Mantle peridotite xenoliths in basaltic lavas of the Lake Nyos volcanic unit are predominantly spinel-bearing lherzolites and harzburgites (Teitchou, 2008). Most of them (more than 60% of collected samples) are spinel lherzolites, composed of olivine (55–70%), orthopyroxene (7–30%), clinopyroxene (6–20%) and spinel (1–6%). Less abundant (30%) harzburgites are composed of olivine (71–83%), orthopyroxene (10–20%) and clinopyroxene (2–4%). The mineralogy and geochemistry of these xenoliths have been recently thoroughly investigated (Teitchou, 2008; Teitchou et al., accepted). A number of these samples was then investigated for fluid inclusions, focusing on lherzolites as these contain by far the most abundant inclusions (see below). As all samples present more or less the same characteristics, it was felt that five samples would give a good representation of the xenolith population as a whole: four lherzolites (NK05, NK09, NK11 and NK14) and one harzburgite (NK10). Texture of the latter is protogranular, against porphyroclastic for the lherzolites (for NK14, transitional porphyroclastic to equigranular). In fact, sample NK14 turned out to be the most typical and interesting, and it has served as a basis for this presentation. Other lherzolite samples are however rather comparable, showing basically the same features than those described in the present work.

For all xenoliths, details of the geochemistry (with the same samples numbers) are given in Teitchou (2008) and Teitchou et al. (accepted). It will be sufficient here to say that REE-contents in clinopyroxenes, normalized to primitive mantle, point to the existence of two groups of xenoliths:

• (i) group 1, with “spoon” shaped patterns (HREE enrichment compared to MREE and LREE; [La /Sm] _N: 0.08–5.50; [Sm /Yb] _N: 0.47–1.17 and [La/Yb] _N: 0.03–6.45). In the present study, this group includes lherzolites NK05, NK09 and NK11, as well as harzburgite NK10;
• (ii) group 2 (in the present study: lherzolite NK14) shows “LREE-enriched” clinopyroxenes; [La /Sm] _N: 0.73–1.35; [Sm /Yb] _N: 3–6.25 et [La /Yb] _N: 3.25–4.5. HREE are also significantly fractionated, as evidenced by [Dy/Lu] _N ratios ranging from 1.6 to 3 (group 1: 1–1.1).

As discussed in detail elsewhere (Teitchou, 2008; Teitchou et al., accepted), these geochemical features can be explained by a low degree of mantle melting (around 5%) followed by extensive metasomatic processes, which for group 2 can even erase the partial melting signature. In all cases, candidate for metasomatic agent has been estimated to be more or less alkaline- and carbonated – mafic silicate melts.

As usual, inclusions were observed in double polished plates, in the present case roughly 60 μm thick. Compared to most inclusion plates, this relatively small thickness (approximately the double of an ordinary thin section) has the disadvantage that dense fluids may have escaped from the cavity, the internal pressure in the inclusion, which as indicated below can reach up to 1 GPa at the time of the volcanic eruption, being still of the order of 100's bars at room temperature. But most inclusions would have leaked anyway, even in much thicker plate. A relatively thin one, much closer to an ordinary thin section, offers a better possibility to observe detailed relationships between the inclusions and the host mineral, in the present case the most interesting aspect of our study.

Compared to many other cases, e.g. in the French Massif Central (Bournac: Bilal and Touret, 1976) or southern Italy (Heblei: Sapienza and Scribano, 2000), Lake Nyos xenoliths contain relatively few inclusions. None were found in a number of investigated samples and, in those retained for the present investigation, inclusions occur only in few spots of any thin section (Fig. 1). Many mineral sections are inclusion-free, but in some of them inclusions, not especially small (size typically in the 5–30 μm range), can be found by careful observation, either along grain boundaries or, more interestingly, within mineral grains. Like in most lower crustal- or mantle xenoliths, these inclusions belong broadly to two categories:

