The Water That Refuses to Freeze: How Scientists Decode a Hidden Molecular World

Beneath every frozen protein, polymer, and cell lies a stubborn minority of water molecules that simply refuse to crystallize. Scientists call it uncrystallized water, or UCW: the fraction of water in hydrated materials that stays amorphous and mobile even well below the nominal freezing point. Instead of joining the rigid lattice of bulk ice, these molecules keep moving, trapped in nanoscale pockets, clinging to charged functional groups, or squeezed into the hydration shells that surround proteins and polymer chains. Their mobility sits in a strange middle ground, slower than bulk liquid water but vastly faster than a glassy solid, and that in-between character is precisely what makes UCW so consequential and so difficult to study.

A new review published in Frontiers in Analytical Science by Kitsakorn Locharoenrat of King Mongkut’s Institute of Technology Ladkrabang in Bangkok argues that the field’s central challenge is not a lack of data but a lack of coordination. Because UCW is inherently heterogeneous, spanning environments from tightly bound interfacial sites to mesoscopic confinement, no single instrument can fully characterize it. Molecular motions associated with UCW stretch across an enormous range of timescales, from hydrogen-bond fluctuations on the picosecond scale to glass-like structural relaxation lasting seconds. The review’s core message is that techniques like dielectric spectroscopy, nuclear magnetic resonance, differential scanning calorimetry, quasielastic neutron scattering, terahertz and infrared spectroscopy, and molecular dynamics simulations are complementary rather than redundant, and that the field will only mature when studies combine them on the same samples under strictly controlled conditions.

The terminology itself is a source of confusion that the review untangles carefully. Hydration water refers to the interfacial layer surrounding a solute or matrix. Bound water denotes molecules engaged in specific interactions such as hydrogen bonding or ionic contacts, which restrict their mobility. Non-freezable water describes the fraction that remains amorphous under a given experimental cooling protocol, often attributed to strongly interacting interfacial molecules. Crucially, the classification is method-dependent: faster cooling rates suppress nucleation and crystal growth, inflating the apparent non-freezing fraction, while slower cooling or extended isothermal holding allows more complete crystallization and shrinks it. This makes direct comparisons between differential scanning calorimetry, which uses controlled ramp rates, and dielectric or NMR methods, which may involve long equilibration at a fixed temperature, surprisingly treacherous.

Dielectric spectroscopy offers one of the widest windows onto these dynamics. By measuring the complex permittivity across frequencies from roughly 10^-2 to 10^9 Hz, corresponding to characteristic times from about 100 seconds down to 10^-11 seconds, DS tracks the collective reorientation of molecular dipoles. Bulk water shows a relatively narrow, Debye-like relaxation in the microwave range, but UCW typically produces a broader, non-Debye loss peak at lower frequencies, signaling slower and more heterogeneous motion. Crystalline ice, by contrast, largely suppresses the liquid-like dipolar response. A key diagnostic of mobile water is therefore a broad relaxation that persists below the freezing point, shifted one to several orders of magnitude toward slower timescales than bulk water at the same temperature. In partially crystallized polymer-water mixtures such as poly(vinylpyrrolidone), broadband spectra resolve into contributions from ice, UCW, and the polymer matrix, with the UCW relaxation falling neatly between the other two, evidence of intermediate mobility shaped by water content and thermal history.

Nuclear magnetic resonance brings molecule-specific chemical and isotopic sensitivity to the problem. Relaxation times T1 and T2, pulsed-field-gradient diffusion measurements, and chemical-shift analysis each report distinct aspects of water mobility from microseconds to seconds, and from atomic coordination shells up to micrometer-scale diffusion. In UCW, slower motion shortens T2, alters T1 dispersion, and reduces apparent self-diffusion, while crystalline ice yields signals so broad they often vanish from liquid-state experiments entirely. Emerging zero- to ultralow-field NMR techniques promise to extend these measurements to soft materials where conventional high-field methods struggle. The review highlights a recent combined NMR-DSC study of gelatin methacrylate hydrogels, in which T1 and T2 relaxometry linked bound and non-freezing water to microviscosity and diffusion, illustrating how dynamic and thermodynamic measurements together sharpen the identification of UCW in complex biomaterials.

Differential scanning calorimetry plays a different role: it does not measure molecular motion at all. Instead, DSC records heat flow during controlled heating and cooling, and comparing the melting enthalpy of a hydrated sample to that of bulk ice yields the fractions of freezable and non-freezable water. Glass transitions, cold crystallization, and depressed melting temperatures add further fingerprints. But the measured non-freezable fraction is not a fixed material constant; it varies with scan rate, equilibration time, sample mass, pan sealing, and the reference enthalpy used for confined ice. In sucrose-water systems, for example, slower cooling shifts crystallization to higher temperatures and lowers the apparent non-freezable fraction. DSC thus defines the thermodynamic boundary conditions that any mechanistic interpretation from other techniques must respect.

