PUBLICATIONS. Journal of Geophysical Research: Planets. Differentiation of the South Pole Aitken basin impact melt sheet: Implications

PUBLICATIONS Journal of Geophysical Research: Planets RESEARCH ARTICLE Key Points: SPA impact melt differentiation accommodates observed noritic surface materials SPA formed before LMO overturn to produce noritic material at the lunar surface Norite-bearing materials represent key samples for dating the age of SPA Supporting Information: Readme Table S1 Table S2 Text S1 Figure S1 Figure S2 Figure S3 Figure S4 Figure S5 Figure S6 Figure S7 Figure S8 Correspondence to: D. M. Hurwitz, hurwitz@lpi.usra.edu Citation: Hurwitz, D. M., and D. A. Kring (2014), Differentiation of the South Pole Aitken basin impact melt sheet: Implications for lunar exploration, J. Geophys. Res. Planets, 119, 1110 1133, doi:10.1002/ 2013JE004530. Received 13 SEP 2013 Accepted 30 APR 2014 Accepted article online 10 MAY 2014 Published online 5 JUN 2014 Differentiation of the South Pole Aitken basin impact melt sheet: Implications for lunar exploration Debra M. Hurwitz 1 and David A. Kring 1 1 Center for Lunar Science and Exploration, Lunar and Planetary Institute, Houston, Texas, USA Abstract We modeled the differentiation of the South Pole Aitken (SPA) impact melt sheet to determine whether noritic lithologies observed within SPA formed as a result of the impact. Results indicate differentiation of SPA impact melt can produce noritic layers that may accommodate observed surface compositions but only in specific scenarios. One of nine modeled impact melt compositions yielded layers of noritic materials that account for observations of noritic lithologies at depths of ~6 km. In this scenario, impact occurred before a hypothesized lunar magma ocean cumulate overturn. The 50 km deep melt sheet would have formed an insulating quenched layer at the surface before differentiating. The uppermost differentiated layers in this scenario have FeO and TiO 2 contents consistent with orbital observations if they were subsequently mixed with the uppermost quenched melt layer and with less FeO- and TiO 2 -enriched materials such as ejecta emplaced during younger impacts. These results verify that noritic lithologies observed within SPA could have formed as a direct result of the impact. Therefore, locations within SPA that contain noritic materials represent potential destinations for collecting samples that can be analyzed to determine the age of the SPA impact. Potential destinations include central peaks of Bhabha, Bose, Finsen, and Antoniadi craters, as well as walls of Leibnitz and Schrödinger basins. Additionally, potential remnants of the uppermost quenched melt may be preserved in gabbroic material exposed in Mafic Mound. Exploring and sampling these locations can constrain the absolute age of SPA, a task that ranks among the highest priorities in lunar science. 1. Introduction The existence of a large basin on the lunar farside was first predicted based on telescopic observations of huge mountains near the Moon s south pole [Hartmann and Kuiper, 1962; Wilhelms, 1987] and was later confirmed when those south polar mountains were linked to other mountains across the southern farside, forming a large ring-like structure [e.g., Howard et al., 1974; Schultz, 1976; Stuart-Alexander, 1978; Wilhelms et al., 1979; Wilhelms, 1987]. The full lateral and vertical dimensions of the basin were deduced from Clementine altimetry data [Zuber et al., 1994] and the diameter of the basin was later refined to 2050 2400 km as geochemical signatures of impact-derived lithologies based on Lunar Prospector observations were integrated with Clementine altimetry data [Garrick-Bethell and Zuber, 2009]. The South Pole Aitken (SPA) basin is generally recognized as the oldest and largest relict impact basin on the Moon. A radiometric age of the SPA basin is not known, however, and a mission to determine its age is one of the highest lunar science priorities [National Research Council (NRC), 2007, 2011]. Thus far, crater-counting techniques indicate SPA is pre-nectarian [Wilhelms, 1987] and recent modeling suggests that it may have been produced prior to 4.3 Ga [Morbidelli et al., 2012]. The SPA impact occurred before or possibly during the earliest stages of an epoch of increased impact activity known as the lunar cataclysm that is characterized by a peak in intensity at 3.9 4.0 Ga [e.g., Turner et al., 1973; Tera et al., 1974;Cohen et al., 2000]. Although the peak and termination of this epoch of enhanced impact activity are somewhat constrained by analyses of samples collected during the Apollo missions [e.g., Turner et al., 1973; Tera et al., 1974] and from meteorites [e.g., Cohen et al., 2000], it is currently definedbyonlythelatterthirdof the basin-forming impacts. It remains unclear whether the SPA impact and those that occurred between thetimeofthespaandnectarisimpactsarepartofthelunarcataclysm[ryder, 1990;Kring, 2003]. Determining the age of SPA has, therefore, been identified by the planetary science community as the second highest priority in lunar science [NRC, 2007] and the highest priority New Frontiers class lunar mission [NRC, 2011]. HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1110

Figure 1. Extent of materials enriched in Th and FeO within SPA from Lunar Prospector data. The outermost black circle represents the approximate boundary of SPA as defined by Garrick-Bethell and Zuber [2009]. The hatched area within the white border represents the extent of material with an enriched Th signature indicating Th contents of ~2 4 μgg 1 [Lawrence et al., 2000]. The central hatched area within the black border represents the extent of material with a highly enriched FeO signature indicating FeO contents of ~10 15 wt %, and the outer speckled area within the black border represents the extent of material with an enriched FeO content of 7 10 wt % [Lucey et al., 1998]. Areas enriched in TiO 2 generally form discontinuous patches that fall within the region enriched in FeO, and signatures indicate a TiO 2 content of ~0.5 1.5 wt % [Lucey et al., 1998]. The most direct and reliable approach for determining the age of the SPA basin is to analyze a sample from within the basin that formed as a result of the SPA impact event or that has an age that was reset during the SPA impact event [NRC, 2007, 2011; Duke et al., 2000;Duke, 2003]. Before such a mission can be planned and executed, however, landing sites that host suitable samples must be identified. Observations of SPA indicate that the basin floor is dominated