Tourmalines are accessory minerals occurring in most crustal environments. They mainly appear in igneous, metamorphic, or metasomatic contexts and occasionally in sedimentary settings. Very common in granitic pegmatites, tourmaline crystals often reach pluri-centimetric to metric sizes, unmatched qualities, and beautiful colours, therefore making them highly prized on the gemstone market. But, beyond this economic value, their real interest lies in their ability to provide a wide range of information about the geological context from which they originate. Some tourmaline solid solutions like the schorl–elbaite series can, for example, be used as differentiation markers within lithium-enriched pegmatites. Moreover, being the most widespread borosilicate phase in nature, they allow extensive comprehension of the phenomenon leading to the formation of their geological deposits.

The nomenclature and classification of the tourmaline supergroup, developed by Hawthorne and Henry (1999) and approved by the Commission on New Minerals, Nomenclature and Classification of the International Mineralogical Association (IMA-CNMNC), define tourmalines as trigonal borosilicates with a R3m space group (Buerger and Parrish, 1937; Hamburger and Buerger, 1948) and with a general formula XY3Z6[T6O18](BO3)3V3W (Hawthorne and Henry, 1999), where X= Na, Ca, K, □; Y= Li, Mg, Fe²⁺, Mn²⁺, Zn, Al, Cr³⁺, V³⁺, Fe³⁺, Ti⁴⁺; Z= Mg, Al, Fe²⁺, Fe³⁺, Cr³⁺, V³⁺; T= Si, Al, B; B= B; V= OH, O; and W= OH, F, O (Henry et al., 2011). The tourmaline structure allows a wide range of substitutions, leading to numerous end-members. Most of the compositional variations occur on the X, Y, Z, W, and V sites, whereas the T and B sites are less likely to accept different cations (Ertl et al., 2018). Amongst the anionic sites, the O3 and O1 positions are occupied by anions from the V and W groups, respectively. Hydrogen atoms H3 and H1 are chemically bonded to the oxygen atoms of these O3 and O1 sites, respectively (Bosi, 2018; Henry and Dutrow, 2018).

Due to their capacity to generate strong economic and scientific interests, tourmalines rank amongst the most extensively studied minerals in the world. The understanding of their highly complex crystal chemistry is refined day by day. This is especially the case for pegmatitic tourmalines from Minas Gerais (Brazil), as the region has proven itself to be a particularly fertile field of experimentation to work on these minerals. This is explained mainly by the wealth in terms of high-quality crystals and in terms of various geological contexts in the region but also by the ease of access to samples formed in these deposits. However, gaps remain concerning some aspects of these minerals, since most publications have been dedicated to the determination of major-element concentrations in crystals from the main deposits of the country. This implies not only that the majority of small deposits are very poorly known in terms of tourmaline crystal chemistry, but also that the trace-element concentrations and behaviours have almost never been studied in most deposits from Minas Gerais.

With the present paper, we attempt to fill this gap by investigating 17 tourmaline crystals from eight pegmatite deposits in Minas Gerais. These samples will be analysed by electron microprobe (EMP) and by laser ablation inductively coupled plasma time-of-flight mass spectrometry (LA-ICP-TOF-MS) to determine their major- and trace-element concentrations and by single-crystal X-ray diffraction (SCXRD) to assess their structural features. Relationships between these data and the geological context will also be discussed.

Covering about 589 000 km² of south-eastern Brazil, the mining state of Minas Gerais (Fig. 1) has been a major tourmaline producer worldwide since the end of the 17th century. Amongst the wide diversity of geological contexts encompassed in this Brazilian state can be found four major units: the tectonic belts of Tocantins and Mantiqueira that are mainly composed of Proterozoic granites and granitoids as well as metamorphic rocks; the São Francisco Craton, defined by sedimentary rocks from the Neoproterozoic and Cretaceous periods; and the Paraná Basin, which gathers sediments from the Ordovician to the Cretaceous (Moreira and Camelier, 1977; Pedrosa-Soares et al., 1994; De Souza et al., 2015).

