Introduction

Increasing demand for lithium (Li), primarily due to its use in Li-ion battery systems, is driving exploration for new sources1. The primary hard rock source of Li is peraluminous granitic bodies derived from partial melting of metasedimentary rocks2,3. Competing models to explain the enrichment of Li in granitic melts invoke, variously, Li-rich source rocks4, degree and number of melting episodes5,6, and/or extent of fractionation7. The central factor determining the viability of these models is the choice of partition coefficients (\({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\)) used to describe Li behaviour. This choice is not straightforward, as literature values can vary by over an order of magnitude for a given phase. For example, for phases and compositions relevant to granitic systems, reported \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) are 0.05–0.45 for quartz, 0.02–0.50 for plagioclase feldspar, 0.01–0.14 for alkali feldspar, 0.50–1.12 for muscovite, 0.8–32.8 for biotite and 0.14–4.0 for cordierite (Supplementary Table 1).

Of these phases, biotite has been highlighted as the dominant host for Li within metasedimentary assemblages8, and confers the advantage that it is stable at solidus conditions across a wide pressure and sedimentary protolith range (e.g. greywacke and semipelite), compared with more restricted ranges for other potentially influential lithium-rich phases such as muscovite and cordierite9. Therefore, biotite breakdown reactions plausibly represent the most important source for the generation of lithium-rich melts. Many previous enrichment models have considered Li as compatible in biotite6,10, typically with a static \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) of ~1.67 (ref. 11). However, a recently developed composition- and temperature-dependent partitioning model12 predicts that Li can be incompatible in biotite for typical metasedimentary biotite compositions at standard partial melting conditions (Supplementary Fig. 1). The above study also shows that Li is extremely compatible in fluorine-rich biotite (\({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) up to 32.8). It is timely, therefore, to analyse the impact of partitioning model choice on melt lithium contents and in doing so test prevailing models to understand magmatic Li enrichment in granites.

To test these models, we use the Li-rich granites of the Cornubian granite batholith (United Kingdom) as a case study (Fig. 1). The Cornubian granite batholith is a peraluminous granite body located in southwest England, which formed between ~295–275 Ma during the Variscan orogeny13,14. The batholith outcrops as six major centres (Fig. 1), continuous at depth, forming a 250-km-long intrusion that is purported to be ~10-km thick15. Based on textures and mineral assemblages, the batholith has been divided into five categories (G1–G5)4. The areally-dominant G1 and G3 granites have been ascribed to muscovite- and biotite-dehydration melting of a Proterozoic greywacke source, respectively4,16. The G2 and G4 granites are interpreted as evolved versions of these G1 and G3 granites4. The areally minor G5 granites have been attributed to partial melting of a biotite-rich residue following G1–G4 melt extraction, with a lack of evidence linking the G1–G4 granites and G5 granites through fractionation4,10,17,18, and contain magmatic fluorite and topaz, with subsidiary aplites and pegmatites4,19. Melting pressures are poorly constrained, with G1 estimated at >5 kbar and G3 at <4 kbar4. The presence of minor magmatic andalusite4 indicates that some crystallisation occurred at <4.5 kbar20. Lithium concentrations vary systematically, with the lowest concentrations in the G1 and G3 granites (≤450 ppm), intermediate concentrations in more evolved granites (G2, G4; ≤832 ppm) and the highest concentrations in G5 granites (≤1699 ppm, up to 3619 ppm in aplitic veins)4,10. The G5 granites are also rare earth element-depleted relative to G1–G4 (Supplementary Fig. 2)10.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Geological map of the Cornubian granite batholith, southwest England, with granite location data taken from ref. 4.

Here, we integrate phase equilibrium modelling with the range of published \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) to explore the controls on the petrogenesis of Li-rich granites, and introduce a model for the production of metasedimentary-derived Li-rich melts through dehydration melting of fluorine-rich biotite.

Results

We investigate the major and trace element evolution of the Cornubian granite batholith by modelling partial melting and subsequent fractional crystallisation using MAGEMin21 (see Methods), with a starting bulk composition of an average Precambrian greywacke22. This composition is consistent with the inferred rock type and age of the protolith, which is not exposed at the current erosion level16. Modelling was also performed with a global average greywacke22 and an average pelite23 (Supplementary Data 1) to test the sensitivity of our results to this parameter. Water contents were set to ensure a minimally fluid-saturated solidus for the pressure of interest. Calculations were performed across a range of crustal pressures (2–12 kbar), with a focus on temperatures between the solidus and complete biotite-dehydration melting.

