American Mineralogist, Volume 76, pages 1092-1100, 1991

Lubrication, gasketing, and precision in multianvil experiments

D.q,vro WalrpnLamont-Doherty Geological Observatory and Department of Geological Sciences, Columbia University,

Palisades, New York 10964, U.S.A.

Ansrnacr

Lubrication, gasketing, and pressure medium efects on pressure generation and repro-ducibility were investigated with 35 experiments on the Bi transitions in a multianvilapparatus. The system of integral gasket fins on octahedral pressure media and sliding,cylindrical anvil-driving wedges within a retaining ring, similar to that of Walker et al.(1990), was used for all experiments. Internal friction in this device of -100/o of thrust,the existence of which was previously known from ring-strain hysteresis, has been foundto be at the interface between the anvils and their cylindrical driving wedges. This frictionis a common property of all MA6/MA8 devices, not just the design employed here. Trun-cated edge length (TEL) anvil facets of 8 mm on MgO (50/o CrrOr) octahedra with epoxy-AlrO, gaskets achieved Bi III - Y at 242 + 6 tons (equivalent to 2.15 + 0.05 MN, I tonforce : 2000 lb force : 8.90 kN), in six experiments between 233 and 250 tons withstandard deviation of the mean, o : !2.5o/o. This gasket system gives excellent pressuri-zation efficiency and precision at room temperature and also affords nearly perfect blowoutsuppression. A procedure for fabricating the pressure medium and gasket fins from castableceramics in a single operation was developed for high-temperature experiments. Using anMgO-based castable ceramic at room temperature, Bi III - V on 8 mm TEL was foundat 266 + l1 tons (eight experiments,249-282 tons, o + 4o/o), whereas for 6 mm TEL itwas found at 159 + 6 tons (eight experiments, 149-168 tons, o + 4o/o). The advantagesofcastable ceramics, in addition to ease offabrication and rigorous dehydration, includetheir excellent high-temperature characteristics of thermal insulation, pressure stability,and thermocouple survival. Calibrations on the coesite to stishovite transition at 1200 "Cconfirm the good performance of these ceramics. Experiments of several days' durationwithin the melting range have been routinely achieved with this single-substance gasketand pressure medium system.

INrnooucrroN

Growing interest in mineralogical research at very highpressures (tens of GPa) on relatively large volumes ofmaterial (cubic millimeters) has prompted developmentof simplified equipment and procedures for multianvilexperiments. Walker et al. (1990) reported a cost-effec-tive device based on a system ofsliding, split-cylindrical,anvil-driving wedges within a deformable retaining ring.The wedges guide a uniaxial compressive force onto thefaces of a standard MA6/MA8 payload, i.e., a cubic clus-ter of eight separated cubic WC anvils converging on oc-tahedral pressure medium.

The pressurization of samples by such multianvil tech-niques requires judicious use of lubricating and gasketingmaterials. Some requirements of the material propertiesmay be contradictory. For instance, for compressive stroketo occur and cause pressurization, sufficient lubricationbetween sliding stages of the apparatus and sufficientcompressibility of gaskets between converging anvils mustbe available. However, good lubricants and materials oflow compressive strength, conducive to these require-

ments for pressurization, tend to be weak in shear, there-by promoting material extrusion or leakage with dimi-nution ofthe desired result. Successfirl experimental designis often therefore an exercise in seeking the proper bal-ance among competing efects.

One way to monitor the efficiency of pressurization andthe relative merits of different assemblages of materialsis to observe the thrust necessary for particular phasechanges to occur. The Bi transitions at room temperature(Lloyd, l97l) and the coesite - stishovite transition at1200 'C (Yagi and Akimoto, 1976) arc convenient ref-erences. Repeated observations of these transitions shouldgive some indication of the precision in the pressuriza-tion achieved from a specific thrust. The series of obser-vations served as a calibration study of thrust vs. samplepressure for the studies underway in this new laboratoryand as an opportunity to explore the effect ofchanges oflubricants, gaskets, and pressure media upon pressuriza-

tion in the search for more favorable assemblage of ma-

terials. The studies of reproducibility were undertaken toestablish the precision ofpressurization achieved with the

0003-004x/9 I /0708- l 092$02.00 r092

new gasket and pressure medium configurations; thesemay be of general interest to students of multianvil tech-nique because such studies at other laboratories aresomewhat inaccessible.

