screening of covalent organic frameworks for adsorption

9
Screening of CovalentOrganic Frameworks for Adsorption Heat Pumps Wei Li, ,,§ Xiaoxiao Xia, ,,§ and Song Li* ,,,§ State Key Laboratory of Coal Combustion, School of Energy and Power Engineering, China-EU Institute for Clean and Renewable Energy, and § Nano Interface Centre for Energy, School of Energy and Power Engineering, Huazhong University of Science and Technology, Wuhan 430074, China * S Supporting Information ABSTRACT: Exploring high-performing adsorption-driven heat pumps (AHPs) remains a challenging task owing to the low working capacity, high regeneration temperature, and low energy eciency of conventional adsorbents. Quick discovery of the novel promising adsorbents could help to improve the coecient of performance of AHPs for heating (COP H ) and cooling (COP C ). Herein, we reported an approach to identify the high-performing covalentorganic frameworks (COFs) for heating, cooling, and ice making by high-throughput computational screening based on grand canonical Monte Carlo simulations and, for the rst time, machine learning. It was demonstrated that compared with metalorganic frameworks (MOFs), COFs were more suitable adsorbents of AHPs for cooling because of their weak interaction toward ethanol that favors stepwise adsorption. Structureproperty relationship analysis revealed that the average enthalpy of adsorption commensurate with the enthalpy of evaporation will benet the performance of AHPs besides the high working capacity and low step positions of adsorption isotherms. In order to reduce the computational cost of screening, a random forest model was developed to successfully predict the COP C of both COFs and MOFs. KEYWORDS: adsorption heat pumps, high-throughput computational screening, machine learning, covalentorganic frameworks, coecient of performance 1. INTRODUCTION The building energy consumption will remain a major part of world energy consumptions even in 2040, predominantly for space heating, hot water production, and air conditioning. 1 As essential devices, heat pumps played a vital role in space heating and cooling. Thus, the development of advanced heat pump technology is an eective strategy for alleviating energy crisis. 2 Among various heat pumps powered by electricity, chemical reaction, absorption, and adsorption, adsorption- driven heat pumps (AHPs) are promising technologies owing to the low regeneration temperatures 3 powered by renewable solar energy or industrial waste heat. Nevertheless, the coecient of performance (COP) that describes the energy conversion eciency of AHPs is notoriously low 4 because of the unsatisfactory adsorption capability of conventional adsorbents such as activated carbon and zeolites. Therefore, exploring high ecient porous materials with favorable adsorption properties is the key to improve the COP of AHPs. In recent decades, metalorganic frameworks (MOFs) consisting of inorganic metal nodes and organic linkers have emerged as potential adsorbents for AHPs because of their ultralarge surface area, high pore volume, and outstanding adsorption performance. 3 Among various working uids (i.e., water, alcohols, ammonia) of AHPs, water is widely used because of its low cost and environmental-friendliness. 5,6 In recent years, the water adsorption capability of multiple MOFs has been investigated, including MIL-101(Cr), 7 Al-fumarate, 8 and CAU-10-H. 9 A newly synthesized MOF, MIP-200, was recently reported to exhibit a benecial S-shaped adsorption isotherm and high water uptake (0.39 g/g), demonstrating its potential as a candidate for cooling. 10 Nevertheless, the poor water stability of a majority of MOFs is still a concern for their applications in AHPs. 11 On the contrary, the solvothermal stability of MOFs in alcohols including methanol and ethanol is less of an issue, 3 beneting the application of MOFalcohol, especially MOFethanol working pairs because of the low toxicity of ethanol. Thus, compared with water, ethanol can be a good alternative owing to its high vapor pressure, low freezing point, fast heat and mass transfer, and large critical diameter for condensation that can utilize large pores of adsorbents more eciently. 12 However, given the large number of synthesized MOFs, experimental evaluation of the AHP performance of each MOFethanol working pair is extremely impractical. Recently, Received: November 16, 2019 Accepted: December 23, 2019 Published: December 23, 2019 Research Article www.acsami.org Cite This: ACS Appl. Mater. Interfaces 2020, 12, 3265-3273 © 2019 American Chemical Society 3265 DOI: 10.1021/acsami.9b20837 ACS Appl. Mater. Interfaces 2020, 12, 32653273 Downloaded via HUAZHONG UNIV SCIENCE & TECHNOLOGY on March 7, 2020 at 13:16:58 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

Upload: others

Post on 25-Feb-2022

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Screening of Covalent Organic Frameworks for Adsorption

Screening of Covalent−Organic Frameworks for Adsorption HeatPumpsWei Li,†,‡,§ Xiaoxiao Xia,†,‡,§ and Song Li*,†,‡,§

†State Key Laboratory of Coal Combustion, School of Energy and Power Engineering, ‡China-EU Institute for Clean and RenewableEnergy, and §Nano Interface Centre for Energy, School of Energy and Power Engineering, Huazhong University of Science andTechnology, Wuhan 430074, China

*S Supporting Information

ABSTRACT: Exploring high-performing adsorption-drivenheat pumps (AHPs) remains a challenging task owing tothe low working capacity, high regeneration temperature, andlow energy efficiency of conventional adsorbents. Quickdiscovery of the novel promising adsorbents could help toimprove the coefficient of performance of AHPs for heating(COPH) and cooling (COPC). Herein, we reported anapproach to identify the high-performing covalent−organicframeworks (COFs) for heating, cooling, and ice making byhigh-throughput computational screening based on grandcanonical Monte Carlo simulations and, for the first time,machine learning. It was demonstrated that compared withmetal−organic frameworks (MOFs), COFs were more suitable adsorbents of AHPs for cooling because of their weakinteraction toward ethanol that favors stepwise adsorption. Structure−property relationship analysis revealed that the averageenthalpy of adsorption commensurate with the enthalpy of evaporation will benefit the performance of AHPs besides the highworking capacity and low step positions of adsorption isotherms. In order to reduce the computational cost of screening, arandom forest model was developed to successfully predict the COPC of both COFs and MOFs.

