REPORTS

that proposed to explain, in terms of sp-d bond- down to about 5% at 300 K in Co/STO/
ing, the positive polarization at the Co-ALO LSMO (4). However, other types of oxides of
interface (8). However, there is no general the- the double-perovskite family (for example,
ory predicting the trend of the experimental Sr2FeMoO6) combine electronic properties
results for Co—that is, a negative polarization similar to those of manganites with a defi-
with oxides of d elements (STO, CLO, Ta2O5) nitely higher Curie temperature (15). Their
and a positive one when there are only s and p use in magnetic tunnel junctions is promising
states (ALO). It is likely that the spin polariza- for a new generation of tunnel junctions with
tion should also depend on the position of the very high magnetoresistance for room-tem-
Fermi level with respect to the electronic levels perature applications.
of each character above and below the gap of
the insulator. In addition, as an evanescent References and Notes
1. J. S. Moodera, L. R. Kinder, T. M. Wong, R. Meservey,
wave in an insulator is a Bloch wave with an Phys. Rev. Lett. 74, 3273 (1995).
imaginary wave vector, one can expect differ- 2. J. De Boeck, Science 281, 357 (1998).
ent decay lengths for Bloch waves of different 3. M. Sharma, S. X. Wang, J. H. Nickel, Phys. Rev. Lett.
82, 616 (1999).
character. This means that the final polarization 4. J. M. De Teresa et al., ibid., p. 4288.
could also depend on the thickness of the bar- 5. P. M. Tedrow and R. Meservey, Phys. Rev. B 7, 318
rier, as illustrated by the calculations of Mac- (1973); R. Meservey and P. M. Tedrow, Phys. Rep.
Laren et al. for Fe/ZnSe/Fe junctions (14). 238, 173 (1994).
6. M. B. Stearns, J. Magn. Magn. Mater. 5, 187 (1997).

