Chemical Engineering Science 66 (2011) 171182

Contents lists available at ScienceDirect

Chemical Engineering Science

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

Modelling and experimental verication of cellulose substitution kinetics

Tapio Salmi a, Pia Damlin a, Jyri-Pekka Mikkola a,b, Matias Kangas a,n
a
Abo Akademi, Process Chemistry Centre, Laboratory of Industrial Chemistry and Reaction Engineering, FI-20500 Abo/Turku, Finland
b
Umea University, Department of Chemistry, Chemical-Biochemical Center, Technical Chemistry, SE-90187 Umea, Sweden

a r t i c l e in f o abstract

Article history: Cellulose and starch derivatives are often prepared by substitution reactions to the hydroxyl groups (OH-2, OH-3
Received 21 May 2010 and OH-6) of the corresponding anhydroglucose units. A general kinetic model for the substitution of cellulose
Received in revised form and starch was developed. The model assumes different reactivities of the hydroxyl groups and a decrease in the
4 September 2010
reactivity as the substitution proceeds. Both phenomena are taken into account in the model. The model predicts
Accepted 6 October 2010
the detailed distribution of the different mono (2, 3, 6)-, di (23, 26, 36)and trisubstituted (236) units as a
Available online 13 October 2010
function of the reaction time. The classical Spurlin distribution is obtained as a special case of the general model.
Keywords: Numerical strategies were developed for the solution and computer simulation of the model. Simulation results
Cellulose substitution were shown to be valid for various reactivity ratios of the hydroxyl groups. The modelling concept was veried
Kinetics
with experimental data obtained for carboxymethylation of cellulose. It should be emphasized, however, that
Modeling
the same general approach is also applicable in studies of any hemicelluloses or oligo- or monosaccharides. The
Non-uniform reactivity
Product distribution detailed data, which revealed the non-uniform reactivities of the different hydroxyl groups and the decline of
Simulation the substitution rate with time, were successfully described by the mathematical model.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction
Substitution of native cellulose, the most abundant biopolymer
on the earth, is a classical process. Cellulose can be obtained from
different origins although that derived from soft- and hardwood via
various pulping processes represents the most common one. The
hydroxyl groups adjacent to carbon atoms 2, 3 and 6 in the
anhydroglucose unit of cellulose (Fig. 1) can undergo several
substitution reactions, some of which are of high technical
importance. The other important native polymer, starch, has an
analogous structure to that of cellulose: three hydroxyl groups can
undergo substitution reactions. The most common cellulose
derivatives are cellulose esters and cellulose ethers. Organic
cellulose esters are prepared by letting acid-pretreated cellulose
to react with an anhydride or acid chloride. The esterication
reaction is schematically presented as follows:
alkene oxides to react with an alkali cellulose. The cellulose bers
are pretreated by an alkali to promote swelling of the bers and
ionization of the hydroxyl groups. With alkyl halogenides, the
reaction proceeds as follows:
M + CellO  +RX-CellOR+M + X 
where cellO is an ionized cellulose and an RX is the alkyl halogenide.
M + denotes an alkali metal ion, typically Na + . Depending on the
halogenide, various products are obtained: with methyl (RX CH3Cl)
and ethyl chloride (RCH3CH2Cl), methyl and ethyl cellulose are
obtained as main reaction products, whereas with a-monochloro-
carboxylic acids carboxyalkylcelluloses are obtained, such as carboxy-
methyl and carboxyethyl celluloses. The most important of them is
carboxymethylcellulose (CMC), which is obtained by letting alkali
cellulose to react with monochloroacetic acid (Fig. 1)

CellOH+ RCOOH-RCOOCell+H2O M + CellO  +CH2ClCOOH-CellOCH2COOH+M + Cl  .

where cellOH denotes the hydroxyl group adjacent to the
anhydroglucose unit. Most common esterication products are
cellulose acetate and cellulose xanthate. They are used in lacquers,
lms, in acetate silk as well as in the preparation of regenerated
cellulose in the viscose process (Sjostrom, 1993). Cellulose ethers
are prepared by letting alkyl halogenides or allyl halogenides or
n
Corresponding author.
E-mail address: Matias.Kangas@abo. (M. Kangas).
The sodium salt of CMC is widely used, for instance, in detergents,
alimentary products, textiles, glues as well as emulgators in various
products, such as dyes, ceramic products and cosmetics.
Alkene oxides, such as ethene and propene oxide react with
cellulose forming hydroxyl alkyl ethers of cellulose. The reaction
mechanism is complicated, leading also to a growth of the hydroxyalkyl
side chains. Only catalytic amounts of OH ions are needed. The main
reaction can be represented in a simplied manner as follows:
CellOH+alkene oxide-CellOROH

0009-2509/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2010.10.013
172 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182
OH
4
HO O HO
HO HO
O O
OH
OR
O
O RO
RO
OR
Fig. 1. Structure of anhydroglucose units of cellulose and carboxymethylcellulose (CMC), R H or CH2COOH.
where RCH2CH2 and RCH2CH(CH3) for ethene and propene oxides,
respectively. Hydroxyalkyl celluloses have widespread applications as
additives in paints, paper adhesives, in ceramics industry as well as in
the preparation of photographic lms.
A common feature for the esterication and etherication reactions
is that their substitution mechanisms can vary between SN1 and SN2.
This implies that the reaction kinetics can vary between rst order
kinetics with respect to cellulose to second order kinetics, to which
both cellulose and the substitution reagent contribute.
The substitution reactions are traditionally carried out with
dispersed cellulose in a solventfor example, in acetic acid or
dichloromethane in acetylation and in aqueous 2-propanol in
carboxymethylation. At least in the initial stage of the reaction,
the process is a heterogeneous one, where the reagent diffusion
into the bers might inuence the reaction rate.
The diffusion limitations can be surmounted by using solvents
for cellulose. In this way, the heterogeneous reaction is trans-
formed to a homogeneous one and the substitution kinetics is
expected to be accelerated. Typical examples of classical water-free
solvent systems for cellulose are based on the use of dimethyl
sulphoxide (DMSO), such as DMSO-paraformaldehyde. Recently,
the discovery of ionic liquids, which are able to dissolve cellulose
among other biomaterials, has opened an entirely new perspective
to the preparation of cellulose derivatives in homogeneous envir-
onments. For example, Heinze et al. (2005) have shown that
acetylation of cellulose is possible in ionic liquids and Mikkola
et al. (2007) have demonstrated that cellulose can be dissolved in
several ionic liquids, such as [Amim][Cl] and [Bmim][Cl], in a more
efcient manner: the dissolution of cellulose in ionic liquids can be
considerably enhanced by an ultrasound treatment (acoustic
cavitation) as was discovered by Mikkola et al. (2007).
Conventionally, the chemical analysis of cellulose substitution
has been restricted to the determination of the overall degree of
substitution (DS), but it has the limitation that the individual
reactivities of the hydroxyl groups are not revealed. However, it is
possible to reveal the details of substitution by an appropriate
analytical method. This has been demonstrated, for instance by
Reuben and Conner (1983) and by Reuben (1986), who applied
NMR spectroscopy on hydrolyzed carboxymethyl cellulose and
methyl cellulose. Kragten et al. (1992) and Saake et al. (2001) used
anion-exchange chromatography with pulsed amperometric
detection to determine the distribution of the substituents in
OH
6 OH
5 O
OH
2 O
3 1
OH
n OH
+ CH2ClCOOH
OR
O
O
O
OR
CMC. The liquid-chromatographic method approach used
recently by us (Damlin et al. 2010) enables the distinction of the
OH-groups and the determination of the concentrations of
substituted hydroxyl groups separately. In conclusion, the
availability of a sophisticated analysis method is the basis for a
detailed kinetic analysis of the substitution process.
Pioneering models for the substitution kinetics of cellulose have
been proposed by Spurlin (1939), who derived the classical distribution
between mono-, di- and trisubstituted anhydroglucose units for the
case that all of the hydroxyl groups (OH-2, OH-3, OH-6) have equal
reactivities. Timell (1950) presented differential equations for the
substitution reactions for cases that rst order kinetics is obeyed
and Salmi et al. (1994) have developed kinetic models for
carboxyalkylation of cellulose with a-chlorocarboxylic acids.
Diffusion phenomena inside the bers was treated (Salmi et al.
1994, Salmi et al. 2003) with an empirical model and with a
reaction-diffusion model. A detailed study, including a comparison
of three analytical methods, on carboxymethylation of starch was
carried out by Lazik et al. (2002), who concluded that the reactivities of
the different hydroxyl groups in starch are different, but their ratio is
constant; thus, the classical model of Spurlin (1939) was applicable.
The aim of the present paper is to develop a generalized
simulation model for the substitution of cellulose and starch,
including both the reactivity differences and concentrations of
the individual hydroxyl groups as well as the prediction of the
product distribution of the substituted anhydroglucose units. Some
simplied cases with rst order kinetics and equal reactivities of
the hydroxyl groups will be presented, too. The models are veried
with experimental results obtained for carboxyalkylation of cellu-
lose in a slurry reactor. The experimental results are based on a
chromatographic approach, which allows us to obtain the detailed
substitution kinetics, and not only the overall degree of substitu-
tion (DS), as was the case in our previous study (Salmi et al. 1994).
2. Reactivities of hydroxyl groups
2.1. General kineticsnon-equal reactivities
The anhydroglucose unit of cellulose has three functional
hydroxyl groups (OH) adjacent to carbon atoms 2, 3 and
6 (OH-2, OH-3, OH-6) (Fig. 1). The reactivity of the hydroxyl
 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182 173