KEYWORDS: adsorption heat pumps, high-throughput computational screening, machine learning, covalent−organic frameworks,coefficient of performance

1. INTRODUCTION

The building energy consumption will remain a major part ofworld energy consumptions even in 2040, predominantly forspace heating, hot water production, and air conditioning.1 Asessential devices, heat pumps played a vital role in spaceheating and cooling. Thus, the development of advanced heatpump technology is an effective strategy for alleviating energycrisis.2 Among various heat pumps powered by electricity,chemical reaction, absorption, and adsorption, adsorption-driven heat pumps (AHPs) are promising technologies owingto the low regeneration temperatures3 powered by renewablesolar energy or industrial waste heat. Nevertheless, thecoefficient of performance (COP) that describes the energyconversion efficiency of AHPs is notoriously low4 because ofthe unsatisfactory adsorption capability of conventionaladsorbents such as activated carbon and zeolites. Therefore,exploring high efficient porous materials with favorableadsorption properties is the key to improve the COP of AHPs.In recent decades, metal−organic frameworks (MOFs)

consisting of inorganic metal nodes and organic linkers haveemerged as potential adsorbents for AHPs because of theirultralarge surface area, high pore volume, and outstandingadsorption performance.3 Among various working fluids (i.e.,water, alcohols, ammonia) of AHPs, water is widely used

because of its low cost and environmental-friendliness.5,6 Inrecent years, the water adsorption capability of multiple MOFshas been investigated, including MIL-101(Cr),7 Al-fumarate,8

and CAU-10-H.9 A newly synthesized MOF, MIP-200, wasrecently reported to exhibit a beneficial S-shaped adsorptionisotherm and high water uptake (0.39 g/g), demonstrating itspotential as a candidate for cooling.10 Nevertheless, the poorwater stability of a majority of MOFs is still a concern for theirapplications in AHPs.11 On the contrary, the solvothermalstability of MOFs in alcohols including methanol and ethanolis less of an issue,3 benefiting the application of MOF−alcohol,especially MOF−ethanol working pairs because of the lowtoxicity of ethanol. Thus, compared with water, ethanol can bea good alternative owing to its high vapor pressure, lowfreezing point, fast heat and mass transfer, and large criticaldiameter for condensation that can utilize large pores ofadsorbents more efficiently.12

However, given the large number of synthesized MOFs,experimental evaluation of the AHP performance of eachMOF−ethanol working pair is extremely impractical. Recently,

Received: November 16, 2019Accepted: December 23, 2019Published: December 23, 2019

Research Article

www.acsami.orgCite This: ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

© 2019 American Chemical Society 3265 DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

Dow

nloa

ded

via

HU

AZ

HO

NG

UN

IV S

CIE

NC

E &

TE

CH

NO

LO

GY

on

Mar

ch 7

, 202

0 at

13:

16:5

8 (U

TC

).Se

e ht

tps:

//pub

s.ac

s.or

g/sh

arin

ggui

delin

es f

or o

ptio

ns o

n ho

w to

legi

timat

ely

shar

e pu

blis

hed

artic

les.

Page 2: Screening of Covalent Organic Frameworks for Adsorption

high-throughput computational screening (HTCS) of MOF−ethanol working pairs from Computational-Ready Experimen-tal (CoRE) MOF database13 has been reported. Erdos et al.evaluated 2930 MOF−methanol/ethanol working pairs usingthe predicted working capacity (ΔW) and step locations ofadsorption isotherm (α) of each MOF structure from grandcanonical Monte Carlo (GCMC) simulations, both of whichcan favor the COP of AHPs.14 Nevertheless, as the mostimportant evaluation criterion, COP should be used to directlyassess the cooling performance of MOF−ethanol workingpairs. Very recently, we developed a computational screeningstrategy to quickly identify the top-performing MOF−ethanolworking pairs by calculating the coefficient of performance forcooling (COPC) of 1527 CoRE MOFs using integratedGCMC simulations and thermodynamic cycles of AHPs.15 Itwas revealed that COPC was not positively dependent on theworking capacity (ΔW): as ΔW < 0.2 g/g, COPC increasedwith the working capacity, beyond which COPC was notremarkably influenced by the increase of ΔW. Moreover, thesmall-pore MOFs exhibited poor COPC because of the stronghost−adsorbate interaction, resulting in type I isotherm andthus low COPC. In contrast, large-pore MOFs (>1.2 nm)generally exhibiting low host−adsorbate interaction and type Visotherms possessed high COPC. Thus, exploring the noveladsorbents exhibiting relatively weak host−adsorbate inter-action and S-shaped isotherm could be a more effectiveapproach to enhance AHP performance from the perspectiveof material science. However, given the limited number oflarge-pore MOFs in the CoRE MOF database that can meetthe above requirements, searching for more suitable candidatesfrom other nanoporous materials can not only add to thenumber of top performers but also increase the possibility ofdiscovering more efficient adsorbents for AHPs.Covalent−organic frameworks (COFs)16 are potentially a

class of promising candidates for AHPs. Most of COFs possessthe large pore size and preferentially exhibit weak host−adsorbate interaction strength resulting from the fully organic

unit of frameworks, leading to stepwise adsorption (S-shaped)isotherms. Such possible features expected for COFs canperfectly meet the requirements for high-performing adsorb-ents. Furthermore, Biswal et al.17 and Stegbauer et al.18

reported that COFs can be promising adsorbents for AHPsowing to their high porosity and chemical stability. Veryrecently, Perez-Carvajal et al.19 found that TpPa-1 was anefficient adsorbent for cooling, which can be regenerated bysunlight through the experimental measurement, suggestingthe great potential of COF adsorbents for AHPs. Therefore, weperformed a HTCS of 275 experimentally synthesized COFsfor adsorption heating, cooling, and ice making for the firsttime. One of the selected top performers was synthesized andtested, and its COPC based on the experimentally measuredethanol adsorption isotherm showed good agreement withpredicted results. By comparing their structure−propertyrelationship with MOFs for cooling, it was revealed thatCOFs were more promising adsorbents for AHPs. Finally,machine learning (ML) was also implemented to predict theCOPC of COFs and MOFs, demonstrating a more efficientapproach to identify top performers from a given databasewithout exhaustive computation. This work opens up thepossibility of using COF−ethanol working pairs as potentialcandidates for AHPs, and the extracted structure−propertyrelationship provides insightful guidance for quickly discover-ing and designing high efficient COFs and MOFs for heating,cooling, or ice making.

2. MATERIALS AND METHODS2.1. High-Throughput Computational Screening. 2.1.1. Struc-

tural Properties’ Calculation. The CoRE-COF 2.0 database20

carrying extended charge equilibration method (EQeq) charges21

was adopted in this work. In the initial screening stage consisting of 4× 104 MC cycles, four COFs exhibiting no uptake (W = 0)throughout the pressure range, and one COF exhibiting disorderedstructure were removed from the database. Thus, 275 structures fromthe CoRE-COF 2.0 database were eventually used for screening.Helium void fraction (VF) and Henry’s law constant (KH) of each

Figure 1. (a) Schematic AHPs based on COFs and (b) thermodynamic cycles of AHPs for heating, cooling, and ice making.

