Interactive comment on “ Picroilmenites in Yakutian kimberlites : variations and genetic models ” by I . V . Ashchepkov

I found this manuscript very difficult to evaluate. It certainly contains an immense quantity of data and from the list of references and the depth of the discussion, it represents a profound study of an important phase in kimberlites. The problem is that, try as I might, I was not able to obtain a clear picture of exactly what the authors were trying to say. I will explain the problems in the following text and make many suggestions on what must be done to improve the manuscript. The list is long, but I want to emphasize that the paper has great potential and should certainly be published, but only after major revision and total rewriting. Only after this is done will it be possible (for me at least) to properly evaluate the scientific content of the manuscript and assess the authors’ model for ilmenite formation.

Cr-poor ilmenite megacrysts (group 2) trace the hightemperature path of protokimberlites developed as result of fractional crystallization and wall rock assimilation during the creation of the feeder systems prior to the main kimberlite eruption.Inflections in ilmenite compositional trends probably reflect the mantle layering and pulsing melt intrusion during melt migration within the channels.Group 2 ilmenites have inclined REE enriched patterns (10-100)/PM with La / Yb n ∼ 10-25, similar to those derived from kimberlites, with high-field-strength elements (HFSE) peaks (typical megacrysts).A series of similar patterns results from polybaric Assimilation + fractional crystallization (AFC) crystallization of protokimberlite melts which also precipitated sulfides (Pb < 1) and mixed with partial melts from garnet peridotites.Relatively low-Ti ilmenites with high-Cr content (group 3) probably crystallized in the metasomatic front under the rising protokimberlite source and represent the product of crystallization of segregated partial melts from metasomatic rocks.Cr-rich ilmenites are typical of veins and veinlets in peridotites crystallized from highly contaminated magma intruded into wall rocks in different levels within the mantle columns.Ilmenites which have the highest trace element contents (1000/PM) have REE patterns similar to those of perovskites.
Low Cr contents suggest relatively closed system fractionation which occurred from the base of the lithosphere up to the garnet-spinel transition, according to monomineral thermobarometry for Mir and Dachnaya pipes.Restricted trends were detected for ilmenites from Udachnaya and most

Analytic methods and data set
Ilmenite grains from seven kimberlite fields were analyzed in the Institute of Geology and Mineralogy (Novosibirsk) using CamebaxMicro and Jeol Superprobe electron microprobes.All analyses were made in wavelength-dispersive (WDS) mode with accelerating voltage 15 kV and a focused beam current of 20 nA.A set of natural reference minerals (natural picroilmenite GF −55) and synthetic glasses for calibration (Lavrentyev et al., 1987) and reduction procedure using "Karat program" (Lavrent'ev and Usova, 1994) was applied to obtain the analyses.The relative standard deviation did not exceed 1.5 %; the precision was close to 0.02-0.015% for minor elements.Xenolith analyses include data for 180 veined peridotite and websterite xenoliths from the Sytykanskaya pipe (Fig. 2).
Trace element analyses for > 90 ilmenites from Yakutian kimberlite concentrates and xenoliths and ∼ 1200 for other minerals (garnet (Gar), clinopyroxenes (Cpx), phlogopites, etc.) were obtained in the Analytic Centre of IGM on a Finnigan Element ICPMS with a UV LaserProbe using international reference standards NIST 612-614.
The major and trace element compositions and the program for the calculations of pressure-temperature (PT) conditions are available in Supplement, files 1-3.surround ilmenite grains in Upper Muna pipes (Kostrovitsky and Bruin, 2004).

