The Rose Model of Water: Linking Theory and Simulation Authors: Peter Ogrin, Tomaz Urbic Published: 2026 ABSTRACT Water plays a fundamental role in countless natural and technological systems, where its unique properties are connected with those of the surrounding environment. The water’s anomalous behaviors arise from the directional nature of hydrogen bonding between molecules. To understand these anomalies, numerous molecular models have been developed, ranging from detailed atomistic descriptions to coarse-grained, conceptually simple representations. Among the latter, the two-dimensional Rose model offers a minimal yet physically meaningful framework that reproduces key thermodynamic and structural anomalies of real water while remaining analytically tractable. In this work, we present a comprehensive review and comparison of results obtained for the Rose water model using Monte Carlo and molecular dynamics simulations, thermodynamic perturbation theory, integral equation theory (both orientation-averaged and orientation-dependent), and an analytical model. The study encompasses the thermodynamic and structural properties of pure Rose water and of systems containing nonpolar solutes. Moreover, the anomalous regions and phase behavior of the model are systematically explored. The combined results demonstrate that the Rose model successfully captures the essential physics of water’s anomalies within a simple and computationally efficient framework, providing a valuable bridge between theory and simulation. 1. Introduction Water is likely the most extensively researched substance and has been the subject of investigation for centuries, as its fundamental role in both biological and non-biological systems has long been recognized. Water’s special importance arises from the combination of its abundance and a variety of anomalous properties [1], which strongly influence nearly every system in which it is present. The anomalous behavior of water is evident in its thermodynamic properties (such as the density maximum at 4 °C, the negative thermal expansion coefficient, and the unusually large heat capacity of the liquid), dynamic properties (such as the non-monotonic temperature dependence of the diffusion coefficient [2,3,4,5,6,7]), and structural properties (including the remarkably rich phase diagram with 20 known crystalline ice forms [8,9,10], and regions of unusual local structural order [11]). Although water exhibits a wide range of anomalous properties, their underlying origin is generally attributed to the ability of water molecules to form hydrogen bonds. The locally tetrahedral arrangement of water molecules is believed to be closely connected with many of these anomalies [12,13]. Evidence for this link is provided by the fact that other liquids with tetrahedral local structures also display similar anomalous behavior [14,15]. Examples include molten GeO2 [16], liquid BeF2 [17,18], liquid silicon [19], and liquid silica [20,21]. This indicates that the local structural ordering of water molecules plays a crucial role in nearly all of its (anomalous) properties. The properties of water confined at the nanoscale deviate from those of bulk water. For example, in monolayer confinement, the dielectric constant of water decreases [22], the phase behavior differs considerably from that of bulk water [23,24,25,26,27,28,29], the freezing point is altered [30,31], and diffusion is enhanced [32]; while confined water retains anomalous thermodynamic behavior comparable to that of the bulk, these anomalous regions shift to lower temperatures, and its structural and dynamical features resemble those of bulk water at higher temperatures [33]. Moreover, ab initio studies suggest that water monolayers can adopt structural motifs belonging to either a square or honeycomb hydrogen-bonded network [34], and a liquid–liquid phase transition has also been predicted at low temperatures [35]. In short, confinement modifies the properties of water monolayers quantitatively, for example, by shifting phase transitions and anomalous regions, but the qualitative behavior remains consistent with bulk water, as it still forms the same hydrogen-bonding interactions that make water such a unique liquid. The two-dimensional model presented in this work could, in principle, also be employed to represent monolayers. Nevertheless, the particular model used here has not been designed or validated for the description of water monolayers. A crucial distinction is that monolayers are subject to nanoscale confinement, which limits molecular motion and reduces available degrees of freedom. In contrast, a two-dimensional water model is not inherently confined within 2D space and thus cannot be directly interpreted as representing a water monolayer. Rather, such models are generally regarded as simplified 2D analogues of water that reproduce certain essential