Distribution of Stresses in the Descending Lithosphere from a Global Survey of Focal&Hyphen;Mechanism Solutions of Mantle Ea

REvli•ws OF GEOPHYSICSAND SPACE PHYSICS, VOL. 9, NO. 1, FESltUAIt¾1971

Distribution of Stressesin the Descending Lithosphere from a Global Survey of Focal-MechanismSolutions of Mantle Earthquakes

BRYAN ISACKS

Earth Sciences Laboratories National Oceanographicand AtmosphericAdministration Lamont-DohertyGeological Observatory o[ ColumbiaUniversity Palisa'des,New York 10964

P•,T•,I• MOLNAI• La•nont-DohertyGeo•og•ca• Observatory oJ Columbia University Palisades,New York 1096•

A region-by-region analysis of 204 reliable focal-mechanismsolutions for deep and intermediate-depth earthquakes strongly supports the idea that portions of the lithospherethat descendinto the mantle are slablike stressguides that align the earthquake- generating stressesparallel to the inclined seismic zones. At intermediate depths extensional stressesparallel to the dip of the zone are predominant in zones characterized either by gaps in the seismicityas a function of depth or by an absenceof deep earthquakes.Compressional stresses parallel to the dip of the zone are prevalent everywhere the zone exists below about 300 kin. These results indicate that the lithosphere sinks into the asthenosphereunder its own weight but encountersresistance to its downward motionbelow about 300 km. Additionalresults 'indicate contortions and disruptions of the descendingslabs; however,stresses attributable to simple bendingof the plates do not seem to be important in the generationof subcrustalearthquakes. This summary, intended to be comprehensive,includes nearly all solutionsobtainable from the World-Wide StandardizedSeismograph Network (WWSSN) for the period 1962through part of 1968plus a selectionof reliablesolutions of pre-1962events, and it includesdata from nearly every region in the world where earthquakesoccur in the mantle. The double-coupleor sheardislocation model of the sourcemechanism is adequatefor all the data.

INTRODUCTION The idea that deep and intermediate-depthearthquakes mark the location of a descendingtongue of lithosphereprovides a simple explanationfor the peculiarorientation of subcrustalearthquake mechanisms; the eventsare not part of a deepmegathrust fault but insteadoccur inside the descendinglitho- spherein responseto compressionalor extensionalstresses aligned parallel to the slablike geometryof the lithosphere.This explanation[Oliver a•d Isacks, 1967; Elsasset,1967; Isacks et al., 1968, 1969] suggeststhat focal-mechanism data may thereforeyield informationabout the forcesapplied to the lithosphere 103 104 ISACKS AND MOLNAR and about the contortions and disruptions of its descendingparts. In our pre- vious paper [Isacks and Molnar, 1969] we used focal-mechanismdata to support the ideas that at intermediate depthsthe lithospheresinks into the asthenosphere underits (•wnweight and at gi:eatdepths encounters stronger or densermaterial. Our evidence came from regionswhere the planar two-dimensional structure of the arc is clear and well defined. In this paper we compile and summarize reliable focal-mechanismsolutions for deep and intermediate-depth earthquakes (Figure 1) in nearly every region in the world where subcrustalearthquakes are known to occur. We present further evidence for the gravitational sinking of the lithosphereand, in addition, showthat even in regionswhere the distribution of hypocentersand inferred stressesis complex,the idea of a stressedslab of lithosphere still providesthe simplestinterpretation of the results. The standard classificationsof shallow, intermediate, and deep loci [•uten- berg a•d Richter, 1954] also correspondto the main types of focal-mechanism solutions in island arcs and thus appear lo be of major tectonic significance. The boundary of 60-70 km between shallow and intermediate depths approximately dividesthe earthquakesthat occur in the shallowzone of thrust faulting between the convergingplates of lithosphere from the earthquakes that occur inside the underthrust and descendingplates. An overlapping of both types, although limited to few events, shows that the boundary is somewhat diffuse and perhaps slightly variable in depth from region to region. The boundary of approximately 300 km between intermediate and deep loci marks the most outstanding feature of the distribution of the stress orientations inside the descendinglithosphere; the stress parallel •o the dip of the zone tends to be compressionalbelow 300 km and extensional or variable above 300 km. Our purposeis to place emphasison individual solutionsthat are reliable and well determined rather than to extract average features from a larger number of poorly determinedsolutions. We thereforeinclude only solutionswhose quality we are able to determine clearly (Figure 1, Table 1, and appendix). One hundred seventy-nineare basedon data from the World-Wide Standardized SeismographNetwork (WWSSN) and includenearly all the deepor intermediate- depth earthquakesthat occurredfrom 1962 throughpart of 19•8 and that were large enoughthat a reliable solutioncould be obtained.The additional solutions t•orevents that occurredprior to 1962 are relatively few in number (25) because of the scarcity of reliable data for that period [Ritsema, 1964; Stevens and Itodgson, 1968] and becausein certain casesthe data from which an assessment of quality couldbe madewas not availableto us. Ninety-five of the solutionsare new and have not beenpublished previously. The solutionsfor the New Guinea-New Britain-Solomons,New Hebrides, and South Sandwichregions are the first to be reportedthat rely upon WWSS•N data. We also report numerousnew data for the Peru-Chile and the Kurile-Japan regions. We summarizeand' discuss the main resultso• this study on • global basis in the first two sections,and in'..'the final sectionpresent the data and its analysis in detail for each region.The region-by-regionanalyses, besides providing the basis for our generalizationsand inferences,yield information about the extent o •

¸ r,.. •- r,.- oo o •', •', oo

OO 0 ½:D "• ½:D OO ½:D "'*

O0 O0 0 0 • 0 0 0 N 0 0 ,"• 0 0

z <

o z o

o • N

•: • • o 6 6 6 6 6 6 •: • • , :• o' 6 6 6 •: 6 •: 6 6 •: 6 o'

!•- •1• O •I' •0 !•- O e0 •1' •0 •'• •I' O •'• ,,0 !•- N e• t'- e• O N "'• N U'• !•. O •½ N "• ß-• • u• • • •I' •1' m0 !•- N U• N •'• • N O0 U• •' O e• ½• t'- N O e0 ,•½ N N "-• e•

O •0 ,,0 t'- O' N N N • O N 00 •I' t'- • •1' N e• O O t'- • •0 ½• •'• !•' •0 !•. •0

O o0 o0 ,,,O •'• O e• •I' O N • N •1• 00 N ,.D O O • I"- ,,,O O O", O mJq 00 O ,•O O 00

ß-• O O N

O !"- O o

O O O ff") ,,,-m O ,,-'• N ,,,-m O • ,,• ,,-'• • O O N N O • O • O "'• "-'• O

ff") 00 mJq N 00 •1• ,,,O • •'• !'"- O',, • ,,,-m • e,q e,q • O !'-. 00 ,•O • 00 ,,,O O',, N ,•O • 0 •

O "• m4q ,,• O N '•' tn • O • O O o0 .q• • ,4:) O',, O- o0 t•. N m4q O- • ,,• O',, • O O ß,,O • o0 • ,,,O O • N if') ,,,O O • m4q • • • O m4q e,q I"- • • u") • O • • m4q N I"- e• O O'•

,4D t• o • o t• ,-• oo •.. ,,o • ,,o N •'• •0 ff'• N ,,0 o o • o,. tw,,) o,• N oo ,,,-m m4q O • o0 0 • •1• c•0 •n 0 0 ,.0 • 0 0'• •.• • O• 0 I•- •'• 0 0 0 ,• 0 .• • .• c• o0 o0 .• ,-• • •.• ,-• c• • •t• o • • o0 o0 c• • o o o o o o o o o o •-• o •i •i ,-• • •i •i o

• o o •t • o o0 • o0 o o c•o o o0 u• • o u• o0 o .• .• o o •t• • t• u• .• • • .• • .• o0 o • o • • •t • • • • ,-• ,-• • ,-• o

o0 • o0 •c• • o0 t• t• •t• t• • o o o o o • o o o o •i

• o • • • o o0 •i o o • o0 c• c• • u• o o0 o •i •t • • o c• o o o • • o o ,-• • ,-• • •t • • • • •t • o o • o o o o

• • •o o •,- • o o • o 0 • o uq ,• • • 00 •- • c• •o • •o •-• •o •o • • 0 • •,. • o o •- o •o • • • •o • •-• o •0 m • • • • 0

N o0 • C• • •O 00 • CO •,- w• w• • •,- • o 00 • •o • 0 • • u• • N • N N • • •'* N N N

o o o o • o o •o o w• o •- o N o • u• •, •- • ,• • • • N ,•D •-- o • • o-, •- • •- • N ,•D •z• N • •'• ,.0 N •0 • o 0q •

N ,•D 00 o0 N • • •-, o o o • N • o e• o N • • o o o o o •-• o o 0 0 0

oo oo N o o • o • 0q •D 0• t• uq o o o o o o o 00 •o • N o •-• N •-• o o •0 o • ,-• ,• • • N N N o 0q N o o N •'• e• • o

o ,-• • uq 00 N 0 N u• • • •0 00 w• • •2' •2' ,-' • • • •'• • 0 I I I I I I I

o o • o o • o •o • o •1 o •o o c0 • o o •o •o •1 • • • c0 • o •1 o o • o wp o o • • o o •1 • o • • o o • • • • • • N • •1 •

<• • o• o• o • o o •-• • N

..o 00 .,o u• .4D N O e,I O 1'- wl• e,I ..0 ..0 o oo 0o • • r• ea o• ..o 1-- o o 00 o • N o o • ,--, • ,--, N ea N • 0O • O

• N 0 • O O !'- -,0 I'-- -,0 N N O 0 • O I'-- N O O O O • m ,-* ea ,--, ,--½ O

N O 0 • u• O I'-- -.0 I-- ..o O 00 O 0o ,'• I'.- c•D N 0 • N e,I '-• 0o • 0o • 0o -• • N • O

O wl• O m ..o 0o N 0o N m 00 0• I-- 00 • o o • wl• 00 • m ..o ..o ..o .,o u• wl• • o• ,--, o r-- 114 ISACKS AND MOLNAR and configurationof the parts of lithosphericplates that have descendedinto the mantle during the past 10 m.y. or more.

DOUBLE-COUPLE MODEL OF THE SOURCE MECI•ANISM The double-couplemodel [Honda, 1962] is adequatefor all the solutionswe obtainedor examined.Alternate models,such as a suddenvolume change,if presentin combinationwith a double-couplecomponent, would cause the nodal planesto be nonorthogonaland would affect the directionsof first motionsof shear waves. We find no evidencefor such effects [see also Ritsema, 1964], althoughwe cannot,in general,exclude minor contributionsfrom such com- ponents [Randall, 1968]. We adhereto the followingstandard terminology for the parametersof the double-couplesolutions: The two orthogonalaxes that bisectthe quadrantsof dilatationaland compressionalfirst motionsare the compressional•or P, axis and the tensional,or T, axis,respectively. A third axis,perpendicular to boththe P and T axesand coincidentwith the intersectionof the nodal planes,is the B, or null, axis. In the dislocationmodel equivalent to the doublecouple, one of the nodalplanes is the fault plane,and the poleof the othernodal plane is parallelto the slip vector. The correspondenceof the P, B, and T axesand the principal axesof maximum, intermediate,and minimum compresslyestress is an assumption,justified basically by the results-(describedin the next section),that for deep and intermediate-depthearthquakes the P, B, and T axesare the most closelygrouped of the double-coupleparameters and are most nearly parallel to the planar geometry of the seismiczones [Isacks et al., 1968, 1969; McKenzie, 1969a; Isacks and Molnar, 1969]. Hence, the physicalmechanisms of mantle earthquakesmay be suchthat the shearingor faultingforms a• an angleof about45 ø to the axisof maximum compressirestress. If the angle betweenthe P axis and the axis of maximumcompressire stress were as much as 15ø, however,detection of suchan effect [Isacks et al., 1969'] would be difficultwith the presentresolution of our data. Thus, in terms of a Coulomb-Mohr-typeprocess, we can only suggestthat the effectivecoefficient of internal friction is probablyless than 1 and may be 0.

STRESS WITHIN DESCENDING LITHOSPHERIC PLATES

Parallelism between Stress Axes and Inclined Seismic Zones The parallelismbetween the axes of compressionor tensionof the double- couplesolutions and the seismiczones is the primary evidencethat the earthquakes in the mantle occur inside descendinglithospheric plates in responseto stresseswithin the plates.The nonparallelismbetween the nodal planes (one of whichis the fault plane) and the seismiczones shows that the solutionscannot be interpretedas simpleshearing parallel to the zones.We demonstratedthese rela- tionshipsin our previouspaper (Figure 1 of Isacks and Molnar [1969]) by plottingthe axesand polesof the solutionson a commonprojection oriented relative to the local strike and dip of the well-definedinclined seismic zones. We do STRESSES IN THE DESCENDING LITHOSPHERE 115 not usethis methodof presentationfurther in this paper because(1) the precision with which the orientation of a zone can be determinedvaries greatly from region to region,and (2) a major causeof variationsamong solutions for a regionmay be variations in the structure of the zone rather than some inherent feature of the mechanism.Thus, furtherquantitative study of the closenessof groupingof the axesrequires detailed resolution of the distributionof hypocentersnear the events with we!l-determinedsolutions. In this paper we seek instead a broad global surveyof the outstandingfeatures that canbe discernedwith presentlyavailable information. The parallelismbetween the stressaxes and the seismiczones is demonstrated individually for eachregion in Figures7 through29. In most caseseither the P or the T axis is parallel to the zone.Often either the P or the T axis may group stronglyabout a commondirection, while the othertwo axesinterchange posi: tions or form a girdle about the stably oriented axis. This feature suggeststhat the intermediateprincipal stressis nearly equal in magnitudeto one of the other principal stresses[Balakina, 1962] In some.regions all three stressaxes inter- changepositions. Although the causeof sucha marked spatial variability in the distribution of stressesmay be obscure,the idea of a stressedplate doesoffer a generalexplanation for this otherwisepuzzling result. Hypothesesfor deep earthquakesof the type proposedby Sugimura and Uyeda [1967] and Savage[1969] do not accountfor the resultsof the preceding paragraph.In thesehypotheses the shearingand faulting is constrainedto be parallelto planesof pre-existingweakness in the material; thus the polesof the nodal planes,or a• least one set of poles,rather than the P or T axes,should be the most closelygrouped parameter of the solutionsand have the most consistent relationshipto the orientationof the seismiczone. Savage recognizes this problem and proposesthat the interchangingof stressaxes is characteristiconly of more contortedparts of slabs.In ourstudy, however, we canfind little evidenceto supportthis; the interchangingof axesoccurs in zonessuch as thosein parts of the Izu-Bonin,Tonga, and New Hebridesregions that are relativelyun½ontorted. Ritsema [1964, 1966, 1970] reportsthat the polesof the nodal planesfor earthquakesin the mantle tend to be horizontalor vertical and that the sense of shearingacross the more horizontalnodal plane reversesfrom intermediateto greatdepths. To explainthese results Ritsema proposes a predominantlyhori- zontal motion of material through the gap of seismicity between deep and intermediate-depthearthquakes (see also Harrington [1963]). This reversal in orientationis in agreemen•with our findingsexcept that we interpret this as a changefrom extensionto compressioninside an inclinedslab of lithosphere.The main differenceis that we find that the nodal planesor their polesdo not tend to be horizontal or vertical. A histogramof the plungesof the poleslisted in Table 1 showsa nearly uniformdistribution of anglesbetween 0 ø (horizontal)and 90ø (vertical).The difference seems to bea resultof Ritsema's[1970, p. 504]rejec- tion of solutionsin which the pattern of firs• motionsis dominatedby onepolarity. His selectionwould thus be biased against solutionsthat dø not have a nearly verticalnodal plane. Thelack of a strongtendency for the nodal planes to bevertical and: hori_ 116 ISACKS AND MOLNAR zontal also arguesagainst the hypothesis(e.g., Lliboutry [1969]) that the lith- ospherebreaks up and sinks vertically along en echelonvertical faults. A few solutionsfor intermediate-depthearthquakes can be interpretedas a downwardcontinuation of the thrust faulting characteristicof shallowportions of inclined seismiczones. The depthsof theseevents are lessthan about 100 km. Althoughin somecases either or both the hypocentraldepth and the dip of the seismiczone are uncertain,further resultsof this type may give someindication of the thicknessof the overthrustplate of lithosphere.

Predominanceand Variationso• Down,-DipStress Orientations For brevity we refer to orientations of the stress axes in which the P or T axis is parallel to the local dip of the seismiczone as 'down-dip compression'or 'down-dip extension,'respectively. The prevalenceof these orientationsis demonstrated by the fact that two-thirds of the 204 solutions listed in Table I are of one type or the other. Included are solutionsfrom nearly every region where mantle earthquakesoccur, so that the generalizationis global in scope.In regions where the structureof the zoneis known best,the axestend to scatterwithin 25.0 or less of the direction of dip; it is not certain how much of the scatter is due to complexitiesin the structureof the zonesor small systematicdifferences between the P, T, and B axes and the principal axes of stress in the material. In other regions,where the orientationsof the zonesare not so well known, the available evidence is still sufficient in many casesto conclude that the parameter most nearly parallel to the dip of the zone is either the P or the T axis. The remaining third of the solutionsare for events in markedly contortedparts of zonesor in zonesof undeterminedstructure, or are more or less well-establishedexceptions. Figure 2 is a summaryof the variations of down-dipstress type as a function of regionand depth.Also shownare the approximateconfigurations of the deep seismiczones, the maximum depths,and the locationsof marked gaps in seismic activity as a function of depth. The data and interpretations upon which this figure is basedare describedin detail in the regional sectionsand are summarized in the next-to-last column of Table 1. Not included in the figure are certain complexregions near prominent junctions or endsof zones (discussedin the next section) and a few regionswhere the structureis very poorly known. The figure illustrates the following major results of this study: 1. There are remarkablesystematic differences between intermediate-depth and deepsolutions: (a) The deepsolutions are very stronglydominated by down-dipcompression. (b) The intermediate-depth solutions are more variable and include down-dip compressionand extension,as well as many solutions that are clearly neither (X's in Figure 2).