Minas Gerais owes most of its opulence to the Eastern Brazilian Pegmatite Province (EBPP) (Fig. 1), defined by Paiva (1946) and which mainly extends east of the São Francisco Craton in an orogenic unit called the Araçuaí Orogen (Fig. 1). The 150 000 km² EBPP is therefore characterized by granitic rock formations dating from the late Neoproterozoic to the early Ordovician into which a series of pegmatites are inserted. These rocks are gathered into super-suites named G1, G2, G3, G4, and G5 on the basis of the relation between their establishment and the advancement of the orogeny. G1 includes the pre-collisional formations (630 to 585 Myr), G2 is syn-collisional (585 to 560 Myr), G3 is late- to post-collisional (545 to 520 Myr), and G4 and G5 are post-collisional (535 to 490 Myr) (Pedrosa-Soares et al., 2001a, 2009). The EBPP is itself subdivided into a series of pegmatitic districts, and the main ones are given in Fig. 1.

These pegmatite districts are where most of the gemstone production in Minas Gerais is concentrated. A wide variety of mineral species can be found in important quantities, like beryl, tourmaline, topaz, chrysoberyl, or quartz. In minor quantities, spodumene, amblygonite, brazilianite, or garnets are also represented as well as a multitude of other rarer minerals (Barreto and Bittar, 2010).

There are two types of pegmatites in these Eastern Brazilian Pegmatite Province districts: the anatectic ones which formed by partial melting of the parent rocks (DPA pegmatites, direct products of anatexis; Wise et al., 2022) and the residual ones that crystallized from the residual silicate melts left by the fractional crystallization of the parent granite (RMG pegmatites, residual melts of granite magmatism; Wise et al., 2022). Most of the pegmatites from the Araçuaí, the Ataléia, the Conselheiro Pena, the Espera Feliz, the Padre Paraíso, the Pedra Azul, and the São Jose de Safira districts (Fig. 1) are of RMG type. The Caratinga, Santa Maria de Itabira, and Espírito Santo districts (Fig. 1) are dominated by DPA pegmatites resulting from the partial melting of gneisses. Finally, some hydrothermal deposits from Malacacheta and Santa Maria de Itabira are gathered under the pegmatite term despite the fact that they are not pegmatites s.s. (Correia-Neves et al., 1986; Pedrosa-Soares et al., 2009).

Most of the DPA pegmatites formed during the collisional phase of the Araçuaí Orogen. They are usually associated with migmatites and granulites, and often they constitute corundum and beryl deposits (Correia-Neves et al., 1986; Morteani et al., 2000; Pedrosa-Soares et al., 2009). RMG pegmatites, more abundant in the EBPP, have as parent rocks the different syn- (G2) and post-collisional (G4, G5) granites. They can be called “intragranitic” if, as in the Araçuaí, the Ataléia, the Conselheiro Pena, and the São Jose de Safira districts, they are hosted inside the parent granite or “extragranitic” if they are surrounded by the host rocks, as in the Espera Feliz, the Padre Paraíso, and the Pedra Azul districts. Extragranitic pegmatites are usually considered better gemstone deposits than intragranitic pegmatites (Pedrosa-Soares et al., 2009).

The granitic super-suite G2 mainly encompasses peraluminous (Type S) and some metaluminous granites. Their setup followed the superposition of continental margins during the collisional phase of the Araçuaí Orogen. This superposition led to a thickening of the crust followed by a rise in the temperature conditions at the origin of the production of huge quantities of magmas. Two types of granites were generated: a superficial two-mica granite and a deep granite hosting biotites and garnets. The first of these granites, despite being less abundant than the later one, was favourable to the production of pegmatites like those of Conselheiro Pena. In the meantime, a migmatization of the paragneiss complexes is at the origin of the genesis of pegmatites in the Caratinga, Espirito Santo, and Santa Maria de Itabira districts (Pedrosa-Soares et al., 2009).

