The analyzed flows sampled during Legs 152 and 163 (Larsen, Saunders, Clift, et al., 1994; Duncan, Larsen, Allan, et al., 1996; Larsen et al., 1998; Larsen et al., this volume) have been recast into the same normative components using the same calculation procedure as for illustrating the experimental phase equilibria. The results are shown in Figure 7 and highlight several important features. As was observed by Thy et al. (1998), the primitive flows from Hole 917A in the Lower and Upper Series plot away from the experimentally determined multiply-saturated cotectic. This is consistent with the phenocryst assemblages found in these lavas (Larsen, Saunders, Clift, et al., 1994; Demant, 1998). Thy et al. (1998) used inferred shifts in the locations of the olivine-plagioclase cotectic to suggest that some differentiated Lower Series lavas were modified by crustal contamination and concluded that excessive olivine fractionation caused by suppressed plagioclase crystallization at elevated partial water pressures could explain some of the compositional features of the Lower Series lavas. In general, however, both the Lower and Upper Series were controlled by near-anhydrous crystallization conditions. In contrast, lavas of the Middle Series show large variations in the normative projections and plot systematically below the projected olivine-plagioclase-augite cotectic in Figure 7B. This position implies fractionating phase assemblages of olivine or low-Ca pyroxene and plagioclase. Such fractionating assemblages would not lead to the large variation in SiO₂ content shown by the Middle Series lavas.

Recovery at Site 988 consisted of differentiated olivine, plagioclase, and augite phyric lava from only one fresh flow. The analyzed samples from this flow cluster near the multiply-saturated experimental cotectic, consistent with the evolved nature of the flow and its phenocryst assemblage. In these respects, the Site 988 lava is similar to the onshore plateau lavas in the Scoresby Sund area (Larsen et al., 1989; Fram and Lesher, 1997). The flows analyzed from Sites 989 and 990 are compositionally very similar to each other and narrowly cluster along the multiply-saturated cotectic, consistent with their aphyric to highly olivine, plagioclase, and augite phyric nature.

The multiply-saturated liquid line of descent for the Site 918 sample is slightly offset from the trend determined for the basaltic Site 917 samples by lower normative diopside and plagioclase (Fig. 6). The flows recovered from the oceanic succession at Site 918 have compositions that deviate from their experimentally determined liquid line of descent (compare Fig. 6 and Fig. 7). The majority of the Site 918 lavas are aphyric and contain microphenocrysts of olivine and plagioclase and, occasionally, augite phenocrysts (Larsen, Saunders, Clift, et al., 1994). Therefore, these are likely to have been controlled by olivine and plagioclase crystallization and thus plot away from the olivine, plagioclase, augite cotectic toward lower normative diopside (Fig. 7). Lesher et al. (Chap. 12, this volume) discuss important kinetic controls on the nucleation of augite that further help to explain the discrepancies between the petrographic observations and the expectations based on the equilibrium phase relations for the basalts of the oceanic series.

Quantitative modeling of fractional crystallization is presented in Figure 8. The modeling is accomplished by incrementally removing the calculated solid fractionate in equilibrium with the residual liquid. The calculation of the solid fractionate is based on the experimentally determined mineral compositions and their proportions from this study, Thy et al. (1996, 1998), and Toplis and Carroll (1995). The first model illustrates the effect of fractionation on the estimated primary liquid in equilibrium with mantle olivine (Fo₉₁) for Section 152-917A-86R-7 and uses the crystallization order experimentally determined (Thy et al., 1998). The second model shows the results for the section from Hole 989B using the crystallization order from Table 2.

We can draw several conclusions from this modeling. Most of the compositional variation within groups of flows is accounted for by <15% fractionation. The exceptions are a few TiO₂-rich flows from Site 917 that suggest extensive olivine, plagiocase, and augite removal approximating 50%-60% fractionation. The flows from Site 988, as well as a small subset from Site 918, also suggest derivation by extensive fractionation and may require a more Ti-rich parental composition than used in the calculations (Fig. 8).

The Middle Series at Site 917 deviates from the experimental liquid lines of descent (Fig. 7), as well as the modeling of compositional effects of fractional crystallization (Fig. 8). This is especially true for FeO and TiO₂ that do not show increasing concentrations with decreasing MgO as predicted for typical tholeiitic evolution trends. An alternative fractionation model involving early low-Ti magnetite saturation (~1180º-1200ºC) could account for an early decrease in FeO and only modest variation in TiO₂. However, early saturation of Fe-Ti oxides is neither supported by the present study nor predicted based on Fe-Ti oxide solubility in basaltic systems (e.g., Toplis and Carroll, 1995). Furthermore, magnetite and ilmenite are conspicuously absent in the lavas of the Middle Series (Larsen, Saunders, Clift, et al., 1994).