2. There is a correlationbetween the down-dipstress type at intermediate depthsand the grossnature of the distributionof seismicityas a functionof depth: STRESSES IN THE DESCENDING LITHOSPHERE 117

Fig. 2. Global summary of the distribution of down-dip stresses in inclined seismic zones. The stress axis that is approximately parallel to the dip of the zone is rep- resented by an untilled (open) circle for the compressional or P axis and a 5OOf• NORTHCENTRALSOUTH filled (solid) circle for the 7001- "•'" tensional or T axis; an X indicates that neither the P nor the T axis is approximately parallel to the dip. The data plotted a•re tabulated in the second-to-last 700F.... NCHILE NE•'HEBRIDES-- column of Table 1. For each region the line represents the seismic zone in a vertical section aligned perpen- 3OO. ? ...... dicular to the strike of the 500[•/ ' / (• ...... RMANi • zone. The lines show ap- 7oo[-¾..... NEw- --NEW ..... •E.5•-- _ _ ._.••US.. _ _ S__?._WE..__ ._U__.M•_- _ --E _ A-4 ISUNDAMINDANAO ZEALAND BRITAIN SOLOMONS ALEUTIANS SUMATRA AEGEAN HINDUKUSH / proximately the dips and lengthsof the zoneand gaps in the seismicactivity as a function of depth.

Down-dip extensionis predominantin regionswhere either marked gapsin seismicityare presentbetween deep and intermediate-depth eventsor where no deep earthquakesare present. (b) Down-dip compressionis predominant where the zone is continuous, but this correlation is tenuous. 3. 'There is no obvious correlation between the bends in the vertical sections shown in Figure 2 and the down-dip stress type. Down-dip compressionis dominant below 300 km regardlessof the configurationof the zone, whereasat intermediatedepths the variations in down-dip stresstype are not related in any obviousway to the structure of the zone.

Sinking o[ the Lithosphere Sinkingat intermediatedepths. We claimedin our previouspaper that gen- eralizations i and especially 2 listed above constitute evidence for the models shownin Figure 3 in which lithosphericplates, more densethan the surrounding mantle,sink into the asthenosphereunder their ownweigh[ and are therebyplaced underextension. Thus the sectionsshown in the middleand lowerparts of Figure 2 correspondto modelsa, b, or possiblyd in Figure3, whereasthose in the upper par[ correspondto c (or possiblyb in the Kurile-Kermadecsections). The idea that the suboceaniclithosphere is gravitationally unstable and can sink into the asthenosphereis physicallyplausible and supportedby seismologicaland gravitydata (e.g.,Elsa,sser [1967], Oliver and Isacks [1967,1968], McKenzie 118 ISACKS AND MOLNAR a b c d

LOW INCREASINGSTRENGTH•) •o 0 HIGH STRENGTH Fig, 3. A model showingplausible distributions of stresseswithin slabs wheregravitational forces act on excessmass within the slabs.A filled circle representsdown-dip extension, an untilledcircle represents down-dip compression,and the size of the circle qualitatively indicatesthe relative amount of seismicactivity. In (a) the slab sinks into the asthenosphere, and the load of excessmass is mainly supportedby forces applied to the slababove the sinkingportion; in (b) the slabpenetrates stronger material, and part of the load is supportedfrom below,part from above; the stress changesfrom extension to compressionas a functionof depth.In' (c) the entire load is supportedfrom below, and the slab is under compression throughout.In (d) a piecehas brokenoff. A gap in seismicityas a function of depth would be expectedfor (b) and (d), whereasno deep earthquakesoccur beneaSh (a). The horizontaldashed lines in the figuresin- dicatepossible phase changes in the uppermantle near 350-400km and 650-700km [Anderson,1967]. (Figure after Isacksand Molnar, 1969.)

[1969b],Press [1969], Hath•rton [1969, 1970], Jacoby [1970], and Griggs [19711). Mos• of •he seismiczones shown in Figure 2 appear •o be in descending piecesof suboceaniclithosphere. Beneath several continental regions in theAlpide bel•, suchas Burma,•he Hindu Kush, and Rumania,however, •he subcrustal zonescanno• be •raced•o any nearbypiece of oceaniccrusL Although •here is no problemin havingoceanic lithosphere sink in•o •he mantle,continental lithosphereis probablybuoyan• and would •end •o remainon •he surface.Geological evidenceindicates that much of the Alpide bel• may be beneath or near the suture where an ocean has closed and where collision of continental pla•es has taken place [Holmes,1965]. Thus •he prevalenceof nearly vertical extensional stressesin the mantle beneaththese regions suggests •ha• piecesof oceaniclithosphere,remnants of a consumedsuboceanic pla•e, are now pullingaway and downwardfrom the continentalpiece •o which i• is a•ached. The imperfectionof the correlationbetween down-dip s•ress type at in•er- mediatedepths and the distributionof seismicityas a functionof depth--notably, the variability of s•resseswithin •he Kurile arc and the southernNew Hebrides, •he down-dipcompressions in the Sou•hSandwich and Ryukyu arcs,and the complexorientations tha• are foundin severalother regio.ns--sugges• tha• a• intermediatedepths stresses other •han the simpletwo-dimensional •ype implied in Figure3 are presen•in certainregions. This is no• surprising,since events a• intermediatedepths are neares••he regionwhere the pla•,ebends downward STRESSES IN THE DESCENDING LITHOSPHERE 119 into the mantle; thus in certain regionsgeometric adjustments to this major deformationmay locally controlthe orientationof stressinside the plate. Other complicationsmay include effects of seamountsand other topographic features on the forcesapplied to the descendingslab in the underthrustingzone; effectsof fault zones and other mechanicalheterogeneities inherited from the formation and deformation of the slab on the surface prior to its descent; variations in the gravitational force due to a heterogeneousdistribution of excessmass in the slab; and localizedeffects of chemicalor phasetransitions that are not distributed uniformly through the descendingslab. Some of these effectscould also expla.in the nestsof intenseactivity or remarkablegaps in activity that characterizethe distribution of earthquakesalong the strike of many zones. Down-dip compressionbetween 300 and 700 km. The most consistentresult of this investigation is the predominanceof down-dip compressioneverywhere below about 300 km. One interpretation of this result is that the forces on the lower portions of the slab that resist its motion become relatively important below 300 km. The compressionalstress inside the slab would be the result of.the downward-directed forces applied to the upper portions of the slab and the upward-directedresisting forces applied to the lower parts of the slab. The cause of this resistancemight be either an increasein strengthin the surroundingmantle or a buoyant effect if the density of the slab were lower than that in the surrounding mantle. The interval of depth in which down-dip compressionis prevalent is bounded approximately by the depths of 350-400 and 650-700 km at which major phase changesare thought to occur in the normal mantle [Anderson, 1967]. If the shallower transition (near 350-400 km) occurs at shallower depths in the slab than in the adjacentmantle [Griggs,1971], •he additionalload of densematerial would further drive the lower portions of the slab downward and therefore accentuatethe compressionalstress inside the slab below 300 km. This effect is supportedby both the focal-mechanismdata and the estimatedparameters of the phasechange [Anderson, 1967; Akimoto and Fujisawa, 1968]. On the other hand, the parametersgiven by Andersonsuggest that the deeper phase change (near 650-700 km) would occur at greater depth in the slab than in the surroundingmantle. This would provide an upward buoyant force that resiststhe sinking of the lithosphere.The critical parameters in this case are the slope of the pressure-temperaturecurve at the phase boundary and a possible increasein the iron-magnesiumratio with depth. Unfortunately, however,to our knowledgethese parameters are not well determined. If the parameters of the phase changeare such that buoyant resistancedoes not occur,the prevalenceof down-dip compressionprovides strong evidencefor an increasein strength in the mantle. Alternatively, if the strength does not increase enough to support the heavy slabs,then the focal-mechanismdat• supportthe idea that the buoyant resistingforce is important. In Figure 3 we suggesta zone of increasingstrength below depths of about 350 km to account for the compressionalstresses in sqismic zones that reach depthsof only 500 km. This is reasonablein view of the indicationsof • minimum of creep strength as a function of depth in the upper 300-400 km of the mantle 120 ISACKS AND MOLNAiR [Gordon, 1965; Anderson,1966; McConnell, 1968; Weertman,1970]. However, the strength of the material through which the slab penetratesmust remain significantlysmaller than the strengthof the slab if the plate is to maintain its identity as suchto the depthsof the deepestearthquakes. 'Strength' as usedhere is creepstrength as definedby Weertman[1970]. Gaps in the lithosphere. Figure 3 showstwo possibleexplanations for the gaps in seismicactivity as a function of depth; in Figure 3b t.he lithosphere is continuous,but the deviatoricstress goes to zero in changingfrom compression to extension.In Figure 3d the lithosphereis discontinuous,and the gap in seismicity correspondsto a gap in the lithosphere.In this model a piece of litho- spherehas becomedetached and is sinking independently. There is evidencethat a[ least in someregions a piece of lithospheremay be detached as shown in Figure 3d. The remarkable configurations of the zones beneath Peru and western Brazil and the New Hebrides (Figures 10 and 14) sugges[this. The deep earthquakesbeneath the North Island of New Zealand [Adams, 1963] represen[a significantdiscrepancy in the otherwiseregular relationship between the lengLhof the seismiczone and the rate of convergencealong the Tonga-I(ermadec-New Zealand island arc [Isacks et al., 1968; McKenzie, 1969b] and therefore may be located in a detached piece of lithosphere.The solutionfor the Spanishdeep ear[hquake,in comparisonwith result•sfor nearly every other region of deep-focusactivity in the world, is not explicable as down-dip compressionin a lithosphericslab that is traceable t•o a presently active island arc on the surface; this earthquake may thus also be located in a pieceof lithospherethat is detachedfrom a surfaceplate. Observationsof the frequenciesand amplitudes of seismicshear waves from deep earthquakesrecorded a[ suitable locationsmight indicate the presenceor absenceof a gap in the lithosphere.In New Zealand [Mooney, 1970], the New Hebrides (Barazangi, personalcommunica.tion, 1970), New Brit•ain,the Solomons, and Chile [Molnar and Oliver, 1969; Sacks,1969], regionswhere notable gapsin seismicity are present, high-frequencyshear waves pass through the gap. In the caseof the Peru-westernBrazil region [Molnar and Oliver, 1969; Sacks,1969], the results are as yet ambiguous.The clearly recordedhigh-frequency phases mean either that the lithosphere is continuousthrough these gaps or that the zone of large attentuation in the adacen[ normal mantle is confined primarily to the region above the gaps, i.e., above about 300 km. Sincethe last alternative cannot be excluded,the questionremains open. Contortionsand Disruptions of the.Descending Lithosphere The two major world-encircling belts of island-arc or island-arc-like structures, the circum-Pacific and Alpide belts, are divided into more or less well- defiheddiscrete segments. The junctionsbetween these segmentsare characterized by abrup[ changesor offsetsin trend, marked changesin the dip of the seismic zones,gaps in the seismicactivity, shallowingof the oceanictrench, and intersection of major transverse features. The curvature of the island arc on the surface,though generally convex toward the oceanicside, varies considerablyamong the segmentsof island arcs. In somecases the arc endsabruptly and the inclined STRESSES IN THE DESCENDING LITHOSPHERE 121 seismic zone has a well-defined lateral edge. These features imply that the descendingedges of lithosphericplates are contortedand segmented.Thus, it would be surprisingif the distribution of stressesinside the descendingportions couldbe completelyunderstood in terms of the two-dimensionalmodels of Figure 3. In particular, the solutionsexcluded from Figure 2 are for eventslocated near major junctions and contortededges of seismiczones, and are discussedbelow. Moreover, someof the solutionsshown in Figure 2 by an X (i.e., solutionsinter- pretable as neither down-dipcompression nor extension),as well as someof the anomalousdown-dip types, can be attributed to deformationsof the slabsthat result from three-dimensionalgeometrical effects rather than the two-dimensional effectsof Figure 3. Bendingof lateral edgesof inclinedseismic zones. The distributionof hypo- centersat the northern end of the Tonga island arc (Figure 11) and beneaththe Banda Sea at the easternend of the Sunda island arc (Figure 26) showthat the lateral edgesof the seismiczones in thesetwo regionsare bent upward. A simpli- fied picture of such a deformation is shown in Figure 4. Limited evidence suggeststhat bent edgesare also located at the northeastern end of the seismiczone beneath Colombia, South America (near the Bucaramanga 'nest'), the northern ends of the West Indies a.ndthe South Sandwicharcs, the southernend of the New Hebrides arc, and the northern end of the Izu-Bonin arc. Direct effectsof the bending,i.e., a compressionor an extensionperpendicular to the axis of bending,are indicatedclearly by only two solutionsfor deepevents beneaththe Banda .Seaand by solutionsfor intermediate-depthevents in Colom- bia. In the latter regionthe interpretationis very uncertain.In contrast,the solutions for events near the northern end of the Tonga arc appear to be merely rotated versionsof the solutionsin the unbent part, wherein the axis of rotation correspondsto the axis of bending.This relationshipimplies that the bending

Fig. 4. Diagram showing a bent edge of a descendinglithospheric plate. The surficial portion of the plate is to the left; on the right side the plate bends down and de- scends into the mantle. The lower corner of the descendingpart of the slab curls upward, as shown in the lower right-hand side of the model. Such a structure is postulated, for example, for the eastern, northern, and southern ends of the Sunda, Tonga, and New liebrides arcs, respectively. (Figure after Fitch and Molnar [1969].) 122 ISACKS AND MOLNAR doesnot producethe stressreleased by the earthquakesbut merely changesthe orientation of the stressguide. Oblique-angledjunctions. Three notable oblique-anglejunctions, all con- cave toward the oceanicside, are located respectivelya• the junctions of the Kurile, North I-Ionshu,and Izu-Bonin arcs and betweenthe island-arc-like structures of Peru and Chile in westernSouth America. Only near the junction between the North I-Ionshuand Izu-Bonin segmentsdo we find goodevidence for localized effects on the focal-mechanismsolutions. The data suggesttha• the lithosphere may be tearing beneath central I-Ionshuand descendingbeneath the Sea of Japan and south of I-Ionshuas separateslabs (Fig. 5b). Wyss [1970] suggeststha• the low apparent stresscalculated for intermediate-depth events near the Peru-Chile corner may be due to a tear in the slab. However, the solutionsfor intermediate-depthevents exhibit no obvious effects of the corner,bu• seemto be related to stressesinside one of the segmentsrather than to stressesattributable to the presumedtearing.

•C•'HINGEFAULTING

Fig. 5. Contortion (a) or disruption (b) betweentwo segmentsof a slab with different dips. In (a) the portion between the two segmentsis bent and stretched. In (b) the two segments tear apart along a vertical hinge fault. In this type of faulting the motion would be nearly vertical and such that the more steeply dipping side moves down relative to the more gently dippingsid e. An examplemay be locatedbeneath south-central Honshu (Figures 21 and22). STRESSES IN THE DESCENDING LITHOSPHERE 123 This may also be the casefor the junction betweenthe Kurile and North Itonshusegments, where •he solutionsseem more rela•ed •o pa•ternson either side than •o effectsof •he corneritself. However,•he interpretationfor •his regionis more complexbecause of •he variability of s•ressorientations within •he KUrile zone. Junctionsbetween the Tonga and Kermadec island arcs and the North Island of New Zealand, and betweenthe Izu-Bonin and Marianas, are characterizedby low seismicactivity; and few solutionsnear •hese junctionswere obtained.The distributionof hypocenterssuggests •ha• the Tonga-Kermadeczone is disrupted in•o two segmentscorresponding to a Tonga slab and a Kermadec slab. Convexcurvature. Frank [1968] pointsout that unlessa simplegeometrical relationship holds between •he dip 'and the curvature of •he surface trace of the seismiczone, surface area is no• conserved,and •he descendingpla•e will be either laterally compressedor s•re•ched.$tauder [1968b] suggeststha• such an effe½• could account for extension parallel to •he strike of •he seismic zone •ha• he inferred from •he solutionof an intermediate-deptheven• in the Aleu•ans (Fig- ure 19). AlthoughStauder's argumen• is basedon a fia•-earthmodel (as depicted in Figure 6), extensionis s•ill to be expectedthere becauseof •he s•eepdip of the seismiczone [Davies and McKenzie, 1969; Murdock, 1969]. A similar effect may accountfor solutionsof severalevents in •he Sunda,Mindanao, and New Hebrides arcs.

