P1: GAE/LSK P2: FLV Final Pages Qu: 00, 00, 00, 00

Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes
Ronald W. Rousseau
Georgia Institute of Technology

I. Objectives of Crystallization Processes

II. Equilibrium and Mass and Energy Balances
III. Nucleation and Growth Kinetics
IV. Purity, Morphology, and Size Distributions
V. Crystallizer Configuration and Operation
VI. Population Balances and
Crystal Size Distributions

GLOSSARY tallizing species; includes both homogeneous and het-

erogeneous nucleation mechanisms.
Crystallizer The vessel or process unit in which crystal- Secondary nucleation The formation of crystals through
lization occurs. mechanisms involving existing crystals of the crystal-
Growth The increase in crystal size due to deposition of lizing species.
solute on crystal surfaces. Solubility The equilibrium solute concentration. The di-
Magma The mixture of crystals and mother liquor in the mensions in which solubility is expressed include, but
crystallizer. are not limited to, mass or mole fraction, mass or mole
Mode of crystallization The means by which a thermo- ratio of solute to solvent, and mass or moles of solute
dynamic driving force for crystallization is created. per unit volume of solvent or solution.
Mother liquor The liquid solution from which crystals Supersaturation The difference between existing and
are formed. equilibrium conditions; the quantity represents the
MSMPR crystallizer A vessel operating in a continuous driving force for crystal nucleation and growth.
manner in which crystallization occurs and whose con-
tents are perfectly mixed. As a result of perfect mixing,
all variables descriptive of the mother liquor and crys- CRYSTALLIZATION PROCESSES addressed in this
tals are constant throughout the vessel and are identical discussion are used in the chemical, petrochemical, phar-
to corresponding variables in the product stream leav- maceutical, food, metals, agricultural, electronics, and
ing the vessel. other industries. Moreover, the principles of crystalliza-
Nucleation The formation of new crystals. tion are important in all circumstances in which a solid
Primary nucleation The formation of crystals by mech- crystalline phase is produced from a fluid, even when the
anisms that do not involve existing crystals of the crys- solid is not a product of the process. Much has been done

91
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

92 Crystallization Processes

in recent years to improve the understanding of crystal- product that meets purity specifications. For example, in
lization, and a large portion of the research on the topic the manufacture of the amino acid L-isoleucine, the prod-
has dealt with mechanisms of nucleation and growth. Es- uct is first recovered in acid form, redissolved, neutralized,
pecially important has been elucidation of the effects of and then recrystallized in order to exclude the impurity
process variables on the rates at which these phenomena L-leucine and other amino acids from the product.
occur. Additionally, extensive progress has been achieved A simple change in physical properties also can be
in modeling both steady-state and dynamic behavior of achieved by crystallization. In the process of making soda
crystallization systems of industrial importance. The pri- ash, referred to earlier, the sodium bicarbonate crystals are
mary elements of the discussion that follows are the princi- subjected to heat that causes the release of carbon dioxide
ples that influence yield, morphology, and size distribution and produces low-density sodium carbonate crystals. The
of crystalline products. density of these crystals is incompatible with their use in
glass manufacture, but a more acceptable crystal can be
obtained by contacting the sodium carbonate crystals with
I. OBJECTIVES OF CRYSTALLIZATION water to form crystalline sodium carbonate monohydrate.
PROCESSES Drying the resulting crystals removes the water of hydra-
tion and produces a dense product that is acceptable for
Several examples of objectives that may be satisfied in glass manufacture.
crystallization processes are given in the following dis- Separation of a chemical species from a mixture of sim-
cussion. Soda ash (sodium carbonate) is recovered from ilar compounds can be achieved by crystallization. The
brine by contacting the brine with carbon dioxide that re- mode of crystallization may fall in the realm of what
acts with sodium carbonate to form sodium bicarbonate. is known as melt crystallization. In such processes, the
Sodium bicarbonate, which has a lower solubility than mother liquor largely is comprised of the melt of the crys-
sodium carbonate, crystallizes as it is formed. The pri- tallizing species, and, subsequent to its crystallization,
mary objective of the crystallizers used in this process crystals formed from the mother liquor are remelted to
is separation of a high percentage of sodium bicarbonate produce the product from the crystallizer. Para( p)-xylene
from the brine in a form that facilitates segregation of the can be crystallized from a mixture that includes ortho
crystals from the mother liquor. The economics of crystal and meta isomers in a vertical column that causes crys-
separation from the mother liquor are affected primarily tals and mother liquor to move countercurrently. Heat is
by the variables that control the flow of liquid through the added at the bottom of the column to melt the p-xylene
cake of crystals formed on a filter or in a centrifuge. For crystals; a portion of the melt is removed from the crys-
example, the flow rate of liquid through a filter cake de- tallizer as product and the remainder flows up the col-
pends on the available pressure drop across a filter, liquid umn to contact the downward-flowing crystals. Effluent
viscosity, and the size distribution of crystals collected on mother liquor, consisting almost entirely of the ortho and
the filter. With a fixed available pressure drop and defined meta isomers of xylene, is removed from the top of the
liquid properties, the crystal size distribution controls filter column.
throughput and, concomitantly, the production rate from Production of a consumer product in a form suitable
the process. for use and acceptable to the consumer also may be an ob-
Crystallization can be used to remove solvent from a jective of a crystallization process. For example, sucrose
liquid solution. For example, concentration of fruit juice (sugar) can be crystallized in various forms. However, dif-
requires the separation of solvent (water) from the natural ferent cultures are accustomed to using sugar that has a
juice. The common procedure is evaporation, but the de- particular appearance and, unless the commodity has that
rived juices may lose flavor components or undergo ther- appearance, the consumer may consider the sugar to be
mal degradation during the evaporative process. In freeze unacceptable.
concentration, the solvent is crystallized (frozen) in rela- In all these processes, there are commmon needs: to
tively pure form to leave behind a solution with a higher form crystals, to cause them to grow, and to separate the
solute concentration than the original mixture. Significant crystals from the residual liquid. While conceptually sim-
advantages in product taste have been observed in the ap- ple, the operation of a process that utilizes crystallization
plication of this process to concentrations of various types can be very complex. The reasons for such complexity in-
of fruit juice. volve the interaction of the common needs and process re-
The elimination of small amounts of an impurity from a quirements on product yield, purity, and, uniquely, crystal
product species may be an objective of crystallization. In morphology and size distribution. In the following discus-
such instances, a multistep crystallization–redissolution– sion, the interactions will be discussed and general princi-
recrystallization process may be required to produce a ples affecting crystallizer operation will be outlined. More
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 93

