Subscriber access provided by Kaohsiung Medical University

Article
Quantitative Nano-Structure-Property Relationships for the Nanoporous
Carbon: Predicting the Performance of Energy Storage Materials
Maike Kaarik, Uko Maran, Mati Arulepp, Anti Perkson, and Jaan Leis
ACS Appl. Energy Mater., Just Accepted Manuscript • DOI: 10.1021/acsaem.8b00708 • Publication Date (Web): 26 Jul 2018
Downloaded from http://pubs.acs.org on July 27, 2018

Just Accepted

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

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

Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.
Page 1 of 36 ACS Applied Energy Materials

1
2
3
4
Quantitative Nano-Structure-Property Relationships for the
5
6 Nanoporous Carbon: Predicting the Performance of Energy Storage
7
8
9 Materials
10
11
12
13
14
15
16 Maike Käärika, Uko Marana, Mati Aruleppb, Anti Perksonb, Jaan Leisa,b,*
17
18
19
20 a
21 Institute of Chemistry, University of Tartu, Ravila 14a, 50411 Tartu, Estonia
22
b
23 Skeleton Technologies, Valukoja 8, 11415 Tallinn, Estonia
24
25
26
27
28
29
30 *corresponding author, e-mail: jaan.leis@ut.ee
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 1
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 2 of 36

1
2
3 Abstract
4
5
6 Nanoporous carbon-based energy storage is a fast-growing research field thanks to high
7
8
energy densities of carbon electrodes with nanoporous amorphous texture. To support the
9
10
11 developments on electrical double-layer based ultra-capacitors it is necessary to improve
12
13 understanding about relationships between the porous structure and energy storage
14
15 behavior of carbon materials. This can be facilitated by the analysis of complex data sets
16
17
18 and the development of corresponding descriptive and predictive models. Related to that
19
20 this paper presents a in silico regression model to predict the suitability of various carbon
21
22 materials for energy storage, thus being probably the first time a quantitative nano-
23
24
25
structure-property relationship (QnSPR) approach is applied to the nanoporous carbon
26
27 materials. With this study, which is based on the experimental data of 100 carbide-
28
29 derived carbon materials, it has been shown that the electrical double-layer capacitance of
30
31
carbon electrode in a nonaqueous electrolyte can be predicted using experimentally
32
33
34 determined specific surface area and a volume of certain pore size fraction of carbon and
35
36 a bulk density of carbon electrode. The three-parameter QnSPR model for volumetric
37
38 cathodic capacitance of carbon in triethylmethylammonium tetrafluoroborate / propylene
39
40
41 carbonate electrolyte, CV,NEG = f(SBET, Vd<1.14, Del), comprising the above-mentioned
42
43 parameters and characterized by R2=0.94 and s2=8.7, confirms the important role of
44
45 carbon pore size for the double layer capacitance. It was shown that carbon pores with a
46
47
48
size below 1.1 nm have the most significance for achieving high energy densities in the
49
50 nonaqueous electrochemical systems studied. Putting the results of this research into
51
52 wider perspective, it has been shown that the QnSPR approach provides a useful tool for
53
54
describing and predicting the variable performance-related physical properties of
55
56
57
58
59 2
60 ACS Paragon Plus Environment
Page 3 of 36 ACS Applied Energy Materials

1
2
3 nanoporous carbon and nanomaterial properties in general. The models are available in
4
5
6 the QsarDB repository (http://dx.doi.org/10.15152/QDB.205).
7
8
9
10
11
12
13
14
15 Keywords: nanoporous carbon, QnSPR, pore size distribution, capacitance, CDC,
16
17 structure-property relationship
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 3
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 4 of 36

1
2
3 Introduction
4
5
6 The nanoporous carbon-based energy storage is a rapidly growing research
7
8 field1,2,3. A flagship of these developments is an ultracapacitor, for which the application
9
10 list extends every year4. To enhance the energy storage capability of ultracapacitors, the
11
12
13
capacitance (C) of carbon electrodes must be increased by enhancing the architecture and
14
15 composition of nanoporous carbon5,6.
16
17 Several studies have demonstrated the importance of specific surface area (SA)
18
19
and average pore size (APS) on capacitance7,8. As a rule, the capacitance increases
20
21
22 proportionally with the increase of total surface area of carbon electrode9. However, in
23
24 the case of BET surface area (SBET), which is the most common characteristic of surface
25
26 properties, the capacitance attains a plateau when SBET becomes more than 1600 m2g-1
27
28 10,11
29 . This kind of capacitance limitation has been explained by a space constriction for
30
31 charge accommodation inside the pore wall11.
32
33 On the other hand, different pore sizes have different impact into the electrical
34
35
36
double layer (EDL) capacitance12, whereas the capacitance per pore volume increases
37
38 with decreasing pore size13. Steric restrictions to ion adsorption have been found to be a
39
40 limiting factor for the capacitance in small micropores14. The dependencies between
41
42
specific capacitance and APS run over the maximum as seen with Mo2C-15 or TiC-
43
44
45 derived carbon16. Yet, the APS does not tell much about the preferred pore size as it
46
47 averages the whole range of pore size distribution, which in turn may vary significantly
48
49 on different origin of carbon materials and their manufacturing history. Important
50
51
52 question thus is whether we can establish the “best pore size” and can we quantify the
53
54 range of suitable pore sizes?
55
56
57
58
59 4
60 ACS Paragon Plus Environment
Page 5 of 36 ACS Applied Energy Materials

1
2
3 All the above considerations give rise to a thorough study of the relationship
4
5
6 between the EDL capacitance and respective nanoporous structure to explain how the
7
8 structure of the material determines the capacitance and which nanostructure properties
9
10 can have the greatest effect on the EDL capacitance. An appropriate method for
11
12
13
conducting such a research is quantitative structure-property relationship (QSPR), which
14
15 began a prominent development in the context of physical organic chemistry (i.e.
16
17 chemical reactivity)17, medical chemistry (drug design)18,19,20, environmental
18
19
chemistry21,22, etc. But not only, the QSPR has an important place in the modeling of
20
21
22 technologically relevant properties23 and materials24, where direct numerical calculations
23
24 are not possible, but fitting methods serve as invaluable tool. QSPR has found
25
26 applications in the study of the properties of materials, precisely because of the triumph
27
28
29 of nanomaterials in various applications and the need to learn how nanostructures work in
30
31 technological processes and influence the surrounding environment. Since the structure
32
33 (or material) is now on a nanoscale, then the method can be shortened subsequently,
34
35
36
QnSPR (Quantitative nano-Structure-Property Relationship), and the structure can be
37
38 characterized both by the experiment-derived structure parameters and/or by the
39
40 calculated parameters.
41
42
In QnSPR, two example pioneering works are on the solubility of fullerene and
43
44
45 polycyclic aromatic hydrocarbons25 and the polarizability of polyaromatic hydrocarbons
46
47 and fullerenes26. However, over the last decade, QnSAR (here Property in abbreviation is
48
49 replaced by Activity) has come a long way in the field of nanomedicine and
50
51
52 nanotoxicology, in last instance particularly due to the regulatory and legislative interest.
53
54 A number of good reviews have been written on this topic covering various models and
55
56
57
58
59 5
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 6 of 36

