Earth and Planetary Science Letters 276 (2008) 152–166 Contents lists available at ScienceDirect Earth and Planetary Science Letters j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e p s l Post-emplacement melt flow induced by thermal stresses: Implications for differentiation in sills Ingrid Aarnes ⁎, Yuri Y. Podladchikov, Else-Ragnhild Neumann Physics of Geological Processes, Postbox 1048 Blindern, N-0316 Oslo, Norway a r t i c l e i n f o Article history: Received 12 November 2007 Received in revised form 5 September 2008 Accepted 14 September 2008 Available online 28 October 2008 Editor: C.P. Jaupart Keywords: thermal stresses differentiation melt flow sill D-shape a b s t r a c t We present the first steps of a new explanation model for differentiation in sills, using a combination of geochemical data and field observations, numerical modeling and dimensional analysis. Geochemical data from a saucer-shaped dolerite sill intruded into the Karoo basin, South Africa reveal a differentiation process which causes D-shaped profiles. The geometry name is based on the variation in whole-rock Mg-number (Mg# =Mg/(Mg+ Fe)) from floor to roof in a sill; the D-shaped geochemical profiles represent sheet-intrusions with the most primitive composition (i.e. high Mg#) in its center, and progressively more evolved composition (i.e. low Mg#) towards the upper and lower margins. The differentiation is reversed compared to the normal differentiation produced by fractional crystallization (C-shaped profiles). C-shaped profiles are believed to be formed by segregation of crystals from the magma. We propose that the opposite, the D-shaped profile, may result from melt segregation from the crystal mush. This is achieved by porous melt-flow through a consolidated crystal network after the main phase of emplacement, and before complete solidification. We show that a significant flow is feasible under natural occurring conditions. An underpressure of magnitude 108 Pa develops at the cooling margins due to volume reduction of the crystallizing porous melt. The resulting pressure gradient is the driving force for the melt-flow towards cooling margins considered in this work. As a result the margins will be enriched in the incompatible elements associated with the melt phase, while the center will be depleted. We show that the amount of flow is primarily a function of viscosity of the melt and permeability of the crystal network, which in turn is a transient phenomenon dependent on a number of parameters. Diagrams have been constructed to evaluate the feasibility of substantial melt extraction in terms of these poorly constrained parameters. Data from the Golden Valley Sill and many other natural occurrences of D- and I-shaped geochemical profiles show a reasonable agreement with our predictions of melt flow potential, and are thus well explained by the presented model. We conclude that in order to fully understand differentiation processes occurring in sheet intrusions, we need to account for post emplacement segregation of melt from crystals, and not only segregation of crystals from melt. © 2008 Elsevier B.V. All rights reserved. 1. Introduction Extensive sill and dyke complexes emplaced in sedimentary basins are common in large igneous provinces found in several places around the world, like the Karoo basin, South Africa, and the North Atlantic igneous province (e.g. Chevallier and Woodford, 1999; Du Toit, 1920; Planke et al., 2005). Detailed geochemical studies of such intrusive complexes reveal the complexity of the processes associated with cooling and crystallization of magmatic sheet-intrusions (Galerne et al., 2008). Profiles sampled from the floor to the roof of sills show variations in geochemistry which cannot be explained by the existing theories for the emplacement and crystallization processes of intrusive bodies. Some geochemical profile geometries are repeatedly observed in sills and referred to as D-, I and S-shaped profiles (Fig. 1) (e.g. Marsh, 1996; Latypov, 2003a and references ⁎ Corresponding author. E-mail address: ingrid.aarnes@fys.uio.no (I. Aarnes). 0012-821X/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.epsl.2008.09.016 therein). The geometry-notations are based on the variations in the whole-rock Mg-number (Mg# = 100 ⁎ Mg/(Mg + Fetotal)) from floor to roof of a sill. The whole rock Mg# is used because magnesium is preferentially partitioned into early-forming crystals such as olivine and pyroxene as compared to iron. A C-shaped profile shows the most evolved composition (e.g. lowest Mg#) in the center of the sill (Fig. 1). A D-shaped profile, in contrast, shows the most evolved composition at the sill-margins and progressively less evolved compositions (e.g. increasing Mg#) towards the center of the sill. The I-shaped profile shows no variation throughout the sill and is in general believed to be a result of closed-system crystallization (e.g. Mangan and Marsh, 1992; Marsh, 1996). The S-shaped profile is an intermediate shape, with an Ior C-shaped upper part and a D-shaped lower part within the same sill, resembling a “beer belly”. One of the main problems with the formation mechanism of D-shaped profiles is that they cannot be explained by classical fractional crystallization theory as developed for large plutons. Plutons are known to have mafic margins and felsic cores (normal zoning), forming C-shaped geochemical profiles, like the I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 153 Fig. 1. A schematic drawing of a sill at one time-step during crystallization. The end- member geochemical profiles are I, D, C and S. Skaergaard intrusion (Wager and Brown, 1968; Naslund, 1984). The formation of C-shaped profiles is interpreted as the result of magmatic differentiation by fractional crystallization (Wager and Brown, 1968). The presence of D-shaped profiles in many sills imply that we cannot directly apply theories developed for large magma chambers (3D structures of several kilometers thickness) to sheet-intrusions like sills (essentially 2D structures of commonly 10–200 m thickness). Several models have been proposed for the formation of D-shaped profiles, such as crystal settling and convection (e.g. Wager and Brown, 1968), multiple injections (e.g. Gibb and Henderson, 1992; Gibb and Henderson, 2006), flow differentiation causing phenocryst to concentrate in the center of the sill (e.g. Simkin, 1967; Richardson, 1979; Marsh, 1996), Soret fractionation in combination with in situ crystallization (Latypov, 2003a,b) or melt-flow into the margins (Cherepanov et al., 1982; this study). There is no general agreement of one particular model. We will use our data to discuss the feasibility of these mechanisms. We here present the first step in developing an alternative formation mechanism of D- shaped geochemical profiles. The model also explains Fig. 2. The Karoo basin, South Africa. The Golden Valley Sill Complex chosen for this study is located close to Queenstown in the Beaufort Group. 154 I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 the formation of I- and S-shaped profiles, and has the potential of explaining other common phenomena observed in sills. Differentiation in a magma may occur when crystals and melt are separated. C-shaped profiles are believed to be formed by segregation of crystals from the magma. We propose that the opposite, the D-shaped profile, may result from melt segregation from the crystal mush. Melt can segregate from its equilibrium crystals by post-emplacement porous flow, thus causing differentiation by advection. We explore the possibility for such flow to be induced by thermal stresses associated with cooling and crystallization of an intrusion. We have developed a numerical model to constrain the conditions under which a substantial flow may occur. We assume that a stationary crystal network is formed during cooling and crystallization, and that melt may flow through this porous network. We also assume that the rigidity of the crystal network only allows a negligible volume change from the density change associated with crystallization of the interstitial melt. The prevented volume change is accommodated by an underpressure and subsequent flux of melt into the zones of fastest crystallization along the sill margins. This paper has three main sections: 1) Presentation of geochemical and petrological data from a representative D-shaped profile, sampled from a tholeiitic sill in the Karoo igneous province, South Africa. The data are provided as a basis for the proposed differentiation model and are, along with similar profiles reported in the literature, used to constrain the validity of the numerical model. 2) Presentation of the numerical model developed to evaluate the feasibility of postemplacement flow under natural conditions. 3) Discussion of the potential formation mechanisms of D- and I-shaped profiles based on dimensional analysis, numerical modeling and geochemical data. 2. Geochemistry and petrography of a D-shaped profile 2.1. Geological setting A large volume of mafic magma was emplaced as sills and dykes in sedimentary strata during the Karoo volcanism (183 ± 1 Ma ago; (Duncan et al., 1997). The Karoo intrusive event is a part of the Karoo– Ferrar large igneous province and has affected nearly two thirds of southern Africa (e.g. Marsh et al., 1997; Chevallier and Woodford, 1999; Le Gall et al., 2002). Extensive erosion has removed much of the extrusives and revealed the underlying sill intrusions. The Karoo basin has not been affected by major tectonic activity following emplacement and the sills have thus kept their original geometry. The diameter and geometry of the sills are related to the stratigraphic level of intrusion. The sills forming the larger, subhorizontal structures are intruded at the base of the Karoo sequence, while the smaller, sills (b10 km in diameter) intruded the upper part of the sequence (i.e. the Beaufort Group; Fig. 2) and they are also commonly saucer-shaped (Chevallier and Woodford, 1999). A group of saucer-shaped sills, the Golden Valley Sill Complex, located near Queenstown, South Africa, was chosen for this study (Fig. 2). 