To simulate partial melting, we apply a stepwise-batch model, with melts extracted from the system once the melt reaches a 7 vol% threshold (the melt connectivity transition24). Figure 2a shows a composite pseudosection for successive bulk compositions produced from melt extraction events at 8 kbar (orange circles). At the 8 kbar fluid-saturated solidus (668 °C), the metagreywacke phase assemblage comprises plagioclase, quartz, biotite and amphibole with minor rutile and fluid, with biotite present at ~20 vol% (Fig. 2b). The amphibole is pargasitic hornblende in composition, consistent with the moderate CaO content (~3 wt%) of the metasedimentary bulk composition (see Supplementary Data 1 and 2). Plagioclase is An31–38, becoming more calcic with increasing temperature until 840 °C, when Ca content remains at An38 up to the fourth melt extraction step. Biotite has an Mg# of 0.52–0.64, with a moderate Ti content (0.11–0.27 Ti a.p.f.u.), with both Mg and Ti increasing in concentration with increasing temperature. Minimal melt is produced up temperature until the onset of amphibole- and subsequently biotite-dehydration melting, leading to three melt extraction steps in close succession at 803, 829 and 836 °C. Biotite is fully exhausted at 840 °C, soon after the third melt extraction step, at which point melt production rates decrease such that the fourth melt extraction step is not reached until 909 °C.

Fig. 2: Mineral assemblages and melt compositions.
Fig. 2: Mineral assemblages and melt compositions.The alternative text for this image may have been generated using AI.
Full size image

a Composite P--T pseudosection for an average Precambrian greywacke composition (Supplementary Data 1 and 2; minimally fluid-saturated at 8 kbar) showing melt extraction events (orange circles) at 7 vol% melt intervals (i.e. stepwise-batch melting). Zero mode isolines are highlighted for liquid (solidus; pink) and biotite (green). The solidus is marked at 8 kbar by an orange cross. b Phase abundances (vol%) during stepwise-batch melting at 8 kbar. c Melt compositions during stepwise-batch melting (dotted lines, with the solidus composition marked by a cross, and melt extraction steps marked by open circles) and isobaric fractional crystallisation of melts from each third melt extraction step (solid lines, with 75 vol% fractionation marked by a vertical black bar for reference) from 2 to 12 kbar. Background data were reported compositions from the Cornubian granite batholith4,10, with the legend shown in (d). d Melt composition plot as (c), but only showing 8 kbar melting. The melt composition during subsequent fractional crystallisation (for the melt produced from the third melt extraction step, as closest to complete biotite dehydration) is shown for fractionation at 8 kbar (solid line as per panel (c)) and at 4 kbar (dashed line), with the shaded region representing viable melt compositions produced for intermediate fractionation pressures. Phase abbreviations: afs alkali feldspar, amp amphibole, bt biotite, fl fluid, ilm ilmenite, liq liquid, mag magnetite, opx orthopyroxene, pl plagioclase, qz quartz, rt rutile.

Melt compositions produced during stepwise-batch partial melting at 2–12 kbar are shown in Fig. 2c (dotted lines), with additional oxides shown in Supplementary Fig. 3. For each melt extraction step (open circles), we also calculate the melt composition evolution during subsequent fractional crystallisation in an isobaric system (solid lines). To capture the principal effects of biotite breakdown (and for figure clarity), we only show the results of fractionation for extraction event three, as this is the final melt extraction event before complete biotite dehydration, but all follow similar trajectories. To consider scenarios where melting and crystallisation occur at different pressures, as suggested by existing estimates for the Cornubian granite batholith, we also show melt evolution for fractional crystallisation at 4 kbar (dashed line) from an 8 kbar partial melt (Fig. 2d). Additional scenarios are detailed in Supplementary Fig. 5.

The results show that partial melt compositions are pressure sensitive, with lower-pressure melts more enriched in SiO2 and exhibiting a smaller MgO range (Fig. 2c). We observe a good match between the modelled melt compositions and the majority of reported Cornubian granite batholith data (within ~1 wt% for all oxides; see Supplementary Fig. 4), while also taking into account mineralogy (e.g. magmatic andalusite4), for scenarios involving partial melting at 8 kbar, and fractionation at a range of pressures down to 4 kbar (shaded orange region, Fig. 2d). Similar conclusions are obtained using the alternative greywacke or pelite compositions (Supplementary Fig. 6). The partial melting pressure estimate of ~8 kbar is consistent with pressures inferred for G1 granites, but higher than previous G3 estimates. The role of biotite-dehydration melting is in accord with previous studies4. However, the absence of muscovite-dehydration melting is counter to previous suggestions4, but consistent with the typical scarcity of muscovite in metagreywacke compositions25.

