Content uploaded by Dapeng Zhao
Author content
All content in this area was uploaded by Dapeng Zhao on Jun 15, 2019
Content may be subject to copyright.
Physica C 191, 485 (1992).
20. C. J. Muller et al.,Phys. Rev. B 53, 1022 (1996), and
references therein.
21. J. M. Tour et al.,J. Am. Chem. Soc. 117, 9529
(1995).
22. C. Zhou, C. J. Muller, M. R. Deshpande, J. W.
Sleight, M. A. Reed, Appl. Phys. Lett. 67, 1160
(1995).
23. V. Mujica, M. Kemp, A. Roitberg, M. Ratner, J.
Chem. Phys. 104, 7296 (1996).
24. M. P. Samanta et al.,Phys. Rev.53B, R7626 (1996).
25. S. Datta et al.,Phys. Rev. Lett., in press; W. Tian et
al.,Physica E, in press.
26. We thank M. R. Deshpande for fruitful discussions
and the Defense Advanced Research Projects
Agengy for support.
25 April 1997; accepted 22 August 1997
Depth Extent of the Lau Back-Arc Spreading
Center and Its Relation to Subduction Processes
Dapeng Zhao,* Yingbiao Xu, Douglas A. Wiens, LeRoy Dorman,
John Hildebrand, Spahr Webb
Seismic tomography and wave form inversion revealed that very slow velocity anomalies
(5 to 7 percent) beneath the active Lau spreading center extend to 100-kilometer depth
and are connected to moderately slow anomalies (2 to 4 percent) in the mantle wedge
to 400-kilometer depth. These results indicate that geodynamic systems associated with
back-arc spreading are related to deep processes, such as the convective circulation in
the mantle wedge and deep dehydration reactions in the subducting slab. The slow
regions associated with the Tonga arc and the Lau back arc are separated at shallow
levels but merge at depths greater than 100 kilometers, suggesting that slab components
of back-arc magmas occur through mixing at these depths.
Knowledge of the seismic structure be-
neath back-arc spreading centers is impor-
tant because the width and depth of the
slow-velocity regions below spreading cen-
ters provide constraints on the origin of
back-arc spreading (1,2), the geochemical
source of arc and back-arc magmas (3), the
interaction between subduction and back-
arc spreading (1), whether the mantle up-
welling beneath spreading centers is passive
or active, and to what depth the upwelling
persists (2). A subduction zone with an
associated back-arc spreading center and
the existence of deep earthquakes immedi-
ately beneath the center provide an ideal
geometry to image and understand back-arc
spreading processes. The Tonga-Fiji region,
which contains two-thirds of all deep earth-
quakes in the world, represents an optimal
region for exploring these questions. Previ-
ous studies have discussed the seismic ve-
locity anomalies due to the Tonga slab (4,
5), but this work has been hampered by the
poor distribution of seismic stations.
The installation (6) of 12 broadband
stations in the Tonga and Fiji islands from
November 1993 through December 1995
and a related 3-month deployment of 25
ocean bottom seismometers (OBS) (7)in
the Lau back arc and the Tonga forearc
provided a unique opportunity to determine
high-resolution three-dimensional (3D)
structure in this region (Fig. 1A). We used
41,471 arrival times from 926 earthquakes
that occurred in the Tonga-Fiji region dur-
ing the seismic experiment (Fig. 1B). Most
of the events were associated with the sub-
duction of the Tonga slab; they had a rela-
tively uniform distribution in the entire
upper mantle. This uniform distribution is
an advantageous feature over other subduc-
tion zones, such as Japan and Alaska, where
most of the seismicity is concentrated at
depths shallower than 250 km (8,9). We
picked about 8200 arrival times at the 12
land stations from the 926 earthquakes and
about 2900 arrivals at the 25 OBS stations
from 250 earthquakes that occurred during
the OBS deployment. The picking accuracy
is estimated to be 0.05 to 0.3 s. The remain-
ing arrival times were recorded by stations
reporting to the Preliminary Determination
of Epicenters (PDE) with epicentral dis-
tances up to 90°. The PDE arrival times
have lower quality (picking accuracy of 0.2
to 0.5 s), so they were assigned less than
half the weight of the local data. All of the
926 earthquakes were recorded by more
than 20 stations, and their hypocentral lo-
cations have a statistical accuracy of 63to
9 km. We also picked 450 arrival times at
the 12 land and 25 OBS stations from 45
large (magnitude of 6.0 to 8.0) teleseismic
events with epicentral distances from 30° to
90°, which were assigned the same weight
as the local data in the inversion.
