American Mineralogist, Volume 64, pages 1107-1114, 1979
Nickel partitioning betweenolivine and silicate melt: Henry's law revisited
BronN O. MyseN
Geop hysical Lab or at ory, Ca r negie I nst itut i on of lV'ashingt on
lhashington, D. C. 20008
Abstract
Partition coefficients for nickel between olivine and an aluminous anhydrous silicate melt
have been determined as a function of nickel content under isothermal conditions (1300"C)
at pressures between I atm and 20 kbar. The partition coefficients are independent of nickel
content until 1000 ppm Ni is dissolved in the forsterite. No pressure-dependence of this concentration limit was observed. It is concluded that a maximum of about 1000 ppm Ni can dissolve in forsterite according to Henry's law. The partition coefficients are pressure-dependent
at both higher and lower nickel concentration. The pressure-dependence is less pronounced
in the olivine with nickel content above 1000 ppm because the activity coefficient of nickel in
the coexisting liquid varies with pressure, presumably as a consequence of pressure-induced
structural changes of the silicate melt.
Fractional crystallization of olivine results in a lowering of the nickel content of the
magma, but the extent to which such fractionation results in a lowered nickel content depends on both the nickel content of the silicate magma and the confining pressure. It is concluded that the nickel content of natural magma derived from the upper mantle is commonly
controlled by fractional crystallization that involves minerals other than the major rockforming minerals.
Introduction
The nickel content of olivine-bearing natural magmas is considered a sensitive petrogenetic indicator
(e.g. Hiikli and Wright, 1967; Leeman, 1974; Hart
and Davis, 1978;O'Hara et al., 1975)becauseof the
importance of olivine in igneous processes and because of the large values of olivine-liquid partition
coefficientsfor nickel (D,ii "'). Recently, the interpretation of nickel partitioning data when applied to
such problems has been questioned (Mysen, 1978a)
because of the apparent dependence of D$]'o on temperature (Hekh and Wright, 1967; Leeman, 1974),
bulk composition (Hart and Davis, 1978; Arndt,
1977), and pressure (Mysen and Kushiro,1979), and
because of the possibility that Di'."o is a function of
the nickel content of olivine in the natural abundance range of nickel in magmatic olivine (Mysen,
1976a, 1978a). There is little agreement about the
quantitative values of any of these functional relationships.
One of the least well defined variables is the dependence of Dil "q on the nickel content of olivine. It
has been suggested that nickel dissolves according to
000t-0a4x/ 79/ 09I 0- l I 07$02.00
Henry's law only at or below about 1000 ppm Ni in
olivine (Mysen, 1976a,1978a).Hart and Davis (1978)
worked at I atm in the nickel concentration range
measurable with the electron microprobe and reported no dependence of D$]'o on nickel concentration. On the basis of their results, they suggested that
previous data (Mysen, 1976a, 19782) obtained at
lower nickel concentration and at higher pressure
might need reinterpretation. Lindstrom and Weill
(1978), using a technique similar to that of Hart and
Davis (1978), also argued that the data of Mysen
(1976a, 1978a) might need reevaluation. Finally,
Shaw (1977), summarizing some pertinent experimental data on element partitioning, concluded that
the data of Mysen (1978a) were open to question but
did not attempt to explain how he reached this conclusion.
Experiments involving nickel partitioning are
complicated because of the extensive solubility of
nickel in most of the sample containers used. This
problem existed in all the published work on nickel
partitioning summarized above, although the extent
ofthe problem depended on the experirnental conditions.
I 107
I 108
MYSEN: NICKEL PARTITIONING
In view of the above arguments, it was decided to
conduct experiments where at least some of the experimental problems could be avoided. Experiments
were conducted at pressures ranging from I atm to 20
kbar, because all previous data reporting concentration dependenceof Dil "o were at high pressure,
whereas all those claiming no dependence of the
nickel partition coefficients on the nickel concentration were at I atm. The present data would therefore
help rectify the lack of overlap between the published data sets.
Experimental technique
The starting composition (Fo,oJdro,by weight) was
chosen becauseat 1300'C (the experimental temperature) the liquidus of this composition does not
change within experimental uncertainty with pressure up to 20 kbar (Schairer and Yoder, 196l; Mysen
and Kushiro, 1979). The uncertainty of the melt
composition with this technique correspondsto +1.5
percent forsterite component. Electron microprobe
analysis of the melt was not employed because of
problems with loss of Na by volatilization even with
0.01 pA beam current at l5 kV accelerationpotential.
