American Mineralogist, Volume 78, pages 132-142, 1993

Melting experiments on a SiOr-pooro CaO-rich aphanitic kimberlite from 5-10GPa and their bearing on sources of kimberlite magmas

Ar,lN D. Elcan, Hnarnrn E. Cn,q,nnoNNEAUDepartment of Geology, University of Western Ontario, London, Ontario N6A 5B7, Canada

Ansrnc.cr

Phase relations in a SiOr-poor, CaO-rich aphanitic kimberlite from the Wesselton mine,Kimberley, South Africa, were determined in a multianvil apparatus at 5.0-10.0 GPa and1300-1670'C. The experiments were done with the 6.20wto/o HrO and 4.'77 wto/oCO2(X.or= 0.24) present in the rock. Olivine is the primary liquidus phase up to the maximumpressure investigated. An olivine + garnet + liquid assemblage is present at about 70 'C

below the liquidus at low pressures to nearliquidus conditions at >100 GPa. No otherprimary phase is observed in these experiments, but a metastable (quench) phase resem-bling a high aluminum clinopyroxene occurs. Orthopyroxene is absent in all experiments,and the only carbonate phase is calcite, possibly a quench phase. In lower pressure exper-iments at 1.0-4.0 GP4 orthopyroxene was also absent, but primary clinopyroxene andcalcite occurred, suggesting that their absence in this study may be the result of well-documented experimental difficulties with these phases in a multianvil apparatus.

The absence oforthopyroxene is not probably a result ofdecarbonation reactions suchas those that occur in the peridotite-Hro-Co, system, but it could possibly be the resultof degassing during ascent of the Wesselton kimberlite, causing variation in the HrOlCO,in this rock. The most likely reason for the absence of both orthopyroxene and a Mg-richcarbonate is that the bulk composition of this kimberlite has insufrcient SiO, combinedwith high CaO relative to most kimberlites.

This study shows that the aphanitic kimberlite magma is unlikely to have equilibratedwith a carbonated (magnesite or dolomite) garnet lherzolite or harzburgite in which or-thopyroxene is an essential phase. Near-liquidus assemblages suggest that an olivine-bear-ing or olivine-free garnetite (garnet * clinopyroxene + olivine), partially melted, is themost likely source to form this kimberlite. Furthermore, the depth of partial melting mayhave been considerably greater than the 150-250 km commonly accepted for kimberlitemagma formation from a lherzolitic source.

Many kimberlites have the characteristic low SiO, and high CaO compositions foundin the aphanitic kimberlite. If these kimberlites originate at the deep mantle levels sug-gested by this study, and if they have not been extensively changed by fractionation orloss of CO, on degassing, then they may be more likely to be diamondiferous, as they passthrough a larger volume of potentially diamond-bearing mantle than kimberlites formedat shallower levels.

INrnooucrroN the difficulty in recognizing primitive kimberlite com-

Suprasolidus experiments were done at 5-10 GPa on positions: the rarity of kimberlites without xenocrysts and

a low Sio,- and Mgo-poor and cao-rich uprrurriic [iir' xenoliths'^and' until recently' the absence of apparatus

berlite from the Wesselton mine, South Africa. il;;;; capable of attaining pressures equivalent to the depths of

is an extension of earlier experiments up to 4.0 a;;;- ongin of kimberlite magmas' The nature and chemistry

formed on the same composition, in wtrlcrr tnl^rlsir-tts of the sources and the melting process to form kimberlite

indicated that a kimberlite magma of this r"-o"dri"" magma il ":ty poorly constrained in models of kimber-

may not have equilibrated with an assemblag. "*,"i"i"g

lite genesis (see Mitchell' 1986' p' 395)'

orthopyroxene, i.e., a lherzolite or a harzburgite (Edgaret al., 1988). The objectives of the present study were to ARE LHERZOLTTES AND HARZBURGITES THE ONLY

determine whether results of the lower pr"rrrr* e*peri- souRCES oF KTMBERLTTE MAGMA?

ments could be extended to higher pressures and to assess It is commonly accepted that all kimberlite magmasthe depth of origin and ascent path of such kimberlites. can be derived by partial melting of a carbonate-bearing

Experimental data for kimberlites are scarce because of (dolomite or magnesite) garnet lherzolite or harzburgite

0003-004x/93l0 l 02-0 l 32$02.00 r32

EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE 1 3 3

on the basis of very extensive experimentation in modelperidotite-Hro-Co, systems (e.g., Eggler, 1975, 1978;Wyllie, 1977 , 1978, 1979a, 1979b, 1980; Boettcher et al.,1980; Brey et al., 1984; Huang and Wyllie, 1984; Wyllieand Huang, 1976; Canil and Scarfe, 1990) that demon-strated the importance of CO, in forming dolomite andmagresite by carbonation reactions involving orthopy-roxene. For example,

olivine + clinopyroxene * CO,

: orthopyroxene + dolomite (l)

o l iv ine+dolomite+CO,

: orthopyroxene * magnesite (2)

orthopyroxene + dolomite + CO,

: clinopyroxene * magnesite. (3)

Loss of CO, from kimberlite melts favors olivine * cli-nopyroxene (Eq. l). Hence, carbonated garnet lherzoliteneed not be the only source for kimberlite magmas, and,indeed, unequivocally carbonated (dolomite-magnesite)mantle samples are virtually unknown, the only possibleexception being inclusions of calcite and possibly dolo-mite in garnet (McGetchin and Besancon, 1973).

HrO is also essential for partial melting of likely sourcesof kimberlite magma. Phlogopite and amphibole are like-ly HrO reservoirs, and both are relatively common inmetasomatised mantle material (Lloyd, 1987). The com-positions of partial melts of such mantle material is un-known because they cannot be quenched to a glass(Wendlandt and Eggler, 1980a) and, although they areprobably KrO-, MgO-, and carbonate-enriched, the ex-tent depends on the degree of phlogopite and carbonatein the source.