2.2. Sampling The Golden Valley Sill Complex was sampled in detail (Galerne et al., 2008). Geochemical profiles display varying shapes, including D-, S- and I-shapes. We chose to focus on the profile showing the most prominent D-shape. This profile is sampled on the western inclined sheet of the Golden Valley Sill (Fig. 3a). The Golden Valley Sill is the main sill of the complex. It is saucer-shaped, consisting of an inner, flat sill which continues into sheets that transgress radially upwards with a 5–10° angle (Fig. 3a). The diameter across the shortest axis of the sill is ~ 10 km and the transgressive sheets are on average 100 m thick, but the thickness varies up to tens of meters along the transgressive sheets. The sampling of a vertical wall ~ 100 m thick provided some difficulties. The sampling of the profile was therefore done in a gully, which made us able to climb and sample with fairly uniform intervals (Fig. 3b). A sampling interval of 10–15 m was judged to be sufficient to display the general trend. The thickness of the inclined sheet in the gully was approximately 70 m. The erosional level of the inclined sheet is on average less than 1 m. However, at the profile location some part of the sill roof (less than 10 m) was eroded away due to the gully-effect. The upper sample of the profile is therefore located only 65 m above the lower contact of the sill and the total number of samples is six. The chemical variations of the presented profile from the Golden Valley Sill are identical to other reported D-shaped profiles (e.g. Richardson, 1979; Marsh, 1996; Gibb and Henderson, 1992 among others). We therefore regard the Golden Valley profile as representative Fig. 3. a) The Golden Valley Sill with latitudinal and longitudinal coordinates. The Golden Valley Sill is approximately 10 km across in the shortest direction. The profile was sampled on the west inclined sheet of the sill. b) The profile was sampled in an eroded gully. The upper contact between the sill and the sediments has been eroded away and could not be sampled. Sampling points are shown with stars. I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 of D-shaped profiles in general, although we were not able to sample the very top of the sill. The profile is representative of the process we want to study, i.e. the mechanism responsible for more primitive composition in the center of a sill relative to the margins. Furthermore, the Golden Valley profile makes a good basis for developing and constraining the numerical model because we have access to a complete set of information from mapping of the area, collecting the samples and doing textural studies of thin-sections, along with the geochemical data. It is important to note that although D-shaped profiles have been reported in the literature from sills in several areas, the study of the Golden Valley Sill Complex shows that a variety of different profile shapes occur in nature and may be found within a single sill. Furthermore, our data suggest that D- and I- shapes are less common (6 and 4 profiles, respectively) than S-shapes or other forms of “zigzag” profiles (N16 profiles) (data from Galerne et al., 2008). 2.3. Analytical methods The samples discussed in this paper were analyzed for whole-rock major and trace element compositions by XRF at the University of Bergen. The XRF is a sequential spectrometer, Philips PW1404, with LIF200, LIF220, PE, GE, PX1 and PX5 crystals and a sample changer with 12 positions, using the wavelength-dispersive system. Major element compositions are given in weight percent (wt.%) and trace elements are given in parts per million (ppm). The detection limit for most of the trace elements is approximately 5 ppm. The accuracy of the XRF analysis was measured using the USGS reference rock standards W2 (a basaltic rock) and BCR1 (a monzonitic rock). Electron microprobe (EMP) analyses of minerals were conducted at the University of Oslo using a Cameca SX100 with integrated energy dispersive spectrometer and 5 wavelength-dispersive crystal spectrometers. The focused beam is 2 μm in diameter. Three measurements were done within a small area at each point chosen for analysis, and the average was used. For each analyzed mineral in each sample at least two different grains were analyzed. The detection limit of the analyses is on average 0.05 wt.%. 155 samples actually have slightly less content of Na2O (Fig. 4b). K2O shows a clear C-shape (Fig. 4b). The relative abundance of K2O compared to Na2O is higher at the margins than in the center. K2O behave like an incompatible element, while Na2O is preferred over K2O in the plagioclase structure. Ni and Cr, which are strongly compatible elements with olivine and the pyroxenes, respectively, show identical D-shaped profiles (Fig. 4c). Ni is ranging from 63 and 69 ppm at the upper and lower margins, and 84 ppm in the center. Cr is ranging from 269 and 296 ppm to 384 ppm. Zr is a strongly incompatible element, and thus shows the opposite curvature, with 101 and 96 ppm at the upper and lower margins, and 87 ppm in the center (Fig. 4d). Y (Fig. 4d) is also strongly incompatible and shows the same curvature as Zr, although the curvature is sharper. Olivine shows no zoning in individual crystals, suggesting that this mineral has re-equilibrated with the melt over time. The Mg# of olivine shows a D-shape with the samples at the upper and lower margins having Mg# of 33.5 and 33.0 respectively, and the central samples 39.3–41.5 (Fig. 4e). In the olivine profile the central samples have similar Mg# while the two margin samples deviate strongly. The pyroxenes show normal zoning from core to rim of individual crystals although the average compositions vary little through the profile. The plagioclase (measured in the core of the crystals) tends to be more anorthitic in the center and more albitic at both margins (Fig. 4e). 3. Numerical modeling of post-emplacement melt flow 3.1. Presentation of the numerical model Before we can start to argue whether post-emplacement flow can cause differentiation or not, we need to test the feasibility of such flow occurring under natural conditions. In order to carry out the feasibility study, we have developed a numerical model using the Finite Element Method in Matlab. The numerical model couples three main processes: (1) Sill cooling and crystallization; (2) development of pressure anomalies due to cooling; (3) melt flow driven by pressure anomalies (Fig. 5). We have chosen a simple case of sill cooling in 2.4. Geochemistry and petrography The profile samples are medium to fine grained dolerite of tholeiitic composition. The main phases are plagioclase, clinopyroxene and orthopyroxene, with some olivine and late-stage oxides. Microscope analysis indicates that the profile has an overall homogeneous modal composition through the profile. However, small differences in modal compositions cannot be excluded. Plagioclase is found both as large crystal aggregates (~ 2 mm) and small groundmass laths (~ 0.5 mm). Olivine occurs as sub-hedral, single crystals (~ 0.5 mm) or clusters of generally smaller crystals (~0.2 mm). The pyroxenes are found mostly as large oikicrysts (~ 2 mm), enveloping and filling the spaces between olivine and plagioclase laths. The oxides are only found as oikicrysts (~0.5 mm). Olivine and plagioclase were the first minerals on the liquidus, preceding the onset of crystallization of pyroxenes; the oxides were latest. The textures and grain sizes are overall homogeneous through the profile, although there is a slight tendency for larger plagioclase groundmass laths towards the center. Compositional variations along the profile are presented in Tables 1 and 2, and plotted for some elements (Fig. 4). Both strongly incompatible (i.e. TiO2, K2O, V, Cu, Y and Zr) and strongly compatible (i.e. MgO, CaO, Cr and Ni) elements show the most primitive composition in the center and the most evolved compositions at the sill margins. Elements which are not favored by either the melt or crystal phases show no particular trends. Mg# varies from 51.9 and 52.2 at the upper and lower margins to 54.8 in the center, giving a Dshape (Fig. 4a). CaO (Fig. 4a) shows a similar trend. Na2O shows a tendency for C-shape in the four central samples, while the two margin Table 1 Whole-rock XRF data for the presented profile Meter above contact K04 AA-17 K04 AA-16 K04 AA-15 K04 AA-14 K04 AA-13 K04 AA-12 1 % STD 18 30 39 47 65 Major elements (wt.%) 51.82 SiO2 1.01 TiO2 15.87 Al2O3 Fe2O3 total 11.66 MnO 0.18 MgO 6.44 CaO 10.64 2.18 Na2O 0.68 K2O 0.15 P2O5 L.O.I 0.15 Sum 100.78 52.03 0.96 16.09 11.16 0.18 6.58 10.76 2.26 0.64 0.15 0.09 100.90 51.57 0.95 15.36 11.67 0.18 7.13 10.81 2.12 0.61 0.13 0.02 100.55 51.44 0.93 15.57 11.41 0.18 6.98 10.89 2.14 0.60 0.13 0.00 100.27 52.02 0.92 16.20 10.99 0.17 6.65 10.86 2.23 0.64 0.15 0.11 100.94 52.31 0.97 15.28 11.80 0.19 6.42 10.37 2.18 0.71 0.16 0.19 100.58 0.09 0.08 0.44 0.11 0.00 0.35 0.15 0.40 0.00 6.29 0.00 0.06 Trace elements (ppm) V Cr Co Ni Cu Zn Rb Sr Y Zr Nb 286 341 46 76 91 89 15 206 24 88 6 286 384 49 84 95 92 16 196 22 87 8 274 362 48 81 92 89 15 198 21 89 6 273 346 45 77 83 85 14 207 24 90 7 280 269 48 63 98 94 16 198 26 101 7 1.80 1.70 1.20 1.26 0.86 1.65 4.09 0.80 4.25 1.68 6.42 277 296 45 69 102 90 17 204 25 96 7 156 I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Table 2 EMP analysis giving the composition of the main mineral phases through the presented profile Sample K04 AA-17 K04 AA-16 K04 AA-15 K04 AA-14 K04 AA-13 K04 AA-12 K04 AA-17 K04 AA-16 K04 AA-15 K04 AA-14 K04 AA-13 K04 AA-12 Meter above contact 1 18 30 39 47 65 1 18 30 39 47 65 50.93 0.42 0.77 26.44 0.53 16.02 4.60 0.06 0.00 0.02 0.03 99.85 49.52 0.32 0.38 32.74 0.68 13.27 2.27 0.02 0.01 0.01 0.03 99.25 49.74 0.18 0.35 32.23 0.73 14.01 1.76 0.02 0.01 0.01 0.03 99.08 50.08 0.66 1.36 15.52 0.33 13.21 16.90 0.22 0.00 0.14 0.02 98.45 51.07 0.55 1.44 15.21 0.36 13.29 17.32 0.23 0.00 0.16 0.04 99.67 50.32 0.84 1.62 16.20 0.36 12.93 16.84 0.21 0.01 0.04 0.02 99.40 49.78 0.77 1.15 17.44 0.35 12.30 16.12 0.19 0.00 0.03 0.04 98.16 32.99 0.04 0.00 48.09 0.64 17.33 0.12 0.00 0.00 0.00 0.04 99.26 32.94 0.03 0.03 48.26 0.60 17.52 0.15 0.01 0.01 0.02 0.05 99.62 33.62 0.02 0.01 47.28 0.58 18.24 0.11 0.00 0.00 0.00 0.04 99.91 32.41 0.03 0.00 51.91 0.72 14.51 0.12 0.01 0.00 0.01 0.03 99.76 Orthopyroxene crystal-core (wt.%) SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2O K2O Cr2O3 NiO Total 51.67 0.33 0.71 24.73 0.51 18.16 3.61 0.04 0.01 0.02 0.03 99.83 49.73 0.30 0.43 31.50 0.67 14.58 2.07 0.02 0.01 0.01 0.01 99.32 34.65 0.18 0.56 15.18 0.31 12.97 2.70 0.04 0.00 0.02 0.04 66.66 Orthopyroxene crystal-rim (wt.%) 52.41 0.25 0.64 25.58 0.53 18.86 2.41 0.03 0.00 0.02 0.02 100.75 50.36 0.43 0.48 30.04 0.65 15.23 2.65 0.03 0.00 0.02 0.03 99.90 51.42 0.34 0.65 25.05 0.58 17.91 3.47 0.04 0.01 0.02 0.02 99.50 52.64 0.33 1.63 8.47 0.22 17.24 18.62 0.22 0.01 0.47 0.02 99.86 52.49 0.37 1.69 8.90 0.25 16.83 18.79 0.19 0.01 0.29 0.02 99.83 51.92 0.38 1.59 9.31 0.24 16.89 18.48 0.20 0.00 0.13 0.01 99.16 Clinopyroxene crystal-core (wt.%) SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2O K2O Cr2O3 NiO Total 52.76 0.27 1.12 10.01 0.30 15.58 20.00 0.14 0.01 0.08 0.05 100.32 47.18 0.67 3.20 13.44 0.29 12.52 16.11 5.74 0.03 0.21 0.04 99.43 51.34 0.50 1.49 10.98 0.29 15.39 18.58 0.19 0.01 0.19 0.03 98.99 49.19 0.27 0.29 33.25 0.68 13.22 1.61 0.01 0.02 0.00 0.04 98.58 50.03 0.39 0.64 29.59 0.62 14.67 3.46 0.03 0.00 0.01 0.00 99.44 Clinopyroxene crystal-rim (wt.%) Olivine crystal-core (wt.%) SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2O K2O Cr2O3 NiO Total 50.85 0.31 0.62 28.02 0.61 15.88 3.18 0.03 0.00 0.01 0.02 99.54 50.56 0.79 1.21 18.34 0.37 11.89 16.28 0.20 0.00 0.01 0.01 99.66 49.99 0.63 0.96 17.68 0.41 11.58 17.57 0.22 0.02 0.01 0.02 99.10 Olivine crystal-rim (wt.%) 32.49 0.03 0.01 51.36 0.73 14.47 0.27 0.01 0.00 0.01 0.02 99.41 32.55 0.02 0.00 47.56 0.62 18.19 0.18 0.00 0.00 0.01 0.05 99.19 32.86 0.03 0.01 48.17 0.63 17.53 0.21 0.00 0.01 0.00 0.05 99.50 33.14 0.02 0.01 48.26 0.60 17.56 0.16 0.02 0.00 0.01 0.03 99.81 33.84 0.02 0.01 47.19 0.61 18.76 0.21 0.00 0.01 0.00 0.06 100.71 32.21 0.02 0.01 52.01 0.72 14.38 0.18 0.02 0.00 0.01 0.06 99.62 52.36 0.05 29.04 0.69 0.01 0.13 12.71 4.13 0.28 0.01 0.01 99.43 52.46 0.06 28.61 0.60 0.01 0.04 12.60 4.29 0.31 0.01 0.03 99.02 49.23 0.02 30.87 0.52 0.02 0.03 15.10 2.94 0.18 0.01 0.02 98.93 50.88 0.05 30.19 0.57 0.01 0.04 13.98 3.58 0.22 0.01 0.01 99.56 51.75 0.05 29.76 0.47 0.02 0.04 13.66 3.72 0.20 0.00 0.00 99.66 52.29 0.04 29.53 0.69 0.02 0.02 13.09 4.08 0.19 0.01 0.01 99.97 32.53 0.03 0.01 52.30 0.72 13.81 0.16 0.02 0.01 0.02 0.05 99.66 32.68 0.01 0.00 46.71 0.58 18.42 0.15 0.00 0.01 0.02 0.06 98.66 Plagioclase (wt.%) SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2O K2O Cr2O3 NiO Total Detection limit (wt.%) SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2O K2O Cr2O3 NiO opx cpx olivine plag 0.02 0.02 0.02 0.06 0.06 0.03 0.03 0.02 0.02 0.05 0.08 0.02 0.02 0.02 0.07 0.06 0.03 0.03 0.03 0.02 0.05 0.08 0.03 0.02 0.01 0.06 0.06 0.03 0.03 0.03 0.02 0.05 0.08 0.02 0.02 0.02 0.06 0.05 0.02 0.03 0.03 0.02 0.06 0.08 I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 157 Fig. 4. Whole rock and olivine compositions showing the D-shaped profile. a) D-shape recognized by Mg#. CaO (5⁎wt.%) is superimposed to show that it follows approximately the same trend. b) Na O (wt.%) and K2O (3.5⁎wt.%) show a similar trend for the central samples, but deviates at the margin samples where the relative amount of K O is higher. c) The 2 2 trace elements Ni (ppm) and Cr (ppm/4.5) show clear, and similar, D-shaped trends through the profile. d) The incompatible elements Zr (ppm) and Y (4⁎ppm) show opposite trends of the D-shape, this is in agreement with less evolved melt in the center than at the margins. e) Mg# of olivine and Ca#/2 of plagioclase from the crystal cores. The olivine compositions in the center are similar (Mg# around 40), but have a strong deviation at both margins (Mg# around 33). Plagioclase is more albitic at the margins and more anorthitic in the center. Both are following an approximate D-shaped trend. order to visualize and explain the key aspects of the model. We consider a 2D sill which may be infinitely long in the lateral direction. All physical properties are non-dimensionalized and the procedure is described below. Description of physical parameters can be found in Table 3. The model and its applications are discussed further in Section 4 (Discussion). The model is used to test whether melt displacement may arise from thermal stresses associated with melt-to-crystal transition, and to investigate factors controlling the maximum segregation of melt from crystals. In the model, cooling progresses inwards from the upper and lower margins, causing a vertical melt-pressure gradient from the center to the margins. Thus the melt segregation is calculated for vertical flow in the sill. We would like to point out that in more complex systems, e.g. large sills, there can be several other possibilities for melt to flow than just towards the margins, and that in a dynamic system there may be other driving forces like e.g. buoyancy promoting lateral or upward moving melt (e.g. Galerne, 2008; Tait and Jaupart, 1992). Heat (step 1) and pressure (step 2) are solved using conduction type equations for both sill and host-rock. Porous melt flow is solved using the Darcy equation, and is only solved for within the sill at the condition of 55–90% crystals by volume (Fig. 1). The maximum melt segregation potential, or cumulative displacement, is calculated by integrating vertical melt flow velocity over total time (step 3). In Section 4 we present maximum melt displacements resulting from several runs as a function of the most important parameters. 3.2. Equations and non-dimensionalization The equations are solved on a 2D square grid with a resolution of 100 by 400 elements. We solve the equations only before total solidification. The boundaries (no-flux condition) are located far away from the intrusion to assure that they are not influencing our results. Step 1 in the model is computed using the heat diffusion equation, AT ¼ KTeff r2 T; At ð1aÞ where T is the temperature and Keff is the thermal diffusivity coefficient. The del-operator ▿ indicates that the equations can be solved for all spatial dimensions, however in this study we focus on Fig. 5. Schematic setup for the numerical model. Cooling and crystallization (1) is linked to development of pressure anomalies (2) which in turn drives a porous flow through the partly crystallized network (3). The boxes illustrate the processes at one time- step during cooling of the sill. 158 I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Table 3 Description of physical parameters Symbol Description Units X d P Tm Thr TS TL ΔT t KT kheat Cp ρ KH Χ µ β α Utotal ϕ D L Ste System size Sill thickness Pressure Initial temperature of melt Initial temperature of host rock Solidus temperature of melt Liquidus temperature of melt Initial difference; Tm − Thr Time Thermal diffusivity (kheat/Cp/ρ) Thermal conductivity Specific heat capacity Density Hydraulic diffusivity (χ/μ/β) Permeability of crystal network Viscosity of melt Isothermal compressibility Thermal expansion coefficient Total melt displacement Porosity (T − TS) / (TL − TS) Grain size Latent heat of crystallization Stefan number L/Cp/ρ / (TL − TS) m m Pa K K K K K s m2 s− 1 J m− 1 K− 1 s− 1 J kg− 1 K− 1 kg m− 3 m2 s− 1 m2 Pa s Pa− 1 K− 1 m % m J/m3 – Note that porosity is a linear function of temperature, where T = TL represents 100% melt, and correspondingly T = TS represents 100% crystals. the vertical y-direction. The effective thermal diffusivity accounts for the latent heat of fusion: KTeff ¼ KT for ðTs bTbTL Þ 1 þ Ste ð1bÞ KTeff ¼ KT for ðTs NT Þ Non-dimensional ratio quantifying the effect of the latent heat is the Stefan number, Ste, given by Ste ¼ L ðTL −Ts ÞρCP where KH is the hydraulic diffusivity, P the pressure, α the volumetric coefficient of thermal expansion and β the isothermal compressibility. The first part on the right hand side of Eq. (2b) describes the pressure diffusion (similar to heat conduction Eq. (1a)); the second part describes the development of pressure anomalies due to changes in temperature. The initial pressure is zero because the flow only depends on the developed pressure anomalies. In step 3 we use the developed pressure gradients to calculate melt flow, following Darcy's law for flow in porous media, Y v χ ¼− rP ¼ −KH rP β μβ ð3Þ → where v is the average velocity of the melt and β the isothermal compressibility. Note that the standard equation is divided by β on both sides to reduce the uncertainty of having permeability over viscosity (χ/ μ) as a separate parameter in the model. The Darcy velocity is integrated over time into total length of melt-displacement, U, U ¼ β Z t 0 jvy j dt β ð4Þ As a consequence of Eq. (3), the left hand side of Eq. (4) is also divided by β. The chosen values of physical parameters for Eqs. (1)–(4) (Table 3), along with initial conditions, will greatly influence the magnitude of the calculated flow. We introduce dimensional analysis to reduce the number of variables needed to solve the system of equations, and thus reduce the uncertainties of the model. The main goal of the analysis is to identify the key parameters that influence the calculated total displacement, Utotal. To non-dimensionalize the equations we choose a set of independent new units (Table 4), ΔT = Tm − Thr , KH (hydraulic diffusion), d (sill thickness) and α/β (expansion coefficients). These independent parameters are used to non-dimensionalize all other (dependent) parameters used in the following equations (Table 4). By introducing ð1cÞ where ρ is melt density, CP heat capacity and L the latent heat of fusion per volume (note that KT also include ρ and CP). The initial conditions are: 1) The sill has an initial higher temperature, Tm, than the host rock, Thr. 2) The upper and lower margins of the sill (0.2% of the sill thickness) are initially set to a temperature intermediate between the initial temperature in the sill and that of the host-rock in order to account for chilled margins formed during emplacement (Galushkin, 1997). 