review article

1
Mantle geochemistry: the message
from oceanic volcanism
A. W. Hofmann

Basaltic volcanism 'samples' the Earth's mantle to great depths, because solid-state convection transports deep
material Into the (shallow) melting region. The Isotopic and trace-element chemistry of these basalts shows that the
mantle contains several Isotopically and chemically distinct components, which reflect Its global evolution. This
evolution is characterized by upper-mantle depletion of many trace elements, possible replenishment from the deeper,
less depleted mantle, and the recycling of oceanic crust and lithosphere, but of only small amounts of continental
material.
I
Geochemists began to study the composition and evolution of the differences between MORB and OIB appeared to indicate that the II
Earth's mantle in the 1960s, when it became clear that plate tectonics former come from an upper, depleted layer, whereas the latter rise J
is driven by solid-state mantle convection, which carries deep- from a deeper, undepleted layer (see, for example, ref. 12). This is I
mantle material upwards until it begins to melt (at about 30- the geochemical 'standard model' shown in Fig. 1a. Apparent 1
100 km depth). This melt rises to the surface where it extrudes as support came from geophysical evidence showing a sharp increase II

basaltic lava and delivers a chemical 'message' from the mantle to in the velocity of seismic waves at a depth of 660 km (the 660-km
the geochemist. seismic discontinuity). This indicated that the mantle consists of
When a mantle region melts, it loses incompatible (see Box 1 for two layers, which differ in either chemical or mineralogical com-
explanation of professional jargon) trace elements, such as position or both.
uranium, thorium and potassium, to the melt, which transports This standard model is now untenable because OIB are highly
them to the oceanic or continental crust. The continental crust now variable in composition, so that they could not be derived from a
contains about half of the Earth's total inventory of Th and U, chemically uniform, primitive layer of the mantle. This has led to a
leaving at least the upper mantle correspondingly depleted. This proliferation of speculative models about the origin and dis-
depleted region is sampled globally by basalts along the 60,000-km- tribution of the different types of material in the mantle. Some of
long mid-ocean-ridge system. Other types of volcanism are fed by these have retained the concept of a two-layer mantle (Fig. 1b): the
less depleted or even enriched regions in the mantle. A mantle two convecting layers are kept separate by an intrinsic com-
region can 'remember' its depletion or enrichment history, because positional density contrast or by a negative pressure-temperature
the accumulation of daughter products of radioactive decay records slope of an isochemical phase transition at 660 km (ref. 13). Plumes
the changes in parent-to-daughter concentration ratios which occur originate from the base of the upper layer but may entrain material
during melt removal (or addition). Because of this, isotopic 'leaking' from the lower layer 14 .1 5 • Other models are based on 'whole-
measurements have become the major (though not the only) tool mantle' convection (see, for example, ref. 16 and Fig. 1c). In this
of the mantle geochemist. This allows us to 'map' the geochemical class of models, subducted lithosphere penetrates the 660-km
structure of the Earth's interior by interpreting the isotopic com- boundary, and all plumes originate from the core-mantle boundary.
position of mantle-derived basalts. Thus isotope geochemistry, Recently, hybrid models have been developed in which the two,
having established the chronology of the Earth's accretion and normally separately convecting layers are intermittently perturbed
continent formation, is now beginning to decipher its internal when sinking pieces of subducted lithosphere and rising plumes
differentiation history as well. occasionally penetrate the 660-km boundary (Fig. 1d; refs 5, 17, 18).
I review here the main geochemical evidence bearing on mantle It will be seen further below that geochemical mass-balance con-
structure and processes, and some of the models based on it. I siderations strongly favour some type oflayered convection (models
conclude that much of the chemical heterogeneity evident in in Fig. 1b or d).
oceanic basalts is caused by recycling of oceanic, and to a much
lesser extent continental, crustal material, which resurfaces pre- Geological background
dominantly in the 'mantle plumes' that create volcanic islands. The following geological observations and inferences derived from
Whether these plumes rise from the base of the upper mantle or the them provide a framework for interpreting the geochemical
base of the lower mantle remains a hotly debated issue. Earlier messages from the mantle:
reviews of this and related topics may be found in refs 1- 9. ( 1) Mid-ocean ridges migrate laterally in a pattern controlled
largely by the surface geometry of the plates. Therefore, they should
Overview of mantle models normally 'sample' the uppermost mantle (that is, the
In the 1970s, geochemists developed the idea of a chemically layered asthenosphere), which rises passively under the spreading ridge.
mantle, with an upper depleted and a lower undepleted (also called Occasionally, a migrating ridge intersects a plume, which rises from
primitive) layer, because it was found that there is a fundamental the deeper mantle. When this happens, such as on Iceland, oceanic
geochemical difference between mid-ocean-ridge basalts (MORB), crust of anomalous chemistry and thickness is produced.
which come from mantle regions that are depleted in all incom- (2) Mantle plumes are thought to generate about 20- 40 stationary
patible elements, and ocean island basalts (OIB), which are less 'hotspots', which are also used as a reference frame to determine
depleted or even enriched. Ocean islands often form long lines of absolute plate motions. Plumes probably originate from boundary
volcanoes of progressively increasing ages, which are created by an layers in the mantle7' 19' 20, which may be located either above the 660-km
apparently stationary heat source ('hotspot' ) under a moving seismic discontinuity or above the core- mantle boundary (at 2,900 km ).
plate10• Morgan 11 proposed that hotspots are created by narrow In either case, heating from below lowers the density until the layer
'plumes' rising from the deep mantle. The chemical and isotopic becomes unstable and forms a rising column or 'plume'.

NATURE IVOL 3851 16 JAN UARY 1997 219

review article
(3) Not all intra-plate oceanic volcanoes originate from plumes.
Box 1 Geochemical and geophysical jargon Thousands21 of small, isolated volcanic seamounts litter the oceanic
crust and are probably formed by local melting anomalies. In
Asthenosphere. Relatively soft, hot mantle region (at or near the melting addition, there are 'hot lines' of volcanoes that show no age
point) underlying the colder, harder lithosphere. Its thickness is of the progression but erupt roughly simultaneously along the entire
order of 100 km but is poorly defined because stiffness increases extent of the line (for example, the Cameroon Line volcanoes22 ) .
gradually with depth. Most of these geochemical and melting anomalies probably come
Depleted (enriched) reseiVolr. Region in the mantle that is depleted from the upper mantle.
(enriched) in incompatible elements relative to a primitive reservoir. (4) Continental flood basalts and oceanic plateaus are, according
Hotspot. Locus of volcanism that remains nearly stationary relative to the to some authors23 , formed by the surfacing of a large, mushroom-
moving lithospheric plates. Classic hotspots form long chains of like head of a 'starting' plume, which has entrained large amounts of
volcanoes (such as the Hawaiian-Emperor chain) which become mantle material on the way from its origin to the surface. Often, a
progressively older as a function of distance from the presently active progressively 'younging' line of volcanoes connects an old flood
volcanism. basalt province (for example, the Deccan Traps of India) with a
currently active hotspot (for example, Reunion Island). This is
Incompatible element. A chemical element that is excluded from solid
interpreted as the surface expression of the much narrower conduit
minerals of the upper mantle and therefore preferentially. enters an
or 'stem' of the plume, which may be active for over a hundred
available melt phase. Reasons for incompatibility include large ionic
million years.
radius and high ionic charge, both of which cause misfits in the existing
(5) Sea-floor spreading generates large chemical heterogeneities
crystal structures of mantle minerals. Incompatible elements have, by
by extracting melt from the mantle, thereby forming a basaltic crust
oefinition, low partition coefficients, usually D.;; 0.1.
and a refractory residue. Subduction reinjects these heterogeneities
Lithosphere. Stiff, cold layer comprising the uppermost mantle and back into the mantle, where they are gradually rehomogenized by
overlying crust. The lithosphere is about 100 km thick in oceanic regions convection24 • Subduction also causes melting, resulting in volcanic
and 100-400 km in continental regions. The moving plates created by sea- 'island arcs', which are ultimately accreted to the continental crust.
floor spreading consist of lithosphere.
Mantle plume. Solid-state, narrow upwelling current in the mantle witH a The geochemical evidence
diameter of the order of 100 km and originating from a hot, low-density To characterize the mantle source regions of basaltic lavas, the
boundary layer located either above the seismic discontinuity at 660 km geochemist uses a variety of geochemical tracers. Such tracers are
depth or near the core-mantle boundary at 2,900 km depth. either isotope abundance ratios of daughter elements of radioactive
nuclides, or concentration ratios of incompatible trace elements.
Metasomatism. Process of changing the bulk chemical composition of a
The melt 'copies' these tracer ratios from the source and delivers
rock, usually by infiltration of an aqueous or carbonaceous fluid or melt.
them to the surface with little or no distortion. In spite of occasional
MORB. Mid-ocean-ridge basalt. MORB forms the upper part of the assertions to the contrary, the geochemical tracing approach has
oceanic crust. withstood the test of time quite well (see, for example refs 1 and 25)
018. Ocean island basalt, restricted to islands that are not related to and will be used here without apologies. Table l shows the radio-
subduction. Typical 018 occur on Hawaii, Iceland and the Polynesian active decay systems used in this Review, their half-lives, and
islands, but not in so-called island arcs such as the Al.eutians, Marianas or (radiogenic to non-radiogenic) isotope ratios of the daughter
Japan. elements, which will vary as a function of age and parent/daughter
Partition coefficient. Description of the equilibrium distribution of a trace element ratio (discussed below). I focus here purely on oceanic
element between different phases. According to Henry's law, the partition basalts (MORB and OIB) rather than mantle xenoliths or volcanic
coefficient is independent of the absolute concentration of the trace rocks from island arcs or continental regions, because oceanic
element. For solid-melt systems, the coefficient is defined as basalts represent relatively large volumes of mantle and carry the
D "' C501"'tcm•", where C is the concentration of the trace element in smallest risk of being contaminated during magma transport
question. through the crust to the surface.
Elemental chemistry of MORB, OIB and continental crust.
Primitive mantle (sometimes called 'bulk silicate Earth'). Silicate portion
General chemical characteristics of average continental crust, aver-
of the Earth as it existed after. separation of the core but before it was
age oceanic crust (= MORB), and selected OIB are shown in Fig. 2.
differentiated into crust and present-day mantle. A primitive reseiVolr is
Concentrations are normalized to those of the primitive mantle
any region of the mantle that has retained this composition. The
and plotted in the order of increasing compatibility, which, except
composition of the primitive mantle is inferred from the composition of
for the 'anomalous' elements Nb, Pb and Ti (see below), coincides
the stony meteorites known as chondrites'24·' 25.
with the order of descending abundances in the continental crust.
Recycling. Introduction of (oceanic or continental) crust or lithosphere All types of crustal rock (MORB, OIB and continental crust) are
into the mantle by subduction or delamination, storage in the mantle, and enriched in incompatible elements (relative to the primitive mantle)
reappearance in volcanism. because they originate as melts, which concentrate incompatible
Sea-floor spreading. Creation of new oceanic lithosphere by solid- elements from the mantle. However, in the oceanic crust
mantle upwelling, consequent partial melting, extrusion and cooling of (= MORB), the most highly incompatible elements (for example,
MORB at a mid-ocean ridge, and lateral movement of the new lithosphere Rb, Ba and Th ) are actually relatively depleted in comparison with
away from the ridge. the moderately incompatible elements (for example Sm and Hf),
because this crust is formed by melting of a mantle region that has
Subduction. Destruction of lithosphere by sliding back into the mantle
previously been depleted by extraction of the continental cruse .
beneath island arcs, often at an angle of 45• and reaching depths of at
This effect also explains the mutually complementary positive and
least 600 km.
negative anomalies of Nb and Pb.
66G-km seismic discontinuity. A narrow zone of sharp increase in Ocean island basalts are much more enriched in incompatible
seismic wave velocity at a depth of 660 km. It may be caused by a elements than MORB, and they show a large variation in com-
mineralogical phase change of upper-mantle silicates to denser struc- position. This variability will be especially obvious in the isotopic
tures, or by a sharp change in major-element composition. compositions discussed below. Here I show average concentration
patterns for basalts from Mauna Loa (Hawaii) and two of the

