The Eighth Industrial Fluids Properties Simulation Challenge

Abstract

The goal of the eighth industrial fluid properties simulation challenge was to test the ability of molecular simulation methods to predict the adsorption of organic adsorbates in activated carbon materials. In particular, the eighth challenge focused on the adsorption of perfluorohexane in the activated carbon BAM-P109. Entrants were challenged to predict the adsorption in the carbon at 273 K and relative pressures of 0.1, 0.3, and 0.6. The predictions were judged by comparison with a benchmark set of experimentally determined values. Overall, good agreement and consistency were found between the predictions of most entrants.

Introduction

Background and motivation

As with all previous challenges (Case et al., 2004, 2005, 2007, 2008, 2009, 2011; Industrial Fluid Properties Simulation Collective, 2014; Ross et al., 2014), the eighth industrial fluid properties simulation challenge (IFPSC) is focused on prediction of properties that are both academically challenging and industrially relevant. The eighth challenge is an extension of the seventh challenge (Ross et al., 2014), which focused on adsorption in zeolite materials. The eighth challenge continues this work by focusing on adsorption in carbon materials. In both challenges, the adsorbate was perfluorohexane. The seventh challenge was relevant from an industrial standpoint because of an increasing number of societal needs for removal of volatile or toxic chemicals from air and water. The ability to predict the performance of adsorbent materials for these applications would aid in materials development and ultimately reduce the number of product development cycles. The eighth challenge is industrially important for the same reasons as the seventh, but it has the added difficulty of a less well-defined structure. The lack of a well-defined structure is a common difficulty for simulation, but it is particularly acute for industrial research. Although there have been many publications reporting simulated adsorption results for zeolites (Fuchs and Cheetham, 2001; Smit and Maesen, 2008), metal-organic frameworks (Düren et al., 2009; Keskin et al., 2008), nanotubes (Shi and Johnson, 2003), and porous carbons (Davies and Seaton, 2000; Jain et al., 2006; Sweatman and Quirke, 2005) organizing a challenge for a less well-characterized carbon will help to address a gap in adsorption research.

The use of molecular simulation has benefits beyond quantitative predictions. Qualitatively, molecular simulation methods can be used to provide atom-level insights into a wide variety of phenomena, including transport and adsorption mechanisms in porous solids. Many of these mechanisms cannot be directly determined via experimental efforts. Thus, simulated results can be used to supplement experimental results and can be useful in distinguishing between contradictory experimental measurements (Chialvo et al., 1998; McCabe et al., 2001; Siepmann et al., 1993).

History of the IFPSC

The first industrial fluids simulation challenge resulted from discussions held at the workshop on predicting “Thermophysical Properties of Fluids by Molecular Simulation” in May 2001. The workshop, held at the National Institute of Standards and Technology (NIST), was attended by scientists from a variety of companies, universities, and national laboratories.

The first challenge, held in 2002, focused on predicting viscosity, density, and vapor–liquid equilibria (Case et al., 2004; Friend et al., 2004; Klamt and Eckert, 2004; Martin and Thompson, 2004; Morrow and Maginn, 2004; Pollock et al., 2004; Rigby, 2004; Sun, 2004; Zhang and Ely, 2004). Challenge entrants employed molecular simulation to predict a Px curve for a mixture of dimethyl ether and propylene. In addition, entrants were asked to predict the composition and pressure of an azeotropic point for a mixture of nitroethane and propylene glycol monomethyl ether.

The second challenge, held in 2004, focused on problems involving heats of mixing, Henry’s law constants, heats of vaporization, and vapor pressure (Boutard et al., 2005; Case et al., 2005; Cichowski et al., 2005; Dai et al., 2005; Friend et al., 2005; Martin and Biddy, 2005; Schnabel et al., 2005, 2006; Zhang and Siepmann, 2006). Challenge entrants employed molecular simulation to predict heats of vaporization and vapor pressures for acetone and butyramide; heats of mixing for n-butylamine and n-heptane; and Henry’s law constants for N₂, CO₂, CH₄, and O₂ in ethanol. Predicting heats of mixing proved to be especially challenging for alkylamine/water systems.