extensive discussion of the subject matter can be found in conditions and the limitations placed on the operation by
the bibliography at the end of the chapter. the system properties. The curves AB and BC represent
solution compositions that are in equilibrium with solids
whose compositions are given by the lines AD and CE,
II. EQUILIBRIUM AND MASS AND respectively. If AD and CE are vertical and coincident
ENERGY BALANCES with the left and right extremes, the crystals are pure S and
R, respectively. Crystallization from any solution whose
A. Solid–Liquid Equilibrium equilibrium composition is to the left of a vertical line
through point B will produce crystals of pure S, while
The solubility of a chemical species in a solvent refers
solutions with an equilibrium composition to the right of
to the amount of solute that can be dissolved at constant
the line will produce crystals of pure R. A solution whose
temperature, pressure, and solvent composition (including
composition falls on the line through B will produce a
the presence of other solutes). In other words, it is the
mixture of crystals of R and S.
concentration of the solute in the solvent at equilibrium.
Now suppose a saturated solution at temperature T1 is
As with all multiphase systems, the Gibbs phase rule
fed to a crystallizer operating at temperature T2 . Since it is
provides a useful tool for determining the number of inten-
saturated, the feed has a mole fraction of R equal to xF . The
sive variables (ones that do not depend on system mass)
maximum production rate of crystals occurs when the so-
that can be fixed independently:
lution leaving the crystallizer is saturated, meaning that the
NDF = Nc − Np + 2 (1) crystal production rate, m prod , depends on the value of T2 :
NDF is the number of degrees of freedom, Nc is the num- m prod = m F xF − m L xL (2)
ber of components, and Np is the number of phases in the where m F is the feed rate to the crystallizer and m L is the
system. The number of degrees of freedom represents the solution flow rate leaving the crystallizer. Note that the
number of independent variables that must be specified lower limit on T2 is given by the eutectic point, and that
in order to fix the condition of the system. For example, attempts to operate the crystallizer at a temperature other
the Gibbs phase rule specifies that a two-component, two- than the eutectic value will result in a mixture of crystals
phase system has two degrees of freedom. If temperature of R and S.
and pressure are selected as the specified variables, then When certain solutes crystallize from aqueous solu-
all other intensive variables—in particular, the composi- tions, the crystals are hydrated salts, which means that
tion of each of the two phases—are fixed, and solubility the crystals contain water and solute in a specific stoichio-
diagrams of the type shown for a hypothetical mixture of metric ratio. The water in such instances is referred to as
R and S in Fig. 1 can be constructed. water of hydration, and the number of water molecules
Several features of the hypothetical system described associated with each solute molecule may vary with the
in Fig. 1 illustrate the selection of crystallizer operating crystallization temperature.
Potassium sulfate provides an example of such behav-
ior. When it crystallizes from an aqueous solution above
40◦ C, the crystals are anhydrous K2 SO4 , while below
40◦ C each molecule of K2 SO4 that crystallizes has 10
molecules of water associated with it. The hydrated salt,
K2 SO4 ·10H2 O(s), is called potassium sulfate decahydrate.
Another solute that forms hydrated salts is magnesium
sulfate, which can incorporate differing amounts of water
depending upon the temperature at which crystallization
occurs (see Table I).
The solubility diagrams of several species are shown
in Fig. 2, and these illustrate the importance of solubility
behavior in the selection of the mode of crystallization.
For example, consider the differences between potassium
nitrate and sodium chloride: The solubility of potassium
nitrate is strongly influenced by the system temperature,
whereas the opposite is true for sodium chloride. As a
FIGURE 1 Hypothetical solubility diagram of eutectic-forming consequence, (1) a high yield of potassium nitrate crys-
system. tals can be obtained by cooling a saturated feed solution,
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

