilibria, Jd-DiHd and CaTs-DiHd, require
panied by a large change in molar volume and provides a pressure-dependent term for complete description.
barometric information for igneous processes. The thermometers are presented both with and without
The thermobarometers are thermodynamic rather pressure terms; ignoring the pressure term facilitates
than empirical. They expand on the approaches taken application of the thermometers and results in only
by Roeder and Emslie (1970) and Drake (1976). Equi- a slight increase in the uncertainty of the estimate.
librium constants (K ) describing pyroxene solution As many of the pertinent thermodynamic data are
%2
are calibrated for both pressure (P) and temperature lacking for the appropriate pyroxene and liquid com-
(¹ ). The advantage of calibrating equilibrium con- ponents, the thermobarometers are constructed from
stants is that thermodynamic relationships exist which regression analysis of experimental clinopyroxene-
relate K to ¹ and P. A thermodynamic approach also liquid equilibrium compositions. In an attempt to ex-
%2
allows extrapolation to pressures, temperatures, and plore compositional complexities, new experiments
compositions outside the experimental range used for were performed in the melting interval 8—30 kbar at
calibration. 2 kbar intervals on a range of natural basalt composi-
The large partial molar volumes for Na and Al tions. Several solution models were tested. The result-
oxides in basaltic liquid (Lange and Carmichael 1987) ing thermobarometers recover P and ¹ for
combined with the small partial molar volume of clinopyroxene-liquid pairs synthesized from other labs
jadeite (NaAlSi O ; Jd) result in a relatively large P- and yield thermodynamic quantities that compare well
2 6
dependency for jadeite formation compared to other with known values.
mineral-liquid equilibria (Table 1; see also Robie et al. It is unclear whether mafic volcanic rocks are erup-
1979). The Jd-liquid equilibrium is also temperature- ted directly from substantial depths or are invariably
sensitive and is thus calibrated as a thermobarometer. processed in shallow-level magma reservoirs. Also un-
The volume changes for the solution of Jd and calcium- certain is the depth of stagnation during different stages
Tschermak (CaAlAlSiO ; CaTs) into diopside-heden- of volcanism; are the parental magmas of alkalic and
6
bergite (Ca(Mg,Fe)Si O ; DiHd) are significantly tholeiitic suites extracted from different depths? Fur-
2 6
ther, what controls depth of magma ponding: density
contrasts or mechanical behavior of the lithosphere?
Table 1 Thermodynamic properties! These problems will be explored through application of
Anticipated thermodynamic properties the thermobarometers to pyroxene-bearing rocks from
Mauna Kea, Hawaii.
Equilibrium *S" *S# *V$ *H%
1573 1573
Jd-liq !120 !113 !23.5 !154
Di-Jd 56.7 47.1 !10.8 !7.7 Thermodynamic approach
Di-CaTs 54.0 — !1.33 —
Di-liq !63.3 !65.9 — !146 Below we present the equilibria and thermodynamic
equations used for thermobarometry. The thermodyn-
Thermodynamic properties recovered from models
amic database is presently inadequate to yield all the
K Model d *S *V *H coefficients appearing in these equations. The following
%2 functions are thus used for regression analysis of ex-
Jd-liq P1 !71.6 !23.2 !128 perimental equilibrium clinopyroxene-liquid composi-
Jd-liq P2 !95.6 !23.4 !129 tions.
Di-Jd T1 193 — 288
Di-CaTs T3 283 — 449
As anticipated from existing molar volume data,
a useful equilibrium for barometry involves jadeite
! Crystalline data are from Robie et al (1979); crystal molar volumes formation (K[Jd-liq]). The exchange equilibria, diop-
are corrected to 1573 K assuming aJ$"4.0]10~5/K and side-jadeite (K[DiHd-Jd]), and diopside-calcium-
aD*@C!T4"3.0]10~5/K; *H is in kJ/mol, *S is in J/mol*K and Tschermak (K[DiHd-CaTs]), are less sensitive to P but
volumes are cc/mol. SEE for model *H are $20 kJ/mol and for
model *S$5 J/mol*K are sensitive to ¹, providing pressure-sensitive ther-
" Calculation uses partial molar entropy of oxides at 298 K (Richet mometers. These equilibria may be expressed as
et al. 1993), integrated to 1600 K using partial molar C data from
1
Stebbins et al. (1984) who determined C with data ranging from
1 2SiO-*2#NaO-*2 #AlO-*2 "NaAlSi O19 (1)
1200—1850 K 2 1@2 3@2 2 6
# From C and S -S data of jadeite and diopside glass reported by
1 298 0 CaFmSi O19#NaO-*2 #AlO-*2 "NaAlSi O19
Richet et al. (1993, Table 7) 2 6 1@2 3@2 2 6
$ Uses partial molar oxide data of Lange and Carmichael (1987)
% Derived from H and H data of Navrotsky (1981) and molar #CaO-*2#FmO-*2 (2)
7 &
volume of jadeite glass (Richet et al. 1993); the estimate for
HJ$ assumes that H ./H is the same for jadeite and diopside CaFmSi O19#2AlO-*2 "CaAlVIAlIVSiO19#FmO-*2#SiO-*2 (3)
& &,T 7,985 2 6 3@2 6 2
94

where superscripts denote the phase (liq"liquid; where the coefficient A is determined by multiple linear
px"pyroxene) and the symbol FmO refers to total regression. Equation 9 is equivalent to Eq. 8 when
(FeO#MgO). The methods used to calculate liquid A"0.
and pyroxene components are discussed below.
As equilibrium constants will vary with P and ¹,
expressions must be developed to quantify these de- Activity modifying terms
pendencies. The relationship between K and Gibbs
%2
free energy may be written as Mixing properties of pyroxene and liquid components
are unlikely to be perfectly ideal; such non-idealities
!R¹ lnK "*G0 (4) may reveal themselves as compositional dependencies
%2 3 residual from the ¹ and P dependencies of K antici-
%2
where R is the gas constant, ¹ is temperature (Kelvin) pated from the equations presented above. To compen-
and *G0 is the Gibbs free energy change for the sate for non-ideality and to produce expressions that
3 are useful as geothermobarometers for natural systems,
stoichiometric equilibrium at a reference pressure
P and temperature, ¹. Rearranging and expanding additional compositional terms were added to Eqs 6,
0
*G0 while holding *V constant yields 8 and 9. Compositional terms that most successfully
3 3 improve regression statistics and the predictive power
of the models are those that appear in the calibrated
*HP0 *SP0 (P!P )*V
ln K "! 3 # 3 ! 0 3 (5) equilibria. Formally, equilibrium constants are the ra-
%2 R¹ R R¹ tio of activities, a , (products divided by reactants) each
*
raised to the power of the corresponding coefficient in
where *HP0, *V , and *SP0 are, respectively, the differ-
3 3 3 a balanced chemical equilibrium. Equilibrium con-
ences (products!reactants) of enthalpy, volume and stants may be rewritten in terms of K where K "
entropy of a given equilibrium. Rearranging Eq. 5 to D %2
K Kc and a "X c . K is the ratio %XJ (prod)/
give P as the dependent variable gives DJ * * i D *
%X (react), where X are mole fractions of components
* *
i, and the exponent J represents the coefficient of i in
R¹ *SP0 *HP0 a balanced chemical equilibrium. Similarly, Kc is a ratio of
P"! ln K # 3 ¹! 3 (6)
*»
3
%2 *V
3
*V
3
activity-modifying constants, %J c (prod)/%c (react).
J
* *
Terms of the form ¹ ln K (Eq. 6) combine both
where P "0. Holding *HP0, *SP0 and *V constant, %2
0 3 3 3 temperature and implicit compositional information.
Eq. 6 provides the simplest function that satisfactorily Since ¹ ln K "¹ ln (K Kc )"¹ ln K #¹ ln Kc ,
describes variations in the experimental data. Temper- %2 D D
any model regressed only with ¹ ln K terms may ap-
ature and pressure were measured directly in each D
pear to have non-ideal compositional dependencies
experiment. K is computed from microprobe analyses upon the components appearing in the K . If activity
%2 %2
of coexisting pyroxenes and glass and is based on coefficients are known beforehand, Eq. 6 could be writ-
Equation 1, 2 or 3. The remaining quantities, *HP0, ten as
3
*SP0 and *V , are determined by regression analysis.
3 3
Equation 5 may be rearranged to give 1/¹ as the R¹ *SP0 *HP0 R¹
dependent variable, P"! ln K # 3 ¹! 3 ! ln Kc (10)
*V D *V *V *V
3 3 3 3
1 R *SP0 Note that the coefficients of ¹ ln K and ¹ ln Kc are
ln K # 3 (7) D
"!
(*HP0#P*V ) %2 (*HP0#P*V ) both !R/*V . In the present case, however, the c are
¹
3 3 3 3 3 *
unknown. Compositional dependencies may be written
If *V is zero (or if only data at P"0 are utilized) then as terms of the type (!R/*V) ¹ ln Kc"
3 Y*¹ ln [%XJ]. The %X for any particular Y would
Eq. 7 reduces to * *
include all components with the same activity coeffic-
1 R *SP0
ient (as, for example, appears to be the case for Na and
"! ln K # 3 (8) Al) possibly reflecting coupling. Thus, for a single non-
¹ *HP 0 %2 *HP0
3 3 ideal component j, (!R/*V) ¹ ln Kc "Y*¹ ln [X J]
+ +
(let Jbe positive or negative if j is on the products or
As with Eq. 6, the values for R/*HP0 and *SP0/(*HP0) in
3 3 3 reactants side respectively). Equation 10 in this case
Eq. 8 may be derived through regression analysis. Since
becomes,
Eq. 7 is not easily rearranged for 1/¹ with P and ln K
%2
as separable, linear independent variables, the pressure
R¹ *SP0 *HP0
dependency may be added to Eq. 8 empirically as: P"! ln K # 3 ¹! 3 !Y*¹ ln [%XJ] (11)
*V D *V *V *
3 3 3
1 R *SP0
ln K # 3 #A*P
The c may be determined by finding the value that
"! (9) *
¹ *HP 0 %2 *HP0 %XJ must have so that Y has the value !R/*V .
3 3 * 3
 95

