sign in sign up
<<
Home , Scotia Sea

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 87, NO. B8, PAGES 6697-6710, AUGUST 10, 1982

GeologicalEvidence For The GeographicalPattern of Mantle Return Flow and the Driving Mechanism of Plate Tectonics

WALTER ALVAREZ

Departmentof Geologyand Geophysics,University of California,Berkeley, California 94720

Tectonic features at the 's surface can be used to test models for mantle return flow and to determine the geographicpattern of this flow. A model with shallow return flow and deep con- tinental roots places the strongestconstraints on the geographicalpattern of return flow and predictsrecognizable surface manifestations. Because of the progressiveshrinkage of the Pacific (averaging0.5 km2/yr over the last 180 m.y.) this model predicts upper mantle outflow through the three gapsin the chain of continentsrimming the Pacific (,,Australian- Antarcticgap). In this model, upper mantle return flow streamsoriginating at the westernPacific trenches and at the Java Trench meet south of , filling in behind this rapidly northward- moving and providingan explanationfor the negative bathymetricand gravity anomalies of the 'Australian-Antarctic Discordance'. The long-continued tectonic movements toward the east that characterize the Caribbean and the easternmost may be produced by viscous cou- pling to the predictedPacific outflow through the gaps, and the Caribbean floor slopes in the predicteddirection. If mantle outflow does passthrough the gapsin the Pacificperimeter, it must passbeneath three seismiczones (, Lesser Antilles, ScotiaSea): none of these seismic zones shows loci below 200 kin. Mantle material flowing through the Caribbean and Drake Passagegaps would supplythe Mid-Atlantic Ridge, while the JavaTrench suppliesthe Indian ridges,so that deep-mantleupweilings need not be centeredunder spreadingridges and therefore are not required to move laterally to follow ridge migrations. The analysisup to this point suggests that upper mantle return flow is a responseto the motion of the . The secondpart of the paper suggestsa possibledriving mechanismfor the plate tectonic processwhich may explain why the continentsmove. This hypotheticaldriving mechanismhas four aspects:(1) the lower mantle convects, (2) this convectiondrives the continentsby drag on their deep roots, (3) return flow in the upper mantle providesa volumetric balancefor the motion of the continental masseswithout passingunder them, and (4) oceanicplates are effectivelydecoupled from the asthenosphereand are driven largely by slab pull. This mechanismaccounts for the openingof the Atlantic, the ability of spreadingridges to stay on the midlines of ,and the penetrationof India into follow- ing their collision. Continental motions strongly imply lower mantle upwelling and divergence beneath the Atlantic and southeast Indian Oceans, and convergence and subsidence along the Tethyan belt. This picture disagreeswith the concept,derived from hot spot studies, of an unde- forming, absolute reference framework at depth, but weaknessesin current hot spot theory would make a rejection of the present model on this ground premature. If the present model is generally correct, a fairly simple pattern of lower mantle convection, with as few as four cells, may explain much of the tectonic complexity of the earth.

INTRODUCTION example is the flow-roll model of Richter and Parsons [1975] and Richter [1978], which has been discussed At the present time the geometry of plate move- by Marsh and Marsh [1976, 1978] and Watts[1978]. ments is largely understood, but the driving mechan- The present paper gives an alternative model for the ism of plate tectonics remains elusive. There has been driving mechanism, derived by a two-step procedure. much discussion of convection in the mantle, in some The first step involves a consideration of the geo- casesinvolving the entire mantle [Kanasewich,1976; graphic pattern of mantle return flow, without regard Gough,1977; Davies,1977] and in other caseswith the to driving mechanism. A model is developed in which upper and lower mantle separate [Richter, 1979; Liu, mantle return flow extends no deeper than the mid- 1979; Chase,1979b]. In some treatments,mantle con- mantle transitionzone (420-670 km; Dziewonskiet al., vection is dominated by forces generated in or by the [1975]) and is probablyconfined to the asthenosphere. plates themselves, in particular, slab pull, ridge push, This flow is restricted to suboceanicpaths becauseof and sliding of the plate away from the elevated ridge the presence of continental roots. This model explains [Forsythand Uyeda,1975]. None of the suggested a number of tectonic features, including the unusual driving mechanismsseems to be capableof explaining bathymetry south of Australia and eastward move- all the observed features of plate kinematics. The ments in the Caribbean and easternmost Scotia . variety of such features is probablytoo great to be The second step explores the possibility that the plates explained by any single, simple mechanism. The which contain continents are driven by coupling answer may well lie in a more complex model, com- between lower mantle convection cells and continental bining various aspectsof the simpler models. An roots. Oceanic plates, underlain by the weak astheno- sphere, are considered to be driven primarily by slab Copyright 1982 by the American GeophysicalUnion. pull. This model appears to be capable of explaining Paper number 2B0395. many aspects of plate kinematics. 0148-0227/82/2B-0395505.00 At this point the rationale behind the present study

6697 6698 ALVAREZ:MANTLE RETURNFLOW AND PLATE-DRIVINGMECHANISM

should be emphasized. The physical conditions within the model considered here and will be discussed at the mantle are controversial, and many conceptual various places below. models for the movement pattern in the mantle have To approachthe return flow questionfrom a geologi- been proposed. It is generally difficult either to prove cal viewpoint, one might ask whether mantle return or disprove any model on the basis of existing geophy- flow would leave any traces recognizableat the surface. sical data. The approach here is to consider one model The question of depth to the top and bottom of the which predicts observable geologic effects at the sur- return flow layer remains unresolved, and one can face of the earth. At least some of the predictedgeo- consider, therefore, models based on various logic features are found to exist, and in view of this configurationsof the return flow layer (Figure 1). If partial support, the model deserves further examina- the whole mantle convects and the return currents are tion. at great depth, surface manifestations would be MANTLE RETURN FLOW unlikely (Figures1 c and 1 d). If, however,return flow occurs only in the upper mantle, it could well leave Modelsfor Mantle Return Flow tracesin surfacemorphology (Figures 1a and 1 b). In this case the depth to the bottom of the lithosphere is The movement of plates at the earth's surface of great importance, for if subcontinental lithosphere is requires a return flow of mantle material, but the pat- several hundred kilometers thick, return flow could be tern of this flow is unknown. Recent debate on flow restrictedto suboceanicpaths (Figure 1 a). in the mantle has concentrated on the depth to which Mass Balance and the Shrinkageof the Pacbqc convectiveflow extends [Smith, 1977; Elsassetet aL, 1979; Busse,1981], with opinionpolarized between the Mantle return flow must transport material from whole-mantle convection view and the shallow convec- areas of lithosphere consumption toward areas where tion view. Evidence from observational and theoreti- new lithosphere is being generated. Despite rapid cal geophysicsis ambiguous on this question, and no spreadingat the East PacificRise, net lithospherecon- consensus has been reached. sumption is presently concentrated, in a general way, A second critical question, which has received less in the Pacific, while lithosphere production is dominant attention, is the depth to the top of the layer in which in the hemisphere surrounding . One would mantle return flow takesplace [Chapmanand Pollack, therefore expect mantle return flow to proceed from 1977]. If the lithosphereis everywhereabout 100 km the Pacific hemisphere to the African hemisphere. thick, as shown in the conventional cross-sectional diagrams, then mantle return flow will pass beneath both continentsand oceans. However, Jordan[1975a, deep roots b] and Sipkinand Jordan[1975] have madea casethat yes no subcontinental mantle may differ from suboceanic a mantle down to depths of at least 400 km. If this is so, it indicates that in order to maintain the continent-ocean contrast, the continents must have deep, permanent roots and that lithosphere in the tec- tonic sense(Jordan's 'tectosphere') is severalhundred kilometers thick beneath the continents. A similar conclusion has been reached on the basis of strontium isotopicratios [Brookset al., 1976] and on the basisof heat flow considerations[Pollack and Chapman,1977; Chapman and Pollack, 1977]. However, Anderson [1979]disagrees, finding no seismologicalevidence for continental roots deeper than 150-200 km. The ques- tion will not be debated here; the point is that in view of this unresolved controversy, one can entertain models either with or without deep continental roots. The question of the geographicpattern of mantle return flow was first investigatedby Garfunkel[1975] and subsequentlytreated by Chase [1976, 1979a], Hager and O'Connell[1976, 1979], Harper [1978], Alvarez [1978], and Parmentierand Oliver [1979]. These subsequent studies support the conclusion of Fig. 1. Four models for mantle return flow, dependingon Garfunkel[1975] that return flow cannot be accom- whether the return flow is in the uppermantle (top) or the lower mantle (bottom), and on whether the continents plished by closed convection cells; mantle material have deep lithosphericroots (left) or whether the litho- must flow from shrinking to expandingreservoirs in sphereis everywhereabout 100 km thick (right). Because responseto seafloorspreading, subduction, and lateral of its severe constraints, mantle return flow according to plate motions. Parmentier and Oliver [1979] evaluated the model in Figure 1 a is the most likely to leave traces at the earth's surface. The model in Figure l a, in which the effect of differing lithosphere thickness beneath crossed circles are flow lines passing through the gap, is continents and oceans. Their results are relevant to investigated in this paper. ALVAREZ:MANTLE RETURN FLOW AND PLATE-DRIVINGMECHANISM 6699