Melting of model carbonated garnet lherzolite at highpressure (Wendlandt and Eggler, 1980b) and of a ksothokimberlite in which Co substituted for Fe in experimenlscarried out up to 5.0 GPa (Eggler and Wendlandt, 1979)suggested that kimberlitic melts may be formed at about150 km. Canil and Scarfe (1990) experimented in thesynthetic peridotite-CO2 system up to 12.0 GPa and pro-posed that kimberlite magmas may form between 5.0 and7.0 GPa and protokimberlite above 7.0 GPa. Experimen-tal data imply that eutectic-like melting of phlogopite +magnesite * garnet + lherzolite (an assemblage contain-ing essential orthopyroxene) at increasing pressures anddegrees of partial melting can produce a series of meltswhose compositions cannot be determined experimen-tally, but which probably range from carbonatite to car-bonatitic kimberlite. Such melts, however, need not bein equilibrium with orthopyroxene in the source (see Caniland Scarfe, 1990, their Fig. 5a).

Ano rrrrnBnr,rrEs pRrMrrIvE MANTLE MELTS?

Experiments on kimberlite compositions, as opposedto model (and actual) peridotite systems used to infer

C ( c l ) L s r n i t e S

Fig. l. Analyses of various kimberlites plotted on the CMSsystem. Uncontaminated kimberlites (Mitchell, 1986) plot alongthe tie line from olivine (ol) to diopside (Diop cpx) and maycontain orthopyroxene (opx). Kimberlite compositions W andA-K are SiO.-poor, CaO-rich kimberlites that are incompatiblewith opx and fall in the ol + cpx + ct (calcite) assemblage inthis projection. W : Wesselton aphanitic kimberlite (this study),A-K : kimberlite from various localities given by Mitchell (1986,his Table 7.1). FuIl lines indicate stable assemblages, dashedlines represent incomparible assemblages.

lherzolitic or harzburgitic mantle sources for kimberlitemagmas, are severely hampered by the inability to deter-mine if kimberlite compositions represent actual melts,primitive or derivative. The abundance of xenoliths andxenocrysts in most kimberlites has probably modifiedcompositions relative to the original kimberlite melt, andhence such kimberlites cannot be used for experimentalstudies.

Based on the absence of xenoliths and xenocrysts, andthe kimberlite's fine-grained nature, Edgar et al. (1988)have argued that the hypabyssal facies aphanitic kimber-lite (Clement et al., 1984) from the Wesselton mine, Kim-berley, South Africa, may represent a composition closeto that of a primitive kimberlite magma. Mitchell (1986)and Shee (1986) considered aphanitic kimberlites to beformed by near-surface flow differentiation or by crystalfractionation at depth. For the purposes of the presentstudy, the actual origin of the Wesselton aphanitic kim-berlite is not as important as the similarity in its com-position to that of other kimberlites, some of which maybe primitive.

The aphanitic Wesselton kimberlite is characterized bylow SiO, and high CaO (Edgar et al., 1988, their Table1), relative to many kimberlites (Mitchell, 1986). Thiscomposition and those of similar kimberlites (Mitchell,1986, his Table 2) are plotted in the CMS system (Fig. l)and fall within a narrow compositional range that is re-moved from the range of uncontaminated kimberlitecompositions. None of the kimberlite compositions sim-ilar to the aphanitic variety is compatible with enstatite

ff analyses ol nalural krmberlles(() tuncon\emnated) tom Milchell {1986)

H'Shandong Liaoning (av o{ 14)I 'Siberian kimb€rl i te (av ol 63)

J- Alakil, Malaya, Botuoba andrelaled li6lds. Siboria (av ot 232)

K- 6s abov€ (av ot 42)W- Wesselton aphanit ic k imberl i t€A- Lesotho kimberl i t€, Pipe 300B- Cumulale B€nlont€inC- Benlont€in Srl l (composi le)D' Orocoo dikeE- Holsteinsborg drkeF- Micaceous kimb€rl i t€G- Holst€insborg diko (av of 11)

tM

134 EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE

(Fig. l), and none contains normative hypersthene. Al-though the CMS system is very simplistic relative to thecomplex chemistry of kimberlites, the high CaO and verylow SiO, in kimberlites, akin to that of the aphanitic kim-berlite used in this study, make it highly unlikely thatthey were ever equilibrated with an assemblage contain-ing orthopyroxene. The comparison in Figure I indicatesthat the aphanitic Wesselton kimberlite composition isreasonably representative of SiOr-poor and CaO-richkimberlites. The major difference between the aphaniticWesselton kimberlite and the other kimberlites (Fig. l) isits much lower COrl(H,O + COr). The effect of this onphase relations is discussed below.

Suitability of the Wesselton kimberlite for experimen-tal studies based on its texture and the absence ofxeno-crystal olivine has been discussed by Edgar et al. (1988).The primitive value of 100 Mg/(Mg * Fe,",) : 83.9, thevalues of Ni (8 l0 ppm) and Cr (24 l0 ppm), and its sim-ilarities to the average of 44 other Group I kimberlitesfrom Wesselton (Edgar et al., 1988, their Table l) suggestthe aphanitic kimberlite has not undergone appreciablefractionation, although some olivine or pyroxene frac-tionation is possible on the basis of the data in Figure Iofthis paper.

In previous experiments on the same kimberlite up to4.0 GPa at CO, = 0.24 and 0.52, Edgar et al. (1988) didnot find orthopyroxene in any experiment but found cal-cite as the only carbonate in the lower XCO, conditionand dolomite only at temperatures well below that of theIiquidus and at >3.5 GPa. On the basis of these experi-ments, Edgar et al. (1988) suggested that the Wesseltonaphanitic kimberlite magma was not generated by partialmelting of a source containing orthopyroxene, that themagma may have exceeded 1300'C at depths equivalentto about 1.2 GPa, and that calcite was the onlv carbonatein the source.