A consequence is that the major element variation observed for the Middle Series can neither be accounted for by the phenocryst assemblages nor by the experimental phase equilibria. The flows of the Middle Series are aphyric to plagioclase phyric basalts with embayed plagioclase and rare augite phenocrysts. The dacites and silicic tuffs are generally restricted to the lower and uppermost parts of the Middle Series and contain abundant embayed and sieved textured plagioclase phenocrysts and minor amounts of augite phenocrysts. Olivine and quartz are only observed as rare xenocrysts. The occurrence of augite phenocrysts, although not common, is inconsistent with the low-pressure phase relations shown in Figure 7, and it is possible that these augites are also xenocrystic. Alternatively, Thy et al. (1998) proposed that differentiation of Lower Series basaltic magmas was influenced by the addition of water, possibly during the early stages of crustal contamination. The effect of elevated water content is to reduce the stability of plagioclase relative to olivine and augite. This could explain the occurrence of pyroxene phenocrysts and sieve textured plagioclase phenocrysts.

The Leg 152 Shipboard Scientific Party concluded that flows and tuffs of the Middle Series were contaminated by significant amounts of crustal material and suggested a process of magma mixing between basaltic and silicic components (Larsen, Saunders, Clift, et al., 1994). Fitton et al. (1998a, 1998b) argued that the flows of the Lower Series at Site 917 were contaminated by granulite facies leucogneiss, while the Middle Series was contaminated by amphibolite facies leucogneiss. This temporal change in crustal contaminant is interpreted as caused by a shallowing of magma chambers during development of the Lower and Middle Series succession. The modeling of Fitton et al. (1998a) implies relatively high extents of fractionation (50%-70%) and/or assimilation to account for the dacitic flows of the Middle Series. The contaminants were suggested by Fitton et al. (1998a) to have been relatively similar to average Lewisian leucocratic gneisses analyzed by Hamilton et al. (1978) and Dickin (1981). Fitton et al. (1998a) proposed a contaminant for the Middle Series flows with high ⁸⁷Sr/⁸⁶Sr (0.7166) and low ¹⁴³Nd/¹⁴⁴Nd (0.5104) and, for modeling purposes, used a late Archean granitic gneiss from the Skjoldungen area of Southeast Greenland (Blichert-Toft et al., 1995). Another possible contaminant is granophyric material described by Blichert-Toft et al. (1992) from the Tertiary Miki Fjord macrodike of central East Greenland. This granophyre represents anatectic melt of Precambrian basement with a ⁸⁷Sr/⁸⁶Sr ratio of 0.7544 and a ¹⁴³Nd/¹⁴⁴Nd ratio of 0.5109. The major element compositions of these two potential contaminants are given in Table 7, and their chondrite-normalized rare-earth element (REE) concentrations are given in Figure 9A. Figure 9B-9D illustrate the results of combined fractional crystallization and assimilation processes (AFC) (DePaolo, 1981; Albarède, 1995) using these two possible contaminant types and Unit 89 of the Lower Series as the differentiating magma. The fractionation assemblage is 30% augite and 70% plagioclase, as experimentally determined by Beard and Lofgren (1991) and Thy et al. (1998). The results of the calculations are illustrated in Figure 9 for ranges in liquid remaining (F = 1 to 0.5) and constant ratios of assimilation and crystallization masses (R = 0 to 0.5). Figure 9B-9C show the effects of using the granophyre and Figure 9D of using the gneiss as contaminants. The AFC modeling of the REEs illustrates that it is possible to account for the observed variation in the heavy rare-earth elements (HREEs) for small amounts of fractionation and assimilation. However, none of the models will account for the relatively steep slope in the light rare-earth elements (LREEs). In addition, the model compositions show large negative Eu anomalies that are inconsistent with the REE abundance data for the Site 917 dacites.

