Tectonic activity in convergent plate boundaries commonly involves backward migration (rollback) of narrow subducting slabs and segmentation of subduction zones through slab tearing. Here we investigate this process in the Italian region by integrating seismic tomography data with spatiotemporal analysis of magmatic rocks and kinematic reconstructions. Seismic tomography results show gaps within the subducting lithosphere, which are interpreted as deep (100–500 km) subvertical tear faults. The development of such tear faults is consistent with proposed kinematic reconstructions, in which different rates of subduction rollback affected different parts of the subduction zone. We further suggest a possible link between the development of tear faults and the occurrence of regional magmatic activity with transitional geochemical signatures between arc type and OIB type, associated with slab tearing and slab breakoff. We conclude that lithospheric-scale tear faults play a fundamental role in the destruction of subduction zones. As such, they should be incorporated into reconstructions of ancient convergent margins, where tear faults are possibly represented by continental lineaments linked with magmatism and mineralization.
 Slab tearing plays a major role in the development of segmented subduction zones [Woodcock and Daly, 1986], particularly during subduction rollback of narrow slab segments [Govers and Wortel, 2005; Schellart et al., 2007]. Slab tear faults are likely to link two adjacent segments of the subduction zone and to accommodate horizontal movements if different rollback velocities operate in the two segments (Figures 1d and 1e). Govers and Wortel  have identified the existence of such faults in the North Fiji Basin (Hunter fracture zone [Schellart et al., 2002]), on the island of Sulawesi (Palu fault), and at the edges of the Calabrian, Caribbean and Scotia arcs.
 Seismic tomography data indicate that the subducting lithospheric slab beneath Italy is not continuous, and that a prominent slab window exists beneath the central Apennines [Lucente et al., 1999; Wortel and Spakman, 2000; Piromallo and Morelli, 2003]. The crustal response to the development of this slab window has been shown to be associated with lateral changes in the architecture of foredeep basins [van der Meulen et al., 1998], rapid uplift [van der Meulen et al., 2000], and possible tear-related magmatism [Marani and Trua, 2002; Faccenna et al., 2005; De Astis et al., 2006]. However, the geometry of associated tear faults, and their role during the development of the central Apennine slab window are not clear. These issues are addressed here by coupling kinematic reconstructions [Rosenbaum and Lister, 2004] with the geometry of relatively cold lithospheric slab segments as deduced from seismic tomography data [Lucente et al., 1999]. This approach enables us to reconstruct the development and propagation of slab tear faults during the process of subduction rollback, and to better understand the distribution of Plio-Quaternary magmatism in Italy in the context of the regional geodynamics.
 Slab tearing is commonly used in the literature to describe a propagating tear in the slab. Such tears can be either vertical (Figure 1b) or horizontal (Figure 1c). The process of slab tearing as described in this paper is associated with vertical tears that laterally propagating along slab tear faults (Figures 1d and 1e). Such structures are similar to those termed by Govers and Wortel  as Subduction-Transform Edge Propagator (STEP) faults.
3. Seismic Tomography
 The upper mantle velocity structure beneath Italy and the Tyrrhenian Sea is now well established, and all recent tomographic models of the area are consistent in the gross features [Lucente et al., 1999; Wortel and Spakman, 2000; Piromallo and Morelli, 2003]. The two most prominent features are (1) a narrow lithospheric slab segment (the Ionian slab) subducting steeply beneath Calabria down to a depth of 670 km and becoming horizontal parallel to the 670 km discontinuity (where it fails to penetrate the lower mantle); and (2) a remnant slab segment (the Adriatic slab) subducting subvertically beneath the northern Apennines. The absence of high-velocity perturbation between these two slab segments (in the central and southern Apennines) is interpreted to represent the earlier cessation of subduction in this area [Lucente and Speranza, 2001]. At depth of >250 km, the northern Apennine and Ionian slabs possibly join into a single fast anomaly (Figure 3).
 The tomographic images (Figures 3–5) are based on the distribution of P wave positive velocity anomalies [Lucente et al., 1999]. The 3-D morphology of subducting lithospheric slab beneath Italy (Figure 3), based on an enveloping surface defined by the 1.5% positive anomaly contour, shows a strongly segmented complex structure. The actual tears of the lithospheric slab are recognized in zones of negative velocity anomalies in cross sections that run parallel to the strike of the subduction zone (Figure 4). A more detailed mapping of the geometry of such tears is done by projecting only positive (> 1%) velocity anomalies down to a depth of 670 km (Figure 5).