ExpnnnnrNTAL AND ANALvTTcAL TECHNTeuES

All experiments were done on a USSA-2000 fully au-tomated uniaxial, split-sphere multianvil apparatus at theUniversity of Alberta, Edmonton. The sample was placedin a Pt capsule that was 2.2 x 5 mm, predried at 120 "C,and welded shut. The capsule was then placed in the fur-nace assemblage, which consisted of a stepped graphiteheater surrounded by aZrO, sleeve, which was containedwithin a semisintered MgO-CrrO. (50/o) octahedron of 12-mm edge length. The octahedron was then placed in acavity formed by truncating the corners of eight WC an-vils; the truncation edge length was I I mm.

Temperatures were controlled to within +5 "C of thevalues quoted by a Eurotherm 8l8P controller interfacedwith a 80286 microcomputer. A Pt-PtrrRh,. thermocou-ple placed axially and in contact with the tip of the sam-ple capsule recorded the temperature. No correction wasmade for the effects of pressure on the emf output of thethermocouple. Independent measurements of the axialtemperature profile at the center of the furnace indicated

that the center of the sample may be -10'C hotterthan the ends of the capsule, where the temperature ismeasured.

Pressure was calibrated at room temperature at thephase transitions BiI-BiII ar 2.55 GPa (Hall, 1971) andBiII-BiV at 7.7 GPa (Holman, 1975). At 1000'C, thecalibration was based on the transitions: quartz-coesiteat2.95 GPa (Bohlen and Boettcher, 1982), fayalite-spinelat 5.27 GPa (Yagi et al., 1987), garnet-perovskite at 6.1GPa for the composition CaGeO, (Susaki et al., 1985),and coesite stishovite at 9.1 GPa (Yagi and Akimoto,1976). The estimated pressure uncertainty is

EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE

Fig. 2. Backscattered electron images of quench (a) and primary @) products of experiments showing typical textures. Darkgrains are olivine, gray grains are garnet, and lighter grains are the quench "clinopyroxene" (see text). Part a is from experimentno. 10 (Table l) and Part b from experiment no. 17 Oable l).

1 3 5

gested that ,6, in diamond- and carbonate-bearing kim-berlites ranged from that of the WM to the QFM buffer.More recently, Luth et al. (1990) have proposed a moreoxidized mantle correspondingto an fo2between the QFMand WM buffers based on Fe3* in mantle-derived garnets.The /o, values ofboth Bulanova (1986) and Luth et al.(1990) fall within the minimum possible /o, calculatedby O'Neill and Wall (1987), on the basis of calculationsinvolving the Ni content of olivine in equilibrium withorthopyroxene in a typical mantle peridotite. These stud-ies suggest that the present experiments were done underhigher /o, than those generally accepted for a peridotitemantle composition.

At the end of the experiment, the capsule was removed,embedded in epoxy, and sliced with a diamond saw. Thesurface was ground, and a polished thin section made forelectron microprobe analysis. The products of each ex-periment were identified by EDS analysis and quantita-tively by WDS analysis. Textures were observed by back-scattered electron imagery.

In most experiments, the products consisted of quenchmaterial, composed of elongated crystals and blades ofacicular olivine with interstitial garnet and a quench phasethat resembled a very AlrO.-rich clinopyroxene (13-16wto/o AlrOr) (Fig.2a), and of primary mineral assemblageswith characteristic granuloblastic texture (Fig. 2b). Thelatter assemblage occurs close to the tip ofthe recordingthermocouple, whereas the quench assemblage with spi-nifex-type texture (Fig. 2a) is observed farther away andindicates large thermal gradients along the capsule. Noglass is detected in these experiments, as the melts havevery low viscosities relative to those from the same com-positions at lower pressures (

136 EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE

TABLE 1, Results of experiments

Productst

(min)t

fc)P

(GPa)Exp.

q-ol(93.0), q-g(79.6)ol(94.6), q-g(86.8)ol(94 4), 9(90.4), q-"cpx", ap, sp(?)ol(91.0), 9(78.4), q-"cpx", pv, Mg-ferrite, Mg-Ti-mtol(94.1), g(81.6), q-"cpx", ap, ctol(94.8), q-g(86.1) Mg-chr, ruol(94.0), g(83.1), pvol(93.1), g(81.0), q-"cpx", ap, ctol(94.0), q-"cpx", cr-sp, Mgjerrite, ap, pvol(92.5), g(78.4), q-"cpx", Mg-Ti-mt, pvo(9a.9), q-g(78.8), q-"cpx", Mg-TLmt, Mg-chr, apol(94.2), 9(83.3), q-"cpx", Mg-Ti-mt, ap, ct, pvol(93.5), 9(76.9), q-"cpx", ap, ctol(92.5), 9(76.4), q-"cpx", sp

No te . 'Abb rev ia t i ons :q :quench ,o l : o l i v i ne ,g :ga rne t , " cpx " : quenchc l i nopy roxene l i kephase (see tex t ) , c t : ca l c i t e , ap :apa t i t e , pv=perovskite, MgJerrite: magnesioferrite, mt : magnetite, sp : spinel, Mg-chr: magnesiochromite. Numbers in parentheses represent average Mg'from several grains. Only olivine and garnet are considered to be primary phases.