3) The thermal diffusivities are unequal for the sill and the host-rock, due to their different thermal properties. In Fig. 6 we show how the numerical model can be used to solve the heat conduction equation on the 2D grid for a dimensional case. The initial values used in this example are of little importance, as it is just one out of many possible scenarios. In step 2, Eq. (1) is coupled with pressure through thermal stresses, dP ¼ α dT β ð2aÞ as described by e.g. Turcotte and Schubert (2002), p. 172, assuming isochoric conditions for crystallization. Taking the partial derivative of Eq. (2a) with respect to time, our hydraulic equation becomes AP α AT ¼ KH r2 P þ At β At ð2bÞ Fig. 6. Typical evolution of temperature in a 2D simulation in a dimensional version. The boundary conditions are the same as in Fig. 8. I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Table 4 Parameters used for non-dimensionalization, and the resulting non-dimensional parameters Scale Independent dimensional parameters Temperature Diffusion Expansion Length Latent heat⁎ ΔT = Tm − Thr, initial temperature difference KH, hydraulic diffusion coefficient α/β, expansion coefficients D, sill thickness Ste, ⁎already non-dimensional Scale Dependent non-dimensional parameters Temperature System lengths Diffusion Time Pressure Velocity Displacement T′ = T/ΔT ∇′ = ∇/d (i.e. x′ = x/d, y′ = y/d) K′ = KT/KH t′ = tKH/d2 P′ = P/(α/βΔT) v′ = v · d/KH U′ = U /d For symbol descriptions see Table 3. the non-dimensional parameters T′ = T / ΔT, ∇′ = ∇/ d, K′ = KT / KH and t′ = tKH/d2, Eq. (1a) becomes AT 0 2 ¼ K 0r0 T 0 At 0 ð5Þ Similarly, by introducing P′ = P/(α/βΔT), Eq. (2) becomes AP 0 2 AT 0 ¼ r0 P 0 þ 0 At 0 At 159 3.3. Assumptions The numerical model is focused on the main processes associated with post-emplacement melt flow, and does not attempt to describe the full natural system. The model cannot predict the geochemical Dshape directly because it is not coupled to chemical evolution of the system. The equations are based on well defined physical principles, and the only major assumption in our model is isochoric condition. In other words, we assume that thermal contraction or mechanical compaction of the crystal network is negligible for the area where we calculate melt flow. The flow is calculated where the crystals form a stationary, solid network with interstitial melt, i.e. 55–90% crystals by volume. It is now recognized that magma chambers can be crystallized to a rigid, but not completely solid, crystal mush. A rheological “locking-point”, i.e. a stationary crystal network, forms when the volume fraction of crystals exceeds 55% (Marsh, 1996), although for a plagioclase-rich system a considerable strength of the network can be reached with as low as 30% crystals (Philpotts and Carroll, 1996). The stationary crystal network is an essential criterion for the geochemical differentiation; if the crystals move along with the melt, no differentiation by flow occurs. The Peclet number is set to 0, i.e. no feedback to the main temperature profile due to advection of fluids in the contact aureole or internal advection of melt. It would be interesting to investigate further implications of substantial melt-advection on the cooling and ð6Þ and by introducing v′ = v · d/KH Eq. (3) becomes v0 ¼ −r0P 0 ð7Þ Finally, the maximum total displacement is calculated by introdu→ cing U′ = U /d into Eq. (4) U0 ¼ Z t0 0 jv0y jdt 0 ð8Þ The dimensional analysis gives that the only important parameters determining the system is K′ = KT/KH and an unknown function of Ste. Thus, the total displacement becomes U KH f ðSteÞ ¼ d KT ð9aÞ Varying the ratio KT/KH we find that displacement is a linear function of KH/KT (i.e. flipped) in the pressure domain (KH dominates) (Fig. 7a). Flow is expected when pressure build-up is dominating the system, and cooling is slow, i.e. the pressure domain. The Ste has an influence on displacement as long as we are in the regime dominated by temperature (i.e. KT). Varying Ste corresponds to varying proportion of initial crystal content and thus varying release of latent heat of crystallization. At the crossover, where pressure becomes the governing factor (and flow is expected), the data collapses and the coefficient replacing f (Ste) is approximately 0.06 (Fig. 7b). This gives us the formula, U KT ¼ f ðSteÞ≈0:06 d d KH ð9bÞ Or U≈0:06 KH d KT which will be elaborated on in the discussion. ð9cÞ Fig. 7. a) Varying the ratio KT/KH gives a linear relationship of KH/KT (note that the ratio is flipped) with displacement, resulting in Eq. (9a). The range of possible combinations of parameters is wide, and does not necessarily correspond to conditions occurring in nature. b) Plotting of Eq. (9a) for different values of Ste gives a data-collapse for values where pressure (i.e. KH) is more dominant than temperature (i.e. KT). We get the relationship f(Ste) ~ 0.06 (Eq. (9b)). Ste = 0 corresponds to results with no latent heat. 160 I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Table 5 Values used in numerical calculation (new basic parameters are all 1) Nondimensional numbers Dimensional values assumed Values (KT/KH)melt (KT/KH)host-rock (TS − Thr)/ΔT (TL − Thr)/ΔT Length in y-direction Length in x-direction Ste Total time 1 × 10− 6/2.0 × 10− 6 1 × 10− 6/2.2 × 10− 6 (900 − 100) / 1100 (1200 − 100) / 1100 2.5 × sill thickness 5 × sill thickness 300 / (1200 − 900) 1/2.0 × 10− 6 0.5 0.45 0.72 1 2.5 × d 5×d 1 1/KH crystallization history of sills. However, this study is concentrated on the implications of melt flow on chemical differentiation. Gravity is neglected because it is not a triggering mechanism for the flow. Gravity may, nevertheless, affect the flow pattern in natural systems. Eq. (1) assumes simplified linear dependency of melt fraction with temperature in order to calculate the effect of the latent heat of crystallization. The effect of latent heat is important because it slows down the cooling and crystallization. Otherwise the cooling would be at least twice as fast, which would imply less time for melt flow to occur. Nevertheless we show that for a realistic range of Ste, (where Ste = 0 corresponds to no latent heat) the effect of latent heat (Ste = 1) on calculated displacement is negligible (Fig. 7a). The effect of latent heat varies in the temperature domain where cooling dominates over flow (Fig. 7b). Ste has a constant value of ~0.06 in the pressure domain where dominant flow is expected. For simplicity we excluded temperature dependency on the material properties, i.e. the diffusivity ratio for the host-rock is constant with exception of the interval of released latent heat. The validity of these assumptions is discussed in Section 4.1. 4. Discussion 4.1. Validity of model assumptions The assumption of isochoric condition is obviously not true for all stages in the cooling process of a sill. Columnar jointing is, for example, a strong indication of thermal contraction. The reasoning behind the assumption of isochoric condition is that the density change due to temperature changes in a stationary, solid crystal network is negligible compared to the density change caused by meltto-crystal transition in the interstitial melt due to the cooling. As crystallization continues in the interstitial melt, the latter “wants” to contract much more that the rigid crystal network will allow, because no voids can develop. Since the volume is constant due to the rigidity of the network, an underpressure develops. To allow for the crystal network to contract thermally will probably give a minor correction to the developed underpressure, but would not change the main results of the calculations. The implication of the isochoric condition for the host-rock is that pore-fluids want to expand much more than the surrounding rock due to the increased temperature. The result will be a hydrostatic overpressure. Evidence for overpressure in the contact aureole of sills can be observed, for example in vent-structures associated with sill intrusions (Svensen et al., 2006; Svensen et al., 2007). Linearity of Ste is a crude assumption, especially for eutectic crystallization. However, the effect on cooling time by using a linear Ste is negligible for melts of mafic composition in comparison with more elaborated models for the Stefan number (Podladchikov and Wickham, 1994; Turcotte and Schubert, 2002). Latent heat of crystallization is known to be one of the main processes related to sill cooling and crystallization. However, we show that processes like release of latent heat, with only minor contributions compared to conduction, will have negligible influence on our final predictions. This strengthens our assumption that we have included the most important processes which will affect porous melt-flow by thermal stresses. Constant diffusivity independent on temperature is obviously not true in nature. We could, for example, have made the permeability evolve with porosity (melt fraction). Melts tend to maintain an interconnected network even though porosity decreases (e.g. Shirley, 1986). Also, hot inflow of melt may dissolve crystals on its way in order to maintain local equilibrium thermodynamics, thus sustaining a local high permeability (Tait and Jaupart, 1992). Thus, the relationship between permeability and porosity may not be trivial, and by including it we would have had to include additional assumptions. Fig. 8. Numerical results of a vertical cross-section of the sill intrusion. a) Temperature profile: the sill is cooling, while host-rock is heating according to the heat diffusion equation, accounting for latent heat of fusion. b) Melt percentage profile: the phase transition from melt to crystals is progressing inwards as the sill cools. c) Pressure anomaly profile: an underpressure develops where the sill cools, while the heating of the host-rock results in a positive pressure, according to the constitutive equation of thermal stresses. d) Melt displacement profile: because the underpressure is larger at the margins (faster cooling) than in the center (slower cooling), melt flows towards both margins. I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Likewise, the effective thermal expansion coefficient should vary, taking into account the density change of crystallization. Instead of having parameters changing with temperature we solve the problem by varying our restricted set of key constants systematically by several orders of magnitude to capture all possible combinations of e.g. permeabilities and expansion coefficients. 