EART

H,A

TMO

SPH

ERIC

,A

ND

PLA

NET

ARY

SCIE

NCE

S

Chemical trends in ocean islands explained byplume–slab interactionJuliane Dannberga,b,1,2 and Rene Gassmollera,b,1

aDepartment of Mathematics, Colorado State University, Fort Collins, CO 80523-1874; and bDepartment of Mathematics, Texas A&M University, CollegeStation, TX 77843-3368

Edited by Norman H. Sleep, Stanford University, Stanford, CA, and approved March 13, 2018 (received for review August 9, 2017)

Earth’s surface shows many features, of which the genesis canbe understood only through their connection with processesin Earth’s deep interior. Recent studies indicate that spatialgeochemical patterns at oceanic islands correspond to struc-tures in the lowermost mantle inferred from seismic tomo-graphic models. This suggests that hot, buoyant upwellingscan carry chemical heterogeneities from the deep lower man-tle toward the surface, providing a window to the compositionof the lowermost mantle. The exact nature of this link betweensurface and deep Earth remains debated and poorly under-stood. Using computational models, we show that subductedslabs interacting with dense thermochemical piles can triggerthe ascent of hot plumes that inherit chemical gradients presentin the lowermost mantle. We identify two key factors control-ling this process: (i) If slabs induce strong lower-mantle flowtoward the edges of these piles where plumes rise, the pile-facing side of the plume preferentially samples material origi-nating from the pile, and bilaterally asymmetric chemical zoningdevelops. (ii) The composition of the melt produced reflectsthis bilateral zoning if the overlying plate moves roughly per-pendicular to the chemical gradient in the plume conduit. Ourresults explain some of the observed geochemical trends ofoceanic islands and provide insights into how these trends mayoriginate.

mantle plumes | hotspots | bilateral zoning | geodynamic modeling |plume generation zones

When projected to the lowermost mantle, many eruptionsites of oceanic islands overlie the margins of large zones

of reduced seismic shear-wave velocities (1), the large low shearvelocity provinces (LLSVPs). Some of these islands display par-allel volcanic chains with distinct geochemical trends, exposinga different composition in the part of the dual chain closer tothe LLSVP compared with the chain farther away from it. Themost prominent example, the Hawaii-Emperor chain, overliesthe northern edge of the Pacific LLSVP. Its southern “Loa”trend has been associated with a greater prominence of recycledmafic material in the melt source compared with the northern“Kea” trend (2), and both trends show very little compositionaloverlap (3). Recent studies suggest that more ocean islands fea-ture geochemical asymmetry, including Marquesas (4, 5), Samoa(4, 6), Society (7), Galapagos (8), Easter (8), and Tristan daCunha/Gough (9, 10).

Mantle plumes—buoyant, hot upwellings from the Earth’slowermost mantle (11)—could explain this phenomenon. Theyare thought to ascend from the edges of thermochemical piles(1, 12) (corresponding to the LLSVPs) and might inherit deepmantle chemical heterogeneities. When approaching the sur-face, plumes melt due to decompression and generate Earth’soceanic island chains (13). The islands’ distinct geochemicaltrends may arise from melting of a bilaterally zoned underlyingmantle plume, one side sampling the LLSVP and the other onesampling the ambient mantle composition (2, 4, 14).

Supporting this hypothesis, geodynamic modeling studies ofthermally buoyant plumes show that if chemical gradients arepresent at the core–mantle boundary, they can be preserved

during plume ascent (14–16) and would be reflected in magmacompositions (14, 17). However, these models disregard materialentrainment from thermochemical piles into plumes and con-sider compositional heterogeneity as a passive feature withouteffects on physical properties like density. As stable thermo-chemical piles are expected to be at least 2% denser than thebackground mantle (18–22) (SI Materials and Methods), the aris-ing buoyancy difference crucially influences the entrainment andascent of chemically heterogeneous material (23, 24). A recentstudy including this effect (25) has shown that—in the absence ofambient mantle flow driven by other phenomena than the plumeitself—already small density differences focus entrainment intothe hot plume center, creating concentric zoning instead ofbilaterally asymmetric plumes. Consequently, plumes would notpreserve the spatial distribution of geochemical domains inthe lowermost mantle, suggesting that chemical anomalies atthe surface are not directly related to lower-mantle structure.Another recent study suggested that even if plume tails wereto be bilaterally zoned, melt generated from the side of theplume containing more mafic material might be dominated bymelting of peridotite due to its higher temperatures, thus revers-ing the trend present in the plume tail (26). Thus, it remainsunclear how bilateral chemical zoning develops in mantleplumes and how surface observations are linked to deep mantlestructures.

