letters to nature

damage5), and is strongly model-dependent4. Tougaards approach3 (quantitative analysis of signal-to-background correlation) requires minimal interference of neighbouring lines across a wide spectral range, and is therefore less effective with small signals. CSC is basically non-destructive, allowing fast and convenient data collection. It enables differentiation of spectrally identical atoms at different locations. It is applicable to thicker structures than ARXPS, as the signal is not subject to increased attenuation associated with off-normal measurements. Substrate roughness, severely affecting ARXPS depth analysis by smearing the angular scale, would only have a minor effect on CSC depth proling. The linear dependence on depth, found in the present systems, is an attractive feature of CSC. Note, however, that other systems may involve more elaborate conduction processes, possibly causing deviations from linearity. Such deviations may allow the exploration of additional characteristics of the systems, such as charge distribution and conduction mechanisms. As a high-resolution depth-proling method, CSC is applicable to a large variety of non-conducting layers less than about 10 nm thick. Moreover, the present results suggest applications of CSC as a contactless electrical probe, capable of direct detection of local potentials in thin overlayers. In the light of recent progress in enhancing XPS lateral resolution, CSC may be an effective tool for studying three-dimensional nanostructures and molecular architectures.
Received 14 March; accepted 26 May 2000. 1. Briggs, D. & Seah, M. P. (eds) Practical Surface Analysis Vol. 1, 2nd edn (Wiley, New York, 1990). 2. Hofmann, S. in Practical Surface Analysis Vol. 1, 2nd edn (eds Briggs,D. & Seah, M. P.) 143199 (Wiley, New York, 1990). 3. Tougaard, S. Quantitative analysis of the inelastic background in surface electron spectroscopy. Surf. Interface Anal. 11, 453472 (1988). 4. Tyler, B. J., Castner, D. G. & Ratner, B. D. RegularizationA stable and accurate method for generating depth proles from angle-dependent XPS data. Surf. Interface Anal. 14, 443450 (1989). 5. Frydman, E., Cohen, H., Maoz, R. & Sagiv, J. Monolayer damage in XPS measurements as evaluated by independent methods. Langmuir 13, 50895106 (1997). 6. Seah, M. P. in Practical Surface Analysis Vol. 1, 2nd edn (eds Briggs, D. & Seah, M. P.) 541554 (Wiley, New York, 1990). 7. Tielsch, B. J., Fulghum, J. E. & Surman, D. J. Differential charging in XPS. 2. Sample mounting and xray ux effects on heterogeneous samples. Surf. Interface Anal. 24, 459468 (1996). 8. Barr, T. L. Studies in differential charging. J. Vac. Sci. Technol. A 7, 16771683 (1989). 9. Lewis, R. T. & Kelley, M. A. Binding energy reference in XPS of insulators. J. Electron Spectrosc. Relat. Phenom. 20, 105115 (1980). 10. Miller, J. D., Harris, W. C. & Zajac, G. W. Composite interface analysis using voltage contrast XPS. Surf. Interface Anal. 20, 977983 (1993). 11. Beamson, G. et al. Characterization of PTFE on silicon wafer tribological transfer lms by XPS, imaging XPS and AFM. Surf. Interface Anal. 24, 204210 (1996). 12. Barr, T. L. in Practical Surface Analysis Vol. 1, 2nd edn (eds Briggs, D. & Seah, M. P.) 370 (Wiley, New York, 1990). 13. Shabtai, K., Rubinstein, I., Cohen, S. & Cohen, H. High-resolution lateral differentiation using a macroscopic probe: XPS of organic monolayers on composite Au-SiO2 surfaces. J. Am. Chem. Soc. 122, 49594962 (2000). 14. Hatzor, A. et al. Coordination-controlled self-assembled multilayers on gold. J. Am. Chem. Soc. 120, 1346913477 (1998). 15. Moav, T. et al. Coordination-based symmetric and asymmetric bilayers on gold surfaces. Chem. Eur. J. 4, 502507 (1998). 16. Yang, H. C. et al. Growth and characterization of metal(II) alkanebisphosphonate multilayer thinlms on gold surface. J. Am. Chem. Soc. 115, 1185511862 (1993). 17. Hatzor, A. et al. A metal-ion coordinated hybrid multilayer. Langmuir 16, 44204423 (2000). 18. Umemura, Y., Yamagishi, A. & Tanaka, K.-I. X-ray photoelectron spectroscopic study of alternately layered zirconium and hafnium phosphate thin lms on silicon substrates. Bull. Chem. Soc. Jpn 70, 23992403 (1997).