Downloaded from http://science.sciencemag.org/ on February 17, 2021

The influence of the barrier on the spin 7. J. A. Hertz and K. Aoi, Phys. Rev. B 8, 3252 (1973).
polarization opens new ways to shape and op- 8. D. Nguyen-Mahn et al., Mat. Res. Soc. Symp. Proc.
timize the TMR. Interesting bias dependencies 492, 319 (1998).
9. J. H. Park et al., Nature 392, 794 (1998).
can be obtained with barriers selecting the d 10. J. Nassar, M. Hehn, A. Vaurès, F. Petroff, A. Fert, Appl.
Fig. 3. Bias dependence of the TMR ratio in (A) electrons and probing the fine structure of the Phys. Lett. 73, 698 (1998).
Co/STO/LSMO and (B) Co/ALO/STO/LSMO 11. J. S. Moodera et al., ibid. 81, 1953 (1998).
d-DOS, as in Fig. 3A. The DOS of a d-band can
tunnel junctions. 12. K. Wang, thesis, New York University (1999).
also be easily tailored by alloying (for example, 13. S. Zhang, P. M. Levy, A. C. Marley, S. S. P. Parkin, Phys.
by introduction of virtual bound states) to pro- Rev. Lett. 79, 3744 (1997).
level of LSMO is situated above the Fermi duce specific bias dependencies. Although here 14. J. M. MacLaren, X. G. Zhang, W. H. Butler, X. Wang,
Phys. Rev. B 59, 5470 (1999).
level of Co and a maximum of inverse TMR we concentrated on the problem of the spin 15. K. I. Kobayashi, T. Kimura, H. Sawada, K. Terakura, Y.
is expected when the Fermi level of LSMO is polarization of the Co electrode and regarded Tokura, Nature 395, 677 (1998).
approximately at the maximum of the spin 2 the strongly spin-polarized LSMO only as a 16. We thank the group of A. Revcoleschi for providing
DOS of Co. This is consistent with the max- LSMO targets, R. Lyonnet and A. Vaures for their
useful spin analyzer, the large TMR ratios ob-
experimental help and J. Nassar for fruitful discus-
imum of inverse TMR observed at ⫺0.4 V tained by combining Co and LSMO electrodes sions. Supported by the European Union and the New
for Co/STO/LSMO junctions (Fig. 3A). For a (50% with a STO barrier) are also an interesting Energy and Industrial Technology Development Or-
positive bias, the TMR is expected to change result. The drawback arising from the low ganization of Japan.
sign and become normal above 1 V when the Curie temperature of LSMO (⬃350 K) is the
Fermi level of LSMO goes down into the reduction of the TMR at room temperature, 1 July 1999; accepted 2 September 1999
energy range of the majority spin d-band of
Co. This is also observed in Fig. 3A.
For ALO and ALO/STO barriers, a predom-
inant tunneling of s-character electrons (see ar- Emergence of Scaling in
row in Fig. 2B) is the usual explanation of the
positive polarization (6 – 8). The rapid drop Random Networks
with bias (Fig. 3B) is similar to what has been
observed in most junctions with ALO barriers, Albert-László Barabási* and Réka Albert
and completely different from what is obtained
when the tunneling is predominantly by d-char- Systems as diverse as genetic networks or the World Wide Web are best
acter electrons (Fig. 3A). The origin of this described as networks with complex topology. A common property of many
rapid decrease of the TMR at relatively small large networks is that the vertex connectivities follow a scale-free power-law
bias has never been clearly explained. This is distribution. This feature was found to be a consequence of two generic mech-
roughly consistent with the energy dependence anisms: (i) networks expand continuously by the addition of new vertices, and
of the DOS induced by sp-d bonding effects on (ii) new vertices attach preferentially to sites that are already well connected.
the first atomic layer of ALO in the calculation A model based on these two ingredients reproduces the observed stationary
of Nguyen-Mahn et al. (8) for the Co-ALO scale-free distributions, which indicates that the development of large networks
interface. But Zhang et al. (13) have also shown is governed by robust self-organizing phenomena that go beyond the particulars
that a large part of the TMR drop can be of the individual systems.
attributed to the excitation of spin waves.
The experiments reported here and in sev- The inability of contemporary science to de- actions currently limits advances in many
eral recent publications (3, 4) demonstrate the scribe systems composed of nonidentical el- disciplines, ranging from molecular biology
important role of the electronic structure of the ements that have diverse and nonlocal inter- to computer science (1). The difficulty of
metal-oxide interface in determining the spin describing these systems lies partly in their
polarization of the tunneling electrons. The neg- Department of Physics, University of Notre Dame, topology: Many of them form rather complex
ative polarization for the Co-STO interface has Notre Dame, IN 46556, USA. networks whose vertices are the elements of
been ascribed to d-d bonding effects between *To whom correspondence should be addressed. E- the system and whose edges represent the
Al and Ti (4). This interpretation is similar to mail: [email protected] interactions between them. For example, liv-

www.sciencemag.org SCIENCE VOL 286 15 OCTOBER 1999 509

REPORTS
ing systems form a huge genetic network these two ingredients, we show that they are cited in a paper. Recently Redner (11) has
whose vertices are proteins and genes, the responsible for the power-law scaling ob- shown that the probability that a paper is
chemical interactions between them repre- served in real networks. Finally, we argue cited k times (representing the connectivity of
senting edges (2). At a different organization- that these ingredients play an easily identifi- a paper within the network) follows a power
al level, a large network is formed by the able and important role in the formation of law with exponent ␥cite ⫽ 3.
nervous system, whose vertices are the nerve many complex systems, which implies that The above examples (12) demonstrate that
cells, connected by axons (3). But equally our results are relevant to a large class of many large random networks share the com-
complex networks occur in social science, networks observed in nature. mon feature that the distribution of their local
where vertices are individuals or organiza- Although there are many systems that connectivity is free of scale, following a power
tions and the edges are the social interactions form complex networks, detailed topological law for large k with an exponent ␥ between
between them (4), or in the World Wide Web data is available for only a few. The collab- 2.1 and 4, which is unexpected within the
(WWW), whose vertices are HTML docu- oration graph of movie actors represents a framework of the existing network models.
ments connected by links pointing from one well-documented example of a social net- The random graph model of ER (7) assumes
page to another (5, 6). Because of their large work. Each actor is represented by a vertex, that we start with N vertices and connect each
size and the complexity of their interactions, two actors being connected if they were cast pair of vertices with probability p. In the
the topology of these networks is largely together in the same movie. The probability model, the probability that a vertex has k
unknown. that an actor has k links (characterizing his or edges follows a Poisson distribution P(k) ⫽
Traditionally, networks of complex topol- her popularity) has a power-law tail for large e ⫺␭ ␭ k /k!, where