ACS Applied Materials & Interfaces Research Article

DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

3266

Page 3: Screening of Covalent Organic Frameworks for Adsorption

COF toward ethanol were calculated by the Widom particle insertionmethod22 of RASPA 2.0,23 respectively. A nitrogen probe with aradius of 1.86 Å was used to predict the accessible surface area (ASA),available pore volume (Va), largest cavity diameter (LCD), and pore-limiting diameter (PLD) of each structure by Zeo++.24

2.1.2. COP Calculation Based on Thermodynamical Cycles ofAHPs. The coefficient of performance (COP) of all COFs under threedifferent working conditions was calculated using the calculatedadsorption parameters by GCMC simulations. According to the idealthermodynamical cycle of AHPs in Figure 1, COPH and COPC can becomputed according to eqs 1 and 2, respectively. The detailedinformation can be found in the Supporting Information.

H T m W

M C T T H WCOP

( )

( )pC

vap ev liqwf

sorbent

wsorbent

des con liqwf

ads

ρ

ρ=

Δ Δ

− − + ⟨Δ ⟩Δ (1)

H T m W

M C T T H WCOP 1

( )

( )pH

vap con liqwf

sorbent

wsorbent

des con liqwf

ads

ρ

ρ= +

Δ Δ

− − + ⟨Δ ⟩Δ(2)

ΔHvap is the evaporation enthalpy of working fluids at evaporationtemperature (Tev) or condensation temperature (Tcon), ρliq is theworking fluid density in the liquid state, and msorbent is the mass ofadsorbents. ΔW denotes the working capacity, that is, the uptakedifference between maximum isostere of adsorption stage (i.e., I andII in Figure 1) and minimum isostere of desorption stage (i.e., III andIV in Figure 1). Cp is the heat capacity of adsorbents (assuming thatthe heat capacity of COFs varies in the range similar to that ofMOFs). Previous study has demonstrated that the variation in Cp ofMOFs will not impose significant impacts on COPC.

3 Mw representsthe molar mass of working fluids. ⟨ΔHads⟩ is the loading averagedenthalpy of adsorption defined as below.

H H W W W W( )d /( )W

W

ads ads max minmin

max∫⟨Δ ⟩ = Δ −(3)

According to the above equations, COP of AHPs based on varyingCOFs for heating, cooling, and ice making can be obtained. Thespecific parameters defining typical heating, cooling, and ice makingconditions are provided in Table 1. For example, in ice making, the

adsorption temperature is 298 K (I of the thermodynamical cycle ofFigure 1), and the corresponding pressure is 1.1 kPa. The desorptiontemperature is 353 K (III of the thermodynamical cycle of Figure 1),corresponding to the pressure of 7.8 kPa. According to Trouton’srules,25 the minimal preheating temperature (II of the thermody-namical cycle of Figure 1) is 331 K. The evaporation temperature is268 K, which is the target temperature for ice making. The adoptedcondensation temperature that is usually identical to the adsorptiontemperature is 298 K. Afterward, GCMC simulation was performed toobtain the working capacity (ΔW) and averaged enthalpy ofadsorption (⟨ΔHads⟩) at predefined conditions to predict COP.2.1.3. GCMC Simulations. GCMC simulations were carried out for

275 COFs to calculate their working capacity and isosteric heat ofdesorption (−⟨ΔHads⟩) using RASPA 2.0.23 Universal force fieldparameters were adopted to describe the Lennard-Jones interactionsof all COFs.26 EQeq atomic partial charge was used for all COF

atoms.21 The force field parameters of ethanol were from TraPPE(Table S3).27 The reliability of TraPPE force field of ethanol has beendemonstrated in previous studies.14,28,29 MC cycles were performedfor all COFs under heating, cooling, and ice making conditions,respectively, to estimate their ethanol adsorption capacity. Insertion,deletion, rotation, and translation moves were implemented inGCMC simulation with equal probability. 4 × 104 MC cycles wereconducted for 275 COFs in the initial round, in which 2 × 104 cycleswere used for equilibration and 2 × 104 cycles were performed forproduction. After obtaining their COP from the first round, structureswith COPC ≥ 0.85 for cooling and ice making and structures withCOPH ≥ 1.85 for heating were selected for the second-roundscreening consisting of 4 × 105 MC cycles. In the final round, COFstructures with COPC ≥ 0.85 and COPH ≥ 1.85 from the secondround were chosen for extended GCMC simulations until completeequilibration. Similar approach has been adopted in previousstudies15,30 in order to complete the screening process at lessexhaustive computational expense. Eventually, 8 COFs with COPH ≥1.85 for heating (Table S6), 32 COFs with COPC ≥ 0.85 for cooling(Table S7), and 12 COFs with COPC ≥ 0.85 for ice making (TableS8) from the final stage were identified.

2.2. Experimental Synthesis and Vapor Adsorption. Allchemicals required in this study were purchased from commercialsources and used without any further purification. 2,3,6,7,10,11-Hexahydroxytriphenylene (HHTP, 96%) was purchased fromZhengzhou Alfachem Co., Ltd. 1,4-Benzene diboronic acid (BDBA,97%) was obtained from Shanghai Aladdin Bio-Chem TechnologyCo., Ltd. Absolute ethanol, methanol, and acetone were purchasedfrom Sinopharm Chemical Reagent Co., Ltd. (Shanghai, China, AR).Nitrogen (N2, 99.999%) gas was purchased from Huaerwen IndustrialCo., Ltd. COF-5 was synthesized using a slightly modified method ofprevious report.31 HHTP (112 mg, 0.345 mmol), BDBA (86 mg, 0.52mmol), and methanol (0.21 mL, 5.2 mmol) were added in a dioxane/mesitylene mixture solution (4:1 v/v, 43 mL) at room temperatureand sonicated for 30 min under the N2 atmosphere. The solution wasthen heated to 90 °C for 20 h in an oil-bath oven with stirring underthe N2 atmosphere. After cooling to room temperature, the solid wasisolated by centrifugation and washed three times with acetone (30mL). Subsequently, the solid was dried under vacuum at roomtemperature for 12 h. COF-320 was synthesized according to theprevious study32 in which biphenyl-4,4′-dicarbaldehyde (BPDA, 100mg, 0.476 mmol) and tamoxifen (100 mg, 0.263 mmol) weredissolved in 5 mL of anhydrous dioxane in a 15 mL glass tube undersonication. After quick freezing the solution with liquid nitrogen, 1mL of aqueous acetic acid (3 mol/L) was added into the tube. Then,the frozen tube was vacuumed and sealed. After heating at 120 °C for72 h, yellow solid was isolated by centrifugation, which was thenwashed three times with acetone (30 mL) and dried under vacuum atroom temperature for 12 h. Ethanol adsorption isotherms weremeasured at 288 and 298 K on an Autosorb-iQ2 from QuantachromeInstruments. In each measurement, approximately 100 mg sampleswere activated at 393 K for 24 h under vacuum. Adsorption isothermswere collected over the relative pressure range from 0.01 to 0.9 P/P0.Then, a universal adsorption isotherm model33 was adopted to fit themeasured isotherms, from which the adsorption isotherms of varyingtemperatures can be predicted. According to the Clausius−Clapeyronequation, the load averaged enthalpy of adsorption can be computedby the adsorption isotherms measured at different temperatures.