The granitic super-suite G4 consists of a series of balloon-shaped plutons with zonations ranging from granites with biotites in the core to muscovite-bearing granites towards the rim. These plutons are capped with pegmatoidic domes. The peraluminous (Type S) granites, sometimes slightly metaluminous, appeared during the post-collisional phase of the Araçuaí Orogen. Several pegmatites from the Araçuaí, São Jose da Safira, and Malacacheta districts derive from these plutons (Pedrosa-Soares et al., 1987; Basílio et al., 2000; Pinto et al., 2001; Pedrosa-Soares et al., 2009).

The granitic super-suite G5 is characterized by granitic, charnockitic, or dioritic balloon-shaped intrusions located at depth (25 km). These plutons, which were established during the post-orogenic phase, are the source of residual pegmatites in the Ataléia, Espera Feliz, Padre Paraíso, Pedra Azul, and Espírito Santo districts (Fig. 1). These pegmatites are usually enriched in biotites but poor in lithium minerals. They form veins of a few metres thick inside the granites and the charnockites (Gandini et al., 2001; Ferreira et al., 2005; Pedrosa-Soares et al., 2009).

Samples were selected in the collections of the Laboratory of Mineralogy, University of Liège (prefix “U” in the sample number), and of the Natural History Museum of Luxembourg (prefixes “WR”, “KM”, and “KF”). Selected tourmalines are constituted by single crystals carefully extracted from larger collection pieces. They were chosen on the basis of their quality, their colour, and their occurrence. Their respective origins, which are known with variable degrees of precision, as well as their macroscopic descriptions and the associated mineral assemblages, are detailed in Table 1. Accurate localization of most pegmatites can be found in Cassedanne and Philippo (2015). A preliminary optical description has been conducted for each sample under a binocular microscope coupled with a microscope-adapted numerical camera for capturing images. Magnifications used vary between 6× and 9×, depending on the sample. Results are gathered in Table 1 and Fig. 2.

Tourmaline fragments of about 2 to 3 mm in diameter were then handpicked, sealed in epoxy resin, and polished to perform electron-microprobe analysis (EMPA) at the Geosciences Department of Oslo University. The instrument used was a CAMECA SX100 electron microprobe, working under an acceleration voltage of 15 kV, a beam current of 15 nA, and a beam diameter from 5 to 10 µm and equipped with five WDS (wavelength-dispersive X-ray spectroscopy) spectrometers with the following diffracting crystals: PET (pentaerythritol), LPET (pentaerythritol – large crystal), TAP (thallium acid phthalate), LTAP (thallium acid phthalate – large crystal), and LLIF (lithium fluoride – large crystal). Standards used for the calibration were wollastonite (Si, Ca), pyrophanite (Ti, Mn), orthoclase (K), albite (Na), fluorite (F), and a series of synthetic standards – Al₂O₃, MgO, and Fe. All analytic lines were Kα, and detection limits were 2000 ppm (F), 615 ppm (Fe, Mn), 400 ppm (Ca), 335 ppm (Na, Si, Al), 265 ppm (Mg, Ti), and 235 ppm (K). Five measurements were performed on each sample, and their average value was used during data treatment.

Since boron and lithium are not determined by the EMP analyses, they were analysed by LA-ICP-TOF-MS (see below). In order to verify the reliability of these values, they were compared to the B₂O₃ concentrations calculated by considering an ideal value of three atoms per formula unit (apfu) on the B site and to the Li₂O concentrations necessary to fill the 3 apfu on the Y site. The populations of the different crystallographic sites were calculated on the basis of 15 Y+Z+T cations (Table 2) based on the method developed by Henry et al. (2011). The totals on the X site did not reach the ideal value, which can be explained by the presence of vacancies in the crystal structure. The H₂O levels were calculated assuming that the V site was filled by 3 OH groups pfu and that the W site was occupied by 1 (OH,F,O) anion pfu. Considering that iron is mostly divalent (except in sample KM132), the OH content was calculated as 62 - total cationic charges - F, taking into account the positive charges corresponding to the ideal boron content.

