Melting temperatures of MgO under high pressure by micro-texture analysis

Periclase (MgO) is the second most abundant mineral after bridgmanite in the Earth's lower mantle, and its melting behaviour under pressure is important to constrain rheological properties and melting behaviours of the lower mantle materials. Significant discrepancies exist between the melting temperatures of MgO determined by laser-heated diamond anvil cell (LHDAC) and those based on dynamic compressions and theoretical predictions. Here we show the melting temperatures in earlier LHDAC experiments are underestimated due to misjudgment of melting, based on micro-texture observations of the quenched samples. The high melting temperatures of MgO suggest that the subducted cold slabs should have higher viscosities than previously thought, suggesting that the inter-connecting textural feature of MgO would not play important roles for the slab stagnation in the lower mantle. The present results also predict that the ultra-deep magmas produced in the lower mantle are peridotitic, which are stabilized near the core–mantle boundary.

The authors are using two methods to infer melting of MgO: laser power-temperature relationship and quenched micro-texture. While these are common ways that some studies have inferred melt, the methods have problems, as they show in their manuscript. Quenched texture may reflect solid-solid structural phase transitions or even "shear-induced plastic flow." Discontinuities in laser power vs. temperature profiles, used frequently to infer letting temperature, may also be caused by recrystallization or grain growth prior to melting. They did not mention, perhaps on purpose, that many think that the plateau in laser power vs. temperature may be due to latent heat, although this has been shown to not likely be the case (e.g., Geballe & Jeanloz, 2012).
Their experiments are similar to those run by (Zerr & Boehler, 1994) in which they use singlecrystal MgO in an Ar pressure medium/thermal insulation and heating by CO2 laser. In the earlier study, Zerr & Boehler used the visual observation of a large increase in temperature to mark the onset of melting inferring that the melt could now absorb the laser energy more readily. The current study uses temperature plateaus and quenched texture. The initial test of the system of melting MgO at room pressure yielded a promising temperature in that it was very similar to literature values. More information is needed on this test. Was the melting done in air or under an inert atmosphere? Is there chemical information of the quenched sample (does the sample oxidize or hydrate)? What does the texture look like?
Now moving on to the experiments at high pressures. In these experiments, the authors claim to observe two plateaus in the laser-power vs. temperature profiles: the first they infer to be when the Re gasket flows due to weakening of the material under high temperatures. A second plateau was observed some 1500-1700 degrees greater than the first plateau. This plateau, they claim, is when the sample melts. But in lines 79-81, they say "no clear change was observed" in the sample chamber or sample morphology. But I thought the premise of this study was to look at the microtexture upon quench?
This takes me to Figure 3 in which back-scattered electron (BSE) images and TEM images are shown. BSE gives info on atomic number Z: typically brighter regions are represented by higher Z material thus giving a sense of composition. As such I'm confused on what I'm looking at in the BSE images. Why the color change in Figure 3i or between the crystals in HT? Shouldn't the color be the same throughout given it is only MgO? Or is there contamination from Ar? Water? Carbon (diamond)? Or something else? In Figure 3a,b,d, we see very different textures in HT and LT for what should be all be below the melting temperature. In Figure 3f,g,i, we see much different texture. Additionally, what is happening in the middle of Figure 3f, the region between "HT" and "LT"? The "chilled margins" make sense since that is likely self-insulation layers from thinning Ar layers, but there isn't such a feature in the other regions? Why not?
Why are the "HT" regions so different in extent? It looks like a factor of 2. Is this reasonable? What are the expected temperature gradients between the "HT" and "LT" regions? How long did the heating experiments last? Did they try to just heat to at high laser power, rather than ramping up to see if the texture was the same with melting, rather than an effect of just grain growth with time?
What do the electron diffraction images show? Is it MgO? Or something else?
Now on to thermal pressure... This is tricky. There have been several studies that have claimed that thermal pressures are negligible when using a soft pressure medium such as Ar (e.g., Fischer et al, 2013;Zerr & Boehler, 1994) and others that suggest that Pth=~0.5 αKΔT (e.g., Goncharov et al., 2010), thus when ΔT is large, Pth may also be large. In any case, when comparing to Zerr & Boehler, who didn't add Pth, the agreement becomes even worse since Zerr & Boehler do not include thermal P.
Almost as an aside, the authors make mention of the controversies in melting temperature in refractory metal by LHDAC experiments. The possibility that it is occurring due to "shear-induced anisotropic plastic flow" is an intriguing idea... But shouldn't this happen along the same P/T path for all samples since it would be dictated by the Re gasket? If this method proves to be reliable (I'm not yet convinced), this may be good to state it in the discussion, but not the abstract as it is off topic. Figure 2b, c, d: How much time was the sample heated for between c and d? The temperature holds steady, but as the gasket got weak, the sample appear to expand. What is causing the browning of the sample? Did the diamonds burn (especially shown in part d)? is rather striking, although the estimates of the melting temperature at the CMB are different by nearly 1000 K (8000 vs 8900 K). Why the discrepancy? Figure S3: There appears to be a slight shift in the MgO peaks between the "HT" and "LT" regions. What do you attribute this to? Are the MgO (and Ar) volumes consistent for this pressure? What are the uncertainties? Why are the "HT" peaks broader? Supplementary Table S2: I'm very puzzled at the relevance on the values given in this table. The average over such a large region (20 um x 20 um) is problematic. I'd prefer to see a compositional map. Is the melt region enhanced in Ar or the other way around? How was Ar quantified? What were the "real" totals before normalization?
Line 11: "mantle-core boundary" should be "core-mantle boundary" to go with standard convention.
Line 175: missing "nm" after "500 to 800" I really want to like this paper, but unfortunately I am not convinced by the data shown.
second abundant mineral in the lower mantle, under high pressure condition by using 8 experimental method. The authors address important issue, but I feel that the current manuscript 9 is too specific to be published in Nature Communications. I can recommend this paper for the 10 publication only if the manuscript is substantially re-written with additional experimental data.

