00 голосов за00 голосов против

7 просмотров17 стр.Dec 18, 2017

© © All Rights Reserved

PDF, TXT или читайте онлайн в Scribd

© All Rights Reserved

7 просмотров

00 голосов за00 голосов против

© All Rights Reserved

Вы находитесь на странице: 1из 17

journal homepage: www.elsevier.com/locate/micromeso

the way

Martijn F. De Lange a,b,, Thijs J.H. Vlugt b, Jorge Gascon a, Freek Kapteijn a,

a

Catalysis Engineering, Chemical Engineering Department, Delft University of Technology, Julianalaan 136, 2628 BL Delft, The Netherlands

b

Process & Energy Laboratory, Delft University of Technology, Leeghwaterstraat 39, 2628 CB Delft, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: Adsorptive characterization using nitrogen at 77 K is one of the most widely used techniques to assess

Received 3 February 2014 textural properties of porous solids, such as pore volume, specic surface area and pore size distributions.

Received in revised form 11 August 2014 Based on a thorough error analysis the inuence of experimental uncertainties on the accuracy of volu-

Accepted 22 August 2014

metric nitrogen adsorption isotherms and derived properties using the most popular methods is analyzed

Available online 3 September 2014

in detail, comprising the pore volume and specic surface area determined using the method posed by

Brunauer, Emmet and Teller (BET) and the pore size distribution according to the method developed

Keywords:

by Barrett, Joyner and Halenda (BJH). Based on series of isotherms measurements with different sorbents

Surface area

Texture characterization

(MOFs, zeolite, activated carbon and alumina) and on examples from literature (MIL-101), the extensive

BrunauerEmmetTeller error analysis shows that these methods may yield highly inaccurate or even statistically irrelevant (BJH)

Guidelines results. To improve the meaningfulness of derived properties and to minimize statistical uncertainties,

Error analysis practical recommendations and guidelines are proposed for experimental operation variables and data

analysis.

2014 Elsevier Inc. All rights reserved.

errors). Furthermore, the results obtained with these methods

Adsorptive characterization using probe gas molecules is one of are prone to misinterpretation. As it will be demonstrated, not

the most widely used techniques to assess textural properties of adhering to underlying assumptions, denitions and guidelines,

porous solids [1]. The most commonly used adsorbate for this pur- might lead to erroneous results and/or large uncertainties in

pose is nitrogen, recommended by IUPAC for porous materials with obtained values (human errors). The detailed analysis is based

a specic surface area, S, larger than 5 m2 g1 [2,3]. From measur- on ve notably different materials. Model adsorbents have deliber-

ing adsorption of N2 at its normal boiling point, information about ately not been chosen, but instead a selection of widely different

the total pore volume, specic surface area and pore size distribu- porous materials often reported in literature is made.

tion can be derived [4]. This characteristic information is vital for Two materials selected are metalorganic frameworks (MOFs),

application of porous materials in heterogeneous catalysis and porous crystalline materials that have gained increasing interest

adsorptive separation or storage, amongst others. In spite of the in the past decade because of unprecedented topological richness

importance of this technique, in general, little or no attention is and comprising large specic surface areas and pore volumes.

paid to the accuracy and relevance of the obtained quantitative The combination of organic and inorganic building blocks offers

characteristics reported in literature. an almost innite number of combinations, resulting in enormous

In this work, uncertainties in nitrogen adsorption isotherms and variation in pore size, shape, and structure. These materials have

derived textural properties using the most commonly applied pro- found application in adsorptive separation [57], storage [8],

tocols are thoroughly analyzed. An indication of the uncertainty in encapsulation [9] and catalysis [10]. MIL-101(Cr) [11] is among

these quantities is a requirement to draw sound conclusions about the most famous structures, it contains both meso- and microp-

ores, and displays high stability and interesting properties [12

17]. Second MOF is the fully microporous UiO-66 [18], which also

Corresponding authors. Address: Catalysis Engineering, Chemical Engineering

gained signicant attention because of excellent stability and

Department, Delft University of Technology, Julianalaan 136, 2628 BL Delft, The

Netherlands (M.F. De Lange).

interesting properties [1922]. Sigma-1, a microporous zeolite

E-mail addresses: m.f.delange@tudelft.nl (M.F. De Lange), f.kapteijn@tudelft.nl and member of the DDR structural topology, known for its high

(F. Kapteijn). separation performance in membrane and adsorptive processes,

http://dx.doi.org/10.1016/j.micromeso.2014.08.048

1387-1811/ 2014 Elsevier Inc. All rights reserved.

200 M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215

List of symbols

Greek

Latin q density (g ml1)

Acs cross-sectional area (m2) r standard deviation or uncertaintya,b

C dimensionless BET parameter () r2 variancec

D diameter (m)

E adsorption energy (kJ mol1)

Subscript

I intercept (g ml1STP)

ads adsorbed

K Langmuir equilibrium constant (bar1)

BET Brunauer, Emmett and Teller (method)

M N2 molar mass of nitrogen (g mol1)

cell sample cell

N number ()

D.O.F. degree(s) of freedom

NA Avogadros constant (mol1)

K Kelvin

p pressure (bar)

m monolayer

p/po relative pressure ()

man manifold

q adsorbed amount (mlSTP g1)

nbp normal boiling point

R universal gas constant (J mol1 K1)

p Pore

r radius (m)

res residual(s)

R2 coefcient of determination ()

sample sample of (porous) material

res residual (a)

sat at saturation

S specic surface area (m2 g1)

STP standard temperature and pressured

s slope (g ml1

STP)

SS sum of squares (varies)

T temperature (K) Superscript

t thickness of adsorbed layer (m) liq liquid phase

t Student t-distribution () S Studentized

V volume (m3) vap vapour phase

a

Vp specic pore volume (ml g1) Notes: Same units as the property it is related to;

b

Vl liquid molar volume (m3 mol1) Standard deviation if based solely on measured values,

w weight (g) else uncertainty;

c

Squared units of the property it is related to;

d

273.15 K and 1 bar

incorporates aluminium into the framework, making the material investigation the tting strategy applied (number of data points,

also suitable for catalysis [2325]. c-Alumina is chosen as repre- part of the isotherm and tting method) strongly inuences the

sentative for mesoporous metal-oxide supports, which are used value obtained for the BET area. Furthermore, not only the absolute

frequently in heterogeneous catalysis [2628]. From the group of value for BET area but also its uncertainty is investigated as func-

activated carbons, widely used for gas separation and storage tion of tting strategy.

[29,30] and water purication [31], the commercially available The most commonly applied method to determine the pore size

Norit RB2, frequently also used as reference carbon material, is distribution for mesopore containing materials is the one devel-

selected [32]. oped by Barrett, Joyner and Halenda in 1951 (BJH-method) [38].

As the nitrogen adsorption isotherm forms the basis of the tex- This method is based on the Kelvin equation and modied to

ture characterization, rstly the accuracy and reproducibility of the include multilayer adsorption. In this work it is investigated how

adsorption isotherm measurements on these materials is assessed. the uncertainty in a measured isotherm propagates in the pore size

From these isotherms the pore volume is determined, simply distribution, something not published in prior literature.

derived from the amount of N2 adsorbed inside the pores of the

material, and assuming that the density therein is that of liquid

nitrogen, as it is seen most often in literature [4].

The most popular method to determine the specic surface area

of a porous solid, despite profound criticism on underlying

assumptions [33], is the one put forward by Brunauer, Emmett

and Teller in 1938 (BET-method) [34], a multi-layer extension of

Langmuirs monolayer description of adsorption [35]. Although

the underlying assumptions of the BET-method suggest that this

method cannot be used for microporous materials, Rouquerol

et al. demonstrated its applicability, albeit that the physical mean-

ing of the resulting surface area is weaker than for mesoporous

materials [33]. Furthermore, Walton and Snurr have shown that

BET areas determined for microporous MOFs can correspond well

to geometrically accessible surface areas, as calculated by molecu-

lar simulations [36]. The method relies on curve-tting the BET-

equation on a specic part of the adsorption isotherm. The absolute

value of the obtained surface area is, even for a material that

behaves very much BET-like, dependent on which part of the iso- Fig. 1. Reported values of BET surface area as function of reported pore volume for

therm is used [37]. It will be shown that for the materials under MIL-101(Cr), from various literature sources [11,12,4066].

M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215 201

Furthermore, despite the highly appreciated work of Groen et al. and synthesis was carried out at 453 K under autogenous pressure

[39], the current literature is still plagued by erroneous conclusions for 6 days under 200 rpm of rotational speed. Obtained powder is

drawn from BJH pore size distributions. rinsed with water thoroughly and nally rinsed with EtOH to

MIL-101(Cr) is one of the most reported metalorganic frame- remove residual ADA. Calcination of the powder is carried out at

works in scientic literature. Because of the large availability of 923 K for 7 h, with an initial heating rate of 2 K min1.

nitrogen derived material characterization data [11,12,4066] this

structure lends itself well to an analysis of the reported scatter in 2.2. Sorbents commercial samples

pore volume and BET surface area (Fig. 1). As these parameters

refer to the same material, a strong correlation between these Activated carbon (Norit RB 2) and c-alumina (CK-300), were

quantities should be expected. The origin of the scatter will be both purchased from their respective suppliers, Cabot Norit and

illustrated, and based on the proposed guidelines it will be shown Akzo Nobel. A second sample of c-alumina (000-3p, Akzo Nobel)

how standardization can improve the correlation between pore was used to investigate the effect of sample cell volume on mea-

volume and BET surface area for the same material. surements (S-8, S.I), to which is referred as c-alumina(2) if used.

Throughout this work, it is tacitly assumed that during an

adsorption measurement for each measured point adsorption equi- 2.3. Instrumentation measurement procedure

librium is reached or approached closely, so deviations from equi-

librium are not addressed here. With this assumption and error Nitrogen physisorption measurements were performed at 77 K

propagation analysis the uncertainty in volumetric isotherm mea- using a Quantachrome Autosorb-6B unit gas adsorption analyzer,

surements and in the derived properties are estimated. Based on using an equilibration time of 2 min, meaning that a data point is

these ndings, guidelines are proposed to improve the experimen- considered in equilibrium when the pressure varies less than

tal operation and to assist in the determination of the adsorption 80 Pa in 2 min. All samples were pretreated ex-situ for at least

isotherm, pore volume and BET area, in order to decrease their 16 h under vacuum at 473 K, except for Sigma-1 (573 K instead),

uncertainty and hopefully also the variation in absolute values before adsorption measurements. Sample amounts used and sam-

reported for the same material in literature. ple cell volumes determined are given in Table S 1.