Lithium enrichment during source melting

Melt lithium contents can be forward modelled by integrating the above-described phase equilibrium modelling with constraints on lithium distribution between minerals and silicate melt (\({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\)). Our calculations consider the full range of published \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) relevant to granitic systems (Supplementary Table 1), which, aside from one parameterisation of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) and one of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{cordierite/melt}}}}}\), are static values. We consider four combinations: (i) the values we consider to be most reasonable for our modelled melt composition (with justifications provided in Supplementary Table 1); (ii) the same suite of reasonable values, but with a dynamic12\({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) in place of the 1.67 value11 used in (i), to highlight the effect of using this new model; and (iii–iv) the minimum and maximum reported \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) for phases aside from biotite, which is kept constant at 1.67, to highlight the effect of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) choice for generally Li-poor phases. These combinations are referred to as (i) reasonable-static, (ii) reasonable-dynamic, and (iii–iv) minimum and maximum hereafter.

We first quantify the Li enrichment of melts produced from a Precambrian greywacke source during stepwise-batch melting at 8 kbar (following the major element pressure constraints above). Using the reasonable-static suite (solid blue line, Fig. 3a), the Li enrichment factor (melt Li concentration, relative to the initial source concentration) increases with overall melt fraction from ~1.7 to 2.4. This behaviour is driven by the breakdown of biotite (green line), which has a \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\)  > 1, leading to a progressive decrease in bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) (solid orange line) and increase in melt Li enrichment as melting progresses. By contrast, the reasonable-dynamic combination leads to the enrichment factor decreasing from ~3.2–1.6 (dashed blue line). This behaviour is driven by a low \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) of 0.29–0.37 (i.e. Li behaving incompatibly), which is similar to the remaining bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\). Therefore, bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) (dashed orange line) is approximately uniform during melting, and melt enrichment decreases with increasing melt fraction as expected for an incompatible element. Finally, the minimum and maximum suites yield a wider range of melt enrichment factors from 1.3 to 6.1 (grey shading, Fig. 3b); a range consistent with previous modelling of metasedimentary protoliths6,10,26. The dashed vertical lines in Fig. 3 denote the melt extraction events, with the corresponding enrichment factors used during subsequent fractionation modelling.

Fig. 3: Li enrichment, bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) and biotite wt% during stepwise-batch partial melting of a Precambrian greywacke at 8 kbar, with different sets of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) (see main text).
Fig. 3: Li enrichment, bulk $${D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}$$ D Li solid / melt and biotite wt% during stepwise-batch partial melting of a Precambrian greywacke at 8 kbar, with different sets of $${D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}$$ D Li mineral/melt (see main text).The alternative text for this image may have been generated using AI.
Full size image

a Li enrichment in the melt using the reasonable-static (solid line) and reasonable-dynamic (dashed line) suites of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\). b Li enrichment using the minimum and maximum suites of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\), along with the reasonable-static suite for reference. The shaded areas represent the range of potential values between the minimum and maximum lines. See Supplementary Table 1 for the compilation of partition coefficients used.

Lithium enrichment during fractionation

Next, we model Li enrichment during fractional crystallisation of the extracted melts, building on enrichment achieved during stepwise-batch melting (Fig. 3). For each extracted melt, we calculate enrichment using the same \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) sets as used for stepwise-batch melting: reasonable-static (Fig. 4a), reasonable-dynamic (Fig. 4b), and the minimum and maximum suites (Fig. 4c). We also show the modelled onset of fluid exsolution (blue circles), which leads to a slight increase in bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid* /melt}}}}}\) (with solid* indicating solid phases plus a fluid phase) as we use a \({D}_{{{{\rm{Li}}}}}^{{{{\rm{fluid/melt}}}}}\) of 0.65 (ref. 27), which is higher than the bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) prior to exsolution. Reported maximum lithium concentrations for Cornubian G1–G5 granites are shown for reference (horizontal lines, scaled relative to a 100 ppm Li source, based on reported Cornubian metasedimentary Li concentrations28), alongside commonly cited upper limits to fractionation extent: 75 vol%, the particle-locking threshold29 (PLT) for felsic systems; and 90 vol%, based on this value (in wt%, but similar owing to densities of melt and solid in granitic systems) being referenced as extreme and potentially unrealistic30,31. Evidence from the Cornubian granite batholith suggests some Li-rich G5 samples have been enriched in Li by secondary processes32, meaning the reported G5 concentrations (Fig. 4) represent upper limits for target enrichment in our models. Fractional crystallisation is modelled at 4 kbar for consistency with the major element and mineral assemblage constraints (following 8 kbar partial melting), but Li enrichment in the melt is broadly insensitive to pressures of either partial melting or fractionation (Supplementary Fig. 7), so these model choices are not an influential factor.

