LETTERdoi:10.1038/nature09507

Deformation of the lowermost mantle from seismicanisotropyAndy Nowacki1, James Wookey1 & J-Michael Kendall1

The lowermost part of the Earths mantleknown as D0showssignificant seismic anisotropy, the variation of seismic wave speedwith direction15. This is probably due to deformation-inducedalignment of MgSiO3-post-perovskite (ppv), which is believed tobe the main mineral phase present in the region. If this is the case,then previous measurements of D0 anisotropy, which are generallymade in one direction only, are insufficient to distinguish candidatemechanisms of slip in ppv because the mineral is orthorhombic.Here we measure anisotropy in D0 beneath North and CentralAmerica, where material from subducting oceanic slabs impinges6

on the coremantle boundary, using shallow as well as deep earth-quakes to increase the azimuthal coverage in D0. We make morethan 700 individual measurements of shear wave splitting in D0 inthree regions from two different azimuths in each case. We showthat the previously assumed2,3,7 case of vertical transverse isotropy(where wave speed shows no azimuthal variation) is not possible,and that more complicated mechanisms must be involved. We testthe fit of different MgSiO3-ppv deformation mechanisms to ourresults and find that shear on (001) ismost consistent with observa-tions and the expected shear above the coremantle boundarybeneath subduction zones. With new models of mantle flow, orimproved experimental determination of the dominant ppv slip

systems, this method will allow us to map deformation at thecoremantle boundary and link processes in D0, such as plumeinitiation, to the rest of the mantle.

Studies of D0 anisotropy in the Caribbean are numerous24,79

because of an abundance of deep earthquakes in South America andseismometers in North America, and show approximately 1% shearwaveanisotropy.Thesestudiesmostlycompare thehorizontallypolarized(SH) and vertically polarized (SV) shear waves, assuming verticaltransverse isotropy, a kind of anisotropy in which the shear wavevelocityVS varies only with the angle away from the vertical. With thisassumption, SH leads SV here, corresponding to Q95 190u in ournotation (Fig. 1c). A further limitation is their use of only one azimuthof rays in D0: this cannot distinguish vertical transverse isotropy fromthe case of an arbitrarily tilted axis of rotational symmetry in whichwave speed does not vary when the axis dips towards the receivers orstations (tilted transverse isotropy). An improvement on this situationcanbemade byusing crossing raypaths inD0 (ref. 10), but this relies onhaving the correct sourcereceiver geometry, which is not possiblebeneath North America using only deep earthquakes. We address thisissue beneath the Caribbean by incorporating measurements fromshallow earthquakes in our data set, and thus reduce the symmetryof the anisotropy which must be assumed.

1Department of Earth Sciences, University of Bristol, Wills Memorial Building, Queens Road, Bristol, BS8 1RJ, UK.

a

c

Mantle

Outer coreD

S

North

Backazimuth

= backazimuth

Receiver

Coremantle boundary

Source

ScSSK

S

b

Figure 1 | Sourcereceiver geometry, and explanation of Q9. a, Earth sectionwith ray paths for the S, ScS and SKS phases. The stippled upper mantle andgrey D0 are anisotropic. S turns above D0; ScS samples it. b, Shown are seismicstations (triangles), earthquake epicentres (yellow circles), ray paths (thin blacklines) and ray paths in a 250-km-thick D0 (blue lines). The measured source-side shear-wave splitting parameters for shallow earthquakes are shown asblack bars beneath circles (bar length corresponds to delay time, orientationrepresents fast direction, largest delay time is 2.4 s). We note that fastorientations of shear-wave splitting in the upper mantle beneath shallowearthquakes on plate boundaries are either generally very closely parallel to the

plate-spreading direction (the East Pacific Rise and theMid-Atlantic Ridge), orto the subduction zone trench (Central America). c, Relationship of themeasured fast directions in the geographic (Q) and ray (Q9) reference frames.Because the ScS phase is nearly horizontal for most of its travel through D0, wedefine Q95 backazimuth2Q, which corresponds to the polarization awayfrom the vertical of the fast shear wave. In terms of transverse anisotropy,Q95 690u is compatible with vertical transverse isotropy, and 90u,Q9, 90uimplies tilted transverse isotropy. This can also be thought of as the planenormal to the rotational symmetry axis being tilted from the horizontal, ordipping, at (902Q9)u.

