Fact-checked by Grok 2 weeks ago

Non-random two-liquid model

The non-random two-liquid (NRTL) model is a thermodynamic activity coefficient model designed to represent the non-ideal behavior of multicomponent liquid mixtures, particularly for calculating phase equilibria such as vapor-liquid (VLE) and liquid-liquid (LLE) equilibria.^[1] It assumes that the local composition around a central molecule in the liquid phase differs from the bulk composition due to differences in intermolecular interactions, leading to non-random molecular arrangements.^[2] Developed by Henri Renon and John M. Prausnitz in 1968, the NRTL model builds on the two-liquid theory proposed by R.L. Scott and the local composition concept introduced by Grant M. Wilson, extending these ideas to derive an expression for the excess Gibbs free energy of mixing. The model's core formulation expresses the molar excess Gibbs energy (G^E) as \frac{G^E}{RT} = \sum_i x_i \frac{\sum_j x_j G_{ji} \tau_{ji}}{\sum_k x_k G_{ki}}, where x_i is the mole fraction of component i, G_{ji} = \exp(-\alpha_{ji} \tau_{ji}), and \tau_{ji} = \frac{g_{ji} - g_{ii}}{RT}, with g_{ji} representing the energetic interaction between molecules j and i, R the gas constant, and T the temperature.^[1] Activity coefficients (\gamma_i) are then obtained by differentiating this expression, enabling predictions of phase behavior from binary interaction parameters.^[2] The NRTL model requires three adjustable parameters per binary pair: two binary interaction parameters (\tau_{ij} and \tau_{ji}, related to energy differences) and a non-randomness factor (\alpha_{ij}, typically fixed between 0.2 and 0.47 to reflect the degree of local order, though it can be adjustable).^[1] These parameters are regressed from experimental data, such as VLE or LLE measurements, making the model empirical yet versatile for systems exhibiting partial miscibility or azeotropes.^[2] Widely applied in chemical engineering for process simulation and design— including distillation, extraction, and reactive systems—the NRTL model excels in handling both polar and non-polar mixtures but can face challenges with highly nonlinear behavior, parameter inconsistency across temperature ranges, and convergence issues in multicomponent calculations.^[1] Extensions like the electrolyte NRTL (eNRTL) have adapted it for ionic solutions, enhancing its utility in areas such as CO₂ capture^[3] and battery electrolytes.^[4]

Introduction

Overview

The non-random two-liquid (NRTL) model is an activity coefficient model designed to correlate mole fraction-based activity coefficients (\gamma_i) in multicomponent nonelectrolyte liquid mixtures, enabling accurate representation of non-ideal solution behavior.^[5] Developed by Renon and Prausnitz in 1968, it builds on local composition concepts to address limitations in prior models that assumed random molecular mixing, such as the Margules and van Laar equations.^[5]^[6] At its core, the NRTL model accounts for non-random local compositions around a central molecule, which arise due to differing interaction energies between like pairs (U_{ii}) and unlike pairs (U_{ij}) in the mixture.^[5] This approach recognizes that molecules tend to associate preferentially with similar species in non-ideal systems, leading to deviations from bulk composition.^[5] The model incorporates two binary energy parameters \tau_{ij} and \tau_{ji} (reflecting interaction energy differences) and a non-randomness parameter \alpha_{ij} (typically between 0.2 and 0.47, controlling the degree of local order), where the term G_{ij} = \exp(-\alpha_{ij} \tau_{ij}) accounts for non-randomness in local composition.^[5] In chemical engineering, the NRTL model is widely applied to predict phase equilibria, including vapor-liquid equilibria for distillation processes, liquid-liquid equilibria for extraction operations, and solid-liquid equilibria for crystallization and solubility calculations.^[5]^[7] These applications rely on the model's ability to fit experimental data with a minimal number of adjustable parameters, making it suitable for multicomponent systems without requiring additional terms beyond binary interactions.^[5]

Historical Development

The non-random two-liquid (NRTL) model was introduced in 1968 by Henri Renon and J. M. Prausnitz in their seminal paper published in the AIChE Journal. This work derived a new expression for the excess Gibbs energy of liquid mixtures based on local composition concepts, extending prior thermodynamic frameworks to better handle nonideal behaviors. The primary motivation for developing the NRTL model stemmed from the limitations of the Wilson equation, introduced in 1964, which excelled at correlating vapor-liquid equilibria in fully miscible systems but failed to predict phase splitting or partial miscibility in strongly nonideal liquids. Renon and Prausnitz addressed this by incorporating a nonrandomness factor into a two-liquid approximation, enabling the model to represent liquid-liquid equilibria and ternary systems using only binary parameters, thus providing greater versatility for complex mixtures. By the 1980s, the NRTL model had gained widespread adoption in chemical process simulation, with integration into commercial software such as Aspen Plus, facilitating its use in industrial design and optimization of separation processes. This early incorporation into simulation tools accelerated its application in engineering practice, where it remains a standard for activity coefficient calculations in nonideal systems.^[8] In recent years, advancements have focused on enhancing parameter estimation efficiency, exemplified by machine learning approaches that predict NRTL interaction parameters from molecular representations like SMILES strings. A notable 2023 study introduced the SPT-NRTL model, a physics-guided neural network that ensures thermodynamic consistency while enabling rapid predictions for diverse mixtures, building on the original framework for modern computational workflows.^[9]

