New Linear Partitioning Models Based on Experimental Water

Subscriber access provided by UNIV OF CALIFORNIA SAN DIEGO LIBRARIES

Article

New Linear Partitioning Models based on Experimental Water – Supercritical CO2 Partitioning Data of Selected Organic Compounds Aniela Burant, Christopher J. Thompson, Gregory V. Lowry, and Athanasios K Karamalidis Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.6b00301 • Publication Date (Web): 15 Apr 2016 Downloaded from http://pubs.acs.org on April 18, 2016

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 free 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 accessible to all readers and 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

Environmental Science & Technology 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.

Subscriber access provided by UNIV OF CALIFORNIA SAN DIEGO LIBRARIES

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.

Environmental Science & Technology 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 22

Supercritical CO2

Anisole

Environmental Science & Technology

Thiophene

Pyrrole

Water

ACS Paragon Plus Environment


1 2

New Linear Partitioning Models based on Experimental Water – Supercritical CO2 Partitioning Data of Selected Organic Compounds

3 4

Aniela Burant,1 Christopher Thompson,2 Gregory V. Lowry,1 and Athanasios K. Karamalidis1* 1Department

5 6

of Civil and Environmental Engineering, Carnegie Mellon University, Pittsburgh, PA 15213.

7

2Pacific

8 9

*Corresponding Author: Address: 123L Porter Hall Civil & Environmental Engineering Carnegie Mellon University 5000 Forbes Avenue, Pittsburgh, PA 15213-3890

Northwest National Laboratory, Richland, WA 99352.

10

Phone Number: 412.268.1175

11

Fax Number: 412.268.7813

12

Email address: [email protected]

13

Abstract

14

Partitioning coefficients of organic compounds between water and supercritical CO2

15

(sc-CO2) are necessary to assess the risk of migration of these chemicals from subsurface

16

CO2 storage sites. Despite the large number of potential organic contaminants, the current

17

data set of published water-sc-CO2 partitioning coefficients is very limited. Here, the

18

partitioning coefficients of thiophene, pyrrole, and anisole were measured in situ over a

19

range of temperatures and pressures using a novel pressurized batch reactor system with

20

dual spectroscopic detectors: a near infrared spectrometer for measuring the organic

21

analyte in the CO2 phase, and a UV detector for quantifying the analyte in the aqueous

22

phase. Our measured partitioning coefficients followed expected trends based on volatility

23

and aqueous solubility. The partitioning coefficients and literature data were then used to

24

update a published poly-parameter linear free energy relationship and to develop five new

25

linear free energy relationships for predicting water-sc-CO2 partitioning coefficients. Four

26

of the models targeted a single class of organic compounds. Unlike models that utilize

27

Abraham solvation parameters, the new relationships use vapor pressure and aqueous

28

solubility of the organic compound at 25 °C and CO2 density to predict partitioning

29

coefficients over a range of temperature and pressure conditions. The compound class 1 ACS Paragon Plus Environment

Page 2 of 22

Page 3 of 22


30

models provide better estimates of partitioning behavior for compounds in that class than

31

the model built for the entire dataset.

32 33

Introduction

34

Carbon capture, utilization, and storage (CCUS) encompasses capturing CO2 from

35

point sources of emission, utilizing the CO2 in a process such as enhanced oil recovery

36

(EOR), and safely depositing the CO2 in underground formations for long term storage.

37

CCUS is considered a viable and economical technology due to the use of CO2 for EOR. As

38

the discovery of and production from conventional oil wells decreases1 and the cost of

39

anthropogenic CO2 decreases, CO2-driven EOR will likely increase.2 CO2-EOR involves the

40

injection of supercritical CO2 (sc-CO2) into oil formations to increase the amount of crude

41

oil recovered. CO2-flooding, as it is typically called, is used for light to medium crude oils.

42

sc-CO2 dissolves in the oil, making the oil less viscous and therefore more extractable. Low

43

molecular weight petroleum hydrocarbons being soluble in sc-CO2, increase the

44

extractability of crude oil.

45

CO2 injected into the aforementioned formations will exist as a supercritical fluid

46

due to the increased temperatures (>304 K) and pressures (>75 bar) encountered at

47

typical reservoir depths (> 800 m).3 Under those conditions, sc-CO2 is an excellent solvent

48

for organic compounds with high volatility and low aqueous solubility.4 However, there

49

are a limited number of partitioning coefficients for organic compounds of interest, e.g.

50

benzene, toluene, ethylbenzene and xylenes (BTEX), polycyclic aromatic hydrocarbons

51

(PAHs), and nitrogen, sulfur, and oxygen (NSO) containing organic compounds, including

52

phenols, carboxylic acids, pyrroles/pyridines, organosulfurs, and other polar organics

53

typically found in oilfield brines.5 Partitioning coefficients are needed as inputs to reactive

54

transport models, such as STOMP and TOUGHREACT,6,7 which can be used to predict what

55

compounds may transport with CO2 if leakage occurs.

