Composition of the basaltic samples applied for modeling. Oxides and LOI are given in wt. %, Ni and Cr in ppm. GPS: Coordinates of sample locations.
Besides mantle peridotites primary basaltic melts are the best tool to investigate upper mantle petrology and geochemistry. However, de facto primitive melts are hard to found, as basaltic melts usually go through a fractionation process during their ascent towards the surface. Most primary melt calculators are based on the major or trace element compositions of olivine-phyric ocean island basalts and peridotites and are less accurate if clinopyroxene fractionation occurred. In this chapter a new fractionation modeling method of alkaline basalts will be introduced, which has been published earlier only in Hungarian. Olivine ± clinopyroxene fractionation of four basaltic volcanoes have been modeled from different Miocene-Quaternary volcanic fields from the Carpathian-Pannonian Region (Stiavnica (Selmec) VF, Novohrad-Gemer (Nógrád-Gömör) VF, Perşani Mts. (Persányi Mts.) VF and from the Lucaret-Sanoviţa (Lukácskő-Sziklás) volcano.
- alkaline basalt
Intracontinental monogenetic alkaline basaltic volcanic fields consist of various types of small-scale volcanoes (e.g. scoria cones, tuff rings, maars, diatremes and related lava flows) . However, the size of the volcanic edifices is usually smaller than 1 km3 and they are short-lived volcanoes, the total magma output rate of a monogenetic volcanic field can be compared with a polygenetic volcano . The lifespan of the volcanic field can last for several millions of years (e.g. [2, 3]). Based on the geochemical and petrographic characteristics of the individual volcanoes, monogenetic basaltic volcanic fields could be very diverse both spatially and temporally (e.g. ). This diversity depends on several factors. Basaltic melt could be generated in the asthenosphere or in the lithosphere. Partial melting of the asthenosphere could be caused by a mantle plume (e.g. [5, 6, 7]), or mantle flows related to active rifting, collision or lithospheric delamination (e.g. [8, 9, 10]). Melt generation in the lithospheric mantle is often related to metasomatized peridotites and pyroxenites which are rich in volatiles and alkalis (e.g. [3, 11, 12, 13]). Primitive alkaline basalts are excellent tools to study the petrological and geochemical heterogeneity and thermal characteristics of the upper mantle and the type of melt generation (e.g. [5, 14, 15, 16, 17]), however a seemingly primitive basaltic magma could have undergone fractionation . Only with fairly accurate knowledge on the primary melt composition it is possible to reveal the whole story and characterize the source region of the basalts (e.g. [14, 19, 20]).
To find primary melt compositions we have to study their ascent history and characterize the melt evolution from the source to the surface. Most common methods are based on the main and trace elements composition of the bulk rock (e.g. [14, 18, 19, 20, 21]), and these methods are often less accurate if clinopyroxene fractionation occurred or if the source rock contained pyroxenite. To avoid the “pyroxene problem” a new olivine +/− clinopyroxene fractionation method has been developed by  and first published by  in Hungarian. Here we will show the method accuracy and some fractionation calculation results of alkaline basalts from [22, 23].
2. Concept of olivine and clinopyroxene fractionation modeling
Primitve magma compositions derived from less fractionated alkaline basalts are key tools to characterize the source region of upper mantle melts (i.e. petrology and geochemistry, depth and degree of melting, potential temperature of the upper mantle) (e.g. [7, 14, 17, 19, 24, 25]).
Most well-known primitive melt calculations are based on olivine addition or subtraction from bulk rock compositions of alkaline basalts from OIBs and LIPs [14, 19, 20, 24]. These methods could be misleading if the source rock of the basalt wasn’t purely peridotitic and especially if not only olivine but clinopyroxene fractionation occurred . Olivine addition to rocks that were generated from eclogitic/pyroxenitic mantle or went through clinopyroxene fractionation can produce unrealistically primitive melt compositions with too high MgO content, Mg# and mantle potential temperature .
Even primitive, high Mg# intracontinental alkaline basaltic rocks could be derivative melts after olivine and deep-seated clinopyroxene fractionation . To develop a more realistic fractionation modeling method instead of pure olivine addition/subtraction the degree of clinopyroxene fractionation should be taken into account. During fractional crystallization both melt and newly-formed crystals’ composition will change continuously. If this process can be modeled, an estimate for the degree of crystallization and for the main elements composition of the parental melt can be given.
