Si-Disordering in MgAl2O4-Spinel under High P-T Conditions, with Implications for Si-Mg Disorder in Mg2SiO4-Ringwoodite

A series of Si-bearing MgAl2O4-spinels were synthesized at 1500–1650 ◦C and 3–6 GPa. These spinels had SiO2 contents of up to ~1.03 wt % and showed a substitution mechanism of Si4+ + Mg2+ = 2Al3+. Unpolarized Raman spectra were collected from polished single grains, and displayed a set of well-defined Raman peaks at ~610, 823, 856 and 968 cm−1 that had not been observed before. Aided by the Raman features of natural Si-free MgAl2O4-spinel, synthetic Si-free MgAl2O4-spinel, natural low quartz, synthetic coesite, synthetic stishovite and synthetic forsterite, we infer that these Raman peaks should belong to the SiO4 groups. The relations between the Raman intensities and SiO2 contents of the Si-bearing MgAl2O4-spinels suggest that under some P-T conditions, some Si must adopt the M-site. Unlike the SiO4 groups with very intense Raman signals, the SiO6 groups are largely Raman-inactive. We further found that the Si cations primarily appear on the T-site at P-T conditions ≤~3–4 GPa and 1500 ◦C, but attain a random distribution between the T-site and M-site at P-T conditions ≥~5–6 GPa and 1630–1650 ◦C. This Si-disordering process observed for the Si-bearing MgAl2O4-spinels suggests that similar Si-disordering might happen to the (Mg,Fe)2SiO4-spinels (ringwoodite), the major phase in the lower part of the mantle transition zone of the Earth and the benchmark mineral for the very strong shock stage experienced by extraterrestrial materials. The likely consequences have been explored.


Introduction
Spinel (Sp; AB 2 O 4 ) sensu lato plays a crucial role in Earth sciences.The so-called 2-3 Sp, A = 2 + cations and B = 3 + cations, is ubiquitous in most terrestrial rocks [1,2].With significant compositional complexity and a wide P-T stability field, it participates in many phase equilibria, which can be calibrated as geothermometers, geobarometers and oxybarometers [3][4][5], and therefore In Sp, the size of a cation has a profound influence in determining its site preference, with larger ions preferring the T-site of 2-3 Sp, but the M-site of 4-2 Sp [59].With a relatively small size difference between the Mg and Al cations in the MgAl2O4-Sp, the cation disorder achieved under high P-T conditions can be partially preserved [42,45,48,50,51,53].In contrast, the relatively large size difference between the Mg and Si cations in the Rw may strengthen this size-dependent site preference and accelerate the cation-redistribution process during cooling, so that the cation disorder attained under high P-T conditions can be easily lost, leading to null signals for cation disorder, as observed experimentally [61,64].To circumvent this obstacle, we have taken an indirect approach by doping the MgAl2O4-Sp with some Si, and examined whether Si can be disordered.It was expected that silicon could readily enter the MgAl2O4-Sp, for the SiO2 in natural 2-3 Sp can reach up to ~5.3 wt % (Figure 1).In this study, we first synthetized the Si-bearing MgAl2O4-Sp at high P.We then analyzed the experimental products using Raman spectroscopy, a powerful method for studying cation-disordering [49,57].To facilitate data interpretation, natural Si-free MgAl2O4-Sp (N-Sp), natural low quartz (N-Qz), and synthetic Si-free MgAl2O4-Sp, coesite (Coe), stishovite (St) and forsterite (Fo) were similarly analyzed.Here we report the first experimental evidence for 6-coordinated Si in the Sp structure.[65], Sobolev & Nikogosian [66], Kamenetsky et al. [67], Franz & Wirth [68], and Chistyakova et al. [69].

Experimental and Analytical Methods
High-P experiments were conducted on a cubic press at the High-Pressure Laboratory of Peking University [70] and a multi-anvil press at the Geophysical Laboratory, Carnegie Institution of Washington [71].With the experimental charges encapsulated in sealed Pt tubes, a series of Si-bearing MgAl2O4-Sp were synthesized in the system CaO-MgO-Al2O3-SiO2-K2O-CO2 at 3-6 GPa and 1500-1650 °C by employing a conventional electrical resistance heating technique (Table 1).In addition, we used high-P experimental techniques to separately synthesize Si-free MgAl2O4-Sp, Coe and St (Table 1).The P and T uncertainties in our high-P experiments should be better than ~0.[65], Sobolev & Nikogosian [66], Kamenetsky et al. [67], Franz & Wirth [68], and Chistyakova et al. [69].