94 Crystallization Processes

TABLE I Water of Hydration for MgSO4

Form Name wt% MgSO4 Conditions

MgSO4 Anhydrous magnesium sulfate 0.0 >100◦ C

MgSO4 ·H2 O Magnesium sulfate monohydrate 87.0 67 to 100◦ C
MgSO4 ·6 H2 O Magnesium sulfate hexahydrate 52.7 48 to 67◦ C
MgSO4 ·7 H2 O Magnesium sulfate heptahydrate 48.8 2 to 48◦ C
MgSO4 ·12 H2 O Magnesium sulfate dodecahydrate 35.8 −4 to 2◦ C

but (2) cooling a saturated sodium chloride solution ac- sion of the complex solubility behavior in such systems
complishes little crystallization, and vaporization of water by graphical means usually is limited to systems of two
is required to increase the yield. solutes. The interaction of added solutes on solubility is
The effect of water of hydration on solubility can be seen illustrated by the plot of equilibrium behavior for potas-
in Fig. 2. Note, for example, that sodium sulfate has two sium nitrate–sodium nitrate–water in Fig. 3. As before,
forms in the temperature range of the solubility diagram: the curves in the diagram trace solution compositions that
sodium sulfate decahydrate (Na2 SO4 ·10H2 O), which is are in equilibrium with solid solutes. Points A, D, G, and
known as Glauber’s salt, and anhydrous sodium sulfate. J are based on the solubilities of pure potassium nitrate,
Since a transition from Glauber’s salt to anhydrous sodium while points C, F, I , and L are based on solubilities of
sulfate occurs at approximately 34◦ C, crystals recovered pure sodium nitrate. Curves AB, DE, GH, and JK repre-
from a crystallizer operating above about 34◦ C will be sent compositions of solutions in equilibrium with solid
anhydrous, but those from a crystallizer operating below potassium nitrate at 30, 50, 70, and 100◦ C, respectively.
this temperature will contain 10 waters of hydration. Also Curves BC, EF, HI, and KL represent compositions of so-
observe the effect of water of hydration on solubility char- lutions in equilibrium with solid sodium nitrate. Should
acteristics; clearly, cooling crystallization could be used to the solution condition, including temperature, correspond
recover significant yields of Glauber’s salt but evaporative to points B, E, H , K or any condition on the curve con-
crystallization would be required to obtain high yields of necting these points, crystals of both solutes would be
the anhydrous salt. formed by cooling.
Mixtures of multiple solutes in a single solvent are en- A second type of solubility behavior is exhibited by
countered in a number of processes—for example, in the mixtures that form solid solutions. Consider, for exam-
recovery of various chemicals from ores or brines. Expres- ple, a hypothetical system containing R and S whose

FIGURE 2 Solubility diagram for several common substances.

P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 95

creases and at TC the mass fraction of R in the liquid is yC

and in the solid it is xC . At TD the liquid phase disappears,
leaving a solid with a mass fraction of R equal to z M .
Systems that exhibit behavior of the type illustrated
in Fig. 4 cannot be purified in a single crystallization
stage. They represent situations in which multiple stages
or continuous-contacting devices may be useful. The prin-
ciples of such operations are analogous to those of other
countercurrent contacting operations—for example, dis-
tillation, absorption, and extraction.
Variables other than temperature and the presence of
other solutes can influence solubility. For example, the ef-
fect of a nonsolvent on solubility sometimes is used to
bring about recovery of a solute. Figure 5 shows the sol-
ubility of L-serine in aqueous solutions containing vary-
ing amounts of methanol. Note that increasing methanol
content reduces the solubility by more than an order of
magnitude, and this characteristic can be used to obtain a
FIGURE 3 Solubility diagram of KNO3 and NaNO3 mixtures in
high yield in the recovery of L-serine.
water.
There is increasing interest in the crystallization of so-
lutes from supercritical-fluid solvents. In such instances,
equilibrium behavior is described in Fig. 4. The phase en- solubilities often are correlated by an equation of state.
velope is drawn based on the compositions of coexisting Such concepts are beyond the scope of the current discus-
liquid and solid phases at equilibrium. The pure compo- sion but are presented elsewhere in the encyclopedia.
nent R has a melting point at pressure P equal to T2 while Although this discussion provides insight to the types
the melting point of pure S is T1 . The system behavior can of solubility behavior that can be exhibited by various sys-
best be described by the following example: Consider a tems, it is by no means a complete survey of the topic. Ex-
mixture of R and S at temperature TA and having a mass tensive solubility data and descriptions of more complex
fraction of R equal to z M . From the phase diagram, the equilibrium behavior can be found in the literature. Pub-
mixture is a liquid. As the liquid is cooled, a solid phase lished data usually consist of the influence of temperature
forms when the temperature reaches TB and the system is on the solubility of a pure solute in a pure solvent; seldom
allowed to come to equilibrium; the solid-phase composi- are effects of other solutes, co-solvents, or pH considered.
tion corresponds to a mass fraction of R equal to xB . On As a consequence, solubility data on a system of interest
cooling the liquid further, the ratio of solid to liquid in- should be measured experimentally, and the solutions used
in the experiments should be as similar as possible to those
expected in the process. Even if a crystallizer has been de-
signed and the process is operational, obtaining solubility
data using mother liquor drawn from the crystallizer or
a product stream would be wise. Moreover, the solubility
should be checked periodically to see if it has changed due
to changes in the upstream operations or raw materials.
There have been advances in the techniques by which
solid–liquid equilibria can be correlated and, in some
cases, predicted. These are described in references on
phase-equilibrium thermodynamics.