Viewed in another way, the has been contracting for the last 180 m.y. Since the growth of the has roughly balanced the disappear- ance of the Tethys Ocean during this time [Smithand Briden,1977], the Pacifichas contractedby roughlythe area of the Atlantic. As a result, the subduction zones surrounding the Pacific have been forced to retreat . G toward the center of the Pacific;EIsasser [1971] termed this process 'retrograde motion.' This again leads to the conclusion that net mantle return flow must move material from the Pacific to the Atlantic. Measurements on the equal-area paleocontinental maps of Smith and Briden [1977] show that since the beginning of Atlantic seafloor spreading at 180 m.y. B.P., the area of the world ocean has decreased by the areaof theformer Tethys Ocean (5.0 x 107 km2), and increasedby the area of the present Atlantic (7.5 x 107km 2) and Indianoceans (6.6 x 107km2). The netgain in area(9.1 x 107km 2) haspresumably been balanced by loss of area from an originally larger Pacific Ocean over the last 180 m.y., at an average rate of 0.5 km2/yr.Garfunkel [1975] gives the following rates of plate area change for the last 5-10 m.y., in km2/yr:Pacific,-0.45; Nazca,-0.11; Cocos,-0.08; Antarctic, +0.50. Roughly one quarter of the Antarc- tic plate accretion affects the Pacific Ocean, so these valuesindicate a current shrinkage rate of 0.52km2/yr for the Pacific. Using numbers from Minster and Jor- dan [1978], one finds that the is presentlygrowing ata rateof about0.45 km2/yr, while the IndianOcean is growingat about0.15 km2/yr (accretion in the Indian Ocean is nearly compensated by subduction at the Java Trench); these values require that the Pacific Ocean be shrinking at a rate of about0.6 km2/yr.Thus, although the shrinkage rate of the Pacific has varied through time as spreading rateschanged [Larson and Pitman,1972; Baldwinet al., 1974], the current rate is in good agreementwith the long-term average rate.

Gapsin the PacificRims

In a model with a uniform 100-km lithosphere, material forced away from the Pacific Basin could escape anywhere around the perimeter. In a model with upper mantle return flow and thick subcontinental lithosphere (Figure la), the continental roots would form a barrier to flow away from the Pacific, and the escapingmaterial would be funneled through the gaps between the continental masses that nearly encircle the Pacific (Figure 2). Continental crust is continuous between Alaska and and continuous or nearly so between the Sunda Shelf and Australia. There are only three gaps in the continental perimeter of the Pacific Ocean -- the Drake Passage between and Antarctica, the southeastern Indian Ocean between Australia and Antarctica, and the . Central America has a crystallinebase- ment of lower Paleozoic and possibly older rocks as far Fig.2. Obliquecylindrical proiection tangent to theglobe south as central Nicaragua; the rest is underlain by a alonga greatcircle (a-a') approximatingthe perimeterof younger volcanic basement which cannot be con- thePacific Ocean (pole at 15øN,0øE). Large,open arrows showthe suggestedmantle return flow through gaps in the sideredcontinental [Dengo, 1969]. Pacificperimeter. C, Caribbean;S, Scotiaarc-Drake Pas- Thick subcontinental lithosphere, if it does exist, sageregion; A, Australia-Antarcticagap. 6700 ALYAREZ:MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM certainly does not underlie all continental crust. This Outflow through the Caribbean and Drake Passage is shown, for example, by the active, shallow-dipping gaps would be expected, but in the third model of Par- slabs moving several hundred kilometers eastward mentierand Oliver[1979, Figure 7] this doesnot occur, from the trench off South America at depths up to 700 apparently because their boundary conditions provide km [Barazangiand Doretan, 1969; Stauder,1975] and other gaps in the Pacific rim. One gap is through the by the probable former presence of similar, shallow- Bering Straits, Ocean, and ; man- dipping slabs under the during tle flow through this route feeds the North and Central much of the Tertiary [Coneyand Reynolds,1977]. The Atlantic ridges. A second gap, through the Mediter- former slabs reconstructed by Coney and Reynolds ranean, feeds the Central and South Atlantic ridges, [1977] on the basis of volcanic episodesapparently with some contribution from flow passing south of passed at shallow depth beneath Arizona, an area of Africa. However, continental crust is continuous Precambriancratonic crust [Burch/ieland Davis, 1972, across both these paths; if they are closed at depth by 1975]. This situation seems to be peculiar to the continental roots, upper mantle return flow will be young orogenic belts bordering the Pacific; thick litho- forced to pass through the Caribbean and Drake Pas- sphere, if it exists, would underlie the more ancient sage gaps. Geological evidence argues that this is the crust of cratons, undisturbed by young tectonic and case. thermal events [Brookset al., 1976]. The third model of Parmentierand Oliver[1979, Fig- The Caribbean and Drake PassageRegions ure 7] has very thick subcontinentallithosphere, so it is comparable to the situation considered here, yet it The Caribbean Sea and the Drake Passage are both shows no outward flow through any of the three gaps characterized by complex tectonic histories which are in the Pacific rim. In the case of the Australia- not yet completely understood. Continental recon- Antarctic gap this is because the Java Trench provides structions show that both have opened up by a major source of return flow material on the west side extensionsince the breakupof Pangea[Bu!lard et al., of Australia. Flow lines beginning at the Java Trench 1965; Barker and Gr.iffiths,1977]. However, the north and the Western Pacific trenches stream around the and south boundaries of the two regions were not sim- western and eastern sides of Australia, respectively, ply passive trailing margins; there is much evidence for converging on the southern side, where this flow pat- lithospheric consumption and compressional tectonics tern produces a sharp negative perturbation of about along these margins at various times during their his- 25-30 regal in the calculatedgravity field. Weisseland tories [Bell, 1972; Mattson,1973, 1979; Maresch,1974; Hayes [1974], using the satellite-derivedgravity field Ladd, 1976; Mattsonand Pessagno,1979]. of Gaposchkinand Lambeck[1971], show a negative However, here we are most concerned with the evi- gravity anomaly of 30-35 mGal immediately south of dence for eastward motion in both regions. In the Australia, in almost exactly the position predicted by Caribbean, long-continued eastward motion is shown the third model of Parmentier and Oliver [1979]. by the system of right-lateral faults along the northern Weisseland Hayes[1971, 1974] and Hayes[1976] have edge of South American and by the left-lateral fault made detailed studies of the Australian-Antarctic system extending from Guatemala to Puerto Rico. Discordance, a low saddle on the southeast Indian The Lesser Antilles subduction zone and the short spreading ridge, immediately south of the negative spreading ridge segment in the Bartlett Trough indicate gravity anomaly just mentioned. The discordance is that most of the Caribbean oceanic crust and parts of characterized by lineated topography oriented north- the adjacent continent belong to a Caribbean plate south in a band between the ridge and the south Aus- which is presently moving eastward at about 2 cm/yr tralian margin, indistinct magnetic anomalies, and an with respect to North and South America [Jordan, unusual history of asymmetric seafloor spreading. At 1975c; Minster and Jordan, 1978]. a given crustal isochron, the sea floor is systematically The situation is less clear in the Drake Passage. deeper on the north side than on the south side of the Although the South Sandwich subduction zone ridge. These features are not easily explained, but the geometrically resembles that of the Lesser Antilles, topographic saddle at the discordance may mark the there is a north-south spreading center only a few hun- convergence of the flow paths around Australia, the dred kilometers to the west, so that only the very south-to-north depth variation suggests a component small Sandwich plate in the easternmost Scotia Sea is of asthenospheric flow in that direction [Weissel and currently moving east with respect to South America Hayes, 1974], and the band of lineated topography [Barker,1972]. Magneticlineations in the Drake Pas- may mark the path of this flow. Although Parmentier sage show southeastward motion of West Antarctica and Oliver[1979] did not discussthe regionsouth of away from South America [Barker and Burrell, 1977]. Australia, its peculiar features provide dramatic sup- Major eastward movements in the past are required by port for the way their third model treats this the close structural and sedimentologicalties between and for the closely similar model considered in the South Georgia Island and southernmost South Amer- present paper. An important point is that although the ica [Dalzielet al., 1975; Winn,1978]. ocean south of Australia is the biggest gap in the rim Since the Atlantic Ocean is not being subducted of the shrinking Pacific, upper mantle outflow will not under any other part of North or South America, the occur there because of the presence of another major lithosphere of the Caribbean and the easternmost source, the Java Trench, outside the gap. Scotia Sea must be moving eastwardrelative to the two ALVAREZ: MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM 6701

