Journal of

Materials Chemistry A
PAPER View Article Online
View Journal | View Issue

Can COFs replace MOFs in flue gas separation?

This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

Cite this: J. Mater. Chem. A, 2020, 8,

high-throughput computational screening of COFs
for CO2/N2 separation†
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

14609

Omer Faruk Altundal, Cigdem Altintas and Seda Keskin *

Covalent organic frameworks (COFs) are under study as adsorbent and membrane candidates for gas
separation applications. However, experimental testing of all synthesized COF materials as adsorbents
and membranes under different operating conditions is not practical. Herein, we used a high-throughput
computational screening approach to investigate adsorption- and membrane-based flue gas separation
performances of 295 COFs. Adsorption selectivity, working capacity, percent regenerability and
adsorbent performance score of COFs were calculated for separation of CO2/N2 mixture for three
different cyclic adsorption processes, pressure swing adsorption (PSA), vacuum swing adsorption (VSA)
and temperature swing adsorption (TSA). The top performing COFs were identified for each process
based on the combination of several metrics. Selectivities of the top COFs were predicted to be greater
than those of zeolites and activated carbons. Molecular simulations were performed considering the wet
flue gas for the top COF adsorbents and results revealed that most COFs retained their high CO2
selectivities in the presence of water. Using COFs with detailed geometry optimization and high-
accuracy partial charges in molecular simulations led to lower selectivities and adsorbent performance
scores compared to using experimentally reported COFs with approximate charges. Membrane-based
flue gas separation performances of COFs were also studied and most COFs were found to have
comparable CO2 permeabilities with metal organic frameworks (MOFs), up to 3.96
106 barrer, however
Received 1st May 2020
Accepted 16th June 2020
their membrane selectivities were lower than MOFs, 0.38–21, due to their large pores and the lack of
metal sites in their frameworks. Structure–performance relations revealed that among the COFs we
DOI: 10.1039/d0ta04574h
studied, the ones with pore sizes <10 Å, accessible surface areas <4500 m2 g1 and 0.6 < porosity <0.8
rsc.li/materials-a are not only highly selective adsorbents but also CO2 selective membranes.

robust under cyclic operation conditions. Activated carbons and

1. Introduction zeolites have been utilized as adsorbents but they have limited
Carbon dioxide (CO2) capture from ue gas, CO2/N2 mixture, is selectivity and regenerability.7 Therefore, identication of new
environmentally, industrially, and economically important.1,2 adsorbent materials offering the best combination of CO2
Adsorption-based gas separation techniques such as pressure capacity, CO2/N2 selectivity and regenerability has been a long-
swing adsorption (PSA), vacuum swing adsorption (VSA), and standing goal.
temperature swing adsorption (TSA) have been used for ue gas Metal organic frameworks (MOFs) are porous structures in
separation.3,4 An ideal adsorbent should simultaneously offer which nodes consisting of metal clusters are connected through
high selectivity and high working capacity5,6 in addition to being organic linkers.8 The large variety of metal ions and organic
linkers enabled the synthesis of many different materials with
tunable chemistries, low densities (0.2 g cm3), large surface
Department of Chemical and Biological Engineering, Koc University, Rumelifeneri areas (>6000 m2 g1), various pore sizes and high porosities
Yolu, Sariyer, 34450, Istanbul, Turkey. E-mail: [email protected]; Tel: (0.3–0.9).9,10 MOFs have been considered as promising materials
+90-212-338-1362
for adsorption-based CO2/N2 separation due to these structural
† Electronic supplementary information (ESI) available: Comparison of CO2
uptakes obtained from GCMC simulations with experimental results;
and chemical aspects.11,12 Several computational studies used
distribution of atoms for screened COFs; comparison of R% and Ndes values for Grand Canonical Monte Carlo (GCMC) simulations to predict
TSA and VSA conditions; comparison of CO2 uptakes of CURATED and CoRE CO2/N2 separation performances of various MOFs.13,14 For
COFs obtained from GCMC simulations for CO2/N2 : 15/85 mixture at VSA, PSA, example, early studies showed CuBTC had a slightly higher CO2/
and TSA conditions; comparison of calculated Sads and DN values of CURATED
N2 selectivity (33) than zeolite MFI (27) for separation of
and CoRE COFs for VSA, PSA, and TSA conditions according to their ASA ratios;
comparison of calculated APSs of CURATED and CoRE COFs according to their
CO2/N2 : 15/85 mixture at PSA condition.13 Krishna and van
density and porosity ratios. See DOI: 10.1039/d0ta04574h Baten15 showed that MgMOF-74 has a similar selectivity to that

This journal is © The Royal Society of Chemistry 2020 J. Mater. Chem. A, 2020, 8, 14609–14623 | 14609
 View Article Online

Journal of Materials Chemistry A Paper

of NaX (200) with a higher working capacity (3 mol kg1) CO2/CH4, and CO2/H2 mixtures at PSA condition and showed
than that of NaX (2 mol kg1) for CO2/N2 : 15/85 mixture at that COFs offer better CO2 working capacities (>3 mol kg1)
VSA condition at 300 K. With these, the promise of MOFs in than many conventional zeolites and common MOFs. A
adsorption-based CO2/N2 separation has been proven.16–18 computation-ready experimental COF database (CoRE COF)
Due to the acceleration in the synthesis of MOFs, experi- consisting of 187 solvent-free COFs was created,40 later
mental investigation of MOFs for ue gas separation has expanded41 and nally reported to include 309 COFs which
become challenging. High-throughput computational facilitated high-throughput computational screening of these
screening methods have been very useful for efficiently materials.42 Using this database, 290 functionalized COFs were
screening large numbers of MOFs to guide experimental studies designed and 137 of these COFs were predicted to perform
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

to more promising MOFs. The rst high-throughput screening better than conventional polymers for membrane-based CO2/
of MOFs for CO2/N2 separation was carried out by studying 489 CH4 separation.42 Ongari et al.43 recently reported a publicly
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

MOFs and two MOFs were highlighted with their high CO2/N2 available clean, uniform and rened with automatic tracking
mixture selectivities (269 and 197) at 1 bar, 303 K.19 Wu et al.20 from experimental COF database (CURATED COFs) consisting
studied 105 different MOFs and showed that higher CO2/N2 of 324 optimized COFs with high-quality partial charges. Smit's
selectivities (>120) can be obtained at 1 bar, 298 K if the isosteric group44 predicted CO2 parasitic energy of the CURATED COFs43
heats of adsorption of CO2 and N2 is highly different in MOFs and hypothetical COFs using molecular simulations and
having low porosities. Wilmer et al.21 examined ue gas sepa- showed that many COFs have lower energy than that of tradi-
ration potentials of 130 000 hypothetical MOFs at VSA condition tional amine scrubbing process. As can be seen from this
at 298 K and showed that the best materials have pore sizes <5 literature summary, we have limited information on the
Å, porosities in the range of 0.3–0.4. Qiao et al.22 screened 4764 adsorption and membrane-based CO2/N2 separation perfor-
MOFs23 for ue gas separation at VSA condition and showed mances of COFs and the high-throughput computational
that most of the promising MOFs have lanthanides. Our group screening of COFs is essential to compare their adsorbent and
recently computed several adsorbent performance evaluation membrane performance evaluation metrics with those of MOFs.
metrics of 3816 MOFs for separation of CO2/N2 mixture at VSA Motivated from this, in this work, we searched the answer to
conditions and reported that the best MOF candidates have 3.8 the following question: can COFs replace or at least compete
Å < pore limiting diameter <5 Å and surface area <1000 m2 g1.10 with MOFs as adsorbent and membrane materials in ue gas
Molecular simulations were also used to investigate membrane- separation processes? We rst performed a high-throughput
based ue gas separation potential of MOFs. For example, computational screening study on CoRE COF database to
Krishna and van Baten24 screened a large number of MOF, unlock both adsorption-based and membrane-based CO2/N2
zeolitic imidazolate framework (ZIF), and zeolite membranes separation potentials of all experimentally synthesized COFs.
using Congurational-Bias Monte Carlo (CBMC) and Molecular CO2/N2 mixture adsorption in COFs was computed using GCMC
Dynamics (MD) simulations. They reported that CO2/N2 selec- simulations and results were used to evaluate selectivity,
tivity (20) and CO2 permeability (8
105 barrer) of Mg-MOF- working capacity, adsorbent performance score and regener-
74 was higher than those of many zeolites (DDR, ERI, MFI, TSC, ability of COF adsorbents at three different cyclic adsorption
and FAU). Watanabe and Sholl25 calculated CO2 permeabilities conditions, PSA, VSA, and TSA. Efficiencies of these three
and CO2/N2 selectivities of 179 MOFs at innite dilution and processes were then compared using adsorbent performance
reported that MOF membranes offer superior CO2 permeabil- evaluation metrics to select the most promising COFs for each
ities (104 barrer) and CO2/N2 selectivities (>100) than polymeric operating condition. The effect of humidity on the ue gas
membranes. Our group recently studied 3806 MOF membranes separation performance of the most promising COF adsorbents
at innite dilution and the top 15 MOFs having high CO2/N2 was also investigated by performing simulations for adsorption-
selectivity in the range of 16–820 at 1 bar, 298 K were identi- based CO2/N2/H2O mixture separation. We also utilized
ed.26 All these works revealed the high potential of MOF CURATED COF database for molecular simulations and
adsorbents and membranes for CO2/N2 separation. However, compared the results with those of CoRE COF database to
MOFs offering the highest CO2 separation performances under understand how structural curations on the experimentally re-
dry ue gas condition may suffer from a decrease in CO2 ported COFs affect their predicted CO2/N2 separation perfor-
selectivity and/or a stability issue under humid environment mances. We nally assessed membrane-based CO2/N2 mixture
and therefore, the search for robust MOFs and MOF-like separation performances of COFs by combining the results of
materials is still continuing.27–32 GCMC and MD simulations and compared gas permeabilities
Covalent organic frameworks (COFs) are a class of porous and selectivities of COF membranes with those of polymers. The
materials containing light elements (hydrogen, boron, carbon, large-scale screening of COFs performed in this work allowed us
nitrogen, oxygen, silicon) linked by strong covalent bonds, like to elucidate the structure–performance relations that can help
in diamond.30,33–37 Since COFs are light materials and have to describe the structural features leading to high-performance
strong covalent bonds, they show low density, good thermal and COF adsorbents and membranes. All these results will (i)
chemical stability, and permanent porosity.37 Compared to provide the rst comparison for the ue gas separation
MOFs, 3-dimensional COFs can offer a higher fraction of performances of COFs, MOFs, and established adsorbents and
accessible surface area for gas adsorption.38 Tong et al.39 membranes including zeolites and polymers, (ii) guide the
computationally studied 46 COFs for separation of CH4/H2, future studies to the most promising COFs identied from high-