B. Mass and Energy Balances

Illustrating the formulation of mass and energy bal-
ances is simplified by restricting the analysis to sys-
tems whose crystal growth kinetics are sufficiently fast
FIGURE 4 Hypothetical solubility diagram of mixture without a to utilize essentially all of the supersaturation provided
eutectic at constant pressure: x, solid; y, liquid; z, combined. by the crystallizer; in other words, the product solution
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

96 Crystallization Processes

FIGURE 5 Effect of methanol on solubility of L-serine.

is assumed to be saturated. Under such conditions (re- If specific enthalpies are unavailable, they can be es-
ferred to in the crystallization literature as Class II or fast- timated based on defined reference states for both solute
growth behavior), the solute concentration in the mother and solvent. Often the most convenient reference states are
liquor can be assigned a value corresponding to satura- crystalline solute and pure solvent at an arbitrarily chosen
tion. Should the supersaturation in the mother liquor be reference temperature. The reference temperature selected
so great as to affect the solute balance, the operation is usually corresponds to that at which the heat of crystal-
said to follow Class I or slow-growth behavior. In Class I lization,  Ĥ c , of the solute is known. (The heat of crystal-
behavior, the operating conditions affect the rate at which lization is approximately equal to the negative of the heat
solute is crystallized, and an expression coupling the rate of solution.) For example, if the heat of crystallization is
of growth to a solute balance must be used to describe known at Tref , then reasonable reference conditions would
the system. Such treatment will be considered beyond the be the solute as a solid and the solvent as a liquid, both at
scope of this discussion. Tref . The specific enthalpies could be estimated then as:
The solution of mass and energy balances requires sol-
ubility and enthalpy data on the system of interest. Various Ĥ F = xF  Ĥ c + C pF (T − Tref ) (4)
methods of presenting solubility data were given earlier, Ĥ C = C pC (T − Tref ) (5)
and the use of solubilities to estimate crystal production
rates from a cooling crystallizer was demonstrated by the Ĥ L = xL  Ĥ c + C pL (T − Tref ) (6)
discussion of Eq. (2). Subsequent to determining the yield,
the rate at which heat must be removed from the crystal- where xF and xL are the mass fractions of solute in the
lizer can be calculated from an energy balance: feed and mother liquor, respectively. All that is required
now to determine the required rate of heat transfer is the
m C Ĥ C + m L Ĥ L − m F Ĥ F = Q (3) indicated heat capacities, which can be estimated based
on system composition or measured experimentally.
where m F , m C , and m L are feed rate, crystal production Now suppose some of the solvent is evaporated in the
rate, and mother liquor flow rate, respectively; Ĥ is spe- crystallizer. Independent balances can be written on total
cific enthalpy of the stream corresponding to the subscript; and solute masses:
and Q is the required rate of heat transfer to the crystallizer.
As m F , m C , and m L are known or can be calculated from a mF = mV + mL + mC (7)
simple mass balance, determination of Q requires estima- xF m F = xL m L + xC m C (8)
tion of specific enthalpies. These are most conveniently
obtained from enthalpy-composition diagrams, which are Assuming that the streams leaving the crystallizer are in
available in the general literature for a number of sub- equilibrium, there is a relationship between the temper-
stances. ature (or pressure) at which the operation is conducted
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 97