We used a tomography method (9)to
determine the 3D Pwave velocity structure
in the Tonga-Fiji region (9,10) (Figs. 2 and
3). To confirm that the major velocity fea-
tures were adequately resolved by the inver-
sion, we conducted checkerboard resolution
tests (11) (Fig. 4). The checkerboard test
with a grid spacing of 50 km indicates good
resolution for the area in and around the
subducting Tonga slab and along the main
line of OBSs (Fig. 4, A and B). For the test
with a grid spacing of 70 km, the resolution
is good for all the areas discussed (Fig. 4, C
and D). We also conducted a number of
inversions and resolution tests by changing
the grid spacing, the grid configuration, and
the initial model (10). The results show
that the velocity structure in the study area
(Fig. 3) can be resolved with a resolution of
50 to 70 km. This resolution scale is better
than the 100-to 200-km resolution ob-
D. Zhao, Southern California Earthquake Center and De-
partment of Earth Sciences, University of Southern Cali-
fornia, Los Angeles, CA 90089, USA.
Y. Xu and D. A. Wiens, Department of Earth and Plane-
tary Sciences, Washington University, St. Louis, MO
63130, USA.
L. Dorman, J. Hildebrand, S. Webb, Scripps Institution of
Oceanography, University of California, San Diego, La
Jolla, CA 92093, USA.
*To whom correspondence should be addressed. E-mail:
dzhao@usc.edu
Fig. 1. (A) Map showing the seismometer deploy-
ments in the Fiji-Tonga region. Twelve broadband
instrument island sites, 25 OBS sites, and two
PDE sites (at Samoa and Fiji) recorded the data
used in this study. A 2-year sample of deep earth-
quakes (depths of 300 to 680 km and m
b
.4.8)
(dots) delineates the deep Tonga slab. PASSCAL,
Program for Array Seismic Studies of the Conti-
nental Lithosphere; IRIS, Incorporated Research
Institutions in Seismology; GSN, Global Seismo-
graphic Network. (B) Hypocentral distribution of
the 926 earthquakes used in this study.
SCIENCE zVOL. 278 z10 OCTOBER 1997 zwww.sciencemag.org254
tained in previous studies (5).
The subducting Tonga slab was imaged
as a 100-km-thick zone with a Pwave
velocity that is 4 to 6% higher than the
surrounding mantle (Fig. 2). Beneath the
Tonga arc and the Lau back arc, low-
velocity anomalies of up to 7% are visible
(Figs. 2 and 3). The slow-velocity anomaly
beneath the Tonga arc represents a dip-
ping region about 30 to 50 km above the
slab, extending from the surface to about
140-km depth (Fig. 2). This feature is
similar to the low-velocity features found
beneath the Japan and Alaska volcanic
fronts (8,9). This slow anomaly probably
represents the source zone for island arc
magmas. Volatiles released from the sub-
ducting slab may reduce the melting point
of the rock above and allow partial melt-
ing to produce arc magmas (3,12). Slow
anomalies beneath areas of the active
Central Lau Spreading Center (CLSC)
and the Eastern Lau Spreading Center
(ELSC) extend to depths of about 100 km.