3.1 Late melt (mainly glass) inclusions

3.2 Early CO₂ inclusions

Far more interesting is the second type of inclusions, dark and monophase (gaseous), which as first sight look like empty cavities (Figs. 1 and 2). Most of them are indeed now empty, fluids inside the cavity having escaped during the eruption or sample preparation. However, the shape (elongated tubes or, more typically, negative crystal), general appearance, and location within the mineral host are not ambiguous: as first shown in 1965 by E. Roedder (1984), these inclusions do or did contain a supercritical fluid, which in most cases investigated so far turned out to be pure (or almost pure) CO₂. A complete investigation would require microthermometry measurements, in order to determine the fluid density, as well as Raman analyses to check its purity. These measurements were not done in the present study, but an independent investigation (Pintér et al., 2009), that was brought to our attention after the submission of this manuscript, provided us with additional information, which could be used in our discussion. In the literature, average maximum CO₂ densities recorded so far in mantle xenoliths are close to 1 g/cm³, corresponding to an homogenization temperature (to liquid) of about −10 °C (Bilal and Touret, 1976, Evans et al., 1994, Honda et al., 1995). This gives an internal overpressure of roughly 1 GPa at the surface at the time of the eruption, illustrating the remarkable strength of the mineral host in the rapidly evolving volcanic system. These values would correspond to about 30 km depth for the formation of these inclusions. These estimates are however very minimal. We discuss below the reasons why some inclusions at least were made at the time of mineral equilibration in the xenoliths, with P = 2 GPa and T = 1300 °C as reasonable estimates (Teitchou, 2008, Teitchou et al., accepted). The depth of inclusion formation would then be about 60 km, with a CO₂ density, estimated from the PVT equation of state (Touret and Bottinga, 1979), above 1.2 g/cm³.

It is known by experience that, even for volcanoes occurring in apparently similar settings, the number of inclusions occurring in any xenolith varies grossly from one volcano to the other. In some cases (e.g. Bournac, French Massif Central and Heblei, Sicily), peridotite xenoliths contain quite abundant inclusions, occurring in most mineral phases virtually, whereas others are virtually inclusion-free. This is notably the case for peridotites with equigranular texture. The case of Lake Nyos is intermediate, in this respect most interesting. CO₂-inclusions may be locally abundant, but only within some mineral grains, which in all cases observed so far in the samples we have investigated, turned out to be clinopyroxenes. Pintér et al., 2009 report “… an older inclusion generation occurring randomly in olivine and orthopyroxene, partly or completely decrepitated…” (p. 189, slightly summarized). We have not observed such inclusions in the samples we have studied.

In the example shown on Fig. 1 (sample NK14), only two pyroxene grains in the whole thin section did contain inclusions. These are shown in Figs. 1 and 2 A and C, respectively. Inclusions are isolated or, more typically, they form tridimensionnal “clusters” (Touret, 2001) dispersed within the mass of the mineral host. Some clusters, containing few 10’ inclusions, are limited to few spots in the mineral host, whereas others contain many more inclusions, giving to the host a vacuolar, spongy appearance (Fig. 2A). In this case, neighboring inclusions are often disposed along a roughly planar surface (“trails”), which may eventually cross the grain boundary and grade into the next crystal. Such a trail is shown in Fig. 2B.

The most typical example of well-defined cluster that we have found, in lherzolite sample NK 14, is shown in Figs. 1 and 2C, respectively. About 30 inclusions, all almost identical in size (diameter about 30 μm) and shape (roughly spherical, close to negative crystal) are regularly disposed in the volume of the mineral host, not along any planar surface. The center of the cluster does not contain any inclusion. It is occupied by a low-polarizing zone with indistinct boundaries, which includes an intricate mixture of microcrystals and low-birefringence mineral phases, possibly devitrified glass. A total of 33 microprobe analyses have been done on three clinopyroxene crystals from samples NK14: 25 within the cluster shown in Fig. 2C, the other eight on two crystals containing spinel, either in inclusions (analyses 26 to 29), or in exsolutions (analyses 30 to 33). All analyses (available upon request to the corresponding author) give roughly comparable results, with the following composition, expressed in % of clinopyroxene solid solution end members (wollastonite [Wo(Ca)], enstatite [En (Mg)] and ferrosilite [Fs(Fe2+)]: Wo [Ca]) between 44.6 and 46.2, En (Mg) between 47.7 and 49.7, and Fs(Fe2+) between 4.3 and 6.8. The Mg number (X _Mg = Mg /Mg+Fe2+) is also relatively constant at 0.8 to 0.9. For this last parameter, the most sensible indicator of xenolith mineral equilibration, this small variation range is not at random. The 25 analyses mentioned above have been done along two orthogonal profiles, indicated as lines 1 and 2 in Fig. 3. Analyzed spots, with corresponding numbers, are reported on the profiles. As shown on the graph of Fig. 3, the X _Mg number along these profiles shows a distinct decrease towards the center of the “fusion” zone, indicating that the composition of the mineral host has not fully homogenized after the formation of the melting zone and related inclusion cluster.

