Subtle changes in hydrogen bond orientation result in glassification of carbon capture solvents

Article abstract of DOI:10.1039/d0cp03503c

Water-lean CO2 capture solvents show promise for more efficient and cost-effective CO2 capture, although their long-term behavior in operation has yet to be well studied. New observations of extended structure solvent behavior show that some solvent formulations transform into a glass-like phase upon aging at operating temperatures after contact with CO2. The glassification of a solvent would be detrimental to a carbon-capture process due to plugging of infrastructure, introducing a critical need to decipher the underlying principles of this phenomenon to prevent it from happening. We present the first integrated theoretical and experimental study to characterize the nano-structure of metastable and glassy states of an archetypal single-component alkanolguanidine carbon-capture solvent and assess how minute changes in atomic-level interactions convert the solvent between metastable and glass-like states. Small-angle neutron scattering and neutron diffraction coupled with small- and wide-angle X-ray scattering analysis demonstrate that minute structural changes in solution precipitae reversible aggregation of zwitterionic alkylcarbonate clusters in solution. Our findings indicate that our test system, an alkanolguanidine, exhibits a first-order phase transition, similar to a glass transition, at approximately 40 °C - close to the operating absorption temperature for post-combustion CO2 capture processes. We anticipate that these phenomena are not specific to this system, but are present in other classes of colvents as well. We discuss how molecular-level interactions can have vast implications for solvent-based carbon-capture technologies, concluding that fortunately in this case, glassification of water-lean solvents can be avoided as long as the solvent is run above its glass transition temperature.

Full text of DOI:10.1039/d0cp03503c

PCCP
View Article Online
View Journal
PAPER
Subtle changes in hydrogen bond orientation
result in glassification of carbon capture solvents†
Cite this:DOI: 10.1039/d0cp03503c
a
Jose Leobardo Banuelos,
Mal-Soon Lee,b Manh-Thuong Ngyuen,b
´
˜
Difan Zhang,b Deepika Malhotra,c David C. Cantu,b Vassiliki-Alexandra Glezakou,b
Roger Rousseau,b Thomas F. Headen,
Robert M. Dalgliesh,d
d
ce
be
b
David J. Heldebrant,
Steven R. Saunders
*
Trent R. Graham,
Kee Sung Han
and
e
Water-lean CO2 capture solvents show promise for more efficient and cost-effective CO2 capture,
although their long-term behavior in operation has yet to be well studied. New observations of
extended structure solvent behavior show that some solvent formulations transform into a glass-like
phase upon aging at operating temperatures after contact with CO2. The glassification of a solvent
would be detrimental to a carbon-capture process due to plugging of infrastructure, introducing a
critical need to decipher the underlying principles of this phenomenon to prevent it from happening.
We present the first integrated theoretical and experimental study to characterize the nano-structure of
metastable and glassy states of an archetypal single-component alkanolguanidine carbon-capture
solvent and assess how minute changes in atomic-level interactions convert the solvent between
metastable and glass-like states. Small-angle neutron scattering and neutron diffraction coupled with
small- and wide-angle X-ray scattering analysis demonstrate that minute structural changes in solution
precipitae reversible aggregation of zwitterionic alkylcarbonate clusters in solution. Our findings indicate
that our test system, an alkanolguanidine, exhibits a first-order phase transition, similar to a glass
transition, at approximately 40 1C—close to the operating absorption temperature for post-combustion
CO2 capture processes. We anticipate that these phenomena are not specific to this system, but are
present in other classes of colvents as well. We discuss how molecular-level interactions can have vast
implications for solvent-based carbon-capture technologies, concluding that fortunately in this case,
glassification of water-lean solvents can be avoided as long as the solvent is run above its glass
transition temperature.
Received 30th June 2020,
Accepted 10th August 2020
DOI: 10.1039/d0cp03503c
rsc.li/pccp
Earth ‘‘well below’’ a 1.5 1C above pre-industrial level warming
scenario to minimize the effects of a warming climate on
Introduction
The Earth has warmed by approximately 1.0 1C since the human life.1 Ultimately, negative emission technologies will
industrial revolution due to anthropogenic CO2 emissions. be needed to remove the CO2 already in the atmosphere, but
As global emissions continue to rise, warming is expected to before those technologies come online, post-combustion carbon
reach 2 1C by the end of this century. The recent report from the capture is needed to stop future anthropogenic emissions from
Intergovernmental Panel on Climate Change (IPCC) identifies increasing even further and exhausting Earth’s carbon budget.2
carbon capture and sequestration as critical to keeping the Post-combustion capture technologies include solvents,
membranes, and porous materials.3 Regardless of technology,
the ultimate driver that has limited widespread adoption of
a The University of Texas at El Paso, El Paso, Texas 79968, USA
post-combustion carbon capture and storage (CCS) is cost. CCS
is still far from affordable; commercial technologies cost at
least $58 per tonne CO2.4 The U.S. Department of Energy
has set an ambitious cost target of $40 per tonne CO2 for
post-combustion technologies by the year 2030. Solvent-based
technologies are the most mature post-combustion CCS technol-
ogy and are expected to be the quickest to market as they are
b Physical Sciences Division, Pacific Northwest National Laboratory, Richland,
Washington 99352, USA
c Energy Processes and Materials Division, Pacific Northwest National Laboratory,
Richland, Washington 99352, USA. E-mail: David.Heldebrant@pnnl.gov
d ISIS Neutron and Muon Source, Rutherford Appleton Laboratory,
Harwell Campus, Didcot OX11 0QX, UK
e The Gene and Linda Voiland School of Chemical Engineering and Bioengineering,
Washington State University, Pullman, WA, 99164, USA
† Electronic supplementary information (ESI) available. See DOI: 10.1039/d0cp03503c undergoing testing at large-scale demonstrations.5–10
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
Paper
PCCP
The main energetic driver for solvent-based technologies We recently showed that such a heterogeneous structure could
is the energy-intensive thermal regeneration of chemically- be responsible for the anomalously high CO2 transport in many
selective solvents. This energy can be reduced by lowering the water-lean solvents,24,25 as the disparate regions of ionic clusters
water content of the solvent as it minimizes energy lost due to and aggregates provides pockets or channels within the fluid
boiling and condensing of water. The past decade has seen a where CO2 can readily diffuse and react with CO2-free solvent.17
shift in solvent formulation toward ‘‘water-lean’’ solvents that Ultimately, the nanoscale heterogeneity of the solvent structure is
comprise similar formulations to first- and second-generation a double-edged sword and needs to be fully characterized so that
aqueous amines, albeit with as little water as possible. it can be controlled.
All formulations of water-lean solvents are either (1) organic
In this work, we employ new spectroscopic measurements
bases that chemically react with CO2 to form either carbamate and computations at length scales above and below previous
or alkylcarbonate ionic liquids or (2) solid/liquid carbamate or studies to thoroughly measure the changes in the mesoscopic
alkylcarbonate salts dissolved in co-solvents.11,12 The low-water (4nm, omm) molecular structure of alkanol guanidines as a
content in these solvents enables lower reboiler duties than function of CO2 loading and temperature. We present a com-
first- and second-generation aqueous amines, making them bined experimental (neutron, X-ray) and theoretical study of
potentially more efficient,11,13 though some recent lost-work 1-IPADM-2-BOL as an archetypal water-lean solvent and assess
analyses have questioned the performance benefits of water- how changes in the mesoscopic structure occur. We also
lean solvents using first-generation CCS infrastructure when present evidence for a previously undiscovered glass transition
compared against second-generation amines using 21st century within 1-IPADM-2-BOL (and two other alkanolguanidine deri-
configurations.14
vatives), indicating that this behavior may be common to other
In the absence of water as a co-solvent, water-lean solvents water-lean formulations. We close with a discussion of how
have been shown to exhibit unique physical and thermo- minute changes in hydrogen bond partners within the fluid can
dynamic properties that may be exploitable for efficiency or influence the viability of these solvents for CO2 capture in
cost gains.11 Properties exploited for efficiency gains include: addition to ways to mitigate the potential for catastrophic
polarity-swing-assisted regeneration to reduce reboiler tempera- plugging of a CCS unit during operation.
tures by as much as 72 1C; higher CO2 solubility to drive faster
absorption kinetics; and phase changes to concentrate the
CO2-rich solvent so less exogenous material is heated during
Results and discussion
Viscosity anomalies
regeneration.15,16 Not surprisingly, not all physical and thermo-
dynamic properties of these solvents are beneficial to performance;
thus, understanding the molecular-level behavior of these solvents Our first molecular simulations22,26 predicted the transition
is critical to maximizing their full potential while minimizing any of free-flowing 1-IPADM-2-BOL and other alkanolguanidine
limiting behaviors.
