New Raman measurements for H₂O ice VII in the range of 300 cm⁻¹ to 4000 cm⁻¹ at pressures up to 120 GPa Free

The hydrogen-bond behavior in the structure of water ice has been of major interest of the multiple scientific fields in the past decades. Especially, the possibility to form symmetrical bonding under high pressure has attracted many experimental and theoretical investigations.^1,2 Theory indicates that the energy barrier between double-well energy potential for the proton motion in the oxygen-oxygen (O–O) line will be depressed with increasing pressure and eventually the two energy minima may converge to a single minimum shape, and the proton will occupy the central position of the O–O distance, molecular solid ice will become ionic materials (phase X) at sufficiently high pressure.^3,4 X-ray diffraction experiments demonstrated that the H₂O ice structure retains body-centered close-packed oxygen sublattice from above 2.1 GPa up to over a megabar (100 GPa).^5–7 Spectroscopic experiments including infrared, Raman, Brillouin scattering, and other optical methods also have been used extensively for studying this material,^8–20 and revealed a wealth of information for the possible transitions, some of which were explained as the clue that the hydrogen-bond has symmetrized around 60–80 GPa.^{9,12–14,17,18,21–23} However, the evidence is not direct because no position information for hydrogen nuclei was determined. Therefore, at present there is no definitive evidence to support this transition. Furthermore, numerous modern quantum-mechanical model calculations have proposed the possibility that intermediate phases exist prior to the VII to X transition.^4,24 Meanwhile, there are numerous reported experimental anomalies, both in x-ray diffraction and spectroscopic experiments, indicate a number of transitions likely happened at moderate pressure ranges.^{7,10,19,25–27} Most recently, new neutron diffraction observations reveal structure anomalies starting from 13 GPa that cannot be explained by conventional structure network above 26 GPa.²⁸ The structure evolution of H₂O ice under pressure remains in mystery because of the complication of hydrogen-bonding variation.

In principle, Raman spectroscopy is a sensitive probe for phase transitions, especially in transformations involving the change of the nature of chemical bonds. Since the Raman intensities for H₂O ices are extremely weak; it was very difficult to measure these signals under high pressures, in particular, the O–H stretching modes merge with the diamond second-order Raman above ∼23 GPa (see Figure 1(a)). This problem has prevented the observation of further pressure change of these modes in all previously reported Raman measurements. However, this information is critical for tracking the change of the hydrogen bonding behavior under pressures. The predicted symmetrization transition is related to hydrogen movements,^4,29 a displacive-type transition feature, i.e., soft mode, should be observed in the hydrogen vibrations. A monotonically decreasing stretching mode with cascading interactions (Fermi resonance) to other vibrations has been proposed, and the possibility of a hydrogen-bond symmetrization transition at ∼60 GPa has been evaluated in the previous experiments.^{12,13,17,18,21–23} Fermi resonance effect³⁰ seems to successfully explain the speculative soft mode behavior during the process of symmetric hydrogen-bond transitions; however, because of the faint peak shape of the stretching mode in infrared spectra, and the lack of measurable Raman signals, the real feature and entire trace of the development of the soft stretching mode under pressure have never been observed directly.

Here we report a new Raman measurement, which overcomes the problem associated with the overlap of the diamond-sample signals by using both moissanite anvil cell (DAC) and diamond anvil cell (MAC) (see Figures 1(a)-1(d)) in multiple measurements. The full continuous spectral range of H₂O ice VII from 300 cm⁻¹ to 4000 cm⁻¹, including the O–H stretching modes, bending modes, as well as many lattice modes, and rich combinations, was obtained up to 120 GPa for the first time. A wealth of vibrational information for this important hydrogen-bonded compound over a wide pressure range will greatly contribute to revealing the dynamics and kinetics of all possible transitions predicted by various experiments and theoretical computations.

Both diamond anvil cell (DAC) and moissanite anvil cell (MAC)³¹ were used. Double-distilled deionized water together with a few tiny ruby chips strategically located at different spots were loaded to the gasket hole. The ruby chips were used as the pressure sensor using ruby fluorescence pressure scale.³² Three laser excitations of 457, 532 (∼100 mW), and 633 (∼30 mW) nm were used for measurements in order to unambiguously distinguish any new signals. A confocal optical system with a tiny spatial filter designed to provide a sampling area of only ∼5 μm at the sample and a highly focused laser beam was used at the backscattering geometry. The exposure time is 10 s repeated for 5 times.

