We have earlier (Fitton et al., 1995, 1998b; L.M. Larsen et al., 1998a; Saunders et al., 1998) demonstrated the differences in chemical composition between the relatively uniform postbreakup oceanic basalts at Sites 915 and 918 and the much more variable compositions in the pre- and synbreakup volcanics at Site 917. All the lava flows recovered at Site 989 and 990 have chemical compositions similar to the oceanic basalts at Site 915 and 918. This is illustrated by the four variation diagrams in Figure 3. In all four plots, the basalts from Sites 915, 918, 989, and 990 lie together in a single, well-defined tight cluster, which is completely separated from both the continental and the synbreakup volcanics at Site 917. The basalts at Site 989 and 990 are of indisputable oceanic character, with higher Sc and lower Ce (and other light REEs) than the continental and synbreakup volcanics. They also have lower MgO and higher SiO₂ than the synbreakup volcanics.

The volcanics from the 63ºN transect include two dikes, one in the Site 917 middle series (Unit 917-39; L.M. Larsen et al., 1998a, included here in Table 1) and one that cuts Unit 990-12. Both of these dikes have an oceanic chemical character, and in Figure 3 they plot within the tight cluster of basalts from Sites 915, 918, 989, and 990.

Figure 4 shows TiO₂ vs. Mg# for the basalts of Sites 989, 990, 915, 918, and the two dikes. The synbreakup series (Site 917 upper series) is shown for comparison. The basalts plot in a number of parallel trends that can be explained as fractionation trends, as detailed later. Site 917 upper series comprises two parallel trends, interpreted as belonging to magmas formed by different degrees of partial melting, smaller degrees of melting producing higher contents of TiO₂ in both primary and derivative melts (L.M. Larsen et al., 1998a; Fitton et al., 1998b; Fram et al., 1998). The most evolved basalts from the synbreakup series show similar degrees of fractionation to the least evolved basalts from Site 990 (Mg# = 60-63). All the postbreakup lavas and the dike from Site 917 plot within one, or perhaps two, trends with some of the basalts from Site 918 having the highest TiO₂. The dikelet in Unit 990-12, with its higher contents of incompatible elements, seems to have arisen from a separate magma batch produced by smaller degrees of melting or from a different source (Saunders et al., Chap. 8, this volume).

Figure 5 shows the upsection development in Mg# in a combined section through the syn- and postbreakup succession at Sites 917 and 990/915. A gap is inserted between the two to show the undrilled interval. The single lava flow from Site 915 occupies a stratigraphic position immediately above Site 990.

The other oceanic rocks are also shown in Figure 5, placed on top of the 917-990/915 succession in accordance with their younger age. Site 918 is clearly the youngest, being situated 70 km farther offshore from Site 990, and the stratigraphical gap between these two sites is considerable. Ar/Ar dating of the two flows from Site 989 (Tegner and Duncan, Chap. 6, this volume) has shown that they have the same age as those from Site 990 (~56 Ma), thus confirming that the flows at Site 989 belong to the postbreakup stage despite their position in the innermost part of the SDRS. The upper flow at Site 989 shows normal magnetic polarity, whereas the lower flow shows reversed magnetic polarity. The two uppermost flows at Site 990 also show normal magnetic polarity, whereas the other flows at Site 990 and the one at Site 915 show reversed magnetic polarity (Duncan, Larsen, Allan, et al., 1996). The normal magnetic polarity interval represented is probably C25n (Tegner and Duncan, Chap. 6, this volume). It is thus possible that the flows at Site 989 are exactly coeval with the upper flows at Site 990, but, for the sake of comparison, they have been placed above Site 990/915 in Figure 5.

The minimum ages of the two dikes are unconstrained, and for comparison they have been placed in Figure 5 in the "void" between Sites 989 and 918.