2. Materials and methods

One frequently derived property from a nitrogen adsorption

2.1. Sorbents material synthesis isotherm is the total pore volume, Vp. Frequently it is assumed that

at saturation the adsorbed nitrogen has the same density as in the

All chemicals were obtained from SigmaAldrich and were used liquid phase at the same temperature [4], which makes that the

without further purication. MIL-101(Cr) was synthesized as pre- pore volume (Vp) can be calculated by making use of the Gurvich

viously reported in literature [11]. 1.63 g of chromium nitrate principle [68]:

[Cr(NO3)39H2O, 97%], 0.70 g of terephthalic acid [C6H4-1,4-

(CO2H2)2, 97%], 0.20 g of hydrouoric acid (HF, 40%) and 20 g of qvap:

STP

V p qsat 1

distilled water were added to a Teon container that was inserted qliq:

nbp

in a stainless steel autoclave. The autoclave was heated for 8 h at

493 K in an oven under static conditions. After synthesis, the solid Herein, q is the density for the vapour phase (vap.) at standard tem-

product was ltered from the synthesis solution. The as-synthe- perature and pressure (STP) and for the liquid phase (liq.) at normal

sized material was activated solvothermally using ethanol (EtOH, boiling point (nbp), and qsat is the loading at saturation expressed in

95%) at 353 K for 24 h. The resulting solid was exchanged in a mlSTP g1. Note that the assumption of the adsorbed phase having

1 M solution of ammonium uoride (NH4F) at 343 K for 24 h and liquid density is strictly never exactly true but more questionable

was immediately ltered off and washed with hot water. MIL- for microporous materials [4]. Hence the obtained values for such

101(Cr) was nally dried overnight at 433 K in air and stored under materials should be interpreted with care. As it will be shown later,

air atmosphere. the choice of the exact loading of saturation is not always trivial.

Synthesis of UiO-66 was carried out according to literature as Generally it is recommended to calculate the pore volume at the

well [18]. 0.053 g of zirconium chloride [ZrCl4, 99%], 0.034 g tere- plateau in adsorption, for type I (microporous) and type IV (meso-

phthalic acid [C6H4-1,4-(CO2H2)2, 97%] and 24.9 g N,N0 -dimethyl- porous) materials (IUPAC classication [2,3]) to avoid including

formamide (DMF) were added to a Teon container, inserted in a inter-particle nitrogen condensation [4]. This means in practice that

stainless steel autoclave. The autoclave was heated at 393 K for relative pressure should not be close to unity, (p/po) 6 0.9 is used

24 h under static conditions. After cooling in air to room tempera- often in literature (po is the saturated vapour pressure of nitrogen

ture, the resulting solid was ltered, washed with DMF and dried at @ 77 K).

room temperature.

Synthesis of Sigma-1 was performed adapting the method pat- 2.5. Adsorption derived properties BET surface area

ented by Stewart [67] with minor modications, keeping the syn-

thesis composition as adamantylamine (ADA):N a2O:SiO2:Al2O3 Nitrogen physisorption is a key technique to obtain the specic

:H2O = 20:3:60:1:2400 (Ex. 1 of patent). 5.39 g of Ludox HS-40 surface area of a material. Most commonly, the theory developed

was added to vial A and was diluted with 20 g of deionized water. by Brunauer, Emmet and Teller (BET) is used [34] for the isotherm

After 5 min of stirring, 3.62 g of ADA was added and the obtained data interpretation, though occasionally surface areas are calcu-

solution further diluted using 19.4 g of deionized water. Finally, lated using the Langmuir isotherm [35]. Since the rst publication

contents of vial A were stirred for over 15 min, until a homogenous of the BET equation, it has been the major tool to assess specic

solution was obtained. In vial B, rstly 0.2 g of NaAlO2 was dis- surface area, despite criticism towards its derivation and underly-

solved in 5.41 g of water. After obtaining a clear solution, 0.19 g ing assumptions [1], of which the more important ones are briey

of NaOH and 4 g of deionized water was added. After stirring vial addressed. Firstly, the adsorbent is assumed to have a homoge-

B for 15 min, vial B is mixed over vial A and aging was continued neous surface, onto which molecules adsorb in multiple layers.

for 30 min. Obtained solution was transferred to Teon containers This is notably different to the method developed by Langmuir,

202 M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215

which is limited to a single layer of adsorbate molecules on a sur- The extent to which the BET parameters vary as function of the

face [35]. Secondly, to the second layer and onwards, molecules pressure range and the degrees of freedom used for tting is calcu-

can be adsorbed before complete lling of the lower layers. There lated for the materials under investigation, including the variation

is an innite amount of layers when (p/po) reaches unity. Thirdly, of uncertainty in the BET parameter values.

there are no lateral interactions between molecules located in

the same layer, making the molar adsorption energy within one 2.6. Adsorption derived properties BJH pore size distribution

layer constant. For the second and further layers the molar adsorp-

tion energy is assumed to be equal (E2), and differs from that for The pore size distribution is calculated based on the method

the rst layer (E1). These assumptions give rise to the well-known developed by Barrett, Joyner and Halenda (BJH) [38]. Herein it is

BET relation, that can be formulated as [34]: assumed that the total amount adsorbed in a pore of a material

0 1 is based upon two separate consequent contributions. Firstly, the

C ppo pores contain a surface on which layers of adsorbate molecules

q qm @n o n oA 2

1 pp C pp 1 pp can be formed, consistent of a certain thickness. The thickness of

o o o

this layer on the pore surface increases with increasing (p/po). Sec-

ondly, there is an inner capillary radius in this pore of which the

Here qm is the BET monolayer capacity and C the dimensionless BET

volume is lled by condensation of the adsorbate and no longer

parameter, calculated as the ratio between the adsorption constants

by prolonged layer formation. For a given adsorbate species, the

of the rst and second and further layers, often approximated by

relative pressure at which this volume condensation occurs, is

[37]:

determined by the size of this capillary radius, and can be calcu-

E1 E2 lated with the Kelvin equation (rK). The thickness of adsorbate mol-

C exp 3 ecules attached to a pore surface (t) was originally estimated for

RT

different relative pressures based on experiments by Shull [75]

Note that negative values of C are physically meaningless. The but can currently be calculated with a variety of equations, includ-

specic surface area can be calculated after the monolayer capac- ing those of De Boer [76] and HarkinsJura [77,78]. Here the latter

ity, qm, has been determined, via [34]: is applied, as it is used in the accompanying software of the

adsorption equipment. For a given relative pressure thus, the vol-

qm qvap:

STP N A ACS

SBET 4 ume of adsorbate present inside a porous material is the sum of

M N2

(i) the amount of adsorbate present in all pores that are already

Here qvap fully lled via condensation, for which the pore radius must be

STP is the density of nitrogen vapour at standard temperature

and pressure (STP), NA is Avogadros constant, M N2 is nitrogens smaller than or equal to the Kelvin radius for the given relative

molar mass and ACS is the cross-sectional area of a nitrogen mole- pressure (ri 6 rK (p/po)), and (ii) the amount of adsorbate that is

cule. The current standard value of the latter is 0.162 nm2 [69], present in the layers of certain thickness on the walls of the pores

derived from the density of liquid nitrogen assuming a hexagonally for which the radius larger than the Kelvin radius (ri > rK (p/po)).

closed packed system. Historically this value has varied between Summing up over all pore sizes present in the material, this can

0.13 and 0.20 nm2 [70]. One could directly obtain both C and qm from be written as:

nonlinear tting the BET equation to adsorption data directly. How- vap: X k

p qSTP p

ever, it is common practice to obtain these parameters from a line- q D V r 6 r

po k qliq:

p;i i K

arized form of the BET equation: i1

p o k

nbp

Xn

p p

1 C1 p p DSi t i r i > r K 7

po Is 5 po k

p

q 1 po Cq m Cq m po po

ik1

The left-hand side of Eq. (5) is plotted versus relative pressure. tive pressure, DVp,i are the incremental pore volumes that are

The intercept (I) and slope (s) can be obtained using a simple linear already completely lled, associated with radii ri, DSi are the

least squares tting routine. From these two parameters, the BET incremental pore surface areas that belong to pore radii, ri, that

C-parameter and monolayer capacity, qm, can be back-calculated are not yet completely lled and only contain layers of adsorbate

via: molecules, ti the layer thicknesses thereof and rK the Kelvin radius

for a given relative pressure. Furthermore, the kth pore size is the

1 Is

qm ; C 6 largest pore lled completely via condensation (for given (p/po))

Is I

and the nth pore size is the largest present. As q is given in vol-

As mentioned above, intercept I should have a positive value. ume of N2 vapour at standard temperature and pressure (STP), this

Furthermore, the linearization does not hold for the entire pressure requires conversion to liquid phase at measurement conditions, as

range. Originally, based on their own experimental results, has been done for the calculation of the total pore volume (Eq.

Brunauer et al. indicated that Eq. (5) should only be applied for (1)). To be able to apply Eq. (7) to determine the pore size distri-

0.05 < (p/po) < 0.35, as outside these boundaries the left-hand side bution, calculations should be started for a measured point at sat-

of Eq. (5) was found to strongly deviate from linearity [34]. No uration (adsorption plateau in type IV isotherms) [2,3] and an

physical phenomena were mentioned as reason for these limita- adjacent data point at lower relative pressure. It is thus tacitly

tions. Later, IUPAC recommended the use of a slightly narrower assumed (see Eq. (7)) that the difference in loading between these

pressure window, 0.05 < (p/po) < 0.30 [2,3]. The linear tting two points is only caused by depletion of the completely lled

method may be preferred over directly tting C and qm because largest pore. Subsequently, from the difference in loading, one

of visual tractability and simplicity of tting, not because of pro- can determine the incremental pore volume (DVp) for the largest

found physical insights or statistical benets. Regarding the latter, pore in the adsorbent. The radius of this pore follows directly from

the error distribution is changed by linearization [71], similarly as the relative pressure, as it is the sum the thickness (t) and Kelvin

in the determination of (bio)catalytic reaction kinetic parameters radius (rK). The difference in loading between the second and third

(HougenWatson, LineweaverBurk approach) [7274]. point is not only assumed to originate from a smaller pore but

M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215 203

also from the surface of the larger one, of which in the previous nitrogen adsorption isotherms were determined in threefold by

step the size was determined. Thus starting from saturation, the repeating the measurement with the same sample in the same

distribution of pore sizes can be recursively calculated following sample holder and identical pre-treatment protocol, and subjected

the desorption branch, for which it was derived. It is also feasible to a detailed error propagation analysis. Results are shown in Fig. 2.

to apply this approach to the adsorption branch, although this is Uncertainties in relative pressure (x-axis) are insignicant, except

not advised based on the underlying assumptions of the model. for the lowest relative pressures, and therefore not depicted.