by low-ca pyroxene- (i.e., orthopyroxene and/or pigeonite) bearing noritic materials, high-ca pyroxene-bearing gabbroic materials, and basaltic lithologies [Pieters et al., 2001] and is characterized by anomalously high concentrations of FeO, TiO 2 [Lucey et al., 1998], and Th [Lawrence et al., 2000] compared to the adjacent lunar highlands (Figure 1). Although the basaltic and gabbroic lithologies have generally been interpreted to represent products of later volcanic activity, thus having ages that do not reflect the age of SPA formation, the noritic lithologies have been interpreted to represent a melt sheet derived during the impact event from exposed lower crustal material [Pieters et al., 1997, 2001] or from a melted mixture of mantle and lower crustal materials [Lucey et al., 1998]. Hydrocode models of the SPA impact [Potter et al., 2012; Wieczorek et al., 2012], however, indicate the impact melt should be dominated by ultramafic mantle compositions. The hydrocode results can be reconciled with observations if the SPA impact melt sheet differentiated, producing noritic surface or near-surface materials [Morrison, 1998; Kring, 2012]. A preliminary study [Hurwitz and Kring, 2013] indicates that a noritic lithology is a feasible differentiation product of an ultramafic melt. In a parallel study that assumed a simplified initial impact melt composition, Vaughan and Head [2013a, 2013b] and Vaughan et al. [2013] also found that the differentiation of such an impact melt would indeed be expected to generate an upper layer of noritic material. Here we expand our preliminary results [Hurwitz and Kring, 2013] and complement the work by Vaughan and Head [2013a, 2013b] and Vaughan et al. [2013] with a model that begins with melt compositions that are specifically tailored to the SPA impact event. Because the age of SPA remains uncertain, we explore the consequences of an impact at different times during the evolution of the lunar interior, as summarized in section 2. The differentiation sequences of several potential impact melt compositions are then investigated using Petrolog3 [Danyushevsky and Plechov, 2011], as described in section 3. Results are analyzed and compared with observations of SPA in sections 4 and 5, where they are used to constrain the timing of SPA formation and to identify sites of interest for future lunar missions. 2. Defining the Initial Composition of the SPA Impact Melt Sheet 2.1. Lunar Structure Before the SPA Impact The discovery of anorthositic clasts in the regolith of the Apollo 11 landing site spawned the idea that the uppermost portions of the Moon were once completely molten, allowing plagioclase to crystallize and float to the surface of a Lunar Magma Ocean (LMO) [e.g., Wood et al., 1970; Smith et al., 1970]. The LMO differentiated HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1111

Table 1. Published Compositions Used To Calculate Initial Impact Melt Compositions Average Crustal Composition Initial Bulk Moon Compositions Residual LMO Liquid Compositions From Experiments Bulk LMO Cumulate Compositions From Experiments Korotev et al. [2003] Elardo et al. [2011] LPUM, P = 20 kbar, z = 420 km Elardo et al. [2011] TWM, P = 20 kbar, z = 420 km Rapp and Draper (personal communication, 2012) TWM, P = 18 kbar, z = 350 km Elardo et al. [2011] LPUM, P = 20 kbar, z = 420 km Elardo et al. [2011] TWM, P = 20 kbar, z = 420 km Rapp and Draper (personal communication, 2012) TWM, P = 18 kbar, z = 350 km LPUM [Longhi, 2006] TWM [Taylor, 1982] SiO2 44.4 46.1 46.43 46.27 48.15 36.54 41.5 42.1 44.96 TiO 2 0.31 0.17 1.34 0.57 0.27 0.09 0.02 0.01 0.22 Al 2 O 3 6.1 3.9 9.56 10.93 7.08 2.03 1.13 0.29 28.54 FeO* 10.9 7.6 12.58 12.03 10.24 11.45 8.95 6.03 4.01 MnO 0.15 0.13 0.27 0.18 0.16 0.18 0.11 0.08 0.06 MgO 32.7 38.3 18.89 20.45 27.98 44.33 47.5 51.0 5.31 CaO 4.6 3.2 9.42 8.53 5.23 0.21 0.35 0.19 16.42 Na 2 O 0.09 0.05 0.52 0.36 0.25 0.01 0 0 0.34 K2O 0.01 0.01 0.06 0.02 0.02 0 0 0 0.03 P 2 O 5 0.01 0.01 0.26 0 0 0.01 0 0 0.02 Cr2O3 0.61 0.5 0.68 0.62 0.60 5.14 0.41 0.27 0.09 Mg# 84 90 73 75 83 87 90 94 70 Density 3.26 3.18 3.1 3.08 3.05 3.29 3.27 3.26 2.58 as it cooled, forming ultramafic cumulatesat the base of the LMO and more evolved cumulates at the top. The compositions of the LMO products depend on the initial bulk composition of the Moon as well as the nature of the crystallization process. Two plausible bulk Moon compositions are an Al, Fe-enriched composition (relative to the terrestrial mantle), referred to as the Taylor Whole Moon (TWM) [Taylor, 1982; Taylor et al., 2006] model, and an Al, Fe-depleted composition (similar to that of the terrestrial mantle), referred to as the Lunar Primitive Upper Mantle (LPUM) [Longhi, 2006] model. The differentiation products of both bulk LMO compositions have been investigated experimentally to determine the crystallizing mineral species at decreasing pressures [Elardo et al., 2011; Rapp and Draper, 2012, 2013, 2014]. The experiments conducted by Elardo et al. [2011] investigated the solidification of both a TWM and a LPUM initial bulk composition, with the first 50% of solidification assumed to have occurred by equilibrium crystallization before shifting to fractional crystallization. Alternatively, the experiments conducted by Rapp and Draper [2012, 2013] investigated the solidification of a TWM initial bulk composition, with all solidification assumed to have occurred by fractional crystallization. The differentiation products identified in all three experimental cases, hereafter designated Rapp TWM, Elardo TWM, and Elardo LPUM, indicate that at least the lower 40% ( 460 km depth) of the LMO cumulates are dominated by monomineralic olivine. However, differentiation products at shallower depths vary between the three studies, where results from the Rapp TWM experiments produced more evolved minerals, such as clinopyroxene, plagioclase, and ilmenite, whereas the Elardo TWM and LPUM experiments produced only orthopyroxene and olivine in the lowest pressure experiments conducted [Elardo et al., 2011; Rapp and Draper, 2013]. The amount of plagioclase predicted to crystallize in the Rapp TWM case [Rapp and Draper, 2013] would have produced a crust with a thickness that is consistent with the average crustal thickness of 34 43 km that was recently interpreted from the Gravity HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1112