The polished sections used for EMP analyses were sent to the Swiss Gemmological Institute SSEF of Basel to quantify trace elements. A LA-ICP-TOF-MS instrument, also nicknamed GemTOF, has been used for this purpose. It includes an ArF excimer laser of 193 nm equipped with a TWoVol2 ablation chamber (NWR193UC, Elemental Scientific Lasers, USA). The laser is coupled with a commercial ICP-TOF-MS unit (icpTOF, Tofwerk AG, Switzerland), modified from an ICP-Q-MS (Q denotes quadrupole) optimized unit (iCAP Q, Thermo Fisher Scientific). The ICP power ranges from 1300 to 1350 W, and the instrument is able to measure a complete mass spectrum from ⁷Li⁺ to ²³⁸U⁺, thus allowing the determination of all elements of the periodic table, except hydrogen and elements heavier than uranium. Measurements have been conducted under the following analytic conditions: two points have been measured in each fragment, and the average value of the two measurements will be used in the discussion below. Each point underwent laser ablation of 50 µm in diameter with a repetition rate of 20 Hz and a fluence of 5.6 J cm⁻². Helium has been used as carrier gas (0.8 to 0.85 L min⁻¹). Five pre-ablation cleaning shots were realized before each measurement (30 s of background measurement, 30 s of ablation). Every element between lithium and uranium was measured simultaneously. The quantification was realized with a MATLAB script developed at the SSEF (Wang and Krzemnicki, 2021). NIST610 and NIST612 glasses were used as standards and underwent 75 µm laser ablation in the same conditions as previously described. Data obtained through LA-ICP-TOF-MS analyses, concerning trace elements and rare earth element (REE) concentrations in tourmalines from Minas Gerais, are given in Tables 3 and 4, respectively. Elements with concentrations below the detection limit in all samples are not included in the table for ease of reading.

The samples finally underwent structural analyses by single-crystal X-ray diffraction methods. Fragments were carefully handpicked for each tourmaline and were then glued on a glass capillary using a transparent polish. The whole mounting was introduced in a Rigaku Xcalibur 4-circle diffractometer using a kappa geometry, a CCD (charge-coupled device) EOS detector, and MoKα (λ=0.7107 Å) radiation. Several pre-experiments were sometimes necessary to choose the best single crystal. Data reduction was performed with the RED program (Oxford diffraction, 2007) by subtracting the background from the raw intensities and correcting the Lorentz and polarization factors, as well as the absorption. Structure refinements were achieved with the OLEX² suite of programs, which uses SHELXS to solve the structure (direct methods) and then SHELXL to refine the model (Sheldrick, 2015). The VESTA program was finally used to obtain a drawing of the three-dimensional structural models. The KM058 tourmaline sample was zoned from the core (pink) to the rim (green), and an analysis was conducted for each zone. Selected fragments were named KM058_R and KM058_V, respectively. More details concerning data collection and crystal structure refinements are given in Table 5.

Average major-element compositions for the tourmaline samples, calculated from the electron-microprobe and LA-ICP-TOF-MS analyses, are shown in Table 2. The totals reach values close to 100 wt % (98.68 wt % to 99.99 wt %), which attests to the reliability of the calculation method. The major-element compositions (Table 2) are very close to those usually given in the literature for tourmalines from the Eastern Brazilian Pegmatite Province (Bilal et al., 1998; Federico et al., 1998; Bosi and Lucchesi, 2004; Bosi et al., 2005; Andreozzi et al., 2025).

In order to accurately determine the cation populations on each crystallographic site of the tourmaline samples, cations were distributed among those sites in order to reach the best fits between several structural (see below) and chemical parameters: the calculated site scattering (CSS) and the refined site scattering (RSS) values, the mean bond lengths (MBLs) and the calculated bond lengths (CBLs), the bond-valence sums (BVSs), and the theoretical valence (TV). The results of these distributions are shown in Table 6, and generally, the good agreement between those parameters confirms the validity of the assigned site populations (Table S1 in the Supplement). The B site is fully occupied by boron, the T site is mainly occupied by Si and sometimes by minor amounts of B, and the Z site is mainly occupied by Al with sometimes minor amounts of Fe³⁺ (sample KM132). The X site contains vacancies, Na, K, and Ca, and the Y site is occupied by Li, Al, Fe²⁺, and minor amounts of Mn²⁺, Ti, Mg, and Zn. In the two samples KM120 and WR045, a significant disorder between Li and Ca on the X and Y sites has been observed, necessary to reach a good fit between the RSS and the CSS values (Table S1).