1
2
3 modelling approaches27,28,29,30,31. Good examples of specific recent work could be related
4
5
6 to experimental detection and computational prediction of nano-bio interactions of gold
7
8 nanoparticles32 and understanding toxicity mechanism of metal oxide nanoparticles to the
9
10 human keratinocyte cell line33. Despite the active development, QnSAR/QnSPR is still in
11
12
13
infancy, literally in fifties and sixties of classical QSAR/QSPR path, since there are few
14
15 data series suitable for modeling, along with missing mechanistic knowledge of processes
16
17 and lack of computational methods that reach to acceptable level for molecular descriptor
18
19
calculation. This makes QnSPR an evolving and perspective method for describing and
20
21
22 modeling properties of nanostructured materials, including nanoporous carbon.
23
24 Classically, molecular descriptors derived from theoretical calculations (e.g., graph
25
26 theory and quantum mechanics) are used to describe molecular structures in the frame of
27
28
29 QSPR. On the other hand, for molecular systems of nanoscale materials as heterogeneous
30
31 structures, today's knowledge does not allow us to calculate the theoretical descriptors
32
33 that we would like or know for small molecules. Therefore, an alternative option is to
34
35
36
compute descriptors describing the structure from experimental measurement data - e.g.,
37
38 porosity characteristics from adsorption measurements - the path that we have chosen in
39
40 this study.
41
42
Reliable modeling and estimation of energy storage properties requires the
43
44
45 rational training data series, large and systematic variability of carbon samples in these
46
47 series. Herein, the developments of last decades made in the field of carbide-derived
48
49 carbon (CDC)34 offer a unique possibility for building large experimental data sets from
50
51
52 carbon materials with a similar history but with notable variability in pore structure and
53
54 pore-size distribution35,36. Beside of the wide choice of possibilities in varying the
55
56
57
58
59 6
60 ACS Paragon Plus Environment
Page 7 of 36 ACS Applied Energy Materials

1
2
3 nanostructure and pore size distribution of CDC, many of these materials already have
4
5
6 been proven as very promising materials for the ultracapacitors37,38,39,40.
7
8 In this study, the CDC data set was assembled from 100 items with a varied
9
10 structure and electrochemical performance giving special attention to reproducibility and
11
12
13
uniformity of experimental properties and measured structural characteristics of carbon
14
15 materials. The measured specific capacitance value derived for each carbon sample was
16
17 correlated to the large number of porosity characteristics, which were recorded by the
18
19
same equipment at the same experimental conditions. The main goal of the research was
20
21
22 to confirm if the experimental adsorption-based characteristics describing the structure of
23
24 nanoporous carbon material can be used in QnSPR models. Another, equally important
25
26 goal was to build a simple descriptive and predictive in silico model for connecting pore
27
28
29 size distribution and the EDL capacitance of carbon materials.
30
31
32
33 Experimental
34
35
36
Data set and experimental descriptors
37
38 The full set of carbon samples for the QnSPR study was composed of 100 CDC
39
40 materials. Powdered carbon samples were synthesized via chlorination of titanium,
41
42
aluminum, silicon, boron and molybdenum carbides at variable reaction conditions.
43
44
45 Pore structure characteristics of powdered carbon samples were evaluated by
46
47 means of low-temperature nitrogen adsorption using Gemini 2375 (Micromeritics).
48
49 Before measurement, the carbon samples (à ~60 mg) were pre-treated in argon flow at
50
51
52 300ºC for 1h. Specific surface area was derived according to the BET theory41 up to the
53
54 nitrogen relative pressure (p/p0) of 0.2, total pore volume was calculated at p/p0 = 0.95,
55
56
57
58
59 7
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 8 of 36

1
2
3 and pore size distribution from the adsorption isotherm according to Barret-Joiner-
4
5
6 Halenda (BJH) theory42. Different pore size fractions were calculated from the
7
8 cumulative PSD data.
9
10 The EDL capacitance of disc-shaped CDC electrodes (0.43 cm2) was tested in
11
12
13
standard glass cell equipped with large counter electrode (apparent area of ~30 cm2) and
14
15 saturated calomel electrode (SCE). Before testing, the carbon electrodes were vacuum-
16
17 dried at 180°C for 2 hours. During experiments the electrolyte solution was saturated
18
19
with argon (AGA, 99.999%) to avoid the contamination by oxygen. CDC electrodes were
20
21
22 made by roll-pressing technique with using 6% (wt.) of PTFE (Aldrich, 60% suspension
23
24 in water). To secure good reproducibility of electrochemical characteristics, the contact
25
26 resistance between carbon electrode and current collector was improved by thin (~2 µm)
27
28
29 aluminum PVD coating on one side of the carbon electrode. Cathodic and anodic
30
31 capacitances in 1M triethylmethylammonium tetrafluoroborate ((C2H5)3CH3NBF4, Stella,
32
33 H2O<0.005%) solution in propylene carbonate (PC, anhydrous, Merck Selectipur®) were
34
35
36
calculated at 5 mAcm-2 from the constant current discharge (CCD) experiments at the
37
38 potential range of 0 to -2.5 V and 0 to +2.5 V, respectively. The Potentiostat-Galvanostat
39
40 1286 Solartron and FRA 1255 Solartron were used for all electrochemical measurements.
41
42
43
44
45 Formation of QnSPR models
46
47 QnSPR models were developed using best multiple linear regression method.
48
49 BMLR analysis was performed using CODESSA program43. BMLR follows a stepwise
50
51
52 forward algorithm to select significant descriptors while building a multiple linear model.
53
54 During the BMLR procedure, the pool of descriptors is cleaned of non-significant
55
56
57
58
59 8
60 ACS Paragon Plus Environment
Page 9 of 36 ACS Applied Energy Materials

1
2
3 descriptors (R2<0.1) and the descriptors with missing values followed by the construction
4
5
6 of the best two-parameter regression, the best three-parameter regression, etc. based on
7
8 the statistical significance and non-collinearity criteria (R2<0.7) of the selected
9
10 descriptors. The descriptor scales are normalized and centered automatically, with the
11
12
13
final result given in natural scales. The final model has the best representation of the
14
15 activity or property in the given descriptor pool within the given number of parameters44.
16
17 The quality of the models was assessed by the square of coefficient of determination (R2),
18
19
square of cross-validated (leave-one-out) coefficient of determination (R2CV), square of
20
21
22 standard error of the estimate (s2), and the Fisher’s criterion (F)45. The final QSAR
23
24 models (together with data) were provided in the QSAR Data Bank format46 and were
25
26 uploaded to the QsarDB repository (http://dx.doi.org/10.15152/QDB.205)47,48.
27
28
29
30
31 Model diagnostics
32
33 A Williams plot was used for model diagnosis and the applicability domain
34
35
36
analysis of the derived model. This plot is based on the leverage (hi) and standardized
37
38 prediction error (r’) of each compound and generates a graphical representation of
39
40 structurally different and statistically deviating compounds. The leverages (hi) of the
41
42
chemicals (i) in the descriptor space are the diagonal elements of the influence matrix:
43
44
45 H=X(A)-1XT, where A=(XTX). Here, X is the matrix of descriptors (chemicals ×
46
47 descriptors), and the leverage is calculated from the molecular descriptors as follows:
48
49 hi=xi(A)-1xiT, where x is the descriptor row-vector of a query chemical.49,50 Compounds
50
51
52 with leverage values larger than the critical leverage value h*=3(k + 1)/n, where k is the
53
54 number of descriptors in the model and n is the number of compounds in the data set, can
55
56
57
58
59 9
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 10 of 36