Experimental and Analytical Methods
High-P experiments were conducted on a cubic press at the High-Pressure Laboratory of Peking University [70] and a multi-anvil press at the Geophysical Laboratory, Carnegie Institution of Washington [71].With the experimental charges encapsulated in sealed Pt tubes, a series of Si-bearing MgAl 2 O 4 -Sp were synthesized in the system CaO-MgO-Al 2 O 3 -SiO 2 -K 2 O-CO 2 at 3-6 GPa and 1500-1650 • C by employing a conventional electrical resistance heating technique (Table 1).In addition, we used high-P experimental techniques to separately synthesize Si-free MgAl 2 O 4 -Sp, Coe and St (Table 1).The P and T uncertainties in our high-P experiments should be better than ~0.5 GPa and 50 • C [70 -72].
The compositions of the crystalline phases from the high-P experiments were obtained by using a JXA-8100 electron microprobe (EMP) in wavelength dispersive mode (WDS).For all the EMP analyses, the beam current was 10 nA, the accelerating voltage 15 kV, the beam spot size 1 µm, and the counting time 40 s.Calibration was based on optimization to some standards provided by the SPI Corporation (USA), with diopside for Mg and Ca calibrations, jadeite for Si, Al and Na calibrations, chromium oxide for Cr calibration, hematite for Fe, sanidine for K, rutile for Ti, rhodonite for Mn, and nickel silicide for Ni.Data correction was performed with the PRZ method.The results are shown in Table 1 (the CaO and K 2 O contents below the detection limits).Two natural gem-quality mineral samples were employed in this study as well; one was a red, Si-free Sp crystal (N-Sp) with an octahedral shape from Mogok (Burma), and the other was a clear low Qz crystal (N-Qz) from Donghai (China).Both were similarly analyzed for compositions with the EMP in the WDS mode.In addition to the components shown in Table 1, extra components in the N-Sp included 0.06(3)% TiO 2 , 0.95(6)% Cr 2 O 3 and 0.10(1)% FeO, leading to the chemical formula (Mg 0.993 Fe 0.002 Ti 0.001 )(Al 1.983 Cr 0.018 )O 4 (all iron assumed as Fe 2+ ).Extra components in the N-Qz were below the detection limits.
Unpolarized Raman spectra were collected from 100 to 1350 cm −1 with a Renishaw inVia Reflex system in a back-scattering geometry at ambient P-T conditions.A 532 nm laser with an emission power of ~50 mW and a 50× long-distance objective were used in all analyses.Other analytical conditions were ~1 µm light spot, 1 cm −1 spectral resolution, and 20 successive scans for every analysis (10 s for each scan).For every high-P product, multiple analyses were conducted on well-polished and arbitrarily selected Sp, Coe, St and Fo grains with unknown orientations.For comparison, the Raman spectrum of the N-Sp was collected from the (111) plane, whereas that of the N-Qz was from the (001) plane.The Raman data were processed by using the PeakFit V4.12 software (SPSS Inc.).
In addition, we analyzed one fragment of the N-Sp for its order-disorder state by single-crystal XRD method.Data were collected using an Agilent Technologies Rigaku micro-focused diffractometer (Mo Kα radiation; λ = 0.071073 nm), and processed using the SHELXT software included in the SHELXTL package.From the single-crystal XRD data we directly obtained an x value of 0.129, probably with relatively large uncertainty due to the similar scattering factors of Mg and Al.Following the method of Carbonin et al. [73], with the bond distances from Lavina et al. [74] and with x = 0.129 as one of the input variables; further, we calculated a new x value, which was in turn used as an input in the next round of crystal structural analysis.The final cycles of the least-squares refinement, including atomic coordinates and anisotropic thermal parameters for the atoms [I > 2sigma(I)], converged at R 1 = 0.0164, wR 2 = 0.0730 and S = 1.065, and yielded x = 0.162 (see Supplementary Materials for the details).Using the empirical equation proposed by Andreozzi & Princivalle [55], x = 21.396− 80.714u (1) where u is the oxygen positional parameter in the Sp structure (u = 0.26329 (24) for our N-Sp); alternatively, we constrained the x value as 0.145.x = 0.145 is preferred in this study.In total, nine high-P experiments with long durations of 8-36 h were conducted (Table 1): six of them for synthesizing Si-bearing MgAl 2 O 4 -Sp, one for Si-free MgAl 2 O 4 -Sp, one for Coe, and one for St.In the synthesizing experiments for the Si-bearing MgAl 2 O 4 -Sp, a CO 2 -rich melt phase with intense quench-modification texture was always observed.Some other crystalline phases like Fo and garnet (Grt) were occasionally detected.The crystalline phases in all these experiments had large grain sizes of up to ~600 µm, showed sharp grain boundaries and attained homogeneous chemical compositions.Typical electron back-scatter images from some of these experiments are shown in Figure 2. In the experiments for the Si-free MgAl 2 O 4 -Sp, Coe and St, a melt phase was clearly observed in LMD659 only (Table 1).The grain boundaries of the Si-free MgAl 2 O 4 -Sp, Coe and St were well defined, their grain sizes were large (up to ~100 µm in diameter), and their compositions were expected to be homogeneous.
Supplementary Materials for the details).Using the empirical equation proposed by Andreozzi & Princivalle [55], x = 21.396− 80.714u (1) where u is the oxygen positional parameter in the Sp structure (u = 0.26329 (24) for our N-Sp); alternatively, we constrained the x value as 0.145.x = 0.145 is preferred in this study.