Comparison of variation trends for different regions
Short continuous trends (55 to 45 wt % TiO 2 ) for ilmenites from the Daldyn pipes (Amshinsky and Pokhilenko, 1983;Rodionov et al., 1988;Pokhilenko et al., 1999;Ashchepkov et al., 2003;Kostrovitsky et al., 2003;Alymova et al., 2004) with negative trends for FeO and positive for MgO are determined by crystal chemistry.Plots of TiO 2 vs. Al 2 O 3 , NiO, Cr 2 O 3 and V 2 O 5 consist of several separate arrays with varying inclinations (Fig. 5a).Increase in Fe and incompatible elements accompanies decrease of MgO, TiO 2 and NiO controlled by olivine fractionation.Decrease of Al 2 O 3 content probably relates to contamination and precipitation of garnet.The very high Cr 2 O 3 in the MgO-TiO 2 -rich part corresponds to metasomatites from the lithosphere base.Three levels from 0.5 to 2 wt % Cr 2 O 3 for ilmenites from the middle part in Zarnitsa (Amshinsky and Pokhilenko, 1983) (Fig. 1a, Supplement 4) and nearby pipes are regulated by different extents of contamination of parental melts.The similarities of the diagram for other pipes in the cluster (Alymova et al., 2004) (Fig. 5a) suggest the same source.Shorter trends were found for ilmenites from the satellite bodies of Zarnitsa pipe.In the heavy mineral concentrate of the micaceous type II kimberlite from the Bukovinskaya pipe (Kostrovitsky and Bruin, 2004), Ti chromites are more common than ilmenites.Ilmenite trends from kimberlite pipes in the Alakit field show higher Cr 2 O 3 (Fig. 4, Supplement 4, Fig. 1b).For ilmenites from Sytykanskaya pipe, the Cr content increases towards the low-Ti part of the diagram is explained by contamination of hydrous parental melts in peridotites accompanied by abundant metasomatism beneath Sytykanskaya and Komsomolskaya pipes (Fig. 4b).The subtrends with 2-6 wt % Cr 2 O 3 ranges for Ilm are found within some veined metasomatic Phl in peridotite xenoliths from Sytykanskaya.For the Aykhal pipe, the restricted (46-53 wt % TiO 2 ) ilmenite trend may be subdivided into 7-8 groups.NiO and Al 2 O 3 contents decrease rapidly with TiO 2 .

Ilmenites from Kimberlite Xenoliths
Ilmenite trends from Malo-Botuobinsky (Mir and Dachnaya) pipes (Ashchepkov et al., 2010(Ashchepkov et al., , 2013a) (Fig. 4, Supplement 4, Fig. 1c), are long (31-50 wt % TiO 2 ) and were formed by continuous fractionation, as suggested for South Africa (Moore et al., 1992;Griffin et al., 1997) and Canada (Wyatt et al., 2004).Ti-rich and Cr-poor ilmenites from pyroxenite xenoliths (Alymova, 2006) at the right part of diagram correspond to the SCLM base according to thermobarometry.Low Ni and Cr contents are typical.Cr contents increase at the beginning and end and in the middle part of the crystallization trend.Continuous rise of Al 2 O 3 suggests lack of garnet among the main precipitating phases.Ilmenites from Amakinskaya and International'naya pipes reveal shorter TiO 2 trends divided into four separate intervals.
In the Nakyn field (Lapin et al., 2007) (Fig. 5a), picroilmenites are abundant only in the tuffisitic kimberlites from the Nyurbinskaya pipe showing 55 to 48 wt % TiO 2 interval, while ilmenites found in placer deposits reveal a wider range with a decrease of NiO and MgO accompanied by an increase in FeO and V 2 O 5 which may be explained by olivine fractionation (Fig. 5a).Metasomatic increase in Cr is accompanied by a decrease in Al.
Picroilmenites from the Deimos pipe in Upper Muna field (Fig. 5b) show 3 levels of Cr 2 O 3 enrichment which are higher in left part of the trend.Longer and more discrete trends were determined for ilmenites from Zapolyarnaya pipe.Ilmenites from the Poiskovaya pipe are higher in Al 2 O 3 .In the other Upper Muna kimberlite pipes, ilmenites are rare.