1
2
3 be considered structurally different and thus may result in unreliable predictions51,52.
4
5
6 Standardized residuals indicate moderate outliers with ±2r’ (95% confidence limit) and
7
8 strong outliers with ±3r’ (98% confidence limit) according to the Gaussian distribution.
9
10
11
12
13 Results and discussion
14
15
16 Nanoporous carbons
17
18 CDC materials used have been synthesized from several metal or metalloid
19
20
carbides by removing the non-carbon elements in high-temperature chlorine atmosphere.
21
22
23 Since formation of carbon-carbon bonds depends on the distance and chemical
24
25 surrounding of carbon atoms in carbide crystal lattice, the nanostructure and pore-size
26
27 distribution of CDCs can be significantly influenced by varying the elemental
28
29
30 composition and crystal structure of parent carbide. Another important and fortunately
31
32 well-controlled parameter, which is responsible for the structural order and porosity of
33
34 CDC, is a temperature of chlorination. A rule of thumb is that higher temperature
35
36
37
produces higher structural order, while also increasing the average pore-size of CDC.
38
39 Altogether, these factors give a large matrix of variables for synthesizing a vast number
40
41 of carbon samples – all CDC, but still different materials with well controlled structural
42
43
variability.
44
45
46 Current selection of CDC materials reflects the wide variability of nanoporous
47
48 carbons: 53 samples were derived from TiC, 14 samples from Mo2C, 19 samples from
49
50 Al4C3, 10 samples from SiC and 4 samples were made from B4C. From them 19 samples
51
52
53 were treated during chlorination in situ with TiO2 and 21 samples were made via step-by-
54
55 step chlorination at varied temperatures, giving representatives of CDC with hierarchical
56
57
58
59 10
60 ACS Paragon Plus Environment
Page 11 of 36 ACS Applied Energy Materials

1
2
3 pore size distribution, i.e., the CDC where the sizes of pores in different layers of carbon
4
5
6 particle are different53. All CDC samples were additionally annealed in hydrogen at
7
8 800ºC and approximately half of them, i.e., 53 carbons, were post-activated at 900ºC by
9
10 H2O trapped in nanopores54.
11
12
13
14
15 Volumetric capacitance of CDC
16
17 Electrochemical performance of CDC materials at positive (anodic) and negative
18
19
(cathodic) potentials was evaluated in three-electrode cells using 1M
20
21
22 triethylmethylammonium tetrafluoroborate (TEMABF4) solution of propylene carbonate
23
24 (PC) as the electrolyte. The range of variability and distribution of volumetric
25
26 capacitance, CV [F cm-3], of all CDC samples of this study is shown in Figure 1.
27
28
29 Why volumetric capacitance? It is true that gravimetric capacitance, Cg [F g-1], is
30
31 often preferred for characterization of the electrode materials, because the weighing is
32
33 more precise procedure than volume calculations due to the inhomogeneity in the
34
35
36
thickness of the electrode film. However, the Cg value, although related to the specific
37
38 surface area, SA [m2 g-1], of the electrode, has almost no relationship to the efficient
39
40 packaging of a surface area in volume and, therefore, is of significantly less value from
41
42
the practical point of view. It must be noticed that the size of the ultracapacitor is
43
44
45 inversely related to the density of electrodes, which in turn is proportional to the
46
47 volumetric capacitance.
48
49 In Figure 1, all 100 carbon materials have been plotted on x-axis in a sequence of
50
51
52 increasing cathodic capacitance (CV,NEG). At lower values of volumetric capacitance,
53
54 which mostly corresponds to the CDC materials with very low content of pores in a
55
56
57
58
59 11
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 12 of 36

1
2
3 nanometer range (cf. Table S2), the anodic and cathodic capacitances are almost equal,
4
5
6 whereas the gap between them notably increases at higher capacitance values. This
7
8 phenomenon can be explained by the match between size of the pore and the electrolyte
9
10 ion, that is, the larger the coincidence, the higher the capacitance. In the electrolyte of
11
12
13
present study, the tetrafluoroborate (BF4-) anions are smaller than
14
15 triethylmethylammonium ((C2H5)3CH3N+) cations, approximately 0.46 nm vs. 0.66 nm,
16
17 respectively55. Therefore, the anodic capacitance exceeds the cathodic one, when pore-
18
19
size becomes the limiting factor in nanoporous carbon electrodes.
20
21
22
23
24 Experiment-derived structural descriptors for nanoporous carbon
25
26 For the QnSPR analysis the set of 100 CDC samples was characterized with the
27
28
29 124 experimentally determined descriptors, i.e., characteristics describing the
30
31 nanostructure of the nanoporous carbon. 123 of descriptors were derived from the N2
32
33 adsorption measurements and include: BET surface area (SBET [m2 g-1]), the total pore
34
35
36
volume (Vt [cm3 g-1]) and volumes of 121 different pore size fractions (Vx-y and V<y [cm3
37
38 g-1], where x and y indicate the lowest and highest pore size of the fraction in
39
40 nanometers) derived from BJH pore size distribution (see Table S1). When all above pore
41
42
size descriptors characterize the carbon in powder form, then the measurable property –
43
44
45 volumetric capacitance – is related to the volume of carbon electrode, i.e., the carbon
46
47 powder pressed into carbon film using polymeric binder. Therefore, one additional
48
49 descriptor, bulk density of carbon electrode (Del), was included to describe the
50
51
52 compressed conditions of the carbon particles in the electrode film. Here, the bulk density
53
54
55
56
57
58
59 12
60 ACS Paragon Plus Environment
Page 13 of 36 ACS Applied Energy Materials

1
2
3 is defined as Del = Sel h / m, where Sel is a surface area, h is a thickness and m is a weight
4
5
6 of the disc-shape electrode sample after drying in vacuum and backfilling with argon.
7
8 ε
9 C= A (1)
4πd
10
11 According to theory, the EDL capacitance of a carbon electrode is directly related
12
13
to the specific surface area of carbon. The capacitance (Eq. (1)) of parallel-plate capacitor
14
15
16 is proportional to the area (A) of the electrode plate56. In the case of EDLC, the A can be
17
18 defined as the surface area of the carbon/electrolyte interface, i.e., the accessible surface
19
20 area of the carbon.
21
22
23 There are many physisorption and calorimetric methods for quantifying the
24
25 accessible surface area of porous materials57. The problem is that none of those is
26
27 universal and completely reliable over the wide range of pore sizes. In this study, the
28
29
30
SBET, even though known to overestimate the surface area of low-density carbon
31
32 materials9,58, was chosen as the most conventional molecular structure characteristic of
33
34 accessible surface area, and therefore, readily available for majority of porous carbon
35
36
materials.
37
38
39 Nevertheless, it was quite surprising to see no correlation between specific
40
41 capacitance and BET surface area for the full set of 100 CDC materials (see Figure 2).
42
43 Visual analysis indicates that no correlation is due to the group of data points that have
44
45
46 systematically higher surface areas, while capacitance is moderate. At first glance, this
47
48 result warns about the use of SBET for structure-property analysis. On the other hand, the
49
50 lack of correlation is extremely informative since only a certain fraction of SBET is
51
52
53
effective for EDL formation. Relative amount of effective surface area is different in each
54
55 material and depends on the contributions from other structural characteristics. The
56
57
58
59 13
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 14 of 36

1
2
3 distribution of pore size is presumably the one that plays a crucial role here and,
4
5
6 therefore, a combined effect of surface area and pore size distribution is expected to
7
8 improve mechanistic description for the capacitance.
9
10
11
12
13
Intuitive QnSPR for full set of CDC materials
14
15 QnSPRs can be derived in multiple ways from manual to fully automatic manner.
16
17 At first the improvement for the relationship between specific capacitance and BET
18
19
surface area (see Figure 2) is sought via intuitive-manual step-by-step improving of linear
20
21
22 regression models. For this the number of the descriptors is increased in the model one by
23
24 one and statistical characteristics are verified thereafter. The additional descriptor to SBET
25
26 were selected from pore size fractions (V<y) calculated from the cumulative pore size
27
28
29 distribution according to BJH. The index y in V<y values corresponds to the upper value
30
31 of the pore size fraction selected in between 1.0 nm and 1.4 nm.
32
33 For predicting volumetric capacitance of positively charged electrode the top five
34
35
36
two-parameter multi-linear regression models (Table 1) obtained in this manner are
37
38 ranked according to the square of the correlation coefficient, R2. The values of the R2
39
40 equations are very similar, which impedes the comparison. The difference of the
41
42
correlation equations has been brought out by s2 values that differ significantly from each
43
44
45 other and are recommended to be used in analogous situations by Winkler and co-
46
47 workers59. The best model (Table 1: #2, Eq. (2)) includes BET surface area and volume
48
49 of pore size fractions less than 1.06 nm.
50
51
52 CV,POS = (1.11×10-2±1.9×10-3) SBET
53 + (119±3) Vd<1.06
54 (2)
55 + 21.1±3.3
56
57
58
59 14
60 ACS Paragon Plus Environment
Page 15 of 36 ACS Applied Energy Materials