冉 冊
ogy have been described with the random k, following P(k) ⬃ k ⫺␥ actor, where ␥actor ⫽
N⫺1
2.3 ⫾ 0.1 (Fig. 1A). A more complex net-

Downloaded from http://science.sciencemag.org/ on February 17, 2021

graph theory of Erdős and Rényi (ER) (7),
but in the absence of data on large networks, work with over 800 million vertices (8) is the ␭⫽N p k 共1 ⫺ p兲 N⫺1⫺k
k
the predictions of the ER theory were rarely WWW, where a vertex is a document and the
tested in the real world. However, driven by edges are the links pointing from one docu- In the small-world model recently intro-
the computerization of data acquisition, such ment to another. The topology of this graph duced by Watts and Strogatz (WS) (10), N
topological information is increasingly avail- determines the Web’s connectivity and, con- vertices form a one-dimensional lattice,
able, raising the possibility of understanding sequently, our effectiveness in locating infor- each vertex being connected to its two
the dynamical and topological stability of mation on the WWW (5). Information about nearest and next-nearest neighbors. With
large networks. P(k) can be obtained using robots (6), indi- probability p, each edge is reconnected to a
Here we report on the existence of a high cating that the probability that k documents vertex chosen at random. The long-range
degree of self-organization characterizing the point to a certain Web page follows a power connections generated by this process de-
large-scale properties of complex networks. law, with ␥www ⫽ 2.1 ⫾ 0.1 (Fig. 1B) (9). A crease the distance between the vertices,
Exploring several large databases describing network whose topology reflects the histori- leading to a small-world phenomenon (13),
the topology of large networks that span cal patterns of urban and industrial develop- often referred to as six degrees of separa-
fields as diverse as the WWW or citation ment is the electrical power grid of the west- tion (14 ). For p ⫽ 0, the probability distri-
patterns in science, we show that, indepen- ern United States, the vertices being genera- bution of the connectivities is P(k) ⫽ ␦(k ⫺
dent of the system and the identity of its tors, transformers, and substations and the z), where z is the coordination number in
constituents, the probability P(k) that a ver- edges being to the high-voltage transmission the lattice; whereas for finite p, P(k) still
tex in the network interacts with k other lines between them (10). Because of the rel- peaks around z, but it gets broader (15). A
vertices decays as a power law, following atively modest size of the network, contain- common feature of the ER and WS models
P(k) ⬃ k ⫺␥ . This result indicates that large ing only 4941 vertices, the scaling region is is that the probability of finding a highly
networks self-organize into a scale-free state, less prominent but is nevertheless approxi- connected vertex (that is, a large k) decreas-
a feature unpredicted by all existing random mated by a power law with an exponent es exponentially with k; thus, vertices with
network models. To explain the origin of this ␥power ⯝ 4 (Fig. 1C). Finally, a rather large large connectivity are practically absent. In
scale invariance, we show that existing net- complex network is formed by the citation contrast, the power-law tail characterizing
work models fail to incorporate growth and patterns of the scientific publications, the ver- P(k) for the networks studied indicates that
preferential attachment, two key features of tices being papers published in refereed jour- highly connected (large k) vertices have a
real networks. Using a model incorporating nals and the edges being links to the articles large chance of occurring, dominating the
connectivity.
There are two generic aspects of real net-
works that are not incorporated in these mod-
els. First, both models assume that we start
with a fixed number (N) of vertices that are
then randomly connected (ER model), or re-
connected (WS model), without modifying
N. In contrast, most real world networks are
open and they form by the continuous addi-
tion of new vertices to the system, thus the
number of vertices N increases throughout
the lifetime of the network. For example, the
actor network grows by the addition of new
actors to the system, the WWW grows expo-
Fig. 1. The distribution function of connectivities for various large networks. (A) Actor collaboration
nentially over time by the addition of new
graph with N ⫽ 212,250 vertices and average connectivity 具k典 ⫽ 28.78. (B) WWW, N ⫽ Web pages (8), and the research literature
325,729, 具k典 ⫽ 5.46 (6). (C) Power grid data, N ⫽ 4941, 具k典 ⫽ 2.67. The dashed lines have constantly grows by the publication of new
slopes (A) ␥actor ⫽ 2.3, (B) ␥www ⫽ 2.1 and (C) ␥power ⫽ 4. papers. Consequently, a common feature of