12
We made efforts to expand our temperature and pressure range toward ~6000K and 13 ~50GPa, respectively, which are certainly the limits in the current state of the art of the LHDAC 14 technology combined with CO 2 laser heating, and extensively re-written the manuscript 15 according to the referee's recommendation (see below).   temperatures of MgO by about 1000K at ~30 GPa than those of the earlier study, which is really 31 new and has significant implications for the determination of melting temperatures of other pressure region to ~50 GPa, which further confirm our results in the earlier version of our paper, demonstrating that our experimental results should be used for the melting curve of MgO. It is 35 fortune that our data are quite consistent with the theoretical predictions and dynamic 36 compression data, but we believe carefully determined experimental data are most valuable in 37 scientific research.

107
The authors also offer an explanation as to why the early results reported in Ref. 6 may have 108 underestimated the MgO melting curve, which would be down to the incorrect identification of 109 the onset of melting. The criterion used to locate the melting transition is the plateau in the 110 temperature vs power curve: when the system melts the temperature stops increasing for a while, 111 due to the latent heat of melting. However, the authors observe two plateaus in their data. When 112 they recover the sample after reaching the first plateau, located at the lower temperature, they 113 find that the sample has expanded in the direction perpendicular to the axial compression, We thank the referee, who adequately addresses the important points in our paper.
The authors also point out that there is a whole class of transition metals for which DAC melting slopes, and they suggest that also these experiments may have been affected by wrong

135
We agree that the earlier work by Boehler on the melting temperatures of Fe could have 136 been also underestimated because of a similar reason as discussed here. However, we note the 137 first plateau occurs at temperatures between 3500-4000K, in the pressure range studied here.

138
The melting temperatures of Fe determined by this author is significantly lower than these 139 temperatures, and may not have directly relevance to the present phenomenon of plastic flow of 140 Re gasket. Actually, a recent study using X-ray absorption spectroscopy coupled with DAC 141 proposed lower melting temperatures, which is consistent with those of Boehler (G. Aquilanti et 142 al., Proc. Natl. Acad. Sci. 2015). This is the reason why we limit our discussion on the 143 refractory metals whose melting temperatures higher than ~3000K at the ambient condition.

145
The other comment that I would make is that it is somewhat far fetched to extrapolate the 146 present results, up to 32 GPa, to the core mantle boundary pressure of 135 GPa, and so the 147 authors may want to add some words of caution.

149
In response to the reviewer's comment, we have eventually determined the T m up to ~50

191
The point in this study is that for accurate determination of the melting temperature of

264
In the present study, we used a relatively low accelerating voltage (5 kV) and high beam current 265 (3 nA) to enhance the contribution of crystallographic orientation contrast. We also checked by 266 SEM-EDS mapping analysis that there is neither chemical contamination (such as by Ar or C) 267 nor impurities in the samples, although small Ar inclusions were found to be present within 268 individual crystals and their boundaries by STEM-EDS.

269
To avoid such confusion, we modified the caption of Fig. 3 Figure 3f, the region between "HT" and "LT"? The "chilled >Why are the "HT" regions so different in extent? It looks like a factor of 2. Is this reasonable?
What are the expected temperature gradients between the "HT" and "LT" regions? How long 280 did the heating experiments last? Did they try to just heat to at high laser power, rather than 281 ramping up to see if the texture was the same with melting, rather than an effect of just grain 282 growth with time?