liquids into what is best described as a ‘‘glassy state.’’ Initial
Our group has spent the past few years studying how and simulations were performed at a representative ‘‘CO2-lean’’ of
why these properties arise in concentrated solvents such as 0.25 mol CO2 per mol solvent and a ‘‘CO2-rich’’ solvent loading
CO2-binding organic liquids (CO2BOLs). These solvents are of 0.50 mol CO2 per mol solvent to span the solvent’s expected
single-component systems comprised of alkanolguanidine working range in a working process. Initial simulations of these
chemical moieties. Our recent studies have linked the kinetic two loadings showed higher viscosity values than experimental
and thermodynamic properties of CO2BOLs to a heterogeneous measurements of 1-IPADM-2-BOL. At the time, there had
(disparate ionic/non-ionic domain) structure that begins forming been no prior evidence (visual or measured) of glassification of
upon CO2 fixation.17,18 1-((1,3-Dimethylimidazolidin-2-ylidene)- 1-IPADM-2-BOL at B40 1C during continuous flow CO2-
amino)propan-2-ol (1-IPADM-2-BOL) is an archetypal alkanol- absorption/desorption testing in our lab.27 Our original obser-
guanidine derivative that we have studied extensively, noting vations of the glass-like appearance of 1-IPADM-2-BOL at room
that the molecular-level structure17 is markedly different temperature, was attributed to the temperature dependence of
than that of non-ionic liquids (e.g., alcohols) and ionic liquids viscosity of CO2BOLs and other water-lean solvents.28–32 Thus,
(e.g., [bmim][BF4]).19,20 Recently, the heterogeneous solvent it was not clear why simulations were predicting glassification
structure was experimentally confirmed using two- and three- at 40 1C when it had not been measured experimentally.
dimensional speciation mapping of 1-IPADM-2-BOL measured
To ascertain the nature of this dual state of the solvents, we
from in situ liquid state secondary ion mass spectrometry synthesized a fresh batch of 1-IPADM-2-BOL and subjected it to
(SIMS) data.19 Molecular dynamic (MD) simulations have shown measurements using a custom pressure–volume temperature
that the formation of strong hydrogen bond networks and clusters (PVT) cell described previously.18 The cell that was specifically
of zwitterionic carbonate greatly impact viscosity of the CO2-rich designed to enable simultaneous vapor–liquid equilibria (VLE)
solvents, limiting their performance with respect to heat exchange was used for viscosity and kinetics measurements as a function
and pumping.16,21,22 Alternately, the judicious placement of these of CO2 loading. 1-IPADM-2-BOL was tested at 40 1C (the esti-
hydrogen bonds, through careful solvent design, may also reduce mated absorption temperature in a working process33), with
the viscosity, by containing the hydrogen bonding within the one hour between CO2 injections to allow the solvent to
molecule,22 or by favoring a high population of neutral species.23 reach equilibrium. Interestingly, the viscosity of 1-IPADM-2-BOL
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
PCCP
Paper
pressure of CO2 (P*) (Fig. 1, and Table S1, ESI†) indicating this
phenomenon may be more common to alkanolguanidines than
previously known.
In earlier wok, we reported that the viscosity increase in
alkanolguanidine solvents is linked to the orientation of hydro-
gen bonding in solution, but we had never seen samples transi-
tion to a glassy state during measurements.18,26 The discovery of
a second, more viscous, viscosity profile with no associated
change in P* suggests that there must be a transformation in
the molecular-level ordering within the fluid. We had previously
observed 4500 nm aggregates of zwitterionic guanidinium
carbonates for 1-IPADM-2-BOL using two- and three-dimensional
time-of-flight (ToF)-SIMS measurements, albeit at room
temperature.17 It was not clear whether such clustering con-
tributed to the original or glassy-state viscosity profile. Further,
it was unclear how and why these large aggregates form in
solution, and perhaps even less clear why the glassification was
being observed. As indicated above, initial simulations had not
yet observed changes in longer range order, which may be
attributed to the simulation boxe size (B8 Â 8 Â 8 nm). Thus,
little was known about how the structural features within the
solvent at length scales greater than our box size could be
Fig. 1 Viscosity profile for the metastable and glassy state for three
alkanolguanidines as a function of CO2 loading. Each derivative is repre-
sented in the same color as either circles (G = glassy) or diamonds
(M = metastable).
(Fig. 1, yellow circles, and Table S1, ESI†) was greater than our impacting the physical and thermodynamic properties of the
previously reported viscosity profile34 and data for the solvent solvent. Scattering methods that could identify such changes
in continuous-flow operation. The higher viscosity profile was in the mesoscopic structure were needed to decipher how
peculiar, as the viscosities for all three measurements were structural changes within the fluid were occurring and assess
performed on the same viscometer at the same temperature on which structural features were converting these three alkanol-
the same solvent. The difference in viscosity profiles was initially guanidines into a newly discovered ‘‘glassy state.’’
thought to be due to different CO2 loadings, though the accom-
Neutron diffraction and scattering of 1-IPADM-2-BOL
panying vapor liquid equilibrium curves that were measured in
both systems were well within experimental error.
Neutron diffraction (ND) and small-angle neutron scattering
The mismatch between the VLE and viscosity data led us (SANS) were employed to investigate the short-range molecular-
to wonder whether or not the time between injections and scale interactions and nanoscale structural changes within the
measurement was the culprit. The previous viscosity profiles for fluid to shed light on the structural details of 1-IPADM-2-BOL.
1-IPADM-2-BOL34 were measured with a B10 minute equili- SANS has been successfully used to characterize liquid–liquid
bration time between CO2 injections, and in our continuous structural heterogeneity (mass-fractal dimension and aggrega-
flow testing,27 the residence time for the solvent leaving the tion behavior in water–alcohol mixtures).35–37
absorber before regeneration in the stripper column was
Measurements to observe the solvent’s proposed meso-
B12 minutes. A second viscosity profile for 1-IPADM-2-BOL scopic heterogeneity require sufficient scattering contrast between
was measured under isothermal conditions (40 1C) in our PVT the ionic and non-ionic domains within 1-IPADM-2-BOL when CO2
cell, performing manual injections at forced 10 minute injec- is present. 1-IPADM-2-BOL was synthesized in various levels of
tion intervals, ensuring that the VLE and loading profile deuteration to enhance the contrast between the CO2-reacted and
matched those for the measurement with an hour CO2 injection unreacted phases. Our efforts to synthesize fully deuterated target
interval.
were not successful due to challenges with ring closure in the
Because one anomaly may not indicate a broader trend, we penultimate synthetic step, leaving us with a perdeuterated
investigated whether disparate viscosity profiles were present in analogue, dimethyl (ethane-1,2-diyl-d4)bis((methyl-d3)carbamate).
other alkanolguanidine formulations. Of the dozens of alkanol- We note that alkanolguanidines always have at least 50% non-
guanidine derivatives we have synthesized previously,18 we ionic co-solvent. This is in part due to alkanolguanidines reaching
chose two other well-characterized derivatives (1-IPADM-3-BOL and CO2 saturation at 0.5 mol CO2 per mol solvent, leaving half of the
1-((1,3-dimethylimidazolidin-2-ylidene)amino)-3-methoxypropan-2-ol solvent unreacted. Details of the synthetic methodology for the
{1-MEIPADM-2-BOL}) and subjected them to the same viscosity derivatives of 1-IPADM-2-BOL can be found in the experimental
and VLE measurements at 40 1C as a function of CO2 loading section below. In this work we hereafter refer to the tri-nitrogen
with 10 minute and 60 minute equilibration times between guanidine core as the ‘‘head’’ and the alcohol chain as the ‘‘tail.’’
injections. 1-((1,3-Dimethyltetrahydropyrimidin-2(1H)-ylidene)-
ND measurements were performed on the Near and Inter-
amino)propan-2-ol {1-IPADM-3-BOL} and 1-MEIPADM-2-BOL Mediate Range Order Diffractometer (NIMROD)38 and SANS
also showed different viscosity profiles at the exact same partial measurements on the ‘‘Larmor’’ (see ESI†) instrument, both at
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
Paper
PCCP
Fig. 2 1-IPADM-2-BOL with varied levels of deuteration used in this
1
2
1
2
study: (a) per-proteo, (b) proteo, deuteron.
Target Station 2 of the ISIS neutron spallation and muon source
in Oxfordshire, UK experiments were performed at CO2 loadings
ranging in mole fractions from 0 to 0.5 and at temperatures
ranging from 20–100 1C, covering the expected loading and
temperature operating ranges of a working process.