Fig. 4: Li enrichment during 4 kbar fractional crystallisation of melts produced from the 8 kbar stepwise-batch melting of a Precambrian greywacke.
Fig. 4: Li enrichment during 4 kbar fractional crystallisation of melts produced from the 8 kbar stepwise-batch melting of a Precambrian greywacke.The alternative text for this image may have been generated using AI.
Full size image

The Li melt concentration required for spodumene saturation is shown for reference (5000 ppm, refs. 6,67) alongside published granite Li data4,10, all scaled to a source metasediment concentration of 100 ppm Li (see main text). G1/G3 and G2/G4 granites are grouped as the latter are thought to be fractionated versions of the former4,10. Note that hydrothermal processes may have acted to increase Li concentrations above those achieved during magmatic differentiation, so these targets represent the upper limits for magmatic enrichment30,32. Vertical lines at 75 vol% and 90 vol% represent the particle-locking threshold (PLT) and extreme fractionation (EF), respectively29,30. The point of fluid exsolution in each crystallising melt is shown by a blue circle on the bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid* /melt}}}}}\). Solid* highlights that after fluid exsolution has occurred, the bulk DLi value includes fluid as part of the solid assemblage. a Results using the reasonable-static \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) suite. b Results using the reasonable-dynamic \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) suite. c Results using the minimum and maximum \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) suites (range of grey shading), alongside the reasonable-static suite (purple line), shown only for melt extraction step 3, as this step is the closest to complete biotite dehydration in the protolith.

Using the reasonable \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) suites, reported G1–G4 Li concentrations can be reached before 90 vol% fractionation, and in some cases before the particle-locking threshold, depending on the combination of extracted melt step and \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) suite used (Fig. 4a, b). However, the G5 concentrations require more extensive fractionation at or beyond 90 vol%. Even considering the set of minimum \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\), leading to maximum Li enrichment (Fig. 4c), G5 aplite concentrations still require ~85 vol% fractionation (i.e. close to extreme) from their source melt. Collectively, our results show that fractional crystallisation is the dominant driver of Li enrichment in granites, but that the highest reported Cornubian granite batholith Li concentrations may still be difficult to achieve without considering exceptional circumstances, which we explore below.

Discussion

Our modelling suggests that reported G5 granite Li concentrations (and higher) could be achieved via extensive fractionation and/or particular Li partitioning behaviour (Figs. 3 and 4), with the caveats that such extreme degrees of fractionation may be unrealistic and that other partition coefficient choices may be more reasonable. Alternative suggestions have focussed on source controls, such as increased concentrations of Li in the source5 (i.e. greater than the 100 ppm used in Fig. 4), or a particular protolith that yields higher enrichments during source melting4,33,34 (i.e. higher than achieved in Fig. 3). Considering the first source suggestion, Supplementary Fig. 8 shows the effect of increasing protolith Li concentrations, with 200 and 400 ppm Li still requiring ~92 and 81 vol% fractionation, respectively, to reach G5 aplite concentrations (using the reasonable-static suite of partition coefficients). Given that the average upper continental crust contains around 20 ppm Li35, this explanation is unlikely to be widely applicable. For the latter source suggestion, remelting of a biotite-rich restite has been invoked to explain G5 petrogenesis4,33,34. However, forming such restite requires melt to crystallise in situ to liberate water into the wall rock36, which is incompatible with the large-scale extraction of granitic melts.