It has been experimentally demonstrated that dehydration melting of amphibolitic protoliths will produce partial melts ranging from rhyolitic near the solidus to dacitic at slightly higher extent of melting (~10%-15%) (Beard and Lofgren, 1989, 1991; Wolf and Wyllie, 1991, 1994; Rushmer, 1991; Rapp et al., 1991; Sen and Dunn, 1994; Patiño Douce and Beard, 1995; Springer and Seck, 1997). The composition of such melts is strongly dependent on the extent of melting and the restitic phase assemblage. Sen and Dunn (1994) showed that melts produced at up to 15% melting in 10-kbar melting experiments are quartz saturated and broadly dacitic in composition, while melts produced between 20% and 30% melting are andesitic. The SiO₂ contents of melts produced in experiments conducted at lower pressures (<3 kbar) are relatively similar to those produced at higher pressures for identical melt fractions (5-15 kbar) (e.g., Beard and Lofgren, 1989; Springer and Seck, 1997). It was illustrated by Beard and Lofgren (1989), Thy et al. (1990), and Wolf and Wyllie (1994) that the Al₂O₃ concentrations of partial melts were indicative of the stability of plagioclase relative to pyroxene and amphibole in the source and consequently of the partial pressure of H₂O (pH₂O). High pH₂O would tend to stabilize amphibole and destabilize plagioclase resulting in melts with high Al₂O₃. In contrast, low pH₂O destabilizes amphibole and thus reduces the Al₂O₃ content of a partial melt (Beard and Lofgren, 1989). The relatively low Al₂O₃ contents of the andesites and dacites recovered from the Middle Series at Site 917 are similar to partial melts experimentally produced by dehydration melting of amphibolitic protoliths (Table 7).

Drilling of the SDRS at both Site 642 on the Vøring Plateau and in the Rockall Trough recovered cordierite-bearing dacite flows (Edholm, Thiede, Taylor, et al., 1987; Morton et al., 1988) with average ⁸⁷Sr/⁸⁶Sr of 0.7120 and ¹⁴³Nd/¹⁴⁴Nd of 0.5122 (Taylor et al., 1989). These dacite flows were interpreted to represent partial melts of upper crustal metasedimentary rocks (Viereck et al., 1988; Morton et al., 1988; Parson et al., 1989; Taylor and Morton, 1989). Melting of mica-bearing metasediments and gneisses produces large melt fractions at relatively low temperatures (Patiño Douce and Johnston, 1991). For identical melting temperature and pressure (e.g., 950ºC, 10 kbar), metapelites and tonalite gneisses produce 30%-60% melt, while amphibolite protoliths may produce around 10% melt (Gardien et al., 1995). Therefore, the compositions of partial melts of metasedimentary rocks are strongly controlled by the protolith and often highly silicic, aluminous, and alkalic (Johnston and Wyllie, 1988; Patiño Douce and Johnston, 1991; Skjerlie and Johnston, 1993; Patiño Douce and Beard, 1995; Scaillet et al., 1995; Singh and Johannes, 1996; Stevens et al., 1997). For example, a biotite gneiss melted by Patiño Douce and Beard (1995) produced 42% melt at 925ºC, which is granitic in composition, containing >6 wt% K₂O (Table 7). These melts differ from the tonalite, trondhjemite, or granodiorite melts produced from melting quartz amphibolite at similar experimental conditions (Patiño Douce and Beard, 1995). The silicic volcanic rocks of Site 917 are tonalitic or granodioritic using the Ab-Or-An classification (Baker, 1979), and, therefore, more akin to melts produced from amphibolite than from leucogneiss protoliths. The dacitic flows recovered from the Vøring and Rockall Plateaus (Table 7) have relatively high Al₂O₃ contents and are corundum normative (3-8%). Elevated Al₂O₃ concentrations can reflect the presence of a vapor phase during partial melting of an amphibolitic source (Beard and Lofgren, 1989) and may not uniquely identify a metasedimentary source as suggested by Viereck et al. (1988) and Morton et al. (1988). In terms of SiO₂ and K₂O contents, both the Vøring and Rockall dacites are relatively similar to the Site 917 dacites and to the melts produced from amphibolitic protoliths (Beard and Lofgren, 1989).

The lower crust along the prerifted continental margin may act as a barrier to magma migration and may consequently trap large volumes of mantle-derived magma (Fram and Lesher, 1997; Larsen and Saunders, 1998). Repeated injection and underplating may provide the heat for partial melting of lower to middle crustal granulites and amphibolites (Hildreth and Moorbath, 1988; Kay and Kay, 1991). Wolf and Wyllie (1995) suggest that the amphibolite dehydration melting solidus is equivalent to the H₂O-saturated basalt solidus and show that liquid interconnectivity can be achieved at temperatures <900°C and with <5 vol% liquid. The melting is related to the (simplified) breakdown of amphibole at low pressure to pyroxene + liquid and at pressures >10 kbar to pyroxene + garnet + liquid (Sen and Dunn, 1994; Wolf and Wyllie, 1995).