Monomineral ilmenite thermobarometry
The empirical dependence of the geikilite component in ilmenites on pressure was used to indicate the diamond grade of kimberlites (Sobolev, 1977;1980;Sobolev and Yefimova, 2000).Ilmenite has two types of positions in structure.The distance between Fe (or Mg) and O is highly dependent from the temperature by 1 order of magnitude more than the distance between Ti and O (Wechsrel et al., 1984).
The FeTiO 3 -MgTiO 3 internal exchange accounting for the influence of the other components was used for Ol-Ilm thermometry (Taylor et al., 1998), which is cited here: all components are given in formula units (f.u.); To obtain agreement with Cpx thermometry (Nimis and Taylor, 1998), the following corrections were introduced The monomineralic formulation takes into account the forsterite content where Fe#Ilm = Fe / (Fe + Mg).
A further correction: Fe#Ol = 0.1027ln(Fe#Ol(Ilm)) + 0.365(see Fig. 8c) Evaluation of the single grain thermobarometry for olivine (De Hoog et al., 2010) shows that, if the concentration of an element in one of the minerals is negligible compared to the variation in the other mineral, the equation simplifies to where C i is concentration of trace elements.The −lnC i is positive for minor components.But for major components forming the crystal structure as Ti the logarithmic equation may be simplified.
Close linear correlations are found between the cell sizes and experimental pressures (Fig. 7b) (Reynard et al., 1996;Karki et al., 2000) (Fig. 8).Rutile has a significantly lower size cell unit (Liu, 1975) than ilmenite.Many ilmenites have exsolutions of rutile in natural assemblages (Ashchepkov et al., 2001) and experimental products (Bataleva et al., 2012).Admixture of rutile suggests significant reduction in linear correlation with pressure in this exchange.The simplified equation is V/V 0 Karki atal., 2000Reynard et al., 1996 Figure 6.Correlations between the compression of unit cell volumes V/Vo in ilmenite structure vs pressure (Karki et al., 2000;Reynard et al., 1996).
xenoliths with ilmenites varying in Cr and xenocrysts with zonal cores in rims.Pressure estimates were produced with clinopyroxene thermobarometry (Ashchepkov, 2003;Ashchepkov et al., 2010;Nimis and Taylor, 2000) for 110 ilmenite-bearing associations (Reimers, 1994;Alymova, 2006;Ashchepkov, 2006b;Ashchepkov et al., 2010).Ilmenites often are not in complete equilibrium with all minerals in associations.We selected mainly analyses from closely located Ilm and Cpx.Best agreement between pressure estimates based on Cpx and Ilm was obtained for the peridotite xenoliths from Sytykanskaya pipe.The approximation allowed us to calibrate the barometer as follows: and yielded further corrections P 1 = P 0 + (60 − P 0 )/6.2 and next P 2 = P 1 +(35.4−FeO)/0.9+(0.20−Fe/(Fe+Mg))×0.1×P 1 (all components are in wt % and P in kbar).This yields a better agreement than the previous version (Ashchepkov et al., 2010).Nearly linear correlation of pressure between clinopyroxene and ilmenite estimates (using the same temperatures) was found within 7.0-3.5GPa interval (Fig. 7b).This equation is calibrated for the associations crystallized in Mg-rich protokimberlite melts and peridotite rocks.
The equation derived from the correlation with pressure values based on clinopyroxenes using the database for xenoliths from Mir and Udachnaya pipes (Alymova, 2006), data for xenoliths from the Sytykanskaya pipe (Reimers, 1994;Reimers et al., 1998;Ashchepkov et al., 2014, submitted) and the Dalnyaya pipe (Rodionov et al., 1988(Rodionov et al., , 1991)), yielded good agreement for the pressure using the same temperatures .The correlations between the methods of mineral thermobarometry and PT diagrams based on pyroxenes and ilmenites show good agreement (Fig. 9b).