group varies, depending on the substitution reaction and the By denoting a2/6 k2/k6 and separating the variables, we obtain
conditions. The most acidic hydroxyl group is OH-2, which
dcP2 dcP6
implies that it is claimed to be the most reactive one, for a2=6 11
c0 cP2 c0 cP6
instance, in hydroxymethylation. On the other hand, as the size
of the substituent increases, steric factors become important, thus which is easily integrated with the limits [0, cP2] and [0, cP6]. The
preferring the reactivity of OH-6. result becomes
The substitution reactions to OH-2, OH-3 and OH-6 can con-
ln1cP2 =c0 a2=6 ln1cP6 =c0 12
veniently be listed as
OH2 R-P2 which gives
OH3 R-P3 cP2 =c0 11cP6 =c0 a2=6 13
OH6 R-P6 1a2c
Analogously, we obtain for cP3 (a3/6 k3/k6)
where R is a substitution reagent and P2, P3 and P6 denote the
product molecules. In the beginning of the reaction, all of the cP3 =c0 11cP6 =c0 a3=6 14
groups are assumed to be non-substituted. Their initial concentra-
tions are thus equal ( c0)
The overall degree of substitution (DS) is dened as
c02 c03 c06 c0 2 X
DS cPj =c0 15
The total balance for each group can be written as
ci cPi c0 3 DS obtains values between 0 and 3. The expression for cP2 and
cP3, Eqs. (13 and 14) are inserted in Eq. (15) giving
where ci refers to unsubstituted OH-groups and cPi represents the
substituted groups (i2, 3, 6). DS 2 cP6 =c0 1cP6 =c0 a2=6 1cP6 =c0 a3=6 16
In the spirit of Paul Flory (1953), we assume that the reactivity of
each group is independent of the overall degree of substitution
The relative product distribution is obtained as follows. DS is
(DS). Consequently, the substitution reactions to the OH-groups 2, 3
assumed, cP6/c0 is calculated iteratively from Eq. (16), after which
and 6 are characterized by individual rate constants k2, k3 and k6.
cP2/c0 and cP3/c0 are obtained directly from Eqs. (13) and (14).
The formation mechanism for each group is presumed to be similar.
The time dependence of DS is obtained by adding the balances
Based on these hypotheses, the formation rates of the products can
for dcPi/dt, Eq. (9) (ScPj c0DS)
be written as X
rP2 k2 c2 cRa dDS=dt c0R DSc0 a kj c0 cPj =c0 17

rP3 k3 c3 cRa
rP6 k6 c6 cRa 4a2c We denote a c0R/c0 and k0 j kjc0, after which Eq. (17) becomes
X 
dDS=dt kuj c0 cPj =c0 aDSa c0a1 18
where cR denotes the concentration of the substitution reagent and
a is the reaction order depending on the detailed substitution
mechanism (SN1 or SN2). Recalling the ratios between cP2, cP3 and cP6, Eqs. (13) and (14),
The reaction stoichiometry reveals that the amount of reacted we obtain for dDS/dt
substrate (R) (c0R  cR) is equal to the amounts of products X 
formed. Since no products were assumed to be present in the dDS=dt kuj 1cP6 =c0 aj=6 aDSa c0a1 19
beginning of the reaction, the reaction stoichiometry gives
X
c0R cR cP2 cP3 cP6 cPj 5 Eqs. (16) and (19) form a differential-algebraic system with two
unknowns, cP6 and DS. A more straightforward way is to insert the
relations between cP2, cP3 and cP6 in the balance equation for cP6,
The mass balances of the product groups in a constant-volume
which becomes
batch reactor are given by
dcP6 =c0 =dt ku6 1cP6 =c0 acP2 =c0 cP3 =c0 cP6 =c0 a c0a1 20
dcPi =dt rPi 6
where i2, 3, 6. The reactant group concentrations (ci) are solved where cP2/c0 1 (1  cP6/c0)a2/6 and cP3/c0 1 (1  cP6/c0)a3/6.
from Eq. (3) and the reagent concentration is obtained from Eq. (5): The differential Eq. (20) is solved with the initial condition t 0,
cP6/c0 0 after which cP2/c0, cP3/c0 and DS are obtained from
ci c0 cPi 7 Eqs. (13) to (15). It should be mentioned that the approach
X presented is a general one: instead of rst solving cP6/c0, the
cR c0R  cPj 8 calculation can be started with cP2/c0 or cP3/c0 as well.