14610 | J. Mater. Chem. A, 2020, 8, 14609–14623 This journal is © The Royal Society of Chemistry 2020
 View Article Online

Paper Journal of Materials Chemistry A

throughput screening and (iii) contribute to the intuition at the implemented in Materials Studio.51 The Ewald summation was
molecular scale for the design and advancement of new COF implemented to incorporate long–range interactions.52 10 000
adsorbents and membranes with more efficient CO2 separation cycles were set for initialization and 20 000 cycles were set for
potentials. taking ensemble averages in GCMC simulations. Potential
parameters for CO2 molecule were taken from the TraPPE53
force eld. For CO2 molecule, a three-site rigid model with LJ
2. Computational details 12–6 potential was used where the partial point charges were set
2.1. COF database at the center of each site.54 Similarly, for N2 molecule, a three-
site model was used with partial charges at the center of
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

We used the most recent CoRE COF database in the literature

consisting of 309 COF structures.42 Structural properties of mass.55 DREIDING56 was used for the potential parameters of
COFs; pore limiting diameter (PLD), largest cavity diameter COF atoms since this force eld was shown to give a good
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

(LCD), porosity (f), density (r) and accessible surface area (ASA) agreement with the experimentally reported single-component
were computed employing Zeo++ soware.45 A probe diameter CO2 and CH4 adsorption isotherms of COFs in the literature.42
of 3.72 Å which corresponds to the kinetic diameter of N2 We also showed the good agreement between our simulations
molecule was used for surface area determination whereas using DREIDING and experimentally measured CO2 adsorption
a probe with zero diameter was used for the pore volume isotherms of several COFs in Fig. S1 of the ESI.† To understand
calculations. CoRE COF database was then rened to only the effect of humidity, CO2/N2/H2O : 15/82/3 ternary mixture
include the structures with non-null ASA so that both gases can simulations were performed. The partial pressure of H2O was
adsorb in the pores. Consequently, 295 COFs remained in the kept at 3 kPa, which is 70% of the vapor pressure of H2O in
database offering a diverse range of structural properties. TIP4P57,58 model at 298 K (4.3 kPa). The Henry's constants of
Porosities of these COFs were computed to be between 0.44– H2O in COFs were also calculated to quantify the affinity of
0.96 while their PLDs and ASAs were calculated to be in the COFs to water at 298 K by conducting simulations with only
range of 4.19–44.50 Å and 245–8561 m2 g1, respectively. We Widom particle insertion move in RASPA using 105 cycles.59
note that atomistic coordinates of HAT-NTBA-COF was cor- Using the results of GCMC simulations, adsorption selectivity,
rected by using the coordinates taken from its experimental working capacity, adsorbent performance score and percent
synthesis paper46 following the warning in the literature.43 To regenerability were computed as shown in Table 1 to assess the
examine the effect of structural curations performed on the performances of COF adsorbents for gas separation. CO2 and N2
experimental CoRE COF database, we also studied 23 materials self-diffusivities were calculated using MD simulations which
from the CURATED COF database. were performed for 5
106 cycles in the NVT ensemble with
a time step of 1 fs. 106 cycles were set both for equilibration and
initialization of NVT-MD simulations in which the Nosé–Hoo-
2.2. Molecular simulations ver60 thermostat was utilized. Self-diffusivities of gases were
We used the high-throughput computational screening calculated from the slope of their mean square displacements
approach that we introduced and used for MOFs in our previous according to the Einstein's relation.61 Gas permeabilities and
works10,47 where we showed that the results of our molecular selectivities of COF membranes were predicted by combining
simulations were in accordance with experimental CO2 and N2 the results of GCMC and MD simulations as shown in Table 1.
uptakes of MOFs. To compute the CO2/N2 mixture adsorption Two feed pressures, 1 bar and 10 bar, were used while the
data of COFs, GCMC simulations were performed with RASPA permeate pressure was set as vacuum. Gas loadings obtained
2.35 simulation code.48 We specied the composition of CO2/N2 from GCMC simulations at 1 bar and 10 bar were used as the
as 15/85 to represent dry ue gas. GCMC simulations were input of MD simulations.62 Atomic coordinates of COF atoms
performed at three different operating conditions; VSA, PSA, were xed during calculations to save computational time.
and TSA. For VSA and PSA processes, temperature was xed at Since pore sizes of COFs studied in this work are large enough
298 K while adsorption (desorption) pressure was set as 1 (0.1) with respect to the kinetic diameters of both gases, exibility
bar at VSA and 10 (1) bar at PSA. For TSA process, pressure was was expected to not considerably effect our results as we dis-
set as 1 bar and temperatures for adsorption and desorption cussed before.63
were set as 298 K and 393 K, respectively.49 Peng–Robinson
equation of state was used for pressure to fugacity conversion. 3. Result and discussion
Non-bonded interactions were dened with Lennard-Jones (LJ)
potential and a cut-off distance of 14 Å was specied for trun- 3.1 Adsorption-based separation performances of COFs
cation of these interactions. The number of unit cells in the We rst examined adsorption-based CO2/N2 separation poten-
simulation box was adjusted according to the cut-off distance. tials of 295 COFs. Fig. 1 represents CO2/N2 selectivity of COFs as
Electrostatic interactions between adsorbate–adsorbate and a function of their CO2 working capacity at (a) VSA (b) PSA and
adsorbate-COF atoms were considered due to the quadrupolar (c) TSA conditions. CO2/N2 selectivity and CO2 working capacity
moments of CO2 and N2. Partial atomic charges of COF atoms of 3808 MOFs which were previously computed at VSA condi-
were assigned via the charge equilibration method (Qeq).50 The tion10 were also shown in Fig. 1(a). CO2 selectivities were
partial charges of atoms in a COF, COF-SDU1, did not converge computed to be between 1–105 and 2–6107 for COFs and MOFs,
using the Qeq in the RASPA and assigned by the Qeq respectively, whereas CO2 working capacities were calculated to

This journal is © The Royal Society of Chemistry 2020 J. Mater. Chem. A, 2020, 8, 14609–14623 | 14611
 View Article Online