4.2. Application of the numerical model The main aim of the numerical study is to find under which conditions we can expect substantial flow and insignificant flow, respectively. D-shaped profiles can only result from a porous meltflow if our calculations show that substantial flow in sills is possible. The modeling of one case study uses the values listed in Table 5. It is important to remember that the results for one chosen set of parameters is not sufficient to decide whether or not this happened in the Golden Valley Sill, because the details of the natural occurring parameters and their time evolution is not well enough constrained by our data. The first stage of the modeling is characterized by cooling of the sill and heating of the host rock as shown in Fig. 8a. With progressive cooling the crystallization front moves from the contacts towards the sill center (Fig. 8b). The second stage is characterized by the development of pressure anomalies, P, due to changes in temperature (Fig. 8c). Where the sill is cooling an underpressure develops. Where the host rock is heated (in the contact aureole), an overpressure develops. The non-dimensional pressure P′ = 0.1 corresponds to a physical pressure of P ~ 108 Pa, for the chosen values of α = 10− 5 K− 1, β = 10− 11 Pa− 1 and ΔT = 103 K (see Section 3.2). The expansion coefficients used are typical values for rocks (e.g. Turcotte and Schubert (Turcotte and Schubert, 2002, p. 172). The pressure anomaly is of the same magnitude as other studies associated with magmatic processes (Bachmann and Bergantz, 2006; Voight et al., 2006). The uniaxial compressive strength of fresh basalt exceeds 250 MPa which is sufficient to support our estimates of pressure anomalies (~100 MPa) (Hoek and Brown, 1997). This suggests that for the system to re-equilibrate from the large developed thermal-stress anomalies, it is favorable to induce melt-flow as opposed to deform the rigid network. However, we cannot exclude some degree of compaction in the partly crystallized regions. There is a strong gradient going from the heated host-rock into the low-pressurized sill margins, but we only consider flow within the sill. The cooling in the central parts of the sill is minor compared to that at the margins, thus there is little or no anomaly developing in the center. The melt is “sucked” into the low-pressure margins from the less affected central parts as the cooling front moves inwards. The resulting displacement, U/d, is shown in Fig. 8d. The maximum total displacement corresponds to 20 m in a 100 m thick sill. Twenty meters of displacement is a measure of the potential change in concentration. Advection can only change the composition if there is a gradient in the melt composition, which is likely to be present due to the thermal gradient and different degrees of crystallization. We evaluate the instantaneous rate of melt extraction (Eq. (7)) and integrate over time in order to get maximum possible melt displacement from a given point (Eq. (8)). We therefore do not trace the exact trajectories of the melt patches, but evaluate extraction potential of melt at the given point. The sensitivity of extraction potential to the variations in the input variables (e.g. the host-magma heat diffusivity ratio) that vary within one order in magnitude is not considered here. Commonly, in studies dealing with magma crystallization, melt pressure is assumed lithostatic and thermal stresses are not considered. There are exceptions, however, where non-lithostatic and coupled models for pressure evolution are used in e.g. numerous studies of fluid circulation in contact aureoles (Delaney, 1982; Barton et al., 1991; Podladchikov and Wickham, 1994; Ingebritsen and Sandford, 1998) among others. Some studies document significant 161 transient pressure anomalies, which are comparable in magnitude to the lithostatic pressure. In the specific context of dike or sill intrusions these were discussed by e.g. Litvinovski et al. (1990); Tommasini et al. (1997) and Jamtveit et al. (2004). Litvinovski et al. (1990) documented quartz grain crushing and melting of the host-rock, combined with contact melt injecting into intruded magma, as evidence for increased pressure in the contact aureole. Tommasini and Davies (1997) also argue for pressure rise in order to explain contact melting. Jamtveit et al. (2004) focus on explaining vent overpressure associated with sill intrusions. All these previous studies are focused on transient pressure anomalies developing in the contact aureoles. In this study we examine pressure anomalies developing within the crystallizing magma. 4.3. Model feasibility using dimensional analysis To test the feasibility of the model, we have to relate the relationship obtained in the dimensional analysis to physical parameters within a feasible range of values. Rewriting Eq. (9c), the relationship between displacement and the diffusivity ratio becomes U χ ≈0:06 d μβKT ð10Þ Two of the least constrained parameters in Eq. (10) are the permeability of the crystal network, χ, and the viscosity of the melt, μ (parts of KH, Table 3). These parameters indeed may vary by several orders of magnitude within one magmatic system, depending on factors like temperature and crystal size. We choose the values for thermal diffusivity (K) of 10− 6 ms2 s− 1 (Delaney, 1988), isothermal compressibility (β) of 10− 11 Pa− 1, and put them into Eq. (10) to get U χ ¼ 6 1015 d μ ð11Þ which shows how total melt displacement responds to different values of viscosity (μ) and permeability (χ) in the calculations. This relationship is plotted in Fig. 9, where the contours represent the magnitude of the displacement. We set a rough transition zone between the D-shape and I-shape regimes at above 10% of the sill thickness (U/d N 0.1). The results (Fig. 9) suggests that in a melt of basaltic viscosity (i.e. ~102 Pa s), we have significant flow when the permeability of the crystal network is above 10− 14 m2, corresponding to ~99% crystallinity of a basaltic rock (Hersum et al., 2005) showing flow is feasible for natural occurring conditions. S-shaped profiles, or other intermediate shapes, probably form in the transition zone between the D-shape and the I-shape regime, requiring less melt flow than D-profiles. There is also a possibility that the Sshaped, or beer belly, profiles are a combination of porous melt segregation with some other processes, like compaction. In general, other formation mechanisms for the S-shaped profile cannot be excluded by the results in this study. However, only the classical and well documented process for magma evolution in large plutons, that is, gravitational settling, is needed in addition to the process advocated in this study to explain the formation of more complex shapes. Indeed downward settling combined with D-shape-porous flow would result in a beer belly-profile. 4.4. Formation mechanisms of D-shaped geochemical profiles The D-shaped and I-shaped geochemical profiles are end members used to categorize observed profiles. The D-shaped profile represents 162 I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Fig. 9. The response in melt displacement for different viscosities and permeabilities. A melt of basaltic viscosity requires a permeability of the crystallizing network having a magnitude above 10− 14 m2 to be able to create a substantial flow. the process we aim at understanding, i.e. why a sill can have the least evolved composition in the center and the most evolved composition at its upper and lower margins. Petrographic observations from the Golden Valley Sill show no evidence for larger concentrations of phenocrysts in the center of the sill as compared to its margins. The occurrence of D-shaped profiles in phenocryst-poor melts has previously been pointed out as a weakness of the flow differentiation model (e.g. Latypov, 2003a,b). In addition Barriere (1976) has shown that crystal segregation is practically inoperative for sills with thickness of 100 m or more. Olivine is only a minor phase, so the phenocryst assemblage consists mostly of plagioclase. Plagioclase does not partition MgO from FeO and can therefore not be responsible for the observed Mg# profile. Thus, flow segregation is unlikely to be the major differentiation mechanism, although some variations in the phenocryst distribution cannot be excluded. There is no evidence for a second intrusion in the Golden Valley Sill. The samples are all homogeneous, and the geochemical profiles show no abrupt changes. The field observations strongly suggest one intrusion causing the Golden Valley Sill. Where the Golden Valley Sill is in direct contact with another sill there is clear evidence of two separate intrusions, both by observations and geochemistry (Galerne, in press). We cannot exclude the possibility of continuous infill, although this requires a source of continuously more primitive magma in order to explain the D-shape. However, D-shapes are occurring in lava flows where there is clearly one single pulse (Latypov, 2003b and references therein). Soret