Peng–Robinson or Redlich–Kwong? Twu or Soave α-Function? Which Combination of Cubic Equation of State (CEoS) and α-Function Produces More Accurate and Consistent Results for Pure Components Authors: Twaha Mohamed, Janusz Kozinski, Francisco Ramos-Pallares Published: 2024 ABSTRACT This study tested the accuracy and thermodynamic consistency of four CEoS/α-function models. The objective was to find the most suitable CEoS/α-function combo for producing accurate and consistent physical and derivative properties for nonpolar, polar and hydrogen bonding components at subcritical conditions. The models tested were PR-Twu, PR-Soave, RK-Twu, and RK-Soave. The first term in the model’s name refers to the CEoS used: Peng–Robinson (PR) or Redlich–Kwong (RK). The second term indicates the α-function used, i.e., Twu’s or Soave’s. The models were tested on a data set containing saturation pressure, enthalpy of vaporization and saturated liquid heat capacity of 147 pure components classified as polar, nonpolar, and hydrogen bonding. The three Twu α-function parameters were fitted to data and constrained to produce thermodynamic consistent values across the phase diagram; and, the Soave α-function parameter was predicted from a well-known correlation. The thermodynamic consistency of the models was assessed by calculating the Waring number and the saturated liquid speed of sound of 147 and 79 pure components, respectively. The results showed that PR-Twu and RK-Twu produced more accurate pure component properties compared to those from PR-Soave and RK-Soave. However, there was not a significant difference between the performance of PR-Twu and RK-Twu for calculating pure component properties. The same result was obtained when comparing PR-Soave and RK-Soave. Interestingly, the consistency analysis showed that only PR-Twu and PR-Soave produced consistent Waring number trends for components with acentric factors below 0.7. It was also observed that the saturated liquid speed of sound calculated from all four models tested was not accurate as the models cannot produce precise −(∂P∂v)T and liquid volumes. Besides, using volume translation is detrimental to the accuracy of the calculated saturated liquid speed of sound. The most accurate and consistent model was PR-Twu; however, caution should be exercised when modeling the saturated liquid heat capacity of hydrogen bonding components. Introduction Cubic equations of state (CEoS) play a very important role in process design, simulation and optimization because of their simplicity, accuracy, and computational efficiency.1,2 A CEoS calculates the total pressure of a system from contributions arising from repulsive and attractive interactions between molecules. The term accounting for repulsive interactions (or repulsive term) is derived from the kinetic theory of hard spheres3−5 and consequently is theoretically sound. However, the term accounting for attractive interactions (attractive term) is empirical and needs to be adjusted to improve the performance of the CEoS in matching pure component saturation pressure. This adjustment is carried out by introducing a temperature-dependent, component-specific parameter known as the α parameter. The α parameter is calculated by fitting the CEoS to experimental pure component vapor pressure (Psat), enthalpy of vaporization (ΔHvap) and saturated liquid heat capacity (Cpsat).6−8 The empirical expression for the calculation of the α parameter proposed by Soave9 has been adopted as the basis for calculations using the Redlich and Kwong (RK)10 and the Peng and Robinson (PR)11 CEoS, because of its simplicity and predictive capability. However, reported results suggest that the Soave’s α-function does not produce accurate Psat results for polar components,12−14 and is inconsistent at high temperatures.2,15,16 Numerous α-functions have been proposed and extensive reviews have been reported elsewhere.2,14,15 Among those α-functions, the expression proposed by Twu13 has been used in engineering calculations because of its accuracy and consistency. A brief review of the Soave and Twu α-functions is presented here as they were used in this study. The α-function proposed by Soave9 calculates α using the pure component’s acentric factor and reduced temperature (Tr) as inputs. This α-function is empirical and was obtained by plotting the square root of the fitted α parameters for 9 components (linear/branched alkanes, and aromatics) versus the square root of the reduced temperature. The resulting trends were linear and constrained to produce a value of α = 1 at Tr = 1 to ensure the CEoS fits the critical point as required. The slope of the lines, also known as the m parameter, was component-specific and a generalized correlation for its calculation was proposed. This generalized correlation is a quadratic polynomial of the acentric factor with coefficients fitted to match the saturation pressure at Tr = 0.7. The RK CEoS when combined with the Soave α-function is commonly referred to as the Soave–Redlich–Kwong (SRK) CEoS.9,10 Peng and Robinson11 used the same α-function proposed by Soave but the coefficients of the quadratic polynomial were fitted to match pure component saturation pressure from the normal boiling point up to the critical point. The motivation behind this fitting approach was to improve the performance of the PR CEoS at high temperatures. These original expressions for the calculation of the slope of the Soave’s α-function, or m parameter, have been modified through the years to improve their accuracy. Graboski and Daubert17 updated the original Peng and Robinson expression for the calculation of m by fitting the vapor pressure data of 41 pure components including alkanes (C1–C20), olefins (C2–C10), cycloalkanes (C3–C10), aromatics (C6–C10), ammonia, carbon dioxide, hydrogen sulfide, sulfur dioxide, nitrogen, and water. According to Graboski and Daubert, the PR CEoS with the updated expression for m produced more accurate vapor–liquid partitioning coefficients than those from the same CEoS with the original expression for m; however, the authors did not present a comparison of the deviations between the updated and original expression for m. Peng and Robinson18 updated their original expression for m by proposing a piecewise function for its calculation according to the acentric factor of the component. The updated piecewise function produced results with lower deviations for heavy hydrocarbons with an acentric factors higher than 0.491 compared to those from the original expression for m.19,20 Piña-Martinez et al.21 proposed an updated expression for the calculation of m for the RK and PR CEoS by fitting these equations to match pseudo experimental (rather than experimental) saturation pressure, enthalpy of vaporization and saturated liquid heat capacity of 1,721 pure compounds, including hydrocarbons, alkanes, aromatics, esters, ethers, nitrogen compounds, acids, alcohol, peroxides, polyfunctional, and silanes/siloxanes. Piña-Martinez et al. reports that using the updated expressions for m produced lower deviations than those obtained when the original expressions were used, especially for components with acentric factors higher than 0.7. Despite the simplicity of the Soave α-function, this function is not flexible enough to capture the properties of polar and hydrogen-bonding components2,14 and does not produce consistent results at extreme temperatures as reported by a handful of authors.2,15,16,22 Twu et al.13 developed a 3-parameter exponential α-function that produced more accurate results for polar components than the Soave α-function. The Twu α-function is temperature-dependent and uses three component-specific parameters as inputs. Twu et al.13 proposed an α-function that satisfied the following thermodynamic consistency constraints regarding the value of α: (1) that it is always positive; (2) that α equals unity at the critical point; and (3) that α approaches zero as the temperature approaches infinity. Twu et al. reported three component-specific parameters for nine polar (acetone, methanol, ethanol, water, ethylene glycol, diethylene glycol, and triethylene glycol) and twenty-two nonpolar (linear alkanes C1–C20, cyclohexane, and benzene) components. All the three component-specific parameters were calculated by fitting the RK CEoS with the Twu α-function to saturation pressure and saturated liquid heat capacity data. Only heat capacity data at low temperatures was used in the fitting. Twu et al. claimed that the proposed α-function fitted the saturation pressure of pure components from the triple point to the critical point within the error of the measurement.13 The main disadvantage of the Twu α-function is that there is not a predictive correlation for the three component-specific parameters. This limitation makes the equation less attractive for general use in engineering calculations. However, several studies have investigated the Twu α-function due to the function’s accuracy and flexibility.14,15,23,24 In addition to the higher accuracy and flexibility of the Twu α-function, compared to that of Soave’s, the main advantage of the Twu α-function is that it can be easily constrained to ensure thermodynamic consistency across the entire phase diagram. Le Guennec et al.15 reported a set of conditions that any α-function must satisfy to produce consistent thermodynamic results. These authors showed that the Twu α-function can be fully constrained to ensure thermodynamic consistency, but the Soave α-function cannot be constrained and therefore produces inconsistent results at high temperatures. Le Guennec et al.15 calculated the three component-specific parameters