American plates, and overriding the Atlantic crust. It tal barriers. The mantle outflow idea was anticipated is difficult to consider the Caribbean and eastern Scotia by Hamilton[1963, p. 14], who suggestedthe possibil- Seas to be typical marginal basins of western Pacific ity that the type, for the latter are apparently the consequence of subduction of Pacific lithosphere all along the western ...represents disruption and scattering of continental margin of the Pacific Ocean, whereas the former are material, whereby a sort of subcrustal jet stream, mov- local perturbations of a passive margin and require a ing eastward from the Pacific, tYagmentedand stretched out an initially compact land mass, by strike-slip faulting different explanation. To emphasize this difference, and tensional rifting. A similar explanation appears the opening of the western Pacific marginal basinscan applicable to the Caribbean Sea. be considered a second-order effect produced by the downgoingslab [Karig, 1974, Figures7A and 7D]; it is The outflow hypothesis requires that mantle flow more difficult to explain the eastward motion of the pass at shallow depth beneath three active subduction Caribbean and Scotia Seas relative to South America in zones -- those of the Lesser Antilles and Panama in this way, since there would be no downgoing slab the Caribbean region and the South Sandwich subduc- without the eastward motion. tion zone in the Scotia area (Figure 2). If the earth- The maximum eastward transport in both the Carib- quake foci marking these subduction zones extended bean and Scotia regions has I•een about 3000 km. In to several hundred kilometers in depth, this would the southern Caribbean, right-lateral strike slip motion clearly invalidate the hypothesis for it would be evi- began in the Eocene or Oligocene in northern Colum- dent that the descending slabs were passing undis- bia [,41varez,1971], and in the middle Eocene in turbed through areas where lateral flow should occur. Venezuela [Be!!, 1972]. In the northern Caribbean, It is noteworthy, therefore, that seismicity in these left-lateral motion certainly began by middle Eocene three zones extends no deeper than 200 km, which time, although the geological evidence is also compati- places them among the shallowest of the descending ble with initiation of this motion as early as 85 m.y. lithosphereslabs [Barazangiand Dorman,1969; Tomb- B.P. [Mattson,1979]. Also in the northern Caribbean, !in, 1975; Bowin, 1976; Forsyth, 1975; Gutenbergand left-lateral motion on the Cayman Trough apparently Richter,1949; Rothk,1969]. The seismicityinformation began in the Eocene [Per.fit and Heezen, 1978]. In thus indicates that there is no lithospheric slab curtain their tectonic reconstruction of the Caribbean, Malfait in the way of the inferred mantle return flow. and Dinkelman [1972] show eastward motion of the In contrast to these shallow seismic zones, the zone Caribbean Plate by the beginning of the Oligocene, but dipping north beneath Indonesia reaches depths of they show northeastward penetration of Pacific litho- 600-650 km from western Java to Timor [Fitch, 1970; sphere into the Caribbean region from the Late Creta- Cardwelland Isacks, 1978]. Although the boundary ceous on. The situation is less clear in the Drake conditions in the third model of Parmentier and Oliver Passage-Scotia Sea area. South Georgia Island has [1979, Figure 7] permit mantle flow beneathSoutheast probably moved 1500 kilometers to the east [Dalziel et Asia and Indonesia, the Java-Sumatra lithospheric slab a!., 1975] by left-lateral strike slip motion along the curtain shows that there is no Pacific mantle outflow in northern side of the Scotia Sea, with separation of this region. South Georgia and occurring between Rates of upper mantle flow through the gaps in the the early Oligoceneand the middle Miocene [W inn, Pacific rim would best be obtained by a calculation of 1978]. The south margin of the Scotia Sea does not the type presented by Parmentierand Oliver [1979], mirror the north margin; recent reconstructions using boundary conditions appropriate to the model require left-lateral strike slip motion on both margins discussedhere, but a rough estimate will be sufficient [Barker and Griffith&1972; DeWit, 1977], and seismic at present. Using values calculated on the basis of the information shows this to be the present pattern as work by Minster and Jordan [1978], it appears that well [Forsyth,1975]. mantle supply required by Indian Ocean spreading Jordan [1975c] showed that the Caribbean plate is (IND-ANT:0.52 km2/yr; IND-AFR: 0.11 km2/yr) is nearly fixed with respect to the 'absolute' mesosphere metby inputat theJava Trench (0.48 km2/yr) aug- reference frame deduced from hot spot data, and he mented by a contribution from the Pacific, passing suggested that the subduction zones flanking the around the east side of Australia. Thus, to a first Caribbean plate on the east and the west pin it to the approximation,Atlantic expansion (0.45 km2/yr) mesosphere. This may be correct, but an alternative would be supplied by flow through the Caribbean and possibility is that the Caribbean plate is disconnected Scotia gaps, each of which is about 600 km wide. If from the deeper mantle and moves eastward relative to return flow takes place everywhere in the same inter- the at a rate which happens to compensate val, the average outflow rate through the gaps is about roughly for their westward movement relative to the 38 cm/yr. In plate tectonic terms this is a high velo- hot spot framework. Problems with the 'absolute' city, but it is still not certain that the Caribbean and reference frame are discussed below in connection Scotia lithospheres (if they are underlain by normal with the hot spot data. The eastward motion in the oceanicasthenosphere) would be carried eastwardby Caribbean and easternmost Scotia Seas is just what is drag resulting from eastward mantle outflow. A more to be expected if the outflow of Pacific mantle material effective coupling between mantle outflow and the required by the contraction of the Pacific is concen- lithosphere in gaps may result from the presence of trated and funneled through the gaps in the continen- seismic slabs down to 200 km and from the lateral 6702 ALVAREZ: MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM contact between the outflowing mantle and the flank- Ludwig, 1977] and gravity field [Bowin, 1976] of the ing continental keels along the north and south mar- Caribbean before this could be taken as evidence for gins of the gaps. In favor of this view is the observa- the present hypothesis. When considering the ocean tion that the major strike slip faults bounding the floor outside the Caribbean, thermal effects are impor- CaribbeanSea (e.g., the Oca and Cuisa faults [A!varez, tant because the Cocos plate is quite young. Thermal 1971]) are located within the continentalcrust, 100- effects have been removed in the calculation of depth 200 km from the ocean rather than at the ocean- anomalies [Cochran and Talwani, 1977], which show continent boundary. that the Pacific Ocean immediately west of Central If there is uppermantle flow-passing through the America is several hundred meters shallower than gaps in the Pacific rim, there should be a difference in expected and that the Atlantic Ocean adjacent to the bathymetric level between the upstream and down- Antilles is slightly deeper than expected. These effects stream ends of the gap, representing the head that may be unrelated to the outflow mechanism, but they drives the flow. To test whether this difference in are correct in sign and of reasonable magnitude and level should be large enough to detect, the problem would seem to merit further study from this viewpoint. can be treated as flow between parallel plates, ignoring The presence of an active spreading center within the changes in the north-south direction, a simplification Scotia Sea and the lack of depth anomaly information which introduces much less error than that due to makes it difficult to look for a gradient in lithostatic large uncertainties in the parameters of the problem, head in this area. Return flow outside the constricted and disregarding motion between the upper plate gaps would be much slower and less likely to be (lithosphere)and lower plate (base of the return flow marked by recognizable depth variations. layer), sincethis velocityis an order of magnitudeless than the return flow velocity. For these conditions the Supply of Material to the Ridges discharge per unit width is given by a 3 dP Mass balance requires that mantle material expelled from the shrinking Pacific Ocean be transported to the Q=12txdx expanding Atlantic Ocean. In the present model, where a is the thicknessof the return flow layer, /x is Pacific mantle outflow escapesonly through the gaps, its viscosity,and dP/dx is the pressuregradient. For a gradient manifested by a slope in regional bathy- metry, f dP= pgAh 6OO 0 1 at 2821 dx L where L is the length of the gap, about 3000 km for both the Caribbean and Scotia regions, and A h is the Ah= 5 km elevation difference between the upstream and down- 5OO stream ends of the gap. Thus,