Figure 2. Schematics of the uppermost portion of the pre-spa Moon (a) before and (b) after LMO cumulate overturn. The pre- SPA Moon was characterized by an ~45 km thick lunar crust [Wieczorek et al., 2013] overlying an ultramafic lunar mantle. Before LMO overturn, the lunar mantle was characterized by basal layers dominated by olivine and orthopyroxene and by shallower layers dominated by clinopyroxene and materials enriched in incompatible elements. This initial sequence would have generated an unstable density profile in the LMO cumulates, thus generating Rayleigh-Taylor instabilities that drove LMO overturn [Hess and Parmentier, 1995]. LMO overturn occurred as dense materials enriched in incompatible elements formed diapirs that sank from the upper remnants of the LMO into the lower lunar mantle, dislodging hot, less dense materials in the lower LMO cumulates that in turn formed ascending diapirs [e.g., Hess and Parmentier, 1995; Solomatov, 2000; Elkins-Tanton et al., 2003; Parmentier, 2009]. Impact into each of these pre-spa Moon scenarios would produce impact melts of distinct compositions. Recovery and Interior Laboratory (GRAIL) data [Wieczorek et al., 2013]. It should be noted that the experiments by Elardo et al. [2011] only account for the lower 50% (to a depth of 360 km) of the LMO, whereas the experiments by Rapp and Draper [2013] progress to lower pressures (to a depth of ~65 km). Therefore, experiments investigating further LMO solidification by fractional crystallization using the residual liquid composition in Elardo et al. [2011] might also produce other crystallizing species such as clinopyroxene and plagioclase at shallower depths. The experimental liquid compositions at depths of ~350 km from the Rapp and Draper [2013] scenario and ~420 km from the Elardo et al. [2011] scenarios (Table 1) were used as inputs into Petrolog3, described in section 3, to complete crystallization of the LMO to shallower depths. In summary, the structure of the early, pre-spa Moon can be characterized by an outer plagioclase-rich crust with a thickness of ~45 km that overlies a mantle of differentiated LMO cumulates (Figure 2). The shallowest mantle is expected to have been composed of predominantly clinopyroxene and other materials enriched in elements that were incompatible during LMO crystallization. Orthopyroxene and olivine contents are expected to have increased with depth and the deepest mantle is expected to have been composed of predominantly monomineralic Mg-rich olivine. This structure represents the Moon at the end of LMO crystallization (e.g., Figure 2a). However, the predicted LMO crystallization sequence would have resulted in HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1113

an inverted density profile and, thus, a cumulate overturn event may have occurred to stabilize the inverted density structure of the lunar interior [e.g., Ringwood and Kesson, 1976;Spera, 1992;Elkins-Tanton et al., 2003, 2011]. During this overturn process, the composition of deep cumulates dominated by olivine would have been mixed with the composition of shallow cumulates dominated by clinopyroxene (e.g., Figure 2b). This cumulate mixture represents a post-cumulate overturn Moon. Both pre-overturn and post-overturn initial lunar structures are considered in our modeling because the uncertainty in the age of SPA does not constrain whether this basin formed before or after LMO cumulate overturn. These different conditions resulted in six distinct pre-spa scenarios: both a pre-overturn and a post-overturn scenario applied to each of the three considered initial bulk Moon compositions of Rapp TWM, Elardo TWM, and Elardo LPUM. These six scenarios define the target for the impact event that produced the SPA basin. 2.2. Formation of the Basin The SPA basin is a 2400 km by 2050 km impact structure centered at 53, 191 E on the farside of the Moon [Garrick-Bethell and Zuber, 2009] with an average rim-to-floor depth of 13 km [Spudis et al., 1994]. Recent hydrocode models indicate SPA formed when a projectile with a diameter of either 170 km [Potter et al., 2012] or 200 km [Wieczorek et al., 2012] impacted the Moon at a velocity of 10 km s 1 [Potter et al., 2012] or 15 km s 1 [Wieczorek et al., 2012], respectively. The Wieczorek et al. [2012] model assumed a more probable oblique impact angle of 45 [Gilbert, 1893; Shoemaker, 1962] as opposed to the idealized vertical impact angle assumed in the Potter et al. [2012] model. However, the Wieczorek et al. [2012] study was focused on the distribution of ejecta and associated magnetic anomalies outside of SPA and did not specify characteristics of the impact melt sheet within the basin, such as the source depth of melted material and the depth of the resulting melt sheet. In contrast, the alternative model by Potter et al. [2012] analyzed the formation of the basin and should have produced an azimuthally averaged melt sheet even though it did not explicitly model a 45 impact. Because the Potteretal.[2012] model adequately defines an azimuthally averaged melt sheet formed during the SPA impact event, we used those results to characterize the impact melt within the 200 km wide, 50 km deep pool that we investigate in the current study. It is important to explain that this impact melt volume is suitable for our analyses, regardless of the impact angle, because it is firmly constrained by the basin diameter. Although a 45 impact angle will generate less melt than a vertical impact with the same impact conditions (i.e., same projectile diameter, velocity) [Pierazzo and Melosh, 2000; Abramov et al., 2012], this shallower impact will also generate a smaller transient crater and, thus, basin diameter than the vertical impact. Therefore, to produce the observed basin dimensions, a larger projectile colliding at a faster impact velocity is required. This is evident in the results of an oblique impact generated by Wieczorek et al. [2012], where a larger projectile with a faster velocity is required to produce the same basin as the vertical impact scenario modeled by Potter et al. [2012]. This larger, faster, oblique impact produces approximately the same volume of melt (e.g., ~7.7 10 7 km 3 melt volume) [Kring et al.,2012]astheslower, smaller, vertical impact (e.g., 6.7 10 7 km 3 melt volume) [Potter et al., 2012]. Hydrocode model results [Potter et al., 2012] indicate the upper 100 km of the Moon was excavated as the transient crater (radius of 410 km) developed. The lunar interior was completely melted to a depth of 250 km (corresponding to a shock pressure of >80 GPa) and incompletely melted (>50%, corresponding to a shock pressure of >60 GPa) to a depth of 400 km (Figure 3a) [Potter et al., 2012, also personal communication, 2013]. The liquid resulting