Zr multilayer Ce multilayer 1.6

1.2 V0 (V) slope = 0.148 0.8

Acknowledgements
This work was supported by the Israel Science Foundation, and the Tashtiot Infrastructure Program of the Israel Ministry of Science.

0.4 slope = 0.051 0

Correspondence and requests for materials should be addressed to I.R. (e-mail: [email protected]). or H.C. (e-mail: [email protected]).
10

Number of repeat units

Figure 3 Plot of V0 versus overlayer thickness. V0 is the total potential difference across the overlayer, and the overlayer thickness is given as the number of molecular layers. V0 is derived from the low-binding-energy side of the peaks of the respective ions. Inset, the electrical analogue of the overlayer, with a simple parallel-plate capacitor and a shunt resistor.

.................................................................
Signaturesofgranularmicrostructure in dense shear ows
Daniel M. Mueth*, Georges F. Debregeas*, Greg S. Karczmar, Peter J. Eng, Sidney R. Nagel* & Heinrich M. Jaeger*
* The James Franck Institute and Department of Physics, Radiology Department, Center for Advanced Radiation Sources (CARS), University of Chicago, 5640 South Ellis Avenue, Chicago, Illinois 60637, USA
.......................................... ......................... ......................... ......................... .........................

1.2 1 0.8 V M/V 0 0.6 0.4 0.2 0

V(Hf)/V(P) V(P)/V(Zr)

1 0 -1 -2 -3 ln[I(Hf)/I(P)] ln[I(Hf)/I(Au)] -4 -5 12 In(intensity ratio)

10

Marker position

Figure 4 Depth proling with accurately positioned markers. Solid lines show the normalized local potential of the moving marker, VM/V0, versus marker position, for P and Hf markers in the multilayers shown in Fig. 1b and c, respectively (linear scale). Dashed lines show depth analysis via line intensities for the multilayers shown in Fig. 1c (log scale). The lms with Hf(IV) in position 2 (Fig. 1c, left) displayed weak Hf signals that required second-derivative data processing for reliable determination of the energy shifts.
NATURE | VOL 406 | 27 JULY 2000 | www.nature.com

Granular materials and ordinary uids react differently to shear stresses. Rather than deforming uniformly, materials such as dry sand or cohesionless powders develop shear bands15 narrow zones of large relative particle motion, with essentially rigid adjacent regions. Because shear bands mark areas of ow, material failure and energy dissipation, they are important in many industrial, civil engineering and geophysical processes6. They are also relevant to lubricating uids conned to ultrathin molecular layers7. However, detailed three-dimensional information on motion within a shear band, including the degree of particle rotation and interparticle slip, is lacking. Similarly, very little is known about how the microstructure of individual grains affects movement in densely packed material5. Here we combine
385