Synthetic MgAl2O4-Sp and Its SiO2
In total, nine high-P experiments with long durations of 8-36 h were conducted (Table 1): six of them for synthesizing Si-bearing MgAl2O4-Sp, one for Si-free MgAl2O4-Sp, one for Coe, and one for St.In the synthesizing experiments for the Si-bearing MgAl2O4-Sp, a CO2-rich melt phase with intense quench-modification texture was always observed.Some other crystalline phases like Fo and garnet (Grt) were occasionally detected.The crystalline phases in all these experiments had large grain sizes of up to ~600 μm, showed sharp grain boundaries and attained homogeneous chemical compositions.Typical electron back-scatter images from some of these experiments are shown in Figure 2. In the experiments for the Si-free MgAl2O4-Sp, Coe and St, a melt phase was clearly observed in LMD659 only (Table 1).The grain boundaries of the Si-free MgAl2O4-Sp, Coe and St were well defined, their grain sizes were large (up to ~100 μm in diameter), and their compositions were expected to be homogeneous.With up to ~1 wt % SiO 2 (Table 1), the compositions of the Si-bearing MgAl 2 O 4 -Sp are shown in Figure 3.A primary observation here is that one Si 4+ and one Mg 2+ substitute for two Al 3+ , Si 4+ + Mg 2+ = 2Al 3+  (2) In detail, the (Si pfu ) values seem slightly lower than the (Mg pfu -1) values, which perhaps relates to the compositional characteristics of the coexisting phase(s).Nevertheless, the effects of P, T and the coexisting phases on this cation substitution reaction are not clear, but are presently undergoing thorough experimental investigation.
Figure 3.A primary observation here is that one Si 4+ and one Mg 2+ substitute for two Al 3+ , Si 4+ + Mg 2+ = 2Al 3+  (2) In detail, the (Sipfu) values seem slightly lower than the (Mgpfu-1) values, which perhaps relates to the compositional characteristics of the coexisting phase(s).Nevertheless, the effects of P, T and the coexisting phases on this cation substitution reaction are not clear, but are presently undergoing thorough experimental investigation.
The cation radii of Mg (rMg), Al (rAl) and Si (rSi) are very different, rMg = 0.585 Å > rAl = 0.39 Å > rSi = 0.275 Å on the T-site and rMg = 0.715 Å > rAl = 0.53 Å > rSi = 0.40 Å on the M-site under ambient conditions [59].Since larger ions prefer the T-site of the 2-3 Sp, the Si-free MgAl2O4-Sp should generally adopt a normal Sp structure, as verified by some studies on natural Sp with compositions close to the MgAl2O4 formula (x = ~0.02-0.04 in Schmocker & Waldner [47]; x = 0.05 in Maekawa et al. [51]).By the same token, Si in the MgAl2O4-Sp should occupy the M-site.However, existing single-crystal XRD studies on natural 2-3 Sp locate Si on the T-site [73,[75][76][77].The coupled substitution of Si and Mg for 2Al as observed in our high-P synthetic MgAl2O4-Sp and the site-occupation knowledge to be revealed by our Raman spectroscopic data should shed light on the Si distribution.
Our N-Sp displays four sharp peaks at ~312, 407, 664 and 766 cm −1 , compatible with the Raman features established for normal MgAl 2 O 4 -Sp (Figure 4).Furthermore, two weak and broad peaks are observed at ~222 and 715 cm −1 , which are attributable to the slightly disordered structural feature (x = 0.145).The small peak at ~715 cm −1 was also evident in the Raman spectra of the natural MgAl 2 O 4 -Sp studied by Chopelas & Hofmeister [78] and by Cynn et al. [86].Both samples attained some structural disorder: using Equation (1), the x value of the former sample was calculated as  [78] had an x value comparable to our N-Sp, so a weak peak at ~222 cm −1 should be expected.Chopelas & Hofmeister [78], however, did not report any Raman data below ~250 cm −1 .
Our N-Sp displays four sharp peaks at ~312, 407, 664 and 766 cm −1 , compatible with the Raman features established for normal MgAl2O4-Sp (Figure 4).Furthermore, two weak and broad peaks are observed at ~222 and 715 cm −1 , which are attributable to the slightly disordered structural feature (x = 0.145).The small peak at ~715 cm −1 was also evident in the Raman spectra of the natural MgAl2O4-Sp studied by Chopelas & Hofmeister [78] and by Cynn et al. [86].Both samples attained some structural disorder: using Equation (1), the x value of the former sample was calculated as ~0.144 (u = 0.2633); the x value of the latter sample was claimed to be ~0.02,which might have been slightly underestimated (more discussion later).On the other hand, it was not observed for the natural MgAl2O4-Sp studied by Cynn et al. [49], Van Minh & Yang [87] or Slotznick & Shim [57], implying x values smaller than at least ~0.145.No Raman spectra previously collected on unannealed natural MgAl2O4-Sp showed the weak peak at ~222 cm −1 .The sample studied by Chopelas & Hofmeister [78] had an x value comparable to our N-Sp, so a weak peak at ~222 cm −1 should be expected.Chopelas & Hofmeister [78], however, did not report any Raman data below ~250 cm −1 .In situ high-T Raman spectroscopic investigations on natural MgAl2O4-Sp were conducted by Cynn et al. [49,86], Van Minh & Yang [87], and Slotznick & Shim [57].The weak peak at ~715 cm −1 evidently emerged or intensified at high T, and persisted to ambient T after cooling, so that it could be confidently attributed to the high-T structural disorder process.Theoretical investigations have confirmed this attribution [81,83].In comparison, an even weaker Raman peak at ~222 cm −1 was detected at high T by Slotznick & Shim [57] only, and was similarly attributed to the high-T structural disorder process.Additionally, it was observed by Cynn et al. [86] on the natural MgAl2O4-Sp after, rather than before, their high-T Raman spectroscopic experiments.
The two Raman peaks at ~222 and 715 cm −1 directly observed on our N-Sp (x = ~0.145)may provide a convenient and inexpensive method to quantify the disorder extent of natural 2-3 Sp.Recording rich genetic conditions such as chemical environment, geological setting, and cooling history [77,88], natural 2-3 Sp commonly has an x value ranging from 0 to ~0.23 ([89]; and references therein).The x parameters are usually constrained by applying the single-crystal XRD method, powder neutron diffraction or nuclear magnetic resonance spectroscopy, which is often instrumentally unavailable, technically challenging, requires a large quantity of homogeneous sample, and/or costs too much in terms of funds and time.Raman spectroscopy is, however, the exact opposite.The Raman feature at ~715 cm −1 has high intensity, and is well separated from the A1g band at ~766 cm −1 , so that it can be readily used to estimate the disorder extent (Figure 4).With fixed analytical conditions in the Raman spectroscopic experiments, the intensity ratio of these two peaks should reflect the inversion extent according to the following equation [86]: (3) In situ high-T Raman spectroscopic investigations on natural MgAl 2 O 4 -Sp were conducted by Cynn et al. [49,86], Van Minh & Yang [87], and Slotznick & Shim [57].The weak peak at ~715 cm −1 evidently emerged or intensified at high T, and persisted to ambient T after cooling, so that it could be confidently attributed to the high-T structural disorder process.Theoretical investigations have confirmed this attribution [81,83].In comparison, an even weaker Raman peak at ~222 cm −1 was detected at high T by Slotznick & Shim [57] only, and was similarly attributed to the high-T structural disorder process.Additionally, it was observed by Cynn et al. [86] on the natural MgAl 2 O 4 -Sp after, rather than before, their high-T Raman spectroscopic experiments.
The two Raman peaks at ~222 and 715 cm −1 directly observed on our N-Sp (x = ~0.145)may provide a convenient and inexpensive method to quantify the disorder extent of natural 2-3 Sp.Recording rich genetic conditions such as chemical environment, geological setting, and cooling history [77,88], natural 2-3 Sp commonly has an x value ranging from 0 to ~0.23 ([89]; and references therein).The x parameters are usually constrained by applying the single-crystal XRD method, powder neutron diffraction or nuclear magnetic resonance spectroscopy, which is often instrumentally unavailable, technically challenging, requires a large quantity of homogeneous sample, and/or costs too much in terms of funds and time.Raman spectroscopy is, however, the exact opposite.The Raman feature at ~715 cm −1 has high intensity, and is well separated from the A 1g band at ~766 cm −1 , so that it can be readily used to estimate the disorder extent (Figure 4).With fixed analytical conditions in the Raman spectroscopic experiments, the intensity ratio of these two peaks should reflect the inversion extent according to the following equation [86]: where c is an unknown coefficient presumably dependent to the analytical setups, and x = 0.02 by assuming c = 1.We prefer the larger x value, simply because a disorder extent of 0.02 in the MgAl 2 O 4 -Sp structure may not be high enough to bring forth the Raman peak at ~715 cm −1 .

Mg-Al Order-Disorder State of Synthetic MgAl 2 O 4 -Sp
The Mg-Al order-disorder states of our synthetic MgAl 2 O 4 -Sp can be estimated using the results from the in situ observations under high P-T conditions made by Méducin et al. [45], as shown in Figure 5.  6)).Applying this value to the Raman data of the unannealed natural MgAl2O4-Sp of Cynn et al. [86] leads to an x value of ~0.06 (or 0.09).Cynn et al. [86] obtained x = 0.02 by assuming c = 1.We prefer the larger x value, simply because a disorder extent of 0.02 in the MgAl2O4-Sp structure may not be high enough to bring forth the Raman peak at ~715 cm −1 .