Equation 11 forms the basis of the ‘activity-modifying’ pyroxene and/or plagioclase in addition to clinopyroxene. Many of
form of the thermobarometer (model P2). This proced- the experiments performed by Walter and Presnall, whose (1994)
ure does not account for the P- or ¹-sensitivity of the data are incorporated into the models, are also multiply saturated.
The presence of additional phases other than pyroxene and liquid
activity coefficient and merely assumes that some num- should not affect the equilibria calibrated.
ber Y exists such that c "(XJ )Y. Experimental bulk compositions were selected to explore an-
+ +
The compositional dependencies may be regressed ticipated compositional dependencies (Table 3). The Woodlark
as ‘empirical’ terms such as, Basin basalt is both Al O - and Na O-rich. A ugandite (a mafic,
2 3 2
peralkaline volcanic rock) is low in SiO and high in CaO, TiO ,
2 2
alkalis, and volatiles. An ankaramite (a mafic, pyroxene-phyric vol-
R¹ *SP0 *HP0 canic rock) is high in FeO#MgO and extends pyroxene-liquid
P"! ln K # 3 ¹! 3 !C*[%X ] (12)
*V D *V *V * equilibria to high pressures. A mid-ocean ridge basalt from the
3 3 3 KANE Fracture Zone in the mid-Atlantic provides a less extreme,
Equation 12 resembles Eq. 6, but with K replacing MgO-rich bulk composition. The basalts studied span a wide range
D of TiO , Cr O , and FeO contents. Since such elements may aid Na
K and %X as an added empirical parameter with 2 2 3
%2 * or Al substitution in pyroxene, they may affect the equilibria of Eqs.
corresponding coefficient C determined through regres- 1, 2 and 3. Phase equilibria data from experiments in the synthetic
sion (model P1). Note that exponents are assumed to be SiO #CaO#Al O #MgO#Na O (SCAM#Na) system by
2 2 3 2
unity to simplify application. The thermodynamic in- Walter and Presnall (1994) provide a study of the equilibria in the
terpretation of C in Eq. 12 is less clear than for Y in Eq. absence of these elements.
11, but this form avoids introducing temperature
a third time, facilitating P!¹ estimates with minimal
error when ¹ is unknown. The right-most terms in Eqs. Modifications to experimental procedure
11 and 12 may be repeated, as necessary, for all i compo-
nents that have dissimilar activity coefficients. If two Experiments performed at pressures above 12 kbar produced a no-
ticeable quench crystalline phase along the crystal/liquid interface if
components are coupled (or have the same activity quenched isobarically. This problem was minimized by employing
coefficients) then regression of the natural log of the the following modification. For experiments performed above 12
terms individually would produce identical coefficients kbar, the pressure was dropped to 8—10 kbar immediately prior to
for each term, in which case they could then be com- shutting off the power. The entire procedure, dropping pressure then
shutting off the power, takes 1—2 s. In experiments quenched in this
bined. A similar approach to compositional depend- manner, the quench phase is negligible (Fig. 1). Re-equilibration
encies was followed for the thermometric functions. during quench could potentially cause crystal rims to re-melt. This is
of critical concern since the rims are most likely to be in equilibrium
with surrounding glass. Pyroxene dissolution rates are expected to
be diffusion-controlled (Zhang et al. 1989) and for quench times of
Experimental approach 1—2 sec the distance affected by dissolution both within the glass and
crystal should be negligible. The diffusion coefficient of Al O (D )
2 3 A-
Although many pyroxenes have been synthesized at high pressure, in liquid coexisting with diopside at 1375° C, 13 kbar is 7.3 l2/s
few studies have focused on growing pyroxenes from a single liquid (Zhang et al. 1989). The distance (d) through which the liquid is
composition over a wide pressure range. Further, many experiments affected during a 2 s quench (d"(D t)1@2) is 3.8 lm. As a distur-
A-
have been geared toward understanding mid-ocean ridge mag- bance in liquid composition on this scale is not observed we con-
matism and thus encompass a narrow range of starting composi- clude that resolution of pyroxene has not affected the interface
tions. As the goal of this study is to produce thermobarometers of composition. None of the crystals show significant melting and the
general applicability, new experiments were performed to broaden glass in all experiments is homogeneous throughout the charge. The
the existing data base. These experiments were performed at temper- K derived from such experiments also lie on trends continuous and
%2
atures at or near the pyroxene liquidus in order to maintain near- overlapping with lower pressure experiments where the pressure-
uniform liquid composition while varying P and ¹. Close attention quench procedure was neither needed nor used.
was paid to determining equilibrium pyroxene and glass composi- This technique was found to suppress quench growth success-
tions. Some experimental modifications, discussed below, improved fully in komatiitic melts, where the quench problem is otherwise
our ability to recover glass compositions unmodified by quench most severe (Fig. 1). Cores and pistons also survived this procedure
pyroxene overgrowths. with no ill-effects. Figure 1A shows the pitfall of dropping the
Experiments were run in Boyd-England piston-cylinder presses pressure too rapidly to 1 bar: the liquid in this case has expelled its
with a 1/2 inch bore. The pressure medium was BaCO wrapped in volatiles. In some experiments disruption of the crystal pile took
3
Pb-foil. Samples were loaded into a graphite capsule surrounded by place and glass was preserved within crevasses. In experiment WB-
an alumina sleeve and situated between two Al O end-plugs; these 17 (Fig. 1F) melt was injected within two halves of a broken garnet
2 3
materials were placed within a graphite heater. Pressure was applied and small pyroxene crystals were swept into the crevasse.
using the cold piston-in technique with a friction correction made
for BaCO (see Fram and Longhi, 1992). No pressure correction was
3
made to the thermocouple emf. Temperatures were measured with
type-D thermocouples (W#3%Re/W#25%Re). Thermal gradi- Analytical procedure
ents across the sample are 5°C (Fram and Longhi 1992). Runs were
performed at 2 to 3 kbar intervals between 8 and 30 kbar to establish Mineral and glass analyses were performed on a CAMECA MBX
clear trends with respect to pressure and K for each bulk composi- wavelength dispersive electron microprobe at Lamont-Doherty
%2
tion, and to provide numerical leverage for regression analysis. Earth Observatory (Table 4). The accelerating voltage was 15 kV
Experimental conditions and products are listed in Table 2. Run and a 1—2 lm beam diameter was used. Counting times were 30 s
times vary from 1 to 6 days, most lasting 3 days. Volume percent and a PAP correction procedure was applied. Multiple analyses of
liquid was determined optically; where reported as less than 100%, a pyroxene standard (Kakanui) yield standard deviations of 0.17
accuracy is $10%. Several runs crystallized garnet, olivine, ortho- wt% for all oxides except SiO , which has a standard deviation of
2
96

Table 2 Experimental conditions

P T T Time Run Percent
Run d (kb) (K) (C) (hours) quench Products liquid

A-11 8 1473 1200 67 T cpx#ol#gl 90

A-3 8 1523 1250 67 T gl 100
A-2 10 1573 1300 61 T cpx#ol#gl 85
A-10 10 1598 1325 48 T ol#gl 90
A-8 10 1623 1350 74.5 T ol#gl 99
A-4 12 1593 1320 70 T cpx#ol#gl 60
A-7 17 1643 1370 66 P cpx#gl 60
A-6 20 1663 1390 24 P cpx#gl 45
A-9 30 1748 1475 48 P cpx#gar#gl 45
MA-9 8 1498 1225 94 T cpx#gl 90
MA-5 8 1523 1250 72 T gl 100
MA-8 10 1523 1250 73 T cpx#gl 80
MA-6 10 1543 1270 72 T cpx#gl 90
MA-11 12 1548 1275 63 T cpx#plag#gl 70
MA-10 15 1573 1300 93 T cpx#gl#pl 60
MA-12 20 1613 1340 96 T gar#cpx#gl 55
WB-13 8 1473 1200 72 T cpx#gl 95
WB-15 10 1373 1100 166 T cpx#opx#gl 10
WB-7 10 1498 1225 120 T gl 100
WB-21 12 1498 1225 72 T cpx#gl 80
WB-9 12 1498 1225 67 T cpx#gl 80
WB-17 15 1423 1150 95 P cpx#gar#gl 55
WB-14 15 1473 1200 73 P cpx#gar#gl 70
WB-11 15 1513 1240 46 T cpx#gl 80
WB-18 20 1573 1300 70 P cpx#gar#gl 85
WB-20 25 1623 1350 72 T gar#gl 90
U-1 8 1423 1150 46 T cpx#gl 85
U-2 10 1443 1170 48 T cpx#gl 90
U-3 12 1473 1200 49 T cpx#gl 90
U-4 15 1498 1225 70 T cpx#gl 90
U-5 20 1543 1270 70 P cpx#gl 85

Quench: ‘T’ refers to experiments quenched at indicated pressure by shutting off power. ‘P’ refers to
experiments where pressure was dropped prior to power shut-off