We propose that the key to solving this problem is to considerthe interaction of subducted slabs arriving at the core–mantleboundary, thermochemical piles, and starting mantle plumes.Ambient mantle flow associated with realistic plate motions

Significance

The composition of ocean island basalts is known to corre-late with zones in the lowermost mantle almost 3,000 kmbelow, thought to represent piles of hot, dense material. Oneimportant open question in the Earth sciences is the mecha-nism behind this link between the surface and Earth’s deepinterior. Here, we use high-resolution 3D geodynamic modelswith a realistic subduction history, plate geometries, and platemotions to identify a dynamically feasible mechanism for howrising hot material can inherit the lower-mantle geochemicalstructure despite variations in material density. Our findingsprovide a framework for mapping chemical anomalies at thesurface to the deep mantle and illuminate the composition ofone of the least well-understood regions of the Earth.

Author contributions: J.D. and R.G. designed research, performed research, analyzeddata, and wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.1 Present address: Department of Earth and Planetary Sciences, University of California,Davis, CA 95616.

2 To whom correspondence should be addressed. Email: judannberg@gmail.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1714125115/-/DCSupplemental.

Published online April 9, 2018.

www.pnas.org/cgi/doi/10.1073/pnas.1714125115 PNAS | April 24, 2018 | vol. 115 | no. 17 | 4351–4356

Dow

nloa

ded

by g

uest

on

June

20,

202

0

controls the location of subduction zones and the evolutionof the associated slabs in the lower mantle—a process that isthought to be intimately connected to LLSVP formation andthe origin of plumes (20, 27). This highlights a probable closelink between slabs and LLSVP processes, suggesting that slab–LLSVP–plume interaction has a significant effect on the thermo-chemical structure of plumes and that considering slabs is crucialto understanding LLSVP processes. Investigating this inherently3D interaction and the material entrainment in mantle plumesrequires high-resolution models with a realistic spatial distri-bution of slabs and their appropriate ages and is the noveltyof this study. Our findings advance the current understandingof how geochemical anomalies at the surface are related tothe deep mantle and hence provide a framework for mappingthe composition of one of the least well-understood regionsof the Earth.

Development of Chemical Zoning in PlumesTo model this process, we use the mantle convection codeASPECT (Advanced Solver for Problems in Earth’s Con-vecTion) (28, 29) to combine high-resolution 3D regional plumemodels and global mantle convection simulations. We assumean average mantle composition of 82% harzburgite and 18%recycled oceanic crust (30) and compute the material proper-ties using the software package Perple X (31, 32). The globalmodel (29) uses plate reconstructions for the last 250 My (33)to create a realistic subduction history. As found in earlier stud-ies, sinking slabs push chemically dense material (in our modelsassumed to be recycled crust) at the core–mantle boundaryinto piles near the location of today’s LLSVPs and subse-quently initiate the ascent of plumes at positions comparableto present-day hotspots (27, 33, 34). Using these simulations asvelocity, temperature, and composition boundary conditions for

Fig. 1. Time snapshot of a simulation of a bilaterally asymmetric zoned plume in the geographical setting of the South Pacific. (Top Left) The main plotshows isosurfaces of excess temperature and composition, together with arrows indicating the direction of flow in the deep mantle (black) and the directionof plate motion (red). Subducted slabs (blue) arriving from the south push material against the thermochemical pile in the lowermost mantle (green) andtrigger the ascent of a bilaterally zoned plume (orange and green). (Bottom) This process is illustrated in the drawing. Top Right Inset displays a top viewwith isosurfaces representing the fraction of melt: melt generated from average mantle (peridotite melt) shown in orange and melt generated from recycledcrust (pyroxenite melt) shown in green. The black sphere has a radius of 500 km and indicates the volume we use to compute the bilateral asymmetry numberin the melting zone Zm (SI Materials and Methods), which compares the melt composition in the two hemispheres, divided by a vertical plane parallel to thedirection of plate motion (marked by the black dashed line). Red and blue areas inside the light orange and blue dashed lines illustrate plume temperatureand composition at 500 km depth, using the same color scale as in Fig. 2A. The full model development is presented in Movie S1.