56

The partitioning coefficients of organic compounds between water and sc-CO2 are

57

dependent on pressure, temperature, and the concentrations of salts. The ability to

58

measure or predict partitioning coefficients for petroleum-related compounds over a range 2 ACS Paragon Plus Environment


Page 4 of 22

59

of temperature, pressure, and salinity conditions typical of oil and gas reservoirs is needed

60

for modeling because reservoir conditions are site-specific and can vary vastly; with

61

measured temperatures up to 423 K, pressures up to 500 bar, and total dissolved solids

62

concentrations up to 350 g/L.8 These partitioning coefficients are difficult to measure

63

accurately, especially for volatile organic compounds, due to the high pressures and

64

temperatures required and the corresponding potential for artifacts in the measurements

65

(e.g. losses during depressurization). Consequently, the available partitioning data are

66

sparse (partitioning coefficients have been reported for only ~37 organic compounds),4

67

and typically are not available over the entire range of temperature and pressures of

68

interest. Notably, partitioning coefficients of organic compounds with NSO moieties are

69

necessary to increase the breadth of compound classes in the available predictive models,

70

which will allow for more accurate predictions.

71

For organic compounds that have no reported partitioning coefficients, models that

72

are able to estimate these values based on their physicochemical properties are necessary

73

for providing input parameters into reactive transport models for risk assessment

74

purposes. Equations of state, such as Soave-Redlich-Kwong and Peng-Robinson, are often

75

used for the interpolation of partitioning coefficients between data points. However, they

76

are not generally used to predict partitioning coefficients of new compounds because they

77

require empirically derived (from data) binary interaction parameters to correct model

78

data for experimental data. Another type of model used for estimating partitioning

79

coefficients of organic compounds from water to sc-CO2 is a poly-parameter linear free

80

energy relationship (pp-LFER). Timko, et al.,9 building on Lagalante and Bruno’s work,10

81

developed a pp-LFER (Eqn. 1) that incorporates a CO2 density term. This allows the pp-

82

LFER to be used over a range of temperatures and pressures for prediction of partitioning

83

coefficients of organic compounds: log K i , c / w = 3 .810 − 1 .230 π 2 − 3 .110α 2 − 2.010 β 2 + 0 .110V 2 + 2 .450 π 1

Eqn. 1.

84 85

This model requires inputs of Abraham solvation parameters, which include: R 2 , the

86

index of refraction of the organic compound; π 2 , the polarizability of the organic compound; 3 ACS Paragon Plus Environment

Page 5 of 22


87

α the hydrogen bonding acidity value of the organic compound; β , the hydrogen bonding

88

basicity value of the organic compound; and V2 , the McGowan’s molar volume of the

89

organic compound, and a CO2 polarizability term ( π 1 ), calculated from the CO2 density.9,11

90

The calculation of π1 requires an additional calculation, based on the reduced density ( ρ r )

91

of CO2 (Eqn. 2), which is a function of the density at the temperature and pressure of

92

interest ( ρ ) and the critical density of CO2 ( ρ ).12

2

2

T ,P

C

ρ =

ρ

Eqn. 2.

T ,P

ρ

r

c

π = 1.15ρ − 0.98

( ρ r < 0.7)

Eqn. 3.

π = 0.173ρ − 0.37

( ρ r > 0.7)

Eqn. 4.

1

1

r

r

93 94

This pp-LFER (Eqn. 1) was developed based on partitioning coefficients of 33

95

compounds (providing 332 data points), with R2adj = 0.88 and an average absolute

96

deviation (AAD) of the log values of the partitioning coefficients of 0.29. A root mean

97

square error (RMSE) was not reported.

98

This study has two objectives: 1) measure partitioning coefficients over a range of

99

temperatures and pressures for three NSO-containing compounds, thiophene, pyrrole and

100

anisole; and 2) improve and develop LFERs for more accurate water – sc-CO2 partitioning

101

coefficients predictions. These compounds were selected for study because they contain

102

key nitrogen, sulfur, and oxygen moieties present in organics found in oil & gas reservoirs

103

including. Thiophene and pyrrole are structural analogs, which provides a comparison of

104

how those a S or N moietiy results in differences in partitioning behavior. Thiophene and

105

pyrrole are also found in oil and gas reservoirs, and are precursors for larger compounds,

106

such as dibenzothiophene, or carbazole. In addition, thiophene has a relatively high vapor

107

pressure, and relatively low aqueous solubility, while pyrrole has a relatively high aqueous

108

solubility, and lower vapor pressure. Anisole was chosen because it is less volatile and has

109

lower aqueous solubility than both thiophene and pyrrole. These compounds therefore

110

provide a range of volatility and aqueous solubility to assess how these properties 4 ACS Paragon Plus Environment


Page 6 of 22

111

influence their water-sc-CO2 partitioning. For the second objective of this study, the pp-

112

LFER model described above was updated with new data from this study, along with

113

literature data that has been reported since the publication of the original pp-LFER. New

114

pp-LFERs, based on inputs of vapor pressure, aqueous solubility, and CO2 density, were

115

developed and trained using available literature values (hundreds of measurements for 37

116

compounds) and data from this study.

117

Materials and Methods

118

Carbon dioxide was supercritical fluid chromatography grade (99.999%) from

119

Matheson Tri-Gas. Thiophene (≥99%) was obtained from Sigma Aldrich. Pyrrole (99%) and

120

anisole (99%, extra dry) were obtained from Fisher Scientific. Water used in this study was

121

treated with a Thermo Scientific Barnstead water purification system and had a resistivity

122

of 18.2 MΩ-cm.

123

Partitioning coefficients here are given as a ratio of the mole fraction of organic

124

compound of interest ( i ) in the sc-CO2 phase ( y ) and aqueous phase ( x ) (Eqn 5). They

125

are calculated as functions of the mole concentration of the organic compound in both the

126

CO2 and H2O phases, respectively ( CCO and C H O ), as well as the densities ( ρ CO and ρ H O )

127

and molecular weights (MW) of the pure fluids.