Fig. 6. Illustration of how lateral extensioncan be producedin the descendingportion of a plate near a corner (or bend) that is concavetoward the direction of dip of the slab. The dotted line representsthe axis of the trench and the axis of bending. In 6A a horizontal, undeformedslab is shown.Bending of the underthrustedge of the slabalong the trenchaxis produceslateral extension or separation near the corner; in 6B a separation is shown for clarity. An example of possible lateral extension is shown in Figure 19 for the Aleutian arc [Stauder, 1968b]. • 124 ISACKS AND MOLNA:R On the other hand, if the dip of the seismiczone is lessthan twice the radius of curvature (in degrees)of the surfacetrace of the arc, then lateral compression shouldresult. One solutionin agreementwith this is found at intermediatedepths in the nearly straight Izu-Bonin arc. By the same reasoning,lateral compressionshould be presentnear the con- cave cornersdiscussed in the last section.In those cases,however, the zoneson either side differ appreciablyin dip sothat a contortionor disruptionof the types depictedin Figure 5 may apply instead. Bending as the source o• earthquake-producing stresses. The evidence reviewed above, plus •he lack of effect on the mechanismsolutions of the bends in the seismiczones depicted in Figure 1, suggeststha• s•ressesdue •o bendingof •he lithosphericplates do no• play an important role in the generationof earthquakesin the mantle. This conclusionis no• surprisingin view of the lack of any appreciable seismic activity that can be associatedwi•h •he very pronounced bending of the lithosphere as it turns downward beneath the island arc. The only earthquakes•ha• appear •o be clearly associatedwith •his bend are shallowear,h- quakeslocated beneath or slightly seawardof the axis of the •rench and characterized by horizontal extension perpendicular •o the trench. These are mos• prominentin the Aleutian arc [Stauder, 1968b], althougha few examplesare reportedfor o•her arcs [Molnar and Sykes, 1969; Katsumata a•d Sykes, 1969; Fitch, 1970a; Isacks, 1970]. An even• a• 80 km beneath the axis of the Kurile trench may also be associatedwith compressionals•ress at the bottom of •he bendinglithosphere, although in this case•he interpretation is uncertain (Figure 20). The par• of •he under•hrus• lithospherebetween the trench axis and •he volcanoes,where the bendingis presumablymost severe,appears to be relatively aseismic.The relief of extensionalstress near the upper par• of the ben• suboceaniclithosphere by earthquakesmus• thereforebe suppressedwhen the upper surfaceof the plate movesbeneath the arc. The implication of these observationsis that the generationof intermediate- depth and deep earthquakesis governednot only by the distribution of stress within the lithospherebut also by unusualphysical propertiesor conditionswithin the downgoingpart of the plate. This is further suggestedby the fact that earth- quakesinside the lithosphereare apparently rare exceptin the downgoingpart. Of three generaltypes of hypothesesthat have beenproposed to explainthe proc- essof shearfracturing in a deepearthquake--(1) an embrittlementor weakening [Raleigh and Paterson, 1965] or some other effect (e.g., Griggs and Handin [1960] and Savage [196.9]) associatedwith a fluid phase; (2) a high-temperature instability in the creeprate (e.g., Orowan [1965], Griggsand Baker [1969]); and (3) a high-pressure,low-temperature mechanical instability [Byeflee and Brace, 1969]--the first might be expectedto apply more to the sidesthan to the centerof the descendingslab. The lack of eff•btsattributable to simplebending of the slabs suggeststhat the earthquakesare located nearer the center than the sides, and further suggeststhat the earthquakes occur in the coldest material. Theseinferences concur with (3), althoughboth the theoreticaland the experi- mentalbases for that hypothesiswith respectto deepearthquakes are uncertain. STRESSES IN THE DESCENDING LITHOSPHERE 125 Implications for Plate Tectonicsand Mantle Convection In adding œur•hersuppor• •o •he concep•oœ subcrustal earthquake zones as par•s oœcoheren• and s•rongpla•es of lithosphere,our resultscontribute •o •wo major conditions•ha• a convection•heory shouldsafisœy. The firs• is •ha• litho- sphericmaterial can descend•o depthsoœ.a• leas• 650-700 km. Furthermore,•he predominanceoœ compressive s•resses a• grea• depthsand •he correspondenceoœ •he depth of •he sharp cutoff in world seismicityas a œuncfionoœ depth wi•h a major discontinuityor •ransi•ionregion in •he man•le [Engdahland Flinn, 1969] sugges• bu• do no• prove •ha• •he lithospheric material does no• penetrate deeper•han 700 km. The secondcondition is •ha•, because•he descendingslabs would be impenetrable, the configuration oœ•he inclined seismic zones places s•rong constraints on •he movemen• oœmaterial in •he as•henosphere;any flow of material mus• •urn down near •he zonesand canno•pass •hrough •hem. By •he same reasoning,•he contortionsoœ •he slabs contain inœorma•ionabou• •he integrated history of •he relative motions between •he lithosphere,•he as•henosphere,and, possibly,•he mesospherein •he casesoœ •he deepes•zones. Although •he contortionsare no• oœsufficien• magnitude •o obscure•he basic inclined-plane asymme•ry oœmos• zones,•hey do a•es• •o relative mo•ionsbetween •he upper and lower par•s oœslabs, including bo•h •ranslafionsand ro•a•ions. The forcesinvolved in •he gravitational sinking a• intermediatedepths, if •ransmi•ed pas• •he bendingand under,brushingpar• oœ•he lithospherenear •he •rench and island arc, couldbe an impor•an• contribution•o •he driving mechanism of sea-floorspreading and continentaldriœ•. Moreover, •he increasedresis- tance •o •he downwardmo•ion oœ•he lithospherea• depthsgrea•er •han 300 km migh• causealterations in •he pa•ern oœmovements oœ surficial pla•es oœlitho- sphere.Our results for •he Tonga region may have a direc• bearing on •hese questions.The Tonga arc includes•he cleares• example oœa continuouszone under compression•hrough a range oœdepths between abou• 80 km and grea•er •han 600 km; in •his case•he resisting•orces applied •o •he lowerpar• oœ•he slab apparenfiy suppor• nearly •he entire load oœexcess mass within •he slab. Thus (1) •he excessmass above 80 km is sutTicien••o drive •he slab downward; (2) the slab is being pusheddownward by forcesapplied mainly to the surfaceportions; or (3) •he downwardmo•ions are essentiallys•opping. The high level oœshallow seismicity [Brine, 1968; Davies and Brune, 1971], •he lack oœsediments in •he Tonga •rench,•he depthoœ •he •rench,and o•her geophysicaland geologicalobser- vations argue agains• bu• do no• disprove (3). The firs• is no• likely because•he lithospherei•selœ may be nearly 80 km. Thus the results for •his region (and probably for North I-Ionshu also) sugges••ha• •he gravitational sinking force and resis[antea[ grea[ depth, allhoughprobably impor[an[ in [he dynamicsof [he descendingpor[ions of [he pla[es, eanno[ be [he only importan[ forcesdriving [he horizontal moJiohsof [he li[hosphere.

REGION-BY-REGION PRESENTATION OF DATA A de[ailed region-by-regionanalysis of •he rela[ionshipsbe[ween [he siress axes inferred from the focal-mechanism solu[ions and [he local s[rue[ures of [he seismic zones is presen[edin •his see[ion. The eri[ieal information is •he dis[ri- 126 ISACKS AND MOLNAR butionof hYpocentersnear the earthquakesfor whichœocal-mechanism solutions are available.Three basicsources of data usedœor all the regionsare Gutenberg and Richter [1954],the locationsoœ •he Coastand Geode•icSurvey (ESSA) as compiledby Barazangiand Dorman [1969], and Duda [1964]. In particular, enlargedversions of Barazangiand Dorman'smaps of epicenterswithin intervals of 100-km depth were very helpful. These data are supplementedby more detailed s•udiesof particular regionsif available. In eachregion •he solutionsare divided into groupsaccording •o their similarity and the locationsof •he earthquakesinvolved. If the structureof the seismiczone is reasonablywell determined,the orientationof •he stressaxes can then be comparedwith that of a planethat bestrepresents the local structureof the zone.This provides•he basictest of the hypothesisthat deepand intermediate-depthearthquakes indicate stresses within thin lithosphericplates as well as of additionalhypotheses discussed in the previoussection. On the otherhand, if •he structureof the seismiczone is poorlyknown, the procedurecan be reversed; the stressaxes of •he solutionsdetermine planes whose orientations can then be comparedwith the availableinformation on the distributionof hypocentersand wi•h other regional•rends. This comparisonyields additionalweak testsfor the hypothesisof stressedplates as well as informationon •he detailedregional tectonics. In addition, the featuresused in •he constructionof Figure 2 are described for each region; these include the maximum depth of hypocentersand the pres- enceand range of depthsof a gap in the distributionof hypocentersas a function of depth. The determinationof gaps is subject to some uncertainties;it is not possibleto discriminatebetween • truly aseismicgap and a gap•hat is actually a minimum in a continuousbu• insufficientlysampled distribution. The presentations•ar•s with the New Guinea region and proceedscounter- clockwisearound the Pacific and •hencethrough Indonesiaand the Himalayan- Alpide belt •o the MediterraneanSea. The numbersused in Table I and the figures refer •o either the. earthquake or its mechanismsolution.

New Guinea-New Britain-SolomonIslands (Figures7 and 8) Deep sea •renches,belts of volcanoes,and inclined zonesof shallow- and deep-focushypocenters form island-arc structuresparallel to •he New Britain and SolomonIslands. These •wo structuresmeet and form a sharp corner east of the southeastern•ip of New Ireland. Like the New Hebrides arc farther east and unlike all other circum-Pacificarcs, the New Britain and Solomonarcs are convexto the south and face away from the Pacific basin. Apparently, litho- spherebeneath the Coral and SolomonSeas is being consumedin these arcs. Southwestand west of New Britain subcrustalearthquakes occur beneath New Guinea,but •he distributionof hypocentersis not well defined.Several additional zonesof shallow earthquakesare present in the region. The most striking is a' well-definedlinear alignment of shallow epicenterstha• crossesthe Bismarck Sea north of New Britain. Another follows the Bismarck archipelago,while a third appears•o crossthe SolomonSea from the southeasternend of New Guineato the SolomonIslands. These zones suggest •ha• in this region•he boundarybetween STRESSES IN THE DESCENDING LITHOSPHERE 127

5-7 8-10 ß x•

•00• o

NEWBRiTAiN

NEW GUINEA

! i 140 E 150 160

Fig. 7. The Solomons and the New Britain island arcs and the region of New Guinea. The following symbols apply to Figures 7-29- The large circles are equal-area projections of the lower hemisphereof a focal sphere and are oriented accordingto the directions of the map projection; in all casesthe top and right-hand side correspondto north and east, respectively. A filled circle representsthe axis of tension, T; an untilled circle representsthe axis of compression,P; and an X representsthe null axis, B. A solid line in the projection is the trace of the plane that best fits the orientation of the seismic zone; a dashed line is the trace of a plane parallel to two of the stress axes (or the average positions of groups of axes) and perpendicular to the third axis. If earthquakes occur in thin, slablike stress guides, this plane should give the orientation of the slab. The numbers near the projections refer to the solutions plotted in the projection and listed in Table 1. On the map the contoursof hypocentral depth (in kilometers) are shown by solid lines. The epicenters of the events are shown by filled, upward-pointing triangles for hypocentral depths of 70-229 kin; untilled, upward- pointing triangles for depths of 300-499 kin; and filled, downward-pointing triangles for depths of 500-700 kin. The landward side of coastlinesare shown by dots. In this figure only, the dashedlines on the map representzones of shallow earthquakes.

the large Australian and Pacific plates is complexand probably includesthree o.r more additional small plates. An additional reference used for this region is Denham [1969]. Solomonarc. A zoneof intermediate-depthearthquakes with depthsmainly lessthan 150 km parallelsthe southeasternpar• of the Solomons.Deeper earth- quakesappear •o be confinedto the westernmos•parb (wesbof abou• 155øE,)and definea very steeplydipping zone •ha5 reaches depths of 500-600km (Figure8). 128 ISACKS AND MOLNAR

NW -.-* N E ---.• v ¾ o oo b ' c '

2OO _ Cib 0 KM

400 - 4;;) Fig. 8. Vertical sectionsthrough the New Britain are (A) and the northwestern end

600 of the Solomons are (B). Sections include the most accuratelylocated events I I I based on data reported in the Earth- quake Data Reports for 1961-1965.Events within 100 km of the sections A and B are plotted.

! i 145E 1•5

The deepestshocks are displacednortheast of the nearly vertical distributionof hypocentersabove about 400 km. Denham'ssections and Figure 8 suggestthat a gap or a minimumin seismicitymay be presentbetween depths of about200 and 350 km. Becauseearthquakes are few in number at all depthsbelow 150 km, however,discrimination between the two possibilitiesis quite uncertain. Earthquake I is locatedin the easternpart of the arc (Figure 7) wherethe inclinedzone appears to dip steeplynorth. The T axisof I is nearlyvertical and thus couldindicate extension parallel to the dip of a nearly vertical zone.The B axis is approximatelyparallel to the trend of the arc. Althoughthe depthgiven in the PreliminaryDetermination of Epicentersof the Coast and GeodeticSurvey (ESSA) is 95 km, this depth is revised to 52 km in the Coast and GeodeticSur- vey's SeismologicalBulletin. The shallowerdepth is in better accordwith the characterof the seismogramsfor this event.Moreover, the solutionfor I is consistent with solutionsfor other shallowearthquakes in the region (Molnar, unpub- lished data, 1969). Thus, the alternate and preferable interpretationis that 1 indicatesunderthrusting of the Coral Sea beneaththe Solomons;i.e., slippage between two lithospheric plates. Solutions2-4 for earthquakesin the westernmostpart of the zone exhibit a marked variation with depth. The solutions3 and 4 for the events below 300-km depth have nearly vertical P axes, whereasthe solution for the event at 118 km has a nearly vertical T axis. The zone above400 km, includingearthquakes 2 and 3, is nearly vertical and strikes approximatelynorthwest; thus solutions2 and 3 indicate,respectively, compression and extensionparallel to the dip of the zoneat depthsof 118 and 392 km, respectively. A plane parallel to the T and P axes and perpendicularto the B axis of 4 dips steeplybut has a more northerly strike than that of the zoneat intermediate depths.The distributionoœ hypocenters near 4, althoughnot well defined,suggests STRESSES IN THE DESCENDING LITHOSPHERE 129 such• changein s•rike. The da• m•y •hus be ½onsis•en•wi•h •n inœerenceof down-dip compressionnear •he bottom oœ•he zone. New Britain arc. Figure 8 and the sectionsgiven by Denham show an inclined seismiczone •h•t dips north-northwesterlybeneath •he New Britain islands.These d• plus recentlo½•fions not plo•ed in •he figureshow that this zonere•½hes depths gre•er •h•n 500 km, •l•houghhypocen•ers deeper •h•n •bou• 200 km occur very inœrequen•ly.Toward the wes• •he line oœ•½tive volcanoes •nd •he bel• oœshallow •nd intermediate-depthe•rthqu•kes bend •round •o • more westerlytrend •nd in•erse½••he northern ½o•stoœ New Guinea. In •his regionthe zone appearsto steepenin dip. Earthquakes 5-7 are located beneath eastern New Britain, where the seismiczone dips north-northeas•at about 45ø. The T axesof 5 and 6 are parallel to the dip of the zone,and the B axesare parallel to the strike oœthe zone.These resultsthus indicate down-dip extensionalstress. The T axis of 7, however,is nearly vertical, whereasone of the nodal planes is most nearly parallel to the seismic zone. This relationship is mos• simply accountedfor as lithosphere beneath •he Solomon Sea being thrust under New Britain. The. depth oœthis earthquake (103 km) appearsto be reliable. The solution for 7 may thus represen• a downward continuation of the thrust œaulting •ypical of shallow events in the region (Molnar, unpublisheddata, 1969). The solutions for 8-10 for events located in the western par• oœthe arc are similar to one another. A plane through the B and T axes strikes approximately parallel to the local alignmentoœ earthquakes and volcanoesand dipsvery steeply north. This plane is also a reasonablerepresentation of the distribution of hypo- centersshown by Denham. These solutions,like those for 5 and 6, are thus consistentwith down-dipextensional stress a• intermediatedepths. New Guinea. The seismicity of New Guinea is very complex. A zone oœ subcrustal earthquakes with depths less than 200 km trends west-northwest beneaththe northern and westernparts of New Guinea and appearsto dip sou•h or southwest.This zone also appearsto be nearly continuouswith the steeply dipping zone of the New Britain arc. Near the intersectionoœ •hese •wo zones, where the dip apparently reverses,the pattern is complicatedby a •hird, very poorly definedzone that •rendseast-southeast. If •he solutionfor 11 is indicative oœdown-dip extension,then this zone would have a westwarddip. However, the solutionsfor three, closelygrouped, intermediate-depth shocks (12-14) are qui•e differen• from one another, although one of these (12) is similar •o 11. The data plotted in the appendixshow that thesedifferences are real. The variability in orientationsexhibited by 11-14 and the unresolvedcomplexities in the configuration of the seismiczone make œur•herinterpretation pointless.

New Hebrides (Figures9 and 10) Near the Santa Cruz Islands the seismicbel• of •he Solomonsbends abruptly and continuesas the linear south-southeasterlytrending seismiczone of the New Hebrides island are. The inclined seismic zone is very active a• intermediate depths.Figure 10 shows tha• it hasa slablikestructure with a dipoœ a'bout 65-70 ø. The zoneis remarkably linear in strike •nd is parallel to •he line of active volca- 130 ISACKS AND MOLNAR •

I I i i

_ '"'•....Lx._ , -

Fig. 9. The New Hebrides island arc. In this figure and in others, except where 15 noted, the heavy dashedline ' NEW600-650 KM - on the map represents the /" • • \_%• HEBRIDES FIJI• axis of the oceanic trench associated with the arc. The light dashed line encloses the epicentersof deep earth- 20 quakes with depths between about 600 and 650 km. / X \ -• [ '-o./_' , J 4- F, I- '----:{ _x''-----t-.... \ ,, '•-• / \,'

, I 160 E 165 170 1'75

WSW Trench •Volconicline ( N 20øW, thru 15 5øS, 168 OE) ENE I i I I I I o ob e' o• o•O o ;

I00 ' o• ¸ KM

2O0 ' 15 -

:50O

4OO - NEWCALED*ONIA i""' "" f 500- I l•o5E ITO 175_

o 60( - o o o- o Ooø 8 o o

->_00,, -I00I .... 0I, +100I KM 200I , •)0 400I 500I 600I , Fig. 10. Vertical sectionthrough the New Hebridesarc, including the most accuratelylocated events that occurredbetween 1963 and 1967,selected from the same sourcesused in Figure 8 plus •ykes [1964]. The sectionin- cludesevents within the rectangulararea shown on the map inset. STRESSES IN THE DESCENDING LITHOSPHERE 131 noeson the surface.Near its southernend the zone appearsto bend slightly east and then terminatesabruptly; many featuresof the seismicityand bathyme•ry [Sykeset al., 1969] suggestthai the are is •ermina•edby a transformfaull Irend- ing east-northeast. The evidencefor a gap in activity betweendepths of about300 to 500 km is good.At greaterdepths •he distributionof hypocen•ersis qui•eremarkable (Fig- ure 10). Figure 10 showsthai hypocenterswith depthspredominanfiy between about600 and 650 km occurin a nearly horizontalplanar zonetha• in map view (Figure 9) is elliptical and has i•s longestdimension •rending west-northwest. Only one deep earthquake (32) is known to have occurredsouth of •his zone. Intermediate-depthearthquakes. Solutions 15-23 for earthquakesin •he northern half of the linear belt of intermediate-depthearthquakes all have very steeplydipping T axes•hat are more nearly parallel •o the dip of •he zone•han are the polesof •he nodalplanes. These solutions are thereforein agreementwi•h an inferenceof extensionalstress parallel to •he dip of •he s•eeplydipping slab. Although four of these solutionshave B and P axes approximatelyparallel, respectively,to •he s•rike and normaldirections of the inclinedzone, the B and P axesof 17 and three othersare disfin½fiynonparallel •o thosedirections. The solutions24-29 for earthquakesin •he southernpar• of •he zone differ appreciably from those in •he northern part. Solution24 is nearly identical •o 25 and 26, even though earthquake24 is locatedclosest •o earthquakesof the northern group and separated by more •han 300 km from earthquakes 25 and 26. Solutions24-26 are similar •o 15-25 in •hat •he T and B axes are parallel •o •he inclined zone, but differ in that •he T and B axesare distinctly not parallel •o the dip and strike directions. Earthquakes 28 and 29 are located nearestthe southernend of •he intermediate-depth zone,where the arc curveseastward. Although the B and P axesinter- changepositions from one solutionto •he o•her, both indicate extensionals•ress parallel •o •he s•rike of the zone,and may •hus indicale the type of deformation shownin Figure 6. Solution 27 is different from •he o•hers in •ha• •he T axis is perpendicular rather than parallel to the inclined zone. The solution indicates that the zone is under compressionparallel •o i•s s•rike. The depth of •he shock,249 km, is about 100 km greater than the depthsof the o•hers in the intermediate-depthgroups except17. Bo•h 27 and 17 are near the lower par• of •he intermediate-depthzone, i.e., near •he gap in seismicactivity. Although •he solutionsfor 24-29 differ remarkably, •he s•ressaxes remain closely grouped and oriented parallel or perpendicula.rto •he over-all planar geometry of the inclined zone. This result offers strong support •o the idea that the earthquakesresull from stressesin a thin plate. The causesof •he spatial variations of the stressorientations, however, are not known. Deep earthquakes. Solutions 30-31 for •wo earthquakes located in the nearly horizontal zone of deep 'earthquakesalso exhibit a remarkable variation in orientation. Although all of •he s•ress axes interchange positions from one solution•o •he other, the axes remain approximately parallel or perpendicularto the planar geometryof the zone.A plane throughthe more nearly horizontal axes 132 ISACKS AND MOLNAR (the dashedline in the projectionin Figure 9) actually dips abou• 10ø wes• •nd is consis•en•wi•h the distribution oœhypocen•ers shown in Figure 10. These resultssuppor• the concep•oœ a s•ressedplate and sugges•a nearly horizontal, possibly isolated, piece oœlithosphere. The causesoœ •he spatial variations in stress are no• known. The solution œorthe single deep even• (32) in the southernpar• of •he region is shownin Figure 9 togetherwi•h •he solutionœor 33, the westernmos•deep even• of •he Tonga zone.Earthquake 33 appearsto be locatedin a bel• oœdeep ear•h- quakes (34-38) •ha• trendswestward •rom •he northernend oœ•he Tonga deep zone,and its solutionis similar •o those (34-38) in •ha• zone (seeFigure 11). The orientation oœsolution 32 wi•h respec•to the southernend oœthe •NewHebrides arc is similar to tha• oœ33 (and 34-38) wi•h respec•to •he northern end oœthe Tonga arc. These relationshipsmay thus indicate a bend in •he lower southern edge oœa •New Hebrides deep zone similar to •ha• a• •he northern end oœ•he Tonga zone. In summary, the da•a œorthe northern par• oœ•he •New Hebrides arc clearly indicate down-dip extensiona• intermediate depths. In the southernend oœthe zone and a• grea• depths below a marked gap in seismicactivity, the orientation oœs•ress, although highly variable, is still parallel or perpendicularto •he planar geometryoœ the zone,as would be expectedœor s•resses in a •hin plate.