2.3. Data Mining and ML. Data mining and ML wereimplemented by the python code.34,35 In total, 1527 MOFs and275 COFs were used for principal component analysis (PCA) anddecision tree (DT). In PCA, ΔW, −⟨ΔHads⟩, LCD, and Va were takenas input parameters, which is normalized between 0 and 1. As for DT,a three-depth layer was used in DT, in which ΔW, −⟨ΔHads⟩, LCD,ASA, and Va were taken into account. In order to efficiently predictthe cooling performance by ML, multiple linear regression (MLR),DT, support vector machine (SVM), gradient boosting machine(GBM), and random forest (RF) were tested for the databaseconsisting of MOFs and COFs. In each algorithm, 80% of sampleswere randomly chosen for training, and the rest was used as the test

Table 1. Working Conditions of AHPs for Heating, Cooling,and Ice Making

working condition I II III

heating T (K) 318 351 373P (kPa) 4.2 22.8 22.8

cooling T (K) 303 325 353P (kPa) 3 10.4 10.4

ice making T (K) 298 331 353P (kPa) 1.1 7.8 7.8

ACS Applied Materials & Interfaces Research Article

DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

3267

Page 4: Screening of Covalent Organic Frameworks for Adsorption

set. The samples were randomly chosen from both MOF and COFdatabases to ensure the inclusion of both MOFs and COFs in thetraining and test sets. LCD, ASA, helium VF, Va, KH, and adsorbentdensity (ρ) were used as descriptors. R2 was adopted to describe theaccuracy of each model. Each ML model was trained 100 times to findout the best option to represent the quantitative correlation betweendescriptors and COP.

3. RESULTS AND DISCUSSION3.1. Structure−Property Relationship of COFs. The

correlations between LCD and working capacity (ΔW) that isvital for the COP of heating, cooling, and ice making arepresented in Figure 2a−c. It was found that COFs generally

exhibited smaller pore sizes than the critical diameter forethanol condensation (i.e., ∼45.2 Å at 318 K for heating,∼43.8 Å at 303 K for cooling, and ∼41.0 Å at 298 K for icemaking), indicating the reversible ethanol adsorption withinthese COFs. A vast majority of COFs with pore sizes of 15−25Å possessed the low working capacity (ΔW) of <0.2 g/g underthree working conditions. The working capacity increased withLCD until 10 Å and then decreased with the increase of poresizes, similar to our previous finding in MOFs15 (Figure S3b).Notably, the structures with ΔW > 0.2 g/g exhibited the poresize between 10 and 15 Å, and there is obviously a largernumber of COFs exhibiting the higher ethanol uptakes forcooling than those for heating and ice making. Similartendency in the correlation between COP and LCD underthree working conditions is demonstrated in Figure 2d−f,suggesting that COFs are more suitable for cooling. Moreover,the high-performing COFs for cooling (COPC ≥ 0.8)

possessed a wide range of pore sizes (5−30 Å) comparedwith those for heating (5−15 Å; COPH ≥1.8) and ice making(5−15 Å; COPC ≥ 0.8), indicating the preferential small pore-sized COFs for heating and ice making, whereas both smalland large pore-sized COFs were suitable candidates forcooling, which can be ascribed to the different temperaturelifts. According to the previous study,3 the temperature lift thatis the difference between evaporation (Tev) and condensation(Tcon) temperatures played an important role in determiningthe applicable operation conditions of adsorbents. Small-poreframeworks that could enable high-temperature lift due to thestrong host−adsorbate interaction are favorable for high-temperature-lift applications, that is, heating and ice making(Tlift = 30 K) in this work. In contrast, large-pore frameworksexhibiting a relatively weak host−adsorbate interaction aremore suitable for low-temperature-lift applications, that is,cooling (Tlift = 20 K). Furthermore, 72 out of 275 COFs(26.18%) exhibited COPC ≥ 0.8 in contrast to 94 out of 1527MOFs (6.16%) of previous study15 (Figure S4b), validatingthe merit of COFs for cooling.Comparing the dependence of COP on the working capacity

(ΔW) of COFs with that of MOFs reported previously,3,4,14,36

similar tendency can be observed. The quantitative correlationwas recently demonstrated by MOF−ethanol working pairs forcooling, in which no remarkable dependence of COPC on ΔWwas observed when ΔW was greater than 0.2 g/g.15 In otherwords, COPC could not be dramatically enhanced by merelyimproving the working capacity as ΔW ≥ 0.2 g/g for MOFs.Similarly, COP of COFs for heating, cooling, and ice makingdisplayed identical dependence on ΔW (Figure 3a−c). Thereliability of our predictions was further validated by theconsistency between simulated predicted and experimentallymeasured COPC of COF-5, COF-320, and TpPa-1 displayedin the inset of Figure 3b, even though there is discrepancy inethanol adsorption isotherms from experiment measurementsand simulations (Figure S13) because of the poor crystallinityof COFs.37

Regarding the impacts of different working conditions onCOP, it was revealed that the number of selected high-performing COFs (COPC ≥ 0.8 or COPH ≥ 1.8) for cooling(72 structures) is larger than that for heating (21 structures)and ice making (30 structures), implicating the greaterpotential of COFs for cooling. Moreover, the predicted steppositions (α) of COFs in Figure 3d−f demonstrated that allhigh-performing COFs generally exhibited lower step positions(α ≤ 0.4) under varying working conditions, consistent withthe preferential stepwise adsorption or type V isotherm fromboth the thermodynamic and energetic perspectives.14,36,38,39

Specifically, COFs with 0.2 < α ≤ 0.4 are favorable for heatingand cooling, whereas COFs with 0 < α ≤ 0.2 are bettercandidates for ice making, which is resulted from the highworking capacity at defined operating conditions (Figure S6).Besides, the top-performing COFs exhibited not only the highworking capacity (ΔW) and relatively low step positions butalso the moderate averaged enthalpy of adsorption (−⟨ΔHads⟩∼40 kJ/mol), both of which are favorable for the COP ofMOFs (Figures S7 and S8).Because of advantageous structure properties (i.e., LCD, Va)

and adsorption characteristics (ΔW, −⟨ΔHads⟩), their evolu-tionary trends from three-round screening for cooling (FigureS16) revealed that the screening evolved toward the COFswith medium pore sizes (8 Å < LCD ≤24 Å), moderate porevolumes (0 < Va ≤ 3 cm3/g), high working capacities (ΔW >

Figure 2. Adsorption and AHP performance of COFs. The numberdistributions of COFs with varying working capacity and LCD for (a)heating, (b) cooling, and (c) ice making; the number distribution ofCOFs with different COP and LCD for (d) heating, (e) cooling, and(f) ice making. The number of COFs in each square was calculatedwith an interval of LCD = 5 Å and ΔW = 0.2 g/g.