These relations are inserted in the mass balance, which becomes 2.2. Second order kinetics and equal reactivities
X
dcPi =dt ki c0 cPi c0R  cPj a 9
Some special cases are of interest. Assuming that all of the
groups are equally reactive, k0 2 k0 3 k0 6 k0 , the expression for DS,
The calculation of the product distribution in time scale requires Eq. (19) is simplied to (cP6/c0 DS/3)
a numerical simulation, but the relative product distribution can be
obtained from Eq. (9), for instance by dividing dcP2/dt by dcP6/dt, dDS=dt ku3DSaDSa c0a1 21
which yields
c0 cP2 Analytical solutions can be developed for integer values of a. The
dcP2 =dcP6 k2 =k6 10
c0 cP6 most probable for a bimolecular reaction is that a 1 (SN2), which
174 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182

enables a simple separation of variables and integration The denition of DS, Eq. (15) gives
Z Z X
dDS DS 3 ekit 29
ku dt 22
3DSaDS
Consequently, in case that the hydroxyl groups are equally
After developing the left-hand side to partial fractions and reactive, ki k0 for i2, 3, 6 and DS becomes
integrating with the limits for t and DS; [0, t] and [0, DS], the DS 31ekut 30
solution becomes
3aDS It is important to emphasize that Eqs. (26)(30) are also valid for
ln a3kut 23 the case of a large excess of substitution reagent present in the
a3DS
reaction mixture. In this case, the rate constants (ki and k0 )
from which DS is solved explicitly: incorporate the reagent concentration and the experimentally
a1ea3kut observed kinetics is of pseudo-rst order.
DS 24
1a=3ea3kut
3. Distribution of mono-, di- and trisubstituted units
If a 43, the substitution reagent is available in excess, a 3 40
0 0
and DS approaches ae(a  3)k t/((  a/3)e(a  3)k t)3 with increasing 3.1. Probabilistic distributiongeneral considerations
(a  3)k0 t
t. On the other hand, if a o3, a  3o0 and e approaches zero
with increasing t. The limiting value of DS becomes a, i.e. it is equal The distribution of mono-, di- and trisubstituted units can be
to the dimensionless concentration ratio c0R/c0. obtained by probabilistic considerations. According to Fig. 2, the
Eq. (24), cannot be directly used for the stoichiometric amount concentrations of mono (I)-, di (II)- and trisubstituted (III) units are
(a3). Instead, Eq. (22) can be integrated directly or lHopitals rule dened as
can be applied to Eq. (24) (a  3y, y-0). The result becomes
  cI cu2 cu3 cu6 31
3kut
DS 3 25
1 3kut cII cu23 cu26 cu36 32

cIII cu236 33
The value of DS approaches 3 with an increasing reaction time as
0
easily seen from Eq. (25). where c k, k 2, 3, 6, 23, 26, 36, 236 denote the concentrations of the
corresponding substituted anhydroglucose units (Fig. 2).
2.3. First and pseudo-rst order kinetics The probability that a hydroxyl group (OHi) is substituted in an
anhydroglucose unit is proportional to the dimensionless concentra-
tion of the corresponding substituted group, cPi/c0. In an analogous
In the case of zero order kinetics with respect to the substitution
manner, the probability that a hydroxyl group has remained unsub-
reagent (a 0), the overall reaction order with respect to the
stituted is proportional to the dimensionless concentration of the
product is 1 in Eq. (9)
unsubstituted groups, 1 cPi/c0. Thus, the probability for the appear-
dcPi =dt ki c0 cPi 26 ance of a monosubstituted unit is (cPi/c0)(1 cPj/c0)(1 cPk/c0), where i
which is easily integrated with the limits for t and cPi: [0, t] and [0, refers to the substituted group, and j and k to the unsubstituted ones.
cPi]. The resulting formula reads The probability of a disubstituted unit is (cPi/c0)(cPj/c0)(1cPk/c0) and of
a trisubstituted unit (cPi/c0)(cPj/c0)(cPk/c0).
c0 cPi
ln ki t 27 Application of the reasoning to the mono (I)-, di (II)- and
c0
trisubstituted (III) units gives the relations below, respectively,
from which the dimensionless product concentration can be solved cI =c0 cP2 =c0 1cP3 =c0 1cP6 =c0 1cP2 =c0 cP3 =c0 1cP6 =c0
cPi =c0 1ekit 28 1cP2 =c0 1cP3 =c0 cP6 =c0 34

Fig. 2. Reaction scheme for the formation of substituted anhydroglucose units. Notation: 0, unsubstituted glucose unit; 2, 3, 6, monosubstituted glucose units (substituted at
hydroxyl groups 2, 3 and 6); 23, 36, 26, disubstituted glucose units (substituted at hydroxyl groups 2 and 3, 3 and 6 and 2 and 6); 236, trisubstituted glucose units.
 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182 175

cII =c0 cP2 =c0 cP3 =c0 1cP6 =c0 cP2 =c0 1cP3 =c0 cP6 =c0 r36 k3 cu6 k6 cu3 k2 cu36 cRa 48
1cP2 =c0 cP3 =c0 cP6 =c0 35
r236 k2 cu36 k3 cu26 k6 cu23 cRa 49

cIII =c0 cP2 =c0 cP3 =c0 cP6 =c0 36
The concentration of the unsubstituted units (c0 0) is given by
cu0 =c0 1cP2 =c0 1cP3 =c0 1cP6 =c0 37
The degree of substitution can now be obtained as a weighted
average from the fractions of mono-, di- and trisubstituted units
DS cI =c0 2cII =c0 3cIII =c0 : 38
3.2. DS for equal reactivities
An interesting special case is obtained, if all of the hydroxyl
groups are equally reactive. In this case, cP2/c0 cP3/c0 cP6/c0 x.
The concentrations of mono-, di- and trisubstituted units and
unsubstituted units are obtained from Eqs. (34)(37)
2
cI =c0 3x1x
cII =c0 3x2 1x
cIII =c0 x3
cu0 =c0 1x3 39a2d
The sum of cI/c0 cII/c0, cIII/c0 and c0 0/c0 is 1. The denition of DS,
Eq. (38) gives in this case combined with Eq. (39):
DS 3x(1 x)2 + 6x2(1  x)+3x3 3x. De facto, xDS/3 and Eq. (39)
can be rewritten to cI/c0 3(DS/3)(1  DS/3)2, etc. These relations
can be generalized to
cn =c0 gn DS=3n 1DS=33n 40
where n 0, I, II, III. The coefcients gn are obtained from the third
row of Pascals triangle, gn (3n)3!/n!/(3 n)!, n 0, 1, 2, 3.
Eq. (40) represents the classical distribution of Spurlin (1939),
which has been used throughout the years by numerous authors to
analyze the results of substitution experiments (e.g. Saake et al.
2001). The reality is, however, in most cases very different: the
substitution activities are typically very reagent-dependent and
rather unequal for the hydroxyl groups (OH-2, OH-3 and OH-6).
Thus, a general simulation approach is needed.
3.3. Calculation of detailed product distribution: general case and rst
order kinetics
The consumption rate of the substitution reagent is obtained from
rR k2 k3 k6 cu0 k3 k6 cu2 k2 k6 cu3 k2 k3 cu6
k2 cu36 k3 cu26 k6 cu23 cRa 50
If the substitution reagent (R) undergoes side reactions, their effects
are added to the expression of rR. Typical side reactions are a
substitution of hydroxyl groups and elimination reactions in an alkaline
environment, as discussed by Salmi et al. (2003). After solving the mass
balance Eq. (41), the fractions of mono-, di- and trisubstituted units are
obtained from Eqs. (31) to (33) and DS from Eq. (38).
Furthermore, the reactivity of a hydroxyl group diminishes due
to the substitution to another group in the same anhydroglucose
unit, the effect can be taken into account in the rate constant (Salmi
et al., 1994). The simplest way is to introduce a function
kj k0j ed0 DS 51
which predicts the decline of the reactivity as the substitution
progresses. Alternatively, the individual concentrations of the
substituted hydroxyl groups can be introduced in the rate constants.
For example, if the reactivity of OH-3 is diminished by the substitution
to OH-2, we get k3 k03 exp( d2cP2/c0). This approach can be
generalized to cover the contributions of all groups to the retardation
P
kj k0j e dk cPk =c0 52
where cPk denotes the substituted group concentration.
3.3.1. First order kinetics and equally reactive hydroxyl groups
If the hydroxyl groups are equally reactive and (pseudo-)rst
order kinetics is applicable, a remarkable simplication is possible:
k2 k3 k6 k and the balances can be added together to get the
concentrations of mono-, di- and trisubstituted units
dc2 =dt dc3 =dt dc6 =dt dcI =dt 3kc0 2kcI 53
dc23 =dt dc26 =dt dc36 =dt dcII =dt 2kcI kcII 54
dc236 =dt dcIII =dt kcII 55
dc0 =dt 3kc0 56
A Damkohler number, y kt is introduced and the above
differential equations are transformed to
dcu0 =dy 3cu0 57