fractionation (thermal diffusion) is a process that causes the heavy components (e.g. Fe) to migrate toward the colder end of a thermal gradient, and the lighter components (e.g. Mg) to migrate towards the hotter end. The geochemical data on the Golden Valley Sill show strong evidence against Soret fractionation. If heavy elements migrate towards the cooling margins, we would expect the same trend for all elements of differing molecular weight. In contrast, Ca increases towards the center of the sill and Na towards the margin, although Ca is heavier than Na. Also Ni and Cr are enriched in the center although the molecular weights are within the same range as Fe (Fig. 4). The presented model of post-emplacement flow explains well the symmetrical behavior of the geochemical data, and strengthens the assumption that gravity is not an important factor. Gravity settling is expected to result in more asymmetric profiles (e.g. Gibb and Henderson, 1992). The overall efficiency of gravity settling as a differentiation mechanism in sills is largely questioned (Shirley, 1987; Marsh, 1988, 1989). In large magma chambers, however, settling of crystals behaving as effective dense plumes may occur (Brandeis and Jaupart, 1986). It is widely recognized that the main cause of melt differentiation occurs when melt and crystals are segregated, for example by crystal settling, or by compositional convection (Kerr and Tait, 1986; Tait and Jaupart, 1992; Jaupart and Tait, 1995). Post-emplacement melt-flow has the potential of causing differentiation by advective transport (Fig. 10a). An outwards flux of melt from the central parts of the sill can increase the content of compatible elements in the center, and the content of incompatible elements in the distal parts. The resulting profile of whole-rock geochemistry will be D-shaped (Fig. 10b). Normal zoning observed in the pyroxenes suggests that fractionation of the melt occurs continuously throughout the whole magma body implying that melt becomes increasingly enriched in incompatible elements compared to the crystals. This favors the presence of a compositional melt-gradient from the center to the margins. For example, plagioclase is more anorthitic in the center and more albitic in the margins. The proportion of olivine may be too small to influence the whole-rock Mg# significantly, although the olivine crystals also Fig. 10. a) A schematic drawing of the flow process at one time-step during crystallization. Early forming crystals are mainly plagioclase and do not affect the Mg#. Melt flows from the central parts and into the margins, thus depleting the center and enriching the margins in incompatible and less incompatible elements, and vice versa for the compatible elements. The resulting geochemical profile will be according to b). I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 show the D-shaped trend with the largest changes towards the margins. This is in agreement with the pressure gradients being largest at the margins where cooling is most rapid, thus giving the strongest driving force for the melt flow. Assuming that we start with a homogeneous intrusion, no flow will result in an I-shaped profile (closed system crystallization). Melt can ultimately be drawn from areas that have less than 55 vol. % crystals, and mechanical compaction or inflow from other parts of the sill will make sure no voids develop where the last melt extraction occurs. If the entire sill, contemporaneously and in all directions, has a crystal-content above the limit of mechanical compaction, the sill will have a relatively flat cooling gradient. Hence, there will be no strong internal pressure gradients. In such settings post-emplacement flow should not be expected. There are several studies suggesting that reactive-advective meltflow will cause changes in geochemistry (McKenzie, 1984; Navon and Stolper, 1987; Spiegelman, 1996; Hauri, 1997; Spiegelman et al., 2001; Steefel et al., 2005). Weinberg (2006) stresses that segregation processes are very important for our understanding of the physical processes that lead to chemical and mineralogical differentiation in magmas. A study by Johnson et al. (2003) suggests that also the thermal, compositional and rheological gradients in the host-rocks may influence directions of pluton expansion. Rabinowicz et al. (2001) ask for physical processes that are capable of inducing strong variations in the melt/rock ratio. 4.5. Application to the Golden Valley Sill The result of the numerical modeling (Fig. 9) can be applied to the Golden Valley Sill, and other natural systems in terms of measurable quantities. The permeability, χ, of the crystal network is a function of grain size and porosity (i.e. melt percentage), which in turn are functions of cooling time. To get a rough measure of permeability expressed in average grain size we use the Carman (1956) relationship 2 χ¼ D /3 114 ð1−/Þ3 ð12Þ — where 114 is a fit parameter (Hersum et al., 2005), D is mean crystal length in meters and ϕ is porosity of the crystal network (melt fraction). We set the porosity to 10%, i.e. 90% crystals. Viscosity is a function of the chemical composition of the melt. We use the model of Giordano and Dingwell (2003) to describe the viscosity of multicomponent silicate melts. Viscosity is linked to composition by log10 μ ¼ c1 þ c2 c3 c3 þ SM 163 A representative I-shaped profile from another part of the Golden Valley Sill has an average grain size of ~ 0.5 mm and a SM number of ~25 (Galerne et al., 2008). A D-shaped profile from a sill underlying the Golden Valley Sill has an average grain size of ~0.75 mm and an average SM number of ~ 32 (Galerne et al., 2008). The local variations from sills that are overall homogeneous seem to be sufficient to separate between substantial flow and insignificant flow. Thus, the model can explain occurrences of different profiles within one sill by varying degrees of flow. In Fig. 11 we have also plotted examples of D- and I-shaped profiles reported in the literature. Richardson (1979) reports a D-shape from the Tandjiesberg sill (80–110 m thick), South Africa. The SM number ~26 is calculated on basis of bulk compositions and average of solid (from Table 5 in the paper Richardson, 1979). Grain sizes vary from 0.2 at the margins to 2–3 mm in the center. An average of ~1 mm is used. Gunn (1966) gives data from a D-shape (with some tendencies towards S-shape) from the Lake Vanda sill (332 m thick), Antarctica. The SM number varies from ~ 25–32 (Table 8 in the paper of Gunn, 1966), and the average is used. Reported grain sizes are 1–4 mm. We have used 1.25 mm. Gunn (1962) gives data for a profile with limited variation (I-shape) from the Peneplain sill (330 m thick), Antarctica. The SM number is ~23 (at the center). Crystal sizes vary from 0.4– 1 mm. An average of 0.7 mm is used. Gibb and Henderson (2006) published data from the Shiant Isles Main Sill, Scotland, composed of multiple intrusions. One batch of crinanite shows an approximate Ishape, while a lower batch of picrite shows a nice D-shape. The SM numbers of the two batches are clearly different, the I-shape having a SM number of ~27 and the D-shape having a SM number of ~ 41. The crinanite has an average grain size of b0.75 mm (Gibb and Henderson, 1996). We have visualized the crinanite data for a grain size of ~0.6 mm. The grain size for picrite is reported as “large”, hence a dashed star is used at ~ 0.8 mm, as a conservative value. The natural examples fit very well with the predictions of the model. The examples were all from relatively low-viscosity melts of mafic compositions. A viscous granitic melt with higher SiO2-content (lower SM) is less likely to segregate from the crystal network, although evidence for segregation in granitic bodies has been reported (Weinberg, 2006). It is not obvious that our transition zone between substantial and insignificant flow is correct. The boundary between the D- and I-regime should therefore be shifted if other observations show a significant misfit between observed data and our model. The ð13Þ where SM = Σ(Na2O + K2O + CaO + MgO + MnO + FeOtot/2). Lower SM number corresponds to higher viscosity. c1, c2 and c3 represent adjustable parameters, c1 ¼ −17:80106 þ 0:018708103T ð-CÞ 1−2:2869 10−3 T ð-CÞ c2 ¼ 0:02532 þ 2:5124expð−6:3679 10−3 T ð-CÞ þ40:4562 10−6 T ð-CÞ−1 c3 ¼ : −3 1−1:6569 10 T ð-CÞ 0:017954−63:90597 10−6 T ð-CÞ The result of these conversions is plotted in Fig. 11, and can be used directly to compare with natural cases. The samples in this study have an average grain size of 1 mm, and an average SM of ~31. Thus, the profile plots within the regime of significant flow where D-shape is feasible. Fig. 11. Permeability as a function of grain size and viscosity as a function of composition. The probability of getting a significant flow increases with longer cooling time and lower silica content. For 10% porosity the various occurrences of D- and I-shaped profiles plot within the domains of significant and no flow, respectively, showing a good agreement with our predictions. 