from incomplete melting of the target would have the same composition as the bulk target rock, because the energy generated during an impact with a velocity of 10 km s 1 would have resulted in a propagating shock wave that induced melting of all silicate minerals within the target rock [Stöffler, 1971; Ahrens and O Keefe, 1972; Kring and Boynton, 1992], even if the entire rock was incompletely melted. Therefore, the bulk composition of the resulting melt is assumed to reflect the composition of the bulk target cumulates once the average composition of the crust (Table 1) [Korotev et al., 2003] was removed to account for excavation of the crust upon impact, in contrast to the scenario considered by Vaughan and Head [2013a, 2013b] and Vaughan et al. [2013] in which no crust was removed. Melt sourced from both 250 km and 400 km was considered in each pre-lmo overturn scenario to determine how depth of melting would affect the products of the differentiated impact melt sheet (increasing the considered initial melt compositions from six to nine; Table 2). As indicated above, the volume of melt produced by the impact was approximately 7 10 7 km 3 [Potter et al., 2012;Kring et al., 2012],though~20%[Potter et al., 2012, also personal communication, 2013; Vaughan et al., 2013] to ~25% or 45% [Cintala and Grieve, 1998;Kring, 2012;Kring et al., 2012] of this material HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1114

Figure 3. Schematics of the uppermost portion of the Moon (a) during and (b) after the SPA-forming impact event. During the SPA-forming impact event, the projectile formed a transient crater with a diameter of 820 km and a depth of 390 km [Potter et al., 2012]. This model assumed a vertical impact event though the impact most likely occurred at 45 as shown in Figure 2; see section 2.2 in the text for a more detailed justification for using model results from an idealized vertical impact. Hydrocode models indicate that the upper 100 km of the Moon was excavated and that the upper 250 km of the lunar interior was completely melted whereas the upper 400 km of the lunar interior was incompletely (>50%) melted. After the impact, the impact melt formed a melt sheet with a depth of 50 km and a radius of 200 km within the transient crater of a 2050 2400 km basin with a rim-to-floor depth of 13 km [Spudis et al., 1994;Garrick-Bethell and Zuber, 2009; Potter et al., 2012]. The impact resulted in the excavation of crustal material as far as 630 km away from the basin center [Pieters et al.,2001;potteretal., 2012]. would have been ejected from the transient crater. After the basin formed, melt remaining within the transient crater pooled to form an impact melt sheet with a radius of ~200 km and a depth of 50 km (Figure 3b) [Potter et al., 2012]. This superheated mantle-derived melt was initially relatively homogeneous because of the thorough mixing typical of impact melt [e.g., Bostock, 1969;Currie, 1970, 1972; Grieve, 1975; Grieve et al., 1977; Table 2. Nine Scenarios Considered To Calculate Initial Impact Melt Composition Model Case Description Pre-overturn Pre-overturn Post-overturn # a b c Rapp TWM 1 An Al, Fe-enriched bulk Moon, LMO solidification 100% by fractional crystallization Elardo TWM 2 An Al, Fe-enriched bulk Moon, LMO solidification by 50% equilibrium and 50% fractional crystallization Elardo LPUM 3 An Al, Fe-depleted bulk Moon, LMO solidification by 50% equilibrium and 50% fractional crystallization 100% shallow cumulates melted to a depth of 250 km (completely melted whole rock) 100% shallow cumulates melted to a depth of 400 km (incompletely melted whole rock) 50% deep cumulates + 50% shallow cumulates, melted to a depth of 250 km (completely melted whole rock) HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1115

Grieve and Floran, 1978] and because the viscosities of the melts produced from the target sequence would have been similar (within a factor of ~2 between, for example, a high-ca pyroxene-bearing melt [Dingwell, 1989] and a peridotite melt [Liebske et al., 2005]). This is in contrast to an emulsion between crustal-derived mafic and felsic melts (viscosity contrast of 10 15%) hypothesized to have been produced at Sudbury [Zieg and Marsh, 2005]. It should be noted that the particle content and temperature of the melt have significant effects on the viscosity of the melt [e.g., Warren et al., 1996; Petford, 2009; Öhman and Kring, 2012], where an ~30% increase in clast content can increase the viscosity by a factor of 5 and an ~400 C decrease in melt temperature can increase the viscosity by 2 orders of magnitude [e.g., Öhman and Kring, 2012]. Effects of contamination on impact melt differentiation are explored in section 5.1. 2.3. Observed Lithologies in the Center of SPA Basin In addition to the noritic, gabbroic, and basaltic materials described in section 1, central SPA is characterized by a unique compositional signature (Figure 1) defined by elevated FeO (7 14 wt %) [Lucey et al., 1998], TiO 2 (0.5 1.5 wt %) [Lucey et al., 1998], and Th (2 4 ppm) [Lawrence et al., 2000] abundances compared to the adjacent highlands terrain [Jolliff et al., 2000]. Non-mare concentrations of high-ca-bearing pyroxene have been identified as possible remnants of a more mafic melt sheet or of plutons or dikes that intruded after SPA formation and before being exposed by younger craters [Pieters et al., 2001; Moriarty et al., 2013a, 2013b]. These younger craters can be used to probe the basin interior to determine the structure of the near-surface lithologies within SPA basin. One such crater, Bhabha (64 km in diameter), is fortuitously located in the center of the SPA basin. Noritic materials observed in crystalline form in the central peak of Bhabha crater [Nakamura et al., 2009] have been interpreted to represent subsurface lithologies that were excavated from a depth of ~6 km (calculated from Cintala and Grieve [1998]). Thus, noritic materials might extend from the surface to a depth of 6 km or more. Interpretations of GRAIL data indicate that central SPA has a crustal thickness of ~12.5 km, though this estimate is based on a global model and, therefore, may have local uncertainties up to a few kilometers [Wieczorek et al., 2013]. This crustal thickness estimate may, therefore, indicate the presence of an interface of contrasting densities at a depth ~13 km below the surface of central SPA. Additionally, anorthositic materials have been identified within craters that formed just inside the rim of SPA. The innermost occurrence of observed anorthosite is 630 km from the center of the basin [Pieters et al., 2001], well outside of the expected 410 km radial extent of the transient crater [Potter et al., 2012]. Together, these observations indicate that a reasonable model of impact melt differentiation must produce an upper layer of noritic materials ~6 km thick and possibly as thick as ~13 km. The successful model must also result in the absence of an upper layer of anorthosite. 