ACS Applied Materials & Interfaces Research Article

DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

3268

Page 5: Screening of Covalent Organic Frameworks for Adsorption

0.25 g/g), and moderate average enthalpy of adsorption (40kJ/mol < −⟨ΔHads⟩ ≤ 50 kJ/mol). Similar tendency wasobserved for heating and ice making (Figures S15 and S17).Eventually, 8 COF candidates with COPH ≥ 1.85 wereselected for heating (Table S6), 32 COFs with COPC ≥ 0.85were identified for cooling (Table S7), and 12 COFs withCOPC ≥ 0.85 were chosen for ice making (Table S8). It isnoticed that a considerable number of top performers selectedunder one fixed working condition also perform well underother working conditions (i.e., heating, cooling, and icemaking), as shown in Tables S6−S8. All the selected topthree COFs for heating (Figure S18), cooling (Figure S19),and ice making (Figure S20) show stepwise adsorptionisotherms (type V). The only exception is Ph-AnCD-COF(Figure S20b) that exhibits multiple steps and extremely highsaturation uptakes, which also meets the requirements for high-performing AHPs.3.2. Comparison between COFs and MOFs. To

quantitatively demonstrate the discrepancy in the coolingperformance of COFs and MOFs, herein we compared thestructure−property relationship of COFs reported in this workand MOFs reported in previous study in Figure 4. The formeranalysis has demonstrated that COFs can be potentially bettercandidates for cooling considering the larger percentage of top-performing COFs (72 out of 275 COFs with COPC ≥ 0.80)

than MOFs (94 out of 1527 MOFs with COPC ≥ 0.80). Tointerpret such a difference, ethanol Henry’s constant (KH) thatdescribes the affinity of adsorbents toward ethanol at extremelylow pressure was taken into account (Figure 4a). It was foundthat COPC of both COFs and MOFs increased with theincrease of Henry’s constant until KH > 10−1 mol/(kg·Pa). Ingeneral, COFs exhibited the lower KH or weaker interactionwith ethanol than MOFs, which favored the stepwiseadsorption of COFs. Similar to MOFs, the preferential KHfor COPC ranges from 10−4 to 10−1 mol/(kg·Pa), and the KHof a majority of COFs is located in this region, suggesting thehigher possibility of identifying high-performing candidatesfrom COFs. Figure 4b manifests the correlation between KHand the step position (α) of adsorption isotherms, in whichstructures with KH between 10−4 and 10−1 mol/(kg·Pa)exhibited stepwise adsorption (predominantly 0 < α ≤ 0.4 forCOFs and 0.2 < α ≤ 0.4 for MOFs). Either too low or too highKH may give rise to poor ethanol uptake or require highregeneration energy because of the strong host−adsorbateinteraction, leading to unsatisfactory AHP performance forcooling (i.e., low COPC).From the perspective of favorable heat of adsorption,

previous study15 has demonstrated that the moderate heat ofadsorption or the slightly higher averaged enthalpy ofadsorption (−⟨ΔHads⟩) than the enthalpy of evaporation

Figure 3. Relationship between AHP performance and adsorption properties. The predicted COP of 275 COFs as a function of working capacity(ΔW) for (a) heating, (b) cooling, and (c) ice making colored by the average enthalpy of adsorption. The step position (α) distribution of 275COFs for (d) heating, (e) cooling, and (f) ice making by boxplots. Boxplots show the distribution characteristics of each data set, in which themean (the line in the rectangular box), average (small squared symbols in the rectangular box), outliers (aster symbols distributed in the top andbottom), and interquartile ranges were included. The bottom and upper fences of the rectangular box represent the first and third quartiles of thedata set, respectively. For heating, there are 54 COFs with 0 < α ≤ 0.2, 51 COFs with 0.2 < α ≤ 0.4, 35 COFs with 0.4 < α ≤ 0.6, 56 COFs with 0.6< α ≤ 0.8, and 79 COFs with 0.8 < α ≤ 1.0. For cooling, there are 54 COFs with 0 < α ≤ 0.2, 51 COFs with 0.2 < α ≤ 0.4, 38 COFs with 0.4 < α ≤0.6, 62 COFs with 0.6 < α ≤ 0.8, and 70 COFs with 0.8 < α ≤ 1.0. For ice making, there are 53 COFs with 0 < α ≤ 0.2, 41 COFs with 0.2 < α ≤0.4, 51 COFs with 0.4 < α ≤ 0.6, 41 COFs with 0.6 < α ≤ 0.8, and 89 COFs with 0.8 < α ≤ 1.0. The corresponding simulation and experimentalresults for COF-5, COF-320, and TpPa-1 are displayed in the inset of Figure 3b.

ACS Applied Materials & Interfaces Research Article

DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

3269

Page 6: Screening of Covalent Organic Frameworks for Adsorption

[ΔHvap(Tev)] is favorable for the regeneration of adsorbents.Figure 4c presents the correlation between COPC, the ratio of−⟨ΔHads⟩/ΔHvap(Tev), and step positions (α), which demon-strated that the structures with comparable −⟨ΔHads⟩ andΔHvap(Tev) exhibited favorable step positions (0 < α ≤ 0.4 forCOFs and 0.1 < α ≤ 0.2 for MOFs) and COPC. Either too lowor too high ratio of −⟨ΔHads⟩/ΔHvap(Tev) will lead to lowCOP, which is an essential aspect to be taken into accountwhen choosing high-performing adsorbents for AHPs. Thisparameter can be used as a novel criterion for identifying high-performing COFs and MOFs. Comparison of the total host−adsorbate interaction energy (Qtotal) of COFs and MOFs(Figure S22) also demonstrated that neither too weak nor toostrong host−adsorbate interaction is favorable for the coolingefficiency.