properties of real, three-dimensional liquid water. Many different water models have been developed in attempts to capture the unique behavior of water in various systems. The Bernal–Fowler model [36] was the first realistic water model, using point charges and a repulsion term. As computational sciences developed, similar models began to be used in simulations [37]. The models from the TIP [38,39,40,41,42,43] and SPC [44,45] families were developed some time ago for use in simulations of water and aqueous systems and are still widely used today. Most of these models are empirical potentials parametrized to replicate the properties of experimental water. These models consist of a Lennard–Jones potential and differently placed point charges; the interactions of these point charges are also responsible for hydrogen bonding. More recently, more complex and accurate models have been developed, enabling accurate calculation of a wide range of properties and effects that were neglected in classical empirical models. Such advanced models often rely on ab initio quantum mechanical calculations for high accuracy and machine learning methods to speed up the calculations [46,47,48,49,50,51]. The crucial properties of water can also be captured with much simpler models that lack atomistic detail. The main strength of simple models lies in their ability to investigate the fundamental physical origins and mechanisms underlying diverse phenomena. By stripping away secondary effects and interdependencies, such models allow the relationships between the studied properties to emerge more transparently. Their simplicity also makes them valuable tools for developing and testing new computational methods of varying complexity, from advanced machine learning algorithms to statistical mechanical theories at different levels of sophistication. In addition, when the model is formulated in two dimensions, the resulting structures are easier to visualize, enabling clearer identification of structural changes and effects. Well-known examples of such two-dimensional models include the Mercedes–Benz (MB) model [52,53] and the Rose water model [54]. These two models are closely related, as the Rose model was introduced as a computationally more efficient mimic of the MB model. Another representative of this class is the coarse-grained mW water model [55], which, like the MB model, neglects long-range electrostatic interactions and instead incorporates hydrogen bonding by explicitly favoring tetrahedral molecular arrangements. More recently, another coarse-grained model has been proposed, also based on explicit tetrahedral hydrogen bonding [56]. In this article, we review the research conducted using the Rose water model. The Rose water model is a simple water-like model that describes water molecules as two-dimensional Lennard–Jones disks. The disks have an explicit hydrogen bonding potential added, enabling the formation of hydrogen bonds between particles. The hydrogen-bonding potential has an orientation-dependent term that uses the Rose function, from which the model gets its name. Due to the formulation of the model, each interacting molecule has an independent contribution to the hydrogen bond energy. Consequently, hydrogen half-bonds can be formed, which are not physically correct. Two different parametrizations of the Rose model have been most commonly used. One parametrization, called the MB parametrization, is such that the properties of the model are more similar to the MB model, while the other parametrization, called the real parametrization, more closely resembles real experimental water. The Rose model qualitatively predicts the anomalous properties of pure water and the anomalous solvation thermodynamics of nonpolar solutes [57]. Wertheim’s thermodynamic perturbation theory (TPT) and integral equation theory (IET) were applied to the Rose water model. The IET successfully predicted pair correlation functions between Rose particles, while thermodynamic properties calculated with IET and TPT were in semi-quantitative agreement with those obtained from Monte Carlo (MC) simulations [57]. To improve IET results, an orientation-dependent IET was developed for the Rose water model [58]. Incorporating orientational dependence in IET improves agreement between theoretical and simulation results and enables the calculation of orientation-dependent properties such as spatial distribution functions. The orientation-dependent IET was also extended to include nonpolar solutes in water [59]. Properties of systems with nonpolar solutes of different sizes solvated in Rose particles were calculated by the theory and compared to simulations; the effect of nonpolar solutes on the local structure of the solution was also investigated [59]. The system of nonpolar solutes in water modeled with the Rose water model was also studied using TPT, focusing on thermodynamic