2.4. Summary of Initial SPA Impact Melt Compositions Nine possible initial compositions (Table 3) representing distinct formation scenarios (Table 2) of the 50 km deep impact melt sheet were derived from experimental results of LMO solidification [Elardo et al., 2011; Rapp and Draper, 2012, 2013] and from hydrocode models of the SPA impact event [Potter et al., 2012]. The nine defined compositions were developed from three cumulate sequences generated during LMO solidification that each evolved via three distinct scenarios. The three LMO cumulate sequences considered include (1) an initially Al, Fe-enriched (TWM) bulk Moon that solidified entirely by fractional crystallization (Rapp TWM), (2) an initially Al, Fe-enriched (TWM) bulk Moon that solidified initially by equilibrium crystallization followed by fractional crystallization (Elardo TWM), and (3) an initially Al, Fe-depleted (LPUM) bulk Moon that solidified initially by equilibrium crystallization followed by fractional crystallization (Elardo LPUM). The three scenarios applied to each LMO cumulate sequence include (a) a pre-cumulate overturn Moon completely melted by the impact to a depth of 250 km, (b) a pre-cumulate overturn Moon completely melted by the impact to a depth of 250 km and incompletely melted (>50%) to a depth of 400 km, and (c) a post-cumulate overturn Moon with deep cumulates mixed with shallow cumulates, completely melted by the impact to a depth of 250 km. In all nine cases, a 45 km crust of average feldspathic composition was removed due to impact excavation during the basin-forming event, as illustrated by hydrocode simulations [Potter et al., 2012; Wieczorek et al., 2012]. Any uncertainties of that excavation are discussed in section 5.1. HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1116

Table 3. Calculated Initial Impact Melt Compositions Rapp TWM Melt at 250 km Rapp TWM Melt at 350 km Rapp TWM PostOT at 250 km Elardo TWM Melt at 250 km Elardo TWM Melt at 400 km Elardo TWM PostOT at 250 km Elardo LPUM Melt at 250 km Elardo LPUM Melt at 400 km Elardo LPUM PostOT at 250 km 1a 1b 1c 2a 2b 2c 3a 3b 3c SiO 2 49.3 46.6 41.7 49.1 47.0 44.7 53.9 48.9 47.4 TiO 2 2.21 1.51 1.05 0.99 0.64 0.46 0.47 0.29 0.20 Al 2 O 3 9.4 6.7 3.2 13.2 8.9 3.7 7.5 4.4 0.9 FeO* 14.9 13.9 13.6 14.4 13.3 12.1 13.3 11.3 9.6 MnO 0.34 0.30 0.26 0.23 0.20 0.17 0.22 0.18 0.15 MgO 9.9 20.9 31.8 8.9 21.3 34.1 16.8 30.0 39.5 CaO 12.5 8.4 4.5 12.5 7.8 4.2 6.4 3.9 1.5 Na 2 O 0.83 0.54 0.36 0.62 0.38 0.26 0.41 0.25 0.16 K 2 O 0.08 0.06 0.04 0.03 0.02 0 0.03 0.02 0 P 2 O 5 0.45 0.29 0.21 0 0 0 0 0 0 Cr 2 O 3 0.11 0.77 3.26 0.06 0.43 0.28 0.90 0.69 0.58 Mg# 54 73 81 52 74 83 69 83 88 Density 3.22 3.27 3.33 3.16 3.23 3.3 3.2 3.27 3.31 3. Modeling the Evolution of the SPA Impact Melt Sheet 3.1. Introduction to Petrolog3 The Petrolog3 code [Danyushevsky and Plechov, 2011] was used to investigate the evolution of the SPA melt sheet. Petrolog3 is a software program that models fractional, equilibrium, and semifractional crystallization using an algorithm that allows the user to select from available published mineral-melt equilibrium models for each crystallizing species of interest. The code incorporates 46 models for eight different anhydrous minerals in mafic and andesitic systems, including olivine (ol), plagioclase (plag), clinopyroxene (cpx), pigeonite (pig), orthopyroxene (opx), spinel (sp), ilmenite (ilm), and magnetite (mag). The user can select a model for each mineral individually, increasing the ability of the model to simulate the evolution of complex melt compositions. Alternative models were also considered for this study, including MELTS [e.g., Ghiorso and Sack, 1995] and MAGPOX [Longhi, 1991]. However, neither MELTS nor MAGPOX encompasses the compositions relevant in this study, specifically, shock melt derived from shallow LMO cumulates. MAGPOX, for example, was created for models of lunar mare basalt petrogenesis with olivine on the liquidus [Longhi, 1991, also personal communication, 2012]. Because olivine is not currently observed at the surface of the SPA interior, olivine was expected to come off the liquidus over the course of differentiation and, thus, Petrolog3 is the principal tool used for this study. Because MAGFOX/MAGPOX was designed for certain lunar scenarios, we also used FXMO, a combined MAGFOX/MAGPOX model [Longhi, 1991, also personal communication, 2012], for model verification and model output comparisons where possible. To run Petrolog3, an initial melt composition was first input into the model, then crystallization models were selected for each desired mineral, the oxidation state and pressure gradient were defined, and the mode of crystallization was selected. The crystallization models are based on experimentally derived data; those selected for this study include Beattie [1993] for ol, Ariskin et al. [1993]for plag, Ariskin et al. [1986]for cpx, Ariskin et al. [1993] for pig, Beattie [1993] for opx, and either Nielsen [1985] or Ariskin and Nikolaev [1996] for sp where appropriate (mag was not included in this study). These models were chosen because they best accounted for the pressure, temperature, and redox conditions (where available) expected in the SPA melt sheet. The oxidation state was defined as Fe W 1 to represent the oxygen-poor conditions present on the Moon [Papike et al., 1998]. The pressure gradient was defined in five 25% increments, with pressure decreasing as melt content decreased from 100% to 0%. The initial pressure was defined to be equivalent to the desired initial depth of the melt, with a depth of 50 km corresponding to a pressure of 2.77 kbar (calculated from Elkins-Tanton et al. [2011]). Finally, the mode of crystallization was set to be either 0%, representing equilibrium crystallization, or 100%, representing fractional crystallization, for each mineral of interest. Intermediate percentages could be selected to represent semifractional crystallization, but the current study investigated the end-member cases to determine how the mode of crystallization affected results. Results are presented in section 4 from models that simulate 100% fractional crystallization; these results are then compared to those from models that simulate first equilibrium HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1117

