Naroozi

Subscriber access provided by UNIV AUTONOMA DE COAHUILA UADEC
A: New Tools and Methods in Experiment and Theory

An Efficient Molecular Simulation Methodology for Chemical Reaction
Equilibria In Electrolyte Solutions: Application to CO Reactive Absorption
2
Javad Noroozi, and William Robert Smith

J. Phys. Chem. A, Just Accepted Manuscript • DOI: 10.1021/acs.jpca.9b00302 • Publication Date (Web): 05 Apr 2019
Downloaded from http://pubs.acs.org on April 7, 2019
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a service to the research community to expedite the dissemination
of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in
full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully
peer reviewed, but should not be considered the official version of record. They are citable by the
Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore,
the “Just Accepted” Web site may not include all articles that will be published in the journal. After
a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web
site and published as an ASAP article. Note that technical editing may introduce minor changes
to the manuscript text and/or graphics which could affect content, and all legal disclaimers and
ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or
consequences arising from the use of information contained in these “Just Accepted” manuscripts.
is published by the American Chemical Society. 1155 Sixteenth Street N.W.,

Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.
Page 1 of 44 The Journal of Physical Chemistry
1
2
3
4
5
6
7
8 An Efficient Molecular Simulation Methodology
9
10
11
12
for Chemical Reaction Equilibria In Electrolyte
13
14
15 Solutions: Application to CO2 Reactive
16
17
18 Absorption
19
20
21
22 Javad Noroozi† and William R. Smith∗,‡,¶,§
23
24
25 †Dept. of Chemical Engineering, Un. of Waterloo, Waterloo ON, N2L 3G1, Canada
26
27 ‡Department of Mathematics and Statistics, University of Guelph, Guelph ON N1G 2W1,
28
29 Canada
30
31 ¶Department of Chemical Engineering, University of Waterloo, Waterloo ON N2L 3G1,
32
33 Canada
34
35 §Faculty of Science, University of Ontario Institute of Technology, Oshawa ON L1H 7K4,
36
37 Canada
38
39
40 E-mail: bilsmith@uoguelph.ca
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 1
60 ACS Paragon Plus Environment
The Journal of Physical Chemistry Page 2 of 44
1
2
3
Abstract
4
5
6 We develop a computationally efficient molecular–based simulation algorithm for
7
8 chemical reaction equilibria in liquids containing neutral and ionic species, which is
9
10 based on the combination of classical force field and ab initio calculations, and per-
11
mits calculations involving very small species concentrations. We show its application
12
13 to the reactive absorption of CO2 in aqueous monoethanolamine (MEA) solvent as
14
15 a benchmark case, the first time that a quantitatively accurate predictive approach
16
17 requiring no experimental data has been successfully applied to calculate all solution
18
19 species concentrations for this system, including the partial pressure of CO2 above the
20
21 solution.
22
23 The Reaction Ensemble Monte Carlo algorithm, the only other generally applica-
24
25
ble approach, requires special system–dependent Monte Carlo enhancements for its
26 implementation, and to detect species with very small concentrations requires long
27
28 simulation times and/or large system sizes. In contrast, the proposed algorithm can be
29
30 straightforwardly implemented for systems of any molecular complexity using a stan-
31
32 dard Molecular Dynamics (MD) simulation package capable of calculating free energy
33
34 changes, and can calculate small species concentrations with normal simulation times
35
36 and system sizes. In addition, the inherent parallelization capability of MD (which is
37
38 problematic for MC–based approaches) enables the algorithm’s computationally effi-
39
cient implementation.
40
41 The H2 O–MEA–CO2 benchmark system has been the subject of many previous
42
43 studies based on macroscopic thermodynamic modeling, which primarily involve fitting
44
45 their parameters (of which reaction pK values are the most important) to experimental
46
47 data measurements. To make contact with such approaches, we show the translation
48
49 of the molecular–based quantities to the direct prediction of these parameters, and cal-
50
51 culate reaction equilibrium in the framework of a Henry–Law–based chemical potential
52
53
model. We consider both the ideal solution form and its extension using the Davies
54
equation for the species activity coefficients. We study a range of temperatures and
55
56 CO2 solution loadings in a 30 weight % MEA solution, and incorporate an Uncertainty
57
58
59 2
1
2
3
Analysis in our methodology. We find that the uncertainties of the simulated solution
4
5 species compositions are comparable to those of the available experimental data. We
6
7 report predictions of minor species compositions of very small magnitude, for which
8
9 experimental measurements are typically extremely challenging.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 3
1
2
3
4
1 Introduction
5
6 Calculation of the equilibrium composition of chemically reacting systems is of great interest
7
8 and importance in chemistry, biology and chemical engineering. Systems involving ionic
9
10 species (especially aqueous electrolyte systems) are of particular significance, and have long
11
12 been studied using macroscopic thermodynamic tools. 1,2
13
14 Carbon Capture and Storage (CCS) methodology applied to fixed CO2 sources is cur-
15
16 rently considered to be one of the most advanced programs to reduce greenhouse gas emis-
17
18 sions, and the development of more economically effective technologies is crucial to encourag-
19
20 ing its future deployment (for reviews, see, e.g., Kumoro et al. 3 and Song et al. 4 ). Strategies
21
22 based on CO2 capture from point sources using reactive absorption in solvents is considered
23
24 to be a viable option, and aqueous solutions of monoethanolamine (MEA, also referred to
25
26 herein as RNH2 ), where R = (CH2 )2 OH) were one of the first solvents used for this purpose.
27
28 Its drawbacks include the large energy cost of regeneration from the CO2 –loaded, and this
29
30 has spurred the search for improved solvents.
31
32 The search for such solvents has involved the use of chemical reaction equilibrium (CRE)
33
34 calculations based on thermodynamic models for the involved aqueous CO2 –aqueous elec-
35
36 trolyte solutions. 5–16 Model development is complicated by the fact that its parameters are
37
38 not directly measurable quantities, but must be obtained indirectly by fitting to experimen-
39
40 tal data. Thus, most CO2 reactive absorption models in the experimental literature can be
41
42 considered to be primarily correlations rather than predictions, and the resulting parameter
43
44 values show considerable variation among different research groups.
45
46 A less expensive and more fundamentally based approach to CRE modeling involves the
47
48 use of molecular–based methodology, which in principle requires little or no experimental
49
50 data and is thereby predictive in nature. Several groups have focussed attention on the
51
52 calculation of solution properties at infinite dilution (for example, reaction pK values) using
53
54 approximate quantum mechanical (QM) approaches (recent examples are the works of Gupta
55
56 et al. 17 and Taranishiet al. 18 ). Although such properties are important, these methods are
57
58
59 4
1
2
3
computationally infeasible for calculating solution properties at higher concentrations due
4
5
to their computational complexity.
6
7
Although classical force field (CFF) methodology has been used extensively for non–
8
9 reacting systems, there have been relatively few studies of its application to systems un-
10
11 dergoing chemical reactions, and fewer still involving systems containing reacting ionic
12
13 species. CRE implementation by means of the Reaction Ensemble Monte Carlo (REMC)
14
15 algorithm 19,20 and its variants (for a review, see Turner et al. 21 ) has been used for multi–
16
17 reaction plasma systems, 22,23 to a single-reaction ionic liquid system. 24 It has also been
18
19 applied in a preliminary study of CO2 solubility modeling in aqueous monoethanolamine
20
21 involving four reactions, 25 with which we make comparisons in this study. Since REMC
22
23 implementation relies on a system–specific combination of alchemical and insertion moves,
24
25 special system–dependent techniques such as Continuous Fractional Component (CFC) re-
26
27 action moves 25–28 must be used at liquid densities. However, the recent study of Mullen et
28
29 al. 24 found CFC to fail due to the large variance in insertion free energies, and they replaced
30
31 it with an alternative (system–dependent) approach. Another challenge in molecular-based
32
33 CRE simulations is calculations involving very small species amounts (e.g., concentrations
34
35 of 10−4 or less), which at first sight would seem to require infeasibly long simulation times
36
37 and/or large system sizes. 25
38
39 The primary goal of this study is to develop a new CFF-based chemical reaction equi-
40
41 librium algorithm that overcomes the indicated limitations of the REMC algorithm, and to
42
43 illustrate its implementation for an important example of the class of electrolyte systems
44
45 arising in CO2 capture. The proposed algorithm requires only ideal–gas standard reaction
46
47 free energy changes, which are available from thermochemical tables or may be calculated
48
49 using standard quantum mechanical software, and well-established solution phase free en-
50
51 ergy change methodologies implemented in readily available Molecular Dynamics software.
52
53 The recently developed ReMD algorithm of Smith and Qi, 29 based on macroscopic ther-
54
55 modynamic principles, minimizes the Gibbs energy by calculating an iterative sequence of
56
57
58
59 5
1
2
3
compositions, which are used as reference states for an ideal solution extrapolation model.
4
5
We develop a variant of it here and apply it to the aqueous MEA–CO2 system, using the
6
7
infinite dilution reference state as the initial approximation and extrapolating the chemical
8
9 potentials using the Henry Law ideality model and its extension incorporating the Davies
10
11 equation for the activity coefficients. 30 This technique can also be used to calculate the con-
12
13 centrations of species present in very small concentrations. In this paper, we implement the
14
15 ReMD algorithm using a combination of QM and MD procedures, resulting in an approach
16
17 that is entirely predictive and utilizes no experimental data.
18
19 In the course of our study, we provide uncertainty estimates of the predicted partial
20
21 pressure of CO2 above the solution by propagating the uncertainties in the underlying pa-
22
23 rameters. We remark that the importance of uncertainty analysis applied to thermodynamic
24
25 models of CO2 reactive absorption has also recently been emphasized by Morganet al. 31 and
26
27 by others. 32,33
28
29 The paper is organized as follows. The next section briefly reviews CRE calculation
30
31 methodology in the context of conventional macroscopically based Henry-Law-based chem-
32
33 ical potential models, and the subsequent section relates the relevant molecular simulation
34
35 quantities to the macroscopic parameters and outlines the CRE calculation procedures. The
36
37 following section describes the systems studied, and the relevant molecular models and sim-
38
39 ulation protocols used. The next section give our results and their discussion, followed by a
40
41 section presenting our conclusions and recommendations.
42
43
44
45
46 2 Reaction Equilibrium Using Henry-Law-Based Chem-
47
48
49 ical Potential Models
50
51
52 In Section 3, we translate our molecular simulation CRE model to the Henry–Law-based
53
54 model, which is commonly used in macroscopic theoretical and experimental studies of elec-
55
56
57
58
59 6
1
2
3
trolyte solutions and is briefly reviewed here.
4
5
6 m
i
7 µi (T, P ; m) = µ†i (T ; P ) + RT ln + RT ln γi (T, P ; m); i = 1, 2, . . . , Nsolu (1)
8 m0
9
10
11 where T is the absolute temperature, P the pressure, R the universal gas constant, µi (T, P ; m)
12
13 is the chemical potential of species i, m is the vector of molalities, γi is the species activity
14
15 coefficient, and Nsolu is the number of solute species. µ†i (T ; P ) is the standard state chemical
16
17 potential of solute species i in a hypothetical ideal solution of m0 = 1 molal. In the CO2 -H2 O
18
19 system, we take H2 O as the solvent and consider all other species to be solutes.
20
21 The chemical potential model for the solvent based on its pure component reference state
22
23 is normally used:
24
25
26 µsolv (T, P ; m) = µ∗solv (T, P ) + RT ln asolv (T, P ; m) (2)
27
28
29
30 where µ∗solv (T, P ) is the chemical potential of the pure solvent at the solution (T, P ) and asolv
31
32 is its activity in the solution.
33
34 Reaction equilibrium in a closed system at specified (T, P ) can be calculated by minimiz-
35
36 ing its Gibbs free energy subject to the conservation of mass constraints, the latter of which
37
38 may be implemented by means of a set of R linearly independent stoichiometric reactions,
39
40 where
41
42 R = Ns − rank(A) (3)
43
44
45 Ns is the total number of species (including the solvent), A is the Ns × M matrix of chemical
46
47 formulae and M is the number of chemical elements. 34 It is important to note that the
48
49 reaction set is independent of any underlying reaction mechanism, and may be selected
50
51 based on numerical convenience.
52
53
54
55
56
57
58
59 7
1
2
3
The equilibrium composition is then determined by the solution of the equations
4
5
6 Ns
X
7 ∆Gj ≡ νij µi = 0; j = 1, 2, . . . , R (4)
8 i=1
9
10
11 where νij is the stoichiometric coefficient of species i in reaction j. Substitution of Eqs. (1)
12
13 and (2) in Eq. (4) yields the final working equations:
14
15
16 ej (T, P ) NX
∆G solu m NX
i
solu
17 + νij ln 0
+ νij ln γi (T, P ; m)+νsolv,j ln asolv (T, P ; m) = 0; j = 1, 2, . . . , R
RT m
18 i=1 i=1
19 (5)
20
21 where
N
22 X solu
23 ∆G
ej (T, P ) = νij µ†i (T, P ) + νsolv,j µ∗solv (T, P ) (6)
24 i=1
25
26 The G
ej (T, P ) values are commonly expressed in terms of equilibrium constants Kj via
27
28
29 ∆G
ej (T, P )
30 ln(10)pKj (T, P ) = − (7)
RT
31
32
33 where pKj = −log10 Kj .
34
35 The solute activity coefficients γi (T, P ; m) in Eq. (1) may be modelled in various ways.
36
37 In this paper, we consider (1) the ideal Henry-Law solution model with ln γi = 0, and (2)
38
39 the empirical Davies model: 30
40
41 √ !
42 I
43 ln γi = −zi2 A(T, P ) √ − 0.3I (8)
44 1+ I
45
46
47 where zi is the valence, and A(T, P ) is related to the pure solvent dielectric permittivity and
48
49
density by 35,36
50 [ρsolv (T, P )]1/2
51
A = 0.13287 × 106 (9)
[r,solv (T, P )T )]3/2
52
53
where ρsolv (T, P ) is the density of the solvent in kg m−3 . r,solv is the solvent’s relative static
54
55
56
57
58
59 8
1
2
3
permittivity and I is the solution’s ionic strength, defined by
4
5
6 sN
1X
7 I= mi zi2 (10)
8 2 i=1
9
10
11 ln asolv in Eqs. (2) and (5) for these models is obtained by means of the Gibbs–Duhem
12
13 equation:
14
15
16 1 − xsolv
17
ln aIdeal
solv = − (11)
xsolv
18 " √ ! #
19 Msolv 1+2 I √ 1 − xsolv
20
ln aDavies
solv = A(T, P ) 2 √ − 4 ln(1 + I) − 0.3I 2 − (12)
1000 1+ I xsolv
21
22
23
24 3 Molecular-Based Calculation of pKj
25
26
27
28 3.1 Relation to Ideal–Gas and Hydration Free Energy Quantities
29
30 µ†i values for the solute species in Eq. (1) are calculated by molecular simulation using the
31
32 following expression 37–40 (see also Section 2 of the Supporting Information):
33
34
35
36 µ†i (T, P ) = µ0i (T ; P 0 ) + ∆Ghyd corr
i (T, P ) + ∆Gi (T, P ) (13)
37
38
39
40 where µ0i (T ; P 0 ) is the species ideal–gas standard chemical potential at T and the reference
41
42
state pressure P 0 = 1 bar, ∆Ghyd corr
i (T, P ) is its absolute hydration free energy, and ∆Gi is
43
44
a (typically small in magnitude) correction term that includes 39 (1) a finite–size correction
45
46
term to account for the size dependence of the simulations; (2) a correction for the mean
47
48
effect of electrostatic interactions beyond the simulation cutoff distance, depending on the
49
50
relative dielectric permittivity of the water model used in the simulation; (3) a quantum
51
52
correction term.
53
54
55
56
57
58
59 9
1
2
3
∆Ghyd
i (T, P ) is given by
4
5
6
7 ∆Ghyd std intr
i (T, P ) = ∆G (T, P ) + ∆Gi [T, ρsolv (T, P )] (14)
8
9
10
11 where ∆Gstd (T ; P 0 ) accounts for the change from the reference state pressure of P 0 = 1 bar
12
13 in the ideal–gas state to a reference state solution concentration of 1 molal, ∆Gintr
i is the
14
15 intrinsic solvation free energy of species i, and ρsolv (T, P ) is the pure solvent density at the
16
17 specified (T, P ). The first two quantities are given by
18
19
RT ρsolv (T, P )m0