of the Twu α-function for 1197 components from different chemical families including alkanes, alkynes, aromatics, ethers, esters, acids, ketones, aldehydes, alcohols, sulfur and nitrogen-containing compounds. The three component-specific parameters were calculated by fitting the RK and the PR CEoS to pseudoexperimental vapor pressure, enthalpy of vaporization and/or saturated liquid heat capacity when available. Le Guennec et al. used a constrained optimization algorithm to ensure the fitted Twu α-function parameters satisfied the proposed thermodynamic consistency conditions. Bell et al.,6 using a similar constrained optimization algorithm, calculated consistent Twu α-function parameters for 2500 pure components from the same chemical families as those investigated by Le Guennec et al. The Twu α-function parameters reported by Bell et al. were obtained by fitting the PR CEoS to actual experimental pure component vapor pressure, enthalpy of vaporization and/or saturated liquid heat capacity rather than pseudoexperimental data as done by Le Guennec et al. Both, Le Guennec et al. and Bell et al., used a weighted objective function and set the weighting factors for saturation pressure and enthalpy of vaporization equal to one and adjusted the weighting factor for the saturated liquid heat capacity to produce fitted properties. When the weighting factor for the saturated heat capacity was set to one, the accuracy of the fitted vapor pressure and enthalpy of vaporization was typically low because CEoSs are not suited to accurately fit saturated liquid heat capacities as noted by Piña-Martinez et al.8 To date, one of the few available comparisons between the RK and the PR CEoS with α parameter calculated from the Soave and Twu α-functions has been published by Piña-Martinez et al.25 Piña-Martinez et al. reported deviations between fitted and pseudoexperimental pure component saturation pressure, enthalpy of vaporization and saturated liquid heat capacity data and concluded that coupling either the RK or PR CEoS with the Twu α-function produces lower deviations compared to those when coupled with the Soave α-function. This conclusion was drawn by comparing the deviations for 1800 pure components from assorted chemical families including alkanes, aromatics, alcohols, ketones, aldehydes, amines, amides, etc. However, lower deviations do not necessarily imply a model is capable of producing property trends consistent with those experimentally measured. Therefore, in this study, the performance of the RK or PR CEoS was not only assessed based on fitting capabilities and deviations but also by assessing the thermodynamic consistency of the calculated property trends as well as the consistency of derivative properties such as the saturated liquid speed of sound and the Waring number (Section 3.2). The purpose was to gain a better understanding of the quality, reliability and performance of CEoS. The objectives of this study were to (1) fit the RK and PR CEoS with α parameters from the Twu α-function; (2) compare the deviations and trends produced by the RK and PR CEoS with α parameter from the Twu and Soave α-functions on pure component saturation pressure (Psat), enthalpy of vaporization (ΔHvap) and saturated liquid heat capacity (Cpsat); and, (3) assess the consistency of the results by calculating derivative properties such as the Waring number and the saturated liquid speed of sound. To do so, the three component-specific parameters of the Twu α-function were calculated by fitting the RK and PR CEoS to experimental saturation pressure, enthalpy of vaporization and saturated liquid heat capacity of 147 pure components (objective 1). Saturated liquid density was not used for fitting because CEoS are unable to accurately match this property unless a temperature dependent volume translation expression is used. These expressions are thermodynamically inconsistent as they produce incorrect isotherms as well as negative heat capacities as reported by Jaubert et al.7 Most of the experimental data were collected from the NIST database,26 with some components utilizing supplementary ΔHvap and Cpsat data from the databases compiled by Zábranský et al.;27 and Acree and Chickos.28 The Soave α-function was used as reported by Piña-Martinez et al.21 To help in the analysis of the results the pure components were classified into three groups: Group 1, nonpolar components (50 components); Group 2, polar components (61 components); and Group 3, hydrogen-bonding components (36 components). The saturation pressure, enthalpy of vaporization, and saturated liquid heat capacity were calculated using the RK and PR CEoS coupled with the corresponding α parameter from the Twu and Soave α-functions; the resulting deviations were tabulated; and the experimental and calculated trends were compared (objective 2). Finally, the consistency of the models was assessed by applying a qualitative test based on the Waring number and by examining the calculated and experimental saturated liquid speeds of sound (objective 3). The experimental saturated liquid speed of sound data used in this study were collected from the NIST Database.26 Modeling Four different models were tested in this study: PR-Twu, PR-Soave, RK-Twu, and RK-Soave. The first term in the model’s name refers to the CEoS used: Peng–Robinson (PR) or Redlich–Kwong (RK). The second term indicates the α-function used. These four models can be written using the generalized CEoS as follows:1where P, T, and v are the pressure, the absolute temperature, and the molar volume respectively; R is the gas constant; and, δ1 and δ2 are CEoS-specific parameters summarized in Table 1. The expressions for b and ac are given by23where Ωb and Ωa are CEoS-specific parameters presented in Table 1. The α parameter in eq 1 was calculated using the following α-functions: Table 1Parameters for the Generalized Cubic Equation of State (Eq 1)parameterPeng–Robinson (PR)Redlich–Kwong (RK)Ωα0.4570.427Ωβ0.0780.087δ11 + (2)0.51δ21 – (2)0.50 Parameters for the Generalized Cubic Equation of State (Eq 1) Soave α-Function:9 4 The parameter m in eq 4 is a CEoS-specific parameter and was calculated as follows for the PR and RK CEoS:2156where ω is the Pitzer acentric factor and subscripts PR and RK refer to the CEoS. Twu α-Function:13 7 Parameters α1, α2, and α3 in eq 7 are component-specific parameters calculated by fitting the PR-Twu and RK-Twu models to experimental Psat, ΔHvap, and Cpsat as recommended by Piña-Martinez et al.8 The generalized CEoS (eq 1) can be converted into the PR-Twu, PR-Soave, RK-Twu, or RK-Soave by selecting the appropriate CEoS parameters (Table 1) and α-function (eqs 4 or 7), respectively. Flash Calculations The Saturation pressure (Psat), enthalpy of vaporization (ΔHvap) and saturated liquid heat capacity (Cpsat), needed for fitting the Twu α-Function parameters (eq 7), were calculated from the isothermal two-phase flash algorithm described elsewhere.29 The flash algorithm solves the equal fugacity coefficient requirement of equilibrium in the case of pure components using the accelerated substitution method proposed by Michelsen and Mo̷llerup.29 The pure component fugacity coefficient was calculated as8where Z is the compressibility factor. The residual Helmholtz energy (Ar) is given by299 The flash algorithm iterates until it finds the Psat and the compressibility factors for the vapor, Zv, and liquid, Zl, phases that satisfy the equal fugacity coefficient requirement for pure components. The flash algorithm was implemented in MATLAB.30 The average speed of the algorithm was 1000 flash calculations per second with an Intel Core i7-9700 CPU@3.00 GHz processor. The enthalpy of vaporization, ΔHvap, was calculated using Psat, Zv, and Zl as inputs, using the following equations:10 The enthalpy of either the vapor, Hv, or the liquid, Hl, phase was calculated as2911121314where Sr is the residual entropy; and, Hideal gas is the enthalpy of the ideal gas at the system temperature. The Hideal gas cancels out when evaluating eq 10; and therefore, does not need to be calculated. The Cpsat was calculated as follows:291516171819where the superscript r refers to a reduced property; and, Cpidealgas is the heat capacity of the ideal gas at the system’s absolute temperature and was calculated using the following correlation:20where A, B, C, and D are component-specific parameters reported by Yaws.31 The flash algorithm implemented in this study calculates Psat, ΔHvap, and Cpsat using the following inputs: (1) the absolute temperature of the system; (2) constants A, B, C, and D in eq 20; (3) the critical temperature (Tc) and pressure (Pc) of the pure component for the calculation of b and ac using eqs 2 and 3, respectively; and (4) parameters α1, α2, and α3 for the calculation of α from the Twu α-function (eq 7). All the required inputs are known except for parameters α1, α2, and α3. Fitting Algorithm The parameters α1, α2, and α3 from the Twu α-function (eq 7) were calculated by minimizing the following objective function (OF):21where Wf is the property weighting factor; and subscripts Psat, ΔHvap, and Cpsat refer to pure component saturation pressure, enthalpy of vaporization and saturated liquid heat capacity, respectively. The MSLE is the mean squared log error calculated as follows:22where subscripts calc and exp refer to the calculated and experimental property, respectively; X is the property under study; and NP is the number of experimental data points. The MSLE was chosen in this study because it is appropriate to quantify deviations for data sets containing small and large values as in the case of the saturation pressure (low values near