Mg-Al Order-Disorder State of Synthetic MgAl2O4-Sp
The Mg-Al order-disorder states of our synthetic MgAl2O4-Sp can be estimated using the results from the in situ observations under high P-T conditions made by Méducin et al. [45], as shown in Figure 5.There has been excellent agreement on the T effect on the Mg-Al disorder process of the MgAl2O4-Sp at ambient P: x increases as T increases [42,47,48,50,51,[53][54][55][56][57].As to the P effect at ambient T, discrepancy presumably exists because the order-disorder reaction could not be readily activated and did not adequately approach its equilibrium during the course of a conventional high-P study [52,90,91].Thanks to Méducin et al. [45] who conducted an investigation under simultaneously high-P and high-T conditions (up to 3.2 GPa and 1318 °C), the P effect at relatively high T has been well established: x increases as P increases.It is thus clear that our synthetic There has been excellent agreement on the T effect on the Mg-Al disorder process of the MgAl 2 O 4 -Sp at ambient P: x increases as T increases [42,47,48,50,51,[53][54][55][56][57].As to the P effect at ambient T, discrepancy presumably exists because the order-disorder reaction could not be readily activated and did not adequately approach its equilibrium during the course of a conventional high-P study [52,90,91].Thanks to Méducin et al. [45] who conducted an investigation under simultaneously high-P and high-T conditions (up to 3.2 GPa and 1318 • C), the P effect at relatively high T has been well established: x increases as P increases.It is thus clear that our synthetic MgAl 2 O 4 -Sp, formed under high P-T conditions, should attain large degrees of cation disorder, which should be well preserved due to the quick quench process in the cubic press experiments (T decreased to <600 • C in ~20 s).
Claimed by Méducin et al. [45], the heating-up experiments at T ≥ 500 • C closely reached their cation order-disorder equilibrium, with the P almost linearly correlating with the T (Figure 5a).Since both P and T promote Mg-Al disorder under simultaneously high-P and high-T conditions, the effects of P and T can be lumped together and adequately accounted for by using just one independent variable.Here, we have chosen T (Figure 5b).Coincidently, our synthesizing experiments at 4 GPa and 1500 to 1550 • C (Table 1) plot rather near the P-T locus defined by those heating-up experiments at T ≥ 500 • C (Figure 5a), suggesting that, with a short-distance extrapolation, the x values of the MgAl 2 O 4 -Sp from our experiments at 4 GPa could be accurately estimated.Using the equation shown in Figure 5b, the derived x values are from 0.70 (15) to 0.73 (15); therefore, the true x values should be close to 0.667 (random Mg-Al distribution).In addition, the x values of our synthetic MgAl 2 O 4 -Sp at 5 and 6 GPa should also be ~0.667due to the even higher experimental P and T (Figure 5a).Furthermore, the x values obtained for the P-T conditions of 2.8 GPa and 1163 • C, and 3.2 GPa and 1318 • C by Méducin et al. (2004) [45] were 0.571(49) and 0.633 (50), respectively, implying that the x of our MgAl 2 O 4 -Sp at a similar P of 3 GPa but a much higher T of 1500 • C (LMD565; Table 1) should be close to 0.667, as well.
Assuming no effect of the additional Si with abundances ≤~0.025 pfu (Figure 3), we conclude that our synthetic MgAl 2 O 4 -Sp should achieve a nearly random Mg-Al distribution.

Raman Features of Fully Disordered MgAl 2 O 4 -Sp
The Raman spectrum of our synthetic Si-free MgAl 2 O 4 -Sp (LMD487) is compared to that of our N-Sp in Figure 4.It similarly shows six peaks at slightly different wavenumbers, although with all peaks significantly broadened.Compatible with the observations made by Cynn et al. [49,86] and Slotznick & Shim [57], the A 1g , E g and T 2g(1) modes shift slightly to lower wavenumbers, whereas the T 2g(2) mode shifts slightly to higher wavenumbers, as x increases from ~0.145 to 0.667.In addition, the E g band becomes not only very broad, but highly asymmetric, as well, indicating a possible hiding Raman peak.According to Caracas & Banigan [84], a very intense Raman feature should occur at the lower wavenumber side of the E g peak when the MgAl 2 O 4 -Sp disorders.Moreover, the two weak, broad, and Mg-Al disorder-related peaks at ~723 and 225 cm −1 become much more distinct in the Raman spectrum of the synthetic Si-free MgAl 2 O 4 -Sp.All these are diagnostic features for a high degree of Mg-Al disorder.
With Equation (3), and adopting x = 0.667, the peak height data (or integrated area data) of our synthetic Si-free MgAl 2 O 4 -Sp, I 723 = 7148(215) and I 765 = 10,986(228) cps (or I 723 = 181,810(5451) and I 765 = 240,020(6329) cps cm −1 ), lead to a c value of 0.32(2) (or 0.38(2)), which is again much smaller than the assumed value of 1 in Cynn et al. [86].Combining this result with that determined by the Raman data of our N-Sp, 0.33 (9) or 0.20 (6), the c coefficient appears generally constant for a large range of x, supporting the constant c assumption made by Cynn et al. [86].To confirm this, more investigation on the MgAl 2 O 4 -Sp with different disorder extents using jointed experimental methods to simultaneously obtain Raman spectroscopic data, chemical compositional data and crystal structural data like what we have done in this study is highly desirable.