1
2
3 For the negatively charged electrode, the best two-parameter multi linear
4
5
6 regression model (Table 1: #8) includes BET surface area and volume of pore size
7
8 fraction less than 1.14 nm (Eq. (3)).
9
10 CV,NEG = (1.23×10-2±1.5×10-3) SBET
11
12 + (89±2) Vd<1.14 (3)
13
14 + 17.4±2.6
15 It is important to note that in both cases, for anodic and cathodic capacitance, the
16
17
18 quality of two-parameter QnSPR model improves with decreasing the pore-size, whereas
19
20 the best correlation is achieved with the pores smaller than 1.06 nm and 1.14 nm, for
21
22 positively and negatively charged electrode, respectively.
23
24
25 This result contradicts the thinking that the ions can electrosorb on the surface of
26
27 porous carbon electrodes only with solvent shells intact, because it would require the
28
29 pores larger than 2 nm to fit the solvated organic ions of aprotic electrolytes60. On the
30
31
32
contrary, a very good correlation (cf. Table 1) achieved for the pores with a size below
33
34 1.1 nm supports the more recent observations of increased capacitance in subnanometer
35
36 pores61.
37
38 Based on the differences in anodic and cathodic capacitance observed in Figure 1,
39
40
41 we could expect a bigger gap between preferred pore sizes at negatively and positively
42
43 charged electrodes, because the tetraalkylammonium cations of the electrolyte used are
44
45 _
46 by ~0.2 nm larger than BF4 anions. Statistical models and particularly molecular
47
48 descriptors in the models do not show such a clear difference for the preferred pore sizes
49
50 by anions and cations. It may be that the smallest pore size, which was possible to
51
52
53 describe with BJH data in this research, is close to but still larger than the “best fitted”
54
55 pore size for anions.
56
57
58
59 15
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 16 of 36

1
2
3
4
5
6 Automatic QnSPR for full set of CDC materials
7
8 In order to check that the previous derived intuitive QnSPR-s were not a
9
10 coincidence, and whether there are other combinations of descriptors for the regression
11
12
13
equations the automatic Best Multi Linear Regression (BMLR) approach was used44. The
14
15 BMLR uses systematic forward selections of descriptors to analyze the set of 124
16
17 experiment-derived descriptors and results in the best possible in silico model for the
18
19
given set of descriptors.
20
21
22 A question that often rises with multiple-parameter QSPR models is the
23
24 reasonable number of parameters for describing the property chosen. To keep the
25
26 predictive models simple and easy to use it is desirable to involve as few parameters as
27
28
29 possible. However, the decision is very much intuitive and depends on the reasonable
30
31 interpretation of selected descriptors and the improvement of model`s quality with each
32
33 additional descriptor in terms of R2, RCV2 and s2. A simple method to help in finding
34
35
36
correlation improvement cut-off has been introduced by us earlier62 and helps to verify
37
38 the impact of each additional parameter to the quality of regression model, i.e., allows
39
40 determining the crossing point (correlation improvement cut-off) of trend lines on the
41
42
plot of R2 vs. descriptor number. According to this, the three descriptors is the optimum
43
44
45 number for the BMLR model of the volumetric capacitance for 100 CDC materials
46
47 (further identified as set (A)), and the third descriptor has small improvement in terms of
48
49 R2 value (see Figure S1).
50
51
52 The three descriptors in this model (Eq. (3)) for the cathodic EDL capacitance
53
54 comprise of BET surface area of carbon powder (SBET), a volume fraction of the pores
55
56
57
58
59 16
60 ACS Paragon Plus Environment
Page 17 of 36 ACS Applied Energy Materials

1
2
3 with a size of less than 1.14 nm (Vd<1.14) and a bulk density of the carbon electrode film
4
5
6 (Del). Numeric values of descriptors for all 100 CDC samples can be found in Table S2.
7
8 This result is in good agreement with the two-parameter models (Eqs. (2) and (3))
9
10 worked out by the intuitive-manual step-by-step improving approach, which confirms
11
12
13
that the pores with a size below 1.1 nm have the highest significance in predicting the
14
15 EDL capacitance of carbon materials in TEMABF4/PC electrolyte.
16
17
18 CV,NEG = (1.84×10-2±2.2×10-3) SBET
19
20
21 + (73±5) Vd<1.14
22 (4)
23 + (39.6±10.7) Del
24
25 – 11.1±8.1
26
27
28
29 A correlation between calculated and experimental values of volumetric cathodic
30
31 capacitance is graphically represented on Figure 3a. Statistical impact of descriptors in
32
33 this model is given in Table 2 (http://dx.doi.org/10.15152/QDB.205). As one can observe
34
35
36
all descriptors in the equation do have positive correlation with volumetric cathodic
37
38 capacitance. This means that when the quantity of pores with a size of less than 1.14 nm
39
40 (Vd<1.14) increases, also the volumetric cathodic capacitance increases. Same applies to
41
42
BET surface area of carbon powder (SBET) and bulk density of the carbon electrode film
43
44
45 (Del). The coefficients of the descriptors and t-test values show that the Vd<1.14 has also
46
47 highest contribution to the equation and describes therefore most of the structural
48
49 variation in the model. The high impact of Vd<1.14, however, is quite expected if to
50
51
52 analyze respective single descriptor correlations presented for all three descriptors on
53
54 Figure 2. According to the t-test the second descriptor, SBET, considerably improves
55
56
57
58
59 17
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 18 of 36

1
2
3 description of structural variation of the model via increasing statistical performance of
4
5
6 the model by 0.059 R2 units. Smallest improvement to the equation is provided by the
7
8 third descriptor, Del, which eventually turns out to be important, because the density of
9
10 the carbon electrode film includes into the equation compressed conditions of the carbon
11
12
13
particles in this way accounting also the bulk properties of the carbon material in addition
14
15 to powder properties exposed by two previous descriptors. In case of all three descriptors
16
17 one can conclude that to improve volumetric cathodic capacitance of the nanoporous
18
19
carbon material one needs to improve all three structural parameters simultaneously and
20
21
22 in combination. Having said that, one should remember that beside of specific double-
23
24 layer capacitance, there are also other issues, like electrical conductivity and surface
25
26 morphology of carbon particles, which should be carefully controlled and adjusted for the
27
28
29 carbon electrode material. Furthermore, the electrochemical performance and suitability
30
31 of carbon material for the energy storage depends on the electrode manufacturing
32
33 technology and the binding materials used in the electrode composition63.
34
35
36
Validation statistics expressed through leave-one-out cross-validation shows very
37
38 good agreement with square of correlation coefficient (Table 2), both coefficients differ
39
40 in 0.004 R2 units. Analysis of Williams plot (Figure 3b) shows that the model derived is
41
42
stable, without strong outliers and none of the samples are with high leverage. Five data
43
44
45 points are moderate statistical outliers and five nanomaterials are with higher leverage
46
47 than critical leverage, but still very close to the threshold line. Going into more details,
48
49 the outliers and the samples with higher leverage value mostly are related to the
50
51
52 materials, which are partly mesoporous or have larger micropores. Apparently, these kind
53
54 carbons are poorly described by given three-parameter QnSPR model, which may be
55
56
57
58
59 18
60 ACS Paragon Plus Environment
Page 19 of 36 ACS Applied Energy Materials