Journal of Materials Chemistry A Paper

Table 1 Calculation of metrics used to evaluate separation performances of COF adsorbents and membranesa

Metrics Formula

Adsorption selectivity NCO2 yN2

Sads;CO2 =N2 ¼

NN2 yCO2
Working capacity (mol kg1) DN ¼ Nads,CO2  Ndes,CO2
Adsorbent performance score (mol kg1) APS ¼ Sads,CO2/N2
DNCO2
Percent regenerability DNCO2
R% ¼
100%
Nads;CO2
Dself;CO2
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

Diffusion selectivity
Sdiff;CO2 =N2 ¼
Dself;N2
Membrane selectivity Smem,CO2/N2 ¼ Sads,CO2/N2
Sdiff,CO2/N2
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

Permeability (barrer) ci;ads
Dself;i

Pi ¼
fi
a
i: gas species, CO2 or N2. Nads (mol kg1): gas uptake at adsorption conditions. Ndes (mol kg1): gas uptake at desorption conditions. y: composition
of the gas species in the bulk phase. f (Pa): partial pressure of gas species in the mixture. c (mol m3): gas concentration obtained from GCMC
simulations. D (m2 s1): self-diffusivity of gas obtained from MD simulations. 1 barrer ¼ 3.348
1016 mol
m (m2
s
Pa)1.

be between 0.02–1.69 mol kg1 and 0.01–4.16 mol kg1 for COFs COFs have LCDs > 15 Å. Large pores of COFs allow both gas
and MOFs, respectively. This comparison shows that COFs have molecules to be adsorbed in the framework, reducing the
lower CO2/N2 selectivities than most of the MOFs at VSA selective adsorption property of materials. In addition to this,
condition which can be explained with larger pore sizes of COFs MOFs have metal sites leading to strong coulombic interactions
compared to MOFs and the absence of metal sites in COFs. Only between the framework atoms and CO2 molecules.64 Most COFs
5% of MOFs shown in Fig. 1(a) have LCDs > 15 Å while 70% of do not have those metal sites. For example, in the whole set of

Fig. 1 Predicted selectivity and working capacity of COFs for separation of CO2/N2 : 15/85 mixture computed at (a) VSA, (b) PSA, (c) TSA
conditions. Separation performances of MOFs (black dots) are also shown in (a) for comparison.

14612 | J. Mater. Chem. A, 2020, 8, 14609–14623 This journal is © The Royal Society of Chemistry 2020
 View Article Online

Paper Journal of Materials Chemistry A

295 COFs that we examined, only 24 COFs have metal sites with 85% at VSA and PSA conditions while this ratio was 18% at TSA
relatively high partial charges (8 with Cu sites, 6 with Ni sites, 5 condition. At VSA and PSA conditions R% values were generally
with Zn sites, 4 with Co sites and 1 with Li sites). As a result of high but tend to decrease as APS increases, which was previ-
this, most COFs have weaker interactions with the adsorbate ously observed for MOFs in adsorption-based CH4/H2 separa-
molecules compared to MOFs. Detailed information about the tion.66 This indicates that many COFs with high APSs suffer
distribution of elements in COFs that we examined in this work from low R%. For example, ICOF-2 would be a promising
can be seen in Fig. S2.† Performances of the adsorbents are material at all cyclic adsorption process conditions we studied
generally assessed by considering both selectivity and working in this work if we only focused on selectivity and APS. However,
capacity because a promising adsorbent should offer both high R% was calculated to be very low for this COF (23.12% for VSA,
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

working capacity and high selectivity for an efficient and 45.98% for PSA and 61.60% for TSA), which makes it practically
economic gas separation process. We used the adsorbent unusable under these operating conditions.
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

performance score (APS), which is the multiplication of Fig. 2(c) shows that R% values computed for TSA condition
adsorption selectivity and working capacity as shown in Table 1, were generally lower than the ones obtained at VSA and PSA
to appraise the performances of COFs for CO2/N2 separation. conditions. Also, interestingly, these R% values showed
We arbitrarily set APS to 100 mol kg1, as shown with the red a different trend than those computed for VSA and PSA condi-
curves in Fig. 1, to identify good performing materials. Fig. 1(a) tions and increased as a function of APS. When we consider the
shows that COFs have relatively lower APSs than MOFs mainly calculation of APS (Sads
DN), R% (DN/Nads), and DN (Nads 
due to their lower selectivities at VSA condition. Even the COF Ndes), only the amount of gas at the desorption condition (Ndes)
with the highest APS value (ICOF-2, APS: 59.59 mol kg1) cannot is different for VSA and TSA processes as both share the same
compete with MOFs, of which the APS can be higher than adsorption temperature and pressure (1 bar, 298 K). Therefore,
2000 mol kg1. We, therefore, concluded that MOFs perform a higher R% at either of these conditions indicates that a higher
better than COFs for adsorption-based separation of dry ue gas amount of CO2 is desorbed. At the desorption condition of TSA
at VSA condition. (1 bar, 393 K), adsorbed gas molecules did not desorb as much
Calculated CO2 selectivities of COFs at PSA and TSA condi- as at desorption condition of VSA (0.1 bar, 298 K) for most of the
tions were between 1–66 and 1–105, respectively, whereas CO2 COFs, especially the ones with APS <10 mol kg1. This is shown
working capacities were in the range of 0.18–6.36 mol kg1 and in Fig. S3† where higher Ndes indicates that less amount of gas is
0.02–1.83 mol kg1 as shown in Fig. 1(b) and (c), respectively. desorbed. Based on this, we can conclude that pressure have
APSs of MOFs at PSA (TSA) condition were calculated to vary a more pronounced effect on gas desorption than temperature
between 0.65 and 153.64 mol kg1 (0.04 and 158.79 mol kg1). for most COFs.
Although selectivities of COFs at PSA were computed to be lower We aimed to identify the most promising adsorbent candi-
than those calculated at VSA and TSA conditions, APSs of COFs dates in each process by specically focusing on COFs with R%
were found to be the highest at PSA because of the high CO2 > 85% and ranking them based on their APSs. COFs with R% >
working capacities. None of the COFs was able to surpass APS ¼ 85% were found to have APSs in the range of 0.07–58.95 mol
100 mol kg1 curve at VSA condition while 5 COFs were iden- kg1 at VSA, 0.65–39.70 mol kg1 at PSA and 0.74–63.43 mol
tied to have APSs > 100 mol kg1 at PSA condition as shown in kg1 at TSA conditions. 10 COFs with the highest APSs were
Fig. 1(b), and only one COF exceeds APS ¼ 100 mol kg1 target selected for each operating condition as the most promising
at TSA condition as shown in Fig. 1(c). The COF that exceeds the materials and shown by red stars in Fig. 2. Among these, COF-
APS target in TSA (ICOF-2) has Li+ cations which signicantly 42-gra, FLT-COF-1-staggered, COF-JLU3 and NPN-2 were the
contributed to coulombic interactions between COF atoms and common top materials for VSA and TSA processes. Calculated
CO2 molecules. Our simulation results showed that electrostatic adsorbent performance metrics and the names of the top COFs
interactions were responsible for 89% of the total interaction are given in Table 2 along with the selectivities and working
energy between this COF and guest molecules. Since the capacities of some commercial adsorbents used in the industry
quadrupole moment of N2 (4.65
1040 C
m2) is smaller to compare ue gas separation performances of different
than that of CO2 (14.27
1040 C
m2),65 electrostatic inter- materials. Unit cell representations of the most promising COFs
actions become more pronounced for CO2 than N2. for each process condition are given in Fig. S4.† Table 2 shows
Regenerability (R%) is an essential factor when assessing the that many of the promising COF adsorbents have higher CO2/N2
practical usage of adsorbents for cyclic processes and it should selectivities than commercial adsorbents while having compa-
be taken into account when selecting the most promising rable working capacities to them. For example, zeolite 13X
materials for cyclic VSA, PSA and TSA processes.66 This is provided a selectivity of 17 at 1 bar, 298 K and working capacity
because adsorbents offering high selectivities generally suffer of 2.30 mol kg1 at VSA conditions. The selectivities and
from low R% values. Calculated R% of COFs are shown as working capacities of the top COFs of VSA condition were in the
a function of their APSs in Fig. 2 where the red dashed line range of 14–40 and 0.58–1.67 mol kg1, respectively. This
represents R% ¼ 85 which we set as the minimum acceptable indicates that COFs have the potential to replace zeolites in
R%. High R% of COFs were generally obtained at VSA and PSA adsorption-based CO2/N2 separation. The top COFs identied
conditions as shown in Fig. 2(a) and (b), respectively, and COFs for TSA process have higher APSs compared to the top COFs
were found to have higher R% than most MOFs as shown in identied for VSA and PSA conditions, suggesting that when
Fig. 2(a). More than 80% of COFs were computed to have R% > APS and R% were both considered, TSA can be the best process