Figure 4. Model verification 1, solidification of the LMO. These graphs compare the (a) temperature, (b) FeO, (c) TiO 2, and (d) MgO contents predicted by both the Petrolog3 (gray lines) and FXMO (crosses) models to the temperature and compositions observed in experiments conducted by Elardo et al. [2011]. Although the Petrolog3 model overestimated the extent of crystallization at a given temperature, the model accurately predicted the FeO, TiO 2, and MgO contents expected during crystallization. Results were consistent for the three oxygen fugacities tested, including IW + 2, IW + 1, and IW 1; comparisons were made for these three oxygen fugacities because of experimental uncertainties [Elardo et al., 2011]. In comparison, the FXMO model precisely predicted the temperature at which 55% crystallization had occurred, but because the model was not isobaric, this temperature occurred at a much lower pressure (3.4 kbar) than observed in the model. The majority of the major oxides predicted by FXMO matched experimental results, with the exception of FeO (+3 wt %) and MnO (not shown, +0.5 wt %). However, the different pressure conditions suggest that these oxides should not be equivalent, as different phases are stable at different pressures. The Petrolog3 model is favored because the results from this model matched experimental compositions for all major oxides at the same pressure. crystallization then fractional crystallization in section 5.2. At each calculation step (set at the default of 0.01% crystallization), the model noted the magma and melt oxide compositions, the compositions of the crystallizing minerals, the bulk cumulate and individual cumulate mineral compositions, and the temperature, pressure, density, viscosity, and the cumulative weight percentages of each mineral that crystallized. The model then used the output conditions from the previous step as the input into the next calculation step, and the model continued iteratively until the melt was completely crystallized or until the MgO content in the output melt composition reached 0 wt %, forcing the model to stop running before the melt completely solidified. The FXMO model also worked iteratively but it employed different crystallization models that have generally been effective in predicting solidification of hypersthene-normative basalts [e.g., Longhi, 1991]. Similar input and output information was generated for the FXMO model and results from each model were compared to determine which model was optimal for modeling the impact melt compositions desired. To verify the veracity of the model output, two tests were conducted. In the first test, the models were run with starting conditions comparable to those used in the LPUM LMO solidification experiments (e.g., same HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1118

a b Figure 5. Model verification 2, liquid line of descent (LLD) for the impact melt of scenario 2b. The evolution of the liquid in scenario 2b (Table 2) is plotted on (a) an Ol-Plag-Sil {Wo} ternary and (b) an Ol-Sil-Wo {Plag} ternary. Results are shown from both Petrolog3 (closed diamonds) and FXMO (open diamonds) for comparison; Petrolog3 results are shown for each 0.1% crystallization step while FXMO results are only shown when a new phase begins to crystallize. Phase boundaries on the {Wo} ternary were calculated (solid lines) or estimated (dashed lines) from Longhi [1991]; phase boundaries were not included on the {Plag} ternary because the boundaries were not available for the specific melt composition considered. Phase boundaries in Figure 5a shift as the liquid composition evolves (gray arrows); the numbers within the enlarged symbols represent the Mg# of the liquid at that stage. (a) Both models predict a similar LLD when projected from Wo, with ol crystallizing first, followed by low-ca pyroxene, then plagioclase. (b) The LLD predicted by Petrolog3 follows typical phase boundaries [e.g., Longhi, 1991], with the downward turn in the LLD approximately following the ol + plag pig + plag opx + plag peritectic as it shifts with the evolving liquid. In contrast, the LLD predicted by FXMO follows a divergent path. initial melt composition, constant P = 10 kbar) conducted by Elardo et al. [2011]. Although the Petrolog3 model overpredicted the percent crystallization at any given temperature compared to the experiments, the modeled oxide weight percentages matched the experimental compositions (Figure 4) and, thus, we conclude that the Petrolog3 model accurately predicted the crystallization sequence. The correlation between model and experimental results improved as pressure decreased, indicating the model could be used to investigate the shallow SPA impact melt sheet scenarios. The FXMO model precisely predicted the temperature at which 55% crystallization had occurred (Figure 4), but it produced the relevant assemblage of oxide compositions at a much lower (3.39 kbar) pressure than the experiments (10 kbar). Moreover, the FXMO model produces a liquid with 3 wt % more FeO than observed in experimental results. These comparisons suggest that the Petrolog3 model, which matched experimental results very closely for all major oxides at the same pressure, is the most reliable for our study of SPA. In the second verification test, liquid lines of descent (LLD) were calculated to investigate the evolution of the melt composition resulting from impact into a pre-overturn, equilibrium and fractionally crystallized, Fe, Al-enriched target (Elardo TWM) melted to a depth of 400 km (Figure 5, scenario 2b). Compositions were projected onto two ternaries, the Ol-Plag-Sil ternary projected from Wollastonite (Figure 5a, {Wo}) and the Ol-Sil-Wo ternary projected from Plag (Figure 5b, {Plag}). The phase boundaries shown on the {Wo} ternary diagram are either calculated (solid lines) or estimated (dashed lines) from Longhi [1991]. These boundaries shift as the liquid composition evolves; the initial pressure that we are considering for these models is low enough (<2.8 kbar) that the effects on phase boundary location due to decreasing pressure are insignificant. Phase boundaries are not included in the {Plag} ternary, because existing projections [e.g., Longhi, 1991], created largely to evaluate mare basalt petrogenesis, are for melt compositions with lower Na content (NAB = 0.05) [Longhi, 1991] and slightly higher K content (NAB = 0.01) relative to the impact melt compositions considered in this study (NAB = 0.9, NOR = 0.0). The LLDs projected from Wo follow very similar paths in both the Petrolog3 (solid diamonds) and FXMO (open diamonds) model results (note that Petrolog3 results are shown for every 1% crystallization, while FXMO results can only be shown when a new phase crystallizes). However, when these results are projected from Plag (Figure 5b), the two models predict different trends in liquid composition. The Petrolog3 results appear to follow typical phase boundaries fairly accurately [e.g., Longhi, 1991], with the downward portion of the trend approximately tracing the phase boundary as the opx + plag and pig + plag fields diminish. HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1119