The third challenge, held in 2006, focused on assessing the transferability of molecular simulation methods and force fields (Case et al., 2007; Christensen et al., 2007; Eckl et al., 2007; Gordon et al., 2007; Kelkar et al., 2007; Klamt and Eckert, 2007; Kleiner and Sadowski, 2007; Rai et al., 2007; Yazaydın and Martin, 2007; Zhao et al., 2007). Entrants were challenged to predict bubble point pressures for mixtures of 1,1,1,2,3,3,3-heptafluoropropane and ethanol at 343 K, given data at 283 K. In addition, entrants were asked to predict shear viscosities across a range of diols and triols as well.

The fourth challenge, held in 2007, focused on testing the transferability of molecular simulation methods and force fields when applied to the prediction of a wide variety of physical properties for ethylene oxide, an industrial relevant molecule (Case et al., 2008; Eckl et al., 2008; Ketko et al., 2008; Li et al., 2008; Mountain, 2008; Müller et al., 2008; Olson and Wilson, 2008). The fifth challenge was held in 2008. This challenge focused on predicting infinite-dilution activity coefficients for 1-ethylpropylamine and 3-methyl-1-pentanol and predicting octanol–water partition coefficients (Case et al., 2009; Hsieh and Lin, 2009; Klamt et al., 2009; Liu et al., 2009; Olson et al., 2009). The sixth challenge, held in 2010, focused on prediction of the mutual solubilities of a glycol ether (PROGLYDE DMM) and water in liquid–liquid equilibria as a function of temperature (Case et al., 2011; Bai and Siepmann, 2011; Donate et al., 2011; Köddermann et al., 2011; Lin et al., 2011; Reinisch et al., 2011; Zhao et al., 2011). The seventh challenge, held in 2012, focused on adsorption of the moderately complex chemical, perfluorohexane, in a zeolite (Bai et al., 2014; Demir and Ahunbay 2014; Ross et al., 2014; Wang et al., 2014; Yang et al., 2014).

Challenge

The eighth challenge was the prediction of the adsorption isotherm of n-perfluorohexane (n-C₆F₁₄) on the Certified Reference Material BAM-P109 carbon (Panne and Thünemann, 2010) at a temperature of 273 K and at relative pressures of 0.1, 0.3, and 0.6. The relative pressure is defined as that relative to the bulk saturation pressure predicted by the model for the given temperature.

At relative pressures of 0.1 and higher, the micropores of BAM-P109 are expected to be filled with adsorbate. An accurate simulation model for adsorption in the carbon micropores will be necessary for successful predictions at a relative pressure of 0.1. Evaluating this capability was the primary goal for this challenge. Additional adsorption over the range of 0.1 to 0.6 relative pressure (i.e. in the pseudo-plateau region of the adsorption isotherm) is expected to occur in the mesoporous regions of BAM-P109. Molecular simulation methods used to model adsorption in the micropores may not be directly applicable to predict adsorption in the mesopore regions. However, challenge entrants were encouraged to develop additional approaches to account for mesoporous adsorption (for example, via correlations to the provided experimental benchmark adsorption data) as a secondary aspect of this challenge.

Challenge champions along with second and third places were awarded. Entries were scored by comparing the predicted adsorption isotherm data with experimentally measured data. For adsorption isotherms, the amount of perfluorohexane experimentally adsorbed at the defined set of pressures was compared with the amounts predicted at the same pressures.