This journal is © The Royal Society of Chemistry 2020 J. Mater. Chem. A, 2020, 8, 14609–14623 | 14613
 View Article Online

Journal of Materials Chemistry A Paper

This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

Fig. 2 Calculated R% and APS of COFs for CO2/N2 : 15/85 mixture under (a) VSA, (b) PSA, (c) TSA conditions. The red dotted line represents R% ¼
85%. Blue dots represent COFs while red stars represent the top 10 COFs identified at each condition. R% and APSs of MOFs (black dots) are also
included for comparison in (a).

option for adsorption-based ue gas separation using COFs. wet ue gas, i.e. CO2/N2/H2O mixture. Calculated CO2 selectiv-
Most COFs have high thermal stability37 but to ensure if the top ities of hydrophobic COFs for wet ue gas were found to be only
COFs that we identied for TSA process can be robust under slightly lower (6–56 and 7–61 at 1 and 10 bar, respectively) than
temperature changes, we conrmed the thermal stabilities of those computed for dry ue gas (7–77 and 7–66 at 1 and 10 bar,
these top COFs by checking their experimental synthesis respectively). However, COFs having higher KH, water values were
papers. For instance, one of the promising COFs, 3D-COOH- found to have signicantly lower CO2 selectivities for wet ue
COF, was reported to be thermally stable up to 723 K (ref. 67) gas compared to those calculated for dry ue gas. For example,
and another promising COF, COF-42-gra, was thermally stable CO2 selectivities of CuP-TFPh COF (KH, water ¼ 3.12
104 mol
up to 553 K.68 These temperatures are much higher than the kg1 Pa1) decreased from 17 to 1 at 1 bar and from 13 to 3 at 10
desorption temperature of the TSA condition, 393 K, that we bar when ternary CO2/N2/H2O mixture was considered. We,
considered in this work. therefore, concluded that humidity decreases the CO2 selectiv-
Aer identifying the top COF adsorbents, we investigated the ities of hydrophilic COFs in agreement with the results obtained
separation performances of these materials under humid when MOFs were studied for wet ue gas separation,71 but
conditions by performing GCMC simulations for ternary CO2/ hydrophobic COFs retained their high CO2 selectivities in the
N2/H2O mixture. We rst calculated the Henry's constants of presence of humidity.
water (KH, water) for these COFs and found that KH, water values of
15 of the 26 distinct top COFs (top 10 COFs were identied for
each process and 4 top materials were common for VSA and 3.2 Effects of structural curations on the simulated
TSA) were smaller than that of ZIF-8 (6.21
106 mol kg1 performances of COFs
Pa1), which is known to be a hydrophobic MOF.69,70 Therefore, We aimed to compare performances of COFs for CO2/N2 sepa-
it was seen that most of the top COFs we studied were also ration with those of MOFs. Therefore, COF structures in this
hydrophobic. Then, we performed adsorption simulations for work were taken from the CoRE COF database and used without

14614 | J. Mater. Chem. A, 2020, 8, 14609–14623 This journal is © The Royal Society of Chemistry 2020
 View Article Online

Paper Journal of Materials Chemistry A

Table 2 Separation performances of the top 10 COFs computed at structural differences due to different optimization methods
VSA, TSA, and PSA conditions and their comparison with some will affect simulation results and hence predicted ue gas
conventional adsorbents
separation performances of COFs. To investigate this, we
DN focused on the top COFs identied from the CoRE COF data-
Top COFs Sads (mol kg1) APS (mol kg1) R% base, computed their adsorbent performance metrics, and
compared them with those computed for CURATED COFs. 9 (8)
Vacuum swing adsorption (VSA) of the top COFs identied for PSA (VSA) process were available
COF-42-gra 40.20 1.41 56.70 87.54
COF-43-gra 24.75 0.81 20.11 89.07
in CURATED COF database and all top performing COFs iden-
tied for TSA were available in the CURATED COF database.
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

COF-JLU3 24.91 1.43 35.65 87.74

CuP-TFPh COF 16.78 0.94 15.71 86.35 We arranged the size of the data points in Fig. 3 according to
FLT-COF-1 staggered 35.28 1.67 58.95 87.30 the ratio of a structural property (ASA, r, and f) of a CURATED
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

NPN-2 25.26 1.47 37.11 86.64 COF to that of a CoRE COF and related the change in the
Ph-AnCD-COF 14.03 0.82 11.51 86.51
calculated performance metrics to the change in the COF
PyTTA-BFBIm-iCOF 16.96 0.88 14.87 87.33
TpMA 21.77 0.47 10.15 87.83 structure. The most signicant change was observed for ASA of
Tp-Por COF-AB 19.98 0.58 11.58 87.98 COFs. Fig. 3 shows the change in calculated APSs of COFs at
VSA, PSA and TSA conditions according to ASACURATED/ASACoRE.
Pressure swing adsorption (PSA) The minimum, average, and maximum ratio of ASAs of
COF-102 6.16 2.63 16.19 90.38
CURATED and CoRE COFs (ASACURATED/ASACoRE) were calcu-
COF-300 9.52 2.65 25.28 86.63
COF-6 10.84 1.86 20.14 89.01 lated as 0.30, 0.69, and 1.16, respectively. If a COF is shown with
COF-DL229-6 fold 10.40 2.40 24.92 87.21 a clearly different symbol size that means its ASA was signi-
EB-COF:F 18.52 1.86 34.38 87.08 cantly different when it was taken from CoRE COFs or
SIOC-COF-4-AB 7.17 2.27 16.29 85.53 CURATED COFs. APSs of COFs signicantly changed especially
TPE-COF-I 12.03 3.30 39.70 88.63
TPE-Ph COF 8.07 2.79 22.52 87.95
when the structure optimization led to very different ASAs as
TpPa-1-F2 12.83 1.60 20.52 89.28 presented with the symbol size in Fig. 3. It is important to note
TpPa-F4 18.17 1.88 34.21 88.21 that the CURATED COFs do not only differ from CoRE COFs
because of the detailed geometry optimization but they were
Temperature swing adsorption (TSA) also reported with high accuracy partial charges. In order to
3D-COOH-COF 76.52 0.78 59.58 91.89
isolate the charge effect, we compared performance metrics
COF-42-gra 40.20 1.49 60.05 92.70
COF-JLU3 24.91 1.47 36.55 89.97 under three different cases in Fig. 3: (i) when the coulombic
CoPc-PorDBAc 27.89 1.74 48.48 88.48 interactions between adsorbents and gas molecules were
FLT-COF-1 staggered 35.28 1.73 61.11 90.51 neglected (blue points), (ii) when the charges of both COFs were
NPN-1b 30.45 1.68 51.27 89.90 assigned using the Qeq method (black points), and (iii) when the
NPN-2 25.26 1.54 38.95 90.94
PI-COF 34.71 1.83 63.43 89.23
charges of CoRE COFs were assigned using the Qeq method
POR-COF 26.09 1.39 36.18 86.82 while the charges of CURATED COFs were taken as the density-
TThPP 28.51 1.42 40.47 86.82 derived electrostatic and chemical charges-DDEC (red points).
a With case (i) and (ii) we examined only the effect of structural
Calculated using Ideal adsorbed solution theory (IAST).87 b This COF
was reported to lose its crystal structure upon solvent removal.88 discrepancies on performance metrics while with case (iii)
c
This COF was reported without Co atoms in databases.89 combined effect of different structural properties and different
partial charges on performance metrics was investigated.
Conventional adsorbents
Fig. 3(a) shows that APSs of most COFs signicantly change for
Adsorbents Sads DN (mol kg1) Ref. all three cases at VSA conditions. The coefficients of determi-
nation (R2) between APSs of CoRE and CURATED COFs were
NaX 226.00 (IASTa, 15 : 85, 298 K) 1.47 (1–0.1 bar) 81
calculated as 0.35 for case (i), 0.10 for case (ii) and 0.07 for case
NaY 14.84 (ideal, 1 bar, 303 K) 1.87 (1–0.1 bar) 82
Zeolite 13X 17.45 (ideal, 1 bar, 298 K) 2.30 (1–0.1 bar) 83 (iii) at VSA condition. In case (i), only the effect of geometry
CHA 27.00 (ideal, 1 bar, 300 K) 3.60 (10–1 bar) 84 optimization was considered and the highest changes in APSs
Hb 11.33 (ideal, 1 bar, 300 K) 1.43 (1–0.1 bar) 85 were observed for the COFs with the highest ASACURATED/
Activated carbon 8.86 (ideal, 1 bar, 293 K) 2.58 (1–0.1 bar) 86 ASACoRE values. This can be explained with the change in the
connement and interaction of gas molecules with the COF's
pores as these interactions are dominant at low pressures as in
VSA process. The change in APSs of COFs taken from different
further optimization as we previously did in our large-scale databases was mainly due to working capacities rather than
screening of the MOF database.10 While we were working on selectivities as shown in Fig. S5.† The difference in structural
CoRE COFs, CURATED COF database, geometry optimized by properties between the CURATED and CoRE version of a COF
considering symmetry and stacking of structures with a more signicantly affected the CO2 uptakes (given in Fig. S6†),
detailed procedure than that of CoRE COFs,40,43 was reported. leading to considerably different working capacities. When we
The main input of our molecular simulations is the crystal included the contribution of electrostatic interactions between
structures of COFs. Therefore, we aimed to understand how COFs–guests in case (ii), larger changes were observed in APSs.