the triple point and large values near the critical point). The minimization of the objective function (eq 21) was constrained to satisfy the following thermodynamic consistency requirements proposed by Le Guennec et al.:1523242526where α is the Twu α-function (eq 7). The weighting factors in the objective function (eq 21) were normalized as follows:27 To ensure high accuracy for Psat (required for calculating accurate phase boundaries), its weighting factor was constrained to vary within the following interval determined by trial and error:28 The weighting factor for the ΔHvap was constrained to vary within the following interval also determined by trial and error:29 The weighting factor for the Cpsat was calculated from eq 27. The weighting factors for the Psat and the ΔHvap were varied by 0.1 increments within the intervals shown in eqs 28 and 29, respectively, and the objective function (eq 21) was minimized for each combination of weighting factors. The solutions were assessed using Pareto analysis where the optimum set of weighting factors was selected as that producing the minimum Euclidean distance to the aspiration point. The aspiration point represents an unobtainable best-case scenario defined in this study as the set of weighting factors that produced a mean squared log error (MSLE) equal to zero for Psat and Cpsat. The MSLE for ΔHvap was excluded from the definition of the aspiration point because it was found that the weighting factors for this property do not significantly affect the accuracy of the fitted ΔHvap. It was also found that setting the weighting factor for ΔHvap equal to zero does not have a significant effect on the accuracy of ΔHvap or the accuracy of Psat and Cpsat. However, the accuracy of the fitted Psat and Cpsat was highly sensitive to their weighting factors. Regardless of this observation, the three properties (Psat, ΔHvap, and Cpsat) were used in the minimization of eq 21. Data Sets Three data sets were used in this study: one Development data set and two Test Data sets. The Development Data set was used to fit the three component-specific parameters of the Twu α-function (eq 7) for the PR-Twu and RK-Twu models. The Test Data sets were used to assess the performance of the models (PR-Twu, RK-Twu, PR-Soave, and RK-Soave) in predicting derivative thermodynamic properties such as the Waring number (Test Data set 1) and the saturated liquid speed of sound (Test Data set 2). Development Data Set The data in this data set was used for fitting the three component-specific parameters, α1, α2, and α3, in eq 7, for the PR-Twu and RK-Twu models. It contains the experimental saturation pressure (Psat), enthalpy of vaporization (ΔHvap) and saturated liquid heat capacity (Cpsat) of 16 n-alkanes, 7 branched alkanes, 12 cyclic alkanes, 15 aromatics, 9 heterocyclic sulfur-containing compounds, 14 thiols, 13 pyridines, 3 aldehydes, 16 ketones, 6 nitriles, 9 amines, 3 heterocyclic nitrogen-containing compounds, 12 alcohols, 10 carboxylic acids, water, and ammonia. Most of the experimental data were collected from the NIST database,26 with some components utilizing supplementary ΔHvap and Cpsat data from the databases compiled by Zábranský et al.;27 and, Acree and Chickos.28 The PR-Twu and RK-Twu models were fitted to the data using the optimization procedure described in Section 2.2. Pure component inputs properties required for the flash algorithm described in Section 2.1, such as Pc, Tc, and the component-specific coefficients needed for the calculation of Cpidealgas (eq 20), were collected from the Yaws’ Handbook of Thermodynamic and Physical Properties of Chemical Compounds.31 Values of the pure component’s acentric factor, ω, were also included in this data set as they were used as an input for the calculation of the parameter m (eqs 5 and 6) for PR-Soave and RK-Soave models. The acentric factors were also collected from the Yaws’ Handbook of Thermodynamic and Physical Properties of Chemical Compounds.31 All the data in the Development Data set were screened to detect and eliminate obvious outliers using a quantile regression algorithm implemented in MATLAB.32 Additionally, manual screening was performed to ensure all obvious outliers were eliminated. To aid in the analysis of the results, the components in this data set were classified into three groups based on the nature of their intermolecular interactions: (1) Group 1: nonpolar, (2) Group 2: polar, and (3) Group 3: hydrogen bonding components. The components in Group 2 are polar but do not exhibit classical hydrogen bonding interactions defined as the attractive force between a hydrogen atom bonded to any of the following electronegative atoms: oxygen, nitrogen, fluorine, and chlorine.33Table 2 summarizes relevant information about the Development Data set. The fitted Twu α-function parameters, α1, α2, and α3, for all the components in the Development Data set are summarized in the Supporting Information. Table 2Number of Data Points (NP) and Reduced Temperature Range (Tr range) for the Saturation Pressure (Psat), Enthalpy of Vaporization, (ΔHvap), and Saturated Liquid Heat Capacity (Cpsat) Data for the Different Chemical Families in the Development DatasetafamilyNCSCN rangePsatΔHvapCpsatNPTr rangeNPTr rangeNPTr rangegroup 1: nonpolar componentsn-alkanes161–1620,6360.23–1.005840.37–0.9719340.22–0.80branched75–821180.39–1.001660.45–0.937100.16–0.72cyclics123–1649360.34–1.001640.39–0.816200.25–0.63aromatics156–1212,9240.31–1.005880.37–0.9414860.27–0.79group 2: polar componentsHSCCb92–814280.33–1.001120.41–0.844660.25–0.80thiols141–78020.43–1.00960.43–0.9111580.18–0.64pyridines135–935240.37–0.991780.39–0.829440.32–0.79aldehyde33–63260.43–0.78320.50–0.64720.36–0.57ketones163–1368140.33–1.003720.35–0.968860.25–0.80nitriles62–710960.33–1.00720.43–0.861920.30–0.64group 3: hydrogen bonding componentsamines91–619040.41–1.00700.48–0.902820.38–0.67HNCCc34–57600.38–1.00320.52–0.641720.38–0.63alcohols121–1217,9120.30–1.014640.41–0.948080.30–0.50H2O/NH320–045100.39–1.001240.46–0.803140.31–0.79acids104–1136260.37–0.751120.39–0.901720.38–0.62overall1470–1683,3160.23–1.0131660.35–0.9710,2160.16–0.80aNC and SCN range stands for the number of components and the single carbon number range of the components in each chemical family, respectively. Branched and cyclics refers to branched and cyclic alkanes, respectively.bHSCC: Heterocyclic Sulfur-Containing Compound.cHNCC: Heterocyclic Nitrogen-Containing Compound. Number of Data Points (NP) and Reduced Temperature Range (Tr range) for the Saturation Pressure (Psat), Enthalpy of Vaporization, (ΔHvap), and Saturated Liquid Heat Capacity (Cpsat) Data for the Different Chemical Families in the Development Dataseta NC and SCN range stands for the number of components and the single carbon number range of the components in each chemical family, respectively. Branched and cyclics refers to branched and cyclic alkanes, respectively. HSCC: Heterocyclic Sulfur-Containing Compound. HNCC: Heterocyclic Nitrogen-Containing Compound. Figure 1 shows the distribution of the data points in the Development Data set for Psat, ΔHvap, and Cpsat, respectively, versus fixed ranges of reduced temperature (Tr). The Tr range for Psat and ΔHvap data was from 0.3 to 1.0, which covers temperatures from slightly above the triple point to up to the critical point, Figure 1a,b. The PR-Twu and RK-Twu were fitted to Psat and ΔHvap data at reduced temperatures from 0.3 to 1; and the deviations between experimental and fitted Psat and ΔHvap, presented in Section 4, were calculated over the same reduced temperature range. For Cpsat,Figure 1c, only data in a Tr range from 0.3 to 0.8 was considered because, as described by Kontogeorgis and Folas,1 CEoS are unable to capture Cpsat in the vicinity of the critical point (Tr > 0.8). The PR-Twu and RK-Twu were fitted to Cpsat data at reduced temperatures from 0.3 to 0.8, except for alcohols. In the case of alcohols, PR-Twu and RK-Twu were fitted to Cpsat data at reduced temperatures from 0.3 to 0.5 because, at higher temperatures, changes in self-association patterns produce an inflection point in the experimental Cpsat which cannot be captured by CEoS as they do not account for component self-association.34−36 However, the deviations between experimental and fitted Cpsat, presented in Section 4, were calculated over the entire reduced temperature range shown in Figure 1c for all components in the Development Data set, including alcohols. Distribution of data points versus reduced temperature (Tr) ranges for (a) saturation pressure (Psat), (b) enthalpy of vaporization (ΔHvap) and (c) saturated liquid heat capacity (Cpsat). G-1, G-2, and G-3 refer to Group 1 (nonpolar), Group 2 (polar), and Group 3 (Hydrogen Bonding), respectively, in the Development Data set. Test Data Sets: Assessment of the Thermodynamic Consistency of the Models The thermodynamic consistency of the PR-Twu, PR-Soave, RK-Twu, and RK-Soave models was assessed by the calculation of the Waring number and the saturated liquid speed of sound for the pure components in Test Data sets 1 and 2, respectively. Test Data Set 1: The Waring Number This data set includes the same 147 components included in the Development Data set. For each one of the components in this test data set, the Waring Number (W) was calculated at reduced temperatures from 0.3 to 1 according to the following expression:3730where the derivative on the right-hand side of eq 30 was calculated from the PR-Twu, PR-Soave, RK-Twu, and RK-Soave models. The calculated Waring number was then plotted against the reduced temperature and the resulting trend was used to assess the thermodynamic consistency of the