dcI =dy 3cu0 2cI 58

The detailed product distribution in the time scale can be simulated
by using the mass balance Eq. (41) and the reaction scheme displayed dcII =dy 2cI cII 59
by Fig. 2. In the actual case, the balance equations of the components in
a constant-volume batch reactor are written as dcIII =dy cII 60
dcui =dt ri 41
The solution of this differential equation system is straightfor-
where i0, 2, 3, 6, 23, y, 236, R. The generation rates of different
ward. The rst differential equation is solved by separation of the
anhydroglucose units are listed below
variables and with the initial condition y 0, c0 c0. The other
r0 k2 k3 k6 cu0 cRa 42 differential equations are of the type y0 + f(y)yg(y), which has a
standard solution. The initial condition is: y 0, cI cII cIII 0.
r2 k2 cu0 k3 k6 cu2 cRa 43 The solution is briey summarized below

r3 k3 cu0 k2 k6 cu3 cRa 44 cu0 =c0 e3y 61

r6 k6 cu0 k2 k3 cu6 cRa 45 cuI =c0 31ey e2y 62

r23 k2 cu3 k3 cu2 k6 cu23 cRa 46 cuII =c0 312ey e2y ey 31ey 2 ey 63

r26 k2 cu6 k6 cu2 k3 cu26 cRa 47 cuIII =c0 13ey 3e2y e3y 1ey 3 64
176 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182

The degree of substitution is dened by Eq. (38), which gives It should be pointed out that 1 a36 a2 and the solution takes
nally the very simple form
DS 31ey 65 cu2 =cu0 cu0 =c0 a2 1 72
It can be shown that Eqs. (61)(65) are consistent with the which leads to
previously obtained production distribution for equally reactive
hydroxyl groups. By denoting e  y y, Eqs. (61)(65) can be cu2 =cu0 cu0 =c0 a2 1cu0 =c0 73
rewritten as
cu0 =c0 y3 The simplied notation X c0 0/c0 is introduced, after which the
solution is written in a compact form. All of the monosubstituted
cuI =c0 31yy2
units behave in a similar manner. The solution can be generalized
cuII =c0 31y2 y into a set of relations
cuIII =c0 1y3 66a2d
cu2 =c0 X a2 1X 74

DS 31y 67 cu3 =c0 X a3 1X 75

cu6 =c0 X a6 1X 76

A comparison of Eqs. (39) and (67) reveals that x1  y in the
actual case. By differentiating the expressions with respect to x where ai ki/Skj.
( DS/3) and setting the derivatives to equal to zero, the minima An analogous, but mathematically slightly more complicated
and maxima of cI and cII can be found: cI has a maximum at x 1/3, treatment can be applied to the di-substituted units. For the di-
DS1, cII has a maximum at x 2/3, DS 2, while cIII increases substituted unit (23), the following expression can be obtained
monotonously as a function of DS. For unequal k-values, the after dividing the balance Eq. (46) by Eq. (42)
maxima of cI and cII are shifted from these ideal values, as will dcu23 k2 cu3 k3 cu2 k6 cu23
be shown by numerical simulations. 77
dcu0 k2 k3 k6 cu0
It should be noticed that the product distribution in the time
scale was determined for rst order kinetics only. In a relative which is rewritten as
sense, the distribution is, however, valid for arbitrary kinetics with dcu23 =c0
a2 c03 =cu0 a3 cu2 =cu0 a6 cu23 =cu0 78
respect to the substitution reagent (a a0), provided that the dcu0 =c0
reaction orders coincide for all of the substitutions. Eqs. (57)
(60) then have an additional factor in the right-hand side, namely
The following substitutions are introduced: X c0 0/c0 and
caR. If the balances for cI, cII and cIII are divided by the rst balance
yc0 23/c0. After some rearrangement, the differential Eq. (78)
(c0), the reagent concentration disappears and the product
becomes
distribution is obtained by solving the differential equations
with respect to the dimensionless concentrations (cI/c0, cII/c0 and dy=dXa6 =Xy a2 X a3 1a3 X a2 1 79
cIII/c0), nally giving the product distribution according to Eq. (40). 0
which is of the type y + f(X)yg(X). The differential equation has a
standard Rsolution R
R
3.3.2. Relative product distributiongeneral approach y e  fdX(C + e fdX gdX). Alter performing the integrations, we
The relative product distribution, i.e. the distribution of the get the solution of Eq. (79)
mono, di- and tri-substituted units related to the concentration of
y CX a6 X 1a2 X 1a3 X 80
the unreacted units can be derived by dividing the balance
equations of the substituted unit by the balance equation of the
unreacted unit. For instance, for the mono-substituted unit (2), we The integration constant (C) is obtained from the initial condi-
get after division of Eq. (43) by Eq. (42) tion t 0, X 1 and y 0 (no substituted units present in the
beginning of the reaction). This gives C 1.
dcu2 k2 k3 k6 cu2 =cu0
68 All of the disubstituted units behave analogously and the
dcu0 k2 k3 k6
solution can be generalized as follows

The following notation is introduced: a2 k2/(k2 + k3 + k6) and cu23 =c0 X X a6 X 1a2 X 1a3 81
a36 (k3 + k6)/(k2 + k3 + k6). The balance equation now becomes
cu26 =c0 X X a3 X 1a2 X 1a6 82
dcu2
a2 a36 cu2 =cu0 69
dcu0 cu36 =c0 X X a2 X 1a3 X 1a6 83

The substitution yc0 2/c0 0 is introduced, which gives dc0 2/ Now, the total number of mono- and di-substituted units can be
dc0 0 (dy/dc0 0)c0 0 +y and the differential Eq. (69) is easily solved calculated simply by adding Eqs. (7476) and (8183), respectively
by separation of variables and integration
Z Z cI =c0 X 1a2 X 1a3 X 1a6 3X 84
dy dcu0
70
a2 a36 1y cu0 cII =c0 3X X a2 X a3 X a6 2X 1a2 X 1a3 X 1a6 85

