Palmer et al . reply
Jeremy C. Palmer,Fausto Martelli,Yang Liu,Roberto Car,Athanassios Z. Panagiotopoulos,Pablo G. Debenedetti
DOI: https://doi.org/10.1038/nature16540
IF: 64.8
2016-01-01
Nature
Abstract:replying to D. Chandler Nature 531, http://dx.doi.org/10.1038/nature16539 (2016) We reported1 for the ST2 model of water advanced free energy calculations using six sampling techniques, all of which show the existence of a low-density liquid (LDL) as well as a high-density liquid (HDL) and a liquid–liquid phase transition (LLPT) between them. In the accompanying Comment2, Chandler contends that fundamental arguments3 preclude an LLPT in water and reiterates his claim4 that the LDL phase is an artefact associated with poor equilibration. We point out that although the fundamental argument3 concerns the question of whether critical fluctuations can be detected in metastable systems despite nucleation of the stable phase, it was explicitly stated3 that it has firm implications only for the detection of a critical point, but does not preclude liquid–liquid phase separation. When applying this argument, Chandler2 concludes that critical fluctuations larger than 2–3 nm cannot be equilibrated in deeply supercooled water. Following the same analysis but using different values for relevant timescales (τR ≈ 10−10 s from experimentally derived correlations5 and τMS ≈ 10−3 s from experiment6, as defined in ref. 2), we estimate that critical fluctuations at 229 K can reach ~100 nm—potentially large enough to characterize experimentally. Regarding putative artefacts arising due to poor equilibration, the LDL persisted in our simulations1, 7 after relaxing all accessible fluctuations by sampling reversibly between the liquid and crystal regions, and two liquid basins were obtained independently of the sampling method and duration. The LDL basin did not disappear over time when sampling to and from the crystal, as had been predicted when incorrectly4 assuming our calculations to be poorly equilibrated. The free energy exhibits scaling consistent with an LLPT over the range of system sizes that can be explored computationally1, 7. Each of these facts is inconsistent with poor equilibration. Moreover, the salient features of our free energy calculations have been reproduced by others8, 9, and our code has been publicly available since 2014 (http://pablonet.princeton.edu/pgd/html/links.html). The recent demonstration9 that adjustment of a single model parameter in ST2 (the hydrogen-bond angular flexibility) makes the LLPT thermodynamically stable with respect to ice Ih/Ic disproves the claim4 that crystallization was mistaken for an LLPT. The main issue is the irreconcilable difference between seemingly identical free energy calculations for the same water model: these identified either two liquids and a crystal as we reported1, or only one liquid and one crystal4. Chandler argues2 that LLPT-like artefacts arise from limiting relaxation of fluctuations, but this was only observed when transforming simulation data using a theory whose key assumption is that density fluctuations in HDL decay much faster than bond-orientational fluctuations4. In contrast, molecular dynamics simulations show that density is the slowly relaxing variable in the HDL region10. Chandler’s explanation is therefore contradicted by the reversible phase behaviour1, 8, 9 and equilibrium dynamics1, 10 of ST2. Ultimately, we are confident that continued scrutiny of codes and methods used in the free energy calculations will reveal the cause of the different behaviours predicted for ST2. The question of which one occurs in real water must await an answer by experiment, not by theory or simulation. Download references