164 I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Table 6 Calculations of bulk composition in the margin and center using Eqs. (14a) and (14b) Method Element Xinitial Xmargin Xcenter Calculated 14a Calculated 14b Data Calculated 14a Calculated 14b Data Zr Zr Zr Y Y Y 92 92 101 101 101 26 26 26 87 87 87 21 23 21 24 24 10% influx to the margins and 5% outflux of the center correspond very well to the data from the Golden Valley. model predicts that flow is likely to occur under natural conditions. However, prediction of the geochemical effects of such flow in nature is not trivial. Such prediction can only be made when a proper closed set of equations coupling flow, mass balance, thermodynamic equations and element partitioning has been established. Some recent progress has been made in the direction of developing a fully coupled set of equations solving the scenario of differentiation due post-emplacement melt flow (Tantserev, 2008), but this is still work in progress. We will therefore provide simple mass balance calculations for the incompatible elements Zr and Y in order to illustrate the possibility of melt flow causing chemical differentiation. We calculate the bulk concentration X of the element by weighting the fraction F of the phase and concentration C of the element in that phase Xmargin ¼ Fc Cc þ Fm Cm þ Fin Cin Xcenter ¼ Fc Cc þ Fm Cm −Fout Cout ð14aÞ with indexes c, m, in and out indicating crystals, melt, influxed and outfluxed melt, respectively. For comparison we adapt the equations of specifically developed for differentiation by infiltration by Jellinek and Kerr (2001), Xmargin ¼ ðFin e þ 1ÞCm Xcenter ¼ ð/ þ ð1−/ÞDÞð1−Fout eÞCm ð14bÞ where ε is enrichment factor of the moving melt and ϕ is melt fraction. The results and data are shown in Table 6. We let the average compositions represent the bulk composition of the melt before differentiation (I-shape), Cm = Xinitial. We use an influx of 10% melt to the margins, Fin = 0.1, and 5% outflux from the center Fout = 0.05. We assume that the elements are perfectly incompatible in the crystal network of plagioclase and olivine, i.e. Cc = 0 and D = 0. We use an enrichment factor of 1ε = 1, and ϕ = 1, i.e. pure melt, as a conservative value. Our results fit the data very well, and show that the compositional trends can readily be explained by 10% net influx of melt at the margins, and 5% net outflux in the center. For fluxes of more enriched melt, ε N 1, which is to be expected in nature, the differentiation trend is calculated to be more pronounced than what the data shows. However, a retardation factor keeping the local thermodynamic equilibrium would slow down the differentiation and requires more melt to see the same trends as observed. It is important to note that these calculations are not showing the full “truth” of what is actually going on. The estimates show that a feasible amount of melt segregation is needed to explain our data. An interesting implication of the formation of pressure anomalies in and around sills is that this may explain the occurrence of sandstone-dykes intruding into the sill roof of the Golden Valley Sill, a phenomenon described by Planke et al. (2000) and Van Biljon and Smitter (1956). The heating results in a large overpressure in the contact aureole, and combined with the underpressure developing in the sill roof, there is a strong driving force for sucking fluidized sediments into the sill as soon as a crack develops. There are still unresolved problems associated with the cooling and crystallization of sheet-intrusions exemplified by the variety of profile-shapes within the limit of one sill. The proposed postemplacement flow-model may explain some of these varieties by varying degrees of melt displacement, and thus differentiation. In our numerical model we have demonstrated that thermal stresses develop in the cooling margins of the sill (Fig. 8), but the margins are not necessarily the only domains where thermal stresses can develop in an extensive intrusion like the Golden Valley Sill. A several kilometers wide saucer-shaped sill can have local differences in degree of cooling and crystallization both in lateral (or sub-lateral) and vertical directions. Melt may also be transported laterally or sublaterally, e.g. by flow of melt from hotter parts closer to the feeder towards cooler parts of the sill (Fig. 12). Dissimilar flow patterns in different parts of a sill open for the possibility that different geochemical profiles may develop in different parts of a single sill. Different chemical profiles are indeed found in different parts of the Golden Valley Sill and in the York Haven sheet intrusion (Mangan et al., 1993; Galerne, pers. comm.). Fig. 12. a) Scenario of lateral flow. More melt is leaving from the center than the margins due to longer cooling time. b) More differentiation is occurring in the center due to more flow. Margin samples correspond to the original composition in this scenario. I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Post-emplacement melt flow may explain geochemical variations in several magmatic intrusions were former models have failed. With very few assumptions we have shown that post-emplacement flow is feasible under natural conditions, and also compatible with geochemical observations from several sills and the formation of D-shaped profiles. 165 grain sizes); 5) data from this study combined with other studies coincide well with the predicted conditions for substantial and insignificant flow. Our results strongly suggest that post-emplacement melt flow is an important, and previously overlooked, factor in differentiation of sill intrusions. Acknowledgements 4.6. Geological significance Zones adjacent to the basal contact of large intrusions are expected to be fully crystallized in situ before the onset of compositional convection. If negligible mass transport occurs, the same final rock composition should be found at all levels (I-shape). The basal zone of the Stillwater Complex (Page, 78, Jaupart and Tait, 1995) becomes progressively enriched in pyroxene and olivine up from the contact (D-shape). This progression may be interpreted as recording the advection of fresh melt towards more crystallized cooling margins (see Jaupart and Tait, 1995). For large-scale differentiation to be efficient a large amount of melt is expected to “flush” through the system. An extensive influx of melt towards the margins would require a corresponding outflux. Such outflow is expected to be localized into narrow zones (Jaupart and Tait, 1995). Vertical or subvertical segregation veins, pegmatitic structures and evolved pipe-like bodies are indeed found to crosscut the horizontal layering in several sills, like the Bushveld intrusion, South Africa (McKenzie, 1984; Hauri, 1997), the Stillwater Complex, Montana (Navon and Stolper, 1987), the Peneplain Sill, Antarctica (Spiegelman, 1996), and also in the upper and lower part of 1959 Kilauea Iki lava lake (Spiegelman et al., 2001; Steefel et al., 2005; Weinberg, 2006), or as “sandwich horizon” of incompatible elements in the Palisades Sill (Shirley, 1987). Flow induced by thermal stress, combined with local buoyancy effects from either composition or thermal differences, may help to explain these occurrences in the ca.100 m basal zones of large intrusions. These are not directly linked to formation of Dshape, but confirm that segregation processes are highly present in sills. Modal mineral zoning is the ultimate degree of compositional segregation that can be achieved by reactive melt flow. Chromatographic columns are commonly found in sills (e.g. Jaupart and Tait, 1995; Latypov, 2003b, and references therein), and can also display Dshaped geochemical profiles. The modal layers in such sills can be regarded as metasomatic columns where the chemical fronts are formed due to a differential movement of melt relative to crystals (Guy, 1993; Korzhinskii, 1973). Thermal stress induced by cooling is a strong driving force with the potential of segregating large amounts of melts from its equilibrium solids. Thus, post-emplacement flow can readily explain D-shaped profiles in sheet-intrusions of variable types and sizes. 5. Conclusions D-shaped geochemical profiles in sills are formed by a differentiation process which causes the upper and lower margins to have a more evolved composition than the center of the sill. Where no differentiation occurs we get I-shaped profiles. We propose that the differentiation is caused by a separation of melt from crystals by a porous melt flow through a rigid crystal network. Thermal stresses associated with the cooling and crystallization of the sill is a feasible driving force for substantial porous-flow. Numerical modeling and dimensional analysis have shown that 1) cooling of sills leads to large underpressure within the sill, while heating of the host-rock leads to large over-pressure in the contact aureole; 2) the underpressure can drive a substantial melt flow after the main emplacement episode of a sheet-intrusion and 3) a substantial melt flow is feasible under natural occurring conditions, like the case of the Golden Valley Sill; 4) the main factors controlling the flow are the viscosity of the melt (i.e. geochemistry) and the cooling time (i.e. This study was supported by the Norwegian Research Council (grant 159824/V30, “Emplacement mechanisms and magma flow in sheet intrusions in sedimentary basins”) through PGP, a Center of Excellence at the University of Oslo. We thank professor Julian (Goonie) Marsh, Rhodes University, Department of Geology in Grahamstown, South Africa, for supplying field equipment, the farmers whose land we were accessing, and K. Haaberg, S. Polteau and C. Galerne for assistance and collaboration during and after field work. H. Svensen, O. Galland, G. Gisler and B. Jamtveit are thanked for assistance and discussions. We would also like to thank Richard Kerr and two anonymous reviewers for constructive comments which led to substantial improvement of the manuscript. Thanks to M. Erambert for electron microprobe analytical support and R. Latypov for discussions and useful references. References Bachmann, O., Bergantz, G.W., 2006. Gas percolation in upper-crustal silicic crystal mushes as a mechanism for upward heat advection and rejuvenation of near-solidus magma bodies. J. Volcanol. Geotherm. Res. 149, 85–102. Barriere, M., 1976. Flowage differentiation — limitation of Bagnold effect to narrow intrusions. Contrib. Mineral. Petrol. 55, 139–145. Barton, M.D., Staude, J.M., Snow, E.A., Johnson, D.A., 1991. Aureole systematics. Rev. Miner. 