Table 3 Starting compositions (wt%)

Sample SiO TiO Al O FeO MnO MgO CaO Na O K O Cr O total

2 2 2 3 2 2 2 3
MAR Basalt! 49.79 1.38 15.92 9.38 0.18 8.88 11.28 3.00 0.11 0.00 99.92
Woodlark Basin Basalt" 48.26 2.31 16.97 10.02 0.13 6.47 9.78 5.08 0.09 0.02 99.11
Ugandite# 36.85 6.09 9.83 13.74 0.26 5.53 13.52 2.51 6.13 0.01 94.47
Ankaramite$ 47.85 3.16 11.76 13.14 0.21 10.55 10.03 2.20 0.65 0.10 99.55

! Mid-Atlantic Ridge Basalt (near Kane Fracture Zone) donated by Jennifer Reynolds
" Woodlark Basin Basalt"sample G32-8 from Perfit et al. (1987)
# Ugandite donated by Richard Sack
$ Ankaramite"sample LP-5 from Frey et al. (1991)

0.42. Alkali-loss was minimized by first determining Na O and K O

2 2 Data sets of other studies
at 5 nA, then determining all other elements at 25 nA. Glass analyses
are homogeneous throughout each charge. Pyroxenes, however, are Although no systematic study has been previously performed in-
not. Many experiments show thermal compaction of crystalline volving the pressure-sensitivity of pyroxene-liquid equilibria, pyro-
material at the cold end of the capsule. Crystals far from the xene-liquid pairs from existing experimental work are available.
crystal/liquid interface have variable composition and were deemed These data allow an independent test of the thermobarometric
unlikely to be in equilibrium with the melt. Intracrystalline variation models derived from the data of this study.
also occurs at the interface but core-to-rim variations are not syste- Several caveats must accompany the use of previously published
matic. For a given experiment, the most consistent pyroxene ana- data. (1) Although many published studies show general trends of
lyses are from rim compositions of crystals next to the crystal/liquid increasing Al- and Na-contents of pyroxene as pressure increases,
interface. Selecting analyses in this manner is necessary when the only in the Walter and Presnall (1994) data set were the analytical
closest possible approach to equilibrium crystal compositions in procedures outlined above ostensibly followed. (2) For runs at high
crystallization experiments is required. pressure ('12 kbar), if an isobaric quench procedure was followed,
 97

Fig. 1A and B show the effect of reducing pressure on komatiite

bulk compositions just prior to thermal quench (scale in C). In A the Component calculations and experimental clinopyroxene
pressure was dropped (accidentally) to 1 bar; note the vesicles. In compositions
B the pressure was dropped to the target pressure of 8 kbar; the
entire upper portion of the charge recovered glass with relatively As all models are based on equilibrium constants, appropriate
little crystalline quench. A ‘‘normal’’, thermal quench performed on components for the liquid and clinopyroxene phases must be chosen.
a chondrite bulk composition is shown in C for comparison. D and Pyroxene components were initially assumed to be ideal (a "X ).
E compare the effects of thermal and pressure quench procedures on * *
Although clearly incorrect, abandoning this assumption led to little
basalt compositions. D is rund MA-12 where a thermal quench improvement in the models. Liquid components were ultimately
procedure was used; note the rim of quench growth separating computed as a cation fraction. Ideal mixing of oxides in the liquid
pyroxenes from glass. E is rund WB-18 where a pressure quench was also used as a starting point in calculating K . Compositional
procedure was used; note the absence of quench and also the %2
terms were added if they improved both the quality of regression
disruption of the crystal pile. F is rund WB-17, also pressure- statistics and predictive power of the models. Interpretation of these
quenched. This sample shows a garnet crystal that was split. Melt compositional terms is given in the Activity-modifying Terms section.
was injected into the crevasse carrying with it several small pyroxene Pyroxene cations are calculated per 6 oxygens. The Jd compo-
crystals. See text for description of the procedure nent is the amount of Na or octahedral Al (AlVI"Al505!AlIV;
AlIV"2!Si), whichever is less. The CaTs component is equal
the presence of a quench phase may have hindered determining the to any remaining AlVI(CaTs"AlVI!Jd). AlIV in excess of that re-
equilibrium glass and pyroxene-rim compositions. (3) For 1 bar quired to form CaTs is used to form CaTiAl O
2 6
experiments, possible Na-loss from the melt during the experiment (CaTiAl O "[AlIV!CaTs]/2). Where there is insufficient AlVI to
2 6
(Tormey et al. 1987) may lead to spuriously high K for the equili- form CaTs, the CaTs component is set to zero. Calcium remaining
%2
bria of Eqs. 1 and 2. after forming CaTs and CaTiAl O (CaTi) gives the CaFmSi O
2 6 2 6
98

Table 4 Compositions of pyroxenes and coexisting glass (wt %, NA not analyzed, standard deviations in italics)

P T d
Rund (kb) (C) phase SiO TiO Al O FeO MnO MgO CaO Na O K O Cr O Total Avgd
2 2 2 3 2 2 2 3
MAR-(Kane Fracture Zone) Data
MA-9 8 1225 pyx 49.94 0.91 7.96 6.90 0.17 16.80 16.93 0.51 0.02 0.16 100.30 2
0.86 0.02 0.80 0.31 0.02 1.19 0.98 0.07 0.01 0.04
gl 49.95 1.48 17.78 10.33 0.18 7.11 9.89 3.06 0.14 0.02 99.94 5
0.26 0.02 0.16 0.09 0.02 0.18 0.11 0.09 0.02 0.02
MA-8 10 1250 pyx 49.85 0.69 8.40 6.37 0.15 16.09 17.39 0.66 0.00 0.24 99.84 4
0.62 0.05 0.58 0.20 0.01 0.56 0.28 0.04 0.01 0.04
gl 49.76 1.50 17.27 10.26 0.17 7.36 10.00 3.08 0.10 0.00 99.50 3
0.19 0.00 0.07 0.05 0.02 0.30 0.09 0.03 0.01 0.01
MA-6 10 1270 pyx 50.17 0.88 9.08 7.78 0.14 15.44 15.86 0.63 0.00 0.08 100.05 2
0.36 0.02 0.10 0.77 0.02 1.15 1.14 0.05 0.01 NA
gl 49.90 1.59 17.16 10.31 0.16 7.21 9.95 2.95 0.12 0.01 99.37 3
0.12 0.03 0.17 0.06 0.02 0.07 0.06 0.19 0.01 NA
MA-11 12 1275 pyx 48.66 1.19 10.45 9.29 0.22 14.57 15.26 0.87 0.02 0.06 100.60 1
NA NA NA NA NA NA NA NA NA NA
gl 49.84 1.58 18.06 10.70 0.19 6.76 9.41 3.19 0.16 0.03 99.91 4
0.10 0.02 0.03 0.08 0.02 0.01 0.06 0.05 0.01 0.02
MA-10 15 1300 pyx 49.16 0.71 10.35 6.59 0.14 14.36 16.99 1.00 0.01 0.14 99.46 5
0.65 0.03 0.15 0.28 0.02 0.17 0.47 0.09 0.00 0.02
gl 49.11 1.63 17.69 10.77 0.18 6.51 9.08 3.60 0.16 0.01 98.74 4
0.33 0.02 0.11 0.05 0.02 0.06 0.10 0.06 0.04 0.01
MA-12 20 1340 pyx 50.12 0.61 11.17 6.19 0.13 13.06 17.13 1.57 0.01 0.17 100.16 4
0.54 0.13 0.24 0.43 0.02 0.08 0.52 0.04 0.01 0.01
gl 49.55 1.66 17.46 10.44 0.17 6.48 9.29 3.46 0.17 0.02 98.69 4
0.14 0.04 0.08 0.06 0.03 0.04 0.06 0.09 0.03 0.02
Woodlark Basin Data
WB-13 8 1200 pyx 51.01 0.95 5.43 6.59 0.21 17.68 17.78 0.72 0.00 0.21 100.58 3
0.37 0.21 0.49 0.17 0.03 0.32 0.12 0.04 0.01 0.03
gl 46.89 2.37 17.26 10.32 0.18 6.88 8.76 4.30 0.08 0.02 97.07 3
0.43 0.10 0.14 0.14 0.01 0.02 0.12 0.07 0.16 0.03
WB-15 10 1100 pyx 48.99 1.63 8.63 10.58 0.24 13.16 16.06 0.87 0.00 NA 100.17 2
0.15 0.15 0.22 0.05 0.04 0.10 0.21 0.02 0.00 NA
gl 50.82 2.24 19.34 11.09 0.19 3.78 6.41 5.60 0.12 NA 99.60 4
0.26 0.02 0.09 0.07 0.02 0.07 0.05 0.15 0.03 NA
WB-21 12 1225 pyx 48.09 1.44 9.10 7.16 0.20 15.50 17.10 0.86 0.03 0.11 99.58 3
0.29 0.09 0.41 0.51 0.02 0.56 0.50 0.03 0.01 0.02
gl 47.94 2.55 19.23 11.08 0.16 6.52 7.96 4.51 0.08 0.02 100.04 4
0.85 0.05 0.13 0.10 0.02 0.04 0.06 0.08 0.01 0.01
WB-9 12 1225 pyx 49.70 1.12 8.38 6.52 0.16 14.66 17.85 0.98 0.01 NA 99.38 3
0.10 0.06 0.47 0.25 0.03 0.31 0.35 0.07 0.01 NA
gl 47.17 2.53 18.02 10.47 0.18 6.37 8.43 3.77 0.11 NA 97.03 3
0.45 0.00 0.05 0.07 0.01 0.08 0.03 0.09 0.02 NA
WB-17 15 1150 pyx 48.81 1.66 10.50 7.20 0.18 12.41 18.22 1.72 0.02 0.12 100.84 2
0.10 0.06 0.17 0.15 0.01 0.05 0.13 0.02 0.01 0.00
gl 49.70 2.79 18.72 10.20 0.16 5.07 7.00 3.58 0.11 0.03 97.35 2
0.94 0.22 0.58 0.14 0.06 0.52 0.02 0.10 0.03 0.02
WB-11 15 1240 pyx 48.11 1.39 10.30 7.23 0.19 13.66 17.15 1.34 0.00 0.14 99.52 3
0.32 0.32 0.28 0.18 0.01 0.42 0.16 0.16 0.00 0.02
gl 47.15 2.64 18.24 10.84 0.15 6.43 8.08 4.19 0.10 0.00 97.81 2
0.25 0.01 0.08 0.16 0.00 0.00 0.01 0.00 0.02 0.00
WB-14 15 1200 pyx 49.25 1.37 10.78 6.95 0.07 12.89 17.12 1.50 0.01 0.07 100.00 7
0.31 0.10 0.25 0.26 0.07 0.25 0.48 0.06 0.01 0.03
gl 49.22 2.78 18.15 9.98 0.14 5.21 7.78 4.17 0.08 NA 97.50 4
0.32 0.02 0.18 0.11 0.02 0.03 0.06 0.05 0.01 NA
WB-18 20 1300 pyx 50.98 0.96 10.50 5.90 0.11 12.23 17.80 1.82 0.03 0.10 100.41 2
0.59 0.01 0.00 0.01 0.02 0.09 0.00 0.25 0.02 0.01
gl 48.34 2.43 16.70 10.36 0.18 7.23 8.90 3.86 0.07 0.03 98.10 3
0.37 0.09 0.78 0.33 0.02 0.32 0.34 0.15 0.02 0.03
Ankaramite Data
A-11 8 1200 pyx 50.64 1.33 5.86 10.77 0.19 17.63 13.52 0.54 0.02 0.23 100.49 4
0.56 0.19 0.80 0.58 0.03 0.36 0.93 0.07 0.02 0.04
gl 46.67 4.34 16.08 14.20 0.17 6.27 8.08 3.29 1.20 0.01 100.31 5
0.20 0.03 0.11 0.09 0.03 0.06 0.04 0.06 0.03 0.02
 99