220 I
NATURE VO L 385 116 JAN UARY 1997
 review article
isotopically 'extreme' types of mantle plume (see below and Figs 3- Two features are noteworthy. ( 1) The Pb data do not form a
6), namely Tubuai, Austral Islands (HIMU) and Tristan-Inacces- simple trend extending from depleted MORE through OIB to the
sible Island (EM-1). (See Isotope taxonomy section, below, for continental crust. Thus, OIB sources cannot be explained simply by
explanation of HIMU, EM -1 and EM-2). Their greater enrichments back-mixing of continental crust (as might have been inferred from
in incompatible elements cause them superficially to resemble the Sr-Nd-Hf data). (2) Given the relative compatibilities ofU and
continental crust, but their Nb and Pb anomalies are similar to Pb from Fig. 2, Pb isotope data in Fig. 4a for MORE should fall well
those of the oceanic crust (=MORE) and opposite to those of the to the left of the 'geochron' (locus of primitive-mantle Pb composi-
continental crust. Thus, except for their grossly different relative tions). In other words, the Pb data are expected to show roughly the
levels of enrichment (which can be explained by differences in the same topology as the Sr-Nd data in Fig. 3, but in fact most MORE
melting process), all these basalt types have fundamentally similar data are roughly centred on the geochron, or even lie on its right-
trace-element patterns. In contrast, the trace-element patterns of hand side. Moreover, the bulk continental crust (located somewhere
EM-2 basalts (not shown for reasons of clarity) differ from those of between UCC and LCC) also lies close to the geochron rather than
other OIB and resemble that of continental crust (see section on
trace-element ratios).
Isotope chemistry. The radioactive parent nuclides listed in Table 1 a MOAB OIB Arc
are all very long-lived, so the daughter isotope ratios reflect the o km
Continent
Continental
long-term history of the basalt-source reservoirs in the mantle. lithosphere
Depleted
Figures 3-6 show Sr, Nd and Pb isotope data for over 1,100 samples upper mantle
660 km
of MORE and OlE.

c
Strontium, neodymium and hafnium. Isotope data for Sr and Nd in
MORE, OIB and continental crust are summarized in Fig. 3. The
MORE points form a relatively tight cluster that defines the upper
left-hand corner of the array. In general, 143 Nd/ 144 Nd correlates
~1 Primitive
lower mantle

negatively with 87 Sr/ 86 Sr, and positively with 176Hf/ 177 Hf (not
shown). The isotopic composition of the primitive mantle,
marked by a square, is normally estimated from this correlation
and the Nd value of meteorites 26 • These correlations are consistent 2,900 km
Liquid outer core
with the relative compatibilities shown in Fig. 2, namely Rb < Sr, MORB OIB Arc
b
Sm > Nd, and Lu > Hf. Thus, mantle regions depleted in incom- Okm Continental
patible elements have low Rb/Sr and therefore low 87 Sr/ 86Sr ratios, lithosphere
Depleted
coupled with hifh Sm/Nd and Lu/Hf, and therefore high upper mantle
143 Nd/ 144 Nd and 76 Hf/ 177 Hf ratios. Continental crust, which is 660 km
Entrainment

c
enriched in incompatible elements, shows the opposite isotopic
signatures. The position of OIB between the depleted MORE source
and the continental crust suggests that the OIB sources might just be Primitive
lower mantle
the result of back-mixing of various types of continental material
into the mantle. It will be seen below that this is, for the most part,
not the case.
Lead. The Pb isotope data (Fig. 4) are more complex than those for
2,900 km
Sr, Nd and Hf, in part because the geochemical behaviour oflead is Liquid outer core
anomalous (see below and Fig. 2). The U/Pb ratio of the Earth c MORB OIB Arc

cannot be inferred directly from meteorite data because the silicate Okm

portion of the Earth is severely depleted in lead (either because it has

been lost from the Earth along with other volatile elements or
660 km
because much of the lead has entered the core )27 • However, possible
primitive mantle compositions are constrained to lie along a line
(the 'geochron') determined by the age of the mantle. Also shown
are estimates for the average upper (UCC) and lower (LCC) ~1
continental crust.
I I
I I
//
I I

Figure 1 Models of mantle circulation. a, Old 'standard model' of two-layer Recycling ? /

II

circulation. The upper layer has been depleted in incompatible elements by the 2,900 km
Liquid outer core
formation of continental crust. The layers are separated by either an endothermic
phase change or by an intrinsic dens1ty contrast at a depth of 660 km. Plumes rise d MORB 018 Arc

from the base of the lower, primitive (or less depleted) layer. The counterflow into 0 km

the lower layer is not specified. (MORB, mid-ocean-ndge basalt; OIB, ocean
Upper mantle
island basalt.) b, Two-layer circulation with nearly complete isolation between
660 km
upper and lower layers. Plumes rise from the base of the upper layer but may
entrain small amounts of material from the lower layer. Plume sources are created
by recycling of oceanic or continental lithosphere, which creates reservoirs that
are enriched in incompatible elements. c, Whole-mantle, single-layer circulation ~1
with plumes rising from the core-mantle boundary. Plume sources are created as Lower mantle

in b. d, Hybrid model, with circulation occurring primarily in two separate layers

with small plumes rising from 660 km depth as in b, but with limited exchange by
means of occasional foundering of subducted lithosphere and rise of strong
plumes from the core-mantle boundary. Liquid outer core

NATURE IVOL 385116 JANUARY 1997 221

review article

0 Cont. crust EM-1 (Tristan) 0 FOZO, C • Pac. MORB

__...,._ MORB - HIMU (Tubuai) • HIMU o Atlan . MORB
- Hawaii o EM-2 • Indian MORB
• EM-1 + OtherOIB
Nb

100
0 .5134

c
0 0 .5132
-~
-o
cQ) 10
z
"<!"
0.5130
u "<!"
c
u
0
"0
Q)
-~
--z
-o
0 .5128

(ij
C') 0 .5126
"<!"
E
0 Primitive mantle
z 0 .5124

0.5122
•
0 .702 0 .704 0 .706 0 .708

~ Th U ~ ~ P ~ TI ~ 8 ~ ~ ~
Ba Nb La Pb Sr Sm Eu Gd Y Lu Ca Si

Compatibility
Figure 3 Nd and Sr isotopic compositions, assembled from literature sources, of
MORB and 0 18, w ith extreme HIMU, EM-1 and EM-2 sam ples marked in red,
Figure 2 Concentrations of selected trace and major elements, arranged in the brown and yellow colours, respectively. Only those samples are shown for w hich
order of ascending compatibility and normalized to primitive-mantle concentra- Pb isotope data are also available (see Figs 4-6). HIMU samples are arbitrarily
tions, in average continental crust'". average MORB4 , average Mauna Loa, defined by having 206 PbJ204 Pb .:; 20 (Fig . 4). EM -1 and EM-2 samples are also
Hawaii 119, and three types of OIB: average Tristan and Inaccessible islands arbitrarily defined by their low 143Nd / 144 Nd and high 87 Sr/ 86Srvalues as shown on
representing EM-1120 ·m , and Tubuai representing HIMU islands49. The patterns this diagram. The arrow points to the com position of average continenta l crust
for MORB and the oceanic island basalts differ by their different enrichments (or ( 87 Sr/86 Sr = 0.72, 143 Nd/144 Nd = 0.5118). A lso marked are the compositions repre-
depletions) in incompatible elements, but they are sim ilar with respect to t heir senting the primit ive mantle (PRIMA) and t he proposed common mantle compo-
positive Nb and negative Pb anomalies. The continental crust has opposite Nb nents of most plumes 'FOZO" ' and 'C' 40 . See Supplementary Information for
and Pb anomalies. A fourth type of OIB, called EM-2, is similar to continental crust references used in compiling Figs 3- 7.
and has been omitted for c larity. (See text for explanations of EM-1, EM-2 and
HIMU.)