1
2
3 caused by incompatibility of SBET descriptor. It is known that the BET theory estimates
4
5
6 surface area values somewhat higher in case for carbon materials with a pore size greater
7
8 than 1 nm 9. A quality of the QnSPR model is also likely influenced by the degree of
9
10 purity of the carbon material that may have been subject to variation from sample to
11
12
13
sample.
14
15
16
17 Influence of TiO2 derivatives to the capacitance.
18
19
Several carbon materials of this study were derived from carbide/TiO2
20
21
22 composites, which may leave in the carbon some residues or derivatives of TiO2 and by
23
24 that affect the electrochemical capacitance by adding possible faradic components to the
25
26 EDL capacitance. Therefore, it was reasonable to verify if these materials and possible
27
28
29 impurities from TiO2 are not affecting the quality of prediction model (Eq. (4)). For this,
30
31 a modified data set of 81 CDC materials (set (B)) was separated with excluding all those
32
33 made from carbide/TiO2 composites. Statistical analysis of data set (B) by means of
34
35
36
BMLR produced the three-parameter model comprising the same descriptors as for full
37
38 set (A) of 100 CDC. Comparison of R2 values, 0.941 and 0.943 for the data sets (A) and
39
40 (B), respectively, confirms that the CDC made from carbide/TiO2 composite, even if it
41
42
contains traces of the oxide, does not have a notable negative impact on the quality of
43
44
45 prediction model for a set (A).
46
47
48
49 Influence of H2O-activation of nanopores: external validation case.
50
51
52 The reliability of three-parameter model for the prediction of volumetric double
53
54 layer capacitance of porous carbon was additionally studied through partition of data set
55
56
57
58
59 19
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 20 of 36

1
2
3 into training and external validation sets. For this the full set (A) of 100 CDC was divided
4
5
6 into training set (C) including 53 carbon samples, which had been post-treated with H2O-
7
8 activation in nanopores54, and external validation set (D) of the rest 47 samples
9
10 comprising the carbon materials without H2O-activated nanopores. This allowed to
11
12
13
understand the influence of H2O-activation to nanopores, particularly, to see if there is
14
15 need for introducing the additional descriptors into the QnSPR model (Eq. (4)) for
16
17 describing the changes in surface chemistry due the oxidation of carbon surface, and to
18
19
see if newly derived model bares similarity with previous models and what is the external
20
21
22 prediction quality for the model. Indeed, the following search for the best multi-linear
23
24 regression model validated the reliability of the QnSPR model (Table 3,
25
26 (http://dx.doi.org/10.15152/QDB.205)). The descriptors for the training set are exactly
27
28
29 the same as those used for the full set of 100 CDC samples (Table 1 and Table 2): Vd<1.14,
30
31 SBET, and Del. The best three-parameter multiple-linear regression model (Figure 4, Table
32
33 3) derived with BMLR for CV,NEG of a training set (C) of 53 CDC is characterized by R2
34
35
36
of 0.944, F = 276 and s2 = 7.63. Also, the internal validation statistics expressed through
37
38 leave-one-out cross-validation (R2CV = 0.936) shows good agreement with square of
39
40 correlation coefficient (Table 3).
41
42
The model obtained for the training set (C) was used to calculate the volumetric
43
44
45 cathodic capacitance for the external validation set (D) carbons, which were compared
46
47 with the respective experimental values (Table S3). The square of the standard error for
48
49 the predicted values of the validation set compounds is s2 = 11.28. Correlation coefficient
50
51
52 between calculated and experimental properties of the external validation set is R2ext =
53
54 0.932.
55
56
57
58
59 20
60 ACS Paragon Plus Environment
Page 21 of 36 ACS Applied Energy Materials

1
2
3
4
5
6 Conclusion
7
8
9 In this research specific electrical double-layer capacitance of wide variability of
10
11 porous carbide-derived carbon materials was analyzed using the QnSPR approach. The
12
13
14
results allow general conclusion that the volumetric capacitance of porous carbon
15
16 materials can be estimated with a reasonable accuracy using the surface area and pore
17
18 size distribution data combined into one in silico model. Analyzed and studied QnSPR-s
19
20
show that double layer capacitance is greatly influenced by the amount of nanopores of
21
22
23 less than 1.1 nm, which strongly supports the hypothesis that the electrolyte ions are at
24
25 least partly desolvated when electrosorbed on the surface of nanoporous carbon.
26
27 The BMLR study revealed that the volumetric capacitance of nanoporous carbon
28
29
30 electrode can be predicted using a three-parameter QnSPR equation comprising a
31
32 combination of characteristics of carbon powder and compacted electrode: specific
33
34 surface area and a volume of pore size fraction of the carbon and a bulk density of
35
36
37
compacted carbon electrode. Hence, the QnSPR model proposed involves
38
39 mechanistically relevant descriptors, which can be easily derived from the routine
40
41 nitrogen adsorption measurements, and therefore do offer a good assistance for
42
43
developing the high-surface area carbon electrodes.
44
45
46 Importantly current QnSPR shows that when structure of nanoporous carbon
47
48 material is well quantified and data set systematically covers that structural variation the
49
50 derived in silico model allows mechanistically justified explanation on the level of
51
52
53 physico-chemical phenomenon. Also the experimental adsorption-based characteristics
54
55
56
57
58
59 21
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 22 of 36

1
2
3 describing the structure of nanoporous carbon material can be successfully used in
4
5
6 QnSPR models.
7
8 Obviously, the quality of QnSPR prediction models for nanocarbons can be
9
10 further improved by introducing the descriptors from theoretically more consistent and
11
12
13
more detailed pore size distribution calculations instead of BJH. Further developments
14
15 should also consider characteristics for subnanometer pores by using micropore analysis
16
17 techniques64 and the data reduction methods based on a density functional theory65.
18
19
20
21
22
23 Supporting material
24
25
26
Dependence on the quality of QnSPR model on the number of parameters in BMLR
27
28 equation (Figure S1). A list of descriptors for nanoporous carbons calculated from
29
30 cumulative pore size distribution (Table S1). Properties and selected descriptors for the
31
32
100 CDC materials used in QnSPR models (Table S2). External validation of QnSPR
33
34
35 model (Table S3).
36
37
38
39
40 Acknowledgements
41
42
43 Estonian Ministry for Education and Research (grant IUT34-14). Former colleagues from
44
45
Tartu Tehnoloogiad are warmly acknowledged for the assistance in collecting the CDC
46
47
48 database for this research. Many thanks go to Mrs. Liina Grauberg for the invaluable aid
49
50 in electrochemical testing of CDC materials.
51
52
53
54
55
56
57
58
59 22
60 ACS Paragon Plus Environment
Page 23 of 36 ACS Applied Energy Materials

1
2
3 Figures
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26 Figure 1. Experimental values of anodic (CV,POS) and cathodic (CV,NEG) volumetric
27
28 constant current discharge (CCD) capacitance of a set of 100 CDC materials.
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53 Figure 2. Correlation between CV,NEG and single descriptors involved in QnSPR models.
54
55
56
57
58
59 23
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 24 of 36

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23 Figure 3. Experimental CV,NEG vs. QnSPR calculated CV,NEG for 100 CDC (set A)
24
25
obtained with BMLR (a), and corresponding Williams plot (b).
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50 Figure 4. Correlation between experimental and calculated CV,NEG values of the training-
51
52 set C (53 structures) and validation set (D) (47 structures).
53
54
55
56
57
58
59 24
60 ACS Paragon Plus Environment
Page 25 of 36 ACS Applied Energy Materials