4352 | www.pnas.org/cgi/doi/10.1073/pnas.1714125115 Dannberg and Gassmoller

Dow

nloa

ded

by g

uest

on

June

20,

202

0

EART

H,A

TMO

SPH

ERIC

,A

ND

PLA

NET

ARY

SCIE

NCE

S

regional models allows us to study material entrainment fromthe piles into rising plumes in high resolution. These regionalconvection simulations feature plumes reaching the surface inthe general vicinity of the locations of four present-day hotspots,Samoa, Easter, Galapagos, and the Azores (Fig. S1), and in thefollowing text we refer to the model plumes using those names.SI Materials and Methods describes the modeling technique andhow we selected the geographic settings for our regional modelsin more detail.

Our models show that subducted slabs arriving in the lower-most mantle can induce bilateral asymmetry in rising plumes,even for significant chemical density contrasts. Examples arethe Pacific hotspots, which are surrounded by subduction zones.When reaching the core–mantle boundary, these slabs displacethe hot thermal boundary layer and push it toward a pile ofintrinsically dense, chemically distinct material, which resem-bles the Pacific LLSVP. This triggers the ascent of plumes nearthe edges of the pile (Fig. 1), with material originating fromwithin the pile rising at the side of the plume facing awayfrom the slab. Correspondingly, the top of the thermal bound-ary layer above the core–mantle boundary is predominantlypushed upward at the slab-facing side of the plume (Fig. S2Aand Movie S2). This creates an asymmetric and bilaterally zonedplume. Because of the predominantly laminar flow inside theplume, this zoning is preserved during plume ascent. The Pacifichotspots displaying bilateral zoning, such as Society, mostly lieon the northern or southern edge of the LLSVP and develop achemical gradient in the north–south direction. As the overlyingPacific plate moves toward the west-northwest—forming a large

angle between chemical gradient and plate motion—chemicaltrends in the plume remain horizontally separated (as opposedto overlying each other) (14) when the moving plate deflectsthe plume. Thus, they are visible in the generated melt (Fig. 1and Fig. S3).

Conditions for Bilaterally Asymmetric Zoning in thePlume TailA series of models with different plume temperatures and differ-ent geographical settings reveals the controlling factors for thedevelopment of zoning in rising plumes (Fig. 2). While someplumes feature a bilaterally asymmetric structure, illustratedby the fact that regions of highest plume temperature (whiteareas) and highest content of recycled crust (blue contours) donot match (Fig. 2 A and J), other plumes are concentricallyzoned, even though they rise from the edges of thermochemi-cal piles (Fig. 2 E and H). As indicated in previous studies (25),a higher plume temperature (entailing a lower buoyancy ratioB , defined as the ratio between the chemical and the thermaldensity contrast driving the flow; SI Materials and Methods) gen-erally promotes bilateral chemical zoning (Fig. 3A). However,we find appreciably bilaterally zoned plumes up to much higherbuoyancy ratios (B = 2.3, corresponding to a chemical densitycontrast of ∆ρC = 2.8% and a thermal density contrast of ∆ρth ≈1.2% between pile material and background mantle; Fig. S4)compared with the results of previous 2D models (25) (B ≈ 0.5,∆ρC = 0.75%, ∆ρth = 1.5%). In contrast to this previous study,we include the effect of ambient mantle flow controlled by pastplate motion, and we discuss the effect of this below.

A B

D E F

I JH

C

G

K

Fig. 2. Zoning of the plume tail for different plume temperatures and geographic settings. (A, B, D–F, and H–J) Plume excess temperatures (background)and fraction of recycled oceanic crust (contours) on slices through the plume tail at 500 km depth. Note that the temperature color scale is different foreach column. Bilateral asymmetry number of the plume tail Zt as plotted in Fig. 3 is given for reference. (C, G, and K) Chemical piles and velocity field in amap view at 20 km above the core–mantle boundary. Background colors illustrate the content of recycled crust at that depth: Dark blue colors represent thechemical piles, light gray colors represent background mantle, and the white areas represent the edges of the regional models. Arrows show the directionof lower-mantle flow, with darker colors representing higher velocities. The yellow and red contours illustrate the position of the plume tail at the topof the thermochemical piles. All panels are in north-up orientation, and panels in the same row show the same plumes, but for different initial plumetemperatures.