2

K i ,c / w =

2

yi CCO ρ H O MWCO = ⋅ ⋅ xi CH O ρCO MWH O 2

2

2

2

2

2

2

Eqn. 5

2

128 129

Partitioning coefficients were measured in-situ using a custom batch reactor system

130

equipped with dual spectroscopic detectors (Figure 1). The apparatus is described in detail

131

elsewhere by Bryce et al. (in review).13 In brief, the system is comprised of a titanium Parr

132

reactor, near-infrared and UV spectroscopic detectors, high-pressure pumps, and tubing

133

and switching valves that enable quantitative injection of organic liquids into the

134

pressurized reactor. The reactor has an internal volume of approximately 121 mL and is

135

fitted with quartz windows on opposite sides for optical measurements. Organic

136

concentrations in the CO2-rich phase are measured by a Bruker IFS 66/S FT-IR 5 ACS Paragon Plus Environment

Page 7 of 22


137

spectrometer equipped with a tungsten source and a silicon diode detector for near

138

infrared (NIR) measurements. Concentrations in the aqueous-rich phase are measured by

139

circulating fluid from the bottom of the reactor past a Gilson model 151 UV detector fitted

140

with a high-pressure flow cell. An organic liquid reagent can be injected into the reactor by

141

filling a small loop of PEEK tubing with the reagent, switching a valve to place the loop in-

142

line with the reactor, and using CO2 supplied by the syringe pump to force the organic into

143

the reactor.

144 145 146

Figure 1. Schematic of the experimental apparatus used to measure partitioning coefficients.

147

Protocols for calibrating the detectors and measuring partitioning coefficients are

148

described in Bryce et al.;13 only limited details are provided here. A measured volume (21-

149

22 mL) of water was added to the reactor, and the system was pressurized with CO2 using

150

the syringe pump. After the pressure and temperature have stabilized, the organic

151

compound of interest was titrated into the reactor. NIR spectra were then collected every 5

152

minutes until both the NIR spectra and UV signal had stabilized, indicating that equilibrium

153

been reached. Spectra were recorded over the range 12,000 – 5,000 cm-1, and 128 scans

154

were co-added for each spectrum. A linear baseline correction was applied to all NIR

155

spectra by subtracting the a line fit through the average absorbances between 6255-6282

156

cm-1 and 5736-5752 cm-1. Temperature, pressure, and UV absorbance data were

157

continuously collected during the entire partitioning experiment. Equilibrium partitioning

158

of compounds between the two phases was typically achieved in two hours (See supporting

159

information, SI 1). 6 ACS Paragon Plus Environment


Page 8 of 22

160

The two spectroscopic detectors were calibrated independently with each organic

161

compound of interest before partitioning coefficients were measured. NIR-spectral

162

calibrations for the organic compound in sc-CO2 were performed at each temperature and

163

pressure of interest (i.e., six calibration curves per organic compound). This was done by

164

adding a small amount of water (~0.2 mL; enough to fully saturate the CO2 with water

165

according to the Spycher, et al.14 model for predicting aqueous solubility in sc-CO2) to the

166

reactor, repeatedly injecting organic into the reactor, and measuring the NIR absorbance

167

after each injection. The UV detector was calibrated by using a glass syringe to flow

168

aqueous standards through the UV detector's flow cell at the temperatures of interest.

169

Wavelengths used for the UV measurements were 225 nm for thiophene, 205 nm for

170

pyrrole, and 269 nm for anisole. All calibration curves contained 4-5 points and were linear

171

over the entire concentration range of interest.

172

Masses of pure organic compound injected into the reactor were chosen to avoid

173

creation of a third, neat organic phase. To do this, organic compound concentrations after

174

injection were kept to less than ~30% of the aqueous solubility. This corresponded to

175

maximum aqueous phase concentrations of 60 mg/L for thiophene, 870 mg/L for pyrrole,

176

and 120 mg/L for anisole. The aqueous calibration ranges for each of the organic

177

compounds, along with their aqueous solubilities are given in Table 1.

178 179

Table 1. Properties and Ranges of Concentrations for Organic Compounds in the Partitioning Experiments Organic Compound

MW (g/mol)

Melting Point (⁰C)

Thiophene Pyrrole Anisole

84.1 67.1 108.1

-39.4 -23.4 -37.3

log VP at 25 ⁰C (Pa)

Aqueous Solubility (g/L)

4.0215 3.0415 2.3118

316 47.517 1.718

KH (Pa L g-1)

3490 23 120

Range for Aqueous Calibration Curve (g/L) 0 – 0.37 0 – 1.7 0 – 0.18

180 181

Molar concentrations of organic compounds in the partitioning experiments were

182

calculated using the equilibrium absorbance measurements and the appropriate

183

calibration curves for each phase. These concentrations were then used to calculate the

184

mole fraction partitioning coefficients. The densities of CO2 were calculated using the Span

185

and Wagner EOS,19 which is available online (http://webbook.nist.gov/chemistry/fluid/) 7 ACS Paragon Plus Environment

Page 9 of 22


186

The Span and Wagner EOS is endorsed by the United States’ National Institutes of

187

Standards and Technology (NIST) and can calculate data for CO2 density over a range of

188

temperatures and pressures.20

189

Duplicate partitioning experiments were conducted for every temperature and

190

pressure point, and the data show a high degree of reproducibility. The pressure variability

191

was relatively low, with a mean standard deviation between duplicates of ±1.3 bar, ranging

192

from ±0.5 – 3.5 bar. Temperature measurements were within ±1 ⁰C. Standard deviations of

193

the measured partitioning coefficients were therefore relatively low as indicated by the

194

small error bars in Figure 2, which portrays the partitioning coefficients determined in this

195

study.