Alternatively, to explain G5 petrogenesis (and Li-rich granitic melts in general), we propose an alternative metasedimentary protolith: one that contains fluorinated biotite. This suggestion is supported by the presence of primary fluorine-rich phases (e.g. fluorite) in G5 granites (and in Li-rich melts globally37,38,39), yet not in G1–G410,19,40. The key characteristics of fluorinated biotite are its order of magnitude greater \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) than non-fluorinated biotite12, and its stability to higher temperatures, of order 100 °C41,42,43. These attributes mean that during partial melting of a source region that contains variably fluorinated biotite, non-fluorinated biotite breaks down first (at a given structural level), with fluorinated biotite retaining Li in the residual solid assemblage. Lithium- and fluorine-rich melt is produced as temperatures increase sufficiently to destabilise the fluorinated biotites. In turn, fluorine is a network modifier in high-silica melts, reducing melt viscosity (of magnitude similar to H2O44,45) and facilitating melt extraction after extreme degrees of fractionation. In granitic melts, fluorine does not partition preferentially into exsolved aqueous fluids at magmatic conditions46,47, meaning that the melt remains depolymerised by abundant fluorine beyond the onset of fluid saturation. Additionally, fluorine acts to suppress the solidus, therefore allowing fractionation to continue to lower temperatures46,48.

To test the effect of this concept for the Cornubian granite batholith, we modify \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) during stepwise-batch melting in the reasonable-static and minimum/maximum suites of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) used above, using 32.8 (the highest experimental \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) reported in fluorinated biotite12). As the thermodynamic model system does not contain fluorine, we cannot model the resultant higher-temperature biotite stability or lower temperature solidus, but the principles remain similar. Lithium is retained in the residual solid during melting (bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) ~6, orange lines, Fig. 5a) while (fluorinated) biotite is stable. As biotite-out is reached, Li becomes incompatible in the residual solid. Consequently, the melt Li enrichment factor increases rapidly to 3–14 (depending on the choice of partition coefficients) as Li is quickly released from the solid. The propagated effect of this partial melting enrichment during fractional crystallisation is shown in Fig. 5b (using \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) = 1.67, i.e. crystallising non-fluorinated biotite) and Supplementary Fig. 9b (using \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) = 32.8, i.e. crystallising fluorinated biotite). These two scenarios represent end-members depending on the behaviour of fluorine in the fractionating system. In Fig. 5b, G5 granite Li concentrations can be achieved prior to 90 vol% fractionation using reasonable \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) choices. However, if early-formed biotite is highly fluorinated (Supplementary Fig. 9), Li is instead removed from the melt during early crystallisation, and lower Li enrichments are reached during fractionation. The scenario in Fig. 5b is likely more realistic for G5 granites, where primary magmatic fluorite suggests that fluorine is not sequestered in biotite during early crystallisation.

Fig. 5: Li enrichment during stepwise-batch melting (8 kbar) and subsequent fractional crystallisation (4 kbar) mimicking the effect of fluorinated biotite in the protolith.
Fig. 5: Li enrichment during stepwise-batch melting (8 kbar) and subsequent fractional crystallisation (4 kbar) mimicking the effect of fluorinated biotite in the protolith.The alternative text for this image may have been generated using AI.
Full size image

a Li enrichment during melting for modified versions of the reasonable-static and minimum/maximum \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) suites (\({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) = 32.8, see text). The horizontal grey dotted line marks the point where \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) switches from compatible to incompatible, driving the increase in melt Li enrichment. b Li enrichment during subsequent fractional crystallisation of the melt extracted from panel (a) at the point of complete biotite breakdown, using the reasonable-static and minimum/maximum suites of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) (with \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) = 1.67 for all scenarios). The terminology \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid* /melt}}}}}\) in panel (b) reflects the presence of fluid in the fractionating assemblage, with solid*/melt referring to ((solid + fluid)/melt). The horizontal grey dotted line marks the point where \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid* /melt}}}}}\) switches from compatible to incompatible.

While G5 aplite Li concentrations still require ~93 vol% fractionation using the reasonable-static \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) suite (or ~60 vol% for the minimum suite), the noted behaviour above of fluorine as a network modifier increases the viability of extensive fractionation beyond suggested limits. A source rock that contains a higher biotite abundance than modelled here (~20 vol%; Fig. 2b) would also yield higher overall enrichments. Finally, this melting scenario can also explain the relatively rare earth element depletion of G5 compared to G1–4 (Supplementary Fig. 2), as the G5 melts are extracted from an already depleted source as a result of earlier G1–G4 melt extraction.