Figure 5 shows that there is a slight but distinct decrease in Mg# upsection through the flows at Site 990, from Mg# = 60-62 to Mg# = ~50. The Shipboard Scientific Party noticed that Zr increases and Cr and Ni decrease with height upsection, indicating increasing degrees of evolution (Duncan, Larsen, Allan, et al., 1996). Site 990 appears to consist of three flow groups: Units 13 to 6, with Mg# = 62.3-58, Zr = 39-49 ppm, Cr = 250-212 ppm, Ni = 125-83 ppm; Units 5 to 2, with Mg# = 58-55, Zr = 52-57 ppm, Cr = 158-128 ppm, Ni = 87-70 ppm; and Unit 1 with Mg# = ~49, Zr = 67-72 ppm, Cr = 36 ppm, Ni = 52-54 ppm. The flow from Site 915 is similar to Units 990-5 to 990-2. Site 915 is situated only 150 m northeast of Site 990, and it is difficult to imagine that Units 990-1 and 990-2 (both normally magnetized), with a combined thickness of ~90 m, should not be present at Site 915 below the drilled flow. Unit 990-1 is probably an unusually differentiated flow. It also has a very low Cr/Ni ratio of 0.7, an unusual (but not exceptional) value for a North Atlantic Tertiary basalt. Unit 989-1 likewise has Cr/Ni = 0.7, despite higher Mg# (and MgO) than Unit 990-1, which corroborates the possibility that these two flows could be coeval and even related. Unit 989-2 is most similar to Units 990-13 to 990-6.

The flows at Site 918 cover close to the same range in major and trace element compositions as those at Site 990 and, despite the signs of systematic development upsection at Site 990, we must conclude that the conditions that only allowed magmas within a fairly narrow compositional range to be erupted had been established before the flows at Site 990 were deposited. As in earlier works (Larsen, Saunders, Clift, et al., 1994; Fitton et al., 1995, 1989b), we believe this filtering happens in magma chambers in the new-formed oceanic crust. Thus, the allegedly short, undrilled interval between the Site 917 upper series and Site 990 must comprise the transition from syn- to postbreakup conditions. There is little indication of whether the transition is abrupt or gradational. Fram et al. (1998) suggest an increase in the degree of melting with time, but most of this increase takes place between the Site 917 upper series and Site 915. Data such as the Zr/Sc relations shown in Figure 6 suggest completely separate magmatic systems. Perhaps lavas from the two magmatic systems form inter-fingering flows over a short interval, as seen in the Faeroe Islands upper series (Gariépy et al., 1983; Waagstein, 1988).

Figure 6 is a plot of Zr vs. Sc for the syn- and postbreakup basalts. This was used by Fitton et al. (1998b) to demonstrate that the much higher Sc in the oceanic rocks can be an effect of thinning of the lithospheric lid with time, thus allowing more of the melting to take place in spinel-facies mantle, where Sc is relatively incompatible during melting. The new data for the oceanic basalts from Site 989 and 990 fall on exactly the same trend as Site 915 and 918; the less evolved lowermost units 990-13 to 990-6 just extend the trend to less fractionated compositions. Thus, after breakup no further gradual shallowing of the melting column is indicated.

The same conclusion can be reached from considerations of the major element compositions. SiO₂ is sensitive to pressure during melting (McKenzie and Bickle, 1988; Niu and Batiza, 1991; Kinzler and Grove, 1992, 1993), and the high and similar SiO₂ contents (50% irrespective of slight contamination; see below) in the oceanic rocks from all sites indicate final melt segregation at low pressures corresponding to conditions beneath an oceanic ridge, right from the beginning of the postbreakup stage. Synbreakup basalts at a similar stage of differentiation (Mg# = 63-60) have lower SiO₂ contents, average 48.9%, compared to an average of 51.2% in the oceanic rocks (calculated free of volatiles). Even though the Site 990 basalts are slightly contaminated (see below), the difference is in the order of 2% SiO₂, corresponding to a relative shallowing of the average melting pressure in the order of 3-5 kbar during the final breakup (Niu and Batiza, 1991; Kinzler and Grove, 1993).