Ah= 12gLla 2 p ga 400 where V is the average outflow velocity, estimated 0 above as 38 cm/yr. The value for A h obtained in this way is model 300 dependent, and Figure 3 shows the effect of choosing differentvalues for/z and a• Elevationdifferences less L = 3000 km than a few hundred meters would not be detectable; differencesgreater than 3 or 4 km would scarcelyfit in 200 = 38 cm/yr oceanic depths. Although a wide range of acceptable models (for example, 5 of the 7 modelsused by Hager and O'Connell [1979]) yield values for A h that are impossibly large, there is a set of other acceptable I00 models,with viscosities of 10 i9 to a fewtimes 102øP o in return flow layers greater than 100 km thick, which IE,Vll give A h values that are both possibleand detectable. 0 It is therefore interesting to note that the floor of the 18 19 20 21 22 23 oceanic part of the Caribbean does in fact slope down log P from west to east, with mean depths of about 3500 m Fig. 3. Expected bathymetric difference, 3, h, over the in the Colombian Basin and 4500 m in the Venezuelan 3000-km length of the Caribbean and Scotia gapsas a func- Basin [Saunderseta!., 1973, Figure 2]. This slope is tion of return flow layer thickness,a (km), and viscosity,• (poise), for an average outflow velocity of 38 cm/yr. consistent with the range of calculated values, but Models used in previous calculationsare shown by triangles careful consideration would have to be given to the [Chase, 1979a], circles [ttager a,d O'Connel( 1979], and crustal structure [Ludwig eta!., 1975; Houtz and square[Parmentier and Oliver,1979]. ALVAREZ' MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM 6703 and it is this outflow that must feed the spreading continental roots, but it does not offer an explanation Mid-Atlantic Ridge. Depth anomalies in the North for why the plates are moving. This is true also of the Atlantic Ocean [Cochran and Talwani, 1977] suggest recentmodels of Harper[1978], Chase[1979a], Hager that Iceland may mark an additional source of and O'Connell [1979], and Parrnentier and Oliver material, unrelated to the mechanism under discussion [1979], which, like the present paper, treat mantle here. return flow as a response to movements of the plates. In the present model, mantle return flow above the Shrinkage of the Pacific and the resulting mantle midmantle transition zone feeds the Atlantic and outflowsuggest that motions of the continentsmay be Indian spreading ridges, but this motion does not drive the dominant control on return flow in the mantle. the plates apart. If the continents bordering the The evidence for deep continental roots derived from spreading oceans are moving apart for other reasons, seismology[Jordan, 1975a, b], from geochemistry as discussedbelow, mantle return flow simply fills in [Brookseta!., 1976],from heatflow results [Chapman the zone of separation. Since the area of passive dike andPollack, 1977], and from the considerationsgiven injection at the rift axis is the hottest and therefore the in this paper suggestsa possibledriving mechanism weakest place, it will continue to be the locus of that has apparentlynot previously been proposed. One separation, spreading will be symmetrical, and the can envision a model in which (1) the lower mantle ridge will remain on the medial line of the growing (belowthe midmantletransition zone) undergoescon- ocean [Morgan, 1971]. In a few places,other factors vective overturn, (2) this convection drives the con- intervene, producing discontinuous ridge jumps or tinents by viscous coupling to their roots, (3) return their continuous equivalent, asymmetric spreading flow in the upper mantle providesa volumetric com- [Hayes, 197 6]. pensation for the motion of the continental masses In the present model, the Atlantic ridge system is without passingunder them, and (4) oceanicplates, fed by material derived from the Pacific, which has underlain by weak asthenosphere,are decoupledfrom escaped through either the Caribbean or the Scotia the lower mantle and are driven largely by slab pull gap. These gaps are positioned in such a way that they (Figure 4). can feed all partsof the Mid-Atlantic Ridge (Figure 2); 1. Radioactiveheat sourcesdistributed through the in fact, they flank the SAM-AFR segment which mantle and the heat output of the core are sufficient to presently accommodates two thirds of the Atlantic ensure that the entire mantle convects [Verhoogen, expansion. However, the geographic pattern of this 1980], althoughit is not clearwhether the upperand flow is obscure downstream from the gaps becauseof lower mantles convect separately or together. Plate the effective decoupling of the lithosphere from the motions are sufficientlycomplex to precludea simple deeper levels that apparently occurs everywhere in the pattern of whole-mantle convection; they suggest, oceanic regions except possibly above the rapidly mov- rather, a more complicatedpattern, perhapsinvolving ing flow within the gaps. It thus seems probable that two-tiered convection. Jeanloz and Richter [1979] have once the gap is traversed, the flow lines pass beneath considered the thermal profile of the earth; in their the Atlantic Ocean lithosphere in a pattern determined model a thermal boundary layer must be present at the by the volumetric requirements of the spreading ridges base of the lower mantle. Furthermore, unless the and by the differential pressure between the gaps and core temperature is considerably lower than expected, the various ridge segments. a second thermal boundary layer would be required. If the present Atlantic ridge is being fed by mantle This could be either near the base of the lower mantle return flow through the gaps in the Pacific rim, how or at the top of the lower mantle. Although they was the Atlantic ridge fed in its early phases, before could not choose between these two alternatives, the these gaps developed? This problem is most notice- latter possibility would suggest that the lower and able for the Jurassic opening of the Central Atlantic, upper mantle may be dynamically and chemically dis- before the North or South Atlantic Oceans or Carib- tinct systems. Richter [1979] also reachedthis conclu- bean gap existed. However, at that time the Tethvs, sion on the basisof seismicevidence from the Tonga- separating from the southern con- Ker•adec region. As the question of whole-mantle tinents, formed a wedge-shaped ocean reaching west- versus two-tiered convection remains unsolved, it is ward to Spain, where the Central Atlantic opening justifiable to consider here whether a two-tiered model began [Smithand Briden,1977]. Early openingof the can account for known plate motions. In the present Central Atlantic required a connection eastward to model, continent-bearingplates are driven by a simple Tethys [Deweyet al., 1973]. This oceanic pathway pattern of lower mantle convection cells, but oceanic through Tethys would have allowed a mantle return plates are not. flow feed to the new ridge in the Central Atlantic. 2. The model makes use of the possibilitythat deep continental roots may exist. It specifiesthat the con- THE DRIVING MECHANISM tinents are forced to move laterally by viscouscoupling between their roots and the convecting lower mantle. The correspondence between the predicted Pacific Oceanic lithosphere included in the same plate as a mantle outflow and the geological and geophysical continent moves together with the continent (for character of the Australian-Antarctic gap, the Carib- example, the western North Atlantic moves with the bean, and the easternmost Scotia areas provides sup- North Americanplate) but is not driven directlyby the port for the model of shallow return flow with deep lower mantle convection. Oceanic plates with no con- 6704 ALVAREZ:MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM

Africa

o

O/ •o r'n

// /

Fig.4. Equatorialsection through the earth,looking north, showing the mantleflow pattern and driving mechanismconsidered in thispaper. Arrows show movement direction (not rate) relative to thegrid of axesof convergenceand divergence at the top of thelower mantle; this grid should deform slowly compared to theother motionsshown. In the mantle above 670 km, the lower arrowsshow the subasthenosphereupper mantle, entrainedby movementot' the uppermost lower mantle; the upper arrows show return flow in theasthenosphere. Black:continental crust; vertical lines: lithosphere and thickcontinental tectosphere. The linksat 'y' indicatethat South Americaand Africa move with the lower mantle['low. The modelplaces stronger constraints on the lower mantleconvective pattern in the continentalhemisphere than in the Pacifichemisphere, where the patternshown is extremely speculative.

tinental crust (Pacific,Nazca, Cocos, Philippine)are viscosity increase at the base of the asthenosphereand underlain by weak asthenosphereand thus have only a another at the top of the lower mantle (e.g., modelsVi weak viscous coupling to lower mantle movements. and VII of Hager and O'Connell[1979, Figure 2]), it Viscous coupling between lower mantle and would be sufficient for the continental roots to extend continental roots could occur in at least two ways. If below the asthenosphere and be embedded in the continental roots extend to 670 kin, they may be more viscous upper mantle entrained by lower mantle anchored directly to the top of the lower mantle con- movement [cf. Chase,1979a, Figure 1]. This situation vection cells. In a more realistic model, with one is shown in Figure 4. ALVAREZ: MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM 6705