models. A model produces a consistent Waring Number trend for a given component if it is convex, continuous, and shows a minimum at a temperature below the critical point.6,38−41 The temperature at which the minimum value of the Waring number is found is known as the Waring temperature (TW). The reciprocal of TW is the condition at which the inflection point in a log vapor pressure-inverse temperature curve is observed.42 Therefore, the analysis of the Waring number is an indirect approach for testing the thermodynamic consistency of CEoS-calculated vapor pressures. In addition, the values of TW reported by Velasco et al.40 for 105 pure components were also included in this test data set for comparison purposes. The results of the analysis of the Waring Number for the components in the Test Data set 1 is presented in Section 4.5.1. Test Data Set 2: Saturated Liquid Speed of Sound This data set contains the experimental saturated liquid speed of sound for 79 pure components classified as nonpolar (Group 1), polar (Group 2), and hydrogen bonding (Group 3). Additional data such as the pure components’ experimental saturated liquid density, required for calculating the volume translation parameter; critical temperature, critical pressure and acentric factor, required as input for the CEoS; and the molecular weight, required for the calculation of the speed of sound (eq 31), were also included in this test data set. The experimental saturated liquid speed of sound and density data were collected from the NIST Database,26 and the components’ critical properties, molecular weights and acentric factors were collected from the Yaws handbook.31 Relevant details about this test data set are presented in Table 3. The data set was screened to eliminate outliers using the method described in Section 3.1. Table 3Number of Data Points (NP) and Reduced Temperature Range (Tr range) for the Experimental Saturated Liquid Speed of Sound (u) and Density (ρ) of the Components in the Test Dataset 2afamilyNCSCN rangeu, m s–1ρ, kg m–3NPTr rangeNPTr rangegroup 1: nonpolar componentsn-alkanes141–1616760.30–1.0049900.30–1.00branched45–71060.43–0.626700.30–1.00cyclics76–102100.45–0.975940.31–1.00aromatics96–137980.31–1.0022600.31–1.00group 2: polar componentsHSCCb22–4460.40–0.48660.41–0.99pyridines55–91760.37–0.542720.31–1.00aldehydes17–7180.42–0.469420.36–1.00ketones113–83380.40–0.891180.40–0.86nitriles43–71200.42–0.871940.38–1.00group 3: hydrogen bonding componentsamines62–5900.49–0.631940.38–1.00HNCCc34–5700.46–0.62 alcohols111–1212240.34–1.0023000.30–1.00H2O/NH320–0800.49–1.006100.37–1.00acids44–880.42–0.482520.35–0.87overall791–1649600.30–1.0013,2680.30–1.00aNC and SCN range stand for the number of components and the single carbon number range of the components in each chemical family, respectively. n-Alkanes, branched and cyclics refer to normal, branched and cyclic alkanes, respectively.bHSCC: Heterocyclic Sulfur Containing Compound.cHNCC: Hydrogen-Bonding Heterocyclic Nitrogen-Containing Compound. Number of Data Points (NP) and Reduced Temperature Range (Tr range) for the Experimental Saturated Liquid Speed of Sound (u) and Density (ρ) of the Components in the Test Dataset 2a NC and SCN range stand for the number of components and the single carbon number range of the components in each chemical family, respectively. n-Alkanes, branched and cyclics refer to normal, branched and cyclic alkanes, respectively. HSCC: Heterocyclic Sulfur Containing Compound. HNCC: Hydrogen-Bonding Heterocyclic Nitrogen-Containing Compound. The thermodynamic consistency of the PR-Twu, PR-Soave, RK-Twu, and RK-Soave models was assessed by comparing experimental and calculated saturated liquid speed of sound (u). This property was chosen because is highly sensitive to the first and second derivatives of an EoS; in other words, the calculation of the saturated liquid speed of sound is a useful indirect test for checking the consistency of the derivatives calculated from the PR-Twu, PR-Soave, RK-Twu, and RK-Soave models.1,43−46 The saturated liquid speed of sound was calculated as follows:2931where Cp and Cv are the isobaric and isochoric saturated liquid heat capacities, respectively; MW is the molecular weight; ∂P∂v is the derivative of the pressure with respect to the saturated liquid volume; and, v is the saturated liquid volume. Cp, Cv, and ∂P∂v were calculated from eqs 15 to 19 using the first and second derivatives of the residual Helmholtz energy function (eq 9) corresponding to the PR-Twu, PR-Soave, RK-Twu, and RK-Soave models. The saturated liquid volume was calculated from the Peneloux volume translation expression:4732where superscript model refers to the PR-Twu, PR-Soave, RK-Twu, or RK-Soave models; and c is a temperature-independent, component-specific parameter used to improve the accuracy of the calculated volume from a CEoS. As proved by Jaubert et al.,7 the application of the Peneloux volume translation, eq 32, does not introduce inconsistencies into the CEoS and does not affect the calculated Cp, Cv, and ∂P∂v. Therefore, as seen in eq 31, of the variables used in calculating u, only v is affected by volume translation. Therefore, the magnitude of u is also affected by volume translation. The parameter c in eq 32 was calculated by fitting the saturated liquid volume calculated from the model (PR-Twu, PR-Soave, RK-Twu, and RK-Soave) to that calculated from the modified Rackett equation48 at a reduced temperature of 0.8 as proposed by Le Guennec et al.:4933The Rackett equation requires the following parameters as inputs: critical temperature and pressure; and, an adjustable parameter, ZRa, tuned to match the experimental saturated liquid volume at Tr = 0.8 according to the procedure described elsewhere.50Equation 33 was used for calculating the parameter c for all the components in the Test Data set 2 The deviations between experimental and calculated saturated liquid densities are presented in Section 4.5.2. The calculated c parameters for the components in the Test Data set 2 are presented in the Supporting Information. For comparison purposes, the saturated liquid speed of sound for the components in Test Data set 2 was calculated using untranslated (c = 0) and translated (c from eq 33) saturated liquid volumes. The translated models will be referred to as PR-Twu-t, PR-Soave-t, RK-Twu-t and RK-Soave-t in Section 4.5.2. In addition, to help in the analysis of the results presented in Section 4.5.2, the following version of eq 31 was used:34eq 34 was obtained by applying the logarithm on both sides of eq 31. According to eq 34, the log of the squared speed of sound is the result of the summation of contributions arising from the saturated liquid molar volume, the ratio of heat capacities and the first derivative of pressure with respect to volume. As it will be shown in Section 4.5.2, by using eq 34 it is possible to assess the role of these different contributions on the saturated liquid speed of sound. Results The following nomenclature was used for the models tested in this study: PR-Twu, PR-Soave, RK-Twu, and RK-Soave. The first term in the model’s name refers to the CEoS used: Peng–Robinson (PR) or Redlich–Kwong (RK). The second term indicates the α-function used: Twu’s (eq 7) or Soave’s (eq 4). The three component-specific parameters in the Twu α-function were calculated by fitting PR-Twu and RK-Twu to experimental saturation pressure, enthalpy of vaporization and saturated liquid heat capacity as described in Section 2. All the fitted Twu α-function parameters are summarized in the Supporting Information. In the case of the PR-Soave and RK-Soave models, the component-specific parameter m, required as input for the Soave α-function (eq 4), was predicted from the expressions proposed by Piña-Martinez et al. (eqs 5 and 6) using the component’s acentric factor as input. The performance of the PR-Twu, PR-Soave, RK-Twu, and RK-Soave on calculating pure component properties is presented first and then a discussion on their consistency, assessed by the calculation of the Waring number and saturated liquid speed of sound, is presented. The following error metric was used to assess the deviations between calculated (or fitted) and experimental values:35where AARD is the average absolute relative deviation, NP is the number of data points, X is the property under study and the subscripts exp and calc indicate the experimental and calculated property. Psat, ΔHvap, and Cpsat for the Non-Polar Components in Group 1 Figure 2 shows the overall AARD for the saturation pressure, Psat (panel a), enthalpy of vaporization, ΔHvap (panel b) and saturated liquid heat capacity, Cpsat (panel c), for the nonpolar chemical families in Group 1 in the Development Data set (Table 2). In the case of Psat, (Figure 2a), lower AARDs were obtained when using the Twu α-function for all the chemical families in Group 1. For normal alkanes and branched alkanes, there was not a significant difference among the four models tested; however, for cyclics and aromatics, PR-Twu and RK-Twu produced results with significantly lower AARD compared to those from PR-Soave and RK-Soave. The results also showed that there are no significant differences between the PR-Twu and RK-Twu models when used for the same chemical family. The same observation applies to the PR-Soave and RK-Soave models. Overall AARD (%) for (a) saturation pressure, Psat; (b) enthalpy of vaporization, ΔHvap; and (c) saturated liquid heat capacity, Cpsat, for the nonpolar chemical families in Group 1 in the Development Data set. The chemical families are arranged by increasing polarity from left to right. Alkanes, branched and cyclics refer to