Significant broadening of the electron beam to lower
the temperature on the surface during analysis could
not be done with confidence because of interference
from neighboring forsterite crystals. Only a small
percentage(10 percent) of pure, quite small (-10 pm)
forsterite crystals is formed, thus making attainment
of equilibrium less cumbersome than in previous experiments (e.9. Mysen, 1978a). The experimental
conditions were the same as those of Mysen and
Kushiro (1979), with the exception that l-atm experiments were added to the program and a range of
nickel concentrations was employed(2-32N ppm Ni).
Two sets of starting materials were used. For forward experiments, the starting material was a glass of
ForoJdro composition with l0 ppm u'Ni and various
amounts of nonradioactive nickel. This glass was
made by fusing oxide mixtures, thoroughly ground,
for 2 hr at 1450'C in air. The material was checked
for homogeneity of all elements prior to the experiments. For the reversal experiments, 20 weight percent crystalline forsterite with 10-16,000 ppm Ni in
solution was mixed with 80 weight percent nickelfree glass of NaAlSi,Ou composition (10-16,000 ppm
Ni in forsterite corresponds to 2-3200 ppm in forsterite glass mixture). The forsterite was synthesized
from an oxide mixture at 2O kbar and 1650"C for I
hr and checked for homogeneity. In all experiments,
the forsterite contained 10 ppm u'Ni and different
amounts of nonradioactive nickel to make up the total nickel concentration.
The experiments at l0 and 20 kbar were conducted
in the solid-media, high-pressure apparatus (Boyd
and England, 1960), whereas the l-kbar experiments
were carried out in the gas-media, high-pressure apparatus (Yoder, 1950). The sample containers were
BN capsules sealed inside Pt capsules. The BN capsules had been fired at 1000'C and stored at llOoC
before use to decompose the reducing ammoniabased binder of the BN and to avoid adsorption of
HrO from the moisture of the air before the experiments. With this sample-container configuration, the
experimental charges were never in physical contact
with the outside platinum capsule; thus the problem
of Ni loss from the experimental charge to the container was eliminated. At the same time, no reducing
agent was present that could cause reduction of Ni'*
to metallic Ni. Mass-balance calculations were conducted on a routine basis to assess the extent to
which this experimental design was successful. The
results of these calculations are shown in the run tables. The uncertainty (+lo) of the mass-balanced
nickel values includes all analytical uncertainties
(about 3 percent of the individual analyses), weighing errors (negligible), and other experimental errors
( 1 kbar and +lOoC at P > l0 kbar, negligible pressure uncertainty and a5oC at P: I kbar).
The l-atm experiments were conducted in a vertical quench furnace in air with a temperature uncertainty of -r2"C using 0.1 mm Pt loops as sample
holders. The high air pressure inside the BN-Pt capsules at 1300'C and I atm resulted in rupture of the
capsule during the experiment. With open capsules,
the sample migrated out of the capsule during the experiment. The technique was therefore not used in latm experiments.
About 20 mg of sample was used in these experiments. Larger samples tended to fall off the loop because of the low viscosity of the melt. Theoretically,
there is a small nickel loss to the Pt loop during the latm experiments. The absolute amount of Ni thus
lost appears to have been so small, however, that it
did not appear in the mass-balance calculations
within the quoted uncertainty. It should also be
noted that because there may have been some Na
loss by volatilization, the proportions of phases may
be slightly different from the theoretical value. At the
same tfune the liquid composition may have been
slightly off the join. This problem is not considered
I r09
MYSEN: NICKEL PARTITIONING
serious,however, in the light of experimental data
relevantto the problem by Seifertet al. (1979).
The nickel analyseswere carried out with the betatrack mapping technique (Mysen and Seitz, 1975),
using u3Nias the sourceof beta particles and llford
K-5 nuclear emulsionsas detectors.The applicability
and accuracyof this techniquefor nickel analysisof
experimental run products have previously been
demonstrated(Mysen and Seitz, 1975; Mysen,
1978a).
Reversal experimentswere conductedat all pressures and with three different nickel concentrations
in olivine to ensurethat D$;'tc at low, intermediate,
and high nickel concentration representedequilibrium values. The low nickel concentration chosen
was well within the range where results of forward
experimentshad indicatedthat D;', "o was independent of nickel content.An intermediateconcentration
(1000-3000ppm Ni in olivine) was chosenbecause
this was the range where forward experimentshad
indicatedthe most rapid changeof Ditatiqwith nickel
concentration.The third reversalexperimentat each
pressurewas carried out in the high nickel concentration range (=10,000 ppm Ni in olivine), wher.ethe
value of Di,tr"oreacheda minimum level and was relatively insensitiveto changingnickel concentration.