Raman Features of Si-Bearing Fully Disordered MgAl 2 O 4 -Sp
The octahedra in the Sp structure share six edges with six neighboring octahedra, resulting in an extensively edge-linked structure in three dimensions [92].In comparison, the tetrahedra are fully isolated from each other, with their four oxygen atoms linking to four neighboring octahedra.If Si occupied the M-site of the MgAl 2 O 4 -Sp, its Raman signals would be much analogous to those of St, which similarly places Si in edge-shared octahedra [93].If Si occupied the T-site, alternatively, its Raman signals would resemble those of Fo because Si in Fo also adopts an isolated T-site and forms a separate SiO 4 group, with the oxygen atoms being shared between neighboring octahedral [94].On the other hand, Si atoms in low Qz [95] and Coe [96] are 4-coordinated, but the SiO 4 tetrahedra are fully polymerized into a three-dimensional framework, so that the Raman features of low Qz and Coe should be very different to those of potential SiO 4 groups in the Sp structure.
Apart from those six bands previously described, the Si-bearing MgAl 2 O 4 -Sp shows a new set of well-defined Raman bands at ~610, 823, 856 and 968 cm −1 (Figure 6).These peaks are distinctly different to the Raman features of St, Coe and N-Qz, but highly resemble those of Fo.Furthermore, a less well-defined peak with low intensity occasionally appears at ~920 cm −1 , and perfectly matches the relatively weak 920 cm −1 Raman peak of Fo (Figure 6).In analogy with the Raman features of Fo [97], we tend to attribute these five peaks to potential separate SiO 4 groups in our Si-bearing, fully Mg-Al disordered MgAl 2 O 4 -Sp, and assign the peaks at ~968, 920 and 856 cm −1 to the asymmetric stretching of the SiO 4 groups, the peak at ~823 cm −1 to the symmetric stretching, and the peak at ~610 cm −1 to the bending.It follows that at least some Si atoms adopt the T-site.Apart from those six bands previously described, the Si-bearing MgAl2O4-Sp shows a new set of well-defined Raman bands at ~610, 823, 856 and 968 cm −1 (Figure 6).These peaks are distinctly different to the Raman features of St, Coe and N-Qz, but highly resemble those of Fo.Furthermore, a less well-defined peak with low intensity occasionally appears at ~920 cm −1 , and perfectly matches the relatively weak 920 cm −1 Raman peak of Fo (Figure 6).In analogy with the Raman features of Fo [97], we tend to attribute these five peaks to potential separate SiO4 groups in our Si-bearing, fully Mg-Al disordered MgAl2O4-Sp, and assign the peaks at ~968, 920 and 856 cm −1 to the asymmetric stretching of the SiO4 groups, the peak at ~823 cm −1 to the symmetric stretching, and the peak at ~610 cm -1 to the bending.It follows that at least some Si atoms adopt the T-site.Furthermore, two weak and diffusive Raman peaks have occasionally been observed at ~560 and 1010 cm −1 for our Si-bearing MgAl 2 O 4 -Sp (Figure 6), with the former attributable to the usually undetected fifth fundamental Raman band of the MgAl 2 O 4 -Sp (T 2g(3) ) and the latter likely featured as a combination band/overtone.
The intensities of the Raman peaks attributable to the SiO 4 groups show interesting behavior.Considering the very low SiO 2 contents in the MgAl 2 O 4 -Sp from LMD563 and LMD565 (0.30 (7) wt % and 0.39(5) wt %, respectively; Table 1), the low intensities of the new Raman peaks at ~610, 823, 856 and 968 cm −1 can be readily explained by the small amounts of the SiO 4 group (Figure 6).As the SiO 2 contents increase, one would anticipate these peaks to grow if some of the added Si entered the T-site.Surprisingly, the Raman spectra of our MgAl 2 O 4 -Sp with higher SiO 2 contents, from 0.65 (7) to 1.03 (7) wt %, show distinctly divergent behaviors (Figure 6), with the new Raman peaks at ~610, 823, 856 and 968 cm −1 intensifying for the MgAl 2 O 4 -Sp synthesized at relatively low P-T conditions (4 GPa and 1500 • C for LMD564, and 4 GPa and 1550 • C for LMD558; Table 1) but increasing little for the MgAl 2 O 4 -Sp synthesized at relatively high P-T conditions (5 GPa and 1630 • C for LMD578, and 6 GPa and 1650 • C for LMD568).Evidently, some of the Si atoms added into the MgAl 2 O 4 -Sp did take the T-site under relatively low P-T conditions, but most them did not under relatively high P-T conditions.It follows that some Si atoms in the MgAl 2 O 4 -Sp from LMD578 and LMD568 must have adopted the M-site and formed SiO 6 groups (Figure 6).
The SiO 6 groups seem Raman-inactive.With similar amounts of SiO 2 , the MgAl 2 O 4 -Sp from LMD558 shows much stronger Raman peaks for its SiO 4 groups than that from LMD578 (Figure 7), suggesting that the former generally contains more SiO 4 groups, but the latter contains more SiO 6 groups.In both cases, no new Raman peaks can be confidently identified, implying that the SiO 6 groups in the MgAl 2 O 4 -Sp are by and large Raman-inactive.Different crystallographic orientations are unlikely to affect this conclusion.As shown in Figure 7a, the two sets of unpolarized Raman spectra for the MgAl 2 O 4 -Sp in LMD558 (Set A and Set B), taken from the only crystal shown in Figure 2a, but with crystallographic orientations normal to each other, do display some variations in the intensities of the Raman peaks for the SiO 4 groups, but overall exhibit very similar patterns.Furthermore, the 10 unpolarized Raman spectra taken from 10 randomly-selected MgAl 2 O 4 -Sp grains in LMD578 do not show much variation in their overall appearance as well (Figure 7b).

Si-Disordering in Fully-Disordered MgAl 2 O 4 -Sp
In the MgAl 2 O 4 -Sp with SiO 2 contents as low as ~0.65-0.76wt %, the Raman peaks for the minor SiO 4 groups can be as intense as those for the major (Mg,Al)O 4 groups (Figures 6 and 7a), so that the relationships among the Raman intensity, SiO 2 content, Si disorder state and P-T condition are worth of further exploration.
We can write the formula [4] (Mg 0.333 Al 0.667 ) [6] (Al 1.333 Mg 0.667 )O 4 for a Si-free Mg-Al fully disordered MgAl 2 O 4 -Sp (x = 0.667).Ignoring the effect of small amounts of Si, one obtains [4] (Mg 0.333 Al 0.667 Si y ) [6]   [86]; c i is a constant), leading to where [Si y ] + [Si z ] = Si total = 0.0237 × SiO 2 wt % for cases with small amounts of SiO 2 (as implied by Equation ( 2)).With the SiO 4 groups represented by the Raman peaks at ~823 and 856 cm −1 and the (Mg,Al)O 4 groups by those at ~725 and 766 cm −1 , we obtain The term c SiO 4 ×0.0237 I 725 +I 766 is essentially a constant (C), so that Equation ( 5) can be briefed as Evidently, the variable I 823 +I 856 I 725 +I 766 of the Mg-Al fully-disordered MgAl 2 O 4 -Sp with certain y should be linearly correlated with the SiO 2 , and the curve should pass through the origin (the case of zero SiO 2 ).