Theoretical Basis

Local Composition Concept

The local composition concept in the non-random two-liquid (NRTL) model describes how the composition of a liquid mixture varies at the molecular level, specifically around a central molecule of one component. The local mole fraction of component j surrounding a central molecule of component i, denoted x_{ji}, deviates from the overall bulk mole fraction x_j because molecules tend to associate preferentially with neighbors that minimize the system's energy, favoring unlike interactions over random distribution. This idea accounts for the non-ideal behavior in mixtures where intermolecular forces lead to clustering or segregation at short ranges.^[10] In contrast to random mixing assumptions, where local and bulk compositions are identical—as in ideal solutions with no energetic biases—the NRTL model incorporates non-randomness driven by differences in pairwise interaction energies. Under random mixing, molecules distribute uniformly without preference, resulting in zero excess Gibbs energy beyond entropy of mixing. The NRTL approach, however, posits that stronger attractive forces between unlike pairs (or weaker repulsive forces) cause deviations, making the local environment richer or poorer in certain components compared to the bulk. This non-random arrangement is assumed to be proportional to the relative energy differences, providing a more realistic depiction of real liquid mixtures exhibiting azeotropes or phase splitting. The NRTL model builds upon the local composition framework first proposed by Wilson in 1964, which introduced the idea that activity coefficients could be derived from local rather than bulk compositions to better capture non-idealities. While Wilson's model assumes a regular, random local structure adjusted only by interaction energies, NRTL extends this by adding a non-randomness parameter \alpha_{ij} (typically between 0.2 and 0.47) to explicitly measure the extent of deviation from randomness, allowing it to handle systems with significant local order, such as those forming two liquid phases. This enhancement makes NRTL particularly suitable for partially miscible systems.^[10] Intuitively, the local composition in NRTL is captured by the relation x_{ji} = \frac{x_j \exp(-\alpha_{ji} \tau_{ji})}{\sum_k x_k \exp(-\alpha_{ki} \tau_{ki})}, where \tau_{ji} = \frac{g_{ji} - g_{ii}}{RT} quantifies the energetic disparity between j-i and i-i interactions relative to thermal energy RT, with R the gas constant and T the temperature. The parameter \tau_{ji}, tied to binary interaction energies, is explored further in the model parameters section. This formulation ensures that local compositions sum to unity and reflect the model's two-liquid hypothesis of microphase separation.^[10]

Key Assumptions

The Non-random two-liquid (NRTL) model is grounded in the two-liquid hypothesis, which conceptualizes a liquid mixture as a superposition of two immiscible "pure" liquid phases, each characterized by local enrichment in one component, thereby accounting for compositional heterogeneities at the molecular level. This approach, originally developed by Scott and adapted by Renon and Prausnitz, enables the model to describe both miscible and partially immiscible systems by treating the overall mixture as a blend of these hypothetical liquids.^[10] A core assumption is that local compositions are derived from bulk compositions using a quasi-lattice fluid approximation that posits molecules occupy lattice sites with surrounding coordination shells. This simplification allows for the definition of local mole fractions to capture non-ideal behavior without requiring detailed knowledge of the global arrangement, facilitating practical computations.^[10] The model further assumes that molecular interactions occur exclusively in pairwise fashion, focusing on binary energy parameters between unlike species and disregarding higher-order multiplet effects to maintain tractability. Temperature and pressure effects are incorporated primarily via the model's adjustable parameters, such as the dimensionless energy difference \tau_{ij}, which exhibits temperature dependence through forms like \tau_{ij} = a_{ij} + \frac{b_{ij}}{T}, where a_{ij} and b_{ij} are fitted constants, R is the gas constant, and T is temperature.^[10] The NRTL framework is designed for systems with moderate to strong deviations from ideality, such as those prone to phase splitting, but presumes no dominant associative interactions (e.g., strong hydrogen bonding) that would require additional modifications for accurate representation. The degree of non-randomness in these local environments is quantified by the parameter \alpha_{ij}, typically ranging from 0.2 (for hydrocarbons with polar liquids) to 0.47 (for self-associating systems).^[10]

Derivation of the Model

Excess Gibbs Energy Expression

The molar excess Gibbs free energy in the non-random two-liquid (NRTL) model is derived from principles of local compositions, assuming that the local composition around a central molecule differs from the bulk due to non-random molecular arrangements from differences in intermolecular interactions. This extends earlier two-liquid theories by incorporating a non-randomness factor to account for preferential interactions in the coordination shells around each central molecule.^[5] The full expression for the molar excess Gibbs free energy in a multicomponent mixture is given by

\frac{G^{ex}}{RT} = \sum_i x_i \frac{\sum_j \tau_{ji} G_{ji} x_j}{\sum_k G_{ki} x_k},

where x_i is the global mole fraction of component i, \tau_{ji} = \frac{g_{ji} - g_{ii}}{RT} represents the dimensionless energetic interaction parameter between molecules j and i relative to the pure i component (with g_{ji} denoting the molar interaction energy of j-i pairs, and g_{ii} for i-i pairs), and G_{ji} = \exp(-\alpha_{ji} \tau_{ji}) is the nonrandomness factor incorporating the parameter \alpha_{ji} (typically 0.2–0.47, reflecting the degree of local order). This formulation arises from approximating the residual excess Gibbs energy using local mole fractions, where the first summation term in the equation captures deviations from random mixing due to preferential interactions.^[5] The NRTL expression relates to the classical two-liquid model by treating the mixture as a weighted superposition of local "liquid" environments around each component, with corrections for nonrandomness that adjust the effective compositions beyond simple volume averaging of pure component properties. This approach enables the model to predict liquid-liquid immiscibility and other phase behaviors by allowing G^{ex} to exhibit common tangent constructions. From this excess Gibbs energy, activity coefficients can be obtained via the Gibbs-Duhem relation, though detailed formulations appear elsewhere.