3. If lower mantle convection does move continents Figure 4 is an equatorial section through the earth, by drag on deep roots, upper mantle return flow would and Figure 5 is a sketch map showing the pattern of be a necessity. This flow would be most likely to pro- movement envisioned in this paper. Ascending lower duce recognizable effects where it squeezes through mantle limbs apparently underlie the Atlantic and the gapsin the Pacificrim, and the expectedeffects are southeast Indian oceans, with a long, transform-type exactly what one observes in the tectonics of the offset between Africa and Antarctica. The activity of Caribbean, Scotia and Australia-Antarctica regions. this limb began in the central segment of the Atlantic This subject was discussedextensively in the first part about 180 m.y.B.P. and more recently extended north- of the paper. ward (NAM-EUR: 80 m.y.B.P.) and southeastward 4. The final feature of the model is that oceanic (SAM-AFR: 120 m.y.B.P.; AFR-ANT: 120 m.y.B.P.; lithosphere, being underlain by the asthenospheric low ANT-IND: 80 m.y.B.P.; ANT-AUST: 85 or 53 m.y. viscositylayer and having no deep roots (exceptwhere B.P.) [Smith and Briden, 1977; Norton and Sclater, it is attached to seismic slabs), is affected neither by 1979; Weisseland Hayes, 1974; Cande et al., 1981]. the lower mantle convection nor by the upper mantle This behavior may well reflect the gradual growth of return flow. Because of this decoupling, the oceanic the region in the lower mantle that was organized into lithosphere would behave as in the model examined by this particular pair of convection cells. This growth Forsythand Uyeda[1975], in which the driving forces and the variability in continental 'absolute' velocities for plate motion are generated by the plate itself. In indicate time-dependent lower mantle convection, their analysis the dominant force is slab pull due to the which is to be expected[Verhoogen, 1980, Ch. 5]. negative buoyancy of the cold, dense, descending slab. The east-west Tethyan zone from Spain to India and This downward pull is balanced by a viscuous resis- beyond, with its long history of convergence, would be tance, and the resulting 'terminal velocity' is 6-9 underlain by a descending limb. This provides an cm/yr, the observed velocity of plates that are con- explanation for one of the most uncomfortable con- nected to downgoing slabs and which have little or no tradictions in current plate tectonic theory -- the pro- continental area. As discussed below, ridge push at tracted collision between India and Asia. That the two the East Pacific Rise may also help drive the oceanic continents should collide by subduction of the inter- plates of the Pacific. Slab pull and ridge push are, vening ocean is reasonable; that India should continue however, simply aspects of thermal convection in a to drive northward into Asia for some 38 m.y. after broad view [ Verhoogen,1980, Ch. 1]. Becauseof the the collision [Molnar and Tapponnier,1975] is not. effective decoupling of the oceanic lithosphere from Buoyancyconsiderations predict that shortly after such the deeper levels, oceanic plate motions do not reflect a continent-continent collision, a new subduction zone lower mantle convection. should form; in this case the logical place would be along the southwest coast of India, from Karachi to Sri GLOBAL MOVEMENT PATTERN Lanka, facing the Carlsberg Ridge. This has not occurred, and of the apparently important driving Three kinds of geological evidence can be used to mechanisms for plate tectonics considered by Forsyth infer the pattern of movements at depth. Tectonic and Uyeda[1975], slab pull clearlycannot be forcing disruptions as in the Caribbean may indicate shallow India deep into Asia, and ridge push is generally return flow passing through constrictions. In the thought to be too weak to accomplishsuch a task [For- present model, the movements of continents reflect sythand Uyeda,1975]. The problemis resolved,how- the motion of the top of the lower mantle. Finally, ever, if the two continents are being pushed together hot spot tracks should indicate the motions of the lev- by drag due to a pair of converging lower mantle con- els where their heat sources reside. vection cells. Molnar and Tapponnier[1975] have Ideally one would determine lower mantle motions made a strong case that the Asian continental crust from continental movements and then use hot spot north of the Indus Suture has responded to the indent- information to test whether the pattern was valid. ing action of India by deforming along a set of major Unfortunately, this is not possible in view of the transcurrent faults. Most of these faults trend east- uncertaintiesin understandinghot spot phenomena,as ward from the suture zone, indicating easiest relief in discussed below. the direction of the nearest oceanic region. Both the continuing collision and the way in which it is being Lower Mantle MovementJkom Continental Motions accommodated by deformation of Asia are well explained by the present model. The motions of continentsprovide a hazy picture of the pattern of lower mantle convection. Forsythand Information From Hot Spots Uyeda[1975] showedthat most plates that contain continents move at average rates of 1-2 cm/yr with Hot spots and their trails would appear to offer an respect to the approximatelyrigid hot spot framework; ideal way to test the model developed here, and early a major exception is the Indian Plate, carrying India drafts of this paper contained extensive discussionsof and Australia, with an average 'absolute' velocity of hot spot data. Although the depth of hot spot heat 6.1 cm/yr. Gordonet al. [1979] have shown that rms sources is not known, a position below the astheno- 'absolute' velocities of continents have been at least 5 sphere is reasonable, and if heat sources reside any- cm/yr over periodsof about 30 m.y. in the past. where from the lower half of the upper mantle down 6706 ALVAREZ:MANTLE RETURNFLOW AND PLATE-DRIVINGMECHANISM

o.

Fig. 5. Convective flow pattern at the top of the lower mantle based on the model developed in this paper. Open trends are axes of upwelling and divergence in the uppermost lower mantle; solid trends are axes of con- vergence and sinking. The Atlantic-Indian and Tethyan trends are determined by continental motions; the Pacific trends are very tentatively suggested, on the basis of much weaker evidence, as explained in the text. Arrows show inferred flow lines of the uppermostlower mantle relative to the more slowly deforming grid of conver- gence and divergence axes.