Each derivative of 1-IPADM-2-BOL (Fig. 2) was handled
under an inert atmosphere. The samples (as shown in Fig. 2)
were synthesized (CO2-free) and fully saturated with CO2 to
represent extremes in CO2 loading for the solvent. Individual
and mixed samples were loaded into a cleaned and dry cell. The
samples were inserted into the TiZr null scattering alloy holder
in NIMROD and ND measurements were performed at constant
temperature using a circulating oil bath connected to the cell.
Measurements were performed to observe any changes in the
solvent’s structure as temperature was increased or decreased.
Diffraction data from the fully protonated sample (H) is very
similar with and without CO2 loading with the first diffraction
peak at Q E 1.18 ÅÀ1. The low-Q intensity increase in sample
12D:12D exhibits a large increase at low Q, a sharp diffraction peak
near 0.85 ÅÀ1, and a broad g(r) peak at 7–9 Å, indicative of
structural heterogeneity. The majority of atom–atom correla-
tions o4 Å in ND measurements were performed on the Near
and InterMediate Range Order Diffractometer (NIMROD)38 and
SANS measurements on the ‘‘Larmor’’ (see ESI†) instrument,
both at Target Station 2 of the ISIS neutron spallation and
muon source in Oxfordshire, UK experiments were performed
at CO2 loadings ranging in mole fractions from 0 to 0.5 and at
temperatures ranging from 20–100 1C, covering the expected
loading and temperature operating ranges of a working process.
Fig. 3D are intramolecular in nature and have the expected
values for the chemical groups comprising the system. Temperature-
dependent measurements over the solvent’s expected temperature
range in operation (40–100 1C) yielded a slight increase in the
differential scattering cross section (DCS) with temperature at
Q-values below the first sharp diffraction peak, as expected for
liquids because the DCS at Q = 0 is proportional to the iso-
thermal compressibility (see Fig. S1 in the ESI†). Mixtures with D
showed the most variation with temperature, though the exact
shape of the DCS for Q o 0.2 ÅÀ1 may be impacted by some errors
in the subtraction of the incoherent scattering contribution to the
DCS which is particularly high for the light hydrogen isotope.
ND measurements of mixtures of the CO2-bound protiated,
Fig. 3 Experimental and simulated ND of H and 21D 1-IPADM-2-BOL
mixtures with and without CO2. The data in this figure were all collected
at 40 1C. Unless otherwise noted, the second component in the 1 : 1 molar
mixture is fully loaded with CO2. (A) Differential scattering cross-section
(DCS) plots over the entire measured Q-range (log scale). (B) Simulated
neutron diffraction at different deuteration levels showing qualitative
agreement with the experimental data. (C) DCS plots zoomed in over
Q = 2–20 ÅÀ1. (D) Plot of 4prrh(r) vs. r correlation, where h(r) is the Fourier
transform of the DCS; the plot is magnified Â4 from 3 to 12 Å.
We then compared the experimental data to simulation.
Simulated ND data (Fig. 3B) was obtained by analyzing the
structures from classical molecular dynamic simulations.39,40
Consistent with experiment, all systems exhibit a strong peak is
identified at B1 ÅÀ1 for the 21D:21D and B1.1 ÅÀ1 for no CO2 and
full CO2 loading structures. The simulated ND results success-
fully captured the peak shift as well as the peak width of
1-IPADM-2-BOL with different levels of deuteration at B1 ÅÀ1
in agreement with experimental results (Fig. 3A). Furthermore,
our simulations are also in agreement with the diffraction data
for Q 4 1, see Fig. S2 in the ESI.† This further supports our
previous observations17 that the nanostructuring observed in
these simulations is consistent with that of the CO2-loaded
solvent.
As observed in Fig. 3B, there is a qualitative difference
between the diffraction peaks at Q o 0.7 ÅÀ1 that we attribute
to the simulation size effect. However, due to limitations in
simulation cell size (8 Â 8 Â 8 nm). Fig. S3 in the ESI† shows
how the low Q structure becomes smoother in the larger
simulation cells. Understanding the composition and structure
of these species could help describe the mechanisms for the
proposed glassification within the solvent, though observing
any large-scale supramolecular assembly was beyond the size
resolution limit of both the ND experiment and the simulations.
Small angle neutron scattering (SANS)
partially deuterated (tail), and fully deuterated 1-IPADM-2-BOL Lower Q data was needed to assess how nm-scale interactions
liquid each highlight specific correlations that play a role in impact larger scale ordering in solution. As such, SANS mea-
forming the disparate nano-environment within the fluid. For surements on Larmor sought to describe the behavior and
the first time, we characterized the interface morphology at the properties of these aggregates and nanoclusters as a function
nanometer scale as a function of temperature and CO2 loading. of temperature and CO2 content. Initial measurements carried
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
PCCP
Paper
out from 22–100 1C in 20 1C increments showed no large
aggregates in the measured Q range (details of SANS experi-
ment contained in the ESI†). CO2-loaded hydrogenous and
neutral deuterated mixtures (Fig. 2a and d, respectively) showed
only a liquid background signal that decayed on average as
Q
À1.65, with a radius of gyration, Rg, of 10 to 11 Å. A scattering
exponent of QÀ5/3 is expected for swollen chains in solution, so
the observed dependence could indicate that molecular dimers
characterized by dimension Rg may cluster to form chains
throughout the liquid. Beyond this, the inability to detect any
larger scale clustering was likely due to a combination of having
short-lived intermolecular association due to the system’s high
thermal energy and that the density of such clustering was too
low to result in any appreciable difference in scattering length
density contrast between clusters and the surrounding homo-
geneous solvent.
Hypothesizing that at low temperatures the zwitterionic
dimers would not have enough kinetic energy to overcome
aggregation forces and result in an aggregate structure that is
stable in time, we proceeded to carry out measurements from
À5 1C (the limit of the cooling bath) to 40 1C. Our results show
the gradual formation of large aggregates with decreasing
temperature, while the aggregates reversibly diminished as
temperature was increased.
Fig. 4 (A and B) Two-level unified exponential model fit parameters help
describe the fractal nature of the large aggregates (p1) and relative
orientation of the molecules (p2), the increase in aggregate concentration
(G1), and the static size of molecular clusters over the entire temperature
range (Rg2). (C) Schematic illustrates changes in the hierarchical structure
of the material in transition between low and high temperature.
A two-level unified exponential power model41,42 was applied
to the 1 : 1 deuterated : hydrogenous CO2-bound system tempera-
ture cycling scattering data to extract details of the reversible
aggregate structure (details of the UEP model described in the
ESI†). Fig. 4 shows the model parameters as a function of time
(and temperature cycling). Parameters from measurements
collected in decreasing temperature directions are shaded in blue
and measurements collected as a function of increasing tempera-
ture are shaded in red.
The first level, i = 1, describes scattering from aggregates
with size 450 nm. Because the low-Q signal which increases
with low temperature is due to aggregates that are much larger
than the SANS technique’s observation window, Rg1 was fixed at
50 nm as this was the smallest value that would allow the model
to fit the data satisfactorily. Parameter G1 (Fig. 4A) is related to
the concentration of large aggregates as a function of time
during the temperature cycling experiment. The G1 parameter
is generally higher for measurements carried out as tempera-
ture was decreased and lower when temperature was increased.
The Rg2 parameter (Fig. 4B), which likely characterizes the
dimensions of dimers, remains relatively constant at approxi-
mately 10 Å for all temperatures. Similarly, the p2 parameter
that describes the arrangement of the dimers has an average
value of 1.61, which is slightly lower than 5/3, the expected
value for a chain-like structure. This suggests that locally the
1.4 to 2.5, which is characteristic of mass fractal and polymer-
like structures. Thus, during heating the p1 exponent nearly
matches the p2 exponent and G1 is at a minimum indicating a
homogeneous liquid at large length scales 450 nm, which is
similar to the local structure characterized by the i = 2 para-
meters. During cooling, exponent p1 4 3 and maximum values
of G1 indicate the presence of 450 nm regions with a density of
1-IPADM-2-BOL-CO2 zwitterions that is high enough to form
a well-defined albeit rough interface with the surrounding
primarily neutral liquid (Fig. 4C). The rough interface between
the neutral and zwitterionic regions matches our previous
ToF-SIMs data, which showed that the interface of the
4500 nm aggregates (25 1C, 0.25 mol CO2 per mol solvent)
was not completely smooth with the non-ionic regions of the
fluid.17 These observations during heating and cooling also
suggest aggregation of the small clusters is reversible and
temperature dependent, much like a glass transition (further
details contained in the Fig. S4 and S5, ESI†).