Occupancies of the X crystallographic sites (Table 6) show that they are dominated by large amounts of Na in almost every sample. These crystals therefore belong to the alkali group (Fig. 3a) defined by the new classification of tourmalines developed by Hawthorne and Henry (1999) and reviewed by Henry et al. (2011). These results are coherent with those obtained by Bosi and Lucchesi (2004), Bosi et al. (2005), and Andreozzi et al. (2025) on tourmalines from the Cruzeiro mine. In most of our samples, significant numbers of vacancies occur on the X site, leading to an alignment of the compositions along the solid solution between the alkali and the X-vacant group (Fig. 3a). Three crystals, however, escape this rule: first, the KM132 sample from the Clementina mine, in which vacancies (0.592□ pfu) exceed the proportion of (Na + Ca + K) on the X site – this sample consequently belongs to the X-vacant group (Table 6); secondly, the WR045 tourmaline from Urucum, which is still Na-dominant but displays a strong enrichment in Ca (0.216 apfu) and a very low number of vacancies; finally, the sample KM120 from Baixão, which shows an intermediate behaviour characterized by an enrichment in Ca (0.150 apfu) and in vacancies (0.288 pfu), coupled to a slight depletion in Na (Fig. 3a).

In order to calculate the W-site populations, the V-site occupancies were constrained to 3 OH⁻ pfu, following the consensus from the literature (Grice and Ercit, 1993; Hawthorne and Henry, 1999). The occupancies of the W sites are relatively heterogenous, with F concentrations ranging from 0.111 to 0.689 apfu and OH contents between 0.228 and 0.594 apfu (Table 2). A majority of our samples belong to the fluor series, whereas many other samples can be classified in the hydroxy series (Fig. 3b). It is noteworthy that these OH-rich tourmalines show a composition very close to that of pure elbaite and therefore occur on the same point in Fig. 3c. KM132, KF081, and WR044 show different behaviours, since they contain significant amounts of O²⁻ on the W sites (between 0.238 and 0.366 O pfu, Fig. 3b). The presence of the three types of tourmaline series amongst the samples from Minas Gerais had already been highlighted by previous studies (Bosi and Lucchesi, 2004).

Previous authors also agree on the fact that in most tourmalines from the Eastern Brazilian Pegmatite Province, the boron contents on the B sites can be considered very close to 3 apfu. However, in our case, boron has been directly measured by LA-ICP-TOF-MS, and the resulting data (Table 2) are consistently below the stoichiometric values (2.70 to 2.84 B pfu). Similar results were obtained by Bosi et al. (2005) when they measured, by secondary ion mass spectrometry (SIMS), the B₂O₃ contents of tourmalines from Cruzeiro. According to these authors, such low B contents can be attributed to analytical errors, which is also certainly the case in our study. This assumption is further supported by the fact that the 〈B–O〉 bond lengths measured in our samples (see below) are within the range of those calculated for boron in triangular coordination in tourmalines (Pieczka, 1999). In the final assigned site populations (Table 6), we consequently considered the theoretical occupancy of 3 B pfu on the triangular sites.

Tetrahedral T sites often display stoichiometric amounts of Si (6.000 apfu), but a careful examination of Si–O average bond lengths (see below) indicates that minor amounts of boron certainly occupy this site, as already described by Ertl et al. (2018). Starting from these observed Si–O mean bond distances, we recalculated the B contents of the tetrahedral sites, according to the correlation proposed by these authors, and we obtained values ranging from 0.020 to 0.300 B pfu (Table 6).