2 8 O C T O B E R 2 0 1 0 | V O L 4 6 7 | N A T U R E | 1 0 9 1

Macmillan Publishers Limited. All rights reserved2010

Wemeasure anisotropy inD0 using differential splitting in S and ScS(respectively direct and reflected from the coremantle boundary)phases using an approach described by refs 10 and 11. Both phasestravel through the same region of the upper mantle, but only ScSsamples D0 (Fig. 1a). Given that the majority of the lower mantle isrelatively isotropic12, by removing the splitting introduced in the uppermantle we can measure the splitting that occurs only in D0 (seeSupplementary Information). Earthquakes in South and CentralAmerica, Hawaii, the East Pacific Rise and the Mid-Atlantic Ridge,detected at North American stations, provide a dense coverage ofcrossing rays that traverse D0 beneath southern North America andthe Caribbean (Fig. 1b). Three distinct regions are covered (Fig. 2),each sampled along two distinct azimuths. The Caribbean (region S)has previously beenwell studied1,4,8, but the northeast (E) and southwest(W) USA have not.

Stacked results along each azimuth in the three regions give splittingparameters shown in Fig. 2 and listed in Supplementary Table 3. Wediscuss results in terms of the delay time (dt) and ray frame fast orienta-tion (Q9; Fig. 1c). The primary observation is that D0 everywhere showsanisotropy of between 0.8% and 1.5% (assuming a uniform 250-km-thickD0 layer).Along southnorth (region S) and southeastnorthwest(region E) ray paths, fromdeep SouthAmerican events (approximately200 measurements), dt5 (1.456 0.55) s, implying shear wave aniso-tropy of about 0.8%. Fast orientations are approximately parallel tothe coremantle boundary (Q9< 90u). This agrees with previous studiesmade along similar azimuths4,79, including the presence of some smallvariation in Q9 of up to 615u (refs 4 and 8). Such variations could beapproximated as vertical transverse isotropy over the region. Detailedresults are shown in Supplementary Figs 1 and 11. Notably, however,oblique to the approximately southnorth raypaths in the Caribbean,fast directions are at least 40u away from parallel to the coremantleboundary (region S: dt5 1.68 s, Q9

rigorous flow modelling in the region is required to resolve this issueconclusively.

The slip systems predicted for perovskite andMgO (SupplementaryFig. 12) seem less likely, particularly where the measured splitting ishigh. The presence of perovskite versus ppv in D0 in region S, forinstance, cannot account for the high anisotropy inferred, and shearplanes and directions for MgO are mostly very steep.

D0 anisotropy might also arise from shape-preferred orientation ofseismicallydistinctmaterial over sub-wavelength scales. Thiswould leadto tilted transverse isotropy behaviour2, withwhich our observations arecompatible. In this case, we can interpret our results simply by findingthe common plane, normal to the rotational symmetry axis, from thetwo azimuths and Q9. These planes are shown in Supplementary Fig. 2.

In each region, the tilted transverse isotropy plane dips approxi-mately in the same way as for the [100](010) case, that is, southwest,southeast and south in regions W, S and E respectively, by between

2652u (Supplementary Fig. 2). However, there is no constraint on theslip direction, and especially in regions S and E, where the dip isabout 50u, it is hard to correlate the transverse isotropy planes with acandidate plane of deformation based on VS, and models of deforma-tion suggest that strain in such slab-parallel orientations is unlikely.For this reason and because the post-perovskite phase explains otherD0 properties25, we favour the mineralogical interpretation at present,in which all tested ppv mechanisms are in some agreement with ourresults, and the [100](001) slip system in ppv is most compatible withour observations.