Based on trace elements, Fram et al. (1998, fig. 6) modeled the oceanic basalts (Sites 915 and 918) as generated at 25-17 kbar by 10%-12% melting. The models of Niu and Batiza (1991), Kinzler and Grove (1992, 1993), and Langmuir et al. (1992) for calculating the conditions of generation (for polybaric incremental melting) of oceanic basalts of known major-element composition can also be applied to the fractionated magma composition from Site 990 with ~8.0% MgO (Table 2, column 2). Although the results vary between the models, all three models indicate that melting started at high pressures (33-20 kbar) and high temperatures (1580º-1460ºC), resulting in high degrees of melting (15-21 wt%). The results for Site 990 are very similar to the results for a basalt from the Reykjanes Ridge calculated by Kinzler and Grove (1993), but higher pressures and degrees of melting are indicated than for the trace-element-based model of Fram et al. (1998).

Figure 7 shows the variation in Ba/Zr within the same composite succession as in Figure 5. Increased Ba/Zr has been used as an indicator of crustal contamination (Larsen, Saunders, Clift, et al., 1994; Fitton et al., 1998a, 1998b; L.M. Larsen et al., 1998a). Figure 7 shows that, while the Site 918 basalts consistently have a Ba/Zr of <0.3, many Site 917 upper series basalts and almost all Site 989 and 990 basalts have a Ba/Zr of >0.3, although not as high as the Ba/Zr values of 2-5 (up to 22) found in the continental series. Could this slight elevation in Ba/Zr be caused by crustal contamination of the syn- and earliest postbreakup basalts? Saunders et al. (Chap. 8, this volume) present REE and isotope data showing that this is indeed the case. Most of the Site 989 and 990 basalts for which there are REEs and isotope data have elevated light REEs, low Pb isotope ratios, low Nd and high Sr isotope ratios. There is a very good correlation between high Ba/Zr and low ²⁰⁶Pb/ ²⁰⁴Pb, indicating that the elevated Ba/Zr is not caused by secondary introduction of Ba during alteration. Only one isotopically analyzed lava flow, Unit 989-1, and the dikelet in Unit 990-12 appear to have uncontaminated isotopes, and these are also among the few rocks with low Ba/Zr (<0.42). If Ba/Zr >0.42 is used as an indicator of contamination, then Figure 7 shows that all of the lava flows at Site 990 and also Unit 989-2 are crustally contaminated, as are several of the Site 917 upper series basalts. Both dikes appear to be uncontaminated. As shown by Fitton et al. (1998a), contamination by subcrustal lithospheric mantle is a possibility that cannot be excluded, but it is not required to explain the data.

Based on trace elements and isotope ratios, Saunders et al. (this volume) model a contamination process where the primary magma is bulk contaminated with 2%-5% gneiss, followed by ~30% olivine fractionation. Most trace elements are explained by addition of up to 2% gneiss. The gneiss sample used has 63.6% SiO₂, and incorporation of 2% of such material should give rise to an increase in SiO₂ of about 0.3% before and 0.4% after fractionation. An increase in SiO₂ of this size is on the limit of what is detectable in a natural data set. Unit 989-1 and the two dikes do have SiO₂ contents which are lower than in most of the other rocks, but in total there is no significant difference between the SiO₂ contents of the contaminated oceanic basalts and the uncontaminated ones from Site 918 (Fig. 3). The major elements thus suggest a maximum of 2% bulk gneiss contamination in accordance with the conclusions from the trace elements. The 5% contamination indicated by the Pb isotopes is probably a result of selective contamination with crustal Pb because of the great contrast in the contents of Pb in the two rock types.