20 std
∆G (T, P ) = RT ln (15)
21 105 P 0
22 res,N V T,∞
23
∆Gintr
i (T, P ) = µi [T, ρsolv (T, P )] (16)
24
25
26 where ρsolv (T, P ) is the pure solvent density in kg m−3 , P 0 is in bar, and µres,N VT
[T, ρ(T, P )]
i
27
28 is the residual chemical potential of species i at infinite dilution in the pure solvent calculated
29
30 by molecular simulation from Eq. (16).
31
32 Excluding the relatively small correction terms, µ†i for a solute may be conveniently
33
34 expressed as
35
36
37 µ†i (T, P ) µ0i (T ; P 0 ) ∆Gintr

RT ρsolv (T, P ) i (T, P )
38 = + ln + ln + (17)
39
RT RT 100P 0 1000 RT
40
41
We remark that µ†i (T, P ) values for individual ions calculated from the above expressions
42
43
are different from the “conventional” experimental values given in thermochemical tables
44
45
(e.g., Wagman et al. 41 ); however, values for charge–conserving linear combinations of ions
46
47
agree with the results using corresponding tabular values. Such linear combinations always
48
49
occur in chemical reactions involving ions, as is the case in this study. (In order to match
50
51
the conventional individual ionic values, a term involving the electrostatic Galvani potential
52
53
at the vacuum/water interface, and a term involving the intrinsic hydration free energy of
54
55 the proton must be included; see, e.g., Majer et al. 42 )
56
57
58
59 10
1
2
3
The pure solvent chemical potential is given by
4
5
6 µ∗solv (T, P ) µ†solv (T, P )

1000

7 = + ln (18)
8 RT RT Msolv
9
10
11 where
12
13
µ†solv (T, P ) µ0solv (T ; P 0 ) ∆Gintr

14 RT ρsolv (T, P ) solv (T, P )
= + ln + ln + (19)
15 RT RT 100P 0 1000 RT
16 res,N V T
17 ∆Gintr
solv (T, P ) = µsolv [T, ρsolv (T, P )] (20)
18
19
20 ∆G
ej (T, P ) and pKj (T, P ) in Eqs. (6) and (7) can then be obtained from
21
22
23

ej (T, P ) = ∆G† (T, P ) + νsolv,j RT ln 1000
24 ∆G j (21)
25
Msolv
26
27
28
where
29
30
ρsolv (T, P )m0

RT
31 ∆G†j = ∆G0j (T ; P 0 ) + RT ν j ln + ln
32 100P 0 1000
33 +∆∆Gintr
j (T, P ) (22)
34
Ns
35 X
36 νj = νij (23)
37 i=1
38 Ns
39
X
∆G0j (T ; P 0 ) = νij µ0i (T ; P 0 ) (24)
40
i=1
41
Ns
42 X
43 ∆∆Gintr
j (T, P ) = νij ∆Gintr
i (T, P ) (25)
44 i=1
45
46
47 3.2 Ideal–Gas (IG) Reaction Free Energy Changes
48
49
The IG µ0i (T ; P 0 ) values may be expressed in terms of statistical mechanical expressions and
50
51
calculated from
52
53
54 qi0 (T, V ) RT