We have made significant progress towards using D0 anisotropy tomeasure deformation in the lowermost mantle. Assuming that aniso-tropy in D0 is caused by the alignment of ppv, we may suggest whichslip system dominates LPO, though without more detailed models ofmantle flow there is still doubt as to the likely orientation of slip planesand directions in the lowermost mantle. As more reliable estimates of

3,600200

4,000600

4,4001,00

04,8001,

400

5,2001,80

05,6002,

200

6,0002,60

0

Radiu

s (km

)

Height above coremantle

boundary (km)

W S

3060

90 0

E

FS

[110

](110

)-

[100

](010

)[1

00](0

01)

Ap

pro

xim

ate

stra

in

a

b

e

c

f

d

g

h i j

North

X X

High

Low

W S E

Slip direction

Shear plane

Figure 3 | Section through study region and compatible shear planes forcandidate ppv slip systems. a, Cross-section through VS model S20RTStraversing the study region, as indicated in Fig. 2. The approximate regions W,S and E in D0 are drawn. Colours indicate VS as for Fig. 2. The inferredlocation of the Farallon slab from high VS is labelled FS. bj, Orientations ofpotential elasticmodels that are compatible with the observed anisotropy inD0.Shown are upper-hemisphere equal-area projections looking down the Earthradial direction (vertical) of the possible shear planes (coloured lines) and slip

directions (black circles) in ppv for each slip system. The colour of the shearplanes indicates the amount of strain required to produce themaccording to thearbitrary colour scale, right. The three slip mechanisms [110](110) (bd), [100](010) (eg) and [100](001) (hi) are tested in each region (left to right,W, S, E). Up is north. There are usually two sets of planes, because twoazimuths of measurements are not sufficient to define the planes uniquely inthe orthorhombic symmetry of the models.

LETTER RESEARCH

2 8 O C T O B E R 2 0 1 0 | V O L 4 6 7 | N A T U R E | 1 0 9 3

Macmillan Publishers Limited. All rights reserved2010

the type of deformation we expect in well-studied regions becomeavailable, and as numerical and physical experiments further indicatethemechanisms bywhich thematerial inD0 deforms, our observationsof seismic anisotropy may become very useful in the mapping ofdynamic processes at the coremantle boundary.

METHODS SUMMARYWe measured differential shear wave splitting between S and ScS (reflected fromthe coremantle boundary) recorded at about 500 seismic stations in North andCentral America, using events of moment magnitude Mw$ 5.7 and epicentraldistance 55u82u (Supplementary Table 3). Data were bandpass-filtered between0.001Hz and 0.3Hz to remove noise.We analysed splitting in the phases using theminimum eigenvalue technique (Supplementary Fig. 3). We correct for uppermantle anisotropy using published31,32 SKS (seismic waves travelling as shearwaves in themantle, compression waves in the outer core) splittingmeasurementsat stations showing little variation of parameters with backazimuthcorrespondingto simple upper mantle anisotropywhere there are measurements along similarbackazimuths to SScS used here. Measuring splitting in S with a receiver-sidecorrection gives an estimate of the source-side splitting beneath the earthquake(Fig. 1b; Supplementary Table 3). Both corrections are applied when analysingScS: the measurement is thus of splitting in D0 alone.

We confirm that the only source of splitting in our measurements is D0 bycomparing: (1) splitting in S from a deep eventwith that in SKS; (2) the source-sideanisotropy with SKS measurements at the source; (3) the initial polarization of Safter analysis with that predicted by the GlobalCMT moment tensor solution(http://www.globalcmt.org/); (4) the consistency of measurements when correct-ing with real SKS and randomized receiver corrections; (5) Q9 and dt along thesame ray paths for deep and shallow events, correcting the latter for upper mantleanisotropy. (See online-only Methods and Supplementary Figs 59 for details.)

Orientations of shear planes and slip directions in each slip system of ppv arecomputed by grid search over the elastic constants16,20,25, which are rotated aboutthe three principal axes. Shear wave splitting is calculated, and orientations whichare compatible with the observations are plotted. The constants are scaled linearlyaway from the isotropic case to fit the observations, and this scaling is shown bycolour (Fig. 3bj), qualitatively representing strain.

Full Methods and any associated references are available in the online version ofthe paper at www.nature.com/nature.

Received 21 January; accepted 6 September 2010.

