The dependence of the spectral band intensities on the crystal orientation relative to the direction of polarization for the incident laser beam has been shown for many minerals, e.g., phlogopite (Loh, 1973; Tlili et al., 1989), pyrite (Bryant et al., 2018), and zircon (Kolesov et al., 2001). The crystal structures of sidorenkite, bradleyite, and bonshtedtite contain valence bonds of different lengths between cations (Na+, Mn2+, Mg2+, Fe2+) and oxygen atoms in [PO4]3- and [CO3]2- groups (Krivovichev et al., 2013; Tkhy et al., 1984). We suggest that change in band intensity for modes depends on the orientation of chemical bonds and vibrational interactions with the electronic state of cations in crystals. The same has been revealed for SiO4 in zircon (Kolesov et al., 2001). In carbonophosphates, the crystallographic-orientation effect is most evident in the variability in the intensity ratios of the main modes υ1(P–O) and υ1(C–O). We determined the relative orientation of the sidorenkite crystal using EBSD (Fig. 4a). As it turned out, the intensity of the υ1(C–O) band increases when the laser polarization direction coincides with the b axis. With such a crystallographic orientation, the υ1(C–O) band is the most intense and exceeds the υ1(P–O) band by a factor of 1.5–2 (Fig. 4b). When the polarization direction deviates from the b axis, the intensity of the υ1(C–O) band is markedly reduced (Fig. 4c). We obtained similar results from the EBSD study for all other carbonophosphates. Additional equipment is required to elucidate the nature of the relationship between the mode intensity and crystallographic orientation (e.g., Fleck et al., 2020). In this study, we approached this issue only phenomenologically. The spectra in which the intensity ratio of the main peak from υ1(C–O) to the main peak from υ1(P–O) are maximal (Fig. 4b) are hereafter referred to as “A type”. The spectra with the minimum ratio of these peaks (Fig. 4c) are hereafter designated as “B type”. Note that we registered both A and B spectra at each point, and the transition between them was carried out by rotating the sample by 90∘.

As the spectral profile transforms from A to B type, many bands change in intensity, and some bands completely disappear (Fig. 5). During the experiment, the sample's rotation significantly affected the intensity of the main peak of the υ1(C–O) band. However, the absolute intensity value of the main peak of the υ1(P–O) band remained almost unchanged; therefore, we used this value as a standard to assess the change in the intensity of other bands.

The differences between type-A and type-B Raman spectra were considered at three spectral segments: (1) the 900–1150 cm-1 range, where υ1 and υ3 stretching modes from the [PO4]3- group and υ1 from [CO3]2- group localize; (2) the 400–800 cm-1 range, which contains υ2 and υ4 bending modes from the [PO4]3- and [CO3]2- groups; and (3) the 70–350 cm-1 range, where lattice modes occur.

The two most intense modes of carbonophosphates υ1(P–O) and υ1(C–O) are positioned in the Raman spectra at 964±5 and 1074±3 cm-1, respectively. Several weak bands located between them are interpreted as the υ3(P–O) modes from the [PO4]3- group (Amaral et al., 2019; Frost et al., 2015). A similar position of the υ3(P–O) mode was found in the Raman spectra of many phosphate phases (Frezzotti et al., 2012; Litasov and Podgornykh, 2017; Pye and Rudolph, 2003). There are several peaks of the υ3(P–O) mode instead of one in this range because of the primitive (monoclinic) symmetry of carbonophosphates. The symmetry of the phosphate ion decreases in carbonophosphates compared with isolated [PO4]3-, which leads to the splitting of vibrational modes. For example, this phenomenon was found in vivianite spectra (Piriou and Poullen, 1984), which showed three peaks of the υ3(P–O) at 990, 1053, and 1018 cm-1. Let us consider the spectra of the studied carbonophosphates in order.

In both A and B spectra of sidorenkite, the right shoulder of the υ1(P–O) band has a flatter slope than the left one (Fig. 6a). This band incorporates a high peak at 957 cm-1 and a smaller peak at 959 cm-1, with the former being 7–30 times as intense as the latter. The υ1(C–O) band is closer to symmetry but decomposes into two peaks localized at 1071 and 1073 cm-1, which are similar in height (0.4–0.9 to 1, respectively). Between the double main bands, three small but well-pronounced bands occur at frequencies of 1010, 1034, and 1042 cm-1. They are present in both types of sidorenkite spectra. The intensities of these bands are relatively stable and do not exhibit any dependence on the crystallographic orientation.