properties [60]. Using MC simulations, the properties of the Rose model were investigated over a wide range of conditions, and regions of anomalous properties were identified. Based on the size of these regions, the anomalous properties were ordered into a hierarchy [61]. The phase behavior of the Rose water model was also investigated. Using MC simulations and TPT, the liquid part of the phase diagram was determined [62]. In search of an automatic method for easy determination of the phase diagram from MD simulation data, the entire phase diagram of the Rose model was also determined. First, the phase diagram was determined based on data obtained by nested sampling (NS) and molecular dynamics [63,64]. Subsequently, an unsupervised machine learning approach was developed to determine the phase diagram of the Rose model in a more automated way [64]. The behavior of the Rose water model confined in random porous media was investigated using associative replica Ornstein–Zernike theory [65], and the results were compared to MC simulations. The effects of static and alternating electric fields on the properties of the Rose water model were also studied using MD simulations [66,67]. When the model was exposed to an alternating electric field, an interesting resonance effect was observed. To describe this effect, a model connecting hydrogen bonding and resonance in an electric field was developed [67]. Conceptually based on the Rose model and MB model, an analytical model of water was developed. This model enables the calculation of thermodynamic, dynamic, and structural properties of water [68]. The model successfully predicts all the properties obtained from simulations of the Rose model, but with only a fraction of the computational time. In this article, we review the calculations and results for the Rose water model, focusing on the real parametrization of the model. The same properties of the Rose model calculated using different methods are collected and compared to highlight the advantages and disadvantages of each method. This paper is organized as follows. In the next section, the Rose model is described (Section 2). In Section 3, the different methods used to calculate the properties of the Rose model are outlined. The results are presented and discussed in Section 4 and summarized in Section 5. 2. The Model The Rose water model is a simple two-dimensional water model. Its name originates from the use of Rose functions in the hydrogen bonding potential [54]. In this model, each water molecule is represented as a two-dimensional Lennard–Jones (LJ) disk augmented with a hydrogen bonding potential that enables directional bonding:(1)U(X→i,X→j)rose=ULJ(rij)+UHB(X→i,X→j), where rij is the distance between the centers of molecules i and j, and X→i and X→j are vectors describing the positions and orientations of molecules i and j. The LJ interaction has its standard form:(2)ULJ(rij)=4ϵLJσLJrij12−σLJrij6. Each molecule contributes independently to the hydrogen bonding potential. Specifically, the contribution from molecule i is independent of molecule j, and vice versa:(3)UHB(X→i,X→j)=UHB(r→ij)+UHB(r→ji).Because each molecule contributes independently to the hydrogen-bonding potential, the model formally allows the formation of half hydrogen bonds, where one molecule is oriented favorably for bonding while the partner is not. These configurations are unphysical in real water, where hydrogen bonds are inherently cooperative. However, their energetic contribution is limited because the total hydrogen-bond energy is halved compared to a fully formed bond, and their statistical weight is small at typical simulation temperatures. As a result, half bonds slightly broaden the angular and distance distributions of hydrogen-bonded pairs but do not significantly alter the overall structural or thermodynamic behavior of the liquid. This simplification is retained to maintain computational efficiency and analytical tractability of the model. The hydrogen bonding contribution from a single molecule is the product of a distance-dependent term s(rij) and an orientation-dependent term U(θij):(4)UHB(r→ij)=ϵHB2s(rij)U(θij), where r→ij is the vector pointing from molecule i to j in the body frame of molecule i, rij=|r→ij|, and θij is the angle of this vector in the same frame. The factor 1/2 normalizes the potential so that the maximum contribution from both molecules is 2. The orientational term uses a three-petal Rose function:(5)U(θij)=a2sin2(3θij)+a1sin(3θij), where a1 and a2 control the angular shape of the potential. In Cartesian coordinates of the molecule j relative to i, this reads:(6)UHB(r→ij)=ϵHB2s(rij)a2(3xij2yij−yij3)2rij6+a13xij2yij−yij3rij3, with xij and yij the Cartesian components of r→ij in the body frame of molecule i. The distance-dependent term is modeled using a symmetric double-sided cubic switching function:s(rij)=0,rij