55 µ0i (T ; P 0 ) = −RT ln − ∆Hai (0)
56 V P0
57
58
59 11
1
2
3
= [Gi (T ) − Hi (0)] − ∆Hai (0) (26)
4
5
6
7 where qi0 (T, V ) is the molecule’s internal partition function and ∆Hai (0) is its atomization
8
9 energy at 0 K. Finally, we may write ∆G0j (T ; P 0 ) as
10
11
12 Ns
X Ns
X
13 ∆G0j (T ; P 0 ) = νij [Gi (T ) − Hi (0)] − νij ∆Hai (0) (27)
14 i=1 i=1
15 = ∆j [G(T ) − H(0)] + ∆Gj (0) (28)
16
17
18
19 where the quantities on the right are calculated at P 0 = 1 bar. The quantities in Eq. (27)
20
21 may be obtained from quantum mechanical calculations, as described in Section 4.3.
22
23
24
25
26
4 Systems Studied and Computational Details
27
28
Our aqueous MEA-CO2 system model contains the species {H2 O, RNH2 , CO2 , RNH+
3,
29
30
RNHCOO− , H3 O+ , OH− , HCO− 2−
3 } (CO3 is neglected as being vanishingly small), where
31
32
R = HO(CH2 )2 . The CO2 absorption process is modelled as a solution in equilibrium with a
33
34
vapour phase at a total pressure of 1 bar, and we calculated solution compositions at several
35
36 temperatures, solvent weight fractions, ω (kg MEA/kg solvent), and CO2 loadings, L(mol
37
38 CO2 /mol MEA).
39
40 The algorithm numerically solves the reaction equilibrium conditions of Eqs. (5) in the
41
42 context of a specified set of chemical reactions (these are given for the aqueous MEA-CO2
43
44 system of interest in Section 5). The required quantities are obtained as follows:
45
46
47 1. The ideal–gas reaction free energies ∆G0j (T ; P 0 ) in Eq. (24) are calculated using the
48
49 methodology described in Section 4.3.
50
51
52 2. The density of the solvent in kg m−3 , ρsolv (T, P ), is calculated, followed by the calcu-
53
54 lation of its relative static permittivity, r,solv , required for the calculation of A(T, P )
55
56 in the Davies model of Eq. (8). The simulation of r,solv is described in Section 5.4.
57
58
59 12
1
2
3
3. The solute and solvent intrinsic hydration free energies ∆Gintr quantities in Eq. (16)
4
5
and (20) respectively, are calculated using the methodology of Section 4.2. Both are
6
7
then used to calculate ∆∆Gintr
j (T, P ) in Eq. (25).
8
9
10 4. ∆G
ej (T, P ) are calculated using Eqs. (21)–(25)
11
12
13 5. Eqs. (5) are solved numerically. In the case of the ideal solution model, ln γi (T, P ) = 0
14
15 is used for the solutes and Eq. (11) is used for the solvent; in the case of the Davies
16
17 model Eq. (8) is used for the solutes and Eq. (12) is used for the solvent.
18
19
20
21
4.1 Force Fields
22
23 All nonbonded intermolecular interactions were modeled using a standard 12-6 Lennard-
24
25 Jones (LJ) plus Coulombic potential with fixed point charges
26
27 " 6 #
28
12
σij σij 1 qi qj
29 Unb (rij ) = 4εij − + (29)
30
rij rij 4πε0 rij
31
32
33
where εij and σij are the LJ energy and size parameters, respectively, qi is the atomic
34
35
partial charge of atom i, and rij is the separation distance between atoms i and j. Unlike
36
37
intermolecular interactions were calculated using Lorentz-Berthelot combining rules.
38
39
For MEA, HCO− + −
3 , RNH3 and RNHCOO , we used a modification of the original General
40
Amber Force Field (GAFF) 43 procedure in conjunction with the TIP3P water model. We
41
42
first note that the TIP3 model has been shown to give better ∆Gintr
i (P, T ) values in Eq.
43
44
(16) in conjunction with GAFF than those obtained with other water models. 44 Although
45
46
there are no widely accepted standards for determining a force field’s partial charges, it has
47
48
been shown that ∆Gintr
i (P, T ) values strongly correlate with the solute’s dipole moment.
44
49
50
To determine the partial charges, following the recent approach of Zhang et al., 45 who had
51
52
found in an earlier study 46 on a set of organic electrolyte solvents that a higher–level of
53
54
quantum theory (B3LYP/6-311++g(d,p)) than employed in the original GAFF (HF/6-31G*)
55
56
yielded reliable charges. We thus proceeded as follows: first, the gas–phase geometry of
57
58
59 13
1
2
3
each molecule was optimized using the Gaussian09 package 47 at the MP2/cc–pVTZ level of
4
5
theory/basis set, followed by a single point energy calculation on the gas-phase optimized
6
7
structure at the B3LYP/6-311++g(d,p) level. The partial charges were then determined
8
9 using the two–stage restrained electrostatic potential (RESP) procedure implemented in the
10
11 Antechamber package. 43 Zhang et al. 45 found that this procedure resulted in similar dipole
12
13 moments to those resulting from the original GAFF charge procedure; we thus expect similar
14
15 values of ∆Gintr +
i (P, T ), which we verified for MEA and RNH3 . The gas–phase geometries
16
17 of the MEA, RNH+ −
3 and RNHCOO molecules are shown in Fig. 1. The original GAFF
18
19 intramolecular bonded and Lennard–Jones (LJ) parameters were used, with the 1–4 LJ and
20
21 electrostatic interactions scaled by 1/2 and 5/6 respectively. Since the Coulomb contributions
22
23 to ∆Gintr
i (P, T ) for the indicated ionic species far outweigh the LJ contributions, we did not
24
25 pursue the possibility of modifying the original GAFF LJ parameters.
26
27 CO2 was modelled using the TraPPE force field of Potoff; 48 although the TIP3P/TraPPE
28
29 combination is not typically used for CO2 –water interactions, 49 we expect little variation of
30
31 its ∆Gintr
i (P, T ) value across different water models.
32
33 Finally, the partial charges for H3 O+ and OH− ions were taken from the polarizable
34
35 model used by Dang et al. 50 The LJ parameters of these two ions were manually optimized
36
37 to approximate the experimental hydration free energies of the individual ions referenced to
38
39 the 1–molal gas phase and the Henry–Law 1–molal aqueous phase standard state at 298.15
40
41 K and 1 bar in TIP3 water according to
42
43
44
45 ∆G∗i = ∆Gintr
i + zi eψs (30)
46
47
48
49
where ∆Gintr
i is the species intrinsic hydration free energy calculated in this work, zi is its
50
51
valence, e is the electronic charge, and ψs is the Galvani potential of TIP3P water, taken
52
53
from Zhang et al. 38
54
55
56
57
58
59 14
1
2
3
4
4.2 Intrinsic Hydration Free Energies, ∆Gintr
i (T, P )
5
6 The initial structure was generated by adding a single ion/solute in a cubic box of 888 TIP3P
7
8 water molecules employing periodic boundary conditions using Packmol. 43 A short energy
9
10 minimization was followed by a 1 ns NVT simulation and a 10 ns NPT equilibration was
11
12 performed to prepare the initial structure for the free energy calculations using an N P T
13
14 simulation. ∆Gintr
i was calculated by first linearly decoupling the electrostatic ion–water
15
16 interactions according to
17 1 qi qj
18 Uelec (rij ; m) = λelec
m (31)
19 4πε0 rij
20
21 over 12 equally spaced λ windows {0.00, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.00} This
22
23 was followed by decoupling the LJ ion-water van der Waals interactions over the 19 λ windows
24
25 {0.00, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.5, 0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.95, 1.00} us-
26
27 ing the soft-core potential
28
29 ( )
30 sc
σij12 σij6
31
ULJ (rij ; m) = 4λLJ
m εij − (32)
6 2 6 6

(1 − λLJ
m ) αLJ σij + rij
6
(1 − λLJ
m ) αLJ σij + rij
32
33
34 where αLJ = 0.5
35
36 At each λ window, we performed an 8 ns N P T simulation to collect the derivative
37
38 of the Hamiltonian every 0.2 ps, with the first 2.5 ns discarded from the analysis. The
39
40 free energy change resulting from decoupling the particle from the solvent was calculated
41
42 using the multi-state Bennett acceptance ratio method 51 as implemented in GROMACS
43
44 2018, 52 with all post-processing performed by the Python script Pymbar. 53 The GROMACS
45
46 stochastic dynamics integrator with time step of 2 fs and Langevin dynamics thermostat
47
48 with friction coefficient of 1 ps were used. For N P T simulations, the pressure was set to
49
50 1 bar using the Parrinello–Rahman barostat with a characteristic oscillation time of 5.0 ps
51
52 and the compressibility set to 4.5 × 10−5 . Short–range LJ interactions were truncated at 1.2
53
54 nm and standard mean–field long–range corrections were applied for the energy and pressure
55
56 truncation of these interactions. Long–range electrostatic interactions were evaluated with
57
58
59 15
1
2
3
the smooth particle-mesh Ewald (SPME) method with tin–foil boundary conditions, with
4
5
real space interactions truncated at 1.2 nm, a SPME B-spline of order 4, a grid spacing of
6
7
0.12 nm, and a relative accuracy of 10−5 . (Following Joung and Cheatham, 54 for single ion
8
9 hydration the net charge of the simulation box was not set to zero, since this is implicitly
10
11 accounted for using Ewald summation with tin–foil boundary conditions.) The solution free
12
13 energy change calculation was followed by a single molecule simulation of the free energy
14
15 in vacuum (in a non–periodic simulation box with infinite force–field cutoffs), which was
16
17 subtracted from the free energy change calculation in the solution.
18
19 All GROMACS input files and force field parameters are provided in the Supplementary
20
21 Information file.
22
23
24
25 4.3 Ideal–Gas Reaction Free Energy Changes, ∆G0j (T, P )
26
27
28 To obtain the initial structure of the flexible molecules (MEA, RNH3+ and RNHCOO− ), a
29
30 conformational search was initially performed with the MMFF94 force field using Spartan’18
31
32 software, 55 and the resulting geometries were then further optimized at increasing QM lev-
33
34 els of accuracy. For RNH+ −
3 and RNHCOO , the energy differences between the lowest and
35
36 next lowest energy conformers were found to be significant; therefore only the lowest energy
37
38 conformer was adopted for each of these molecules. For MEA, however, the energy differ-
39
40 ences were relatively small, and contributions from other conformations also were taken into
41
42 account using Boltzmann averaging, as described in Results and Discussion section.
43
44 To calculate the IG quantities for each species at the temperature of the interest and 1
45
46 bar, the selected geometries were further optimized with quantum chemical methods followed
47
48 by frequency calculations to ensure that the optimized structures were true minima on the
49
50 potential energy surface (no negative eigenvalues in the Hessian matrix). All QM calculations
51
52 were performed with Gaussian09. 47 To assess the sensitivity of the calculations to the QM
53
54 method and to calculate uncertainty estimates, the G(0) quantities were calculated using sev-
55
56 eral theory levels and basis sets, namely wB97XD-aug-cc-pVTZ, wB97XD/6-311++(2d,2p),
57
58
59 16
1
2
3
wB97X/6-311++G(3df,3pd), B3LYP/6-311++g(d,p), CAM-B3LYP/6-311++g(2d,2p), CAM-
4
5
B3LYP-aug-cc-pVTZ, G4, CBS-APNO and CBS-QB3.
6
7
8
9
10 5 Results and Discussion
11
12
13 For the equilibrium calculations, we may use any set of 4 linearly independent reactions; we
14
15 use the set
16
17
18 2RNH2 + CO2 = RNH+ −
3 + RNHCOO (R1)
19
20
RNH+ +
3 + H2 O = H3 O + RNH2 (R2)
21
22
23 RNHCOO− + H2 O = RNH2 + HCO−
3 (R3)
24
−
25 RNH2 + H2 O = RNH+
3 + OH (R4)
26
27
28
29
The equilibrium compositions are obtained from the solution of Eqs. (5). The following
30
31
two subsections describe our methodology for calculating the input quantities and estimating
32
33
their uncertainties. We estimated the uncertainties in the resulting equilibrium compositions
34
35
by propagating those of the input quantities.
36
37
38 5.1 Ideal–Gas Quantities
39
40
41 The only previous simulation study of the reactive absorption of CO2 in the aqueous MEA
42
43 system is that of Balaji et al., 25 who used the REMC algorithm, 19,20,56 which also requires the
44
45 ideal–gas quantities ∆G0j (T ; P 0 ) in Eq. (27). They calculated the [Gi (T ) − Hi (0)] quantity,
46
47 but neglected to include the atomization energy contribution. This is verified in Table 1
48
49 at T = 293.15 K, where our calculations of the former quantities are compared with their
50
51 results and with available JANAF data. 57 It is seen that the three sets of [Gi (T ) − Hi (0)]
52
53 results are in good agreement.
54
55
56
57
58
59 17
1
2
3
4 Table 1: Values of the dimensionless ideal–gas free energy function [G(T ) −
5
6
H(0)]/RT at T = 293.15 K and P 0 = 1 bar for the species involved in the MEA-
7 CO2 -H2 O system. The second column contains values calculated in this work,
8 the third column contains values calculated previously by Balaji et al., 25 and
9 the fourth column contains results from the JANAF thermochemical tables. 57
10
11
12 h i
qi0 (T )RT
13 Species [Gi (T ) − Hi (0)]/RT (a) − ln P0
(b)
[Gi (T ) − H(0)]/RT (c)
14
15 RNH+ 3 -28.977 -28.912 -
16 RNHCOO− -33.075 -32.961 -
17 HCO− 3 -27.250 -27.233 -
18 H 3 O+ -19.118 -19.097 -19.083
19
OH− -17.166 - -17.272
20
21 RNH2 (d) -28.757 -28.671 -
22 H2 O -18.628 -18.618 -18.717
23 CO2 -21.865 -21.882 -21.937
24 (a)
25 This work. Calculated using Gaussian09 47 with B3LYP/6-311++G(d,p).
(b) Calculated using the qˆi (T ) values of Balaji et al. 25
26
(c) Calculated from columns 4 and 5 of the JANAF thermochemical tables. 57
27
28 (d) Calculated for the lowest energy conformer at 293.15 K (see the text.)
29
30
31 Table 2: Ideal–gas reaction free energies at 0 K, ∆G0j (0) of Eq. (28) in kJ mol−1 at P = 1 bar
32
for reactions R1–R4 in the text. For the indicated reaction, the final row gives the average
33
34 value of the methods and the numbers in parentheses denote the standard deviations.
35
36 Method R1 R2 R3 R4
37 B3LYP/6-311++G(d,p) 487.40 233.31 34.61 708.27
38
wB97XD-aug-cc-pVTZ 476.66 232.22 42.12 709.27
39
40 wB97XD/6-311++(2d,2p) 472.58 235.28 42.82 714.57
41 CAM-B3LYP/6-311++(2d,2p) 469.38 234.33 39.60 713.04
42 G4 478.73 231.37 47.69 708.20
43 CBS-APNO 475.00 231.20 48.35 711.78
44
CBS-QB3 474.55 234.43 49.10 720.72
45
46 wB97X/6-311++G(3df,3pd) 470.28 232.10 41.34 716.60
47 CAM-B3LYP-aug-cc-pVTZ 470.83 232.88 39.32 707.10
48 Average(std) 473.50(3.1) 232.98(1.4) 43.79(3.7) 712.66(4.3)
49
50
We found the values of [Gi (T ) − Hi (0)]/RT (the “thermal corrections”) to be relatively
51
52
insensitive to the QM method used, in contrast to the ∆Gj (0) results. In order to account for
53
54
the effects of this sensitivity, we calculated ∆Gj (0) required in Eqs. (26)–(28) in conjunction
55
56
with the range of high–order methods indicated in Section 5.1, in addition to the commonly
57
58
59 18
1
2
3
used B3LYP/6-311++G(d,p) method (e.g., Gupta et al. 58–60 ). For our reaction equilibrium
4
5
calculations, we used the average values of ∆G0j (T ; P 0 ) obtained from the eight indicated
6
7
high–order methods shown in Table 2 and took the standard deviations as their uncertainty
8
9 estimates. The results are shown in Table 2. It can be seen that the B3LYP/6-311++G(d,p)
10
11 results differ markedly from those of the higher order methods, most importantly for reactions
12
13 R1 and R3. The results indicated in the table are in agreement with others 61 who have
14
15 indicated that B3LYP/6-311++G(d,p) results are generally inferior to those of the higher
16
17 order methods for amine–based systems; we thus excluded its results from our calculations.
18
19 It is important to account for the fact that larger molecules typically exhibit more than
20
21 one locally stable structural isomer. The set of ideal–gas isomers can be treated by an exact
22
23 “lumping procedure”, 62,63 whereby the set is replaced by a single species with a standard
24
25 chemical potential µ0I (T ; P 0 ) given by the Boltzmann-weighted sum of the Nc conformer
26
27 chemical potentials. The mole fraction of each conformer in the set is
28
29
30 exp[−µ0i (T ; P 0 )/RT ]
31 xi = PNc (33)
0 0
32 j=1 exp[−µj (T ; P )/RT ]
33
34
35 and
36
37 Nc
38
X
µ0I (T ; P 0 ) xi µ0i (T ; P 0 ) + RT ln xi