The agreementbetweenforward and reversalexperimentswithin experimentaland analytical uncertainty is taken to indicate that equilibrium was
closelyapproximatedin all the experiments.According to Clark and Long (1971),the nickel diffusion coefficient in olivine at l300oC is about 6 x l0-tt cm2/
sec. Using the simple relationship (Hofmann and
Magaitz,1977)
X: JDt
(l)
where 1is the distance,D is the diffusion coefficient,
and r is the time, l equalsabout l0 pm in 4 hr. With
crystalsof about 5 pm radius, the transport distance
is about 5 pm, indicating that a run duration of 4 hr
is more than sufficient to attain equilibrium. This
conclusionis also substantiatedby the homogeneity
of the phasesand the small analytical uncertainty of
the partition coefficients(near the theoretical uncertainty, Mysenand Seitz,1975).
Results
All experimentaldata are shown in Table l, and
the resultsare plotted as partition coefficientsagainst
the nickel content of forsterite(Gl,) in Figure I, in
which reversal experfunentsare shown with open
symbols. The size of the symbols reflects the uncertainty of the data 1+16;.
Two features of these data warrant comment.
First, there is a limited nickel concentration range
where the value of DfrHiq is independent of nickel
concentration (<1000 ppm Ni in forsterite). Second,
the value of the partition coefficient in the concentration range used by Mysen and Kushiro (19'79) (=tOO
ppm Ni) shows the same dependence on pressure observed by them. The limited concentration range of
constant Di|iq occurs at all pressures studied (including I atm). There is no pressure-dependence of the
width of this concentration range. This conclusion
contrasts with that of Mysen (1978a), who suggested
that the nickel concentration range of constant partition coefficient may expand with decreasing pressure
in the system plagioclase (An'oAb'o)-forsterite-silica-water in the range 10-20 kbar at 1025'-1075oC.
It should be remembered, however, that Mysen
(1978a) did not have a close bracket on the absolute
extent of this concentration range. Furthermore,
those data (Mysen, 1978a) were acquired in a hydrous system, whereas the results reported here are in
the absence of HrO.
There is only a limited concentration range of constant Dir|iq even at I atm pressure. At this pressure,
previous results (Leeman, 1974; Hart and Davis,
1978; Lindstrom and Weill, 1978) in the concentration range attainable with the electron microprobe
indicated that the olivine-liquid partition coefficient
remained constant until several weight percent Ni
had dissolved in the olivine. None of those data was
extended significantly below a concentration of 1000
ppm. As can be seen from the data in Figure I, approximately 1000 ppm Ni can dissolve in forsterite
before the partition coefficients become dependent
on the nickel content even at I atm pressure.
The extent of the change of D;;ric with nickel content varies with pressure (Table 2). This change is
likely to reflect the structural change of the liquid
coexisting with forsterite, resulting in varying activity
coefficients of nickel in the melt (Mysen and Kushiro, 1979). From Nernst's distribution law, it can
be shown that
DSt tio- (d".1,/4ii1)/(yi:,/Tli:)
Q)
are activities of nickel in olivine
wherea*, and ail,ro
yil,
and yiil, respectively,are the acand liquid, and
tivity coefficientsof nickel in olivine and liquid. According to the data of Mysen and Kushiro (1979),the
activity coefficient of nickel in silicate melt such as
lll0
MYSEN: NICKEL PARTITIONING
Table l. Experimentalresultsr
Run
No.
9 50 * *
926
925
943*x
934
952
935
933
9 5 4* *
(kbar)
0. 001
0. 001
0. 00r
0.001
0. 001
0. 001
0. 001
0.001
0.001
915**
859
858
857
9 1 6* *
921
856
855
928
r0
Ni added
(ppm)
Ni in olivine
(ppm)
Ni
in liquid
(ppm)
2.7
l0
50
200
300
500
1000
3000
3200
17.4 I 0.08
55x2
276 ! r0
1370 ! 40
1710 1 39
2092 ! L]L
5050 1 300
13980r 500
17853r 706
2
10
50
100
200
300
500
1000
3000
LL.1 t
57.6 r
290 1
607 !
IL27 !
1815 r
2727 !