510 15 OCTOBER 1999 VOL 286 SCIENCE www.sciencemag.org

REPORTS
these systems is that the network continuous- that a new vertex is connected with equal m 2 t/k 2 ) ⫽ 1 ⫺ m 2 t/k 2 (t ⫹ m 0 ). The prob-
ly expands by the addition of new vertices probability to any vertex in the system [that ability density P(k) can be obtained from
that are connected to the vertices already is, ⌸(k) ⫽ const ⫽ 1/(m 0 ⫹ t ⫺ 1)]. Such P(k) ⫽ ⳵P[k i (t) ⬍ k]/⳵k, which over long
present in the system. a model (Fig. 2B) leads to P(k) ⬃ time periods leads to the stationary solution
Second, the random network models as- exp(⫺␤k), indicating that the absence of
2m 2
sume that the probability that two vertices are preferential attachment eliminates the scale- P共k兲 ⫽
connected is random and uniform. In con- free feature of the distribution. In model B, k3
trast, most real networks exhibit preferential we start with N vertices and no edges. At giving ␥ ⫽ 3, independent of m. Although it
connectivity. For example, a new actor is each time step, we randomly select a vertex reproduces the observed scale-free distribu-
most likely to be cast in a supporting role and connect it with probability ⌸(k i ) ⫽ k i / tion, the proposed model cannot be expected
with more established and better-known ac- ⌺ j k j to vertex i in the system. Although at to account for all aspects of the studied net-
tors. Consequently, the probability that a new early times the model exhibits power-law works. For that, we need to model these
actor will be cast with an established one is scaling, P(k) is not stationary: because N is systems in more detail. For example, in the
much higher than that the new actor will be constant and the number of edges increases model we assumed linear preferential attach-
cast with other less-known actors. Similarly, with time, after T ⯝ N 2 time steps the system ment; that is, ⌸(k) ⬃ k. However, although
a newly created Web page will be more likely reaches a state in which all vertices are con- in general ⌸(k) could have an arbitrary non-
to include links to well-known popular doc- nected. The failure of models A and B indi- linear form ⌸(k) ⬃ k ␣ , simulations indicate
uments with already-high connectivity, and a cates that both ingredients— growth and pref- that scaling is present only for ␣ ⫽ 1. Fur-
new manuscript is more likely to cite a well- erential attachment—are needed for the de- thermore, the exponents obtained for the dif-