This journal is © The Royal Society of Chemistry 2020 J. Mater. Chem. A, 2020, 8, 14609–14623 | 14615
 View Article Online

Journal of Materials Chemistry A Paper

This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

Fig. 3 Comparison of APSs of CURATED and CoRE COFs for (a) VSA, (b) PSA, and (c) TSA conditions. Symbols are sized based on ASACURATED/
ASACoRE while the regular size of the symbols can be seen in the legend. Hollow symbols in (b) represent calculated APSs of the top COFs
identified for PSA process at VSA conditions.

The largest deviation in APS was observed in case (iii) due to the with hollow symbols in Fig. 3(b). APSs of these COFs were found
combined effect of structural discrepancy and different charges to be correlated leading to R2 values of 0.78, 0.49, and 0.26 for
of CoRE COFs and CURATED COFs. In Fig. 3(a), the effect of case (i), (ii), and (iii), respectively. R2 values of these COFs were
different partial charges on APS values was clearly isolated for not as low as the ones obtained for the top COFs of VSA process,
the COFs with ASACURATED/ASACoRE value of 1 (shown with indicating that structural properties have less effect on the
standard symbol size). For one of these COFs, APS value performance of the top COFs of the PSA process. Finally, for TSA
decreased to half when the structure was taken from the process, most of the top COFs had high deviations in their ASAs
CURATED database, showing the impact of using high accuracy and APSs upon structural curation as shown in Fig. 3(c) similar
charges on the performance predictions of COFs. to the top COFs of VSA process. The calculated R2 values were
The top COFs selected for PSA process were found to have 0.43 for case (i), 0.08 for case (ii), and 0.10 for case (iii). Both
similar structures in CoRE and CURATED COF databases as working capacity and selectivity changed when the curated
shown by the size of the data points in Fig. 3(b). As a result, their structures were used in molecular simulations but the change
predicted APSs were found to be strongly correlated for case (i) in working capacity was more signicant as shown in Fig. S8.†
and (ii) leading to R2 values of 0.85 and 0.69 (excluding the COF It is not directly possible to apply detailed geometry opti-
with the highest deviation), respectively. The stronger correla- mization to a large number of COF structures and assign high
tion at PSA compared to VSA condition can be due to the COF– accuracy partial charges. However, this comparison between
guest interactions which are less signicant at higher pressure databases suggested that predicted APSs of the top COFs whose
of PSA. However, again higher deviations were observed in APSs structures were taken from CoRE COF database might be
for case (iii) (R2 value of 0.27, excluding the same COF) due to different from those taken from CURATED COF database,
the effect of DDEC charges on selectivity as shown in Fig. S7† especially at low pressures where COF–gas interactions are
with red points. We also computed APSs of the top COFs important and more dependent on the structural properties.
identied for PSA process at VSA conditions and showed them Sharma et al.72 similarly showed that optimized slipped COF

14616 | J. Mater. Chem. A, 2020, 8, 14609–14623 This journal is © The Royal Society of Chemistry 2020
 View Article Online

Paper Journal of Materials Chemistry A

structures have signicantly different CO2/N2 selectivities than 3.3 Membrane-based separation performance of COFs
their counterparts. Therefore, detailed geometry optimization
We also examined membrane-based ue gas separation
of COFs before simulations needs to be considered especially
performances of 295 COFs. Adsorption, diffusion, and
for VSA and TSA processes. We so far discussed the change in
membrane selectivities of COFs computed at two different feed
APS with respect to change in ASA, but change of APS with
pressures, 1 and 10 bar, are shown in Fig. 4(a) and (b), respec-
respect to changes in r and f was also examined as shown in
tively. Adsorption selectivities of all COFs were calculated to be
Fig. S9 and S10.† Overall, except for PSA, higher working
greater than unity since CO2 is more strongly adsorbed than N2
capacities and higher selectivities were predicted when the as discussed in the previous section. Strong adsorption of CO2
structures were taken from CoRE database. As a result of this,
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

molecules resulted in their slow diffusion and diffusion selec-

higher APSs were obtained for CoRE COFs compared to
tivity mostly favored N2. Diffusion of N2 was signicantly higher
CURATED COFs. Assigning partial charges to large numbers of
than that of CO2 for 63 COFs at 1 bar and for 45 COFs at 10 bar
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

COFs using approximate charge methods is computationally

as shown with black points in Fig. 4(a and b), respectively. In 23
very efficient, however we suggest performing molecular simu-
(19) of these COFs at 1 bar (10 bar) CO2 adsorption selectivities
lations using CURATED COFs with high accuracy charges to
dominated high N2 diffusion selectivity, leading to membranes
make a nal assessment about the potential of a COF adsorbent
with CO2 selectivities >2. However, using these COFs as adsor-
for VSA and TSA processes. We nally note that we only bents can still be a better choice since they have higher
compared a small number of COFs from CoRE and CURATED adsorption selectivities than their membrane selectivities. In
databases and there is still much work needed to fully under-
a single COF at 1 bar (Fig. 4(a)) and 4 COFs at 10 bar (Fig. 4(b)),
stand the dependence of simulated separation performance of
CO2 molecules diffuse faster than N2 molecules, shown by blue
COFs on the structures taken from different databases.
dots, resulting in diffusion selectivities for CO2 higher than

Fig. 4 Adsorption, diffusion, and membrane selectivities of COFs computed at (a) 1 bar, (b) 10 bar. Stars represent the top 10 COF membranes. (c)
CO2 permeability and membrane selectivity of COFs for separation of CO2/N2 : 15/85 mixture calculated at 1 and 10 bar presented together with
upper bounds for polymeric membranes.73,74 Separation performances of the top MOF membranes are also shown for comparison of COFs with
MOFs.

This journal is © The Royal Society of Chemistry 2020 J. Mater. Chem. A, 2020, 8, 14609–14623 | 14617
 View Article Online

Journal of Materials Chemistry A Paper

unity. In these COFs, CO2 was favored both by adsorption and detailed information about their calculated membrane metrics
diffusion. The best selectivities for COF membranes were ach- are given in Table 3.
ieved when adsorption strongly favored CO2 and diffusion
slightly favored N2, as shown by red points in Fig. 4. Membrane 3.4 Structure–performance analyses of COFs
selectivities of small number of materials, 49 COFs at 1 bar and Finally, we investigated structure–performance relations of
23 COFs at 10 bar, were calculated to be less than 1, indicating COFs to estimate the best combination of structural properties
that these COF membranes are N2 selective due to their high N2 leading to a good CO2/N2 separation performance of a COF.
diffusion selectivity dominating their CO2 adsorption selec- Fig. 5(a and b) shows the relation between CO2/N2 selectivity of
tivity. As a result of these, membrane selectivities of COFs were
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

COF adsorbents calculated at 1 bar and 10 bar, porosity and

calculated to be between 0.38–21 and 0.63–37 at 1 bar and 10 LCD of 295 COFs. Selectivities of COFs increase as LCD and
bar, respectively. porosity decrease due to the better connement of gas mole-
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