Even using very low-fluorescence anvils fabricated from high-purity synthetic diamond, the Raman signals of H₂O ice are still extremely weak at high pressure. To obtain high quality spectra, it is necessary to carefully subtract any background noise. In principle, an ideal background reference would involve all the measured optical information but without that from the sample. This would require the equality of the excitation conditions when the laser beam is incident upon the sample and the reference areas, respectively. The inner edge of the normally used metallic gasket, if chosen for recording the reference background, would introduce errors as it is not transparent. In this situation, if only one anvil is excited by the incident laser and re-excited by the strong retroflection of laser light, which is quite different from the situation of laser passing through the transparent sample, an incorrect background will result. In this study, a specially designed transparent rim-liner (see inset of Fig. 1(d)), which is optically signal-free at the used laser excitations, was introduced to the inside of the gasket hole with the water sample located at the center of this liner. This transparent liner area served as an in situ background reference at each pressure under the exact same laser focus, power, and exposure time as for the sample. This is essential for reducing the error in the background subtraction in order to reveal the weak signals. It has been suggested that second-order Raman peaks collected from different portions of the diamond anvils may not be the same as they are subjected to different stresses; this could result in imprecise subtraction, and some residual portion of the diamond Raman peaks may be assigned to the H₂O sample. We do not think this is happening because change in pressure should lead to systematic enhancement of their frequencies and intensities. We have been able to trace the peaks in question over a wide range of pressures without observing this effect.

Diamond and moissanite have different Raman spectra; therefore, sample-anvil signal overlapping regions will appear at different spectral range in DAC or MAC, as shown in Figures 1(a) and 1(b), respectively. The use of two different anvils offers a convenient way for checking the background removal procedure by comparing the subtracted overlapping section with the corresponding non-overlapping section obtained from the other type anvils. A typical result on the comparison of the background removed section is shown in Figures 1(c) and 1(d). A total of 15 runs were conducted with 5 runs in MAC and 10 runs in DAC. The consistency between background removed and non-overlapping spectra increases the confidence and possibility for extending pressure range over the limit that moissanite anvils can reach.

Hydrogen disordering in molecular ice VII prohibited precise peak assignment. However, the close similarity of the observed Raman peaks between disordered ice VII and its ordered analogue ice VIII has led to the general agreement that ice VII is the disordered form of ice VIII; this also has been supported by the neutron diffraction experiment.³³ Therefore, we may employ the same assignment as that for ice VIII in the discussion of ice VII as was done in previous reports.^12,15 For convenience, peak assignments in this study are compared with previous work in Table I. It should be noted that in the present study, we revealed peaks that have not been observed previously and cannot be assigned to its analogue in ice VIII.

Figure 2 shows the full range spectra at selected pressures. Measurements with the moissanite anvils have been conducted up to 60 GPa. We first provide a broad overview of the changes in spectral features. In general, the O–H stretching modes in the high frequency region are the strongest features at lower pressure range. The intensities of these modes are found to increase first with pressure and then decrease anomalously upon further compression. Above 44 GPa, the intensity of the strongest peak is reduced to about ∼1/50 from the maximum value at ∼10 GPa. Two of the three stretching modes can be followed up to the highest pressure in this study except for the ν₃Eg (p2) mode that merged with ν₁A₁g (p3) at 44 GPa (see Figs. 3 and 4). The low-frequency lattice modes are the weakest at all pressures. No discernible features were observed if the intensities are plotted in the same scale as that for the high-frequency modes. However, distinctive features are observed with a 50× enlargement scale as shown in the left panel of Figure 2.

