Normal and transform faults originally formed at a spreading-centre can be reactivated in diffuse plate boundary zones and in areas of lithospheric flexure such as at peripheral bulges to subduction zones. Using new seismic reflection profiles and modeling, we investigate how the original oceanic fabric is reactivated in the simple case of fault perpendicular compression. In the Central Indian Basin, well-oriented normal paleofaults were reactivated with reverse motion at the very onset of deformation (9 Ma) but only a small subset remained active past ∼7 Ma, suggesting that most of the densely spaced small-offset faults were abandoned while deformation localized onto fewer faults with larger spacing. We find a similar evolution using a 2D finite element code of lithospheric shortening using a pseudoplastic rheology. Weak zones, 3 km-spaced and 30–40% weaker than the surrounding material, are introduced to simulate the fabric formed at the ridge axis. We show that reactivation and selective abandonment require strain weakening followed by strain-rate weakening once a maturation threshold is reached. A maturation fault slip of less than 50 m is needed to produce a fault network similar to that in the Central Indian Ocean.
 1. A large earthquake (Mw 7.6) ruptured a fossil portion of a transform fault at short distance from the Carlsberg Ridge (A in Figure 1) [Bohnenstiehl et al., 2004]. The sense of slip and the localization of the aftershocks were interpreted as resulting from diffuse intraplate extension [Antolik et al., 2006].
 3. In the eastern CIB and in the Wharton Basin (C), N–S fracture zones are reactivated with a sinistral strike-slip motion decoupling the Australia plate from its Indian counterpart [Deplus et al., 1998; Delescluse and Chamot-Rooke, 2007]. The same network of sinistral faults was reactivated immediately after the great Sumatra earthquake.
 4. The foretrench bulge offshore Sumatra displays active normal faults related to bending. Most of these are neoformed faults (D in Figure 1) since the original fabric is severely misoriented in the Wharton Basin with respect to the flexural stresses as observed and modelled elsewhere [Masson, 1991; Ranero et al., 2003; Billen et al., 2007]. However, a few focal mechanisms indicate slip onto the original E–W faults in the region where the fossil Wharton ridge meets the trench (D′).
 In summary, the oceanic fabric of the Indian Ocean lithosphere from the Carlsberg Ridge to the Sumatra Trench is presently reactivated in a variety of tectonic settings. Intriguingly, this fabric is never totally reactivated. Instead, deformation is localized onto a selection of faults from the original network. The history of reactivation is clearest in the Central Indian Ocean where the lithosphere was formed at spreading-centres with short segment lengths at Cretaceous to Eocene time. There, the sedimentation of the Bengal Fan acts to record the evolution of deformation through time with high fidelity, and seismic reflection imaging can be used to unravel this history.
2. Selective Abandonment of Faults in the CIB
Figure 2 presents portions of one multichannel line (Phèdre Leg 1 survey [Chamot-Rooke et al., 1993]), and two high resolution seismic profiles acquired during the Andaman 2000 survey (see Figure 1 for their location). Figure 2a clearly shows an inverted basin: the original normal fault, traced inside the crust, is presently reused as thrust. Reverse motion on this fault is kilometric, and many others large-offset faults have been imaged through seismic profiling [Bull and Scrutton, 1992; Chamot-Rooke et al., 1993; Van Orman et al., 1995]. Small-offset faults have received less attention: they require high resolution seismic surveys to be imaged, and they are inferred to contribute little to the total shortening [Van Orman et al., 1995]. Figures 2b and 2c show that these small-offset reverse faults are frequent. Their mean spacing (3 km) matches the expected initial normal faults spacing, suggesting that they do represent the original fabric acquired at the ridge axis. This spacing is significantly less than the 5–11 km reported earlier [Chamot-Rooke et al., 1993; Van Orman et al., 1995] due to the incompleteness of the fault catalogues towards the small offsets.
 Careful re-examination of the seismic lines shows that deformation started prior to the main Upper Miocene unconformity drilled at ODP Leg 116 site: the onset of significant deformation is around 7 Ma [Cochran et al., 1989], but faults were active as early as 9 Ma [Delescluse and Chamot-Rooke, 2008], and even possibly earlier but at smaller strain rate (K. S. Krishna et al., Early (pre-8 Ma) fault activity and temporal strain accumulation in the central Indian Ocean, submitted to Geology, 2008). In detail, Figure 2d shows that even after flattening of the 7 Ma unconformity, the earlier 9 Ma presents reverse offsets at small faults that nucleate in the basement. These small-offset faults become inactive soon after the onset of deformation, as the thick sedimentary sequence above the 7 Ma unconformity remained undeformed. High resolution profiler data confirmed that many of these small faults have no present-day surface expression which means that only a widely-spaced subset of these faults remain active to the present. Today, tilted blocks bounded by the largest faults have a longer 20–30 km spacing. Notice that some of these major faults localized deformation from the very onset throughout (Figure 2).