ExpnnrlrnNTAl METHoDSA second soft-shell multianvil module similar to the

one described by Walker et al. (1990) was built at theUniversity of Cambridge with the following dimensionalchanges for installation at Lamont-Doherty Observatory.The height and diameter of the split-cylindrical cluster ofsrx wedges were changed from 4.5 in. to 5.5 in. and from8 in. to 7 in., respectively. The retaining ring was stifenedby increasing the wall thickness of Hl3 steel from 8 in.id to 7 in. id x l0 in. od and increasing its overall lengthto 6.5 in. Al plate was substituted for the Tufnol l0G/40disks at the interface between the base ofthe wedges andthe pressure distribution plates, making it no longer nec-essary to circulate coolant between the wedges within thering during experiments at high temperature, although itwas still possible to do so. In other respects, the Lamont-Doherty version differs little from the Cambridge proto-type, which continues to perform well. (See Fig. l.)

This module was loaded into a uniaxial hydraulic pressof 600-ton capacity normally used for piston-cylinderwork (e.g., Agee and Walker, 1988). Oil pressure in thepress was read on a I kbar Heise Bourdon-tube gauge of14 in. diameter that was readable without vernier or in-terpolation in l-bar increments. Oil pressure was alsorecorded on a strip chart as a voltage across a bridgecircuit in a pressure transducer that was calibrated againstthe Heise gauge. Press thrust was calculated from theproduct of the ram oil pressure and the working area ofthe ram with l6-in. diameter. Ram pressurization wasaccomplished with an air-driven reciprocating oil pump;the rate of pressurization was manually regulated throughtrial and error adjustment ofa series oftwo needle valves.Pressurization to achieve all three Bi transitions was usu-ally accomplished in less than 2 h. Depressurizationthrough the same valves usually proceeded two to threetimes more slowly.

All experiments used the octahedron with integral fingaskets of Walker et al. (1990) with various substitutionsof material for the fins and octahedron. Holes for sampleloading were drilled between opposite faces of the octa-hedron. For the Bi studies, this hole was 1.5 mm in di-ameter and was stuffed with a 4-mm-long cylinder of99.999o/o Bi metal. Excess hole length on eirher side ofthe Bi in the middle of the octahedron was filled with Alor Cu electrodes to allow electrical communication be-tween the Bi and the resistance-measuring circuit outsidethe press. These Al or Cu electrodes contact opposing WCanvils, which become a part of the circuit. Bi resistancewas monitored as a voltage drop across the circuit thatincluded two WC cubes, two Al or Cu electrodes, and theBi cylinder. For a supply voltage of 5 VDC through a 50O resistor, this voltage drop was about 2 mv, about % ofwhich is attributable to the Bi. Voltage changes were re-

WALKER: MULTIANVIL EXPERIMENTS 1093

corded on the same strip chart as the ram oil pressure,and the three Bi transitions at room temperature wereeasily recognized during pressurization.

For the coesite - stishovite transition at 1200 "C, ahole with a diameter of t/s in. was drilled from one faceof the octahedron to the other. A tube fabricated fromtwo turns of 0.001 in. Re foil lined this hole and formedthe heating element. About 2 mm in the center of thistube was filled with SiO, glass powder that had a Pt vs.Pte./Rhr. (type S) thermocouple junction imbedded with-in it. Excess space within the tube on either side of theSiO, powder was filled with crushable AlrOr, AlrO, ther-mocouple ducts, and Zircar AlrO., cement (see Appendixl). The thermocouple leads were brought out through slitsin the ends of the heater tube and run out octahedral finsbetween WC cubes. Samples were pressurized cold, heat-ed to 1200'C for l-2 h, quenched, depressurized in 6-l0 h, and examined optically. Complete devitrificationwas always observed as coarse crystals of low birefrin-gence (coesite), as a mixture of coesite near the heaterand finer grained crystals of refractive index >1.70 andhigh birefringence near the thermocouple (which wereidentified as stishovite) or as stishovite alone. WC cubesof l-in. edge of Hertel grade KMY or KO5HIP (see Ap-pendix l) were used in all these experiments; the onlycube loss in the whole experimental program (three cubes)occurred as the result of a blowout during a too rapiddecompression (100 - 50 kbar in 20 min). At the rela-tively modest pressures examined in this study (less than100 kbar), very little anelastic deformation is sustainedby the carbide cubes. Thus, the differences between car-bide grades that are so obvious in calibrations at morethan 150 kbar are not an important factor here.

Substances used for the octahedral pressure mediuminclude the MgO with 50/o CrrO, used by Walker et al.(1990), Cotronics' compound 809, or Aremco's com-pounds 584, 575, and. 645. (Previous experience withcompound 5ll demonstrated its unsuitability, see Ap-pendix l). Gaskets as fins on the MgO octahedra wereeither epoxy-alumina as described by Walker et al. (1990)or compound 575. Gaskets on castable ceramic octahedrawere of the same material as the octahedron and werefabricated as an integral cast unit in segmented molds(Fie. 2).