2000 Macmillan Magazines Ltd

letters to nature
magnetic resonance imaging, X-ray tomography and high-speedvideo particle tracking to obtain the local steady-state particle velocity, rotation and packing density for shear ow in a threedimensional Couette geometry. We nd that key characteristics of the granular microstructure determine the shape of the velocity prole. In order to probe the role of microstructure inside the narrow granular shear zone, independent determinations are required of the velocity and density proles with spatial resolution well below the size of individual particles. Non-invasive measurements of this type so far have been limited to two-dimensional (2D) geometries where optical tracking of all particle positions is straightforward2,4,811. In a three-dimensional (3D) Couette cell, as sketched in Fig. 1a, a steady-state shear ow can be set up by conning granular material between two concentric, vertical cylinders and turning the inner cylinder at constant velocity vwall while keeping the outer wall at rest. Unlike 2D Couette cells9,10,12, where particles are conned to a single layer with constant volume, there is a free upper surface allowing the packing density to adjust via feedback between shear-induced dilation and gravity. The difculty of imaging the interior has restricted studies of 3D granular Couette systems; such studies have probed only the surface (including its velocity uctuations)13, or tracked coloured tracers in very narrow (a few particles wide) cells14, or measured global quantities, such as the total applied torque. Here we use magnetic resonance imaging (MRI) to obtain ow velocities from the interior of a 3D system15,16. We have used oil-rich seeds as a source of free protons that can be traced using MRI15. Two kinds of seeds were used to explore the role of microstructure: mustard seeds (spherical with mean diameter d = 1.8 mm) and poppy seeds (kidney-shaped

with mean diameter d = 0.8 mm). The wall friction was controlled by gluing a layer of seeds on both cylinders. Using a spin-tagging technique, horizontal slices were imaged, as sketched in Fig. 1a. In the resulting MRI image (Fig. 1b), the shear band shows up as the narrow region of deformed stripes near the inner, moving, cylinder. Imaging slices at different heights, h, we measured these deformations, from which the azimuthal velocity proles were calculated throughout the cell. Similarly, 3D X-ray tomography (Fig. 1d) allowed us to calculate packing-fraction proles at various heights. At the transparent cell bottom, additional velocity and packingfraction information was gathered by direct, high-speed-video particle tracking (Fig. 1c). Before each set of measurements, the cell was run until steady state was reached (as determined by the fact that the packing density proles became stationary). High-resolution images revealed the long-time average local displacements during a time interval, t = 100 ms. Figure 1b represents such displacements averaged over a 17-min acquisition time. The azimuthal velocity at a given distance r from the inner wall was obtained by exploiting the cylindrical symmetry of the Couette geometry: we calculated the MRI intensities along a circle of radius r, Iturning(v) and Istopped(v), for the cell turning and at rest, respectively. The position of the central peak in the cross-correlation between Iturning(v) and Istopped(v) corresponds to the average azimuthal distance travelled by the material during t. This technique yielded the angle- and time-averaged radial prole of the steady-state azimuthal mass ow velocity, v(r), resolving v(r) to within 0.1 mm s1 and r to within 0.1mm. We note that this highresolution mass ow velocity is the average velocity of all material at radius r and thus not only contains information about particle translation, but also about particle spin. This differs from the

Figure 1 Non-invasive MRI, X-ray tomography and high-speed-video probes of granular Couette ow. a, Sketch of the Couette-type shear cell consisting of two concentric cylinders with diameters 51 mm and 82 mm, and lled to a level of 60 mm. The inner cylinder was rotated at velocities 0.6 mm s1 vwall 120 mm s1. The ow velocity was measured using an MRI spin-tagging technique14,15. Before imaging, proton spins were encoded (spin-tagged) so as to display parallel stripes when imaged. Images were taken of 5-mm-thick horizontal slices at various heights, h. b, When the inner cylinder was rotated, distortion of the stripe pattern revealed the displacement of the material that occurred during the 100-ms interval between spin-tagging and imaging. 2,048 spin-tagimage steps were used to assemble each complete image. c, High-speed-video frame, taken at 1/1,000 s, of mustard seeds observed through the cells transparent bottom. To
386

indicate the movement history of individual particles, their centre positions over the preceding 200 frames are traced by black lines. The particle colouring reects the magnitude of each particles average velocity during this 0.2-s interval: fast particles (yellow) near the inner wall appear to move smoothly, while slower particles (orange and red) display more irregular and intermittent motion. In the long-time average, measured by the MRI technique, the ow appears smooth everywhere. d, Two-dimensional, horizontal slice through the cell at half the lling height, taken from the X-ray tomography data set. Particles, in this case mustard seeds, appear bright (as do the walls of the cell). These measurements were performed at the Advanced Photon Source at Argonne National Laboratory in a Couette cell of slightly smaller size but under otherwise identical shearing conditions.
NATURE | VOL 406 | 27 JULY 2000 | www.nature.com