Activity Coefficient Formulation

The activity coefficient for component i in the non-random two-liquid (NRTL) model is obtained by differentiating the molar excess Gibbs energy G^E/RT with respect to the mole number of component i, while holding temperature T, pressure P, and the mole numbers of other components constant: \ln \gamma_i = \left( \frac{\partial (n G^E / RT)}{\partial n_i} \right)_{T,P,n_j}. This partial derivative yields the explicit expression for \ln \gamma_i, which takes a two-term form reflecting the model's local composition and asymmetry corrections. The first term captures the local composition correction around component i:

\frac{\sum_j x_j \tau_{ji} G_{ji}}{\sum_k x_k G_{ki}}

where x_j is the liquid mole fraction of component j, \tau_{ji} = (g_{ji} - g_{ii})/RT with g_{ji} representing the energetic interaction between molecules j and i, and G_{ji} = \exp(-\alpha_{ji} \tau_{ji}) incorporating the non-randomness parameter \alpha_{ji}. This term adjusts the random mixing assumption by accounting for non-ideal local environments. The second term addresses the asymmetry in local compositions for unlike pairs:

\sum_j \frac{x_j G_{ij}}{\sum_k x_k G_{kj}} \left( \tau_{ij} - \frac{\sum_m x_m \tau_{mj} G_{mj}}{\sum_k x_k G_{kj}} \right)

This correction ensures the model can represent systems where the local mole fractions differ between surrounding species, arising from the unsymmetry in interaction energies (\tau_{ij} \neq \tau_{ji}). In limiting cases, the formulation satisfies thermodynamic expectations: as the mole fraction x_i \to 1, \gamma_i \to 1, recovering ideal behavior in pure components; as x_i \to 0, \gamma_i^\infty emerges from the infinite dilution interactions, quantified by the \tau and G terms specific to the surrounding dominant component. The model's structure inherently promotes Gibbs-Duhem consistency at constant temperature and pressure, particularly in ideal or near-ideal limits, by deriving \gamma_i directly from the excess Gibbs energy expression.

Equations for Mixtures

Binary Mixtures

The Non-random two-liquid (NRTL) model simplifies for binary mixtures consisting of components 1 and 2, where the activity coefficients are expressed in terms of mole fractions x_1 and x_2 = 1 - x_1, binary interaction parameters \tau_{ij}, and nonrandomness factors G_{ij}. The natural logarithm of the activity coefficient for component 1 is given by

\ln \gamma_1 = x_2^2 \left[ \frac{\tau_{21} G_{21}}{(x_1 + x_2 G_{21})^2} + \frac{\tau_{12} G_{12}}{(x_2 + x_1 G_{12})^2} \right],

and symmetrically for component 2,

\ln \gamma_2 = x_1^2 \left[ \frac{\tau_{12} G_{12}}{(x_2 + x_1 G_{12})^2} + \frac{\tau_{21} G_{21}}{(x_1 + x_2 G_{21})^2} \right].

Here, G_{ij} = \exp(-\alpha_{ij} \tau_{ij}), with \tau_{ij} = (g_{ij} - g_{jj})/RT representing dimensionless energy differences between interaction parameters g_{ij}, the gas constant R, and temperature T; \alpha_{ij} is the nonrandomness parameter, typically between 0.2 and 0.47 for binary systems. These forms arise from the local composition concept applied to two-liquid theory, enabling representation of nonideal behavior in binary liquid phases.^[11] At infinite dilution, the activity coefficients exhibit limiting behavior that highlights extreme nonidealities. For component 1 as x_1 \to 0,

\ln \gamma_1^\infty = \tau_{21} \exp(\alpha_{21} \tau_{21}) + \tau_{12} \exp(-\alpha_{12} \tau_{12}),

and symmetrically for component 2 as x_2 \to 0,

\ln \gamma_2^\infty = \tau_{12} \exp(\alpha_{12} \tau_{12}) + \tau_{21} \exp(-\alpha_{21} \tau_{21}).

These limits provide key insights into solubility and partitioning, as they quantify the activity coefficient when one component is present in trace amounts.^[11] A representative example is the ethanol(1)-water(2) binary mixture at 25°C, where typical NRTL parameters are \alpha_{12} = 0.3, \tau_{12} = 3.4578, and \tau_{21} = -0.8009. These yield activity coefficient curves showing positive deviations from Raoult's law, with \gamma_1 ranging from approximately 1.5 at equimolar composition to over 4 at low ethanol concentrations, and \gamma_2 peaking near 6 at low water concentrations. The parameters align well with experimental vapor-liquid equilibrium data.^[11] The binary NRTL equations effectively capture both positive and negative deviations from ideality, as well as the formation of azeotropes in systems like ethanol-water, which exhibits a minimum-boiling azeotrope at about 89 mol% ethanol due to the asymmetric activity coefficients promoting phase splitting tendencies in the model framework.^[11]

Multicomponent Systems

The Non-random two-liquid (NRTL) model extends naturally to multicomponent mixtures with n > 2 components through generalized summation forms that account for all pairwise interactions. The activity coefficient for component i in an n-component mixture is given by

\ln \gamma_i = \frac{\sum_j x_j \tau_{ji} G_{ji}}{\sum_k x_k G_{ki}} + \sum_j \frac{x_j G_{ij}}{\sum_k x_k G_{kj}} \left( \tau_{ij} - \frac{\sum_m x_m \tau_{mj} G_{mj}}{\sum_k x_k G_{kj}} \right),

where x_j is the liquid mole fraction of component j, \tau_{ij} is the non-randomness energy parameter between components i and j, and G_{ij} is the local composition Boltzmann factor defined for all pairs as

G_{ij} = \exp(-\alpha_{ij} \tau_{ij}).

Here, \alpha_{ij} represents the non-randomness parameter, typically between 0 and 1, which measures the degree of local order in the mixture. This formulation requires binary interaction parameters \tau_{ij} and \alpha_{ij} (or \tau_{ji}, \alpha_{ji}) for every unique pair of components, resulting in a parameter matrix that scales quadratically with the number of components. In practice, computing activity coefficients for vapor-liquid equilibrium (VLE) in multicomponent systems involves iterative numerical solutions because the vapor mole fractions y_i depend implicitly on the liquid compositions x_i via y_i = x_i \gamma_i P_i^\text{sat} / P, necessitating convergence algorithms like successive substitution or Newton-Raphson methods. A representative example of the model's application to ternary systems is the acetone-chloroform-methanol mixture, where NRTL parameters capture cross-interactions between all pairs, enabling accurate prediction of non-ideal behaviors such as azeotrope formation and deviations from Raoult's law. In this system, the pairwise \tau_{ij} values reflect differing hydrogen-bonding strengths, with chloroform-acetone interactions showing strong negative deviations. The NRTL model's structure supports scalability to complex industrial mixtures with 10 or more components, as implemented in process simulation software for petrochemical and pharmaceutical applications, where it handles extensive parameter matrices efficiently through matrix algebra optimizations.