Data mining of both COF and MOF data was furtherapplied to quantitatively elucidate the correlation betweenvarious descriptors and COP. In PCA (Figure 4d), fourdescriptors, that is, LCD, Va, ΔW, and −⟨ΔHads⟩, wereprojected onto a two-dimensional space defined by two newcoordinate systems (PC1 and PC2). There is a remarkablecorrelation between four descriptors and COPC, suggesting theleast important role of pore volume. In contrast to MOFs,COFs with small pore sizes and relatively high averagedenthalpy of adsorption are preferential for COPC. When COFspossessed the comparable enthalpy of adsorption (−⟨ΔHads⟩)with that of MOFs, ΔW and pore size played dominant roles inthe cooling efficiency, in which the larger the ΔW, the higherthe COPC. The DT model elucidated the route to identify thebest performers from COFs and MOFs. As shown in Figure 5,the larger the Gini entropy, the more complicate the set. Onethousand six hundred and forty two out of the total 1802structures exhibited COPC ≤ 0.8 and ΔW ≤ 0.21 g/g. On thecontrary, 160 structures showed COPC > 0.8 and ΔW > 0.21g/g, indicating the dominant role of working capacity. Besides,115 out of 160 high-performing structures possessing ΔW >0.21 g/g, −⟨ΔHads⟩ ≤ 45 kJ/mol, and Va > 0.72 cm3/g directedthe preferential route for identifying high-performing candi-dates. In summary, the combination of ΔW > 0.21 g/g,−⟨ΔHads⟩ ≤ 45 kJ/mol, and Va > 0.72 cm3/g provides aneffective pathway to explore potential porous adsorbents forcooling.

3.3. Machine Learning. Given the vast number of COFand MOF structures, predicting AHP performance of eachstructure by molecular simulations is extremely time-consuming. ML was frequently employed to accelerate theevaluation process by using the obtained structure−propertyrelationship from the training materials. Herein, MLR, DT,SVM, GBM, and RF were adopted for predicting the COPCbased on computational screening results of COFs andMOFs.34 LCD, ASA, VF, Va, KH, and adsorbent density (ρ)were chosen for ML because of their critical roles indetermining the physical adsorption performance of COFsand MOFs. To our knowledge, KH has been rarely used asdescriptors in ML for predicting the adsorption performance ofCOFs and MOFs. Herein, KH was recognized as a criticaldescriptor given its correlation with step positions (α) ofadsorption isotherm that is essential for COPC. Besides, KH canbe quickly calculated by Widom insertion within trivial

Figure 4. Cooling performance of COFs and MOFs. (a) PredictedCOPC of MOFs (circle symbols shown in the inset) and COFs (plussymbol) as a function of ethanol Henry’s constant (KH). (b)Predicted COPC of COFs (plus symbol) and MOFs (circle symbolsin the inset) as a function of ethanol Henry’s constant (KH) coloredby the step position (α) of COFs and MOFs. (c) Predicted COPC asa function of ratio of −⟨ΔHads⟩/H(Tev) colored by the step position(α) of COFs and MOFs (shown in the inset). (d) Impacts of LCD,Va, ΔW, and −⟨ΔHads⟩ on the COPC of both COFs and MOFs byPCA.

Figure 5. DT analysis of 1527 MOFs and 275 COFs.

ACS Applied Materials & Interfaces Research Article

DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

3270

Page 7: Screening of Covalent Organic Frameworks for Adsorption

computational time. In each ML model, 80% of COF andMOF data were randomly chosen for training and theremaining 20% for testing. The results from the RF modelshowing the highest prediction accuracy (R2 = 0.89) for boththe training and test sets are presented in Figures 6a and S25.

It should be noted that the prediction accuracy of the test set(R2 = 0.85) is lower than that of the training set (R2 = 0.98).Besides, the prediction accuracy (R2) of COFs (R2 = 0.88) islower than that of MOFs (R2 = 0.91) because of the limitednumber of COFs in ML, leading to the relatively largedeviation from GCMC-predicted COPC (i.e., 13.6% of COFsvs 9.6% of MOFs for the training set and 23.6% of COFs vs16.2% of MOFs for the test set).Comparing the prediction accuracy of ML with GCMC, 98

out of 137 top performers from the initial screening based onGCMC (COPC ≥ 0.85) were successfully identified in ML.The deviation of predicted COPC by RF from GCMCsimulation was less than 0.2 for 96.4% of 1802 structures.Only 57 structures exhibited the deviation greater than 0.2(Table S10). Thirty eight out of the 57 structures are withinthe test set and 39 out of the 57 structures are COFs, whichmay be ascribed to the limited sample size of COFs used fortraining. From the perspective of the COPC ranking of the testset, the Spearman’s rank correlation coefficient between ML-predicted and GCMC-predicted results is 0.81, implicating theconsistency between ML and GCMC. Additionally, the goodagreement between the COPC obtained from GCMC and thatpredicted by ML for the additional 10 hypothetical COFs(hCOFs) randomly chosen from the hCOF database40 wasalso observed (Figure S26). From the aspect of computationalefficiency, compared with the average computational time (46days per adsorbent) of calculating COPC of each structure byGCMC simulations, the average calculation time by ML is lessthan 0.01 s for each structure, which greatly accelerates thescreening of a large number of COFs or MOFs for adsorptioncooling. It is noted that if taking the Henry’s constant (KH)calculation time into account, the average prediction time ofthe COPC of each adsorbent is approximately 1.5 h, which isstill 3 orders of magnitude shorter than GCMC-basedscreening. The relative importance of varying descriptors toCOPC in Figure 6b shows that Va contributed most to coolingperformance (38.08%), followed by KH (23.06%) and LCD(14.09%). Such a tendency is reasonable because the porevolume directly determines the maximum uptake of working

fluids in theory. Henry’s constant plays an important role inthe step position of isotherms. LCD can directly affect thecritical diameter for the condensation of working fluids. Otherdescriptors seem to impose less impacts. On the other hand,the total energy stored (Qstored is equal to adsorption energyQads) per unit volume by each structure was also successfullypredicted (Figure S27) by the RF model, in which LCD playeda dominant role, followed by available pore volume (Va).

4. CONCLUSIONSIn this study, a HTCS of 275 COFs was carried out to discoverthe high-performing adsorbents of AHPs for heating, cooling,and ice making. Eventually, 8 COFs with COPH ≥ 1.85 forheating, 32 COFs with COPC ≥ 0.85 for cooling, and 12 COFswith COPC ≥ 0.85 for ice making were identified. It was foundthat large-pore COFs are favorable for cooling compared withthose for heating and ice making. Structure−property relation-ship analysis demonstrated that COFs are preferential forcooling compared with MOFs, which can be ascribed to theweaker host−adsorbate interaction in COFs that favors thestepwise adsorption and the cooling performance of AHPs.PCA and DT analysis based on integrated COF and MOFdatabase revealed that the suitable porous materials for AHPsshould possess the high working capacity, proper enthalpy ofadsorption (−⟨ΔHads⟩) comparable to the enthalpy ofevaporation, and moderate pore size and pore volume. Inorder to accelerate the computational screening of high-performing adsorbents for AHPs, we carried out ML based onsix descriptors to predict the COPC of a large number of COFsand MOFs. The success of the RF model paves the way forquickly discovering high-performing COFs and MOFs forAHPs. It should be noted that the high cost of COFs may notbe a concern with the development of novel synthesistechniques and increasing demand of COFs in variousapplications. In summary, this work highlighted the potentialof COF adsorbents in AHPs for the first time, which are morepromising candidates for cooling than MOFs. The findingsreported in this work also open up the possibility of quickexploration of high-performing COFs and MOFs for AHPs.