Downloaded from http://science.sciencemag.org/ on February 17, 2021

known and thus much-cited paper than its velopment of the stationary power-law distri- ferent networks are scattered between 2.1 and
less-cited and consequently less-known peer. bution observed in Fig. 1. 4. However, it is easy to modify our model to
These examples indicate that the probability Because of the preferential attachment, a account for exponents different from ␥ ⫽ 3.
with which a new vertex connects to the vertex that acquires more connections than For example, if we assume that a fraction p of
existing vertices is not uniform; there is a another one will increase its connectivity at a the links is directed, we obtain ␥( p) ⫽ 3 ⫺
higher probability that it will be linked to a higher rate; thus, an initial difference in the p, which is supported by numerical simula-
vertex that already has a large number of connectivity between two vertices will in- tions (16). Finally, some networks evolve not
connections. crease further as the network grows. The rate only by adding new vertices but by adding
We next show that a model based on these at which a vertex acquires edges is ⳵k i /⳵t ⫽ (and sometimes removing) connections be-
two ingredients naturally leads to the ob- k i / 2t, which gives k i (t) ⫽ m(t/t i ) 0.5 , where tween established vertices. Although these
served scale-invariant distribution. To incor- t i is the time at which vertex i was added to and other system-specific features could
porate the growing character of the network, the system (see Fig. 2C), a scaling property modify the exponent ␥, our model offers the
starting with a small number (m 0 ) of vertices, that could be directly tested once time-re- first successful mechanism accounting for the
at every time step we add a new vertex with solved data on network connectivity becomes scale-invariant nature of real networks.
m(ⱕm 0 ) edges that link the new vertex to m available. Thus older (with smaller t i ) verti- Growth and preferential attachment are
different vertices already present in the sys- ces increase their connectivity at the expense mechanisms common to a number of com-
tem. To incorporate preferential attachment, of the younger (with larger t i ) ones, leading plex systems, including business networks
we assume that the probability ⌸ that a new over time to some vertices that are highly (17, 18), social networks (describing individ-
vertex will be connected to vertex i depends connected, a “rich-get-richer” phenomenon uals or organizations), transportation net-
on the connectivity k i of that vertex, so that that can be easily detected in real networks. works (19), and so on. Consequently, we
⌸(k i ) ⫽ k i /⌺ j k j . After t time steps, the Furthermore, this property can be used to expect that the scale-invariant state observed
model leads to a random network with t ⫹ calculate ␥ analytically. The probability that in all systems for which detailed data has
m 0 vertices and mt edges. This network a vertex i has a connectivity smaller than k, been available to us is a generic property of
evolves into a scale-invariant state with the P[k i (t) ⬍ k], can be written as P(t i ⬎ many complex networks, with applicability
probability that a vertex has k edges, follow- m 2 t/k 2 ). Assuming that we add the vertices reaching far beyond the quoted examples. A
ing a power law with an exponent ␥model ⫽ to the system at equal time intervals, we better description of these systems would
2.9 ⫾ 0.1 (Fig. 2A). Because the power law obtain P(t i ⬎ m 2 t/k 2 ) ⫽ 1 ⫺ P(t i ⱕ help in understanding other complex systems
observed for real networks describes systems
of rather different sizes at different stages of
their development, it is expected that a cor-
rect model should provide a distribution
whose main features are independent of time.
Indeed, as Fig. 2A demonstrates, P(k) is
independent of time (and subsequently inde-
pendent of the system size m 0 ⫹ t), indicat-
ing that despite its continuous growth, the
system organizes itself into a scale-free sta-
tionary state.
The development of the power-law scal-
ing in the model indicates that growth and
preferential attachment play an important role
in network development. To verify that both
Fig. 2. (A) The power-law connectivity distribution at t ⫽ 150,000 (E) and t ⫽ 200,000 (䊐) as
ingredients are necessary, we investigated obtained from the model, using m 0 ⫽ m ⫽ 5. The slope of the dashed line is ␥ ⫽ 2.9. (B) The
two variants of the model. Model A keeps the exponential connectivity distribution for model A, in the case of m 0 ⫽ m ⫽ 1 (E), m 0 ⫽ m ⫽
growing character of the network, but prefer- 3 (䊐), m 0 ⫽ m ⫽ 5 ({), and m 0 ⫽ m ⫽ 7 (‚). (C) Time evolution of the connectivity for two
ential attachment is eliminated by assuming vertices added to the system at t 1 ⫽ 5 and t 2 ⫽ 95. The dashed line has slope 0.5.