= (34)
39
i=1
40 "N #
c
41 X
42 = −RT ln exp[−µ0i (T ; P 0 )/RT ] (35)
43 j=1
44
45
For a given species, we first determined the conformer with the lowest chemical potential,
46
47
µ0∗ (T ; P 0 ), and the relative free energy differences of the other conformers from its minimum
48
49
value, δµ0i (T ; P 0 ), where
50
51
52
53 δµ0i (T ; P 0 ) = µ0i (T ; P 0 ) − µ0∗ (T ; P 0 ) (36)
54
55
56
57
58
59 19
1
2
3
The standard chemical potential µI (T ; P 0 ) of the isomer group is then calculated by
4
5
rearranging Eq. (35) to the form
6
7
8 Nc Nc !
µ0I (T ; P 0 ) X µ0i (T ; P 0 ) µ0∗ (T ; P 0 ) δµ0i (T ; P 0 )

9
X
= = − ln exp − (37)
10 RT i=1
RT RT i=1
RT
11
12
13 where Nc is the number of conformers considered. We found that the final term was van-
14
15 ishingly small for all species except MEA, for which we found 13 conformers, the same set
16
17 reported by Xie et al. 64 Table 3 shows the mole factions of the individual conformers and
18
19 the final term of Eq. (37) at each temperature considered.
20
21 Table 3: Values of the fraction of the total population of the MEA conformers with the 13
22 lowest values of µ0 (T ) at the temperatures considered in this work, calculated using M06/6-
23
311++G(3df,3dp); the conformer notation is that of Xie et al. 64 The final row “Adjustment”
24
25 values refer to the argument of the logarithmic term in Eq. (37)
26
27
28
29 Conformer 293.15 298.15 313.15 333.15 353.15
30
31 g’Gg’ 0.6177 0.6139 0.5746 0.5255 0.4811
32 gGg’ 0.1279 0.1304 0.1413 0.1541 0.1650
33 tGt 0.0638 0.0640 0.0685 0.0737 0.0778
34 gGt4 0.0506 0.0508 0.0552 0.0603 0.0647
35
36 tGg 0.0371 0.0373 0.0412 0.0459 0.0502
37 gGg 0.0330 0.0330 0.0362 0.0401 0.0436
38 tGg’ 0.0254 0.0255 0.0288 0.0330 0.0369
39 tTt 0.0116 0.0117 0.0139 0.0112 0.0201
40
g’Tt 0.0091 0.0092 0.0109 0.0134 0.0159
41
42 tTg 0.0082 0.0083 0.0100 0.0124 0.0148
43 gTg’ 0.0073 0.0074 0.0090 0.0170 0.0135
44 gTg 0.0072 0.0073 0.0089 0.0111 0.0134
45 g’Gt 0.0011 0.0012 0.0016 0.0022 0.0030
46
Adjustment 1.593 1.628 1.740 1.902 2.078
47
48
49
50
51
52
53
54
55
56
57
58
59 20
1
2
3 Table 4: Water density and dimensionless species intrinsic hydration free energies,
4
∆Gintr
i (T, P )/RT , defined in Eq. (16), at the temperatures, T , considered in this study
5
6 and pressure P = 1 bar, using the TIP3P water model and the species FFs described in the
7 text. A subscript indicates the uncertainty (one standard deviation) in the final number of
8 digits.
9
10 293.15 K 298.15 K 313.15 K 333.15K 353.15 K
11 H2 O Density(kg m−3 ) 991.8398 987.3598 973.886 954.319 932.618
12
13 Species
14 RNH+ 3 -98.354 -96.434 -95.674 -89.133 -83.513
15 RNHCOO− -150.908 -148.108 -143.969 -134.347 -125.207
16 HCO− -158.892 -155.922 -143.072 -133.382 -124.772
3
17
H3 O+ -168.903 -165.723 -156.953 -146.502 -137.312
18
19 OH− -199.963 -196.383 -185.713 -173.013 -161.762
20 MEA -12.543 -12.143 -12.263 -10.952 -9.802
21 H2 O -11.091 -10.901 -10.031 -9.041 -8.221
22 CO2 0.391 0.411 0.674 0.821 0.931
23
24
25 5.2 Intrinsic Hydration Free Energies
26
27
28 The intrinsic hydration free energies of the species, ∆Gintr
i (T ; P ) defined in Eq. (16) are
29
30 shown in Table 4. As noted in Section 3, for our purposes we only require the charge-
31
32 conserving linear combinations of the intrinsic hydration free energies, ∆Gintr
j , that are cal-
33
34 culated from the individual ∆Gintr intr
i (T ; P ) values. The individual ∆Gi (T ; P ) values show
35
36 relatively small uncertainties, which we propagated to the ∆Gintr
j quantities shown in Table
37
38 2.
39
40
41
42 5.3 pK Values
43
44
Table 5 shows the equilibrium constants and their constituent contributions for reactions
45
46 R1–R4 at the temperatures of this study, along with estimates of their uncertainties (one
47
48 standard deviation). It is difficult to compare our pK results with those of experiment,
49
50 since the latter are typically obtained indirectly by fitting an activity coefficient model to
51
52 experimental data, and the uncertainties are rarely discussed. Since reaction R3 involving
53
54 HCO− −
3 and RNHCOO is one of the most important, we compare our pK value at 298.15 in
55
56 the table with its values from several sources, 8,65–67 which yield an average value of 1.58 with
57
58
59 21
1
2
3
a standard deviation of 0.19. This overlaps with our pK value and its uncertainty range.
4
5
6
7 5.4 Davies Parameter A
8
9
10 r,solv for TIP3P water was obtained using
11
12
β < M 2 > − < M >2