A frequently observed phenomenon for many desorption hystere- Clearly for each material under investigation, the three isotherms

sis branches is that they are not extended below a certain critical and their condence intervals are very similar, showing very good

(p/po) [4]. This lower limit is only dependent on temperature and reproducibility of the measurement procedure. This reproducibil-

used adsorbate, and thus independent of the material under ity is also shown by the fact that for all investigated materials none

investigation. For nitrogen adsorption at 77 K, this limit is at of the isotherms is outside of the condence interval of the other

(p/po) = 0.42 [4]. In general one should not make use of the two. A closer investigation of this condence interval clearly shows

isotherm below (p/po) = 0.42, when determining a BJH-pore size the cumulative nature of the propagation of uncertainties, see Eq.

distribution [4]. This in turn means that the BJH-pore size distri- S-2.14. For each additional measured point, the interval widens

bution is limited to Dp P 3.4 nm and thus should strictly not be slightly. At low relative pressures, the condence intervals are

applied to the microporous region. In this work, the BJH pore size insignicant. At relative pressures above 0.3, the growing con-

distribution was calculated as described in this section and subse- dence intervals become clearly visible and are the largest for the

quently the uncertainty in the pore size distribution was analyzed, last point measured during desorption. The calculated absolute

details of which can be found in the Supporting information (S-3). condence interval is generally below 10 mlSTP g1 for adsorption

and below 20 mlSTP g1 for desorption (see Fig. S 1). A detailed

2.7. Uncertainty analysis analysis of the different contributions to the overall uncertainty

shows that an increase in accuracy of the adsorbed amount can

To assess the uncertainty in measured nitrogen adsorption iso- be realized by increasing the accuracy of the pressure sensor used,

therms, the theory of propagation of uncertainties is applied (see, a more accurate calibration of the manifold volume (see S-6) or by

e.g. Taylor [79]). For independent random errors, the variance optimizing the ratio of manifold volume and sample volume. To

can be formulated as: determine this optimal ratio, the uncertainty in pore volume is

X @y 2 determined as function of (Vman/Vcell), using a representative Lang-

r2y r2xi 8 muir-Type isotherm (qm = 500 mlSTP g1, K = 10 bar1). Results, as

@xi

depicted in Fig. 3, show that preferably (Vman/Vcell) is between 2

Here y is a variable calculated from i measured variables xi, and ry is and 3 (see S-7 for calculation details). As pressure sensors often

the uncertainty in this variable y, clearly a function of the uncertain- have an accuracy, which is a percentage of the full range, an

ties in xi, rxi. Applying Eq. (8) consecutively on all calculated vari- increase in adsorption measurement accuracy might be best real-

ables, will ultimately lead to the variance in the adsorbed amount ized by using multiple pressure sensors with different pressure

as a function of relative pressure (calculation details are given in ranges, although it might be rather difcult to retrot this in

S-2), from which the absolute uncertainty (square root of variance) already existing adsorption equipment.

and consequently the absolute condence interval can be calcu- In the preceding discussion, sample mass and cell volume have

lated. All condence intervals in this work are calculated for a been xed purposely to investigate reproducibility of the measure-

95% condence level. ment procedure. Separate efforts have been conducted to envisage

Assuming no uncertainties in the determination of density, the the inuence of these two variables experimentally (S-8). Firstly,

variance in pore volume can be related directly to the variance in decreasing sample mass and increasing cell volume both cause a

the measured isotherm (adsorbed amount) via: widening in the condence interval (see Fig. S 10). More interest-

!2

qvap: ingly, an articially increased desorption hysteresis can be

r2V p r2qsat STP

9 observed when the cell volume is increased, becoming more

qliq:

nbp noticeable for lower sample masses (see Fig. S 11). For the smallest

The variance in the adsorbed amount of nitrogen is also cell volume (Vcell 10 ml) under investigation, the closure of the

required to determine the uncertainty in the BJH-pore size distri- hysteresis loop at (p/po) 0.42 can be observed, where for the larg-

bution (details given in S-3). The uncertainty in BET surface area est cell (Vcell 35 ml) the hysteresis loop is not yet closed at

is directly determined from the tting procedure (See S-4). The (p/po) 0.2, suggesting unphysical desorption behaviour. This is

number of degrees of freedom used (i.e. the difference between because both a larger cell volume and lower sample mass make

the number of data points and the number of model parameters that pressure in the cell volume varies less during measurements

to be estimated) is important for the data tting. Obviously, in of the different data points. The pressure of the cell volume is mon-

order to have a meaningful tting, one should, at least, have one itored to determine whether a measured point has approached

degree of freedom (ND.O.F.). As the BET-equation contains two equilibrium sufciently. So, when variation in pressure during

parameters, either C and qm or I and S, depending on the applied measurements decreases, sensitivity towards equilibrium is lost

approach, at least three data points are needed for a t. 95% Con- and, especially for measured points with signicant ad- or desorp-

dence intervals in measured temperatures (0.1 K), pressures tion, a deviating trend can be observed (see S-8).

(0.1% of measurement range), weighted amounts (0.1 mg) and Importantly, the absolute volume of the sample cell thus deter-

manifold volume (5%), as reported by respective suppliers, are mines whether an erroneous hysteresis between ad- and desorp-

used in the error analysis. tion occurs. Clearly, a small sample cell (Vcell 10 ml) is desired.

However, minimization of this cell is not unconstrained. Obviously,

3. Results and discussion the cell should have a nite inner diameter to be able to load the

actual sample. Furthermore, and less trivial, the cell length cannot

3.1. Uncertainty in adsorption measurements be shortened, because of liquid nitrogen level control. Keeping the

liquid nitrogen level constant is crucial, as minor deviations of this

To properly assess the uncertainties in variables derived from level have a signicant inuence on adsorption measurements, as

physisorption measurements, e.g. pore volume and surface area, has been shown by Pendleton and Badalyan [80] As nitrogen is

204 M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215

Fig. 2. Repeated isotherm measurements and condence intervals calculated using error propagation for MIL-101(Cr) (a), UiO-66 (b), Norit RB 2 (c), c-alumina (d) and Sigma-

1 (e). Key: rst (j), second ( ) and third ( ) measurement (same sample and holder) depicted with closed symbols, condence intervals given with lines and open symbols.

continuously evaporating, the liquid level is naturally decreasing 3.2. Uncertainty in pore volume

over time. To counter this, the liquid nitrogen vessel is moved

upwards with respect to the sample cell, making that the cell Using the measured adsorption isotherms and error analysis,

should be sufciently long. Alternatively, a sleeve around the one can directly determine the pore volume and its uncertainty.

sample cell and capillary suction may keep the effective liquid Results are given in Table 1, for the third measurement of each

nitrogen level around the cell constant. To obtain a smaller cell vol- material. The relative condence interval in pore volume can be

ume, it is recommended to insert a (glass) ller rod in the sample as large as 10%, as is the case for Sigma-1. Both the absolute value

cell after loading the sample, as was used in this work. of the pore volume and its uncertainty are very similar for the

M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215 205

adsorption to avoid arriving at a too low value for the pore volume.

For example for MIL-101(Cr) the latter would be feasible. However,

as shown above, two other parameters can be used to decrease the

uncertainty in the measured isotherm and thus in the derived pore

volume: the mass of the material under investigation and the vol-

ume of the sample cell used relative to that of the manifold. Vary-

ing both parameters for e.g. a Langmuir isotherm with qm = 500

mlSTP g1 and K = 10 bar1, has shown that (Vman/Vcell) should be

between two and three. A much larger or smaller cell volume will

have a detrimental inuence on the accuracy of the measurement

and therefore on the pore volume (See S-7). Furthermore, a sample

mass below 0.05 g leads to a prohibitively high uncertainty. With

increasing sample mass, the uncertainty is lowered, although this

may result in much longer measurement times, and diffusion

issues of heat and mass.

Fig. 3. Simulated 95% condence interval for the calculated pore volume at A Langmuir isotherm was chosen to model adsorption behav-

(p/po) = 0.9 as function of (Vman/Vcell) based on a Langmuir isotherm iour, because of the inability of the BET equation to describe satu-

(qm = 500 mlSTP g1, K = 10 bar1) and a sample mass of 0.2 g. For calculation details, ration and thus a well-dened pore volume. Experimentally, these

see S-7 (S.I). conclusions are conrmed. Increasing cell volume ((Vman/Vcell) < 2),

and decreasing sample mass both show an increase in both con-

dence interval and variation in obtained pore volumes from differ-

Table 1 ent measurements (See S-8).

Estimated pore volume at (p/po) = 0.9 and its 95% condence interval for the third Variation of both K and qm, representing the difference in

isotherm measurement of each material. adsorptive properties of various materials, shows that the relative

Material Vp/cm3 g1 95% Conf. int./cm3 g1 uncertainty in pore volume is especially high when both parame-

ters are small, as in case of poorly adsorbing materials (See S-7).