www.sciencemag.org SCIENCE VOL 286 15 OCTOBER 1999 511

REPORTS
as well, for which less topological informa- References and Notes diagram of a computer chip (see http://vlsicad.cs.
1. R. Gallagher and T. Appenzeller, Science 284, 79 ucla.edu/⬃cheese/ispd98.html). We found that P(k)
tion is currently available, including such for both was consistent with power-law tails, despite
(1999); R. F. Service, ibid., p. 80.
important examples as genetic or signaling the fact that for C. elegans the relatively small size of
2. G. Weng, U. S. Bhalla, R. Iyengar, ibid., p. 92.
networks in biological systems. We often do the system (306 vertices) severely limits the data
3. C. Koch and G. Laurent, ibid., p. 96.
quality, whereas for the wiring diagram of the chips,
not think of biological systems as open or 4. S. Wasserman and K. Faust, Social Network Analysis
vertices with over 200 edges have been eliminated
growing, because their features are genetical- (Cambridge Univ. Press, Cambridge, 1994).
from the database.
5. Members of the Clever project, Sci. Am. 280, 54
ly coded. However, possible scale-free fea- 13. S. Milgram, Psychol. Today 2, 60 (1967); M. Kochen,
( June 1999).
ed., The Small World (Ablex, Norwood, NJ, 1989).
tures of genetic and signaling networks could 6. R. Albert, H. Jeong, A.-L. Barabási, Nature 401, 130 14. J. Guare, Six Degrees of Separation: A Play ( Vintage
reflect the networks’ evolutionary history, (1999); A.-L. Barabási, R. Albert, H. Jeong, Physica A Books, New York, 1990).
272, 173 (1999); see also http://www.nd.edu/~net- 15. M. Barthélémy and L. A. N. Amaral, Phys. Rev. Lett.
dominated by growth and aggregation of dif- works. 82, 15 (1999).
ferent constituents, leading from simple mol- 7. P. Erdős and A. Rényi, Publ. Math. Inst. Hung. Acad. 16. For most networks, the connectivity m of the newly
ecules to complex organisms. With the fast Sci. 5, 17 (1960); B. Bollobás, Random Graphs (Aca- added vertices is not constant. However, choosing m
demic Press, London, 1985). randomly will not change the exponent ␥ ( Y. Tu,
advances being made in mapping out genetic 8. S. Lawrence and C. L. Giles, Science 280, 98 (1998); personal communication).
networks, answers to these questions might Nature 400, 107 (1999). 17. W. B. Arthur, Science 284, 107 (1999).
not be too far away. Similar mechanisms 9. In addition to the distribution of incoming links, the 18. Preferential attachment was also used to model
WWW displays a number of other scale-free features evolving networks (L. A. N. Amaral and M. Bar-
could explain the origin of the social and characterizing the organization of the Web pages thélémy, personal communication).
economic disparities governing competitive within a domain [B. A. Huberman and L. A. Adamic, 19. J. R. Banavar, A. Maritan, A. Rinaldo, Nature 399, 130
systems, because the scale-free inhomogene- Nature 401, 131 (1999)], the distribution of searches (1999).
[B. A. Huberman, P. L. T. Pirolli, J. E. Pitkow, R. J. 20. We thank D. J. Watts for providing the C. elegans and
ities are the inevitable consequence of self- Lukose, Science 280, 95 (1998)], or the number of

Downloaded from http://science.sciencemag.org/ on February 17, 2021

power grid data, B. C. Tjaden for supplying the actor
organization due to the local decisions made links per Web page (6). data, H. Jeong for collecting the data on the WWW,
by the individual vertices, based on informa- 10. D. J. Watts and S. H. Strogatz, Nature 393, 440 and L. A. N. Amaral for helpful discussions. This work
(1998). was partially supported by NSF Career Award DMR-
tion that is biased toward the more visible
11. S. Redner, Eur. Phys. J. B 4, 131 (1998). 9710998.
(richer) vertices, irrespective of the nature 12. We also studied the neural network of the worm
and origin of this visibility. Caenorhabditis elegans (3, 10) and the benchmark 24 June 1999; accepted 2 September 1999