13
14
r,solv =1+ (38)
30 V
15
16
17 where M = [Mx , My , Mz ] is the total dipole moment and 0 is the permittivity of free space.
18
19 We equilibrated systems of 2000 water molecules for 2 ns and used 10 ns production periods
20
21 in the NPT ensemble to obtain the system box size, which was followed by 10ns N V T
22
23 simulation to calculate r,solv , saving configurations for analysis every 0.1 ps. r,solv and A
24
25 values for TIP3P water in Eq. (9) are shown in Table 6.
26
27
28
29 5.5 Solution Compositions
30
31
32 We calculated the solution equilibrium compositions using the thermodynamic model of
33
34 Eqs. (5) and (6), with the µ†i parameters obtained by the simulation methodology of
35
36 Section 3. The conservation of mass constraints are expressed in terms of the quantities
37
38 {N 0 (RNH2 ), N 0 (H2 O), N 0 (CO2 )}, which are in turn defined in terms of the weight fraction
39
40 ω and CO2 loading L as follows:
41
42
43 0 ωMH2 O
N (RNH2 ) = N 0 (H2 O) (39)
44 (1 − ω)MRNH2
45
ωLMH2 O

0
46 N (CO2 ) = N 0 (H2 O) (40)
47 (1 − ω)MRNH2
48
49
50 where N 0 is a particle/mole number and M is a molecular weight (our simulations used
51
52 N 0 (H2 O) = 888.)
53
54 We performed two sets of calculations, both implemented by means of a locally developed
55
56 Python script: (1) based on the Henry–Law model of Eq. (1) in the case of an ideal solution
57
58
59 22
1
2
3
(ln γi ) ≡ 0); and (2) using the Davies approximation 30 for the activity coefficients:
4
5
6 √ !
I
7 ln(γi ) = −zi2 A √ − 0.3I (41)
8 1+ I
9
10
11 where zi is the valence and I is the ionic strength. Numerical results for all systems studied
12
13 are given in the Supporting Information.
14
15 Simulation results for a 30% weight fraction aqueous MEA solvent in conjunction with the
16
17 Davies approximation are shown in Fig. 2 at a range of temperatures and CO2 loadings for
18
19 a 30 weight % MEA/H2 O solvent, where they are compared with experimental results from
20
21 the literature. The curves are obtained from smoothing polynomial fits to the simulation
22
23 data points; numerical results for all species compositions are provided in the Supplementary
24
25 Information, for both the Henry–Law ideal solution and the Davies approximation. A loga-
26
27 rithmic concentration scale is used in the figure to visually emphasize species concentration
28
29 differences, which would be barely visible on a linear scale over the four orders of magni-
30
31 tude of the data. The combination MEA/RNH+
3 is also shown, because in most cases only
32
33 this quantity is available experimentally. The insets in the subfigures show simulated minor
34
35 species concentrations, which are compared with experimental data at 293.15 K, the only
36
37 state point for which such data are available. The methodology for the error (uncertainty)
38
39 bars on the simulation results (indicating one standard deviation) is discussed in Section 5.7.
40
41 The uncertainties of the experimental results can be inferred from the degree of scatter of
42
43 the results from different groups.
44
45 The simulation results for the major species combination RNH+
3 /MEA are in good agree-
46
47 ment with the experimental results at all temperatures, with both sets of results showing
48
49 relatively small uncertainties.
50
51 For RNHCOO− at 293.15 K (Fig. 2(a)), the black simulation curves incorporating their
52
53 uncertainties are consistent with all the filled black experimental data points at loadings
54
55 L / 0.7. At higher loadings, the simulation curves agree with the results of Böttinger et
56
57
58
59 23
1
2
3
al. 8 (filled black circles) except at the two highest loading values, and lie slightly below
4
5
the results of Jakobsen et al. 7 (filled upward triangles). At 313.15 K, agreement of the
6
7
simulation results with the experimental results of Jakobsen et al. 7 and of Hilliard 9 (filled
8
9 squares) remains good at L / 0.4, is increasingly poorer at higher loadings. At 333.15 K
10
11 and 353.15 K, reasonable agreement of the simulation results with those of Böttinger et al. 8
12
13 is maintained.
14
15 At all temperatures the HCO−
3 concentrations are very small at low loadings, and the
16
17 experimental concentration data show considerable scatter, reflecting the inherent experi-
18
19 mental difficulties in determining species concentrations of very small magnitude. At 293.15
20
21 K and L / 0.6,, the simulation results incorporating their uncertainties are in agreement
22
23 with the experimental results of Matin et al. 11 (filled red diamonds), and at L / 0.4 are
24
25 up to almost two orders of magnitude higher than the other experimental data. The HCO−
3
26
27 concentrations rise to significant values above L u 0.5, where the experimental data show
28
29 generally less scatter. At 293.15 K for these values of L, the simulation results incorpo-
30
31 rating their uncertainties are in reasonable agreement with those of Böttinger et al. 8 (filled
32
33 red circles) and of Jakobsen et al. 7 (filled and open upward red triangles). At 313.15 K,
34
35 the agreement of the simulations with the (only available) data of Jakobsen 7 is similar. At
36
37 T = 333.15 K, there is no experimental data above L = 0.6 and similarly at T = 353.15 K
38
39 above L = 0.5. At these temperatures and lower loadings, the only experimental data are
40
41 those of Böttinger et al. 8 (filled red circles) and of Jakobsen et al. 7 (filled and open upward
42
43 red triangles), which lie considerable below the simulation results.
44
45 The scatter of the experimental results, particularly for HCO−
3 , indicate that it is prob-
46
47 lematic to experimentally determine small species concentrations with reasonable precision,
48
49 particularly at mole fractions less than about 0.01. We conclude that the simulation results
50
51 predict equilibrium compositions of similar quality to the experimental measurements; in ad-
52
53 dition, estimates of the very small species compositions (CO2 , H3 O and OH− ) are available
54
55 from our simulations, unlike the case for the experiments.
56
57
58
59 24
1
2
3
We now compare our results with those of the only previous simulation study of this
4
5
system, 25 who used a variant of the REMC algorithm. 19–21 We first remark that the REMC
6
7
algorithm (in common with many other simulation algorithms) is intrinsically unsuited to
8
9 calculations involving systems with species of very low concentrations. For example, using a
10
11 simulation box with Nbox particles, an equilibrium mole fraction of zi would require 1 particle
12
13 of the species to be observed on average every 1/(Nbox zi ) reaction moves. In our simulations,
14
15 a mole fraction xCO2 of order 10−10 is not uncommon. To observe such a concentration, the
16
17 REMC algorithm would require a CO2 particle to be observed on average every 20 million
18
19 reaction moves (and a reasonable uncertainty would require it to be observed multiple times)
20
21 using the Nbox ≈ 500 value of Balaji et al., 25 who displayed graphical results for L ≤ 0.7
22
23 for the concentrations of {MEA + RNH+ − − −
3 , RNHCOO , HCO3 }, and for HCO3 only at high
24
25 loadings where its concentration is non–negligible.
26
27 We also note that in the work of Balaji et al., 25 the ideal–gas quantities ∆Gj (0) in Eq.
28
29 (28) were erroneously omitted in the calculation of the values of ∆Gj (T ; P 0 )/RT in Table
30
31 5; these quantities provide by far the largest contributions to ∆Gj (T ; P 0 )/RT . Their study
32
33 also modelled the system using a different species set than used here, omitting OH− and
34
35 including CO2−
3 ; however, since these species are present in very small amounts, they have
36
37 only a minor effect on the concentrations of the remaining species. In addition, in our study
38
39 different species force fields were used. Differences in our results could be due to any of these
40
41 factors; it could also be due to the nonconvergence of their REMC simulations.
42
43
44
45 5.6 CO2 Partial Pressure
46
47
48 The partial pressure of CO2 in the vapour phase at a total pressure of 1 bar is determined
49
50 from the equality of its solution and vapour phase chemical potentials. At the relatively low
51
52 pressure considered here, the vapour phase may be treated as an ideal gas, which yields its
53
54
55
56
57
58
59 25
1
2
3
partial pressure, PCO2 , as
4
5
6
! !
µres,N V T,∞

RT ρsolv (T, Pe) CO2 (T, ρ(T, Pe)
7 PeCO2 = mCO2 exp (42)
8 100P 0 1000 RT
9
10
11 We ignore the small pressure dependence of the quantities in this expression, and use our
12
13 simulation values calculated at P = 1 bar.
14
15 The dependence of P (CO2 ) on loading at the temperatures considered is shown in Fig.
16
17 3. Error bars for our results are shown at 298.15 K and 353.15 K, and those at the other
18
19 temperatures are similar. The methodology used is described in the next Section. We note
20
21 that, although we only show P (CO2 ) data from one source 5 to simplify the graph, other
22
23 literature results 68 imply experimental error bars of similar magnitude to those shown for
24
25 our simulation data in Fig. 3.
26
27
28
29
5.7 Uncertainty Analysis
30
31 Although the propagation of uncertainties in the values of ∆G
ej in Table 5 to the uncertainties
32
33 ∆ ln(x∗i ) in the equilibrium species mole fractions and to the uncertainty ∆ ln(PeCO2 ) in the
34
35 CO2 partial pressure can be performed using methodology similar to that of Morgan et al., 31
36
37 we take a simpler numerical approach here.
38
39 Since the equilibrium concentrations of H3 O+ and OH− are always very small, their
40
41 equilibrium mole fractions are primarily determined by reactions R2 and R4, and those of
42
43 all the other species are primarily determined by reactions R1 and R3. We thus estimated
44
45 the uncertainties in the equilibrium compositions arising from the uncertainties ∆∆G/RT
e
46
47 in the reaction ∆G/RT
e values using
48
49 1/2
50
 !2 !2
51
 ∂ ln(x∗i ) ∂ ln(x∗i ) 
∆ ln(x∗ ) = ej /RT )2 +
(∆∆G ek /RT )2 (43)
(∆∆G
52  ∂(∆G
ej /RT ) ∂(∆G
ek /RT ) 
53
54
55 where (j, k) = (2, 4) for H3 O+ and OH− , and (j, k) = (1, 3) for all other species. We
56
57
58
59 26
1
2
3
numerically calculated the relevant compositional derivatives.
4
5
Applying the same approach for PCO2 gives
6
7
8

 ∂ ln msoln (T, P ; m∗ ) 2
! !2
∗
9 CO2 e1 /RT )2 + ∂ ln msoln
CO2 (T, P ; m ) e3 /RT )2
∆ ln(PeCO2 ) = (∆∆G (∆∆G
10  (∆G1 /RT )
e (∆Ge3 /RT )
11
12 2 )1/2
∆µres