Various lubrication arrangements were tried at the slid-ing interfaces exhibited in Figure l. At the A-B interfacebetween the wedges and the ring, a double thickness of0.004 in. glossy polyester sheet, with PTFE mold-releaseagent sprayed on the plastic to plastic interface, providedlubrication in all but one experiment. This experimentemployed a single layer of the plastic to illustrate thelosses experienced to friction on the A-B bore withoutthe double-plastic lubrication. At the A-C interface be-tween the wedges and the cubic anvils, I used pads offiberglass impregnated with epoxy resin, fiberglass im-pregnated with silicone resin, Nylon 66 impregnated withmolybdenum disulfide, a single thickness of 0.014 in.Mylar polyester, or double thicknesses of 0.007 in. Mylar

1094

Fig. I . Axial cross section of cylindrical module used to per-form multianvil experiments, closely similar to the unit de-scribed in Walker et al. (1990). A,, Ar, etc. are tool steel, anvil-driving wedges guided by the bore of alloy steel containmentring B. The upper set of wedges, A,-A., and lower set, Ao-Au,converge on a cluster of eight tungsten carbide cubic anvils, C,when activated by a uniaxial thrust, U, delivered through thealuminum alloy pressure distribution plates, G. The sample re-sides in the octahedral pressure medium at the convergence point

WALKER: MULTIANVIL EXPERIMENTS

of the wC cubes. The cubes have l-in. edge dimension. Theinterfaces that must slide past one another during sample pres-

surizations are indicated by traction shear couples at sliding in-terfaces. Friction on these interfaces should be minimized foroptimal sample pressurization. A-B and A-G tractions are in-trinsic to this design. The A-C friction is shared by all MA6/MA8 devices and appears to be the source of the residual strainhysteresis observed during loading and unloading of ring B oncethe A-B interface has been lubricated with double-plastic sheet.

with interfacial PTFE spray. Experimental results are giv-en in Table l

DrscussroN oF RESULTS: LUBRTcATToN

Inspection of Table I and Figure 3 shows a good sep-aration of the forces necessary to observe the Bi transi-tions for 8-mm truncated edge length (TEL) anvils andfor 6-mm TEL. The forces required are reasonably re-

producible and are observed to be roughly in the ratio of1.5/l for 8 mm/6 mm TEL. However, the expected ratioof forces for achieving some particular pressure is (8 mm/6 mm)'z : 1.8. This discrepancy from the observed valueof 1.5 reflects the fact that the forces are not supportedentirely by the truncation on the anvils but also by somemargin of the anvil flanks through the gaskets.

The control ofTEL upon the force required to achieve

WALKER: MULTIANVIL EXPERIMENTS 1095

Fig.2. (A) Partially assembled mold for casting pressure media and gaskets. White teflon sheet provides separation of the darkPVC cubes to form gasket fins. Cubes have various truncations for casting of appropriate TEL dimensions. @) Mold assembled.Final cube will be inserted after mix is poured into cavity. (C) Finished pressure medium with 6 mm TEL and gasket fins 3 mmwide by 2 mm thick. A hole 0. 128 in. in diameter has been drilled after firing at 1000 "C. Grid mesh 0. I in. in each picture.

1096 WALKER: MULTIANVIL EXPERIMENTS

TABLE 1, Bi transition results

Experiments tor I mm truncated edge length anvils

TonsExper.group Serial no. Notes Pads Gaskets B i l ' l l B i l l - l l l B i l l l - V