The units for comparing between experiment and simulation were chose to be (cm³ g⁻¹) at STP. The scoring was computed as a mean percent error, M%E, between the computed value and the experimental value at the specified relative pressures. The M%E was computed for n points using,

$\text{M}\%\text{E}=\sum _{\text{i}=1}^{\text{n}}|V_{\mathrm{ab},\exp }-V_{\mathrm{ab},\mathrm{sim}}|V_{\mathrm{ab},\exp }$

where V _{ab ,exp} and V_ab,sim are the experimental and simulated volumes of absorbed gas, respectively.

The M%E were weighted such that the prediction of adsorption at 0.1 relative pressure accounted for 60% of the score, whereas the predictions for adsorption at 0.3 and 0.6 relative pressures accounted for 20% of the score each. The weighting reflects the adsorption filling of the microporous region below 0.1 relative pressure followed by further adsorption in the mesoporous region above 0.1 relative pressure.

Results and discussion

Five entries were received for the eighth challenge as shown in Table 1. The predicted isotherms along with the experimental adsorption isotherm for n-perfluorohexane in BAM-P109 activated carbon are plotted in Figure 1. As can be seen in the figure, all of the entrants predict adsorption in the correct order of magnitude. For four out of five entrants, the maximum deviation from experiment at 0.1 relative pressure is within ≈25 cm³/g.

Figure 1.

Predicted and experimental adsorption isotherms of n-perfluorohexane (n-C₆F₁₄) on the Certified Reference Material BAM-P109 Carbon at a temperature of 273 K.

Table 1.

Entrants and their affiliations.

Team members	Affiliation
Herdes, Forte, Jackson, and Müller	Imperial College London
Gonciaruk, Runcieman, Avendaño-Jiménez, and Siperstein	University of Manchester
Demir and Ahunbay	Istanbul Technical University
Cao, Jing, and Sun	Shanghai Jiao Tong University
Sarkisov	University of Edinburgh

Mean percent errors (M%E) for the molecular simulation predictions are shown in Figure 2. For four out of five entrant groups, the M%E is within ∼25%. The team of Demir and Ahunbay from Istanbul Technical University were the winners of the challenge with an M%E of less than 15%. The teams of Sarkisov (University of Edinburgh) and Cao, Jing, and Sun (Shanghai Jiao Ting University) won the second and third places.

Figure 2.

The M%E for the five entries.

All of the entrants employed the grand canonical Monte Carlo simulation method. The team of Cao, Jing, and Sun used a combination of the grand canonical and Gibbs ensemble Monte Carlo simulation methods. A variety of force fields were employed ranging from newly developed to those available in the literature.

A significant difference between the entrants was in how they chose to represent the microporous carbon region. The groups of Demir and Ahunbay, Sarkisov, and Gonciaruk represented the microporous region as a collection of discrete molecules whose packing geometries were optimized against experimental data. The group of Gonciaruk also employed a slit-pore model for comparison. The groups of Ca and Herdes represented the microporous region as a distribution of slit-pore models.

An additional difference between the entrant groups was in the use of the provided benchmark XPS data and levels of oxygenation in their models. The groups of Dimer and Ahunbay and Sarkisov, which had the lowest mean percent errors incorporated oxygen into their models based on the provide benchmark experimental XPS data. The remaining three entrant groups did not incorporate oxygen into their models.

Additional details for the individual simulation entries can be found in the complimentary individual articles published by each entrant group in this special journal section.

Representatives of two of the five entrant teams were present at a special session of the American Institute of Chemical Engineers (AIChE) 2014 Annual Meeting in Atlanta, GA. Figure 3 shows a photo of Goktug Ahunbay of the team of Demir and Ahunbay accepting the first place award from Daniel Siderius of the National Institute of Standards and Technology. The session was sponsored by the Computational Molecular Science and Engineering Forum (CoMSEF: http://comsef.org.) of the American Institute of Chemical Engineers. At the session, the entrants presented their predictions and discussed various aspects of their methods as well as challenge problems.

Figure 3.

M. Goktug Ahunbay accepts the award for first place for the eighth industrial fluid properties simulation challenge from Daniel Siderius of NIST.