Most of the major compositional variations affecting the tourmalines of this study take place on the Y crystallographic sites, which can incorporate various types of cations (Table 6). Ti and Mg are almost absent from our samples, with Ti appearing sporadically in very small proportions (up to 0.03 apfu) and Mg reaching maximum values of 0.21 wt % MgO, indicating that the dravitic component is low amongst tourmalines from the EBPP (Fig. 3c). Higher values are obtained for ^YMn²⁺ (Table 6), with concentrations from 0.009 to 0.291 apfu. Although manganese does not become the dominant cation on the Y sites of these Mn²⁺-rich samples, its presence indicates a non-negligible tsilaisitic component (samples KM066, KM076, WR037, WR044, WR045, and U20662).

Al, Li, and Fe² are, by far, the most represented cations on the octahedral sites (Table 6). The dominant cation on these sites is Al in most samples, with values ranging from 0.915 to 1.608 apfu. In some cases (WR065, KM066, KM120, WR045), Li can also be the dominant cation, sometime reaching 1.491 apfu. KM132 is the only example of ^YFe-dominant (Fe = 1.377 apfu) tourmaline amongst our samples; this Fe enrichment is accompanied by strong depletion in lithium (Li = 0.147 apfu). Based on these data, the tourmaline species of this study have been identified using the ternary (Al3++Li⁺)–Mg²⁺–Fe²⁺ diagram developed by Henry et al. (2011) to classify the tourmaline species from the alkali and X-vacant groups.

The results are given in Fig. 3c, and the identified species are given in Table 6. Most of our tourmalines are fluor-elbaites or elbaites, where similar proportions of Al and Li share the Y positions. Elbaites are located close to the elbaitic end-member, with Fe²⁺ strongly depleted and the schorlitic component negligible (Fig. 3c). On the other hand, fluor-elbaites tend to align themselves along the fluor-elbaite to fluor-schorl solid solution. The only exception is sample KM132, which shows a dominance of vacancies on the X site (Fig. 3a), as well as of Fe²⁺ on the Y site (Fig. 3c). Such a composition indicates that this sample can be considered a rossmanite close to the foitite composition field (Fig. 3c).

Crystal structure refinements were performed in the R3m space group. Cation occupancies for Si on the T site, as well as for B on the B site, were constrained to 1, as well as oxygen occupancies. Occupancies for Na on the X site, for Al or Fe on theY site, and for Al on the Z site, were generally refined, except for the Z-site occupancies, which were constrained to 1 when the preliminary refinements indicate that this site was fully occupied. For the last refinement cycles, all atoms except hydrogens were refined with anisotropic displacement parameters. Atomic coordinates and occupancy factors and anisotropic displacement parameters, as well as bond length distances and interatomic angles, are available in the Supplement (Tables S2 to S4). Average bond lengths are given in Table 7.

Unit-cell parameters (Table 5) are in the ranges usually admitted for tourmalines (15.7 Å <a<16.2 Å; 7.05<c<7.25 Å; Hawthorne and Dirlam, 2011), and they are similar to the values obtained for Brazilian tourmalines of the elbaite–schorl solid solution (Bosi et al., 2005; Gatta et al., 2012; Bosi and Lucchesi, 2004). The a parameters of our samples vary between 15.8187(3) and 15.9309(3) Å, whereas their c parameters range from 7.0924(2) to 7.1279(2) Å. None of the sample reaches the schorl–dravite solid solution reference values. The KM132 crystal is, by far, the sample showing the most important schorlitic component. KM117 and WR045 are located in the centre of the schorl–elbaite solid solution, whereas WR037, KM075, and KM076 have weak schorlitic components but still keep their distances from the elbaitic pole. The rest of the samples are relatively close to the elbaite end-member (Fig. 3). The specific case of the zoned KM058 crystal is interesting as it displays an evolution of the c parameter. The pink part of the sample shows unit-cell parameters close to those of the elbaitic end-member, while the green part of the sample shows unit-cell parameters close to those of an elbaite, with a strong schorlitic component. Four tourmalines (U13757, KF081, KM118, and KM066) also display particularly low a and c parameters, below the reference values of elbaites. In the case of the first three samples, this behaviour can be explained by the presence of high amounts of aluminium, with a small ionic radius, on the Y site of tourmaline, but this explanation does not apply to the KM066 crystal, which is enriched in Mn²⁺ and Fe²⁺ on this site.