1
2
3 Tables
4
5
6
7 Table 1. The two-parameter regression models for volumetric anodic capacitance
8
9
(CV,POS) and volumetric cathodic capacitance (CV,NEG) obtained with improving the one-
10
11
12 parameter regression model.
13
14
15 # ND desc1 improved R2 F s2 RCV2
16
volumetric anodic capacitance (CV,POS), regression with BET surface
17
18 1 1 SBET 0.01 1 211.3 0.03
19
20 volumetric anodic capacitance (CV,POS), 2-parameter regression
21
2 2 SBET Vd<1.06 0.93 624 15.6 0.93
22
23 3 2 SBET Vd<1.14 0.92 576 16.8 0.92
24
25 4 2 SBET Vd<1.22 0.91 515 18.6 0.91
26 5 2 SBET Vd<1.28 0.91 462 20.6 0.90
27
28 6 2 SBET Vd<1.35 0.90 418 22.5 0.90
29
30 volumetric cathodic capacitance (CV,NEG), regression with BET surface
31 7 1 SBET 0.00 0 144.7 0.50
32
33 volumetric cathodic capacitance (CV,NEG), 2-parameter regression
34
35 8 2 SBET Vd<1.14 0.93 673 9.8 0.93
36 9 2 SBET Vd<1.06 0.93 657 10.1 0.93
37
38 10 2 SBET Vd<1.22 0.93 635 10.4 0.93
39
11 2 SBET Vd<1.28 0.92 583 11.2 0.92
40
41 12 2 SBET Vd<1.35 0.92 531 12.2 0.91
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 25
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 26 of 36

1
2
3
4 Table 2. The best 3-parameter regression model for volumetric cathodic capacitance
5
6
7 (CV,NEG) of CDC materials (set (A)), obtained with BMLR.
8
9
# X DX t-test descriptor R2 s2 R2cv
10
11
12 0 -1.109×10+1 8.049×100 -1.4 Intercept
13
14 1 7.293×10+1 4.980×100 14.7 Vd<1.14 0.89 16.4 0.88
15
16
2 1.838×10-2 2.171×10-3 8.5 SBET 0.93 9.8 0.93
17
18
19 3 3.957×10+1 1.066×10+1 3.7 Del 0.94 8.7 0.94
20
21
22
23
24
25
26
27
28
29 Table 3. The best 3-parameter regression model according to BMLR for volumetric
30
31 cathodic capacitance (CV,NEG) of the training set (C) of 53 CDC.
32
33
34 # X DX t-test descriptor R2 s2 R2cv
35
36
0 -1.789×10+1 1.079×10+1 -1.7 Intercept
37
38
39 1 7.160×10+1 6.394×100 11.2 Vd<1.14 0.90 13.2 0.84
40
41 2 1.926×10-2 3.129×10-3 6.2 SBET 0.93 9.3 0.90
42
43 3 4.906×10+1 1.408×10+1 3.5 Del 0.94 7.6 0.94
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 26
60 ACS Paragon Plus Environment
Page 27 of 36 ACS Applied Energy Materials

1
2
3 References
4
5
6 1
Kötz, R.; Carlen, M. Principles and Applications of Electrochemical Capacitors.
7 Electrochim. Acta 2000, 45, 2483–2498.
8 2
9 Burke, A. R&D Considerations for the Performance and Application of
10 Electrochemical Capacitors. Electrochim. Acta 2007, 53, 1083–1091.
11 3
Beguin, F.; Frackowiak, E. Supercapacitors, John Wiley & Sons, 2013.
12 4
13 Yu, A.; Chabot, V.; Zhang, J. Supercapacitors for Energy Storage and Delivery,
14 CRC Press, 2017; pp. 317–335.
15 5
Simon, P.; Gogotsi, Y. Materials for Electrochemical Capacitors. Nat. Mater.
16 2008, 7, 845–854.
17 6
18 Gu, W.; Yushin, G. Review of Nanostructured Carbon Materials for
19 Electrochemical Capacitor Applications: Advantages and Limitations of
20 Activated Carbon, Carbide-Derived Carbon, Zeolite-Templated Carbon, Carbon
21 Aerogels, Carbon Nanotubes, Onion-Like Carbon, and Graphene. WIREs Energy
22 Environ 2014, 3, 424–473.
23 7
24 Chmiola, J.; Yushin, G.; Dash, R.; Gogotsi, Y. Effect of Pore Size and Surface
25 Area of Carbide Derived Carbons on Specific Capacitance. J. Power Sources
26 2006, 158, 765–772.
27 8
Raymundo-Piñero, E.; Kierzek, K.;Machnikowski, J.; Béguin, F. Relationship
28
29 between the Nanoporous Texture of Activated Carbons and Their Capacitance
30 Properties in Different Electrolytes. Carbon 2006, 44, 2498–2507.
31 9
Lobato, B.; Suàrez, L.; Guardia, L.; Centeno, T.A. Capacitance and Surface of
32 Carbons in Supercapacitors. Carbon 2017, 122, 434–445.
33 10
34 Wang, L.; Toyoda, M.; Inagaki, M. Dependence of Electric Double Layer
35 Capacitance of Activated Carbons on the Type of Pores and Their Surface Areas.
36 New Carbon Mater. 2008, 23, 111–115.
37 11
Barbieri, O.; Hahn, M.; Herzog, A.; Kötz, R. Capacitance Limits of High Surface
38
39
Area Activated Carbons for Double Layer Capacitors. Carbon 2005, 43, 1303–
40 1310.
12
41 Jäckel, N.; Rodner, M.; Schreiber, A.; Jeongwook, J.; Zeiger, M.; Aslan, M.;
42 Weingarth, D.; Presser, V. Anomalous or Regular Capacitance? The Influence of
43 Pore Size Dispersity on Double-Layer Formation. J. Power Sources 2016, 326,
44
45
660–671.
13
46 Centeno, T.A.; Stoeckli, F. The Volumetric Capacitance of Microporous Carbons
47 in Organic Electrolyte. Electrochem. Comm. 2012, 16, 34–36.
48 14
49
Segalini, J.; Iwama, E.; Taberna, P.-L.; Gogotsi, Y.; Simon, P. Steric Effects in
50 Adsorption of Ions from Mixed Electrolytes into Microporous Carbon.
51 Electrochem. Comm. 2012, 15, 63–65.
52 15
Leis, J.; Arulepp, M.; Käärik, M.; Perkson, A. The Effect of Mo2C Derived
53 Carbon Pore Size on the Electrical Double-Layer Characteristics in Propylene
54
55
Carbonate-Based Electrolyte. Carbon 2010, 48, 4001–4008.
56
57
58
59 27
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 28 of 36