far on its right-hand side. A solution to this puzzle (the 'lead excellent tracer for recycled basaltic material in the mantle and
paradox' 28 ) will be suggested below. current results have generally been interpreted in this way. Mantle
Correlations between Pb and other isotopes. Figure 5 shows relation- xenoliths from ancient subcontinental lithosphere have consistently
ships between 206PbP04Pb and Sr isotope ratios. Although these lower-than-primitive 1870s1 1880s values36, and this constitutes an
isotope ratios correlate positively in MORB from the Atlantic and excellent potential tracer for recycled ancient lithosphere. However,
Pacific oceans, they do not correlate in MORB from the Indian no such low Os values have been found in oceanic basalts as yet.
Ocean and in the overall OIB- MORB array. This indicates that Noble gas isotopes. The radiogenic nuclides 4He and 40Ar are
U-Pb and Th-Pb systems in the different mantle reservoirs evolve produced by the decay of U and Th, and 4°K, respectively, and
rather differently from the Rb-Sr (and Sm-Nd and Lu- Hf) m ay be degassed from the Earth's interior into the atmosphere. The
systems. accumulation of 40Ar in the atmosphere will be further discussed in
In contrast, the parameter 208 Pb*I 206Pb* (the ratio of the radio- the section on mass-balance considerations. Oceanic basalts also
genic additions to the initial terrestrial lead, defined as ie08 Pbl contain the non-radiogenic isotopes 3He and 36Ar (as well as Ne, Kr
204Pb) - eospb I zo4Pb)init. l I{ eo6pb I zo4Pb) - eo6Pb I zo4Pb)imt. l and Xe). Atmospheric helium is not recycled into the mantle
does correlate with 143Ndi 144Nd (and with Sr and Hf isotopes); see because it is continually lost to space from the atmosphere, and
Fig. 6. This parameter depends primarily on ThiU (and age) but not nearly all 3He now coming out the mantle is primordial, thus
on UIPb or ThiPb. The topology of Fig. 6 is rather similar to that of demonstrating that the Earth has never been completely degassed.
Fig. 3. This means that ThiU correlates with SmiNd (as well as with (This nearly universally accepted view has been challenged by
LuiHf and RbiSr) in mantle evolution, but UIPb and ThiPb do not. Anderson37, who suggested that cosmic-dust-derived 3He has been
This identifies Pb (rather than Thor U) as the anomalous element29 • supplied to the mantle through subduction of pelagic sediments.
Osmium. Recent improvements in analytical methods have made However, high 3Hei4 He ratios do not correlate with other tracers of 1

osmium isotope analyses of oceanic basalts feasible. The primitive- recycled subducted sediments in OIB38 , a fact that seriously weakens
mantle value of 1870sl 1880s (or 1870sl 1860s) is identical or similar to Anderson's argument.)
the meteorite value30 •3 1• This means that the Earth's mantle has the Relative to the atmospheric 3Hei4He ratio (RA CHei4 He).,m =
same ReiOs ratio as meteorites, which is best explained by con- = 1.4 X 10 - 6 ), continental crust has low ratios CHe/ 4 He"'= 0.01 RA)
tinued infall of meteoritic material after the main accretion and core because it is enriched in Th and U, which produce 4 He during decay.
formation of the Earth. Current results indicate the presence of MORB have rather uniform values of (8 ::t: l)RA (ref. 39), and OIB
higher-than-primitive 1870si 1880s in many OIB32 - 35 • Basalts have range from 5 to 30~ (ref. 40). The existence ofhigh-3Hei4He islands
much higher (up to a factor of a hundred or more) ReiOs ratios than (Hawaii, Iceland, Bouvet, Galapagos, Easter, Juan Fernandez,
their mantle sources, so that growth of 1870sl 188 0 s in the oceanic Pitcairn, Samoa, Reunion and Heard islands; see, for example,
crust is quite rapid. Therefore, Os isotopes should furnish an ref. 40) is consistent with the layered-mantle model: plumes

222 NATURE! VOL 385 ! 16 JAN UARY 1997

 review article
might rise directly from the lower, high-jHetHe reservoir through enriched and a more depleted component originally thought to be
the upper, low- 3 He/4 He layer38 ' 41 - 43 , or they might start from the the DMM. However, recently published data (for example, refs 38,
base of the upper layer, in which case the helium must migrate into 48, 49) show arrays in which DMM is clearly not a mixing end-
the plume source from the lower mantle2.44 • member. This has led to the definition of a new component, called
Neon isotope data from basalts are still scarce because of experi- FOZO ('focal zone') 16, whose composition is defined by the point of
mental difficulties. Recent work has shown that Ne from the convergence, in three-dimensional isotope diagrams, of linear data
Hawaiian plume and from MORB have different 2 1Ne/22Ne and arrays for individual ocean islands. A similar point of convergence
20 Ne/22Ne correlations, indicating the presence of a high- 20Ne of isotope arrays for MORB defines the component C ('common') 40 .
reservoir45 in the plume source and a high- 21 Ne reservoir in the In the two-dimensional diagrams shown here, the convergence of
MORB source46 • Both components form mixing lines with a HIMU and MORB arrays is most clearly seen in Fig. 4b.
low- 21 Ne/22Ne and 20Ne/22Ne reservoir of atmospheric origin. This There is much current debate about the meaning of this taxon-
provides additional evidence for a layered mantle with restricted omy. Authors disagree as to whether the linear isotopic trends seen
exchange between layers. in many hotspots are caused by the existence of two components in
Isotope taxonomy. The deviations from simple linear correlations in the plume sourceM9, whether isotopically distinct but internally
Figs 3- 6 are not random but show systematic regularities 1' 14 ' 47 , uniform plume sources are mixed during ascent by entraining
which are illustrated in these figures by colour coding. For example, relatively depleted lower-mantle materiaF 6 ' 18 , possibly producing a
all samples possessing 206 PbP04 Pb :;, 20 (Fig. 4) are shown as red zoned plume 50 , or whether similarly uniform plumes are mixed only
dots and labelled 'HIMU' (see key in Fig. 3). They also form well- with upper-mantle asthenosphere and/or lithosphere5 1• In general,
defined clusters on isotope diagrams not involving the element lead individual hotspots contain only a limited range of isotopic com-
(Figs 3, 5, 6), even though they do not form an obvious endmember positions and most hotspots have distinct isotopic 'flavours'. On a
in them. Samples with this 'isotopic colour' occur in several oceanic much larger geographical scale, EM-1 and EM-2 compositions are
islands in completely different parts of the world. This isotopic- concentrated in OIB just south of the Equator. This feature has been
chemical regularity demonstrates the existence of reproducible called the 'DUPAL anomaly' 52 and has been correlated with large-
differentiation processes in the mantle and justifies the isotopic scale seismic anomalies in the lowermost mantle. Also, nearly all
taxonomy discussed in the following paragraphs. Indian Ocean MORB have distinctly high 208PbP04 Pb, 87Sr/ 86Sr and
Zindler and Hare suggested that the mantle contains the follow- 208Pb' / 206 Pb* values relative to Pacific and Atlantic MORB (Figs 4-

ing isotopically extreme 'mantle components' (Figs 3-5), which 6). All of this provides strong evidel}ce that the different isotopic
contribute to the (usually mixed) sources of oceanic basalts. (1) flavours are not randomly distributed in the mantle but can be
HIMU ('high p.'; p. = 238 U/204 Pb) has the highest Pb ratios and the mapped. Nevertheless, conclusive interpretations are hard to come
lowest 87Sr/86Sr of any OIB (almost as low as MORB). Examples are by because we cannot look into the mantle directly but must rely on
St Helena, Austral Islands, Balleny Islands and the Azores. (2) EM-1 simulations53 or correlations between geophysical and geochemical
('enriched mantle 1')occupies the lower left-hand corner of the Sr- observations. Until recently, the spatial resolution of both geo-
Nd arra& (Fig. 3), and is concentrated in the right-hand cornerofthe physical observations such as seismic tomography and numerical
208 Pb* / 2 6 Pb*- Nd array (Fig. 6), and the upper left-hand corner of convection 'experiments' has been insufficient to 'see' or model
the Pb-Pb array (Fig. 4b). Representatives are the Pitcairn and relatively small features such as plumes. However, this resolution is
Tristan hotspots. (3) EM-2 ('enriched mantle 2' ) has the highest improving rapidly and much progress can be expected within the
87 Sr/86Sr and relatively high 207 Pb/ 204Pb ratios. Representatives are coming decade.
the Societies and Samoa hotspots. (4) DMM ('depleted MORB Quite distinct from the spatial distribution of plume sources and
mantle') is not specifically marked on the figures. It is defined by the other heterogeneities within the mantle is 'Darwin's question': what
most depleted MORB samples with the highest 143Nd/ 144 Nd and the is the origin of the isotopic mantle species? To answer this, we must
lowest 87Sr/86 Sr, 206Pb/204 Pb, 207Pb/ 204Pb and 208 Pb/ 204Pb ratios. consider other aspects of mantle chemistry, in particular the
Many of the isotope data arrays for individual islands or hotspots concentration ratios of highly incompatible elements, which
are roughly linear, indicating that they represent mixtures of a more characterize the different species and should help to identify the

a 15.9 b

41
15.8

.0 .0
40
a..
'<:!'
15.7
a..