2000 Macmillan Magazines Ltd

letters to nature
average velocity of particle centres that typically is obtained by video particle-tracking techniques. We found v(r) highly reproducible from run to run, even though on shorter timescales there are rapid velocity uctuations from point to point within the material that, at the top and bottom boundaries, can be observed with high-speed video (Fig. 1c). Within the resolution of our measurements, v(r) for both types of seeds did not vary with height, including the regions near the top and bottom surfaces. (We note that this indicates that nonazimuthal, secondary ow was small and did not affect v(r), in accordance with tracer-bead studies13.) Apart from an overall scale factor, we also did not detect any shear-rate dependence to the velocity prole over the entire range 5 mm s1 vwall 120 mm s1 explored with MRI. (With video imaging we veried rate independence down to velocities as small as 0.6 mm s1). For (nearly) monodisperse smooth, spherical particles, the decay of v(r) away from the shearing wall is dominated by abrupt drops at integer multiples of r/d as shown in Fig. 2a for mustard seeds. In the normalized velocity gradient g(r) (d/v) v/r (Fig. 2b) these drops appear as deep narrow valleys. They are correlated with pronounced oscillations in the packing density, r(r), which signal the presence of well dened, single-grain-wide layers near the moving wall (Fig. 2c). (We note that the average density approaches the random close packing value at larger r in a manner that is consistent with an exponential form10.) The shear-induced layering is reminiscent of that seen17 in the collisional regime of dilute ows, but occurs here in the high-density limit of rate-independent, frictional ow. Because the velocity drops are highly localized, we associate them with slipping at the interface between adjacent layers. Either directly from v(r) or by comparing the integrated areas of each valley in g(r) we nd that across each slip zone the velocity decreases by approximately the same factor, b = 0.36 0.13. In addition to slip between layers, MRI resolves a non-zero velocity gradient g(r) within each layer. This could be caused by either particle rotation or disorder in the layering along the radial direction. (Along the azimuthal and axial directions, particles within layers certainly show packing disorderas can be seen from the maxima in r(r) in Figs 2c and 3c, whose values lie signicantly below those for a crystalline conguration). For perfectly arranged layers, g(r) at the layer centres would be determined solely by particle rotation (spin) within each layer. The presence of particles which do not lie perfectly within the layers (revealed by the non-zero values of r(r) between layers in Fig. 2c and seen directly from the tracks in Fig. 1c) reduces the gradient in the slip region and increases it at the layer centres. In the limit of complete disorder, we would expect the staircase shape of v(r) to vanish completely. Using information about the density, rc(r), and velocity, vc(r), of the particle centres, as determined by the highspeed-video and X-ray experiments, we nd that approximately 90% of g(r) at the layer centres is due to the radial disorder. In order

40 0.4 0.2 0.0 -0.2 -0.4 0 10 0 1 2 3 4 5 6

30

v (mm s1)

20

d/v

40

r/d
v (mm s1) 30

b
0.0

20

10 -0.5 0 -1.0 0.0 -1.5 -0.5

0.8

c
0.6 -1.0

0.4 0.8 0.2 0.6 0.0 0 1 2 3 r/d 4 5 6 0.4 0.2 0.0 0

r/d
Figure 2 Radial proles of azimuthal velocity, spin and packing density for spherical mustard seeds. a, Steady-state, angle-averaged azimuthal velocity v(r) across the shear band at h = 30 mm, halfway below the lling level. The layer of seeds glued to the inner cylinder wall extends to the dotted vertical line at r/d = 1. The lower inset shows a photograph of the seeds. The upper inset shows the normalized particle spin rate, qd/v, as a function of distance from the moving wall (where q is the angular velocity of a particle about its centre). b, Normalized velocity gradient g (d/v)v/r for the data in the main panel of a. c, Angle-averaged, steady-state radial packing density prole r(r) computed from X-ray tomography data as in Fig. 1d, measuring the volume fraction occupied by seed material. For 0 r/d 1, the seeds glued to the wall contribute to r.
NATURE | VOL 406 | 27 JULY 2000 | www.nature.com