Table 4 Continued

P T d
Rund (kb) (C) phase SiO TiO Al O FeO MnO MgO CaO Na O K O Cr O Total Avgd
2 2 2 3 2 2 2 3
A-2 10 1300 pyx 52.65 0.71 4.09 8.84 0.18 20.72 12.01 0.50 0.02 0.45 99.70 3
0.19 0.03 0.28 0.17 0.02 0.60 0.49 0.03 0.02 0.09
gl 46.59 3.54 13.55 13.38 0.20 8.90 9.46 2.76 0.86 0.07 99.31 3
0.31 0.01 0.11 0.24 0.01 0.32 0.14 0.08 0.03 0.01
A-4 12 1320 pyx 51.58 0.85 5.17 9.10 0.17 19.20 12.80 0.71 0.01 0.27 99.58 4
0.36 0.08 0.20 0.14 0.02 0.41 0.63 0.06 0.02 0.02
gl 45.06 3.82 13.39 14.36 0.18 8.98 8.72 3.16 0.92 0.04 98.62 4
0.39 0.06 0.11 0.08 0.01 0.06 0.04 0.07 0.05 0.02
A-7 17 1370 pyx 50.82 0.77 5.77 10.23 0.20 20.91 10.06 0.90 0.04 0.23 99.66 5
1.02 0.22 0.56 0.34 0.02 0.76 0.69 0.09 0.05 0.03
gl 44.16 4.11 12.83 14.99 0.18 10.14 8.62 2.80 1.03 0.04 98.90 4
0.20 0.02 0.04 0.35 0.03 0.25 0.03 0.04 0.04 0.02
A-6 20 1390 pyx 51.49 0.75 7.16 9.54 0.17 18.52 11.41 1.07 0.01 0.27 100.11 5
0.44 0.10 0.45 0.14 0.02 0.57 0.58 0.14 0.01 0.06
gl 45.86 3.93 13.25 15.00 0.18 9.76 8.71 2.45 0.94 0.05 100.14 4
0.22 0.06 0.13 0.43 0.02 0.08 0.22 0.40 0.11 0.03
A-9 30 1475 pyx 52.50 0.77 8.15 9.01 0.17 15.81 12.36 1.76 0.02 0.19 100.53 5
0.36 0.04 0.14 0.04 0.02 0.18 0.25 0.14 0.02 0.02
gl 45.38 4.49 11.60 15.14 0.18 8.88 8.53 2.57 1.14 0.03 97.94 3
0.27 0.03 0.13 0.30 0.02 0.16 0.09 0.05 0.01 0.02
Ugandite Data
U-1 8 1150 pyx 49.66 2.66 4.42 5.58 0.10 13.53 23.59 0.36 0.05 0.03 99.98 4
0.55 0.31 0.36 0.25 0.05 0.31 0.19 0.05 0.05 0.01
gl 36.91 6.13 10.07 13.83 0.25 5.41 13.35 2.53 6.30 0.00 94.79 3
0.29 0.09 0.10 0.26 0.03 0.09 0.13 0.20 0.11 0.00
U-2 10 1170 pyx 49.46 2.30 4.29 5.57 0.16 13.90 23.56 0.39 0.01 0.00 99.64 2
0.03 0.10 0.27 0.07 0.02 0.51 0.14 0.03 0.02 0.00
gl 36.81 6.18 9.98 13.38 0.27 5.45 13.42 2.49 6.05 0.01 94.03 3
0.06 0.01 0.03 0.22 0.03 0.04 0.06 0.08 0.13 0.01
U-3 12 1200 pyx 51.33 2.16 4.69 5.56 0.12 13.22 23.41 0.49 0.03 0.05 101.06 3
0.68 0.20 0.24 0.23 0.01 0.15 0.07 0.09 0.01 0.01
gl 36.80 5.94 9.68 13.24 0.26 5.51 13.60 2.55 5.94 0.01 93.54 3
0.25 0.02 0.09 0.14 0.02 0.02 0.19 0.17 0.02 0.01
U-4 15 1225 pyx 48.91 2.24 6.13 6.24 0.11 12.51 22.93 0.55 0.04 0.01 99.67 3
0.92 0.19 0.55 0.06 0.01 0.24 0.27 0.03 0.01 0.01
gl 36.94 6.15 9.84 14.60 0.24 5.64 13.54 2.48 6.41 0.02 95.87 4
0.10 0.00 0.09 0.23 0.02 0.06 0.07 0.05 0.13 0.03
U-5 20 1270 pyx 47.02 2.90 8.65 6.85 0.11 11.62 22.07 0.86 0.64 0.00 100.71 3
0.21 0.19 0.22 0.28 0.03 0.18 0.11 0.03 0.00 0.00
gl 37.21 5.79 9.59 13.89 0.26 5.49 13.52 2.65 6.63 0.02 95.06 2
0.99 0.31 0.40 0.26 0.01 0.50 0.28 0.00 0.41 0.03

(DiHd) component (DiHd"Ca!CaTs—CaTi). The enstatite-fer- balance calculations (Lindsley 1983) that revealed minimal Fe3` in
rosilite (EnFs) component is equal to one-half the Fm component pyroxene. All pyroxenes of this study and in Walter and Presnall
remaining after forming DiHd (EnFs"[Fm-DiHd]/2). If the calcu- (1994) contain sufficient Al to form a CaTs component. As might be
lation is performed correctly and the cation sum is four, the pyro- expected for pyroxenes in the SCAM#Na system, no significant
xene components in this scheme should sum to unity (see also AlIV in excess of CaTs exists; AlIV and CaTs are highly correlated
Lindsley (1983), for calculation of pyroxene components). In calcu- (AlIV vs CaTs has a slope of 0.9, an intercept of 0, and R"0.97)
lating K pyroxene component activities are set equal to the indicating that charge balance for AlIV in the tetrahedral site is taken
%2 up by AlVI rather than vacancies.
amounts calculated from the normative procedure.
Titanium contents of experimental pyroxenes have a one-to-one
correlation with CaTi components as calculated above (Ti vs CaTi
has a slope of one, an intercept of zero and R"0.88). This correla-
tion indicates a coupled substitution of the type TiAl (MSi )
Results
2 2 ~1
where M is a divalent cation (Sack and Carmichael 1984; Gee and
Sack 1988). Experimental pyroxenes contain very little Cr and the Thermobarometers employing various activity models
thermobarometers predict the test data best when no attempt is (Ghiorso et al. 1983; Nielsen and Drake 1979; simple
made to account for a Cr-component in pyroxene. Pyroxenes were cation fraction), compositional corrections and vari-
also analyzed for Mn and K; these elements were not used in
calculating pyroxene components. Since all experiments were per- ations on the expansion of Eq. 4, were initially ranked
formed in graphite capsules the acmite (NaFe3`Si O )component on the basis of regression statistics. Models were then
2 6
was considered negligible; this expectation was borne out by charge compared and tested on their ability to predict P and
100