1. Kendall, J.-M. & Silver, P. Constraints from seismic anisotropy on the nature of thelowermost mantle. Nature 381, 409412 (1996).

2. Kendall, J.-M. & Silver, P. G. in The CoreMantle Boundary Region (eds Gurnis, M.,Wysession, M. E., Knittle, E. & Buffett, B. A.) 97118 (American Geophysical Union,1998).

3. Lay, T., Williams, Q., Garnero, E. J., Kellogg, L. &Wysession,M. E. in The CoreMantleBoundary Region (eds Gurnis, M., Wysession, M. E., Knittle, E. & Buffett, B. A.)299318 (American Geophysical Union, 1998).

4. Maupin, V., Garnero, E. J., Lay, T. & Fouch,M. J. Azimuthal anisotropy in theD0 layerbeneath the Caribbean. J. Geophys. Res. Solid Earth 110, B08301 (2005).

5. Long, M. D. Complex anisotropy in D0 beneath the eastern Pacific fromSKSSKKSsplitting discrepancies. Earth Planet. Sci. Lett. 283, 181189 (2009).

6. Ren, Y., Stutzman, E., van der Hilst, R. D. & Besse, J. Understanding seismicheterogeneities in the lower mantle beneath the Americas from seismictomography and plate tectonic history. J. Geophys. Res. Solid Earth 112, B01302(2007).

7. Kendall, J.-M. & Nangini, C. Lateral variations in D0 below the Caribbean. Geophys.Res. Lett. 23, 399402 (1996).

8. Garnero, E. J., Maupin, V., Lay, T. & Fouch, M. J. Variable azimuthal anisotropy inEarths lowermost mantle. Science 306, 259261 (2004).

9. Rokosky, J. M., Lay, T. & Garnero, E. J. Small-scale lateral variations in azimuthallyanisotropic D0 structure beneath the Cocos plate. Earth Planet. Sci. Lett. 248,411425 (2006).

10. Wookey, J.&Kendall, J.-M.Constraintson lowermostmantlemineralogyandfabricbeneath Siberia from seismic anisotropy. Earth Planet. Sci. Lett. 275, 3242(2008).

11. Wookey, J., Kendall, J.-M. & Rumpker, G. Lowermost mantle anisotropy beneaththe north Pacific from differential SScS splitting. Geophys. J. Int. 161, 829838(2005).

12. Meade, C., Silver, P. & Kaneshima, S. Laboratory and seismological observations oflower mantle isotropy. Geophys. Res. Lett. 22, 12931296 (1995).

13. Karki, B., Wentzcovitch, R., de Gironcoli, S. & Baroni, S. First-principlesdetermination of elastic anisotropy and wave velocities of MgO at lower mantleconditions. Science 286, 17051707 (1999).

14. Long, M. D., Xiao, X., Jiang, Z., Evans, B. & Karato, S. Lattice preferred orientation indeformed polycrystalline (Mg,Fe)O and implications for seismic anisotropy in D0.Phys. Earth Planet. Inter. 156, 7588 (2006).

15. Yamazaki, D. & Karato, S. Fabric development in (Mg,Fe)O during large strain,shear deformation: implications for seismic anisotropy in Earths lower mantle.Phys. Earth Planet. Inter. 131, 251267 (2002).

16. Merkel, S. et al. Deformation of (Mg,Fe)SiO3 post-perovskite and D0 anisotropy.Science 316, 17291732 (2007).

17. Merkel, S. et al. Plastic deformation of MgGeO3 post-perovskite at lower mantlepressures. Science 311, 644646 (2006).

18. Oganov, A., Martonak, R., Laio, A., Raiteri, P. & Parrinello,M. Anisotropy of EarthsD0layer and stacking faults in the MgSiO3 post-perovskite phase. Nature 438,11421144 (2005).

19. Carrez, P., Ferre, D. & Cordier, P. Implications for plastic flow in the deep mantlefrommodelling dislocations in MgSiO3 minerals. Nature 446, 6870 (2007).

20. Yamazaki, D., Yoshino, T., Ohfuji, H., Ando, J. & Yoneda, A. Origin of seismicanisotropy in the D0 layer inferred from shear deformation experiments on post-perovskite phase. Earth Planet. Sci. Lett. 252, 372378 (2006).