The fact that CO₂-inclusions occur dominantly in clinopyroxenes, also that they are not related to local melting in basaltic lavas, precludes the possibility that fluids were issued from the xenolith host-basalts. If this was the case, the late melt inclusions would contain CO₂, which would also occur in secondary trails, cutting all minerals. This situation has been observed in a number of volcanic xenoliths, notably in the French Massif Central (e.g. Bournac: Bilal and Touret, 1976). It is possible that some late trails reported by Pintér et al., 2009 for their “younger generation” have this origin. We believe however that it is more probable that they are “pseudo-secondary” features (Roedder, 1984), generated inside the mineral host. What we have observed in our samples is different: all inclusions that we have seen, either primary, made at the time of the mineral equilibration of the pyroxene host, or, if occurring in clusters, “pseudo-secondary”, resulting from the decrepitation of a former primary inclusions. The “melted” zone shown in Fig. 3 is a good example. It contains only inclusions which have been formed inside the crystal host, without evidence of fluids brought from outside. CO₂ there must have been generated “in-situ”, from a mineral reaction which has led to the local melting of the pyroxene. A most plausible explanation is the destabilization of a calcium-rich carbonate (calcite) at the time of pyroxene melting. This would not only liberate CO₂, but also Ca ions, which by diffusion in the pyroxene structure could depress X_Mg, as shown in Fig. 3. Such a mechanism is further supported by a number of arguments. We know that peridotites under the Cameroon Volcanic Line show a strong evidence of mantle metasomatism, caused by carbonate-rich percolating melts (or fluids) (Teitchou et al., accepted). We know also that clinopyroxenes are the “sensible” minerals in “fertile” peridotites, being the first mineral phases destroyed during partial melting and formation of basaltic melts. The cluster shown in Fig. 3, with its central melted zone, is then believed to be a direct illustration of this process.

In conclusion, we believe that CO₂ was first produced at mantle depth by carbonate destabilization during interaction between percolating metasomatic melts (or fluids) and mantle peridotites. Phase stability in a carbonated mantle (i.e. in presence of CO₂ fluids, carbonate melts, and /or carbonates) has been thoroughly studied experimentally (e.g. Frezzotti et al., 2009). At “normal” mantle temperatures (i.e. about 1250 °C), carbonates, either occurring as mineral phases or dissolved in melts, decompose at a pressure of about 2 GPa (depth about 60 km). As indicated above, cations (Ca, Mg, Fe) can be incorporated into silicate structures notably pyroxenes, whereas free CO₂ considerably depresses the dry peridotite solidus (by about 200 °C, leading to partial melting and formation of alkaline basalts, see e.g. Fig. 6 in Frezzotti et al., 2009). Free CO₂ appears thus when alkali basalt melts are first generated, eventually supplying a considerable amount of potential energy to boost the basalts towards the surface. Even if, in the present case, there is no reliable mineral P–T indicators in the xenoliths themselves, we can have an idea of possible P–T conditions at which this sequence of events has happened from commonly assumed conditions of mantle partial melting during the formation of basaltic melts: T about 1250–1300 °C, P = 2–2.5 GPa. CO₂ equation of state would then give a CO₂ density (or molar volume), of about 1.2 g/cm³ (35 cm³/mole) (CO₂ PVT extrapolated from Touret and Bottinga, 1979, pers. com. M.L. Frezzotti). Microthermometry measurements made by Pintér et al., 2009, even if they raise new problems, are in line with these estimates. They have measured in their investigated samples, both in Nyos and Barombi, low homogenization temperatures to liquid, typical for dense CO₂-rich fluids. For the pure CO₂ system, the relation between homogenization temperature and fluid density is immediate, from CO₂-PVT data. But the result in terms of density (or molar volume) can be somewhat modified, due to the possible occurrence of other gases. Some melting temperatures measured by Pintér et al. (2009) are very low, between −68 and −95 °C for some Nyos samples. These very low values are certainly due to the presence of a polar gas, such as CH₄ or, more probably N₂. However, in all cases of comparable melting temperatures so far encountered by the senior author, homogenization did occur to gas, indicating low-density fluids. On the other hand, fluids which homogenize to liquid in the vicinity of CO₂ triple point (−56.6 °C), as measured by Pintér et al., 2009 in Barombi samples, have always proven to be almost pure CO₂, which can reasonably well be expressed in term of “CO₂-equivalent. The “quite low homogenization temperature of CO₂-rich phase to liquid phase, ranging from −53 to −44 °C in Nyos and −52 to −38 °C in Barombi xenoliths” (p. 190, Pintér et al., 2009) corresponds presumably to this type of fluids. For Nyos, these data would correspond to a CO₂-equivalent density of 1.15 and 1.13 g/cm³, respectively (Touret and Bottinga, 1979). Raman measurements would be very desirable, and hopefully they will be done in the future. We feel however confidant that they will not change the main conclusion of the present work for the scope of this paper, namely that the CO₂ density at the source was well above that of liquid water.