Fig. 4(c) represents CO2/N2 membrane selectivities of COFs cules inside the narrow pores.75 However, some COFs with LCD
as a function of CO2 permeabilities. CO2 selectivities and > 9 Å and f > 0.6 did not follow the same trend, and their CO2/N2
permeabilities of the best performing MOF membranes iden- selectivities were relatively high although they had large pores.
tied in a previous work of our group26 at 1 bar were also shown To further investigate this, we performed GCMC simulations for
in Fig. 4(c). COF membranes offer similar permeabilities with these COFs by switching-off the coulombic interactions between
the top MOF membranes, however selectivities of COF COFs and adsorbate molecules. The triangles in Fig. 5(a and b)
membranes were calculated to be lower, 0.38–21, than those of show selectivities of these COFs which were calculated by
MOFs, 29–449. This can be explained by the large pore sizes neglecting the coulombic interactions in molecular simula-
and/or lack of metal sites in COFs as we discussed above. tions. Selectivities signicantly decreased when the electrostatic
Robeson's upper bound,73 which is the upper limit for the interactions were neglected and we changed the color of the
separation performance of polymer membranes, and the new symbols in Fig. 5(a and b) according to the coloring scale used
upper bound,74 which was recently updated considering CO2/N2 for selectivity when all the interactions were considered in
separation performances of ultrapermeable polymers of molecular simulations. The most remarkable difference was
intrinsic microporosity, are also shown in Fig. 4(c). Due to their observed for the selectivity of ICOF-2, which decreased from 105
low selectivities, only a small number of COFs exceeded the to 4, when the coulombic interactions between this COF and
upper bounds. On the other hand, we extended these upper CO2 molecules were neglected. Electrostatic interactions
bounds because COFs have very high CO2 permeabilities, between the host and adsorbates signicantly contributed to
ranging from 5.12
104 to 3.96
106 barrer at 1 bar and from the total energy of this COF, as discussed before, due to the
2.39
105 to 9.30
106 barrer at 10 bar. The most promising existence of B and Li ions with large partial charges (1.28e for B
COF membranes offering CO2 permeabilities >106 barrer and ions and 1.12e for Li ions). Therefore, we concluded that COFs
the highest selectivities are shown by stars in Fig. 4(c) and with large pores generally suffer from low adsorption selectivity

Table 3 Performances of the top 10 COFs for membrane-based CO2/N2 separation

Top COFs Dself,CO2 (105 cm2 s1) Dself,N2 (104 cm2 s1) PCO2 (106 barrer) PN2 (105 barrer) Sads,CO2/N2 Sdiff,CO2/N2 Smem,CO2/N2

1 bar
CC-TAPH-COF 7.27 1.97 1.16 1.27 24.51 0.37 9.03
COF-JLU3 8.88 2.49 1.27 1.43 24.91 0.36 8.87
CTF-FUM 28.50 4.18 2.12 3.48 9.02 0.68 6.15
MPCOF 55.20 5.08 1.03 2.11 4.55 1.09 4.94
NPN-1 9.98 1.44 3.91 1.83 30.45 0.69 21.13
NPN-2 12.00 1.78 3.96 2.32 25.26 0.67 16.98
NPN-3 12.33 1.12 1.61 1.24 12.00 1.10 13.18
PyTTA-BFBIm-iCOF 8.01 1.66 1.39 1.69 16.96 0.48 8.18
TEMPO-COF 33.24 9.02 1.51 3.77 10.90 0.37 4.02
TpMA 17.16 2.46 2.27 1.50 21.77 0.70 15.19

10 bar
CC-TAPH-COF 6.58 1.40 4.28 5.52 16.58 0.47 7.78
COF-JLU3 5.90 1.80 3.58 5.39 20.31 0.33 6.66
CTF-1 26.44 3.20 5.46 9.90 6.64 0.83 5.49
CTF-FUM 6.33 6.20 2.22 2.00 10.91 1.02 11.14
FLT-COF-1 staggered 4.82 1.90 3.81 6.21 24.16 0.25 6.14
NPN-1 1.15 1.23 1.09 0.30 39.39 0.93 36.56
PyTTA-BFBIm-iCOF 4.81 1.38 3.56 5.85 17.49 0.35 6.09
TPE-COF-I 4.34 7.11 2.46 3.35 12.03 0.61 7.34
TpMA 5.08 5.23 2.29 0.89 26.48 0.97 25.76
TpPa-F4 5.13 1.55 2.22 3.71 18.17 0.33 6.02

14618 | J. Mater. Chem. A, 2020, 8, 14609–14623 This journal is © The Royal Society of Chemistry 2020
 View Article Online

Paper Journal of Materials Chemistry A

This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

Fig. 5 LCDs and porosities of COFs as a function of their adsorption selectivities calculated at (a) 1 bar, (b) 10 bar. COFs are colored according to
their adsorption selectivities in (a) and (b). Triangles represent calculated selectivities of COFs when the coulombic interactions were neglected.
PLDs and porosities of COFs as a function of their membrane selectivities calculated at (c) 1 bar and (d) 10 bar, respectively. Colors represent
membrane selectivity of COFs in (c) and (d).

but the ones having metal atoms with relatively large partial COFs that we studied in this work, the top 10 promising COF
charges could offer high selectivities even if they are large- adsorbents identied for each cyclic adsorption process, VSA,
pored. Fig. 5(a) has a wider distribution of CO2/N2 selectivities PSA, and TSA, and the top 10 promising COF membranes. For
and more outliers than Fig. 5(b) because the coulombic inter- both adsorption- and membrane-based ue gas separation,
actions between the adsorbent and adsorbates become more COFs with pore sizes (PLD and LCD) <10 Å offer the best
pronounced at low pressures. Similar to adsorption selectivity- performances. For adsorption-based separation, COFs with
LCD relation, we also studied membrane selectivity-PLD corre- 2000 m2 g1 < ASA <4500 m2 g1 and 0.6 < f <0.8 were identied
lation. Fig. 5(c and d) shows the relation between CO2/N2 to achieve the highest APS values whereas for membrane-based
selectivity of COF membranes calculated at 1 bar and 10 bar, separation, COFs with ASA <2000 m2 g1 and f < 0.6 were
porosity and PLD of 295 COFs. Membrane selectivities of COFs identied to have the highest selectivities. Previously, it was
decrease as their PLDs and porosities increase because both shown that MOFs with pore sizes <7.5 Å, ASA <1000 m2 g1, and
gases can easily diffuse through the large pores, limiting the 0.5 < f < 0.75 can have high CO2/N2 adsorption and membrane
separation capacity of membranes. Results in Fig. 5(a–d) selectivities.10,26 Promising COF adsorbents and COF
showed that pore size has a greater importance than porosity on membranes have slightly larger pore sizes, similar porosities,
the CO2 selectivity of COFs since for a specic porosity a wide and higher surface areas compared to promising MOFs for CO2/
variety of selectivity values could be obtained according to the N2 separation. We nally note that chemical properties of COFs,
pore sizes of the COFs. e.g. type of framework atoms, presence of metal atoms and
We nally aim to elucidate the optimum structural proper- functional groups in the frameworks, also effect CO2 adsorp-
ties of COFs to achieve high CO2/N2 separation performances tion, but we solely focused on the effects of structural proper-
and to compare these optimum structural properties with those ties, such as pore size and surface area which are easily
of MOFs. Fig. 6 shows computed structural properties of all calculated with computational methods, on the predicted

This journal is © The Royal Society of Chemistry 2020 J. Mater. Chem. A, 2020, 8, 14609–14623 | 14619
 View Article Online

Journal of Materials Chemistry A Paper

This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

Fig. 6 Distribution of calculated PLD, LCD, ASA and f of all 295 COFs and the top 10 COFs identified for each separation process. Numbers in the
charts show the number of COFs.

separation performances of COFs in this work. These structure– COFs have lower selectivities (1–105) than MOFs (2–6107) due to
performance relations will provide valuable information to their larger pores and the lack of metal sites in their frame-
experimentalists for the design and advancement of unique works. On the other hand, COFs offer working capacities
COF structures which realize highly selective CO2/N2 between 0.02–1.69 mol kg1 while this range was 0.01–4.16 mol
separations. kg1 for MOFs at VSA conditions. Results also showed that
COFs can have higher selectivities than currently used adsor-
bents, such as zeolites and activated carbons, while having
4. Conclusion similar working capacities to them. Therefore, COFs have the
In this work CO2/N2 mixture separation potentials of 295 COFs potential to replace conventional adsorbents for CO2 separation
were examined both for adsorption- and membrane-based from ue gas. For CO2/N2 separation using COFs as adsorbents,
applications by combining GCMC and MD simulations. TSA process was suggested because more COFs were identied
Results of molecular simulations were utilized for predicting to have a combination of high R% (>85%) and high APS
adsorption selectivity, working capacity, R%, APS of COFs for (>10 mol kg1) compared to VSA and PSA conditions. Molecular
three distinct cyclic adsorption processes, VSA, PSA, and TSA. simulations also showed that selectivity of COFs do not signif-
Comparison of COFs with MOFs at VSA condition revealed that icantly change in the presence of humidity, making them