Regularities of the distribution of ilmenite associations in the mantle sections worldwide
The conditions obtained from pyroxene thermobarometry for ilmenite-bearing associations shown on the PT diagram (Fig. 8) reveal a similarity with those produced by ilmenite PT equations.The highest pressure (7.0-6.0GPa) estimates for the HT pyroxenites and eclogites (Boyd and Nixon, 1973) trace the convective branch of the geotherm corresponding to deformed garnet peridotites (Boyd et al., 1997).PT conditions for hybrid pyroxenites and HT metasomatic and MARID-type xenoliths (Dawson and Smith, 1977;Hamilton et al., 1998;Gregoire et al., 2002) plot within the transition between the SCLM base and lithospheric mantle (5.5 to 6.5 GPa), Commonly PT clusters for Ilm-bearing metasomatic veins and megacrysts in the lower part of lithospheric mantle columns coincide in pressure but differ in temperature and Fe# of coexisting Ol (Figs. 9-11).
New data for the Dalnyaya pipe based mostly on mantle xenoliths (> 110) (Ashchepkov et al., 2013) and for concentrates as well as previous data for pyroxenites (Rodionov et al., 1988(Rodionov et al., , 1991;;Alymova, 2006) give the deepest and highest temperature plot of the mantle section in Yakutia (Fig. 9b).Estimates for porphyroclastic and sheared peridotites in 7-5 GPa interval are overlapped by the PT values for ilmenites.Ilm megapyroxenites sometimes with garnet (Rodionov et al., 1988) are lower temperature.Metasomatic peridotites with Cr-bearing ilmenites close in Fe# to common garnets (0.8-0.11) are located in three SCLM levels and correspond to the 45 m Wm −2 geotherm.
In the PT diagram for the SCLM beneath the Mir pipe (Fig. 12a), the PT values for ilmenite megacrysts also trace a HT 39-45 m Wm −2 from 7.0 to 2.0 GPa.The inflection of geotherm called the convective branch (Boyd, 1973;Boyd et al., 1997) at 7.0 to 6.0 GPa is traced by ilmenites which overprint the PT conditions for the Fe-enriched minerals.Nearer 4.0 GPa, the geotherm splits and ilmenites trace the HT PT advective arrays close to PT conditions of clinopyroxenes from pyroxenites.The long stepped HT PT path refer to megacrysts geotherm (Eggler and Mccallum, 1976)  but several estimates in the lower part for the pyroxenites are close to the 37 m Wm −2 low temperature (LT) geotherm.Symplectites with pyroxenes are found in the lower part of the convective branch.
In the PFe#Ol diagram (Fig. 10a), the values for garnets, chromites and ilmenites form fluctuating lines, possibly reflecting primary layering of the SCLM formed by paleosubduction slabs (Pearson et al., 2003).The clots of ilmenite PT estimates may trace the boundaries of the primary layering.There are at least three PFe#Ol arrays.The lower Fe# values refer to the contact zones of megacrystalline associations while the middle one corresponds to the common megacrysts.The most Fe-rich array is for ilmenite megapyroxenites.Cr-bearing metasomatic associations are found in the 3-5 GPa interval.
PT estimates for minerals from concentrates and xenoliths from the Sytykanskaya pipe (Fig. 10b) can be used to reconstruct the SCLM section in the northern part of the Alakit region.The PT paths for ilmenite megacrysts create a stepped trend divided into three parts: two broad clusters at the base of the SCLM (7.0-6.5 GPa) and in the 5.0-3.0GPa pressure interval, and a narrower chain connecting them.A broad scatter of the calculated Fe#Ol values is typical for the upper part near 4.0 GPa where the dense cluster suggests a large propor-tion of metasomatic veins (Reimers et al., 1998;Ashchepkov et al., 2010).The increase in Fe# and Cr for the ilmenites from the bottom to the upper part of the mantle section is typical for most of mantle sections beneath this kimberlite and most pipes in Yakutia especially in the Alakit field (Ashchepkov et al., 2010).The abrupt increase of Cr and decrease of Fe# near 3.5 GPa and above in the mantle section corresponds to ilmenite crystallization mainly within metasomatic veins forming dense stockworks as seen in xenoliths.The ilmenite-bearing xenoliths with the veinlets of Ilm and Cpx are widespread, mainly in the middle part of the mantle column.
PT values for ilmenites from the three kimberlite phases found in the Yubileinaya pipe quarry, i.e., autholithic kimberlite (Kurszlaukis et al., 2008;Zinchuck and Koptil, 2008), the Ozernaya pipe, Ottorzhenets kimberlite body and the Aprelskaya pipe represent four stages of kimberlite activity (Fig. 11a, b; Supplement 4, Fig. 3a, b).Ilmenites in heavy mineral fractions from the autholithic kimberlite show the deepest (∼ 7.5 GPa) and relatively HT conditions.The highest temperature values at the base of the SCLM correspond to a geothermal gradient of 43 m Wm −2 with deviations to the 35 m Wm −2 conductive geotherm.The continuous increase of Cr with decreasing pressure suggests continuous PT estimates for the Ozernaya mantle column (Fig. 11b) also reveal that the base of the SCLM is situated at 7.5 GPa.A wide scatter of temperature estimates between the 35 and 45 m Wm −2 geotherms which is close to those of the Feenriched Cpx corresponds to the veined metasomatities.The ilmenite PT clusters are located at 7.5-5.5GPa and around 4.5 GPa.

Trace element geochemistry of ilmenite megacrysts
The LAM ICP-MS study of ilmenite compositions (Ashchepkov et al., 2008b) reveals three main groups .The first one (group 1) has low rare earth element (REE) concentrations and flatter, spoon-like or complex curved REE patterns.They have high peaks for Ta and Nb and slightly lower in Zr and Hf, similar to those for ilmenites from South Africa (Dawson et al., 2001;Gregoire et al., 2003).The Pb peaks are their characteristic feature.For the second group (gr2), inclined linear REE (La / Yb n ∼ 10-15) patterns similar for those of alkali basaltic melts and temperate peaks of high-field-strength elements (HFSE) are typical.The last group (group 3) is highly enriched in REE and other trace elements revealing smoother trace element diagrams with HFSE peaks.The multi-element diagrams are generally similar for different pipes, showing peaks in TaNb and ZrHf and fluctuating Pb anomalies which are positive due to melting processes and negative for fractionation.
In Udachnaya group 1 ilmenites are abundant (Fig. 12a), with compositions of REE (0.1/C1) forming a series of patterns from spoon-like to flatter slightly REE-enriched patterns.Very high TaNb peaks are in contrast to low REE concentrations.Ilmenites from the Dalnyaya pipe have the lowest REE content and close to W-shaped REE patterns (Fig. 12b).Small negative Y anomalies were found only in ilmenites from Dalnyaya pipes.Ilmenites from the Festivalnaya pipe (Fig. 12c) also show inclined La −− Yb n patterns, possibly produced by partial melting.The high Hf peak relative to Zr, as well as enrichment of Ba and Th found in some of these compositions, are unusual for ilmenites from the other pipes.
In Alakite, field ilmenites from the Aykhal pipe (group 1) show slightly inclined REE patterns for the most Ti-rich (deepest) samples (Fig. 13a).Ilmenites from the Komsomolskaya pipe have W-shaped inclined REE patterns with an inflection at Gd and a peak at Pb (Fig. 13b).The mildly enriched group ∼ 100 relative to PM primitive mantle   (McDonough and Sun, 1995) (100/PM) of picroilmenites (group 2) also shows trace element patterns typical for group 2 (Fig. 15b, c, d).Ilmenites from Yubileynaya have REE patterns (Fig. 12c) that are almost S-shaped, similar to those determined for chromites.