-
0 '<:!'
C\J 0
..._ 15.6 N 39
..0
a.. . .0
,.._
0
15.5 a..
a)
38
C\J 0
N
15.4
37

15.3
16 17 18 19 20 21 22 36
16 17 18 19 20 21 22
2o6pb I 2o4Pb

Figure4 Pb isotopes forthe samples shown in Fig. 3, using the same colour code. and LCC mark average compositions of upper and lower continental crust,
The solid line labelled " geochron" in a marks the locus of possible primitive respectively '23 .
mantle values assuming an overall age of the mantle of 4.50 Gyr (refs 27,122). UCC

I
N ATU RE VO L 385 1 16 JAN UARY 1997 223
review article
Table 1 Long-lived radioactive decay series used as tracers
D FOZO, C • Pac. MOAB
Parent nuclide Daughter nuclide Half-life (yr) Tracer ratio • HIMU o Allan. MOAB
(radiogenic/
nonradiogenic) • EM-2 • Indian MOAB
>43Nd • EM-1 + OtherOIB
147Sm
........................ ....... 106 X 109 ' 43 Nd/ 144 Nd
87F{i) 878 ~···
48.8 X 109 .......... . .......s;s ~j86s~....
;;sHf ................. 35 7x iii' ····················· ,y;;Hiii 7'Hi
•
;·;;;L:~
0.708

0.707
238u
2j5iJ" ................ ......................... ,_ 0.706
2D7pb (/)

processes that ultimately created the isotopic differences.

CD
co
-....
.....
(/)
0.705
..
r-...
co
0.704 ••
Trace element ratios. Concentration ratios (hereafter called
'ratios') of incompatible trace elements can serve as tracers of 0.703
source chemistry and complement the information gained from
radiogenic isotopes. However, they must be used with care because, 0.702
unlike isotope ratios, they may be fractionated during igneous
processes. The ratio of two elements in a melt in equilibrium with 16 18 20 22
its solid residue is given by
C1 Di1 - F) + F _C 1.o
cz Dl(l- F)+ F c 2.o Figure 5 87Sr/ 86 Sr versus 206 Pb/ 204Pb fo r the sa me samples as in Figs 3 and 4.
where C1 and C2 are the concentrations in the melt, C1,0 and C2,0 the
concentrations in the original source, D 1 and D2 the bulk partition
=
coefficients, D C,01;d/ Cliquid , and F is the melt fraction. If Du << F, tectonic processes58 ; ( 4) it provides excellent tracers to test for the
that is, if the elem ents are highly incompatible or the melt fraction is presence of recycled continental crust in OIB sources.
large, the first term on the right-hand side vanishes and one obtains Questions about the 'constancy' of these ratios and the above
C 1I C2 = C~.0 1 C2•0 , so that the ratio in the melt will be the same as the interpretation have recently been raised59 ' 60 • The validity of the use
ratio in the source. Such trace-element ratios can be just as useful as of Nb/U as an important tracer of mantle sources is reaffirmed here
isotope ratios in serving as tracers of mantle chemistry. by a new compilation, shown in Fig. 7, ofNb/U versus 143Nd/ 144 Nd
Indicators of source depletion. The bulk partition coefficients of the (expressed as ENd; see legend for definition). This shows average
most highly incompatible elements increase in the approximate values for basalts from 33 ocean islands (or hotspots) and MORE,
order Ba = Rb ~ Th < U = K < La < Ce. Therefore, Ba/La, Rb/ compiled from high-quality data published mostly since 1986. The
La, Th/U and Th/La ratios in basalts are generally greater than (or great m ajority of all these basalt suites have average Nb/U between
equal to) the respective source ratios. Thus, when these ratios are 40 and 60, independent of the source depletion or enrichment
less than the primitive-m antle values, they document a previous expressed by ENd> and are in good agreement with the value of
depletion of the source rocks. This effect produces the leftward- 47 :::'::: lO proposed in 198654 • This is significantly higher than the
descending branch in the MORB pattern seen in Fig. 2. A relative primitive mantle ratio of Nb/U = 30 and the mean value of
depletion for Rb and Ba is also seen in the Tubuai pattern; it is Nb/U = 10 for continental crust and island arc rocks. Thus, most
characteristic of all HIMU basalts and distinguishes them from EM- of the sampled mantle has been enriched in Nb (and Ta, which has
type basalts. It is consistent with the low Rb/Sr of HIMU sources similar properties to Nb) relative to U and other similarly incom -
inferred from their 8 7Sr/86Sr ratios, which indicate a 'depleted patible elements. Possible mechanism s for this enrichment are
heritage' for both MORB and HIMU sources. discussed in the final section preceding the conclusion.
Globally 'unifo rm' ratios and the significance of N b, Ta and Pb. Figure 7 shows that normal OIB, including the HIMU islands and
Certain trace-element ratios are comparatively uniform in most EM-1 islands, representing the full range ofNd-isotopic composi-
oceanic basalts, irrespective of their absolute concentrations, which tions in OIB, have Nb/U ratios very close to those of average MORE.
may range over two orders of m agnitude from the most depleted Therefore, these OIB types cannot be generated by mixtures of
MORB to the most enriched OIB. In such cases, the bulk partition depleted mantle and either primitive mantle or recycled continental
coefficients of the two elem ents involved are nearly identical, or crust (although the slightly low Nb/U ratios of EM -1 basalts permit
both are much smaller than the melt fraction F (see above) 54 • Such the presence of a small sedimentary component; see below).
uniform ratios include Ba/Rb, K/U (except in HIMU basalts), Nb/ Obvious exceptions are the extreme EM-2 OIB from the Society
Ta, Zr/Hf, Y/Ho, Ti/Sm, Sn/Sm and P/Nd (see, for example, refs 55- and Marquesas hotspots (selected by having the highest 87Sr/ 86Sr at
57). These ratios are also similar in the. average continental crust, each hotspot) and the highly anomalous MORB from segment 3 of
and for the highly refractory element pairs Nb- Ta, Zr- Hf and the Chile ridge (ref. 61 and E. Klein, unpublished data). This
Y- Ho, they are identical to the respective ('chondritic') ratios confirms that, except for EM-2, all types of oceanic basalt
of the prim itive m antle (see Box l ). and their sources have very similar Nb/U values, which are sig-
Some ratios, namely Nb/U = 47, Ta/U = 2.7 and Ce/Pb = 25 are nificantly higher than primitive-mantle Nb/U. EM -2 OIB, and Chile
similar in MORB and OIB, but their values are much higher than ridge MORB are unique in that they, and only they, lie on mixing
those of the continental crust and island arc volcanics (by factors of t rajectories between DMM or HIMU mantle and average continen-
4- 5) and those estimated for the primitive mantle (by factors of tal crust (probably in the form of mantle-recycled sediments).
1.5- 2.5) 54• The significance of this result is fourfold: (l ) it shows An analogous but converse relationship holds for lead. Using Ce/
that ocean island basalts do not sample predominantly primitive Pb as a tracer, Hofmann et al. 54 and Newsom et a/. 6 2 showed that lead
m antle reservoirs; (2) it links the origins ofMORB and OIB sources; is systematically depleted in oceanic basalts, but overabundant in
(3) it confirms independently the link between island arc mag- island arc rocks and continental crust (see also Fig. 2). A compila-
matism and formation of continental crust as a consequence of plate tion of recent, more extensive data 63 shows that the bulk partition

224 NATURE IVO L 385 116 JANUARY 1997

 review article
coefficient ofPb is closer to that ofNd than to that ofCe. Therefore,
although Ce/Pb (or better, Nd/Pb) ratios scatter considerably more 0.5134
than Nb/U ratios59, they nevertheless appear to be good tracers of
source ratios. For reasons of clarity, no example of an EM-2 basalt is 0.5132
"0
shown on Fig. 2. However, on the Society Island (EM-2) hotspot, z
'<!' 0.5130
Ce/Pb and Nd/Pb ratios are anomalously low and well correlated

--z
'<!'