normal, branched, and cyclic alkanes, respectively. As seen in Figure 2a, the AARD for Psat of normal alkanes is higher than that of branched alkanes for all four models tested. It was found that the overall AARD within a chemical family increased with the carbon number of the component regardless of the model tested. Therefore, the higher AARD found for normal alkanes compared to that for branched alkanes is because the normal alkanes included in Group 1 cover a wider range of carbon numbers (from 1 to 16). The branched alkanes in Group 1 cover a carbon number range ranging from 5 to 8. The highest AARD for Psat were found for the aromatics in Group 1, Figure 2a. This result might suggests that the PR and RK CEoS, regardless of the α-function, are unable to fully capture the stronger interactions exhibited by aromatics due to the stacking of aromatic rings.51 PR-Twu and RK-Twu fitted the Psat data in Group 1 with overall AARD values of 3.9 and 3.8%, respectively; and PR-Soave and RK-Soave with overall AARD values of 10.0 and 9.5%, respectively. The Psat calculated from PR-Twu and RK-Twu for selected n-alkanes is shown in the Supporting Information, Section S1. In the case of ΔHvap (Figure 2b), in general, PR-Twu and RK-Twu produced lower overall AARD than PR-Soave and RK-Soave. PR-Twu and RK-Twu fitted the ΔHvap of the components in Group 1 with an overall AARD of 1.5 and 1.6%, respectively. PR-Soave and RK-Soave predicted the ΔHvap of the components in Group 1 with an overall AARD of 2.8 and 2.7%, respectively. In the case of Cpsat (Figure 2c), PR-Twu and RK-Twu produced results with significantly lower AARD compared to those from PR-Soave and RK-Soave suggesting that the Twu α-function is more suitable than the Soave α-function for the calculation of Cpsat due to its greater flexibility resulting from its three component-specific parameters. PR-Twu and RK-Twu fitted the Cpsat of the components in Group 1 with an overall AARD of 2.7 and 2.8%, respectively; and the PR-Soave and RK-Soave predicted the Cpsat with an overall AARD of 6.9 and 5.0%, respectively. Psat, ΔHvap, and Cpsat for the Polar Components in Group 2 Figure 3 shows the overall AARD for the Psat (panel a), ΔHvap (panel b) and Cpsat (panel c) for the chemical families in Group 2 in the Development Data set (Table 2). All the components in Group 2 are polar but unable to self-associate via hydrogen bonding; and, they are ordered in Figure 3 from left to right in terms of increasing polarity. In general, for all three properties tested, PR-Twu and RK-Twu produced results with lower AARD compared to those from PR-Soave and RK-Soave. In the case of Psat, PR-Twu and RK-Twu produced significantly lower AARD compared to those from PR-Soave and RK-Soave showing that the Twu α-function is more appropriate for polar components due mainly to its flexibility conferred by its three parameters. However, as the polarity of the components increases, the AARD for Psat from PR-Twu and RK-Twu tends to increase, Figure 3a. A similar observation can also be made for PR-Soave and RK-Soave. PR-Twu and RK-Twu fitted the Psat of the components in Group 2 with an overall AARD of 4.8 and 4.7%, respectively; while PR-Soave and RK-Soave predicted Psat of the components in Group 2 with an overall AARD of 27.6 and 28.0%, respectively. Overall AARD (%) for (a) saturation pressure, Psat; (b) enthalpy of vaporization, ΔHvap; and (c) saturated liquid heat capacity, Cpsat, for polar chemical families in Group 2 in the Development Data set. All the components in Group 2 are polar but unable to self-associate via hydrogen bonding. The chemical families are arranged by increasing polarity from left to right. HSCC refers to Heterocyclic sulfur-containing Compounds. In the case of ΔHvap (Figure 3b) and Cpsat (Figure 3c), PR-Twu and RK-Twu produced results with lower AARD compared to those obtained from PR-Soave and RK-Soave for all the chemical families in Group 2. PR-Twu and RK-Twu fitted the ΔHvap of the components in Group 2 with an overall AARD of 2.8 and 2.7%, respectively; and fitted their Cpsat with an overall AARD of 3.4 and 3.3%, respectively. PR-Soave and RK-Soave predicted ΔHvap with an overall AARD of 7.4 and 8.5%, respectively; and Cpsat with an overall AARD of 8.3 and 7.0%, respectively. Psat, ΔHvap, and Cpsat for the Hydrogen-Bonding Components in Group 3 Figure 4 presents the overall AARD for the Psat (panel a), ΔHvap (panel b) and Cpsat (panel c) for the chemical families in Group 3 in the Development Data set (Table 2). These chemical families contain components that self-associate through hydrogen bonding; and, they are ordered in Figure 4 from left to right in terms of increasing hydrogen bonding strength. The Cpsat of alcohols was the most challenging property to fit because when fitted to data within a reduced temperature range from 0.3 to 0.8, PR-Twu and RK-Twu considerably overestimated Cpsat at low reduced temperatures where CEoS usually produce acceptable results for this property. To address this issue, the fitting approach for the Cpsat of alcohols was modified so only data at reduced temperatures from 0.3 to 0.5 was utilized for fitting. However, the AARD for Cpsat was calculated including data up to a reduced temperature of 0.8 for assessing the performance of the models at higher temperatures. Overall AARD (%) for (a) saturation pressure, Psat; (b) enthalpy of vaporization, ΔHvap; and (c) saturated liquid heat capacity, Cpsat, for the chemical families in Group 3 in the Development Data set. All the components in Group 3 can self-associate through hydrogen bonding. The chemical families are ordered from left to right in terms of increasing hydrogen bonding strength. HNCC refers to Heterocyclic Nitrogen-Containing Compounds. In general, PR-Twu and RK-Twu produced lower AARD for Psat, ΔHvap, and Cpsat than those from PR-Soave and RK-Soave for all the chemical families in Group 3 except for ΔHvap of carboxylic acids, Figure 4b. For carboxylic acids, PR-Soave and RK-Soave produced ΔHvap with slightly lower AARD than those from PR-Twu and RK-Twu. The PR-Twu model fitted the Psat, ΔHvap, and Cpsat of the components in Group 3 with an overall AARD of 11.0, 4.5, and 5.4%, respectively; and the RK-Twu model fitted the same three properties with an overall AARD of 10.4, 4.4, and 3.6%, respectively. The PR-Soave model predicted the Psat, ΔHvap, and Cpsat of the components in Group 3 with an overall AARD of 30.1, 7.6, and 13.9%, respectively; and the RK-Soave model with an overall AARD of 26.1, 6.9, and 14.8%, respectively. All these deviations are comparatively higher than those obtained for the components in Groups 1 (nonpolar) and 2 (polar). Figure 5 shows the Cpsat of the alcohols in Group 3 fitted to PR-Twu and predicted by PR-Soave. RK-Twu and RK-Soave were not included in Figure 5 because they showed similar performance to PR-Twu and PR-Soave, respectively. Figure 5a includes alcohols with carbon numbers between 1 and 5 as well as ammonia; and, Figure 5b includes alcohols with carbon numbers between 6 and 12 as well as water. Ammonia and water were included for comparison purposes. In general, the overall AARD between experimental and modeled Cpsat increases with increasing carbon number regardless of the model used. However, the PR-Twu model produces results with lower deviations compared to those from the PR-Soave model, especially for low carbon number alcohols such as methanol, ethanol and propanol, Figure 5a. As expected, none of the models tested could capture the inflection point in the Cpsat curve observed in Figure 5b for high carbon number alcohols. This inflection point is the result of changes in self-association patterns which cannot be captured by classic CEoS regardless of the α-function used.36,43,45,52 Note, despite of the model’s deficiencies in capturing Cpsat trends for alcohols, they all produced accurate Psat trends as shown in the Supporting Information, Section S2, for the alcohols in Group 3 in the Development Data set. Experimental and modeled saturated liquid heat capacity (Cpsat) of (a) light alcohols (C1–C5) and ammonia; and, (b) heavy alcohols (C6–C12) and water from Group 3 in the Development Data set. Tr is the reduced temperature. The results of this study indicate that the Twu α-function is more adequate than the Soave α-function to capture the effect of hydrogen bonding on Cpsat when hydrogen bonding is the dominant intermolecular interaction as in the case of ammonia, water, and low molecular weight alcohols such as methanol and ethanol (see Figure 5). This observation does not imply that the Twu α-function captures hydrogen bonding but rather produces a more appropriate α-parameter for low carbon number alcohols, water and ammonia. In the case of higher carbon number alcohols, the Twu and Soave α-functions do not produce accurate results because, as suggested by Cerdeiriña et al.,34 Bureš et al.,52 and Zábranský et al.,36 these alcohols do not only interact through hydrogen bonding but also through stronger dispersion forces resulting from the longer paraffinic chains. This combination of hydrogen bonding and stronger dispersion forces might produce clusters with the paraffinic chains oriented toward the center and hydroxyl groups at the outer surface of the cluster interacting through hydrogen bonding with other hydroxyl groups in other clusters. This nonrandom arrangement renders the repulsive-attraction interaction approach of CEoS unsuitable; with the α-function unable to correct this fundamental deficiency. More sophisticated models that account for self-association, such as SAFT53 and CPA54 should be used for alcohols. Figure 6 shows the experimental and