Wide-angle X-ray scattering
molecules associate asymmetrically with one another, forming Wide angle X-ray scattering (WAXS) of CO2-free and CO2-bound
a chain-like motif which persists throughout the measured (0%, 25%, and 50% by mole CO2) 1-IPADM-2-BOL were carried
temperatures (Fig. 4C). Regions in the liquid that are richer out over a Q-range of 0.1–2.6 ÅÀ1 along with measurements
in the chain-like associations are characterized by the G1 and p1 from 0 to 60 1C for 0% and 50% CO2-loaded samples (Fig. 5).
parameters (Fig. 4A). During the cooling cycles, p1 varies The CO2-bound liquid shows a structural correlation near
between 3.0 and 3.6, characteristic of surface scattering from Q = 0.36 ÅÀ1 which is absent in the CO2-free liquid (Fig. 5A).
a large structure, while for the heating cycles p1 varies from This type of intermediate range order corresponds to dimensions
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
Paper
PCCP
increases for both the metastable and glassy states. We then
cooled the metastable state and observed the spectra change to
match that of the glassy state, indicating the glassification is due
to changes in long range order.
Our fitting suggested that the glassy state appears at tem-
peratures near 40 1C or below. To confirm this, we checked the
viscosity profile of 1-IPADM-2-BOL in our previously reported
continuous flow testing,27 noting the measured viscosities were
comparable to the metastable state (Fig. 1). This finding was
attributed to being above the transition temperature because
the absorber (while jacketed to 40 1C) had fluid exiting the
bottom of the column at 45 1C. We then measured an isotherm
at 42 1C for 1-IPADM-2-BOL using our PVT cell (Fig. S9 in the
ESI†). The viscosity data measured in this isotherm correlated
with the viscosity of the metastable state, not the glassy state
(both shown in Fig. 1). Ultimately, PVT and continuous-flow
testing of 1-IPADM-2-BOL at greater than 42 1C confirmed the
theoretical prediction that glassification occurs at approxi-
mately 40 1C.
Fig. 5 WAXS temperature dependence of molecular length scales.
(A) CO2-loaded sample has a correlation near Q = 0.36 ÅÀ1 which is
completely absent in the CO2-free material. (B) The d1 correlation varies
from B17.0 Å to 18.6 Å from 0 to 60 1C and exhibits a steep 45-fold
increase in the rate of change from 30 to 60 1C.
Transport and rheological properties
Our focus then shifted to identifying changes in the simulated
transport and rheological properties of 1-IPADM-2-BOL-CO2 in
both states. The diffusion coefficients of glassy and metastable
states were calculated from molecular dynamics simulations
and results are provided in Fig. 6a. As expected, the diffusion
coefficients increase with temperature (noting that the glassy
states show lower diffusion coefficients than metastable states
at temperatures below 50 1C). There is an abrupt leveling of
the diffusion coefficient in the metastable state at 40 1C until
approximately 50 1C, where the coefficient in the meta and
glassy states begin to converge.
We then aimed to experimentally validate the diffusion
coefficients from the molecular dynamics simulations by using
pulsed field gradient, diffusion-oriented nuclear magnetic reso-
nance spectroscopy on 1-IPADM-2-BOL, which is a technique
sensitive to diffusion over millisecond timescales. Two series of
experiments were performed; the first applied 13C PFG NMR at
14.1 T on a sample loaded with 13C-enriched CO2 to directly
resolve zwitterionic species via their 13C resonances ca. 160 ppm.
The results (Fig. S11, ESI†) show that the absolute value of the
diffusion coefficients obtained with PFG NMR and that of
molecular simulations are in fair agreement, where specifically,
13C resonances of 1-IPADM-2-BOL exhibit diffusion coefficients
which are similar to some low-molecular-weight conventional
ionic liquids.43–45 The degree of disorder indicated by the peak
widths of individual correlations is similar to that of ionic liquids.
All WAXS data was fitted with a model consisting of 3 Lorentzian
peaks for the 0%-CO2 data and 4 peaks for the 50%-CO2 data (see
Fig. S6A in the ESI†).46 The largest correlation, d1, varies from
B17.0 Å to 18.6 Å over the 0 to 60 1C range. From 0–30 1C, d1
increases slowly at a rate of 0.0011 Å 1CÀ1, yet increases 45-fold to
a rate of 0.0483 Å 1CÀ1 from 30–60 1C (Fig. 5B). Conversely,
correlations d2, d3, and d4 in the 50% CO2-loaded liquid exhibit
higher rates of change from 0–30 1C followed by slower rates from
30–60 1C (Fig. S6B in the ESI†). The variation in dn at low
temperature occurs as expected for simple volume expansion of
the liquid. As temperature is increased, the d2–d4 correlations for
the 50% CO2-loaded samples approach the correlations of the
CO2-free liquid (Fig. S6C in the ESI†), indicating that CO2 is
gradually being removed from the solvent.
To further validate the structural model represented by the
MD, we further analyzed our molecular simulations to deter-
mine whether they could identify the specific interactions that
drive this phenomenon. The ND data illustrating both the
‘‘glassy’’ and ‘‘metastable’’ fluid states at three temperatures
is provided in Fig. S7, ESI,† while X-ray diffraction spectra
of them at 40 1C are shown in Fig. S8, ESI.† The simulated
diffraction curves of the ‘‘glassy’’ state closely align with those
of the neutron and X-ray measurements. It is reasonable that
the experimental data likely represent the glassy-state of the
fluid because the required measuring times do not allow
measuring the short-lived metastable states. Otherwise, the
simulations reasonably reproduce the trends in the experimental
data, showing an increase in Q (smaller range) as temperature
Fig. 6 Comparison of (a) diffusion coefficients and (b) viscosity between
glassy and metastable states.
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
PCCP
Paper
ranging from ca. 0.01–0.03 Â 10À5 cm2 sÀ1 at 40 1C and
ca. 0.02–0.07 Â 10À5 cm2 sÀ1 at 60 1C at a loading of 0.2.
The second series of experiments were composed of 1H PFG
NMR at 17.6 T on 1-IPADM-2-BOL loaded with natural abun-
dance CO2 at a loading of 0.2. Analysis was focused on the
methyl proton occurring at ca. 1.2 and 1.4 ppm34 at tempera-
tures ranging from 25 to 100 1C, with diffusion coefficients
shown in Fig. S12 (ESI†), where diffusion coefficients at
40 and 60 1C of the methyl resonance were ca. 0.01 and 0.03 Â
10À5 cm2
s
À1, respectively. Given that molecules in glassy
regions exhibit efficient relaxation47,48 and the PFG NMR measures
a convoluted relaxation and population weighted diffusion
coefficient, it is likely that the PFG NMR diffusion coefficients
acquired are of mestastable states.
Fig. 7 The partial radial distribution function of zwitterion–zwitterion (gZ–Z
in the glassy and metastable states.
)
aggregation of zwitterions as illustrated in Fig. S14 (ESI†). The
aggregation in both states is reduced at temperatures 450 1C, and
eventually a similar behavior was observed in both systems at
100 1C. The g(r) results suggest that different types of clustering
are present in both glassy and metastable states at 25 1C.
A stronger aggregation of zwitterions exists in the glassy state as
the temperatures increase to 40 1C, and such effect is reduced at
the temperatures above 50 1C. The ‘‘glassy transition’’ between
40 1C and 50 1C results in different distributions of zwitterion/
neutral neighbors in the glassy and metastable states, causing
distinct behaviors of the two states (as discussed in the previous
sections).
We also calculated the viscosity of these systems from
the molecular trajectories using Stokes–Einstein equation as
shown in Fig. 6b. The radius of gyration is used to approximate
the radius of a sphere particle in the Stoke–Einstein equation.