All petrographical evidences outlined above indicate that CO₂ first liberated at the mantle source had a density at least equal, most probably significantly higher that of liquid water (1 g/cm³). This result is by no mean specific to lake Nyos: from the experience of the first author, high-density CO₂ inclusions have been found (notably in the French Massif Central) or can be suspected in most mantle xenoliths investigated so far. In most eruptions, the volcanic column continues without interruption up to the surface. Volume is not limited, and deep CO₂ progressively expands, with a considerable expansion and rapid decrease of the density. This phenomenon is very general, explaining the considerable amount of mantle-derived CO₂ released during or after the eruptions in some volcanic areas (e.g. southern Italy, Sapienza and Scribano, 2000). The case of a crater lake, like in Nyos, is different: water may infiltrate in the chimney at some kilometer depth, forming a fixed volume in which deep CO₂ has no possibility to expand. Then, like for any inclusion (also a constant volume system), CO₂ has to follow an isochore (equal density line). For a reference density of 1 g/cm³, the CO₂ isochore starts in the vicinity of CO₂ triple point, at about −50 °C. It crosses the almost vertical liquid H₂O isochore at a temperature of about 30 °C and a pressure of about 0.1 GPa (Fig. 4). These P–T conditions (I in Fig. 4) define the density inversion for pure H₂O (liquid) and 1 g/cm³ CO₂ systems. At higher temperature, the relevant CO₂ isochore crosses successively H₂O isochores of lower density: CO₂ fluids are heavier than H₂O, they tend to sink. At lower temperature, they are lighter. CO₂-fluids tend then to rise to ultimately reach the surface. Intersections of CO₂ and H₂O-isochore pairs of identical densities define the complete density inversion curve, shown as l in Fig. 4. The situation that we know at the Earth's surface, namely gaseous CO₂ bubbles rising in liquid H₂O, is then only valid for a limited domain of geological interest, not exceeding a few kilometer depth. Below this level, CO₂ tends to sink in liquid water, as long as the temperature remains above few 10's °C. Such a situation has been observed on several occasions in deep-sea ocean waters. It had been reported in early (unfortunately secret) Soviet literature (A. Chupin, Novosibirsk, pers. com. to J.T.). But it has also been found more recently (1995) during submersible experiments in Japan (Shinkai 6500 expedition, Honda et al., 1995). H₂O–CO₂ density inversion was then observed to occur at a depth of 2569 m at the low temperature (about 4 °C) of deep ocean water. This low temperature may somewhat complicate the problem, notably through the possible formations of gas hydrates (clathrates). Nevertheless, it gives a reasonable order of magnitude, which confirms the data shown in Fig. 4.

In conclusion, we propose that the sudden gaseous bursts observed in the Cameroon volcanoes are caused, not by one, but by two successive density inversion phenomena. First of all, as advocated by all students of these remarkable eruptions, a shallow one, indicated as d2, in Fig. 4. Strictly confined to the lakes, it corresponds to a density inversion between deep, CO₂-loaded waters near the bottom of the lake, and denser, CO₂-free waters near the surface. But the results of the present study suggest that this d2 density inversion has been preceeded (and fed) by a deeper density inversion (d1, Fig. 4), at the level between infiltrating lake waters in the volcanic pipe, to a depth of about 3 km. Trapped dense CO₂, issued from mantle depth during the last eruption, some 500 Ky ago (Aka et al., 2008), could quietly remain in a deep reservoir, between the magma chamber and the water level, as long as the regional temperature at depth remains above the density inversion temperature. The heat accumulated during the eruption, the high geothermal gradients in rifted volcanic areas, suggest that the major part of this deep CO₂ is still probably under these conditions. Only in the contact zone with descending waters, temperature may be low enough to initiate the d1 inversion, leading to a regular flow of gaseous CO₂ towards the surface. Understandably, the present-day gas recharges in both Nyos and Monoun lakes are carefully controlled, notably through the international Nyos and Monoun Degassing Project (NMDP) (Issa et al., 2006). Measurements between August 1987 and January 2004 have indicated a recharge in lake Monoun of 8 Mmol/year, constant during this entire time interval. This constancy indicates that a large part of the “deep” reservoir remains untouched at depth, with (in the absence of external hazards, such as earthquakes or new eruptions) no reason for the recharges to change significantly in a foreseeable future.