 We performed a systematic binning of the cumulative fault slip along seismic lines that cross the deformation zone in order to quantify shortening accommodated by different fault-offset ranges (Figure 1). As previously discussed by a number of authors, the total amount of shortening is increasing from west to east as a result of India-Australia kinematics [DeMets et al., 2005]. However, faults with small offsets do not follow this eastward increase: they accommodate the same amount of shortening (∼1%) in the west (where they accommodate most of the deformation) and in the east (where they accommodate only a small part of deformation). Note that small offset faults are not resolved in the Phèdre lines as only low frequency seismic data are available.
 Based on the observations, we hypothesize the existence of a threshold process whereby most reactivated faults are abandoned after a given amount of slip. Deformation then concentrates on fewer still active larger faults. We call this process selective abandonment.
3. Modeling Strain and Strain-Rate Weakening With LAYER
 We model the selective abandonment of reactivated faults using the Lagrangian finite element code LAYER [Neumann and Zuber, 1995]. This software utilizes a viscous-pseudoplastic rheology [Chen and Morgan, 1990], whereby the viscosity in each element, η, is adjusted from a value consistent with power law creep, ηpl to limit strain to a yield strength σy
where II is the second invariant of the strain rate in each element.
 At variance with Montési and Zuber [2003a], who suggested reactivation only in the vicinity of predicted localization instability wavelength, our new data lead us to consider the near complete reactivation of the spreading ridge normal faults. To capture this behavior, we modified LAYER to consider both strain weakening and strain-rate weakening (Figure 3). The yield strength increases with depth according to
where C0 is the cohesion, C the frictional term, ρ the density, g the acceleration of gravity, and K is the weakening factor. There are two sources of weakening. Initially, the yield strength decreases linearly with strain, by a factor Ws, up to a critical strain ɛs. When strain exceeds a maturation threshold ɛm (equivalent to a critical fault slip), the weakening factor becomes strain-rate dependent [Marone and Cox, 1994; Beeler et al., 1996].
 The complete expression of K is as follows
where 0 and Wr are normalization values (parameter values in auxiliary material Table S1).
 The preexisting fabric is represented by patches with non-zero initial strain dipping uniformly and spaced every 3 km (see auxiliary material Text S1 and Figure S1). These weak patches are equivalent to the reduced friction necessary to reactivate slightly oblique preexisting normal faults at trenches [Billen et al., 2007]. The entire model is 225 km long and 80 km thick, with 900 × 160 elements. Element height increases progressively with depth. The modeled lithosphere is shortened at a constant rate with time (∼6.3 × 10−9 yr−1).
 A model that best resembles the central Indian Ocean fault network is shown in Figure 4. Initially, the network of initial strain patches is reactivated uniformly so that after 1.8 Ma of shortening (top), all the initial faults display a small displacement. Subsequently, only a few of the reactivated faults accumulates significant displacement, leading to a network of major faults and pop-up structures with a spacing that is consistent with the localization instability of Montési and Zuber [2003c]. A wide variety of models were run to test the sensitivity to the model parameters. Timing of the transition from reactivation to selection is mainly controlled by the maturation threshold. A ∼2 Myr maturation time, as we document in the CIB, requires a maturation slip of less than 50 m. This short period is similar to the stress build-up time required to form the long wavelength lithospheric buckling in previous modeling [Gerbault, 2000]. A longer maturation period is possible in the model, but it would require very slow deformation rate in the early stage to keep the contribution of the small faults around 1% of the total strain, as observed. After 9 Myr of shortening (middle), only major faults are active. Nevertheless, the initial fault network has accumulated sufficient strain (about 1.1% of the total shortening, bottom) to produce structures visible in high resolution seismic data.
 Our new data point to an early onset of significant deformation in the Central Indian Ocean around 9 Ma. At that time, the entire fault network formed at the ridge axis was reactivated. Few of these faults remained active beyond the time of the major Upper Miocene unconformity (around 7 Ma) following a selective abandonment process. Our modeling offers a physical mechanism for such a process. Strain weakening controls the localization of deformation in the early stage, until strain-rate weakening becomes dominant. The remaining major active faults bound blocks that are 20 to 30 km spaced in the model, with regularly-spaced internal faulting within the blocks themselves. Seismic lines show tilted blocks of the same size bounded by major faults with hundreds of meters of cumulated slip. The maximum offset of the finely spaced fault set constrains the critical offset over which strain weakening is active to be less than ∼50 m. The resulting shortening is significant since it reaches at least 1%. The final picture is a set of sealed small faults and active large faults.
 We thank Jonathan Bull and Magali Billen for their very constructive reviews. L. G. J. M. was supported by grant OCE-0623188 and a visiting professorship from the École Normale Supérieure.