196

Modeling Methods

197

Multi-parameter LFERs in this paper were all developed using ordinary least

198

squares regression techniques using data from the literature and this study. The models

199

were validated using parameter uncertainty using repeated k-fold cross-validation (CV).21

200

The data are randomly divided into k-blocks (these are the partitioning coefficients and

201

inputs) of nearly equivalent size. One block is randomly withheld from each linear

202

regression – this block becomes the test set, and the models are then trained on the

203

remaining, k-1, blocks. There were 10 k-blocks used in this model. The predictive accuracy

204

(root mean square error, RMSE) of the model is reported here using the test set (i.e.

205

excluded data). This is then repeated 30 more times to account for the relatively small

206

sample size, and those are averaged to determine the final parameters of the model, as well

207

as the RMSE and R2adj that are reported in Table 3. This leads to 300 parameter estimates,

208

which are averaged and reported. CV calculations were performed using MATLAB R2013A.

209

Akaike information criteria (AIC, Eqn. 6) and Bayesian information criteria (BIC.

210

Eqn. 7) were implemented here to accurately compare models. AIC = 2k − 2 ln(σˆ e2 )

Eqn. 6

BIC = n ⋅ ln(σˆ e2 ) + k ln(n)

Eqn. 7

8 ACS Paragon Plus Environment


211 212

Page 10 of 22

Here k is the number of predictors in the model, including the intercept, σˆ e2 is the error variance of the model, and n is the number of observances in the model.

213

RMSE and R2 cannot be used to compare models with different parameters. The

214

model that minimizes the AIC and BIC is considered the best model, because that model

215

achieves the least amount of error with the least number of predictive variables, i.e. it

216

penalizes complex models. BIC is considered a harsher penalty than the AIC, because it also

217

accounts for sample size.

218

Results and Discussion

219

Experimental

220

Results of the partitioning experiments of the organic compounds from water to sc-

221

CO2 are shown in Table 2, along with the temperatures, pressures, and CO2 densities in

222

which these partitioning coefficients were measured. The pp-LFER predictions for the

223

partitioning coefficients are also listed and are discussed below.

224

Similar to air-water partitioning, the partitioning of these compounds between sc-

225

CO2 and water will depend on both their volatility and aqueous solubility. Higher volatility

226

and lower aqueous solubility will favor partitioning to sc-CO24, whereas lower volatility and

227

higher aqueous solubility will favor partitioning to the aqueous phase.

228

compounds of interest in this study, thiophene, pyrrole, and anisole, were chosen to span a

229

range of vapor pressure and aqueous solubility (Table 1). This in turn provides

230

approximately a two order of magnitude range of air-water partition coefficients at 25 ºC,

231

KH (Table 1), that should also trend with the sc-CO2-water partition coefficients.

The organic

232

The results show that the partitioning behavior followed expected trends based on

233

volatility and aqueous solubility. Thiophene had the highest partitioning coefficients for the

234

range of temperatures and pressures measured; while pyrrole had the lowest due to its

235

high aqueous solubility. Anisole, which has comparably lower volatility and aqueous

236

solubility than thiophene and pyrrole, had sc-CO2 partitioning coefficients in between the

237

two, just comparable to air-water partitioning.


Page 11 of 22


238

Thiophene had comparable experimental water-sc-CO2 partitioning coefficients to

239

benzene, which is similar to thiophene in terms of magnitude of vapor pressure and

240

aqueous solubility. Using the same in-situ batch reactor, Bryce, et al.,13 found that the

241

water-sc-CO2 partitioning coefficients were lower than previously reported water-sc-CO2

242

partitioning coefficients.22,23 This is likely due to the fact that benzene had volatile losses

243

from both phases in those previous partitioning experiments. Anisole and pyrrole, which

244

have lower vapor pressures, had comparable experimental partitioning coefficients to

245

similar compounds in terms of properties, (i.e. benzaldehyde and aniline)24 measured at

246

similar temperatures and pressures.

247

An isothermal increase in pressure led to an increase in partitioning from water to

248

sc-CO2 for each compound. An isobaric increase in temperature led to decrease in

249

partitioning from water to sc-CO2 for each compound; this is because an isobaric increase

250

in temperature leads to a decrease in CO2 density. In fact, temperature and pressure effects

251

can be attributed to changes in the CO2 density. Partitioning from water to sc-CO2 increased

252

with increasing CO2 density. The increase in partitioning was log-linear over the CO2

253

density range (~190 – 680 kg/m3) studied here. (Figure 2). The linear coefficient of

254

determination values (R2) for each compound’s log K

255

0.879, 0.967, and 0.949 for thiophene, anisole, and pyrrole, respectively.

i ,c / w

value versus CO2 density was

256



Page 12 of 22

257 258 259 260 261

Figure 2. Partitioning of thiophene, anisole, and pyrrole (this study) versus CO2 density (kg/m3). Masses of organic compound added to the system (total volume = 121 mL) ranged from 6 – 40 mg for thiophene, 20 – 73 mg for pyrrole, and 6 – 29 mg for anisole. This corresponded to maximum aqueous phase concentrations of 60 mg/L for thiophene, 1500 mg/L for pyrrole, and 120 mg/L for anisole.