After performing the integration and inserting the limits [0, c0 2/ The sum of unsubstituted (X), mono- (cI), di- (cII) and tri-
c 0] and [c0, c0 0], the result becomes
0
substituted (cIII c236/c0) units is 1; thus, the relative concentration
of tri-substituted units is obtained from the total balance
     a361
1a36 cu2 cu0
ln 1 ln 71
a2 cu0 c0 cIII =c0 1XX a2 X a3 X a6 X 1a2 X 1a3 X 1a6 86
 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182
The degree of substitution is expressed as the weighted sum of
the mono-, di- and tri-substituted units (n I, II, III)
X f u 121ai X ai 0
DS ncn =c0 87
The addition of Eqs. (8486) gives
DS 3X a2 X a3 X a6 88
In the beginning of the reaction X1 and DS0, after an innitely
long reaction time, X0 and DS approaches 3. An interesting special
case is valid for hydroxyl groups with equal reactivities:
a2 a3 a6 1/3, which gives DS3(1 X1/3), i.e. Xc0 0/c0 (1
DS/3)3, which coincides with the previously presented result,
Eq. (40) (Spurlin distribution). On the other hand, if only one of the
hydroxyl groups (i) is reactive, whereas the other ones are completely
nonreactive (aj ak 0), Eq. (88) gives DS1 Xaj.
3.3.2.1. Concentration maxima for various cases. For the mono- and
di-substituted units, concentration maxima are expected for cer-
tain values of X and DS. The maxima can be revealed by classical
analysis. Differentiating the expression for the mono-substituted
unit, for instance, Eq. (74) and setting the derivative to zero gives
the maximum at
1=ai
Xmax 1ai
where i 2,3 or 6. The maximum of the mono-substituted unit is
cui =c0 1ai 1 11ai 1=ai 90
Some special cases are of interest. If all the rate constants are equal,
aj 1/3 and Xmax (2/3)3. Eq. (88) then gives DS3(1 X1/3)1; the
maximum of the mono-substituted unit is obtained at DS1 and its
numerical value is calculated from Eq. (90). We get c0 i/c0 (3/21)
(2/3)3 4/270.148 and the total concentration of mono-substituted
units is cI/c0 0 4/90.444 at the maximum. In case that one of the
hydroxyl groups is dominating in the reactivity, ai-1, X-0 and DS-1
according to Eq. (88): DS-3 (Xai +1+1)1Xai.. For small values of
ai, the expression (89), (1 ai)1/ai-e  1 0.368, which implies that the
mono-substituted units have a maximum at Xe  1 or at higher
X-values always.
Analogous reasoning can be applied to the di-substituted units.
For instance, for a di-substituted unit (ij), we have
f cuij =c0 X X ak X 1ai X 1aj
Differentiation gives
f u 1 ak X ak1 1ai X ai 1aj X aj 0 92
from which Xmax can be solved iteratively for known values of a2, a3
and a6. An interesting special case is that the a-constants are equal
( 1/3). Eq. (92) now gives 1+(1/3)X  2/3  (2/3)(X  1/3 +X  1/3)0. This
is in fact a second-degree equation with respect to X  1/3 having the
roots X1/3 1 and X1/3 3. The latter root is relevant for the
concentration maximum giving Xmax 1/27. This is inserted in the
expression of DS, Eq. (88) giving DS2, i.e. the maximum of di-
substituted units is obtained at DS2 for the case of equal rate
constants. The relative concentration of each di-substituted unit is then
c0 ij/c0 4/27 and the total concentration of di-substituted units is
cII/c0 4/9 at the maximum.
For situations appearing in practice, another special case is,
however, even more interesting. If two of the three hydroxyl groups
(e.g. OH-2 and OH-6) are much more reactive than the third one
(OH-3), we have for the di-substituted unit (ij)
f cuij =c0 X X ak 2X 1ai
177
Differentiation of Eq. (93) and recalling that ak 0 gives
94
as the necessary condition for a concentration maximum. After
setting ai aj 1/2, Eq. (94) becomes 1 X  1/2 0, which has the
solution X1. As X diminishes from 1 towards 0 during the
reaction, the function f 1+ X 2x1/2 (Eq. (93)) increases from 0
to 1, all units become disubstituted after an innitely long time. The
degree of substitution is calculated from Eq. (88): DS2(1 X1/2),
DS increases monotonously from 0 towards 2 as the reaction
progresses.
3.3.2.2. Special case: rst and pseudo-rst order kinetics. For rst and
pseudo-rst order kinetics, the differential Eq. (42) can be solved
analytically (a 0) giving the exponential function
cu0 =c0 X ek2 k3 k6t 95
After this, the concentrations of mono-, di- and tri-substituted
units are calculated from Eqs. (7476), (8183) and (86). After
inserting the solution (95) in the denition of DS, Eq. (88), and
remembering the denitions of a2, a3 and a6, we get
DS 3ek2t ek3t ek6t 96
89 for rst and pseudo-rst order kinetics. The result coincides with
Eq. (29) as it should.
3.4. General numerical approach and determination of kinetic
parameters
The general approach is based on the numerical solution of the
differential Eq. (41) for each component having the generation
rates given by Eqs. (42)(50). In a mathematical sense, this case
represents a system of ordinary differential equations (ODEs), an
initial value problem. In recent years, a lot of numerical methods
and computer codes have been developed for the solution of this
kind of problems, particularly for the case that the system is stiff,
i.e. that the values of the rate constants are very different. The
methods applied most often to stiff ODEs are semi-implicit Runge
Kutta methods (SIRK) (Kaps and Wanner, 1981) and backward
difference (BD) methods (Gear, 1971, Hindmarsh, 1983). These
methods are even implemented in high-level programming
language systems, such as Matlab and Mathematica. They are
91
also included in the simulation and parameter estimation software,
such as Modest (Haario, 2007).
In our case, the model was implemented in the form
dy=dt f y 97
in the computer. The dependent variables were y[c0 0, c0 2, y,
c0 236, cR]T and the right-hand sides of Eq. (97) were represented by
the corresponding generation rates (4250). The stiff ODE-solver
based on the BD method was used in the Matlab application. The
estimation of the rate parameters was based on the kinetic data
obtained with High Performance Liquid Chromatography (HPLC)
(see Section 4.2).
The differences between the observed anhydroglucose units
(c0 0, c0 2, y, c0 236) and those predicted by the model (Eqs. 4150)
was minimized by nonlinear regression. The various concentra-
tions recorded at different times were weighted equally. The
objective function to be minimized assumed the form
XX
Q yi, t, exp yi, t 2 98
i t
where yit,exp and yit denote the experimentally recorded and
93 modeled concentrations, respectively. The best values of the rate
178 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182
constants (k2, k3, k6) were obtained through a minimization of the
objective function with respect to these parameters.