3 r/d

Figure 3 Radial proles of azimuthal velocity, spin and packing density for aspherical poppy seeds. a, Steady-state, angle-averaged azimuthal velocity v(r) across the shear band at h = 30 mm. The inset shows a photograph of the seeds. b, Normalized velocity gradient for the data in a. c, Corresponding angle-averaged, radial packing density prole r(r) from X-ray data. For 0 r/d 1, the seeds glued to the wall contribute to r.
387

2000 Macmillan Magazines Ltd

letters to nature
for the MRI and high-speed-video experiments to give the same g(r), particle spin must be considered, revealing the spin prole seen in the inset of Fig. 2a. We note that the spin is small, increasing slowly with distance from the shearing wall to a value at r/d = 4.5 of less than one full particle rotation per 13d translation. The message of Fig. 2 is that the overall shape of v(r) across the shear band can be understood as arising from two main contributions to the velocity gradient g(r): a slip contribution in the presence of layering, and a second contribution associated with radial disorder. These two pieces are distinguished by the characteristic r-dependence they produce in v(r) on length scales larger than a single grain. A constant slipping fraction, interface gslip r dr b, across each interface leads (by itself) to a velocity prole with an exponential decay, v(r) = voexp(br/d). The constant inter-layer slip appears on a background due to radial disorder, gdisorder(r) = 2c(r ro)/d, that starts at ro within the glued-on layer and increases linearly in strength with slope 2c as indicated by the dotted line in Fig. 2b). Such linear r-dependence in g(r) results in a gaussian prole, vr vo exp cr=d r o =d2 ). The sum of the two contributions to g(r), after integration, leads to a product of an exponential and gaussian term for the overall, averaged velocity prole: vr vo exp br =d cr =d r o =d2 . Fits of this function to mustard-seed data yield b = 0.36 0.13, c = 0.06 0.03 and ro/d = 0.6 0.8 independent of height and shear rate. The decomposition of g(r) allows the quantitative tracking of the slip inside the shear band for different particle types. We nd a greatly reduced slip rate b when smooth spherical particles are replaced by roughened ones. More dramatic differences result from changes in the particle shape. For kidney-shaped poppy seeds we observe a comparatively smooth overall velocity prole (Fig. 3a). Much smaller modulations in g(r) (Fig. 3b) together with little layering in r(r) (Fig. 3c) indicate that interlayer slip is greatly reduced. The linear trend g(r) = 2c(r/d ro/d) is still present, however, and leads to the gaussian shape for v(r) seen in Fig. 4 over several orders of magnitude. These data also demonstrate explicitly that the shear rate xed by vwall , while setting the overall scale, leaves the shape of the prole unchanged. This allows for a collapse of all poppy-seed data for different shear rates and heights in the cell onto a single gaussian described by c = 0.11 0.02 and ro/d = 0.1 0.5. Layering can also be suppressed and radial disorder be promoted in sphere packings if wide particle size distributions are used. We nd that these, too, produce essentially gaussian velocity proles. From these results, the gaussian component of the velocity prole emerges as a robust, generic feature; slip, if it occurs, is seen to augment v(r) by an additional, exponential factor. In contrast to the translationally invariant slip, the explicit dependence of the gaussian on the distance from the shearing wall indicates correlations. There are several possible mechanisms for radial correlations5,9,10,1820, including the formation of stress chains5,9,10 or particle clusters18,20, both of which require high packing densities and the absence of slip or some degree of interlocking. However, a detailed mechanism that would explain the observed gaussian dependence does not exist at present. A successful theory must not only account for the linear increase in the magnitude of the velocity gradient with r, but also show how the gaussian shape is independent of the presence of layering. We speculate that both of these might be provided by a coupling of g(r) to the average packing fraction and its gradients (for 2D Couette systems, a