Tonga-Kermadec-New Zealand Region The nearly linear island-•rc system •ha• ex•ends•rom Samoa •hrough •he North Island oœNew Zealand can be divided in•o •hree distinc• segmentson the basisof seismicityand o•her •ec•onicœea•ures: •he Tonga arc, the Kermadec arc, •nd •he •Nor•h Island oœNew Zealand [Sykes, 1966; Hamilton and Gale, 1968]. Tonga a'rc (Figure 11). Figure 11 summarizesthe results of Isaclcs et al. [1969] plus eigh• new solutionsno• reported in •ha• paper. The con•oursin the figure summarize•he loc•tions oœSykes [1966] and Sykes et al. [1969] and show tha• the zone can be representedby an inclined plane tha• has contor4ionsalong its lower and lateral edges.Between depthsof 450 and 650 km the northern part of the zone bends sharply toward the northwest.At depths lessthan 450 km the northwesterly bending is probably present but considerablyattenuated. The distribution of seismicity is continuousat depths lessthan about 680 kin. Partly as a resul• of the contortionsalong the lower edge of the zone, hypocentral depths decreasesouthward to less than 600 km and then abruptly increase to greater •han 600 km near 26øS. The main consistencyamong •he solutionsof bo•h deep and intermediate- depth shocksis •he parallelism between •he P axes and •he local dip of •he seismiczone. This is shown clearly by solutions40-48 and 51-55 for the ear•h- quakes located away from •he northern and southern lateral boundariesof •he zone. The solutionsfor earthquakesnear •he lateral boundariesare morecomplex. The orientationsof solutions33-39 can be explained,however, by a simple geometriceffect. The s•ruc•ureof •he zonenear earthquakes33-39, •houghcom- pleg in detail, can be representedapproximately as a bend of •he lower northern STRESSES IN THE DESCENDING LITHOSPHERE 133

i i i I i I i i i

15 5(

% i 5 FIJI I ! I I ,-•, ... ..: •,• I I ,1:3:3"'•...' , i I 40 TONGA I TRENCH Fig. 11. Tongaisland are. 2o • I I i

.Jr_•

•49

I I ' 180 175 W cornerof the inclined zone (as in Figure 4). The plane shownby the dashed line in the equal-area,projection of 33-39 (Figure11) is approximatelyparallel to the groupingsof P and B axesand perpendicularto the groupingof T axes. This planeis a reasonablerepresentation of the part of the inclinedzone bent aroundto the west.Thus, the main differencesbetween the solutionsfor 33-39 and thosefor 40-48 in the unbentpart can be accountedfor by a rotation about the axis of bendingand not by stressescaused by the bending. Solution37, in the densecluster of hypoeentersthat includes36 and 38, is differentin respectto an interchangeof the B and T axes.This relative instability of the orientation of the B and T axes is also characteristic of the shocks in the main part of the are. However,39, locatednearest the bend and deeperthan the others,differs from the solutionsfor nearbyevents in that the B and P axeshave interchanged.This one solutionmay be related to the deformationnear the end. The solution for 33 is similar to others in the group 34-38, although the location of 33 is more than 400 km to the west. of 34-38. Several deep earth- quakesreliably locatedbetween 33 and 34-39 indicate that the zone is continuous to 33; this is shownby the 600-km contour in Figure 11. If this bend were straightenedout, the deep zone would extend considerablynorth of the edgeof the zone as definedby the distributionof hypocentersabove 450 km. It is interestingthat if the whole inclined zone were unbent and moved to the surface,the part of the zone correspondingto the 600-km contourwould then extend considerablynorth of the presently active transform fault located along the 15øSlatitude line. That part correspondingto the zone at depths less than 134 ISACKS AND MOLNAR 450 km would,however, nearly coincidewith the transformfault. Theserelation- ships,as well as •he discrepancybetween •he locationof the presentlyae•ive transform fault and the expectedlocation for this feature farther north along the borderof the Fiji plateau,can be explainedby a changein •he directionof relative motionbetween the Pacific and Australianplates from a west-northwesterly directionperpendicular to the Tonga trench •o a westerly direction parallel[o •he seismiczone along 15øS. The slipvectors inferred from mechanism solutionsof shallowearthquakes [Isacks et al., 1969]do not clearlydiscriminate betweenthese two directions.The appropriatelengt. h alongthe northernedge of the Tongaslab is about750 km, which,if dividedby LePichon's[1968] rate for the northernend of the are, yields a time of approximately8 m.y. ago for the changein direction.Instead of a changein direction,C. Chasesuggested to the authorsthat the hinge-transformfault systemthat terminatesthe northern endof the Tonga lares'migrated southward andthereby shortened thelateral dimensionof the deseelndingslab. Earthquake 50 is locatedat in[e}mediatedepth near the northern edge of the zone,where the westwardcurvature is presentbut not pronounced.Although the stressaxes of this solutiondo not appearto be closelyparallel or perpendicular to the seismiczone, their orientationswith respectto the northernend of the Tonga are are similarto thoseof 28 (Figure9) with respectto the southern end of the New Hebrides are. Solution49 is anomalousin that the T axis,rather than the P axis,appears to be parallelto the seismiczone (but not to its dip). This solutionmay indicate a localizedreorientation of stressnear the lower edgeof the zone. Solutions56-58 for intermediate-depthevents located nearest the junction betweenthe Kermadee and Tonga ares resembleothers in the region and indicate no profound effect of the junction, although the northerly deflectionof the P axes of 56 and 58 from the dip of the zone may indicate a minor effect. In summary,the contortionsof the seismiczone do not seemto have large effectson the focal-mechanismsolutions; instead, the stressaxes tend to follow the local structureas if embeddedin the zoneand suggesta compressionalstress transmittedthrough nearly all parts of the slab. This is contrary to what would be expectedif the mechanismswere related primarily to a localizedstress near a contortion of the slab. Kermadecarc (Figure12). The inclinedzone in thisregion dips more steeply than that of the Tonga are and doesnot appearto extendto depthsgreater than 500-550 km [Sykes, 1966]. A plane striking N 20øE and dipping 60ø west is a goodrepresentation of this zone. Solutions60-61 for earthquakesat depths near 230 km both have T axes parallel to the dip of the seismiczone, whereas 59, for an earthquakelocated north and 120 km deeperthan 60-61, showsdown-dip compression.Although the data are limited, theseresults suggest a changefrom down-dipextension to down-dip compressionsomewhere between 230 and 350 km. An unusualfeature is that the other stressaxes, although closely grouped, are not parallel to the strike and normal directionsof the seismiczone. North Island,New Zealand(Figure 12). Hamiltonand Gale [1968,1969] STRESSESIN THE DESCENDINGLITHOSPHERE 135 ix ,rl! / ;5s/ I

. // TRENCH

Fig. 12. Kermadec island arc and New Zealand. NEW /

I i i?o E 180 showedthat the seismicactivity at intermediatedepths beneath the North Islandof New Zealandhas a slablikestructure. Below depths of about350 km onlythree earthquakes are known[Adams, 1963]. These three, near a depth of 600kin, are locatedalmost vertically beneath the southernpart of the intermediate-depthzone. Near 62 the zonecan be approximatelyrepresented by a planestriking aboutN 45øEand dipping northwesL Both the T andthe P axesof 62 appear to parallelthe zone;the T axisis approximatelyparallel to the dip, andthe P axis is parallel to the strike. Adams [1963] reported a solution different from 62 for an event located in the southernpart of the intermediate-depthzone. Adams found that first motionsrecorded at) New Zealandstations from the threedeep shocks are gen- erallythe oppositein polarityof thosefrom the intermediate-depthshocks. This resultmay indicatea variationin mechanismswith depthsimilar to that in the Kermadecregion. The predominantlydilatational readings reported by Adams for the deepshocks are consistentwith a P axisoriented more nearly vertical thanhorizontal, i.e., a P axisparallel to a steeplydipping deep zone.

South America The disSribuQonof earthquakesand o•her island-arc-like tectonic features indicates•hat •he coas• region of western South America north of the Chile rise is a majorzone of convergencebetween an oceanicand a continentallithospheric pla•e.Mechanisms of shallowearthquakes [Isacks, 1970] indicate•hat between Ecuador and the Chile rise the oceanicNazca plate underthruststhe oontinen• of South America in an easS-northeasterlydirection. North of Ecuador an 136 ISACKS AND MOLNAR island-arc-like structure continuesthrough Colombia, but the plate tectonics of this region are complexand have yet to be fully worked out [Molnar and Sykes, 1969]. In the following,South America is consideredin two parts, one north and one southof Ecuador. South of the Chile rise the seismicitydecreases markedly and provides little evidence of the nature of the connectionbetween the South Sandwich arc and the main seismically active structuresof Chile. Region of the Peru-Chile trench (Figures 13 and 1•). Additional sources of informationon the distributionof earthquakesin this regionare Ocala [1966] and Santo [1969a]. A• intermediatedepths the seismiczone is slablikeand dips gently beneath the continent. The section in Figure 13 showsthis for the inclined zone beneathPeru; the slablike zone of intermediate-depthhypocen•ers, separated from a zone of shallow,probably crustal earthquakesthat lies above i•, has a dip of only 10ø to 15ø and reachesdepths of about 200 km. Santo [1969a] showsthat. the zone beneath northern Chile dips more steeply a• abou• 25-30 ø and reaches depths of abou• 300 to 350 km. Between latitudes of abou• 16øS and 25øS the zone of earthquakes with depths between abou• 70-150 km is extremely active; •his is illustrated by the large number of earthquakesin this region for which focal-mechanism solutions were obtained. Near the border betweenChile and Peru the seismiczone, topographicfeatures, and the line of active volcanoes all appear to curve around toward the northwest. Just north of this bend, near 16øS in southernPeru, the seismicactivity at intermediate depths decreasesmarkedly and the zone of historically active volcanoestermi- nates. Beneath depths of about 200 km in Peru and abou• 300 km in Chile the seismicactivity decreasesremarkably. Santo's plo• of •he frequencyof earth-

Fig. 13. The Peru-Chile trench and the adjacent region of western South Amer- ica. Coastline shown extends from Ecuador to central Chile. The northern deep zone (600-km contour) is 2O beneath westernBrazil, and the southerndeep zone (500- ß õ ß km contour) is beneath west- o o ! ern Argentina.

TRENCHCHILE

, :•90 80 W 70 , 60 STRESSES IN THE DESCENDING LITHOSPHERE 137

_AXIS OF PERU TRENCH At -200 0(B) +200 +400 7600 C I I i I I I 0 i i i i

øOoo% o o øø O0 0 Fig. 14. Vertical section o(P ø o through Peru including 200 oo ooo•O oQ••• events within the area shown KM ! in the inset. The events are those with magnitudes greater than about 5 reported in the Preliminary Determi- nation o• Epicenters [1961- 1967].

I I I I 80 W 75 70 quakesas a functionof depth showsa pronouncedgap betweenabout 350 and 500 km. This gapis especiallyprominent because the seismic-energyrelease a't great depths in this region is so high; Santo estimatesthat during the past 50 years it is larger in South America than that at great depths in any other single region. Santo tabulates only six small earthquakeslocated by the Coast and GeodeticSurvey (ESSA) with depthsbetween 300 and 500 km. These locations, however, are determined by few and poorly distributed data and could probably be moved either above or below the gap in seismicity.These six are all in the Chile-Argentina region; none have been reported for the gap in the P'eru-western Brazil zone. At depths of 500-650 km, beneath the gap in seismicactivity, the hypocenters are confined to •two' remarkably linear belts, one beneath western Argentina and one benBath western Brazil. These belts are parallel and trend slightly west of nor•l•, but the one beneath western Brazil is offset about 500 km west of that beneath western Argentina. Between these zonestwo large earthquakes occurred in 1958 and, 1963 (no. 91) at nearly the same location beneath the Peru-Bolivia border. These eventsare closestto the deep •arthquakes beneath western Brazil and may thus be on a Southeastwardextension of that zone. Recent studies of focal-mechanismsolutions reported by Stauder [1969] and Ritsema [1970] appear to 'be in goodagreement with our analysis, although Ritsema placesa different interpretation on the results. The zone of intense intermediate-depth activity beneath northern Chile and southern'Peru, including earthquakes70-84, is representedon the equal-area plot in Figure 13 by two planes. One has the N 7øE strike of the linear coast of Chile, and the secondhas the northwest strike of southeastPeru;-both have a dip of 28ø. The orientation of the zone in the bend between Chile and. Peru probably varies between these two extremes.Although the stress axes scatter somewhat,partially as a result of the inclusionof solutionsof varying quality (see appendix),the T and P axesof 70-84 showa distinct grouping.The plunges 'of the T axes vary between 7ø and 54ø but have an average of 31ø, which is 138 ISACKS AND MOLNAR very closeto Santo'sestimate of 28ø for the dip of the zone.The trends of the T and B axes, although somewhatscattered, do not vary systematicallywith respec••o •he locationof the event and •hus do not follow the major bend of the coastline and other tectonic features in the region of the Peru-Chile border. Instead, the axes trend approximatelyparallel or perpendicularto a direction between the strikes of the coastlines of southern Peru and northern Chile. The east-northeasterlytrend of the T axes is approximatelyparallel to the direction of underthrustinginferred from the solutionsof shallow earthquakes [Isacks, 1970] and approximatelyperpendicular to the alignment of active volcanoesin northern Chile. These results are in good agreement with down-dip extension in the zone but do not reveal any marked effect of the bend in the coastline between Chile and Peru. Although less closelygrouped than 70-84, the T axes of 64-69 also tend to parallel the dip of the intermediate-depthzone beneathPeru. From the epicentral locationsalone it is not clear whether earthquakes63 and 64 (beneath Ecuador) are at the northern end of the Peruvian zone, at the southernend of the northeastward-trendingzone i• Colombia,or in a bendbetween two zones.The simi- Iarity between 64 and those for other events in Peru suggeststhat this earthquake may be in the northernpart of a coherentslab beneathPeru. The solution for 63 is quite different in orientation.This event could be related to deformation near a junction between the inclined zonesof Peru and Colombia. The solutionfor 67 is also different from others beneathPeru, althoughthe earthquake is located in the central part of Peru. Nevertheless,the B and T axes,although not parallel to the dip and strike directions,remain approximately parallel •o the inclined seismiczone. Included in this analysis are mechanismsof several earthquakes (68, 69, 72, 79, 82, 84) whosedepths vary ñ20 km around the 70-km borderline between intermediate and shallow depths. In some casesthe depths are controlled •airly well by pP readings,but in other casesthey are uncertain. In all cases,however, some evidenceo• • depth greater than normal was •ound. The additional basis for inclusion in the analysis is the similarity between the solutions •or these earthquakesand those for the well-located events at greater depths.Because of the gentle dip of the seismiczones, these solutionshave a distinctly different orientation •rom that characteristic o• •he underthrus• •ype o• shallow earthquakesin the region. Solutions 85-90 for the deep earthquakes located beneath western Brazil are nearly identical to one another.The P axes are all nearly vertical, whereas the B and P axes are approximatelyparallel and perpendicularto the linear trend of the deepseismic zone. A plane •hroughthe B and P axesis more nearly parallel •o the N 10øW •rend of the belt o• deepearthquakes than •o the northwest trend of the Peru trench and coastline.This relationshipsuggests that the plane throughthe B and P axesrepresents the local structureof •he deep zone and is thus consistentwith an inferenceof down-dipcompression within a nearly vertical zone.The locationof the deep eventsnearly vertically beneaththe lower par• of the intermediate-depthzone further supportsthis interpretation. • Solutions92-97 •or deep earthquakesbeneath western Argentina h•ve B •nd STRESSES IN THE DESCENDING LITHOSPHERE 139 T axesthat are alsomore nearly parallel to the belt of deephypocenters than to the Chile trench and coastline,but the differencein this caseis small. A plane throughthe B and P axesdips about 70ø east and is approximatelyparallel to a plane joining the deep and intermediate-depthearthquakes through the gap in seismicity.These resultsare in agreementwith an inferenceof down-dip compressionwithin a steeplydipping deep slab. Solution 91 for the large multiple event beneath the Peru-Bolivia border [Chandra, 1970b] appearsto indicate down-dipcompression in a steeplydipping zone. The B and T axes, however, have significantly different trends than those in the deep zonesbeneath either western Brazil or western Argentina. The sim- plest interpretation is that the seismiczone near 91 strikes northwest,and the T rather than the B axis is parallel to the zone. A straightforward connection between91 and the zone beneath western Brazil would require this strike. The parallelism of the T axis and the inferred strike of the zone may be related to a stretchingor someother contortionin this region. In summary, the most outstandingresult for the Peru-Chile region is the changefrom down-dip extensionat intermediate depths to down-dip compressionat great depths. Colombia: the Bucaramanga nest (Figure 15). The island-arc-like features of the Peru-Chile region bend around in Ecuador and trend east-northeast in westernColombia. Earthquakes 98-100 are locatedin the zoneof intenseactivity at depthsnear 150 km beneathBucaramanga. This zonemay be at the northern end of a continuous inclined zone that trends parallel to the island-arc-like featuresof Ecuador and Colombia.Molnar and Sykes [1969], for example,show

Fig. 15. Region of northwestern South America showing the zone of intermediate-depth earthquakes in Ecuador and Colombia. Earthquakes 98-100 are located in the Bucaramanga nest. The lines of alternating long and short dashes represent major mountain ranges and structural trends. 140 ISACKS AND MOLNAR a zonedipping eastwardbeneath the coas•of Colombianear 5øS. Althoughlarge numbersof earthquakesare very concentratedspatially near earthquakes98-100 [Trygsvasona•d Lawson, 1970], shallowerbut subcrustalearthquakes are reliably located northwest to north-northeast of the Bucaramanga region, and deeper earthquakes (depths near 200 km) are located slightly southeastof the region [Sykesand Ewiag, 19•5]. A vertical sectionaligned northwest-southeast through the Bucaramanga 'nest' [Santo, 1969a] also shows a zone dipping southeast. These locations sugges•tha• near Bucaramanga the zone may bend around to a more easterly strike than in Ecuador and southern Colombia. Solutions98 and 100 are nearly the same; the T axes could be parallel to a zone dipping southeast.The third solution (99), however, is not in agreemen• with tha• interpretation. Alternatively, a plane perpendicularto the closegroup- ing of northward-plungingaxes (representedby the dashedline in Figure 15) is parallel to all the other axes.The orientation of •his plane is consistentwith the other evidence cited above for a zone near Bucaramanga tha• dips more •oward the sou•h than toward the eas•. If this is the case,then the data indicate compressionals•ress oriented more nearly parallel to the strike •han to the dip of the zone. Moreover, the approximate interchangeof •he positionsof •he B and T axes of 99 wi•h respec•to 98 and 100 suggeststhai; compressionals•ress is dominant. One interpretation of •hese relationshipsis tha• the lateral com- pressionresults from a bending of the northern end of the slab.