These depths correspond to regions where
the primary magma genesis is expected to
take place beneath an oceanic spreading
center (13,14). The maximum heteroge-
neity of Pwave velocity between the Lau
back-arc basin and the Pacific Plate is
about 13% at these depths. Slow anoma-
lies are located to the west of the CLSC
and ELSC (Figs. 2 and 3). Beneath 100-
km depth, the amplitude of the back-arc
anomalies is reduced, but a moderately
slow anomaly (22to24%) exists down to
a depth of at least 400 km. To investigate
the depth extent of slow anomalies in the
Lau back arc with a different methodolo-
gy, we inverted 16 wave forms from seven
regional earthquakes recorded at the land
stations to determine the 1D Swave struc-
ture beneath the Lau Basin (15). The
inversion results (Fig. 5) show a similar
level of velocity heterogeneity and depth
distribution of the back-arc anomalies to
that found in the Pwave tomography. The
level of Swave velocity heterogeneity
reaches a maximum of about 18% between
the Lau Basin and the old Pacific litho-
sphere at depths of 40 to 90 km (Fig. 5).
The velocity difference decreases to about
2% at 180-km depth, but a small, poorly
resolved difference persists to greater
depths (15). There has been disagreement
concerning the depth extent of slow-ve-
locity anomalies at mid-ocean ridges
(MORs) (13,16). Our results show that,
at least for back-arc spreading centers,
moderately slow velocity anomalies ex-
tend to depths of at least 400 km. These
anomalies may reflect either the depth
extent of oceanic spreading centers due to
the depth of the associated upwelling pat-
terns or processes endemic to back-arc
spreading centers, perhaps due to interac-
tions between the slab and the back arc.
The slow-velocity anomalies at depths
of 300 to 400 km (Fig. 2) could be caused
by upwelling flow patterns in the back-arc
region or by volatiles resulting from the
deep dehydration reactions occurring in
the subducting Tonga slab. Volatiles
would have the effect of lowering the
melting temperature and the seismic ve-
locity and may produce small amounts of
partial melt (17). Temperatures in fast
subducting slabs like Tonga are low
enough for water to reach the stability
depths of dense hydrous magnesian silicate
phases (18), which may allow water pen-
etration down to depths of 660 km (18,
19). The phase diagrams of important hy-
drous phases, the associated reaction ki-
netics, and the relevant mantle conditions
(slab temperature and composition) are
not known sufficiently well enough to pre-
dict the depth at which dehydration would
occur. Partial melting of the back-arc re-
gion by volatiles from the deep slab may
be important in localizing low seismic ve-
locities in the back arc; the slow anomalies
we observed at depths of 300 to 400 km
may represent this process.
The slowest anomaly in the back-arc
region is not found beneath the spreading
center, but rather to the west. This is sim-
0
100
200
300
400
500
600
700
Depth (km)
-6% -3% 0% 6%3%
Fiji CLSC ELSC
Lau
ridge Tonga arc
volcanoes
Fig. 2. East-west vertical
cross section of a Pwave ve-
locity image from 0- to 700-
km depth along the line AB
(1220-km length) in Fig. 3A.
Red and blue colors denote
slow and fast velocities, re-
spectively. Solid triangles de-
note active volcanoes. CLSC
denotes the location of the
Central Lau Spreading Center
and ELSC denotes the loca-
tion of the Eastern Lau
Spreading Center. Earth-
quakes within a 40-km width
from the cross section are
shown as white circles. The
velocity perturbation scale is
shown at the bottom.
177˚E 179˚W 175˚W 171˚W
A
C
EF
B
A
B
D
177˚E 179˚W 175˚W 171˚W
15˚S
19˚S
23˚S
15˚S
19˚S
23˚S
15˚S
19˚S
23˚S
Fig. 3. Pwave velocity
images at (A) 25-, (B) 60-,
(C) 100-, (D) 140-, (E)
230-, and (F) 430-km
depths. Earthquakes
within a 20-km depth
range of the slice are
shown as white circles.
The long contour line to
the right shows the Tonga
trench. The lines in the
middle show the back-arc
spreading centers. The
short contour lines to the
left and in the upper right-
hand corner show islands.
Line AB in Fig. 3A shows
the location of the cross
section in Fig. 2. All other
labeling is the same as in
Fig. 2.
REPORTS
www.sciencemag.org zSCIENCE zVOL. 278 z10 OCTOBER 1997 255
ilar to observations from several recent ex-
periments along MORs, which showed
smaller delay times for arrivals at stations
near the MORs than for stations on the
flanks (20,21). The faster arrivals near the
ridge axis have been attributed to the align-
ment of anisotropic minerals in the mantle,
with fast propagation for vertically traveling
Pwaves caused by focused vertical flow
beneath the spreading center. This effect
may cause the arrivals at OBS stations near
the spreading center to be faster than those
off the ridge, causing the slowest anomalies
to be displaced off the spreading center.