The main feature of the tourmaline structure (Fig. 4) is the presence of six [TO₄] tetrahedra linked by their corners to form [T6O₁₈] rings, perpendicular to the c axis. The presence of these rings implies that tourmalines belong to the cyclosilicates sub-class (Mills et al., 2009). The rings are linked to three [YO₆] octahedra bonded together by their edges to form trimers. The second type of octahedra hosts the Z sites and is bonded to the periphery of these trimers by the edges. B sites are in triangular coordination, therefore forming [BO₃] groups that are linked by their corners to the two types of octahedra. This arrangement therefore leaves cavities in the crystal structure that are occupied by the X sites, in 9-fold coordination. These positions are located along the c axis, outside the plan containing the [T6O₁₈] rings. The [XO₉] polyhedra are usually occupied by cations with large ionic radius such as Na or Ca, but they are also often empty. These polyhedra are strongly distorted (Fig. 5a). The W anions occupy the O1 sites, located along the c axis, in a central position, and every V anion located in O3 is shared between a [YO₆] and a [ZO₆] octahedra (Hawthorne and Dirlam, 2011). The unidirectional orientation of the [TO₄] tetrahedra constituting the rings implies that tourmalines are both non-centrosymmetric and polar. These minerals are, therefore, pyroelectric and piezoelectric (Hawthorne and Dirlam, 2011). The structural formula, deduced from these data, can be written XY3Z6[T6O18](BO3)3V3W (Hawthorne and Henry, 1999).

Table 7 indicates that the measured mean bond lengths are relatively coherent for each site, among the different samples. In the B sites of these tourmalines, the B–O bond lengths show variations of a few thousandths of ångströms, indicating that these sites are geometrically stable. Their coordination polyhedra are slightly distorted, compared to a perfect triangle, as one of the three B–O bonds (Table S3) is 0.020 to 0.040 Å shorter than the two other bonds. This distortion also produces a significant tightening of the O8–B–O8 interatomic angle (Table S3). Bond length distortion values (BLD, Table 7), allowing an estimation of the crystallographic site deformation based on the measured bond lengths, were calculated using a method similar to the one used by Hatert et al. (2004) and based on the Renner and Lehmann (1986) approach. For the triangular BO₃ sites, the BLD factors vary between 0.49 % and 1.43 %, which confirms the low deformation degree of the polyhedra. These parameters are even lower for the tetrahedral TO₄ positions, thus implying that these sites are close to the perfect tetrahedral shape. Based on their BLD values of around 2.37 % and 1.79 %, respectively, the Y and Z sites appear to be more distorted (Fig. 5b, c; Table 7). The XO₉ coordination polyhedron, with BLD values above 5.19 %, appears to be the most distorted in the tourmaline structures of our samples (Fig. 5a, Table 7).

The trace-element contents of tourmalines from Eastern Brazilian Pegmatite Province (Table 3) were normalized to the Upper Continental Crust (Rudnick and Gao, 2014). Variations in these element ratios are shown in Fig. 6a, where they are gathered based on their dominant valence states (monovalent to pentavalent) and then by increasing ionic radii (VI coordinated to O; Shannon, 1976), according to a procedure defined by Trumbull et al. (2018). These variations are relatively strong, since the diagram obviously shows a sawtooth pattern. Such a pattern is difficult to decipher, but certainly both the pegmatitic fractionation and the crystal–chemical constraints influence the observed variations.