Atlantic Island Arcs South Sandwicharc (Figure I6). The seismicityof •his region is poorly known, partly becauseof •he lack of nearby seismographs•ations. As in other arcs, most of the shallow earthquakes appear to occur beneath the landward wall of the trench; and in general,the intermediate-depthevents are located west of the main zone of shallow activity. Earthquakes appear to be restricted to

I I .I. I

EORGIA 55- S

":) - Fig.wich16. islandSouth arc.Sand-

60- /

",....,•x •1o5 I I I ! 50 W 20 STRESSES IN THE DESCENDING LITHOSPHERE 141 depthsless than about 200 km. At the northernpart of the are, where the features bend toward the wes[, an intense concentrationof hypoeentersis located at a depth of about 100 km (near earthquakes 101-103). A vertical section through this part of the are, oriented about N 60øE, shows an inclined zone dipping very steeply southwest[Engdahl et al., 1969, and E. R. Engdahl, per- sonal communication,1970]. The data are no[ adequate to define the dip of the zone farther south along the are. Solution 104 indicates down-dip extension.Solutions 101-103 for the earthquakes located doses[ to the westward bend also indicate extensional stress parallel to the zone, but the axes are not parallel to the dip. This could be an effect of the westwardbend, i.e., an effect of a contortionat the northernend of the zone. In contrast, solution 105 for an earthquake located in the middle of the are exhibits simple down-dip compressionand is nearly the opposite of 104. Solution 105, plus two others (148-149) for earthquakesin the Ryukyu are, are the only easesin which down-dip compressionwith the B axis parallel to the strike of the zone occurs in a seismic zone that does not reach depths greater than 300 km. West Indian arc (Figure 17). The distributionof hypoeenters[Sykes and Ewing, 1965] and the focal mechanismsof shallow earthquakes [Molnar and Sykes, 1969] demonstratean eastward rigid-body motion of the Caribbean with respectto North America, South America, and the westernAtlantic. In the Lesser Antilles the Atlantic sea floor underthrusts the Caribbean toward the west. Near the Puerto Rico trench, the Atlantic sea floor is also underthrustingthe Caribbean, but the motion is nearly parallel to the strike of the trench. The seismicity of the Lesser Antilles portion of the are clearly shows a westward- dippingzone of aet.ivity,but at the northernend of the are the activity changes orientation abruptly and trends east-west, parallel to the Puerto Rico trench. This east-wes[belt has a more complexstructure. No earthquakesin either portion of the are have been located deeperthan 250 km. An intensesource of intermediate-depthearthquakes is located beneath the easterntip of Hispaniola.Although the distributionof hypoeentersis not simple, Sykesand Ewing'ssection suggests a nearly vertical distributionof hypoeenters near that source.Molnar and Sykesreport solutions for three intermediate-depth

I I I I .i'.'•_•=½HISPANIOLA ......

Fig. 17. West Indies island are. I \ "-t ß '"•' ..lk•"-'•, '• •'

70 W •0 142 ISACKS AND MOLNAR earthquakesin this region,but only one (106) is well determined.A plane throughthe B and T axesoœ 106 has a strike that is Slightlynorth o• the westerly trend o• the Puerto Rico trench but is approximatelyparallel to the belt o• seismicactivity alongthe northerncoast of Hispaniola.The vertical dip of this plane is consistentwith the distributionof hypocenters.A nearly vertical slab underextension approximately parallel •o its dip may thus be locatedbeneath the easternend of Hispaniola.It is interestingthat both the locationsand the solutionsof 106 and 101-103 (Figure 16) have approximatelysimilar relation- shipsto the West Indian and the SouthSandwich arcs, respectively. Molnar and Sykes'two other solutions,though not well determined,are different from one another and from that of 106. This variability plus the com- plexity in the distributionof hypocentersin the regionmay alsoindicate a contortion or disruption of the slab. Sykesand Ewing reportedevents with reliably determineddepths between 70 and 120 km located at the southern end of the Antilles arc near Trinidad. Molnar and Sykesfound evidence that someof the eventsin this regionmay be relatedto hinge faulting,whereby the portionof the westernAtlantic-South Americanplate that underthrustsand descendsbeneath the Antillesarc tears awayfrom the portionremaining on the surface.It is not clear,however, whether the intermediate-depthevents are involvedin this type of faulting.The hinge faultingat the northernend of the Tonga arc occursat shallowdepths [Isacks et aI., 1969].

Middle America Arc (Figure 18) Molnar and Sykes [1969] showthat the oceaniclithosphere is underthrusting Mexicoand CentralAmerica in a northeastdirection. Between about 86øW and 96øW the structure of the arc is particularly simple. The surface features are nearlystraight, and the slablikezone of subcrustalearthquakes is relatively planar and free of major contortions.This regionprovides a particularlygood

, I I I

Fig. 18. Middle America trench and adjacent region of Central America. STRESSES IN THE DESCENDING LITHOSPHERE 143 tessof the idea that, as a result of gravitational body forceson excessmass in the sinkinglithosphere, extensional stress predominates in the slabstha5 are relativelyuncontorted and that do not reachgreat depths(Figure 3a). The plane that fits the closegroupings of T and B axes of 107-110 strikes N 40øW and dips about 60øNE. The 20-25 ø difference between the trends of the B axesand the strike of the trench and seismiczone appears to be significant. The T axis of 111 has a plungeof about 40ø. The differencebeSween the plunges of 111 and 107-110 is determinedby numerousreadings in the central parts of the focal spheresand is probably real. Earthquake 111, at 85-kin depth, is shallowerthan the other four. These resultsare consistentwi•h down-dipextension in a zone tha5 increasesin dip from about 40ø to 60ø a5 depths greater •han 85 km. Although sucha changecannot be resolvedby the data of Molnar and Sykes,their results do not excludeit. The data are thus in agreementwith the model shownin Figure 3a, where the slab is sinking bus has not descended beneaShthe asthenosphere.

Cascades In the northwesternUnited Statesan island-arc-likestructure trends parallel to the coastsof Oregonand Washingtonand is definedapproximately by the volcanoesof the Cascaderange. Subcrustalearthquakes may occurin this region [Tobin an.d Sykes, 1968], but the data are inadequateto define an inclined seismiczone. Several studiesof the motionsof the lithosphericplates in this region indicate that liShospherehas been consumedin the pass and that the downgoingslab will dip east [Morgan, 1968; McKenzie and Morgan, 1969; Silver, 1969]. Althoughthe depth of earthquake112 (59 km) [Algerm.issena•d Harding, 1965] is shallowerthan other eventsin this study, it is significantly deeperthan othersin the westernUnited States. •/Ioreover, its solutionhas an eastward-dippingT axis and thereforecould indicate extensionalstress parallel to an eastward-dippingzone.

Aleutian Arc (Figure 19) The underthrustingof the Pacific plate northwestwardbeneath the Aleutians andAlaska is describedby McKenzieand Parker [1967] and Stauder[1968a, b]. Focal depths are generally less than 200 kin. A5 the western end of the zone of active volcanismand subcrustalseismicity (near 178øE) the dip of the zone, althoughnot too well determined,appears to be abou560 ø [Davies and Mc- Kenzie, 1969; Murdock, 1969]. A plane through the P and T axes of 113 is approximatelyparallel to the seismiczone, and the solutionindicates compression parallel to the dip or extensionparallel to the strike of 5he zone. Sincethis earthquakeis locatedwhere the curvatureof the arc increases,Stauder [1968b] suggestedthat the extensionalstress parallel to the strike resultsfrom the type of deformationof the descendingslab shown in Figure6. Solutionsof this type are found in other regionswhere convexcurvature is pronounced. Solution 204 is similar to those for shallow events in the region and is interpretedby Stauder [1968b] as underthrustingon a nearly horizontal fault plane. A15houghthe depShof 60 km for 204 is no5 qui•e intermediate,its loca- 144 ISACKS AND MOLNAR

55 Fig. 19. Aleutian island arc. N

5O I I I 1'75'E 180 1'75W 1'70 160 tion is well determined;and the event is probablylocated in the part of the seismiczone that has a dip considerablygreater than that requiredfor the underthrustinginterpretation. The solutionis thereforemore consistent with an inferenceof down-dipextension inside a northward-dippingslab. The dip of the zoneinferred from the plungeof the T axisis about40 ø. Down-dipextension is alsoa reasonableinterpretation of the solutionof the January1, 1963,earthquake [Stauder,1968b] locatedeast of 204. The intermediatedepth of 75 km for this eventis determinedby numerouspP readingsand is preferableto the depthof 51 km listedby the InternationalSeismological Summary. Thus, down- dip extensionmay be prevalentin the part of the descendinglithosphere east of earthquake 113 that has a more nearly linear strike.

Northwest Pacific' Kamchatka-Japan-Marianas A nearly continuouszone of deepand intermediate-depthearthquakes can be tracedfrom Kamchatkathrough Japan and thenceto the Marianas.This vast descendingedge of the northwestPacific lithospherecan be divided into four segments,each characterizedby distincttrends of the oceanictrench, of the line of activevolcanoes, and of the inclinedzones of earthquakes.The Kurile- Kamchatka,North Ito•shu, and Izu-Bonin segments,trending northeastward, north-northeastward,and north-northwestward,respectively, form two oblique- angledcorners near Itokkaido and east centralHonshu. The fourth segment,the Marianas arc, continuesthe trend of the Izu-Bonin arc but is offsetto the east and is separatedfrom the Izu-Bonin segmentby a gap in seismicityand by a shoalingof the oceanictrench. The dips of the inclined seismiczones differ notably amongthe four segments.The structureof the zonebetween the segments,whether continuous(and bent) or disrupted,cannot be clearly determined by the available data on locationsalone. Additional sourcesused in determiningthe contoursshown in Figures20, 21, and 22 includeFedotov [1965, 1968] and Sykes [1966] for the Kurile-Kamchatkaregion; Katsumata [1966, 1967a,b] and Ichikawa [1961] for North Itonshu and the Sea of Japan; and Katsumataand Sykes [1969] for the Izu-Bonin and Marianas regions. STRESSES IN THE DESCENDING LITHOSPHERE 145 As a resultof the pioneeringresearch of It. Honda and his colleagues,reliable solutions of many earthquakes that occurred during the last 40 years are available for the Japaneseregion. Although only solutionsbased on the most numerousand consistentdata are listed in Table 1, other data, summarizedby Balalcina [1962] for the Kurile-Kamchatka region and by Honda et al. [1956, 1967], Ichilcawa [1966], and Alci [1966] for the Japaneseregion, are in sub- stantial agreement with the results selectedfor Table 1. The Kurile-Kamchatka arc (Figure 20). Between Kamchatka and the Hokkaido-Sakhalin region the inclined seismic zone appears to be relatively simple in structureand can be representedby a plane with a dip of 45ø and a strike varying from about N 35øE near Kamchatka to N 60øE near Hokkaido. Nearly all the large, well-locateddeep events northeast of the Itokkaido-Sakhalin region have depths between400 and 500 km. Examination of reported locations suggeststhat the maximum depth of this zone is probably lessthan 600 km and may be near 500 km. A minimum in activity as a function of depth occurs betweenabout 200 and 400 km, but apparently reliable locationsof events with depthsin this range (e.g., Savarenslcyand Golubeva [1969]) indicate that this minimum is not an aseismicgap in activity. In nearly all the solutionsfor this region either the P or the T axes tend to

.o x

128-154 ! ! 5O N ,/

! 118

'4-

I I I I --]-- o I x o i / 125-127

140 E 150 16 0 Fig. 20. Kurile island arc and the region near Hokkaido, North Honshu, and the Sea of Japan. The discontinuity in the contours of hypocentral depths representsa possible disruption between the Kurile and North Honshu seismic zones. 146 ISACKS AND MOLNAR beparallel to thedip of the zone,while the B axesare approximately parallel to i•s strike.However, the pa•ternof down-dips•ress in the regionis complexand not jus• a œunctionoœ depth. Solutions 123-127 œor events near the junction between•he Kurile and North Honshu zones all indicatedown-dip extension, as do 119-121for threeintermediate-depth events located south oœ Kamchatka. In contrast,solutions 117-118 œorintermediate-depth events located north and southoœ 119-121 and •wo solutions (114-115) œor deep events all showdown-dip compression. Solution116 for a deepearthquake has • differen•orientation from the othersin the regionand doesnot appear•o be closelyparallel to the over-all planargeometry oœ the zone.However, the parameteroœ the solutionthat is mostnearly parallel to •he dip oœthe zoneis s•ill •he P axis. The peculiar location oœ122 a• 80 km beneaththe axis o• the Kurile trench doesnot appear to be an errorin •helocation. Shallow events characterized by solutionsoœ the normalœaulfing type are foundnear the axisoœ trenches in other arcs,notably the Aleutians[Startder, 1968b], and canbe interpretedas exten- sionnear the upperpar• oœthe oceanicplate as it bendsbeneath the islandare. By the samereasoning, compression would be expectednear the lowerpart oœ the plate.Although the •rendoœ the P axisoœ 204 is not perpendicularto the axisof thetrench, the P axisis morenearly horizontal than the T axisand could indicatecompression oœthe lowerpar• oœthe plateas it bendsdownward near the trench. The Hotet•aidocor•er (Figure 20). Solutions123-129 for the eventslocated in •heregion between the gently dipping North Honshu zone and the more steeply dippingKurile zoneare oœtwo types.Solutions 123-127 for eventslocated on theKurile side oœ the cornerall indicatedown-dip extension and are thus similar to 119-121 œorevents in the northern Kuriles. Solutions 128 and 129 •or events located on the North I•Ionshuside oœthe corner are similar to 130-134 and are thusapparently related to the down-dipcompression in the North Honshuzone. The solutions,thereœore, appear to be relatedmore to patternson eitherside of •he cornerthan to any specialeffect of the corneritself. Hondaet al. [1956]and lchit•awa []966] repor•o•her solutions (no• included in Table1) •or earthquakesnear •he corner, some oœ which can be interpreted to suppor•the lateral s•retchingor tearingoœ • slab as shownin Figure5. For example,•hree oœtheir solutionsœor earthquakes near 123 have T axesthat plungebetween the dip directionand the s•rike direction oœ the zone;these solu- tionscould be taken to indicatethe obliques•retching oœ a slab as shownin Figure5a. Also,one of •he solutionshas a verticalnodal plane striking north- northwes•that, iœtaken as the œaultplane, indicates tha• the Kurile slabmoves downwith respe½• to the North Honshu slab. This is in agreementwith •he hinge faultingshown in Figure5b, and,iœ correct, i• couldindicate a disruptionoœ the zone. However, examinationoœ the solutionsoœ Honda et al. and Ichikawa in these casessuggests to us tha• the differencesbetween their solutionsand thoseœor 123-127 may no• be real. In fact, it seemspossible to fi• mos•oœ the datagiven by Hond•e• al. andIchikawa, with minor modifications, to the solutions œor123-127. STRESSES IN THE DESCENDING LITHOSPHERE 147 The distribution of earthquakesplus the remarkable difference between solutions127 and 128 suggestthat the main junction betweenthe North Itonshu and the Kurile zonesis located beneath western Itokkaido (between 127 and 128) and beneaththe northern Sea.of Japan east of 129. Whether the zone is con- tinuousor disruptedin this regionis not dear from the distributionof hypoeenters. The reorientationof stressbetween 127 and 128, however,may be easierto understandif the eventsresult from stressesin separateslabs. Although 125 and 126 are near the zone of shallow underthrusting,their depthsnear 80 km are reliably determined,and their solutionsare quite similar to those for the deeper events. Solutions 125 and 126 thus appear to indicate stresswithin the descendingplate rather than slip betweenplates. Moreover,the directionsof slip inferred from the solutions,were the earthquakesshallower, indicatemotion of the Pacificplate relative to the Asian plate that is significantly more northerly than the northwesterly direction indicated by McKenzie and Parker [1967] ; Isacks et al. [1968]; Le Pichon [1968]; and Stauder [1968b]. Inclusion of such events probably accountsfor the diserepan[ directions near Hokkaido shownin the worldwideplot of slip directions(Figure 3 of Isacks et al. [1968]). North Honshu (Figures 20 and 21). The deep •nd in•ermediMe-depth earthquakes•ha[ occur beneath [he Sea of Japan are par[ of an inclined zone rela[ed to •he ac[ive volcanoesand offshoreIrerich thai parallel the nor•heas[ coast;of Honshu. Ichikawa [1961] and Katsumata [1967a] show •hat; •he zone has a dip of abou[ 30ø. Allhough seismicac[ivi[y betweendep[h• of 200 and 500