¹ from the test data. The test data set comprises 129 Thermobarometers
experimental pyroxene/glass pairs synthesized from
1 bar to 50 kbar. Sources for test data are listed in the On the basis of regression statistics, slope and inter-
appropriate figure captions. Satisfactory thermo- cept values of predicted vs measured P, and predic-
barometers were constructed from regressions based tion of test data, model P1 is our preferred model
only on the data of Table 4 and from Walter and for P estimation. Model P1 yields the lowest standard
Presnall (1994) and thus do not include any 1 bar data. error of estimate (SEE) of all thermobarometers tested.
Thermometric functions, however, were greatly im- The SEE represents the formal error of a model and
proved when randomly selected 1 bar data were incor- is analogous to the standard deviation of a mean.
porated into the regressions. All 1 bar data incorpor- In the present context, it is the number E for which
ated into the thermometers for regression analysis were the probability is 68% that any individual P (or
removed from the data set used to test the thermom- ¹ ) estimate is within $E of the true (model) value.
eters, thus avoiding overlap between test and regression If errors in pressure, temperature and compositional
data sets. As a final consistency check for the thermo- measurements are random, the mean of predictions
barometers, thermodynamic quantities were computed of a set of isobaric experiments should be normally
from regression coefficients and compared to existing (1 distributed about the reported value. Error from
bar) thermodynamic data. any other well-characterized set of equilibrium pyro-
We observed that use of liquid components ex- xene-glass pairs with similar analytical and experi-
pressed as a cation fraction led to the most successful mental uncertainty should be comparable to the model
thermobarometers and thermometers. The ther- SEE.
mobarometers are presented in two forms (Table 5): P1
presents compositional dependencies in their ‘empiri-
cal’ form as shown in Eq. 12, while model P2 uses the 8—50 kbar
explicit activity-modifying form of Eq. 11. The product,
Na-*2*Al-*2, (models P1 and P2) implies identical acti- Predictions of models P1 and P2 against the regression
vity coefficients for Na-*2 and Al-*2. Nielsen and Drake data are shown Fig. 2. Figures 3 and 4 show how well
(1979), motivated by the work of Bottinga and Weill these models predict the test data. Model P1, employ-
(1972), found that combining Na- and Al-activities as ing the empirical form of the compositional term
an NaAlO-*2 component was useful in describing 1 bar Na-*2*Al-*2, predicts P somewhat better than model
2
pyroxene-liquid equilibria. Application of their silicate P2 (SEE"1.36 vs 1.51 kbar). The lower apparent
melt activity model was, however, less successful than accuracy of model P2 may result from error added
simple cation units in predicting test data. by introducing temperature a third time (in the

Table 5 Thermobarometers (¹
in units of Kelvin, P in kbar, Fm Thermobarometric models

C D
FeO#MgO, DiHd Di#Hd; ¹ ¹ Jd19
superscripts denote phase—px P1 P"!54.3#299 #36.4 ln #367 [Na-*2*Al-*2]
104 104 [Si-*2]2*Na-*2*Al-*2
pyroxene, liq liquid; liquid
components in cation fraction;

C D C D
¹ ¹ Jd19 ¹ 1
see text for calculation of P2 P"!50.7#394 #36.4 ln !20.0 ln
pyroxene components) 104 104 [Si-*2]2*Na-*2*Al-*2 104 [Na-*2*Al-*2]

Thermometric models

C D C D
104 Jd19*Ca-*2*Fm-*2 Mg-*2
T1 "6.73!0.26*ln !0.86*ln #0.52* ln [Ca-*2]
¹ DiHd19*Na-*2*Al-*2 Mg-*2#Fe-*2

C D C D
104 Jd19*Ca-*2*Fm-*2 Mg-*2
T2 "6.59!0.16*ln !0.65*ln #0.23* ln [Ca-*2]!0.02p
¹ DiHd19*Na-*2*Al-*2 Mg-*2#Fe-*2

C D C D C D
104 CaTs19*Si-*2*Fm-*2 Mg-*2 1
T3 "6.92!0.18*ln !0.84*ln !0.29* ln
¹ DiHd19*[Al-*2]2 Mg-*2#Fe-*2 [Al-*2]2

C D C D
104 CaTs19*Si-*2*Fm-*2 Mg-*2
T4 "7.20!0.04*ln !0.59*ln
¹ DiHd19*[Al-*2]2 Mg-*2#Fe-*2

C D
1
!0.18* ln !.03*P
[Al-*2]2
 101

Fig. 2 Models P1 A and P2 B are used to recover pressures for the

Fig. 3 Models P1 A and P2 B are used to predict pressures from
data used in the regressions. The lines shown are one-to-one correla-
published experimental data (data not used in the regressions) from
tion lines. The statistics in each figure are for regressions on the
8 to 25 kbar. The lines shown are one-to-one correlation lines. The
model data
data are from Bartels et al. (1991), Elthon and Scarfe (1984), Falloon
and Green (1987), and Kinzler and Grove (1992). Also included are
unpublished data from John Longhi and Rosamond Kinzler

activity-modifying parameter). Model P1 is thus rec- ing that the model may not extrapolate to these com-
ommended for P!¹ estimates of natural samples. positions. Also, predictions based on Elthon and Scarfe
The mean values of predicted pressures using model data are consistently low by 3—4 kbar. Run durations
P1 for data in the 8—25 kbar range match the measured were very short for these experiments (1 h) and equilib-
values with a SEE of 0.3 kbar (Fig. 4A). Takahashi’s rium may not have been achieved. Because the 20—25
(1986) 50 kbar experiment provides the only kbar portion of the test data set is small, statistics on
clinopyroxene-liquid pair synthesized from a mafic mean values are also reported for the 1 bar—16 kbar
bulk composition at a significantly greater pressure portion of the test data set in the appropriate figure
than those used for calibration. The model recovers captions.
a pressure of 50 kbar indicating that model P1 may To understand the uncertainty in P!¹ estimates
extrapolate to pressures higher than the P-range of the of natural samples it is necessary to understand the
regression data. sources of model error. Interlab pressure-calibration
Standard deviations for the 20 and 25 kbar portion problems, if systematic, might contribute to model in-
of the test data set are large compared to other pres- accuracy. The models are based on data from two
sures. These data comprise experiments performed on different laboratories using different pressure calib-
lunar analogs (John Longhi, unpublished data) and ration procedures (piston-in at LDEO vs piston-out
oceanic basalts (Elthon and Scarfe 1984). In the first at UTD) and yet the data regress well together. Fur-
study two runs at 20 and 25 kbar respectively with very thermore, test data from Kinzler and Grove (1992)
FeO-rich (33 and 27 wt%) and SiO -poor (36 and 34 fall within the SEE of the models. At least for these
2
wt%) liquids yield erroneously high pressures, indicat- three data sets little indication exists of a substantial P
102

nation of either analytical uncertainty in necessary

pyroxene and liquid compositions or from failure
to achieve reproducible equilibrium among the phases,
no matter how accurately they are analyzed. The
analytical strategy of choosing pyroxene rim composi-
tions is common to both model studies. The strategy
of Kinzler and Grove (1992) is to average a large
number of pyroxene analyses (usually '10) which
include some core compositions. As already noted,
their data fall within the SEE of the models (see 11,
13 and 16 kbar in Figs. 3 and 4). Except where noted,
error on P-predictions from other labs is normally
distributed about mean values that are identical to
measured pressure. Propagation of compositional ana-
lytical imprecision through the model shows that this
source of noise accounts for only about $0.5 kbar of
error, and is thus an inadequate source for the size of
the model SEE. Since the residuals from regression
analysis of the models are uncorrelated with P, ¹, or
composition, it is concluded that disequilibrium pyro-
xene compositions are the most significant source of
experimental error.

1 bar

One bar data were not incorporated into the thermo-

barometers and thus provide a test of how well the
models extrapolate to pressures below the calibration
range. The mean of the estimates (0.09 kbar) is quite
Fig. 4A, B The isobaric means of the predicted pressures of data close to 1 bar with a standard deviation of 2.3 kbar and
shown in Fig. 3A and B are plotted against measured pressure. a standard error of 0.3 kbar. Predictability of 1 bar data
A and B correspond to models P1 and P2 respectively. Also shown
are the mean values for 1 bar data. A one-to-one correlation line is was found to be study-dependent. Some individual
shown. The statistics given are for the regression of mean predicted studies yield mean estimates that are consistently and
vs measured pressures. For the 1 bar to 16 kbar portion of the data significantly higher than 1 bar.
R and SEE for P1 are 1.00 and 0.26 kbar respectively; for model P2 Systematically high pressure estimates in the 1 bar
R and SEE are 1.00 and 0.29 kbar. The 1 bar data are from Bender et
al. (1984), Grove and Juster (1989), Grove et al. (1992), Juster et al.
data set probably do not reflect analytical bias (i.e.,
(1989), Sack and Carmichael (1984), Sack et al. (1987), and Tormey et Na-loss from the glass during microprobe analysis) as
al. (1987). Also included are unpublished 1 bar data from John this error was not observed in the 8—25 kbar data.
Longhi and Rosamond Kinzler. The standard deviations (in kbar) Possible sources of error for the 1 bar data may involve
for model P1 (A) are as follows: 1 bar, 2.3; 8 kbar, 1.0; 10 kbar, 2.2; 11 the following, (1) Na-loss by volatilization may occur
kbar, 0.7; 12 kbar, 2.0; 13 kbar, 0.9; 15 kbar, 2.8; 16 kbar, 1.5; 20 kbar,
2.6; 25 kbar, 4.8
during long runs, especially if the sample size/surface
area ratio is small (see Tormey et al. 1987). Although no
continuous correlation between run time and error on
the pressure estimates exists, all experiments that ex-
inaccuracy in the model although precision of the ceed 500 h in length yield high predicted pressures.
estimates cannot be better than the $0.5 kbar quoted Reasonable estimates are also obtained in studies
for the two model studies. Temperature uncertainties where efforts were made to minimize Na-loss. (2) All
are more difficult to asses because experimental ¹ experiments run at ¹(1050° C have negative pressure
errors are vulnerable to cryptic difficulties like ther- estimates. The near-solidus temperatures required for
mocouple placement shifts, wire batch differences, and pyroxene saturation at 1 bar may lead to sluggish
contamination. Even so, little evidence exists to suggest diffusion rates, although it is unclear why this sort of
¹ inaccuracies in the models more serious than the error should be systematic. (3) Sector-zoning is a ubi-
quoted experimental precision of $10°C. And yet quitous feature of 1 bar experiments, but is undetected
SEEs of the models (1.36—1.51) are clearly larger in the near-liquidus experiments of this study. The
than the P!¹ uncertainties of the model input equilibrium pyroxene composition is obscure when sec-
data. Thus, uncertainties are introduced from a combi- tor-zoning is present.
 103