5913 t
16210r
0.4
L8
t0
22
36
31
LOL
257
550
0.53 ! 0.02
2.63 1 0.08
LL.] ! 0.4
25.5 r r.0
59r1
1 0 0 . 1t 1 . 8
r48!6
429 ! t4
1480 1 71
0.96 ! 0.05
3.L9 ! 0.77
15.9 1 0.9
79 ! 4
11114
L44!4
427 ! 2L
1232! 44
1497 ! 85
Dl,ri
o I-Iio
l8.l 1
7 7. 2 !
t7.4 x
17.3 1
15.4 t
14.5 t
11.8 r
1 1. 3 1
11.9 1
1.3
I.0
L.I
1.0
0.7
0.8
0.9
0.6
0.8
Ni by mass balance
(pp*)
?1+n?
9.9 ! 0.9
50 14
24t ! 25
3t9 ! 22
400 ! 35
1028 r 103
2890 ! 2890
3600 r 400
22.M.L
2L.9 ! I.O
24.8 ! r.2
23.8 ! r.3
19.1 ! 0.7
L8.2 ! O.4
18.4 r 1.0
13.8 l 0.8
1 1 . 0 1 0 .7
2.0 ! 0.2
9.5 1 0.9
46!5
98i9
r98 t 16
323 ! 27
470 ! 42
1115 1 ll0
3400 I 300
907**
854
852
9L7
880**
929
848
931
949*x
10
10
10
10
l0
10
10
10
2
l0
50
100
200
300
500
3000
3200
10.5 t 3
4 7. 6 ! L . 3
242!8
576 t t5
rL27 ! 36
L529 ! 67
2496 ! 87
110111 511
12400 ! 479
0.50 r 0.02
2.34 ! 0.06
12.0 1 0.3
28.2 ! L.9
59 1l
85!2
L45!5
1187 1 36
1578 I 51
21.0 r
20.3 t
20.2 !
20.4 !
19.1 1
18.0 t
L 7. 2 !
9.3 t
7.9 t
1.0
0.8
0.8
r.5
0.7
0.9
0.8
0.5
0.4
1.8 r 0.2
8.2 r 0.8
42t4
99 18
198 1 16
213 ! 22
245L ! 40
2464! 222
2985 x 269
875**
853
851
849
879
930
847
845
932
948**
20
20
20
20
20
20
20
20
20
20
2
10
50
100
200
300
500
r000
3000
3200
8.9 r
55.1 !
273 !
536 !
854 !
t-141 1
1791I
2595 I
9800 t
1 0 8 0 0r
0.88 1 0.03
5.2r ! 0.14
2 7. 2 L 0 . 7
51.1 1 1.8
10113
r37!6
246!7
519 r 16
7397 ! 36
L769 ! 8L
10.1 1
r0.6 1
10.0 1
10.5 I
8.5 r
8.3 !
7.3 t
6.9 !
7.O!
6.1 1
0.5
0.4
0.4
0.5
0.4
0.6
0.3
O.4
O.4
0.3
I.9 ! O.2
11 11
57!5
111 1 10
199 ! 20
268 x 24
440 x 40
904 i 90
2489 ! 200
2940 ! 260
*? = 1300"C and rime
*JtReversal
run.
= 4 hr
in
all
0.3
1.6
LO
20
32
36
53
130
400
316
experiments.
that usedhere increaseswith pressureup to about 5l0 kbar before it decreaseswith further pressureincrease.If it is assumedthat yil shows the same dependence on nickel composition at all pressures,
equation 2 shows that the degree of alteration of
D$i"o with changing nickel concentrationwould increasewith increasingpressureuntil about l0 kbar
and then begin to decreaseas the pressureis increasedfurther. This prediction is in accordancewith
observation (Table 2). lf it is assumedthat the expressionof Hart and Davis (1978)relating MgO contents of melts to D3t.tic,
D31-'c-#-on
(3)
is valid for the present compositions, the uncertainty
in MgO content (+0.9 weight percent, derived from
+1.5 weight percent uncertainty of Fo content of the
melt) over this pressure interval results in about a 20
percent uncertainty of the pressure-dependence of
D$, "o. The data in Figure I show that the pressure effects discussed here are greater than2O percent. Consequently, the variations cannot be explained in
terms of analytical uncertainty.
Because the rate of change of DflHiq as a function
of nickel concentration is also a function of pressure
at constant temperature (1300"C) and liquid bulk
composition, the pressure-dependence of D*i'tq depends on the nickel content of the system (Fig. 2).