As related in the summary of earlier results, the magmas of the continental series are considered to be fractionated and contaminated in magma chambers in the continental crust; the synbreakup series is variable and Mg-rich because magma chambers at breakup were transient or nonexistent; and the limited compositional range of the erupted oceanic rocks was acquired when magma chambers became re-established, this time in oceanic crust. However, despite their oceanic-type chemistry, all the flows at Site 989 and 990 are crustally contaminated except one (Unit 989-1, ironically one of the two flows resting on relatively thick continent in the inner part of the continental edge; Fig. 2). If the magmas were contaminated in the magma chambers, then these chambers must sit in, or be in contact with, continental crust. Alternatively, the magmas may have been contaminated when passing through the continental crust on their way to the surface.

Based on a deep seismic profile, regional gravity data, and an Iceland-type crustal accretion model for the postbreakup volcanic rocks, H.C. Larsen et al. (1998) estimated that the thinned wedge of the continent that constitutes the continent-to-ocean transition zone is ~45 km wide on the 63ºN transect and that Site 917 is situated on thinned continent 25 km from the outer edge (Fig. 8). The syn- and postbreakup volcanics were envisaged to have accumulated, fractionated, and extruded within the new oceanic rift zone at the outer edge of the continent. The rift zone was envisaged as standing above sea level so that the lavas flowed subaerially inland to their present positions of ~25 km from the rift. If this were so, then the magmas could not have become crustally contaminated. On the other hand, the young oceanic rift zone, just after breakup, must have been situated immediately adjacent to the continental edge. There are three possible ways in which the early oceanic magmas could have come in contact with continental crustal material.

1. The magma accumulation zone would still have extended beneath the continental wedge. Magma that failed to be focused into the rift zone could have been trapped in magma chambers within the wedge and been extruded up through it, thus bypassing the rift itself. This mechanism is expected to produce magmas more similar to those of the continental series than to those of the oceanic series. Magmas that bypass the center of a rift zone and are erupted on older crust are known from other areas including oceanic ridges, and they tend to be compositionally different from those being erupted at the rift axis. We do not believe that this mechanism gave rise to the contamination in the oceanic basalts.
2. The young oceanic rift zone could still have contained torn slivers and foundered blocks of continental material. This is a simple and straightforward explanation that fits the observations. The magmas could be contaminated or not, depending on the character of the sidewall in the local magma chambers. The slivers of continental material need not be so large as to show up on seismic profiles; however, the seismic profile given by H.C. Larsen et al. (1998) does show evidence that the continental edge was ragged and that a megasliver had formed (Fig. 8).
3. The magmas could have been emplaced laterally from the rift into the continent and extruded locally. The two dikes provide evidence that such a process did take place, in addition to the second possibility favored above. The dike from Site 917 sits in the strongly contaminated middle series, yet it is clearly oceanic in chemistry and is even uncontaminated. This strongly suggests lateral injection into the continent from the oceanic rift zone, and it also shows that this can take place without contamination of the magma. The dikelet in Unit 990-12 is clearly from another melt batch (Fig. 4) from a different mantle type (mid-ocean-ridge basalt [MORB], Saunders et al., Chap. 8, this volume) than the surrounding flows, and it would similarly have been intruded at a later stage in the development of the rift.

Lesher et al. (Chap. 12, this volume) interpret the oceanic flow of Unit 989-1 as erupted from a local center at the inner edge of the continent and not in the rift zone 30 km away. This also requires lateral injection of oceanic magma into the continental crust.

Oceanic basalts on a continental margin are rare but are also known from the Faeroe Islands. Here, Gariépy et al. (1983) and Hald and Waagstein (1991) described low-Ti basalts with oceanic chemistries very similar to that of the oceanic basalts described here. They form lava flows and are intruded as dikes within the lava plateau. The rift zone along the continent split runs west-east ~50 km north of the Faeroes (Smythe, 1983), and some of the dikes run west-east for 10-20 km and dip to the north (Hald and Waagstein, 1991). They probably fed lava flows. A similar situation can be envisaged for the Southeast Greenland margin.