The possibility that the andesites and dacites of the Middle Series at Site 917 are produced by partial melting of mafic continental sources can be tested by examining the rare-earth systematic of the Middle Series dacites (Fitton et al., 1998b). The model calculations shown in Figure 10 assume nonmodal batch melting of continental mafic amphibolite (either amphibolite with 27% plagioclase, 70% amphibole, and 3% quartz or metagabbro with 35% plagioclase, 48% amphibole, 14% augite, and 3% quartz). Further, we assume that likely protoliths are represented by a range of rare-earth compositions from average depleted to relatively enriched lower basaltic crust (Rudnick and Fountain, 1995). The melting reactions used in the calculations are from Sen and Dunn (1994) and the partitioning coefficients from Martin (1987). The REE analyses available of Middle Series silicic flows are also shown in Figure 10B (Units 47 and 55) (Fitton et al., 1998b). The first set of modeling of partial melting of mafic amphibolite assumes the melting reaction quartz + plagioclase + amphibole = clinopyroxene + melt (Fig. 10B), while the second set uses the reaction quartz + plagioclase + amphibole = clinopyroxene + garnet + melt (Fig. 10C). The source modes only have minor effects on the model results (Fig. 10D) (e.g., Martin, 1987; Rapp et al., 1991; Sen and Dunn, 1994; Johnson et al., 1997; Springer and Seck, 1997). Of most importance in governing the rare-earth pattern is the proportion of garnet in the residue. To account for the relatively flat slopes of the HREEs and the steepening of the LREEs, a garnet-absent residue is required (Fig. 10B). This restricts the melting regime to middle and upper crustal conditions (<8 kbar; Wolf and Wyllie, 1995). It can be seen that the results based on both amphibolitic and gabbroic crustal xenolith compositions reproduce relatively well the general pattern observed in the REEs of the dacites (Fig. 10B-10C). However, only for very low extents of melting (<5%) or by assuming an enriched protolith composition can the dacitic flow Unit 55 be reproduced. In order to reproduce Unit 47, a source with slightly higher LREE/HREE ratios is required (Fig. 10B) and suggests some variability in the source LREE/HREE ratio.

The model suggested by Fitton et al. (1998a) requires contaminants with ¹⁴³Nd/¹⁴⁴Nd ratios <0.5109 and ⁸⁷Sr/⁸⁶Sr ratios >0.7140 to account for the isotopic relations seen in the Middle Series dacites (Fig. 11). The Southeast Greenland coastal areas are composed of Archean to Proterozoic granulite and amphibolite gneiss complexes that are dominated by tonalitic to granodioritic ortho- and paragneisses, metapelitic supercrustral rocks, metadolerite dikes, and granitic, syenitic, and basic intrusives (Taylor et al., 1992; Kalsbeek et al., 1993; Blichert-Toft et al., 1995). The isotopic signature of many of these orthogneisses and metagneisses fulfill the basic requirement for Fitton et al. (1998a, 1998b) contamination model (Fig. 11). However, the validity of the alternative proposal argued in this paper can equally be substantiated by the available isotope data. Metadolerite dikes and basic intrusions present in the basement suggest late Archean and Proterozoic underplating and intrusion of basaltic magma into the continental crust. Many of these basic dikes and intrusions have Nd-Sr isotope relations relatively similar to the dacites from Site 917 and the Vøring Plateau (Fig. 11). In particular, a Proterozoic basic dike swarm in the Ammassalik area reveals Nd-Sr isotope ratios (Kalsbeek et al., 1993) very similar to the Vøring dacites and to one of the Site 917 dacites (Fig. 11). Further, the Proterozoic Ammassalik igneous complex and the late Archean Skjoldungen complex, at about the same geographic latitude as Site 917, contain basic and intermediate intrusive components (Kalsbeek et al., 1993; Blichert-Toft et al., 1995) that isotopically compare well to other Site 917 dacites. Hence, there is evidence that old mafic crustal material was available for remelting during an early Tertiary continental breakup and that this can account for the high-Sr and low-Nd isotope ratios observed in the dacites.

The experimental and compositional modeling of the Site 917 dacites provide support for their formation as partial melts from basic protoliths. The relatively steep REE patterns and the lack of Eu anomalies can be closely modeled by partial melting using reasonable amphibolitic source compositions and melting reactions. In contrast, modeling of combined fractionation and assimilation with local gneissic contaminants produces relatively flat REE patterns and strong Eu anomalies that are not observed for the dacites. The origin, as partial melts from amphibolitic protoliths, is further attractive because it does not require extensive fractional crystallization of Fe-Ti oxide-containing assemblages. This latter point is important since Fe-Ti oxides do not occur as phenocrysts and are not saturated in the low-pressure melting experiments.