Osmium Isotope Constraints on suprasubduction xenolith locality, the Tubaf

seamount in the Lihir island group of the

Ore Metal Recycling in Tabar-Lihir-Tanga-Feni island arc in Papua

New Guinea (Fig. 1). This xenolith locality is
important for the following reasons: (i) it
Subduction Zones contains samples that represent a complete
section of oceanic lithosphere at an intraoce-
Brent I. A. McInnes,1* Jannene S. McBride,2 Noreen J. Evans,1 anic convergent margin, (ii) it is located ad-
David D. Lambert,2 Anita S. Andrew 3 jacent to one of the world’s largest and
youngest volcano-hosted Au deposits, and
Veined peridotite xenoliths from the mantle beneath the giant Ladolam gold (iii) it contains metasomatized mantle perido-
deposit on Lihir Island, Papua New Guinea, are 2 to 800 times more enriched tite xenoliths with Au-enriched vein minerals
in copper, gold, platinum, and palladium than surrounding depleted arc mantle. that crystallized in the mantle from oxidizing,
Gold ores have osmium isotope compositions similar to those of the underlying alkali- and sulfur-rich hydrous fluids.
subduction-modified mantle peridotite source region, indicating that the pri- During the oceanographic investigation of
mary origin of the metals was the mantle. Because the mantle is relatively submarine hydrothermal systems in Papua
depleted in gold, copper, and palladium, tectonic processes that enhance the New Guinea (8), a submarine cinder cone
advective transport and concentration of these fluid soluble metals may be a (Tubaf volcano, 1280 m below sea level; 3°
prerequisite for generating porphyry-epithermal copper-gold deposits. 15.25⬘ S, 152° 32.50⬘ E) was discovered 14
km southwest of the giant Ladolam gold mine
The tectonic relationship between subduction- to 187Os) is a tracer of metallogenic processes at (⬎40 million oz contained Au) on Lihir Is-
related magmatism at convergent margins and convergent margins because both elements land. Dredge and video-grab sampling of the
porphyry copper-gold (Cu-Au) ore formation have geochemical properties similar to metals 1-km-diameter volcanic cone returned 130
has long been recognized (1). However, the that occur in porphyry ore deposits (2, 3). Be- ultramafic, mafic, and sedimentary xenoliths.
physical and chemical processes that govern cause Re is highly concentrated in crustal rocks The study of these samples has provided an
Cu-Au metallogeny and the ultimate source(s) and Os is concentrated in the mantle (4, 5), this unprecedented view of the source region of
of the metals in these ore deposits are poorly isotopic system is particularly useful for quan- an island arc magmatic system with a propen-
understood. The rhenium-osmium (Re-Os) iso- tifying the flux of ore elements in island arc sity to produce giant porphyry-epithermal ore
topic system (based on the ␤⫺ decay of 187Re settings where the two principal reservoirs for deposits. The xenolith assemblage includes
metals are subducted crust and mantle wedge spinel lherzolite, harzburgite, websterite, or-
1
Commonwealth Scientific and Industrial Research peridotite. thopyroxenite, clinopyroxenite, syenite, ser-
Organization Exploration and Mining, Post Office Box Os isotope studies in subduction zones are pentinite, gabbro, hornblende gabbro, dia-
136, North Ryde, New South Wales 1670, Australia; currently limited because of the rarity of base, basalt, pelagic deep-sea sediment, and
2
Department of Earth Sciences, Monash University, deep-seated suprasubduction samples (rocks shallow-water volcaniclastic sediment as well
Clayton, Victoria 3168, Australia; 3Commonwealth
Scientific and Industrial Research Organization Petro-
overlying a subducted slab). Previous studies as coralline and coralgal limestone. These
leum Resources, Post Office Box 136, North Ryde, (6, 7) have demonstrated that radiogenic Os lithologies represent a cross section of the su-
New South Wales 1670, Australia. is introduced into the subarc mantle by hy- prasubduction assemblage and can be reassem-
*To whom correspondence should be addressed. E- drous, oxidizing fluids derived during slab bled into an “ophiolite-type” model of oceanic
mail: [email protected] dehydration. We report on results from a lithosphere (Fig. 1) (9).

512 15 OCTOBER 1999 VOL 286 SCIENCE www.sciencemag.org

Emergence of Scaling in Random Networks
Albert-László Barabási and Réka Albert

Science 286 (5439), 509-512.

DOI: 10.1126/science.286.5439.509

Downloaded from http://science.sciencemag.org/ on February 17, 2021

ARTICLE TOOLS http://science.sciencemag.org/content/286/5439/509

RELATED http://science.sciencemag.org/content/sci/309/5736/877.full
CONTENT

REFERENCES This article cites 13 articles, 3 of which you can access for free
http://science.sciencemag.org/content/286/5439/509#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. The title Science is a registered trademark of AAAS.
Copyright © 1999 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science.
No claim to original U.S. Government Works.