MgO (5% Cr2Os) pressure mediumA1 15 xx

12 xxxx'xxxxXXXX

Notes

M66G 1 0G 1 0G10G l 0G 1 0G 1 0

S ILG l 0

M 1 4G 1 0G10G 1 0

BDHBDH5750.3 pmBOH0.3 pmBDH

Summary of A1BDHBDH

F TTt - lF T

H F TH F TH F TH F T

Summary of Bj and B,

F TT I

71 9182 10567 92 250

919289

727373

2332392412!3248

A2 18 xx1 X

584 pressure medium and gasket

CUI

1 1

4

Bl 59234951

7 3 + 5 9 3 + 6 2 4 2 + 6100 134 34084 108 350

73 92 25874 92 25870 90 27379 99 282

9 3 + 480848292

7 4 + 464676874

8 4 + 58 9 + 6

869089

6 8 + 47 1 + 5

707471

xx'XXxx'XX

xxxx'xxXX

5558co57

B2 M 1 4M 1 4M 1 4M 1 4

266 + 12249267270277

266 + 12267 + 11

575 pressure medium and gasketC 2 2 x x

52 xx41 xx*

G 1 0MM7

232270284

7 2 + 27 2 + 4

8 8 + 29 0 + 6

262 + 27257 + 17Summary of A1, 81, Br, and C

Experiments for 6 mm truncated edge length anvils

Pads Gaskets B i t - t l

TonsExper.group serial no B i i l - i l l B i l l l - V

MgO (5% CrrO3) pressure mediumD, 20

2153

D, 171 9

584 pressure medium and gasketE 3 0

25244042444647

BDH5755757BDHBDH

F

FF TFF TF T

Summary of E

F TT I

Summary of D1, E, and F

XXxxxxxxxx

xxxxXXxx'xx*xxxx A'XX A*

M66G10G l 0S ILS IUG lO

G10M66

MM7MM7G10G 1 0G 1 0

M 1 4MM7

1 ;'176

197199

61668277

5261605758585 /

60

52cz525960

4550504748474649

149152158160160162163168

4 8 + 2 5 8 + 3 1 5 9 + 6575 pressure medium and gasket

F 5 4 x x39 xx

45 52 14946 55 176

48 + 2.5 5 8 + 4 1 6 1 + 9

/Vofe..xx : Double plastic liner on ring bore at A-B interface. X : Single plastic liner on ring bore at A-B interface. M-66: Pads of Nylon 66' 0.015

in. thick, impregnated with motybdenum ;isulfide. G1 0 : Epoxy-impregnated iiOerglass pads, 0.01 6-0.020 in. thick. SIL : Silicone-impregnated fiberglasspaOs, O.OtS in]tnict. M14 : Mylar polyester pads of single tnicfness, 0.014 in. MM7 : Vylar polyester pads of dor.rble thickness of 0.007 in. material.'BDH.:

Co"rse, gritty Al,Os in 6poiy. b.3 pm : Buehleipolishing Al,O3 in epoxy. 575 : Aremco's compound 575 used as gasket fins. F : G.aslets

and pressure meAiu; fir;d-at 1 OOO ;C. A : Crushabte aiumina Atl900 core around Bi cylinder. H : 10% Fe,O3 added to 584 castable ceramic. T :

Teflon tape wrapped on gasket extremities.- PTFE mold release agent added to A-C interfacial pads.

" Measurement fails or resistance trace uninterDretable.

the Bi transitions is the most obvious (and expected) as-pect of the data in Table l. Another control that is ob-vious from even the limited amount of data collected isthe sensitivity of the results to particular, poor-lubricat-

ing substances at the A-B and A-C interfaces. Sections A,

and D, of Table I give the results for experiments con-ducted with MgO (50/o CrrO.) pressure media that arerelevant to this proposition. Experiment I (section Ar)

differs from all other experiments reported in having asingle thickness of plastic lining the A-B ring-bore inter-face instead of a double layer with pTFE mold-releasespray between them. Walker et al. (1990) analyzed thehysteresis of strain observed during loading and unload-ing of the containing ring with the single plastic linerconfiguration; they concluded that about 300/o of thrustwas being lost to frictional effects. The hysteresis and pre-sumed loss of thrust to friction decreased but was notentirely eliminated by the use of the double-plastic ringliner configuration. A comparison of the 350 tons forcenecessary to produce the Bi III - V transition in exper-iment l, section Ar, with the mean of the results in sec-tion A, of 242 tons (for experiments using a double ringliner) confirms in a rough way the conclusion based onnng strain hysteresis: pressurization improvement of-300/o is realized with care in lubrication at the A-B in-terface.

Insight into possible causes of the residual strain hys-teresis over and above that ofbore friction may be gainedfrom considering experiment l8 in section A, of Table Iand section D, of Table l. These experiments all em-ployed the double-plastic ring liner but substituted Si-impregnated fiberglass pads for the material at the A-Cinterface between the wedges and cubes. This silicone-based material was tried because of its substantially high-er thermal stability than epoxy-impregnated fiberglass. Itcan be seen that this choice was not a happy one fromthe point of view of pressurization efficiency. Evidently,there is as much scope for introducing frictional problemsat the A-C interface as there is at the A-B ring bore.Furthermore, friction at the A-C interface is a circum-stance common to all MA6/MA8 devices, unlike the fric_tion at the A-B ring bore, which is specific to the splitcylinder design of Walker et al. (1990). Thus it is of gen_eral interest to become aware of the frictional propertiesof materials to be used at this interface, silicone fiberglasslaminate being a poor choice whatever its benefits of ther-mal stability may be.

