Earthquakes and coseismic slip on faults are the common response of Earth's crust to plate-tectonic forces. Here we demonstrate, using three-dimensional numerical experiments, that pulses of magmatic activity may alter the slip behavior of nearby tectonic faults by causing unusual aseismic creep and even reversals in the sense of slip. We apply our results to the Teton normal fault, Wyoming, which experienced hitherto unexplained episodes of reverse and normal creep between 1988 and 2001, to show that its anomalous behavior can be explained by inflation and deflation of two magma chambers beneath the Yellowstone caldera. Our findings imply a strong coupling between magmatism and tectonic faulting, which requires coordinated monitoring of both processes to improve our understanding of the resulting spatial and temporal strain pattern.
 Slip on active faults in the Earth's crust generally occurs during earthquakes, when elastic strain energy is suddenly released after long periods of interseismic stress build-up, during which the fault does not slip. In contrast, some faults - e.g. a segment of the San Andreas fault in central California [Schulz et al., 1982] - are known to slip continuously over decades without generating large earthquakes, a process referred to as aseismic creep. Regardless of seismic or aseismic slip accumulation, the sense of slip, i.e. the relative displacement between two crustal blocks, does usually not change as long as the plate-tectonic framework remains the same. A switch in the sense of slip occurs only when a fault is reactivated after the tectonic boundary conditions have changed [e.g., Kelly et al., 1999].
Table 1. Vertical Displacement of the Eastern Yellowstone Caldera Between 1923 and 2001 in the Yellowstone-Teton Experiment Based on a Compilation of Leveling, GPS and InSAR Data
 Against this background, the behavior of the Teton normal fault in the actively extending northeastern Basin-and-Range Province is remarkable in two ways: not only did the fault creep aseismically in the last two decades, it also changed its sense of slip from normal to reverse and vice versa, as measured by repeated leveling surveys [Sylvester et al., 1991, 2001a, 2001b]. These surveys were carried out between 1988 and 2001 and revealed two phases of hanging-wall uplift relative to the foot wall, i.e. reverse slip, alternating with two phases of normal slip at the fault center (Figure 1) [Sylvester et al., 1991, 2001a, 2001b]. The occurrence of net hanging-wall uplift and horizontal contraction across the Teton fault has recently been confirmed by GPS data that integrate over the period from 1987 to 2003 [Puskas et al., 2007].
2. Setup and Results of the Model With a Periodically Inflating and Deflating Magma Chamber
 To evaluate if a causal relationship between slip reversals on an aseismically creeping fault and the magmatic activity beneath a nearby caldera exists, we constructed - using the commercial software ABAQUS - a three-dimensional finite-element model that includes a creeping fault and a magma chamber. The model lithosphere is divided into upper crust, lower crust and lithospheric mantle (for details see Figure 2a). The layering of the model lithosphere reflects the geology of the northeastern Basin-and-Range Province, where a 40-km-thick crust and a 100-km-thick lithosphere have been constrained by gravity and teleseismic data [Saltzer and Humphreys, 1997; Peng and Humphreys, 1998]. Following Nishimura and Thatcher , the upper crust in our model is elastic, whereas the lower crust and lithospheric mantle have a viscoelastic rheology. A 60°-dipping fault, which is frictionless and simulates aseismic creep by continuous accumulation of slip, is embedded in the upper crust. Length and dip of the fault (Figure 2) are constrained by geologic and geophysical data from the Teton fault [Byrd et al., 1994]. The sides of the model that are parallel to the fault are extended at a velocity of 4 mm/a (Figure 2a), as constrained by GPS data across the northeastern Basin-and-Range Province [Puskas et al., 2007]. The magma chamber and the thermally weakened host rock around it are implemented as two ellipsoids, with the materials inside them having a viscoelastic rheology (see Figure 2a). Their viscosities are similar to those inferred for the magma chamber and the surrounding host rock beneath the Long Valley caldera, California [Newman et al., 2001, 2006]. Inflation and deflation of the magma chamber are modeled by a temporarily variable pressure exerted by the inner ellipsoid on the surrounding crust. To study the effect of the magma chamber position relative to the fault, three different experiments were performed with the ellipsoids located in the prolongation of the fault as well as opposite to its footwall and hanging wall, respectively (Figure 2a).
 In the following, we present a suite of experiments to show the response of the fault to inflation and deflation of the magma chamber (Figure 2). In the first experiment, which serves as control run, the model is extended without inflation or deflation of the magma chamber. In this experiment the fault accumulates 16 mm of normal slip in 50 years (Figure 2c). In the second set of experiments, a periodic inflation-deflation history of the magma chamber is applied. Each cycle starts with five years of inflation, during which the pressure increases linearly from 0 to 30 MPa [cf. Jellinek and DePaolo, 2003], and ends after five years of deflation with a pressure decrease to zero. The mean pressure applied, i.e. 15 MPa, is similar to the pressure inferred for magma chambers beneath other calderas [e.g., Newman et al., 2006]. Inflation and deflation lead to uplift and subsidence of the model surface above the magma chamber (Figure 2b) at a rate of 54 mm/a, which is similar to the maximum ground velocities observed at the Yellowstone [Wicks et al., 1998]. When the magma chamber switches from inflation to deflation and vice versa, the fault changes its sense of slip with inflation causing normal slip and deflation leading to reverse slip (Figure 2c). In contrast to the symmetric slip distribution on the fault in the control run, the largest amount of slip occurs at point N, which is closer to the magma chamber than point C. At the end of the experiment, the fault has accumulated a net amount of normal slip as a result of the horizontal extension. Regardless of the magma chamber position, the rates of normal and reverse slip are much higher than the slip rate in the control run (Figure 2c). The position of the magma chamber only influences the amount of slip, which is highest for a magma chamber located beneath point M (Figures 2a and 2c). Similar results have been obtained for a sill-shaped magma body (electronic supplement, Figure S1) and a friction coefficient of μ = 0.4 (electronic supplement, Figure S2).
