Publications

Why Water Doesn’t Behave Like a “Normal” Liquid

Share this post:

Water is essential to life as we know it. Because of this, it has been extensively studied, yet it remains incompletely understood. For example, water does not always behave in accordance with theories of thermodynamics, but deviates from them under particular conditions. Upon cooling, the density of a “normal” liquid increases monotonically and, eventually, the liquid freezes in a denser crystal that sinks. This is not the case for water: upon cooling, the density of liquid water increases but, at about 4 degrees Celsius, the density starts to decrease and the liquid freezes in a crystal that floats instead of sinking. As a result, the surface of water freezes during winter, while the depths maintain a comfortable constant temperature of approximately 4 degrees Celsius that allows life to advance. If water were a “normal” material, it would freeze from the bottom up, killing most aquatic life. These are just two of the many anomalies of water (scientists have now counted at least 72 of them), and scientists are working hard to understand their underlying sources.

Although there still is no full consensus, it is hypothesized that the source of water’s anomalous behavior occurs at very low temperatures (around 90 Kelvin) and high pressures (around 1800 atm, corresponding to 1,800 times the ambient pressure). At this unfamiliar thermodynamic point (known as the liquid-liquid critical point, LLCP), according to the hypothesis, water coexists in two local, different liquid forms: a low-density liquid (LDL) and a high-density liquid (HDL). LDL and HDL differ not just in term of density (LDL being similar in density to ice), but also in terms of order at the molecular level and underlying hydrogen bond network. In HDL each water molecule is surrounded by neighboring molecules displaced in space in a very disordered manner, while in LDL each molecule is surrounded by neighboring molecules arranged in a more ordered manner, creating an underlying network of hydrogen bonds arranged mostly in hexagonal rings, like in ice.

These two forms (or states) exist also at ambient conditions, but they are continuously mixing, transforming into each other, with a strong predominance of HDL because thermal energy causes water molecules to diffuse very fast, breaking local order. Upon cooling and approaching the LLCP, however, the amount of LDL increases and, upon reaching the LLCP, the two liquids should unmix in a 1:1 ratio.

Unfortunately, technical shortcomings cause experiments to fail in investigating liquid water at conditions close to the LLCP. Furthermore, numerical simulations are very hard in such unfamiliar territories because water dynamics become extremely slow and, therefore, incredibly long simulations are necessary. Therefore, the existence of the LLCP in real water (so far, I and colleagues have proven the existence of the LLCP from numerical simulations in a “crude” molecular model of water: J. C. Palmer, F. Martelli, Y. Liu, R. Car, A. Z. Panagiotopoulos, P. Debenedetti, Metastable liquid-liquid transition in a molecular model of water, Nature 510, 385–388, 2014) and the molecular origin of water anomalies remain under debate.

An important concept in the LLCP hypothesis is the Widom line (named for the scientist who located it for the first time), which leads to the LLCP from more familiar territories of higher temperatures and lower pressures and roughly outlines when a liquid behaves more like HDL and when it behaves more like LDL. Along this line, at variance with normal liquids, water thermodynamic response functions show maxima that become more and more pronounced upon approaching the LLCP.

In a recent article, “Unravelling the contribution of local structures to the anomalies of water: The synergistic action of several factors”, I shed light on the microscopic origin of water anomalies by performing extensive numerical simulations on a realistic model of water. I focused on the dynamics of the network of hydrogen bonds that connect water molecules, a new perspective that yielded new information about the heart of the anomalies of water. I found that the maxima in the thermodynamic response functions in correspondence with the Widom line are caused by a “critical” concentration of LDL that causes fragmentation of the large HDL cluster. At this critical concentration, the hydrogen bond network of LDL environments is characterized by an equal number of pentagonal and hexagonal rings, which promotes frustration between crystallizing and fluidizing tendencies.

Hydrogen bond network in water

Two-dimensional slices of the network connecting HDL environments (red) and LDL environments (blue, open circles) at 400 bar. (a) At temperature above the Widom line the sample is dominated by one large HDL cluster whose network spans the simulation box and accommodates a few LDL environments. (b) Upon cooling the temperature, in correspondence of the Widom line, it is possible to observe that the dominating HDL network undergoes fragmentation caused by the increased number of LDL environments. (c) Upon further decreasing the temperature the two networks interpenetrate.

This work shows that the anomalies of water are caused by several factors that act synergistically at the microscopic level. The findings are relevant not just to water, but also to other materials with anomalous behavior, some of which are of incredible interest for technological purposes, such as silicon, silica, carbon, germanium, and phosphorous.

Unravelling the contribution of local structures to the anomalies of water: The synergistic action of several factors” is an invited article for The Journal of Chemical Physics and appears in the Special Topic on “Chemical Physics of Supercooled Water”.

Research Staff Member, IBM Research

More Publications stories

Making and Imaging Cyclocarbon

Scientists have stabilized and imaged a ring of 18 carbon atoms for the first time.

Continue reading

Pushing the boundaries of convex optimization

Convex optimization problems, which involve the minimization of a convex function over a convex set, can be approximated in theory to any fixed precision in polynomial time. However, practical algorithms are known only for special cases. An important question is whether it is possible to develop algorithms for a broader subset of convex optimization problems that are efficient in both theory and practice.

Continue reading

Making Neural Networks Robust with New Perspectives

IBM researchers have partnered with scientists from MIT, Northeastern University, Boston University and University of Minnesota to publish two papers on novel attacks and defenses for graph neural networks and on a new robust training algorithm called hierarchical random switching at IJCAI 2019.

Continue reading