128-134

c) / o

40 i N i l i

x Fig. 21. Izu-Bonin island o are and the region of Honshu o and the Sea of Japan. The area near earthquakes 135- 137 is shown in more detail in Figure 22. 50 ß

•0 144-

130E 140 150 148 ISACKS AND MOLNAR kmis sparse, the zone appears to becontinuous. It does not appear to reachdepths much greater than about 600 km. Nearly all of the P axesof 128-134are approximatelyparallel to the dip of the NorthItonshu zone. Three solutions show an interchangeof B andT axes with respectto the others.The consistencyamong the solutionsis fi•rtherevi- dencefor the coherencyof a singleinclined zone dipping beneath the Sea of Japan.Other solutions reported by Hondaet al. and Ichikawaagree with the resultsshown in Figures20-21. In addition,their results suggest that down-dip compressionmay also be presentat depthsof 100-200km beneaththe northern coastof Itonshu,but the dataare limited.Thus the NorthItonshu slab may be undercompression at all depthsbelow 100 km. Thejunction between the Izu-Bonin and the North Honshu segments (Fig- ures21 and 22). Both the distributionof hypocentersand the focal-mechanism solutionssuggest that the junctionbetween the Izu-Bonin and the North Itonshu zonesis. located near 135-137 (Figures 21-22). The distributionof earthquakes shallowerthan 200 km, the oceanictrench, and the volcanicfront [Sugimura, 1965] form a well-definedcorner in the regionof Tokyo and centralItonshu. Subcrustalactivity is especiallyintense at depthsof 50-100km near 135 and at depthsof 200-300 km near 136. Earthquake 137, located at a depthof 360km, appearsto beat thenorthern end of theIzu-Bonin zone; to thenorth the activity decreasesand appearsto be part of the gentlydipping North ttonshuzone. As shownin Figure21 anddescribed in the next section, solutions for deepearth- quakesin theIzu-Bonin zone located south of 136and 137 are characterized by down-dipcompression and are similar in thisrespect to thosein the Northttonshu arc.Solutions 135-137 for the eventslocated between the two zones, however, havea distinctlydifferent orientation and thereforeprobably reflect localized effectsnear the junction. The solutionsfor 135-137are in eachcase representative of solutionsfor othershocks inthe vicinity of each hypocenter. Honda. etal. and Ichikawa report twoother solutions that agree with 137 and five solutions that agree (or for which

Fig. 22. The corner between the Izu-Bonin and the North tIonshu zones. The dotted lines in[ the

55 projectionof solution136 represent N the nodal planes of the solution. U (up) and D (down) show the sense of motion on the more nearly vertical nodal plane taken as the fault plane. The solution is

30 consistentwith hingefaulting. STRESSES IN THE DESCENDING LITHOSPHERE 149 the data are not inconsistent)with 136. Solution 135 is an average of a group of solutions reported by Ichikawa [1966]. The individual solutions combined in 135 are quite similar to one another and agree with independentresults of Aki [1966] for small earthquakeslocated in the region. Solution 136 is consistentwith the hinge faulting depictedin Figure 5b; in this casethe steeply dipping Izu-Bonin slab tears away and moves downward with respectto the moregently dippingNorth Honshuslab. This type of deformation, however, doesnot account for the orientation of 135 or 137. If the tearing and hinge faulting are located in the zone of intensesubcrustal seismic activity near earthquake 136, then at greater depths the two slabs will be separated. Hence, earthquake 137 may be in the northern edge of a separate Izu-Bonin slab. The differencebetween 137 and 138-146 may be attributable to a rotation of the axes due to a bend at the edge similar to that at the northern end of the Tonga arc (Figures 4 and 11). Further study of the detailed distribution of hypocentersin the region beneath central Honshu is required to test this interpretation. The earthquakescomprising 135 may result from stressesin the portion of the slab abovethe tearingand disruptionnear 136.The orientationof 135, however,is most simply interpreted as a faulting parallel to the inclined seismic zone, i.e., thrust faulting between the convergingPacific and Asian plates. if this is correct,then the earthquakescomprising 135 would be an anomalously deep extensionof the shallowthrust zone characteristicof the inner margins of the trenches.Most of the earthquakescomprising 135 have depthsgreater than 60 km, and many have depthsbetween 80 and 100 km. In most casesthese depths are accurately determined by numerous arrival times from stations within a degreeof the epicenter. Izu-Bonin arc (Figure 2I). Katsumata and Sykes [1969] show that the inclined seismiczone south of Honshu, though slablike in structure,varies as a function of depth and location along the strike. The zone appears to be continuousto. depths of about 550 km. In the separateequal-area plots of Figure 21, the data are shown relative to the local orientation of the zone. The P axes for the deep earthquakes138-147 all appear to parallel the local dip of the seismic zones;i.e., the P axestend to follow the local variations in dip. The T axestend to be perpendicularto the zone except in two casesin which the B and the T axes interchange. The data are sparsefor intermediatedepths. Ichikawa reports two solutions for earthquakesat depthsof 150 and 130 km locatedsouthwest of Tokyo, just southof the junction describedin the precedingsection; both indicate down-dip compression.Solution 148 for an even[ at 80-km depth, however,indicates compressionparallel to the strike of the zone.Mechanisms of this type are rare; the only other clear exampleis solution27 for an intermediate-deptheven[ located at the southernend of the New Hebrides arc (Figure 9). Marianas arc (Figure 23). Katsumata and Sykes [1969] show that the inclined zone beneath the northern par[ of the Marianas arc has a remarkable structure.From shallowdepths near the trench the slablike zone curvesdown- wardand below 250 km hasa nearlyvertical dip. The zoneis continuousto a 150 ISACKS AND MOLNAR

2C N Fig. 23. Marianas island arc.

,5•x •/' , IR..... I0 140E 150

depth of 680 kin. Most of the hypocentersbelow 500 km are confinedto. s narrow range along strike betweenabout 19ø and 20øN; few 'deepevents occur between this zone and the southern end of the Izu-Bonin zone near 27øN. There is evidence [Gutenbergand Richter, 1954], however,for the location of earthquakes at depthsof 500-600 km near about 24øN; a short solid segmentof the 500-kin contour in Figure 23 indicates the location of these events. A continuous and nearly straight 500-km contour that trends north-northwest and connectsthe Izu-Bonin and Marianas deep zonespasses through the deep hypocentersnear 24øN. The distributionof the deepestevents does not thereforeappear to reflect the offsetthat is prominentin the surfacefeatures of the Marianas arc. The P axis of 150 plungesvertically and is thus parallel to the local dip of the Marianas deepzone. The B and T axesare horizontaland trend parallel and perpendicularto the north-northwesterlytrend inferred above for the regionaldistribution of deephypocenters. This relationshipis evidencethat the deep seismic zone is continuousbetween the Izu-Bonin and the Marianas arcs. Solution149 for an event at intermediatedepth indicatesextensional stress approximatelyparallel to the strike of the zone. Note that althoughthe P and the B axesare not parallelto the zone,neither are the nodalplanes. The solution is similar to others for earthquakesin regionswhere the surface features are curved, and can be explainedby the lateral stretchingeffect shownin Figure 6.

Ryukyu Arc (Figure 24) Katsumata and Sykes [1969] showthat an inclinedzone dips beneaththe Ryukyu Islandsat an angleof about45 ø and reachesdepths less than 300 kin. Their solutionsfor shallowearthquakes indicate underthrusting oœ the Philippine Seaplate beneaththe Ryukyus.In Taiwanthe seismicitychanges trend abruptly and trendssouth along the easterncoast. Subcrustal earthquakes are also found beneathTaiwan and may be part of an inclinedzone that dipssteeply eastw•ard. Earthquakes151-152 are locatedaway from eitherend of the arc. The solutions are similar to one anotherand indicate down-dipcompression. In our previouspaper [Isacksand Molnar, 1969] we suggestedthat thesesolutions are similar to 113 in the Aleutiansin respec••o indicatingextensional stress STRESSES IN THE DESCENDING LITHOSPHERE 151

Fig. 24. The Ryukyus island arc and the region near Taiwan.

parallelto •he s•rike.However, as shownin Figure23, •he T axesfor the Ryukyu solutionsare morenearly perpendicularto. the zoneand the B axesmore nearly parallel to i•s strike. Thus, we canno5in•erpre5 5hese solutions by 5he lateral sSretchingeffec• shown in Figure 6. Earthquakes153-155 are locatednear the intersectionof zonesin Taiwan. Two of thesehave depthsof 67 and 70 km and so are near the borderlinebetween shallowand in•ermediaSedepths; the third (153) has a depth of 119 kin. The T axesof all •hree plungesteeply northward, and the B and P axestend to be nearly horizontal.The B and P axesof 154 and 155 approximatelyinterchange positions. The daSafor the three solutionscan thus be 5aken •o sugges5exSension parallel to •he dip of a nearly vertical and possiblycontorted slab near the Taiwan comer. Earthquake 156 is locatedat a depth of 72 km in •he easSward-dippingzone sou•h of the Taiwan intersection. The s[rucSureof the zone is no5 well enough known to interpre• this solution.

Philippines-CeIebes-Halmahera(Figure 25) The •ectonicsof •his region are especially complexand are no5 well under- s•ood.Fitch [1970a] summarizesevidence for a subdivisionof 5he region into three disfinc5 island-arc-like s•ruc•ures: •he Luzon arc, •he Mindanao arc, and •he Talaud-ttalmahera arc. Although all three have zones of subcrustal eaFch- quakes, only •he Mindanao arc has a, fairly well-defined inclined zone tha5 reaches grea5 depths. The solutions for subcrustal earthquakes are also moss simply groupedaccording to 5hesedivisions. Luzon arc. A northward-strikingStench [Hayes and Ludwig, 1967; Ludwig et al., 1967] and the presenceof volcanoesand shallow and in•ermedia.•e-depSh earthquakesindicaSe 5ha5 an island-arc structure exists on •he wesSernside of 152 ISACK$ AND MOLNAR

,x/5B-161

Fig. 25. Region of deep and intermediate-depthearthquakes near the Philippines,I-Ialmahera, and Celebesislands. Talaud Island is locatedbetween the dashedrectangles that encloseearthquakes 158-161 and 163-166.

Luzon.An inclinedseismic zone appears to dip east,but the data are inadequate to determineits dip.The onlysolution (157) for this zonecan be interpretedas extensionparallel to a steeplydipping zone or thrust faultingalong a zone dippingabout 45 ø. Because of thedepth of theevent (166 km), the formerinter- pretationis moreconsistent with the data for otherregions. Mindanao arc. The inclined seismic zone associated with the Mindanao trenchdips westward and reachesdepths greater than 650 km (M. Katsumata, personalcommunication). A gap in activityappears to occurbetween depths of about 300 and 500 km. Shallow earthquakemechanisms in the region indicate a westwardunderthrusting of the PhilippineSea plate beneathMindanao [Fitch, 1970a].In the southernpart of the structurethe zonesof earthquakesand active volcanoesappear to bendwestward and parallel the east-west-trendingpeninsula of northernCelebes. Fitch a.n.dMolnar [1970] showthat the zonemay dip northward near Celebes. Earthquakes158-161 are locatedin the westward-dippingzone, and 167-168 are located where the arc bendsand has a more northerly dip. Earthquake 163, thoughlocated close to 160 and 161,appears to be part of the eastward-dipping zone beneathTalaud and Halmahera; i• is consideredin the next section.The STRESSES IN THE DESCENDING LITHOSPHERE 153 T axes of 158-161 all dip westwardand are consistentwith down-dip extensionalstress. Note that 161is interpretedby Fitch as indicativeof motionbetween plates,i.e., underthrustingof the PhilippineSea plate beneaththe CelebesSea plate.The depthreported in the PrelimiaaryDeterminatioa of Epicentersof the Coastand GeodeticSurvey [1963] is 64 kin, but it is revisedto 83 km in the SeismologicalBulletin of the Coastand GeodeticSurvey [1963] and 122 km in the InternatioaalSeismological Summary [1963]. The data controllingdepth for this eventare poor,but examinationof the seismogramssuggests that the depth is greaterthan normalbut probablynot as greatas that listedby the ISS. If the solutionsfor thrust faults in the shallowportion of the seismiczone have the same orientationas onesindicating extensional stress parallel to the zone at intermediatedepths, the seismiczone must decrease in dip near the thrust zone. Althoughsuch a changein dip is difficultto resolvein the distributionof hypocenters,Mitronovas et al. [1969] demonstrateit for the Tonga arc. The solutionof 162 indicatescompression parallel to the dip and extension parallelto the strike of the zone.This result may thus indicatean effectof the convexcurvature of the arc near Celebes (Figure 6). The solutionsfor 167 and 168 for events located where the arc bends westward are consistent with extensionalstress parallel to a zonedipping steeply northwestward. In summary,solutions 158-161 and 167-168 are consistentwith an inference of down-dip extensionat intermediatedepths, whereas 162 can be interpretedas an effect of the bend near Celebes.The results for great depths are uncertain; Ritsema's solutions [1964] for two deep events in the Mindanao zone are not clearly interpretablein terms of stressparallel to the deepzone. Talaud-Halmahera arc. The distributions of volcanoes and earthquakes with depthsbetween 70 and 200 km suggestan island-arc-likestructure striking south-southeastbetween the islands of Talaud and Italmahera [Hatherton and Dickinson,1969; Fitch and Molnar, 1970]. The apparenteastward dip of the zone is oppositeto that of the adjacentMindanao zone. The T axesof 163-166 tend to be more nearly vertical, whereasthe P and B axes tend to be more nearly horizontal. A nearly vertical plane approximately parallel to the island-arc-likesurface features can be fitted throughthe stress axesof three of the solutions.This result suggeststhe presenceof down-dip extensionin a steeplydipping slab. Celebes. Earthquake169, located beneath southern Celebes, is isolatedfrom the seismicallyactive island-arcstructures or from other zonesof shallowearth- quakesin the region.The planethrough the B and T axesdips nearly vertically and strikesslightly west of north. This would be consistentwit.h extensionin a vertically sinking slab. Alternatively, the solutioncould indicate horizontal compressionwithin a thick lithosphericplate on the surface.Examination of the seismogramsand the locationdata for this event indicatesthat the depth of 81 km is reliable.

Sunda Arc (Figure 26) Sumatra-FloresSea. The Sunda arc is a nearly continuousstructure extend- ing fromBurma to the BandaSea. The portionshown in Figure26 includesthe 154 ISACKS AND MOLNAR

Fig. 26. Sunda island arc. main belt of deep and in[ermediate-depthearthquakes. The distributionof earth- quakeswith depth appearsto vary abrup[ly alongthe are; near the SundaStrait depthsincrease from lessthan 200 km beneath Sumatra to grea[er than 600 km beneaththe Java Sea [Fitch and Molnar, 1970]. Be[weenabou[ 105øEand 126øE [he zone reachesdepths greater [han 650 km and possiblygreater than 700 km [Gutenbergand Richter, 1954]. Benea[h the Flores and Java Seasa pronounced minimum in activity is presentbe[ween depths of abou[ 300 to 500 km. The T axes of 170-171 plunge beneath Sumatra and can be in[erpreted as down-dipex[ension. Moreover, Fitch [1970a] reportsa solution(his number1) for an event at the northern end of Suma[ra which, thoughnot well determined, is alsoin agreemen[with extensionparallel •o [he seismiczone; [he solutionis no• in agreemen[with under[hrust faulting. The in[ermedia[e depth of [hat even[ (70 km) appearsto be reliable. The T axis of 173 plungesgently north, whereas[he dip of the zone near 173 appears to be steeper. The structure of [he seismic zone, however, is not suftieientlywell resolved[o excludethe possibilitythat the zone dips more gently near 173 and steepensin dip at greaterdepth [Fitch and Molnar, 1970; Engdahl et al., 1960; E. R. Engdahl, personaleommuniea[ion, 1970]. The solution of 172 indicates extension parallel to [he strike of [he zone. This earthquake is near the change of s[rike of the are, where the distribution of hypoeentral depths abruptly changes.Filch and Molnar suggestedthat the convexcurvature of the are in this region may be responsiblefor the anomalous orientation of 169 and can be explainedby the model of Figure 6. The solutions174-176 for three deep shocksbeneath [he Java Sea are in good agreementwith an inferenceof simple down-dip compression.The ver[ieal orien- STRESSES IN THE DESCENDING LITHOSPHERE :155 tation of the seismiczone shown in Figure26 is basedon sections[Engdah• et al., 1969; F•tch a•d Mo•nar, 1970] that showa well-definednarrow zone•ha• dips nearly vertically betweendepths of abou•500 and 650 km. The three solutions177-179 for deep events beneath the Flores Sea also indicatedown-dip compression parallel to a nearly vertica! zone,but the trends of the B and T axes suggestthat •he zonemay have a more nearly east-north- easterlystrike rather than the easterly strike of •he surfacefeatures and of the alignment of deep hypocentersfarther west. This relationship plus reliable locations of deep events northeast of 177-179 suggestthat the deep zone bends around in a fashion similar to the shallower zone beneath the Banda Sea (see next section).Solutions for three deep earthquakesreported by Ritsema, [1964] are in approximate agreementwith the solutionsof 174-179. Banda Sea. Fitch and Molnar [1970] argue that both the spatial distribution and the focal mechanismsof earthquakeslocated beneath the Banda Sea can be explained by the upward bending of the western edge of a lithospheric slab dipping beneaththe Sunda are (Figure 4). The earthquakesare not as deep beneath the Banda Sea as those farther west; furthermore, as a resul• of the bending, the activity at intermediate depths and at depths of 300 to 500 km is especiallyintense. This explanation implies that the gap in activity west of the bend correspondsto a minimum in stresswithin a continuousslab rather than an actual gap in the slab, i.e., implies model b rather than model d in Figure 3. Solutions 180-187 indicate two predominant stressorientations in the zone' 180-185 indicate down-dip extensionand an interchangeof the positionsof the B and P axes, whereas 186-187 have compressionalaxes oriented between •he strike and dip of the zone and an interchangeof the T and B axes. Solutions 186-187, for the two eventslocated at depthsgreater than 300 km, couldindicate a ½ompressivestress in the upper part of a bent slab; this might also explain the parallelism of the P axes and the strike directionsin the solutionsof someof the eventsa• intermediatedepths. The predominanceof down-dipextension, however, suggeststha• a• intermediatedepths the stressresults mainly from a stretching rather than a bending of the slab. In summary, the down-dip extensionat intermediate depths and down-dip compressionat great depthspredominate in the Sunda are. The solutionsfor the two deep eventsin the bent edgeof the zone beneaththe Banda Sea are among the few caseswe have found in which the effectsof bendingmay be important.