Note that the maximum D values occur at lower
lltl
MYSEN: NICKEL PARTITIONING
J
a
{
oZ
o
15
6z
'
5
5
15
10
P r e s s u r ek, b o r
20
Fig. 2. PartitioncoefficientDflitto as a function of pressureat
ppmNi in olivine.
300,3000,and 10,000
Fig. l. Partition coefficient Df,fLs as a function of nickel content in olivine at pressuresfrom I atm to 20 kbar and a temperatureof l300oC.
pressurethe higher the nickel content.This shift may
relate to the structural role of Ni in the melt. No data
are available, however,for further discussionof this
point. The data points in Figure 2 have been connected with smooth, dashed lines. Mysen and Kushiro (1979) showed a discontinuouscurve with increasingpressureat Cn| < 100 ppm based on more
closely bracketed data. The main discontinuity was
between 15 and 20 kbar. Such a discontinuity is in
accord with other data involving pressure-induced
structural changesin NaAlSirO.-rich liquids (Mysen,
Table 2. Percentageofchange ofD$;tie between 1000and 10,000
ppm Ni in forsterite as a function ofpressure
P
(kbar )
0 .0 0 r
I
10
20
Percentage
of change relative
value
at 1000 ppn Ni
-33
-47
-55
-36
to
1976b Kushiro, 1976; Mysen and Virgo, 1978;
Sharmaet al.,1978). The presentdata can also be fitted to such a curve. but in the absenceof a closer
bracket sucha fit was not attempted.
Discussion
In view of the somewhat surprising observation
that Dif iqappearsto dependon nickel concentration
at much lower levels than previously expected,it is
useful first to discussthe presentdata in the light of
previous considerations of experimental problems
that could have affectedthe results.
Lindstrom and Weill (1978)arguedthat nickel loss
to the container could be a problem in experiments
such as these, as also discussedby Mysen (1978a).
Mysen (1978a)suggestedthat nickel losswould most
likely causethe partition coefficientsto increasewith
increasing nickel content, whereas Di"'il in fact,
showed a decrease.Hart and Davis (1978) commented, however, that Pt containers effectively becomeNi-saturatedat relatively low nickel concentration. They suggestedthat this apparent nickel
saturation resulted from the formation of a surface
layer of nickel-saturatedplatinum, making the remaining platinum unavailable as a Ni sink. According to the hypothesisof Hart and Davis (1978),the
effectivenickel loss to the container from the experi-
rtt2
MYSEN: NICKEL PARTITIONING
mental chargeswould decreasewith increasingnickel liquid remained constant over extensiveconcentracontent of the system.In view of the more rapid dif- tion intervals. Furthermore, Mysen (1978b) pointed
fusion of nickel in liquid than in crystals(Clark and out that growth-induced metastable defects would
Long, l97l; Hofmann and Magaritz,1977),Hart and anneal with time. The results of time studies by
Davis plausibly arguedthat the relative nickel loss is Mysen (1978a) indicated that the sites on which
reduced as total nickel content of the systemis in- nickel substitutedremained during the time studies.
creased.As a result, the apparent D$:riawould be There were no signs of metastabledefects. Mysen
lowered with increasing nickel content. However (1978b)concludedthat suchstructural featurescould
plausible the argumentsmay sound, they are moot not explain the variation of DS"iqwith nickel concenpoints in view of the present experiments,because tration. The presentexperimentalresultslend further
theseresultswere obtained without the complication credenceto this conclusion.First, the efects were obof nickel loss to the container (Table l), and the de- servedwhether forsteritewas grown in the solid-mependenceof D$ tioon nickel content of the systemre- dia, high-pressureapparatus,in the gas-media,highmained (Fig. l). On the basisof thesenew data, the pressureapparatus,or in the l-atm vertical quench
suggestionthat nickel lossto the samplecontainer is furnace (Fig. l). Second,in the reversalexperiments
the causeof the apparent variations of the partition the forsterite was first grown in the solid-media,
coefficient (Hart and Davis, 1978; Lindstrom and high-pressureapparatus from an oxide mixture.
Weill, 1978)must be ruled our.
Theseforsterite crystalswere then equilibrated with
Lindstrom and Weill (1978)also suggestedthat the the rest of the systemin all the diferent experimental
effectsseenby Mysen(1976a,1978a)
could reflectthe equipment. Growth rates and growth mechanisms
lack of isotopic equilibrium, becausein thoseexperi- are likely to be different in theseapparatus;however,
ments stable nickel was added mostly as NiO the variable D$;"oremained.
whereasu'Ni was added as a dilute chloride solution.