- ~+

..
with Nb/U and 87Srl6Sr (refs 64-66), confirming the inference that 0.5128
EM-2 OIB, but not other OIB, contain small but significant "0 +. +....+
+
amounts of recycled material of continental origin. The reasons (")
'<!' 0.5126 0 *
'~ •
for the anomalous behaviour oflead will be discussed in the section
on the 'lead paradox' below. 0.5124
•
Mass fraction of depleted mantle 0.5122
By calculating how much of the mantle must be depleted in order to 0.9 1.0 1.1 1.2
account for the incompatible-element enrichment in the continen- 2oaP b* I 206pb*
tal crust, one can estimate how much of the mantle could still be
chemically primitive. The mass fraction of the depleted mantle, Xd, Figure 6 143 Nd / 144Nd versus 208 Pb' / 206 Pb' for the same samples as in Figs 3-5
is given by (Here 21ll Pb"/ 200 Pb' = {("" Pb/204 Pb)- (208 Pb/ 204 Pb), 1}/{(200 Pbi"' 'Pb) - (200 Pbi "" Pb),n1J,
where the subscript 'in it.' designates the initial composition ofthe primitive mantle
Xd = {X, C,(Rd- R,)/(Cr(~- Rr)l - X, 4.5Gyr ago.
where the subscripts d, c and prefer to depleted mantle, continental
crust and primitive mantle, respectively, R is a ratio such as
143 Nd/ 144 Nd or Nb/U, and C is the concentration of the element club, the HIMU source regions with the highest 206PbP 04 Pb values,
in the denominator of R. Initial results based on Nd isotopes (for were formed about 2 Gyr ago. Thus, the timescale observed for
example, re£ 67) indicated depleted-reservoir sizes between 25% 'oceanic' mantle differentiation is likely to be significantly shorter
and 30% of the mantle. This corresponds to the mass of the mantle than the mean age of the continental crust (about 2-2.5 Gyr; re£ 58).
above the 660-km discontinuity and is thus consistent with the
layered mantle model. However, other authors estimated the Evaluation of geochemical mantle models
proportion of depleted mantle to be considerably larger, between The large body of geochemical data reviewed above demonstrates
30% and 90% (for example, refs 1, 68- 70), depending on specific that the Earth's mantle has been partially depleted in incompatible
assumptions about the composition of the continental crust and elements by the formation of the continental crust, which has
depending on the inclusion or omission of the OIB source extracted roughly half of the total budget of highly incompatible
reservoir( s). elements from the mantle. This residual, depleted mantle is chemi-
Perhaps the best-constrained mass balance is that for radiogenic cally heterogeneous on local, regional and global scales. Convective
argon. Approximately 40% ofthe 40Ar produced from 4°K in 4.5 Gyr stirring has not homogenized the mantle, presumably because near-
of Earth history now resides in the atmosphere; only about 1Oo/o is in surface processes such as partial melting and subduction continue
the continental crust and the depleted upper mantle, and the to reintroduce chemical and isotopic heterogeneities. The most
remaining 50% must still be in the lower mantlen. In terms of the highly depleted, and least heterogeneous region is located in the
above equation, this corresponds to Xd = 0.5. This is significantly upper 660 km of the mantle and produces MORB by passive
greater than the mass fraction of the mantle above 660 km (about upwelling and partial melting. Plumes are more heterogeneous in
0.27) but much less than the entire mantle. Thus, although the lower composition and presumably originate from a boundary layer
mantle is certainly no longer primitive, it is substantially less somewhere beneath this upper-mantle reservoir. There is a con-
degassed, and thus probably also less depleted in incompatible tinuing debate about the existence of layered mantle convection.
elements, than the upper mantle (which has lost almost all of its Layering may have been caused by an intrinsic, compositional
40 Ar). This constitutes a strong argument against simple whole- density stratification that originated through crystallization of an
mantle convection, because such a circulation type would be early 'magma ocean' formed during segregation of the core73 •
expected to cause similar levels of degassing throughout the entire Current evidence from mineral physics 74 and experimental evidence
mantle. on crystal-liquid partitioning of high-pressure minerals75•76 does
not favour this model and probably precludes the survival of this
Ages of mantle heterogeneities type of mantle stratification. Layered convection may also be caused
The isotopic heterogeneities shown in Figs 3- 6 require time to by the endothermic nature of the phase change from upper-mantle
develop, particularly in view of the extremely long half-lives of the peridotite to a lower-mantle perovskite-wiistite mineralogy13 • In
parent nuclides. If the parent/daughter ratio and the increase in this case, the major-element compositions of the upper and lower
radiogenic daughter nuclide were known, the age of a heterogeneity mantle may be the same, but the trace element and isotopic
could be calculated. In practice, this is difficult for mantle domains, chemistry of the two reservoirs may differ drastically because of
because the parent/daughter ratios of the mantle sources are crust formation from the upper mantle (see section on mass
generally not preserved in the basalts. Lead is an exception, because fraction of depleted mantle).
the production ratio orz07 Pb/ 206Pb depends only on time, so that the Layered mantle, primitive OIB sources. This is the original
slope of a straight line in 207Pb/ 204Pb- 206Pb/204 Pb space can be 'standard' model (Fig. la) in its simplest form 12•77 • Plumes rise
interpreted to yield the age of a single U- Pb differentiation event. from the primitive lower layer and mix with the upper, depleted
Thus, if the OIB source regions had been formed in a single event, layer on their way to the surface. This model is no longer tenable
the slope of the correlation line in Fig. 4a would mean that this event because ( 1) OIB sources require more than just two source compo-
occurred about 2 Gyr ago. However, continuous differentiation and nents; (2) most OIB have distinctly non-primitive Pb isotopes to the
remixing processes have been shown to produce data arrays with a right of the 'geochron' (Fig. 4a); (3) some OIB sources have lower-
similar slope (yielding an apparent age of 2 Gyr), even though the than-primitive 143 Nd/ 144 Nd ratios (Fig. 3); and (4) Nb/U and Ce/Pb
actual mean differentiation age of the individual mantle sources is (or Nd/Pb) ratios in OIB require non-primitive sources (Fig. 7).
only about 1- 1.3 Gyr (refs 53, 72). Only the oldest members of the Several authors have nevertheless recently proposed that some

N ATURE I VOL 385 l t 6 JANUARY 1997 225

review article
OIB may be derived from primitive (or slightly depleted) sources 0 OIB, EM 1 ~!":] Chile ridge,
(for example, refs 38, 42, 43). A particularly popular candidate is segm. 3
0 OIB, EM 2
Hawaiian volcanism, especially the tholeiites from Oahu, Lanai and 0 Prim. mantle
Kahoolawe. Basalts from these islands have close-to-primitive iso- •+ OIB, HIMU
OIB, other
- - -MOAB -
CC mixtures

•
topic compositions of Sr, Nd, Hf and Pb (for example, ref. 78), and
MOAB HIMU-
some Hawaiian volcanoes have high 3He/ 4He ratios41 ' 79. However, CC mhctures
every Hawaiian volcano investigated so far shows the same Nb- Ta
100
excess (and Pb deficiency) 80 that characterizes other OIB, and this

•
precludes the derivation of Hawaiian volcanism from a primitive +
source.
In subsequent variants 14' 15' 81 - 83 of the two-layer model, plumes 00
~
originate from the base of the upper mantle, but they entrain small .0 /
z ., '0 /
amounts of primitive (or slightly depleted) noble-gas-rich material 0
-- -- --
0> .......... 0 /
from the lower mantle, which imparts the high 3He/4 He and ~
20 Ne/ 22 Ne (Fig. 1b). There are two difficulties with these models.

First it is not obvious why some plume types, particularly the HIMU
types40, should be consistently free of entrained, 3He- and 20Ne-rich
lower-mantle material. Second, the amounts of lower-mantle 10 CONT. CRUST
material calculated from noble-gas evolution models are extremely
low83 and appear to be inconsistent with the mass of the less- -5 0 5 10
depleted mantle derived from most recent mass-balance estimates Avg £Nd
(see above).
Other layered mantle models. Anderson (see, for example, refs 84,
85) has proposed some thought-provoking alternatives to the more Figure 7 Nb/ U plotted against 'Nd· where ' Nd is the deviation, in parts per 104 , of the
143 Nd / 144 Nd ratio of a sample from that of present-day primitive mantle (p): ' Nd =
conventional models. He suggested that the upper mantle is
internally differentiated into a lower, garnetiferous, incompatible- 10,0001< " 3 Ndt' 44 Nd) - ('"' Nd/ 44 Nd)0 J 1 ( 143 Ndt"'Nd)0 • The points shown are
average values for suites of MORB and 0 18 using general ly different samples
element-depleted 'piclogite' layer above 660 km (the MORB
source), and an upper (above 220 km), enriched peridotite layer, but the same colou r coding as in Figs 3-6. EM-1: Pitcairn, Tristan-I naccessible,

the 'perisphere', which is also the OIB source. Evidence for such an Kerguelen islands; EM-2: Society, M arquesas islands; HIMU: Tubuai, St Helena;
enriched layer is found in many young rifts, which typically deliver other 0 18 : Cameroon Line islands, Iceland, Hawaii, Reunion. Comores, MacDo-

OlE-type magmas. However, the opening of the Red Sea shows that nald seamount, Easter hotspot, Bouvet, Gran Canaria, Tenerife, Azores, Cape

this enriched rift magmatism is soon replaced by normally depleted Verde, Fernando di Noronha, Trinidade, Madeira, Ascension; MORB: Atlantic

MORB magmatism, suggesting that the enriched perisphere is at MORB 10°- 23° N, Kolbeinsey ridge, Pacific MORB, Galapagos rift. Indian MORB,

best a thin veneer of enriched material that may well produce some Chile ridge segment 3. All these ocea nic basalt suites except EM-2 types have
Nb/ U averages similar to the global average of 47 :+:: 10 given by Hofmann et a/. 54
of the unusual types of OIB, such as the Cameroon Line 'hot line'
using a much smaller set of data. Note that these Nb/ U averages show no
volcanoes59 , but could not sustain voluminous, long-lived, hotspot-
type magmatism such as the Hawaiian- Emperor chain. Further- significant relationship to the degree of source enrichment or depletion as

more, there is scant evidence for the existence of such a perisphere in measured by Nd isotopes. Also, the primitive-ma ntle value lies completely out-

intra-oceanic regions, because laterally moving mid-ocean ridges, side the MORB- HIMU-EM-1 array Only EM-2 type OIB and the segment3 of the

or ridge jumps, generally sample an ordinary, depleted MORB-type Chile ridge appear to be mixtures of MORB- HIMU type ma ntle w ith (recycled)

mantle rather than an enriched perisphere supposedly located in the material. probably sed iments, from the continental crust. For references to this

uppermost mantle region. Finally, recent seismic imaging has traced compilation, see Supplementary Information, except for unpublished data by C. H.

the thermal anomaly beneath Iceland to a depth of 400 km and a Hemond et at. on Atlantic MORB 10°- 23° N, and by E. Klein on the Chile ridge.
diameter of about 100 km (ref. 86), indicating that the Iceland
hotspot represents a mantle plume originating from at least 400 km
depth. These features are difficult to reconcile with the perisphere patible elements and volatiles, until it melts and generates OIB (see
model. ref. 90). Vollmer91 has based a global mantle model on such a
Whole-mantle convection models. In these models, the 660-km mechanism. The effects of metasomatism are certainly evident in
seismic discontinuity represents little or no barrier to convection. many mantle xenoliths92 , but the scale of these processes is unknown
Geophysical support for this comes from evidence for deep, high- and may be small relative to the source volume of OIB80 • The
seismic-velocity structures (for example, ref. 87), interpreted as viability of the plume model has largely obviated the 'need' for
subducted lithosphere in the lower mantle, and from the fixed metasomatic mechanisms to generate OIB directly.
locations of hotspots, which appear to require that plumes originate Metasomatism may also enrich substantial volumes of the mantle
in a lower-mantle region that has a much higher viscosity and moves above a subduction zone. The oceanic crust is hydrated by ridge-
much more slowly than does the upper mantle88. It is also argued crest hydrothermal and sea-floor alteration processes, and it is
that flood basalt provinces, which are erupted during geologically covered by wet sediments. During subduction, this hydrous
very brief time spans and attain enormous volumes and lateral material is heated and dehydrated. The escaping fluids migrate
extent (about 2,000 km), are produced by plume heads that could through the overlying mantle, thereby metasomatizing it. Island arc
have attained such dimensions only by thermal entrainment during volcanism is likely to remove much of the effects of this meta-
plume ascent through the full depth of the mantle89 • Geochemical somatism93·94 (as incompatible elements are partitioned into the
arguments are based on the so-called 'FOZ0' 16 or 'C' 40 isotopic fluid and ultimately into the melt), but some of the metasomatized
component, which appears to be common to many plumes and mantle may be incorporated in the subcontinental lithosphere or
differs isotopically from the MORB source. FOZO is thought to returned to the convecting part of the mantle. In either case, the
reside in the lower mantle where it is entrained by plumes rising metasomatized material may ultimately be recycled through the
from the core- mantle boundary16 • mantle and reappear in intra-oceanic volcanism (see below).
Metasomatic models. Metasomatism has been popular with many A variant of this mechanism is metasomatism by the migration of
workers as a mechanism for enriching the upper mantle in incom- small amounts of melt from the asthenosphere into the oceanic or