Twelve elements generally show concentrations above those of the Upper Continental Crust, namely Be, Mn, Zn, Cd, Pb, Ga, In, Bi, Ge, Sn, Sb, and Ta (Fig. 6a). Some of these elements are clearly enriched by pegmatite fractionation, for example Be, Ga, Sn, and Ta (Simmons and Webber, 2008). On the other hand, divalent cations such as Mn (effective ionic radius, e.i.r. = 0.830 Å; Shannon, 1976), Zn (e.i.r. = 0.740 Å), or Cd (e.i.r. = 0.95 Å), as well as trivalent cations such as In (e.i.r.=0.800 Å), could easily substitute for Fe²⁺ (e.i.r.=0.780 Å) on the octahedral Y site. Larger cations such as Bi (e.i.r.=1.17 Å) and Pb (e.i.r. = 1.19 Å) could substitute for Na (e.i.r.=1.24 Å) on the X [9]-fold site, and smaller cations such as Ge (e.i.r.=0.530 Å) could replace Al (e.i.r.=0.535 Å) on the Z site. The incorporation of Sb (e.i.r.=0.60 Å) could take place on the Z site, but recent studies on melt inclusions by Thomas and Davidson (2016) have shown that this element was also enriched by fractionation in pegmatites.

It is extremely difficult to unravel the behaviour of independent samples in Fig. 6a due to a strong overlap of their geochemical trends. Sample WR045, which appears to be particularly enriched in Ca due to its evolution towards the calcium-rich end-member liddicoatite (Tables 2, 6), is also the richest in other elements such as Sr, Sc, Y, and Sb. The substitution of Ca [9] (e.i.r.=1.18 Å) by Sr (e.i.r.=1.31 Å) and Y (e.i.r.=1.075 Å) is relatively easy due to their close ionic radii, and this steric effect also allows the incorporation of light REEs from La (e.i.r.=1.216) to Dy (e.i.r.=1.083) in this tourmaline sample (Fig. 6b). The presence of REEs in the crystal structure of tourmalines is certainly mostly constrained by geochemical factors, explaining the positive correlation between the Ca and REE contents of tourmalines. As shown in Table 6 and Fig. 6b, the most calcic sample WR045 (1.90 wt % CaO) is the richest in REEs and the less calcic sample KM132 (0.01 wt % CaO) is the poorest in REEs (except sample KM076, which is basically depleted in REEs; Table 3). The calcium and REE enrichment of sample WR045 is obviously linked to a late calcium metasomatic stage, well known to affect some granitic pegmatites (Martin and De Vito, 2014; Pieczka et al., 2019) and responsible for the crystallization of other Ca-rich species associated with these samples, such as fluorapatite and stokesite (Table 1).

The present paper gives a detailed crystal–chemical investigation of 17 tourmaline samples from different pegmatites of Minas Gerais, Brazil. The combined approach, including single-crystal X-ray data, electron-microprobe, and GemTOF analyses, gives new insights in the complex crystal chemistry of this mineral group.

The samples generally belong to the schorl–elbaite and the fluor-schorl to fluor-elbaite series, but some samples clearly show different chemical compositions, for example sample WR045, which is particularly enriched in calcium, and sample KM132, which corresponds to a rossmanite. Major observed substitution mechanisms are from elbaite to schorl (0.5 ^YAl3++0.5YLi+=YFe²⁺), from elbaite to rossmanite (^XNa++0.5YLi+=X□+0.5YAl³⁺), from elbaite to liddicoatite (^XNa++0.5YAl3+=XCa2++0.5YLi⁺), or from schorl to oxy-foitite (0.5 ^XNa++YFe2++0.5W(OH)-=0.5X□+YAl3++0.5WO²⁻). The mechanism ^XNa++YFe2+=X□+YAl³⁺ explains the observed positive correlation between Al and the vacancy content of the X site.

Most chemical and crystal–chemical correlations had already been established in the abundant literature on tourmalines. However, the detailed interpretations of the trace-element behaviours and comparison with their crystal–chemical data allow a better understanding of the boundaries between the pure geochemical pegmatitic influence and the crystal–chemical constraints. Simple correlations, such as those between the (Li + Al) contents and the unit-cell parameters, will certainly be useful to semi-qualitatively determine the lithium content of these minerals using simple X-ray powder diffraction data.