 Owing to the use of high-quality data, our tomographic model suffers from relatively low regularization, which potentially allows a more accurate representation of both the amplitude of the velocity anomalies, and of their spatial distribution (i.e., thickness and lateral continuity) at depths of >50 km [Lucente et al., 1999]. In order to assess this accuracy, we conducted detailed resolution analyses by computing sensitivity tests [e.g., van der Hilst et al., 1993; Piromallo and Morelli, 2003], which address the resolution of the inversion scheme in terms of its ability to retrieve a known input model, given the ray coverage used in the real data inversion. Such sensitivity analyses provide comprehensive visual inspection of the model's spatial resolution, showing a measure of the wavelength of the structures that can be judged as reliable in various portions of the model. The performed synthetic tests are presented in Appendix A and in the auxiliary material Figures S1–S4.
 Results of the sensitivity analysis confirm that nonperipheral areas of the model (i.e., where the diagonal elements of the resolution matrix are higher than 0.6 [cf. Lucente et al., 1999] suffer from very little smearing of the anomalies. Structures are well resolved down to wavelength comparable with the block dimensions in each layer (varying from 52.5 to 95 km with increasing depth). Identified discontinuities and tear faults (Figure 5) are located in the well resolved nonperipheral areas, and are therefore regarded as reliable structures within the resolution power of the inversion method (see Figure A1 and auxiliary material Figures S1–S4). This, in turn, enables us to incorporate the present-day 3-D geometry in a kinematic reconstruction, emphasizing the role of subduction rollback and propagating tear faults.
4. Slab Tear Faults in the Italian Region
4.1. Distribution of Slab Tear Faults as Inferred From Seismic Tomography
 A major feature recognized in the tomographic images (Figures 4 and 5) is the complex geometry of lithospheric slab segments that are possibly separated by slab tear faults. The existence of a slab window down to depths of >200 km is recognized in the central Apennines (section A-A′ in Figure 4). This asthenospheric window is bounded by two tear faults, as mapped in Figure 5. Cross section A-A′ also shows a weak expression of the two southern Apennine tear faults, associated with a zone of lower seismic velocities down to depths of 250 km around latitude 40.5°. In a section parallel to Sicily (section B-B′ in Figure 4), three negative seismic anomalies down to depths of 100–200 km delineate the boundaries of a strongly segmented lithospheric structure. In the northern Tyrrhenian (section C-C′ in Figure 4) possible tear faults bound a low-velocity zone between latitudes 42° and 43°.
 The complex lithospheric structure revealed in Figures 4 and 5 is construed to represent distinct slab segments separated by relatively linear seismic velocity discontinuities. We interpret these lines as lithospheric tear faults, which accommodated strike-slip displacements during slab tearing propagation. The seismic tomography model enables us to estimate the depth of these tear faults on the basis of the depth of the displaced slab-related anomalies. The crustal expression of such faults is possibly related to the existence of orogen perpendicular strike-slip faults within the Apennine fold-and-thrust belt [e.g., Di Luccio et al., 2005; Billi et al., 2006; Scrocca, 2006]. Some of these structures show relatively deep seismicity (∼20 km [Di Luccio et al., 2005]), indicating that they are deep rooted structures (as opposed to thin-skinned accommodation faults).
 The deepest recognized tear fault (down to a depth of 550 km) runs E-W parallel to latitude 42°N (42°10′ ± 10′) (Figure 5). It is indicated by the transition from positive seismic velocities north of the fault to an area that generally lacks positive anomalies (with the exception of a single displaced slab segment around longitude 11–12°E). This tear fault continues at slightly shallower depths (< 340 km) and meets one of the major structural boundary, the so-called Ancona-Anzio line (Figure 2) that crosses the Apennines from north to south and separates the northern Apennine belt (Umbria-Marche arc) from the central Apennines. Another deep (< 440 km) lithospheric tear fault, defined by the southern extension of the displaced slab segment around longitude 11–12°E, runs parallel to latitude 41°N (Figure 5). This lithospheric-scale fault is a well known structure (41°N parallel line), which separates a domain with relatively low crustal extension in the northern Tyrrhenian, from an area that underwent much higher degree of crustal extension, leading to the formation of oceanic crust in the southern Tyrrhenian Sea [Sartori, 1990]. The crustal discontinuity parallel to latitude 41°N was also recognized in seismic profiles and was shown to be associated with a magnetic and free-air gravity anomaly [Spadini and Wezel, 1994; Bruno et al., 2000].