Dal'nyaya
In Malo-Botuobinsky field trace element patterns for ilmenites from Mir (group 1) ∼ 0.1 PM are flat in MHREE but vary in LREE patterns increasing together with most of the trace elements.Typical group 2 patterns are found for two ilmenites from the middle part of the mantle column.The rim of one ilmenite from Mir reveals an abrupt enrichment in trace elements resembling that of group 3 samples, while the inner core is closer to group 2 in composition (Fig. 13a).Ilmenites from the International'naya pipe show a REE pattern similar to garnets of pyroxenitic type with a hump in MHREE.Trace element diagrams reveal Y minima typical for low-Ca garnets (Stachel et al., 2008) and Siberian diamonds (Afanasiev et al., 2005) that show signs of depletion of the melt source.One ilmenite has a small minimum at Eu (Fig. 13b).
In the Nakynsky region (Fig. 13c) most of the highpressure ilmenite (group 1) show flat REE patterns with gentle minima in MHREE and fluctuations in LREE especially in Ce.The enriched (gr2) varieties demonstrate inclined linear trace element patterns similar to those of alkali basalts.In Prianabarie all three populations of analyzed ilmenites show different trace element patterns in the different pipes (Fig. 13d).Ilmenites from the Trudovaya pipe (Fig. 13d) demonstrate different types of trace element patterns like those from Aykhal pipe.Ilmenites having the REE level near 1/PM display slightly inclined distributions with Pb peaks and a mild enrichment in ZrHf and TaNb.Ilmenites with REE levels ∼ 100/C1 (for La) demonstrate an inclination of the REE patterns close to primitive 1 % melts produced in garnet-facies peridotites (McDonough and Sun, 1995).The International'naya  most enriched ilmenites show enrichment up to 1000/C1 for La have La / Yb n ∼ 15.They show no HfZr anomalies.

Model of origin based on PT estimates
The suggestion about the origin of ilmenites at the base of the SCLM (Boyd and Nixon, 1973) from protokimberlites was supported by Gurney et al. (1979) for ilmenites from the Monastery pipe based mainly on the PT estimates for Ilm-Opx symplectites and by the contact relationships with sheared peridotites (Moore and Lock, 2002).A similar model (Moore et al., 1992) suggested that they are products of extensive fractionation in pegmatite-like bodies, which is supported by evidence of fluid crystallization (Kopylova et al., 2009) for megacrystalline ilmenites.Direct calculations for clinopyroxenes coexisting with ilmenites (Dawson and Reid, 1970;Haggerty et al., 1979;Haggerty, 1995;Gurney et al., 1979Gurney et al., , 1998) ) and orthopyroxenes from ilmenite intergrowths and from quenched nodules (Rowlinson and Dawson, 1979) including clinopyroxenes in type II diamond (Moore, 2009) give a pressure interval from 6.0 to 4.0 GPa.Estimation of a HT geotherm based on megacrysts (Eggler and Mccallum, 1976) was followed by the suggestion that megacrysts result from polybaric magma fractionation (Neal and Davidson, 1989;Eggler and Mccallum, 1976) during crystallization of fluid-rich magma.Ilmenite trends from many localities show the gap in the trends between the Mg-rich and Fe-rich varieties, for example in Angola (Robles-Cruz et al., 2009;Ashchepkov et al., 2012).The highest temperature metasomatites show depths of origin near 4.0-7.5 GPa.This coincides well with the conditions determined by most thermobarometric methods and agrees well with the SCLM thickness determined by seismic methods (O'Reilly et al., 2009) near 270 km.Inflections of the geotherms and the major field of the megacryst associations start from 7 to 6.5 GPa and refer to the minimum on the peridotite solidus in the lower part of mantle column (redox melting) (Tappe et al., 2007) and olivine creep under the presence of volatiles (Peslier et al., 2012) and magma fracturing.