To find the degree of olivine ± clinopyroxene fractionation quantitative estimations were performed.
With increasing degree of fractional crystallization of an alkaline basaltic melt the Fo and Ni content of olivine crystals will decrease. As it is a minor/trace element, the concentration of Ni in olivine depends on the . In accordance with results of  it was suggested by  that on the Fo-Ni plot of olivines from an alkaline basalt the olivine-only fractionation would appear as a curve of exponentially decreasing Ni-content with decreasing Fo-content, while co-crystallization of olivine and clinopyroxene draws a linear (diagonal or nearly horizontal) trend on the same plot. The slope of the exponential curve is determined by the , while the slope of the linear section depends on the . It has been shown on olivine and clinopyroxene phyric rock samples that the primitive (Fo and Ni-rich) population of olivine crystals will cluster dominantly around the upper part of the predicted exponential curve on the Fo-Ni plot and with decreasing Ni and Fo content at some point – when the clinopyroxene also begins to crystallize - the curve will change to diagonal . Accordingly if other significant process that can modify the Fo-Ni pattern of olivines than crystallization of olivine and clinopyroxene can be ruled out, the degree of fractional crystallization of the two mineral phases can be estimated based on the composition of the olivine crystals and the bulk rock. The Fo-Ni pattern of fresh, unaltered olivine crystals depends on the Mg# and/or the Ni-content of the basaltic melt. Significant change in Mg# of alkaline basaltic melts – besides olivine ± clinopyroxene fractionation – could be caused by fractionation of amphibole or by incorporation of Mg/Fe rich minerals (olivine, clinopyroxene, amphibole and magnetite). The Ni-content of olivines during fractional crystallization could be changed by precipitation of minerals with higher than 1. This value is the highest for magnetite (up to 30) and olivine (up to 25), and almost a magnitude lower for clinopyroxene (1–3) and amphibole (0.6–3) [28, 29, 30, 31]. To avoid any processes than olivine and clinopyroxene fractionation during magma ascent the following filters have been used during sample selection. 1: only fresh, unaltered olivine and olivine-clinopyroxene phyric basalts without amphibole content were investigated. 2: samples contain only minor magnetite content (<1%) with very low Ni concentration (200–700 ppm). 3: there is no sign of olivine or clinopyroxene incorporation (mega- or antecrysts with significantly different petrographic or geochemical characteristics.
2.1 The steps of the modeling
Creating a corrected major elements composition for each rock sample from LOI-free raw data: redistributing the Fe2O3tot content into FeO and Fe2O3 assuming a Fe3+/Fe2+ ratio of 0.15; recalculating Ni and Cr concentrations from ppm into NiO and Cr2O3 wt. % values. The Fe3+/Fe2+ will be reflected in the of the rock (, where Mg and Fe2+ are cation fractions. A different Fe3+/Fe2+ ratio (and Mg#) of the rock would modify later calculations (e.g. the amount of olivine that is calculated in the 2nd step) as they will lead to another equilibrium olivine composition. Therefore highly oxidized samples cannot be used as they have significantly higher Fe3+ content and lower Mg#rock. If the exact ratio of Fe3+/Fe2+ is known, it must be used instead of 0.15.
Calculating the Fo content of a hypothetical olivine (Fo*) being in equilibrium with the bulk rock composition, using the equation of Fo (mol%) = 100-(X/(1 + X)*100), where X = ([100*]-)/ and = 0.3. If the difference between the calculated equilibrium Fo value (= Fo*) and the Fo content of the most primitive (richest in Ni and probably in Fo) olivine () exceeds 0.1 mol%: the bulk rock composition has to be modified by olivine addition or subtraction in portions of 0.1 wt. % to reach equilibrium as follows (Figure 1A).
If Fo* < : the composition of an olivine crystal (being in equilibrium with the instantaneous melt) has to be added to the bulk rock composition in steps of 0.1 wt. %. For this, real analyzed olivine compositions from the same rock can be used.
If Fo* > : the composition of the most primitive olivine () has to be subtracted from the bulk rock composition in steps of 0.1 wt. %.
The amount of subtracted or added olivine (in wt. %) could represent the difference between the melt composition represented by the bulk rock and by the equilibrium melt related to the most primitive olivine in the sample. We call this ‘initial olivine fractionation’.