262 263 264

Table 2. Experimental partitioning coefficients of organic compounds measured here over a range of temperatures and pressure, along with predictions from pp-LFER. Error is reported as average absolute deviation (AAD). Organic Compound

T (⁰C)

P (bar)

Thiophene

43 44 44 60 61 62

83±2.1 102±0.3 122±0.9 84±0.3 107±1.8 125±1.2

CO2 Density (kg/m3) 281±21 555.3±4 679.1±3 208.2±0.6 325.4±11.7 449.9±7.8

Pyrrole

43 44 44 61 62 62

86±0.7 102±0.5 121±1.6 81±3.5 110±0.6 121±3.4

312.7±17.7 547.9±6.5 673.8±6 191.2±12.1 338.2±3.9 415.6±2.6

Anisole

43 44 44 61 61 62

79±1.2 101±0.2 122±2 81±2.1 104±1.5 123±0.1

247.2±9.6 535.3±2.6 679.5±7.2 194.2±7.1 306.6±9.3 433±0.4

K (yi/xi) 47.1±0.5 109.8±7.6 133.6±7.4 34.4±6.0 38.6±12.1 94.1±2.7

log K

1.68±0.01 2.04±0.03 2.13±0.02 1.54±0.08 1.58±0.13 1.97±0.01 AAD = 0.83±0.04 -0.09±0.02 2.20±0.02 0.34±0.005 2.24±0.06 0.35±0.01 0.64±0.07 -0.19±0.05 1.01±0.003 0.00±0.001 1.19±0.07 0.08±0.02 AAD = 10.8±0.05 1.03±0.002 44.3±2.8 1.65±0.03 72.5±28.3 1.84±0.17 8.8±0.11 0.94±0.006 15.6±0.8 1.19±0.02 35.6±0.2 1.55±0.002 AAD = TOTAL AAD =

log K ppLFER 2.17 2.47 2.58 1.73 2.26 2.38 0.471 0.58 0.69 0.80 -0.15 0.50 0.57 0.439 1.47 1.96 2.09 1.15 1.76 1.87 0.383 0.431

265 266

Poly-parameter linear free energy relationship predictions

267

The pp-LFER model, introduced by Timko, et al.,9 was reasonably accurate at

268

predicting the partitioning coefficients of the organic compounds of interest in this study.

269

The average absolute deviation (AAD) for the organic compounds of interest in this study

270

was 0.431 log units for the pp-LFER. The pp-LFER for these 18 data points, consistently

271

over-predicted the water-sc-CO2 partitioning coefficients. The pp-LFER requires Abraham

272

solvation parameters, which are available for a variety of organic compounds11 but not all

273

compounds of interest. While some of the Abraham solvation parameters are easily

274

calculated, such as molar volume, others require experimental measurement. For example,


Page 13 of 22


275

the hydrogen bonding acidity and basicity values require measurements of the organic

276

compound with a reference base or acid (for acidity and basicity, respectively) in an apolar

277

solvent, typically tetrachloromethane.25–27

278

Update and Development of Models

279

The pp-LFER is capable of predicting partitioning coefficients; however, there is still

280

associated error with those model predictions (Table 2), and it requires Abraham

281

solvation parameters that may not be available for other organic compounds of interest. To

282

solve these problems, the pp-LFER was re-trained to reduce error in the model (ASP-LFER

283

section) and new LFERs were developed for organic compounds with no reported

284

Abraham solvation parameters (New LFERs section).

285

ASP-LFER

286

To reduce error in the pp-LFER, the model was re-trained to include new data from

287

the literature and this study. In addition, three compounds, a total of 36 data points that

288

had been used to develop the original pp-LFER were rejected because very high

289

concentrations of the organic were used, leading to the formation of a separate neat phase

290

and abnormal partitioning behavior. There are now 360 partitioning coefficients in the

291

training set of the model, there were originally 332 data points in the study. The new LFER

292

is termed ASP-LFER (Abraham Solvation Parameters-Linear Free Energy Relationship,

293

Table 3, Eqn. 8) to differentiate it from the previously produced pp-LFER. The new model

294

led to an increase in the adjusted R2 and a lower average absolute deviation (~0.23 log

295

units) for the ASP-LFER. The total AAD from the ASP-LFER fell to 0.358 log units for the

296

organic compounds in this study from the AAD of 0.431 log units for the original pp-LFER.

297

Repeated k-fold cross validation, described above (Modeling Methods section),

298

allows for every organic compound to be in both the training and test set. The RMSE

299

reported in Table 3 is average of all the reported errors of the test set (there were 300

300

iterations of the model, resulting in an average of the RMSEs reported in Table 3). The

301

RMSE is therefore the predictive accuracy of all the test sets. This is the magnitude of error

302

that can be expected for a new organic compound that has not been a part of the training 12 ACS Paragon Plus Environment


Page 14 of 22

303

set. Average absolute deviations are also reported for ease of comparison, since the

304

previously published pp-LFER only reports AAD.