and xL and xC . In addition, an energy balance must be where µi is the chemical potential of solute i at the existing
satisfied: conditions of the system, µi∗ is the chemical potential of the
solute equilibrated at the system conditions, and ai and ai∗
m F Ĥ F + Q = m V Ĥ V + m L Ĥ L + m C Ĥ C (9)
are activities of the solute at the system conditions and at
The specific enthalpies in the above equation can be deter- equilibrium, respectively. Less abstract definitions involv-
mined as described earlier, provided the temperatures of ing measurable system quantities are often used to approx-
the product streams are known. Evaporative cooling crys- imate supersaturation; these involve either temperature or
tallizers (described more completely in Section V) operate concentration (mass or moles of solute per unit volume or
at reduced pressure and may be considered adiabatic. In mass of solution or solvent) or mass or mole fraction of
such circumstances, Eq. (9) is modified by setting Q = 0. solute. Recommendations have been made that it is best to
As with many problems involving equilibrium relation- express concentration in terms of moles of solute per unit
ships and mass and energy balances, trial-and-error com- mass of solvent. For systems that form hydrates, the so-
putations are often involved in solving Eqs. (7) through (9). lute should include the water of hydration, and that water
should be deducted from the mass of solvent.
Consider, for example, a system at temperature T with
III. NUCLEATION AND GROWTH KINETICS a solute concentration C, and define the equilibrium tem-
perature of a solution having a concentration C as T ∗ and
The kinetics of crystallization have constituent phenom- the equilibrium concentration of a solution at T as C ∗ .
ena in crystal nucleation and growth. The rates at which These quantities may be used to define the following ap-
these occur are dependent on driving forces (usually ex- proximate expressions of supersaturation:
pressed as supersaturation), physical properties, and pro-
cess variables, but relationships between these quantities 1. The difference between the solute concentration and
and crystallization kinetics often are difficult to express the concentration at equilibrium, Ci = Ci − Ci∗
quantitatively. As a result, empirical or qualitative links 2. For a solute whose solubility in a solvent increases
between a process variable and crystallization kinetics are with temperature, the difference between the
useful in providing guidance in crystallizer design and op- temperature at equilibrium and the system
eration and in developing strategies for altering the prop- temperature, T = T ∗ − T
erties of crystalline products. 3. the supersaturation ratio, which is the ratio of the
Nucleation and growth can occur simultaneously in a solute concentration and the equilibrium
supersaturated environment, and the relative rates at which concentration, Si = Ci /Ci∗
these occur are primary determinants of the characteris- 4. The ratio of the difference between the solute
tics of the crystal size distribution; one way of influencing concentration and the equilibrium concentration to
product size distributions is through the control of vari- the equilibrium concentration, σi = (Ci − Ci∗ )/Ci∗ =
ables such as supersaturation, temperature, and mixing Si − 1, which is known as relative supersaturation.
characteristics. Obviously, those factors that increase nu-
cleation rates relative to growth rates lead to a crystal size Any of the above definitions of supersaturation can be used
distribution consisting of smaller crystals. In the discus- over a moderate range of system conditions, but as outlined
sion that follows, an emphasis will be given to the general in the following paragraph, the only rigorous expression
effects of process variables on nucleation and growth, but is given by Eq. (10).
the present understanding of these phenomena does not The definitions of supersaturation ratio and relative su-
allow quantitative a priori prediction of the rates at which persaturation can be extended to any of the other variables
they occur. used in the definition of supersaturation. For example,
defining Sai = ai /ai∗ gives:
A. Supersaturation µi γi Ci
= ln Sai = ln ∗ ∗ (11)
RT γi Ci
Supersaturation is the thermodynamic driving force for
both crystal nucleation and growth; and therefore, it is Therefore, for ideal solutions or for γi ≈ γi∗ ,
the key variable in setting the mechanisms and rates by µi Ci
which these processes occur. It is defined rigorously as the ≈ ln ∗ = ln Si (12)
RT Ci
deviation of the system from thermodynamic equilibrium
and is quantified in terms of chemical potential, Furthermore, for low supersaturations (say, Si < 1.1),
ai µi
µi = µi − µi∗ = RT ln ∗ (10) ≈ Si − 1 = σi (13)
ai RT
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

98 Crystallization Processes

The simplicity of Eq. (13) results in the use of relative liquor transformed to a slurry of very fine crystals with
supersaturation in most empirical expressions for nucle- only a slight increase in supersaturation, for example by
ation and growth kinetics. While beguilingly simple, and decreasing the solution temperature.
correct in limiting cases, great care should be taken in ex- The effect of exogenous solid matter (as in heteroge-
tending such expressions beyond conditions for which the neous nucleation) in the supersaturated solution is equiv-
correlations were developed. alent to that of a catalyst in a reactive mixture. Namely,
ν
For ionic solutes, ai = a± , which leads to Sai = it is to reduce the energy barrier to the formation of a
∗ ν
(a± /a± ) and new phase. In effect, the solid matter reduces the interfa-
cial energy surf by what may amount to several orders of
µi γi± Ci
= ν ln Sai = ν ln ∗ ∗ (14) magnitude.
RT γi± Ci The classical nucleation theory embodied in Eq. (16)
∗
Again, for γi± ≈ γi± , has a number of assumptions and physical properties
that cannot be estimated accurately. Accordingly, empir-
µi Ci ical power-law relationships involving the concept of a
≈ ν ln ∗ = ν ln Si (15)
RT Ci metastable limit have been used to model primary nucle-
ation kinetics:

B. Primary Nucleation B ◦ = kN σmax

n
(17)

The term primary nucleation is used to describe both ho- where kN and n are parameters fit to data and σmax is
mogeneous and heterogeneous nucleation mechanisms in the supersaturation at which nuclei are observed when
which solute crystals play no role in the formation of new the system is subjected to a specific protocol. Although
crystals. Primary nucleation mechanisms involve the for- Eq. (17) is based on empiricism, it is consistent with the
mation of crystals through a process in which constituent more fundamental Eq. (16).
crystal units are stochastically combined. Both homo-
geneous and heterogeneous nucleation require relatively C. Secondary Nucleation
high supersaturations, and they exhibit a high-order de-
pendence on supersaturation. As will be shown shortly, Secondary nucleation is the formation of new crystals
the high-order dependence has a profound influence on through mechanisms involving existing solute crystals; in
the character of crystallization processes in which primary other words, crystals of the solute must be present for
nucleation is the dominant means of crystal formation. secondary nucleation to occur. Several features of sec-
The classical theoretical treatment of primary nucle- ondary nucleation make it important in the operation of
ation that produces a spherical nucleus results in the ex- industrial crystallizers: First, continuous crystallizers and
pression: seeded batch crystallizers have crystals in the magma that
  can participate in secondary nucleation mechanisms. Sec-
16π 3
v 2
ond, the requirements for the mechanisms of secondary
B ◦ = A exp − 3 3 surf
3k T [ln(σ + 1)]2 nucleation to be operative are fulfilled easily in most in-
  dustrial crystallizers. Finally, many crystallizers are oper-
σ <0.1 3
16π surf v2 ated in a low supersaturation regime so as to maximize
≈ A exp − 3 3 2 (16) yield, and at such supersaturations the growth of crystals
3k T σ
is more likely to produce desired morphologies and high
where k is the Boltzmann constant, surf is the interfacial purity; these low supersaturations can support secondary
surface energy per unit area, v is molar volume of the nucleation but not primary nucleation.
crystallized solute, and A is a constant.
The theory shows that the most important variables af-
1. Mechanisms
fecting the rates at which primary nucleation occur are
interfacial energy surf , temperature T , and supersatura- Secondary nucleation can occur through several mecha-
tion σ . The high-order dependence of nucleation rate on nisms, including initial breeding, contact nucleation (also
these three variables, especially supersaturation, is impor- known as collision breeding), and shear breeding. Al-
tant because, as shown by an examination of Eq. (16), a though a universal expression for the kinetics of secondary
small change in any of the three variables could produce an nucleation does not exist, a working relationship often can
enormous change in nucleation rate. Such behavior gives be obtained by correlating operating data from a crystal-
rise to the often observed phenomenon of having a clear lizer with a semi-empirical expression. Guidance as to the
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 99