At the other end of the timescale spectrum sits quasielastic neutron scattering, which measures the dynamic structure factor by analyzing the quasielastic broadening of scattered neutrons. Because incoherent scattering from hydrogen dominates, QENS emphasizes water dynamics directly, probing timescales of roughly 10^-12 to 10^-9 seconds and length scales from angstroms to nanometers. Bulk water produces a broad quasielastic signal; ice is essentially elastic; and UCW falls in between, with narrower linewidths, a larger elastic fraction, and reduced jump-diffusion and rotational rates. Selective deuteration makes the technique even more powerful: in phospholipid membranes, QENS has resolved distinct populations of free, loosely bound, and tightly bound hydration water, with the loosely bound fraction moving roughly an order of magnitude more slowly than free interlamellar water, and the most strongly bound molecules moving on timescales comparable to the lipid headgroups themselves. The method’s limitations are practical, including limited neutron facility access, large sample requirements, and strong dependence on the chosen scattering model.

Terahertz and infrared spectroscopies add a structural dimension by interrogating the hydrogen-bond network directly. THz time-domain spectroscopy captures low-frequency collective modes in the roughly 0.1 to 3 THz range, corresponding to sub-picosecond dynamics, while mid-IR spectroscopy tracks the O-H stretching band between 3000 and 3700 cm^-1 and the bending mode near 1640 cm^-1. UCW shows broadened and shifted THz absorption and persistent liquid-like spectral disorder below freezing, reflecting a hydrogen-bond network that remains irregular rather than adopting the ordered structure of ice. Recent work on proteins has shown that tightly and loosely bound hydration water can change in opposite ways during denaturation, a reminder that vibrational measures of boundness and calorimetric definitions of non-freezable water are related but not equivalent.

Molecular dynamics simulations provide the atomistic bridge that ties these observables together. The autocorrelation of the total dipole moment predicts the dielectric response; orientational and exchange correlation functions connect to NMR; the intermediate scattering function links to QENS; and dipole and polarizability correlations generate IR and THz spectra. Simulations can disambiguate otherwise ambiguous experiments, since a broad dielectric loss peak might reflect either a distribution of static environments or rapid exchange between a few distinct states, and a short NMR T2 might indicate truly immobilized water or mere exchange broadening. The challenges remain substantial, however: common fixed-charge force fields often underperform on dielectric relaxation and interfacial ordering, most models neglect nuclear quantum effects, and ab initio MD is restricted to small systems and short timescales. Machine-learning potentials are extending simulations to larger and more realistic systems, but their reliability depends on training data spanning bulk, interfacial, supercooled, and partially crystallized states.

The review’s case studies show what integration can achieve. Combined solid-state NMR, dielectric spectroscopy, and DSC on mesoporous silica revealed that the low-temperature liquid fraction cannot be understood without accounting for water redistribution during drying and rehydration; later work adding field-cycling NMR and MD showed that crystallization selectively removes bulk-like water from pore centers, leaving a liquid enriched in interface-associated molecules with a broader distribution of correlation times. In amino-acid-functionalized silica pores, surface chemistry slowed water reorientation by up to two orders of magnitude, with activation energies rising from about 0.3 eV in the fully liquid regime to roughly 1.0 eV after partial crystallization. These findings overturn the simplistic picture of a generic bound-water layer: the liquid that survives freezing is selectively stabilized by the surface, and its dynamics reflect confinement, crystallization history, and surface chemistry together. The review also cautions that abrupt signal changes across the narrow window between plus one and minus one degree Celsius usually reflect ice-water redistribution rather than the emergence of a new liquid phase. Looking ahead, the author argues for coordinated studies reporting water content, cooling protocols, hold times, and temperature calibration, and for computational spectroscopy, increasingly accelerated by machine learning, to be integrated with experiments from the outset. The payoff extends far beyond the laboratory: a quantitative, multi-technique picture of uncrystallized water could transform cryopreservation, pharmaceutical freezing, antifouling materials, and biomaterial design, wherever living and synthetic matter meets ice.

Subject of Research: Complementary measurement techniques for characterizing the molecular dynamics of uncrystallized water in hydrated materials

Article Title: Complementary techniques for investigating the molecular dynamics of uncrystallized water

Article References: Locharoenrat, K. (2026). Complementary techniques for investigating the molecular dynamics of uncrystallized water. Frontiers in Analytical Science, 6, Article 1941270. https://doi.org/10.3389/frans.2026.1941270

Image Credits: AI Generated

DOI: 10.3389/frans.2026.1941270

Keywords: uncrystallized water, dielectric spectroscopy, nuclear magnetic resonance, differential scanning calorimetry, quasielastic neutron scattering, terahertz spectroscopy, infrared spectroscopy, molecular dynamics simulations, hydration water, cryopreservation, hydrogen-bond dynamics, nanoscale confinement