MIL-101(Cr) 1.51 0.017

UiO-66 0.43 0.016 Of course the concept of pore volume is ill-dened for poorly or

Sigma-1 0.14 0.014 non-adsorbing materials as these materials show little porosity.

c-Alumina 0.40 0.011 Nevertheless, the uncertainty in pore volume is directly propor-

NORIT RB2 0.46 0.010 tional to the uncertainty in adsorbed amounts (at (p/po) = 0.9),

see Eq. (9), and therefore the uncertainty in the measured

adsorption isotherm is especially large for small values for K and

qm. With increasing K and qm, this uncertainty is decreased (See

three measurements of the same sample, showing good reproduc- S-7). Because the uncertainty is a function of the total amount

ibility (see S-15). Furthermore, values for the relative 95% con- adsorbed (q.wsample) the sample mass should be adjusted in rela-

dence obtained here are very similar to those obtained by tion to these parameters. As an indication, the minimal sample

Pendleton and Badalyan for micropore volumes for a different set mass required to obtain a pore volume with a relative condence

of materials, using either the as-plot method or the theory of vol- interval less than 5% has been simulated as function of the

ume lling, methods not discussed in this work [81,82]. This indi- Langmuir parameters (Fig. 4). For poorly adsorbing materials,

cates that the uncertainty in pore volume is not strongly represented by very low qm and K values, this is more than 0.5 g.

dependent on the method used to determine it. For reasonably adsorbing materials, represented by K > 10 bar1

Note that, since the adsorption equipment used does not pro- and 100 < qm < 200 mlSTP g1, the required amount is 0.20.3 g.

vide information about the exact dosing procedure during mea- For more strongly adsorbing materials, this can even become less

surements, the results mentioned in preceding paragraph were than 0.1 g. As it is not always possible to estimate a priori the

obtained under the simplifying assumption that each measured amount adsorbed, the relative condence interval in Vp as function

data point requires only single dosage of nitrogen (S-2). Releasing of the total amount adsorbed (q.satwsample, see Fig. S 8) has been cal-

this assumption, the uncertainty can become more than two times culated. This is also useful as tool to estimate uncertainties of pore

larger, as it is the case for MIL-101(Cr) (S-9), depending on how the volumes reported in literature, provided the sample mass is

number of doses required per measured point is estimated. The reported as well, as is recommended by IUPAC [2,3].

increased uncertainty is caused by the additional doses needed

to encompass a given adsorbed amount and is a function of the 3.3. Uncertainty and variation in BET surface area

total amount adsorbed by a material and of the shape of the nitro-

gen isotherm (S-9). According to our analysis, this increase in The variation in BET parameter values, qm and C, and uncer-

uncertainty is higher for microporous materials, because a signi- tainty therein has been determined as a function of the used

cant fraction of total nitrogen loading is already adsorbed at the degrees of freedom and the applied pressure window for the ve

rst measured point, i.e. at low pressure. This means that this point materials under investigation. Detailed results can be found in

requires a signicant number of doses, generating a large uncer- Figs. S 17S 21. For comparison, the average 95% condence inter-

tainty therein. Obviously, this effect is notably less important in val is calculated for each number of degrees of freedom (N.D.O.F.)

mesoporous materials (containing hardly any micropores), as for under consideration (Fig. 5). It is apparent that one degree of free-

these materials the number of doses at low pressures is lower dom yields an unsatisfactory condence interval. Condence inter-

(S-9). vals become acceptable from three degrees of freedom onwards,

To decrease the uncertainty in the pore volume, one could, for which is in line with the preference put forward by IUPAC to at

example, consider decreasing the number of measured points. least use ve data points (three degrees of freedom) [2,3]. The

However this would reduce the information obtained from the iso- uncertainty is especially high for MIL-101(Cr), attributed to the

therm. One could calculate the pore volume at a lower relative peculiar shape of its isotherm. The different inections caused by

pressure, but this is only possible if there is a distinct plateau in lling the medium and large cavities, make the isotherm poorly

206 M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215

Fig. 4. The required sample mass, depicted both on z-axis and in colour-scale, to obtain a pore volume, calculated at (p/po) = 0.9, with a relative condence interval of 5%, as

function of the Langmuir monolayer capacity, qm, and equilibrium constant, K, when (Vman/Vcell) = 2. A two-dimensional projection is depicted in the xy plane.

represented by the BET equation. The uncertainty is especially high dened as the ratio of minimum and maximum BET surface area

around these kinks for various degrees of freedom used in the t- determined from ts varying the window of adjacent data points

ting (Fig. S 17). Increasing the number of degrees of freedom will of a selected degree of freedom over the relative pressure range

lower the uncertainty found, as the relative inuence of these limited by the upper bound recommended by IUPAC (0.05 6

inections will be suppressed. In case of UiO-66, NORIT RB2 and (p/po) 6 0.3) [2,3]. Clearly, from Fig. 6 can be seen that the variabil-

Sigma-1 this is notably different. For these materials, the average ity can exceed a factor of four difference for low degrees of freedom

uncertainty in the BET area is rather similar. This is in accordance (for MIL-101(Cr). Since it is not common practice to report the

with adsorption isotherms of these materials, which show similar- relative pressure range applied, in spite of the IUPAC recommenda-

ity in adsorption curvature for (p/po) < 0.30. In all cases, a mini- tions [2,3], nor the degrees of freedom used to determine the BET

mum in average uncertainty exists around seven degrees of surface area, a comparison of literature results is not straightfor-

freedom. This is attributed to the opposite effect of (i) the increase ward. Van Erp and Martens [37] proposed the direct tting of the

of degrees of freedom that will decrease the uncertainty in the t- BET equation instead of linearizing the adsorption isotherm. The

ted parameters (roughly proportionally with N1 D.O.F. (see Eqs. S-4.4 BET monolayer capacity, qm, and dimensionless parameter, C, were

S-4.8 (S.I)) and (ii) the sum of squared residuals, SSRES, which obtained directly by non-linear estimation (direct method).

increases with each degree of freedom added. Moreover, they also applied weights to each data point (weighted

As clearly visible in Fig. S 22, the linearized BET equation plots direct method), based on the relative vicinity of each point to near-

are not fully linear. For purely microporous UiO-66, Norit RB2 and est neighbours, effectively penalizing data points that are close to

Sigma-1 the restriction of non-negativity of the BET constant C is each other. This, according to the authors, to better balance the

violated for (p/po) > 0.040.07, (Fig. S 23). This fact limits the quan- relative importance of monolayer and multilayer adsorption in

tity of ts available for certain number of degrees of freedom. From the determination of the BET parameters. Indeed, when the

Fig. 5, Figs. S 18, S 19 and S 21, it becomes apparent that the uncer- unweighted direct method is applied to the materials under inves-

tainty in the BET surface area for ts adhering to C > 0 is smaller tigation, the variability in the absolute value of the BET surface

than for those where C < 0. area is lower (see Fig. S 26) than for the linear method. This is

Finally, for c-alumina, the average uncertainty decreases with due to the fact that linearization puts signicantly more emphasis

the number of degrees of freedom and has the lowest average on measured points at higher relative pressures (see S-16) [37].

absolute uncertainty. This is attributed to the shape of the nitrogen The weighted direct method performs similarly as the unweighted

adsorption isotherm which is, of these ve materials, most in direct method, except for low degrees of freedom, where the per-

accordance with a BET-type isotherm (see Fig. S 24), yielding the formance of the weighted direct method is signicantly worse

most linear BET plot (see Fig. S 22). (see S-17 for more detailed explanation). Using the nonlinear

Table 2 summarizes the BET surface areas found when using all parameter estimation method, the uncertainty in surface area is

degrees of freedom available in the dataset (for (p/po) 6 0.3) for obtained from the t directly. Hence the unweighted nonlinear

each of the materials, as well as the maximum degrees of freedom parameter estimation is preferred (e.g., Fig. S 28). Note, however,

useable when also implementing the constraint of C > 0. Relative that when the t is applied to a region in the isotherm which is

condence intervals are below 10%, except for Sigma-1 for which near or at saturation it becomes insensitive to the BET C parameter,

it is around 20%. increasing strongly the uncertainty in the surface area (see Figs. S

The BET area is often used as quality indicator: comparison of 18, S 19 and S 21 for (p/po) > 0.04-0.07). Using the linear method

samples of a given material, e.g. reported in literature, is expected one does not encounter problems with sensitivity of the parame-

to be conclusive. However, the variability in obtained BET surface ters, but, this method suffers from statistical criticism (error

areas for a given sample and chosen number of degrees of freedom distribution is changed), and might yield negative values for C. This

might severely skew this comparison. Variability, in this work is calls for recommendations on the relative pressure range to be

M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215 207

Fig. 5. The average 95% condence interval determined from ts varying the window of adjacent data points for the selected degree of freedom over the relative pressure

range limited by the upper bound recommended by IUPAC (0 6 (p/po) 6 0.3) [2,3], is depicted as function of the used degrees of freedom for MIL-101(Cr) (a), UiO-66 (b), Norit

RB 2 (c), c-alumina (d) and Sigma-1 (e). Closed symbols are based on full dataset, open symbols, are obtained when the BET C parameter is constrained to positive values, only

if the constraint excludes part(s) of the data set. Results obtained with the linear tting method.

used to obtain BET surface areas not prone to a large the direct method becomes invariant for the BET C parameter at

variability, keeping the uncertainty as low as possible and avoiding higher relative pressures, increasing signicantly the uncertainty

physically unacceptable results. In order to formulate such in the BET surface area. In both cases thus, approaching saturation

recommendations, a distinction between micro- and mesoporous results in tting problems. Therefore measured points that are

materials is made. close enough to saturation to contribute to a convex linearized

In case of microporous materials, when using the linear method, BET curve should be excluded, as saturation is not properly

for (p/po) > 0.040.07 the BET C parameter becomes negative while described by the BET method. A simple mathematical routine using

208 M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215

Table 2

Estimated BET surface area and both absolute and relative condence interval, for the maximum degrees of freedom (for 0 6 (p/po) 6 0.3) and for the additional restriction of non-

negativity on the BET C parameter, for the third isotherm measurement of each material.

2

SBET/m g 1

95% Conf. Int. SBET/m2 g1 95% Conf. Int. ND.O.F./

2

Material m g 1

% m2 g1 %

MIL-101(Cr) 2810 88 3.1 29

UiO-66 860 60 7.0 1070 4.9 0.5 7

Sigma-1 270 53 20 321 2.3 0.7 7

c-Alumina 183 8.2 4.5 29

Norit RB2 930 57 6.1 1080 10 1.0 13

a

For UiO-66, Sigma-1 and Norit RB2 in many ts, including those with the maximum degrees of freedom present in the isotherm data used for BET analysis, the C

parameter obtained is negative, yielding unphysical results. Therefore, in this part of the table for these materials the results are given obtained for the highest degrees of

freedom, ND.O.F., that still yielded a positive C. If multiple ts were available with the same ND.O.F., the one with lowest uncertainty in the BET surface area was selected.

zero degrees of freedom is proposed to calculate the relative pres- that Langmuir parameters from different adsorption measure-

sure for which C becomes negative, so where the linearized form of ments determined using the nonlinear (direct) tting approach

the BET relation breaks down. Thus, for two data points and using yields better parameters and lower uncertainties than using differ-

the linear BET equation, the C parameter can be calculated: ent linearization schemes [71]. Furthermore they concluded that

the direct method is much more robust, as it is less inuenced

1 C1

yi xi 10 by the in- or exclusion of a specic measured point, which can also

Cqm Cqm

|{z

} |{z

} be seen from the results presented here (see Fig. S 30). However,

Ii si when nonlinear parameter estimation procedures would not be

Using two adjacent data points, one can extract intercept, I, and at hand, the linear method but supplemented with weights, xl1,

slope, s, via: (See S-16) as shown by Van Erp and Martens [37], could be used.