Alpide Belt: Asia The mos• active zones of subcrustal seismic activity in the Alpide bel• between Sumatra and the Mediterranean region are located beneath Burma and the Hindu Kush Mountains. Although subcrustal hypocenters have also been reportedfor the regionof North Sumatra and the AndamanIslands, the Hima- layan Front betweenBurma and the Hindu Kush, the Zagrosbel• of Iran, and the region of the Caspian Sea, focal-mechanismdata for the regionsare lacking. Burma (Figure œ7). The zoneof intermediate-depthearthquakes located near the arcuate range of mountainsbetween Assam and Burma appearsto dip eastward and s•rikes parallel to the north-northeasterly•rend of •he mountain 156 ISACKS AND MOLNAR

Fig. 27. Zone of intermediate-depth earthquakes in Burma. The lines of long and short dashes represent major mountain ranges.

oo belt. As Santo [1969b] notes,the level of shallowactivity is low relative to that at intermediatedepths. Focal depthsprobably do not exceed200 km. Fitch [1970b] reports solutionsfor two intermediate-depthevents in this region.Although only one (188) is fairly well determined,the data for the other are in agreementwith it. The T axisplunges eastward and is thus probablyparal- lel to the dip of the inclined seismiczone. Although limited, the data are thus consistentwith down-dip extensionalstress within the eastward-dippingseismic zone. Note that the P axis, not the B axis, is more nearly parallel to the strike of the zone. This orientation, plus the depths of the events, argues against an interpretation of the data as thrust faulting along the inclined seismiczone. Hindu Kush (Figure 28). The remarkable concentrationof intermediate- depth hypocentersbeneath the Hindu Kush is locatednear a complexjunction of tectonic belts, and no simple island-arc-like structuresare apparent. Near. the Hindu Kush the seismicityis greatestat intermediatedepths, whereas the most active nearby zone of shallow earthquakes is associatedwith the east-northeast-trending block-faulted structures of the Tien Shah located farther north. Beneaththe Hindu Kush the largestearthquakes appear to be remarkably concentratedin space.Gutenberg and Richter'smost accurately located events, includingmost of the largestshocks that occurredbetween 1907 and 1950,all have depths between210 and 240 km and occur within a degreeor less of 36•øN and 70%øE. This spatial concentrationof the largest shockshas con- tinued duringthe periodfrom 1954to the present.Locations of earthquakesof smaller magnitude show that the zone has a discernible extent and structure and is not a pointlikesource of earthquakes.The Atlas ZemletryaseniiSSSR [1962] andNowroozi [1971] show that earthquakehypocenters are distributedthrough- out the rangeof depthsless than 300 km. Nowroozi,in particular,clearly de- lineatestwo thin, slablikezones; one is locatedbetween depths of about 70 and 175 km and trendsnortheast, and the secondis lo.cated at depthsbetween about STRESSESIN THE DESCENDING LITHOSPHERE 157

55 N Fig. 28. Zone of intermediate-depth earthquakes beneath the Hindu Kush. The names refer to the major mountain belts in the region.

50-

65 70E 75

175 and 250 km and trends east-west.In Figure 28 thesetwo zonesare indicated by the 100-kmand 200-km contours oœ hypocentral depths, respectively. Focal-mechanismsolutions for the regionare summarizedby Ritsema[1966], Stevens[1966], Shirokova [1967], and Soboleva [1967, 1968]. Stevens shows that solution189 is representativeoœ most oœthe data, if not all oœthe solutions, reported by other workers.Ritsema derivesa commonsolution (192) for earthquakesthat occurredduring the period 1949-1965.Both are plotted in Figure 28 togetherwith well-determinedsolutions for two other earthquakes(190 and 191). The resultsoœ Soboleva and Shirokovaare in generalagreement with the pattern exhibited by these data. Earthquakes 189-191, as well as most oœthe events comprising192, are probably located in the lower east-west-strikingzone. The T axes dip steeply north, but the poles oœone set oœnodal planes dip as steeply south. From this rela- tionshipalone we couldequally well inœereither (a) shearingor slip parallel to the zone or (b) stressinside a thin slab. However, the interchange oœthe B and P axes oœ190 relative to the other solutions,although difficult to explain in terms oœvertical shearingor slip, is consistentwith extensionalstress inside a slab that dips steeply north. Nowroozi'sdata do not excludea northward dip oœthe zone as small as, say, 75ø . Thus, althoughfurther investigationis required, down-dip extensionin this zoneis the preœerredinterpretation oœthe data.

AlpideBelt: MediterraneanRegion (Figure 29) Known subcrustal earthquakes are confined to •our separate areas in the regionoœ the MediterraneanSea: the Vrancearegion oœ Rumania, the southern AegeanSea, the Tyrrhenian Sea northwestoœ Calabria and Sicily, and southern 158 ISACKS AND MOLNAR

193-197 x

201-202

40 N 'xx

I' -I-- ; + [ X ! 50 o ' 03 0 , 199-2( I I I I 0 IOE 20 30

Fig. 29. The Alpide belt in and near the MediterraneanSea, includingthe Vrancearegion of Rumania (193-197),the Hellenic island arc (198-200),the Calabrianisland are (201-202), and the Spanishdeep earthquake (203).

Spain. Solutionsbased on WWSSN data are available only for the Aegeanarea. Becauseof the numerousstations that have long beenin operationin Europe, however,reliable solutions are alsoavailable for previousearthquakes. ¾rancearegion of the easternCarpathians. A concentratedsource of earth- quakeswith depthsbetween about 100 and 150 km is locatedjust east of the sharpbend of the Carpathian Mountains.Although the detailedstructure of this zoneis not well determined,Karnik [1969], in a summaryof Rumanianwork, suggeststhat it reachesdepths of about 200 km and dips steeply.Radt• [1965] showsan inclinedzone striking northeast and dippingnorthwest. The five solutions193-197 plotted in Figure 29 were selectedfrom the literature as the most reliable that we were able to find. The solutions 193-196 are nearlyidentical; 197 differsin that the P and B axesapproximately interchange positionswith respectto others.The T axesof all five dip nearlyvertically, and the nearly horizontalB and P axesare either parallel or perpendicularto the apparentstrike of the subcrustalzone. The resultsthus suggest down-dip extensional stress within a nearly vertical slab.. This inference is similar to that obtainedfor the concentratedsource region beneath the Hindu Kush. Hellenicarc. The distributionsof shallowand intermediate-depthearth- quakesand volcanoes,focal mechanismsof shallowearthquakes [McKenzie, 1970],and the data from marinegeology [Ryan eta!., 1970] indicatean island- STRESSES IN THE DESCENDING LITHOSPHERE 159 arc structurebounding the southernand southwesternAegean Sea. Solutionsfor shallowearthquakes along the arc indicatenortheasterly underthrusting of the Mediterraneansea floor beneaththe Aegean [McKenzie,1970]. Becausethe seismicityis ratherlow for an islandarc, the distributionof hypocentersis poorly known;the availabledata sugges•a zonedipping north bo northeastbeneath the Aegean.Karnik [ 1969] findsbhab hypocenbral depths do not exceed200 km. Solutions199 and 200 are consistentwi•h down-dipextension wibhin a zone dippingabout 60 degreesnortheast. The T axisof 198 dipsgently north and thus may indicateextension within a gentlydipping zone, although •he P andB axes are not parallelor perpendicular5o the eas•-weststrike of the westernend of the arc. SouthernItaly: Calabrian arc. Gutenberga•d Richter [1954] and Peter- schmitt[1956] showthat intermediate-and deep-focusearthquakes beneath the Tyrrhenian Sea are associatedwith an island-arc-likestructure near Calabria and Sicily (seealso Ryan et al., 1970). The inclinedseismic zone dips northwest and reachesdepths of 450 km. Peterschmittfinds evidencetha• seismic-wave velocitiesnear this zoneare anomalouslyhigh, a resultnow bhought to be charac- 5erisSicof •he island arcs of the Pacific [Oliver and Isacks, 1967; Utsu, 1967; Davies and McKenzie, 1969; Mitronovas, 1969]. Solution202 is an averagesolution for subcrusbalearthquakes derived by Ritsema [1969]. Bo•h solutions201 and 202 have P axes•hat plungenortheast and B axesthat are nearly horizontaland parallel•o the s•rike of the Calabrian arc. The data are bhusin good agreementwith down-dip compressionalstress within •he northwests-dippingzone. Southern Spain. A large deep-focus earthquake occurred beneath the Sierra Nevada of southernSpain in 1954 at a depth of 650 km. No other deep seismicactivity is known •o be located in this region, and the event doesnot appeart• berelated •o an obvious island arc or island-arc-like sbructure. Solution 203 is well determined by numerous firs5 motions reported by nearby Europeanstabions. In Figure 28 two planesare drawn throughthe stress axesof this solution.Extrapolation of resultsfrom o•herregions suggests that •he plane •hroughthe P and B axes,a plane bha• s•rikesnorth and dips 45ø eas•, would be •he mos• likely orientation of a seismiczone near 203. In this casean east-dippingslab of lithosphereis undercompression parallel to its dip. The surface brace of •he inferred slab would be located near or wes• of Portugal and Gibraltar and would strike north-south. Tectonic trends of southern Spain are sometimesshown to be curved around through Gibraltar •o join the riff s•ructures of northern Africa; near Gibraltar such a belt would have a north-south strike and could thus be related to •he surface trace of bhe postulabedseismic zone. Alternatively, a vertical plane throughthe P and T axes is parallel to the pre- dominantlyeas•-wes• alignmen• of the belbof shallowearthquakes in •he western Mediterraneanand alongthe Azores-Gibraltarridge. McKenzie [1970] claims •ha• •his bel• marks •he boundarybetween •he convergingAfrican and Eurasian p!a•es.Thus 203 may be :rela•ed•o lithosphericconsumption along that zone, although•he isolation of the event as well as the orientationof stressare anomalous.Until the complex•ech)nic history of the wesSernMediterranean is 160 ISACKS AND MOLNAR better known,the relationshipof the Spanishdeep earthquake to the near-surface structure remains a puzzle.

APPENDIX

Table 1 Table I lists the locations and focal-mechanismsolutions for the earthquakesplotted in Figures 1, 2, and 7 through 29. The location listed in each case is the most reliable one reported in the literature. If a relocation is not available from a special investigatio.n of seismicity,or from the International SeismologicalSummary and the Bulletins o• the Inter- national SeismologicalCenter, the location is taken from the SeismologicalBulletin of the Coast and Geodetic Survey (ESSA); if none of these is available (mainly for the more recent earthquakes), the location is taken from the Preliminary Determination o• Epicenters of the Coast and Geodetic Survey. The trends and plunges listed for the focal-mechanism solutions are measured in degrees clockwise from north and downward from the horzontal, respectively. The second to last column lists the data plotted in Figure 2. 'P' and 'T' indicate down-dip compression and tension, respectively; 'X' indicates that neither the P nor the T axis is approximately parallel to the dip of the zone; and 'n.p.' indicates that the solution is not plo.tted in Figure 3. The data not plotted are for earthquakes in regions where the structure of the zone is poorly known or very complex and no clear determination of down-dip stress type can be made. In the last column o.n the right, the reference from which the solution is taken is tabulated as follows:

BJ Berckhemerand Jacob[1968] M D.P. McKenzie, personalcommunication CE Constantinescuand Enescu [1962] Ch Chandra [1970b] Md Mendiguren [1969] CRE Constantinescuet al. [1966] MS Molnar and Sykes[1969] Fa Fdch [1970a] N New solutionobtained by us for this paper Fb Fitch [1970b] Ra Ritsema [1965] FM Fitch and Molnar [1970] Rb Ritsema [1966] YI Honda and Masatsuka [1962] Rc Ritsema [1969] and Honda et al. [1956] S Stevens[1966] HC Hodgsonand Cock[1956] St Stauder[1958b] I-Ii Hirasawa [1966] SB Stauder and Bollinger [1965] I-It{ Hedayati and Hirasawa [1966] TB Teng and Ben-Mer•ahem[1965] I Ichikawa [1966] VP• Vvedenskayaand R•prechtova[1961] ISO Isacks et al. [1969] KS Katsumata and Sykes [1969] The numbers following these abbreviations in the last column are those emplo,yedby the authors in each case to identify the particular earthquake.

New Focal-Mechanism Solutions The new solutions ('N' in Table 1) are illustrated in appendix plates 1-4 on equal-area projections of the lower hemisphereof the focal sphere.In all casesthe upper and right-hand side of the projections correspo.ndto north and east, respectively. Filled or solid circles and triangles indicate compressionalfirst motions, and-• untilled or open circles and triangles indicate diiatational first motions. Triangles are for data on the upper hemisphere (pP or P at small distances) projected thro.ugh the center onto the lo,wer hemisphere; circles are for data on the lower hemisphere. X indicates compressional-wavedata judged to be near a nodal plane. The direction of first motion of S is shown by an arrow. A small circle, triangle, or x and a dashedarrow indicate uncertain determinations. Solutionsfor some of the events listed in Table I as new have also been reported by other investigators[e.g., $tauder and Bollinger, 1964, 1965; Chandra, 1970a; and Mikumo, STRESSES IN THE DESCENDING LITHOSPHERE 161 162 ISACKS AND MOLNAR

o o o o

oo •

o STRESSES IN THE DESCENDING LITHOSPHERE 163 164 ISACKS AND MOLNAR STRESSESIN THE DESCENDING LITHOSPHERE 165 166 ISACKS AND MOLNAR

x <] • STRESSESIN THE DESCENDING LITHOSPHERE 167 168 ISACKS AND MOLNAR STRESSES IN THE DESCENDING LITHOSPHERE 169

1969]. In most cases the agreement of their solutions and ours is excellent and attests to the reliability of solutionsbased on WWSSN data. The largest discrepanciesare associated with certain of Stauder and Bollinger's solutions for earthquakesin the Peru-Chile region. We attribute these discrepanciesto their use of first-motion data for P waves that a•e reported in bulletins (and are often inconsistent) and their primary, (nearly exclusive in some cases) reliance upon the polarization of shear waves.

Acknowledgments. Discussionswith E. R. Engdahl, T. Fitch, D. T. Griggs, T. Hanks, V. Karnik, D. P. McKenzie, J. Oliver, A. R. Ritsema, C. Scho.lz,and L. R. Sykes were especiallyhelpful. E. R. Engdahl, D. P. McKenzie, and A. Nowroozi kindly supplied data prior to publication of their work. We thank J. Oliver and L. R. Sykes for critically reading the manuscript. This researchwas supportedby NOAA, Department of Commerce, Grants E22-53-69G and E22-73-70G, and by National Science Foundation Grants NSF GA-11152 and NSF GA-1537. Contribution 1615 from Lamont-Doherty Geological Observatory.