Model Parameters

Binary Interaction Parameters

In the non-random two-liquid (NRTL) model, the binary interaction parameters \tau_{ij} quantify the energetic differences in molecular interactions within liquid mixtures. These parameters are defined as \tau_{ij} = \frac{g_{ij} - g_{ii}}{RT}, where g_{ij} is the interaction energy between molecules of components i and j, g_{ii} is the interaction energy for pure component i, R is the universal gas constant, and T is the absolute temperature. This formulation arises from the local composition concept, capturing deviations from random mixing due to differing pair-wise energies.^[5] Physically, \tau_{ij} > 0 signifies that unlike-molecule interactions (i-j pairs) are less favorable than like-molecule interactions in the pure i environment (i-i pairs), promoting phase separation and positive deviations from Raoult's law. In contrast, \tau_{ij} < 0 indicates stronger attractions between unlike molecules, leading to negative deviations and enhanced miscibility. These parameters thus provide a measure of the relative stability of molecular contacts in the mixture. The NRTL model incorporates asymmetry, with \tau_{ij} \neq \tau_{ji} in general, allowing it to account for directional effects such as differences in molecular size, shape, or polarity that influence local environments. This non-symmetry is essential for accurately describing systems where the influence of component i on j differs from that of j on i. The parameter \tau_{ij} relates directly to the Boltzmann factor in the model's local composition expression, where the ratio of local mole fractions X_{ij}/X_{jj} \propto \exp(-\tau_{ij}), representing the relative probability of forming i-j contacts versus j-j contacts based on energetic favorability. Typical values of \tau_{ij} for binary pairs, as compiled in databases like the DECHEMA Vapor-Liquid Equilibrium Data Collection, range from approximately 0.5 to 2 for hydrocarbon-alcohol systems, reflecting the moderate positive deviations typical of such non-polar/polar mixtures.^[12] These values are derived from regression against experimental phase equilibrium data and vary by specific components, but they illustrate the model's applicability to systems with limited miscibility.

Non-randomness Parameter

The non-randomness parameter, denoted as \alpha_{ij}, in the non-random two-liquid (NRTL) model quantifies the degree of deviation from random mixing in the local molecular environment of a binary mixture of components i and j. It is a dimensionless constant typically bounded between 0 and 1, where values greater than 0 introduce non-idealities in local compositions by accounting for preferential interactions between unlike molecules.^[5] This parameter plays a crucial role in the Boltzmann factor of the model, expressed as G_{ij} = \exp(-\alpha_{ij} \tau_{ij}), where \tau_{ij} represents the energetic interaction difference normalized by the thermal energy RT. By scaling the exponent, \alpha_{ij} controls the bias in local compositions, allowing the model to capture a spectrum of mixing behaviors from near-random to highly segregated local environments. When \alpha_{ij} = 0, the model reduces to random mixing assumptions akin to the Margules two-parameter equation, as local compositions become identical to bulk compositions. Conversely, at \alpha_{ij} = 1, it approaches the full two-liquid theory, resembling the UNIQUAC model's combinatorial contribution with maximum non-randomness.^[5] Typical values of \alpha_{ij} range from 0.2 to 0.47, depending on the mixture type; for non-associating or nonpolar systems, it is often fixed at 0.2–0.3, while higher values around 0.4–0.47 are used for polar or self-associating systems like alcohols to better represent strong local ordering.^[5] In practice, \alpha_{ij} is usually assumed symmetric, such that \alpha_{ij} = \alpha_{ji}, simplifying parameter estimation without loss of generality for most applications.^[5] Selection of \alpha_{ij} is guided by the polarity and association tendencies of the components, with lower values for apolar mixtures exhibiting weak interactions and higher values for polar systems where molecular orientation leads to greater non-randomness; it is not always regressed from data but often preset based on empirical guidelines to ensure model robustness.^[5]

Temperature Dependence

The temperature dependence of the non-random two-liquid (NRTL) model parameters is essential for extending its applicability to phase equilibrium predictions over wide temperature ranges, typically beyond 50 K, where fixed-temperature parameters fail to capture variations in molecular interactions. The binary interaction parameters, embodied in the dimensionless τ_ij, are derived from the energy difference Δg_ij = (g_ij - g_ii), where g_ij represents the energetic interaction between species i and j. To incorporate temperature effects, Δg_ij is often expressed using forms that reflect the thermodynamic consistency via the Gibbs-Helmholtz relation, ensuring the model's excess Gibbs energy aligns with experimental enthalpies and entropies. Common formulations include an Antoine-like equation, such as Δg_ij / RT = a_ij + b_ij / T + c_ij \ln T + d_ij T, or a polynomial expansion, Δg_ij = a_ij + b_ij / T + c_ij T, where a_ij, b_ij, c_ij, and d_ij are fitted constants with units reflecting energy scales (e.g., cal/mol or J/mol), and T is in Kelvin. These forms allow τ_ij = Δg_ij / (RT) to vary smoothly with temperature, accounting for changes in interaction strengths, such as the weakening of hydrogen bonds or dispersion forces as thermal energy disrupts local compositions.^[13] The non-randomness parameter α_ij, which measures the degree of non-ideality in local compositions, is typically treated as temperature-independent and fixed at values between 0.2 and 0.47 for most systems, based on the assumption of constant structural disorder. However, in cases requiring finer adjustments over extreme temperature spans, a linear dependence is occasionally employed: α_ij = α_ij^0 + α_ij^1 T, though this is rarely used due to added complexity without significant gains in accuracy for standard applications. This physical basis stems from the model's local composition theory, where rising temperature alters the Boltzmann factor exp(-Δg_ij / RT), reducing the preference for unlike-pair formations and thus modulating activity coefficients γ_i. For instance, in the acetone-water system, which exhibits strong hydrogen bonding, the b_ij and c_ij terms in the polynomial form capture the curvature in ln γ_i versus T plots, improving predictions of vapor-liquid equilibria (VLE) deviations at elevated temperatures.^[13]^[14] Databases like the DECHEMA Chemistry Data Series provide pre-regressed, temperature-dependent NRTL parameter sets for numerous binaries, enabling VLE calculations up to 100°C or higher with average deviations below 2% in bubble points. These sets, derived from curated experimental data, prioritize systems with polar interactions and facilitate reliable extrapolations while maintaining thermodynamic consistency. For the acetone-water example, DECHEMA parameters such as a_12 ≈ 3489 cal/mol, b_12 ≈ 3477 cal/mol, and c_12 ≈ -1582 cal/mol·K yield robust fits across 25–100°C, highlighting the model's utility in capturing temperature-induced shifts in miscibility.^[13]