calculated ΔHvap (panel a) and Cpsat (panel b) of selected components in Group 3. These components are butanoic acid, butanol, pyrrolidine (cyclic secondary amine) and diethylamine (secondary amine); and, they all have a carbon number of four and self-associate through hydrogen bonding. The Cpsat of butanol was not included in panel b as it was shown in Figure 5a. The ΔHvap and Cpsat shown in Figure 6 were calculated from RK-Twu and RK-Soave. PR-Twu and PR-Soave are not shown in Figure 6 as their performance is similar to those of RK-Twu and RK-Soave, respectively. As seen in Figure 6a, RK-Twu and RK-Soave capture the correct ΔHvap trends; however, RK-Soave produces results with lower AARD than those from RK-Twu for butanoic acid as well as for all the organic acids in Group 3, as seen in Figure 4b. In general, the four models tested in this study could capture the correct ΔHvap trends for all the hydrogen-bonding components in Group 3. In the case of Cpsat, Figure 6b, none of the four models produced satisfactory results for hydrogen-bonding components. PR-Twu and RK-Twu produced trends that exhibit a minimum in the Cpsat curve, Figure 6b; and, PR-Soave and RK-Soave produced results with significantly higher deviations although no minimum in the Cpsat curve was detected. These results suggest that the Twu and Soave α-functions are not adequate to qualitatively capture the Cpsat of hydrogen-bonding components; therefore, these α-functions should be fitted to only Psat and ΔHvap data in the case of hydrogen-bonding components. Experimental and modeled (a) enthalpy of vaporization (ΔHvap) and (b) saturated liquid heat capacity (Cpsat) for selected components from Group 3 in the Development Data set. All the components included have a carbon number of four and can self-associate through hydrogen bonding. Comparing the Models Figure 7 shows a summary of the overall AARD per group for the three properties calculated from PR-Twu and RK-Twu; and, from PR-Soave and RK-Soave. In general, regardless of the model used, the overall AARD of Psat, ΔHvap, and Cpsat increases as the nature of the pure components changes from nonpolar (Group 1) to polar (Group 2) to hydrogen-bonding (Group 3). PR-Twu and RK-Twu produce results with lower AARD than those from PR-Soave and RK-Soave. There was not a significant difference when comparing the AARDs obtained from PR-Twu and RK-Twu; and those from PR-Soave and RK-Soave. This result indicates that both cubic equations of state when coupled with the same α-function produce similar Psat, ΔHvap, and Cpsat for pure components. This result is similar to that reported by Ghanbari et al.55 and Twu et al.56 Overall AARD (%) for (a) saturation pressure, Psat; (b) enthalpy of vaporization, ΔHvap; and (c) saturated liquid heat capacity, Cpsat, for the three groups in the Development Data set. Groups 1, 2, and 3 contain nonpolar, polar and hydrogen bonding components, respectively. Thermodynamic Consistency Analysis Waring Number Figure 8 shows the Waring temperature, Tw, calculated from PR-Twu and PR-Soave (Panel a); and, from RK-Twu and RK-Soave (Panel b), versus the critical temperature, Tc, for the components in the Test Data set 1. The solid and dotted lines in Figure 8 correspond to the correlation proposed by Velasco et al.,40 for the prediction of Tw as a function of Tc, and, the Tw = Tc line, respectively. The Tw = Tc line was included for comparison as Tw is typically lower than Tc as reported elsewhere.40 As seen in Figure 8a, there is a good agreement between the Tw values calculated from PR-Twu and PR-Soave and those predicted from the correlation proposed by Velasco et al.40 All the Tw values calculated from the PR-Twu model are lower than the critical temperature indicating a consistent trend for the Waring number with a minimum in the Waring number at temperatures lower than the critical, as described in Section 3.2.1. However, for some components with critical temperatures above 500 K, their Tw values calculated from PR-Soave are equal to Tc indicating that the minimum of the Waring number is at the critical temperature suggesting potential inconsistencies. In the case of RK-Twu and RK-Soave, Figure 8b, the calculated Tw for components with critical temperatures above 500 K is very close, or equal, to their critical temperature indicating potential inconsistencies of RK-Twu and RK-Soave. Calculated Waring temperature, Tw, versus the critical temperature, Tc, for the components in the Test Data set 1. Tw was calculated from (a) PR-Twu and PR-Soave, and (b) RK-Twu and RK-Soave. The solid line corresponds to Tw predicted from the correlation proposed by Velasco et al.40 Figure 9 shows the reduced Waring temperature (TWr = Tw/Tc) versus the acentric factor of the components in Test Data set 1. The corresponding Waring temperatures were calculated from PR-Twu and PR-Soave (Panel a); and, RK-Twu and RK-Soave (Panel b). Figure 9 also includes data reported by Velasco et al.40 for comparison purposes. As seen in Figure 9a, there is a good agreement between PR-Twu and PR-Soave calculated and reported values when the component’s acentric factor is below 0.7. At higher acentric factors, reported TWr values decrease whereas calculated values increase rapidly approaching TWr = 1. This result suggests that PR-Twu and PR-Soave might be inconsistent for components with acentric factors higher than 0.7. In the case of RK-Twu and RK-Soave, Figure 9b, the calculated TWr values are consistently higher than those reported. Moreover, RK-Twu and RK-Soave produce TWr values equal to one for most of the components with acentric factors above 0.5 suggesting potential inconsistencies as the Waring temperature is found at the critical temperature. The scattering in the data from PR-Soave and RK-Soave seen in Figure 9 could be the consequence of using predicted rather than fitted α parameters as described in Section 2. Reduced Waring temperature (TWr = Tw/Tc) versus the acentric factor for the components in Test Data set 1. Tw was calculated from (a) PR-Twu and PR-Soave; and, (b) RK-Twu and RK-Soave. Reported Tw values were collected from Velasco et al.40 Saturated Liquid Speed of Sound Figure 10 shows the different contributions to the saturated liquid speed of sound of n-pentane calculated from the untranslated PR-Twu, PR-Soave, RK-Twu, and RK-Soave models, as described in eq 34, Section 3.2.2. As seen in Figure 10, the most important contributions are those arising from the pressure derivative, −(∂P∂v)T, and from the squared saturated liquid volume term, v2/MW. At reduced temperatures below 0.8, the contribution arising from the Cp/Cv term has a value close to zero and therefore has a negligible contribution to the calculated speed of sound. The calculated –(∂P∂v)T decreases rapidly with the reduced temperature whereas the value of v2/MW slightly increases with the reduced temperature. Comparing −(∂P∂v)T and the squared speed of sound, u2, versus temperature, it is clear that both trends are very similar in shape suggesting that the saturated liquid speed of sound is qualitatively controlled by −(∂P∂v)T with v2/MW acting as a correction factor that shifts the magnitude of −(∂P∂v)T to lower values. Therefore, the consistent and accurate calculation of the saturated liquid speed of sound is highly dependent on the accurate calculation of −(∂P∂v)T and the saturated liquid molar volume. Calculated contributions to the saturated liquid speed of sound of n-pentane according to eq 34. Figure 11 shows experimental and calculated −(∂P∂v)T and untranslated saturated liquid speed of sound for n-pentane (nonpolar, Group 1, Panel a) and methanol (hydrogen-bonding, Group 3, Panel b) from Test Data set 2. The experimental −(∂P∂v)T used here was calculated using the experimental saturated liquid isothermal compressibility coefficient and density from the NIST Database.26 As discussed in Section 3.2.2, −(∂P∂v)T is not affected by volume translation but the saturated liquid speed of sound is affected by volume translation. In the case of n-pentane, it was found that the four models tested (PR-Twu, PR-Soave, RK-Twu, and RK-Soave) underpredict −(∂P∂v)T; and, for methanol, −(∂P∂v)T was overpredicted, Figure 11. Despite the calculated −(∂P∂v)T trend for both components is qualitatively correct, a close examination of the calculated saturated liquid speed of sound (Figure 11) indicates some inconsistencies. At reduced temperatures below 0.9, the experimental saturated liquid speed of sound decreases almost linearly with the temperature whereas the calculated values decrease exponentially with the temperature. This result, and the analysis of Figure 10, indicates that the calculated −(∂P∂v)T decreases too rapidly with the temperature producing an inconsistent trend for the calculated saturated liquid speed of sound as seen in Figure 11 for n-pentane and methanol. In general, for all the components in the Test Data set 2, the four models tested (PR-Twu, PR-Soave, RK-Twu, and RK-Soave) produced one of the following results for the saturated liquid speed of sound: (1) a trend that crossed the experimental data, as seen for n-pentane in Figure 11a; or, (2) overprediction (or underprediction) as seen for methanol in Figure 11b. The reason behind these results are the inaccuracies in the calculated −(∂P∂v)T that, as previously described, is qualitatively controlling the saturated liquid speed of sound trend. Experimental and calculated −(∂P∂v)T and saturated liquid speed of sound for (a) n-pentane and (b) methanol from the Test Data set 2. Untranslated models were used in the calculation of the saturated liquid speed of sound. Figure 12 shows the calculated (untranslated) and experimental saturated liquid volume for n-pentane and methanol to highlight the effect of the volume on the speed of sound. As expected, PR-Twu and PR-Soave produced more accurate saturated liquid volumes compared to those from RK-Twu and RK-Soave. Figure 12 also shows that, for the same CEoS, the α-function does not have a significant effect on the accuracy of the calculated volume. In general, a good agreement between calculated and experimental volumes does not ensure an accurate calculation of the saturated liquid speed of sound. For instance, although the PR-calculated untranslated volumes are more accurate than those from RK, the saturated liquid speed of sound calculated from PR is not significantly more accurate than those from RK as seen in Figure 11. This observation supports the idea that the saturated liquid speed of sound of pure components is controlled by −(∂P∂v)T; which is not correctly predicted by PR and RK CEoS. Experimental and calculated saturated liquid volumes of (a) n-pentane and (b) methanol. Both components are included in Test Data set 2. Table 4 shows the summary of the overall AARD for the saturated liquid speed of sound for the different chemical families in Test Data set 2. These AARDs correspond to the speed of sound calculated using untranslated (AARD u) and translated (AARD u-t) models as described in Section 3.2.2. In the case of the untranslated models, there are no significant AARD differences among them. Therefore, the cubic equation of state (PR or RK) or the α-function (Twu’s or Soave’s) used does not have a significant effect on the accuracy of the calculated saturated liquid speed of sound. A similar conclusion is drawn when comparing the AARD of the translated models, Table 4. When comparing untranslated and translated models, lower AARDs were obtained from untranslated models indicating there is no advantage in using volume translation for the calculation of the saturated liquid speed of sound. Table 5 shows the summary of the overall AARD for the saturated liquid density of the chemical families in Test Data set 2 calculated from untranslated (AARD ρ) and translated models (AARDρ-t). Using volume translation significantly reduces the AARD of calculated liquid densities; however, more accurate liquid densities do not ensure a more accurate saturated liquid speed of sound. Table 4Summary of the Overall Average Absolute Relative Deviation (AARD) for Speed of Sound of the Chemical Families in the Test Dataset 2afamilyAARD u, %AARD u-t, %PR-TwuPR-SoaveRK-TwuRK-SoavePR-TwuPR-SoaveRK-TwuRK-Soavegroup 1: nonpolar componentsn-alkanes12.512.313.413.314.314.112.212.4branched12.312.216.616.213.713.75.75.3cyclics14.013.417.716.912.212.07.27.7aromatics15.914.617.916.614.514.310.210.1group 2: polar componentsHSCCb18.324.012.717.232.839.543.652.7pyridines4.77.47.97.77.612.010.215.6ketones7.69.510.210.18.411.014.118.1nitriles11.910.19.47.424.024.532.333.0group 3: hydrogen bonding componentsaminesc12.912.814.514.820.318.914.613.1alcohol23.922.220.819.728.226.733.932.1H2O/NH333.131.432.529.750.248.262.559.2acids48.946.938.638.258.356.867.667.6overall17.617.617.317.023.724.326.227.2au and u-t refer to the saturated liquid speed of sound calculated from the untranslated and translated models, respectively. Alkanes, branched and cyclics refer to normal, branched and cyclic alkanes, respectively.bHSCC: Heterocyclic Sulfur Containing Compound.cHNCC: Hydrogen-Bonding Heterocyclic Nitrogen-Containing Compound. Summary of the Overall Average Absolute Relative Deviation (AARD) for Speed of Sound of the Chemical Families in the Test Dataset 2a u and u-t refer to the saturated liquid speed of sound calculated from the untranslated and translated models, respectively. Alkanes, branched and cyclics refer to normal, branched and cyclic alkanes, respectively. HSCC: Heterocyclic Sulfur Containing Compound. HNCC: Hydrogen-Bonding Heterocyclic Nitrogen-Containing Compound. Table 5Summary of the Overall Average Absolute Relative Deviation (AARD) for the Saturated Liquid Density of the Chemical Families in the Test Dataset 2afamilyAARD ρ, %AARD ρ-t, %PR-TwuPR-SoaveRK-TwuRK-SoavePR-TwuPR-SoaveRK-TwuRK-Soavegroup 1: nonpolar componentsn-alkanes8.58.516.216.13.03.04.84.9branched4.34.313.113.13.63.66.16.1cyclics8.48.315.515.42.22.56.65.8aromatics6.46.316.216.03.83.86.76.7group 2: polar componentsHSCCb9.08.916.516.37.27.110.510.4pyridines4.64.114.113.52.71.56.95.8ketones12.511.821.821.17.77.511.29.9nitriles9.48.919.418.93.23.57.66.8group 3: hydrogen bonding componentsaminesc5.85.913.814.12.72.75.85.9alcohol6.06.316.616.93.94.05.55.5H2O/NH314.514.424.324.25.65.79.49.7acids11.211.321.221.21.82.13.42.9overall8.98.817.817.63.83.87.06.7aρ and ρ-t refer to the saturated liquid density calculated from the untranslated and translated models, respectively. Alkanes, branched and cyclics refer to normal, branched and cyclic alkanes.bHSCC: Heterocyclic Sulfur Containing Compound.cHNCC: Hydrogen-Bonding Heterocyclic Nitrogen-Containing Compound. Summary of the Overall Average Absolute Relative Deviation (AARD) for the Saturated Liquid Density of the Chemical Families in the Test Dataset 2a ρ and ρ-t refer to the saturated liquid density calculated from the untranslated and translated models, respectively. Alkanes, branched and cyclics refer to normal, branched and cyclic alkanes. HSCC: Heterocyclic Sulfur Containing Compound. HNCC: Hydrogen-Bonding Heterocyclic Nitrogen-Containing Compound. Conclusions Four cubic equation of state (CEoS) models were tested in this study for the calculation of the saturation pressure (Psat), enthalpy of vaporization (ΔHvap) and saturated liquid heat capacity (Cpsat) of pure components at subcritical conditions. Additionally, the thermodynamic consistency of the models was assessed by calculating the Waring number and the saturated liquid speed of sound of pure components. The four models tested were: PR-Twu, PR-Soave, RK-Twu, and RK-Soave. The first term in the model’s name refers to the CEoS used: Peng–Robinson (PR) or Redlich–Kwong (RK). The second term indicates the α-function used. The models were tested in a Development Data set containing the Psat, ΔHvap, and Cpsat of 147 pure components classified into three groups: Group 1, nonpolar components (50 components); Group 2, polar components (61 components); and Group 3, hydrogen-bonding components (36 components). For PR-Twu and RK-Twu, the three component-specific parameters of the Twu α-function, eq 7, were calculated by fitting the models to the data in the Development Data set. The calculated Twu α-function parameters are applicable across the entire phase diagram as they were constrained to the thermodynamic consistency requirements proposed by Le Guennec et al.15 (eqs 23–26). For PR-Soave and RK-Soave, The component-specific parameter of the Soave α-function, eq 4, was predicted using the correlations proposed by Piña-Martinez et al.,8eqs 5 and 6. The thermodynamic consistency of the four models was assessed by calculating the Waring number of 147 pure components (Test Data set 1) and the saturated liquid speed of sound of 79 pure components (Test Data set 2). Details about the data sets are presented in Section 3. The conclusions from this study are summarized below: Conclusions Regarding the Fitting Capabilities of the Models For the four models tested, the highest and the lowest deviations between experimental and fitted Psat, ΔHvap, and Cpsat were found for the hydrogen bonding (Group 3) and the nonpolar components (Group 1), respectively.PR-Twu and RK-Twu fitted Psat, ΔHvap, and Cpsat with lower deviations compared to those from PR-Soave and RK-Soave.There was not a significant difference in deviations between PR-Twu and RK-Twu when fitting Psat, ΔHvap, and Cpsat. The same result was found for PR-Soave and RK-Soave.Using the Twu α-function produced significantly lower deviations when fitting pure components Psat, ΔHvap, and Cpsat compared to those from the Soave α-function, regardless of the CEoS used.Although the Psat, ΔHvap, and Cpsat of the components in the Development Data set were used in the fitting of PR-Twu and RK-Twu (Section 2.2), it was found that including or not including ΔHvap in the fitting does not have a significant effect on the accuracy of the two models. For the four models tested, the highest and the lowest deviations between experimental and fitted Psat, ΔHvap, and Cpsat were found for the hydrogen bonding (Group 3) and the nonpolar components (Group 1), respectively. For the four models tested, the highest and the lowest deviations between experimental and fitted Psat, ΔHvap, and Cpsat were found for the hydrogen bonding (Group 3) and the nonpolar components (Group 1), respectively. PR-Twu and RK-Twu fitted Psat, ΔHvap, and Cpsat with lower deviations compared to those from PR-Soave and RK-Soave. PR-Twu and RK-Twu fitted Psat, ΔHvap, and Cpsat with lower deviations compared to those from PR-Soave and RK-Soave. There was not a significant difference in deviations between PR-Twu and RK-Twu when fitting Psat, ΔHvap, and Cpsat. The same result was found for PR-Soave and RK-Soave. There was not a significant difference in deviations between PR-Twu and RK-Twu when fitting Psat, ΔHvap, and Cpsat. The same result was found for PR-Soave and RK-Soave. Using the Twu α-function produced significantly lower deviations when fitting pure components Psat, ΔHvap, and Cpsat compared to those from the Soave α-function, regardless of the CEoS used. Using the Twu α-function produced significantly lower deviations when fitting pure components Psat, ΔHvap, and Cpsat compared to those from the Soave α-function, regardless of the CEoS used. Although