Note that these linearization weights, xl, are different from those

yi1 yi applied in the weighted direct method.

si1 ; Ii1 yi xi si1 11

xi1 xi For mesoporous materials, the original restriction posed by

Subsequently the BET C parameter follows from: Brunauer, Emmet and Teller [34] is a starting point for reasoning.

Based on a large set of experimental adsorption measurements,

I i si they suggested to use 0.05 < (p/po) < 0.35. Later, IUPAC decreased

Ci 12

Ii the upper limit to 0.30 [2,3]. The upper limit can be explained by

Eqs. (1012) can be applied to each pair of data points of an iso- the linearized BET curve becoming more convex, as was found

therm recursively. The calculated C values as function of relative for microporous materials, though at signicantly higher relative

pressure are depicted in Fig. S 23 and compared with those values pressures. It is proposed here to replace this upper limit by one

obtained from the previously discussed linear tting procedure. determined by the same mathematical procedure that was suc-

Strikingly, for all the materials under investigation the two-point cessfully applied for microporous materials using pairs of data

procedure yields exactly the same relative pressure at which C points. The lower limit, as mentioned by the original authors

becomes negative as was obtained by the linear tting method, [34], was put in place because of the observation that below it

without the laborious efforts of tting different parts of the iso- the transformed BET data showed for many materials strong devi-

therm with different degrees of freedom. By applying this routine, ations from linearity. This is most probably because of the surface

not only a fundamental physical reason is given to reject data heterogeneity, violating the assumption of a homogenous surface

points, but also a fast mathematical calculation procedure is pro- [34]. Indeed, Salvador et al. showed that the BET C parameter varies

vided to lter out these data points effectively. The rst recom- heavily as function of loading at low surface coverages, yielding

mendation, especially for microporous materials, is to apply this extremely high C values for (q/qm) < 0.50.7, for different mesopore

lter to the isotherm of the material under investigation. This lter containing materials [84]. These high values would erroneously

will lead to a very similar relative pressure window as proposed by indicate highly microporous materials (see Fig. S 24). Also, as C

Rouquerol et al. [33], based on the work of Keii et al. [83]. Here it and qm are negatively correlated, one would underestimate the

was suggested to plot q(pop) versus (p/po) and to limit the relative surface area as well (Van Erp and Martens [37]). In line with these

pressures to an interval where q(pop) shows a continuous previous reports, results indicate the lower surface area and higher

increase. By following this approach BET surface areas were shown C values at (p/po) < 0.030.07, for both MIL-101(Cr) and c-alumina

to be in good agreement with those obtained from molecular sim- (see Figs. S 17, S20 and S 23). This indeed corresponds roughly to

ulation [36]. Added benet of the approach opted here is that C is relative loadings (q/qm) < 0.50.7. Despite the physical sense of

intrinsically constrained to positive values. Also the current constraining the pressure range to a minimum for (q/qm), practi-

method can be more easily implemented in automated software cally this is not easily implemented because it is not trivial to

for routine determination of BET surface areas, as no derivative determine a priori an exact ratio (q/qm) and it depends on the t-

needs to be calculated, as required for the method put forward ting result. Another option might be to eliminate exorbitant high

by Rouquerol et al. [33]. values of C, which is again not feasible a priori. Though Eq. (12)

Furthermore, it is advised to use the maximum degrees of free- is helpful in determining the relative pressure for which there is

dom available within the window to ensure minimal variability a transition from positive to negative C, it does not work to deter-

and uncertainty in the BET area (see Fig. S 30). Since the direct t- mine a low relative pressure limit (see Fig. S 23), because of the

ting method results in a smaller variation in BET area and con- high values of C and strong uctuations therein at low relative

dence interval for data inside the recommended pressure pressures. It is thus proposed to use the direct tting method for

window, it is recommended to use the direct tting procedure. This the data up to the upper relative pressure limit determined above

conclusion is in line with recent work of Osmari et al., who show and to investigate the residuals as function of relative pressure; in

M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215 209

Fig. 6. The ratio of minimum and maximum BET surface area determined from ts varying the window of adjacent data points of the selected degree of freedom over the

relative pressure range limited by the upper bound recommended by IUPAC (0 6 (p/po) 6 0.3) [2,3] (henceforth termed variability), depicted as function of the used degrees of

freedom for MIL-101(Cr) (a), UiO-66 (b), Norit RB 2 (c), c-alumina (d) and Sigma-1 (e). Closed symbols are based on full dataset, open symbols, are obtained when the BET C

parameter is constrained to positive values only if the constraint excludes part(s) of the data set. Results obtained with the linear tting method.

this way, large residuals will indicate data points that require fur- investigated more closely, still under the assumption that the

ther inspection. As these residuals are scale-dependent, they model used is correct. It is not opted here to automatically remove

should be normalized or Studentized. Studentized residuals are a data point that gives rise to |ressi | > 23 as this would require the

the residuals obtained divided by their estimated standard devia- knowledge a priori that the BET model is a proper description, but

tion [85]. These scale-independent Studentized residuals, ressi , are visual inspection of these Studentized residuals might be a better

assumed to follow a normal distribution with zero mean and stan- aid when selecting and possibly eliminate outliers at the lowest

dard deviation of unity [85]. Large residuals, |ressi | > 23 (9599% relative pressures. For c-alumina, indeed the data points at lowest

condence level), are very likely to be outliers and should be relative pressures have the largest residuals. As depicted in Fig. S

210 M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215

31, the Studentized residuals are signicantly larger at low relative test (See S-18) indicates that the proposed tting strategy signi-

pressures, indicating inaccurate description of adsorption by the cantly improves the quality of the t for the materials under

BET equation in this region. Removing these data points increases investigation.

the quality of the t substantially and reduces the condence

interval of the estimated surface area (see Fig. S 31). The t

becomes impeccable for (q/qm) > 0.9 where indeed (p/po) > 0.05. 3.4. Textural characterization in literature the case of MIL-101

This is supported by the seemingly randomly distributed Studen-

tized residuals and the strong correspondence between the mea- As mentioned in the introduction, the large number of publica-

sured isotherm and tted curve (see Fig. S 31). Furthermore this tions on MIL-101 calls for an investigation of the variation in

can be concluded from the normal probability plots accompanying reported pore volumes and surface areas [11,12,4066]. Fig. 1

the ts (see Fig. S 32). With the decreasing condence interval the showed the scatter in BET surface area as function of the pore vol-

value of surface area increases gradually, as shown in Fig. S 33. umes reported in literature, whereas a clear correlation is expected

Note that initially the Studentized residuals do not necessarily between these parameters. Less than half of the cited papers indi-

decrease signicantly, as they are renormalized each time a mea- cated the relative pressure used for the pore volume determination

sured point is eliminated. Obtaining residuals without any tailing and more than half of these used a relative pressure very close to

at low relative pressures, starts occurring at (p/po) > 0.05 for c-alu- unity. At high relative pressures ((p/po) > 0.9) condensation of

mina, when (q/qm) P 0.9, putting most emphasis on multilayer for- nitrogen in inter-particle spaces may occur, and this contribution

mation during the tting procedure, the essence of the BET theory. should clearly not be included in the pore volume. The extent of

For MIL-101(Cr) the story is completely different. Because of the this depends on the particle size of the material. Therefore the pore

distinct kinks in the adsorption isotherm, the BET equation will volumes from the isotherm data available in literature are recalcu-

inherently yield a poor description of the adsorption behaviour. lated at a xed relative pressure, (p/po) 0.4. This might be an

This is reected in the Studentized residuals (Fig. S 34). Thus there unusual low relative pressure, but the MIL-101 structure should

is no statistical incentive to remove particularly the data at low rel- already be saturated at this pressure, and the uncertainty in pore

ative pressures. If one were to, despite previous statistical argu- volume is lower (see Fig. 2). Furthermore, the adsorption isotherms

ments, eliminate points of largest residuals in an iterative all overlap once rescaled with their pore volume (Fig. S 37), except

fashion, one would need to remove almost the entire dataset for one or two samples with purposely defect-induced mesoporos-

(18 data points) to yield some sort of randomly distributed resid- ity. Comparing this recalculated pore volume with the originally

uals (see Figs. S 34 and S 35). The obtained t parameter values for reported value (see Fig. S 36), shows that the literature value is sig-

the latter do not represent the isotherm for MIL-101(Cr) any better nicantly higher in most cases. This is not caused by the low rela-

than those obtained without any exclusion of data points and even tive pressure chosen, since the pore volume at this relative

have a higher uncertainty (see Table S 6). This corresponds with pressure is at most 5% lower than the one calculated at the more

the notion that the linearized BET curve does not show a signicant often used (p/po) 0.9 for the majority of cases (Fig. S 37). Clearly,

linear section (see Fig. S 22). From both a statistical and physical care must be taken when drawing conclusions based on reported

point of view thus, for materials deviating strongly in adsorption pore volumes in literature. When it comes to the BET surface area,

behaviour from the BET formulation as was exemplied for MIL- for less than one third of the values reported in literature the range

101(Cr), one should be very careful with the removal of data of relative pressures or degrees of freedom used for its determina-

points. tion were stated. Furthermore, BET areas were frequently reported

One might opt to consider using the experimentally found with up to six signicant digits, suggesting an accuracy that, in

uncertainties from adsorption measurements as weights for the view of the ndings in this work, is highly exaggerated (see

determination of the BET surface area, as was previously recom- Table 2). For the cited literature sources the BET surface area is

mended by Pendleton and Badalyan [80]. The use of this approach redetermined using the linear, direct and weighted direct method.

however, is strongly discouraged. As uncertainties are cumulative, In all cases the maximum degrees of freedom available for (p/

the highest weights are given to the rst data points. This in turn po) < 0.3 were used. This because, due to the particular shape of

means that highest importance is given to the data at lowest rela- MIL-101, there is no clear statistically valid reason to eliminate

tive pressures, putting strong emphasis on the region where sur- data points, as discussed above. Moreover, using the maximum

face heterogeneity might signicantly interfere. As a result the

obtained C parameter will be articially increased and the BET sur-

face area decreased. Indeed Pendleton and Badalyan obtained, for a

BET-like reference sample even, lower specic surface areas than

for their unweighted case, both based on the linear tting method

[80]. Because the linear tting method puts, compared to the direct

method, more emphasis on high relative pressure data, as has been

shown by Van Erp and Martens [37], the undesirable inuence of

including experimental uncertainties as weights is likely to be even

larger for the direct method.