Parameter Estimation

Regression from Experimental Data

The regression of NRTL model parameters from experimental data involves non-linear optimization to fit the binary interaction parameters (τ_{ij}) and, occasionally, the non-randomness parameter (α) to phase equilibrium measurements, ensuring the model accurately represents local composition effects in liquid mixtures. This process typically targets vapor-liquid equilibrium (VLE) data, such as pressure-composition (P-x-y) isotherms or temperature-composition (T-x-y) isobars, or liquid-liquid equilibrium (LLE) data in the form of tie-lines connecting coexisting phase compositions. Experimental datasets are sourced from comprehensive repositories like the Dortmund Data Bank (DDB), which contains over 100,000 VLE and LLE entries for binary and multicomponent systems, or the NIST Thermodynamics Research Center database, providing critically evaluated thermophysical properties for thousands of substances. The core of the regression is an objective function designed to minimize discrepancies between experimental and model-predicted values, promoting quantitative accuracy in activity coefficients (γ_i) or phase compositions. For VLE, common formulations include least-squares minimization of relative deviations in vapor mole fractions (y_i) or pressures (P), such as ∑ [ (y_i^{exp} - y_i^{calc}) / y_i^{exp} ]^2, or deviations in activity coefficients γ_i, ensuring the model satisfies the modified Raoult's law (y_i P = x_i γ_i P_i^{sat}). Alternatively, maximum likelihood estimators account for experimental uncertainties by weighting terms inversely proportional to measurement errors, as implemented in standard thermodynamic fitting routines. For LLE, the objective function enforces the isoactivity condition, minimizing differences in component fugacities across phases (x_i^I γ_i^I = x_i^{II} γ_i^{II}), often via sum-of-squared errors in phase compositions along tie-lines. These functions are solved using gradient-based algorithms like Levenberg-Marquardt or sequential quadratic programming to handle the non-linearity inherent in the NRTL excess Gibbs energy expression.^[15] Parameter estimation is conducted via specialized software that automates data import, model setup, and optimization. Tools such as Aspen Plus employ built-in data regression modules to process imported experimental files in formats like DETHERM from DDB, applying weighted least squares with user-defined weights for different data types (e.g., higher weight on y_i than x_i due to measurement precision). Similarly, MATLAB toolboxes like Optimization Toolbox facilitate custom scripts for NRTL fitting, integrating numerical solvers for objective function evaluation and sensitivity analysis to assess parameter correlations. Recent advances include Python-based algorithms that leverage differential evolution for robust estimation, particularly for biochemical systems, achieving lower errors than traditional methods.^[16] Convergence is achieved iteratively, with typical tolerances on the order of 10^{-6} for residuals, yielding parameter sets that reduce average absolute deviations in VLE predictions to below 2-5% for pressure or y_i in well-behaved systems. Initial parameter guesses are crucial for avoiding local minima in the non-convex optimization landscape and are often obtained from predictive group-contribution methods or analogous literature values. The UNIFAC model, which decomposes molecules into functional groups to estimate activity coefficients, provides starting estimates for τ_{ij} by regressing group-group interactions to similar VLE data, offering a physically motivated initialization that aligns with molecular structure. For instance, literature compilations of NRTL parameters for hydrocarbon-alcohol binaries can serve as proxies, adjusted for temperature via van der Waals energy differences. Once initialized, the regression refines these values while maintaining bounds, such as 0.2 ≤ α ≤ 0.47 for most systems, to ensure model stability.^[16]^[17] A representative example is the fitting of NRTL parameters for the binary ethanol-water VLE system at 1 atm using 8-10 experimental points spanning x from 0.1 to 0.9, sourced from DDB. The optimization minimizes deviations in y_i, yielding τ_{12} ≈ 3.5, τ_{21} ≈ 0.8, and α ≈ 0.3, which captures the azeotropic behavior with an average y_i deviation of 1.2%, demonstrating the model's efficacy for polar mixtures after regression. This approach extends to multicomponent systems by sequentially regressing binaries before combining parameters, though care is taken to prioritize dominant interactions.