The objective function was minimized with a combined sim-
plex-LevenbergMarquardt algorithm (Haario, 2007); in the
beginning of the procedure, the slow, but more robust simplex
algorithm was used, but as the minimum was approached, the
more rapid LevenbergMarquardt algorithm was switched on. The
concentration values predicted by the model were obtained by
solving the differential Eq. (97) numerically as described
previously. All calculations were performed on a PC. In general,
the simulations and parameter estimations run smoothly, without
any particular problems.
The quality of the model t was examined with the following
approaches: the standard errors of the parameters, the sensitivities
of the parameters, the correlation matrix of the parameters as well
as the overall degree of explanation (R2). The overall degree of
explanation is dened as
PP
yi,t,exp yi,t 2
R2 1 P P 99
yi,t,exp yi,aver 2
R2 compares the t of the model to a case that the experimental
data are described with the average values only. Consequently, a
high degree of explanation (close to 1) is required from a
deterministic kinetic model in order to be regarded as relevant.
The sensitivities of the parameters were checked by preparing the
sensitivity plots: the value of one kinetic parameter was changed
and the objective function (Q) was calculated, while the values of
the other parameters were kept constant (at the level of giving the
minimum of Q) during this procedure. In this way, it was possible to
check, how sharp the minimum of the objective function was with
respect to the kinetic parameter. This gave the identiabilities of
the individual rate constants.
4. Experimental section
4.1. Experimental conditions and materials
The commercial cellulose from birch hardwood was obtained
from Metsa-Serla chemicals; the cellulose obtained was screened
to a size of 0.35 mm. In the heterogeneous carboxymethylation,
experiments were performed in a glass laboratory-scale autoclave
(500 ml) with xed molar ratios of cellulose-to-monochloroacetic
acid-to-NaOH 1:4:8 according to experimental procedures pub-
lished previously (Salmi et al. 1994). 2-propanol was used as the
solvent and the reaction temperatures during carboxymethylation
experiments were 30, 40 and 60 1C. The reaction time was 120 min
and samples were withdrawn from the reaction mixture at pre-
dened intervals (10, 20, 40 min, etc.).
4.2. Chemical analysis: high-pH anion-exchange chromatography
with pulsed amperometric detection (HPAEC-PAD)
4.2.1. Hydrolysis
Before the HPAEC-PAD analysis, the CMC samples were sub-
jected to acid hydrolysis. In the hydrolysis with triuoroacetic acid
(TFA), 50 mg of CMC was weighed in pear-shaped asks and mixed
with 2 ml of 2 M TFA. The mixture was kept in an oven at 100 1C for
2 h. After cooling, TFA was evaporated with nitrogen gas keeping
the sample in a water bath at 4050 1C.
4.2.2. Instrumentation
Consequently, the hydrolysis products were analyzed by HPAEC
with PAD on a Hewlett Packard 1100 Series LC system with a
Dionex ED50 electrochemical detector, a Dionex CarboPac PA1
anion-exchange column (semi-preparative 9  250 mm and ana-
lytical 4  250 mm combined with guard 4  50 mm) and a Chem-
station chromatographic software. The column was maintained
under a controlled temperature of 30 1C. The separation, using
the analytical column, was performed with a linear gradient of 95:5
(v/v) of eluent A (0.1 M NaOH)/eluent B (0.1 M NaOH containing
1 M NaOAc) to 100% eluent B in 12 min followed by 100% B from 12
to 15 min. Before next analysis, the column was reconditioned for
7 min with the 95:5% mixture of eluent A/B. The ow rate selected
was 1 ml/min. The detection was performed by the PAD with a gold
working electrode and triple pulse amperometry by an application
of the following pulse potentials and durations: EDET 0.05 V
(200 ms), EOX 0.75 V (190 ms) and ERED  0.15 V (380 ms)
with a total response time of 1 s for every pulse. Such a repeating
step of surface oxidation and reduction is necessary to maintain a
highly reproducible state of activity of the gold electrode surface.
An Ag/AgCl combination electrode and the titanium cell body of the
PAD detector was used as reference and counter electrode,
respectively.
4.2.3. Standard preparation
In the preparation of standards, one CMC sample with DS 1.1
was hydrolyzed in 0.5 g scale in TFA, in order to get a sufcient
quantity of mono-, di- and tri-substituted CMG. After the separa-
tion, by using the semi-preparative anion-exchange column, frac-
tions were collected, and concentrated by freeze-drying,
neutralized with 2 M acetic acid and desalted on a column
(430  15 mm) lled with Sephadex G-10 (Sigma-Aldrich). The
sugar-containing fractions were detected using a CDM210 Con-
TM
ductivity Meter as detector (MeterLab ), reconcentrated by freeze-
drying and silylated before analysis with GCMS. The response
factors obtained of the substituted units (2, 3, 6, 23, 26, 36, 236)
were very similar to those reported by Kragten et al. (1992). The
details of the analysis method are reported in a previous paper
(Damlin et al. 2010).
5. Numerical simulation results and comparison with
experimental data
5.1. Theoretical simulations
Some numerical simulations of the product distribution are
displayed in Fig. 3. The simulations were carried out for three
principally different cases: all hydroxyl groups have equal
reactivities, one of them has a superior reactivity compared to
the other ones, and, nally, two of the groups are clearly more
reactive than the third one. No retardation effect was included in
the model at this stage. The case that one of the groups (e.g. OH-6) is
more reactive than the other ones correspond to a case that a bulky
substituent is able to react with one OH-group only, and there is a
strong steric hindrance for the substitution reagent to interact with
the other groups. The simulation results in Fig. 3 are presented as a
function of DS, which implies that curves are independent of the
reaction order of the substitution reagent. Fig. 3 is thus valid both
for SN1 and SN2 mechanisms.
A general observation is that prominent concentration maxima
appear at DS 1 and 2 or in the vicinity of these values. The
numerical simulations show coinciding with the theoretical
considerations that xI,max xII,max 4/9, if all hydroxyl groups
are equally reactive (Section 3.3.2).
For the case of unequal reactivities, one of the maxima (either
mono- or di-substituted units, depending on the reactivity ratio)
shifts towards both higher degrees of substitution and mole
fractions, whereas the other maximum only increases in mole
 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182 179