The simulated viscosity at 40 1C of the glassy state (179 cP) and
the metastable state (105 cP) is in good qualitative agreement
with the experimentally determined values of 410 and 130 cP,
respectively. The underestimation may be due to limitations of
the Stokes–Einstein equation, which only accounts for diffu-
sion without explicit collective motions due to intermolecular
correlations. However, another reason for the non-quantitative
agreement may be that the small simulation cells cannot
capture entirely the interaction between large domains. Never-
theless, the simulations suggest clearly different behaviors
Changes in angle distributions between states
between the two systems when the temperature is below 40 1C To provide additional atomistic-level information regarding
and the observed transition from metastable state to glassy state at how the glass transition impacts the two states, we used two
40 1C increases viscosity in agreement with the experimental data. vector parameters in the zwitterion molecules. The first vector
Similar to the diffusion coefficients, as the temperature further points from the positively charged N to the C connected with
increases above 40 1C, the viscosity of the two systems starts to the negatively charged O. This vector indicates the orientation
converge and the two states evolve to a similar system. Note that of the COOÀ group relative to the guanidium group (H+NR).
the calculated data using the larger and longer simulation boxes at The second indicates the normal direction of the plane defined
40 1C and 100 1C are still close to the results obtained using the by the three N atoms (3-N plane) in the zwitterion. With the aid
original box of 83 nm, validating the data obtained in these of these two vector parameters, we evaluated the angle (y)
simulations. These results again reveal the presence of a transition between the COOÀ group and 3-N plane in each zwitterion,
temperature between the metastable and glassy states at 40–50 1C the angles (a) of the COOÀ groups between zwitterions and
and the differences between these two states can be observed only their zwitterion neighbors, and the angles (b) of the 3-N planes
below such temperatures. Having thoroughly characterized this between zwitterions and their zwitterion neighbors. Fig. S15
transition, it was still unclear which molecular-level interactions (ESI†) provides a visual aid for the vector parameters and angles
were resulting in these structural changes that introduced a glass- we used here.
transition in the fluid.
The full angle distributions of these angles (y, a and b) and
visual examples of their corresponding zwitterion molecular
configurations or zwitterion neighbor pairs with different pairing
Assessing local structure through radial distribution functions
We calculated the radial distribution function, g(r), of zwitterion– angles are illustrated in Fig. S16 and S17 (ESI†), respectively. For
zwitterion, zwitterion–neutral, and neutral–neutral pairs, angles between the COOÀ group and the 3-N plane in zwitterions
respectively, based on the center of mass of molecules to further (y), the glassy state shows slightly lower population of angles
understand the structural differences between the glassy and between 301 and 551, but higher population for angles around
metastable states. We also evaluated the average percentage 551–751 compared to the metastable state at 25 1C. Similarly, at
of zwitterionic neighbors of zwitterions in these structures. 40 1C, the glassy state registers lower polulation of angles below
An example for g(r) of zwitterion–zwitterion is illustrated in 201 and higher population of angles above 201 than the meta-
Fig. 7. The rest of the data are provided in Fig. S13 and S14 stable state. This suggests that the COOÀ groups tend to form
(ESI†). At 40 1C, the glassy state suggests a higher populations larger angles relative to the 3-N planes in the glassy state than
of zwitterionic neighbors at B0.45 nm, indicating a strong those in the metastable state. Both systems show similar angle
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
Paper
PCCP
distributions when the temperatures are above 50 1C. For
Based on the analysis of the angle distribution, more
the angle populations of the COOÀ groups and 3-N planes, zwitterions in the glassy state are slightly folded than those in
respectively, between zwitterions and their neighbors (a and b), the metastable state. In pairs of zwitterion neighbors, the COOÀ
the COOÀ groups of zwitterions in the glassy state tend to form groups in the glassy state tend to form larger populations of
larger populations between 1001 and 1501 and smaller populations anti-parallel pairs with angles around 1301 than in the meta-
below 1001 with the COOÀ groups of their zwitterion neighbors stable state, and the 3-N planes in these zwitterions pair less
at temperatures below 40 1C. The angles of 3-N planes in these parallelly with angles above 501 compared to the metastable
zwitterions also have larger populations above 501 than the state. Such orientational differences are accompanied by the
metastable state. Although the results based on the 0.65 nm distinctions of the amounts of external hydrogen bonds.
cutoff reveal the zwitterion pairs more locally, and the results The system-average numbers of external hydrogen bonds in
based on 1.1 nm cutoff consider larger spaces around zwitter- the metastable and glassy states are provided in Fig. S18 (ESI†).
ions, the differences in the angle populations between the glassy The data suggest that more external hydrogen bonds are
and metastable states obey similar trends regardless of the cutoff. observed between zwitterions in the glassy state compared to
The free energy plots as shown in Fig. 8 suggest a single minimum the metastable state. These minute changes in the hydrogen
in the glassy state, but multiple local (metastable) minima in the bonds lead to the aggregation of zwitterion clusters observed in
metastable state at 25 1C. Such differences at the molecular scale experiment and are ultimately the root cause for the glassifica-
increase the mobility of zwitterions in the metastable state and tion of 1-IPADM-2-BOL.
eventually lead to the lower viscosity of the metastable state
It is likely that similar structural changes in orientation of
observed in the experiments. As temperature increases above hydrogen bonding in solution are occurring in 1-IPADM-3-BOL
40 1C, both states converge to a single minimum.
and 1-MEIPADM-2-BOL. We postulate that other concentrated
or water-lean solvents would exhibit a similar phase transition
if their CO2-containing ions are not solvated by water or other
diluents solution as is the case for aqueous amines.
Discussion: impacts of glassification on a solvent-based CCS
process
The majority of water-lean solvents are configured to run their
absorption cycles at approximately 40 1C due to the availability
of cooling water.11 Plugging a carbon capture unit could be
potentially catastrophic to the capture unit, as gas flow would
have to be halted and millions of liters of solvent would have to
be heated and drained prior to continued use. To our knowl-
edge there has been no reported glassification of water-lean
solvents during testing in routine operation. We attribute the
lack of reported glassification to the (À80 kJ molÀ1 11
)
exo-
thermicty of CO2 capture by thermal-swing solvents. The high
enthalpy of CO2 binding coupled with the low specific heat of
organics results in a larger temperature increase of the CO2-rich
solvent by the time it exits the absorber column as compared to
aqueous solvents. The self-heating of the solvent during absorp-
tion of CO2 would likely bringing any water-lean solvent above a
potential Tg.
Paradoxically, post-combustion solvent researchers and
developers (like ourselves) are assessing the benefits of intro-
ducing intercooling into the absorber column, as lower tem-
peratures have been shown to improve CO2 mass transfer
(smaller and cheaper equipment)24 and condense more water
from the inlet flue gas, reducing reboiler duty due to reduced
water content.13 Thus, the resulting lower CO2 absorption
temperature presents a trade-off: reduced cost and improved
efficiency at the expense of operating solvents precipitously
close to Tg during operation. We are currently assessing
whether other formulations of water-lean solvents (e.g., amines)
exhibit similar phenomena to determine whether glass transi-
tions in water-lean solvents are common and how we could
Fig. 8 Free energy as a function of angles of the COOÀ group between
zwitterion neighbors within cutoff of 0.65 nm.
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
PCCP
Paper
mitigate glass transitions through control of molecular-level Synthesis of 1-((1,3-dimethylimidazolidin-2-
structures within these fluids.
ylidene)amino)propan-1,1,2,3,3,3-d6-2-ol34
To a three-neck round bottom flask fitted with inert nitrogen,
containing ice-cooled solution of 1-aminopropan-1,1,2,3,3,3-d6-
2-ol (7.42 g, 0.0915 mol) in anhydrous dichloromethane (50 ml),
trimethylamine (26 ml, 0.1887 mol) was added followed by the
dropwise addition of a solution of 2-chloro-1,3-dimethyl-
imidazolinium chloride (17 g, 0.107 mol) in dichloromethane
(20 ml). The reaction mixture was warmed to room temperature,
stirred for 24 h, and refluxed for a further 48 h. The reaction was
cooled to room temperature and extracted thrice with 30%
aqueous KOH. The combined organic extracts were dried over
anhydrous MgSO4, filtered, and the solvent was removed under
reduced pressure to afford yellowish liquid which was further
purified via distillation under reduced pressure to yield pure
compound 1-((1,3-dimethylimidazolidin-2-ylidene)amino)propan-
1,1,2,3,3,3-d6-2-ol (66.5% yield, 10.77 g, 0.061 mol). 13C NMR
(CDCl3, 125.7 MHz): 18.8 (sep), 36.0 (m), 49.2 (m), 53.5 (qui.),
66.6 (t), 157.9 ppm. MS calculated [M + H+] for C8H12D6N3O+ is
178.2 and found 178.2 (Fig. S19 ESI†).
Conclusions
In this computational and experimental study, we present the
first evidence that post-combustion CO2 capture solvents, such
as 1-IPADM-2-BOL, undergo a first-order phase transition that
we liken to a glass transition. Ultimately, we stress the need of
molecular-level understanding of solvent behavior to mitigate
any potentially limiting behaviors prior to commercial opera-
tion. Thus, we encourage those in the field of CCS to assess
whether their solvent formulations exhibit a Tg prior to large-
scale testing. Identifying a potential glassification provides
knowledge that could prevent catastrophic failure of carbon
capture infrastructure in large-scale testing or commercial
operation. Fortunately, the exothermicity of CO2 capture in a
working solvent-based process could provide enough heating
for solvents to operate above a potential Tg, lessening the
likelihood of glassification and potential failure in operation.