1
2
3
4 16
Perez, C.R.; Yeon, S.-H.; Ségalini, J.; Presser, V; Taberna, P.-L.; Simon, P.;
5
6 Gogotsi, Y. Structure and Electrochemical Performance of Carbide-Derived
7 Carbon Nanopowders. Adv. Funct. Mater. 2013, 23, 1081–1089.
8 17
Hansch, C.; Leo, A.; Taft, R.W. A Survey of Hammett Substituent Constants and
9 Resonance and Field Parameters. Chem. Rev. 1991, 91, 165–195.
10 18
11 Hansch, C.; Hoekman, D.; Gao, H. Comparative QSAR: Toward a Deeper
12 Understanding of Chemicobiological Interactions. Chem. Rev. 1996, 96, 1045–
13 1075.
14 19
Hansch, C.; Hoekman, D.; Leo, A.; Weininger, D.; Selassie, C.D. Chem-
15
16 Bioinformatics: Comparative QSAR at the Interface Between Chemistry and
17 Biology. Chem. Rev. 2002, 102, 783–812.
18 20
Cherkasov, A.; Muratov, E.N.; Fourches, D.; Varnek, A.; Baskin, I.I.; Cronin, M.;
19 Dearden, J.; Gramatica, P.; Martin, Y.C.; Todeschini, R.; Consonni, V.; Kuz’min,
20
V.E.; Cramer, R.; Benigni, R.; Yang, C.; Rathman, J.; Terfloth, L.; Gasteiger, J.;
21
22 Richard, A.; Tropsha, A. QSAR Modeling: Where Have You Been? Where Are
23 You Going To?. J. Med. Chem. 2014, 57, 4977–5010.
24 21
Dearden, J. C. The History and Development of Quantitative Structure-Activity
25 Relationships (QSARs). Int. J. Quant. Struct.-Prop. Relat. 2016, 1, 1–44.
26 22
27 Dearden, J.C. The History and Development of Quantitative Structure-Activity
28 Relationships (QSARs): Addendum. Int. J. Quant. Struct.-Prop. Relat. 2017, 2,
29 36–46.
30 23
Katritzky, A.R.; Maran, U.; Lobanov, V.S.; and Karelson, M. Structurally Diverse
31
32
QSPR Correlations of Technologically Relevant Physical Properties. J. Chem. Inf.
33 Comput. Sci. 2000, 40, 1–18.
24
34 Le, T.; Epa, V.C.; Burden, F.R.; Winkler, D. A. Quantitative Structure–Property
35 Relationship Modeling of Diverse Materials Properties. Chem. Rev. 2012, 112,
36 2889–2919.
37 25
38 Martin, D.; Maran, U.; Sild, S.; Karelson, M. QSPR Modeling of Solubility of
39 Polyaromatic Hydrocarbons and Fullerene in 1-Octanol and n-Heptane. J. Phys.
40 Chem. B 2007, 111, 9853–9857.
41 26
Martin, D.; Sild, S.; Maran, U.; Karelson, M. QSPR Modeling of the
42
43 Polarizability of Polyaromatic Hydrocarbons and Fullerenes. J. Phys. Chem. C,
44 2008, 112, 4785–4790.
45 27
Worth, A.; Aschberger, K.; Asturiol Bofill, D.; Bessems, J.; Gerloff, K.; Graepel,
46 R.; Joossens, E.; Lamon, L.; Palosaari, T.; Richarz, A. Evaluation of the
47
Availability and Applicability of Computational Approaches in the Safety
48
49 Assessment of Nanomaterials, EUR 28617 EN, Publications Office of the
50 European Union, Luxembourg, 2017, ISBN 978-92-79-68708-2, JRC106386.
51 28
Burello, E. Review of (Q)SAR Models for Regulatory Assessment of
52 Nanomaterials Risks. NanoImpact 2017, 8, 48–58.
53 29
54 Oksel, C.; Ma, C.Y.; Liu, J.J.; Wilkins, T.; Wang, X.Z. Literature Review of
55 (Q)SAR Modelling of Nanomaterial Toxicity. Chapter in Modelling the Toxicity
56
57
58
59 28
60 ACS Paragon Plus Environment
Page 29 of 36 ACS Applied Energy Materials

1
2
3
4
of Nanoparticles. Volume 947 of the Series Advances in Experimental Medicine
5
6 and Biology. 2017, pp. 103–142.
30
7 Gajewicz, A.; B. Rasulev, T.C.; Dinadayalane, P.; Urbaszek, T.; Puzyn, D.;
8 Leszczynska, D.; Leszczynski, J. Advancing Risk Assessment of Engineered
9 Nanomaterials: Application of Computational Approaches. Advanced drug
10
delivery reviews 2012, 64, 1663–93.
11
31
12 Burello, E.; Worth, A.P. QSAR Modeling of Nanomaterials. WIREs Nanomed
13 Nanobiotechnol 2011, 3, 298–306.
14 32
Wang, W.; Sedykh,A.; Sun, H.; Zhao, L.; Russo, D. P.; Zhou, H.; Yan, B.; Zhu,
15
16 H. Predicting Nano-Bio Interactions by Integrating Nanoparticle Libraries and
17 Quantitative Nanostructure Activity Relationship Modelling. ACS Nano 2017, 11,
18 12641−12649.
19 33
Gajewic, A.; Schaeublin, N.; Rasulev, B.; Hussain, S.; Leszczynska, D.; Puzyn,
20
T.; Leszczynski, J. Towards Understanding Mechanisms Governing Cytotoxicity
21
22 of Metal Oxides Nanoparticles: Hints from Nano-QSAR Studies. Nanotoxicology
23 2015, 9, 313–325.
24 34
Yushin, G.; Nikitin, A.; Gogotsi, Y. Carbide-Derived Carbon. In Nanomaterials
25 Handbook (ed. Gogotsi, Y.). CRC Press, Taylor & Francis Group, Boca Raton, Fl,
26
2006; pp. 239–282.
27 35
28 Urbonaite, S.; Juárez-Galán, J.M.; Leis, J.; Rodríguez-Reinoso, F.; Svensson, G.
29 Porosity Development Along the Synthesis of Carbons from Metal Carbides.
30 Microporous Mesoporous Mater. 2008, 113, 14–21.
31 36
32
Gogotsi, Y.; Nikitin, A.; Ye, H.; Zhou, W.; Fischer, J.E.; Yi, B.; Foley, H.C.;
33 Barsoum, M.W. Nanoporous Carbide-Derived Carbon with Tunable Pore Size.
34 Nat. Mater. 2003, 2, 591–594.
35 37
Permann, L.; Lätt, M.; Leis, J.; Arulepp, M. Electrical Double Layer
36 Characteristics of Nanoporous Carbon Derived from Titanium Carbide.
37
38
Electrochim. Acta 2006, 51, 1274–1281.
38
39 Leis, J.; Arulepp, M.; Kuura, A.; Lätt, M.; Lust, E. Electrical Double-Layer
40 Characteristics of Novel Carbide-Derived Carbon Materials. Carbon 2006, 44,
41 2122–2129.
42 39
43 Fernández, A.; Arulepp, M.; Leis, J.; Stoeckli, F.; Centeno, T.A. EDLC
44 Performance of Carbide-Derived Carbons in Aprotic and Acidic Electrolytes.
45 Electrochim. Acta 2008, 53, 7111–7116.
46 40
Lätt, M.; Käärik, M.; Permann, L.; Kuura, H.; Arulepp, M.; Leis, J. A Structural
47
48
Influence on the Electrical Double-Layer Characteristics of Al4C3-Derived
49 Carbon. J. Sol. State Electrochem. 2010, 14, 543–548.
41
50 Brunauer, S.; Emmett, P.H.; Teller, E. Adsorption of Gases in Multimolecular
51 Layers. J. Am. Chem. Soc. 1938, 60, 309–19.
52 42
53 Barrett, E.P.; Joyner, L.S.; Halenda, P.P. The Determination of Pore Volume and
54 Area Distributions in Porous Substances. I. Computations from Nitrogen
55 Isotherms. J. Am. Chem. Soc. 1951, 73, 373–380.
56
57
58
59 29
60 ACS Paragon Plus Environment
 ACS Applied Energy Materials Page 30 of 36