The oceanic lava flows have MgO = 8.23%-6.27% (volatile free) and do not represent primary magmas, which would have had much higher MgO contents. Thy et al. (1998, based on experiments) and Fitton et al. (1998b, based on fine-grained aphyric high-MgO rocks) estimated that the primary magmas for the Site 917 upper series contained up to 18% MgO. Demant (1998) found olivine with up to Fo_92.5 in a picrite from the Site 917 upper series, and, if this is a true phenocryst, it similarly indicates a very magnesian magma. Assuming that the mantle temperature stayed constant from the melting of the primary magmas of the Site 917 upper series to those of the primary magmas of Site 990/915, the removal of the continental lid and the resulting continued melting to shallower levels at Site 990/915 must have led to larger degrees of melting than for Site 917 (Fram et al., 1998). We thus estimate that the primary magma for the oceanic basalts also had around 18% MgO, and we put 20% MgO as a possible upper limit (the melt will then be in equilibrium with olivine Fo_92.0 at the fayalite-magnetite-quartz [FMQ] oxygen buffer). We have calculated possible primary magmas by adding equilibrium olivine to the most magnesian of the oceanic basalts (Unit 990-7) in 0.5% increments, assuming an iron-magnesium distribution coefficient between olivine and melt of 0.30 (Roeder and Emslie, 1970) and all iron as FeO. Unit 990-7 is slightly crustally contaminated (Ba/Zr = 0.53, the next lowest in the Site 990 lavas), but its SiO₂ is among the lowest found and was used without any correction. The calculated primitive magma with 18% MgO is shown in Table 2.

Fractionation calculations were then carried out on the primitive magma compositions with the computer program COMAGMAT (Ariskin et al., 1993), assuming water-free compositions at the FMQ oxygen buffer. This program utilizes a free energy minimization algorithm to calculate mineral-melt equilibria and crystallization temperature at a given pressure and percent crystallized. It is based on experiments conducted at 1 atm pressure but is calibrated at pressures up to 10-12 kbar, and temperatures are reproduced to within 15º-20ºC (Ariskin et al., 1993). Calculations were carried out at a number of different pressures to study the effect of pressure on the results. The first part of the liquid line of descent is not pressure sensitive, but the later parts are. At all pressures, olivine first fractionates alone in amounts depending on the starting composition. At pressures 4 kbar, olivine is then joined or replaced by clinopyroxene. Considering the high Sc contents in the oceanic basalts (Fig. 6) and the high D_sc^cpx-liq (0.51: Ulmer, 1989; up to 2.15: Rocholl et al., 1996), the amount of fractionating clinopyroxene is strongly limited, and pressures <4 kbar are more realistic for the later parts of the liquid line of descent. At 3 kbar, olivine is joined on the liquidus simultaneously by plagioclase and clinopyroxene, giving rise to gabbro fractionation. Irrespective of the MgO content of the starting composition, this happens when the liquid has ~8.0% MgO (justifying the addition of olivine only to Unit 990-7 with 8.23% MgO in the calculation of the primary liquid). A pressure of 3 kbar corresponds to a depth of ~10 km and is a realistic estimate for the depth of the magma chambers in which the gabbro fractionation took place (Fig. 8). The preceding olivine fractionation would have taken place at similar and deeper levels, representing deep parts of the magma chambers and underlying magma conduits. Fram et al. (1998) modeled a pressure of 10 kbar for the top of the oceanic melting column, and this is thus a maximum pressure for the start of crystallization. The calculated proportions of fractionated phases are, however, insensitive to the starting pressure.