However, this is a phenomenon that can help keep the solvent
in a metastable liquid state, by tailoring the molecular inter-
actions tthat minimize clustering altogether.
SANS
Measurements were carried out at the Larmor beamline on
Target Station 2 of the ISIS pulsed neutron source at the
Rutherford Appleton Laboratory, Didcot, UK, under allocation
RB1610024, using a sample changer and 1 mm and 2 mm path
length quartz cuvette cells, for hydrogenous and deuterated
samples, respectively. In SANS, intensity as a function of the
scattering vector Q is collected with Q = (4p/l)sin(y). Here, 2y is
the scattering angle, and l is the neutron wavelength range of
0.9–12.5 Å used simultaneously by time of flight. The Q range
for this experiment was 0.004–0.8 ÅÀ1. The instrument was in
the 4 m sample-detector configuration, with A1 = 20 mm2,
S1 = 14 mm2, and a sample aperture of 6 mm (horizontal) by
8 mm (vertical). Typical data collection times on Larmor were
15 min. Measurements on Larmor were carried out in event
mode and data sets were sliced into 2 minute intervals to
observe any time dependence in the SANS signal. Data reduction
was performed using Mantid50 and scattering simulations fitted
using SasView v4.2.2.0.45,46 Raw data for the SANS measurements
can be found at https://doi.org/10.5286/ISIS.E.RB1610024.
Experimental
General experimental
All the reactions were conducted under an atmosphere of nitrogen
unless otherwise indicated. Solvents were transferred via gas-tight
syringe, cannula, or dropping funnel. The d6-propyleneoxide,
1,3-dimethylimidazolidinone, triethylamine, ammonium hydro-
xide, and dichloromethane were purchased from commercial
sources and used without purification. Carbon-13 nuclear mag-
netic resonance (13C NMR) spectra were recorded with 500 MHz
(125.7 MHz) spectrometer. Chemical shifts are reported in delta (d)
units, parts per million (ppm) relative to the center of the triplet at
77.00 ppm for deuterochloroform. 13C NMR spectra were routinely
run with broadband. Mass spectra (MS) ESI accurate masses are
reported for the molecular ion (M + 1) or a suitable fragment ion.
Synthesis of 1-IPADM-2-BOL
1-IPADM-2-BOL was synthesized by previously reported methods
in ref. 34, while 1-IPADM-3-BOL and 1-MEIPADM-2-BOL were
synthesized as described previously.18
Neutron diffraction
Diffraction data were collected using NIMROD, located in Target
Station 2 at the ISIS Neutron and Muon Source, Rutherford
Appleton Laboratory, Harwell Campus, UK under allocation
Synthesis of 1-aminopropan-1,1,2,3,3,3-d6-2-ol49
To a three-neck round bottom flask containing ammonium RB1610023. NIMROD provides and a wide Q range for accessing
hydroxide solution, d6-propyleneoxide was added and reaction correlations between o0.1 nm and approximately 30 nm. Data
was stirred for 2 h at room temperature. The progress of the were collected over the full Q-range of 0.1 to 50 ÅÀ1. Samples
reaction was monitored by thin-layer chromatography and were loaded into null-scattering, vacuum sealed TiZr alloy
crude NMR. Excess ammonium hydroxide was removed under 1 mm sample pathlength flat plate cells. A vanadium standard,
reduced pressure followed by distillation to get pure compound the empty instrument, and empty sample cells were measured
-aminopropan-1,1,2,3,3,3-d6-2-ol (41.4% yield, 7.85 g, 0.0968 mol). in addition to the samples for data normalization and instru-
13C NMR (CDCl3, 125.7 MHz): 19.4 (sep), 48.1 (qui.), 67.2 (t) ppm ment calibration. The samples were placed on a copper bracket
(Fig. S19 ESI†).
which was temperature controlled using a circulating fluid bath.
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
Paper
PCCP
GudrunN, an analysis suite based on the ATLAS software 0.2 moles CO2 per mole of 1-IPADM-2-BOL. Proton decoupled
package, was used to correct raw neutron total scattering data 13C pulsed field gradient NMR was performed with a 5 mm liquid
for multiple-scattering and beam attenuation and normalize to NMR probe (Doty Scientific Instrument, USA) at a field strength of
an absolute scale.51 Due to the high hydrogen content of the 14.1 T on an Agilent spectrometer, which corresponds with a
samples, particular attention was paid to correcting for inelasticity 13C Larmor frequency of ca. 151 MHz. A vendor-supplied, bipolar
effects, which were removed using an iterative approached devel- gradient pulse pair stimulate echo (Dbppste) pulse sequence was
oped by Soper.52 Raw data for the diffraction measurements can used. For both acquisitions at 40 and 60 1C, the spatially depen-
be found at https://doi.org/10.5286/ISIS.E.RB1610023.
dent gradient pulse duration was 2.0 ms and the delay between
spatially dependent gradient pulses was 50 ms. The delay between
scans was 5 s, the acquisition time was 50 ms, and 512 transients
SAXS
SAXS measurements were carried out using a Xeuss 2.0 HR were collected. The 13C PFG NMR spectra were analyzed in vNMRJ
SAXS/WAXS system (Xenocs, Sassenage, France) with a Cu (v. 4.2) where 50 Hz of line broadening was applied.
source tuned to l = 0.1542 nm. The neat and CO2-loaded
1H pulsed field gradient NMR was performed with a 5 mm
1-IPADM-2-BOL materials were each loaded into 1 mm path broadband observe (BBO) NMR probe at a field strength of
length, thin-wall, boron-rich glass capillaries, which were then 17.6 T on an Agilent spectrometer, which corresponds with a
sealed using hot glue. Two sample-to-detector distances (sd) of 1H Larmor frequency of ca. 750 MHz. A vendor-supplied, DOSY
156 mm and 1209 mm, to span a Q-range of 0.009–2.68 ÅÀ1
.
gradient compensated stimulate echo with spin lock with
In SAXS, X-rays scattered as a function of the scattering angle 2y, convection compensation (DgcsteSL_cc) pulse sequence was
with respect to the transmitted direct beam, are collected on an used. Temperature was calibrated using the 1H chemical shifts
area detector. The two-dimensional data is azimuthally averaged of ethylene glycol54 and diffusion coefficients are referenced to
and plotted as I(Q). Here, Q is given by Q = 4psin y/l, where y is water (H2O) at 25 1C.55 For acquisitions at 20, 25, 40, 50, 60, 70,
half the scattering angle, and Q is related to the size of the 80, 90, and 100 1C, the spatially dependent gradient pulse
scattering object d as Q B 1/d. Data were collected for 1800 s at duration was 11, 11, 7.8, 5.5, 3.8, 2.9, 2.2, 1.9, and 1.7 ms,
sd 1209 mm and 600 s at sd 156 mm to obtain the counting respectively with a delay between spatially dependent gradient
statistics shown in the plots. Scattering the empty glass capillary pulses of 200 ms. Additional experiments with a diffusion delay
was used as the background measurement and subtracted from of 250 ms at temperatures of 25, 40, 50, 60, 80, 90, and 100 1C
the presented data. The SAXS intensity I(Q) was scaled to units of used a gradient pulse duration of 9.8, 7.1, 4.9, 3.4, 2.0, 1.7 and
differential scattering cross section per unit volume (cmÀ1) using 1.5 ms. For experiments with either diffusion delay, the acqui-
a glassy carbon intensity calibration standard.53
sition time was 367 ms, the delay between scans was 1.4 s,
32 increments in gradient strength were used, and 8 transients
were collected for each increment. At each temperature, the
CO2 loading and viscosity measurements
A custom test cell developed in-house was used for vapor liquid pulse length corresponding to a p/2 pulse was calibrated.
equilibrium measurements concurrently with viscosity. In a 1H PFG NMR spectra were analyzed in vNMRJ where 2 Hz of
typical experiment, the liquid sample in its CO2-free form is line broadening was applied. For both 13C and 1H datasets, the
loaded into the temperature-controlled test cell and recircu- spline modelling (SPLMOD) routine was applied to determined
lated by a mechanical gear pump under vacuum to remove any diffusion coefficients from integrated resonances. Additional
dissolved gases. Once the pressure has stabilized, metered experimental details and results for supporting 1H–13C gradient
amounts of CO2 gas are injected into the cell vapor phase heteronuclear single quantum correlation (gHSQC) NMR
through an automated valve manifold from a calibrated 150 ml acquired at a field strength of 9.4 T at 40 1C on an unreacted
volume and allowed to be absorbed by the liquid sample until 1-IPADM-2BOL sample are included in Fig. S10 (ESI†).