the Psat, ΔHvap, and Cpsat of the components in the Development Data set were used in the fitting of PR-Twu and RK-Twu (Section 2.2), it was found that including or not including ΔHvap in the fitting does not have a significant effect on the accuracy of the two models. Although the Psat, ΔHvap, and Cpsat of the components in the Development Data set were used in the fitting of PR-Twu and RK-Twu (Section 2.2), it was found that including or not including ΔHvap in the fitting does not have a significant effect on the accuracy of the two models. Conclusions Regarding the Calculated Psat, ΔHvap, and Cpsat Trends The four models tested produced Psat, ΔHvap, and Cpsat trends qualitatively consistent with the experimental data for the nonpolar (Group 1) and polar (Group 2) components in the Development Data set.For the hydrogen-bonding components in the Development Data set (Group 3), the Psat and ΔHvap trends were qualitatively consistent but those for Cpsat were not.None of the models was able to capture the inflection point exhibited by the Cpsat curve of alcohols with carbon number higher than four (Figure 5).For the hydrogen-bonding components in the Development Data set (Group 3), PR-Twu and RK-Twu produced a minimum in the calculated Cpsat curve which was not observed in the experimental data (Figure 6). The four models tested produced Psat, ΔHvap, and Cpsat trends qualitatively consistent with the experimental data for the nonpolar (Group 1) and polar (Group 2) components in the Development Data set. The four models tested produced Psat, ΔHvap, and Cpsat trends qualitatively consistent with the experimental data for the nonpolar (Group 1) and polar (Group 2) components in the Development Data set. For the hydrogen-bonding components in the Development Data set (Group 3), the Psat and ΔHvap trends were qualitatively consistent but those for Cpsat were not. For the hydrogen-bonding components in the Development Data set (Group 3), the Psat and ΔHvap trends were qualitatively consistent but those for Cpsat were not. None of the models was able to capture the inflection point exhibited by the Cpsat curve of alcohols with carbon number higher than four (Figure 5). None of the models was able to capture the inflection point exhibited by the Cpsat curve of alcohols with carbon number higher than four (Figure 5). For the hydrogen-bonding components in the Development Data set (Group 3), PR-Twu and RK-Twu produced a minimum in the calculated Cpsat curve which was not observed in the experimental data (Figure 6). For the hydrogen-bonding components in the Development Data set (Group 3), PR-Twu and RK-Twu produced a minimum in the calculated Cpsat curve which was not observed in the experimental data (Figure 6). Conclusions Regarding the Thermodynamic Consistency Analysis PR-Twu and PR-Soave produced reduced Waring temperatures (TWr = Tw/Tc) that are in good agreement with those reported by Velasco et al.40 when the component’s acentric factor is below 0.7. This result suggests that PR-Twu and PR-Soave produce consistent vapor pressures for components with acentric factors below 0.7 (Figure 9).RK-Twu and RK-Soave produced reduced Waring temperatures higher than those reported by Velasco et al.40 This result may indicate that RK-Twu and RK-Soave produce inconsistent vapor pressures (Figure 9).The consistency analysis of the four models based on the saturated liquid speed of sound showed that the accuracy of the calculated saturated liquid speed of sound is highly dependent on the magnitude of −(∂P∂v)T and the squared molar volume (eqs 31 and 34).None of the models tested produced an accurate saturated liquid speed of sound because they were unable to produce correct −(∂P∂v)T values.Using a temperature-independent volume translation expression (eq 32) does not improve the accuracy of the saturated liquid speed of sound calculated from the four models tested.The deviations between the experimental and calculated saturated liquid speed of sound were not significantly different among the models (Table 4). This result indicates that the four models are equally deficient for the saturated liquid speed of sound. PR-Twu and PR-Soave produced reduced Waring temperatures (TWr = Tw/Tc) that are in good agreement with those reported by Velasco et al.40 when the component’s acentric factor is below 0.7. This result suggests that PR-Twu and PR-Soave produce consistent vapor pressures for components with acentric factors below 0.7 (Figure 9). PR-Twu and PR-Soave produced reduced Waring temperatures (TWr = Tw/Tc) that are in good agreement with those reported by Velasco et al.40 when the component’s acentric factor is below 0.7. This result suggests that PR-Twu and PR-Soave produce consistent vapor pressures for components with acentric factors below 0.7 (Figure 9). RK-Twu and RK-Soave produced reduced Waring temperatures higher than those reported by Velasco et al.40 This result may indicate that RK-Twu and RK-Soave produce inconsistent vapor pressures (Figure 9). RK-Twu and RK-Soave produced reduced Waring temperatures higher than those reported by Velasco et al.40 This result may indicate that RK-Twu and RK-Soave produce inconsistent vapor pressures (Figure 9). The consistency analysis of the four models based on the saturated liquid speed of sound showed that the accuracy of the calculated saturated liquid speed of sound is highly dependent on the magnitude of −(∂P∂v)T and the squared molar volume (eqs 31 and 34). The consistency analysis of the four models based on the saturated liquid speed of sound showed that the accuracy of the calculated saturated liquid speed of sound is highly dependent on the magnitude of −(∂P∂v)T and the squared molar volume (eqs 31 and 34). None of the models tested produced an accurate saturated liquid speed of sound because they were unable to produce correct −(∂P∂v)T values. None of the models tested produced an accurate saturated liquid speed of sound because they were unable to produce correct −(∂P∂v)T values. Using a temperature-independent volume translation expression (eq 32) does not improve the accuracy of the saturated liquid speed of sound calculated from the four models tested. Using a temperature-independent volume translation expression (eq 32) does not improve the accuracy of the saturated liquid speed of sound calculated from the four models tested. The deviations between the experimental and calculated saturated liquid speed of sound were not significantly different among the models (Table 4). This result indicates that the four models are equally deficient for the saturated liquid speed of sound. The deviations between the experimental and calculated saturated liquid speed of sound were not significantly different among the models (Table 4). This result indicates that the four models are equally deficient for the saturated liquid speed of sound. Based on the results of this study, the following is recommended for modeling the phase behavior of pure components using a CEoS:Use PR-Twu because of its higher accuracy and consistency.Fit the Twu α-function against Psat and Cpsat data.Exercise caution when modeling Cpsat of hydrogen-bonding components. For alcohols, the parameters reported here are recommended for reduced temperatures below 0.5.Use a temperature-independent volume translation expression as that shown in eq 32 to produce accurate densities (see Table 5). Keep in mind that using a temperature-independent volume translation expression does not improve the accuracy of the calculated saturated liquid speed of sound as CEoS, regardless of the α-function, do not produce accurate −(∂P∂v)T values.A temperature-dependent volume translation expression should be used if accurate saturated liquid speed of sound is required. However, caution must be exercised as these expressions introduce thermodynamic inconsistencies into CEoS as described elsewhere.7 Use PR-Twu because of its higher accuracy and consistency. Use PR-Twu because of its higher accuracy and consistency. Fit the Twu α-function against Psat and Cpsat data. Fit the Twu α-function against Psat and Cpsat data. Exercise caution when modeling Cpsat of hydrogen-bonding components. For alcohols, the parameters reported here are recommended for reduced temperatures below 0.5. Exercise caution when modeling Cpsat of hydrogen-bonding components. For alcohols, the parameters reported here are recommended for reduced temperatures below 0.5. Use a temperature-independent volume translation expression as that shown in eq 32 to produce accurate densities (see Table 5). Keep in mind that using a temperature-independent volume translation expression does not improve the accuracy of the calculated saturated liquid speed of sound as CEoS, regardless of the α-function, do not produce accurate −(∂P∂v)T values. Use a temperature-independent volume translation expression as that shown in eq 32 to produce accurate densities (see Table 5). Keep in mind that using a temperature-independent volume translation expression does not improve the accuracy of the calculated saturated liquid speed of sound as CEoS, regardless of the α-function, do not produce accurate −(∂P∂v)T values. A temperature-dependent volume translation expression should be used if accurate saturated liquid speed of sound is required. However, caution must be exercised as these expressions introduce thermodynamic inconsistencies into CEoS as described elsewhere.7 A temperature-dependent volume translation expression should be used if accurate saturated liquid speed of sound is required. However, caution must be exercised as these expressions introduce thermodynamic inconsistencies into CEoS as described elsewhere.7