Dannberg and Gassmoller PNAS | April 24, 2018 | vol. 115 | no. 17 | 4353

Dow

nloa

ded

by g

uest

on

June

20,

202

0

Fig. 3. Plume tail bilateral asymmetry vs. (Top) excess temperature and(Bottom) the angle between the flow in the lower mantle and the edgeof the thermochemical pile. Bilateral zoning is quantified using the bilateralasymmetry number Zt , where Zt = 0 indicates no bilateral zoning, and Zt = 1indicates maximum bilateral zoning. For the detailed definition of Zt see SIMaterials and Methods. Data point colors mark different geographical set-tings (Fig. S1); for each setting we performed a series of models, varying theplume excess temperature. Both high temperatures and steep angles (strongflow toward the pile) are required for the development of bilateral zoning.The excess temperature was computed as the maximum temperature dif-ference between the plume and the mantle adiabat in a depth of 500 km(in the slices shown in Fig. 2). From spatial resolution tests, we estimate therelative error in Zt to be ∼10%. A contour plot is shown in Fig. S8, and allvalues of Zt are listed in Table S2.

High plume temperatures (and low buoyancy ratios B) alonedo not guarantee bilateral zoning near the surface (Figs. 2F and3A). The relative position of chemical pile, plume, and subductedslab additionally controls the plume tail structure. In particular,the angle of lower-mantle flow with respect to the edge of thepile is crucial (Figs. 2 C, G, and K and 3B): An arriving slab mov-ing roughly normal toward the edge of the LLSVP (Fig. 2 C andK) facilitates bilateral zoning in the plume, while for flow paral-lel to the edge of the pile (Fig. 2G) or away from it concentriczoning develops. The important difference is that strong lower-mantle flow toward the pile creates starting plumes directly atthe edge of the thermochemical pile and makes this feeder zonemuch more asymmetric (Figs. S5A and S6 G–I), creating a bilat-erally asymmetric distribution of pile material in the plume tailfor a wider parameter range, including values expected in theEarth. If the lower mantle does not (or only very slowly) flowpredominantly toward the pile (Figs. S5B and S6 A–C), plumesform farther away from the edge of the pile and entrainment ismore symmetric. The same behavior can be observed in simpli-fied 2D Cartesian models (SI Materials and Methods and Fig. S6).Consequently, not all plumes rising from the edges of LLSVPs

are bilaterally zoned; and the orientation of the bilateral asym-metry in the plume tail reflects the direction of flow in thelowermost mantle.

Conditions for Chemical Zoning in the Hotspot TrackAlthough spatially separated chemical trends are generally pre-served during plume ascent, they might end up overlying eachother in the melting region if the plume is sheared by a mov-ing plate (14, 15, 23). Whether bilateral zoning of the plumetail is reflected in a chemical gradient of melt compositionsacross the hotspot track depends on the speed (15) and direc-tion of plate motion (14). Experiments have shown that only ifthe overriding plate moves slowly (15) or if the angle betweenplate motion and the chemical gradient in the plume is large(>45◦) (14), chemical trends in the plume conduit remain hor-izontally separated and parallel volcanic chains develop distinctchemical compositions. Our models reveal the same trend: Fora plate moving roughly perpendicular to the chemical gradi-ent in the plume (Fig. 4F), vertically integrating the amountof generated melt reveals a clear spatial variability in compo-sition on a profile across the hotspot track (Fig. 4C). Con-versely, a small angle between chemical gradient in the plumeand plate motion (Fig. 4E) causes filaments of distinct chemi-cal composition that approach the base of the plate upstreamof the plume center to flow on top of the downstream fila-ments, leading to mostly vertical chemical variations in the melt-ing region. Lateral variations in chemical composition presentin the plume tail would be sampled equally by different vol-canoes and would not be visible in melt compositions (Fig.4B). Finally, concentric zoning of the plume conduit (Fig. 4D)leads to a roughly symmetric distribution of melt compositionacross the hotspot track (Fig. 4A), regardless of the direction ofplate motion.