a new equilibrium state is reached. The measurements are
Classical MD and viscosity calculations
repeated until the sample is fully saturated with CO2. The
viscosity of liquid is measured using an in-line viscometer Classical MD simulations of CO2BOL solvents at various
(Cambridge ViscoPro 2000) installed on the liquid circulation CO2 loadings were done at 40 1C to calculate viscosities, self-
loop. The viscometer had a model 372 sensor with a 50–1000 cP diffusion coefficients, and dielectric constants. CO2-loaded mole-
piston head installed during operation. Accuracy of this instru- cules were modeled by either all zwitterionic, half zwitterionic half
ment is within Æ1% full scale. The cell pressure is measured acid, or all acid species that allowed us to gauge how non-ionic
within an accuracy of 0.2% using Nor-Cal CDG-100 series acid-state CO2-bound CO2BOLs change solvent viscosity.
pressure transducers. The CO2 loading in the liquid phase at
each equilibrium point is determined from the mass balance- classical MD simulations. Optimized geometries and charges
The GROMACS56 package version 4.6.6 was used for all
based gas-phase pressure change and known cell volume.
obtained from electronic structure calculations were used with
the OPLS-AA57 force field. Certain dihedral angle parameters
were obtained from electronic structure calculations (see ESI†).
Diffusion measurements
The 13C and 1H NMR samples were prepared by saturating Mixtures of CO2-bound and unbound CO2BOLs were constructed
a solution of 1-IPADM-2-BOL with CO2 and then adding a CO2 at varying carbon loadings (mol%: 0, 10, 15, 20, 25, 30), with a total
free aliquot to produce a solution with a loading of approximately of 1728 molecules. Energy minimizations and high-temperature
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
PCCP
Paper
Houston, 5th edn, 1997, pp. vii–viii, DOI: 10.1016/B978-
088415220-0/50000-8.
runs for proper mixing were done, followed by NPT ensemble
equilibrations at 1 bar and 40 1C until the volume and total energy
display steady-state behavior, typically occurring between 5 and
10 ns of simulation.
Viscosities were calculated using the Stokes–Einstein rela-
tion. The radius of gyration of the solvent molecule is used to
estimate the radius of particles in the equation. The recently
proposed approach58 to calculate viscosity in two-component
ionic liquids, based on ion pair lifetimes, is not compatible
with single-molecule solvent systems. Self-diffusion coefficients
and dielectric constants were also calculated starting from
equilibrated frames with programs in GROMACS suite.
5 Carbon Capture & Sequestration Technologies at MIT, Petra
Nova W.A. Parish Fact Sheet: Carbon Dioxide Capture
and Storage Project, https://sequestration.mit.edu/tools/pro
jects/wa_parish.html, 2016, accessed on January 20, 2017.
6 NRG. News Release: NRG Energy, JX Nippon Complete
World’s Largest Post-Combustion Carbon Capture Facility
On-Budget and On-Schedule, http://investors.nrg.com/phoenix.
zhtml?c=121544p=irol-newsArticleID=2236424, 2017, accessed
on February 6, 2017.
7 U.S. Department of Energy, National Energy Technology
Laboratory (NETL), Recovery Act: Petra Nova Parish Holdings:
W.A. Parish Post-Combustion CO2 Capture and Sequestra-
tion Project, https://www.netl.doe.gov/research/coal/project-
information/fe0003311, 2016, accessed on February 6, 2017.
8 U.S. Office of Fossil Energy, Petra Nova – W.A. Parish
Project. https://energy.gov/fe/petra-nova-wa-parish-project,
2017, Accessed on January 17, 2017.
Conflicts of interest
There are no conflicts to declare.
Acknowledgements
9 Carbon Capture & Sequestration Technologies at MIT,
Kemper County IGCC Fact Sheet: Carbon Dioxide Capture
and Storage Project, http://sequestration.mit.edu/tools/pro
jects/kemper.html, 2016, accessed on February 16, 2017.
10 H. Herzog, Lessons Learned from CCS Demonstration and
Large Pilot Projects: An MIT Initiative Working Paper, MIT
Energy Initiative, Massachusetts Institute of Technology,
2016.
11 D. J. Heldebrant, et al., Water-Lean Solvents for Post-
Combustion CO2 Capture: Fundamentals, Uncertainties,
Opportunities, and Outlook, Chem. Rev., 2017, 117,
9594–9624.
12 M. Lail, J. Tanthana and L. Coleman, Non-Aqueous Solvent
(NAS) CO2 Capture Process, Energy Procedia, 2014, 63,
580–594.
13 P. M. Mathias, et al., Improving the regeneration of CO2-
binding organic liquids with a polarity change, Energy
Environ. Sci., 2013, 6, 2233–2242.
The authors thank the United States Department of Energy’s
(DOE’s) Office of Science Basic Energy Sciences Early Career
Research Program FWP 67038 in the Chemical Sciences,
Geoscience, and Biosciences (CSGB) Division for funding.
Pacific Northwest National Laboratory (PNNL) is operated by
Battelle for the US Department of Energy under contract DE-
AC05-76RL01830. The authors also thank ISIS Pulsed Neutron
and Muon Source for access to Larmor and NIMROD under
allocations RB1610024 and RB1610023. This work benefited
from the use of the SasView application, originally developed
under NSF Award DMR-0520547. SasView also contains code
developed with funding from the EU Horizon 2020 programme
under the SINE2020 project Grant No. 654000. Simulations and
ND and RD calculations were performed using PNNL’s
Research Computing facility and the National Energy Research
Scientific Computing Center (NERSC), a U.S. Department of
Energy Office of Science User Facility operated under Contract
No. DE-AC02-05CH11231. NMR spectroscopy was performed
using resources from both the Nuclear Magnetic Resonance
Spectroscopy Center at Washington State University and the
Environmental Molecular Science Laboratory (EMSL, grid.436923)
at PNNL.
14 Y. Yuan and G. Rochelle, Water-Lean Solvents for CO2
Capture Will Not Use Less Energy than Aqueous Amines.
14th Greenhouse Gas Control Technologies Conference
Melbourne 21-26 October 2018 (GHGT-14). Available at
SSRN: https://ssrn.com/abstract=3366219.
15 M. J. O’Brien, R. L. Farnum and R. J. Perry, CO2 Capture
Using 2,2-Dialkylpropane-1,3-diamines, Energy Fuels, 2013,
27, 467–477.
Notes and references
1 Climate Change 2014 Mitigation of Climate Change, 16 D. Malhotra, et al., Phase-Change Aminopyridines as Carbon
Working Group III Contribution to the Fifth Assessment
Report of the Intergovernmental Panel on Climate Change,
https://www.ipcc.ch/site/assets/uploads/2018/02/ipcc_wg3_
ar5_full.pdf.
2 P. Friedlingstein, et al., Global Carbon Budget 2019, Earth
Syst. Sci. Data, 2019, 11, 1783–1838.
Dioxide Capture Solvents, Ind. Eng. Chem. Res., 2017, 56,
7534–7540.
17 X.-Y. Yu, et al., Mesoscopic Structure Facilitates Rapid CO2
Transport and Reactivity in CO2 Capture Solvents, J. Phys.
Chem. Lett., 2018, 9, 5765–5771.
18 D. Malhotra, et al., Reinventing Design Principles for Devel-
oping Low-Viscosity Carbon Dioxide-Binding Organic Liquids
for Flue Gas Clean Up, ChemSusChem, 2017, 10, 636–642.
3 M. Bui, et al., Carbon capture and storage (CCS): the way
forward, Energy Environ. Sci., 2018, 11, 1062–1176.
4 A. L. Kohl and R. B. Nielsen, in ‘‘Preface’’ in Gas Purification, 19 J. Yao, et al., Two coexisting liquid phases in switchable
ed. A. L. Kohl and R. B. Nielsen, Gulf Professional Publishing,
ionic liquids, Phys. Chem. Chem. Phys., 2017, 19, 22627–22632.
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
Paper
PCCP
20 C. Wang, H. Luo, X. Luo, H. Li and S. Dai, Equimolar CO2 38 D. T. Bowron, et al., NIMROD: The Near and InterMediate
capture by imidazolium-based ionic liquids and superbase
Range Order Diffractometer of the ISIS second target
systems, Green Chem., 2010, 12, 2019–2023.
station, Rev. Sci. Instrum., 2010, 81, 033905.