 Similarly to the 42°N fault further north, the tear fault associated with the 41°N parallel line changes its orientation in the Apennines to almost N-S. It is recognized at depth of <250 km, dividing the two domains of the Roman Province and the Ernici-Roccamonfina volcanoes, which have different petrologic and geochemical signatures [Peccerillo and Panza, 1999; Frezzotti et al., 2007]. The possible surface expression of this fault is the Ortona-Roccamonfina Line (Figure 2).
 Slab tear faults along the margin of the southern Tyrrhenian Sea operated at depths of 100–340 km, and are characterized by NE-SW faults in the southern Apennines, and NW-SE faults in Sicily (see also cross section B-B′ in Figure 4). The existence of such lithospheric-scale tear faults, in particular those rooting the left lateral Sangineto line and the right lateral Taormina line (Figure 2), has been suggested by previous authors [Govers and Wortel, 2005; Panza et al., 2007]. Conversely, other authors interpreted the Taormina line as the edge of the thin-skinned Calabrian nappes [Lentini et al., 1994]. On the basis of the deep structure revealed by seismic tomography, we think that the arrangement of conjugate faults represents lithospheric-scale strain accommodation during the eastward rollback of the Ionian subduction zone. The deep structures roughly mimic crustal structures, which show similar orientations of conjugate strike-slip faults and opposite senses of block rotations around vertical axes [Van Dijk and Scheepers, 1995; Speranza et al., 1999; Renda et al., 2000; Mattei et al., 2004]. The surface expression of the NW-SE right lateral faults in NE Sicily is particularly well documented [Billi et al., 2006], and possibly involves active seismicity [Goes et al., 2004].
4.2. Development and Propagation of Slab Tear Faults
 The kinematic evolution of slab tear faults is schematically illustrated in Figure 6. The reconstruction is based on the detailed spatiotemporal analysis of Rosenbaum and Lister , where an extensive amount of geological data have been used to constrain the kinematic evolution of the Tyrrhenian Sea and the Apennine belt. These data include the migration of shortening and extensional structures, the first appearance of syn-rift sediments, the distribution of palaeogeographic domains, the rotation of crustal blocks as inferred from palaeomagnetic data, and the spatiotemporal distribution of magmatic rocks [Rosenbaum and Lister, 2004]. The reconstruction also shows the possible location of lithospheric slab segments as projected from the present-day position at depths of 100–170 km (see Figure 5).
5. Magmatism in the Italian Region
5.1. Spatiotemporal Distribution
 An important outcome of the kinematic reconstruction is the apparent link between the propagating tear faults and the spatial and temporal distribution of magmatism (Figure 6 and Table 1). Magmatism in Italy is manifested by a large spectrum of magmas ranging from subduction-related calc-alkaline and ultra-potassic magmas to intraplate oceanic island basalts (OIB), and Mid Oceanic Ridge Basalts (MORB) (Figure 7) [Peccerillo, 2005]. Here we show that the geochemical affinities of these magmas are generally consistent with our geodynamic model, thus further supporting the suggestion that the Apennine subduction zone was subjected to segmentation through slab tear faulting. Our model, however, does not attempt to explain all the complex features of the petrology of the volcanic edifices [e.g., Peccerillo, 2001, 2005; Gasperini et al., 2002; Panza et al., 2007].