Walker et al.'s (1990) analysis of the strain hysteresiswith a double-plastic ring liner suggested that - l0o/o ofthrust was still being lost to friction. They conjectured, ifthis residual friction were still to be found in the A-B ringbore, that Japanese-style MA6/MA8 devices based on arigid split sphere, which have no such potentially friction-generating interfaces as A-B or A-G, might show im-proved pressurization of about l0o/o when compared withthe split, sliding cylinder design if the same configurationof an MgO (50/o CrrOr) pressure medium with epoxy-AlrO.gaskets were to be tested. F. R. Boyd (personal commu-nication) recorded the Bi transitions on the Japanese-styleMA6/MA8 at the University ofAlberta (Edmonton) withan 8 mm TEL pressure medium and gasket assemblyfabricated in this lab. His result for Bi III - y e39-242tons) is indistinguishable from the mean in section A,,Table l, although his Bi I - II resulr (91-94 tons) issignificantly higher. Thus it is unlikely that there is sig-nificant friction at the A-B ring bore because a device

WALKER: MULTIANVIL EXPERIMENTS t097

M N t t z

. - t g @

1 I+ ] [

I + I - L

too 200 300TONS FORCE

Fig. 3. Bi transition experiments. Clear separation ofthe forcesrequired to cause transitions to occur is seen as a consequenceof change from 6-mm to 8-mm truncated edge length (TEL) onthe octahedral facets of the cubic anvils. Open circles are forexperiments using Aremco's 584 compound for the pressure me-dium and gaskets. Filled circles are for experiments using MgO(50/o Cr,Or) vdth bonded AlrO, gaskets. Only for 8 mm Bi III -V transitions is there a clear separation ofresults for the differentpressure media. Precision of the pressurization forces requiredfor achieving a transition is typically characterized by a stan-dard deviation + 40lo ofthe average pressure. Considering thatthese results were obtained on a series ofexperiments in whichmany parameters such as mix composition, gasket character, andA-C interface pads were also changed, it is possible that stillhigher precision might be obtained when attention is directedtoward uniformity of procedure rather than the effects of varia-tions. Nevertheless, the precision realized, here is comparable tothat quoted from other laboratories.

without such friction does not perform better. The im-plication of Boyd's result and of the sil icone-(non)lubricated experiments 17, 18, and 19 in Table 1 isthat nontrivial residual friction resides at the A-C anvilto anvil-driver interface. This will be of interest to thosecontemplating adoption of one design of MA6/MA8 de-vice over another. Residual bore friction is clearly not acompromising feature of the single ring design with slid-ing split cylindrical wedges. The residual friction that ispresent is a common feature of all MA6/MA8 designs; itis easily recognized from strain hysteresis analysis of aring assembly, but it would be more difficult to recognizein the same way in a split sphere design, even though itmust be present.

A more subtle aspect of the data in Table I that bearson the question of friction at the A-C interface is the

7.5

s8(,

75a

x

t!EfU)at-dE

LrJJo-

a

1098 WALKER: MULTIANVIL EXPERIMENTS

virtual independence of the results from choice of padmaterials, other than the silicone-impregnated fiberglass.Epoxy-impregnated fiberglass (key Gl0), thick single My-lar pads with or without extra PTFE spray (Ml4), anddouble-thickness thin Mylar pads with intervening PTFEspray (MM7) gave results that were statistically indistin-guishable. The good news is that single thickness (0.014in.) Mylar is commercially available at about 200/o of thecost of epoxy-impregnated fiberglass, and it has betterthermal stability. Unfortunately, at press loads of about400 tons on clusters of l-in. cubes, the (A-C) interfacialforces are sufficient to cause extrusion of the Mylar fromthe interface. Thus Mylar is an admirable substitute, whichcan be used without recalibration, for epoxy-filled fiber-glass interfacial pads, but only at modest press loads.

Of the pad materials in Table l, Nylon 66 filled withmolybdenum disulfide (key M66) performs a little betterthan the other materials in reducing A-C friction. In allthree experiments where this material was tested, nearlythe lowest forces required to cause the transitions wererecorded. This is not altogether a surprise because thismaterial is commercially fabricated expressly for linersand bearings that reduce friction. However, at the com-pressive forces of these experiments, the material is weakenough in shear to extrude even at modest press loads.As a result, M66 is forced between the wedges that areseparated under load. During unloading, the wedges relaxon the intruded M66, catching both the pads and thewedges in a grasp that makes disassembling the moduleextremely awkward. (This is the same problem thatplagues Mylar at press loads above 400 tons with l-in.cubes.) For this reason, further experiment with M66 wasabandoned, although inquiries were made (without suc-cess) for a product composed of fiberglass-reinforcedpolyester sheet filled with molybdenum disulfide.

DrscussroN oF RESULTS: cASKETINGAND PRECISION

The picture that emerges is one in which pressurizationof specimens can be strongly influenced by TEL selectionand by mistakes in the selection of lubricating substancesat various interfaces between stages in the apparatus. Insubsequent discussion, we ignore the data in sections A,and D, of Table I and the distinctions between pads atthe A-C interface. The residual information in Table I isa compilation of results by type of pressure medium andgasket employed, investigations at 8 mm TEL beingslightly more extensive than at 6 mm TEL. Each generalletter category (A, B, etc.) in Table I gives the result fora particular pressure medium and gasket combination.The multiple entries reflect on the precision of pressuri-zation to be expected from the different combinations.