Si-Disordering in Fully-Disordered MgAl2O4-Sp
In the MgAl2O4-Sp with SiO2 contents as low as ~0.65-0.76wt %, the Raman peaks for the minor SiO4 groups can be as intense as those for the major (Mg,Al)O4 groups (Figures 6 and 7a), so that the relationships among the Raman intensity, SiO2 content, Si disorder state and P-T condition are worth of further exploration.
We can write the formula [4] (Mg0.333Al0.667) [6](Al1.333Mg0.667)O4for a Si-free Mg-Al fully disordered MgAl2O4-Sp (x = 0.667).Ignoring the effect of small amounts of Si, one obtains [4] (Mg0.333Al0.667Siy) [6](Al1.333Mg0.667Siz)O4for the Si-containing Mg-Al fully-disordered MgAl2O4-Sp.Without knowing the y value, it is impossible to obtain the value of the constant C, which in turn impairs the application of Equation ( 6).Nevertheless, for the two extreme cases of all Si entering the T-site (y = 1) and Si attaining a fully disordered distribution (y = 0.333), the ratio of the two slopes (C and 0.333C, respectively) should be 3, which in fact represents the maximum ratio of any two slopes.
Our experimental data are summarized in Table 2, and shown in Figure 8.Both LMD563 and LMD558 ran at 4 GPa and 1550 • C, such that they formed a special group (Group 2) that acquired similar Si order-disorder states (identical y values).These two experimental data, plus the zero SiO 2 case, then define a curve for this particular y, with its slope of S 2 = 0.703(248).The uncertainty of the slope is somehow large, reflecting the limited accuracy of the data.5) read as 0.39 ± 0.05.d Ten Raman spectra were collected (Figure 7a), but only four of them were used here.Since the Raman spectra were numerically dominated by those without visible peaks for the SiO 4 groups, we selected four Raman spectra, with the SiO 4 Raman peaks ranging from the lowest to the highest, to derive our result in order to avoid possible data bias.Of course, this procedure might have led to new data bias.
where [Siy] + [Siz] = Sitotal = 0.0237 × SiO2 wt % for cases with small amounts of SiO2 (as implied by Equation ( 2)).With the SiO4 groups represented by the Raman peaks at ~823 and 856 cm −1 and the (Mg,Al)O4 groups by those at ~725 and 766 cm −1 , we obtain Without knowing the y value, it is impossible to obtain the value of the constant C, which in turn impairs the application of Equation ( 6).Nevertheless, for the two extreme cases of all Si entering the T-site (y = 1) and Si attaining a fully disordered distribution (y = 0.333), the ratio of the two slopes (C and 0.333C, respectively) should be 3, which in fact represents the maximum ratio of any two slopes.
Our experimental data are summarized in Table 2, and shown in Figure 8.Both LMD563 and LMD558 ran at 4 GPa and 1550 °C, such that they formed a special group (Group 2) that acquired similar Si order-disorder states (identical y values).These two experimental data, plus the zero SiO2 case, then define a curve for this particular y, with its slope of S2 = 0.703(248).The uncertainty of the slope is somehow large, reflecting the limited accuracy of the data.I 725 +I 766 and SiO 2 content.Using C = 1.02( 14), Equation ( 6) is shown as the pencil of broken lines radiating from the origin, with different y values ranging from 0.33 to 1.For every sample, the error bar of its I 823 +I 856 I 725 +I 766 was directly calculated from the integrated areas of the used peaks of all Raman analyses.
The curve constrained by the experiments of Group 2 divides the remaining four experiments into two groups, with one group including LMD565 and LMD564 conducted under relatively low P-T conditions (Group 1 with larger y), whereas the other group, including LMD578 and LMD568, was conducted under relatively high P-T conditions (Group 3 with smaller y).Due to the good linear relations (Figure 8), we attempted weighted linear least-squares fit and obtained S 1 = 0.935 (17) for the experiments of Group 1 and S 3 = 0.349 (26) for the experiments of Group 3. The assumption behind this practice is that the y values of the MgAl 2 O 4 -Sp from the experiments in either Group 1 or Group 3 are constant.Whether this assumption is justified or not is unimportant, since one can always draw a line through the origin and one single experimental data point, and subsequently define a slope for that particular case.The key observation here is that the ratio between S 1 and S 3 is 2.68 (21), a value close to 3.This means that the curve defined by the experiments of Group 1 generally approximates the case of all Si residing on the T-site (y = 1), and the curve defined by the experiments of Group 3 closely approaches the case of a fully random Si distribution (y = 0.333).It thus follows that with small variations of P and T, from 3-4 GPa to 5-6 GPa, and from 1500 to 1630-1650  (31), respectively.Indeed, the constant C is constant, averagely 1.02 (14), which then allows us to add into Figure 8 a set of curves with fixed y values to show the relationship between the I 823 +I 856 I 725 +I 766 and SiO 2 .Some interesting points emerge from Figure 8. Firstly, the Raman peaks of the minor SiO 4 group are very prominent, compared to those of the major (Mg,Al)O 4 group.For ~1.1 wt % SiO 2 fully ordered on the T-site (y = 1), for example, the Raman peaks at ~823 and 856 cm −1 are generally as intense as the Raman peeks at ~725 and 766 cm −1 .Secondly, the behavior of the Raman peaks of the SiO 4 group strongly correlates with the SiO 2 content: relatively weak and with little change for the SiO 2 -poor MgAl 2 O 4 -Sp, but strong and with significant variation for the SiO 2 -rich MgAl 2 O 4 -Sp.Thirdly, the Si-disordering process is independent of the SiO 2 content, but is controlled by the formation P and T of the MgAl 2 O 4 -Sp.When the P-T conditions change from ~3-4 GPa and 1500 • C to ~5-6 GPa and 1630-1650 • C, the Si cations radically change from fully ordering on the T-site (y = 1) to randomly distributing between the T-site and the M-site (y = 0.333).For the MgAl 2 O 4 -Sp with similar SiO 2 contents, finally, the ones displaying relatively strong Raman peaks at ~823 and 856 cm −1 should have formed under a relatively low P-T environment, and vice versa.