21. Iitaka, T., Hirose, K., Kawamura, K. & Murakami, M. The elasticity of the MgSiO3post-perovskite phase in the Earths lowermost mantle. Nature 430, 442445(2004).

22. Okada, T., Yagi, T., Niwa, K. & Kikegawa, T. Lattice-preferred orientations in post-perovskite-type MgGeO3 formed by transformations from different pre-phases.Phys. Earth Planet. Inter. 180, 195202 (2010).

23. Stackhouse, S., Brodholt, J. P., Wookey, J., Kendall, J.-M. & Price, G. D. The effect oftemperature on the seismic anisotropy of the perovskite and post-perovskitepolymorphs of MgSiO3. Earth Planet. Sci. Lett. 230, 110 (2005).

24. Tsuchiya, T., Tsuchiya, J., Umemoto, K. & Wentzcovitch, R. Phase transition inMgSiO3 perovskite in the Earths lower mantle. Earth Planet. Sci. Lett. 224,241248 (2004).

25. Wookey, J., Stackhouse, S., Kendall, J.-M., Brodholt, J. P. &Price, G.D. Efficacy of thepost-perovskite phase as anexplanation for lowermost-mantle seismic properties.Nature 438, 10041007 (2005).

26. Wookey, J. & Kendall, J.-M. inPost-Perovksite: The LastMantle Phase Transition (edsHirose, K., Brodholt, J., Lay, T. & Yuen, D.) 171189 (American Geophysical Union,2007).

27. Wentzcovitch, R., Tsuchiya, T. & Tsuchiya, J. MgSiO3 postperovskite at D0conditions. Proc. Natl Acad. Sci. USA 103, 543546 (2006).

28. Blackman, D. et al. Teleseismic imaging of subaxial flow at mid-ocean ridges:traveltime effects of anisotropicmineral texture in themantle.Geophys. J. Int. 127,415426 (1996).

29. McNamara, A., van Keken, P. & Karato, S. Development of finite strain in theconvecting lower mantle and its implications for seismic anisotropy. J. Geophys.Res. Solid Earth 108, 2230 (2003).

30. Ritsema, J., van Heijst, H. J. & Woodhouse, J. H. Complex shear wave velocitystructure imaged beneath Africa and Iceland. Science 286, 19251928 (1999).

31. Evans, M., Kendall, J.-M. & Willemann, R. Automated SKS splitting and upper-mantle anisotropy beneath Canadian seismic stations. Geophys. J. Int. 165,931942 (2006).

32. Wuestefeld, A., Bokelmann, G., Barruol, G. & Montagner, J.-P. Identifying globalseismic anisotropy patterns by correlating shear-wave splitting and surface-wavedata. Phys. Earth Planet. Inter. 176, 198212 (2009).

Supplementary Information is linked to the online version of the paper atwww.nature.com/nature.

AcknowledgementsWe thank J. Brodholt and D. Dobson for comments. A.N. wassupported by NERC. Seismic data were provided by I. Bastow, D. Thompson, and theIRIS and CNSN data centres.

Author Contributions A.N. analysed the data and produced the manuscript andfigures. J.W. wrote the analysis andmodelling code and performed the modelling. J.W.and J-M.K. supervised the analysis and commented on themanuscript and figures. Allauthors discussed the results and implications at all stages.

Author Information Reprints and permissions information is available atwww.nature.com/reprints. The authors declare no competing financial interests.Readers are welcome to comment on the online version of this article atwww.nature.com/nature. Correspondence and requests for materials should beaddressed to A.N. (andy.nowacki@bristol.ac.uk).