The presented results show that slab-induced lower-mantleflow enables plumes to entrain even chemically dense materialasymmetrically, facilitating bilateral zoning up to much higherchemical density contrasts than previously thought (25), includ-ing the values expected in the Earth’s mantle. Our modelsexplain a possible cause for bilateral asymmetry of hotspots, ingeneral, and in particular in the South Pacific (Fig. S1), wherethey reproduce, for example, the observed north-side promi-nence of mafic material in the source of the Easter hotspot(8). Furthermore, our models show the observed north–southzoning of Samoa, but contrary to geochemical data (4, 6), ourmodels predict prominence of mafic material in the northernpart—toward the LLSVP—and it remains unclear why Samoafeatures the opposite trend. A potential explanation is providedby a recent study (26), showing that due to lateral temperaturedifferences, bilateral chemical trends in the plume tail might bereversed in the generated melts. An additional complexity in thissetting is posed by the subduction zone nearby. Finally, the pro-posed mechanism could explain why asymmetry has not beenfound at North Atlantic hotspots: Past lower-mantle flow mostlyparallel to the edge of the African LLSVP prevented bilateralchemical zoning in starting mantle plumes.

We note that our models do not take into account melt migra-tion or magma plumbing and its interaction with lithosphericflexure. Our results can explain the chemical trends in oceanislands only if melts do not mix completely on their pathway fromthe source region to the hotspot volcanoes.

Another shortcoming of the current model is that it fails topredict a plume near Hawaii and a number of other observedpresent-day hotspots. This implicates that we cannot test ourhypothesis for all oceanic islands. We mainly attribute this touncertainties in the used plate reconstruction (35) and the mate-rial properties of the mantle: Our modeled thermochemical pileextends farther north than the observed LLSVP (Fig. S1), a fea-ture also observed in other global convection models (27, 34).

4354 | www.pnas.org/cgi/doi/10.1073/pnas.1714125115 Dannberg and Gassmoller

Dow

nloa

ded

by g

uest

on

June

20,

202

0

EART

H,A

TMO

SPH

ERIC

,A

ND

PLA

NET

ARY

SCIE

NCE

S

A B C

D E F

Fig. 4. Conditions for bilateral zoning of generated melt. (A–C) Vertically averaged amount of melt (scale on the left axis) and melt composition(scale on the right axis; SI Materials and Methods) on a profile through the plume head (100 km downstream of the point with the highest melt-ing rate) normal to the plate velocity 40 My after the plume reaches the upper mantle. Distance in radians is counted from the model boundary,but we show only the section with nonzero melt fraction. The position of the profiles is marked in Fig. S9. Bilateral asymmetry of the generatedmelt Zm and the plume tail Zt as plotted in Fig. 3 and Fig. S3 is shown for reference, and the development of Zm over time is shown in Fig. S3.(D–F) Pathlines in the plume head colored with the content of recycled crust displayed together with isosurfaces of excess temperature and com-position. The pathlines reveal whether chemical trends remain horizontally separated (F, blue and red pathlines in distinct regions) or overlie eachother (D and E, blue and red pathlines not or only vertically separated). Corresponding slices through the plume tail are depicted in Fig. 2 A, E,and J.

This means that a plume rising from the edge of this pile wouldnot reach the surface at the present location of Hawaii. The pileshape and position are controlled by the subduction history, andrecently it has been shown that updating the plate reconstruc-tion of ref. 35 allows it to achieve a more realistic location ofthe Hawaiian plume (36). Hence, accurate models of past platemotion are invaluable for relating surface observations to thedeep mantle.

Although our models did not include plumes close to Hawaiior Tristan da Cunha, they provide a general mechanism thatexplains their observed bilateral asymmetry. Hawaii’s high excesstemperature and the presumed steep angle between lower-mantle flow and LLSVP (see, e.g., the models of ref. 36, whichpredict southward lower-mantle flow toward the northern edgeof the Pacific LLSVP for the last 100 My) give reason to expectthe development of bilateral chemical zoning by the mecha-nism proposed in this study. Previously, it was not obvious whyTristan da Cunha and Gough show the observed dual chainswith distinct chemical trends (9, 10), as they are thought to haveonly a moderate excess temperature (37, 38). However, our pro-posed mechanism for asymmetric entrainment suggests that thesupposedly strong lower-mantle flow from the South Americansubduction zone (36) could provide sufficient support to developbilateral asymmetry in the plume.

We conclude that only certain conditions allow mappinggeochemical anomalies from the surface to the core–mantleboundary. Considering the directions of current plate motionand past lower-mantle flow is crucial for inferring lower-mantle

compositional heterogeneities from surface observations. Hence,accurate plate reconstructions and global flow models inferredfrom seismic tomography and advected backward in time mightprove to be valuable tools for deciphering the present and pastlower-mantle chemical structure.