13 + CO2
(44)
14 RT
15
16
17 where the asterisk denotes the equilibrium composition.
18
19 Fig. 2 shows that our simulation results are in good agreement with those of experiment
20
21 when their mutual uncertainties are taken into account.
22
23 The accuracy of our predictions of the species solution concentrations and the CO2 partial
24
25 pressure above the solution, PCO2 , depends on the accuracies of three quantities:
26
27
28 1. the ideal–gas free energy changes for the reactions R1–R4, of which the most important
29
30 are R1 and R3.
31
32 Although our calculations of these quantities have relatively large uncertainties, we
33
34 believe that they are reasonably accurate.
35
36
37
2. the hydration free energy changes for the reactions.
38
39
Although these have relatively low uncertainties, they may not be sufficiently accu-
40
41
rate due to inadequacies in the underlying force fields, and in current work, we are
42
43
investigating improvements to the force fields
44
45 3. the species chemical potential model used to calculate reaction equilibrium.
46
47 The Davies approximation for the activity coefficients in conjunction with the species
48
49 Henry–law–based chemical potential model may not be sufficiently accurate, partic-
50
51 ularly at high CO2 loading values. We are investigating the approach of dispensing
52
53 with a particular algebraic form of chemical potential model, and instead employing
54
55 the strategy of the ReMD algorithm, 29 which recalculates species chemical potentials
56
57
58
59 27
1
2
3
at a sequence of states calculated by ideal solution extrapolations.
4
5
6
7
8 6 Conclusions and Recommendations
9
10
11 We have developed a molecular–based simulation methodology to calculate all species con-
12
13 centrations undergoing chemical reaction equilibrium in electrolyte solutions; an important
14
15 feature is its ability to treat species with very small concentrations. We have applied it to
16
17 the benchmark MEA(RHN2 )–H2 O–CO2 reactive absorption system, which is a key system
18
19 used in CO2 capture. We have also calculated PCO2 , the partial pressure of CO2 above the
20
21 solution. We have used no experimental data in our methodology. Our approach is read-
22
23 ily extended to other systems, and provides a useful computational tool for the in silico
24
25 screening of potential CO2 capture solvents.
26
27 The algorithm presented here is superior to the REMC algorithm, 19,20,56 the only other
28
29 generally applicable approach for molecular simulation of chemical reaction equilibria. For all
30
31 but the simplest of systems, REMC requires the use of special system–specific Monte Carlo
32
33 moves to implement the reactions, and would furthermore require very long computation
34
35 times and/or large system sizes to cope with small species concentrations. Our approach
36
37 overcomes these problems, and can be straightforwardly implemented for systems of any
38
39 molecular complexity using as its underlying computational engine a standard Molecular
40
41 Dynamics (MD) simulation package capable of calculating free energy changes. In addi-
42
43 tion, the inherent parallelization capability of MD enables the algorithm’s computationally
44
45 efficient implementation, which is problematic with MC–based algorithms.
46
47 Our predictions are accompanied by an Uncertainty Analysis based on propagating the
48
49 uncertainties of the underlying parameters to those of the equilibrium species concentrations
50
51 and the CO2 partial pressure above the solution. This allows comparison with the literature
52
53 results, which show mutual disagreement for species with small concentrations due to the
54
55 experimental challenges involved. We believe that the quality of our predictive methodology
56
57
58
59 28
1
2
3
is comparable to that of the experimental studies, both for the species concentrations and for
4
5
the CO2 partial pressure. We recommend that all theoretical and experimental predictions
6
7
be accompanied by an Uncertainty Analysis.
8
9 Our approach makes contact with macroscopic thermodynamic modeling approaches
10
11 by directly translating the relevant molecular simulation quantities to the parameters of
12
13 a Henry-Law–based thermodynamic model, the most important of which are reaction pK
14
15 values. These quantities are obtained experimentally by fitting a thermodynamic model to
16
17 experimentally measured data based on species concentrations. When the relevant concen-
18
19 trations are very small and difficult to determine, this leads to large uncertainties in the
20
21 predicted pK value. The experimental determination of pK1 from the measurement of the
22
23 CO2 partial pressure may be an example of this.
24
25 We compared our results and methodology with those of the only other group to have
26
27 studied the MEA–H2 O–CO2 system using molecular simulation methodology. 25 This group
28
29 inadvertently omitted significant terms in the ideal–gas chemical potential contributions
30
31 involved in their algorithm, and their simulation methodology did not permit the calculation
32
33 of minor species concentrations and of PCO2 .
34
35
36
37
38 Supporting Information
39
40
41 The Supporting Information contains material concerning the calculation of ideal–gas stan-
42
43 dard chemical potentials from electronic structure computations, the calculation of the
44
45 Henry–Law reference state chemical potential, µ† (T, P ) from simulation quantities, tables
46
47 of force–field parameters for the species involved in the simulations, and tabulated values of
48
49 the mole fractions of the species in Fig. 2.
50
51
52
53
54
55
56
57
58
59 29
1
2
3
4
Acknowledgements
5
6 Support for this work was provided by the Natural Sciences and Engineering Research Coun-
7
8 cil of Canada (Strategic Program Grant No. STPGP 479466-15), and by the SHARCNET
9
10 (Shared Hierarchical Academic Research Computing Network) HPC consortium (www.sharcnet.ca)
11
12 and Compute Canada (www.computecanada.ca)
13
14
15
16 Table 5: Quantities contributing to ∆Gej (T, P ), defined in Eqs. (21)–(25) for reactions R1–
17 R4 at the indicated temperatures and P = P 0 = 1 bar. ∆G0j (T ; P 0 ) is calculated from Eq.
18 (28) using the second column of this table and data in Tables 2–4, in conjunction with Eq.
19
(37). ∆∆Gintr is calculated from Eq. (25) and data in Table 4. Uncertainties are given in
20 j
21 parentheses as one standard deviation.
22
23 ∆j [G(T )−H(0)] ∆G0j (T ;P 0 ) ∆∆Gintr
j (T,P ) ∆G
ej
RT RT RT ln Kj ≡ RT pKj ≡ − log10 (Kj )
24
25 293.15 K
26
R1 17.33 212.53(1.27) -224.56(0.11) -15.22(1.28) -6.61(0.55)
27
28
R2 -0.27 94.85(0.57) -72.00(0.06) 18.83(0.58) 8.18(0.25)
29 R3 -4.30 13.20(1.52) -9.44(0.09) -0.26(1.52) -0.11(0.66)
30 R4 1.24 294.09(1.76) -274.68(0.06) 15.40(1.77) 6.69(0.77)
31 298.15 K
32 R1 17.36 209.34(1.25) -220.67(0.11) -14.52(1.26) -6.31(0.55)
33 R2 -0.27 93.23(0.56) -70.52(0.06) 18.69(0.57) 8.11(0.25)
34 R3 -4.28 12.90(1.49) -9.06(0.09) -0.18(1.50) -0.08(0.65)
35
R4 1.25 289.22(1.73) -269.78(0.06) 15.43(1.74) 6.70(0.75)
36
37 313.15 K
38 R1 17.44 200.41(1.19) -209.78(0.12) -12.61(1.20) -5.48(0.52)
39 R2 -0.26 88.67(0.54) -66.66(0.06) 17.99(0.54) 7.81(0.23)
40 R3 -4.23 12.03(1.42) -8.51(0.10) -0.49(1.42) -0.21(0.62)
41 R4 1.26 275.53(1.65) -255.94(0.06) 15.57(1.65) 6.76(0.72)
42 333.15 K
43
R1 17.54 189.77(1.12) -196.29(0.09) -9.80(1.12) -4.26(0.49)
44
45 R2 -0.25 83.22(0.51) -62.19(0.04) 17.01(0.51) 7.39(0.22)
46 R3 -4.16 11.01(1.34) -8.08(0.08) -1.09(1.34) -0.47(0.58)
47 R4 1.27 259.19(1.55) -239.24(0.05) 15.94(1.55) 6.92(0.67)
48 353.15 K
49 R1 17.63 180.35(0.06) -184.64(0.09) -7.60(1.06) -3.30(0.46)
50 R2 -0.25 78.36(0.48) -58.10(0.04) 16.25(0.48) 7.06(0.21)
51
R3 -4.10 10.08(1.26) -7.53(0.08) -1.46(1.26) -0.64(0.55)
52
53
R4 1.29 244.73(1.46) -224.53(0.04) 16.19(1.47) 7.03(0.64)
54
55
56
57
58
59 30
1
2
3 Table 6: Values of r,solv in Eq. (9) and A in Eq. (8) at the temperatures of this study.
4
5 T r,solv A
6
7
8 293.15 102.102 0.350
9 298.15 97.857 0.363
10
313.15 94.067 0.356
11
12
313.15 90.830 0.338
13 353.15 83.094 0.350
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 31
1
2
3
4
References
5
6 (1) Robinson, R. A.; Stokes, R. H. Electrolyte Solutions; Dover Publications: New York,
7
8 2002.
9
10
11 (2) Kontogeorgis, G. M.; Maribo-Mogensen, B.; Thomsen, K. The Debye-Hückel Theory
12
13 and Its Importance in Modeling Electrolyte Solutions. Fluid Phase Equilibria 2018,
14
15 462, 130–152.
16
17
18 (3) Kumoro, A. C.; Raksajati, A.; Ho, M.; Wiley, D.; Hadiyanto,; Roces, S. A.; Yung, L.;
19
20 Rong, X.; Lothongkum, A. W.; Phong, M. T.; Hussain, M. A.; Daud, W. R. W.; Nam, P.
21
22 T. S. Solvent Development for Post-Combustion CO2 Capture: Recent Development
23