form of the expression and the variables that it should in- energy may have a high-order dependence on rotational
clude can be obtained by understanding the various mech- speed and, if that is the case, modest changes in this vari-
anisms of secondary nucleation. able could alter nucleation rates substantially. The fraction
Initial breeding results from immersion of seed crys- of the impact energy transmitted from an impeller to the
tals in a supersaturated solution, and it is thought to be crystal can be manipulated by changing the material of
caused by dislodging extremely small crystals that were construction of the impeller. The influence of using soft
formed on the surface of larger crystals during drying. Al- materials to coat impellers or crystallizer internals may
though this mechanism is unimportant in continuous and vary from one crystalline system to another; those systems
unseeded batch crystallization, it can have a significant in which the crystal face is soft may be more susceptible
impact on the operation of seeded batch crystallizers. The to nucleation rate changes than those crystalline systems
number of crystals formed by initial breeding, has been where the face is hard.
found to be proportional to the surface area of crystals Supersaturation has been observed to affect contact nu-
used to seed a batch crystallizer. Characteristics of the cleation, but the mechanism by which this occurs is not
resulting distribution are affected strongly by the growth clear. There are data that infer a direct relationship be-
kinetics of nuclei resulting from initial breeding, and the tween contact nucleation and crystal growth; these data
phenomenon of growth-rate dispersion (which will be dis- showed that the number of nuclei produced by an impact
cussed later) can lead to erroneous conclusions regarding was proportional to the linear growth rate of the impacted
the nucleation kinetics. face. This could indicate that the effect of supersaturation
Shear breeding results when supersaturated solution is to alter growth rates and, concomitantly, the charac-
flows by a crystal surface and carries with it crystal precur- teristics of the impacted crystal faces; alternatively, what
sors believed formed in the region of the growing crystal appears to be a mechanistic relationship actually could be
surface. High supersaturation is required for shear breed- a result of both nucleation and growth depending upon
ing to produce significant numbers of nuclei. supersaturation.
Contact nucleation in industrial processes results from Another theory that could account for the effect of su-
collisions of crystals with the impeller used for circulation persaturation on contact nucleation is based on the view
of the magma or with other crystallizer internals such as that nuclei formed cover a range of sizes that includes the
baffles, pipe and crystallizer walls, and even other crystals. critical nucleus. Since only the nuclei larger than the crit-
Careful experimental studies have shown that the num- ical nucleus are stable, the relationship of the size of the
ber of crystals produced by collisions between crystals critical nucleus to supersaturation reflects the dependence
and these objects depends upon the collision energy, su- of contact nucleation on supersaturation. This concept,
persaturation at impact, supersaturation at which crystals which has been referred to as a survival theory, seems to
mature, material of the impacting object, area and angle have been refuted by measurements of the sizes of crystals
of impact, and system temperature. The collision energy formed by collisions. These sizes are much larger than the
for contact nucleation is small and does not necessarily critical nucleus, and the survival theory would have little
result in the macroscopic degradation or attrition of the influence on the number of nuclei that survive.
contacted crystal. Evidence of the formation of polymolecular clusters in
Nucleation from collisions between crystals in the cir- supersaturated solutions may provide a mechanistic inter-
culating magma and the rotor in a circulation pump or an pretation of the effect of supersaturation on contact nu-
agitator usually dominate nucleation resulting from other cleation kinetics. These clusters may participate in nucle-
collisions. The operating variables in systems of this type ation, although the mechanism by which this would occur
can be manipulated to some extent, thereby modifying is not clear. One model that has been proposed, however,
nucleation rates and the concomitant crystal size distri- calls for the formation of a semi-ordered region consisting
bution. For example, internal classification can be used to of molecular clusters awaiting incorporation into the crys-
keep larger crystals away from energetic collisions with an tal lattice. Collisions or fluid shear of the region containing
impeller, but doing so may create other problems with sta- high cluster concentrations could then result in these clus-
bility of the crystal size distribution. The rotational speed ters serving as secondary nuclei. In such a model, the vari-
of an impeller can be changed if there are appropriate con- ables that influence formation and diffusion of the clusters
trols on the pump or agitator. Caution must be exercised, also influence crystal growth rates and nucleation.
however, for a reduction in circulation velocity can reduce
heat-transfer coefficients and increase fouling (encrusta-
2. Kinetic Expressions
tion) on heat-transfer surfaces. Moreover, the crystals in
the magma must be kept suspended or crystal morphol- Irrespective of the actual mechanisms by which con-
ogy and growth rates could be affected adversely. Impact tact nucleation occurs, empirical power-law expressions
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