284
According to the temperature distribution measured across the hot spot ( Supplementary Fig.   285 2), the temperature in the LT region is estimated to be ~3000 K when the temperature in the HT 286 region reached at ~5000 K. Since laser-heating at such high temperatures (which is needed for 287 the melting of MgO) may lead to failure of the diamond anvil(s), the heating duration at each 288 laser power had to be short (only for several seconds), then the total heating duration over the 289 first and second plateaus is ~1 min each. However, it is clear from the microtexture that such a 290 short-time heating is adequate to judge melting.

291
In the both samples quenched from the first and second plateaus, the microtexture of 292 MgO is indeed quite different between the HT and LT regions. This is simply due to the large 293 temperature gradient horizontally across the whole sample chamber, since the hot spot is much 294 smaller than that of the sample hole (~ 120 µm), as shown in the 1-D temperature profile 295 ( Supplementary Fig. 2). The LT regions of the both samples (quenched from the first and 296 second plateaus) consist of significantly deformed crystals due to the large shear stress induced 297 by the gasket flow during heating, in which the temperature was, however, not high enough to 298 promote the recrystallization and grain growth. More details are described in line 108-116.

299
On the other hand, the microtexture of the HT regions is quite different between the two 300 samples. The sample quenched from the first plateau shows recrystallization and grain growth 301 features of MgO, while that from the second plateau shows a typical quenching texture from the 302 MgO melt. The latter is characterized by chilled margins and elongated crystals grown inward 303 from the chilled margins ( Fig. 3f and g). Originally, the chilled margins were continuous along 304 the periphery of the lens-shaped melted region, but are now seen only at the lower and upper 305 edges ( Fig. 3f and g). The porous area in the middle of Fig. 3f is also a chilled margin composed 306 of extremely fine grains of MgO, but it has not been well-polished due to drop-out of many 307 grains, since this region was not sufficiently supported by the epoxy resin compared with the from the center of the corresponding bright-field images using a selected-area aperture of a 1.4 >Now on to thermal pressure... This is tricky. There have been several studies that have claimed 318 that thermal pressures are negligible when using a soft pressure medium such as Ar (e.g.,

411
We added it in the sentence (line 9).

414
We removed the word from the sentence (line 10). Line 11: "mantle-core boundary" should be "core-mantle boundary" to go with standard 417 convention.

418
We changed the words in the sentence (line 19).

421
We added it in the sentence (line 214).
now more robust and attractive than previous. However, I'm still not fully convinced by the 9 explanation for lower melting T proposed by Boehler's group. In addition, newly provided 10 T-profile raises one important question, which is not mentioned in the manuscript.

40
>Authors newly provided the T-profile during heating (Fig S2). T value are distributed ± around 41 500 K as usual in the laser-heated DAC experiments. Which points were used for "temperature" 42 in W-T curve (e.g., Fig 2)? This must be described with reasonable explanation, otherwise the 43 temperature has systematic uncertainty of ± around 500 K.

45
In response to the reviewer's comment, we have added a statement to explain how we 46 determined the temperatures in the revised text (Line 216-219). The temperature at each laser power was determined by reading the peak value of the lateral temperature distribution profile, 48 as shown in the revised figure (Suppl. Fig. 2). The temperature difference from the neighboring 49 points is as small as ~100 K, which is substantially smaller than the averaged measurement 50 error of ~200 K. publication.

Response to Referee #3
microscopic image showed that the melting pool is around 40 µm in diameter (Fig. 3). On the 18 other hand, the temperature is about -500 K / 10 µm lower than peak temperature according to

19
the T-distribution provided in the revised manuscript ( Supplementary Fig. 2). At melting 20 temperature, liquid and solid must be coexisting each other, and thus the liquid-solid surface 21 should be at melting temperature. However, the liquid-solid surface is ~20 µm away from the 22 center, where the "melting temperature" was obtained. If I assume T-gradient of -500 K / 10 µm 23 and the diameter of melt pocket = 40 µm, actual melting temperature of MgO can be ~1000 K 24 lower than their result at ~30 GPa.

25
> I believe that the paper is more robust if authors demonstrate the comparison between

26
T-distribution and microscopic image of the recovered sample, which is recently demonstrated 27 by Japanese group using same apparatus (Fig. 4 of Ozawa et al., 2016 EPSL).

29
We checked the radial temperature distribution profiles collected during heating at the second 30 plateau and found that it became flat, while the profiles obtained at lower temperatures showed we don't think our melting temperatures were overestimated due to the lateral temperature 35 gradient. The flattening in the profile is most likely due to a rapid increase in heat transfer 36 caused by the convection of the MgO melt. We added the explanation in the revised manuscript 37 (line 86-88 and 100-104).