Section A (Table l) gives results obtained at Lamontfor the combination of 8 mm TEL MgO (50/o CrrOr) oc-tahedra with epoxy-Al,O, gasket fins described in Walkeret al. (1990). It is clear that there is some variation as-sociated with these transition force measurements. Theaverages of the transition forces required for the Lamont-

based results in Table I (A) are lower than the force for

the single comparable Cambridge experiment reported byWalker et al. (1990). However, the Lamont range in-

cludes the Cambridge result comfortably within 2 sd of

the mean. Attention to the standard deviation of replicate

analyses should give an estimate of experimental preci-

sion. The standard deviations for each of the 8 mm TEL

Bi transitions for MgO in Table I (A) is +5-6 tons; on

the face of it, precision increases with pressure. For theBi III - V transition, the standard deviation ofthe force

for six measurements has diminished to x.2.5o/o of the

average. This is respectable precision. Variation of AlrO,grain size or binder (BDH in epoxy, 0.3 g.m in epoxy, or

575-a castable AlrO, ceramic) does not appear to change

this result.Given this performance characteristic, the ease of prep-

aration compared to some alternatives, and the virtual

absence of blowouts, one might hesitate before seekingadditional pressure medium, gasketing systems. Howev-

er, the high-temperature performance of this combina-tion leaves significant room for improvement. The epoxy

base to the AlrOr-powder gaskets loses strength at high

temperature, so excessive pumping is required to main-tain pressure at high temperature. This circumstance and

the accompanying extrusion of the gaskets and leakage of

the pressure medium has unfortunate consequences for

thermocouple survival rates. MgO, even with 5o/o CrrOr,is also a relatively poor thermal insulator at high tem-perature, leading to high power consumptions and un-necessary carbide and apparatus heating' Furthermore,the epoxy base for the fins precludes a rigorous firingprocedure to dehydrate the MgO once the assembly is

fabricated. To remedy these problems, fabrication ofpressure media and integral gaskets as a uniform ceramicin a single casting operation was explored. Aremco's 584potting compound was most extensively tested, althoughAremco 5l l, 575, and 645 and Cotronics 809 were alsoinvestigated. The fabrication procedure is extraordinarilysimple if appropriate molds are available: mix, pour, set,release from mold, and bake. No particular pains were

taken over mixing procedures or setting times and tem-peratures. Firing at 1000'C for 2 h is also recommendedas this improves the machinability of the product throughthe introduction of -150/o porosity. Firing is done prior

to drilling holes through the octahedra for samples andheaters and slicing slits for running thermocouple leads.Firing is absolutely essential for high-temperature exper-iments. Without firing, evolution of binder residues leadsto blowout upon heating a sample almost every time' Ofcourse, such firing can also be done with sample assemblyin place within the gasket and pressure medium assemblyto insure optimal dehydration.

Composite gasket and pressure media of compound 584were tested for both 8 mm and 6 mm TEL. For Bi I -

II and Bi II - [I, the forces required at 8 mm TEL arenot statistically distinguishable from those found for MgO/5o/o CrrOr, with the same standard deviations' Too fewMgO measurements were taken at 6 mm TEL to make a

statistically significant comparison, although no discrep-ancies in forces required for the Bi transitions at 6 mmTEL appear for the different pressure medium and gasketcombinations tested.

Unlike the case for 8 mm TEL MgO combinations, the584 pressure medium and gasket measurements do notexhibit a constant standard deviation with measuringpressure; they show an increase in standard deviation thatremains a roughly constant percentage ofthe average val-ue (-40lo). Also for 8 mm TEL Bi III - V determinations,it is clear that 584 requires more force than MgO. Re-calling that at 6 mm TEL there was no such obviousincrease of the force required for Bi III - V for 5g4compared to MgO combinations suggests that the depar-ture at 8 mm TEL may be a function of the higher thrustnecessary. Alternatively, the anomalously decreasing per-centage of the standard deviation of the MgO measure-ments with pressure may indicate an unrecognized prob-lem with the 8 mm TEL measurements for MgO.Although it is very difrcult to imagine that highly repro-ducible results should be problematic, the 8 mm TELresults for Bi III - V in MgO are anomalously well clus-tered compared with all the other data combinations, in-cluding Bi I - II and Bi II - I[ in MgO. Most of thedata sets collectively show a standard deviation that is-40lo of the average value. In any case, this lower preci-sion is well within the ranges quoted by other laboratones(e.g., Ohtani et al., 1982).