The actual magma chamber beneath the
spreading center would be expected to be
less than 10 km in width (22), too small to
image in this study.
Anisotropy may explain why the slow-
est velocities are not found immediately
beneath the ridge, but it cannot explain
why the western flank of the spreading
center is slower than the eastern flank.
This observation may be related to the
ongoing tectonic processes in the Lau
Basin. The CLSC, toward the west, is
lengthening southward by rift propagation
at the expense of the older ELSC, trans-
ferring the spreading activity westward
(23). Our results suggest that this transfer
may be favored by the proximity of the
western ridge to upper mantle with the
slowest velocities, which is also presum-
ably the hottest mantle and the best
source region for magma. Thus, the ridge
propagation may be an attempt by the
tectonic system to maintain the spreading
center at or near the upper mantle magma
source region.
The slow-velocity regions beneath the
Tonga arc and the Lau back arc seem to be
separated at shallow levels but merge at
deeper levels (compare Fig. 3, A and C).
This behavior suggests that although the
arc and back-arc magma systems are sepa-
rated at shallow levels, where most of the
magma is generated, there may be some
interchange between the magma systems at
depths greater than 100 km. Interchange
with slab-derived volatiles at depths greater
than 100 km may help to explain some of
the unique features in the petrology of
back-arc magmas relative to typical MOR
basalts, including excess volatiles and large
ion lithophile enrichment (24).
REFERENCES AND NOTES
___________________________
1. S. Uyeda, in Island Arcs, Deep Sea Trenches and
Back-Arc Basins, M. Talwani and W. C. Pitman III,
Eds. (American Geophysical Union, Washington,
DC, 1977), pp. 1–14; M. N. Toksoz and P. Bird, in
Island Arcs, Deep Sea Trenches and Back-Arc Ba-
sins, M. Talwani and W. C. Pitman III, Eds. (American
Geophysical Union, Washington, DC, 1977), pp.
379–393; N. H. Sleep and M. N. Toksoz, Nature 33,
548 (1971).
2. D. W. Forsyth, in Mantle Flow and Melt Generation at
Mid-Ocean Ridges, J. Morgan, D. Blackman, J. Sin-
ton, Eds. (American Geophysical Union, Washing-
ton, DC, 1993), pp. 1–65; D. L. Turcotte and J. P.
Morgan, in ibid., pp. 155–182.
3. S. M. Peacock, Science 248, 329 (1990).
4. L. N. Huppert and C. Frohlich, J. Geophys. Res. 86,
3771 (1981); G. Bock, ibid. 92, 13863 (1987); K. M.
Fischer, K. C. Creager, J. H. Jordan, ibid. 96, 14403
(1991).
5. H. Zhou, Phys. Earth Planet. Inter. 61, 199 (1990); J.
Geophys. Res. 101, 27791 (1996); R. van der Hilst,
Nature 374, 154 (1995).
6. D. Wiens et al., IRIS Newsl. 14, 1 (1995).
7. R. Jacobson, L. Dorman, G. Purdy, A. Schultz, S.
Solomon, Eos 72, 506 (1991).
8. D. Zhao, D. Christensen, H. Pulpan, J. Geophys.
Res. 100, 6487 (1995).
9. D. Zhao, A. Hasegawa, S. Horiuchi, ibid.. 97,
19909 (1992); D. Zhao, A. Hasegawa, H. Kana-
mori, ibid. 99, 22313 (1994). A 3D net of nodes was
set up in the Tonga-Fiji region with a horizontal grid
spacing of 50 km and a vertical spacing of 25 to 50
km. Hypocentral locations of earthquakes and ve-
locities at the grid nodes were taken as unknown
parameters. The velocity at any point was calculat-
ed by linearly interpolating the velocities at the eight
grid nodes surrounding that point. An efficient 3D
ray tracing technique was used to calculate ray
paths and travel times. A conjugate gradient algo-
rithm [C. Paige and M. Saunders, ACM Trans.