Figure 6. Lithology compositions modeled for scenario 1a. The modeled compositions of the crystallizing portion of the melt sheet in scenario 1a (Table 2) are plotted on two ternaries, a px-ol-plag ternary and a opx-cpx-plag ternary. The initial melt in this scenario had a composition that falls in the gabbroic norite field (gray circles). The first crystal to form was ol, followed by pig with an inverted composition dominated by the opx component (80%), then cpx, forming a basal layer of dunite overlain by layers of orthopyroxenite and clinopyroxenite. Plagioclase then began to co-crystallize with cpx, forming a layer of gabbro. Pigeonite was the final crystal to form and, as pig cooled slowly, it is expected to invert to its respective opx and cpx phases. These inverted compositions are plotted to indicate the lithologies expected to be observed at the end of the solidification process (see sections 1 and 3.2 in the text for details). As the opx component became more dominant in the mineral composition, the lithology shifted from gabbro to noritic gabbro, then finally to gabbroic norite. The melt was 86% crystallized when the crystalline component passed into the gabbroic norite field; the final ~10% crystallization may yield anomalous melt compositions due to increasing model uncertainty. Additional crystals that formed but are not included on these ternaries included spinel, which generally crystallized at great depths, and ilmenite, which generally crystallized at shallow depths. In contrast, the results from the FXMO model take a very different path. When ol comes off the liquidus, pig briefly crystallizes before opx begins to crystallize (the two symbols just evident above the Petrolog3 results). Clinopyroxene then begins to crystallize, at which point ol begins to crystallize again, even though the liquid composition has progressed into the aug + plag field. Because the Petrolog3 model more closely corresponds to the expected phase boundaries, it again appears to be the most reliable one for the current work. 3.2. Interpreting Lithologies From Model Outputs Models were run for each of the nine initial melt compositions defined in section 2 and in Table 3. Outputs of melt composition, cumulative weight percentages of each crystallizing mineral, Mg#, and pressure were recorded for each calculation step in each scenario. The initial and final melt compositions in each scenario were estimated by calculating the CIPW norms [e.g., Hollocher, 2013] and the resulting ol, opx, cpx, and plag contents were normalized and projected onto two ternary diagrams that define lunar lithologies, including the px-ol-plag and the opx-cpx-plag ternaries (Figures 6 and S1 S8 in the supporting information). It should be noted that the CIPW normative calculations used do not account for Fe metal, though this component likely represents a minor constituent in the composition. The relative amounts of each mineral that crystallized at each step were calculated from the model outputs and the resulting weight percentages of ol, plag, opx, pig, and cpx were renormalized and projected on the same two ternaries as the initial and final melt compositions. Although Petrolog3 considered pig as a distinct pyroxene phase, pig is generally unstable at low temperatures and pressures and, thus, will exsolve when cooled slowly to form an opx crystal with cpx lamellae with widths dependent on the original pig composition. Inverted pigeonite has been identified in terrestrial basaltic magmas [e.g., Poldervaart and Hess, 1951] as well as in lunar samples (e.g., samples 15459, 15475 [Takeda et al., 1975], 62237 [Dymek et al., 1975], 76255 [Takeda and Miyamoto, 1977], 76255 [Warner et al., 1976], and 79215, [Bickel et al., 1976]). Pig has also been distinguished from opx and cpx in laboratory spectral analyses [e.g., Sunshine and Pieters, 1993; Klima et al., 2011]. To properly represent the proportion of opx and cpx in rocks exposed on the lunar surface after they were exsolved from pig, the relative abundances of opx (0% Wo or wollastonite component) and cpx (50% Wo) were calculated from the Wo component of the modeled pig. These exsolved opx and cpx abundances were added to the primary opx and cpx abundances and these total abundances were plotted (Figures 6 and S1 S8) to determine the lithology that would be observed after the solidification of the melt sheet. The resulting trends in composition were used to determine the lithologies that formed as the melt sheet solidified (for lithology oxide compositions, see Table S1). The lithologies determined from each ternary are HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1120

Figure 7. Lithology sequences modeled for all nine initial melt scenarios considered. The results for the solidification by 100% fractional crystallization of the nine melt scenarios considered (Table 2) are shown. Depths were calculated from modeled pressures, based on the pressure-depth profile of the Moon determined by Elkins-Tanton et al. [2011]. In all scenarios in which pig was predicted, the inverted composition of pig was calculated and the relative proportions of opx and cpx were plotted. Results indicate that a basal layer of dunite was expected in all cases, but the thickness of this layer increased as the ol concentration in the target increased. Scenarios 1a, 3a, and 3b each yield an ~5 km thick layer of noritic materials at the top of the melt sheet, satisfying (within uncertainty) the observed noritic materials beneath the surface and the observed chemical signatures in the SPA basin interior. However, scenarios 3a and 3b result in final melt compositions with high silica contents not observed on the lunar surface and, thus, are not considered as viable scenarios. The scenarios that generate significant (>4 km thick) upper layers of noritic material