Cs
they may be formed as a result of reaction between the protokimberlite melts that formed megacrysts and mantle eclogites.But mantle metasomatites which are close in pressures to the clots of the megacrysts PT points reflect refertilization events in peridotites.

Model of fractionation and melting based on trace elements
Geochemistry of calculated parental melts (Figs. 14 and 15) determined with the clinopyroxene (Hart and Dunn, 1993) and Ilm distribution (partition) coefficients KD (Zack and Brumm, 1998) shows that a significant amount of picroilmenites are the products of differentiation of HFSE-rich kimberlite-like melts.Assimilation + fractional crystallization (AFC) processes (Neal and Davidson, 1989) are indicated by the Cr increase with decreasing pressures during crystallization on the wall of large channels and intermediate chambers accompanied by peridotite assimilation (DePaolo, 1981).Cr-poor ilmenites (Moore and Belousova, 2005) are found mainly in the lower part of mantle section but in some pipes, such as  in Mir and in Anabar fields, they form the main part of the ilmenite trends.The fraction of dissolved and crystallized phases regulated the Cr content of the parental melt.Dissolution of chromite may result in the rapid increase in Cr of the parental liquid.It is likely that protokimberlites assimilated low-degree hydrous partial melts of metasomatic associations.Increase in Al may be formed by garnet dissolution.Megacrystalline ilmenite aggregates near the contacts (Fig. 2a) with the peridotites contain up to 2 % Cr 2 O 3 .The V 2 O 5 content of ilmenites, reflecting variations of oxygen fugacity conditions (V 2 O 5 ) of parental magma (Righter et al., 2006;Canil and Fedortchouk, 2000) and wallrock reactions, generally correlates negatively with MgO.Contamination during growth of veins and at their contacts causes variations in the minor components Cr 2 O 3 , NiO and MnO.Admixture of MnO sometimes is rather significant (Patchen et al., 1997).It is common in ilmenites associated with diamonds (Kaminsky et al., 2009) and in some eclogites (De Stefano et al., 2009) it may indicate presence of subducted material in the mantle.
Calculated parental melts for ilmenites using KD for trace elements for ilmenites (Zack and Brumm, 1998;Klemme et al., 2006) show very variable TRE patterns.The compositions of the kimberlite melt for the Yubileynaya pipe (Kostrovitsky et al., 2007) as well as carbonatites from West Greenland (Tappe et al., 2009) and Oldoinyo Lengai (Bell and Simonetti, 1996) are also shown on the diagram.
It is evident that group 1 ilmenites are lower in trace elements except for HFSE than their host kimberlites.The parental melts for group 2 ilmenites were more evolved and produce the patterns close to carbonatites, associated with ailikites (Tappe et al., 2007) and kimberlites.Mostly TRE patterns of group 2 ilmenites have TaNb and ZHf contents close to kimberlitic values or slightly higher.These common ilmenite megacrysts are similar in isotopic features to their host kimberlites (Noyes, 2000;Nowell et al., 2004).HFSE and other incompatible elements.
The melts that crystallized group 3 ilmenites were more enriched than carbonatite from Oldoinyo Lengai (Bell and Simonetti, 1996).This suggests a high degree of fractionation of parental (protokimberlite) melts responsible for megacryst formation (Griffin et al., 1997;Moore et al., 1992) or splitting of the protokimberlites into carbonatite and silicate melts (Safonov et al., 2007).It is possible that they originated from the melt that dissolved some very enriched TRE metasomatites, which includes ilmenites, rutiles and possibly perovskites.Some group 3 ilmenites definitely have admixtures of perovskites which often substitute for ilmenites in grain boundaries (Amishinsky and Pokhilenko, 1984).Some of them, like those from the Anabar fields, reveal significant ZrHf minima.
We calculated the parental melts for the coexisting Cpx and Ilm from a low-Cr pyroxenite xenolith (UAS1300) from the Udachnaya pipe and an Ilm-Cpx vein in a garnet peridotite xenolith (Stk135) (Fig. 15).The parental melts from the first sample are close to equilibrium in REE but the melt calculated from ilmenite is enriched in U-Ta-Nb.The Cpx's parental melt shows a straight line trace element pattern close to that of the kimberlitic with small peaks in Zr and Sr.The patterns determined for the melts in equilibrium with the metasomatic vein Stk135 show more equilibration but the magma in equilibrium with ilmenite was richer in all HFSE and Ba.This shows that ilmenites crystallized later than Cpx from the melt that was more enriched in HFSE (Zr, Hf, Nb and Ta).
The high inclination of the REE patterns may be produced both by melting and fractionation processes which were modeled using combined partition coefficients for Ol, Ilm, Cpx, Opx and Gar.We used the partition coefficients from Bedard (2006) for silicates, and KDs for ilmenites (Zack and Brumm, 1998;Klemme et al., 2006).The patterns in Supplement 4, Fig. 7a, referring to melting to the rather enriched peridotite Ilm-Ol-Gar websterite, show rather low realistic TRE patterns typical for the metasomatic associations.Melting of the depleted association shows highly variable patterns more highly dependent on the melting degree.The fractionation models where Ol and Ilm are the main precipitating phases give rather simple conformable patterns (Supplement 4, Fig. 7b).The higher HFSE peaks correspond to significant Ol fractionation (Supplement 4, Fig. 7c, d).