Math. Software 8, 43 (1982)] was used to solve the
large sparse system of observational equations by
regularizing the solution in a damped least-squares
fashion [K. Aki and W. Lee, J. Geophys. Res. 81,
4381 (1976)]. The nonlinear tomographic problem
was solved by iteratively conducting linear inver-
sions. Hypocenters were relocated in the inversion
process. For the local data from the 926 events
(Fig. 1), raw arrival times were used in the inversion.
For the teleseismic data with epicentral distances
greater than 30°, we used the relative travel time
residuals by removing the average of the travel time
residuals for each station, to remove the effect of
the structures outside the study area.
10. D. Zhao et al.,Eos (Fall Suppl.) 77, 498 (1996). The
starting 1D model for the inversion was the IASP91
Earth model (25), but the crustal thickness has
lateral variations according to a seismic refraction
survey along the line of OBS stations (26) and the
inversion of land station wave forms (15). The crust-
al thickness varies from 5 to 24 km, with the thick-
est crust beneath the Lau Ridge, Tonga arc, and Fiji
Islands. Numerous previous studies (4,5) have
demonstrated the existence of the Tonga slab,
which is at least 100 km thick with a Pwave velocity
of up to 7% faster. The slab substantially perturbs
ray trajectories, such that the ray path perturbation
can be over 100 km from that in the 1D Earth model
(4,9). Our primary concern in this study is the
precise structure below the Lau Basin and in the
mantle wedge. For this purpose, it was necessary
to use a starting model incorporating the a priori
slab information (8,9) because we hoped to resolve
the anomalies in the Lau Basin with a resolution of
50 to 70 km. We introduced the Tonga slab into the
model using the slab geometry determined by Bill-
ington (27). The introduction of the slab into the
model reduced the nonlinearity of the problem be-
cause seismic rays are traced realistically from the
beginning of the inversion (8,9). We conducted a
number of inversions by changing the initial slab
thickness from 50 to 150 km and the initial slab
velocity from 1 to 9% faster than the normal mantle.
We also conducted inversions without the slab in
the starting model. All of the inversions imaged the
very slow velocity anomalies beneath the Lau
spreading center to about 100-km depth and the
moderately slow anomalies to about 400-km
depth, but there are 1 to 2.5% changes in the
amplitudes of the anomalies and up to 10% chang-
es in the final travel time residuals. The inversion
with the starting model that included a 100-km-
thick slab and that was 6% faster than the normal
mantle resulted in the best fit of model to data. The
results of this inversion are shown in Figs. 2 and 3.
11. To make a checkerboard in the resolution tests, we
assigned positive and negative velocity anomalies
Fig. 4. Results of checker-
board resolution tests for P
wave velocity structure at
25-km (Aand C) and 480-
km depths (Band D). The
grid spacing is 50 km in (A)
and (B) and 70 km in (C) and
(D). Open and solid circles
denote low and high veloci-
ties, respectively. The per-
turbation scales are shown
at the bottom.
Fig. 5. Swave velocity models determined by
inversions of entire regional vertical and radial
wave forms recorded at island broadband seismic
stations for various tectonic regions of the south-
west Pacific.
SCIENCE zVOL. 278 z10 OCTOBER 1997 zwww.sciencemag.org256
with magnitudes of 5% to the 3D grid nodes. Syn-
thetic data were calculated for the checkerboard
model. Then we added random errors to the synthet-
ic data and inverted them with the same algorithm
that we used for the observed data. The inverted
image of the checkerboard suggests where the res-
olution is good and where it is poor. The checker-
board resolution tests and other synthetic tests we
conducted showed that both the high-velocity
Tonga slab and the low-velocity back arc and mantle
wedge were reliably resolved and that there was no
trade-off between them.
12. Y. Tatsumi, J. Geophys. Res. 94, 4697 (1989); J. H.
Davies and D. J. Stevenson, ibid. 97, 2037 (1992).
13. Y. Zhang and T. Tanimoto, Nature 355, 45 (1992); T.
Tanimoto and D. J. Stevenson, J. Geophys. Res. 99,
4549 (1994).