26, 723–847. Brandeis, G., Jaupart, C., 1986. On the interaction between convection and crystallization in cooling magma chambers. Earth Planet. Sci. Lett. 77, 345–361. Carman, P.Z., 1956. Flow of Gases Through Porous Media. Butterworths, London. Cherepanov, A.N., Sharapov, V.N., Krivenko, A.P., 1982. A possible mechanism of the antidromic differentiation in the basic plutons. Dokl. Akad. Nauk SSSR 267, 1204–1207. Chevallier, L., Woodford, A., 1999. Morpho-tectonics and mechanism of emplacement of the dolerite rings and sills of the Western Karoo, South Africa. S. Afr. J. Geol. 102, 43–54. Delaney, P.T., 1982. Rapid intrusion of magma into wet rock — groundwater-flow due to pore pressure increases. J. Geophys. Res. 87, 7739–7756. Delaney, P.T., 1988. Fortran-77 programs for conductive cooling of dikes with temperaturedependent thermal-properties and heat of crystallization. Comput. Geosci. 14, 181–212. Duncan, R., Hooper, P., Rehacek, J., Marsh, J., Duncan, A., 1997. The timing and duration of the Karoo igneous event, southern Gondwana. J. Geophys. Res. 102, 18127–18138. Du Toit, A.L., 1920. The Karoo dolerites of South Africa: a study in hypabyssal injection. Trans. Geol. Soc. S. Afr. 23, 1–42. Galerne, C., in press. Emplacement mechanisms and magmatic differentiation induced by magma flow in sill intrusions in sedimentary basins, University of Oslo. Galerne, C., Neumann, E.R., Planke, S., 2008. Emplacement mechanisms of sill complexes: information from the geochemical architecture of the Golden Valley Sill Complex, South Africa. J. Volcanol. Geotherm. Res. 177, 425–440. Galushkin, Y.I., 1997. Thermal effects of igneous intrusions on maturity of organic matter: a possible mechanism of intrusion. Org. Geochem. 26, 645–658. Gibb, F., Henderson, C., 1992. Convection and crystal settling in sills. Contrib. Mineral. Petrol. 109, 538–545. Gibb, F.G.F., Henderson, C.M.B., 1996. The Shiant Isles Main Sill: Structure and mineral fractionation trends. Mineral. Mag. 60, 67–97. Gibb, F.G.F., Henderson, C.M.B., 2006. Chemistry of the Shiant Isles Main Sill, NW Scotland, and wider implications for the petrogenesis of mafic sills. J. Petrol. 47, 191–230. Giordano, D., Dingwell, D.B., 2003. Non-Arrhenian multicomponent melt viscosity: a model. Earth Planet. Sci. Lett. 208, 337–349. Gunn, B.M., 1962. Differentiation in Ferrar Dolerites, Antarctica. Am. Mineral. 47 191-&. Gunn, B.M., 1966. Modal and element variation in Antarctic tholeiites. Geochim. Cosmochim. Acta 30, 881–920. Guy, B., 1993. Mathematical revision of Korzhinskii theory of infiltration metasomatic zoning. Eur. J. Mineral. 5, 317–339. Hauri, E.H., 1997. Melt migration and mantle chromatography, 1: simplified theory and conditions for chemical and isotopic decoupling. Earth Planet. Sci. Lett. 153, 1–19. Hersum, T., Hilpert, M., Marsh, B., 2005. Permeability and melt flow in simulated and natural partially molten basaltic magmas. Earth Planet. Sci. Lett. 237, 798–814. Hoek, E., Brown, E.T., 1997. Practical estimates of rock mass strength. Int. J. Rock Mech. Min. Sci. 34, 1165–1186. Ingebritsen, E., Sandford, W.E., 1998. Groundwater in Geological Processes. Cambridge University Press. 340 pp. 166 I. Aarnes et al. / Earth and Planetary Science Letters 276 (2008) 152–166 Jamtveit, B., Svensen, H., Podladchikov, Y.Y., 2004. Hydrothermal vent complexes associated with sill intrusions in sedimentary basins. In: Breitkreuz, C., Petford, N. (Eds.), Physical Geology of High-level Magmatic Systems 234, Geological Society. Special Publications, London. Jaupart, C., Tait, S., 1995. Dynamics of differentiation in magma reservoirs. J. Geophys. Res.-Solid Earth 100, 17615–17636. Jellinek, A.M., Kerr, R.C., 2001. Magma dynamics, crystallization, and chemical differentiation of the 1959 Kilauea Iki lava lake, Hawaii, revisited. J. Volcanol. Geotherm. Res. 110, 235–263. Johnson, S.E., Fletcher, J.M., Fanning, C.M., Vernon, R.H., Paterson, S.R., Tate, M.C., 2003. Structure, emplacement and lateral expansion of the San Jose Tonalite pluton, Peninsular Ranges batholith, Baja California, Mexico. J. Struct. Geol. 25, 1933–1957. Kerr, R.C., Tait, S.R.,1986. Crystallization and compositional convection in a porous-medium with application to layered igneous intrusions. J. Geophys. Res.-Solid Earth and Planets 91, 3591–3608. Korzhinskii, D.S., 1973. Theory of metasomatic zoning. A Reply to Hofmann, A., Dr. Am. J. Sci. 273, 958–959. Latypov, R., 2003a. The origin of basic–ultrabasic sills with S-, D- and I-shaped compositional profiles by in situ crystallization of a single input of phenocryst-poor parental magma. J. Petrol. 44, 1619–1656. Latypov, R., 2003b. The origin of marginal compositional reversals in basic–ultrabasic sills and layered intrusions by Soret fractionation. J. Petrol. 44, 1579–1618. Le Gall, B., Tshoso, G., Jourdan, F., Feraud, G., Bertrand, H., Tiercelin, J.J., Kampunzu, A.B., Modisi, M.P., Dyment, J., Maia, M., 2002. Ar-40/Ar-39 geochronology and structural data from the giant Okavango and related mafic dyke swarms, Karoo igneous province, northern Botswana. Earth Planet. Sci. Lett. 202, 595–606. Litvinovski, B.A., Podladchikov, Y.Y., Zanvilevitch, A.N., Dunitchev, V.M., 1990. On the melting of acidic volcanites in the contact of basic magma at shallow depth. Geochimya 6, 807–814. Mangan, M.T., Marsh, B.D., 1992. Solidification front fractionation in phenocryst-free sheet-like magma bodies. J. Geol. 100, 605–620. Mangan, M.T., Marsh, B.D., Froelich, A.J., Gottfried, D., 1993. Emplacement and differentiation of the York Haven diabase sheet, Pennsylvania. J. Petrol. 34, 1271–1302. Marsh, B.D., 1996. Solidification fronts and magmatic evolution. Mineral. Mag. 60, 5–40. Marsh, B.D., 1988. Crystal capture, sorting, and retention in convecting magma. Geol. Soc. Amer. Bull. 100, 1720–1737. Marsh, B.D., 1989. On convective style and vigor in sheet-like magma chambers. J. Petrol. 30, 479–530. Marsh, J., Hooper, P., Rehacek, J., Duncan, R., Duncan, A., 1997. Stratigraphy and age of Karoo basalts of Lesotho and implications for correlations within the Karoo igneous province. In: Mahoney, J., Coffin, M. (Eds.), Large Provinces: Continental, Oceanic and Planetary Flood Volcanism 100. Geophysical Monograph, pp. 247–272. McKenzie, D., 1984. The generation and compaction of partially molten rock. J. Petrol. 25, 713–765. Naslund, H.R., 1984. Petrology of the Upper Border Series of the Skaergaard intrusion. J. Petrol. 25, 185–212. Navon, O., Stolper, E., 1987. Geochemical consequences of melt percolation — the upper mantle as a chromatographic column. J. Geol. 95, 285–307. Philpotts, A.R., Carroll, M., 1996. Physical properties of partly melted tholeiitic basalt. Geology 24, 1029–1032. Planke, S., Malthe-Sørenssen, A., Jamtveit, B., Petroleum implications of sill intrusions: an integrated study, VBPR/TGS-NOPEC Commercial Report, 2000, p. plates (A3). Planke, S., Rassmussen, T., Rey, S.S., Myklebust, R., 2005. Seismic characteristics and distribution of volcanic intrusions and hydrothermal vent complexes in the Vøring and Møre basins. In Petroleum Geology: North-West Europe and Global Perspectives. In: Dore, A., Vining, B. (Eds.), Petroleum Geology: North-West Europe and Global Perspectives, Proceedings of the 6th. Geology Conference, Geological Society, London. Podladchikov, Y.Y., Wickham, S.M., 1994. Crystallization of hydrous magmas — calculation of associated thermal effects, volatile fluxes, and isotopic alteration. J. Geol. 102, 25–45. Rabinowicz, M., Genthon, P., Ceuleneer, G., Hillairet, M., 2001. Compaction in a mantle mush with high melt concentrations and the generation of magma chambers. Earth Planet. Sci. Lett. 188, 313–328. Richardson, S.H., 1979. Chemical variation induced by flow differentiation in an extensive Karoo dolerite sheet, southern Namibia. Geochim. Cosmochim. Acta 43, 1433–1441. Shirley, D.N., 1986. Compaction of igneous cumulates. J. Geol. 94, 795–809. Shirley, D.N., 1987. Differentiation and compaction in the Palisades Sill, New-Jersey. J. Petrol. 28, 835–865. Simkin, T., 1967. Flow differentiation in the picritic sills of north Skye. In: Wyllie, P.J. (Ed.), Ultramafic and Related Rocks. Wiley, New York, pp. 64–69. Spiegelman, M., 1996. Geochemical consequences of melt transport in 2-D: the sensitivity of trace elements to mantle dynamics. Earth Planet. Sci. Lett. 139, 115–132. Spiegelman, M., Kelemen, P.B., Aharonov, E., 2001. Causes and consequences of flow organization during melt transport: the reaction infiltration instability in compactible media. J. Geophys. Res.-Solid Earth 106, 2061–2077. Steefel, C.I., DePaolo, D.J., Lichtner, P.C., 2005. Reactive transport modeling: an essential tool and a new research approach for the Earth sciences. Earth Planet. Sci. Lett. 240, 539–558. Svensen, H., Jamtveit, B., Planke, S., Chevallier, L., 2006. Structure and evolution of hydrothermal vent complexes in the Karoo Basin, South Africa. J. Geol. Soc. 163, 671–682. Svensen, H., Planke, S., Chevallier, L., Malthe-Sorenssen, A., Corfu, F., Jamtveit, B., 2007. Hydrothermal venting of greenhouse gases triggering Early Jurassic global warming. Earth Planet. Sci. Lett. 256, 554–566. Tait, S., Jaupart, C., 1992. Compositional convection in a reactive crystalline mush and melt differentiation. J. Geophys. Res.-Solid Earth 97, 6735–6756. Tantserev, E., 2008. Time reverse methods in modeling of diffusive. Convective and reactive transport. PhD-thesis, University of Oslo. Tommasini, S., Davies, G.R., 1997. Isotope disequilibrium during anatexis: a case study of contact melting, Sierra Nevada, California. Earth Planet. Sci. Lett. 148, 273–285. Turcotte, D., Schubert, G., 2002. Geodynamics. Cambridge. Van Biljon, W.J., Smitter, Y.H., 1956. A note on the occurrence of two sandstone dykes in a Karoo dolerite sill near Devon, South-Eastern Transvaal. Trans. Geol. Soc. S. Afr. 59. Voight, B., Linde, A.T., Sacks, I.S., Mattioli, G.S., Sparks, R.S.J., Elsworth, D., Hidayat, D., Malin, P.E., Shalev, E., Widiwijayanti, C., Young, S.R., Bass, V., Clarke, A., Dunkley, P., Johnston, W., McWhorter, N., Neuberg, J., Williams, P., 2006. Unprecedented pressure increase in deep magma reservoir triggered by lava-dome collapse. Geophys. Res. Lett. 33. Wager, L.R., Brown, G.M., 1968. Layered Igneous Rocks. WH Freeman, San Francisco. Weinberg, R.F., 2006. Melt segregation structures in granitic plutons. Geology 34, 305–308.