Table 1. Magmatic Centers in the Italian Region and Corresponding Geochronological Agesa
 The early stage of magmatic activity in the Tyrrhenian Sea, which followed abundant calc-alkaline magmatism in Sardinia, was associated with 9–6 Ma intrusive and extrusive magmatism in Capraia, Elba, Montecristo and Vercelli seamount (Figures 6a and 6b). Magmatism appears immediately above the Miocene subducting slab and is attributed to the subduction-related magmatic arc. By ∼5 Ma, this magmatic arc has migrated eastward [Civetta et al., 1978] from Capraia, Vercelli Seamount, Montecristo, and western Elba to eastern Elba and Giglio islands (Figure 6c). In the southern Tyrrhenian, direct evidence for subduction-related magmas is less clear, but there have been suggestions for the existence of a submerged Pliocene subduction-related volcanic arc [Sartori, 1986, 2005]. The location of this arc [see Sartori, 1986, Figure 3] exactly corresponds to the geometry of the subduction zone at 6–4 Ma, as shown in Figure 6c.
 Simultaneously with the production of Pliocene subduction-related magmas, further magmatism was generated by decompressional melting of asthenospheric mantle in extensional regions and gaps in the slab corresponding to tear faults. One of these tear faults propagated from the northern tip of Corsica to southern Tuscany, producing, at ∼4.8 Ma, mantle-derived shoshonites in Capraia that are characterized by lower ratios of Large Ion Lithophile Elements versus High Field Strength Elements (LILE/HFSE) compared with the older activity (see section 5.2). Further east, the crustal uplift in southern Tuscany and the production of anatectic felsic magmas (San Vincenzo, Castel di Pietra and Gavorrano) along the same fault (Figure 6c) are also interpreted to result from the combination of slab tearing and arc magmatism. Contemporaneous volcanism located along tear faults also occurred further south in Anchise Seamount and Ponza Island. The latter was subjected to magmatism from 4.2 to 1 Ma [Cadoux et al., 2005], and is considered as the northern end of the southern Tyrrhenian Pliocene arc [Sartori, 1986], which was transected by the tear fault of the 41°N parallel line.
 The distribution of some of the younger (4–2 Ma) magmatic activity was also focused along the deeper tear faults (Figure 6d). This includes a cluster of subduction/tear-related magmatic centers (Tolfa, Manziana and Cerite) in the area where the deep tear fault parallel to latitude 42°N intersected with the Italian peninsula. In the south, the occurrence of magmatism in Volturno (>2 Ma) and Ponza (4.2 to 1 Ma) also coincide with slab tearing. Subsequently, during the last 2 Ma (Figure 6e), the central Apennines were subjected to widespread tear-related magmatic activity and magmatism induced by breakoff of the lithospheric slab [De Astis et al., 2006; Panza et al., 2007]. We interpret this magmatic phase as the geodynamic expression for the formation of the central Apennine asthenospheric window following slab segmentation, and the local destruction of the subduction system. Further south, arc magmatism in the Aeolian Islands has been generated by subduction of the narrow Ionian slab, whereas the combination of rapid slab rollback and slab tearing resulted in asthenospheric upwelling at Mt Etna [Gvirtzman and Nur, 1999; Doglioni et al., 2001; Schiano et al., 2001].
5.2. Geochemical Evidence
 The link between the regional geodynamics and geochemical evidence on the depth of melting and extent of mantle metasomatism is not straightforward. Compositional variations and mantle heterogeneity can be modified by rollback-related horizontal mantle flow [Funiciello et al., 2003; Kincaid and Griffiths, 2003; Schellart, 2004] and are affected by variable types and intensities of metasomatism [Panza et al., 2007]. However, whereas some element abundances and ratios (e.g., LILE/HFSE) and isotopic signatures depend on nature and intensity of metasomatism, other compositional features, such as HFSE ratios do not depend so much on metasomatism but reflect premetasomatic mantle sources. Some major element ratios (e.g., Ca/Al of primary melts, which have not suffered clinopyroxene and/or plagioclase fractionation) depend on source mineralogy. The combination of all these features provides a further support for the geodynamic reconstruction (Figure 6).
 Most significant geochemical evidence includes elemental LILE/HFSE ratios, such as La/Nb and radiogenic isotopes (e.g., 87Sr/86Sr). In Figure 8 we use La/Nb and 87Sr/86Sr values to identify different degrees of mantle metasomatism. Low La/Nb and 87Sr/86Sr values in mafic magmas indicate mantle source, probably asthenosphere, which suffered lesser degree of subduction-related mantle metasomatism. OIB-type magmas typically have low La/Nb. Strongly fractionated and anatectic melts have relatively low (but highly variable) La/Nb and high 87Sr/86Sr values, while magmas derived from a subduction-metasomatized mantle are characterized by high La/Nb and variable 87Sr/86Sr values.