Modeling the decrease of MgO and Fo content in the melt and in equilibrium olivines simultaneously due to fractionation. Starting melts composition (Step 1.) has to be in equilibrium with the most Ni-rich olivine. Fractionation can be modeled by subtracting olivine in portions of 0.1 wt. % from the melt until 20 wt. % of total fractionation. During calculation composition of analyzed olivine crystals (being nearly in equilibrium with the instantaneous melt) should be applied.
Whereas the decrease of Fo content in olivines during fractionation has been calculated with a constant (0.3), the decrease of Ni concentration in the fractionated olivine crystals and in the equilibrium melt have to be modeled for each value between 5 and 20, assuming Rayleigh fractionation. Ni content of the most Ni-rich, analyzed olivine crystal is taken as starting data.
Determination of the of the investigated basalts.
Based on  in the case of pure olivine fractionation the most primitive olivine crystals should cluster along the upper, steep part of an exponential curve on the Fo-Ni plot. The most Mg-rich analyzed olivine crystals (over the 75th percentile by Fo-content and or over Fo84) are compared to the calculated Fo-Ni curves (Step 2.) with the least squares method, and the best-fitting one will be accepted for the of the investigated basalt (Figure 1C).
Estimation of the degree of the pure olivine fractionation.
The hypothetical, pure olivine fractionation trajectory should be compared with the total olivine Fo-Ni dataset of the sample, whether they are fitting well, or at some point the initially exponential trend of analyzed olivine crystals changes to a more straight, diagonal one.
If they are fitting well, only olivine fractionation happened. The olivine crystal with lowest Fo-Ni content will draw the degree of fractionation, which can be easily read from the modeled Fo-Ni curve (Figure 1D). In this case the modeling has been ended here. Jump to Step 7.
If the trend of the analyzed olivine crystals can be divided into an exponential and a linear section, the first part represents the initial pure olivine fractionation and the linear part the later olivine-clinopyroxene co-crystallization. The degree of pure olivine fractionation is determined by the intersection of the modeled olivine Fo-Ni curve and the linear section of the olivine dataset (Figure 2).
Calculating the composition of the equilibrium melt (‘intermediate melt’) for the olivine that represents the end of the initial pure olivine fractionation (Step 4a.) like equilibrium melt calculations in Step 1.
Modeling the evolution of the “intermediate melt” by fractionation of olivine and clinopyroxene together. The decrease of olivines Ni concentration along the diagonal Fo-Ni trend has to be modeled by Rayleigh fractionation presuming different olivine/clinopyroxene ratios. As a thought experiment, the decrease of Ni concentration in olivines could be modeled by a given quantity of olivine (X wt. %) or clinopyroxene (Y wt. %) fractionation only, however it must be a co-crystallization of the two minerals. Therefore, the decrease of Ni by fractional crystallization has to be calculated for each ol/cpx ratio from X wt. % olivine +0 wt. % clinopyroxene to 0 wt. % olivine + Y wt. % olivine, changing the amount of each mineral by steps of 0.1 wt. %.
Estimating the compositional change of the intermediate melt for all ratios of ol/cpx by subtracting these minerals from it. Subtraction of minerals is the same as in Step 1a. The calculated melt compositions are considered “final melts”.
Estimating the Fo-content of equilibrium olivine for each final melt composition and compare them to the analyzed olivine crystals.
That olivine/clinopyroxene fractionation ratio and amount will be accepted, for which the related final melts equilibrium olivine will have the same Fo-content as the lower end of the diagonal Fo-Ni trend of analyzed olivines.
The total degree of olivine and clinopyroxene fractionation will be the sum of the volumes calculated in “Step 4b” and “Step 5e”.
Optional check: it is possible to calculate parental melt composition by another method based only on olivine addition, like the method of PRIMELT2  or Fractionate-PT . If significant amount (>10 wt. %) fractionation of clinopyroxene has been calculated by the presented method, the mentioned programs should calculate a more Mg-rich parental melt (Figure 3 and Table 3).