226 NATURE IVOL 385 116 JANUARY 1997

 review article
continentallithosphere3 ·59 , followed by recycling as above95 • All of enriched as a result of fluid metasomatism during subduction 104•105
these metasomatic models are difficult to test by observations, or melt migration from the asthenosphere 95 , but the scale of such
because the only tangible evidence for them is found in fist-size processes remains largely a matter of speculation. This recycling
or smaller mantle xenoliths, which yield no clues as to the quantities hypothesis can now be tested using osmium isotopes, because
of material transported or the distances travelled. ancient lithosphere is unique in having 187 0s/ 188 0s significantly
Recycling models. The idea that recycling crust through the mantle lower than primitive-mantle values 36 • So far, Os data for only two
could explain the isotopic and chemical diversity of oceanic basalts EM -l type basalts have been published, and both of these have high,
goes back almost 30 years96 but has found wide acceptance only in rather than low, 187 0s/ 1880s (ref. 33).
the past few years, because mantle convection was previously Continental basalts are also commonly thought to sample the
thought to be highly efficient at rehomogenizing the chemical subcontinental lithosphere (see, for example, refs I 04, l 06) and they
and isotopic heterogeneities introduced into the mantle during often resemble EM -l type OIB. However, recent realization that
subduction. continental flood basalts may sample the heads of deep-mantle
Continental crust. Armstrong96 '97 proposed that the continental crust plumes makes the argument somewhat circular. Thus it is not clear
has maintained a constant volume through a steady-state process of whether the EM-l character is due to contamination of a plume
growth (through accretion of island arcs) and destruction through head by the continental lithosphere or whether this signature was
subduction of continent-derived sediments, and suggested that this part of the original plume source 107 •
could also explain the isotopic diversity of oceanic basalts. The
geochemical evidence from Pb isotopes and from Nb/U and Ce/Pb Towards a unified model for mantle evolution
ratios now rules out large amounts of recycled continental material My preferred model for the evolution of the mantle is based on a
in OIB sources (see above), except in the cases of EM-2 OIB and simple-minded, uniformitarian approach that uses known geolo-
Indian Ocean MORB, which have low Nb/U and Ce/Pb values gical processes and avoids exotic processes whenever possible. Thus,
correlated with high 87Sr/85 Sr and 207 Pb/ 204 Pb ratios, and which can I assume that much of the chemical diversity of the isotopic mantle
be explained by additions of small amounts (1-2%) of recycled 'species' is caused by subduction, which introduces oceanic crust
sediments to the source mantle 6 •49 •66 •98 (see Fig. 7). and small amounts of sediments of variable composition into the
Oceanic crust. Subduction of oceanic lithosphere continuously mantle.
recycles oceanic crust and thus injects about 20 km 3 of enriched Timescales of crust-mantle evolution. Continental crust forma-
material per year into the mantle. This is the uniformitarian basis of tion extracts incompatible elements very efficiently from the mantle
the recycling model of refs 99 and 100. Soon after subduction, this and enriches them in the crust roughly 100-fold. Continental crust
crust is transformed into eclogite (and later to higher-pressure may be returned to the mantle by subduction of sediments, but its
equivalents), which is about 2-4% denser than ordinary mantle lifetime is at least 20 times that of the oceanic crust, which has a
(depending on pressure53 ). The assumed segregation of the dense mean life of only about l 00 Myr. The present -day production rate of
former oceanic crust from its lithospheric base and the main mantle continental crust (l km 3 yr -I; ref. 108) is also about 20 times lower
mass is inhibited at normal mantle viscosities, but is enhanced in the than that of oceanic crust (20 km 3 yr- 1 ). Because of the grossly
hot boundary layer at the base of the convection system (660 or different timescales, global differentiation may be approximated
2,900 km). Storage and isotopic evolution of some of this material roughly by a two-stage process 54 :
for periods of 1-2 Gyr, followed ultimately by incorporation into ( l) The 'continental stage' or 'primary differentiation' of primitive
plumes rising from this boundary layer, could generate the MORB- mantle into continental crust with high Rb/Sr and Th/U, and low
HIMU isotope arraf 3 • Sm/Nd, Nb/U and Ce/Pb ratios, and a residual mantle with
Strongly altered oceanic crust is evidently not being recycled, complementary low Rb/Sr and Th/U, and high Sm/Nd, Nb/U and
because alteration adds many 'mobile' elements such as Rb and U Ce/Pb. The mean age of the crust thus produced is 2-2.5 Gyr
(but not Th), which would cause isotopic changes quite unlike the (ref. 58). The trace-element patterns of MORB (Fig. 2) reflect the
observed OIB isotopic arrays 101 • If such alteration has been residual nature of their source mantle. (2) The 'oceanic stage'
ubiquitous in the past, it must therefore be assumed that the altered, continuously differentiates the residual mantle by ridge and sub-
hydrous parts of the crust were efficiently stripped from the system dw;:tion processes and rehomogenizes it by convection. This differ-
during subduction, so that the 'additive' effects of hydrothermal entiation-homogenization process has time constants averaging l-
alteration were removed. Chemical and isotopic evidence from 1.3 Gyr53 ' 72 (see discussion on ages of mantle heterogeneities), and it
subduction-related volcanism supports this 93 ' 94 • In contrast, the creates most of the chemical and isotopic diversity seen in Figs 3-6.
hydrothermal subtraction of certain elements, such as lead, from The complementary Nb(Ta) and Pb anomalies in the oceanic and
the oceanic crust is probably irreversible, and this should contribute continental crusts (Fig. 2) are caused by special processes during
to the high U/Pb and Th/Pb ratios responsible for the extremely continent production. Niobium and tantalum are retained in the
radiogenic lead of HIMU basalts 49 •102 • mantle by Ti minerals 109 or amphiboleno during subduction-related
The model of oceanic crustal recycling is supported by the still volatile transfer 111 and/or magmatism. The continental crust grows,
sparse isotope data for osmium. HIMU basalts in particular have in part, by accretion of subduction-related, Nb-Ta depleted island-
elevated 1870s/ 1880s ratios, which can be explained by an ancient arc magmas and thus acquires the large negative Nb-Ta and smaller
basaltic source componene 2- 35 • Other oceanic basalt types, such as Ti anomalies seen in Fig. 2. The mantle therefore becomes relatively
Hawaiian basalts, which have Pb isotopic compositions closer to enriched in Nb and Ta, and this positive anomaly is homogenized
MORB, also have high 1870s/ 88 0s ratios 50 and may represent through relatively rapid (see above) convective stirring. This
recycled, deeper oceanic crust depleted in Th and U (ref. 103). explains the relatively uniform Nb/U ratios seen in Fig. 7. Oceanic
Subcontinental lithosphere. Old subcontinental lithosphere is com- (MORB and OIB) magmatism does not create new Nb-Ta
paratively cold and dense, so it may be delaminated and sink into anomalies because the minerals or volatiles that could fractionate
the mantle to become a plume source 81 • Because it has been 'pre- Nb/U are absent in the MORB-OIB melting region.
aged' isotopically, while it was still attached to the crust, it does not 'Lead paradox~ The positive lead anomaly in the continental crust
need an extended period of storage in the convecting part of the and the complementary negative anomalies in oceanic basalts
mantle before it acquires the isotopic signatures found in plumes. (Fig. 2) are created by hydrothermal processes in two stages. First,
This is a popular source candidate for EM -l type OIB because some lead is extracted from the basaltic crust and redeposited in metalli-
xenoliths derived from the subcontinental mantle have isotopic ferous sediments by hydrothermal ridge processes 102 • When the
signatures similar to that of EM -l. The lithosphere may well be crust is subducted, additional lead is extracted from the crust and