 Continental arc-related magmatism in central Italy is characterized by moderately elevated 87Sr/86Sr values (around 0.708–0.709), LREE/HREE values (La/Yb ∼ 20–30) and LILE/HFSE values (La/Nb > 2 in mafic-intermediate rocks; Figure 8a). In contrast, 143Nd/144Nd values are relatively low. We find these ratios in 7.5 Ma old intermediate calc-alkaline volcanic rocks at Capraia, mafic inclusions in granites (Elba), and predominantly granitic-anatectic rocks (Elba, Montecristo, Giglio, Vercelli Seamount) derived from the mixing of calc-alkaline melts with crustal anatectic magmas. Some of these geochemical affinities coincide with the characteristics of the younger (supposedly slab breakoff-related) magmas of the Roman Magmatic Province (see below).
 The majority of the Tuscan magmas are strongly evolved, and are largely derived from melting of the Tuscan basement [Poli, 2004]. Mantle-derived mafic magmas contain subduction components but there are rocks with higher La/Nb and Ca/Al (e.g., shoshonites versus older calc-alkaline rocks at Capraia). Higher Ca/Al and lower Sr isotope ratios mean higher proportions of asthenospheric component than metasomatized lithospheric mantle, which agrees with the presence of deep faults allowing deep mantle to contribute to magmatism. We therefore interpret such magmatism as slab tear-related magmatism (Figure 9) involving deeper asthenospheric mantle-derived melts, which either rose to the surface and produced relatively primitive magmas (shoshonites of Capraia and Campiglia dyke) or provided additional heat for the melting of the lithospheric mantle and continental crust (Tolfa-Manziana-Cerite, Roccastrada, San Vincenzo, and southern Tuscany granitoids). We emphasize, however, that the more evolved felsic magmas do not enable us to constrain the geochemical characteristics of the mantle. Their association with possible tear faulting is therefore predominantly based on their temporal and spatial distribution.
 In the central Apennines, slab breakoff following tearing coincided with the generation of magmas in the Roman Magmatic Province (Vulsini, Vico-Sabatini and Alban Hills, Figure 9). These relatively late (<1 Ma) volcanic complexes are characterized by potassic and high-potassic magmas, developed in a zone of NW-SE trending normal faults [Peccerillo, 1990, 2005]. These magmatic centers are situated northwest of the central Apennine asthenospheric window (Figure 6e), and could have possibly been triggered by the final breakoff of the lithospheric slab. Activity at Mt Ernici and Roccamonfina marks the transition between the slab breakoff magmatism and the tear-related activity of the Pontine Islands (Ponza, Ventotene, and Palmarola) and Campania province (Ischia, Procida, Phlegrean Fields, and Vesuvius).
 The two most striking examples of asthenospheric upwelling along slab tear faults are Mt Etna [Gvirtzman and Nur, 1999; Doglioni et al., 2001; Trua et al., 2003] and Mt Vulture [De Astis et al., 2006] (Figure 9). These volcanoes are situated on the margin of the foreland and are therefore much less affected by subduction processes. Mt Etna is characterized by a typical Ocean Island Basalt (OIB) composition with some arc signatures in the younger products [Schiano et al., 2001] (Figure 8b). The source of magmatism in Vulture is similar to the Campanian-Pontine volcanoes (Vesuvius, Phlegraean Fields), but its position over the edge of the Adriatic continental lithosphere resulted in a higher level of intraplate (OIB type) mantle influence [De Astis et al., 2006].