Figure 3 shows details on how the stretching modes change with pressure. Peaks p1, p2, and p3 are the analogues of ν₁B₁g, ν₃Eg, and ν₁A₁g in ice VIII, respectively. The in-phase symmetric stretching mode ν₁A₁g (p3) has the strongest intensity and increases with pressure below 10 GPa but decreases thereafter. On the other hand, the line width of this peak decreases with pressure first and then increases with pressure reaching a minimum at 13 GPa. This observed trend is the same as previously reported.³⁴ The most interesting finding is that the frequency “softens” almost linearly with pressure up to 27 GPa and then sharply increases with pressure. The frequency “hardening” continues to the highest pressure studied here. The anti-symmetric stretching mode ν₃Eg (p2) is a weak but sharp peak. The frequency decreases with pressure almost at the same rate as that of ν₁A₁g below 15 GPa. The downward trend becomes slower and continues to 27 GPa. Above that pressure, the frequency-pressure slope becomes slower and finally merged into the ν₁A₁g vibration around 44 GPa. Since this is a weak peak, the intensity is buried under the broadened neighbor peaks above ∼25 GPa and peak fitting was necessary in order to distinguish it above this pressure. The out-of-phase stretching mode ν₁B₁g is relatively broad. From the beginning, its frequency decreases with pressure at a slower rate than that of other two. A distinct change in the even flatter slope was found at ∼15 GPa. Similar to ν₁A₁g but at a different pressure, the pressure-frequency slope of ν₁B₁g also changes sharply into positive at 44 GPa and persists to the highest pressure of this study.

Peak p4 is a relatively weak peak in the lower-frequency region next to p3. The frequency-pressure shift of this peak is very similar to p3 and became “invisible” between ∼20 and ∼35 GPa because it merges with the broad neighboring peaks. It becomes clearly visible again above 35 GPa up to the highest pressure studied here. Peak p5 did not exist at low pressures; it appears as a sharp peak above 44 GPa. Its intensity grows significantly above 60 GPa, and becomes the dominant feature at and above 90 GPa. As shown in Figure 4 (left panel), peaks p6 and p7 are weak peaks at low pressures. Their intensities are enhanced noticeably above 15 GPa, and become quite “visible” above this pressure (also see Figure 2). These two peaks have not been reported in earlier low-pressure studies for both ice VII and VIII, since they were masked by the diamond signals at lower pressure. The frequency of peak p6 increases slowly and steadily with pressure. In comparison, the frequency shift of peak p7 with pressure behaves very differently. The frequency-pressure slope of peak p7 is steep and positive at the low pressure range, but the slope changes sharply from positive to negative at ∼15 GPa and then it changes back to positive above ∼44 GPa. This peak cannot be seen above 90 GPa. No mode assignments were made for peaks p4–p7 although Raman peaks of similar frequencies (except for p5) do exist in VIII under ambient pressure³⁵ (Table I).

The bending vibration modes p8 and p9 are relatively weak at low pressure (Figure 4). They have been assigned as ν₂B₁g and ν₂A₁g vibrations in ice VIII projected to ambient pressure.³⁵ The frequency and intensity of p8 show no significant change with pressure below 27 GPa, but the intensity jumps up at 27 GPa. Peak p9 is weak at low pressures (see 4th panel of Fig. 5). Its frequency increases with pressure and then merges into peak p10 at ∼24 GPa, as a result the combined intensity becomes stronger at this pressure. The merged peak continuously shifts closer to p8 and both gain intensity at ∼27 GPa. Above 27 GPa, p9 becomes a strong sharp peak at ∼38 GPa (Figure 4). This phenomenon has also been reported for ice VIII at 20 K.¹² This peak becomes broader upon further compression and becomes very weak above 90 GPa. Peak p8 is weak below 90 GPa but becomes strong and broad above 90 GPa. Peak p10 is always weak and gradually separates from p9 with increasing pressure. From the behavior of these two peaks, it is likely both are involved in a resonance phenomenon. These three peaks can be distinguished in the Raman spectrum up to the highest pressure of this study.