Prediction of thermodynamic properties compositional variables CaO-*2 and AlO-*2 for the res-
3@2
pective equilibria K[DiHd-Jd] and K[DiHd-CaTs].
The coefficients derived from regression analysis of Eqs As discussed above, the compositional variables may
6 and 11 may be simply related to known thermodyn- be treated as activity-modifying terms and are thus
amic quantities. The 1 bar values for HD* and HD* (en- presented as ln(X ) (Table 5).
7 & *
thalpy of vitrification and fusion respectively) and
HJ$ (Navrotsky 1981) were used to test thermodynamic
7
quantities recovered from the models. Comparisons are
Predicting experimental data and thermodynamic
made at either 1573 or 1600 K since the median and
properties
mean of experimental data used in the models (1573
and 1602 K respectively) are in this range. HJ$ is infer-
& Figure 5 shows the temperatures predicted for the re-
red from HJ$ assuming that the ratio H /H is the same
7 & 7 gression data for models T1 to T4. Models T1 and T2
for jadeite and diopside. The inferred value of HJ$
& (using K[DiHd-Jd]) are somewhat better than models
compares well with that recovered from models P1 and
T3 and T4 (using K[DiHd-CaTs]) based on slope and
P2 (Table 1). The entropy of fusion, SJ$, from model P2
& intercept values of predicted vs measured temperature.
also compares well with SJ$ calculated from C and
& 1 The SEEs for predictions of the test data are 40 and
entropy data of jadeite glass (Richet et al. 1993). The
30 K respectively for the pressure-independent (T1 and
volume of fusion, VJ$, recovered from model P2
& T3) and pressure-dependent (T2 and T4) models. The
(VJ$"23.4 cm3/mol) is similar to and bracketed by
& pressure-dependent models are thus slightly more accu-
estimates derived from the molar volume of jadeite
rate. The ability of model T2 to recover temperatures of
(Robie et al. 1979) combined with estimates of liquid
the test data is shown in Fig. 6.
partial molar volumes (Lange and Carmichael 1987;
Since most experiments are performed at similar
VJ$"23.5 cm3/mol) and the molar volume of jadeite
& target pressures, the test data may be conveniently
glass (83.16 cm3/mol reported by Richet et al. 1993;
grouped into isobaric data sets. Target temperatures
VJ$"22.7 cm3/mol).
& are much more variable (as can be seen by comparing
the distribution of data in Figs. 3 and 6) and only
a small portion of the test data can be grouped into
Thermometers
isothermal data sets containing more than 4 or 5 sam-
ples. Isothermal separation of the data thus provides an
The data set of this study was explored to determine if
inadequate testing procedure. We can, however, deter-
pyroxene-liquid exchange equilibria involving Mg and
mine the averaging effects on error by predicting both
Fe would provide a practical geothermometer. Func-
¹ and P simultaneously. Pressure predictions for the
tions analogous to the olivine-liquid equilibria of
test data in Fig. 4A utilize measured experimental
Roeder and Emslie (1970, Eqs. 11, 12 and 13) were
temperatures (recall that barometer P1 is temperature-
formulated for pyroxene-liquid equilibria but were not
sensitive). Figure 7 alternatively shows the mean of
nearly as successful as the DiHd-Jd exchange of Eq. 2.
pressure estimates from model P1 when temperatures
Equilibrium 2 was calibrated as a geothermometer
are estimated with model T1. These P-estimates are
(models T1 and T2 in Table 5) using the data of this
thus derived using only compositional information.
study combined with the Walter and Presnall (1994)
Figure 7 gives a measure of the total error when models
data and randomly selected 1 bar data. The DiHd-Jd
P1 and T1 are used in combination, as would be the
exchange equilibrium is also sensitive to pressure,
case with natural samples when only compositional
although the volume change of this exchange equilib-
information is available. Mean values of P-estimates
rium is much smaller than for Jd formation (Table 1).
calculated by this method may be compared to the
Thermometer T2 incorporates an empirical pressure
mean values calculated with reported temperatures
term. Addition of this pressure term reduces the SEE
(compare Fig. 7 to Fig. 4A). The SEE on the mean
from 27 to 24 K. The DiHd-CaTs equilibrium of Eq
values increases from 0.3 to 0.8 kbar on the 1 bar—16
3 was also calibrated as a thermometer (models T3 and
kbar portion of the test data. This 0.5 kbar increase in
T4 in Table 5); T4 incorporates an empirical pressure
SEE may be attributed to the error introduced by the
term. The use of 1 bar data in models T1 to T4 takes
temperature-estimate from model T1, which corres-
advantage of the wide temperature and bulk-composi-
ponds to an average error of about 15 K over the
tion ranges spanned by these experiments. The smaller
amount of error inherent in model P1.
number of data regressed for models T3 and T4 (93 vs
105 for models T1 and T2) reflects the loss of 1 bar data
where AlVI in excess of the amount necessary to form
jadeite did not exist (i.e., CaTs"0). Application to basalts from Hawaii
The compositional variable MgO-*2/(MgO-*2#
FeO-*2) (or Mgd) effectively reduces thermometric Many basaltic rocks lie on low-pressure differentiation
error. Thermometers are also improved by adding the trends suggesting crystallization at shallow levels in the
104

Fig. 5A–D Temperatures are

recovered for the data used in
the regressions. One-to-one
correlation lines and statistics
on model regressions are given
in each figure. A and B compare
the P-independent and
P-dependent models T1 and T2
respectively, both of which are
based on DiHd-Jd exchange.
C and D compare the DiHd-
CaTs-based models T3 and T4,
which are P-independent and
P-dependent respectively. The
data used to form the models are
identical to those used to form
the thermobarometers with
additional 1 bar data selected
from the following studies:
Grove et al. (1992), Juster et al.
(1989), Sack et al. (1987), Tormey
et al. (1987) and unpublished
data from Rosamond Kinzler

Fig. 6 Model T2 is used to predict temperatures of the test data. The Fig. 7 Mean values of predicted pressures are shown for the same
line shown is a one-to-one correlation line. A regression of measured data used to construct Fig. 3 and 4. Model T1 is used to estimate
vs predicted temperatures yields a slope of 0.99 and a zero intercept temperature. This temperature estimate, instead of the measured
of 17 K; the SEE and R are shown. The sources of the high pressure temperature, is then used in model P1 to estimate pressure. Statistics
portion of the test data are the same as those used in Fig. 3. 1 bar for the 1 bar—25 kbar data (mean predicted vs measured) are shown;
data were selected from the sources listed in Fig. 5 and chosen so a regression through the data from 1 bar to 16 kbar give an R of 1.00
that the test and model data sets do not overlap and an SEE of 0.8 kbar with a slope of 0.99 and an intercept of 0.06
kbar. Compare Fig. 4A

lithosphere. Some basalts, however, may bypass shal-

low level storage chambers and erupt directly from
greater depths. Phenocryst compositions may record We apply our thermobarometers to the problem of
these depths and may be used to calculate and compare transport depth by examining pyroxene and whole-
such transport depths from one volcanic suite or tec- rock host compositions from Mauna Kea, Hawaii
tonic environment to another. (Frey et al. 1991). Depth-estimates are then compared
 105