■ ASSOCIATED CONTENT*S Supporting InformationThe Supporting Information is available free of charge athttps://pubs.acs.org/doi/10.1021/acsami.9b20837.

Computational methods; computational screening re-sults of COFs; experimental synthesis, characterization,and vapor adsorption; evolutionary trends of COFs;selected top-performing COFs from computationalscreening; and comparison between COFs and MOFsand ML for MOF and COF data (PDF)

■ AUTHOR INFORMATIONCorresponding Author*E-mail: [email protected].

ORCIDWei Li: 0000-0002-3920-3863Xiaoxiao Xia: 0000-0001-9001-3662Song Li: 0000-0003-3552-3250NotesThe authors declare no competing financial interest.

Figure 6.ML of the MOF and COF database. (a) Predicted COPC bythe RF algorithm vs the COPC from GCMC simulations. Solid circlesymbols represent the training set and the plus symbols are the test set(R2 = 0.89). (b) Relative importance of LCD, ASA, VF, Va, KH, and ρdescriptors obtained from the RF training.

ACS Applied Materials & Interfaces Research Article

DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

3271

Page 8: Screening of Covalent Organic Frameworks for Adsorption

■ ACKNOWLEDGMENTS

This work was supported by Hubei Provincial Nature ScienceFoundation (no. 2019CFB456), double first-class researchfunding of China-EU Institute for Clean and RenewableEnergy (no. ICARE-RP-2018-HYDRO-001) and the Gradu-ates’ Innovation Fund of Huazhong University of Science andTechnology (no. 2019YGSCXCY026). We thank Prof. R.Q.Snurr (Northwestern University) for discussions on dataanalysis and writing of this manuscript. This work was carriedout at National Supercomputer Center in Shenzhen. We alsothank Huazhong University of Science and TechnologyAnalytical & Testing Center for providing support on materialcharacterization.

■ REFERENCES(1) International Energy Outlook; U.S. Energy Information Agency:2017.(2) Chua, K. J.; Chou, S. K.; Yang, W. M. Advances in Heat PumpSystems: A Review. Appl. Energy 2010, 87, 3611−3624.(3) de Lange, M. F.; Verouden, K. J. F. M.; Vlugt, T. J. H.; Gascon,J.; Kapteijn, F. Adsorption-Driven Heat Pumps: The Potential ofMetal−Organic Frameworks. Chem. Rev. 2015, 115, 12205−12250.(4) Demir, H.; Mobedi, M.; Ulku, S. A Review on Adsorption HeatPump: Problems and Solutions. Renew. Sustain. Energy Rev. 2008, 12,2381−2403.(5) Henninger, S. K.; Habib, H. A.; Janiak, C. MOFs as Adsorbentsfor Low Temperature Heating and Cooling Applications. J. Am. Chem.Soc. 2009, 131, 2776−2777.(6) Seo, Y.-K.; Yoon, J. W.; Lee, J. S.; Hwang, Y. K.; Jun, C.-H.;Chang, J.-S.; Wuttke, S.; Bazin, P.; Vimont, A.; Daturi, M. -EfficientDehumidification over Hierachically Porous Metal−Organic Frame-works as Advanced Water Adsorbents. Adv. Mater. 2012, 24, 806−810.(7) Rezk, A.; Al-Dadah, R.; Mahmoud, S.; Elsayed, A. Investigationof Ethanol/Metal Organic Frameworks for Low TemperatureAdsorption Cooling Applications. Appl. Energy 2013, 112, 1025−1031.(8) Elsayed, E.; Al-Dadah, R.; Mahmoud, S.; Elsayed, A.; Anderson,P. A. Aluminium Fumarate and CPO-27(Ni) MOFs: Characterizationand Thermodynamic Analysis for Adsorption Heat Pump Applica-tions. Appl. Therm. Eng. 2016, 99, 802−812.(9) Frohlich, D.; Pantatosaki, E.; Kolokathis, P. D.; Markey, K.;Reinsch, H.; Baumgartner, M.; van der Veen, M. A.; De Vos, D. E.;Papadopoulos, G. K.; Henninger, S. K.; Janiak, C.; Janiak, C. WaterAdsorption Behaviour of CAU-10-H: A Thorough Investigation of ItsStructure−property Relationships. J. Mater. Chem. A 2016, 4, 11859−11869.(10) Wang, S.; Lee, J. S.; Wahiduzzaman, M.; Park, J.; Muschi, M.;Martineau-Corcos, C.; Tissot, A.; Cho, K. H.; Marrot, J.; Shepard, W.;Maurin, G.; Chang, J.-S.; Serre, C. A Robust Large-pore ZirconiumCarboxylate Metal−organic Framework for Energy-efficient Water-sorption-driven Refrigeration. Nat. Energy 2018, 3, 985−993.(11) Henninger, S. K.; Jeremias, F.; Kummer, H.; Janiak, C. MOFsfor Use in Adsorption Heat Pump Processes. Eur. J. Inorg. Chem.2012, 2012, 2625−2634.(12) de Lange, M. F.; van Velzen, B. L.; Ottevanger, C. P.;Verouden, K. J. F. M.; Lin, L.-C.; Vlugt, T. J. H.; Gascon, J.; Kapteijn,F. Metal-Organic Frameworks in Adsorption-Driven Heat Pumps:The Potential of Alcohols as Working Fluids. Langmuir 2015, 31,12783−12796.(13) Nazarian, D.; Camp, J. S.; Chung, Y. G.; Snurr, R. Q.; Sholl, D.S. Large-Scale Refinement of Metal−Organic Framework StructuresUsing Density Functional Theory. Chem. Mater. 2017, 29, 2521−2528.(14) Erdos, M.; de Lange, M. F.; Kapteijn, F.; Moultos, O. A.; Vlugt,T. J. H. In Silico Screening of Metal-Organic Frameworks for