Implications
Electrostatic lattice energy calculations and consideration of the structure of the Sp group of minerals suggest that the larger Mg cations prefer the T-site and the smaller Al cations prefer the M-site, resulting in a generally normal MgAl 2 O 4 -Sp at ambient P and T [59].This principle seems inapplicable to the minor components.The present study indicates that at P-T conditions ≤~3-4 GPa and 1500 • C, covering the P-T range of the top upper mantle of the Earth [98], the even smaller Si cations incorporated by the MgAl 2 O 4 -Sp structure appear on the T-site, rather than on the anticipated M-site (y = 1; Figure 8).This result is compatible with existing single-crystal XRD studies on terrestrial Sp, which locate Si on the T-site [73,[75][76][77]99].The current study further shows that presenting as SiO 4 groups in the Sp, a small amount of SiO 2 like ~1 wt % exhibits very intense Raman peaks at ~823 and 856 cm −1 , and can completely alter the stereotypical overall appearance of the Raman spectra established with some SiO 2 -poor natural 2-3 Sp.Since Si readily enters the 2-3 Sp structure, this result should have important application in identifying the Sp phase, particularly for the circumstances where direct petrographic observation cannot be made.A Raman spectrometer will be launched shortly as part of the ExoMars analytical laboratory and deployed on the Martian surface to investigate the mineralogical and biological aspects of the Mars [100,101].Considering the wide spreading of the 2-3 Sp on the Earth, the Moon, and the extraterrestrial planets, asteroids and meteorites, it will have high chance to encounter some Sp and collect in-situ Raman spectra.A correct interpretation of these Raman spectra must critically evaluate the effect of Si.
Si starts to enter the M-site of the MgAl 2 O 4 -Sp at P-T conditions ≥~3-4 GPa and 1500 • C, and become fully disordered under P-T conditions ≥~5-6 GPa and 1630-1650 • C (Figure 8).However, the 6-coordinated Si may not be easily observable in natural MgAl 2 O 4 -Sp.High-P experimental studies have shown that Al-rich 2-3 Sp is not a stable phase for the upper mantle at P > ~3 GPa [102].On the other hand, adding Cr may stabilize the 2-3 Sp to a much higher P [4], and encapsulating the 2-3 Sp in diamonds may lead to the same result [103].The Cr-rich 2-3 Sp inclusions in diamonds are thus the best targets in which to look for the 6-coordinated Si.
The almost random Si distribution observed for our Si-bearing MgAl 2 O 4 -Sp at P-T conditions ≥~5-6 GPa and 1630-1650 • C strongly hints that at some high P-T conditions the Si cations in the (Mg,Fe) 2 SiO 4 -Sp (Rw) might be disordered to large extents.Mg 2 SiO 4 -Rw has been conventionally regarded as a normal 4-2 spinel with nearly all Si taking the T-site.The single-crystal XRD data of Sasaki et al. [64] and high-resolution 29 Si NMR data of Stebbins et al. [61] did not show any convincing evidence for 6-coordinated Si.In contrast, ~4% Si was inferred to appear on the M-site, based on the systematic deviations of the Si-O bond length determined by new single-crystal XRD data from an average value in silicates [60].Consideration of the bond length systematics and experimentally measured cation distributions led to a similar conclusion [59].However, all these conclusions were drawn from the experimental data collected on quenched samples or based on some crystal structural features established for ambient P. In the former cases, the cation disorder information of the Rw at high P might be completely lost.In analogy to the well-known partial preservation of the high-T equilibrium state of the Al-Mg disorder in the MgAl 2 O 4 -Sp after quenching [42,53], reordering the Si and Mg cations in the Mg 2 SiO 4 -Sp presumably happens fast and proceeds towards its completion as the high-P synthesizing experiment quenches.In the latter cases, the bond length systematics and structural features established for ambient P might not be applicable to the high-P structures.As pointed out by Méducin et al. [45], P has a significant impact on the order-disorder process of the MgAl 2 O 4 -Sp, especially in the T range of 477-1227 • C. Some high-P single-crystal XRD investigations have been conducted up to ~28.9 GPa at ambient T, but could not shed light on the Si disorder issue, partially due to the low experimental T potentially unable to trigger the order-disorder reaction, and partially due to the low data resolution caused by the similar X-ray scattering factors of Mg and Si [104,105].
The most likely evidence in the literature of the presence of 6-coordinated Si in the Rw have come from a high-P Raman spectroscopic investigation on synthetic Mg 2 SiO 4 -Rw [106] and a spectroscopic study on some meteoritic Rw [62].At P > ~30 GPa, a weak and diffusive Raman peak appeared and was interpreted as the signature for the presence of Si-O-Si linkages and/or partial increase in the coordination of Si [106].We propose that this peak might belong to the MgO 4 groups in the Mg 2 SiO 4 -Rw, which would in turn indicate the presence of the SiO 6 groups resulted from the position exchange of the Si and Mg cations.According to Chopelas et al. [107], the MgO 6 groups in the normal Mg 2 SiO 4 -Rw are Raman-silent, and the SiO 4 groups are responsible for all the Raman peaks.Since the order-disorder process in the Sp is non-convergent (i.e., the symmetry of the Sp is maintained at any inversion), no new Raman peaks should be expected from the SiO 6 groups in the disordered Mg 2 SiO 4 -Rw, exactly like what we have observed for the Si-bearing MgAl 2 O 4 -Sp (Figures 6  and 7).On the line of the study about the meteoritic Rw, Taran et al. [62] used a range of analytical methods including optical absorption spectroscopy to investigate some synthetic (Mg,Fe) 2 SiO 4 -Rw, and two compositionally homogenous but doubly-colored meteoritic Rw grains (Grain 1, one part being colorless and the other part blue; Grain 2, one part being blue and the other part dark blue) from two L6-type ordinary chondrites NWA 1662 and NWA 463.They proposed that for the meteoritic Rw, the part with no color was inverse Rw, other parts with various colors were Rw with different amounts of cation inversion.In order to confirm their hypothesis, more investigation should be conducted on the meteoritic Rw, which represents the best natural specimen for studying high-P structural features, including the Mg-Si order-disorder state, due to its having much larger quench rates.Rw of various colors has been documented in many meteorites, including L ordinary chondrites [108][109][110][111][112], LL ordinary chondrites [14,26], and Martian meteorites like the shergottites [32,34,37].If the relationship among the color, composition, inverse magnitude, P and T can be adequately quantified, a fine scale for accurately estimating the shock P-T conditions may be derived, which may serve well the theoretical evolution models of the early solar system.
If the Rw in the LP-MTZ attained substantially higher degrees of inverse than those experimentally observed so far, the mineralogical model of the upper mantle and the nature of the 520-km and 620-km seismic discontinuities would need further careful examination.Some empirical and theoretical studies have demonstrated that the cation disorder process in the Rw leads to significantly larger thermal expansion coefficients, smaller bulk modulus, and smaller shear modulus [19,44,46,63].As a result, a 12.5% Si-Mg disorder can decrease the seismic velocities by ~3-5% [19,46].Direct experimental investigations on the cation inversion of the Rw at the P-T conditions of the LP-MTZ are therefore of high priority.
Author Contributions: Xi Liu (designing the project).Liping Liu (writing the initial draft of the work).Xi Liu (writing the final paper).Xi Liu, Liping Liu and Xinjian Bao (interpreting the results).Liping Liu, Qiang He and Renbiao Tao (conducting high-P experiments).Liping Liu, Wei Yan, Yunlu Ma and Mingyue He (collecting Raman spectra).Xinjian Bao, Liping Liu and Ruqian Zou (collecting and interpreting the single-crystal X-ray data).Liping Liu (performing EMP analyses).Mingyue He (helping in checking the draft of the paper).All authors discussed the results and commented on the manuscript.
(u = 0.2633); the x value of the latter sample was claimed to be ~0.02,which might have been slightly underestimated (more discussion later).On the other hand, it was not observed for the natural MgAl 2 O 4 -Sp studied by Cynn et al. [49], Van Minh & Yang [87] or Slotznick & Shim [57], implying x values smaller than at least ~0.145.No Raman spectra previously collected on unannealed natural MgAl 2 O 4 -Sp showed the weak peak at ~222 cm −1 .The sample studied by Chopelas & Hofmeister

Figure 4 .
Figure 4. Raman features of Si-free N-Sp and synthetic Si-free MgAl2O4-Sp from LMD487.

Figure 4 .
Figure 4. Raman features of Si-free N-Sp and synthetic Si-free MgAl 2 O 4 -Sp from LMD487.

Figure 5 .
Figure 5. (a) Comparison of P-T conditions of our high-P Sp-synthesizing experiments and those of the heating-up experiments closely approaching Mg-Al redistribution equilibrium at T ≥ 500 °C from Méducin et al. [45].The P and T values of the five experiments from Méducin et al. [45] were highly correlated, as shown by the solid line P = 0.0035(2)*T − 1.36(16).(b) x-T relation of those five experiments from Méducin et al. [45], as shown by the solid line x = 0.00043(3)*T − 0.053(32).Filled diamonds are for the five experiments from Méducin et al. [45], whereas the empty triangle is for our experiment synthesizing Si-free MgAl2O4-Sp and the empty squares are for our experiments synthesizing Si-bearing MgAl2O4-Sp.The broken line in (b) is shown for a hypothetical fully disordered MgAl2O4-Sp with x = 0.667.