coupling to the average density alone has been suggested9,10). A gaussian, then, arises naturally for a coupling of type g(r) r(ro) r(r), considering rst-order terms in r. Because all stress-loading is driven by the seed layer glued to the inner wall, its location, ro, serves as a spatial reference point. This argument carries over to the absence of a gaussian contribution to v(r) in less-dense types of ow, where interlocking may be less effective, such as rapid ows down inclines with free upper surface for which power-law forms have been reported4,15,21, or vertical gravity-driven ows through 2D pipes, which appear to follow essentially exponential proles2,22. Our results show that, in this regime of high packing density and slow shearing rate, there is a direct connection between the microstructure and the shape of the velocity prole. For equal-sized spherical particles, where considerable layering is found, a strong exponential contribution is observed. For aspherical seeds, which show no pronounced layering, the prole is almost completely described by a gaussian centred on the shearing wall. But despite the complicated nature of the graingrain interactions, the steady-state velocity prole exhibits robust behaviour: its shapecharacterized by the two length scales, l = d/b and j = d/c1/2 is found to be independent of both height and shear-rate.
Received 13 December 1999; accepted 7 June 2000. otechnique 30, 533536 (1980). 1. Bridgwater, J. On the width of failure zones. Ge 2. Nedderman, R. M. & Laohakul, C. The thickness of the shear zone of owing granular materials. Powder Technol. 25, 91100 (1980). otechnique hlhaus, H.-B. & Vardoulakis, I. The thickness of shear bands in granular materials. Ge 3. Mu 37, 271283 (1987). 4. Drake, T. G. Structural features in granular ows. J. Geophys. Res. 95, 86818696 (1990). 5. Oda, M. & Kazama, H. Microstructure of shear bands and its relation to the mechanism of dilatancy otechnique 48, 465481 (1998). and failure of dense granular material. Ge 6. Scott, D. R. Seismicity and stress rotation in a granular model of the brittle crust. Nature 381, 592595 (1996). 7. Grannick, S. Soft matter in a tight spot. Phys. Today 52, 2631 (1999). zu n, U. & Nedderman, R. M. An investigation of the ow boundary during steady-state discharge 8. Tu from a funnel-ow bunker. Powder Technol. 31, 2743 (1982). 9. Howell, D., Behringer, R. P. & Veje, C. Stress uctuations in a 2D granular Couette experiment: a continuous transition. Phys. Rev. Lett. 82, 52415244 (1999). 10. Veje, C. T., Howell, D. W. & Behringer, R. P. Kinematics of a two-dimensional granular Couette experiment at the transition to shearing. Phys. Rev. E 59, 739745 (1999). 11. Rajchenbach, J. Granular ows. Adv. Phys. 49, 229256 (2000). 12. Karion, A. & Hunt, M. L. Wall stresses in granular Couette ows of mono-sized particles and binary mixtures. Powder Technol. 109, 145163 (2000). 13. Losert, W., Bocquet, L., Lubensky, T. C. & Gollub, J. Partical dynamics in sheared granular matter. Preprint cond-mat/0004401 at http://xxx.lanl.gov (24 April 2000). 14. Khosropour, R., Zirinsky, J., Pak, H. K. & Behringer, R. P. Convection and size segregation in a Couette ow of granular material. Phys. Rev. E 56, 44674473 (1997). 15. Fukushima, E. Nuclear magnetic resonance as a tool to study ow. Annu. Rev. Fluid Mech. 31, 95123 (1999). 16. Ehrichs, E. E. et al. Granular convection observed by magnetic resonance imaging. Science 267, 1632 1634 (1995). 17. Hopkins, M. A., Jenkins, J. T. & Louge, M. Y. On the structure of three-dimensional shear ows. Mech. Mater. 16, 179188 (1993). 18. Debregeas, G. & Josserand, C. A self-similar model for shear ows in dense granular material. Preprint cond-mat/9901336 (cited 24 Feb. 2000) at http://xxx.lanl.gov (1999). 19. Tkachenko, A. & Putkaradze, V. Mesoscopic physics of granular ows. Preprint cond-mat/9912187 (cited 10 Dec. 1999) at http://xxx.lanl.gov (1999). 20. Josserand, C. A 2D asymmetric exclusion model for granular ows. Europhys. Lett. 48, 3642 (1999). 21. Hanes, D. M. & Inman, D. L. Observations of rapidly owing granular-uid materials. J. Fluid Mech. 150, 357380 (1985). 22. Pouliquen, O. & Gutfraind, R. Stress uctuations and shear zones in quasi-static granular chute ows. Phys. Rev. E 53, 557561 (1996).