As demonstrated above (Figs. 3–5), partition coefficient choice variably impacts resulting melt Li enrichment for a given partial melting or fractionation scenario. Biotite-dehydration melting has previously been proposed as an important contributor to Li enrichment in granitic systems8 because of its often-observed mild compatibility for Li (for non-fluorinated scenarios). Therefore, the suggestion that Li may behave incompatibly in biotite during metasedimentary anatexis, as shown in the reasonable-dynamic suite, could be expected to strongly affect the magnitude of Li enrichment in granites. However, our results show that the choice of Li partitioning model for non-fluorinated biotite has a relatively small impact on granitic Li enrichment (Fig. 4a vs b). The reason for this behaviour is that fractional crystallisation, not melting, generates the most substantial increase in Li enrichment above source concentrations, and biotite is produced only in small amounts (<6 wt%) during modelled fractional crystallisation. Therefore, any small predicted difference in melt Li enrichment generated during partial melting due to Li being mildly compatible or incompatible in non-fluorinated biotite (Fig. 3a) is overwhelmed during crystallisation, especially at high degrees of fractionation.

The importance of partition coefficient choice for generally Li-poor phases (e.g. quartz and feldspar) in controlling granitic Li budgets has been poorly explored in the literature. Since these phases are volumetrically dominant in metasedimentary source rocks and resultant granites, models of Li enrichment will be highly sensitive to these choices. Some recent studies modelling Li enrichment within granitic systems6,10 have used a \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) of either 0 or ≤0.02 for quartz, plagioclase, and alkali feldspar. However, these values fall at or below the lower limit of published ranges of relevant \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) for these phases (Supplementary Table 1). In our minimum \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) (and therefore maximum Li enrichment) scenario, we do not consider any phases to have \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) = 0 if non-zero (even if small) contextually-applicable values are reported in the literature. The use of \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) = 0 for volumetrically dominant phases results in an overestimation of Li enrichment at a given melt fraction during both melting and fractional crystallisation, and can highly skew the conclusions drawn from modelling. For example, a recent model6 of two-stage melting does not reach the required Li concentrations if our reasonable, or even minimum, \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) values are applied.

Rather, this study suggests that fluorinated biotite may be responsible for the petrogenesis of some Li-rich granites due to the substantial \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) increase12, and the role of fluorine in facilitating melt extraction after extreme degrees of fractional crystallisation by reducing melt viscosity44,45. This model also explains the common occurrence of fluorine-rich phases in Li-rich granites37,38,39, and the presence of Li-rich granites as late-stage orogenic melts49 due to the timescales of crustal heating50 coupled with the higher-temperature stability of fluorinated biotite41,42,43. Fluorine concentrations are known to be variable in metasedimentary rocks, with localised order of magnitude variations due to processes such as hydrothermal alteration, metasomatism (potentially from mantle-derived fluids), weathering, or fluorine-rich units such as phosphate shales and evaporites51. In addition, evidence of fluorine enrichment in biotite during prograde metamorphism is reported from a number of localities globally52. Therefore, it seems feasible that source rocks of granites, including those for the Cornubian granite batholith, could locally contain regions of fluorine enrichment, hosted within biotite at near-solidus conditions.

Methods

Major-element modelling

Phase equilibria modelling was performed in the ten-component Na2O-CaO-K2O-FeO-MgO-Al2O3-SiO2-H2O-TiO2-Fe2O3 (NCKFMASHTO) model system, using MAGEMin21 (version 1.6.9; Feb 2025) with the thermodynamic dataset ds6.3653. Composition-dependent equations of state (x-eos) used were: silicate melt [liq_G25w], aqueous fluid [fl_G25], biotite [bi_G25] and cordierite [cd_G25] from ref. 54; garnet [g_W24], clinopyroxene [cpx_W24], orthopyroxene [opx_W24] and ilmenite [ilm_W24] from ref. 55; spinel [spl_T21] from ref. 56; plagioclase and alkali feldspar [fsp_H22] from ref. 57; muscovite [mu_W14] from ref. 58; amphibole [amp_G16] from ref. 59; and epidote [ep_H11] from ref. 53. An average greywacke composition (average greywacke D22; calculated from 12 Precambrian greywackes) was used as the starting composition for modelling (Supplementary Data 1), consistent with the inferred age and composition of the source rock for the Cornubian granite batholith16. Modelling was also performed with an additional global greywacke composition (average greywacke A22), and an average pelite composition23, to test the sensitivity of our results to variable bulk compositions (Supplementary Data 1). The bulk compositions were converted to the model system (Supplementary Data 2) by ignoring elements outside the model system (e.g. minor P2O5 and MnO). An apatite correction was not performed as the proportion of apatite to other P-bearing phases, such as monazite, is unknown. This approach is unlikely to impact our results as the reported P2O5 is low (0.14 wt%). The lack of MnO in the model system (NCKFMASHTO) likely results in an underestimation of garnet stability (e.g. ref. 60). However, as the values for \({D}_{{{{\rm{Li}}}}}^{{{{\rm{garnet/melt}}}}}\) are relatively small (≤0.12; Supplementary Table 1), and the amount of garnet stabilised by MnO is minor60, we consider that this effect is of limited importance, and that is was more important to use the latest set of x-eos, which are calibrated in a Mn-free system. For all bulk compositions, water contents were set to ensure a minimally fluid-saturated solidus for the pressure of interest (2–12 kbar), consistent with the general lack of free fluid in the continental crust, aside from in the shallowest few kilometres61. The xFe3+ of the average greywacke A and average pelite compositions were recalculated to 0.20 (from 0.29 and 0.23, respectively) to match the average Precambrian greywacke for more direct comparison. The slight reduction in reported values also negates the likely minor overestimation of ferric iron23, e.g. due to surface oxidation. We also include a T-xFe3+ pseudosection in the supplementary information (Supplementary Fig. 10) showing the minimal change in mineral assemblage of our primary modelled metagreywacke composition with variation of xFe3+ between 0.002 and 0.50 at 8 kbar.