Lastly, the above discussion on the BET surface area determina-

tion is based on the third isotherm measurement of each of the

materials. For the various tting strategies investigated the differ-

ences in specic surface area and 95% condence interval deter-

mined from all three consecutive measurements on the same

sample are minor (See S-15), indicating the reproducibility of the

BET surface area determination procedure. Furthermore, based on

these repeated measurements, a lack-of-t test can be performed

to quantify how well the BET-model can be tted to describe Fig. 7. Recalculated pore volume and BET area for MIL-101(Cr), determined with

adsorption behaviour, in a similar fashion as for the estimation of the linear (j), direct ( ) and weighted direct method ( ) for the literature data

kinetic reaction parameters from repeated experiments [26]. This depicted in Fig. 1.

M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215 211

degrees of freedom will decrease uncertainty. Comparison of the surface area. Furthermore, applying the direct tting method

original and recalculated BET areas, Fig. S 38, shows again that lit- shows the highest correlation between surface area and pore vol-

erature values are exaggerated, albeit less pronounced than it was ume (R2 of 0.96), and the least variability in BET surface area.

the case for the pore volume. Depicting the recalculated surface The weighted direct (R2 of 0.95) and linear method (R2 of 0.93) per-

area as a function of the recalculated pore volume (Fig. 7) gives a form slightly worse. Hence to decrease variability of surface area

signicantly stronger correlation than for the values reported in lit- between different samples of the same compound, one should ide-

erature (Fig. 1). This example clearly shows the necessity of stan- ally use the direct tting method (nonlinear parameter

dardizing conditions for the determination of pore volume and estimation).

Fig. 8. BJH pore size distribution including 95% condence intervals for both the x- and y-coordinate based on the desorption branch of the isotherm for MIL-101(Cr) (a), UiO-

66 (b), Norit RB2 (c), c-alumina (d), Sigma-1 (e) for the third measurement of each material and H-ZSM-5 (f) with articially created mesopores [86]. BJH-calculations

purposely extended to lower relative pressure than is generally recommended (below (p/po) = 0.42) to show trends in distribution and uncertainty.

212 M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215

3.5. Uncertainty in BJH-pore size distribution from pore size distributions. Even for the mesopore-containing

materials under investigation, the 95% condence interval becomes

In Fig. 8, pore size distributions and their 95% condence prohibitively large especially for small pore sizes. This occurs

intervals are given for the materials under investigation. As rst already for Dp > 3.4 nm, indicating also from the error analysis

conclusion, the uncertainty in pore diameter (x-axis) is negligibly perspective that the BJH-method, based on the desorption branch,

small for Dp < 50 nm. Above this pore size, the condence intervals should not be used for (p/po) < 0.42. If one would desire to make

become signicant, showing that indeed the BJH-method is not conclusions based on the pore size distribution quantitatively,

suitable for macroporous materials. This is easily rationalized when one would need to severely decrease the uncertainty therein.

considering that relative pressures corresponding with larger pore This might be accomplished by increasing the sample amount and

diameters are close to unity. The uncertainty in pore diameter is by measuring signicantly less points in the adsorption branch.

roughly proportional to |ln1(p/po)|, see Eqs. S-3.4S-3.5 (S.I), and Furthermore, both Sigma-1 and Norit-RB2 show a small peak in

thus greatly amplied at these relative pressures. Secondly, the the pore size distribution around 3.8 nm, which is, as explained by

condence interval for incremental pore volume per pore diameter Groen et al. [39], due to the so called tensile strength effect (TSE). To

(DVp/DDp) is highest at smallest pore diameters. This is due to the elaborate on this in particular, a sample of mesoporous zeolite H-

recursive nature of the BJH-calculation, which has its starting point ZSM-5 showing distinct type-H2 hysteresis behaviour (IUPAC

at high relative pressure, and thus large pore diameter, and ends at classication) [2,3], obtained via desilication has been included

smallest pore diameters. From Eq. S-3.15 (S.I), it becomes apparent [86]. This because the TSE is particularly visible for this type of

that the uncertainty for a given incremental pore volume is largely hysteresis [4]. The forced closure of the desorption branch near

inuenced by the sum of uncertainties in surface areas of the pores (p/po) = 0.42 indicates the occurrence of this tensile strength effect

larger than the pore size for the data point under investigation. This (TSE), and does not point to a well-dened mesopore size. The BJH

uncertainty is expressed in the pore volume and surface area asso- model applied to the adsorption branch demonstrates this (see

ciated with this pore size, and cumulatively propagates towards Fig. S 39), as the TSE phenomenon is absent. The TSE leads often

smaller pore diameters. As this accumulation will only become to misinterpretation of the pore size distribution: the peak

apparent when pore volume and surface area are of signicant observed at around 4 nm in de BJH from the desorption branch

magnitude, removing from the calculation pores for which (Fig. 8), does not reect the exact porous properties of the material,

Dp > 200 nm will not generate an observable reduction in uncer- but primarily the nature of the adsorptive N2. Still publications in

tainty for smaller pore sizes. Even for the microporous materials highly respected journals appear where the contribution of

under investigation, UiO-66, Norit-RB2 and Sigma-1, that show the TSE is erroneously attributed to the presence of real pores

hardly or no mesopore volume, the uncertainty is signicant. [87,88].

Indeed, the BJH-method is not suited for the microporous region, Because of the smaller uncertainty in the adsorbed quantities in

but these results are included to show that, also for low pore vol- the adsorption branch, uncertainty in the pore volume is generally

umes, the uncertainty is signicant. In general the magnitude of smaller for similar pore sizes. Uncertainties become increasingly

this uncertainty mitigates rm quantitative conclusions drawn large when Dp < 2 nm, the border to the microporous region. Again

Table 3

Recommendations and guidelines for texture characterization using volumetric N2 adsorption @77 K.

When Recommendations

Measuring N2 adsorption Use as minimal sample mass:

isotherm o 0.5 g for weak adsorption

o 0.20.3 g for moderate adsorption (K > 10 bar1 and 100 < qm < 200 mlSTP g1)

o 0.1 g for strong adsorption

Use 2 6 (Vman/Vcell) 6 3 to minimize uncertainty

Keep Vcell small (10 ml) to avoid articially introducing erroneous desorption behaviour

o Use ller rods to decrease Vcell

Use a low but relevant (p/po) to reduce uncertainty, while making sure the material is saturated

The number of signicant digits should correspond with the condence interval of the isotherm

Calculation BET surface Use the direct tting method (nonlinear estimation parameters and condence intervals)

area Before tting, determine upper boundary of relative pressure window using the two-point BET method (for both micro- and mes-

oporous materials)

For mesoporous materials, investigate Studentized residuals to investigate whether low pressure data points are outliers and eli-

gible for exclusion from tting (due to surface heterogeneity)

Within this pressure window, use the maximum degrees of freedom available

Reporting BET surface area Indicate the full tting strategy, including:

o The tting method

o The pressure window

o The degrees of freedom

The number of signicant digits should correspond with the condence interval

BJH pore size distribution Compare the pore size distribution of both the adsorption and desorption branch to make sure that any TSE-related artefact is

(if used)* absent

Keep Vcell small (10 ml) during measurements to avoid articially introducing desorption hysteresis, and an erroneous pore size

distribution

*

In most cases condence interval of pore size distribution statistically undistinguishable from the null hypothesis.

M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215 213

the cumulative nature of the BJH-method is demonstrated. Note o The difference in maximum and minimum estimated BET

that it is strictly not advocated here to apply the BJH-method to surface areas from a single isotherm (variability) is lim-

the adsorption branch of an isotherm, unless it is used to verify ited to 2040% if more than ten degrees of freedom are

the existence of a tensile strength effect related artefact. used. If less degrees are used this can increase up to

As noted before, desorption hysteresis can be articially 400%.

increased by decreasing sample mass and by increasing sample cell o Variability is lowest for the direct method (nonlinear

volume. This artefact has a signicant effect on the BJH pore size parameter estimation) and highest for the linear method,

distribution, when based on the desorption branch (see Fig. S attributed to the higher emphasis on points close to satu-

14). With increasing sample cell volume and decreasing sample ration for the latter.

mass, a strong increase in scatter of the pore size distribution is o To obtain a small uncertainty in BET surface area, one

created. More importantly, the contribution of smaller mesopores should at least use three degrees of freedom (at least 5

is erroneously enhanced. This adverse effect can increase the vol- data points) regardless of tting strategy.

ume contribution of pores with 3.4 nm 6 Dp 6 6 nm up to 5 times o The condence intervals of the BET parameters from the

(S-8, S.I). As an overall conclusion of the error analysis the BJH direct method are generally slightly smaller than those

analysis yield statistically insignicant quantitative information. obtained from the linear method.

At most a qualitative impression of a pore size distribution is Because of the strong inuence of tting strategy on BET surface

obtained. area and uncertainty therein it is strongly recommended to

report the pressure window, degrees of freedom and tting

3.6. Recommendations for adsorptive characterization method applied alongside the obtained BET surface area.

For microporous materials, it is proposed to exclude data points

To optimize the N2 adsorption methodology for the character- close to saturation, as the BET equation is unt to describe

ization of porous materials and to improve the data analysis, the saturation.

ndings of the error analysis are summarized below in detail and

have resulted in recommendations and guidelines collected in o By not adhering to this, one might obtain:

Table 3.

j Negative values for the BET C parameter when apply-

Regarding uncertainty in adsorption measurements:

ing the linear method and thus unphysical results.

Uncertainty propagates cumulatively, and is lowest for the rst j Strongly enlarged uncertainties because of insensi-

point of the adsorption branch and highest for the last point tivity towards C when using the direct method.

measured in the desorption branch. The 95% condence interval j Lower values for the BET surface area for both tting

in the adsorption isotherms found are below 10 mlSTP g1 for methods.

adsorption and below 20 mlSTP g1 for desorption.