 The Aeolian Islands and associated seamounts are typical subduction-related volcanoes, with the oceanic or thinned continental Ionian slab actively subducting beneath their eastern sector. They are characterized by relatively low 87Sr/86Sr values (Figure 8b) in the western sector, and represent oceanic arc volcanism. We recognize, however, an enhanced magmatic activity (Vulcano, Lipari and Salina) and an apparent offset along the tear fault that connects the arc with Mt Etna to the southeast (Figure 5). The westernmost island of Alicudi is characterized by rather low La/Nb and 87Sr/86Sr values, similar to those of Etna and Ustica, suggesting a stronger asthenospheric influence. Also the Island of Stromboli has relatively low LILE/HFSE, resembling Campanian volcanoes for several incompatible element ratios and radiogenic isotope signatures [Peccerillo, 2001]. Its position at the margin of the arc and similar geochemical signatures as the Campanian rocks also support input of OIB-type components for Stromboli. The south Tyrrhenian seamounts and ocean floor with MORB, back-arc, and OIB affinity are also related to the asthenospheric upwelling that accompanied the opening of the Tyrrhenian Sea.
 The role of slab tearing seems to be somewhat less significant in controlling the distribution of kamafugitic-melilititic and lamproites magmas in the northern Apennines. Figure 8a shows the geochemical similarity between the kamafugitic-melilititic magmas and the slab breakoff magmas of the Roman Magmatic Province. This suggests a common metasomatized lithospheric mantle origin [Peccerillo, 2005]. However, their spatial distribution was predominantly controlled by crustal structures associated with a rift-related postcollisional environment [Peccerillo, 2005; Peccerillo and Martinotti, 2006]. Similarly, the ascent of lamproitic magmas, originated from a metasomatized lithospheric mantle [Peccerillo and Martinotti, 2006] with a strong crustal signature (Figure 8a), was controlled by extensional crustal structures.
6. Discussion and Conclusions
 In this study, we combined seismic tomography data and kinematic reconstructions to obtain a deeper insight into the complex geodynamics of Italian magmatism. On the basis of our model, tear faulting and slab breakoff played a fundamental role in controlling the temporal and spatial distribution of young (< 10 Ma) magmatism in this region. We recognize that magmatic activity induced by tear faulting was a transitional phase between subduction-related arc magmatism and postcollisional magmas related to slab breakoff. This tectono-magmatic evolution marks the destruction of subduction zones, indicating the fate of subducting slabs following continental collision.
 The role of components with low LILE/HFSE in the origin of Italian magmatism has been long recognized and has been attributed to various processes. Some authors believe this was a resident end-member in the premetasomatic uppermost mantle [Ellam et al., 1989], while others suggest an origin from a deep-mantle plume [Gasperini et al., 2002]. A mantle inflow from the foreland region has also been suggested [Trua et al., 2003; Peccerillo, 2001, 2005; De Astis et al., 2006]. However, low LILE/HFSE magmas are widespread along the Italian orogenic volcanic belt and mostly concentrated in zones of lithospheric vertical discontinuities, as shown earlier. Therefore, our hypothesis of deep asthenospheric origin of low LILE/HFSE magmas and their rise along slab-tearing discontinuities represents a much more unifying and simple mechanism in comparison with previous suggestions.
 The magmatic signature associated with tear faulting is important because it can considerably enhance our ability to reconstruct modern and ancient convergent plate boundaries. Tear faults, unlike strike-slip faults that are associated with oblique collision or indentation [Woodcock and Daly, 1986], are characterized by magmatic activity induced by upwelling of asthenospheric material. Their geometry is attained by tearing and segmentation of lithospheric slabs during subduction rollback. In most retreating plate boundaries, this process is inevitable. This is because the subducting lithosphere is not uniform but typically consists of numerous seamounts, oceanic plateaus, aseismic ridges and continental fragments [Ben-Avraham et al., 1981; Cloos, 1993]. Since these lithospheric anomalies are relatively buoyant, their arrival at the subduction zone is likely to impede subduction rollback, thus dividing the subduction zone into segments of different rollback velocities. Tomographic evidence from the northwestern Pacific suggests that slab tearing occurred at areas where seamount chains have been subducting [van der Hilst and Seno, 1993; Levin et al., 2002; Miller et al., 2004, 2005, 2006]. Many of the Pacific examples involved subduction of relatively narrow aseismic ridges, which only locally reduced rollback velocities and were therefore responsible for the development of a cuspate shape. Other examples, such as the stepping of the Palau Yap trench in the area where the Caroline Ridge arrived at the subduction zone, could possibly represent slab tear faults. It is emphasized, however, that the response to the arrival of seamounts, oceanic plateaus or island arcs at the subduction zone depends of their relative contribution to the gross lithospheric buoyancy [Cloos, 1993]. Therefore, the dynamic effect associated with the arrival of continental Adria at the subduction zone would be more pronounced than the effect associated with aseismic ridge subduction.