REFERENCES Adams, R. D., Sourcecharacteristics of some deep New Zealand earthquakes,N. Z. J. Geol. Geophys., 6, 209-220, 1963. Aki, K., Earthquake generatingstress in Japan for the years 1961to 1963obtained by smooth- ing the first motion radiation patterns, Bull. Earthquake Res. Inst. Tokyo Univ., 44, 447- 471, 1966. Akimoto, S., and H. Fujisawa, Olivine-spinelsolid solution equilibria in the system Mg•SiO•- Fe2SiO•,J. Geophys. Res., 73, 1467-1479, 1968. Algermissen,S. T., and S. T. Harding, The Puget Sound,Washington Earthquake o• April 29, 1965. Preliminary SeismologicalReport, 26 pp., U.S. Dept. of Commerce, Coast and Geodetic Survey, Rockville, Md., 1965. Anderson,D. L., Earth's viscosity,Science, 151, 321-322, 1966. Anderson,D. L., Phase changesin the upper mantle, Science,157, 1165-1173,1967. Atlas Zemletryase•ii SSSR, Soviet Po Seismologii,338 pp., Akad. Nauk SSSR, Moscow, 1962. Balakina, L. F., General regularitiesin the directionsof the principal stresseseffective in the earthquakefoci of the seismicbelt of the Pacific Ocean,Izv. Acad. Sci. USSR, Geophys. Set., English transl., 918-926, 1962. Barazangi, M., and J. Dorman, World seismicity maps compiled from ESSA, Coast and Geodetic Survey, epicenterdata, 1961-1967,Bull. Seismol.Soc. Amer., 59, 369-380, 1969. Berckhemer,H., and K. H. Jacob,Investigation of the dynamical processin earthquakeloci by analyzingthe pulseshape of body waves,85 pp., Final Rep., ContractAF61(052)-801, Inst. of Meteorology and Geophysics,Univ. of Frankfurt, 1968. Brune,J. N., Seismicmoment, seismicity, and rate of slip alongmajor fault zones,J. Geophys. Res., 73, 777-784, 1968. Byeflee,J. D., and W. F. Brace,High-pressure mechanical instability in rocks,Science, 164, 713-715, 1969. Chandra, U., Analysis of body-wavespectra for earthquakeenergy determination,Bull. Seismol.Soc. Amer., 60, 539-563, 1970a. Chandra,U., The Peru-Boliviaborder earthquakeof August 15, 1963, Bull. Seismol.Soc. Amer., 60, 639-646, 1970b. Constantinescu,L., and D. Enescu,Nature of faulting and stresspattern at the loci of some carpathian-arc-bendearthquakes, Rev. G•ol. G•ogr., Acad. R•publique PopulaireRou- maine, Titage • Part, VI (1), 143-183,1962. Constantinescu,L., L. Ruprechtova,and D. Enescu,Mediterranean-Alpine earthquake mechanismsand their seismotectonicimplications, Geophys. J. Roy. Astron. Soc., 10, 347-368, 1966. Davies,D., and I). P. McKenzie,Seismic travel-time residuals and plates,Geophys. J. Roy. Astron. •goe.,18, 51-63, 1969. . 170 ISACKS AND MOLNAR Davies, G. F., and J. N. Brune, Global and regional fault activity from seismicmoments of large shallow earthquakes,Nature, in press,1971. Denham,D., Distribution of earthquakesin the New Guinea,Solomon Islands region, J. Geo- phys. Res., 74, 4290-4299,1969. Duda, S. J., 8eculaeseismic energy releasein the circum-Pacificbelt, Tectonophysics,•, 409-452, 1965. Elsasser,W. M., Convection and stresspropagation in the upper mantle, in The •4pplic•tion o• Modern Physics to th• Earth and Planetary Interiors, edited by K. Runcorn, pp. 223-246, Wiley-Interscience, 1969. Engdahl, E. R., and E. A. Flinn, Seismicwaves reflected from discontinuitieswithin earth's upper mantle, Science,163, 177-179, 1969. Engdahl, E. R., S. T. Algermissen,and S. T. Harding, Global patterns of earthquakefocal mechanisms,oral presentation, General Scientific Assemblies,IASPEI/IAGA, Madrid, 1969. Fedotov, S. A., Upper mantle propertiesof the southernpart of the Kurile island arc according to detailed seismologicalinvestigation data, Tectonophysics,•, 219-225, 1965. Fedotov, S. A., On deep structure, propertiesof the upper mantle, and volcanism of the Kurile-Kamchatkaisland arc according.toseismic data, in The Crustand Mantle o[ the PacificArea, Geophys.Monograph 1•, editedby L. Knopoff,C. L. Drake, and P. J. Hart, pp. 131-139,AGU, Washington, D.C., 1968. Fitch, T. J., Earthquake mechanismsand island arc tectonics in the Indonesian-Philippine region, Bull. $eismol. $oc. Amer., 60, 565-591, 1970a. Fitch, T. J., Earthquake mechanismsin the Himalayan, Burmese,and Andaman regionsand continental tectonicsin central Asia, J. Geophys.Res., 75, 2699-2709,1970b. Fitch, T. J., and P. Molnar, Focal mechanismsalong inclined earthquake zones in the indonesian-Philippineregion, J. Geophys.Res., 75, 1431-1444,1970. Frank, F. C., Curvature of island arcs,Nature, •0, 363, 1968. Gordon, R. B., Diffusion creep in the earth's mantle, J. Geophys.Res., 70, 2413-2418,1965. Griggs, D. T., The sinking lithosphereand the focal mechanismof deep earthquakes,in Symposiumin Honor of F. Birch, edited by E. Robertson,McGraw-Hill, New York, in press, 1971. Griggs,D. T., and D. W. Baker, The origin of deep-focusearthquakes, in Propertiesof Matter under Unusual Conditions, edited by H. Mark and S. Fernback, 389 pp., John Wiley, New York, 1969. Griggs,D., and J. Handin, Observationson fracture and a hypothesisof earthquakes,in Rock Deformation,edited by D. Griggs and J. Handin, Geol. 5oc. Amer. Memoir 79, 347-373, 1960. Gutenberg,B., and C. F. Richter, 5eismicityof the Earth and AssociatedPhenomena, 310 pp, Princeton University Press,Princeton, N.J., 1954. Hamilton, R. M., and A. W. Gale, Seismicityand structureof the North Island, New Zealand, J. Geophys.Res., 73, 3859-3876, 1968. Hamilton, R. M., and A. W. Gale, Thicknessof the mantle seismiczone beneaththe North Island of New Zealand, J. Geophys. Res., 7•, 1608-1613, 1969. Harrington,H. J., Deep focusearthquakes in SouthAmerica and their possiblerelation to continentaldrift, in Polar Wanderinga•d ContinentalDrift, edited by R. C. Munyan, pp. 55-73, AmericanAssociation of Petroleum Geologists,Tulsa, Okla., 1963. Hatherton,T., Gravity and seismicityof asymmetricactive regions,Nature, •1, 353,1969. Hatherton,T., Upper mantle inhomogeneitiesbeneath New Zealand: Surfacemanifestations, J. Geophys.Res., 75, 269-284, 1970. Hatherton, T., and W. R. Dickinson, The relationshipbetween andesitic volcanismand seismicityin Indonesia,the LesserAntilles, and other islandarcs, J. Geophys.Res., 74, 5301-5310, 1969. Hayes,D. E., and W. J. Ludwig,The Manila trenchand West Luzon trough--2. Gravity and magnetic measurements,Deep Sea Res., 1•, 545-560, 1967. Hedayati, A., and T. Hirasawa, Mechanismof the Hindu Kush earthquakeof Jan. 28, 1964, STRESSESIN THE DESCENDING LITHOSPHERE '171 derivedfrom S wavedata: The useof pP phasefor the focalmechanism determination, Bull.Earthquake Res. Inst. TokyoUniv., 46, 1419-1434,196•. Hirasawa,T, A least squaresmethod for the focal mechanismdeterminations from S wave data(2), Bull.Earthquake Res. Inst. TokyoUniv., •, 919-938,19C•. Hodgson,J. H., andJ. I. Cock,Direction of faultingin the deepfocus Spanish earthquake of March 29, !954, Tellus,$, 321-328,195•. Holmes,A., Principlesof PhysicalGeology, 1288 pp., Ronald, New York, 1965. Honda,H., Earthquakemechanisms and seismicwaves, J. Phys.Earth, 10, 1-97, 1962. Honda,H., andA. Masatsuka,On the mechanismsof earthquakesand the stressesproducing themin Japanand its vicinity,$ci. Rep., TohokuUniv., Set. 5, •eophys.,6, 42-60,1952. Honda,H., A. Masatsuka,and K. Emura,On the mechanismof the earthquakesand the stressesproducing them in Japanand its vicinity, 2, $ci. Rep., Tohoku Univ., •er. 5, G½o- •hys., 8, 186-205,195•. Honda,H., A. Masatsuka,and M. Ichikawa,On the mechanismof earthquakesand stresses producingthem in Japanand its vicinity,3, Geo•hys.Mag., $$, 271-279,1967. Ichikawa,M., On the mechanismof the earthquakein andnear Japan during the periodfrom 1950to 1957,Geophys. Ma•., $0, •5-403, 1961. Ichikawa,M., Mechanismof earthquakesin andnear Japan, 1950-1962, Pap. Meteorol.Geo- phys.(Tokyo), 1•, 201-229,1966. Isacks,B.,•Focal mech•s ofearthquakes in western South America (abstract), EO$, Trans. A GU, 51, 355, 1970. Isacks,B., andP. Molnar,Mantle earthquake mechanisms and the sinkingof the lithosphere, Nature, •$, 1121-1124,1969. Isacks,B., J. Oliver,and L. R. Sykes,Seismology and the newglobal tectonics, J. Geophys. Res., 73, 5855-5899,1968. Isacks,B., L. R. Sykes,and J. Oliver,Focal mechanisms of deep and shallow earthquakes•in the Tonga-Kermadecregion and the tectonicsof islandarcs, Bull. Geol.$oc. Amer., 80, 1443-1470, 1969. Jacoby,W., Instabilityin the uppermantle and global plate movements, J. Geophys.Res., 75, 5671-5680, 1970. Kamik,V., Seismicityo] theEuropean Area, Part 1, 364pp., D. Reidel,Dordrecht, Holland, 1969. Katsumata,M., Seismicactivities in andnear Japan (D--Distribution of epicentersof earthquakesin andnear Japan (in Japanese),J. $eismol.$oc. Japan (Zisin), 19, 237-245,1966. Katsumata,M., Seismicactivities in and near Japan (2)--Vertical distributionof foci of earthquakesin and near the Japaneseislands (in Japanese),J. $eismol.Soc. Japan (Zisin), •0, 1-11, 1967a• Katsumata,M., Seismicactivities in andnear Japan (3)--Seismic activities versus depth (in Japanese),J. $eismol.•oc. Japan(Zisin), 10, 75-84,1967b. Katsumata,M., and L. R. Sykes,Seismicity and tectonicsof the westernPacific: Izu-Mariana- Carolineand Ryukyu-Taiwanregions, J. Geophys.Res., 74, 5923-5948,1969. Le Pichon,X., Sea-floorspreading and continental drift, J. Geophys.Res., 75, 3661-•697,1968. Lliboutry,L., Sea-floorspreading• continental drift and lithospheresinking with an astheno- sphereat meltingpoint, J. Geophys.Res., 74, 6525-6540,1969. Ludwig,W. J., D. E. Hayes,and J. I. Ewing,The Manilatrench and West Luzon trough--1. Bathymetryand sedimentdistribution, Deep-Sea Res., 1•, 533-544,1967. . McConnell,R. K., Viscosityof the mantlefrom relaxation time spectraof isostaticadjust- ment, J. Geophys.Res., 7•, 7089-7105,1968. McKenzie,D. P., The relationbetween fault pl.ah•solutions for earthquakesand the directionsof the principalstresses, Bull. $eismol. •oc. Amer., g9, 591-601, 1969a. McKenzie,D. P., Speculationson the •onsequencesand causes of platemotions, Geophys. J., •$, 1-32, 1969b. McKenzie,D. P., Platetectonics of the Mediterraneanregion, Nature, •6, 239-243,1970. McKenzie,D. P., andW. J. Morgan,Evolution of triplejunctions, Nature, •4, 125-1•, 1969. 172 ISACKS AND MOLNAR McKenzie,D. P., and R. L. Parker,The North Pacific: An exampleof tectonicson a sphere, Nature, 216, 1276-1280, 1967. • Mendiguren,J., Study of focal mechanismsof deepearthquakes in Argentinausing nonlinear particlemotion of S waves,Bull. $eismol.$oc. Amer., 59, 1449-1473,1969. Mikumo, T., Long period P wave forms and the sourcemechanism of intermediateearthquakes,J. Phys. Earth, 17, 169-192, 1969. Mitronovas,W., Seismicvelocity anomalies in the uppermantle beneath the Tonga-Kermadec island arc, Ph.D. thesis, Columbia University, 1969. Mitronovas,W., B. Isacks,and L. Seeber,Earthquake locations and seismicwave propagation in the upper250 km of the Tongaisland arc, Bull. Seismol.$oc. Amer.,59, 1115-1135,1969. Molnar, P., and J. Oliver, Lateral variationsof attenuationin the upper mantle and dis- continuities in the lithosphere, J. Geophys. Res., 74, 2648-2682, 1969. • Molnar, P., and L. R. Sykes,Tectonics of the Caribbeanand Middle Americaregions from focal mechanismsand seismicity,Bull. Geol. Soc. Amer., 80, 1639-1684,1969. Mooney, I-I. M., Upper mantle inhomogeneitybeneath New Zealand' Seismicevidence, J. Geophys., Res., 75, 285-309, 1970. Morgan,W. J., Rises,trenches, great faults and crustal blocks, J. Geophys.Res., 73, 1959-1982, 1968. Murdock,J. N., Crustmantle system in the centralAleutian region--A hypothesis, Bull. $eismol. $oc. Amer., 59, 1543-1558, 1969. Nowroozi,A., Seismotectonicsof the Persianplateau and vicinity, Bull. Seismol. $oc. Amer., in press, 1971. Ocala,L., Earthquakeactivity of Peru,in The Earth beneaththe Continents,edited by J. S. Steinhart and T. J. Smith, pp. 509-528,AGU, Washington,D.C., 1966. Oliver, J., and B. Isacks,Deep earthquakezones, anomalous structure in the upper mantle, and the lithosphere, J. Geophys. Res., 72, 4259-4275, 1967. Oliver, J., and B. Isacks,Structure and mobility of the crust and mantle in the vicinity of island arcs, Can. J. Earth $ci., 5, 985-991, 1968. Orowan,E., Mechanismof seismicfaulting, in Rock Deformation,edited by D. Griggsand J. Handin, Geol. $oc. Amer. Memoir 79, 323-345, 1960. Peterschmitt,E., Quelquesdonn•es nouvelles sur les s•ismesprofonds de la Mer Tyrrh•nienne, Ann. Geofis. (Rome), 9, 305-334, 1956. Press,F., The suboceanicmantle, Science,165, 174-176, 1969. Radu, C., Le r•gime s•ismiquede la r•gion de Vrancea,Rev. Roumaine Gdol. Giophys. Gdogr., 9, 49-64, 196•i. Raleigh,C. B., and M. S. Paterson,Experimental deformation of serpentiniteand its tectonic implications, J. Geophys.Res., 70, 3965-3985, 1965. Randall, M. J., Relative sizesof multipolar componentsin deep earthquakes,J. Geophys. Res., 73, 6140-6142, 1968. Ritsema, A. R., Some reliable fault-plane solutions,Pure Appl. Geophys.,59, 58-74, 1964. Ritsema,A. R., The mechanismof somedeep and intermediateearthquakes in the regionof Japan, Bull. Earthquake Res. Inst. Tokyo Univ., J3, 39-52, 1965. Ritsema,A. R., The fault-planesolutions of earthquakesof the Y[indu Kush centre,Tectono- physics,3, 147-163, 1966: Ritsema,A. R., Seismicdata of the West Mediterraneanand the problem of ocean[zation, Trans. Kon. Ned. Geol. Mynbk., Gen. 26, 105-120, 1969. Ritsema, A. R., The mechanismof mantle earthquakesin relation to phase transformation processes,Phys. Earth Planetary Interiors, 3, 503-510, 1970. Ryan, W. B. F., D. J. Stanley,J. B. I-Iersey,D. A. Fahlquist,and T. D. Allan, The tectonics and geologyof the MediterraneanSea, in The Seas,vol. 4, edited by A. E. Maxwell, John Wiley, New York, in press, 1970. Sacks,I. S., Distribution of absorptionof shear waves in South America and its tectonic significance,Carnegie Inst. Wash. Yearb. 67, 339-344, 1969. Santo,T., Characteristicsof seismicityin SouthAmerica, Bull. EarthquakeRes. Inst. Tokyo Univ., •7, 6•q5•72, 1969a. STRESSESIN THE DESCENDING LITHOSPHERE 173 Santo,T., Regionalstudy on the characteristicseismicity of the world, Part 2, from Burma downto Java,Bull. EarthquakeRes. Inst. Tokyo Univ.,J7, 1049-1061,1969b. Savage,J. C., The mechanicsof deep-focusfaulting, Tectonophysics; 8, 115-127, 1969. Savarensky,E. F., and N. V. Golubeva,Seismicity of continentalAsia and the regionof the Sea of Okhotsk,1953-1965, in The Earth'sCrust and UpperMantle, Geophys.Monograph 13, editedby P. J. Hart, pp. 134-139,AGU, Washington,D.C., 1969. Shirokova,E. I., Generalfeatures in the orientationof principalstresses in earthquakeloci in the Mediterranean-Asianseismic belt, Izv. Acad. Sci., USSR, Phys. Solid Earth, Engl. transl.,no. 1, pp. 12-21, 1967. Silver, E. A., Late Cenozoicunderthrusting of the continentalmargin off northernmost California,Science, 166, 1265-1266,1969. Soboleva,O. V., The stressesin the foci of the Hiudukushearthquakes of July 6, 1962and of March 14, 1965,Izv. Acad.$ci., USSR,Phys. Solid Earth, Engl. transl.,no. 2, pp. 126- 128, 1967. Soboleva,O. V., Specialfeatures of the directionsof the principalstress axes in the foci of Hindu-Kushearthquake, Izv. Acad. Sci., USSR, Phys. Solid Earth, Engl. transl.,no. 1, pp. 36-40, 1968. Stauder,W., Mechanismsof the Rat Island earthquakesequence of February4, 1965,with relationto islandarcs and sea-floorspreading, J. Geophys.Res., 73, 3847-3858,1968a. Stauder,W., Tensionalcharacter of earthquakeloci beneath the Aleutian trench with relation to sea-floorspreading, J. Geophys.Res., 73, 7693-7701,1968b. Stauder,W., The mechanismof South American earthquakes:18øS to 34øS (abstract), Earthquake Notes, JO, 19, 1969. Stauder,W., and G. A. Bollinger,The S wave project for focal mechanismstudies, Earth- quakesof 1962,93 pp., Sci. Rep. Grant AF-AFOSR-62-J58,Project VELA-UNIFORM, St. Louis University, St. Louis, Mo., 1964. Stauder, W., and G. A. Bollinger, The S wave project for focal mechanismstudies, Earth- quakesof 1963,91 pp., Sci. Rep. Grant AF-AFOSR-62-J58,Pro•ect VELA-UNIFORM, St. Louis University, St. Louis, Mo., 1965. Stevens,A. E., S-wavefocal mecha.nismstudies of the Hindu Kush earthquakeof July 6, 1962, Can. J. Earth Sci., 3, 367-384, 1966. Stevens,A. E., and J. H. Hodgson,A study of P nodal solutions(1922-1962) in the Wickens- HodgsonCatalogue, Bull. Seismol.Soc. Amer., 58, 1071,1968. Sugimura,A., Distributionof volcanoesand seismicityof the mantle in Japan, Bull. Volcan. Soc.Japan, 2nd Set., 10, 37-58, 1965. Sugimura,A., and S. Uyeda, A possibleanistropy in the upper mantle accountingfor deep earthquakefaulting, Tectonophysics,5, 25-3.3,1967. Sykes, L. R., Deep-focus earthquakes in the New Hebrides region, J. Geophys. Res., 69, 5353-5355, 1964. Sykes, L. R., The seismicity and deep structure of island arcs, J. Geophys.Res., 71, 2981- 3006, 1966. Sykes, L. R., and M. Ewing, The seismicity of the Caribbean region, J. Geophys. Res., 70, 5065-5074, 1965. Sykes, L. R., B. Isacks, and J. Oliver, Spatial distribution of deep and shallow earthquakes of small magnitude in the Fiji-Tonga region, Bull. Seismol.$oc. Amer., 59, 1093-1113, 1969. Teng, T., and A. Ben-Menahem, Mechanism of deep earthquakesfrom spectrumsof isolated body wave signals,J. Geophys.Res., 70, 5157-5170,1965. Tobin, D. G., and L. R. Sykes, Seismicity.and tectonics of the Northeast Pacific Ocean, J. Geophys. Res., 73, 3821-3845, 1968. Tryggvason,E., and J. E. Lawson,Jr., The intermediate earthquake sourcenear Bucaramanga, Colombia, Bull. Seismol. Soc. Amer., 60, 269-276, 1970. Utsu, T., Anomalies in seismicwave velocity and attenuation associatedwith a deep earthquake zone (1), J. Fac. $ci., Hokkaido Univ., Ser. 7 (Geophys.), 3, 1-25, 1967. 174 ISACKS AND MOLNAR Vvedenskaya,A. V., and L. Ruprechtova,Characteristic features of stressdistribution in the foci of earthquakesat the bend of the Carpathianarc (in Russian),Izv. Akad. Na•k 3,.q3R,•r., Geofiz.,no. 7, 953-965,1961. Weertman,J., The ereepstrength of the earth'smantle, Rev. Geoph•]s.,8, 145-168,1970. Wyss,M., Stressestimates for SouthAmerican shallow and deepearthquakes, J. Geoph•]s. Res., 75, 1529-1544,1970.

(Received July 31, 1970.)