1 0

21 51 2

I1 41 1863

1 31 7

10.010.010.010.09.08.58.08.07 .O7.06.06.06.05.0

1 6751 6001 5601 5001 5001 6001 5001 4001 5401 3601 5301 48014251 300

c

1 51 31 5

c

1 51 51 3

1 61 51 5201 5

oo-(,

oL:o6oL

G

Rrsur,rsPhase relations

Table I lists the results of experiments between 5.0 and10.0 GPa and 1300 and 1675 "C. Figure 4 shows thestable phase assemblages. Only garnet and olivine are

13

12

11

10

9

8

7

6

5

4 r r r | | I | | | | |-13fi'

14oO 15qt 16m 17(XtTemperature ( oC)

Fig.4. Pressure-temperature relationships. Field A : all liq-uid, B : olivine + liquid, C: olivine + garnet + liquid. Thequench "clinopyroxene" occurs in all fields but is most promi-nent in field C. Full lines represent well-established boundaries,dashed lines are extrapolated boundaries.

considered to be stable, on the basis oftheir textural re-lationships and their Mg' values, which decrease with de-creasing temperature and which are based on the averageof 4-20 analyses (normally >10). Wherever possible,grains were analyzed at the liquid-solid interface (Fig. 3)considered to represent nearJiquidus phases (cf. Herz-berg et al., 1990). Throughout the P-Tconditions of theseexperiments, quench clinopyroxene was abundant, par-ticularly in the P-I range of field C (Fig. 4). This mineralhad consistently high AlrO, contents and did not alwaysexhibit textures that could distinguish it as primary orquench. However, the values of Mg', when compared withthose of primary olivine and garnet, were too low to in-dicate that the clinopyroxene was in equilibrium witheither phase. For these reasons, the clinopyroxene formedin these experiments is considered to be quench (Tablel). No glass or orthopyroxene was detected in any of theexperiments. The all-liquid field (Fig. 4, field A) was de-termined on the basis of quench textures (Fig.2a) and onthe Mg' of olivine and garnet (Table l) being lower thanthe Mg' of the primary varieties of these minerals. Forexample, the olivino and garnet in the experiment at 10.0GPa, 1600 oC, have Mg' values of 94.6 and 86.8, respec-tively, whereas the experiment at 10.0 GPa, 1560 oC, hasolivine and garnet with Mg' values of 94.4 and 90.4, re-spectively (Table l). On the basis of this criterion, theolivine is considered to be primary, but the garnet is con-sidered a quench phase at the higher temperature. Botholivine and garnet are primary at 1560 "C. This conclu-sion appears to contradict the textural relations (Fig. 3)in the products ofthis experiment that show garnet sur-rounded by olivine and therefore suggests that garnetformed earlier than olivine. These lines of evidence clear-Iy indicate that both phases are very close to the liquidusat l0 GPa (Fig. a). Minor amounts of various spinels(chromite, magnetite, magnesioferrite), calcite, rutile, ap-atite, and perovskite occur unsystematically in the exper-iments (Table l). For this reason they are considered torepresent metastable quench material.

The boundaries in Figure 4 have been drawn based on

a

o

EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE

TreLe 2, Representative olivine analyses

t37

P (GPa)r(rc)Major coexisting

phasesExperiment no.

10 .01 600

c

6.01 530

o

10.01 500

8.01 500

6.01425

s'

9.01 500

s'1 2

s'1 5

s-1 3

sio,Tio,Alro3Cr2OoFeOMnOMgoCaONaroNio

ToTAL

41.210.000.010.005.510.04

52.320.240.00nd

99.36

0.9970.0000.0000.0000.0000 . 1 1 11.8870.0010.0000.0090.000

94.4

40.650.000 . 1 60.305.620.12

53.210.300.000.03

1 00.19

0.9790.0000.0000.0020.0010.1 131.9090.0020.0010.0080.000

94.3

40.650.000.120.056.240.08

52.940.210.000.03

100.29

0.9800.0030.0000.0000.0010.1261.9020.0020.0010.0050.000

93.7

40.620.040.070.007.490.00

51.960.250.000 . 1 1

r 00.56

0.9820.0020.0000.0010.0000.1511.8720.0020.0000.0060.000

92.4

40.730.300.260.396.210.28

52.O70.440.000.05

100.67

0.9790.0070.0000.0080.0070.1251.8s60.0040.0010.0110.000

93.5

J I

rrtAlr6tAl

Ti

FeMgMnNiCaNaMg'

41.310.020.030.006.540.16

52.450.280 . 1 40.02

100.98Cat ions for4Oatoms

0.9900.0010.0000.0000.0000.1311.8730 0010.0010.0070.007

94.3

A/ote.'For abbreviations see Table 1.'Coexisting garnet analyses given in Table 3.

the first appearance of primary minerals using Mg'. Hencethe boundary between assemblages B and C is well con-strained by experiments 5 and 3 (Table I, Fig. 4). onextrapolation to higher pressures, this boundary meetsthe A-B boundary at about 12.0 GPa and thus representsthe minimum pressure at which olivine and garnet co-exist with liquid. On the basis of textural interpretationofgarnet and olivine at the liquidus boundary (Fig. 3),garnet may be at the liquidus at as low as l0 GPa.

Mineral compositions

Tables 2 and 3list compositions of olivine and garnet,where some of these minerals coexist (Fig. a). The aver-age Mg' values for primary and quench phases used toconstruct the phase diagram (Fig. a) are in Table l.

In backscattered imagery (Fig. 2) olivine appears asdark angular grains that can be distinguished from fre-quently rounded garnet grains by a lighter color (Fig. 2b).Quench clinopyroxene has a color in backscatter imagerythat is intermediate between that of olivine and garnet(Fie. 2b).

Representative olivine analyses are in Table 2. The Mg'values of primary olivine are comparable with those inthe lower pressure experiments (Edgar et al., 1988), withthose of ultramafic assemblages, and with olivine phe-nocrysts in the Wesselton aphanitic kimberlites (Shee,1986). The Mg'values of olivine at the liquidus (Fig. 4)are slightly lower than predicted values for their hightemperatures, suggesting that some Fe may have beenlost.