Thermodynamic Consistency

The Non-random two-liquid (NRTL) model is inherently designed to satisfy the Gibbs-Duhem equation, which ensures thermodynamic consistency by relating changes in chemical potentials across compositions at constant temperature and pressure. For binary mixtures, this requirement translates to the condition x_1 d \ln \gamma_1 + x_2 d \ln \gamma_2 = 0, where x_i is the mole fraction and \gamma_i is the activity coefficient of component i. In practice, fitted NRTL parameters are validated through direct numerical integration of this equation over the full composition range (from x=0 to x=1), confirming that the integral \int_0^1 x \, d(\ln \gamma_1) + (1-x) \, d(\ln \gamma_2) = 0. Failure to meet this criterion indicates inconsistencies arising from parameter estimation errors.^[18] For vapor-liquid equilibrium (VLE) data, thermodynamic consistency is commonly assessed using the area test, which verifies that positive and negative deviations in the activity coefficients balance across the composition space. This test, originally proposed by Herington and refined in subsequent works, involves plotting \ln(\gamma_1 \gamma_2) versus mole fraction x and ensuring the areas above and below the baseline (where \ln(\gamma_1 \gamma_2) = 0) are equal within a tolerance, typically corresponding to the integrated Gibbs-Duhem form \int_0^1 [\ln(\gamma_1 \gamma_2)] \, dx = 0. When applied to NRTL-fitted parameters, the test rejects datasets if the area imbalance exceeds 10-20% of the total area, highlighting imbalances in positive/negative deviations that violate thermodynamic laws. Complementary point-wise tests, such as those by Fredenslund et al., evaluate consistency at individual data points by comparing predicted pressures with experimental values using NRTL-derived activity coefficients.^[19]^[20] In liquid-liquid equilibrium (LLE) applications, consistency checks for NRTL parameters extend to tie-line closure and binodal curve stability. Tie-line closure requires that the sum of mole fractions in each phase equals unity (\sum x_i^I = 1 and \sum x_i^{II} = 1), with deviations quantified by standard error \sigma(x) = \sqrt{\sum (x_i^{\exp} - x_i^{\cal})^2 / (2n)}, where n is the number of tie-lines; values below 0.01 typically indicate acceptable closure. Binodal stability is ensured by confirming positive definiteness of the second derivative of the excess Gibbs energy, preventing unphysical phase predictions. A modified Gibbs-Duhem test for LLE, as detailed by Reyes-Labarta et al., incorporates topological stability analysis across all compositions, rejecting parameters if they predict incorrect miscibility or fail isoactivity conditions. These methods are applied post-fitting to validate NRTL binary interaction parameters.^[21] Common issues compromising NRTL parameter consistency include sparse experimental data, which can lead to overfitting and violation of Gibbs-Duhem integrals, and inappropriate fixing of the non-randomness parameter \alpha, resulting in unbalanced deviations. Reviews of published LLE correlations reveal a quite high percentage of inconsistent parameters, often due to inadequate initial guesses or flat Gibbs energy surfaces producing false tie-lines, necessitating rigorous post-regression checks before database inclusion. Direct integration methods for Gibbs-Duhem and area tests for VLE, alongside LLE-specific validations like those in Reyes-Labarta et al., serve as standard tools to enforce reliability.^[22]

Applications

Vapor-Liquid Equilibria

The Non-random two-liquid (NRTL) model facilitates vapor-liquid equilibria (VLE) calculations by providing activity coefficients \gamma_i that quantify deviations from ideality in the liquid phase, assuming an ideal vapor phase. The fundamental equilibrium relation, derived from the equality of chemical potentials, simplifies under Raoult's law to y_i P = x_i \gamma_i P_i^s, where y_i is the vapor mole fraction of component i, x_i is the liquid mole fraction, P is the total system pressure, and P_i^s is the pure-component saturation vapor pressure at the equilibrium temperature.^[5] This equation enables the prediction of phase compositions for both binary and multicomponent mixtures. Bubble point calculations, used to determine the temperature or pressure at which the first vapor bubble forms given a liquid composition, involve iterating on temperature T (or pressure) until \sum y_i = 1, with y_i = x_i \gamma_i P_i^s(T) / P. Conversely, dew point calculations solve for the liquid composition given vapor composition, iterating until \sum x_i = 1, where x_i = y_i P / (\gamma_i P_i^s(T)). These iterative procedures, often implemented in numerical solvers, are essential for simulating phase separation processes.^[5] The NRTL model's strength lies in its accuracy for azeotropic systems, where activity coefficients capture strong interactions leading to crossings of equilibrium curves. For the ethanol-water binary, a classic minimum-boiling azeotrope, NRTL correlations fitted to experimental data yield low errors in liquid and vapor mole fractions across a range of pressures.^[23] In industrial contexts, such as petrochemical distillation, NRTL is integrated into simulators like Aspen Plus for multicomponent flash calculations, enabling precise design of separation columns for mixtures like hydrocarbons with polar additives.^[24] The temperature dependence of NRTL's binary interaction parameters, typically modeled as \tau_{ij}(T) = a_{ij} + b_{ij}/T, plays a crucial role in predicting azeotrope shifts with changing conditions, such as how the ethanol-water azeotrope composition varies from approximately 89 mol% ethanol at 101.3 kPa to higher ethanol content under vacuum.^[25] The NRTL model outperforms simpler models like the two-parameter Margules in handling non-ideal mixtures by accounting for local composition effects.

Liquid-Liquid Equilibria

The Non-random two-liquid (NRTL) model is applied to liquid-liquid equilibria (LLE) by identifying coexisting phases that satisfy the condition of equal chemical potentials for each component across the phases, derived from the minimization of the total Gibbs free energy of the system. For two immiscible liquid phases I and II, the isoactivity criterion governs the equilibrium: x_i^{\mathrm{I}} \gamma_i^{\mathrm{I}} = x_i^{\mathrm{II}} \gamma_i^{\mathrm{II}}, where x_i is the mole fraction and \gamma_i is the activity coefficient of component i calculated via the NRTL equation. This condition, combined with overall mass balances, allows computation of binodal curves and tie-lines by solving the nonlinear isoactivity equations, often using numerical optimization to ensure the global minimum in Gibbs energy and thermodynamic stability.^[26] A key strength of the NRTL model in LLE predictions lies in its ability to accurately represent systems exhibiting upper critical solution temperature (UCST) or lower critical solution temperature (LCST) behavior, capturing the temperature-dependent miscibility gaps through adjustable binary interaction parameters. For instance, in the binary water + n-butanol system, which displays UCST behavior, the NRTL model correlates experimental binodal curves across a range of temperatures, providing reliable predictions of phase boundaries. This performance stems from the model's local composition concept, which accounts for non-ideal interactions driving phase splitting in partially miscible systems.^[26] In practical applications, such as solvent extraction processes, the NRTL model facilitates the design of separations involving LLE. However, VLE-derived NRTL parameters often inadequately predict LLE due to differences in interaction sensitivities, necessitating dedicated regression from LLE data to ensure accuracy in immiscibility predictions. Extensions of the NRTL model incorporate ternary interaction parameters to handle complex three-phase LLE regions, improving representations in quaternary systems or those with multiple immiscible layers, such as in enhanced extraction processes, while maintaining consistency with binary sub-systems.