305

New LFERs

306

A new type of LFER was developed that does not require Abraham solvation

307

parameters. All of the available partitioning coefficients, including data in this study were

308

used to train a new model with inputs of vapor pressure, aqueous solubility, and CO2

309

density. The new model, VP-AS-LFER (Eqn. 9), was trained with a 369 data point training

310

set. The range of vapor pressure and aqueous solubility of compounds in the training set,

311

and the range of CO2 density, and log K

312

in SI 5. For the VP-AS-LFER (Eqn. 9), the vapor pressures range over eight orders of

313

magnitude and the aqueous solubilities range over five orders of magnitude. This should

314

allow for prediction of partitioning coefficients for organic compounds with a variety of

315

different properties, including differing polarities and sizes. The temperature and pressure

316

dependent partitioning coefficients in the VP-AS-LFER (Eqn. 9) reflects the organic

317

compounds’ relative solubilities in both the CO2 and H2O phases and the inherent volatility

318

of the organic compound. This is similar to use than Eqn. 9 because vapor pressure and

319

aqueous solubility data is widely available or is easy to predicted.18,28,29

i ,c / w

for compounds in the training set is presented

320

To train the model, the vapor pressure and/or aqueous solubility for compounds

321

that have no reported experimental values were calculated using Advanced Chemistry

322

Development (ACD/Labs) Software V11.02.28 Nine compounds had at least one estimate of

323

vapor pressure and/or aqueous solubility found in the literature; these are listed in

324

supporting information (SI 4). The CO2 density term is calculated using the Span and

325

Wagner EOS19 for predicting CO2 phase behavior; described above. The overall fits for the

326

ASP-LFER and the VP-AS-LFER are given in Figure 3.

327

Four additional organic compound class specific VP-AS-LFERs (Eqns. 10 – 13) were

328

developed because prediction of partitioning coefficients from LFERs typically has better

329

agreement with experimental values if they use an organic compound class specific LFER.18

330

This included models for substituted monopolar benzenes, polar-substituted benzenes,


Page 15 of 22


331

chlorinated phenols, and nitrogen-containing compounds (i.e., aromatic/heterocyclic

332

nitrogen compounds) (Eqns. 10 – 13, Table 3). There was not enough data for the

333

development of LFERs for other compound classes. The range of vapor pressure, aqueous

334

solubility, and CO2 density used to develop these LFERs are provided in SI 5, in SI Tables 8 -

335

13. The RMSEs reported in Table 3 represent the magnitude of error that can be expected

336

for a new organic compound that has not been a part of the training set. The compound

337

class specific LFERs have relatively lower RMSE (and AAD), discussed in the Comparison

338

of Models section.

339 340

Table 3. Linear Free Energy Relationships for Predicting Water – Supercritical CO2 Partitioning Coefficients. Name

Formula

N

R2adj

RMSE

AAD

AIC

BIC

360

0.919

0.303

0.230

169

195

Eqn 8.

369

0.872

0.375

0.280

332

352

Eqn 9.

73

0.782

0.263

0.181

17

28

Eqn 10.

146

0.753

0.189

0.148

-64

-49

Eqn 11.

32

0.912

0.245

0.163

3

10

Eqn 12.

35

0.954

0.152

0.114

-26

-18

Eqn 13.

logKi,c / w = 1.31(±0.10) − 0.82(±0.08)R2 − 3.03(±0.07)α2 ASP-LFER

− 3.99(±0.09)β2 + 3.18(±0.13)V2 +1.85(±0.13)π1

log Ki ,c / w = −4.43(±0.29) + 0.43(±0.01) logVP25°C VP-AS-LFER

− 0.83(±0.02) log AS25°C + 1.34(±0.1) log ρCO

2

Monopolar Substituted Benzenes Polar Substituted Benzenes Chlorinated Phenols

log Ki ,c / w = −4.82(±0.45) + 0.43(±0.04) logVP25°C − 1.45(±0.11) log AS25°C + 1.10(±0.12) log ρCO

2


2

log K i ,c / w = −12.62(±1.78) + 0.77(±0.11) log VP25° C − 1.12(±0.11) log AS25° C + 3.99(±0.63) log ρ CO

2

NitrogenContaining Compounds


2

341

Comparison of LFERs

342

Any of the LFERs in Table 3 can be used depending on the organic compound of

343

interest and availability of the ASPs. The compound group-specific LFERs may be more

344

appropriate than the ASP-LFER or the VP-AS-LFER—the overall AICs and BICs are lower

345

for the specialized group LFERs. AIC and BIC both account for error and model complexity,

346

and BIC also accounts for number of observances in the training set, and models that



Page 16 of 22

347

minimize both values are considered more accurate models. The models that minimize AIC

348

and BIC have the best balance of goodness of fit and simplicity. Based on the AIC and BIC

349

analysis, the compound specific VP-AS-LFERs (Eqns. 10 – 13) should be used to predict

350

partitioning coefficients for organic compounds in those classes rather than either of the

351

LFERs created from the total data set.

352

compounds in one of those specific groups. Compounds not in any of the compound classes

353

may be estimated from either of the full LFERs. However, the AIC and BIC values are lower

354

for the ASP-LFER than the VP-AS-LFER, indicating that the ASP-LFER is preferred if the

355

ASPs are available. Since, ASPs are not available for all organic compounds, and the ASPs

356

must be determined experimentally, having all of these models available for estimation

357

widens the number of organic compounds that can have their water-sc-CO2 partitioning

358

coefficients predicted.

However, this is only applicable to organic

359

All of the LFERs in this study have the same dependence on temperature and

360

pressure, which is expressed in the π 1 in the ASP-LFER (Eqn. 8). π 1 is calculated using the

361

CO2 density (Eqns. 2 – 4) and is expressed in the CO2 density term in VP-AS-LFERs (Eqns. 9

362

– 13). An increase in CO2 density results in an increase in partitioning to sc-CO2. However,

363

temperature and pressure have opposing effects on CO2 density, and therefore on water-sc-

364

CO2 partitioning coefficients. An isothermal pressure increase leads to increased

365

partitioning to sc-CO2, while an isobaric temperature increase leads to a decrease in

366

partitioning to sc-CO2. These trends are reflected in these LFERs.