1 1.5
unsubstituted trisubstituted
glucose, glucose,
x0 xIII
0.8 k2:k3:k6 = 1:1:1

Degree of substitution, DS
1
Mole fraction

monosubstituted disubstituted
0.6 glucose, glucose,
xI xII

0.4
0.5

0.2

0 0
0 20 40 60 80 100
0 0.5 1 1.5 2 2.5 3
Degree of substitution, DS Time / [min]
1 Fig. 4. Measured (symbols) and modeled (lines) degrees of substitution (DS) in
1:10:10 k2:k3:k6 1:1:10 carboxymethylation of cellulose. Experimental conditions: c0,cellulose 0.133 mol/l,
c0,acid 0.513 mol/l; m, T 30 1C; , T 40 1C and K, T60 1C.

0.8
1:10:10
30, 40 and 60 1C in the batch reactor. The overall degree of
1:1:10 1:1:10 substitution (DS) and the substitution rate increased as the
1:10:10
Mole fraction

0.6 temperature was augmented, which is clearly revealed by Fig. 4.

1:10:10 1:1:10
0.4
0.2
0 hydroxyl group OH-6 is the most reactive one: the mono-
0 0.5 1 1.5 2 2.5
Degree of substitution, DS
Fig. 3. Simulated mole fractions of substituents per anhydroglucose unit: (a)
distribution of unsubstituted, mono-, di- and tri-substituted glucose units, if all
three hydroxyl groups are equally reactive and (b) distribution of the substituents as
the ratios of the rate constants are k2:k3:k6 1:1:10 and k2:k3:k6 1:10:10.
fraction. If one of the hydroxyl groups is much more reactive than
the other ones, the maximum mole fraction of the mono-sub-
stituted unit in the vicinity of DS1 increases as the rate constant
of the active hydroxyl group increases. The ultimate case is
ki:kj:kk 0:0:very large, for which the mole fraction of the mono-
substituted unit increases monotonously as a function of DS until
DS 1 is reached; at DS1 the mole fraction becomes 1. For values
of rate constants deviating from zero, a maximum is attained for
the mono-substituted units.
In case that two of the hydroxyl groups are much more active
than the third one, the maximum of the di-substituted unit is
higher than that of the mono-substituted unit. In the ultimate case,
ki:kj:kk 0:very large:very large, the mole fraction of the di-
substituted units increases monotonously as DS 2 is approached.
As the numerical values of the two constants of the active groups
increase, the maximum of the disubstituted units increases (Fig. 3).
5.2. Modeling of experimental data
The experimental data are displayed in Figs. 46: the carbox-
ymethylation kinetics of cellulose in 2-propanol was recorded at
The concentrations of unsubstituted, mono-substituted (I), di-
substituted (II) and tri-substituted (III) units are displayed in
Fig. 5. At 30 and 40 1C, the reaction rate was rather slow, and the
concentrations of all of the substituted units (DS) still increased. At
60 1C, however, the rates were clearly higher, and a maximum in
the concentration of the mono-substituted units becomes visible.
The concentration of the tri-substituted units is very low at 30 and
40 1C, while it becomes clearly visible at 60 1C. Fig. 6 reveals that the
3 substituted unit (6) is formed more rapidly, but declining more
slowly than the other mono-substituted units (2 and 3) (Fig. 6a).
A characteristic feature obtained throughout the experimental
domain is that the reaction rate stagnates and a complete
substitution seems to be in practice impossible. In general, the
overall precision of the data is rather good.
In the work of Reuben and Conner (1983), the relative
reactivities of the hydroxyl groups were discussed and literature
data were compared. The general trend was that OH-3 was
regarded to be less reactive than OH-2 and OH-6; concerning the
reactivity of OH-2 and OH-6, no general agreement exists. In the
work of Timell and Spurlin (1952), it was stated that OH-2 and
OH-3 have almost the same reactivity; in the work of Niemela and
Sjostrom (1989), the reactivities of OH-2 and OH-6 claimed to be
very similar. In many cases, the analysis has been based on the end
product only. Factors such as temperature, concentration level and
solvent vary in the works reported. However, if the relative
reactivities of the hydroxyl groups change during the reaction, it
can confuse the interpretation of the experimental data.
A logarithmic plot corresponding to the Spurlin model assuming
constant relative reactivities of the hydroxyl groups was prepared
(  ln(1 ci/c0)  ailn(c0 0/c0)) for our data. The plots showed a
clear deviation from linearity, thus indicating that the simple
model cannot be used in the present case.
Preliminary parameter estimation efforts revealed two impor-
tant features: the reactivities of the hydroxyl groups OH-2 and
OH-3 are rather close and the rate retardation process is compli-
cated, depending both on the overall DS and the specic effect
according to Eq. (51) and (52) (see Fig. 2). It was impossible to
distinguish in practice between the relative reactivities of the
180 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182

1 0.25

0.8 0.2
Mole fraction

Mole fraction
0.6 0.15

0.4 0.1

0.2 0.05

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100
Time / [min] Time / [min]

1 0.14

0.12
0.8
0.1
Mole fraction
Mole fraction

0.6 0.08

0.06
0.4
0.04

0.2
0.02

0
0
0 20 40 60 80 100
0 20 40 60 80 100
Time / [min] Time / [min]

0.1
1

0.08
0.8
Mole fraction

0.06
Mole fraction

0.6

0.04
0.4

0.02
0.2

0
0 0 20 40 60 80 100
0 20 40 60 80 100 Time / [min]
Time / [min]
Fig. 6. Measured (symbols) and modeled (lines) mole fractions in carboxymethyla-
Fig. 5. Measured (symbols) and modeled (lines) mole fractions in carboxymethyla- tion of cellulose. Experimental conditions: c0,cellulose 0.133 mol/l, c0,acid 0.513
tion of cellulose: K, unsubstituted glucose units; , mono-substituted; m, di- mol/l; (a) K, x2; m, x6; , x3; (b) K, x23; m, x36; , x26 and (c) K, x236. Temperature:
substituted; ., tri-substituted; experimental conditions: c0,cellulose 0.133 mol/l, 60 1C.
c0,acid 0.513 mol/l; (a) T30 1C; (b) T 40 1C and (c) T 60 1C.

groups OH-2 and OH-3; therefore, the pre-exponential factors and non-linear regression according to the pattern displayed in Fig. 2.
activation energies for the reactivities of these groups were set The regression analysis revealed that the hydroxyl group OH-6 has
equal. The specic retardation parameters were determined by the highest retardation effect: evidently the substituted group in
 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182 181

Table 1 algorithms were proposed and the software was developed to

Estimated kinetic parameters. simulate the detailed substitution kinetics in batch reactors. The
model was used to describe the carboxymethylation of cellulose in
Parameter Parameter unit Estimated parameter Estimated relative
value standard error (%) 2-propanol. New, detailed data on the substitution of cellulose with
monochloroacetic acid were presented. The data revealed the time
k2,ref k3,ref l/(mol min) 0.238 46.4 dependence of the substituted units (2, 3, 6, 23, 26, 36, 236; Fig. 2). The
k6,ref l/(mol min) 0.326 46.5 experimental data and the model suggested that the hydroxyl group
d2 2.03 52.3
4.42 22.6
OH-6 is the most reactive one in the carboxymethylation of cellulose,
d3
d6 6.46 23.5 whereas the relative reactivities of the hydroxyl groups OH-2 and
Ea,2 Ea,3 kJ/mol 140 11.5 OH-3 were found to be lower and practically equal (Table 1). The
Ea,6 kJ/mol 127 12.5 reaction rate decreased with time, probably due to an increasing steric
d0 4.79 14.8
hindrance, as the substitution reaction proceeds. This was accounted
P
0
where kj kj, ref eEa, j =Ry ed0 DS e dk cPk =c0 For the retardation factors, see Fig. 2. The
for by including a retardation factor in the rate constant. In
reference temperature (Tref) is the average temperature of the experiments (in K). general, the description of the experimental data was good
The transformed temperature (y0 ) is dened as 1/y0 1/T  1/Tref (in K). (Figs. 46), particularly considering the physical and chemical com-
plexities of the experimental system.