Uncertainty can be decreased by increasing the accuracy of the o By adhering to this, one would obtain:

pressure measurements, the calibration of the manifold volume,

j A signicant reduction in uncertainty in obtained

by optimizing ratio of manifold and cell volume and by increas-

parameters.

ing sample mass.

j A signicant reduction in variability.

A large sample cell volume and/or small sample mass can arti-

cially and erroneously enlarge or introduce hysteresis between

The contrived two-point BET method is a useful tool to deter-

ad- and desorption because of reduced sensitivity to determine

mine a priori the upper relative pressure boundary of the BET

equilibration.

window (close to saturation), an alternative to the method

reported by Rouquerol et al. [33]).

For the determination of pore volume:

For mesoporous materials, it is recommended to replace the

The relative condence interval in pore volume can be as large upper boundary of the relative pressure window ((p/po) 6 0.35

as 10% ((p/po) = 0.9, Sigma-1), for the measurements per- as posed by Brunauer et al. [34], or (p/po) 6 0.30 by IUPAC

formed. To reduce the relative condence interval in pore vol- [2,3]) with the relative pressure where the C parameter calcu-

ume ((p/po) = 0.9) to below 5%, one should use: lated with the two-point method becomes negative.

No method was obtained to a priori exclude data for the low rel-

o More than 0.5 g for weakly adsorbing materials (low qm ative pressure range where surface heterogeneity may interfere

and K). strongly for mesoporous materials especially.

o Around 0.20.3 g for moderately adsorbing materials

(K > 10 bar1 and 100 < qm < 200 mlSTP g1). o Here it is suggested to use Studentized residuals, ressi . Pro-

o Less than 0.1 g for strongly adsorbing materials. vided the model isotherm is correct, data points become

For microporous materials, it is recommended to determine the eligible for possible exclusion when |ressi | > 23.

pore volume at relative pressures lower than 0.9. This decreases

the uncertainty in pore volume without inuencing signi- 3.7. Uncertainty in BJH-pore size distribution

cantly the absolute value found and automatically ignores

inter-particle condensation. The magnitude of the 95% condence found for BJH-pore size

The relative uncertainty in pore volume is lowest when distributions severely impedes drawing quantitative

(Vman/Vcell) is between 2 and 3. conclusions.

The articially increased or introduced desorption hysteresis by

When it comes to determination of BET area: unt experimental operation (low sample weight and large

The obtained BET surface area and condence interval are sample volume) has a detrimental effect on the desorption

strongly dependent on applied tting strategy: branch based BJH pore size analysis.

214 M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215

Still the tensile-strength-effect (TSE) related artefact in desorp- The articially increased desorption hysteresis by unt experimen-

tion-branch related pore-size distributions is often overlooked, tation has a detrimental effect on the desorption branch based BJH

yielding erroneous conclusions on seemingly inherent material pore size analysis. Regarding interpretation and reporting of

properties. results human inuences have a strong impact, as exemplied by

the case of MIL-101(Cr). For pore volumes and especially BET sur-

o By comparing the pore-size distribution of the ad- and face areas reported in literature, often the relative pressure (win-

desorption branch one can easily spot the presence of this dow) used and determination strategy are not reported or plainly

artefact. wrong. Using the guidelines posed in this work for the determina-

tion of both parameters, a signicantly better correlation between

Regarding interpretation and reporting of results human inuences both was obtained than was the case for the original values

have a strong impact, as exemplied by the case of MIL-101(Cr). reported in literature.

Less than half of the cited papers indicate the relative pressure

used for the pore volume determination for MIL-101 and of Notes

those that did, more than half erroneously overlooked inter-

particle condensation and overestimated the pore volume The authors declare no competing nancial interest.

signicantly.

Less than one third of reported BET surface areas in literature

Acknowledgments

were accompanied by the relative pressure window and/or

degrees of freedom used for their determination.

Financial support from the Advanced Dutch Energy Materials

BET areas are reported with a highly exaggerated number of

(ADEM) program of the Dutch Ministry of Economic Affairs,

insignicant digits.

Agriculture and Innovation is gratefully acknowledged. Canan

Depicting the recalculated surface area as function of the recal-

Gcyener is gratefully acknowledged for the synthesis of Sigma-1.

culated pore volume gives a signicantly better correlation

between both than was the case for the values reported in liter-

ature, and clearly underlines the potential of standardizing con- Appendix A. Supplementary data

ditions for the determination of pore volume and surface area.

Supplementary data associated with this article can be found, in

Finally, it is realized that the use of nitrogen for this type of adsorp- the online version, at http://dx.doi.org/10.1016/j.micromeso.2014.

tive characterization is under discussion and argon adsorption at 08.048.

liquid argon temperature (87 K) may be preferred as it is slightly

smaller and has no quadrupole moment unlike nitrogen [89,90]. References

However the costs of liquid argon are higher and nitrogen is still

[1] F. Rouquerol, J. Rouquerol, K. Sing, Adsorption by Powders & Porous Solids,

deeply penetrated in daily texture characterization, making that Academic Press, London, 1999.

the presented analysis is nevertheless of high relevance. [2] K. Sing, D. Everett, R. Haul, L. Moscou, R. Pierotti, J. Rouquerol, T.

Siemieniewska, Pure Appl. Chem. 54 (1982) 2201.

[3] R. Pierotti, J. Rouquerol, Pure Appl. Chem. 57 (1985) 603619.

[4] F. Schth, K.S.W. Sing, J. Weitkamp, Handbook of Porous Solids, Wiley-VCH,

4. Conclusions 2002.

[5] J.-R. Li, R.J. Kuppler, H.-C. Zhou, Chem. Soc. Rev. 38 (2009) 14771504.

[6] S. Couck, J.F.M. Denayer, G.V. Baron, T. Rmy, J. Gascon, F. Kapteijn, J. Am.

In this contribution the inuence of uncertainty in adsorptive Chem. Soc. 131 (2009) 63266327.

characterization of porous solids, with special emphasis on the [7] C. Gcyener, J. van den Bergh, J. Gascon, F. Kapteijn, J. Am. Chem. Soc. 132

adsorption isotherm measurements and on the determination of (2010) 1770417706.

[8] L.J. Murray, M. Dinca, J.R. Long, Chem. Soc. Rev. 38 (2009) 12941314.

pore volume, BET area and pore size distribution has been studied. [9] J. Juan-Alcaniz, J. Gascon, F. Kapteijn, J. Mater. Chem. 22 (2012) 1010210118.

Uncertainty in adsorption measurements can be decreased not only [10] J. Gascon, U. Aktay, M.D. Hernandez-Alonso, G.P.M. van Klink, F. Kapteijn, J.

by increasing measuring accuracy or sample mass, but also by Catal. 261 (2009) 7587.

[11] G. Frey, C. Mellot-Draznieks, C. Serre, F. Millange, J. Dutour, S. Surbl, I.

optimizing the ratio of manifold and cell volume (optimum at

Margiolaki, Science 309 (2005) 20402042.

Vman/Vcell is 23). Further, a large sample cell volume and/or small [12] G. Akiyama, R. Matsuda, H. Sato, A. Hori, M. Takata, S. Kitagawa, Microporous

sample mass can articially and erroneously enlarge or even intro- Mesoporous Mater. 157 (2012) 8993.

duce articially apparent hysteresis between ad- and desorption. [13] J. Juan-Alcaiz, E.V. Ramos-Fernandez, U. Lafont, J. Gascon, F. Kapteijn, J. Catal.

269 (2010) 229241.

To reduce the relative uncertainty in the determination of pore [14] E.V. Ramos-Fernandez, C. Pieters, B. van der Linden, J. Juan-Alcaniz, P. Serra-

volume for microporous materials it is benecial to determine the Crespo, M.W.G.M. Verhoeven, H. Niemantsverdriet, J. Gascon, F. Kapteijn, J.

pore volume at relative pressures lower than 0.9. When it comes to Catal. 289 (2012) 4252.

[15] S.N. Klyamkin, E.A. Berdonosova, E.V. Kogan, K.A. Kovalenko, D.N. Dybtsev, V.P.

determination of BET area, obtained surface areas and condence Fedin, Chem. Asian J. 6 (2011) 18541859.

intervals are strongly dependent on applied tting strategy. To [16] D. Jiang, L.L. Keenan, A.D. Burrows, K.J. Edler, Chem. Commun. 48 (2012)

obtain a small uncertainty in BET surface area, one should at least 1205312055.

[17] C.-X. Yang, X.-P. Yan, Anal. Chem. 83 (2011) 71447150.

use three degrees of freedom (at least 5 data points) and apply the [18] J.H. Cavka, S. Jakobsen, U. Olsbye, N. Guillou, C. Lamberti, S. Bordiga, K.P.

direct (nonlinear) tting method. Excluding data points close to Lillerud, J. Am. Chem. Soc. 130 (2008) 1385013851.

saturation, results in reduction in uncertainty and variability. The [19] H. Wu, T. Yildirim, W. Zhou, J. Phys. Chem. Lett. 4 (2013) 925930.

[20] S. Chavan, J.G. Vitillo, D. Gianolio, O. Zavorotynska, B. Civalleri, S. Jakobsen,

contrived two-point BET method is a useful tool to determine a M.H. Nilsen, L. Valenzano, C. Lamberti, K.P. Lillerud, S. Bordiga, Phys. Chem.

priori the upper relative pressure boundary of the BET window. Chem. Phys. 14 (2012) 16141626.

For mesoporous materials, it is recommended to use the same [21] A.D. Wiersum, E. Soubeyrand-Lenoir, Q. Yang, B. Moulin, V. Guillerm, M.B.

Yahia, S. Bourrelly, A. Vimont, S. Miller, C. Vagner, M. Daturi, G. Clet, C. Serre, G.

upper boundary. No method was obtained to a priori exclude data

Maurin, P.L. Llewellyn, Chem. Asian J. 6 (2011) 32703280.

for the low relative pressure range where surface heterogeneity [22] F. Vermoortele, B. Bueken, G. Le Bars, B. Van de Voorde, M. Vandichel, K.

may interfere strongly but it is suggested to use Studentized Houthoofd, A. Vimont, M. Daturi, M. Waroquier, V. Van Speybroeck, J. Am.

residuals for to help locate this boundary. Chem. Soc. 135 (2013) 1146511468.