|Calculation based on||PRIMELT2 software||Fractionation modeling presented in this article|
|Olivin add. by Primelt2||14,20||7,10||23,40||6,60|
2.2 Remarks on limitations of the presented fractionation estimation method
Alkaline basalts that are derivative melts of fertile mantle peridotites will have an MgO content of 8–13 wt. % and equilibrium olivines with a Fo content of 86–89 mol% . It is common that even the most primitive olivine phenocrysts of an alkaline basalt seemingly suitable for modeling (olivine +/− clinopyroxene phenocrysts, no signs of amphibole or early pyroxene crystallization or incorporation of mafic minerals) have quite low Fo and/or Ni-content (e.g. ). In this case a proper fractionation modeling is not possible, only a rough estimation can be performed based on a comparison with similar olivine datasets containing more Fo-rich olivines. An exact limit for proper calculations cannot be specified, only suggestions for it. Based on the empirical observations of  the fractionation modeling is more reliable if the olivine crystals over the 75th percentile by Fo-content draw a steep, semi vertical trend on the Fo-Ni plot and they have 89–82 mol% Fo and 2500–1200 ppm Ni content.
The deep-seated clinopyroxene fractionation  will not affect the composition of later crystallized olivines, while the presented method can reveal clinopyroxene crystallization only if it occurred together with olivine fractionation. Therefore, it cannot indicate the precipitation of clinopyroxene onto the wall rocks at great depth. To identify any modification of primary melts by reaction with mantle minerals is also beyond the limits of the presented method.
Since there are several variable parameters during fractionation modeling (the value, the applied Fe3+/Fe2+ ratio for the bulk rock, the amount and ratio of olivine and clinopyroxene, the composition of the intermediate melt), the calculation is a little bit cumbersome.
3. Case studies from the Carpathian-Pannonian region
3.1 Geological setting
From a volcanological and geodynamical point of view – based on its Neogene-Quaternary volcanism – the Carpathian-Pannonian Region (Figure 4) is a related to the broader Mediterranean area in terms of geodynamic setting . The post-Paleogene volcanic formations of the region can be divided by age and geochemical features into four groups : 1) Miocene Si-rich-; 2) Miocene-Quaternary K- and high-K-; 3) Miocene-Quaternary calc-alkaline- and 4) late Miocene-Quaternary alkaline type. The latter group is represented by several alkaline basaltic monogenetic volcanic fields, some of which are deeply buried by late Miocene sediments [10, 39, 40, 41]. The most intense period of alkaline basaltic volcanism took place between 5 and 3 Ma ago. At that time numerous volcanoes formed, for example, in the Bakony-Balaton Highland Volcanic Field (e.g. ), in the Nógrád-Gömör/Novohrad-Gemer Volcanic Field (e.g. ) and in the Styrian Basin Volcanic Field (e.g. ). The alkaline rocks show a wide range in composition from the primitive, olivine-phyric basanites towards the more differentiated, clinopyroxene-rich phonotephrites (e.g. [10, 43, 44]). Based on their trace element content they show similarities both with the Neogene-Quaternary alkaline rocks of western and central Europe (e.g. [45, 46]) and with alkaline basaltic rocks of the Mediterranean area (e.g. [47, 48]). The driving force of the alkaline basaltic melt generation is still subject of debate as the Neogene-Quaternary basaltic volcanism of the region dominantly postdates the rifting of the Pannonian Basin and took place in the thermal inversion phase of the basin, in a compressional geodynamic regime [10, 37, 49]. There are two dominant theories for basalt generation in the area, both associate it with regional mantle flows, the differences are in the driving forces. 1) In the outer parts of the Pannonian Basin melting could have been associated with mantle upwelling related to the so-called thin-spot event, a dominantly vertical asthenospheric flow from under the deep roots of the surrounding mountain chains or to lithospheric delamination while the sporadic volcanism in the central parts of the basin could have been related to strike-slip tectonics [10, 22, 37]. 2) The other theory comes from , who suggested that basaltic melt generation was caused by rather a regional, mostly horizontal mantle flow related to the north-eastward push of the Adria microplate.
3.2 Analytical methods
Whole-rock compositions were analyzed at Acme Labs (Vancouver, Canada). Major and minor elements were analyzed by ICP-ES, trace elements by ICP-MS. Petrographic descriptions were done with a Nikon YS2-T polarizing microscope using NIS-Elements Br software and an AMRAY 1830 I/T6 scanning electron microscope at the Department of Petrology and Geochemistry, Eötvös Loránd University (Budapest, Hungary). The polished and carbon-coated thin sections were analyzed with a CAMECA SX100 EMPA, equipped with one energy-dispersive and four wavelength-dispersive spectrometers at the Department of Lithospheric Research, University of Vienna (Austria). An accelerating potential of 15 kV and a beam current of 20 nA were used with at least 20 s counting time on peak position and 1 μm beam diameter. PAP correction procedure was applied on raw data. Analyses were done against natural and synthetic mineral standards: olivine (Mg), corundum (Al), quartz (Si), apatite (P), wollastonite (Ca), rutile (Ti), Mg-chromite (Cr), spessartine (Mn), almandine (Fe), Ni-oxide (Ni).