Table 2 shows the results of the COMAGMAT fractionation and temperature calculations starting from the primitive melt with 18% MgO (Table 2, column 1). Crystallization is started at a pressure of 7 kbar where the melt has a temperature of 1449°C, corresponding to a potential temperature of ~1400°C. The melt fractionates olivine during a pressure decrease (ascent of the melt) from 7 kbar to 3 kbar. After crystallization of 30 mol% olivine (Fo_91.2-82.1) the melt has ~8.0% MgO and a temperature of ~1200°C (Table 2, column 2). The calculated melt composition closely matches the measured composition of Unit 990-7 (Table 2, column 3). This is not coincidental since first adding equilibrium olivine to a composition and then removing it again (albeit with two different methods, a spreadsheet and a computer program) is bound to produce a match. The melt then crystallizes olivine, plagioclase, and clinopyroxene simultaneously in near-constant proportions, and an amount of 24% of this gabbro fractionation (Table 2, column 4) is required to reproduce the most evolved basalt, Unit 990-1 (Table 2, column 5). The modal composition (in vol%) of the gabbro cumulate is 8% olivine (Fo_81.9-73.6), 50% plagioclase (An_75.8-65.9), and 42% clinopyroxene (Fs_9.3-14.7).

The erupted oceanic magmas, with 8.23%-6.27% MgO, correspond very closely to those associated with the gabbro fractionation stage. It appears that the magmas had to fractionate almost to the point of three-phase cotectic relations before they were able to erupt. The average MgO content of the oceanic rocks from Sites 989, 990, 915, and 918 is 7.3%, corresponding to an average amount of gabbro fractionation of 14% of the primitive magma.

In conclusion, the COMAGMAT calculations show that the primitive magma with 18% MgO will fractionate into 30 mol% olivine cumulates, 14 mol% gabbro cumulates, and 56 mol% melt, which is intruded as dikes and erupted as lavas. If the magma is more magnesian, the proportion of olivine cumulates will increase (to 35 mol% for a melt with 20% MgO). The balance between the calculated amounts of gabbro cumulates and melt is strongly dependent on the average MgO content of the analyzed samples, but, on the other hand, the average of 7.3% ± 0.5% MgO (1) is fairly robust. The mode of the gabbro cumulates is also robust. What the calculations do not show, however, is the significant amount of melt that must be retained within the cumulates.

The results from the COMAGMAT calculations can be compared with the modeled crustal structure of H.C. Larsen et al. (1998), which is based on geophysical data and shown here in Figure 8. Outside the continental edge, the igneous part of the oceanic crust is constantly ~18 km thick (at 3-21-km depth) and consists of three units (3, 5, and 6 in Figure 8). H.C. Larsen et al. (1998) interpreted units 5 (4 km) and 6 (4.3 km) as respectively lighter and heavier plutonic residues and noted that the boundary between them is uncertain. Unit 3 (9.7 km) was interpreted as SDRS lava flows underlain by a zone of dikes. A depth of 9.7 km to the cumulates corresponds to a pressure of ~3 kbar, the pressure used in our calculations of the gabbro fractionation scheme.

Given a total igneous thickness of 18 km, our fractionation calculations set out above and in Table 2 translate into thicknesses of 5.4 km of olivine cumulates, 2.5 km of gabbro cumulates, and 10.1 km of dikes and lava flows, as set out in Table 3. The cumulate olivine (Fo_91-82) has a density around 3.35, and a gabbro with mineral mode and composition as given above will have an average density of 3.03 (olivine, 3.45; plagioclase, 2.72; clinopyroxene, 3.33). Any amount of melt retained in the cumulates will increase the thickness and decrease the densities of the cumulates and decrease the thickness of the layer of dikes and lava flows. While the total amount of calculated cumulates corresponds to that in the geophysical model, the proportions of light and heavy cumulates are different. However, velocity information from more recent crustal wide-angle seismic data along the same profile indicate that the amount of heavy cumulates in Figure 8 is significantly underestimated, while the amount of light cumulates is correspondingly overestimated. The thickness of the layer with dikes and lava flows is unchanged (T. Dahl-Jensen, pers. comm., 1997). The new data bring the geophysical model and the chemical composition derived model into broad agreement. The chemical data presented here thus support the modeling based on geophysical data and give an independent indication of the nature of the various layers in the oceanic crust.