Mysen (1978b)concludedthat if transmutation of
Mysen (1978a) conducted two experimentswith all u'Ni was the causeof the variation of D$;',o (Navnickel in dilute chloride solution to attain a total rotsky, 1978),different specificactivitiesof 63Niin the
nickel content similar to that of some of the experi- crystalsshould result in a systematicchangeof Diyiq.
ments with NiO and chloride where dependenceof Mysen (1978b), in summarizing the available data,
D$a"oon nickel concentrationwas observed.The two showedthat no such effectswere observedwhen the
new experimentsgave the sameresultsas thosewith specificactivity of u3Niwasvaried by a factor of 6. In
NiO and Ni'* in chloride solution. In the presentex- the presentexperimentsthe specificactivity of urNi in
periments the starting materials were homogenized the reversalexperimentswas only 20 percent of
that
either in the glassor in the synthetic forsterite prior in the forward experiments.The specificactivity
of
to the experimentitself. Consequently,the potential the present experimentswas as much as 20 times
problem of equilibrating u'Ni in solution with stable greater than that of some of the experiments
by
NiO during the experimentis eliminated. The effect Mysen and Kushiro (1979).No relation betweenthe
of nickel concentrationon D$.tiqremained, however. specific activity of u'Ni and the value of D$1"0has
Consequently,the suggestionby Lindstrom and been discerned. I conclude, therefore, that defects
Weill (1978)must be ruled out.
in forsterite crystals induced by the radioactivity of
Navrotsky (1978) suggestedthat the use of radio- u'Ni are not the causeof the variable D$liq.
active 63Nior the growth conditions of forsterite in
Hart and Davis (1978) suggestedthat the presthe experimentsby Mysen (1976a,1978a)
may result ence of HrO in the experimentsby Mysen (1976a,
in formation of metastable defects in the olivine 1978a)could be the causeof the variable
D i',',o.
structure. She argued that the distribution curves of The presentexperimentswere conductedunder
anMysen resembledthoseobservedfor solution of gase- hydrous conditions, and the effect remained.
ous N2 in cold-pressedsteel, where the presenceof
It is appreciatedthat the similar ionic radius and
metastabledefects had been demonstrated(Wriedt electrical charge of Ni2* and Mg2* could indicate
and Darken, 1965).Mysen (1978b)commentedthat that (Ni,Mg) solid solutions behave as Raoult's
or
the loj'j' in the experimentsby Wriedt and Darken at least Henry's law type solutions. Such similarity
changed continuously with nitrogen pressure(con- is no proof that the activity coefficients are concentration), whereasin the results of Mysen (1976a, stant, however. Despite the similarity of Mgr*
and
1978a)the activity coefficientof nickel in olivine and Fe'*, for example,thermodynamicdata on (Fe,Mg)
il13
MYSEN: NICKEL PARTITIONING
solid solutions in olivine (Nafziger, 1973;Olsen and
Bunch, 1970)indicated deviationsfrom ideality.
Observations that olivine-liquid partition coefficients for nickel appear relatively constant at high
nickel concentration (within the range attainable
with the electron microprobe), such as reported by
Leeman (1974\, Hart and Davis (1978), and Lindstrom and Weill (1978), does not rule out different
behavior at lower nickel concentration. Furthermore, observations that a partition coefficient is
nearly constant at high nickel concentrationdo not
constitute proof of behavior according to Henry's
law at lower Ni concentration (see liyama, 1974,
for a discussionof the behavior of elementsin mineral solutions accordingto Henry's law).
In summary, the results reported here (Table l,
are not
Fig. l) and elsewhere(Mysen, 1976a,1978a)
artifacts of experhnental problems. They represent
equilibrium partition coefficientsfor nickel between
olivine and liquid.
Mysen (1976a, 1978a)presentedargumentsin favor of the variationsof D;ftir resulting from the solubility behavior of nickel in forsterite.Paul (1975)has
shownthat y$i may depend on nickel concentration
at low Ni content of borate melts. The observations
and discussionof Mysen (1978a)indicate that the activity coefficient of nickel becomes a function of
nickel concentration at lower Ni content of olivine
than coexisting liquid. The present data cannot be
usedto rule out an effectof the liquid structure.