Adsorption-Driven Heat Pumps and Chillers. ACS Appl. Mater.Interfaces 2018, 10, 27074−27087.(15) Li, W.; Xia, X.; Cao, M.; Li, S. Structure−property Relationshipof Metal−organic Frameworks for Alcohol-based Adsorption-drivenHeat Pumps via High-throughput Computational Screening. J. Mater.Chem. A 2019, 7, 7470−7479.(16) Ding, S.-Y.; Wang, W. Covalent Organic Frameworks (COFs):From Design to Applications. Chem. Soc. Rev. 2013, 42, 548−568.(17) Biswal, B. P.; Kandambeth, S.; Chandra, S.; Shinde, D. B.; Bera,S.; Karak, S.; Garai, B.; Kharul, U. K.; Banerjee, R. Pore SurfaceEngineering in Porous, Chemically Stable Covalent Organic Frame-works for Water Adsorption. J. Mater. Chem. A 2015, 3, 23664−23669.(18) Stegbauer, L.; Hahn, M. W.; Jentys, A.; Savasci, G.; Ochsenfeld,C.; Lercher, J. A.; Lotsch, B. V. Tunable Water and CO2 SorptionProperties in Isostructural Azine-Based Covalent Organic Frameworksthrough Polarity Engineering. Chem. Mater. 2015, 27, 7874−7881.(19) Perez-Carvajal, J.; Boix, G.; Imaz, I.; Maspoch, D. The Imine-Based COF TpPa-1 as an Efficient Cooling Adsorbent That Can BeRegenerated by Heat or Light. Adv. Energy Mater. 2019, 9, 1901535.(20) Tong, M.; Lan, Y.; Qin, Z.; Zhong, C. Computation-Ready,Experimental Covalent Organic Framework for Methane Delivery:Screening and Material Design. J. Phys. Chem. C 2018, 122, 13009−13016.(21) Wilmer, C. E.; Kim, K. C.; Snurr, R. Q. An Extended ChargeEquilibration Method. J. Phys. Chem. Lett. 2012, 3, 2506−2511.(22) Widom, B. Some Topics in the Theory of Fluids. J. Chem. Phys.1963, 39, 2808−2812.(23) Dubbeldam, D.; Calero, S.; Ellis, D. E.; Snurr, R. Q. RASPA:Molecular Simulation Software for Adsorption and Diffusion inFlexible Nanoporous Materials. Mol. Simul. 2015, 42, 81−101.(24) Willems, T. F.; Rycroft, C. H.; Kazi, M.; Meza, J. C.;Haranczyk, M. Algorithms and Tools for High-throughput Geometry-based Analysis of Crystalline Porous Materials. MicroporousMesoporous Mater. 2012, 149, 134−141.(25) Aristov, Y. I. Novel Materials for Adsorptive Heat Pumping andStorage: Screening and Nanotailoring of Sorption Properties. J. Chem.Eng. Jpn. 2007, 40, 1242.(26) Rappe, A. K.; Casewit, C. J.; Colwell, K. S.; Goddard, W. A., III;Skiff, W. M. UFF, a Full Periodic Table Force Field for MolecularMechanics and Molecular Dynamics Simulations. J. Am. Chem. Soc.1992, 114, 10024−10035.(27) Chen, B.; Potoff, J. J.; Siepmann, J. I. Monte Carlo Calculationsfor Alcohols and Their Mixtures with Alkanes. Transferable Potentialsfor Phase Equilibria. 5. United-atom Description of Primary,Secondary, and Tertiary Alcohols. J. Phys. Chem. B 2001, 105,3093−3104.(28) Nalaparaju, A.; Zhao, X. S.; Jiang, J. W. MolecularUnderstanding for the Adsorption of Water and Alcohols inHydrophilic and Hydrophobic Zeolitic Metal−Organic Frameworks.J. Phys. Chem. C 2010, 114, 11542−11550.(29) Bueno-Perez, R.; Merkling, P. J.; Gomez-Alvarez, P.; Calero, S.Cadmium−BINOL Metal−Organic Framework for the Separation ofAlcohol Isomers. Chem. - Eur. J. 2017, 23, 874−885.(30) Wilmer, C. E.; Leaf, M.; Lee, C. Y.; Farha, O. K.; Hauser, B. G.;Hupp, J. T.; Snurr, R. Q. Large-scale Screening of Hypothetical Metal-organic Frameworks. Nat. Chem. 2012, 4, 83−89.(31) Smith, B. J.; Dichtel, W. R. Mechanistic Studies of Two-Dimensional Covalent Organic Frameworks Rapidly Polymerizedfrom Initially Homogenous Conditions. J. Am. Chem. Soc. 2014, 136,8783−8789.(32) Zhang, Y.-B.; Su, J.; Furukawa, H.; Yun, Y.; Gandara, F.;Duong, A.; Zou, X.; Yaghi, O. M. Single-Crystal Structure of aCovalent Organic Framework. J. Am. Chem. Soc. 2013, 135, 16336−16339.(33) Ng, K. C.; Burhan, M.; Shahzad, M. W.; Ismail, A. B. AUniversal Isotherm Model to Capture Adsorption Uptake and EnergyDistribution of Porous Heterogeneous Surface. Sci. Rep. 2017, 7,10634−10644.

ACS Applied Materials & Interfaces Research Article

DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

3272

Page 9: Screening of Covalent Organic Frameworks for Adsorption

(34) Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.;Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.;Dubourg, V. Scikit-learn: Machine Learning in Python. J. Mach. Learn.Res. 2011, 12, 2825−2830.(35) Schaul, T.; Bayer, J.; Wierstra, D.; Yi, S.; Felder, M.; Sehnke, F.;Ruckstieß, T.; Schmidhuber, J. PyBrain. J. Mach. Learn. Res. 2010, 11,743−746.(36) Aristov, Y. I. Challenging Offers of Material Science forAdsorption Heat Transformation: A Review. Appl. Therm. Eng. 2013,50, 1610−1618.(37) Ascherl, L.; Sick, T.; Margraf, J. T.; Lapidus, S. H.; Calik, M.;Hettstedt, C.; Karaghiosoff, K.; Doblinger, M.; Clark, T.; Chapman,K. W.; Auras, F.; Bein, T. Molecular Docking Sites Designed for theGeneration of Highly Crystalline Covalent Organic Frameworks. Nat.Chem. 2016, 8, 310−316.(38) Glaznev, I. S.; Ovoshchnikov, D. S.; Aristov, Y. I. Kinetics ofWater Adsorption/Desorption under Isobaric Stages of AdsorptionHeat Transformers: The Effect of Isobar Shape. Int. J. Heat MassTransfer 2009, 52, 1774−1777.(39) Okunev, B. N.; Gromov, A. P.; Aristov, Y. I. Modelling ofIsobaric Stages of Adsorption Cooling Cycle: An Optimal Shape ofAdsorption Isobar. Appl. Therm. Eng. 2013, 53, 89−95.(40) Martin, R. L.; Simon, C. M.; Medasani, B.; Britt, D. K.; Smit, B.;Haranczyk, M. Silico Design of Three-Dimensional Porous CovalentOrganic Frameworks via Known Synthesis Routes and CommerciallyAvailable Species. J. Phys. Chem. C 2014, 118, 23790−23802.

ACS Applied Materials & Interfaces Research Article

DOI: 10.1021/acsami.9b20837ACS Appl. Mater. Interfaces 2020, 12, 3265−3273

3273