3. Setup and Results of the Model Adjusted to the Yellowstone-Teton Region
 To investigate if the magmatic activity beneath the Yellowstone caldera can explain the slip reversals on the Teton fault, we adjusted the model setup to the Yellowstone-Teton region by using two pairs of ellipsoids and placing them at positions equivalent to the location of the two magma chambers beneath the Yellowstone caldera [Husen et al., 2004] (Figure 3, inset). The pressure exerted by the magma chambers is varied such that the resulting surface displacement corresponds to a simplified uplift-subsidence history of the Yellowstone caldera. We base our model uplift-subsidence history on the results of the leveling surveys across the eastern part of the Yellowstone caldera [Pelton and Smith, 1982; Dzurisin et al., 1990, 1994, 1999], because these are the most comprehensive data covering the time span between 1923 and 1998, and complement them by GPS [Puskas et al., 2007] and InSAR [Wicks et al., 1998, 2006] data for the period 1996–2001.
 The Yellowstone-Teton experiment shows that also for the changed spatial configuration the model fault exhibits enhanced normal slip during inflation and reverse slip during deflation. The largest amount of slip is attained at point N, with 128 mm of normal slip and 19 mm of reverse slip occurring during the first inflation-deflation cycle (Figure 3b). Normal slip during inflation is considerably faster than in the control run with extension only. To compare our modeled fault history with the leveling data from the central Teton fault, we extracted the net slip at point C between the years of the surveys. This shows that the model fault experiences reverse slip between 1988 and 1989, net reverse slip between 1993 and 1997 and net normal slip from 1997 to 2001 (Figure 3b). The modeled phases agree well with the phases of normal and reverse slip on the Teton fault revealed by the leveling surveys [Sylvester et al., 1991, 2001a, 2001b]. Furthermore, our results indicate that despite the repeated phases of reverse creep, normal slip prevails over longer time periods owing to regional extension, which agrees with the occurrence of normal faulting earthquakes on the Teton fault in the Holocene epoch [Byrd et al., 1994].
 Compared to the observed displacement on the Teton fault [Sylvester et al., 1991, 2001a, 2001b; Puskas et al., 2007] described above, the amount of slip on our model fault (Figure 3b) is somewhat lower. Using magma chambers with a more complex geometry, as suggested by seismic tomography [Husen et al., 2004], may resolve this discrepancy. A phase of normal slip on the Teton fault between 1989 and 1993 is not captured by our model. The continuous reverse creep in the model is caused by the application of a constant subsidence rate, which is based on the mean subsidence rate inferred for the time interval between 1985 and 1995 [Dzurisin et al., 1990, 1994, 1999]. As noted by Dzurisin et al. [1994, 1999] the subsidence rate may have varied considerably between 1985 and 1995 but the irremediable uncertainties of the leveling data preclude the derivation of a tightly constrained vertical displacement curve between 1985 and 1995. A temporally limited reduction or even cessation of caldera subsidence may have stopped reverse slip and triggered a short phase of normal slip because regional extension became predominant.
 The slip reversals on the model fault result from the modification of the tectonic displacement field by the inflation and deflation of the magma chambers (Figure 4). Our models reveal that during inflation, crustal material is forced radially away from the magma chamber and the surface above the magma chamber is uplifted (Figure 4a). Therefore a fault oriented perpendicular to the local direction of extension experiences accelerated normal slip owing to the enhanced horizontal stretching. During deflation the pressure exerted by the magma chamber decreases to zero (but remains positive). Hence the surface above the magma chamber subsides and the incremental displacement vectors point toward the magma chamber. This leads to horizontal shortening across the fault and reverse slip (Figure 4b).
 Our model results provide a feasible mechanism to explain the previously enigmatic episodes of reverse slip on the Teton fault and imply a strong coupling between magmatism and faulting. The slip reversals resulting from this coupling are expected to affect the interseismic stress build-up and should be considered when evaluating the seismic hazard in the neighborhood of active calderas. Coordinated monitoring is required to shed more light on the spatial and temporal evolution of the strain patterns resulting from interacting magmatic and tectonic systems.
 We thank T. Walter and an anonymous referee for their constructive reviews. Funding to A. Hampel was provided by the German Research Foundation (DFG) within the framework of an Emmy-Noether fellowship (grant HA 3473/2-1).