to estimates of elastic plate thickness and depth to 190 olivine-liquid pairs synthesized between 1 bar and
seismic activity beneath Hawaii to infer controls 100 kbar. The regression equation used was a variation
on magma transport. P!¹ estimates are compared of Eq. 5. We found compositional dependencies similar
to trace-element geochemistry to constrain magma to those noted by Gee and Sack (1988) and Langmuir et
genesis. al. (1992) as well as a slight P/¹ dependency that is
The compositional data (Frey et al. 1991) consist of consistent with the anticipated *V for the exchange
3
pyroxene rim and core compositions and whole-rock equilibrium (1.5 cm3/mol). The resulting function is
analyses. The rocks include post-shield ankaramites of
alkalic affinity (MU-9 and LP-5), post-shield tholeiitic 11.07*P
basalts (MU-2 and MU-8), a post shield high Fe—Ti ln K0-~-*2 [Mg—Fe]"!1.87#
D ¹
basalt (KI-12), and submarine shield tholeiites (MK1-8
and MK6-18). Together these basalts represent a phase #1.67*[SiO-*2 ]!14.11*[Na-*2*Al-*2] (13)
2
transitional between shield-building tholeiitic and post-
shield alkalic volcanism. To estimate pressure and tem- where P is in kbar and ¹ is in Kelvin (Fig. 9). Applying
perature (Fig. 8) model P1 was solved simultaneously our model to whole-rock compositions from Hawaii
with model T2. over a range of pressures (0—20 kbar) indicates that
equilibrium olivines crystallizing from whole-rock
compositions should yield K0-@8)0-%-3, [Mg-Fe]"0.30
D et al. (1980) model for
for all samples. Use of the Sack
Testing for closed system behavior
estimating ferric/ferrous ratios for an ankaramite
We first checked whether reported whole-rock whole-rock composition (MU-8) also shows that sub-
compositions can represent a liquid from which stantial deviation (16% error) of K0-~-*2[Mg—Fe] from
phenocryst pyroxenes crystallized. Comparing olivine 0.3 will occur for such samples onlyDif oxygen fugacities
and host whole-rock compositions allowed an initial approach Ni-NiO.
check that whole-rock compositions potentially repres- Only one sample for which P!¹ estimation was
ent a liquid. We compare measured versus model attempted, MU-9, showed a large discrepancy between
K0-~-*2[Mg—Fe] ([FeO0-*MgO-*2]/[FeO-*2*MgO0-]). the observed and expected value of K0-@8)0-%~3, [Mg—Fe].
D
The Roeder and Emslie (1970) model, based on experi- Olivine compositions from MU-9 Dyield an average
ments on basaltic bulk-compositions at 1 bar, demon- K0-@8)0-%~3,[Mg—Fe] of 0.42. Since this value is signifi-
D
cantly higher than 0.30, the whole-rock composition
strates a constant value for K0-~-*2[Mg—Fe] of 0.30
D was corrected by subtracting olivine of an average core
$0.03. We performed a multiple linear regression on
composition until K0-@8)0-%~3,[Mg—Fe]"0.30; the
amount of olivine removedD (6%) does not exceed the

Fig. 8 Model P1 is used in combination with model T2 to estimate

the pressure and temperature of pyroxene-bulk rock pairs from data
reported by Frey et al. (1991). Also illustrated is the mantle solidus of Fig. 9 Measured values of K0-~-*2[Mg—Fe] are plotted against
D
McKenzie and Bickle (1988). Sample numbers are from Frey et al. values of K0-~-*2[Mg—Fe] predicted from Eq. 13. The model SEE is
D
(1991). Samples MU-9 and LP-5 are postshield ankaramites. Sam- 0.03 and the correlation coefficient is 0.75. The model is based on
ples MU-2 and MU-8 are postshield tholeiitic basalts. Samples experimental equilibrium olivine-liquid pairs from: Gee and Sack
MK1-8 and MK6-18 are submarine shield tholeiites. Sample KI-12 (1988), Longhi and Pan (1988), Falloon and Green (1987), Bender et
is a postshield high Fe-Ti basalt. ‘‘MU-9 A’’ represents the P!¹ al. (1984), Kinzler and Grove (1992), Takahashi (1986), Grove et al.
estimate for MU-9 after compositional corrections were made; see (1992), Sack et al. (1987), and Tormey et al. (1987). Also included are
text for explanation. Error bars in upper left corner unpublished data from Rosamond Kinzler
106

modal amount of olivine. The resulting P!¹ estimate these samples from 34 km without further stagnation
is 12 kbar and 1550 K (40 km and 1277° C). This and petrogenetic activity is thus indicated.
‘adjusted’ P!¹ estimate is shown in Fig. 8 as MU-9 These parental tholeiite (MU-8) and alkalic basalts
A, and is much nearer to P!¹ estimates for samples (LP-5) have identical SiO , TiO and CaO/Al O , and
2 2 2 3
LP-5 and MU-8. No corrections were made for the have similar Mgds (61 and 63 respectively). This obser-
remaining samples. Notably, all samples except KI-12, vation and the similar P-estimates for these samples
plot to the high-temperature side of the anhydrous suggest that trace- and alkali-element contents are es-
garnet peridotite mantle solidus of McKenzie and tablished below depths of 34 km. Clague (1987) has
Bickle (1988). Thus the P!¹ solutions seem phys- hypothesized the existence of a magma storage zone at
ically reasonable. Frey et al. (1991) presented evidence 20 km (the base of the crust) based on xenolith popula-
for disequilibrium between phenocrysts and whole- tions. Our P estimates are consistent with a deep level
rock for sample KI-12; this sample is probably not of stagnation, but place this ponding level at a depth
amenable to a simple corrective scheme. Disequilib- greater than 20 km.
rium may be the reason sample KI-12 does not plot on
an otherwise coherent P!¹ trend.

Petrological implications Comparisons to geophysical data

Where in the mantle and by what processes do alkalic Controls upon magma transport may be inferred by
and tholeiitic rocks gain their characteristics? Ther- comparing pressure and corresponding depth estimates
mobarometers may provide constraints. Sample MU-8 to seismic data. The maximum depth-estimate (40 km
is argued by Frey et al. (1991) to be the ‘‘the best assuming 3.3 km/kbar) for lavas from Mauna Kea
example of a parental tholeiitic melt’’ while sample corresponds to depths inferred for magma-related
LP-5 is considered ‘‘clearly related to the alkalic lavas’’ tremor (Aki and Koyanagi 1981) and a low velocity
(due to comparatively high La/Yb and Zr/Nb). zone beneath Hawaii (Ryan 1988) and thus, unsurpris-
P!¹ estimates (based on core compositions and ingly, relate such seismic phenomena to the presence of
whole-rock analyses) for LP-5 and MU-8 are identical melt at these depths.
within model error. Interestingly, use of pyroxene rim All magma transport depths, except MU-2, exceed
compositions for both samples results in P-estimates depth estimates for the base of the crust (18 km; Hill
identical to those based on pyroxene cores (the whole- and Zucca 1987). Also, Hill and Zucca (1987) infer
rock composition of LP-5 is corrected for pyroxene a density for the lower crust of 2.9 g/cm3; the base of the
crystallization when pyroxene rims are used in the crust does not provide a level of neutral buoyancy for
model; MU-8 is 98.5% glass). Temperature-estimates basaltic magma with a density"2.7 g/cm3. The range
from rims and cores are also identical for MU-8 and of depth estimates, however, has a shallow bound that
LP-5; the thermometer is insensitive to the small vari- roughly coincides with the long-term elastic thickness
ations in temperature with the existing data set. Since (T ) of the lithosphere (25 km; Watts et al. 1985).
%
pyroxene saturation occurs at lower temperatures for Depths for extraction of magma may thus be sensitive
lower pressures we might expect re-solution of pyro- to the mechanical behavior of the lithosphere. For
xenes rather than growth during upward adiabatic appropriate mantle strain rates (10~14—10~18/s) the
transport, thus, similar P-estimates using rim and core elastic thickness may represent the depth of ductile
compositions are not surprising. Rims yielding lower behavior and a sharp decrease in strength (Kirby 1980;
P-estimates may occur if magma stalls at shallow Bodine et al. 1981). The match of the lower limit of
depths, but time of stagnation must be long enough for depth estimates with T suggests that magma transport
%
temperature to fall and for pyroxene to become re- is controlled, through temperature, by the depth to the
saturated (as possibly occurred for MU-2 (93.2% glass); brittle-ductile transition, and further implies a similar
P!¹ estimates for MU-2 are 3.2 kbar and 1426 K control for both fracture propagation and long-term
using the core composition, and 0.06 kbar and 1402 K lithospheric strength.
using the rim). We suspect that pyroxenes in MU-8 and When P-¹ estimates are obtained from other vol-
LP-5 experienced most of their growth at a single depth canoes and combined with age constraints, both petrol-
and were transported to the surface with insufficient ogic models and comparisons to geophysical data can
time for significant growth and/or re-equilibration. be further refined. Application of the ther-
A combined average taken from core and rim composi- mobarometers to rocks from Mauna Kea nevertheless
tions from both LP-5 and MU-8 likely provides the demonstrate (1) the use of thermobarometry in con-
best P!¹ estimate. Calculated in this way the mean straining the origin of alkalic vs tholeiitic magma, (2)
P and ¹ estimates (with standard deviations) are: that magma may be delivered to a planetary surface
10.2$0.8 kbar and 1530$7.0 K (about 34 km at from substantial depth and (3) ponding of magma may
1257° C; assume 3.3 km/kbar). Direct transport for be controlled by the elastic thickness of the lithosphere.
 107