Representative compositions of garnet are given in Ta-

ble 3. The Mg' of primary garnet is consistently lowerthan that of olivine (Table l). The garnets are subalu-minous, Ti-, Ca-, and Cr-rich members of the grossula-rite-pyrope-almandine series (Table 3).

The compositions of the garnets produced in the pres-ent experiments differ from thoso found in kimberlitesand related rocks. Garnet is absent from the Wesseltonaphanitic kimberlite (Shee, 1986) and in the lower pres-sure experiments (Edgar et al., 1988). The garnet com-positions (Table 3) most closely resemble the Group 6garnet, pyrope grossular almandine garnet, in the statis-tical classification of Dawson and Stephens (1975). Thesegarnets are common in eclogites. The TiO, in the exper-imentally produced garnets is commonly gr€ater (Table3) than that of the megacrysts of Cr-poor garnets listedby Mitchell (1986). Megacryst garnets have lower CaOand higher FeO than those formed in this study.

DrscussroN

Nature of the source region for the Wesselton aphanitickirnberlite and compositionally similar kimberlites

The absence of orthopyroxene and primary clinopy-roxene as suprasolidus phases in the experiments impliesthat, under the conditions used, liquid is not in equilib-rium with a garnet lherzolite assemblage (garnet + oliv-ine + orthopyroxene * clinopyroxene), and therefore thisrock could not be the source of the kimberlite. Beforeconsidering the consequence ofthe absence ofthis assem-blage, it is necessary to consider whether the absence oforthopyroxene is related to either the relatively low CO,

I38 EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE

TraLe 3, Representative garnet analyses

P (Gpa)rfc)Major coexisting

pnaseExperiment no.

1 0 01 500

ol1 4

8.01 500

olq

9.01 500

olol1 5

8.01 400

7.O1 360

ol, cpxa

6.01425

ol1 3

sio,Alr03FeOMnOCaOMgoTio,Cr,03

ToTAL

Si14lAlrotAl

TiFeM nCr

MgMg'

40.4117.74

c . Y /

0.221 9 .1313.312.091.98

100.85

5.9430.0573 0170.231o.734o.0270.2303.0142.918

7 0 2

' | 1.00.4

45.O43.6

39.8617.305.210.27

1 9 . 1 113.062.582.32

99.21

38.3517.874.770.17

20.0512.322.532.42

98.51

5.7890.2112 968o.2870.6020.0220.2893.2482.772

81.6

9 .10.3

48.941.7

40.0316 995 7 8o.24

19.7612 .552 . 1 92.30

99.84

5.9660.0342.9500.2450.7200.0300.2713 .1552.788

7 8 8

t u . / o0.45

47 1441 .65

39.4117.606.000.25

20.061 1 .652.092.00

99.06

5.9780.o723.0480.2360.7550.0320.2383.2332.612

76.9

1 1.380.48

48.7539.39

40.3418.254.840.25

18.7713.372.102.22

1 00 .13Cations lor 24 O atoms

5.925 5.9390.075 0.0612.955 3.1030.288 0.2320.648 0.5980.034 0.0310.273 0.2s83.043 2.9612.893 2.934

80.9 82 4End-members

AlmandineSpessartiteGrossularitePyrope

9.790.51

46.043.7

9 1 440.4845.4044.99

A/ofei For abbreviations see Table 1

in the bulk composition or the possible elimination oforthopyroxene by reaction from a peridotite assemblageat pressures in the range of 5.0-8.0 GPa.

The absence of orthopyroxene

In their study of the aphanitic kimberlite up to 4.0GPa, Edgar et al. (1988) attributed the absence ofortho-pyroxene in experiments at X.o. conditions of 0.24 and0.51 to be caused by the lower SiO, composition of theaphanitic kimberlite relative to many other kimberlites(cf. Mitchell, 1986, his Table 7.3 and Fig. 1). They showedthat the aphanitic Wesselton kimberlite plotted on theCMS and CMS CO, projections (Edgar et al., 1988, theirFigs. 2, 3) was unlikely to have a composition compatiblewith an olivine * clinopyroxene + orthopyroxene sourceassemblage but pointed out that an SiOr-poor melt mightbe derived from a partial melt of an olivine + clinopy-roxene + orthopyroxene (lherzolite) source from whicholivine + orthopyroxene + clinopyroxene had fraction-ated. There is little evidence, however, that the aphanitickimberlite is a particularly evolved rock, rather than aprimitive magma (cf. Edgar et al., 1988).

Alternatively a reaction relationship such as

forsterite + CO, : enstatite + magnesite (4)

may occur under higher Xco, conditions than those usedin the present experiments. Studies of parts of the peri-dotite-COr-HrO system (eg., Eggler and Wendlandt, 19791,Wyllie and Huang, 1976; Olafsson and Eggler, 1983; Breyet al., 1984; Canil and Scarfe, 1990) show that a reaction

such as Reaction 4 will occur in the range of 3.0-4.0 GPa.No evidence of such a reaction occurs in either the rela-tively high Xco: experiments (X.o, e 0.52) of Edgar et al.(1988) up to 4.0 GPa or in the lower X.o, experimentsabove 5.0 GPa in this study, where no bona fide primarycarbonate minerals occuffed. The high amount of CO,required for Reaction 4 and other reactions is due to thenecessity to saturate liquids in COr, based on experimentson the peridotite HrO-CO, system, and hence is probablygeologically unreasonable (cf. Mitchell, 1986). For ex-ample, more than 5olo CO, is required to complete Re-action l, and liquids in equilibrium with phases in theCMS-CO, system may have 35-40o/o COr, values in ex-cess of any believed to be present in the mantle.