3.3 The investigated alkaline basalts
BRE: an olivine phyric massive basanite sample with porphyritic-intergranular texture (Figure 5). Clinopyroxene phenocrysts are very rare (olivine/clinopyroxene ratio is 19/1, total phenocryst content: 6%). Olivine phenocrysts are normal zoned, dominantly hypidiomorphic-idiomorphic with an average size of 650–700 μm, they contain idiomorphic magnetite and Cr-spinel inclusions dominantly in the crystal rims. Largest grains are often resorbed and iddingsitized along crystal rims and cleavage tracks. Hypidiomorphic-idiomorphic pale brown clinopyroxene microphenocrysts are dominantly normal-, rarely sector zoned crystals with a maximal size of 400–500 μm. They often contain idiomorphic magnetite inclusions. The groundmass contains plagioclase, clinopyroxene, olivine, nepheline and accessory magnetite and ilmenite.
RAC2: an olivine phyric trachybasaltic scoria clast with vesicular, porphyritic-intersertal texture (Figure 5). Clinopyroxene microphenocrysts are very rare (olivine/clinopyroxene ratio is 19/1, total phenocryst content: < 5%). Olivine phenocrysts are normal zoned, dominantly hypidiomorphic-idiomorphic, sometimes skeletal; with an average size of 500 μm, largest crystals reach 1250 μm. They are often strongly iddingsitized. Most oxidized grains contain few μm thick iron-oxide needles. Idiomorphic inclusions of Cr-spinel are common in crystal rims. Hypidiomorphic-idiomorphic pale brown clinopyroxene microphenocrysts often form glomerocrysts. They are dominantly sector-, rarely normal zoned crystals with an average size of 300–350 μm. The groundmass contains plagioclase, clinopyroxene, olivine, glass, nepheline and accessory magnetite and ilmenite.
SAN: an olivine and clinopyroxene phyric trachybasaltic lava rock sample with porphyritic-intergranular texture (Figure 5). Clinopyroxene phenocrysts are quite common (olivine/clinopyroxene ratio is 4/1, total phenocryst content: 20%). Olivine phenocrysts are normal zoned, dominantly hypidiomorphic-idiomorphic, often resorbed; with an average size of 500 μm, largest crystals reach 1200 μm. They are often iddingsitized along crystal rims and cleavage tracks. Idiomorphic inclusions of Cr-spinel are common in crystal rims. Hypidiomorphic-idiomorphic pale brown clinopyroxene phenocrysts are sector zoned crystals with an average size of 300–400 μm, largest grains reach 1000 μm. They contain magnetite inclusions and they often form glomerocrysts. The groundmass contains plagioclase, clinopyroxene, olivine and accessory magnetite and ilmenite.
SOR: an olivine and clinopyroxene phyric basanitic dyke sample with slightly vesicular, porphyritic-intergranular/intersertal texture (Figure 5). Clinopyroxene phenocrysts are common (olivine/clinopyroxene ratio is 5/5, total phenocryst content: 8%). Olivine phenocrysts are normal zoned, dominantly hypidiomorphic-idiomorphic, often resorbed; with an average size of 650 μm, largest crystals reach 3000 μm. Idiomorphic inclusions of Cr-spinel and magnetite are common in crystal rims. Idiomorphic-hypidiomorphic pale brown clinopyroxene phenocrysts are sector-, normal- or oscillatory zoned minerals with an average size of 470 μm, largest grains reach 2000 μm. They contain magnetite inclusions. Some clinopyroxene phenocrysts have an olive-green core. Few grains of completely opacitized amphibole megacrysts can be observed. The groundmass contains plagioclase, clinopyroxene, olivine, glass, nepheline and accessory magnetite.