1
2
3
4 43
Katritzky, A.R.; Lobanov, V.S.; Karelson, M. CODESSA Reference Manual
5
6 (version 2.0). Gainesville, FL.
44
7 Draper, N.R.; Smith, H. Applied Regression Analysis, Wiley, New York, 1981.
8 45
Karelson, M. Molecular Descriptors in QSAR/QSPR, WILEY: New York, 2000;
9
10 pp. 141–383.
46
11 Ruusmann, V.; Sild, S.; Maran, U. QSAR DataBank - an approach for the digital
12 organization and archiving of QSAR model information. J. Cheminform. 2014, 6,
13 25.
14 47
15 Ruusmann, V.; Sild, S.; Maran, U. QSAR DataBank repository: open and linked
16 qualitative and quantitative structure–activity relationship models. J. Cheminform.
17 2015, 7, 32.
18 48
QsarDB repository, available at http://qsardb.org/ (accessed 07/08/2018).
19 49
20 Atkinson, A.C. Plots, Transformation, Regression, Oxford, UK: Clarendon Press,
21 1985; p. 282.
22 50
Jaworska, J.; Nikolova-Jeliazkova, N.; Aldenberg, T. QSAR Applicability
23
Domain Estimation by Projection of the Training Set in Descriptor Space: a
24
25 Review, ATLA. 2005, 33, 445–459.
51
26 Tropsha, A.; Gramatica, P.; Gombar, V.K. The Importance of Being Earnest:
27 Validation is the Absolute Essential for Successful Application and Interpretation
28 of QSPR Models, QSAR Comb. Sci. 2003, 22, 69–77.
29 52
30 Netzeva, T.I.; Worth, A.P.;Aldenberg, T.; Benigni, R.; Cronin, M.T.D.;
31 Gramatica, P.; Jaworska, J.S.; Kahn, S.;. Klopman, G.; Marchant, C.A.; Myatt,
32 G.; Nikolova-Jeliazkova, N.; Patlewicz, G.Y.; Perkins, R.; Roberts, D.W.;
33 Schultz, T.W.; Stanton, D.T.; van de Sandt, J.J.M.; Tong, W.; Veith, G.; Yang, C.
34 Current Status of Methods for Defining the Applicability Domain of
35
(Quantitative) Structure-Activity Relationships, ATLA. 2005, 33, 155–173.
36
53
37 Leis, J.; Arulepp, M.; Lätt, M.; Kuura, H. A Method of Making the Porous
38 Carbon Material and Porous Carbon Materials Produced by the Method. US
39 7,803,345, 2010.
40 54
41 Leis, J.; Arulepp, M.; Perkson, A. Method to Modify Pore Characteristics of
42 Porous Carbon and Porous Carbon Materials Produced by the Method.
43 PCT/EP03/04202, 2003.
44 55
Ue, M. Mobility and Ionic Association of Lithium and Quaternary Ammonium
45
Salts in Propylene Carbonate and γ-Butyrolactone. J. Electrochem. Soc. 1994,
46
47 141, 3336–3342.
56
48 Conway, B.E. Electrochemical Supercapacitors. Scientific Fundamentals and
49 Technological Applications. Kluwer Academic / Plenum Publishers, New York,
50 1999.
51 57
52 Lowell, S.; Shields, J.E.; Thomas, M.A.; Thommes, M. Characterization of
53 Porous Solids and Powders: Surface Area, Pore Size and Density. Springer, New
54 York, 2004.
55
56
57
58
59 30
60 ACS Paragon Plus Environment
Page 31 of 36 ACS Applied Energy Materials

1
2
3
4 58
Centeno, T.A.; Stoeckli, F. The Assessment of Surface Areas in Porous Carbons
5
6 by Model-Independent Techniques, the DR Equation and DFT. Carbon 2010, 48,
7 2478–2486.
8 59
Alexander, D.L.J.; Tropsha, A.; Winkler D.A. Beware of R2: Simple,
9 Unambiguous Assessment of the Prediction Accuracy of QSAR and QSPR
10
Models. J. Chem. Inf. Model. 2015, 55, 1316−1322.
11
60
12 Endo, M.; Maeda, T.; Takeda, T.; Kim, Y.J.; Koshiba, K.; Hara, H.; Dresselhaus,
13 M.S. Capacitance and Pore-Size Distribution in Aqueous and Nonaqueous
14 Electrolytes Using Various Activated Carbon Electrodes. J. Electrochem. Soc.
15 2001, 148, A910–A914.
16 61
17 Chmiola, J.; Yushin, G.; Gogotsi, Y.; Portet, C.; Simon, P.; Taberna, P.-L.
18 Anomalous Increase in Carbon Capacitance at Pore Sizes Less than 1 Nanometer.
19 Science 2006, 313, 1760–1763.
20 62
Maran, U.; Katritzky, A.R.; Karelson, M. A Comprehensive QSAR Treatment of
21
22 the Genotoxicity of Heteroaromatic and Aromatic Amines. Quant. Struct.-Act.
23 Relat. 1999, 18, 3–10.
24 63
Mirzaeian, M.; Abbas, Q.; Ogwu, A.; Hall, P.; Goldin, M.; Mirzaeian, M.;
25 Jirandehi, H. F. Electrode and Electrolyte Materials for Electrochemical
26
Capacitors. Int. J. Hydrogen Energy 2017, 42, 25565–25587.
27 64
28 Thommes, M.; Kaneko, K.; Neimark, A.V.; Olivier, J.P.; Rodriguez-Reinoso, F.;
29 Rouquerol, J.; Sing, K.S.W. Physisorption of Gases, with Special Reference to the
30 Evaluation of Surface Area and Pore Size Distribution (IUPAC Technical
31 Report). Pure Appl. Chem. 2015, 87, 1051–1069.
32 65
33 Neimark, A.V.; Lin, Y.; Ravikovitch, P.I.; Thommes, M. Quenched Solid Density
34 Functional Theory and Pore Size Analysis of Micro-Mesoporous Carbons.
35 Carbon 2009, 47, 1617–1628.
36
37
38
39 Graphic Abstract
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59 31
60 ACS Paragon Plus Environment
 90
CPOSACS Applied Energy Materials
Page 32 of 36
CNEG
80
1
CCD capacitance C [F cm-3]

2
3 70
4
5
60
6
7
8
50
9
10
11
40
12
13 ACS Paragon Plus Environment
14 30
15100 Sample ID 1
 SBET Vd<1.14 Del
Page
25 33 of 36
ACS Applied Energy Materials 0.8

Pore size fraction, Vd<1.14 [cm3 g-1]

Surface area, SBET [m2 g-1], x100

Density of electrode, Del [g cm-3]

1 20
0.6
2
3
4 15
5 0.4
6 10
7
8
0.2
9 5
10
11
120 ACS Paragon Plus Environment 0.0
13 20 30 40 50 60 70 80 90
14 Experimental CV,NEG [F cm-3]
 90 4
a) b)
ACS Applied Energy Materials Page 34 of 36
3
80
1
2
2
Calculated C V,NEG [F cm -3]

Standardized residuals
3 70
1
4
5
60
6 0
0.05 0.10 0.15 0.20
7
8 50 -1
9
10 -2
11 40
12
-3
13 ACS Paragon Plus Environment
14 30 h*
15 30 40 50 60 70 80 90 -4
16 Experimental C V,NEG [F cm -3] Leverage
 90
Page 35Training
of 36ACS(C) Applied Energy Materials
Valida�on (D)
80
1
2
Calculated C V,NEG [F cm-3]

70
3
4
5 60
6
7
8 50
9
10
1140
12
13 ACS Paragon Plus Environment
143030 40 50 60 70 80 90
15 Experimental C V,NEG [F cm-3]
 ACS Applied Energy Materials Page 36 of 36

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Experiment-derived
25 CDC & EDLC structure descriptors CNEG = f (structure)
26 Calculated
(C2H5)3CH3N+
27
28 -
29 +
BF4-
30
31 R2=0.94
32
33 Experimental CNEG Experimental

34 0 Surface area
C= SA Pore size fraction
QnSPR
35 Œd Density of electrode

36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 ACS Paragon Plus Environment