In bonshtedtite spectra, the υ1(P–O) band decomposes into a high peak at 958 cm-1 and a smaller peak at around 964 cm-1 (Fig. 6b). Due to the difference of 2–5 times in the intensity of these peaks, the υ1(P–O) band of bonshtedtite is noticeably more asymmetric than that of sidorenkite. As in the spectra of the latter, the right shoulder of this band has a smaller slope. The υ1(C–O) band in the bonshtedtite spectra decomposes into two narrow peaks at 1074 and 1076 cm-1, the former being 6–4 times more intense than the latter. The interval between the main bands differs between types A and B of the bonshtedtite spectra. In both types, two barely noticeable but stable intensity peaks at 1025 and 1051 cm-1 were always observed in this region. However, type-B spectra are complemented by two additional peaks at 1011 and 1059 cm-1. Due to higher intensities, they overlap contours of the weaker “stable” peaks (Fig. 6b).

Bradleyite yields the most complex spectral profile in the considered range. Compared to sidorenkite and bonshtedtite spectra, it has a more asymmetric υ1(P–O) band with a flatter left shoulder (Fig. 6c). This asymmetry results from the spatial closeness of a high peak at 970 cm-1 and a smaller peak at 964 cm-1. The υ1(C–O) band in the bonshtedtite spectrum splits into two peaks at 1076 and 1077 cm-1. The segment between the most intense bands from the [PO4]3- and [CO3]2- groups in the bradleyite spectrum contains five peaks at frequencies of 996, 1016, 1131, 1050, and 1064 cm-1. The last three are well pronounced in the spectra of both A and B types. As the spectrum transforms from type A to type B, the peak positioned at 1131 cm-1 gradually increases. The peaks at 996 and 1016 cm-1 vary in shape with no clear dependence on the crystal orientation. In some spectra, they are hardly discernible, while in other spectra, they have the largest width among the five considered peaks and rarely merge. Two additional peaks were found in some bradleyite spectra, namely, a low-intensity broad peak to the left of the υ1(P–O) band at 927 cm-1 and a weak broad peak near the υ1(C–O) band at about 1100–1110 cm-1. No connection was traced between them and types A and B. The peak at 927 cm-1 tends to appear only with an increase in the intensity of the 996 and 1016 cm-1 peaks. Under the same circumstances, several additional peaks appear both in the region of υ2 and υ4 bending modes from the [PO4]3- and [CO3]2- groups (see Sect. 3.2.3) and in the region of the lattice modes (see Sect. 3.2.4).

The analysis performed showed that the υ1(P–O) and υ1(C–O) asymmetry is characteristic of all studied carbonophosphates. An accurate description of them is possible only by decomposing each band into two peaks. Faintly discernible splitting is manifested in the υ1(C–O) band, which always decomposes into two narrow peaks with a distance between the maxima of only 1–2 cm-1. Considering that the resolution of the instrument used for the study is 2 cm-1, the decomposition of this band into two peaks may be an artifact. However, these two peaks with sustained positions were obtained in all mathematical experiments performed. In the υ1(P–O) band, the splitting is more dramatic (Fig. 7). As shown by the example of sidorenkite, this band has strong asymmetry because of an additional peak at 959 cm-1, which coincides with the higher peak of the main bonshtedtite band. The additional bonshtedtite peak shows the same frequency as the additional bradleyite peak (964 cm-1), while the main bradleyite peak shows a higher frequency (969 cm-1). According to the EDX data, the sidorenkite studied contains iron, bonshtedtite has high magnesium content, and bradleyite is rich in iron (see Fig. 2). Isomorphic substitution of the Mn-Fe-Mg type leads to a change in the lengths of the Me2+–O chemical bonds and a shift in O atom coordinates in the mineral lattice. Simultaneously, the centers of mass of molecular bonds are displaced, and the [PO4]3- tetrahedra and [CO3]2- triangles are spatially distorted with violation of their symmetry. The unit cell parameters also change, which can induce splitting (Chukanov and Vigasina, 2020). Thus, the nature of the splitting of the υ1(P–O) and υ1(C–O) bands most likely lies in the influence of isomorphic impurities in the studied samples. Note that no splitting was observed in chemically pure synthetic carbonophosphates (Amaral et al., 2019). The main peak of the υ1(P–O) carbonophosphate band apparently results from O fluctuations associated with the dominant Me2+. The additional peak is determined by the O fluctuations associated with the Me2+ impurity (Fig. 7). Isomorphism is often accompanied by a regular peak shift, as evidenced by feldspars (Freeman et al., 2008) and sulfides (Kharbish et al., 2007). These features can explain the difference in the peak positions in the Raman spectra of the bonshtedtite–bradleyite pair. However, no similar displacement is observed in the sidorenkite–bonshtedtite pair, although the unit cell parameters of sidorenkite differ significantly from those of other carbonophosphates (see Table 1). The bond lengths (e.g., Mn–O and Fe–O bonds) also vary greatly (Tkhy et al., 1984).

In sum, the main difference between the Raman spectra of carbonophosphate phases in the considered range is the position of the υ1(P–O) and υ1(C–O) bands (e.g., bradleyite highly differs from sidorenkite and bonshtedtite), as well as the number of peaks between these two bands.