21 D. J. Heldebrant, et al., Evaluating Transformational Solvent 39 W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey
Systems for Post-combustion CO2 Separations, Energy
Procedia, 2014, 63, 8144–8152.
and M. L. Klein, Comparison of simple potential functions for
simulating liquid water, J. Chem. Phys., 1983, 79, 926–935.
22 D. C. Cantu, et al., Dynamic Acid/Base Equilibrium in Single 40 B. L. Bhargava, R. Devane, M. L. Klein and S. Balasubramanian,
Component Switchable Ionic Liquids and Consequences on
Viscosity, J. Phys. Chem. Lett., 2016, 7, 1646–1652.
23 D. Malhotra, et al., Directed Hydrogen Bond Placement: Low
Nanoscale organization in room temperature ionic liquids: a
coarse grained molecular dynamics simulation study, Soft
Matter, 2007, 3, 1395–1400.
Viscosity Amine Solvents for CO2 Capture, ACS Sustainable 41 G. Beaucage, Small-Angle Scattering from Polymeric Mass
Chem. Eng., 2019, 7, 7535–7542.
24 P. M. Mathias, et al., Measuring the Absorption Rate of CO2
Fractals of Arbitrary Mass-Fractal Dimension, J. Appl. Crystallogr.,
1996, 29, 134–146.
in Nonaqueous CO2-Binding Organic Liquid Solvents with a 42 G. Beaucage, Approximations Leading to a Unified Expo-
Wetted-Wall Apparatus, ChemSusChem, 2015, 8, 3617–3625.
25 G. A. Whyatt, et al., Measuring CO2 and N2O Mass Transfer
nential/Power-Law Approach to Small-Angle Scattering,
J. Appl. Crystallogr., 1995, 28, 717–728.
into GAP-1 CO2–Capture Solvents at Varied Water Loadings, 43 R. Hayes, G. G. Warr and R. Atkin, Structure and Nano-
Ind. Eng. Chem. Res., 2017, 56, 4830–4836.
structure in Ionic Liquids, Chem. Rev., 2015, 115, 6357–6426.
26 D. C. Cantu, et al., Structure–property reduced order model 44 S. Li, et al., Alkyl Chain Length and Temperature Effects on
for viscosity prediction in single-component CO2-binding
organic liquids, Green Chem., 2016, 18, 6004–6011.
27 F. Zheng, et al., Bench-Scale Testing and Process Perfor-
Structural Properties of Pyrrolidinium-Based Ionic Liquids:
A Combined Atomistic Simulation and Small-Angle X-ray
Scattering Study, J. Phys. Chem. Lett., 2012, 3, 125–130.
mance Projections of CO2 Capture by CO2–Binding Organic 45 S. Li, et al., Toward understanding the structural hetero-
Liquids (CO2BOLs) with and without Polarity-Swing-Assisted
Regeneration, Energy Fuels, 2016, 30, 1192–1203.
geneity and ion pair stability in dicationic ionic liquids, Soft
Matter, 2014, 10, 9193–9200.
28 J. R. Switzer, et al., Design, Synthesis, and Evaluation 46 M. Doucet; J. H. Cho; G. Alina; J. Bakker; W. Bouwman;
of Nonaqueous Silylamines for Efficient CO2 Capture, Chem-
SusChem, 2014, 7, 299–307.
29 V. Blasucci, et al., One-component, switchable ionic liquids
derived from siloxylated amines, Chem. Commun., 2009,
116–118, DOI: 10.1039/B816267K.
P. Butler; K. Campbell; M. Gonzales; R. Heenan; A. Jackson;
P. Juhas; S. King; P. Kienzle; J. Krzywon; A. Markvardsen;
T. Nielsen; L. O’Driscoll; W. Potrzebowski; R. Ferraz Leal;
T. Richter; P. Rozycko; T. Snow and A. Washington, SasView
version 4.2.2, DOI = {10.5281/zenodo.2652478}, Figshare, 2019.
30 V. M. Blasucci, R. Hart, P. Pollet, C. L. Liotta and C. A. 47 W. S. Price, F. Tsuchiya and Y. Arata, Time Dependence of
Eckert, Reversible ionic liquids designed for facile separa-
tions, Fluid Phase Equilib., 2010, 294, 1–6.
31 A. L. Rohan, et al., The Synthesis and the Chemical and
Aggregation in Crystallizing Lysozyme Solutions Probed
Using NMR Self-Diffusion Measurements, Biophys. J.,
2001, 80, 1585–1590.
Physical Properties of Non-Aqueous Silylamine Solvents for 48 W. S. Price, Pulsed-field gradient nuclear magnetic resonance
Carbon Dioxide Capture, ChemSusChem, 2012, 5, 2181–2187.
32 J. R. Switzer, et al., Reversible Ionic Liquid Stabilized
as a tool for studying translational diffusion: Part II. Experi-
mental aspects, Concepts Magn. Reson., 1998, 10, 197–237.
Carbamic Acids: A Pathway Toward Enhanced CO2 Capture, 49 Z. Guangyou, L. Yuquing, W. Zhaohui, H. Nohira and
Ind. Eng. Chem. Res., 2013, 52, 13159–13163.
33 V. Blasucci, et al., Single component, reversible ionic liquids
for energy applications, Fuel, 2010, 89, 1315–1319.
T. Hirose, Resolution of b-aminoalcohols and 1,2-diamines
using fractional crystallization of diastereomeric salts of dehy-
droabietic acid, Tetrahedron: Asymmetry, 2003, 14, 3297–3300.
34 P. K. Koech, et al., Low viscosity alkanolguanidine and 50 O. Arnold, et al., Mantid—Data analysis and visualization
alkanolamidine liquids for CO2 capture, RSC Adv., 2013, 3,
566–572.
package for neutron scattering and m SR experiments, Nucl.
Instrum. Methods Phys. Res., Sect. A, 2014, 764, 156–166.
35 K. Nishikawa, H. Hayashi and T. Iijima, Temperature 51 A. K. Soper, GudrunN and GudrunX: Programs for Correct-
dependence of the concentration fluctuation, the
Kirkwood-Buff parameters, and the correlation length of
tert-butyl alcohol and water mixtures studied by small-angle
X-ray scattering, J. Phys. Chem., 1989, 93, 6559–6565.
ing Raw Neutron and X-ray Diffraction Data to Differential
Scattering Cross Section, Rutherford Appleton Laboratory
Technical, RAL-TR-201, 2011.
52 A. K. Soper, The Radial Distribution Functions of Water
as Derived from Radiation Total Scattering Experiments:
Is There Anything We Can Say for Sure?, ISRN Phys. Chem.,
2013, 2013, 279463.
36 M. Misawa, et al., A visualized analysis of small-angle
neutron scattering intensity: concentration fluctuation in
alcohol-water mixtures, J. Appl. Crystallogr., 2007, 40, s93–s96.
37 G. D’Arrigo and J. Teixeira, Small-angle neutron scattering 53 F. Zhang, et al., Glassy Carbon as an Absolute Intensity
study of D2O–alcohol solutions, J. Chem. Soc., Faraday
Trans., 1990, 86, 1503–1509.
Calibration Standard for Small-Angle Scattering, Metall.
Mater. Trans. A, 2010, 41, 1151–1158.
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020

View Article Online
PCCP
Paper
54 D. S. Raiford, C. L. Fisk and E. D. Becker, Calibration of 57 W. L. Jorgensen, D. S. Maxwell and J. TiradoRives, Develop-
methanol and ethylene glycol nuclear magnetic resonance
thermometers, Anal. Chem., 1979, 51, 2050–2051.
55 M. Holz and H. Weingartner, Calibration in accurate spin-
ment and testing of the OPLS all-atom force field on
conformational energetics and properties of organic
liquids, J. Am. Chem. Soc., 1996, 118, 11225–11236.
echo self-diffusion measurements using 1H and less-common 58 Y. Zhang and E. J. Maginn, Direct Correlation between Ionic
nuclei, J. Magn. Reson., 1991, 92, 115–125.
56 D. Van der Spoel, et al., GROMACS: Fast, flexible, and free,
J. Comput. Chem., 2005, 26, 1701–1718.
Liquid Transport Properties and Ion Pair Lifetimes:
A Molecular Dynamics Study, J. Phys. Chem. Lett., 2015, 6,
700–705.
Phys. Chem. Chem. Phys.
This journal is © the Owner Societies 2020