Our partial melting model uses a stepwise-batch approach, with a melt extraction fraction of 7 vol%, following the melt connectivity threshold24. Our melting model was run at 1 °C temperature steps, with the first step over 7 vol% melt triggering melt extraction. Therefore, some melt extraction steps were slightly larger than 7 vol%, but none were smaller. For model simplicity, we considered that all of the generated melt at each step is fully extracted. As the focus of the partial melting modelling is on the temperature range between the wet solidus and biotite-out, the number of melt extraction steps for each composition was controlled by biotite stability, with the final melt extraction step at the next 7 vol% melt after biotite-out.

For fractional crystallisation models, the starting composition and starting temperature were taken from the stepwise-batch melting results for the pressure/melt extraction step of interest. The melt was cooled with 1 °C steps and the crystallised solid completely removed at each step, up until the point of fluid exsolution. After this point, we switched to 0.04 °C temperature steps, as crystallisation occurs rapidly over only a few degrees when fluid is a free-phase in the assemblage.

Partition coefficient compilation

As discussed in the main text, Li enrichment was calculated using different suites of published mineral/melt partition coefficients (see Supplementary Table 1). Partition coefficients were initially found using the GERM (Geochemical Earth Reference Model) partition coefficients database, accessed in May 2025. This compilation was supplemented by a literature search. Where there were multiple partition coefficients available for a given phase, we selected those for granitic and/or rhyolitic systems. Within this compilation, we then extracted the minimum, maximum and reasonable values, with the most reasonable value based on having the closest pressure, temperature and compositional match to our case study (see Supplementary Table 1 for further details).

Where a range of similarly suitable matches were identified, we selected the intermediate value. Additionally, where a range of values are published within a single study, the mean of all values is taken for natural studies, and the median is taken of all values for experimental studies. This difference in approach reflects that natural studies tend to test minerals of similar compositions, whereas experimental study values often vary based on constrained differences in mineral compositions (e.g. garnet pyrope content, ref. 62). One exception is the values for \({D}_{{{{\rm{Li}}}}}^{{{{\rm{cordierite/melt}}}}}\), as the reasonable and minimum values are represented by a temperature-dependent dynamic model63. Dynamic partitioning models are preferable to static values as they allow for model conditions, such as temperature, to influence partitioning behaviour, as is likely more applicable in nature. The dynamic partitioning model was also used as the minimum value, as no other studies were found that record a \({D}_{{{{\rm{Li}}}}}^{{{{\rm{cordierite/melt}}}}}\) value that is lower than those values determined by the dynamic model (which range from 0.14 to 0.44). As the dynamic partitioning model for cordierite is only calibrated between 700–850 °C, \({D}_{{{{\rm{Li}}}}}^{{{{\rm{cordierite/melt}}}}}\) above 850 °C were fixed to the calibrated 850 °C value of 0.14.

No \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) values were found for rutile, kyanite, sillimanite and epidote, with previous studies having assumed their partition coefficients to be 0, or not considered the phase. These phases are present in accessory proportions in this study. However, rather than normalising these phases out, which would effectively be attributing the bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) of the system to those phases, we used \({D}_{{{{\rm{Li}}}}}^{{{{\rm{mineral/melt}}}}}\) values of 0 to better reflect their contribution of the Li enrichment budget. This choice has a negligible effect in this study due to the minor abundances of these phases in our models, but it would be an important consideration for more rutile or aluminosilicate-rich rocks.