The possibility oforthopyroxene being a reaction prod-uct during peridotite melting at 5.0-8.0 GPa, and wheth-er this melting can control the mineral relations of prim-itive melts, such as those of the aphanitic kimberlite, mustbe considered. Two reactions may take place in the pres-ence of H,O-CO, fluid:

orthopyroxene * magnesite

* clinopyroxene + olivine + fluid

: melt (5)

olivine + magnesite * clinopyroxene + fluid

: orthopyroxene + melt. (6)

Reaction 5 involves a eutectic-like melting of a magne-site-bearing lherzolite source that could crystallize ortho-

EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE 139

pyroxene if the melt separated. Such a melt would alsobe expected to crystallize magnesite or dolomite at lowerpressures. The absence ofthis type ofcarbonate compo-sition in the low-pressure experiments of Edgar et al.(1988) and the total absence ofany Mg-bearing carbonatephases, even in the quench experiments of this study,imply that the magma from which the aphanitic kimber-lite formed was not derived from such a melt as modeledby Reaction 5.

If Reaction 6 is involved, then orthopyroxene mightcrystallize only briefly from the melt, after separation fromthe source, and any evidence of its presence would bedifficult to detect. As in the case of the eutectic-like melt,however, the melt in this reaction might be expected tocrystallize magnesite or dolomite at lower pressures. Theabsence of either of these minerals close to the liquidusin the low-pressure experiments (Edgar et al., 1988) sug-gests that the melt from which the aphanitic kimberliteevolved may not be due to Reaction 6.

Canil and Scarfe (1990) extended experiments on themodel peridotite-CO, systems up to 12.0 GPa. Theyshowed that partial melts from such a model peridotiteat 5-7 GPa would crystallize orthopyroxene, whereas at>7 GPa a Mg-rich kimberlite magma (protokimberlite)might fractionate olivine at lower pressures, producing aprimary kimberlite magma also in equilibrium with or-thopyroxene (Canil and Scarfe, 1990, their Fig. 7). Theseauthors believe that CaO-rich, SiOr-poor kimberlites, suchas the Wesselton aphanitic kimberlite, are the result ofmixing of Si-poor and carbonate-rich melts at lowpressure.

The absence of clinopyroxene

The absence of primary clinopyroxene in the productsof the experiments is partly caused by experimental prob-lems with such Mg-rich melts as the aphanitic kimberliteused in this study. Herzberg et al. (1990) have discussedthis problem in detail and presented methods of obtain-ing analyses of equilibrium (primary) phases between near-solidus and liquidus conditions. The tendency of mixturesof quench and primary minerals to form as mattes con-taining both primary and quench material, particularlynear the liquidus, was observed in nearly all the productsof this study (Fig. 3). In many instances, clinopyroxeneappears to be a primary mineral texturally but is in facta quench phase grown around stable olivine and garnet.In no instance did microprobe analyses of clinopyroxenegrains indicate that they were primary, as the Mg' valueswere too low for the clinopyroxene to be in equilibriumwith primary olivine and garnet (Table l), and the AlrO.contents (12-16 wto/o) were much too high. Using thesecriteria, we do not consider clinopyroxene a stable min-eral in these experiments.

Nevertheless the possibility of some clinopyroxene be-ing primary, even close to the liquidus, cannot be exclud-ed, and this idea receives some support from the presenceof primary clinopyroxene in experiments up to 4 GPa

(Edgar et al., 1988), in which problems in forming stableassemblages are not so great. Estimation of the compo-sitions of residual liquids formed after the onset of garnetformation (Fig. 4, field C) certainly does not precludeclinopyroxene and some Al-bearing mineral (spinel?) asstable phases.

Similar arguments may apply to the absence of car-bonate minerals, which texturally appear to be both pri-mary and quench. Unfortunately no chemical criteria canbe used to assess this. However, the only observed car-bonate mineral is calcite.

The lack of unequivocal stable phases in the experi-ments does not alter the fact that orthopyroxene is not alikely mineral to be in equilibrium with a melt of theaphanitic Wesselton kimberlite composition under theconditions of this study.

Source for the aphanitic kimberlite magma

The source of the aphanitic kimberlite magma can beinferred from multiphase saturation of mineral phases atthe liquidus or from nearJiquidus phases. Figure 4 in-dicates that no multiphase saturation occurs, but garnet,along with olivine, is present near the liquidus at 10.0GPa. If the lines representing the liquidus (first crystal-lization ofolivine) and that ofgarnet (Fig. a) are extrap-olated to higher pressures, the assemblage garnet * ol-ivine occurs at about 12-13 GPa. This assemblage isequivalent to a garnet dunite, an unlikely source for kim-berlite because it is refractory. The presence of quenchmaterial of a composition equivalent to an Al-rich cli-nopyroxene up to the near-liquidus may modify thissource material to that of a garnetite assemblage, a morelikely candidate as a source of the aphanitic kimberlitemagma. This multisaturation point on the extrapolatedliquidus (Fig. a) may imply that kimberlite magma aroseby partial melting at pressures of 12.0-13.0 GPa.

The high pressure at which olivine is on the liquidus(Fig. a) may be a consequence of the low X.o, in theexperiments. Such a low value could be the result of lossof CO, during degassing of the aphanitic kimberlite. Thus,the relatively high H,O or slight changes in the experi-ments increase the size of the olivine + liquid field (Fig.4, field B), as demonstrated by Kushiro (1969) in theforsterite-diopside-SiO, system. If this is the reason forthe stability of olivine to high pressures, then under alarger X.o,the multisaturation point involving olivine *gamet + clinopyroxene(?) might occur at lower pressures.The possibility that the aphanitic kimberlite is a result oflocalized large-scale melting of a garnet * clinopyrox-ene(?) assemblage, representing a garnetite of basalticcomposition from a subduction-related process (Ring-wood, 1989) such as that proposed for the Roberts Victoreclogites by Hatton and Gurney (1987), might exist. Somesupport for this suggestion may come from the similaritybetween the compositions of the garnets found in thisstudy (Table 3) and those present in eclogites (see above).The garnets found in this study do not have the charac-

I4O EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE

teristic high SiO, of very high pressure. Experiments onthe eclogite-garnetite transformation (Irifune et al., 1989)indicate that SiOr-poor gamets occur only above 10.0GPa.