4. Results of olivine and clinopyroxene fractionation modeling
From two samples (BRE and RAC2) olivine crystals are fitting well to the modeled pure olivine fractionation curves that were calculated with = 10. Based on this the calculated degree of olivine fractionation is 12.5 wt. % and 10.35 wt. %, respectively (Figure 8A and B). Adding this amount of olivine to the bulk rock composition the parental melt would have an Mg# of 74.24 and 74.77.
In the case of SAN and SOR only the most primitive, Mg-rich olivine crystals follow the modeled exponential olivine fractionation curve on the Fo-Ni plot and the more evolved ones draw a diagonal trend (Figure 8C and D). A value of 12 and 13 were applied for SAN and SOR, respectively. The calculated degree of olivine fractionation is 9.5 wt. % for SAN and 5.5 wt. % for SOR during the initial, pure olivine crystallization period. Based on the olivine crystals that are falling onto the diagonal trend it is obvious that the evolution of both melts switched at some point from pure olivine fractionation to co-crystallization of olivine and clinopyroxene. The calculated degree of fractionation for olivine and clinopyroxene during this phase are 7 and 5.5 wt. % for SAN and 1.89 and 7.56 wt. % for SOR. The total degree of olivine and clinopyroxene fractionation is 16.5 and 5.5 wt. % for SAN and 7.39 and 7.56 wt. % for SOR. Adding this amount of the two mineral phases to the bulk rock composition the parental melt would have an Mg# of 73.44 and 74.54, respectively (Table 3).
The results of fractionation modeling are consistent with the petrographic observations, i.e. for the olivine phyric basalts (BRE, RAC2) only olivine fractionation have been calculated, while for those which have a considerable amount of phenocrystic clinopyroxene a significant clinopyroxene fractionation have been revealed by the model. The olivine/pyroxene ratio in the rock samples is 4/1 for SAN and 5/5 for SOR, this is fairly the same as the results of modeling (5.5/16.5 for SAN and 7.4/7.6 for SOR).
The calculated parental melts composition may resemble a mafic melt from a fertile peridotite. Parental melts composition and Mg# have been calculated with the Primelt2 software  too. For the olivine phyric rocks it gave a similar Mg# (BRE – 75.13, RAC2–73.27).
In the case of the SAN trachybasalt the Mg# calculated by the Primelt2 software is obviously too high (Mg# = 76.31), which is consistent with the high degree of clinopyroxene fractionation that was indicated by the fractionation modeling. This coincides with the fact that the Primelt2 program does not count on the possibility of clinopyroxene fractionation during primary melt calculations except that it gives a warning about the possible pyroxene fractionation.
The SOR basanite is little bit more interesting as the fractionation modeling have resulted the same amount of olivine and clinopyroxene fractionation, and the Mg# of the parental melts based on the modeling is somewhat higher compared to the result given by the Primelt2 . In this case two important facts have to be considered. First, the modeled mineral fractionation fits very well to the petrographic observations in all case. It suggests that the pyroxene fractionation predicted by the modeling should also be realistic in the case of SOR basanite. Secondly, the Primelt2 program gives us warnings, if the calculation may be inaccurate because of pyroxenite source rocks or because of clinopyroxene fractionation. In the case of SOR basanite a pyroxene fractionation warning was given by the software, although the suggested olivine addition to reach primary melt (6.6 wt. %) is almost the same as it was given by fractionation modeling (7.39 wt. %).
The uncertainty, whether the result of the fractionation modeling or the Primelt2 is the more realistic may have arisen from two factors. Neither olivine, nor clinopyroxene is a dominant phase and probably this amount (<10 wt. %) of pyroxene is close to the limit what can be detected by the difference in primary melts’ Mg#.
Despite its limitations, the presented olivine and clinopyroxene fractionation modeling based on olivine and bulk rock compositional data draw attention that it is not only olivine fractionated basalts that could be useful for primary melt calculations.
It is more important, that all possible tools have to be applied from petrographic observations to quite simple or more complicated geochemical modeling methods. We hope that the modeling presented in this chapter will inspire the reader to develop further ideas and methods to give a more realistic and better description of the fractionation process of alkaline basaltic melts.
First of all Tamás Sági would like to express his special thanks to M. Éva Jankovics, Zoltán Taracsák, Balázs Kiss and Csaba Szabó for the thought-provoking discussions about fractionation of alkaline basalts. The manuscript has been improved by the constructive comments and suggestions of Marco Brenna.