Code AvailabilityThe geodynamic models were computed with the open-sourcesoftware ASPECT (aspect.dealii.org) (28, 29) and visualized withthe open-source programs VisIt (https://wci.llnl.gov/simulation/computer-codes/visit) and GMT (39). The version of ASPECTwe used to run our models is available online (https://github.com/gassmoeller/aspect/releases/tag/plume-zonation).

Data AvailabilityAll of the input files that are required to reproduce this study areprovided in Datasets S1–S19.

ACKNOWLEDGMENTS. We thank Garrett Ito for his very helpful commentsand the PNAS editor and two anonymous reviewers whose suggestionshelped improve and clarify this article. The computational resources wereprovided by the North-German Supercomputing Alliance (HLRN) as partof the project “Plume-Plate interaction in 3D mantle flow—Revealing therole of internal plume dynamics on global hot spot volcanism.” J.D. andR.G. were partially supported by the Computational Infrastructure for Geo-dynamics initiative, through the National Science Foundation (NSF) underAwards EAR-0949446 and EAR-1550901, administered by The University ofCalifornia, Davis, and the NSF under Award OCI-1148116 as part of theSoftware Infrastructure for Sustained Innovation program.

1. Burke K, Steinberger B, Torsvik TH, Smethurst MA (2008) Plume generation zonesat the margins of large low shear velocity provinces on the core–mantle boundary.Earth Planet Sci Lett 265:49–60.

2. Weis D, Garcia MO, Rhodes JM, Jellinek M, Scoates JS (2011) Role of the deep man-tle in generating the compositional asymmetry of the Hawaiian mantle plume. NatGeosci 4:831–838.

3. Abouchami W, et al. (2005) Lead isotopes reveal bilateral asymmetry and verticalcontinuity in the Hawaiian mantle plume. Nature 434:851–856.

4. Huang S, Hall PS, Jackson MG (2011) Geochemical zoning of volcanic chains associatedwith Pacific hotspots. Nat Geosci 4:874–878.

5. Chauvel C, et al. (2012) The size of plume heterogeneities constrained by Marquesasisotopic stripes. Geochem Geophys Geosyst 13:Q07005.

Dannberg and Gassmoller PNAS | April 24, 2018 | vol. 115 | no. 17 | 4355

Dow

nloa

ded

by g

uest

on

June

20,

202

0

6. Jackson M, et al. (2014) Helium and lead isotopes reveal the geochemical geometryof the Samoan plume. Nature 514:355–358.

7. Payne JA, Jackson MG, Hall PS (2013) Parallel volcano trends and geochemicalasymmetry of the Society Islands hotspot track. Geology 41:19–22.

8. Harpp KS, Hall PS, Jackson MG (2014) Galapagos and Easter: A tale of two hotspots.Galapagos A Nat Lab Earth Sci 204:27–40.

9. Rohde J, et al. (2013) 70 Ma chemical zonation of the Tristan-Gough hotspot track.Geology 41:335–338.

10. Hoernle K, et al. (2015) How and when plume zonation appeared during the 132 Myrevolution of the Tristan hotspot. Nat Commun 6:7799.

11. Morgan W (1971) Convection plumes in the lower mantle. Nature 230:42–43.12. Torsvik TH, Smethurst MA, Burke K, Steinberger B (2006) Large igneous provinces

generated from the margins of the large low-velocity provinces in the deep mantle.Geophys J Int 167:1447–1460.

13. Morgan WJ (1981) Hotspot tracks and the opening of the Atlantic and Indian oceans.Oceanic Lithosphere 7:443–487.

14. Farnetani CG, Hofmann AW, Class C (2012) How double volcanic chains sam-ple geochemical anomalies from the lowermost mantle. Earth Planet Sci Lett359:240–247.

15. Kerr RC, Meriaux C (2004) Structure and dynamics of sheared mantle plumes.Geochem Geophys Geosyst 5:Q12009.

16. Farnetani CG, Hofmann AW (2009) Dynamics and internal structure of a lower mantleplume conduit. Earth Planet Sci Lett 282:314–322.

17. Farnetani CG, Hofmann AW (2010) Dynamics and internal structure of the Hawaiianplume. Earth Planet Sci Lett 295:231–240.

18. Tackley PJ (2013) Three-dimensional simulations of mantle convection with a thermo-chemical basal boundary layer: D”? The Core–Mantle Boundary Region, eds GurnisM, Wysession ME, Knittle E, Buffett BA (Am Geophys Union, Washington, DC), pp.231–253.