position 6 can by rotation shield the substitution to OH-groups 2

and 3. Furthermore, this is conrmed by Figs. 6a and b: once the
Nomenclature
substituted OH-6 has been formed, it reacts further rather slowly to
the di-substituted units (26 and 36); the disubstituted unit (23)
is formed most rapidly of the di-substituted units, because the a concentration ratio, Eqs. (17) and (18)
shielding effect of substituted OH-6 is not present. To describe the C integration constant
overall retardation effect, the approach given by Eq. (51) is c concentration
necessary. c0 concentration of a substituted unit
A summary of the parameter estimation statistics is provided by c0 initial concentration of a hydroxyl group
Table 1. As revealed by the table, the overall estimation statistics is c0 concentration of an unsubstituted unit
fairly good, taking into consideration the complexity of the system. DS degree of substitution
The overall degree of explanation (R2, Eq. (99)) exceeded 95%. The Ea activation energy
estimated activation energies of the kinetic constants are f general function
reasonable, around 130 kJ/mol for OH-6 and around 140 kJ/mol f0 derivative
for OH-2 and OH-3. The parameters listed in Table 1 represent the g general function in the solution of differential equations
minimum set of parameters, which is able to describe the behavior k rate constant
of the system. k0 transformed rate constant, k0 kc0
A graphical overview of the model t can be seen in Fig. 4 (DS), 5 Q objective function in nonlinear regression
(unsubstituted, mono- di- and tri-substituted units) and 6 (all R general gas constant (8.3143 J) (mol K)
substituted units separately). The overall DS and the concentrations R2 degree of explanation
of the mono-, di- and tri-substituted units are well predicted as r rate
revealed by Figs. 4 and 5. The concentrations of the mono- (2, 3, 6) t time
and tri-substituted units (236) are very well predicted as X dimensionless concentration of unsubstituted units,
demonstrated by Figs. 6a and c, while the concentrations of the di- X c0 0/c0
substituted units (23, 26, 36) are slightly overestimated by the x dimensionless concentration of a substituted hydroxyl
model according to Fig. 6b. It is difcult to speculate for the reason of group ( cPi/c0)
this deviation; it might depend on the calibration of the chemical y general variable and dependent variable
analysis. However, the general trends of the di-substituted units are a reaction order of the substitution reagent
rather well predicted. It can be concluded that the modeling approach ai fractional rate parameter (i2, 3, 6) of a hydroxyl group
was successful, taking the complexity of the system into account. or mono-substituted unit, ai ki/Skk
Additional experimental work, particularly in completely aij fractional rate parameter (ij 23, 26, 36) of a di-substi-
homogeneous systems, such as substitution of dissolved starch tuted unit, aij (ki + kj)/Skk
or cellulose in solvents such as ionic liquids, will be carried out in ai/j ratio between rate constants, ai/j ki/kj
future, and this is expected to further elucidate the modeling g coefcient in Spurlin distribution of anhydroglucose units
approach: in a homogeneous system, the mass transfer effects are d rate decline parameter, Eq. (52)
completely excluded and the intrinsic kinetics is more easily y Damkohler number, y kt
approached. In the present system, the treatment of the slurry y0 transformed temperature (Table 1)
complicated the chemical analysis and impaired its accuracy.
Subscripts and superscripts

6. Conclusions aver average value

exp experimental data
A detailed mathematical analysis of cellulose and starch substitu- i general index
tion kinetics was provided. The analysis revealed the product i, j, k indices for hydroxyl groups OH-2, OH-3 and OH-6
distributions and concentration maxima of mono-, di- and tri- max maximum
substituted anhydroglucose units in cellulose and starch. The relative n index of mono-, di- and tri-substituted units
product distributions (as a function of DS) are valid for both SN1 and P product, substituted group
SN2 mechanisms. Analytical solutions of time-dependent substitution R substitution reagent; alkyl chain
processes were derived for (pseudo)-rst order kinetics. Numerical ref reference temperature
182 T. Salmi et al. / Chemical Engineering Science 66 (2011) 171182

t time Heinze, T., Schwikal, K., Barthel, S., 2005. Ionic liquids as reaction medium in
0 initial value or unsubstituted unit cellulose functionalization. Macromol. Biosci. 5, 520525.
Hindmarsh, A.C., ODEPACK, A., 1983. Systematized Collection of ODE-Solvers, in
I, II, III mono-, di- and tri-substituted unit Scientic Computing. In: Stepleman, R. (Ed.), IMACS. North Holland Publishing
Company 55-64.
Kaps, P., Wanner, G., 1981. A study of Rosenbrock-type methods of high order. Num.
Abbreviations
Math. 38, 279.
Kragten, E.A., Kamerling, J.P., Vliegenhart, J.F.G., 1992. Composition analysis of
BD backward difference method carboxymethylcellulose by high-pH anion-exchange chromatography with
pulsed amperometric detection. J. Chromatogr. 623, 4953.
Cell anhydroglucose unit of cellulose Lazik, W., Heinze, Th., Pfeiffer, K., Albrecht, G., Mischnick, P., 2002. Starch derivatives
HPLC high-performance liquid chromatography of a high degree of functionalization. VI. Multistep carboxymethylation. J. Appl.
ODE ordinary differential equation Polym. Sci. 86, 743752.
Mikkola, J.-P., Kirilin, A., Tuuf, J.-C., Pranovich, A., Holmbom, B., Kustov, L.M., Murzin,
OH hydroxyl group
D.Yu, Salmi, T., 2007. Ultrasound enhancement of cellulose processing in
SIRK semi-implicit RungeKutta method ionic liquids: from dissolution towards functionalization. Green Chem. 9,
12291237.
Niemela, K., Sjostrom, E., 1989. Characterization of hardwood-derived carboxy-
methylcellulose by gasliquid chromatography and mass spectrometry. Polym.
Commun. 30, 254256.
Acknowledgements Reuben, J., Conner, H.T., 1983. Analysis of the carbon-13 NMR spectrum of
hydrolyzed O-(carboxymethyl)cellulose: monomer composition and substitu-
tion patterns. Carbohydrate Research 115, 113.
This work is a part of the activities at the Abo Akademi Process Reuben, J., 1986. Analysis of the 13C NMR spectra of hydrolyzed and methanolyzed
Chemistry Centre (PCC) within the Finnish Centre of Excellence O-methylcelluloses: monomer compositions and models for their description.
Carbohydr. Res. 157, 201213.
Programmes (20002005 and 20062011) by the Academy of
Saake, B., Horner, S., Puls, J., Heinze, T., Koch, W., 2001. A new approach in the
Finland. Financial support from the Knut and Alice Wallenberg analysis of the substituent distribution of carboxymethylcelluloses. Cellulose 8,
Foundation and Kempe Foundation is gratefully acknowledged. 5967.
Salmi, T., Valtakari, D., Paatero, E., Holmbom, B., Sjoholm, R., 1994. Kinetic study of
the carboxymethylation of cellulose. Ind. Eng. Chem. Res. 33, 14541459.
References Salmi, T., Tirronen, E., Lehtonen, J., Paatero, E., Valtakari, D., 2003. Modeling of
complex organic solidliquid reaction systems in stirred tanks. Ind. Eng. Chem.
Res. 42, 25162524.
Damlin, P., Mikkola, J.-P., Salmi, T., 2010. Characterization of hardwood-derived Spurlin, H.M., 1939. Arrangement of substituents in cellulose derivates. J. Am. Chem.
carboxymethylcellulose by high-pH anion-exchange chromatography using Soc. 61, 22222227.
pulsed amperometric detection. Cellul. Chem. Technol. 44 (13), 6569. Sjostrom, E., 1993. Wood chemistry. Fundamentals and Applications. Academic
Flory, P., 1953. Principles of Polymer Chemistry. Cornell University Press, Ithaca,N.Y.. Press.
Gear, C.W., 1971. Numerical Initial Value Problems in Ordinary Differential Timell, T., 1950. Studies on cellulose reactions. Doctoral thesis. Kungliga tekniska
Equations. Prentice Hall, Englewood Cliffs, N.J.. hogskolan (KTH), Stockholm, Sweden.
Haario, H., 2007. MODESTUsers guide. Profmath Oy, Helsinki. Timell, T.E., Spurlin, H.M., 1952. Sven. Papperstidn. 55, 700708.