24 and Opportunities. MATEC Web of Conferences 2018, 156, 03015.
25
26
27 (4) Song, C.; Liu, Q.; Ji, N.; Deng, S.; Zhao, J.; Li, Y.; Song, Y.; Li, H. Alternative
28
29 Pathways for Efficient CO2 Capture by Hybrid Processes—A Review. Renewable and
30
31 Sustainable Energy Reviews 2018, 82, 215–231.
32
33
34 (5) Jou, F. Y.; Mather, A. E.; Otto, F. D. The Solubility of C02 in a 30 Mass Percent
35
36 Monoethanolamine Solution. Can. J. Chem. Eng. 1995, 73, 140–147.
37
38
(6) Gabrielsen, J.; Michelsen, M. L.; Stenby, E. H.; Kontogeorgis, G. M. A Model for
39
40
Estimating CO2 Solubility in Aqueous Alkanolamines. Ind. & Eng. Chem. Res. 2005,
41
42
44, 3348–3354.
43
44
45 (7) Jakobsen, J. P.; Krane, J.; Svendsen, H. F. Liquid–Phase Composition Determination
46
47 in CO2 –H2 O–Alkanolamine Systems: An NMR Study. Ind. & Eng. Chem. Res. 2005,
48
49 44, 9894–9903.
50
51
52 (8) Böttinger, W.; Maiwald, M.; Hasse, H. Online NMR Spectroscopic Study of Species
53
54 Distribution in MEA–H2 O–CO2 and DEA–H2 O–CO2 . Fluid Phase Equilib. 2008, 263,
55
56 131–143.
57
58
59 32
1
2
3
(9) Hilliard, M. D. A Predictive Thermodynamic Model for an Aqueous Blend of Potassium
4
5
Carbonate, Piperazine, and Monoethanolamine for Carbon Dioxide Capture from Flue
6
7
Gas. Un. of Texas Austin. Ph.D. thesis, 2008.
8
9
10 (10) Arcis, H.; Ballerat-Busserolles, K.; Rodier, L.; Coxam, J.-Y. Enthalpy of Solution of
11
12 Carbon Dioxide in Aqueous Solutions of Monoethanolamine at Temperatures of 322.5
13
14 K and 372.9 K and Pressures Up to 5 MPa. J. Chem. Eng. Data 2011, 56, 3351–3362.
15
16
17 (11) Matin, N. S.; Remias, J. E.; Neathery, J. K.; Liu, K. Facile Method for Determination
18
19 of Amine Speciation in CO2 Capture Solutions. Ind. & Eng. Chem. Res. 2012, 51,
20
21 6613–6618.
22
23
24 (12) Puxty, G.; Maeder, M. A Simple Chemical Model to Represent CO2 –Amine–H2 O
25
26 Vapour–Liquid Equilibria. Int. J. Greenhouse Gas Control 2013, 17, 215–224.
27
28
29 (13) Arcis, H.; Coulier, Y.; Ballerat-Busserolles, K.; Rodier, L.; Coxam, J.-Y. Enthalpy
30
31 of Solution of CO2 in Aqueous Solutions of Primary Alkanolamines: A Comparative
32
33 Study of Hindered and Nonhindered Amine-Based Solvents. Ind. & Eng. Chem. Res.
34
35 2014, 53, 10876–10885.
36
37
(14) Lloret, J. O.; Vega, L. F.; Llovell, F. A Consistent and Transferable Thermody-
38
39
namic Model to Accurately Describe CO2 Capture with Monoethanolamine. J. CO2
40
41
Utilization 2017, 21, 521–533.
42
43
44 (15) Plakia, A.; Pappa, G.; Voutsas, E. Modeling of CO2 Solubility in Aqueous Alkanolamine
45
46 Solutions With an Extended UMR-PRU Model. Fluid Phase Equilib. 2018, 478, 134–
47
48 144.
49
50
51 (16) du Preez, L. J.; Motang, N.; Callanan, L. H.; Burger, A. J. Determining the Liquid
52
53 Phase Equilibrium Speciation of the CO2 –MEA–H2 O System Using a Simplified in Situ
54
55 Fourier Transform Infrared Method. Ind. & Eng. Chem. Res. 2019, 58, 469–478.
56
57
58
59 33
1
2
3
(17) Gupta, M.; da Silva, E. F.; Svendsen, H. F. Computational Study of Equilibrium Con-
4
5
stants for Amines and Amino Acids for CO2 Capture Solvents. Energy Procedia 2013,
6
7
37, 1720–1727.
8
9
10 (18) Teranishi, K.; Ishikawa, A.; Sato, H.; Nakai, H. Systematic Investigation of the Thermo-
11
12 dynamic Properties of Amine Solvents for CO2 Chemical Absorption Using the Cluster-
13
14 Continuum Model. Bull. Chem. Soc. Japan 2017, 90, 451–460.
15
16
17 (19) Smith, W. R.; Trı́ska, B. The Reaction Ensemble Method for the Computer Simulation
18
19 of Chemical and Phase Equilibria. I. Theory and Basic Examples. J. Chem. Phys. 1994,
20
21 100, 3019–3027.
22
23
24 (20) Johnson, K. J.; Panagiotopoulos, Z., A.; Gubbins, E., K. A New Simulation Technique
25
26 for Reacting or Associating Fluids. Molec. Phys. 1994, 81, 717–733.
27
28
29 (21) Turner, C. H.; Brennan, J. K.; Lı́sal, M.; Smith, W. R.; Karl Johnson, J.; Gubbins, K. E.
30
31 Simulation of Chemical Reaction Equilibria by the Reaction Ensemble Monte Carlo
32
33 Method: A Review. Molec. Simulation 2008, 34, 119–146.
34
35
(22) Lı́sal, M.; Smith, W. R.; Bures̆, M.; Vacek, V.; Navrátil, J. REMC Computer Simula-
36
37
tions of the Thermodynamic Properties of Argon and Air Plasmas. Molec. Phys. 2002,
38
39
100, 2487–2497.
40
41
42 (23) Tuttafesta, M.; D’Angola, A.; Laricchiuta, A.; Minelli, P.; Capitelli, M.; Colonna, G.
43
44 GPU and Multi–Core Based Reaction Ensemble Monte Carlo Method for Non-ideal
45
46 Thermodynamic Systems. Comput. Phys. Commun. 2014, 185, 540–549.
47
48
49 (24) Mullen, R. G.; Corcelli, S. A.; Maginn, E. J. Reaction Ensemble Monte Carlo Sim-
50
51 ulations of CO2 Absorption in the Reactive Ionic Liquid Triethyl(octyl)phosphonium
52
53 2-Cyanopyrrolide. J. Phys. Chem. Lett. 2018, 9, 5213–5218.
54
55
56
57
58
59 34
1
2
3
(25) Balaji, S. P.; Gangarapu, S.; Ramdin, M.; Torres-Knoop, A.; Zuilhof, H.;
4
5
Goetheer, E. L.; Dubbeldam, D.; Vlugt, T. J. Simulating the Reactions of CO2 in
6
7
Aqueous Monoethanolamine Solution by Reaction Ensemble Monte Carlo Using the
8
9 Continuous Fractional Component Method. J. Chem. Theory Comput. 2015, 11, 2661–
10
11 2669.
12
13
14 (26) Shi, W.; Maginn, E. J. Improvement in Molecule Exchange Efficiency in Gibbs En-
15
16 semble Monte Carlo: Development and Implementation of the Continuous Fractional
17
18 Component Move. J. Comput. Chem. 2008, 29, 2520–2530.
19
20
21 (27) Rosch, T. W.; Maginn, E. J. Reaction Ensemble Monte Carlo Simulation of Complex
22
23 Molecular Systems. J. Chem. Theory & Comput. 2011, 7, 269–279.
24
25
26 (28) Poursaeidesfahani, A.; Hens, R.; Rahbari, A.; Ramdin, M.; Dubbeldam, D.; Vlugt, T.
27
28 J. H. Efficient application of Continuous Fractional Component Monte Carlo in the
29
30 Reaction Ensemble. J. Chem. Theory. Comput. 2017, 13, 4452–4466.
31
32
33 (29) Smith, W. R.; Qi, W. Molecular Simulation of Chemical Reaction Equilibrium by Com-
34
35 putationally Efficient Free Energy Minimization. ACS Cent. Sci. 2018, 4, 1185–1193.
36
37
(30) Davies, C. W. The Extent of Dissociation of Salts in Water. Part VIII. An Equation
38
39
for the Mean Ionic Activity Coefficient of an Electrolyte in Water, and a Revision of
40
41
the Dissociation Constants of Some Sulphates. J. Chem. Soc. 1938, 2093.
42
43
44 (31) Morgan, J. C.; Chinen, A. S.; Omell, B.; Bhattacharyya, D.; Tong, C.; Miller, D. C.
45
46 Thermodynamic Modeling and Uncertainty Quantification of CO2 –Loaded Aqueous
47
48 MEA Solutions. Chem. Eng. Sci. 2017, 168, 309–324.
49
50
51 (32) Smith, W. R.; Missen, R. W. Sensitivity Analysis in ChE Education. Part 1. Introduc-
52
53 tion and Application to Explicit Models. Chem. Eng. Educ. 2003, 37, 222–227.
54
55
56
57
58
59 35
1
2
3
(33) Smith, W. R.; Missen, R. W. Sensitivity Analysis in ChE Education. Part 2. Application
4
5
to Implict Models. Chem. Eng. Educ. 2003, 37, 254–260.
6
7
8 (34) Smith, W. R.; Missen, R. W. Chemical Reaction Equilibrium Analysis: Theory and
9
10 Algorithms; Krieger Publishing Co.; Reprint of same title, Willey-Interscience, 1982:
11
12 Malabar, Florida, 1991.
13
14
15 (35) Hamer, W. J.; Wu, Y.-C. Osmotic Coefficients and Mean Activity Coefficients of Uni-
16
17 univalent Electrolytes in Water at 25◦ C. J. Phys. Chem. Ref. Data 1972, 1.
18
19
20 (36) Smith, W. R.; Nezbeda, I.; Kolafa, J.; Moučka, F. Recent Progress in the Molecular
21
22 Simulation of Thermodynamic Properties of Aqueous Electrolyte Solutions. Fluid Phase
23
24 Equilb. 2018, 466, 19–30.
25
26
27 (37) Nezbeda, I.; Moučka, F.; Smith, W. R. Recent Progress in Molecular Simulation of
28
29 Aqueous Electrolytes: Force Fields, Chemical Potentials and Solubility. Molec. Phys.
30
31 2016, 114, 1665–1690.
32
33
(38) Zhang, H.; Jiang, Y.; Yan, H.; Yin, C.; Tan, T.; van der Spoel, D. Free–Energy Calcu-
34
35
lations of Ionic Hydration Consistent with the Experimental Hydration Free Energy of
36
37
the Proton. J Phys Chem Lett 2017, 8, 2705–2712.
38
39
40 (39) Höfer, T. S.; Hünenberger, P. H. Absolute Proton Hydration Free Energy, Surface Po-
41
42 tential of Water, and Redox Potential of the Hydrogen Electrode from First Principles:
43
44 QM/MM MD Free–Energy Simulations of Sodium and Potassium Hydration. J. Chem.
45
46 Phys. 2018, 148, 222814.
47
48
49 (40) Lin, F. Y.; Lopes, P. E.; Harder, E. D.; Roux, B.; MacKerell, A. D. Polarizable Force
50
51 Field for Molecular Ions Based on the Classical Drude Oscillator. J. Chem. Inf. Model.
52
53 2018, 58, 993–1004.
54
55
56
57
58
59 36
1
2
3
(41) Wagman, D. D.; Evans, W. H.; Parker, V. B.; Schumm, R. H.; Halow, I.; Bailey, S. M.;
4
5
Churney, K. L.; Nuttall, R. L. The NBS Tables of Chemical Thermodynamic Properties:
6
7
Selected Values for Inorganic and C1 and C2 Organic Substances in SI Units. J. Phys.
8
9 & Chem. Ref. Data 1982, 11.
10
11
12 (42) Majer, V.; Sedlbauer, J.; Wood, R. H. In Aqueous Systems at Elevated Temperatures
13