NATURE I VOL 385116 JANUARY 1997 227

review article
the sediments by dehydration. This lead is ultimately transferred to Atlantic regions in that it has been 'polluted' by ancient sediments 98
the continental crust via island arc magmatism (see above). or subcontinentallithosphere 114 (see section on the lead paradox).
This transfer process provides the key to the solution of the 'lead The former oceanic crust stored as potential plume sources at 660
isotope paradox' and the absence of a global correlation of and 2,900 km depth contains a wide variety of 'enriched' material.
206 PbP 04 Pb with Sr and Nd isotopes (Fig. 5). The hydrothermal HIMU plume sources are enriched in former MORE, in which
processes have enriched the continental crust in lead to a similar hydrothermal ridge processes had caused loss oflead and therefore
degree as in uranium (see Fig. 2). Consequently, the continental an increase in (U, Th)/Pb (refs 6, 49). EM-2 sources represent
crust has, on average, similar U/Pb and therefore 206 Pb/ 204 Pb ratios recycled oceanic crust containing a few per cent of continent-
as the (residual) mantle. Similarly, most of the MORB data also lie derived sediment49 '66 ' 115 , This is consistent with their high
close to the '~eochron' in Fig. 4a, rather than falling to the 'depleted', 207 Pb/ 204 Pb and 87 Sr/ 87 Sr and low Nb/U and Ce/Pb ratios (see

low- 206 Pb/ 20 Pb side (caused by low U/Pb) that would be expected if Fig. 7). EM-I sources are the most enigmatic members of the
the sources were produced by purely magmatic depletion processes. mantle zoo. Their slightly subnormal Nb/U, low 206 Pb/ 204Pb and
(For a detailed discussion of the 'best' position of the geochron, see high 208 Pb* / 206 Pb* ratios are consistent with a contribution from
ref. 27.) Continent-derived ocean-floor sediments have 206 Pb/ 204 Pb either recycled, ancient pelagic sediment6 ' 49 or recycled ancient
ratios similar to average upper oceanic crust (Fig. 4), but possess an subcontinental lithosphere 81 ' 107' 116 • Osmium isotopes (see above)
extremely large range ofU/Pb ratiosll2, which will, in time, generate should resolve the issue.
a large range ofPb-isotopic compositions. When such sediments are
recycled through the mantle, they will destroy the correlations Conclusion and outlook
between the U/Pb, Rb/Sr and Sm/Nd ratios that are produced by Although many parts of the story remain controversial and difficult
purely magmatic differentiation processes. This appears to have to decipher, I regard the following conclusions to be reasonably
happened in EM-2 (and possibly also in EM-I) type OIB, as well as robust. Recycling of oceanic crust and lithosphere plays an impor-
in Indian Ocean MORE, which show no correlations between Sr (or tant role in generating mantle heterogeneities. Available geo-
Nd) and Pb isotopes (Fig. 5). In contrast, Pacific and Atlantic chemical mass balances favour compositional layering, which is
MORE show good correlations and appear to be nearly free of maintained by partial or intermittent convective isolation between
recycled continental material. upper and lower mantle.
Role of plumes and mantle layering. Geochemical evidence, The controversy over the issue of convective layering and depth of
especially that derived from noble gases, favours some type of plume sources is likely to be resolved when the so far rather fuzzy
layered mantle convection, but one in which the layers are not seismic imaging of the Earth's interior becomes comparable in
completely isolated. I therefore suggest that plumes may come from resolution to that achieved by geochemical mapping, so that
both the 660-km discontinuity and the core-mantle boundary. geophysical and geochemical data can be more specifically corre-
Small, short-lived plumes and those carrying no enrichment in lated. The origin of the geochemical 'mantle species' will be clarified
primordial 3 He are most likely to have a shallow origin. Large, long- by a rapidly expanding data base of osmium isotopes and high-
lived, 3He-rich plumes, and especially those that have generated quality data for well-established tracer elements such as Nb and Pb,
enormous volumes of oceanic or continental flood basalts, are more as well as relatively 'new' tracers such as E and Sb, and by high-
likely to have come from the base of the lower mantle89 • precision oxygen isotope microanalysis of igneous minerals 117,
This history of deep-plume activity may well have been episodic capable of recognizing a sedimentary or hydrothermal prehistory
and linked to times of high rates of continent formation 18 ' 108 • Several of the magma. 0
of the large oceanic plateaux were created in the Cretaceous period, A. W Hofman is at the Max-Planck-Institut fur Chemie, Postfach 3060, 55020
and this may constitute an unusual event in Earth history 113 • These Mainz, Germany
plateaux may represent the heads of deep-mantle plumes89 that will 1. Zindler, A. & Hart, S. Annu. Rev. Earth Planet. Sci. 14, 493-571 (1986).
2. Allegre, C. J, Earth Planet. Sci. Lett. 86, 175-203 (1987).
ultimately be accreted to the continental crust. The relatively large 3. Sun, S.-S. & McDonough, W. F. in Magmatism in the Ocean Basins (eds Saunders, A. D. & Norry,
number of 3He-rich plumes may therefore be mostly remnants of M./.) 313-345 (Spec. Pub!., Geological Soc., Oxford, 1989).
4. Hofmann, A. W. Earth Planet. Sci. Lett. 90, 297-314 (1988).
deep-mantle plumes that started in Cretaceous time and are still 5. Olson, P., Silver, P. G. & Carlson, R. W. Nature 344, 209-215 (1990).
active today. Similarly, deep subduction of oceanic lithosphere into 6. Weaver, B. L. Earth Planet. Sci. Lett. 104,381-397 (1991).
the lower mantle, for which there is mounting seismological 7. Davies, G. F. & Richards, M.A. f. Geol. 100, 15\-206 (1992).
8. Carlson, R. W. Rev. Geophys. 32, 337-361 (1994).
evidence87 , may be a sporadic event that balances the occasional 9. \Nhite, W. M. Rev. Geophys. (suppl.): US Natl Rep. to Int. Union Geodesy and Geophysics 1991-1994,
deep-mantle plume flux 17 • 19-24 (1995).
10. Wilson,/. T. Can. f. Phys. 41, 863-870 (1963).
In geologically 'normal' times, both plate-tectonic convection
11. Morgan, W. J. Nature 230, 42-43 (1971).
and plume generation may be confined to the upper mantle. This 12. Schilling, /.-G. Nature 242, 565-571 (1973).
implies further that at present we are seeing both the remnants of a 13. Christensen, U. R. Annu. Rev. Earth Planet. Sci. 23, 65-87 (1995).
14. Allegre, C.). & Turcotte, D. L. Geophys. Res. Lett. 12, 207-210 (1985).
deep-plume and deep-subduction event (in the form of the remain- 15. Galer, S./. G. & O'Nions, R. K. Nature 316, 778-782 (1985).
ing plume trails) and the resumption of isolated upper-mantle 16. Hart, S. R., Hauri, E. H., Oschmann, L.A. & Whitehead,/. A. Science 256, 517-520 (1992).
17. Machete!, P. & Weber, P. Nature 350, 55-57 (1991).
convection dominated by plate tectonics and upper-mantle 18. Stein, M. & Hofmann, A. W. Nature 372,63-68 (1994).
plumes 18 • 19. Whitehead,/. A. & Luther, D. S. f. Geophys. Res. 80, 705-717 (1975).
Depletion, replenishment, recycling. The uppermost mantle, the 20. Loper, D. E. Tectonophysics 187, 373-384 (1991).
2l. Batiza, R. Earth Planet. Sci. Lett. 60, 195-206 (1982).
source of MORE, is depleted in incompatible elements by two 22. Halliday, A. N., Dickin, A. P., Fallick, A. E. & Fitton,/. G.]. Petrol. 29, 181-211 (1988).
processes. (I) Devolatilization and melting during subduction 23. Griffiths, R. W. & Campbell, I. H. Earth Planet. Sci. Lett. 99,66-78 (1990).
24. Allegre, C. J. & Turcotte, D. L. Nature 323, 123-127 {1986).
transfer incompatible elements to long-time (or permanent) 25. Hofmann, A. W. & Hart, S. R. Earth Planet. Sci. Lett. 39,44-62 (1978).
storage in the continental crust and lithosphere; and (2) a portion 26. jacobsen, S. B. & Wasserburg, G./. Earth Planet. Sci. Lett. 67, 137-150 (1984).
of the subducted oceanic crust is stored at the base of the convecting 27. Galer, S. J. G. & Goldstein, S. L. in Earth Processes: Reading the Isotopic Code (eds Basu, A. & Hart,
S. R.) 75-98 (Am. Geophys. Union, Washington DC, 1996).
system and thus sequesters incompatible elements temporarily. The 28. Allegre, C.J. Earth Planet. Sci. Lett. 5, 261-269 (1969).
remainder of this crust is mixed back into the MORE reservoir. 29. Allegre, C.)., Dupre, B. & Lewin, E. Chern. Geol. 56,219-277 (1986).
30. Allegre, C.J. & Luck,/. M. Earth Planet. Sci. Lett. 48, 148-154 (1980).
Consequently, the MORB mantle has a 'marble cake' structure24
31. Meisel, T., Walker, R. /. & Morgan,/. W. Nature 383, 517-520 (1996).
containing schlieren of enriched, recycled material. In addition, 32. Hauri, E. H. & Hart, S. R. Earth Planet. Sci. Lett. 114,353-371 (1993).
plume injection creates 'regional' (I 00-I ,000 km) scale chemical 33. Reisberg, L. et al. Earth Planet. Sci. Lett. 120, 149-167 {1993).
34. Roy-Barman, M. & Allegre, C./. Earth Planet. Sci. Lett. 129, 145-161 (1995).
and isotopic gradients in MORE mantle (see, for example, ref. 12). 35. Marcantonio, F., Zindler, A., Elliott, T. & Staudigel, H. Earth Planet. Sci. Lett. 133, 397-410
The Indian Ocean MORE source region differs from Pacific and (1995).