Limitations and Extensions

Known Shortcomings

The standard NRTL model can exhibit thermodynamic inconsistency when binary interaction parameters are regressed from experimental data without enforcing constraints derived from the Gibbs-Duhem equation. Although the model is theoretically consistent, fitting procedures often yield parameters that violate this fundamental relation, leading to unreliable predictions of phase behavior. For instance, analyses of vapor-liquid equilibrium (VLE) datasets have shown that a significant fraction of binary systems may fail consistency tests when using unconstrained NRTL fits.^[19]^[27] A major practical limitation of the NRTL model is parameter proliferation in multicomponent systems. Each binary pair requires three adjustable parameters: two non-dimensional energy parameters (τ_{ij} and τ_{ji}) and the non-randomness factor (α_{ij}). For a system with n components, the number of binaries is n(n-1)/2, resulting in 3n(n-1)/2 parameters overall; for n > 10, this exceeds 150 values, complicating regression, storage, and transferability across datasets. This scaling issue hinders application to complex industrial mixtures, such as those in petrochemical refining or pharmaceutical formulations.^[28] The NRTL model demonstrates poor extrapolation capabilities beyond the temperature range or system types used for parameter fitting. It frequently fails for electrolytes, where long-range electrostatic interactions are not inherently captured, necessitating extensions like the electrolyte NRTL (eNRTL) variant. Similarly, associating systems (e.g., hydrogen-bonding fluids like alcohols or acids) show degraded performance due to the model's assumption of random local compositions without explicit association terms. In such cases, predictions of activity coefficients can deviate by more than 50% from experimental values outside the fitted conditions. Specific examples highlight these extrapolation flaws. For highly immiscible pairs, such as water-alkane systems, the NRTL model often overpredicts infinite-dilution activity coefficients (γ^∞), leading to inaccurate estimates of mutual solubilities. Additionally, unlike the UNIQUAC model, NRTL neglects combinatorial contributions from molecular size and shape differences (volume effects), resulting in biased predictions for mixtures with disparate molecular volumes, such as polymer-solvent systems.^[29] Historically, the original 1968 formulation of the NRTL model underestimated liquid-liquid equilibrium (LLE) phase envelopes for certain hydrocarbon mixtures, such as alkane-alkane or alkane-aromatic pairs, due to insufficient emphasis on entropic non-idealities in local compositions. This led to narrower predicted immiscibility regions compared to experimental observations, prompting subsequent refinements for better LLE representation.^[30]

Modified NRTL Models

To address limitations in parameter correlation and generalizability of the original NRTL model, several modifications have been developed since the early 2000s, focusing on reducing adjustable parameters, incorporating electrolyte effects, and integrating advanced computational methods.^[31] The one-parameter modified NRTL (mNRTL1) model, introduced in 2014, simplifies the original three-parameter framework by fixing the interaction parameter ratio through pure-component properties and generalizing the non-randomness factor, resulting in a single adjustable energy interaction parameter τ. This reduction mitigates strong correlations between binary interaction parameters, enhancing regression convergence and enabling qualitative classification of vapor-liquid equilibrium (VLE) behaviors, such as nearly ideal or highly non-ideal systems, while maintaining comparable accuracy to the original model across 916 VLE and 20 liquid-liquid equilibrium (LLE) datasets from DECHEMA and NIST-TDE sources.^[31] The electrolyte NRTL (eNRTL) model extends the original to handle ionic solutions by combining short-range local composition effects from NRTL with long-range electrostatic interactions via a Pitzer-like Debye-Hückel term, introducing ion-specific binary interaction parameters for salts in aqueous or mixed-solvent systems. Originally formulated in 1982 and refined in subsequent works, eNRTL provides analytical expressions for excess Gibbs energy and activity coefficients, ensuring thermodynamic consistency in dilute to concentrated electrolyte mixtures, and has become a standard for modeling ion transport and phase behavior in industrial processes. Recent machine learning variants, such as the SPT-NRTL model from 2023, predict NRTL parameters directly from molecular descriptors like SMILES strings and quantum chemical properties using physics-guided neural networks, enforcing thermodynamic consistency through constraints on excess Gibbs energy derivatives and outperforming group-contribution methods like UNIFAC in activity coefficient predictions across diverse functional groups. This approach reduces reliance on experimental fitting, enabling rapid screening for untested mixtures.^[32] Other extensions include temperature-dependent formulations for the non-randomness parameter α, often expressed as α_{ij} = c_{ij} + d_{ij}/T to better capture varying local compositions at different temperatures,^[14] and hybrid models combining NRTL with COSMO-RS, where quantum-derived surface charge densities inform τ parameters for improved predictive accuracy in complex organic systems without extensive regression.^[33] More recent developments as of 2025 include a modified NRTL model focused on improving accuracy in vapor-liquid equilibria predictions^[34] and Python-based algorithms for straightforward estimation of NRTL parameters in aqueous binary systems, enhancing accessibility and thermodynamic consistency in regression tasks.^[35] These modified NRTL models find adoption in advanced simulations, such as eNRTL for CO2 capture in amine-based absorbents and ML-enhanced variants for biofuel mixture design, where they facilitate scalable predictions in sustainable process engineering.^[32]