As mentioned above, out-of-plane (υ2) and in-plane (υ4) bending modes from the [PO4]3- and [CO3]2- groups occur in the 400–800 cm-1 spectral range of carbonophosphates (Amaral et al., 2019; Frost et al., 2015). Frost et al. (2015) indicated that the peaks at 414 and 469 cm-1 correspond to υ2(P–O) and the peak at 579 cm-1 corresponds to υ4(P–O). The positions of the [PO4]3- group modes point to a similarity to the Raman spectrum of the vivianite mentioned above (Piriou and Poullen, 1984). The spectrum of vivianite contains two peaks with wavenumbers close to 458 and 425 cm-1 attributed to υ2(P–O), as well as four peaks at 539, 572, 580, and 620 cm-1 attributed to υ4(P–O). In the spectra of carbonophosphates, the peaks of the υ2(P–O) band occupy positions at 420±7 cm-1 and 465±13 cm-1, and the peaks of the υ4(P–O) band are located at 583±6 and 624±1 cm-1 (Fig. 8). Frost et al. (2015) also emphasized the absence of the υ2(C–O) band around 800 cm-1, which is also typical of most carbonates (Frezzotti et al., 2012), except for cerussite and aragonite. In the spectra of carbonophosphates, we also did not observe these peaks (see Fig. 5). This phenomenon suggests that the [CO3]2- groups do not change the symmetry in the structure of the considered minerals relative to the free state (Chukanov and Vigasina, 2020). The υ4(C–O) band splits into two peaks with wavenumbers 694 and 725±6 cm-1. The second peak in some bradleyite spectra overlaps with an additional (sometimes more intense) peak positioned at 747 cm-1 (Fig. 8c). The latter peak correlates in intensity with the previously described peak at a frequency of 927 cm-1 (see Sect. 3.2.2).

Over the entire range from 400 to 800 cm-1, the peak intensities are 0.1 or less compared to the main peak of the υ1(P–O) band. Only in the type-A spectra bradleyite do they sometimes reach 0.15. When passing from type A to type B, the intensity of most peaks either remains almost unchanged or drops sharply. We observed an intensity decrease for the peaks at 468 and 623 cm-1 in sidorenkite spectra (Fig. 8a), at around 580 and 727 cm-1 in bonshtedtite spectra (Fig. 8b), and at 478 and 694 cm-1 in the spectra of bradleyite (Fig. 8c). The exception is the peak at 427 cm-1 from υ2(P–O) in the bradleyite spectra (Fig. 8c), which is 3 times higher in spectra B than in spectra A. Most bands are quite asymmetric in the spectra of bonshtedtite and bradleyite, which contain abundant impurities (see Fig. 8b, c). They all show splitting, probably for the same reason as that given for the υ1(P–O) and υ1(C–O) bands (see Sect. 3.2.2).

In total, the profiles of the Raman spectra of carbonophosphates in the considered range are generally similar. An unusually small peak at 450 cm-1 from the υ2(P–O) mode distinguishes the bonshtedtite spectrum. The bradleyite spectra exhibit some peak shifts and slightly higher relative peak intensities than other carbonophosphates. Compared to type-B spectra, type-A spectra have a more prominent relief in this range.

The spectral range between 70 and 350 cm-1 contains the maximum number of peaks, which occur close to each other. In this study, we considered such characteristics of the most prominent bands as the position and intensity variations between the A and B spectra. The asymmetry of these peaks was not assessed.

Sidorenkite yields eight distinct peaks in the spectra of both type A and type B in this range. These peaks have the following wavenumbers: 91, 100, 110, 156, 176, 201, 250, and 295 cm-1 (Fig. 9a). The type-B spectra are generally more elevated due to peak growth at 100, 156, and 201 cm-1. Of the listed peaks in the type-A spectra, the maximums are 110 and 176 cm-1, while in the type-B spectra, the maximum is 156 cm-1. In all the spectra of sidorenkite, these peaks are in third place in intensity after the main peaks of the υ1(P–O) and υ1(C–O) bands (about 0.10–0.15 high from the main peak of the υ1(P–O) band).

Bonshtedtite has nine distinct peaks in the type-A spectra and only seven peaks in the type-B spectra (Fig. 9b). Peaks at 99, 130, 157, 181, 208, 255, and 295 cm-1 are common. The latter is very broad and is the sum of an intense peak at 295 cm-1 and an additional peak of lower intensity at 279 cm-1. In the type-A spectra, additional peaks were observed at 92 and 113 cm-1. With a change in the crystallographic orientation, their intensity sharply decreases. Eventually, they overlap with the neighboring peak at 99 cm-1. As the spectra change towards type B, the peak at 130 cm-1 also decreases, while almost all other peaks increase. As a result, the type-B spectra show a total intensity of lattice modes that is higher than that of the type-A spectra. Of all peaks considered here, the highest is the one at 157 cm-1. This peak is the third most intense in the Raman spectra of bonshtedtite and 3–4 times less intense than the main peak of the υ1(P–O) band.