RESEARCH LETTER

1 0 9 4 | N A T U R E | V O L 4 6 7 | 2 8 O C T O B E R 2 0 1 0

Macmillan Publishers Limited. All rights reserved2010

METHODSSScS differential splitting. We measured differential shear wave splittingbetween S and ScS phases recorded at about 500 seismic stations in North andCentral America, according to the method of ref. 10. Events of Mw$ 5.7 in thedistance range 55u82u were used (Supplementary Table 3), because the twophases then traverse very similar regions of the upper mantle. All data werebandpass-filtered between 0.001Hz and 0.3Hz to remove noise. We analysedsplitting in the phases using theminimumeigenvalue technique33, with 100 analysiswindows in each case to estimate the uncertainties in Q and dt using a statisticalF-test34,35. An example is shown in Supplementary Fig. 3. The l2 surfaces formeasurements along each azimuth are stacked34 in three regions (Fig. 2) to reducethe errors greatly.Correcting for upper mantle anisotropy. We correct for upper mantle aniso-tropy using previously published31,32 SKS splitting measurements (distance.90u)at stations that show little variation of splitting parameters with backazimuth,corresponding to simple upper mantle anisotropy, and where there are measure-mentsmade along similar backazimuths to the phases wemeasure in this study (S,ScS). These provide an estimate of the receiver-side anisotropy, and should elimi-nate the chance that lateral heterogeneity or dipping or multiple layers of aniso-tropy beneath the receiver affect our results. Analysing the splitting in S afterapplying a receiver-side correction gives an estimate of the source-side splittingbeneath the earthquake (Fig. 1b; Supplementary Table 3). For nearby stations withno available SKS measurements, measuring splitting in S while correcting for thesource anisotropy gives a receiver-side estimate. Both corrections are then applied(for shallow earthquakes; only a receiver-side correction is applied for very deepevents.550 km, assumingmantle isotropy below this depth) when analysing ScS,so that the remnant splitting occurs in ScS only, and hence results from anisotropyin D0 alone. An example of a measurement where both source and receivercorrections are applied is shown in Supplementary Fig. 4.Testing SKS splitting measurements as upper mantle anisotropy corrections.We test the validity of using SKS measurements as a correction for upper mantleanisotropy. Because the tectonic and geological processes which cause uppermantle anisotropy are unlikely to be determined by structure in D0, we can regardthe two as independent. Hence over broad, continental scales, SKS measurementswill be oriented approximately randomly, and we can check that the consistencyobserved in our results is not due to a systematic error being introduced by uppermantle anisotropy. For the Mid-Atlantic Ridge event of 2008-144-1935 (23May),we analyse the S phase at each station for which we selected reliable SKSmeasure-ments, and replace those with others taken at random. The false corrections aredetermined by allowing the correction fast orientationQcorr to vary between 0u and180u, and the delay time dtcorr between the minimum and maximum values forthose in SKS measurements used in this study (02.5 s). A uniform randomdistribution is used. Supplementary Fig. 8 shows polar histograms of Q0, theprojected fast orientation at the source, for five of the sets of false corrections.Of these, the smallest sample standard deviation sQ 5 47u. Also shown is Q0 forthe true SKS splitting parameters used (sQ 5 33u). Red bars indicate measure-ments of dt. 3.5 s, whichmay correspond to two situations. First, theymay be nullmeasurements,which frequently displayaminimumat the extremeof thepermitteddt (here, 4 s). These arise because by chance the correction applied is the same as thetotal source-side and receiver splitting combined (that is, QSKS

36. Frederiksen, A. W. et al. Lithospheric variations across the Superior Province,Ontario, Canada: evidence from tomography and shear wave splitting. J. Geophys.Res. Solid Earth 112, B07318 (2007).

37. Wolfe, C. & Solomon, S. Shear-wave splitting and implications for mantleflow beneath the MELT region of the East Pacific Rise. Science 280, 12301232(1998).

RESEARCH LETTER

Macmillan Publishers Limited. All rights reserved2010

TitleAuthorsAbstractMethods SummaryReferencesMethodsS-ScS differential splittingCorrecting for upper mantle anisotropyTesting SKS splitting measurements as upper mantle anisotropy correctionsSource-side anisotropy estimatesSource polarization measurementsS-ScS splitting from deep versus shallow earthquakesMineral slip system fitting

Methods ReferencesFigure 1 Source-receiver geometry, and explanation ofFigure 2 Multi-azimuth stacked shear wave splitting results in each region.Figure 3 Section through study region and compatible shear planes for candidate ppv slip systems.