Figure 5 .
Figure 5. (a) Comparison of P-T conditions of our high-P Sp-synthesizing experiments and those of the heating-up experiments closely approaching Mg-Al redistribution equilibrium at T ≥ 500 • C from Méducin et al. [45].The P and T values of the five experiments from Méducin et al. [45] were highly correlated, as shown by the solid line P = 0.0035(2)*T − 1.36(16).(b) x-T relation of those five experiments from Méducin et al. [45], as shown by the solid line x = 0.00043(3)*T − 0.053(32).Filled diamonds are for the five experiments from Méducin et al. [45], whereas the empty triangle is for our experiment synthesizing Si-free MgAl 2 O 4 -Sp and the empty squares are for our experiments synthesizing Si-bearing MgAl 2 O 4 -Sp.The broken line in (b) is shown for a hypothetical fully disordered MgAl 2 O 4 -Sp with x = 0.667.

Minerals 2018, 8 ,
x FOR PEER REVIEW 10 of 21 fully polymerized into a three-dimensional framework, so that the Raman features of low Qz and Coe should be very different to those of potential SiO4 groups in the Sp structure.

Figure 6 .
Figure 6.Raman features of synthetic Si-bearing MgAl 2 O 4 -Sp from our high-P experiments.As comparisons, Raman spectra of N-Qz, synthetic Coe, St and Fo (LMD558; Table1) are shown as well.For the purpose of illustration, some portions of the Raman spectra of St, Coe and N-Qz have been expanded and shown as insets, whereas the entire Raman spectrum of Fo has been compressed by a factor of 60.LMD563(0.30),Exp.# followed by the SiO 2 content of the Sp.To illustrate clearly, only one Raman spectrum is shown for each synthetic phase, although multiple Raman spectra have been collected.
(Al 1.333 Mg 0.667 Si z )O 4 for the Si-containing Mg-Al fully-disordered MgAl 2 O 4 -Sp.The Si disorder state is then defined as y = [Si y ]/([Si y ] + [Si z ]) = [Si y ]/[Si total ], with y = 1 indicating all Si on the T-site, y = 0 indicating all Si on the M-site, and y = 0.333 indicating a random Si distribution.Under certain analytical conditions in the Raman spectroscopic experiments, the intensity of a Raman peak caused by one type of structural unit i (SiO 4 here) is proportional to its abundance ([i]; [Si y ] here),

Minerals 2018, 8 , 21 Figure 7 .
Figure 7. Raman spectra of MgAl2O4-Sp with almost identical amounts of SiO2 from LMD558 (a) and LMD578 (b).The two sets of Raman spectra (A1, A2, A3 and A4 as Set A, and B1, B2, B3 and B4 as Set B) shown in (a) were collected from the only Sp grain shown in Figure 2a, but with their orientations normal to each other.After obtaining the Raman spectra of Set A, we reprocessed the sample to make a new exposure normal to the previous one and then collected the Raman spectra of Set B. Ten Raman spectra shown in (b) were acquired from ten different Sp grains (see Figure 2b for the positions).Due to data compression, the weak Raman peaks for the SiO4 groups of the MgAl2O4-Sp from LMD578, visible in Figure 6, are now barely discernable in (b).

Figure 7 .
Figure 7. Raman spectra of MgAl 2 O 4 -Sp with almost identical amounts of SiO 2 from LMD558 (a) and LMD578 (b).The two sets of Raman spectra (A1, A2, A3 and A4 as Set A, and B1, B2, B3 and B4 as Set B) shown in (a) were collected from the only Sp grain shown in Figure 2a, but with their orientations normal to each other.After obtaining the Raman spectra of Set A, we reprocessed the sample to make a new exposure normal to the previous one and then collected the Raman spectra of Set B. Ten Raman spectra shown in (b) were acquired from ten different Sp grains (see Figure 2b for the positions).Due to data compression, the weak Raman peaks for the SiO 4 groups of the MgAl 2 O 4 -Sp from LMD578, visible in Figure 6, are now barely discernable in (b).

I SiO 4 II
(Mg,Al )O 4 = I 823 + I 856 I 725 + I 766 = c SiO 4 ´y ´0.0237 ´SiO 2 wt% I 725 + I 766 725 + I 766 is essentially a constant (C), so that Equation (5) can be briefed as I 823 + I 856 I 725 + I 766 = C ´y ´SiO 2 wt% (6) Evidently, the variable I 823 + I 856 I 725 + I 766 of the Mg-Al fully-disordered MgAl2O4-Sp with certain y should be linearly correlated with the SiO2, and the curve should pass through the origin (the case of zero SiO2).

Figure 8 .
Figure 8.I 823 +I 856 I 725 +I 766 vs. SiO 2 content of our synthetic Si-bearing MgAl 2 O 4 -Sp.Note that the analytical conditions in the Raman spectroscopic experiments were identical, and all the MgAl 2 O 4 -Sp generally had the maximum amount of Mg-Al disorder (x = 0.667).The experimental P-T conditions are indicated along the symbols; 4/1500, for example, should be read as 4 GPa and 1500 • C. The experiments have been divided into three groups, with Group 1 containing LMD565 and LMD564 (red squares), Group 2 containing LMD563 and LMD558 (black squares), and Group 3 containing LMD578 and LMD568 (blue squares).With the aid of the origin (the zero SiO 2 case), the experiments in each group were used to determine the relationship between the I 823 +I 856I 725 +I 766 and SiO 2 content.Using C = 1.02(14),Equation (6) is shown as the pencil of broken lines radiating from the origin, with different y values ranging from 0.33 to 1.For every sample, the error bar of its I 823 +I 856 I 725 +I 766 was directly calculated from the integrated areas of the used peaks of all Raman analyses.

Table 1 .
(25)rimental conditions, phase assemblages, and compositions of spinels and quartz (wt %).Number in the parenthesis after the name of the phase is the number of successful EMP analyses performed on that phase.Sp, spinel; Melt, silicate melt; Fo, forsterite; Grt, garnet; Coe, coesite; St, stishovite; Qz, quartz.cNumber in the parenthesis is the analytical uncertainty reported as one standard deviation.28.66(25)read as 28.66 ± 0.25.d Starting material is a mixture of dried high-purity MgO and Al 2 O 3 powders, weighted out according to the stoichiometry of the MgAl 2 O 4 spinel.
a P, pressure in GPa; T, temperature in • C; t, time in h.b e Starting material is a dried high-purity SiO 2 powder, with some deionized water added later.

Table 2 .
Ratio of integrated area of the Raman peaks at ~823 and 856 cm −1 for the SiO 4 group to those at ~725 and 766 cm −1 for the (Mg,Al)O 4 group.
a P, GPa; T, o C; SiO 2 , SiO 2 content (wt %) in our synthetic Sp. b Number of Raman spectra collected.c Number in the parenthesis represents one standard deviation; 0.39( • C, Si in the Mg-Al fully-disordered MgAl 2 O 4 -Sp drastically changes from a fully-ordered distribution on the T-site to completely random distribution.With the y values for the Mg-Al fully-disordered MgAl 2 O 4 -Sp from LMD565, LMD564, LMD578 and LMD568, we have calculated the constant C, and obtained 0.93(15), 1.19(58), 0.88(59) and 1.08