Peak p11 does not exist at low pressure. It appears starting from ∼13 GPa (see 4th panel in Figure 5). This peak has not been reported previously because its frequency is close to the Raman band of diamond. Incidentally, its sudden appearance is an indication of a phase transition as has been reported in the x-ray diffraction measurements.²⁵ Below 44 GPa, peak p12 is one of the major features in the low-frequency region (Figure 5), its frequency is close to the rotational vibration ν_RxyEg of ice VIII at ambient pressure,³⁵ and shows almost no change with pressure. This mode was also not reported in previous ice VII studies. Peak p13’s frequency decreases with pressure very slightly below 15 GPa, it merges and crosses over p14 at 15 GPa, then retains a flat slope to the highest pressure of this study. Peak p14 was assigned to the rotational vibration ν_RZB₂g of ice VIII at ambient pressure.⁹ The frequency increases monotonically with pressure and merges into p12 at 60 GPa. Peak p15 is a very weak mode and could be the analogue of ν_RxyEg of ice VIII.³⁵ Its frequency increases with pressure steeply below ∼15 GPa and then decreases slightly with pressure and merges into peak p16 at ∼44 GPa. Peak p16 is one of the strongest of the low-frequency modes, its frequency increases monotonically with pressure up to ∼60 GPa, then the slope flattens for the higher pressure range. Peak p17 has the very similar frequency-pressure behavior as that of p16; however, its intensity is much weaker especially at low pressures.

Except for peaks p5 and p11, which appear above the transitions at 44 and 13 GPa, respectively, mentioned above, three new peaks p18, p19, and p20 start to appear above the transition at 90 GPa. All these three peaks are very weak. The frequency of p20 is very close to that of T_2g mode reported earlier¹² (Fig. 6).

Figure 6 summarizes the pressure dependence of all the vibrational modes obtained in this study. All together, there are 20 “visible” modes designated p1, p2, …, p20 as described above. Additional “invisible” modes, appearing as weak shoulder or curve features, can be resolved by profile fitting and are indicated by thick gray lines.

There have been extensive experimental and theoretical investigations on the ice VII structure and its evolution under high pressure. X-ray and neutron diffraction have confirmed that ice VII is composed of two interpenetrating but not connected tetrahedral hydrogen-bonded networks with a body-centered-cubic oxygen sub-structure. The hydrogen positions are disordered in ice VII and the phase is paraelectric. In comparison, the orientation of the molecules in ice VIII is long-range ordered with the dipole moments on the two sublattices pointing in the opposite directions, i.e., anti-ferroelectric. The structure is formed from ice VII by a small tetragonal distortion of cubic unit cell in the c axis. Theory predicts that the hydrogen-bonds in these lattices will become symmetric under compression,^1,2 i.e., the hydrogen atom must be displaced to the center point between two neighboring oxygens. This phase X is cubic and has cuprite structure (Cu₂O structure). Factor group analysis and symmetry coordinate inspection show phase X has only one Raman active mode T_2g.⁹ Theories have also predicted this new phase should exist from 35 to 80 GPa.² Phase diagram study, however, found that ice VIII is stable in the pressure-temperature domain below 62 GPa.¹⁴ Therefore phase X, if it exists, is expected to be stable above this pressure. Subsequently, IR and Raman studies claimed 60–70 GPa to be the transition pressure for phase X of water ices.^{12,13,17,18,22,23}

One of the most controversial problems in high pressure ice is the hydrogen-bond symmetrization transformation in ice VII. A basic assumption of this transition is that the covalent O–H donor bond length (d_OH) of the nearest O–H⋯O connection will elongate when the d_OO distance decreases under compression. Eventually, the hydrogen atom will be situated at the middle of the O–O contact and water ice will lose the molecular crystal character.^2,36,37 This assumption was supported by the observation on the decrease of stretching frequency with pressure in previous infrared and Raman studies on ice VII and VIII,^{9,12,15,18,23} as this feature is related to a decrease in the force constant of O–H vibration and should be accompanied by bond lengthening. However, this is not direct evidence indicating the hydrogen atom position towards the centro-symmetric transition. It should be mentioned that even this indirect evidence derived from the observation of softening in Raman and IR stretching modes can only be measured within a limited pressure range due to experimental difficulties described above. Previous reported hydrogen-bond symmetrizations were based on the analysis of spectroscopic resonance phenomena in which the stretch mode was assumed to be continuously softened and coupled strongly with extrapolated lower-frequency modes.^12,17,18,22 The speculative transition pressures for the hydrogen-symmetrized phase (ice-X) around 60–70 GPa were derived. One significant finding in this study is that the frequency of stretching mode does not decrease monotonically with pressure but shows a sudden upturn at ∼27 GPa, far away from any proposed ice X transition pressures. This result explicitly demonstrates that some aspects of the assumption for a monotonic hydrogen movement before the transition to phase X are questionable.