367

to the database of existing water-sc-CO2 partitioning coefficients with the measurement of

368

thiophene, pyrrole, and anisole, respectively in water-sc-CO2 systems over a range of

369

temperatures and pressures. Those measured partitioning coefficients highlighted trends

370

in water-sc-CO2 partitioning, with organic compounds with high volatility having higher

371

partitioning into the sc-CO2 phase. This also illustrated that partitioning of organic

372

compounds between water and sc-CO2 is partially controlled by CO2 density.

373

knowledge and the measurement of these partitioning coefficients has helped inform and

374

train new LFERs that can estimate partitioning coefficients without the use of equations of

375

state and binary interaction parameters.


This work has added

This

Page 17 of 22


376

Either the ASP-LFER or VP-AS-LFER can be used depending on the available

377

parameters for other compounds that do not belong in any of those classes of compounds.

378

This has important implications for reactive transport models for CO2-storage. The

379

available water-sc-CO2 partitioning coefficients or the LFER predicted values, from any of

380

the above LFERs, can be easily incorporated into reactive transport models for risk

381

assessment associated with CO2 storage. Since it is not feasible to make measurements for

382

every temperature and pressure point for every compound, these models can be used to

383

make predictions of levels of organic compounds transporting with leaking sc-CO2.

384 385



386

387

388 389 390 391 392 393 394 395 396

Figure 3. Predicted log partitioning coefficients of organic compounds coefficients versus log experimental partitioning coefficients of organic compounds from both this study and literature values for the pp-LFERs in this study: A) Abraham solvation parameter LFER, B) Vapor pressure and aqueous solubility LFER, C) Monopolar substituted benzene LFER, D) Polar substituted benzene LFER, E) Chlorinated phenol LFER, and F) Nitrogen containing compound LFER. The lines indicate the 1:1 fit, blue crosses are literature data points, and red diamonds indicate data measured in this study.


Page 18 of 22

Page 19 of 22


397

Acknowledgements

398

This work was funded under the National Energy Technology Laboratory’s former Regional

399

University Alliance. Part of this research was performed at the Environmental Molecular

400

Sciences Laboratory (EMSL), a national scientific user facility at PNNL that is managed by

401

the DOE's office of Biological and Environmental Research. PNNL is operated for DOE by

402

Battelle Memorial Institute under contract no. DE-AC05-76RLO-1830. Zheming Wang and

403

John Loring provided useful assistance with the FTIR. The Jared and Maureen Cohon

404

Fellowship in Environmental Engineering and the Bradford and Diane Smith Fellowship in

405

Engineering are acknowledged for additional support.

406

ASSOCIATED CONTENT

407

Supporting Information

408

Contains examples of time to equilibrium curves, calculated binary interaction parameters,

409

iterations of the ASP-LFERs, tables of the reported vapor pressures, aqueous solubilities,

410

CO2 densities, ASPs, experimental partitioning coefficients, and the new model predicted

411

partitioning coefficients for all compounds with reported water-sc-CO2 partitioning

412

coefficients, RMSEs for thiophene, pyrrole, and anisole, for each model, and the range in

413

parameters and partitioning coefficients for the new models. This material is available free

414

of charge via the Internet at http://pubs.acs.org.

415 416

References

417 418 419 420

(1)

Murray, J.; King, D. Climate policy: Oil’s tipping point has passed. Nature 2012, 481 (7382), 433–435.

421 422

(2)

DOE/EIA. Annual Energy Outlook 2014 with projections to 2040; 2014.

423

(3)

NETL. Carbon Utilization and Storage Atlas of the United States and Canada, 4th ed.; 18 ACS Paragon Plus Environment


424 425

Morgantown, WV, 2012.

426 427 428 429

(4)

Burant, A.; Lowry, G. V; Karamalidis, A. K. Partitioning behavior of organic contaminants in carbon storage environments: a critical review. Environ. Sci. Technol. 2013, 47 (1), 37–54.

430 431 432

(5)

Neff, J.; Lee, K.; Deblois, E. M. Produced Water : Overview of Composition , Fates , and Effects. Prod. Water 2011, 3–54.

433 434 435 436

(6)

Ward, A. L.; White, M. D.; Freeman, E. J.; Zhang, Z. F. STOMP: Subsurface Transport Over Multiple Phases. Version 1.0: Addendum: Sparse Vegetation, Evapotranspiration Model for the Water-Air-Energy Operational Mode. (PNNL-15465); 2005.

437 438 439 440 441

(7)

Xu, T.; Sonnenthal, E.; Spycher, N.; Pruess, K. TOUGHREACT—A simulation program for non-isothermal multiphase reactive geochemical transport in variably saturated geologic media: Applications to geothermal injectivity and CO2 geological sequestration. Comput. Geosci. 2006, 32 (2), 145–165.

442 443 444 445

(8)

Kharaka, Y. K.; Hanor, J. S. 5.16 - Deep Fluids in the Continents: I. Sedimentary Basins. In Treatise on Geochemistry; Holland, E.-C. H. D., Turekian, K. K., Eds.; Pergamon: Oxford, 2003; pp 1–48.