14. Y. Shen and D. W. Forsyth, J. Geophys. Res. 100,
2211 (1995).
15. Y. Xu and D. Wiens, ibid., in press. To invert region-
al wave forms we used a nonlinear inversion meth-
od that adopts a reflectivity formalism (28) to com-
pute the partial derivatives. This method allows the
entire regional distance (400- to 1500-km range)
wave form to be inverted from Pwave arrival to
surface waves at frequencies between 0.01 and
0.055 Hz. Broadband seismograms from earth-
quakes of 10 to 240 km deep that propagate al-
most entirely within one of the tectonic regions of
the southwest Pacific were used in the wave form
inversion. Parameter variances and resolution tests
suggest that the results are well constrained to
depths of about 200 km. The Swave velocities
from the wave form inversion and the Pwave ve-
locities from the tomography would not necessarily
show the same structure. The larger total hetero-
geneity from the wave form inversion may result
from a greater effect of partial melt beneath the Lau
back arc on Swave velocity than on Pwave veloc-
ity (9,29).
16. W. Su, R. L. Woodward, A. M. Dziewonski, Nature
360, 149 (1992).
17. G. Nolet and A. Zielhuis, J. Geophys. Res. 99, 15813
(1994); G. Nolet, in Processes of Deep Earth and
Planetary Volatiles, K. Farley, Ed. (American Institute
of Physics, New York, 1995), pp. 22–32.
18. A. B. Thompson, Nature 358, 295 (1992); H. Stau-
digel and S. D. King, Earth Planet. Sci. Lett. 109, 517
(1992).
19. A. Navrotsky and K. Bose, in Processes of Deep
Earth and Planetary Volatiles, K. Farley, Ed. (Ameri-
can Institute of Physics, New York, 1995), pp. 221–
228.
20. D. Toomey et al.,Eos (Fall Suppl.) 77, 652 (1996).
21. D. Blackman et al.,Geophys. J. Int. 127, 415 (1996).
22. J. S. Collier and M. C. Sinha, J. Geophys. Res. 97,
14031 (1992).
23. L. M. Parson and I. C. Wright, Tectonophysics 263,1
(1996); B. Taylor, K. Zellmer, F. Martinez, A.
Goodliffe, Earth Planet. Sci. Lett. 144, 35 (1996).
24. J. A. Pearce et al.,inVolcanism Associated with
Extension at Consuming Plate Margins, J. L. Smellie,
Ed. (Geological Society, London, 1992), pp. 53–75;
J. W. Hawkins, in Active Margins and Marginal Ba-
sins of the Western Pacific, B. Taylor and J. Natland,
Eds. (American Geophysical Union, Washington,
DC, 1995), pp. 125–173.
25. B. L. N. Kennett and E. R. Engdahl, Geophys. J. Int.
105, 429 (1991).
26. W. Crawford, S. Webb, J. Hildebrand, Eos (Fall
Suppl.) 77, 478 (1996).
27. S. Billington, thesis, Cornell University, Ithaca, NY,
(1980).
28. B. L. N. Kennett, Seismic Wave Propagation in Strat-
ified Media (Cambridge Univ. Press, Cambridge,
1983); G. E. Randall, Geophys. J. Int. 118, 245
(1994).
29. U. H. Faul, D. R. Toomey, H. S. Waff, Geophys. Res.
Lett. 21, 29 (1994).
30. We thank M. Bevis, W. Crawford, K. Draunidalo, S.
Escher, T. Fatai, H. Gilbert, S. Helu, K. Koper, M.
McDonald, J. McGuire, B. Park-Li, G. Prasad, E.
Roth, A. Sauter, P. Shore, and L. Vuetibau for their
assistance during the seismic experiment and at the
data-processing stage and G. Abers and an anony-
mous referee for thoughtful reviews, which improved
the manuscript. Broadband seismographs were ob-
tained from the PASSCAL program of the Incorpo-
rated Research Institutions in Seismology (IRIS).
Supported by the NSF under grants EAR-9219675,
OCE-9314446, and EAR-9614502. This paper is
Southern California Earthquake Center publication
386.