 It is therefore likely that lithospheric-scale orogen-perpendicular structures exist in other modern and ancient convergent plate boundaries. Evidence for the existence of such structures has been reported, for example, from the New Guinea Fold Belt, where orogen-perpendicular transfer faults were found to be associated with non-subduction-related magmatism and Cu-Au mineralization [Hill et al., 2002]. In addition, there is increasing geochemical evidence for magmatism in orogenic belts that is not directly related to subduction processes, and is attributed instead to the process of slab tearing or slab breakoff [e.g., Caprarelli and Leitch, 2001; Seghedi et al., 2004; Portnyagin et al., 2005; Hoernle et al., 2006]. Such information could possibly be correlated with major crustal lineaments, thus providing a novel approach for reconstructing the geodynamics of ancient plate boundaries.
Appendix A:: Sensitivity Analyses
 In this section we present a set of sensitivity tests [e.g., van der Hilst et al., 1993; Piromallo and Morelli, 2003; Huang and Zhao, 2006] applied to the tomographic model of Lucente et al.  (hereinafter L99). The purpose of these analyses is to address the resolution of the inversion scheme in terms of its ability to retrieve a known input model, given the same ray coverage used for the real data inversion. For a comprehensive evaluation of the resolution power it is necessary to refer to L99, where the method-data apparatus is described in detail. The sensitivity analyses allow a comprehensive, relatively rapid, visual inspection of the model's spatial resolution. However, being a limited representation of the resolution matrix (R), they can only provide a qualitative estimate of the resolution. Therefore, results of the following synthetic tests need to be read together with the analysis of the resolution matrix (R), which represents a higher-rank diagnostic factor for assessing the reliability of the estimated model in an inversion procedure. The resolution matrix (R), (both diagonal and off-diagonal elements) is analyzed and discussed by L99.
 In order to assess the resolution power of the data-method ensemble for structures of different wavelengths, we performed various synthetic tests, modifying size and distribution of the synthetic anomalies in the model. Travel times in the synthetic models are computed by forward calculation with the same theoretical apparatus used for the actual inversion by L99 (i.e., the ACH inversion technique [Aki et al., 1977]). Using this approach, we maintain some of the implicit ACH limitations, such as the use of 1-D ray tracing. The actual inversion by L99 consists of ∼6000 accurately repicked arrival times, for which the scatter of the relative residuals population is estimated to be contained within 0.3 s (see L99). Travel times through the synthetic models are reinverted adding Gaussian random noise with 0.5 s standard deviation. This more conservative value for the random noise added to the synthetic data ensures that the actual noise affecting the inverted data is not underestimated. Input parameters for the synthetic models (i.e., block dimensions, minimum number of rays for each block, damping factor, etc.) are same to those in the actual data and are reported in L99.A detailed representation of the sensitivity analyses is provided in Figure A1 and the auxiliary material Figures S1–S4.
Figure A1 shows results of a chessboard test, where the input heterogeneities of alternating sign (±4%) and wavelength equal to the block dimensions in each layer are superimposed by the ambient mantle velocity at that the layer's depth. Block dimensions increase from 52.5 to 95 km with increasing layer depth and are separated by unperturbed blocks. We observe that horizontal smearing of anomalies mainly occurs in the peripheral regions of the model, where ray sampling is poor and/or strongly anisotropic. In the internal parts of the model, where structural elements discussed in this paper are located, heterogeneities are satisfactorily retrieved down to wavelengths comparable to the block dimension in each layer, and almost in the whole depth range. Only in the lowermost layer (670–800 km depth), we observe smearing effects in the central part of the model and an amplitude decays of anomalies. In this depth range, the central part of the model is characterized by relatively lower values (<0.6) of the resolution matrix diagonal elements (see L99 and auxiliary material Figures S1–S3).
 This research was supported by a University of Queensland Early Career Researcher grant to G. R. We wish to thank R. Govers, K. Collerson, P. Vasconcelos, and T. Uysal for providing critical comments on earlier versions of this manuscript. The paper benefited from critical reviews by A. Malinverno and an anonymous reviewer.