Variations of D3;"o with Ni content may be expected from the phaseequilibria in the systemMgolivine-Ni-olivine (Ringwood, 1956).In the concentration region closeto pure forsterite,the solidusand
liquidus curvesapproachlinear asymptotes,in which
caseDi'i'iq approachesa constant value. This is the
concentrationregion where Henry's law is approximately obeyed.A similar reasoningmay explain the
presentdata.
Petrologicalimplications
is governed
The role ofnickel in igneousprocesses
by a complex function of intensive and extensive
variables,as discussedabove.The presentdata were
obtained in a simplified system,and quantitative application of the data to rock-forming processes
should be exercisedwith care. Some principles can
be derived.
Mysen (1977) and Mysen and Popp (1978) commented that primary partial melts from a peridotite
sourcecontaining 1500-2000ppm Ni (Fishet et al.,
1969)would contain 500-700ppm Ni, dependingon
the modal composition of the source, the mechanisms of melting, and the bulk composition of the
partial melt. This primary rrelt is unlikely to be more
basicthan that of an olivine tholeiite (Yoder, 1976'p.
128-l3l). Tholeiitescontainingmore than 350-400
ppm Ni are not common. Andesitic melts contain
evenlessnickel than olivine tholeiite. Hart and Davis
(1978)suggestedthat nickel partition coefficientsbetween olivine and andesiticmelts were so great that
their nickel content is that of primary partial melts
from peridotite. Mysen and Kushiro (1979)observed'
however, that the partition coefficientsat pressures
correspondingto the depth of partial melting in the
upper mantle are much lower than indicated by ltio
atm experiments.This reductionof D31 is due to
structural changesof the silicate melt as a result of
the increasedpressure(Mysen and Kushiro, 1979).
Consequently,the primary melts are likely to have
precipitated minerals that also act as sinks for nickel
beforethe magmareachesthe earth'ssurface.
Acknowledgments
Critical reviewsby W. Harrison,S. R. Hart, and H. S. Yoder, Jr.
are appreciated.
References
Arndt, N. T. (1977) The partitioning of nickel betweenolivine and
ultrabasic and basic komatiite lnqluids- Caftregie Inst. Wash.
YearBook,76,553-557.
Boyd, F. R. and J. L. Eagland (t960) Apparatusforphase equilibrium measurementsat pr€ssur€sup to 50 kilobars and temperaturesup to 1750'C.J. Geophys.Res.'65,741-748.
Clark, A. M. and J. V. P. Long (1971)The anisotropicdiffusion of
nickel in olivine. In Thomas Graham Memorial Symposiumon
p. 5l l-521. Gordon and Breach,London'
Diffusion Processes,
Fisher,D. E., O. Joensuuand K. Bostrom(1969)Elementalabundances in ultramafc rocks and their relation to the upper
mantle../. Geophys.Res.,74,3865-3873.
Hatti, T. A. and T. L. wrigbt (1967)The fractionation of nickel
betweenolivine and augite as geothermometet'Geochim.Cosmochim.Acta, 3l, 877-88/..
Hart, S. R. and K. E. Davis (1978) Nickel partitioning between
olivine and silicatemelt. Earth Planet. Sci.Lett.,40'203-219.
Hofmann, A. W. and M. Magaritz (1977)Diffusion of Ca, Sr, Ba
and Co in a basalt melt: inplications for the geochemistryof the
mantle../. Geophys.Res.,38,5432-5M1.
Iiyama, I. T . (19'14)Behavior of trace elementsin feldspars under
hydrothermal conditions. In W. S. MacKenzie and J. Zussman"
Eds., The Feldspars,p. 552-573. Manchester University Press,
Manchester,England.
Kushiro, I. (1976) Changesin viscosity and structure of rnelt of
NaAlSi2O6composition at hiSh pressure.J. Geophys.Res.,81,
6347-6350.
Leeman, W. P. (1974) Experimental Determination of Partitionint
MYSEN:
NICKEL
PANTITIONING
of Divalent Cdtions betweenOlivine and Basahic Liquid. part II
(l) systematics of transition metals in crystalline and molten silof Ph.D. Thesis,University of Oregon.
icates and (2) defect chemistry and "the Henry's law problem."
Lindstrom, D. J. and D. F. Weill (19?g)partirioning of transition
Geochim. Cosmochim. Acta, 42, 887-903.
metals between diopside and coexisting silicate liquids. I.