14620 | J. Mater. Chem. A, 2020, 8, 14609–14623 This journal is © The Royal Society of Chemistry 2020
 View Article Online

Paper Journal of Materials Chemistry A

potential adsorbents for wet ue gas separation. Curated, 4 N. Hedin, L. Andersson, L. Bergström and J. Yan, Appl.
optimized COFs with high accuracy partial charges were Energy, 2013, 104, 418–433.
computed to have lower APSs than CoRE COFs having approx- 5 R. Kumar, Ind. Eng. Chem. Res., 1994, 33, 1600–1605.
imate charges, especially if the structure was signicantly 6 Y. S. Bae and R. Q. Snurr, Angew. Chem., Int. Ed., 2011, 50,
affected from the geometry optimization. This showed the 11586–11596.
importance of using the correct COF structure before assessing 7 D. M. D'Alessandro, B. Smit and J. R. Long, Angew. Chem., Int.
its potential for CO2/N2 separation. Gas permeabilities of COF Ed., 2010, 49, 6058–6082.
membranes (5.12
104 to 3.96
106 barrer) were computed to 8 J. Jiang, Mol. Simul., 2014, 40, 516–536.
be similar to those of MOF membranes, however COFs suffered 9 H. Deng, S. Grunder, K. E. Cordova, C. Valente, H. Furukawa,
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

from low membrane selectivities (0.38–21). Due to their low M. Hmadeh, F. Gándara, A. C. Whalley, Z. Liu, S. Asahina,
selectivities, most COF membranes could not exceed the H. Kazumori, M. O'Keeffe, O. Terasaki, J. F. Stoddart and
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

Robeson's upper bound. O. M. Yaghi, Science, 2012, 336, 1018–1023.

These results answered the question raised in the title of our 10 C. Altintas, G. Avci, H. Daglar, A. Nemati Vesali Azar,
work: COFs have the potential to compete with MOFs in S. Velioglu, I. Erucar and S. Keskin, ACS Appl. Mater.
adsorption-based ue gas separations but their potential as CO2 Interfaces, 2018, 10, 17257–17268.
selective membranes is limited. However, several strategies 11 S. Keskin, Nat. Energy, 2019, 5, 8–9.
such as functionalizing COFs by incorporating ionic liquids into 12 H. Furukawa, K. E. Cordova, M. O'Keeffe and O. M. Yaghi,
COFs,76,77 assisting COFs with graphene oxides to fabricate Science, 2013, 341, 974–988.
ultrathin membranes,78 and using COFs as llers in polymers to 13 B. Liu and B. Smit, Langmuir, 2009, 25, 5918–5926.
make mixed matrix membranes similar to MOFs26,79,80 can 14 B. Liu and B. Smit, J. Phys. Chem. C, 2010, 114, 8515–8522.
enhance membrane selectivities of COFs by tuning their pore 15 R. Krishna and J. M. J. S. van Baten, Sep. Purif. Technol., 2012,
sizes and/or chemical properties. Results of structure–perfor- 87, 120–126.
mance analysis that we provided in this work can be used to 16 Z. J. Zhang, Z. Z. Yao, S. C. Xiang and B. L. Chen, Energy
advance the design and development of new COF adsorbents Environ. Sci., 2014, 7, 2868–2899.
and membranes with desired structural properties. We note 17 Q. Wang, J. Bai, Z. Lu, Y. Pan and X. You, Chem. Commun.,
that COF structures that we studied in this work have been 2016, 52, 443–452.
already experimentally synthesized and reported in the 18 R. Ben-Mansour, M. A. Habib, O. E. Bamidele, M. Basha,
computation-ready COF database. Large-scale production of N. A. A. Qasem, A. Peedikakkal, T. Laoui and M. Ali, Appl.
MOFs has already started so we expect to see scalable produc- Energy, 2016, 161, 225–255.
tion and industrial applications of COFs especially in gas 19 E. Haldoupis, S. Nair and D. S. Sholl, J. Am. Chem. Soc., 2012,
storage and separation elds in near future. Finally, high- 134, 4313–4323.
throughput computational screening of COFs that we used in 20 D. Wu, Q. Yang, C. Zhong, D. Liu, H. Huang, W. Zhang and
this work can be applied to unlock the potential of COFs to G. Maurin, Langmuir, 2012, 28, 12094–12099.
separate various gas mixtures in future studies. 21 C. E. Wilmer, O. K. Farha, Y. S. Bae, J. T. Hupp and
R. Q. Snurr, Energy Environ. Sci., 2012, 5, 9849–9856.
Conflicts of interest 22 Z. W. Qiao, K. Zhang and J. W. Jiang, J. Mater. Chem. A, 2016,
4, 2105–2114.
There are no conicts to declare. 23 Y. G. Chung, J. Camp, M. Haranczyk, B. J. Sikora, W. Bury,
V. Krungleviciute, T. Yildirim, O. K. Farha, D. S. Sholl and
Acknowledgements R. Q. Snurr, Chem. Mater., 2014, 26, 6185–6192.
24 R. Krishna and J. M. van Baten, Phys. Chem. Chem. Phys.,
S. K. acknowledges ERC-2017-Starting Grant. This study has 2011, 13, 10593–10616.
received funding from the European Research Council (ERC) 25 T. Watanabe and D. S. Sholl, Langmuir, 2012, 28, 14114–
under the European Union's Horizon 2020 research and inno- 14128.
vation programme (ERC-2017-Starting Grant, grant agreement 26 H. Daglar and S. Keskin, J. Phys. Chem. C, 2018, 122, 17347–
No 756489-COSMOS). 17357.
27 J. Liu, J. Tian, P. K. Thallapally and B. P. McGrail, J. Phys.
References Chem. C, 2012, 116, 9575–9581.
28 A. C. Kizzie, A. G. Wong-Foy and A. J. Matzger, Langmuir,
1 T. Boden, R. Andres and G. Marland, Global, regional, and 2011, 27, 6368–6373.
national fossil-fuel CO2 emissions (1751-2014)(v. 2017), 29 S. J. Datta, C. Khumnoon, Z. H. Lee, W. K. Moon, S. Docao,
Environmental System Science Data Infrastructure for T. H. Nguyen, I. C. Hwang, D. Moon, P. Oleynikov,
a Virtual Ecosystem, 2017. O. Terasaki and K. B. Yoon, Science, 2015, 350, 302–306.
2 M. Songolzadeh, M. Soleimani, M. Takht Ravanchi and 30 Y. Zeng, R. Zou and Y. Zhao, Adv. Mater., 2016, 28, 2855–
R. Songolzadeh, Sci. World J., 2014, 2014, 1–35. 2873.
3 D. Bahamon and L. F. Vega, Chem. Eng. J., 2016, 284, 438–
447.