We also checked the hypothesis of the cognate crystallization trends in Ta n -La n diagrams.Calculations of fractionation using olivine (Ol) (90 %) and ilmenite (Ilm) (10 %) as the main crystallizing phases show enrichment by 2 orders of magnitude during nearly complete crystallization of the protokimberlite or carbonatitic melts.Each fractionation trend is restricted by the line with enrichment of 1.3 log units (Supplement 4, Fig. 8).Hence the ilmenites with extreme enrichment to 1000/PM and more should have another mechanism of HFSE enrichment.It could be remelting of ilmeniterutile-bearing associations very enriched in minerals with high TRE contents (apatites, perovskites).Perovskites (found in rims and intergranular spaces) may be responsible for the very high concentration not only HFSE and REE but also Th, U and LILE.Another mechanism is the immiscible carbonatite melt concentrating TRE.
The trends of the major components (Figs.3-5) reveal separate lines of enrichment in the TiO 2 vs. Cr 2 O 3 diagram.Probably they reflect repeated pulses of melt within the magmatic feeders.This would have been accompanied by partial remelting of previously formed associations (Ashchepkov and Andre, 2002).This model, with melting of earlier cumulates and metasomatic rocks, explains the rapid enrichment in TiO 2 because a simple fractionation model and an origin from the kimberlites could never explain these phenomena.The most realistic mechanism for high enrichment of the incompatible elements including REE and especially HFSE is low-degree partial melting of the contacts with metasomatized associations or the metasomatic front.Mixing of such partial melts with the new portion of protokimberlite melt will cause a strong increase in the trace elements content of the parental megacrystalline liquids.
Contaminated Cr-bearing megacryst varieties deviate in isotopic diagrams.Most ilmenites parental melts (protokimberlites) are higher in TRE than kimberlites.They were the most evolved and enriched in volatiles portions of melts which derived from the original magma.This agrees with the formation of the megacryst from protokimberlite melts which formed the pre-eruption melt feeder systems and evolved separately from the major magma portions.To receive the positive buoyancy protokimberlite melts should be highly volatile rich and essentially carbonatitic (Russell et al., 2012).
For the development of magmatic system in the mantle before the kimberlite eruption, we suggest a three-stage model (Mitchell, 1986;Le Roex et al., 2003).In the initial stage, ultramafic plume-derived melts containing a significant amount of eclogitic material (Sobolev et al., 2009) intruded the carbonated and hydrated base of the SCLM (Tappe et al., 2007(Tappe et al., , 2009(Tappe et al., , 2011)).This produced HT depletion of the peridotite and formation of chromites and ilmenites as precipitates in the feeder channels and surrounding wall rocks (rarely sub-calcic garnets may also have grown within and adjacent to the feeders).Partial melting of the metasomatic pyroxenitic aureole enriched in incompatible elements resulted in a high concentration of carbonates, H 2 O, silica and HFSE in the secondary hybrid melts which were formed during the next pulse of plume-derived melt.
Contamination and fractionation of secondary alkaline ultramafic magma caused separation into immiscible liquids: silicate, carbonatite, sulfide (Irvine, 1976) and probably oxide (Clarke and Mackay, 1990;Hurai et al., 1998) during cooling.Ilmenites may also have been formed by immiscible oxide melts in the final stages of crystallization (Hurai et al., 1998).Evidence for such melts is seen in xenoliths from the Sytykanskaya pipe where monomineral ilmenite aggregates cut large grains of garnets and other minerals in peridotites.
Rising protokimberlite melts formed due to melting of the enriched metasomatized mantle are similar to kimberlites in REE but have higher incompatible elements including HFSE.Their intrusion during formation of the feeder system produced Ti-rich metasomatism in wall rocks near contacts and heating near the feeder channels.In the channels AFC created the megacrystalline ilmenites which show similar trace element patterns.The assimilated material was the partial melts from the wall rocks which contain melted chromites, thus increasing the Cr content.According to thermobarometry the fractionation trends commonly stop at the ∼ 4.0 GPa boundary, forming the veins and veinlets that represent the metasomatic front.Further pulses of melt intrusion caused remelting of metasomatites and high enrichment in TiO 2 .
Step by step melt intrusion with contamination and partial dissolution of wall-rock resulted in continuous enrichment in Cr and TiO 2 and formation of ilmenite megacrysts which show a stepped Cr increase such as that found for ilmenites from Zarnitsa, Yubileynaya and other pipes.Three different Cr trends are found for ilmenites from the Mir pipe.
Possibly contamination and enrichment in water resulted in the formation of the phlogopite-ilmenite veinlets in the peridotites.
The enrichment in TRE may be also the result of incorporation of low-degree melts (∼ 0.01) of hydrous metasomatic rocks in walls of feeders.Fluctuations of Hf, Ba, HFSE, Y, Pb and Sr show that ilmenite crystallization, at least in the last low-temperature stages, was accompanied by growth/dissolution of zircon, rutile, phlogopite, perovskite and apatite (Kalfoun et al., 2002).
During formation of individual pipes, the process of protokimberlite differentiation in the magmatic conduits and chambers may have been a combination of these three simple models.
Variations in Fe at the same level of TiO 2 (Schulze et al., 1995;Nikolenko and Afanasiev, 2008;Robles-Cruz et al., 2009) possibly reflect a multi-stage origin of the ilmenite crystallization trends and reaction with the walls of magmatic conduits.