The experiments described here as well as those of Ed-gar et al. (1988) strongly suggest that the aphanitic Wes-selton kimberlite did not evolve by partial melting of agarnet lherzolite but may have been derived by partialmelting of a source that did not include orthopyroxene,or from which orthopyroxene was consumed during eu-tecticJike melting. The present study also suggests thatformation of the magma from which this kimberliteformed could have occurred at a depth considerablygreater than the range of 150-250 km for kimberlite mag-mas derived from carbonated garnetJherzolite sources. Ifthe olivine field in the experiments is decreased by a high-er X.o, (see above), then the source inferred from theexperiments may be at a depth equivalent to a pressureof < 10.0 GPa ( 1.0 GPa),the magma cooled to below 1300 "C, on the basis of theabsence of clinopyroxene and the presence of monticellitein the aphanitic kimberlite. Rapid uprise and cooling pro-

EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE I4I

duced a matrix of olivine (now as serpentine), monticel-lite, spinel, calcite, and other phases such as phlogopite,formed at P-Zconditions lower than those used by Edgaret a l . (1988).

AcxNowr,nocMENTS

We are grateful to E.M.w. Skinner for donating the aphanitic kimber-lite from Wesselton. The technical help of R. Tronnes at the Universityof Albe(a multianvil laboratory and the assistance of R.L. Barnett, J.Forth, D.M Kingston, and L. O'Connor at the University of WesternOntario is gratefully acknowledged. Versions ofthis manuscript were readby G. Brey, R.H. Sutcliffe, and D. Vukadinovic. Discussions with N.M.S.Rock, W R. Taylor, and R. Ramsey, at the University of Western Aus-tralia, and especially C.J. Hatton of the Anglo-American Research Cor-poration proved very helpful. R. Wendlandt and two anonymous review-ers made many helpful comments. This research was financiaf supportedby De Beers Consotidated Mines and by operating and major installationgrants to the senior author.

RnrnnnNcrs crtnnBoettcher, A L., Robertson, J.K., and Wyllie, P J. (1980) Studies in syn-

thetic carbonate systems: Solidus relationships for CaO-MgO-COrH,Oto 40 kbar and CaO-MgO-SiOr-CO,-H,O to 10 kbar Journal of Geo-physical Research, 85, 6937 -6943.

Bohlen, S.R., and Boettcher, A.L. (1982) The quartz-coesite transition: Aprecise determination and the effect of other components. Journal ofGeophysical Research, 87, 7 07 3-7 07 8

Brey, G., Brice, W.P., Ellis, D.J., Green, D.H., Harris, K L., and Ryabchi-kov, I.D. (1984) Pyroxene-carbonate reactions in the upper mantle.Earth and Planetary Sciences l-Etters, 62, 63-7 4.

Bulanova, G.P. (1986) Compositional evolution of syngenetic inclusionsofultrabasic association in Yukatian diamond. Extended Abstracts, 4thInternational Kimberlite Conference, p. 37 | -37 3. Geological Societyof Australia, Perth.

Canil, D., and Scarfe, C.M. (1990) Phase relations in peridotite + CO,systems to 12 GPa: Implications for the origin of kimberlite and car-bonate stability in the Earth's upper mantle. Journal of CeophysicalResearch. 95. 15805- I 58 16.

Clemenr, C.R., Skinner, E M.w., and Scott-Smith, B.H (1984) Kimber-lite redefined. Journal of Geoloey, 92, 223-228.

Dawson, J B , and Stephens, W.E. (1975) Statistical analysis of garnetsfrom kimberlites and associated xenoliths. Joumal ofGeology, 83, 589-607.

ldgar, A D., Arima, M., Baldwin, D.K., Bell, D R., Shee, S R., Skinner,E.M.W., and Walker, E.C. (1988) High-pressure-high-temperaturemelting experiments on a SiOr-poor aphanitic kimberlite from theWesselton mine, Kimberley, South Africa. American Mineralogist, 73,524-533

Eggler, D.H. (1975) Peridotite-carbonate relations in the system CaO-MgO-SiOyCOr. Camegie Institution of Washington Year Book 74,468-47 4.

-(1978) The effect of CO' upon partial melting of peridotite in thesystem NarO-CaO-AIO,-MgO-SiO,-CO'-HrO to 35 kilobars, with ananalysis of melting in a peridotite-H,O-Co, system American Journalof Science. 278. 305-343.

Eggler, D.H., and Baker, D.R (1982) Reduced volatiles in the systemC-O-H. Implications to mantle rnelting fluid formation and diamondgenesis. In S. Akimoto and M. Manghnani, Eds., High pressure re-search in geophysics, p.237-250 Tokyo Center for Academic Publi-cations, Tokyo.

Eggler, D.H., and Wendlandt, R.F. (1979) Experimental studies on therelationship between kimberlite magma and partial melting of peri-dotite. In F.R. Boyd and H.O.A. Meyer, Eds., Kimberlites, diatremesand diamonds: Their geology, petrology and geochemistry, p. 33 l-378.American Geophysical Union, Washington, DC.

Ford, C.E. (1978) Platinum-iron alloy sample containers for melting ex-periments on iron-bearing rocks, minerals and related systems. Min-eralogical Magazine, 42, 27 1-27 5.