Partition coefficients from ref. 64 were excluded as they are notably different to other values from similar compositions, which we attribute to being based on measurements from melanosomes that were not in equilibrium with the co-existing leucosome. In some instances, no studies of felsic bulk composition were available, in which case we used the next closest match (e.g. intermediate rather than mafic). These values are denoted with a cross in Supplementary Table 1. Only a single \({D}_{{{{\rm{Li}}}}}^{{{{\rm{fluid/melt}}}}}\) value is used, although large variations exist in published values primarily due to fluid salinity variability. We use a \({D}_{{{{\rm{Li}}}}}^{{{{\rm{fluid/melt}}}}}\) value of 0.65 (ref. 27; mean value of two rhyolitic compositions) as this value is for magmatic conditions analogous to those we are modelling. Finally, as discussed in the text, we consider three different partition coefficients for biotite: a static value of 1.67 (ref. 11); a dynamic (composition- and temperature-dependent) \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) (ref. 12) which ranges from 0.29–0.37 for the conditions within our model during metasedimentary melting at 8 kbar, up to a maximum of 1.18 during fractionation at 4 kbar; and a fluorinated-biotite \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) value of 32.8 (ref. 12). We do not separately consider minimum and maximum values for biotite.

The dynamic partitioning model used for \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) in this study is the model recommended for use in phase equilibria modelling12. The dynamic partitioning model is calibrated across a wide compositional range, including granitoid compositions. The static \({D}_{{{{\rm{Li}}}}}^{{{{\rm{biotite/melt}}}}}\) value for F-rich biotite (32.8), from the same study, was chosen to be representative of that for high-F biotite and therefore taken from biotite with greater than 1 F anion per formula unit. The value chosen is from an experiment using a phonolite melt composition, but other melt compositions show the same trends for other elements (e.g. high F biotite in granitic systems have similar Dbiotite/melt to high F biotite in phonolitic systems; see Fig. S4 in ref. 12), suggesting that the melt composition is of subordinate importance to the more influential factor of biotite composition.

Although F is not in the model system, in Supplementary Fig. 11, we use the empirical model of ref. 12 for F-OH exchange in biotite to predict how changing metasedimentary biotite composition in our primary modelled scenario is conducive to developing fluorinated biotite. Supplementary Fig. 11 shows an increase in the predicted compatibility of F within biotite with increasing temperature in our model, controlled by compositional changes such as increasing biotite \({X}_{{{{\rm{Si}}}}}^{{{{\rm{T}}}}}\) and \({X}_{{{{\rm{Mg}}}}}^{{{{\rm{M}}}}}\). This result is consistent with the reported role of prograde metamorphism in increasing biotite F concentrations52, alongside our hypothesis of F-rich biotite being present at higher temperatures and contributing to Li- and F-rich melt generation.

Lithium enrichment modelling

At each temperature step during melting or crystallisation, the phase abundances (as weight fractions) calculated from our phase equilibria modelling were used to calculate a bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) (or \({D}_{{{{\rm{Li}}}}}^{{{{\rm{(solid+fluid)/melt}}}}}\) during fractionation with the presence of free fluid, noted as \({D}_{{{{\rm{Li}}}}}^{{{{{\rm{solid}}}}}^{* }/{{{\rm{melt}}}}}\) in the text). During batch melting, Li enrichment in the melt (concentrationmelt/concentrationinitial) was calculated by 1/(F+D(1-F)), where D is bulk \({D}_{{{{\rm{Li}}}}}^{{{{\rm{solid}}}}/{{{\rm{melt}}}}}\) and F is the weight fraction of liquid present. The corresponding Li depletion in the solid was calculated by mass balance. For a stepwise-batch system where the residual solid after each melt extraction is progressively depleted in Li, the Li enrichment of the melt after a melt extraction step (relative to the initial unmelted protolith) was calculated by multiplying batch melting enrichment by the fraction of initial Li remaining in the residual solid at the point of the previous melt extraction.

Li enrichment during fractional crystallisation was calculated following the approach of ref. 65. Enrichment at each temperature step was considered as a local batch calculation, giving Li enrichment relative to the previous crystallisation step. Therefore, the overall melt enrichment during fractional crystallisation (relative to the initial non-fractionated partial melt) for each step, j, was calculated as 1/(F+D(1-F))j × 1/(F+D(1-F))j−1. This overall crystallisation enrichment was then multiplied by the enrichment factor of the partial melt relative to the protolith for the relevant extraction step, giving overall Li enrichment during crystallisation relative to the initial source protolith.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.