[23] Y. Iwase, Y. Sakamoto, A. Shiga, A. Miyaji, K. Motokura, T.-R. Koyama, T. Baba, J.

The magnitude of the 95% condence limits for BJH-pore size Phys. Chem. C 116 (2012) 51825196.

distributions severely impedes drawing quantitative conclusions. [24] H.-Y. Jeon, C.-H. Shin, H.J. Jung, S.B. Hong, Appl. Catal. A 305 (2006) 7078.

M.F. De Lange et al. / Microporous and Mesoporous Materials 200 (2014) 199215 215

[25] Y. Kumita, J. Gascon, E. Stavitski, J.A. Moulijn, F. Kapteijn, Appl. Catal. A 391 [60] N. Klein, A. Henschel, S. Kaskel, Microporous Mesoporous Mater. 129 (2010)

(2011) 234243. 238242.

[26] F. Kapteijn, H.L. Bredt, E. Homburg, J.C. Mol, Ind. Eng. Chem. Prod. Res. Develop. [61] J. Yang, Q. Zhao, J. Li, J. Dong, Microporous Mesoporous Mater. 130 (2010) 174

20 (1981) 457466. 179.

[27] F. Kapteijn, L. Singoredjo, N.J. Dekker, J.A. Moulijn, Ind. Eng. Chem. Res. 32 [62] T.K. Trung, N.A. Ramsahye, P. Trens, N. Tanchoux, C. Serre, F. Fajula, G. Frey,

(1993) 445452. Microporous Mesoporous Mater. 134 (2010) 134140.

[28] L. Zhou, Y. Guo, Q. Zhang, M. Yagi, J. Hatakeyama, H. Li, J. Chen, M. Sakurai, H. [63] P. Ksgens, M. Rose, I. Senkovska, H. Frde, A. Henschel, S. Siegle, S. Kaskel,

Kameyama, Appl. Catal. A 347 (2008) 200207. Microporous Mesoporous Mater. 120 (2009) 325330.

[29] S. Sircar, T. Golden, M. Rao, Carbon 34 (1996) 112. [64] J.S. Lee, S.H. Jhung, J.W. Yoon, Y.K. Hwang, J.S. Chang, J. Ind. Eng. Chem. 15

[30] D. Lozano-Castello, J. Alcaniz-Monge, D. Cazorla-Amoros, A. Linares-Solano, W. (2009) 674676.

Zhu, F. Kapteijn, J. Moulijn, Carbon 43 (2005) 16431651. [65] S.H. Jhung, J.H. Lee, J.W. Yoon, C. Serre, G. Frey, J.S. Chang, Adv. Mater. 19

[31] C.Y. Yin, M.K. Aroua, W.M.A.W. Daud, Sep. Purif. Technol. 52 (2007) 403415. (2007) 121124.

[32] A.R. Millward, O.M. Yaghi, J. Am. Chem. Soc. 127 (2005) 1799817999. [66] X.X. Huang, L.G. Qiu, W. Zhang, Y.P. Yuan, X. Jiang, A.J. Xie, Y.H. Shen, J.F. Zhu,

[33] J. Rouquerol, P. Llewellyn, F. Rouquerol, Is the BET equation applicable to CrystEngComm 14 (2012) 16131617.

microporous adsorbents?, in: F.R.-R.J.R. P.L. Llewellyn, N. Seaton (Eds.) Studies [67] A. Stewart, Zeolite sigma-1, in: United Kingdom Patent Application, 1988.

in Surface Science and Catalysis, Elsevier, 2007, pp. 4956. [68] S.J. Gregg, K.S.W. Sing, Adsorption, Surface Area and Porosity, Academic Press,

[34] S. Brunauer, P.H. Emmett, E. Teller, J. Am. Chem. Soc. 60 (1938) 309319. London, 1982.

[35] I. Langmuir, J. Am. Chem. Soc. 38 (1916) 22212295. [69] P.H. Emmett, S. Brunauer, J. Am. Chem. Soc. 59 (1937) 15531564.

[36] K.S. Walton, R.Q. Snurr, J. Am. Chem. Soc. 129 (2007) 85528556. [70] A.L. McClellan, H.F. Harnsberger, J. Colloid Interface Sci. 23 (1967) 577599.

[37] T.S. Van Erp, J.A. Martens, Microporous Mesoporous Mater. 145 (2011) 188 [71] T. Osmari, R. Gallon, M. Schwaab, E. Barbosa-Coutinho, J. Severo, J. Pinto,

193. Adsorpt. Sci. Technol. 31 (2013) 433458.

[38] E.P. Barrett, L.G. Joyner, P.P. Halenda, J. Am. Chem. Soc. 73 (1951) 373380. [72] H. Lineweaver, D. Burk, J. Am. Chem. Soc. 56 (1934) 658666.

[39] J.C. Groen, L.A. Peffer, Microporous Mesoporous Mater. 60 (2003) 117. [73] O.A. Hougen, K.M. Watson, Chemical Process Principles, Wiley, New York,

[40] P. Trens, H. Belarbi, C. Shepherd, P. Gonzalez, N.A. Ramsahye, U.H. Lee, Y.K. Seo, 1947.

J.S. Chang, J. Phys. Chem. C 116 (2012) 2582425831. [74] G.F. Froment, K.B. Bischoff, Chemical Reactor Analysis and Design, 2nd ed.,

[41] X. Liu, H. Zhou, Y. Zhang, Y. Liu, A. Yuan, Chin. J. Chem. 30 (2012) 25632566. Wiley, New York, 1991.

[42] K. Munusamy, G. Sethia, D.V. Patil, P.B. Somayajulu, Chem. Eng. J. 195196 [75] C.G. Shull, J. Am. Chem. Soc. 70 (1948) 14051410.

(2012) 359368. [76] J.H. de Boer, B.C. Lippens, B.G. Linsen, J.C.P. Broekhoff, A. van den Heuvel, T.J.

[43] I. Senkovska, E. Barea, J.A.R. Navarro, S. Kaskel, Microporous Mesoporous Osinga, J. Colloid Interface Sci. 21 (1966) 405414.

Mater. 156 (2012) 115120. [77] W.D. Harkins, G. Jura, J. Chem. Phys. 11 (1943) 431432.

[44] S.N. Kim, S.T. Yang, J. Kim, J.E. Park, W.S. Ahn, CrystEngComm 14 (2012) 4142 [78] M. Kruk, M. Jaroniec, A. Sayari, Langmuir 13 (1997) 62676273.

4147. [79] J.R. Taylor, An Introduction to Error Analysis, 2nd ed., University Science Books,

[45] M. Anbia, V. Hoseini, Chem. Eng. J. 191 (2012) 326330. 1997.

[46] M. Anbia, V. Hoseini, J. Nat. Gas Chem. 21 (2012) 339343. [80] A. Badalyan, P. Pendleton, Langmuir 19 (2003) 79197928.

[47] L. Bromberg, Y. Diao, H. Wu, S.A. Speakman, T.A. Hatton, Chem. Mater. 24 [81] P. Pendleton, A. Badalyan, Colloids Surf. A 300 (2007) 253258.

(2012) 16641675. [82] A. Badalyan, P. Pendleton, J. Colloid Interface Sci. 326 (2008) 17.

[48] H.B.T. Jeazet, C. Staudt, C. Janiak, Chem. Commun. 48 (2012) 21402142. [83] T. Keii, T. Takagi, S. Kanetaka, Anal. Chem. 33 (1961) 1965.

[49] Z. Saedi, S. Tangestaninejad, M. Moghadam, V. Mirkhani, I. Mohammadpoor- [84] F. Salvador, C. Snchez-Jimnez, M.J. Snchez-Montero, A. Salvador, A review

Baltork, Catal. Commun. 17 (2012) 1822. of the application of the BET equation to experimental data: the C parameter,

[50] J. Ehrenmann, S.K. Henninger, C. Janiak, Eur. J. Inorg. Chem. 2011 (2011) 471 in: B.M.J.R. F. Rodriguez-Reinoso, K. Unger (Eds.) Studies in Surface Science and

474. Catalysis, Elsevier, 2002, pp. 379386.

[51] Z. Zhao, X. Li, S. Huang, Q. Xia, Z. Li, Ind. Eng. Chem. Res. 50 (2011) 22542261. [85] R.D. Cook, S. Weisberg, Residuals and Inuence in Regression, Chapman and

[52] Z. Zhao, X. Li, Z. Li, Chem. Eng. J. 173 (2011) 150157. Hall, New York, 1982.

[53] Z. Zhang, S. Huang, S. Xian, H. Xi, Z. Li, Energy Fuels 25 (2011) 835842. [86] S. Sartipi, K. Parashar, M. Makkee, J. Gascon, F. Kapteijn, Catal. Sci. Technol. 3

[54] C.Y. Huang, M. Song, Z.Y. Gu, H.F. Wang, X.P. Yan, Environ. Sci. Technol. 45 (2013) 572575.

(2011) 44904496. [87] L.G. Qiu, T. Xu, Z.Q. Li, W. Wang, Y. Wu, X. Jiang, X.Y. Tian, L.D. Zhang, Angew.

[55] J. Shi, Z. Zhao, Q. Xia, Y. Li, Z. Li, J. Chem. Eng. Data 56 (2011) 34193425. Chem. Int. Ed. 47 (2008) 94879491.

[56] P.B. Somayajulu Rallapalli, M.C. Raj, D.V. Patil, K.P. Prasanth, R.S. Somani, H.C. [88] L. Li, S. Xiang, S. Cao, J. Zhang, G. Ouyang, L. Chen, C.-Y. Su, Nat. Commun. 4

Bajaj, Int. J. Energy Res. 37 (2013) 746753. (2013) 1774.

[57] K. Yang, Q. Sun, F. Xue, D. Lin, J. Hazard. Mater. 195 (2011) 124131. [89] M. Thommes, Chem.-Ing.-Tech. 82 (2010) 10591073.

[58] D. Jiang, A.D. Burrows, K.J. Edler, CrystEngComm 13 (2011) 69166919. [90] J. Silvestre-Albero, A. Silvestre-Albero, F. Rodrguez-Reinoso, M. Thommes,

[59] N.A. Khan, J.W. Jun, S.H. Jhung, Eur. J. Inorg. Chem. (2010) 10431048. Carbon 50 (2012) 31283133.

## Гораздо больше, чем просто документы.

Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.

Отменить можно в любой момент.