446 447 448 449 450

(9)

Timko, M. T.; Nicholson, B. F.; Steinfeld, J. I.; Smith, K. A.; Tester, J. W. Partition Coefficients of Organic Solutes between Supercritical Carbon Dioxide and Water: Experimental Measurements and Empirical Correlations. J. Chem. Eng. Data 2004, 49 (4), 768–778.

451 452 453 454

(10) Lagalante, A. F.; Bruno, T. J. Modeling the Water−Supercritical CO2 Partition Coefficients of Organic Solutes Using a Linear Solvation Energy Relationship†. J. Phys. Chem. B 1998, 102 (6), 907–909.

455 456 457 458

(11) Abraham, M. H.; Chadha, H. S.; Whiting, G. S.; Mitchell, R. C. Hydrogen bonding. 32. An analysis of water-octanol and water-alkane partitioning and the Δlog p parameter of seiler. J. Pharm. Sci. 1994, 83 (8), 1085–1100.

459 460 461

(12) Smith, R. D.; Frye, S. L.; Yonker, C. R.; Gale, R. W. Solvent properties of supercritical xenon and sulfur hexafluoride. J. Phys. Chem. 1987, 91 (11), 3059–3062.

462

(13) Bryce, D.; Shao, H.; Cantrell, K.; Thompson, C. Determination of Organic Partitioning 19 ACS Paragon Plus Environment

Page 20 of 22

Page 21 of 22

463 464 465


Coefficients in Water-Supercritical CO2 Systems by Simultaneous In Situ UV and Near-Infrared Spectroscopies. Environ. Sci. Technol. 2016, In Review.

466 467 468 469

(14) Spycher, N.; Pruess, K.; Ennis-King, J. CO2-H2O mixtures in the geological sequestration of CO2. I. Assessment and calculation of mutual solubilities from 12 to 100°C and up to 600 bar. Geochim. Cosmochim. Acta 2003, 67 (16), 3015–3031.

470 471 472 473

(15) Dearden, J. C. Quantitative structure-property relationships for prediction of boiling point, vapor pressure, and melting point. Environ. Toxicol. Chem. 2003, 22 (8), 1696– 1709.

474 475 476 477

(16) Valvani, S. C.; Yalkowsky, S. H.; Roseman, T. J. Solubility and partitioning IV: Aqueous solubility and octanol-water partition coefficients of liquid nonelectrolytes. J. Pharm. Sci. 1981, 70 (5), 502–507.

478 479 480 481

(17) Yaffe, D.; Cohen, Y.; Espinosa, G.; Arenas, A.; Giralt, F. A Fuzzy ARTMAP Based on Quantitative Structure−Property Relationships (QSPRs) for Predicting Aqueous Solubility of Organic Compounds. J. Chem. Inf. Comput. Sci. 2001, 41 (5), 1177–1207.

482 483 484

(18) Schwarzenbach, R. P.; Gschwend, P. M.; Imboden, D. M. Environmental Organic Chemistry, 2nd ed.; John Wiley & Sons, Inc.: Hoboken, NJ, 2003.

485 486 487 488

(19) Span, R.; Wagner, W. A New Equation of State for Carbon Dioxide Covering the Fluid Region from the Triple-Point Temperature to 1100 K at Pressures up to 800 MPa. J. Phys. Chem. Ref. Data 1996, 25 (6).

489 490 491

(20) National Institutes of Standards and Technology. Thermophysical Properties of Fluid Systems http://webbook.nist.gov/chemistry/fluid/.

492 493 494 495

(21) Kohavi, R. A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection. In International Joint Conference on Artificial Intelligence; 1995; Vol. 14, pp 1137–1143.

496 497 498

(22) Ghonasgi, D.; Gupta, S.; Dooley, K. M.; Knopf, F. C. Supercritical CO2 extraction of organic contaminants from aqueous streams. AIChE J. 1991, 37 (6), 944–950.

499 500 501

(23) Yeo, S.-D.; Akgerman, A. Supercritical extraction of organic mixtures from aqueous solutions. AIChE J. 1990, 36 (11), 1743–1747. 20 ACS Paragon Plus Environment


502 503 504 505 506

(24) Wagner, K.-D.; Brudi, K.; Dahmen, N.; Schmieder, H. Partition coefficients of aromatic organic substances in two-phase mixtures of water and carbon dioxide at pressures from 8 to 30 MPa and at temperatures of 313 to 333 K. Part II. J. Supercrit. Fluids 1999, 15 (2), 109–116.

507 508 509 510

(25) Abraham, M. H. Scales of solute hydrogen-bonding: their construction and application to physicochemical and biochemical processes. Chem. Soc. Rev. 1993, 22 (2), 73–83.

511 512 513

(26) Platts, J. A. Theoretical prediction of hydrogen bond donor capacity. Phys. Chem. Chem. Phys. 2000, 2 (5), 973–980.

514 515 516 517

(27) Devereux, M.; Popelier, P. L. A.; McLay, I. M. A refined model for prediction of hydrogen bond acidity and basicity parameters from quantum chemical molecular descriptors. Phys. Chem. Chem. Phys. 2009, 11 (10), 1595–1603.

518 519 520

(28) Substance Identifier, S. S. Advanced Chemistry Development (ACD/Labs) Software V11.02. Columbus, OH.

521 522

(29) Boethling, R. S.; Mackay, D. Handbook of Property Estimation Methods for Chemicals: Environmental and Health Sciences; CRC Press, 2000.

523


Page 22 of 22

New Linear Partitioning Models Based on Experimental Water

Recommend Documents