REPLYING TO W. Zhou et al. Nature 528, 10.1038/nature16145 (2015)

In the accompanying Comment1, Zhou et al. showed that a NaCl solution captured between graphene sheets leads to the formation of few-layer crystals of NaCl with similar structure and lattice constant as for the ‘square ice’ we described2. They suggest that our samples were accidentally contaminated with NaCl or another salt and that the oxygen K-edge in our electron energy-loss spectra (EELS) originates from oxide contaminants on graphene.

We emphasize that at no point were our samples in proximity to NaCl or other salts. All our spectra were obtained in diffraction mode with an effective diameter of 100 nm, not high-resolution imaging mode in which individual crystals may be selected, to decrease the electron dose and allow longer acquisition times. In our EELS analysis, we focused on the energy window in which the oxygen peak was expected; the full energy spectrum comparing regions with and without ice crystals was not acquired in all cases. Unfortunately, this means that we cannot retrospectively prove the absence of NaCl. Nevertheless, following our new simulations of transmission electron microscope (TEM) images of NaCl, the difference in contrast between sodium and chlorine should be visible under our imaging conditions in the case of a mono- or trilayer crystal with a half unit cell. We do not find corresponding differences in contrast in any of our experimental images.

We agree with Zhou et al.1 that our oxygen K spectrum in figure 1b in ref. 2 probably has a contribution from silicon oxide, but we believe this contribution is small. There is disagreement in the literature regarding the peak shape and exact position of the oxygen K-edge for ice. In our paper2, we compared the experimental oxygen K-edge (figure 1b, main oxygen K peak at 540 eV) with a simulated spectrum3 for which the main peak is shifted by approximately 6 eV compared to our experiment. However, other calculations4,5 of oxygen K spectra for hexagonal and cubic ice give the oxygen K peak at 540 eV, in agreement with the spectrum in figure 1b in ref. 2. In addition, in our unprocessed oxygen K spectrum, a pre-edge shoulder is seen that is very similar to those in refs 4 and 5. Unfortunately, these weak features are not visible in figure 1b in ref. 2, owing to smoothing of the raw spectrum. Only EELS in high-resolution imaging mode selecting individual crystals (or scanning TEM-EELS) could unambiguously distinguish such features.

In view of the above, further experiments are needed to rule out the contamination hypothesis.