to roughly the middle of the lower mantle, their tracks lower mantle convergence zone pass west of the would provide valuable information on lower mantle Galapagosand between Yellowstoneand the rest of movements(see Galapagos,in Figure 4). The results , perhapstrending along the East Pacific of the hot spot tests were mixed but, on the balance, Rise and passing just east of the North American more unfavorable than favorable to the present model. (Figures4 and 5). The presenceof a sur- The Chagos-Laccadive and Ninetyeast ridges, attri- face divergence above a lower mantle convergenceis buted to the Reunion and Kerguelen hot spots, are to be expected in a model with chemically distinct particularly unfavorable for the present model, in lower and upper mantles which do not mix, for which India rides northward on a lower mantle cell; subasthenosphereupper mantle will be entrained by hot spotssouth of India should be fixed with respectto the converging lower mantle and forced to escape the continent and ocean of the Indian plate. upward and outward. This raises the possibilityof a In a number of cases the hot spot data were neither fundamental difference between spreading ridges of clearly favorable nor clearly unfavorable, but an the Mid-Atlantic and East Pacific type, with the former attempt to incorporate them in the model led to a being pulled apart above a lower mantle divergence more and more speculative picture, without adequate and the latter pushed apart above a lower mantle con- support. For example, the Pacific hot spots seem to vergence. Ridges of the former type would cease to show little or no present motion with respect to Africa function if subducted; the latter type would continue and [Minster and Jordan, 1978; Duncan,1981 ]. to spread after subduction. Many authors have sug- If a lower mantle upwelling and divergence causesthe gested that subduction of the East Pacific Rise led to 2 cm/yr separation rate of the Americas from Africa extension in the Basin and Range, and the present and Europe, the Pacific hot spots would imply lower scenario would agree with this idea. One may picture mantle convergenceand _descentbetween the Pacific the lower mantle convergence presently lying under and the Americas. If Yellowstone is a valid hot spot, Colorado, with entrained upper mantle rising, uplifting it appearsto be approximatelyfixed with respectto the the High Plains and Rocky Mountains [Suppeeta!., major Pacific hot spots, whereas the Galapagoshot 1975], heatingand softeningthe continentalroots, and spot may be approximatelyfixed with respectto South escaping westward beneath continent already softened America (see below). This would require that the by passage over the lower mantle convergence. This ALVAREZ: MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM 6707 pattern is shown in Figures 4 and 5. It should be dence for monotonic age increasealong the Hawaii- emphasized that this concept is even more speculative Emperor chain is probably the best available. At the than the rest of the paper, but it suggests that atten- other extreme, the Line Islands, thought to be concen- tion should be given to the linkage between flow pat- tric and coeval with the Emperor chain [Morgan, terns in lower and upper mantle. 1972], are evidentlytoo old to fit this interpretation As testing of the model on the basis of hot spot [Haggert?/eta!., 1981]. Duncan[1981, Fig. 1] showsa information proceeded, it became evident that the control point for the Prince Edward hot spot track on quality of information on hot spots was highly varied, WaltersShoal at 55 m.y.B.P. This is basedentirely on ranging from excellent, as in the case of the very weak evidence from DSDP site 2a6, which Hawaiian-Emperorchain [Dalrympleeta!., 1977], to recovered about 16 m of lower Eocene sediments, extremely weak in the cases of many short, poorly including a 4.5 m interval in which a volcanic ash com- known, or dubious volcanic alignments. Eventually, ponent was present, and 0.25 m of volcanic breccia the conclusion was inescapable that hot spot [Simpson,Schlich et al., 1974]. Migration of the phenomena and the pattern of hot spot movements are eastern Australian Cenozoic hot spot is based on not yet well enough understood to be useful as a test diachronoustermination of volcanicactivity which had of the present model. In order to justify this disap- been nearly constant throughout the province for the pointing conclusion, the following points are noted as previous35-70 m.y. [Wellmanand McDougall,1974; weaknesses in the present understanding of hot spots: Pilger,1982]. No systematicreview of the qualityof 1. It is commonly concluded that hot spot heat the evidencesupporting the variousproposed hot spots sources are embedded in a fixed, undeforming, abso- seems to have been undertaken. lute frame of reference, that is, one with 'motion of an 6. The recent hot spot analysesof Morgan [1982] order of magnitude less than the relative motion and Duncan[1981] beginby specifyingthe motionof between plate pairs. In most cases it is concluded that Africa over the hot spotframe, with the Walvis Ridge, inter-hotspot movement cannot be discerned for the interpretedas the track of the Tristan hot spot, provid- period 100 m.y. to Present...' [Duncan, 1981]. This ing key evidence. Their published rotations can be behavior is unlikely in an earth where thermal con- used to predict the path of a plume which might be siderations apparently require convective overturn at located beneath Mount Etna, in eastern Sicily -- an all depths from the top of the inner core to the base of excellent hot spot candidatewhich has apparently not the lithosphere[Verhoogen, 1980]. yet been evaluated as a hot spot volcano. The parame- 2. It is not known at what depth the heat sources ters in either model predictan Etna path startingabout reside [Anderson,1981], or even whether they are all 2700-2900 km north of its present position 100-125 at the same depth -- critical questions in testing a m.y. ago and moving generally southward through the complex model. Cretaceous and Tertiary. However, volcanic rocks 3. Reconstructionsof plate positions tens of millions known from outcrop and drill records show that basal- of years ago are sensitive to the unresolved question tic volcanism, much of it of alkaline character, has of whether Antarctica should be treated as one or occurred repeatedly in southeastern Sicily, within 125 more than one plate [Jurdy, 1978; Dalziel and Elliot, km of Mt. Etna, over the last 220 m.y. -- in the Mid- 1982]. This must also be consideredin investigating dle and Late Triassic, the Early and Middle Jurassic, whether the hot spot framework has deformed over the Late Cretaceous, the Miocene, Pliocene, Pleisto- long periods of time. cene and Recent [ Cristofolini,1966; Pichler, 1970; 4. Uncertainties in relative plate motions [Minster Romanoand Villari,1973]. Yet this part of Sicily,the and Jordan, 1978] produce uncertaintiesin the calcu- Hyblean Plateau, has clearly been an extension of lated motions of hot spots. These uncertainties are African continental crust during this entire time not often reported but may be substantially greater [Charmelland Horvath,1976; Charmellet al., 1979]. than the difference between two alternative Thus, there is at least one probablehot spot source hypotheses. For example, when the motion of the which has long remained roughly fixed with respectto Galapagos hot spot is evaluated relative to various Africa and which, therefore,contradicts the hypothesis reference frames, to test whether it is fixed with of a rigid hot spot framework. respectto South America, as predictedby Figure 4, or 7. As noted above, the Reunionand Kerguelenhot with respect to some 'absolute' frame, the 95% spots are usually considered to have produced the confidence limit on Galapagos movement includes the Chagos-Laccadiveand Ninetyeastridges, which do not best value for South America as well as two proposed fit well in the presentmodel. However,there is some absolute reference frames -- the 'mantle plate' of Hey question whether these features are typical hot spot et al. [1977] and 'AM1-2', the model favored by Min- tracks, since they flank the major transform faults that sterand Jordan[1978]. Figure 7 of Hey [1977] shows bounded the plate that moved rapidly northward with the Galapagos hot spot slowly approachingSouth India during the early Tertiary [Norton and Sclater, America, but in view of the uncertainties, the hot spot 19791. may just as well have remainedfixed with respectto These problems are raised not as a criticism of exist- South America. (A full analysis of this question is ing work on hot spots, or necessarilyas objectionsto available from the author.) the conclusion that there is an absolute reference 5. The evidence for the existence and interpreted frame marked by the hot spots. One cannot fail to be motion of hot spots varies widely in quality. The evi- impressed by the number of observations that fit 6708 ALVAREZ:MANTLE RETURN, FLOW AND PLATE-DRIVING MECHANISM

AcknowledgmentsThe startingpoint of this studywas a dis- togetherin thesynthesis byMorgan [1982], for exam- cussion of Asian tectonics with David W. Simpson in Tadjikis- ple. Theyare raised as an indication of theuncertain- tan in 1977. Since then many colleagues have contributed to tiesstill remainingin hot spottheory. After an inten- the developmentof these ideas,often by pointingout flaws, siveattempt to testthe presentmodel on thebasis of and I thank them all for their help: Subir K. Banerjee, Bruce hot spotinformation, I can only conclude that rejec- A. Bolt, Mark Bukowinski,Lung S. Chan, Clement G. Chase, tion of the model on this basiswould be premature. Gilles M. Corcos, S. Thomas Crough, Garniss Curtis, lan W.D. Dalziel, David Epp, W. Gary Ernst, Raymond Jeanloz, H. Jay Melosh, J. Bernard Minster, Stephen Morris, David Lower Mantle Cells Simpson, and John Verhoogen. I warmly thank three anonymousreviewers whose detailed comments led to major Figure5 showsthe approximate,pattern of move- improvements. mentat the topof thelower mantle, inferred from the REFERENCES foregoingconsiderations. The divergencesin the Atlantic and south of Australia and the convergence Alvafez, W., FragmentedAndean belt of northern Colombia, Geol. Sec. Am. Bull., 130, 77-96, 1971. alongthe Tethyanbelt are stronglyimplied by con- Alvafez, W., The geographicalpattern of mantle return flow tinentalmotions. The EastPacific-Great Basin conver- and the drivingmechanism of platetectonics (abstract), Ees genceis verytenuously suggested,by tectonic and hot Trans. AGU, 59, 1202, 1978. spotconsiderations. Finally, in orderto closethe Anderson, D. L., The deep structure of continents, J. Gee- lower mantleconvection system, an additionaldiver- phys.Res., 84, 7555-7560,1979. Anderson, D. L., Hotspots,basalts, and the evolution of the gencemay be postulated to lie in thecentral Pacific, mantle, Science,213, 82-89, 1981. betweenthe two convergences;by analogywith the Baldwin,B., P. J. Coney, and W. R. Dickinson,Dilemma of a concentrationof hot spotsalong the Atlantic diver- Cretaceous time scale and rates of sea-floor spreading, Geol- gence,the Pacific divergence could perhaps be respon- ogy, 2, 267-270, 1974. sible for the abundant volcanismof the western Barazangi, M., and J. Dorman, World seismicitymaps com- piled from ESSA, Coast and Geodetic Survey, epicenter Pacific.Thus a fairly simplepattern of lowermantle data, 1961-1967,Bull. Seismel. Sec. Am., ,59,369-380, 1969. convectioncells may explainmuch of the tectonic Barker, P. F., A spreadingcentre in the east ScotiaSea, Earth complexityof the earth. Planet. Sci. Lett., 15, 123-132, 1972. Convergentand divergentaxes at the top of the Barker, P. F., and J. Burrell, The openingof Drake Passage, Mar. Geol., 25, 15-34, 1977. lowermantle seem to dieout poleward and to beoffset Barker, P. F., and D. H. Griffiths, The evolution of the Scotia or terminatedin someplaces. The prominent19,000- Ridgeand ScotiaSea, Philos.Trans. R. Sec. London,Ser. A, km lineamentformed by the EltaninFracture Zone 271, 151-183, 1972. and the Africa-Antarctica transform-ridge system Barker, P. F., and D. H. Griffiths, Towards a more certain accommodatesthe offset of the main divergent zone reconstruction of Gondwanaland, Philos. Trans. R. Sec. London,Ser. B, 279, 143-159, 1977. from the Atlantic to the southeastIndian Ocean,and Bell, J. S., Geotectonicevolution of the southernCaribbean mayroughly mark the southwardtermination of the area, Mere. Geol.Sec. Am., 132,369-386, 1972. Pacificconvergence and divergence. Terminations and Bowin, C., Caribbeangravity field and plate tectonics,Spec. offsetswould occur over broad belts in the ductile Pap. Geol.Sec. Am., 169,79 pp., 1976. lower mantle but would be accommodatedby sharp Brooks,C., D. E. James,and S. R. Hart, Ancient lithosphere: Its role in youngcontinental volcanism, Science, 193, 1086- breaksin the rigid lithosphere,and becauseof the 1094, 1976. indirectcoupling of oceaniclithosphere to lowerman- Bullard,E. C., J. E. Everett,and A. G. Smith, The fit of the tle, breaksin the lithospherewill onlyroughly mark continents around the Atlantic, Philos. Trans. R. Sec. Lon- lower mantle features. don, Ser. A, 258, 41-75, 1965. Burchfiel, B.C. and G. A. Davis, Structuralframework and evolution of the southern part of the Cordilleran orogen, CONCLUSION Am. J. Sci., 272, 97-118, 1972. Burchfiel,B.C. and G.A. Davis, Nature and controlsof Cor- In closing,it is worthstressing both the speculative dilleranorogenesis, western United States: Extensions of an earliersynthesis, Am. J. Sci.,275A, 363-396, 1975. natureof the modeldeveloped here and the rationale Busse, F. M., On the aspectratios of two-layer ma.n.t!econ- on which it is based. Geophysicalmodelers are vection,Phys. Earth Planet. Inter., 24, 320-324,1981. developingmethods for calculatingthree-dimensional Cande, S.C., J. Mutter, and J. K. Weissel, A revisedmodel flowpatterns in the mantle[Harper, 1978; Chase, for the break-upof Australiaand Antarctica(abstract), Ees Trans. AGU, 62, 384, 1981. 1979a;Hager and O'Connell, 1979; Parmentier and Cardwell, R. K., and B. L. Isacks,Geometry of subducted Oliver,1979] but are hampered by thelack of firm lithospherebeneath the in eastern Indonesia constraintson many of the physicalparameters from seismicityand fault planesolutions, J. Geephys.Res., involved.Tectonic geologists can contributeto this 83, 2825-2838, 1978. endeavor,both by specifyinggeologically reasonable Channell,J. E. T., and F. Horvath, The African/Adriaticpro- montory as a palaeogeographicalpremise for Alpine oro- boundaryconditions and by lookingfor thetectonic genyand platemovements in the Carpatho-Balkanregion, effectspredicted either explicitly or implicitlyby calcu- Tectenephysics,35, 71-101, 1976. latedflow patterns. The modeldeveloped here is an Channell,J. E. T., B. D'Argenio, and F. Horvath, Adria, the effortin thisdirection. It will certainlyrequire strong African promontory,in Mesozoic Mediterraneanpalaeo- geography,Earth Sci. Rev., 15, 213-292, 1979. modificationand perhapscomplete rejection, but it Chapman,D. S., and H. N. Pollack,Regional geotherms and seemsto accountfor a ratherwide range of geological lithosphericthickness, Geology, 5, 265-268,1977. andgeophysical observations, and it thusdeserves to Chase, C. G., Asthenosphericcounterflow: A kinematic be subjectedto furthertesting. model (abstract),Ees Trans.AGU, ,57,1002, 1976. ALVAREZ:MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM 6709