References

[1]
What does the NRTL model look like? Determination of boundaries for different fluid phase equilibrium regions
### Summary of NRTL Model from https://aiche.onlinelibrary.wiley.com/doi/full/10.1002/aic.17805
[2]
Parametrization of the NRTL Model with a Multiobjective Approach
Non-Random Two Liquids (NRTL) Molecular Model. Renon and Prausnitz [5] developed the NRTL model based on Scott's theory of two liquids [17] and the concept ...
[3]
Open access, thermodynamically consistent, electrolyte NRTL ...
Dec 18, 2024 · This study developed a comprehensive, open-access, thermodynamically consistent, validated vapour-liquid equilibrium model for CO 2 absorption into aqueous ...
[4]
Local compositions in thermodynamic excess functions for liquid ...
A critical discussion is given of the use of local compositions for representation of excess Gibbs energies of liquid mixtures.
[5]
[PDF] Renon H & Prausnitz J M. Local composition in thermodynamic ...
Jan 29, 1979 · Local composition in thermodynamic excess functions for liquid mixtures AICHE Journal 14:135-44, 1968. A new equation for the excess Gibbs.Missing: Solutions | Show results with:Solutions
[6]
Modeling of solid-liquid equilibrium using a modified group ...
Jul 15, 2019 · The non-randomness two-liquid model (NRTL) has been used successfully in chemical engineering to calculate phase equilibria. Different forms of ...
[7]
[PDF] Polymer Characterization and Physical Properties
This document is intended as a guide to using AspenTech's software. This documentation contains AspenTech proprietary and confidential information and may ...
[8]
https://esupport.aspentech.com/FileAttachment?id=a0g0B00000GfNN8
[9]
[PDF] VAPOR-LIQUID EQUILIBRIUM DATA COLLECTION ... - Dechema
Nov 2, 2017 · This collection contains tables and diagrams of vapor-liquid equilibrium data for binary and multicomponent mixtures, including correlation ...
[10]
https://aiche.onlinelibrary.wiley.com/doi/abs/10.1002/aic.1460140517
[11]
NRTL Gibbs Excess Model (thermo.nrtl) - Read the Docs
Class for representing an a liquid with excess gibbs energy represented by the NRTL equation. This model is capable of representing VL and LL behavior.
[12]
[PDF] Quality Assessment Algorithm for Vapor-Liquid Equilibrium Data
Jul 6, 2010 · three-parameter NRTL model. The vapor pressures and three. NRTL parameters (five parameters in total) are fitted to experimental VLE values ...<|control11|><|separator|>
[13]
Python-Based Algorithm for Estimating NRTL Model Parameters ...
Jan 13, 2025 · ... UNIFAC group contribution model and determines the quantity of each functional group. Based on the fragmentation information, the type and ...
[14]
Development of the NRTL functional activity coefficient (NRTL-FAC ...
The NRTL-FAC model can accurately predict the VLE of binary systems, with results in good agreement with experimental data at different temperature and pressure ...
[15]
Gibbs–Duhem-informed neural networks for binary activity ...
Oct 4, 2023 · As the NRTL model fulfills the Gibbs–Duhem equation, the hybrid SPT-NRTL model by design exhibits thermodynamic consistency for the composition ...
[16]
Thermodynamic and statistical consistency of vapor–liquid ...
Mar 25, 2016 · Thermodynamic consistency of VLE data is usually checked with the area test ... The VLE data was described using the NRTL equation, with two ...
[17]
First principles prediction of NRTL binary interaction parameters for ...
Jun 29, 2023 · Most of these methods are implemented in commercial software packages like Aspen Plus to allow straightforward process design. ... NRTL-HOC model ...
[18]
Correlation of the liquid–liquid equilibrium data for specific ternary ...
This paper presents the results of a liquid–liquid equilibrium data correlation for 11 ternary systems which have not been previously fitted using the NRTL ...
[19]
Should we trust all the published LLE correlation parameters in ...
Feb 15, 2017 · This study reveals the necessity of designing a strategy for the validation of the parameters and compositions calculated, specifically in LLE data fitting.
[20]
Experimental Data and New Binary Interaction Parameters for ...
The mean error for NRTL for the variable temperature is 0.033% and for the liquid composition of ethanol in molar fraction the mean error is 0.495%. For the ...
[21]
[PDF] Physical Property Methods and Models 11.1
This manual is intended as a guide to using AspenTech's software. This documentation contains AspenTech proprietary and confidential information and may not ...
[22]
On the calculations of azeotropes from the NRTL and UNIQUAC ...
The local composition models of NRTL and UNIQUAC have been used in predicting of azeotropic data for binary and ternary liquid mixtures of various types.
[23]
Isobaric Vapor–Liquid Equilibrium Prediction from Excess Molar ...
Jul 26, 2023 · This work has systematically assessed the feasibility of determining VLE data based on excess molar enthalpy (h E ) results.
[24]
Liquid–Liquid Equilibrium Data Correlation Using NRTL Model for Different Types of Binary Systems: Upper Critical Solution Temperature, Lower Critical Solution Temperature, and Closed Miscibility Loops
### Summary of NRTL Application to LLE for UCST Systems (e.g., Water-n-Butanol)
[25]
Error (ACS Publications)
**Summary:**
[26]
Thermodynamic consistency tests based on the Gibbs-Duhem ...
The method is based on the Gibbs–Duhem equation, on the fundamental equation of phase equilibrium and on an appropriate combination between equations of state, ...<|control11|><|separator|>
[27]
One-parameter modified nonrandom two-liquid (NRTL) activity ...
Oct 15, 2014 · The nonrandom two-liquid (NRTL) model is utilized widely in phase equilibria calculations and employs three adjustable parameters.
[28]
[PDF] VAPOR LIQUID EQUILIBRIA: A REVIEW | TSI Journals
Jun 3, 2012 · The equations proposed by Van Laar, Margulas, Wilson, NRTL and UNIQUAC, which express the activity coefficients of the components of a mixture ...
[29]
[PDF] Modeling Liquid-Liquid Equilibrium of Ionic Liquid Systems with ...
Three different excess Gibbs free energy models are evaluated: the NRTL, UNIQUAC and electrolyte-. NRTL (eNRTL) models. In the case of eNRTL, a new ...<|control11|><|separator|>
[30]
https://www3.nd.edu/~markst/simoni-lle.pdf
[31]
SPT-NRTL: A physics-guided machine learning model to predict ...
The SPT-NRTL model enables the calculation of thermodynamically consistent, concentration-dependent activity coefficients of binary mixtures in the liquid phase ...<|control11|><|separator|>