Another interesting feature of the data for the ceramicpressure medium and gasket combinations is the appar-ent lack of sensitivity to the details of the constitution ofthe ceramic. For instance, the limited data presentlyavailable for compound 575 do not show any strikingdeviations from the results for compound 584 at roomtemperature. This is somewhat surprising in that 575 isnearly pure AlrO., whereas 584 has hefty admixtures ofSiO, and MgO. Perhaps even more surprising is a com-parison of sections B, and B, of Table l. section B, is forregular 584, whereas B, is for 584 with 100/o FerO, powderadded to the mix. No statistical difference is seen betweenthe results of these two ceramics. Evidently, vigilance overthe proportions ofingredients and the process control inpressure medium fabrication are not key factors inachieving -40lo reproducibility.

The high-temperature performance of the castable ce-ramic gaskets greatly improves upon the epoxy-AlrO,combination. The latter typically dissipate 10-300/o of theapplied thrust by extrusion and gasket flattening when thesample is first heated. This thrust loss is unrecoverablewithout further pumping. By contrast, the 584 gasketsrarely lose more than 3olo of the applied thrust when heat-ed. This small loss is rapidly made up as the press heats.Evidently 584 is a tight gasket at high temperature. Themechanical stability inherent in this tightness is probablya key feature of our almost-perfect thermocouple survivalrecord with these gaskets. The thermal insulation prop-erties of 584 are such that about 340 watts are requiredfor our Re heaters to produce internal temperatures of

WALKER: MULTIANVIL EXPERIMENTS 1099

IONS FORCE

Fig. 4. Performance of Aremco's compound 584 on the Bitransitions at room temperature and for the coesite-stishovitetransition at l2O0 "C. The error bars on the Bi transitions are labounds on the variations observed in multiple experiments. Theerror bars (upon individual experiments) for the six C-S exper-iments reflect the increase in ram pressure experienced duringheating of the experiment with a closed hydraulic system. TheC or S designation corresponds to the SiO, polymorph recoveredat the thermocouple junction even though C may have beenrecovered at the heater marsins.

1200 "C in an 8 mm TEL assembly. By contrast, MgO(50/o CrrOr) pressure media absorb about 700 watts toreach this temperature. These mechanical stability andthermal insulation characteristics make experiments ofseveral days' duration with stable P, I routinely achiev-able.

A further indication of the satisfactory performance ofthis material at high temperature is given by the data inTable 2 and Figure 4 for the coesite - stishovite transi-tion at 1200'C. The forces required to cause pressuri-

Trele 2. Coesite and stishovite results at 1200 "C in castableceramic assemblies

Resu l t : 1200 'C(on heater wall)

6 mm TEL compound 584G10 212 - 226 coesiteM14 236 - 243 stishovite (coesite)M14 248 - 252 stishovite (coesite)

8 mm TEL compound 584366 - 374 coesite379 - 385 stishovite (coesite)390 - 394 stishovite

8 mm TEL compound 575416 - 418 coesite

8 mm TEL compound 645392 - 397 coesite

E mm TEL compound 809383 ' 386 stishovite (coesite)

Serialno. Pads Tons

65 C/S-166 C/S-267 C/S-3

70 c/s-s72 CIS-771 C/S-6

74 C|S-S

75 C/S-l0

85 C/S-12

G 1 0G 1 0G l 0

G10

G l 0

G 1 0

!

UElU)aLrlE(!

LrlJ

(t)

I 100

zatiot at 1200 'C are only minimally larger than for 25"C. This provides additional evidence of the tightness ofthe 584 gaskets. The entry for C/S-9 indicates howeverthat too much gasket stiffness can be counterproductive.Experiment C/S-9 employed the stronger pure AlrO. cast-able ceramic compound 575 for the gasket and pressuremedium combinations. At a force well in excess of thatrequired for the recovery of nothing but stishovite incompound 584, the stiffer 575 yielded only coesite, in-dicating reduced pressurization efficiency with gaskets thatare too strong. Experiment C/S-10 using silica-basedcompound 645 illustrates a similar result.

Experiment C/S-12 in Table 2 gives the result for Co-tronics compound 809, which is MgO based. Its perfor-mance is comparable to Aremco's MgO-based 584. Itditrers in having a coarser grain size for the MgO nuggets,a lower price, and superior mechanical cohesiveness bothbefore and after firing. Compound 584 occassionally givesdifficulty in curing properly without sufficient siliconegrease in the mold, a problem never experienced withcompound 809. However, the superior bonding of 809particles to one another is paralleled by an extraordinarilytenacious bonding to the mold even with silicone grease,with the result that it is somewhat less convenient to cyclethrough the casting procedure when the effort to separateand clean the molds is considered.

Both MgO-based ceramics alone have a useful temper-ature limit of about 1600 "C, when they begin to melt.This limit has been sucessfully extended past 2000'C byinsulating the ceramic from the Re heating tube with aZrO, sleeve. Initial drilling of the heater hole at sfi, in.leaves an oversize cavity that is coated with Cotronicszirconia cement compound 904. Once this paste has beencured and fired, the heater cavity can be redrilled to fin-ished diameter of 7s in., leaving a ZrO, liner to protectthe pressure medium from partial melting.