O'Hara, M. J., M. J. Saunders and E. L. P. Mercy (1975) Garnet
Nickel, cobalt, and manganese.Geochim.Cosmochim.Acta, 42,
peridotite, primary ultrabasic magma and eclogite: inter8 l7-833.
pretation of upper mantle processes in kimberlite. Phys. Chem.
Mysen, B. O. (1976a) partitioning of samarium and nickel be_
Earth,9,571-6M.
tween olivine, orthopyroxeneand liquid: preliminary dataat 20
Olsen, E. and T. E. Bunch (1970) Empirical derivation of activity
kbar and 1025"C.Earth planet. Sci.Leu., 3I, l_7.
coefficients for the magnesium-rich portion of the olivine solid
(1976b)The role of volatilesin silicatemelts: solubility of
solution.,4rn. Mineral.. 55. 1829-1842.
carbon dioxide and water in feldspar, pyroxene and feldspa_ Paul, A. (1975) Activity of nickel in
alkali borate melrs. J. Mater.
thoid meltsto 30 kb and 1625.C.Am. J. Sci.,276,969_996.
Sci.,10,422426.
(1977) Magma genesisin peridotite upper mantle in the
Ringwood, A. E. (1956) Melting relationships in Ni-Mg olivines
light of experimental data on partitioning of trace elements be_
and some geochemical implications. Geochim. Cosmochim. Acta,
tween gamet peridotite minerals and partial nelts. Carnegie
r0,297-303.
Inst. Wash.Year&ook,76,545-550.
Schairer, J. F. and H. S. Yoder, Jr. (1961) Crystallizarion in the
(1978a)Experimentaldeterminationof nickel partition co_
syst€m nepheline-forsterite-silica at one atmosphere pressure.
efficientsbetweenliquid, pargasiteand garnet peridotite miner_
Carnegie Inst. l4tash. Year Book,60, l4l-144.
als and concentration limits of behavior according to Henry's
Seifert, F., D. Virgo and B. O. Mysen (1979) Sodium metasilicate
law at high temperatureand pressure.Am. J. Sci.,27g,2l7_243.
melts in CO2 and CO atmospheres. Carnegie Inst. llash. year
(1978b)Limits of solution of traceelementsin mineralsac_
Book, 78, in press.
cording to Henry's law: review of experimental data. Geochim. Sharma, S. K., D. Virgo and B. O. Mysen (1978) Raman
study of
Cosmochim.Acta, 42, 871-885.
structure and mordination of Al in NaAlSi2Ou glasses syntheand I. Kushiro (1979)The eff€ct ofpressure on the parti_
sized at high pressures. Carnegie Inst. V[/ash. year Book, 77,
tioning of nickel between olivine and alumhous silicate melt.
655462.
Earth Planet. Sci.Lett., 42, 383-i89.
ShaW D. M. (1977) Trace element behavior during anatexis. In
and R. K. Popp (1978)Solubility of sulfur in silicate melts
H. F. B. Dick, Ed., Proceedings of the Chapman Conference on
as a function of /", and silicate melt mmposition at high pres_
Partial Melting in the Earth's Upper Mantle, p. 189-215. Amerisures.CarnegieInst. Vlash.YearBook,72,709-713.
can Geophysical Union, Washington, D. C.
and M. G. Seitz (1975)Trace elementpartitioning deter_ Wri€dt, H. A. and L. S. Darken (1965) Lattice defects and
the somined with beta-trackmapping-an experimentalstudy using
lution of nitrogen in a deformed ferritic steel: part I. Expericarbon and samariumas examples.J. Geophys.Res.,g0,262j_
mental data and thermodynamic analysis. Trans. AIME, 233,
2635.
I I l-133.
and D. Virgo (1978) Influence of prcssure,temlrcrature Yoder, H. S., Jr. (1950) High-low quartz inversion to 10,000 bars.
and bulk composition on melt structures in the system
Trans. Am. Geophys. Union, 3I, 827 -835.
NaAlSi2O6-NaFe3+Si2O6.
Am. J. Sci.,278, l3O7-1322.
(1976) Generation of Basaltic Magma. National Academy
Nafziger, R. H. (1973)High-temperatureactivity-romposition re_
of Sciences, \Iy'a5fiington, D. C.
lations ofequilibrium spinels,olivines and pyroxenesin the sys_
tem Mg-Fe-O-SiOr. Am. Mineral., 58,457465.
Manuscript received, November 13, 1978;
Navrotsky, A. (1978) Thernodynamics of element partitioning:
.acceptedfor publication, June 2, 1979.