For bradleyite, nine pronounced peaks were found in the type-A spectra and eight in the type-B spectra (Fig. 9c). Common are peaks at 95, 104, 114, 158, 194, 215, 259, and 295 cm-1. Like spectra of other carbonophosphates, those of bradleyite in the range of 70–350 cm-1 show a general increase in peak intensity as they approach type B. Only the peak at 95 cm-1 decreases in intensity, while the peaks at 104, 158, and 215 cm-1 increase significantly. Throughout the full Raman spectra of bradleyite, the peaks at 158 and 215 cm-1 rank third in intensity after the main peaks of the υ1(P–O) and υ1(C–O) bands, and in the range under consideration, they are maximal. This is valid for both A and B spectra types. The peak at 134 cm-1 is prominent in the A spectra but hardly distinguishable in the B spectra due to the increased peak at 158 cm-1. Also, several collected spectra showed a peak at 328 cm-1, independent of type A or B. It correlates with the specific additional peak at 927 cm-1 discussed above (see Sect. 3.2.2).

Thus, the profiles of the Raman spectra of carbonophosphates in the considered range have much in common. They primarily differ due to the shift in the frequencies of many peaks. Another notable difference is the lower relative intensity of peaks in sidorenkite spectra compared to other carbonophosphates. Type-B spectra are preferable for studying this range.

The results obtained from the conducted Raman investigation are summarized in Table 2. Let us now analyze these results to create an algorithm for Raman diagnostics of carbonophosphate phases.

Raman spectroscopy provides reliable determination of the belonging of an unknown mineral to the carbonophosphate group. Minerals that show two intense bands in the spectral range of 950–1100 cm-1 are rare (Frezzotti et al., 2012). A search in the RRUFF Raman database using the CrystalSleuth software (Lafuente et al., 2015) showed that only a few minerals (mainly phosphates, e.g., tinticite, bobierrite, and baricite) have peaks in this region that are comparable in intensity with the type-B spectra of the studied carbonophosphates. Peaks in this range are typical of several minerals that contain both [PO4]3- and [CO3]2- groups: devitoite (Kampf et al., 2010), ramikite-(Y) (McDonald et al., 2013), and skorpionite (Krause et al., 2008). However, the studied carbonophosphates differ from these minerals in the position and intensity of other peaks in the Raman spectrum. It is more difficult to distinguish between them.

By the position of the main peaks from υ1(P–O) and υ1(C–O) modes, sidorenkite and bonshtedtite spectra overlap within the resolution of the most commonly used Raman spectrometers. Concerning this characteristic, bradleyite stands out, whose bands have a significantly higher frequency. When comparing the results of our study of natural bradleyite with impurities and a synthetic chemically pure compound (see Table 2), we found that the closer bradleyite is to the ideal formula, the higher the frequency. The studied carbonophosphates differ from each other in the number of peaks between the υ1(P–O) and υ1(C–O) bands at frequencies of 964±5 and 1074±3 cm-1, respectively (see Fig. 6). However, Raman diagnostics of carbonophosphates are often used when studying microinclusions in minerals of carbonatites and kimberlites. In these rocks, the overlap of carbonophosphate peaks with the peaks of host minerals and phases coexisting in the inclusion is highly probable. Apatite, monazite, baryte, xenotime, celestine, zircon, diopside, pargasite, and witherite are common in carbonatites and kimberlites. The most intense bands in their spectra range from 970 to 1080 cm-1 (Hurai et al., 2015). The prominent bands of typical phases found in fluid inclusions are also located there. These are sulfate and carbonate ions, arcanite, thenardite, anhydrite, gregoryite, nahcolite, shortite, gaylussite, and others (Hurai et al., 2015). Thus, diagnostics by the peaks between the υ1(P–O) and υ1(C–O) bands can be successful only if it is possible to collect the reference spectrum of “pure” carbonophosphate and this is not complicated by extraneous peaks of neighboring minerals.

Additional information was obtained by analyzing the spectral ranges of 400–800 and 70–350 cm-1. A peak at 725±6 cm-1 should be noted in the former region, as it shifts markedly in the sidorenkite–bonshtedtite–bradleyite series (see Fig. 8, Table 2). In the range of 70–350 cm-1, the positions of three peaks differ significantly between the carbonophosphate phases: 185±9, 208±7, and 255±5 cm-1 (see Table 2). For the correct analysis of these peaks, the sample should be rotated until type-B Raman spectra are obtained with Iυ1(P-O)≥Iυ1(C-O) (see Fig. 9).