ated by Linnea Norby. Numbers of experiments were carefully and

Summary thoughtfully counted by Noelani Putirka. Supported by grants from
New experiments in the range 8 to 30 kbar produced NSF and DOE. LDEO contribution d5410.
clinopyroxene-liquid pairs whose compositional sys-
tematics were used to construct various geothermo-
barometers. These expressions have a thermodynamic
basis and thermodynamic properties can be calculated References
from coefficients of the expressions. These thermody-
namic properties, *HP0, *SP0 and *V , have values Aki K, Koyanagi R (1981) Deep volcanic tremor and magma ascent
3 3 3 mechanism under Kilauea, Hawaii. J Geophys Res 86 :
similar to 1 bar literature values. As expected, the Jd-liq 7095—7109
equilibrium, which has a high *V , proved to be the
3 Bartels KS, Kinzler RJ, Grove TL (1991) High pressure phase
basis of a useful geobarometer. Exchange equilibria relations of primitive high-alumina basalts from Medicine Lake
involving solution of Jd and CaTs into DiHd are less volcano, northern California. Contrib Mineral Petrol 108 :
sensitive to pressure and were calibrated as thermo- 253—270
Bender JF, Langmuir CH, Hanson GN (1984) Petrogenesis of basalt
meters. glasses from the Tamayo Region, East Pacific Rise. J Petrol
The thermobarometric models reliably reproduced 25 : 213—254
pressures and temperatures of experiments from other Bodine JH, Steckler MS, Watts AB (1981) Observations of flexure
labs. Error analysis indicates that P and ¹ estimates and the rheology of the oceanic lithosphere. J Geophys Res
using the preferred models, P1 and T1, (Table 5) are 86 : 3695—3707
Bottinga Y, Weill DF (1972) The viscosity of magmatic silicate
accurate to at least $1.4 kbar and $27 K on indi- liquids: a model for calculation. Am J Sci 272 : 438—475
vidual pyroxene/liquid pairs, and perhaps as good as Clague DA (1987) Hawaiian xenolith populations, magma supply
$1.0 kbar and $15 K when averages over multiple rates, and development of magma chambers. Bull Volcanol
equilibrium pyroxene/liquid pairs are used. The pres- 49 : 577—587
sure-dependent thermometer, T2, (Table 5) may pro- Drake MJ (1976) Plagioclase-melt equilibria. Geochim Cosmochim
Acta 40 : 457—465
vide more accurate estimates. Since test and model data Elthon D, Scarfe CM (1984) High-pressure phase equilibria of
encompass a wide range of bulk compositions, the a high-magnesia basalt and the genesis of primary oceanic ba-
thermobarometers should be generally applicable to salts. Am Mineral 69 : 1—15
mafic igneous rock compositions. Falloon TJ, Green DH (1987) Anhydrous partial melting of MORB
pyrolite and other peridotite compositions at 10 kbar: implica-
Application of the models to rocks from Mauna tions for the origin of primitive MORB glasses. Mineral Petrol
Kea, Hawaii yield transport depths up to 40 km for 37 : 181—219
ankaramites. Also, samples MU-8, a parental tholeiite Fram MS, Longhi J (1992) Phase equilibria of dikes associated with
(Frey et al. 1991) and LP-5, a parental alkalic basalt, Proterozoic anorthosite complexes. Am Mineral 77 : 605—616
yield identical pressure-estimates (10.2 kbar). The simi- Frey FA, Garcia MO, Wise WS, Kennedy A, Gurriet P, Albarede
F (1991) The evolution of Mauna Kea Volcano, Hawaii: pet-
lar pressure estimates, but different trace-element con- rogenesis of tholeiitic and alkalic basalts. J Geophys Res
tents for these samples, indicate that tholeiitic and 96 : 14347—14375
alkalic volcanic rocks at Hawaii inherit their trace- Gee LL, Sack R (1988) Experimental petrology of melilite
element signatures below 34 km. The agreement of nephelinites. J Petrol 29 : 1233—1255
maximum depth-estimates for mafic rocks from Hawaii Ghiorso MS, Carmichael ISE, Rivers ML, Sack RO (1983) The
Gibbs free energy of mixing of natural silicate liquids; an ex-
with depths observed for tremors and the low velocity panded regular solution approximation for the calculation of
zone suggest that volcanic rocks may be delivered from magmatic intensive variables. Contrib Mineral Petrol 84 :
substantial depths without significant re-equilibration 107—145
of liquid and coexisting pyroxene. Depth estimates also Grove TL, Juster TC (1989) Experimental investigations of low-Ca
have a shallow bound that corresponds to elastic plate pyroxene stability and olivine-pyroxene-liquid equilibria at 1-
atm in natural basaltic and andesitic liquids. Contrib Mineral
thickness. This indicates that the mechanical behavior Petrol 103 : 287—305
of the lithosphere, not density contrasts, controls Grove TL, Kinzler RJ, Bryan WB (1992) Fractionation of mid-ocean
magma storage at depth below Hawaii. This latter ridge basalt (MORB). Geophys Monogr Am Geophys Union
observation implies the utility of igneous thermo- 71 : 281—310
Hill DP, Zucca JJ (1987) Geophysical constraints on the structure of
barometry not only towards disentangling problems of Kilauea and Mauna Loa volcanoes and some implications for
petrogenesis but also regarding tectonic issues. seismomagmatic processes. US Geol Surv Prof Pap 1350 :
903—917
Juster TC, Grove TL, Perfit MR (1989) Experimental constraints on
Acknowledgements We thank Barbara Molfino for advice concern- the generation of FeTi basalts, andesites and rhyodacites at the
ing statistical matters. Comments by, and discussions with Charlie Galapagos Spreading center, 85°W and 95°W. J Geophys Res
Langmuir, George Harlow and Robert Fogel were extremely helpful 94 : 9251—9274
as were discussions with Daniel Farber, Mark Davis and Jennifer Kinzler RJ, Grove TL (1992) Primary magmas of mid-ocean ridge
Reynolds. The manuscript was improved from reviews by Michael basalts 1. Experiments and methods. J Geophys Res 97 :
Walter and Roger Nielsen. The ankaramite, ugandite and Mid- 6885—6906
Atlantic ridge basalt samples used as starting compositions were Kirby SH (1980) Tectonic stresses in the lithosphere: constraints
generously donated by Fred Frey, Richard Sack and Jennifer provided by the experimental deformation of rocks. J Geophys
Reynolds respectively. Electron microprobe work was greatly facilit- Res 85 : 6353—6363
108

Lange RA, Carmichael ISE (1987) Densities of Na O— Roeder PL, Emslie RF (1970) Olivine-liquid equilibrium. Contrib
2
nK O—CaO—MgO—FeO—Fe O —Al O —TiO —SiO liquids: Mineral Petrol 29 : 275—289
2 2 3 2 3 2 2
new measurements and derived partial molar properties. Geo- Ryan MP (1988) The mechanics and three-dimensional internal
chim Cosmochim Acta 51 : 2931—2946 structure of active magmatic systems: Kilauea volcano, Hawaii.
Langmuir CH, Klein EM, Plank T (1992) Petrological systematics of J Geophys Res 93 : 4213—4248
mid-ocean ridge basalts: constraints on melt generation beneath Sack RO, Carmichael ISE (1984) Fe2"Mg2 and TiAl "MgSi
2 2
ocean ridges. Geophys Monogr Am Geophys Union 71 : 183—280 exchange reactions between clinopyroxenes and silicate melts.
Lindsley DH (1983) Pyroxene thermometry. Am Mineral 68 : Contrib Mineral Petrol 85 : 103—115
477—493 Sack RO, Carmichael ISE, Rivers M, Ghiorso MS (1980) Ferric-
Longhi J, Pan V (1988) A reconnaisance study of phase boundaries ferrous equilibria in natural silicate liquids at 1 bar. Contrib
in low-alkali basaltic liquids. J Petrol 29 : 115—147 Mineral Petrol 75 : 369—376
McKenzie D, Bickle MJ (1988) The volume and composition of melt Sack RO, Walker D, Carmichael ISE (1987) Experimental petrology
generated by extension of the lithosphere. J Petrol 29 : 625—679 of alkalic lavas: constraints on cotectics of multiple saturation in
Navrotsky A (1981) Thermodynamics of mixing in silicate glasses natural basic liquids. Contrib Mineral Petrol 96 : 1—23
and melts. In: Newton RC, Navrotsky A, Wood BJ (eds) Thermo- Stebbins JF, Carmichael ISE, Moret LK (1984) Heat capacities and
dynamics of minerals and melts, Springer New York, pp 189—206 entropies of silicate liquids and glasses. Contrib Mineral Petrol
Nielsen RL, Drake MJ (1979) Pyroxene-melt equilibria. Geochim 86 : 131—148
Cosmochim Acta 43 : 1259—1272 Takahashi E (1986) Melting of a dry peridotite KLB-1 up to 14 GPa:
Perfit MR, Langmuir CH, Baekisapa M, Chappell B, Johnson RW, implications on the origin of peridotitic upper mantle. J Geophys
Staudigel H (1987) Geochemistry and petrology of volcanic rocks Res 91 : 9367—9382
from the Woodlark Basin: addressing questions of ridge subduc- Tormey DR, Grove TL, Bryan WB (1987) Experimental petrology of
tion. In: Taylor B, Exon NF (eds) Marine Geology, geophysics, normal MORB near the Kane Fracture Zone: 22°—25° N, mid-
and geochemistry of the Woodlark Basin-Solomon Islands. Cir- Atlantic Ridge. Contrib Mineral Petrol 96 : 121—139
cum Pacific Council for Energy and Mineral Resources, Earth Walter MJ, Presnall D (1994) Melting behavior of simplified lher-
Sci Series 7, Houston Texas, pp 113—154 zolite in the system CaO-MgO-Al O -SiO -Na O from 7 to 35
2 3 2 2
Richet P, Robie RA, Hemingway BS (1993) Entropy and structure of kbar. J Petrol 35 : 329—359
silicate glasses and melts. Geochim Cosmochim Acta 57 : Watts AB, ten Brink US, Buhl P, Brocher TM (1985) A multichannel
2751—2766 seismic study of lithosphereic flexure across the Hawaiian—
Robie RA, Hemingway BS, Fisher JR (1979) Thermodynamic prop- Emperor seamount chain. Nature 315 : 105—111
erties of minerals and related substances at 298.15 K and 1 bar Zhang Y, Walker D, Lesher CE (1989) Diffusive crystal dissolution.
(105 Pascals) pressure and at higher temperatures. US Geol Surv Contrib Mineral Petrol 102 : 492—513
Bull 1452 : 456