This journal is © The Royal Society of Chemistry 2020 J. Mater. Chem. A, 2020, 8, 14609–14623 | 14621
 View Article Online

Journal of Materials Chemistry A Paper

31 O. Shekhah, Y. Belmabkhout, Z. Chen, V. Guillerm, 59 B. Widom, J. Chem. Phys., 1963, 39, 2808–2812.
A. Cairns, K. Adil and M. Eddaoudi, Nat. Commun., 2014, 60 D. Frenkel and B. Smit, Understanding molecular simulation:
5, 4228–4235. From algorithms to applications, Elsevier (formerly
32 P. Nugent, Y. Belmabkhout, S. D. Burd, A. J. Cairns, published by Academic Press), 2002.
R. Luebke, K. Forrest, T. Pham, S. Ma, B. Space, L. Wojtas, 61 A. Einstein, Investigations on the theory of the Brownian
M. Eddaoudi and M. J. Zaworotko, Nature, 2013, 495, 80–84. movement, Courier Corporation, 1956.
33 P. J. Waller, F. Gandara and O. M. Yaghi, Acc. Chem. Res., 62 S. Keskin and D. Sholl, Ind. Eng. Chem. Res., 2009, 48, 914–
2015, 48, 3053–3063. 922.
34 A. P. Cote, A. I. Benin, N. W. Ockwig, M. O'Keeffe, 63 I. Erucar and S. Keskin, J. Membr. Sci., 2016, 514, 313–321.
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

A. J. Matzger and O. M. Yaghi, Science, 2005, 310, 1166–1170. 64 R. Poloni, K. Lee, R. F. Berger, B. Smit and J. B. Neaton, J.
35 C. S. Diercks and O. M. Yaghi, Science, 2017, 355, 923–933. Phys. Chem. Lett., 2014, 5, 861–865.
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

36 U. Dı́az and A. Corma, Coord. Chem. Rev., 2016, 311, 85–124. 65 C. Graham, D. A. Imrie and R. E. Raab, Mol. Phys., 1998, 93,
37 X. Feng, X. Ding and D. Jiang, Chem. Soc. Rev., 2012, 41, 49–56.
6010–6022. 66 Y. Basdogan, K. B. Sezginel and S. Keskin, Ind. Eng. Chem.
38 H. M. El-Kaderi, J. R. Hunt, J. L. Mendoza-Cortes, A. P. Cote, Res., 2015, 54, 8479–8491.
R. E. Taylor, M. O'Keeffe and O. M. Yaghi, Science, 2007, 316, 67 Q. Lu, Y. Ma, H. Li, X. Guan, Y. Yusran, M. Xue, Q. Fang,
268–272. Y. Yan, S. Qiu and V. Valtchev, Angew. Chem., Int. Ed.,
39 M. Tong, Q. Yang and C. Zhong, Microporous Mesoporous 2018, 57, 6042–6048.
Mater., 2015, 210, 142–148. 68 F. J. Uribe-Romo, C. J. Doonan, H. Furukawa, K. Oisaki and
40 M. M. Tong, Y. S. Lan, Q. Y. Yang and C. L. Zhong, Chem. Eng. O. M. Yaghi, J. Am. Chem. Soc., 2011, 133, 11478–11481.
Sci., 2017, 168, 456–464. 69 P. Z. Moghadam, D. Fairen-Jimenez and R. Q. Snurr, J. Mater.
41 M. M. Tong, Y. S. Lan, Z. L. Qin and C. L. Zhong, J. Phys. Chem. A, 2016, 4, 529–536.
Chem. C, 2018, 122, 13009–13016. 70 H. Zhang and R. Q. Snurr, J. Phys. Chem. C, 2017, 121, 24000–
42 T. Yan, Y. Lan, M. Tong and C. Zhong, ACS Sustainable Chem. 24010.
Eng., 2018, 7, 1220–1227. 71 I. Erucar and S. Keskin, Ind. Eng. Chem. Res., 2020, 59, 3141–
43 D. Ongari, A. V. Yakutovich, L. Talirz and B. Smit, ACS Cent. 3152.
Sci., 2019, 5, 1663–1675. 72 A. Sharma, A. Malani, N. V. Medhekar and R. Babarao,
44 K. S. Deeg, D. D. Borges, D. Ongari, N. Rampal, L. Talirz, CrystEngComm, 2017, 19, 6950–6963.
A. V. Yakutovich, J. M. Huck and B. Smit, ACS Appl. Mater. 73 L. M. Robeson, J. Membr. Sci., 2008, 320, 390–400.
Interfaces, 2020, 12, 21559–21568. 74 B. Comesaña-Gándara, J. Chen, C. G. Bezzu, M. Carta,
45 T. F. Willems, C. H. Rycro, M. Kazi, J. C. Meza and I. Rose, M.-C. Ferrari, E. Esposito, A. Fuoco, J. C. Jansen
M. Haranczyk, Microporous Mesoporous Mater., 2012, 149, and N. B. McKeown, Energy Environ. Sci., 2019, 12, 2733–
134–141. 2740.
46 S.-Q. Xu, R.-R. Liang, T.-G. Zhan, Q.-Y. Qi and X. Zhao, Chem. 75 G. Avci, S. Velioglu and S. Keskin, ACS Appl. Mater. Interfaces,
Commun., 2017, 53, 2431–2434. 2018, 10, 33693–33706.
47 Z. Sumer and S. Keskin, Ind. Eng. Chem. Res., 2016, 55, 76 M. Zeeshan, V. Nozari, M. B. Yagci, T. Isık, U. Unal,
10404–10419. V. Ortalan, S. Keskin and A. Uzun, J. Am. Chem. Soc., 2018,
48 D. Dubbeldam, S. Calero, D. E. Ellis and R. Q. Snurr, Mol. 140, 10113–10116.
Simul., 2016, 42, 81–101. 77 L.-G. Ding, B.-J. Yao, F. Li, S.-C. Shi, N. Huang, H.-B. Yin,
49 R. Ben-Mansour and N. A. Qasem, Energy Convers. Manage., Q. Guan and Y.-B. Dong, J. Mater. Chem. A, 2019, 7, 4689–
2018, 156, 10–24. 4698.
50 C. E. Wilmer and R. Q. Snurr, Chem. Eng. J., 2011, 171, 775– 78 Y. Ying, D. Liu, J. Ma, M. Tong, W. Zhang, H. Huang, Q. Yang
781. and C. Zhong, J. Mater. Chem. A, 2016, 4, 13444–13449.
51 R. L. Akkermans, N. A. Spenley and S. H. Robertson, Mol. 79 X. Guo, Z. Qiao, D. Liu and C. Zhong, J. Mater. Chem. A, 2019,
Simul., 2013, 39, 1153–1164. 7, 24738–24759.
52 P. P. Ewald, Ann. Phys., 1921, 369, 253–287. 80 Y. Cheng, L. Zhai, Y. Ying, Y. Wang, G. Liu, J. Dong,
53 B. L. Eggimann, A. J. Sunnarborg, H. D. Stern, A. P. Bliss and D. Z. L. Ng, S. A. Khan and D. Zhao, J. Mater. Chem. A,
J. I. Siepmann, Mol. Simul., 2014, 40, 101–105. 2019, 7, 4549–4560.
54 J. J. Potoff and J. I. Siepmann, AlChE J, 2001, 47, 1676–1682. 81 M. Xu, S. Chen, D.-K. Seo and S. Deng, Chem. Eng. J., 2019,
55 C. Murthy, K. Singer, M. Klein and I. McDonald, Mol. Phys., 371, 693–705.
1980, 41, 1387–1399. 82 P. J. Harlick and F. H. Tezel, Microporous Mesoporous Mater.,
56 S. L. Mayo, B. D. Olafson and W. A. Goddard, J. Phys. Chem., 2004, 76, 71–79.
1990, 94, 8897–8909. 83 S. Cavenati, C. A. Grande and A. E. Rodrigues, J. Chem. Eng.
57 C. Vega, J. Abascal and I. Nezbeda, J. Chem. Phys., 2006, 125, Data, 2004, 49, 1095–1101.
34503–34513. 84 E. Garcı́a-Pérez, J. Parra, C. Ania, A. Garcı́a-Sánchez, J. Van
58 J. L. Abascal and C. Vega, J. Chem. Phys., 2005, 123, 234505– Baten, R. Krishna, D. Dubbeldam and S. Calero,
234517. Adsorption, 2007, 13, 469–476.

14622 | J. Mater. Chem. A, 2020, 8, 14609–14623 This journal is © The Royal Society of Chemistry 2020
 View Article Online

Paper Journal of Materials Chemistry A

85 X. Xu, X. Zhao, L. Sun and X. Liu, J. Nat. Gas Chem., 2008, 17, 88 D. Beaudoin, T. Maris and J. D. Wuest, Nat. Chem., 2013, 5,
391–396. 830–834.
86 X. Dong, J. Zhang, L. Gang, P. Xiao, P. Webley and Y.-c. Zhai, 89 V. S. P. K. Neti, X. Wu, M. Hosseini, R. A. Bernal, S. Deng and
J. Fuel Chem. Technol., 2011, 39, 169–174. L. Echegoyen, CrystEngComm, 2013, 15, 7157–7160.
87 A. L. Myers and J. M. Prausnitz, AlChE J, 1965, 11, 121–127.
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.
Open Access Article. Published on 14 July 2020. Downloaded on 9/28/2023 7:04:05 AM.

This journal is © The Royal Society of Chemistry 2020 J. Mater. Chem. A, 2020, 8, 14609–14623 | 14623