8 May 1997; accepted 18 August 1997
Microscopic Molecular Diffusion Enhanced by
Adsorbate Interactions
B. G. Briner,* M. Doering, H.-P. Rust, A. M. Bradshaw
The diffusion of carbon monoxide molecules on the (110) surface of copper was inves-
tigated in the temperature range between 42 and 53 kelvin. The activation energy for
thermal motion was determined directly by imaging individual molecular displacements
with a scanning tunneling microscope. An attractive interaction between carbon mon-
oxide molecules gave rise to the formation of dimers and longer chains. Carbon mon-
oxide chains diffused substantially faster than isolated molecules although the chains
moved by a sequence of single-molecule jumps. A higher preexponential factor in the
Arrhenius law was found to be responsible for the observed efficiency of chain hopping.
Adsorbate diffusion is of fundamental im-
portance for surface chemistry (1). It is
often the rate-limiting step in catalysis be-
cause adsorbed atoms or molecules first
have to reach a reaction partner or an ac-
tive site (2) on the surface before a reaction
can take place. Efforts to study diffusion on
a microscopic scale are needed to under-
stand how interactions with the surface and
with neighboring adsorbates influence the
way a particle diffuses. This information
forms an indispensable basis to model diffu-
sion on a macroscopic scale under the con-
ditions that prevail in catalysis. This report
focuses on the microscopic diffusion of CO
molecules on Cu(110). Carbon monoxide is
only weakly chemisorbed on Cu(110) (3),
and helium scattering experiments have
suggested very low diffusion barriers (4).
Therefore, CO can serve as a test case to
assess whether diffusion of the often weakly
bound molecules that are of interest in sur-
face chemistry is accessible to microscopic
observation.
All experiments were performed with an
Eigler-type, variable temperature scanning
tunneling microscope (STM), which oper-
ates in an ultrahigh vacuum and can be
cooled down to4K(5). We applied exper-
imental techniques similar to those used in
earlier STM-based studies on diffusion (6),
but the present results differ in two ways
from earlier findings. First, we observed that
the activation energy for CO diffusion was
substantially lower than the barrier heights
that had been determined before with mi-
croscopic imaging techniques. This result in-
dicates that the STM can indeed be used to
probe the motion of weakly bound species
and that the artifacts of STM-induced adsor-
bate motion that were reported in (7) can be
avoided. Second, our study went beyond the
observation of single-particle diffusion. It
was found that CO forms chains on
Cu(110). These chains experienced consid-
erable thermal mobility in the same temper-
ature range in which the diffusion of isolated
molecules was observed. By comparing the
diffusion of single molecules with that of CO
chains, we could investigate the influence of
molecular interactions on the adsorbate mo-
bility. Cluster diffusion was first investigated
by field ion microscopy (FIM) (8). Although
this technique is limited to the study of
strongly bound transition metal adatoms, it
could provide detailed information on the
characteristics of cluster diffusion. In gener-
al, the rule that cluster mobility decreases
strongly with increasing cluster size was con-
firmed, but FIM experiments have also dem-
onstrated that there are exceptions to this
rule. Iridium tetramers have been found to
diffuse faster than trimers (9), and for rheni-
um on tungsten(211), dimers have been
shown to be faster than single adatoms (10).
The reason for this enhanced mobility is a
reduction of the activation energy; adding an
atom to a cluster can strengthen the cluster
bonds at the expense of weakening the
bonds to the substrate (11–13). We observed
that CO chains also experienced an en-
hanced mobility, but in contrast to the metal
clusters described above, no reduced activa-
tion energy for chain diffusion was found.
Samples were prepared by adsorbing CO
onto the clean Cu(110) substrate at a tem-
perature of about 60 K. We found that under
these adsorption conditions, CO still has
substantial mobility. This mobility is inferred
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Fara-
dayweg 4-6, 14195 Berlin, Germany.
*To whom correspondence should be addressed. E-mail:
briner@fhi-berlin.mpg.de
REPORTS
www.sciencemag.org zSCIENCE zVOL. 278 z10 OCTOBER 1997 257