Significance of the ilmenite-bearing association for diamond formation
The PT conditions for the ilmenites and diamond inclusions especially chromites are often very close and trace heated geotherms and convective branches (Logvinova and Ashchepkov, 2008;Ashchepkov et al., 2010).Even diamondbearing eclogites very often reveal irregular heated conditions associated with metasomatism.In addition the carbonate crystals which always occur in the diamonds in fluids (Logvinova et al., 2008) probably were derived from the protokimberlites which, according to TRE reconstructions, reveal the carbonatite signatures at least at the graphitediamond boundary (Dobosi and Kurat , 2002;Rege et al., 2008).So the most realistic scheme of diamond growth is crystallization near the developing magmatic system during formation of the channels.In these conditions, the reaction of relatively oxidized CO 2 -rich fluid and melt with reduced peridotite and eclogite was accompanied by crystallization of newly formed diamonds that coated older microdiamonds.The temperatures and f O 2 of the ilmenites are slightly higher and their associations and inclusions in diamonds are rare because they reflect different conditions.Ilmenite megacrysts were formed in oxidizing conditions at the internal contacts of feeders and in outer contacts in solidliquid mush forming polymict breccias (Morfi et al., 1999;Pokhilenko, 2009;Giuliani et al., 2013) and in surrounding metasomatites.Diamonds should mainly crystallize at a distance from feeders in more reduced conditions from the fluid surrounding protokimberlite veins.

Conclusions
1. Crystallization trends of ilmenites in Siberian kimberlites reflect polybaric fractionation during formation of protokimberlite feeder systems.
2. The correlation of picroilmenite compositions with the mantle layering visible in the Fe# fluctuations in PT diagrams is explained by interaction of the ilmeniteforming magmas with mantle wall rocks.
3. The regular structure of the mantle sections (i.e., a heated SCLM base and a division into two parts with a boundary near 4.0 GPa (Sobolev, 1974;Ashchepkov et al., 2008a) gives three levels and three possible stages of megacryst origin: (1) deep HT metasomatic pyroxenites and crystallization in the feeding systems; (2) fractionation of protokimberlite melts mixed with partial melt from the peridotite wall rocks; (3) low-degree partial melting of hydrous metasomatic fronts and mixing with evolved protokimberlites.
4. Intrusion of protokimberlite melts rapidly changed the mantle beneath the kimberlite pipes due to large-scale percolation of melts and fluids.
The Supplement related to this article is available online at doi:10.5194/se-5-915-2014-supplement.
where B includes corrections to Mg and Cr.These corrections were found using compositions minerals in the

Figure 13 .
Figure 13.(a) REE patterns and trace element spider diagrams for ilmenites from heavy mineral separates from kimberlites from the Malo-Botuobinsky (a Mir and b International'naya pipes), Nakynsky field (c Nyurbinskaya pipe) and Anabar kimberlites (d Trudovaya pipe).