14 and Pressures: Physical Chemistry in Water, Steam and Hydrothermal Solutions;
15
16 Palmer, D. A., Fernandez-Prini, R., Harvey, A. H., Eds.; Elsevier Academic Press,
17
18 Amsterdam, 2004; Chapter 4.
19
20
21 (43) Perlman, D. A.; Case, D. A.; Caldwell, J. W.; Ross, W. S.; Cheatham III, T. E.; De-
22
23 Bolt, S.; ferguson, D.; Seibel, G.; Kollman, P. AMBER, A Package of Computer Pro-
24
25 grams for Applying Molecular Mechanics, Normal Mode Analysis, Molecular Dynamics
26
27 and Free Energy Calculations to Simulate the Structural and Energetic Properties of
28
29 Molecules. Computer Phys. Comm. 1995, 91, 1–41.
30
31
32
(44) Mobley, D. L.; Dumont, E.; Chodera, J. D.; Dill, K. A. Comparison of Charge Models
33
34
for Fixed–Charge Force Fields: Small–Molecule Hydration Free Energies in Explicit
35
36
Solvent. J. Phys. Chem. B 2007, 111, 2242–2254.
37
38 (45) Zhang, Y.; Zhang, Y.; McCready, M. J.; Maginn, E. J. Evaluation and Refinement of
39
40 the General AMBER Force Field for Nineteen Pure Organic Electrolyte Solvents. J.
41
42 Chem. Eng. Data 2018, 63, 3488–3502.
43
44
45 (46) Barbosa, N. S. V.; Zhang, Y.; Lima, E. R. A.; Tavares, F. W.; Maginn, E. J. Devel-
46
47 opment of an AMBER-Compatible Transferable Force Field for Poly(ethylene glycol)
48
49 Ethers (glymes). J. Mol. Model. 2017, 23, 194.
50
51
52 (47) Frisch, M. J. et al. Gaussian 09, revision E.01; Gaussian Inc.: Wallingford CT, 2009.
53
54
(48) Potoff, J. J.; Siepmann, J. I. Vapor-Liquid Equilibria of Mixtures Containing Alkanes,
55
56
Carbon Dioxide, and Nitrogen. AIChE J. 2001, 47, 1676–1682.
57
58
59 37
1
2
3
(49) Liu, Y.; Panagiotopoulos, A. Z.; Debenedetti, P. G. Monte Carlo Simulations of High-
4
5
Pressure Phase Equilibria of CO2 -H2 O Mixtures. J. Phys. Chem. B 2011, 115, 6629–
6
7
6635.
8
9
10 (50) Roy, S.; Dang, L. X. Water Exchange Dynamics Around H3 O+ and OH− Ions. Chem.
11
12 Phys. Letters 2015, 628, 30–34.
13
14
15 (51) Bennett, C. H. Efficient Estimation of Free Energy Differences From Monte Carlo Data.
16
17 J. Comput. Phys. 1976, 22, 245–268.
18
19
20 (52) Abraham, J.; van der Spoel, Lindahl, E.; Hess, B. and the GROMACS development
21
22 team, version 2018. GROMACS User Manual 2018, www.gromacs.org.
23
24
25 (53) Shirts, M. R.; Chodera, J. D. Statistically Optimal Analysis of Samples from Multiple
26
27 Equilibrium States. J. Chem. Phys. 2008, 129, 124105.
28
29
(54) Joung, I. S.; Cheatham III, T. E. Determination of Alkali and Halide Monovalent Ion
30
31
Parameters For Use in Explicitly Solvated Biomolecular Simulations. J. Phys. Chem.
32
33
B 2008, 112, 9020–9041.
34
35
36 (55) Spartan18, Wavefunction Inc., Irvine CA. 2018.
37
38
39 (56) Turner, C. H.; Brennan, J. K.; Lı́sal, M. Replica Exchange for Reactive Monte Carlo
40
41 Simulations. J. Phys. Chem. C 2007, 111, 15706–15715.
42
43
44 (57) Chase, M. W., Jr., Thermochemical Tables, Fourth Edition, Parts I and II. J. Phys.
45
46 Chem. Ref. Data, Monograph #9, 1998, Am. Chem. Society and Am. Physical Society.
47
48
49 (58) Gupta, M.; da Silva, E. F.; Svendsen, H. F. Modeling Temperature Dependency of
50
51 Amine Basicity Using PCM and SM8T Implicit Solvation Models. J. Phys. Chem. B
52
53 2012, 116, 1865–1875.
54
55
56
57
58
59 38
1
2
3
(59) Gupta, M.; da Silva, E. F.; Svendsen, H. F. Comparison of Equilibrium Constants of
4
5
Various Reactions Involved in Amines and Amino Acid Solvents for CO2 Absorption.
6
7
Energy Procedia 2014, 51, 161–168.
8
9
10 (60) Gupta, M.; da Silva, E. F.; Svendsen, H. F. Postcombustion CO2 Capture Solvent Char-
11
12 acterization Employing the Explicit Solvation Shell Model and Continuum Solvation
13
14 Models. J Phys Chem B 2016, 120, 9034–9050.
15
16
17 (61) Orestes, E.; Ronconi, C. M.; Carneiro, J. W. Insights Into the Interactions of CO2 with
18
19 Amines: a DFT Benchmark Study. Phys. Chem. Chem. Phys. 2014, 16, 17213–17219.
20
21
22 (62) Smith, W. R.; Missen, R. W. The Effect of Isomerization on Chemical Equilibrium.
23
24 Can. J. Chem. Eng. 1974, 52, 280–282.
25
26
27 (63) Ho, J.; Coote, M. L.; Cramer, C. J.; Truhlar, D. G. In Organic Electrochemistry, 5th
28
29 edition; Hammerich, O., Speiser, B., Eds.; CRC Press: Boca Raton, FL, 2016; Chapter
30
31 4, pp 229–259.
32
33
(64) Xie, H. B.; Li, C.; He, N.; Wang, C.; Zhang, S.; Chen, J. Atmospheric Chemical
34
35
Reactions of Monoethanolamine Initiated by OH Radical: Mechanistic and Kinetic
36
37
Study. Environ. Sci. Technol. 2014, 48, 1700–1706.
38
39
40 (65) Kim, I.; Hoff, K. A.; Hessen, E. T.; Haug-Warberg, T.; Svendsen, H. F. Enthalpy of
41
42 Absorption of CO2 with Alkanolamine solutions Predicted from Reaction Equilibrium
43
44 Constants. Chem. Eng. Sci. 2009, 64, 2027–2038.
45
46
47 (66) McCann, N.; Maeder, M.; Hasse, H. A Calorimetric Study of Carbamate Formation. J.
48
49 Chem. Thermodynamics 2011, 43, 664–669.
50
51
52 (67) Fernandes, D.; Conway, W.; Wang, X.; Burns, R.; Lawrance, G.; Maeder, M.; Puxty, G.
53
54 Protonation Constants and Thermodynamic Properties of Amines for Post Combustion
55
56 Capture of CO2 . J. Chem. Thermodynamics 2012, 51, 97–102.
57
58
59 39
1
2
3
(68) Wagner, M.; von Harbou, I.; Kim, J.; Ermatchkova, I.; Maurer, G.; Hasse, H. Solubility
4
5
of Carbon Dioxide in Aqueous Solutions of Monoethanolamine in the Low and High
6
7
Gas Loading Regions. J. Chem. Eng. Data 2013, 58, 883–895.
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 40
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20 (a) RNH2
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35 (b) RNH+
3
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50 (c) RNHCOO−
51
52 Figure 1: MP2/aug-cc pVTZ optimized gas phase geometry of ethanolamine (a), protonated
53 ethanolamine (b) and its carbamate form (c) used in this study to derive RESP charges.
54
55
56
57
58
59 41
1
2
3
4
5
6 0.5 0.5
7 +
RNH3 / MEA
+
RNH3 /MEA
1
8 1
9 1.5 +
RNH3 1.5 +
RNH3
10 RNHCOO
- RNHCOO
-
11 2 2
12
-log10(x)
MEA
-log10(x)
2.5 2.5
13 MEA
14 3 3 HCO3
-
15 3.5 HCO3
-
2
2
4
16 6
CO2 3.5 4
6
CO2
17 4 8 4 8
18 10 10
+ -
4.5 + H3O OH
19 12 H3O OH
- 12
4.5
14
20 5
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
5
21 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
22 L(mol CO2/mol MEA) L(mol CO2/mol MEA)
23 (a) 293.15 K (b) 313.15 K
24
25
26
0.5 0.5
27 +
28
+ MEA/RNH3
RNH3 / MEA
1 1
29
30 1.5 +
RNH3
31 1.5
- +
RNHCOO RNH3 -
RNHCOO
32 2
-log10(x)
-log10(x)
33 MEA 2
34 2.5
35 HCO3
- 2.5
-
3 2 2
36 4 CO2
HCO3
4
CO2 MEA
37 6 H3O
+
3 6 +
3.5 H3O
38 8
10 8
39 4 12 OH
-
3.5 10 OH
-
40 14
0 0.2 0.4 0.6 0.8 1 12
0 0.2 0.4 0.6 0.8 1
41 4.5 4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
42 L(mol CO2 /mol MEA) L(mol CO2/mol MEA)
43
44 (c) 333.15 K (d) 353.15 K
45
46 Figure 2: Mole fractions of the major solution species as a function of CO2 loading, L, at the
47 indicated temperatures and P = 1 bar for a 30 weight % aqueous MEA solution. Curves are
48 our simulation results, with error bars indicating their uncertainties (one standard deviation).
49 The main graphs show the major species and the insets show the minor species. Experimental
50
51
data points, whose symbol colors match the corresponding curves, are as follows: filled
52 diamonds: Matin et al. 11 (at 21◦ C); filled circles: Böttinger et al.; 8 open downward triangles:
53 du Preez et al.; 16 filled upward triangles: Jakobsen et al.; 7 open upward triangles: Jakobsen
54 et al., 7 293.15 K and 313.15 K only, indicating the sum of the mole fractions of HCO3 and
55 CO−2 9 −2
3 ; filled squares: Hilliard, indicating the sum of the mole fractions of HCO3 and CO3 .
56
57
58
59 42
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15 4
16
17
18
19
20 2
21
log10(PCO2)/kPa
22 T = 353.15 K
23
24
25
0
T = 313.15 K
26
27
28
29 -2 T = 298.15 K
30
31
32
33 T = 333.15 K
34 -4
35
36
37
38
39 -6
40
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
41 L(mol CO2/mol MEA)
42
43
44
45 Figure 3: Partial pressure of CO2 above the solution as a function of loading, L, at the
46
47
indicated temperatures and P = 1 bar for a 30 weight % aqueous MEA solution. Curves
48 are simulation results of this work, and points are experimental data from Jou et al. 5 The
49 error bars on the simulation curves at 298.15 K and 353.15 K indicate their uncertainties
50 (one standard deviation); the error bars at the other temperatures are not shown, but are
51 of similar magnitude.
52
53
54
55
56
57
58
59 43
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 44

Naroozi

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Naroozi

Uploaded by

Copyright:

Available Formats

Subscriber access provided by UNIV AUTONOMA DE COAHUILA UADEC

A: New Tools and Methods in Experiment and Theory

Javad Noroozi, and William Robert Smith

is published by the American Chemical Society. 1155 Sixteenth Street N.W.,

You might also like