100.0 v (mm s1)

10.0

1.0

0.1 0

20

40 (r/d )2

60

80

Figure 4 Rate-independence of the velocity proles. Shown is a log plot of poppy-seed velocity proles v(r) as a function of (r/d)2 for four different shearing wall speeds (vwall = 120, 52, 33 and 14 mm s1, top to bottom). Straight lines on this plot correspond to a gaussian with origin at r = 0. The slope, c, of these lines corresponds to half the average slope of the velocity gradient in Fig. 3b and characterizes the width j = d/c1/2 of the shear band.
388

Acknowledgements
We thank E. Fukushima, J. Jenkins, C. Josserand, D. Levine, M. Medved, V. Putkaradze,
NATURE | VOL 406 | 27 JULY 2000 | www.nature.com

2000 Macmillan Magazines Ltd

letters to nature
M. Rivers and A. Tkachenko for discussions, and D. Stockwell for the donation of mustard seeds for the experiment. G.F.D. acknowledges a postdoctoral Grainger Fellowship. This work was supported by an NSF research grant and by the MRSEC program of the NSF. Correspondence and requests for materials should be addressed to H.M.J. (e-mail: [email protected]).

.................................................................
Oscillatory cluster patterns in a homogeneous chemical system with global feedback
Vladimir K. Vanag, Lingfa Yang, Milos Dolnik, Anatol M. Zhabotinsky & Irving R. Epstein
Department of Chemistry and Volen Center for Complex Systems, MS 015, Brandeis University, Waltham, Massachusetts 02454-9110, USA
.................................. ......................... ......................... ......................... ......................... ........

Oscillatory clusters are sets of domains in which nearly all elements in a given domain oscillate with the same amplitude and phase14. They play an important role in understanding coupled neuron systems58. In the simplest case, a system consists of two clusters that oscillate in antiphase and can each occupy multiple xed spatial domains. Examples of cluster behaviour in extended chemical systems are rare, but have been shown to resemble standing waves913, except that they lack a characteristic wavelength. Here we report the observation of so-called localized clustersperiodic antiphase oscillations in one part of the medium, while the remainder appears uniformin the BelousovZhabotinsky reactiondiffusion system with photochemical global feedback. We also observe standing clusters with xed spatial domains that oscillate periodically in time and occupy the entire medium, and irregular clusters with no periodicity in either space or time, with standing clusters transforming into irregular clusters and then into localized clusters as the strength of the global negative feedback is gradually increased. By incorporating the effects of global feedback into a model of the reaction, we are able to simulate successfully the experimental data. The BelousovZhabotinsky (BZ) reaction is the best studied chemical oscillator; its dynamics resemble those of many important biological and physical nonlinear oscillators1419. The reaction consists of the oxidation of malonic acid by bromate, catalysed by metal ions or metallo-complexes in acidic aqueous solution. Here we study pattern formation in the photosensitive Ru(bpy)3catalysed BZ reaction2022 in a thin layer of silica gel23,24 with photochemical global negative feedback imposed through illumination (see Fig. 1). The average concentration of Ru(bpy)3+ 3 , Zav, taken over the working area of the gel, is employed to control the intensity I of actinic light (from lamp L2), according to I = Imaxsin2(g(Zav Zt)), where g is the feedback coefcient and Zt is a target concentration. The target Zt is varied by changing the reference voltage (RV) and is set close to the steady state value, Zss, in such a way that the difference Zav Zt is always positive. The angle between polarizers P1 and P2 (that is, g(Zav Zt)) is not allowed to exceed /2. The gain of amplier A5 is used to control the feedback coefcient g. The actinic light produces bromide ion21, which inhibits the oxidation of Ru(bpy)2+ 3 . We investigate how pattern formation depends on g. We choose the initial reagent concentrations in such a way that without any feedback the system generates bulk oscillations. Various travelling wave patterns are found when g is less than g*, a value that depends on the initial concentrations of reagents. In most cases,
NATURE | VOL 406 | 27 JULY 2000 | www.nature.com