On the other hand, theoretical calculations predicted that only one active Raman mode T₂g between 800 and 1000 cm⁻¹ should be present in ice-X with the cuprite structure.^9,12 Coincidentally, the predicted vibration is very close to the p20 mode observed in this study. However, this new peak, confirmed using different laser energies, only appears above 90 GPa and also co-exists with other vibration modes. The new evidence indicates that the predicted phase X does not appear at least up to 120 GPa. However, the high quality spectra obtained here show unambiguously that there are significant spectral changes in both the high and low frequency regions around 90 GPa.

In principle, the direct confirmation for the position of hydrogen atom can only be achieved by neutron diffraction. Unfortunately, so far no strong evidence from this technique to support the proposed transformation even in experiments up to 52 GPa has been reported.²⁸ Instead, older results show almost no change in the d_OH distance up to 10 GPa,^38,39 and the multisite disorder structure is valid up to 20 GPa.⁴⁰ More recently, the analysis of limited Bragg diffraction peaks in a neutron diffraction experiment at 52 GPa²⁸ revealed a quite different bonding picture for D₂O ice VII in which a new structure with interstitial hydrogen in the lattice of ice VII was proposed. In addition, two structural transitions at 13 and 26 GPa were found. These findings are consistent with the appearance of the new Raman peak (see p11 at Fig. 5) at 13 GPa and the change in the pressure slope at ∼27 GPa in this study. The interstitial hydrogen structure model challenges the conventional concept in which the hydrogen atoms should be situated along the closest O–H–O contacts during compression. Although the crystal symmetry was not changed, the occurrence of interstitial hydrogen atoms may alter the local structure and may provide hints to explain the following observations.

There have been several experiments that reported anomalies in water ice occurred at several pressures. From powder X-ray diffraction, Somayazulu et al.²⁵ reported the splitting of the cubic 110 Bragg reflection around 14 GPa. Polian and Grimsditch¹⁰ reported a discontinuity in the pressure dependence of backscattering Brillouin signal at 44 GPa and suggested either the longitudinal sound velocity or refractive index changed anomalously at this pressure. Later, independent experiments on refractive index and sound velocity were reported by Zha et al.¹⁹ and Ahart et al.⁴¹ It was demonstrated that the refractive index is the property that shows discontinuous changes around 40, 60, and 90 GPa. No measurable sound velocity (elasticity) anomaly was found at 44 GPa. Loubeyre et al.⁷ reported the single crystal X-ray diffraction intensity ratio between the 111 and 222 diffraction lines (111/222) for ice VII increases continuously with pressure but reaches a plateau around 30 GPa. This has been attributed to the signature of hydrogen atoms moving to the center position along the oxygen-oxygen directions. Two studies on the equation of state reported by Wolanin et al.⁶ and Sugimura et al.,²⁶ by powder x-ray diffraction method, found two possible phase transformations, respectively, around 12 and 66 GPa, or 40 and 60 GPa.

The present measurements of the full range Raman spectra between 300 and 4000 cm⁻¹ to much higher pressure, except for the pressure dependences of stretching mode contradicted from previous expectation, the experimental results show rich transformations in ice VII including frequency shift rates, splitting, disappearing and appearing of peaks at 13–15, 27, 44, 60, and 90 GPa. These observations are consistent with previous reports and confirm there are indeed several successive phase transitions before the predicted hydrogen symmetrization transition.^4,24,37,42 This study shows the Raman spectra of ice VII under pressure are complex, it reflects the structure evolution is different from the prediction of the conventional model. To some extent, it echoes the new observation from the recent neutron diffraction results.²⁸ As well known, sorting out the detailed structure only from Raman spectra is not practical, so we are not able to give structure explanations for our findings. We hope the work will encourage further study in order to understand the mechanisms responsible for these findings.

See supplementary material for the example of Raman peak verification at different pressures and spectral regions by using different laser excitations.

This work was supported by the National Science Foundation (Grant No. DMR-1106132) and the Department of Energy (DOE)/National Nuclear Security Administration (Grant No. DE-NA-0002006; Carnegie/Department of Energy Alliance Center, CDAC).