19. Deschamps F, Tackley PJ (2009) Searching for models of thermo-chemical convec-tion that explain probabilistic tomography. II–Influence of physical and compositionalparameters. Phys Earth Planet Inter 176:1–18.

20. Tan E, Leng W, Zhong S, Gurnis M (2011) On the location of plumes and lateral move-ment of thermochemical structures with high bulk modulus in the 3-D compressiblemantle. Geochem Geophys Geosyst 12:Q07005.

21. Tackley PJ (2012) Dynamics and evolution of the deep mantle resulting from thermal,chemical, phase and melting effects. Earth Sci Rev 110:1–25.

22. Mulyukova E, Steinberger B, Dabrowski M, Sobolev SV (2015) Survival of LLSVPs forbillions of years in a vigorously convecting mantle: Replenishment and destruction ofchemical anomaly. J Geophys Res Solid Earth 120:3824–3847.

23. Farnetani C, Samuel H (2005) Beyond the thermal plume paradigm. Geophys Res Lett32:L07311.

24. Lin SC, van Keken PE (2006) Dynamics of thermochemical plumes: 1. Plume formationand entrainment of a dense layer. Geochem Geophys Geosyst 7:Q02006.

25. Jones T, Davies D, Campbell I, Wilson C, Kramer S (2016) Do mantle plumes pre-serve the heterogeneous structure of their deep-mantle source? Earth Planet Sci Lett434:10–17.

26. Ballmer MD, Ito G, Cheng C (2015) Asymmetric dynamical behavior of thermochemicalplumes and implications for Hawaiian lava composition. Hawaiian Volcanoes: FromSource to Surface, eds Carey R, Cayol V, Poland M, Weis D (American GeophysicalUnion, Washington, DC; Wiley, Hoboken, NJ), pp 35–57.

27. Steinberger B, Torsvik TH (2012) A geodynamic model of plumes from the margins oflarge low shear velocity provinces. Geochem Geophys Geosyst 13:Q01W09.

28. Kronbichler M, Heister T, Bangerth W (2012) High accuracy mantle con-vection simulation through modern numerical methods. Geophys J Int 191:12–29.

29. Heister T, Dannberg J, Gassmoller R, Bangerth W (2017) High accuracy mantle con-vection simulation through modern numerical methods. II: Realistic models andproblems. Geophys J Int 210:833–851.

30. Xu W, Lithgow-Bertelloni C, Stixrude L, Ritsema J (2008) The effect of bulk com-position and temperature on mantle seismic structure. Earth Planet Sci Lett 275:70–79.

31. Connolly J (2009) The geodynamic equation of state: What and how. GeochemGeophys Geosyst 10:Q10014.

32. Stixrude L, Lithgow-Bertelloni C (2011) Thermodynamics of mantle minerals-II. Phaseequilibria. Geophys J Int 184:1180–1213.

33. Bower DJ, Gurnis M, Seton M (2013) Lower mantle structure from paleogeo-graphically constrained dynamic Earth models. Geochem Geophys Geosyst 14:44–63.

34. Hassan R, Flament N, Gurnis M, Bower DJ, Muller D (2015) Provenance of plumes inglobal convection models. Geochem Geophys Geosyst 16:1465–1489.

35. Seton M, et al. (2012) Global continental and ocean basin reconstructions since200My. Earth Sci Rev 113:212–270.

36. Hassan R, Muller RD, Gurnis M, Williams SE, Flament N (2016) A rapid burst in hotspotmotion through the interaction of tectonics and deep mantle flow. Nature 533:239–242.

37. Putirka K (2008) Excess temperatures at ocean islands: Implications for mantlelayering and convection. Geology 36:283–286.

38. Gassmoller R, Dannberg J, Bredow E, Steinberger B, Torsvik TH (2016) Major influenceof plume-ridge interaction, lithosphere thickness variations, and global mantle flowon hotspot volcanism—The example of Tristan. Geochem Geophys Geosyst 17:1454–1479.

39. Wessel P, Smith WHF, Scharroo R, Luis J, Wobbe F (2013) Generic mapping tools:Improved version released. Eos Trans Am Geophys Union 94:409–410.

4356 | www.pnas.org/cgi/doi/10.1073/pnas.1714125115 Dannberg and Gassmoller

Dow

nloa

ded

by g

uest

on

June

20,

202

0