1 104 M1 g* 2 104 M1. When the feedback coefcient is slightly above g*, either travelling wave patterns or cluster patterns are observed, depending on the initial conditions. The width of this bistability range depends on the initial reagent concentrations. When g exceeds 2 104 M1, standing clusters arise. Figure 2 depicts several patterns of such clusters. Figure 2a shows one period of oscillation of a pattern that arose from initial conditions created by brief illumination of the gel through a striped mask. At t = 4 s, Zav reaches a maximum. After this instant, the white domains start to fade and the system gradually evolves to the uniform reduced (dark) state. During the second half-period, the regions that were dark during the rst half-period become bright, and at t = 29 s the pattern displays another maximum in Zav. As adjacent domains oscillate in antiphase, the period, Tav, of oscillations of Zav is half the period of the local oscillations. Figures 2b and 2c show patterns that developed from a spiral wave and from uniform initial conditions, respectively. Other patterns of standing clusters can be obtained by varying the initial conditions. Standing clusters do not possess intrinsic spatial periodicity, because the global feedback is sensitive only to the spatially integrated characteristics of the patterns. At larger g, two other types of patterns arise. If the initial reagent concentrations lie far from the boundary of the oscillatory region of the parameter space, we observe a transition from standing clusters

9.5 mm

light beam

D b L2 F L3 F SLM P4

CFUR DR P1 C P2 C BS D BS IF PMT BS P3 IF to PC AD2 AL3 RV AD4 A5

L1 IF D CFUR PM A1

Figure 1 Experimental set-up. a, The continuously fed unstirred reactor (CFUR) consists of a 24-ml continuously stirred tank reactor thermostated at 25 C and a thin layer of silica gel polymerized on the reactor optical window (grey circle). The 0.413-mmthick gel has diameter 22 mm, and contains 1.5 mM of immobilized Tris(2,2bipyridyl)ruthenium(II)(Ru(bpy)2+ 3 . Diaphragm D selects an illuminated working area of gel (light grey circle). The rectangular frame shows the eld of view of the CCD camera. A peristaltic pump injects stock solutions of KBr, malonic acid, NaBrO3 and H2SO4 into a premixing chamber (not shown) and then into the CFUR. b, Low-intensity analysing light from the stabilized 45-W light source (L1) passes through the working area of the gel, and is collected by a lens and directed to the photomultiplier tube (PMT). AD2 and AD4 are differential ampliers, AL3 is a logarithmic amplier, A5 is a d.c. amplier. The driver (DR) rotates polarizer P1, and controls the intensity of actinic light from the 450-W Xe arc lamp (L2). P1P4 are polarizers, IFs are interference lters with lmax = 450 nm, Fs are bandpass lters (400500 nm), BSs are beam splitters; lenses and collimators are not shown. The 150-W Xe arc lamp (L3) serves to set patterns of initial conditions. An image of the spatial light modulator (SLM) is focused in the plane of the silica gel. Actinic light intensity (I ) is measured by the power meter (PM). Polarizers P2-P3 and P4-P3 are crossed to separate optical channels. The remaining parasitic signal from scattered actinic light is compensated electronically by ampliers A1 and AD2, and stays below 1% of the photomultiplier working signal when I varies from 0 to Imax = 4.3 mW cm2. Uniformity of illumination of the working area is better than 2%.
389

2000 Macmillan Magazines Ltd

7 mm