Chase, C. G., Asthenospheric counterflow: A kinematic Harper, J. F., Asthenosphere flow and plate motions, Geo- model, Geoph•;s.J. R. Astron.Soc., 56, 1-18, 1979a. phys. J. R. Astro,. Soc., 55, 87-110, 1978. Chase, C. G., Subduction, the geoid, and lower mantle con- Hayes, D. E., Nature and implications of asymmetric sea-floor vection, Nature, 282, 464-468, 1979b. spreading-- 'Different rates for different plates', Geol. Soc. Cochran, J. R. and M. Taiwani, Free-air gravity anomaliesin Am. Bull., 87, 994-1002, 1976. the world's oceansand their relationshipto residualeleva- Hey, R., Tectonic evolution of the Cocos-Nazca spreading tion, Geoph•;s.J. R. Astro,. Soc., 50, 495-552, 1977. center, Geol. Soc. Am. Bull., 88, 1404-1420, 1977. Coney, P. J., and S. J. Reynolds,Cordiileran Benloft zones, Hey, R., G. L. Johnson, and A. Lowtie, Recent plate motions Nature, 2 70, 403-406, 1977. in the Galapagos area, Geol. Soc. Am. Bull., 88, 1385-1403, Cristofolini, R., Le manifestazioni eruttive basiche del trias 1977. superlore nei sottosuolodi Ragusa (Sicilia sud-orientale), Houtz, R. E., and W. J. Ludwig, Structure of Colombia Basin, Period Mineral., 35, 1-28, 1966. Caribbean Sea, from profiler-sonobuoy measurements, J. Dalrymple, G. B., D. A. Clague, and M. A. Lanphere, Geoph•;s.Res., 82, 4861-4867, 1977. Revisedage for Midway Volcano, Hawaiianvolcanic chain, Jeanioz, R. and F. M. Richter, Convection, composition and Earth Planet.Sci. Lett., 3 7, 107-116, 1977. thermal state of the lower mantle, J. Geoph•;s.Res., 84, Daiziel, I. W. D., and D. H. Elliot, West Antarctica: Problem 5497-5504, 1979. child of Gondwanaland, Tectonics,1, 3-19, 1982. Jordan, T. H., Lateral heterogeneityand mantle dynamics, Daiziel, I. W. D., R. H. Dott, Jr., R. D. Winn, Jr., and R. L. Nature, 257, 745-750, 1975a Bruhn, Tectonic relationsof South Georgia Island to the Jordan, T. H., The continental tectosphere, Rev. Geoph•;s. southernmostAndes, Geol. Soc. Am. Bull., 86, 1034-1040, Space Phys., 13, 1-12, 1975b. 1975. Jordan, T. H., The present-day motions of the Caribbean Davies, G.F., Whole mantle convectionand plate tectonics, plate, J. Geophys.Res., 80, 4433-4439, 1975c Geophvs.J. R. Astro,. Soc., 49, 459-486, 1977. Jurdy, D. M., An alternative model for early Tertiary absolute Dengo, G., Problems of tectonic relations between Central plate motions, Geology;,6, 469-472, 1978. America and the Caribbean, Trans. Gull' CoastAssoc. Geol. Kanasewich,E. R., Plate tectonicsand planetaryconvection, Socs., 19, 311-320, 1969. Can. d. Earth Sci., 13, 331-340, 1976. Dewey, J. F., W. C. Pitman Ill, W. B. F. Ryan, and J. Bon- Karig, D.E., Evolution of arc systemsin the western Pacific, nin, Platetectonics and the evolutionof the Alpinesystem, An,u. Rev. Earth Pla,et. Sci., 2, 51-75, 1974. Geol. Soc. Am. Bull., 84, 3137-3180, 1973. Ladd, J. W., Relative motion of South America with respect DeWit, M. J., The evolutionof the ScotiaArc as a key to the to North America and Caribbean tectonics, Geol. Soc. Am. reconstructionof southwesternGondwanaland, Tecto,ophy- Bull., 87, 969-976, 1976. sics, 37, 53-81, 1977. Larson, R. L., and W. C. Pitman, Ill, World-wide correlation Duncan, R.A., Hotspots in the southern oceans -- An abso- of Mesozoic magneticanomalies and its implications, Geol. lute referencefor motion of the Gondwanacontinents, Tec- Soc. Am. Bull., 83, 3645-3662, 1972. tonoph•;sics,74, 29-42, 1981. Liu, L. G., On the 650-km seismicdiscontinuity, Earth Planet. Dziewonski, A.M., A. L. Hales, and E. R. Lapwood, Sci. Lett., 42, 202-208, 1979. Parametricallysimple earth modelsconsistent with geophy- Ludwig, W. J., R. E. Houtz, and J. I. Ewing, Profiler- sical data, Ph•;s.Earth Planet. Inter., lO, 12-48, 1975. sonobuoy measurements in Colombia and Venezuela Elsasset,W. M., Seafloorspreading as thermalconvection, J. basins, Caribbean Sea, Am. Assoc.Petrol. Geol. Bull., 59, Geophys.Res., 76, 1101-1112, 1971. 115-123, 1975. Elsasset,W. M., P. Olson, and B. D. Marsh, The depth of Maifait, B. T., and M. G. Dinkeiman, Circum-Caribbean tec- mantle convection,J. Geophys.Res., 84, 147-155, 1979. tonic and igneous activity and the evolution of the Carib- Fitch, T. J., Earthquake mechanismsand tectonics bean plate, Geol. Soc. Am. Bull., 83, 251-272, 1972. in the Indonesian-Philippineregion, Bull. Seismol.Soc. Am., Maresch, W.V., Plate tectonicsorigin of the Caribbean Moun- 60, 565-591, 1970. tain System of northern South America: Discussion and Forsyth, D. W., Fault plane solutions and tectonics of the proposal, Geol. Soc. Am. Bull, 85, 669-682, 1974. South Atlanticand ScotiaSea, J. Geoph•;s.Res., 80, 1429- Marsh, B. D., and J. G. Marsh, On global gravity anomalies 1443, 1975. and two-scale mantle convection, J. Geophys.Res., 81, Forsyth, D., and S. Uyeda, On the relativeimportance of the 5267-5280, 1976. driving forces of plate motion, Geoph•;s.J. R. Astro,. Soc., Marsh, B. D., and J. G. Marsh, Reply, J. Geophvs.Res., 8.3, 4.3, 163-200, 1975. 3555-3558, 1978. Gaposchkin,E. M., and K. Lambeck, Earth's gravity field to Mattson, P. H., Middle Cretaceousnappe structures in Puerto the sixteenth degree and station coordinates from satellite Rican ophiolitesand their relation to the tectonichistory of and terrestrial data, J. Geoph•;s.Res., 76, 4855-4883, 1971. the Greater Antilles, Geol. Soc. Am. Bull., 84, 21-28, 1973. Gaffunkel, Z., Growth, shrinking,and long-termevolution of Mattson, P. H., Subduction,buoyant braking, flipping, and plates and their implicationsfor the flow pattern in the strike-slip faulting in the northern Caribbean, J. Geol., 87, mantle, J. Geoph•;s.Res., 80, 4425-4432, 1975. 293-304, 1979. Gordon, R. G., M. O. McWiiliams,and A. Cox, Pre-Tertiary Mattson, P. H., and E. A. PessagnoJr., Jurassicand Early velocitiesof the continents:A lower boundfrom paleomag- Cretaceous radiolariansin Puerto Rican ophiolite -- Tec- netic data, J. Geoph•;s.Res., 84, 5480-5486, 1979. tonic implications, Geology, 7, 440-444, 1979. Gough, D. I., The geoid and single-cellmantle convection, Minster, J. B., and T. H. Jordan, Present-dayplate motions, Earth Pla,et. Sci. Lett., .34,360-364, 1977. J. Geophvs.Res., 83, 5331-5354, 1978. Gutenberg, B., and C. F. Richter, Seismici.•ol' the Earth Molnar, P., and P. Tapponier, Cenozoic tectonics of Asia: AssociatedPhe,ome•a, 273 pp., PrincetonUniversity Press: Effects of a continentalcollision, Science,189, 419-426, Princeton, N.J., 1949. 1975. Hager, B. H., and R. J. O'Conneli, Viscous flow in the mantle Morgan, W. J., Convection plumes in the lower mantle, driven by moving plates (abstracts),Eos Tra,s. AGU, 57, Nature, 230, 42-43, 1971. 1002-1003, 1976. Morgan, W. J., Plate motions and deep convection, Mere. Hager, B. H., and R. J. O'Conneli, Kinematic models of Geol. Soc. A•n., 132, 7-22, 1972. large-scaleflow in the earth'smantle, J. Geoph•;s.Res., 84, Morgan, W. J., Hot spot tracksand the openingof the Atlan- 1031-1048, 1979. tic and Indian Oceans,in The Sea, vol. 7, edited by C. Emi- Haggetty, J. A., S. O. Schlanger,and I. Premoli-Siiva,Geo- iiani, pp. 443-487, John Wiley, New York, 1982. logical history of the southern Line Islands, Geol. Soc. Am. Norton, I. O., and J. G. Sclater, A model for the evolution of Abstr. Programs,1.3, 466, 1981. the Indian Ocean and the breakup of Gondwanaland, J. Hamiitoh, W., Tectonics of Antarctica, Mere. Am. Assoc. Geoph•;s.Res., 84, 6803-6830, 1979. Petrol. Geol., 2, 4-15, 1963. Parmentier,E. M., and J. E. Oliver, A studyof shallowglobal 6710 ALVAREZ: MANTLE RETURN FLOW AND PLATE-DRIVING MECHANISM