CoNcr,usroNs

In summary, the use of castable ceramic pressure me-dium and gasket combinations has much to recommendit, especially for experiments at elevated temperatures.Attention to the lubrication between all the sliding inter-faces in MA6/MA8 devices is desirable.

AcxNowr,nocMENTs

I thank F. R. Boyd for sharing cross-calibration results done at Ed-monton, the Earth Sciences Department of the University of Cambridgefor prowiding the module installed at Iamont-Doherty Geological Obser-vatory, Alan Hardstaffof the G. R. Stein Central Research Iab for sup-plying the MgO with 5Vo Cr'O,, and Roland Walker and John Weingastfor technical assistance. I thank Ivan Getting and Dean Presnall for theiruseful reviews and point out that the reviewers are not responsible formy nonuniform adoption of SI units. My excuse is that I report the unitsactually used rather than irrational SI values obtained after conversionfrom other systems. This work was supported by NSF and DOE and isLDGO contribution 4801 .

WALKER: MULTIANVIL EXPERIMENTS

RnnnnpNcns crrED

Agee, C.B., and Walker, D. (1988) Static compression and olivine flota-

tion in ultrabasic silicate liquid. Joumal ofGeophysical Research, 93,

3437-3449.Lloyd, E.C. (l 971) Editorial introduction and summary to: Accurate char-

acterization of the high-pressure environment. U'S. National Bureau

of Standards Special Publication, 326, | -3.

Ohtani, E., Kumazawa, M., Kato, T., and Irifune, T. (1982) Melting of

various silicates at elevated presswes. In S. Akimoto and M' Mangh-

nani, Eds., High pressure research in geophysics, Advances in Ea(h

and planetary sciences, vol. 12, p. 259-270. Wiley, London.Walker, D., Carpenter, M.A., and Hitch, C. (1990) Some simplifications

to multianvil devices for high pressure experiments. American Min-

eralogist, 7 5, 1020-1028.Yagi, T., and Akimoto, S. (1976) Direct determination ofcoesite-stisho-

vite transition by in-situ X-ray measurements. Tectonophysics, 35' 259-

270.

Mnmrscnrsr RECEIvED Ocroasn l, 1990MnNuscnrsr AccEFTED Apml 17. 1991

AppnNorx 1. Ma.rrnrnr,s AND SUPPLIERS

Alumina cement-Bulletlt ZPI-306. Zircar Cotpotation, I 10North Main St., Florida, New York 10921; U.S.A.

Tungsten carbide-Both KMY and KO5HIP nominally 50/o co-balt binder. KMY is mixed grain size, whereas KO5HIP isnominally micrograin size. Hertel AG, Postfach 175 1, D-8510Fiirth, Germany

MgO-5o/o CrrOr-Tiles of mixed grain size pressed and sinteredto 20o/o porosity were sawn into octahedra. Supplied by Mr.Alan Hardstafl G.R. Stein Central Research hboratory, SandyLane, Worksop, Nottinghamshire S80 3EU, U.K.

Castable ceramics - Proprietary formulations by the suppliers.Aremco Products Inc., P.O. Box 429, Ossining, New York10562-0429, U.S.A.

Compound 584 MgO basedCompound 575 AltO, basedCompound 645 SiO, basedCompound 51 I Zircon-MgO based

Cotronics Corp.,3379 Shore Parkway, Brooklyn, New York11235, U.S.A.

Compound 809 MgO basedCompound904 ZrOrcemenl

Semisintered tube and rod:Alumina-grade AN900, Norton Co., Worcester, Massachu-setts 01606, U.S.A.Magnesia-grade HP, Ozark Tech. Ceramics, 402 Ware St.,webb City, Missouri 64870, U.S.A.

Thermocouple junction tubing: 99.80/o alumina, two-hole andfour-hole 0.015 in. id, 0.060 in. od. Bolt Technical Ceramics,Box 718. Conroe, Texas 77305-0718, U.S.A.

Thermocouple lead ducts: mullite, single bore, 0.030 x 0.015in., Coors Porcelain, 600 9th St., Golden, Colorado 80401,u.s.A.

Epoxy-filled fiberglass laminate: GlO grade mfg. by Westing-house, 0.016 in. thick, and Polyester sheet: 0.004, 0.007, and0.014 in. thick Mylar, Ain Plastics, Box 151, Mt. Vernon, NewYork 10550, U.S.A.

MDS-filled Nylon 66: 0.015-0.020 in. thick strip, McMaster-Carr, Box 440, New Brunswick, New Jersey 08903, U.S.A.

Re strip: 0.001 in. sheet for rolling heater tubes, Sandvik Rhe-nium Alloys, Box 245, Elyria, Ohio 44036, U.S.A.