include all pre-lmo overturn scenarios, suggesting that SPA formed before LMO cumulate overturn occurred. For comparisons to results from models that simulate an initial phase of equilibrium crystallization followed by fractional crystallization, see section 5.2. presented in Figure 7, with lithologies organized in the order of crystallization. Mineral assemblages with single crystallizing phases coincide with specific lithologies: Monomineralic olivine forms dunite, monomineralic opx forms orthopyroxenite, monomineralic cpx forms clinopyroxenite, and monomineralic plag forms anorthosite. Mineral assemblages of multiple crystallizing phases form different lithologies (the most significant phase is listed first): cpx + plag forms gabbro, cpx + opx + plag forms noritic gabbro, opx + cpx + plag forms gabbroic norite, and opx + plag forms norite. The purpose of these lithology sequences is to illustrate the mineralogy that might be observed currently from orbit. The pressures from the model output were used to calculate the corresponding depths, according to the published lunar pressure-depth relationship [Elkins-Tanton et al., 2011], and these depths were used to determine the thickness of each identified lithology. The density of each lithology was calculated by first HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1121

multiplying the density of each mineral [Barthelmy, 2010] by the proportion of that mineral in the given lithology and then adjusting for porosity as a function of the depth of the given lithology (see the supporting information and Table S2 for more details on the porosity and density calculations). The sequence of lithologies predicted by each model was compared with observations to determine which scenario(s) best accounted for the features observed within SPA basin. 4. Results The analysis outlined in section 3.2 is described in detail here for one of the nine modeled scenarios, specifically the case in which the projectile, upon impact, melted a pre-overturn, fractionally crystallized, Al, Fe-enriched bulk Moon (Rapp TWM) to a depth of 250 km (scenario 1a). Results are then compared with the other eight scenarios. In all cases, a thin layer of quenched melt with a composition equivalent to that of the initial bulk impact melt is expected to have formed at the surface of the melt sheet, effectively insulating the rest of the melt sheet. The relative abundance of crystallizing minerals affects the rock type produced (Figure 7). Monomineralic lithologies formed from ol are referred to as dunite, from cpx as clinopyroxenite, and from opx as orthopyroxenite. Lithologies that contain cpx and plag are referred to as gabbro; lithologies that contain mostly cpx and plag but also some opx are referred to as noritic gabbro; lithologies that contain mostly opx and plag but also some cpx are referred to as gabbroic norite; and lithologies that contain opx and plag are referred to as norite. Pigeonite forms in most scenarios and, because slow-cooling pig is unstable on the Moon, pig is assumed to exsolve completely into opx and cpx, with relative abundances of these pyroxenes dependent on the Ca content of the pig. The inverted opx and cpx abundances were added to the modeled opx and cpx abundances and were then plotted to determine the lithology that would be observed after solidification. The modeled crystallization sequence for scenario 1a initiated with the crystallization of monomineralic ol (ρ =3.49gcm 3 at Fo 80 )[Barthelmy, 2010], followed by crystallization of pig dominated by the low-ca component (ρ =3.35gcm 3 ), then by crystallization of high-ca-bearing cpx (ρ = 3.4 g cm 3 ), then by cocrystallization of cpx, opx and plag (ρ =2.69gcm 3 ), and finishing with co-crystallization of pig and plag. The lowermost layers in the solidified melt sheet for this scenario included a 3 km thick layer of dunite (ρ =3.48 3.49 g cm 3 ) overlain by a 4 km thick layer of orthopyroxenite (ρ =3.36gcm 3 ) and an 8 km thick layer of clinopyroxenite (ρ =3.4 3.41 g cm 3 ). Plagioclase began co-crystallizing with cpx at a depth of 35 km, and pig (~90% opx component) began crystallizing as well at a depth of 26 km. The relative abundances of these minerals (67 75% cpx, 0 8% pig, ~25% plag) generated a 22 km thick layer of gabbro (ρ = 3.18 3.4 g cm 3 ). At a depth of 13 km, the relative abundance of crystallizing pig (~90% opx component) increased relative to cpx (37 67% cpx, 8 38% pig, ~25% plag), resulting in the formation of an 8 km thick layer of noritic gabbro (ρ = 3.35 3.41 g cm 3 ). Pigeonite continued to crystallize in increasing abundances relative to cpx, forming a 5 km thick layer of gabbroic norite (8 37% cpx, 38 67% pig, ~25% plag; ρ = 2.97 3.09 g cm 3 ) at the top of the differentiating melt sheet (ρ = 3.32 3.35 g cm 3 ). The inverted pig composition continued to be dominated by the opx component until the uppermost kilometer of the melt sheet, at which point the cpx component began to dominate (>50%). It should be noted that Cr sp (Usp 0.18 0.5 Chr 0.47 0.8 Mt.0.02 0.03 [Sack and Ghiorso, 1991a, 1991b]; ρ = 4.48 4.53 g cm 3 ) was also predicted to have crystallized in minor amounts (<1 wt %) from the base of the melt sheet to a depth of ~26 km, and ilm (ρ = 4.72 g cm 3 ) was predicted to have crystallized from depths of 13 km (<1 wt %) to 3 km (~4 wt %). The results for scenario 1a indicate that the noritic-dominated lithology gabbroic norite formed the upper 5 km of the solidified melt sheet. It is worth noting that strongly noritic lithologies do not form after gabbroic lithologies during the crystallization of typical basaltic systems. The evolution of gabbroic noritic units after noritic gabbroic units predicted by Petrolog3 probably reflects the unusual composition of the bulk SPA melt, which was produced by wholesale melting of upper mantle lithologies rather than as a result of equilibrium partial melting processes associated with typical magmatic systems. It is also important to acknowledge that uncertainties accumulate in the iterative modeling of crystallization of the system, resulting in potentially misleading compositions produced after ~90% crystallization. The analysis described for scenario 1a was also conducted for the other eight scenarios considered. Resulting vertical sections of all differentiated melt sheets are shown in Figure 7. Of the eight additional scenarios, six (1b, 1c, 2b, 2c, 3a, and 3b) followed a similar crystallization sequence as described for scenario 1a. These HURWITZ AND KRING 2014. American Geophysical Union. All Rights Reserved. 1122