mantle flow due to the accretion and subduction of continentalmaps, 63 pp., Cambridge University Press, New lithospheric plates, Geophys.J. R. Astron. Soc., 57, 1-21, York, 1977. 1979. Smith, P. J., The return of whole-mantle convection, Nature, Perfit, M. R., and B.C. Heezen, The geologyand evolution 268, 687-688, 1977. of the Cayman Trench, Geol. Soc. Am. Bull,, 89, 1155-1174, Stauder, W., Subductionof the Nazca plate under Peru as 1978. evidenced by focal mechanisms and by seismicity, J. Geo- Pichler, H., Volcanism in eastern Sicily and the Aeolian phys. Res., 8(I, 1053-1064, 1975. Islands, in, Geology and History o.f Sicily, edited by W. Suppe, J., C. Powell, and R. Berry, Regional topography, Aivarez and K. H. A. Gohrbandt, pp. 261-283, Petroleum seismicity, Quaternary volcanism, and the present-day tec- Exploration Society of Libya, Tripoli, 1970. tonics of the western United states, Am. J. Sci., 275A, 397- Pilger, R. H., Jr., The origin of hotspot tracks: Evidence from 436, 1975. eastern Australia, J. Geoph.vs. Res., 87, 1825-1834, 1982. Tomblin, J. F., Caribbean seismicity, in Report of the IDOE Pollack, H. N., and D. S. Chapman, On the regional variation Workshop on Geology and Marine Geophysics of the of heat flow, geotherms, and lithospheric thickness, Tecto- Caribbean Region and its Resources, Kingston, Jamaica, nophysics,38, 279-296, 1977. 1975, Geology, geophysics,and resourceso./' the Caribbean, Richter, F. M., Mantle convection models, Annu. Rev. Earth 63-77, 1975. Planet. Sci., 6, 9-19, 1978. Verhoogen, J., Energetics o./' the Earth, 139 pp., National Richter, F. M., Focal mechanisms and seismic energy release Academy of Sciences,Washington, D.C., 1980. of deep and intermediate earthquakes in the Tonga- Watts, A.B., Comment on 'On global gravity anomalies and Kermadec region and their bearing on the depth extent of two-scale mantle convection' by B. D. Marsh and J. G. mantle tlow, J. Geoph.vs. Res., 84, 6783-6795,1979. Marsh, J. Geoph.vs. Res., at& 3551-3554, 1978. Richter, F. M., and B. Parsons, On the interaction of two Weissel, J. K., and D. E. Hayes, Asymmetric spreadingsouth scales of convection in the mantle, J. Geoph.vs. Res., 80, of Australia, Nature, 21& 518-521, 1971. 2529-2541, 1975. Weissel, J. K., and D. E. Hayes, The Australian-Antarctic Romano, R., and L. Villari, Caratteri petrologici e magmato- discordance:New results and implications, J. Geophvs,Res., iogici del vuicanismo ibleo, Rend. Soc. !tal. Mineral. Petrol., 79, 2579-2587, 1974. 29, 453-484, 1973. Wellman, P., and I. McDougall, Cainozoic igneous activity in Roth{5, J.P., The seismicityof the earth, 1953-1965, UNESCO eastern Australia, Tectonoph.vsics, 2,t, 49, 1974. Earth Sci., 1, 336 pp., 1969. Winn, R. D., Jr., Upper Mesozoicflysch of Tierra del Fuego Saunders, J. B., N. T. Edgar, T. W. Donnelly, and W. W. and South Georgia Island: A sedimentologicalapproach to Hay, Cruise synthesis, Initial Reps. Deep Sea Drill. Pro.j., 15, lithosphere plate restoration, Geol. Soc. Am. Bull, 89, 533- 1077-1111, 1973. 547, 1978. Simpson, E. S. W., R. Schlich et al., Initial ReportsDeep Sea Dr#ting Project,25, 1974. Sipkin, S. A., and T. H. Jordan, Lateral heterogeneity of the mantle determined from the travel times of ScS,J. Geophvs. (Received March 17, 1980: Res., 80, 1474-1484, 1975. revised March 4, 1982; Smith, A. G., and J. C. Briden, Mesozoicand Cenozoicpaleo- accepted March 12, 1982.)