Hall, H.T (1971) Fixed points near room temperature. In E.C. Lloyd,Ed, Accurate characterization ofthe high pressure environment, N.B.S.Special Publication 326, p. 3 I 3-3 14 U.S. Government Printing Office,Washington, DC.

Hatton, CJ., and Gurney, J.J. (1987) Roberts Victor eclogites and theirrelation to the mantle. In P.H. Nixon, Ed., Mantle xenoliths, p. 453-464. Wiley, Chichester. U.K.

Herzberg, C., Gasparik, T., and Sawamoto, H. (1990) Origin of mantleperidotite: Constraints from melting experiments to 16.5 GPa. Joumalof Geophysical Research, 95, I 5779-1 5803.

Holman, C.G. (1975) Phase diagrams of Bi up to 140 kb. Journal of thePhysics and Chemistry ofSolids, 36,1254-1269.

Huang, W.I., and Wyllie, P.J (1984) Carbonation reactions for mantlelherzolite and harzburgite 27th International Geological Congress,Proceedings 9, 455-47 3.

Irifune, T, Hibberson, W.O., and Ringwood, A.E. (1989) Eclogite-gar-netite transformation at high pressures and its bearing on the occur-rence of garnet inclusions in diamond In J. Ross et al., Eds., Kimber-lites and related rocks, vol. 2, p. 877-882. Geological Society ofAustralia,Carlton, Australia.

Kushiro, I. (1969) The system forsterite-diopside-silica with and withoutwater at high pressures American Journal of Science, 267 A, 269-294.

Lloyd, F.E. (1987) Characrerization ofmantle metasomatic fluids in spi-nal lherzolites and alkali clinopyroxeneites from the West Eifel andSouth West Uganda. In M A Menzies and C J. Hawkesworth, Eds,Mantle metasomatism, p. 91-124. Academic, London.

Luth, R W , Virgo, D., Boyd, F R., and Wood, B.J (1990) Ferric iron inmantle-derived gamets: Implications for thermobarometry and for theoxidation state of the mantle Contributions to Mineralogy and Pe-trology, 104, 56-72.

McGetchin, T.R., and Besancon, JP. (1973) Carbonate inclusions inmantle-denved pyropes Earth and Planetary Science Letters, 18,408-4 10 .

Mitchell, R H. (1986) Kimberlites, 442 p Plenum, New York.O'Neill, H.S.C., and Wall, V.J. (1987) The olivine-orthopyroxene-spinel

oxygen geobarometer, the nickel precipitation curve, and the oxygenfugacity of the Earth's upper mantle. Journal of Petrology, 28, 1169-I l 9 l

Olafsson, M., and Eggler, D.H. (1983) Phase relations of amphibole, am-phibole-carbonate and phlogopite-carbonate peridotite: Petrologic con-straints on the asthenosphere Earth and Planetary Science Letters,64,303-3 I 5.

Ringwood, A.E. (1989) Composition and evolution of the mantle. In J.Ross et al.. Eds., Kimberlites and related rocks, vol. l, p 457-485.Geological Society of Australia, Carlton, Australia

Shee, S.R. (1986) The petrogenesis of the Wesselton mica kimberlites,Kimberley, South Africa, 220 p. Ph.D thesis, University of Capetown,Capetown, South Africa

Susaki, J., Akaogi, M., Akimoto, S., and Shimomira, O. (1985) Garnet-perovskite transition in CaGeOr: In situ X-ray measurements usingsynchrotron radiation. Geophysics Research Letters, 12, 729-732.

Takahashi, E. (1986) Melting of dry peridotite KLB-I up to 16 GPa:Implications on the origin ofperidotitic upper mantle Journal ofGeo-physical Research, 9 l, 9367 -9382.

Wendlandt, R.F., and Eggler, D.H. (1980a) The origin of potassic mag-mas. IL The stability ofphlogopite in natural spinel lherzolite and inthe system KAISiO.-MgO-SiOr-H'O-CO, at high pressures and hightemperatures. American Joumal of Science, 280, 421-458.

-(1980b) The origin of potassic magmas. L Melting relations in thesystem KAlSiO4-Mg,SiO"-SiO, and KAlSiO"-MgO-SiO,-CO' to 30 kil-obars. American Joumal ofScience, 280, 385-420.

Wyllie, P J. (1977) Mantle fluid compositions buffered in peridotite-COr

HrO. Journal ofGeology, 85, 187-208.- (1978) Mantle fluid compositions bufered in peridotite-CO,-H,O

by carbonates, amphiboles and phlogopite. Journal of Geology, 86,687-7 13.

-(1979a) Magmas and volatile components. American Mineralo-gist, 64, 469-500.

-(1979b) Kimberlite magmas from the system peridotite-Cor-H,O.2nd Intemational Kimberlite Conference, p 319-329. AGU, Wash-rngton, DC.

142 EDGAR AND CHARBONNEAU: MELTING EXPERIMENTS ON KIMBERLITE

-(1980) The origin ofkimberlite. Jounial of Geophysical Research, Yagi, T., Akaogi,M., Shimomura, O., Susuki, T , and Akimoto, S. (1987)8 5, 6902-6910. In situ observation of the olivine-spinel phase transforrnation in BetSiOn

Wyllie, P.J., and Huang, w L. (1976) Carbonation and melting reactions using synchroton radiation Joumal ofGeophysical Research, 92, 6207-in the system CaO-MgO-SiO,-CO, at mantle pressures with geophysi- 6213.cal and petrological applications. Contributionq to Mineralogy and Pe-trology, 54, 79-100.

Yag, T., and Akimoto, S.L. (1976) Direct determination of the coesite-stishovite transition by in situ X-ray measurements. Tectonophysics, MeNuscnrpr'REcEI\TD Novewrn 7, l99l35.259-27A. M,r.Nuscr.pr AccEprED SrsrsrvrnrR. 16. 1992