228 NATURE I VOL 385116 JANUARY 1997

 review article
36. Pearson, D. G., Carlson, R. W., Shirey, S. B., Boyd, F. R. & Nixon, P. H. Earth Planet. Sci. Lett. 134, 85. Anderson, D. L. Rev. Geophys. 33, 125-149 (1995).
341-357 (1995). 86: Wolfe, C.]., Bjarnason, I. T., VanDecar, J. C. & Solomon, S. C. Nature 385, 245-247 (1997).
37. Anderson, D. L. Science261, 170-176 (1993). 87. Grand, S. P. f. Geophys. Res. 99, 11591-11621 (1994).
38. Farley, K. A., Natland, /.H. & Craig, H. Earth Planet. Sci. Lett.ll1, 183-199 (1992). 88. Steinberger, B. & O'Connell, R. j. Geophys. ]. Int. (submitted).
39. Allegre, C./., Moreira, M. & Staudacher, T. Geophys. Res. Lett. 22, 2325-2328 (1995). 89. Campbell, I. H. & Griffiths, R. W. Earth Planet. Sci. Lett. 99, 79-93 (1990).
40. Hanan, B. & Graham, D. Science 272,991-995 (1996). 90. Bailey, D. K. Nature 296, 525-530 (1982).
41. Kurz, M.D., jenkins, W./., Hart, S. R. & Clague, D. Earth Planet. Sci. Lett. 66,388-406 (1983). 91. Vollmer, R. Geology 13,452-454 (1983).
42. Kellogg, L. H. & Wasserburg, G. j. Earth Planet. Sci. Lett. 99, 276-289 (1990). 92. Menzies, M.A. & Hawkesworth, C.]. Ma11tle Metasomatism, 465 (Academic, London, 1987).
43. Porcelli, D. R. & Wasserburg, G. J. Geochirn. Cosmochim. Acta 59,4921-4937 (1995). 93. Ryan, J. G. & Langmuir, C. H. Geochim. Cosmochim. Acta 57, 1489-1498 (1993).
44. O'Nions, R. K. & Tolstikhin, I. N. Earth Planet. Sci. Lett. 124,131-138 (1994). 94. Ishikawa, T. & Nakamura, E. Nature 370, 205-208 (1994).
45. Honda, M., McDougall, I., Patterson, D. B., Doulgeris, A. & Clague, D. A. Geochim. Cosmochim. Acta 95. McKenzie, D. & O'Nions, R. K.]. Petrol. 36, 133-159 (1995).
57, 859-874 (1993). 96. Armstrong, R. L. Rev. Geophys. 6, 175-199 (1968).
46. Sarda, P., Staudacher, T. & Allegre, C.]. Earth Planet. Sci. Lett. 91, 73-88 (1988). 97. Armstrong, R. L. Phil. Tram. R. Soc. Lond. A 301, 443-472 (1981).
47. White, W. M. Geology 13, 115-118 (1985). 98. Rehkiimper, M. & Hofmann, A. W. Earth Planet. Sci. Lett. (submitted).
48. Barling, j. & Goldstein, S. L. Nature 348, 59-62 (1990). 99. Hofmann, A W. & White, W. M. Yb Carnegie Inst. Wash 79, 477-483 (1980).
49. Chauvel, C., Hofmann, A. W. &Vidal, P. Earth Planet. Sci. I.ett. 110,99-119 (1992). 100. Hofmann, A. W. & White, W. M. Earth Planet. Sci. Lett. 57, 421-436 (1982).
50. Hauri, E. H., Lassiter, j. C. & DePaolo, D. j. f. Geophys. Res. 101, 11793-11806 (1996). 101. Hart, S. R. & Staudigel, H. in Crust/Mantle Recycling at Convergence Zones (eds Hart, S. R. &
51. Class, C., Goldstein, S. L., Galer, S. j. G. & Weis, D. Nature 362, 715-721 (1993). Gwen, L.) 15-28 (Kiuwer, Dordrecht, 1989).
52. Hart, S. R. Nature 309, 753-757 (1984). 102. Peucker-Ehrenbrink, B., Hofmann, A. W. & Hart, S. R. Earth Planet. Sci. Lett. 125, 129-142 (1994).
53. Christensen, U. R. & Hofmann, A. W. f. Geophys. Res. 99, 19867-19884 (1994). 103. Hofmann, A. W. & Jochum, K. P. ]. Geophys. Res.101, 11831-11839 (1996).
54. Hofmann,A. W., Jochum, K.-P., Seufert, M. &White, W. M. Earth Planet. Sci. Lett. 79,33-45 (1986). 104. Carlson, R. W. Geochim. Cosmochim. Acta 48, 2357-2372 (1984).
55. Hofmann, A. W. & White, W. M. Z. Naturforsch. 38, 256-266 ( 1983 ). 105. Hawkesworth, C. J., Rogers, N. W., van Calsteren, P. W. C. & Menzies, M.A. Nature 311, 331-335
56. Jochum, K.-P., Hofmann, A. W., Ito, E., Seufert, H. M. &White, W. M. Nature 306,431-436 (1983). (1984).
57. Jochum, K. P., Hofmann, A. W. & Seufert, H. M. Geochim. Cosmochim. Acta 57,3585-3595 (1993). 106. Carlson, R. W. Austr.]. Earth Sci. 38, 525-544 (1991).
58. Taylor, S. R. & McLennan, S.M. The Continental Crust: its Composition and Evolutio11, 1-312 107. Mahoney,/., Nicollet, C. & Dupuy, C. Earth Planet. Sci. Lett. 104, 350-363 (1991).
(Blackwell Scientific, Oxford, 1985). 108. Reymer, A. & Schubert, G. Tectonics 3, 63-77 (1984).
59. Halliday, A. N. et al. Earth Planet. Sci. Lett. 133, 379-395 (1995). 109. Green, T. H. Chern. Geol. 117, 1-38 (1994).
60. Sims, K. W. W. & DePaolo, D./. Eos 72, 526-527 (1991). 110. lonov, D. A. & Hofmann, A. W. Earth Planet. Sci. Lett. 131, 341-356 (1995).
61. Klein, E. M. & Karsten, j. L. Nature 374, 52-57 (1995). Ill. Keppler, H. Nature380, 237-240 (1996).
62. Newsom, H. E., White, W. M., Jochum, K. P. & Hofmann, A. W. Earth Pla11et. Sci. Lett. 80,299-313 112. Ben Othman, D., VVhite, W. M. & Patchett, J. Earth Pla11et. Sci. Lett. 94, 1-21 (1989).
(1986). 113. Larson, R. L. Geology 19, 547-550 (1991 ).
63. Hofmann, A. W. Abstracts of the 49th IAVCEI Meetir1g (Geol. Soc. Australia, Canberra, 1993 ). 114. Mahoney,/./. et al.]. Geophys. Res. 94,4033-4052 (1989).
64. Devey, C. W. et al.]. Geophys. Res. 95, 5049-5066 (1990). I 15. Wbite, W. M. & Hofmann, A. W. Nature 296, 821-825 (1982).
65. Hemond, C., Devey, C. W. &Chauvel, C. Chern. Geol.115, 7-45 (1994). 116. Milner, S.C. & le Roex, A. P. Earth Planet. Sci.'Lett. 141, 277-291 (1996).
66. \Vhite, W. M. & Duncan, R. A. in Earth Processes: Reading the Isotopic Code(eds Basu,A. & Hart, S. R.) 117. Eiler,]. M. et al. Earth Planet. Sci. Lett. (in the press).
183-206 (Am. Geophys. Union, Washington DC, 1996) 118. Rudnick, R. L. & Fountain, D. M. Rev. Geophys. 33,267-309 (1995).
67. jacobsen, S. B. & Wasserburg, G. j.]. Geophys. Res. 84, 7411-7427 (1979). 119. Hofmann, A. W. et al. Eos 74, 630 ( 1993).
68. Davies, G. F. Nature290, 208-213 (1981). 120. Le Roex, A. P., Cliff, R. A. & Adair, B. /. I.]. Petrol. 31, 779-812 (1990).
69. Allegre, C.]., Hart, S. R. & Minster, J.-F. Earth Planet. Sci. Lett. 66, 191-213 (1983). 121. Cliff, R. A., Baker, P. E. & Mateer, N.J. Chern. Geol. 92, 251-260 (1991).
70. Hofmann, A. W. Phil. Trans. R. Soc. Land. A 328, 425-439 (1989). 122. Faure, G. Pri11ciples of Isotope Geology 1-589 (Wiley, New York, 1986).
71. Allegre, C.]., Hofmann, A. Vv'. & O'Nions, K. Geophys. Res. Lett. (in the press). 123. Zartman, R. E. & Haines, S.M. Geochim. Cosmochim. Acta 52, 1327-1339 (1988).
72. Allegre, C./. & Lewin, E. Earth Planet. Sci. Lett. 136,629-646 (1995). 124. Jagoutz, E. et al. Proc. lOth Lunar Pla11et. Sci Conf, Geochim. Cosmochirn. Acta (Suppl. 11) 2, 2031-
73. Anderson, D. L. Science223, 347-355 (1984). 2050 (1979).
74. Jackson, I. & Rigden, S. M. in The Earth's Mantle: Composition, Structure a11d Evolution (ed. 125. McDonough, W. F. & Sun, S.-S. Chern. Geol. 120, 223-253 (1995).
Jackson, I.) (Cambridge Univ. Press, Cambridge, in the press).
75. Kato, T., Ringwood, A. E. & lrifune, T. Earth Planet. Sci. Lett. 89, 123-145 (1988). Supplementary Information is available on Nature's World-Wide Web site (http://www.nature. com) or
76. Ohtani, E., Kawabe, I., Moriyama, J. & Nagata, Y. Contrib. Mineral. Petrol. 103,263-269 (1989). as paper copy from Mary Sheehan at the London editorial office of Nature.
77. Sun, S. S. & Hanson, G. N. Geology 3, 297-302 (1975).
Acknowledgements: I acknowledge unpublished Nb-U data made available by F. Albarede (Reunion), C.
78. Roden, M. F., Trull, T., Hart, S. R. & Frey, F. A. GeOi·him. Cosmochim. Acta 58, 1431-1440 (1994). Hemond (Atlantic MORB) and E. Klein (Chile ridge). 1 thank R. Carlson and W. White for formal
79. Kurz, M.D. & Kammer, D.P. Earth Planet. Sci. Lett. 103,257-269 (1991). reviews; C. Allegre, R. Carlson, R. Cliff, S. Galer, S, Goldstein, S, R. Hart, K. P. Jochum and \V. McDonough
80. Hofmann, A. W. Chern. Geol. 57, 17-30 (1986). for informal reviews and/or discussions of the issues involved; and H. Hartshorn for reading the
81. McKenzie, D. & O'Nions, R. K. Nature 301,229-231 (1983). manuscript and suggesting numerous improvements in style and clarity. This article was completed while
82. Galer, S.j. G., Goldstein, S. L. & O'Nions, R. K. Chem. Geol. 75, 252-290 (1989). the author was a visiting investigator at the IPGP in Paris as recipient of the MESR-A. v. Humboldt Award.
83. O'Nions, R. K. & Tolstikhin, I. N. Earth Planet. Sci. Lett. 139, 213-222 (1996).
84. Anderson, D. L. Earth Planet. Sci. Lett. 57, 13-24 (1982). Correspondence should be addressed to A.W.H. (e-mail: [email protected]).

NATURE I VOL 3851 16 JANUARY 1997 229