100 Crystallization Processes

provide a useful means of correlating nucleation kinetics Multicrystal-magma studies usually involve examina-
and using the resulting correlations in process analysis and tion of the rate of change of a characteristic crystal di-
control. The correlations generally take the form: mension or the rate of increase in the mass of crystals in
a magma. The characteristic dimension in such analyses
B ◦ = k N σ i MT N k
j
(18)
depends upon the method used in the determination of
where kN , i, j, k are positive parameters obtained from crystal size; for example, the second largest dimension is
data correlation, MT is the magma density (mass of solids measured by sieve analyses, while an equivalent spherical
per unit volume of slurry or solvent in the magma), and N diameter is determined by both electronic zone sensing
is the rotational velocity of the impeller or pump rotor. For and laser light scattering instruments. A relationship be-
convenience, either crystal growth rate or mean residence tween these two measured dimensions and between the
time, both of which are directly related to supersaturation, measured quantities and the actual crystal dimensions can
may be substituted for σ in Eq. (18). be derived from appropriate shape factors. Volume and
If primary nucleation dominates the process, i tends area shape factors are defined by the equations:
to larger values (say greater than 3), j and k approach
vcrys = kvol L 3 and acrys = karea L 2 (19)
zero, and Eq. (18) approaches Eq. (17). Should crystal–
impeller and/or crystal–crystallizer impacts dominate, j where vcrys and acrys are volume and area of a crystal,
approaches 1; on the other hand, if crystal–crystal contacts kvol and karea are volume and area shape factors, and L
dominate, j approaches 2. is the characteristic dimension of the crystal. Suppose an
The ease with which nuclei can be produced by con- equivalent spherical diameter L sphere is obtained from an
tact nucleation is a clear indication that this mechanism electronic zone-sensing instrument, and the actual dimen-
is dominant in many industrial crystallization operations. sions of the crystal are to be calculated. Assume for the
Research on this nucleation mechanism is continuing with sake of this example that the crystals have a cubic shape.
sphere
the objective of building an understanding of the phe- Let L cube be the edge length of the crystal and kvol and
nomenon that will allow its successful inclusion in models cube
kvol be the volume shape factors for a sphere and a cube,
describing commercial systems. respectively. Since the volume of the crystal is the same,
regardless of the arbitrarily defined characteristic dimen-
D. Fundamentals of Crystal Growth sion,
sphere
Crystal growth rates may be expressed as (1) the linear vcrys = kvol L 3sphere = kvol
cube 3
L cube (20)
advance rate of an individual crystal face, (2) the change sphere
Since kvol is π/6 and kvol cube
is 1.0, the numerical rela-
in a characteristic dimension of a crystal, or (3) the rate of tionship between L cube and L sphere is given by:
change in mass of a crystal or population of crystals. These
 sphere 1/3  1/3
different expressions are related through crystal geometry; kvol π
it is often convenient to use the method of measurement L cube = cube
L sphere = L sphere (21)
kvol 6
as the basis of the growth rate expression or, in certain
instances, the method used to analyze a crystallization The rate of change of a crystal mass dm crys /dt can be
process will require that growth rate be defined in a spe- related to the rate of change in the crystal characteristic
cific way. For example, the use of a population balance dimension (d L/dt = G) by the equation:
to describe crystal size distribution requires that growth    
dm crys d ρkvol L 3 dL
rate be defined as the rate of change of a characteristic = = 3ρkvol L 2 (22)
dimension. dt dt dt
Single-crystal growth kinetics involve the advance rate where ρ is crystal density. Since karea = acrys /L 2 ,
of an individual crystal face normal to itself or the rate of
change in crystal size associated with exposure to a super- dm crys
= 3ρ(kvol /karea )acrys G (23)
saturated solution. The advance rate of a single crystal face dt
can be quantified by observation of the face through a cali- At least two resistances contribute to the kinetics of
brated eyepiece of an optical microscope, which allows ex- crystal growth. These resistances apply to (1) integration
amination of the structure of the advancing crystal face and of the crystalline unit (e.g., solute molecules) into the crys-
isolation of surface-reaction kinetics from mass-transfer tal surface (i.e., lattice), and (2) molecular diffusion or bulk
kinetics (these phenomena will be discussed later). An transport of the unit from the surrounding solution to the
additional advantage of single-crystal systems is that it is crystal surface. As aspects of molecular diffusion and mass
possible to examine crystal growth kinetics without inter- transfer are covered elsewhere, the current discussion will
ference from competing processes such as nucleation. focus only on surface incorporation.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 101

1. Mechanisms An empirical approach also can be used to relate growth

kinetics to supersaturation by simply fitting growth-rate
Among the many models that have been proposed to de-
data with a power-law function of the form:
scribe surface-reaction kinetics are those that assume crys-
tals grow by layers and others that consider growth to oc- G = kG σ g (25)
cur by the movement of a continuous step. Each physical
model results in a specific relationship between growth where kG and g are system-dependent constants. Such an
rate and supersaturation and, although none can predict approach is valid over modest ranges of supersaturation,
growth kinetics a priori, insights regarding the effects of and the power-law function approximates the fundamental
process variables on growth can be obtained. Because of expressions derived from the above models.
the extensive literature on the subject, only the key aspects
of the physical models and (in one case) the resulting re- 2. Impurities
lationship between growth and supersaturation predicted
by each theory will be discussed here. The presence of impurities can alter growth rates substan-
The model used to describe the growth of crystals by tially, usually by decreasing them. Furthermore, as de-
layers is based on a two-step, birth-and-spread mecha- scribed in Section IV.B, impurities can alter crystal mor-
nism. In one of the steps (birth) a two-dimensional nu- phology through their effects on the growth rates of crystal
cleus is formed on the crystal surface, and in the sec- faces. Mechanisms include: (1) adsorption of an impurity
ond step (spread) the two-dimensional nucleus grows to on the crystal surface or at specific growth sites such as
cover the crystal surface. When one or the other of the kinks, thereby blocking access to the site by a growth
steps is controlling growth rates, simplifications of the unit; (2) formation of complexes between an impurity and
more complicated dependence of growth rate on super- a growth unit; and (3) incorporation of an impurity into a
saturation can be developed to give what are known as growing crystal and creating defects or repelling the addi-
the mononuclear two-dimensional nucleation theory and tion of a growth unit to the subsequent crystal layer. Few
the polynuclear two-dimensional nucleation theory. of these mechanistic views result in predictive capabili-
In the mononuclear two-dimensional nucleation theory, ties, and it is usual to rely on experimental data that are
surface nucleation occurs at a finite rate while the spread- often correlated empirically.
ing across the surface occurs at an infinite rate. The re- Because impurities most often result in reduced crys-
verse is true for the polynuclear two-dimensional nucle- tal growth rate, feedstocks to laboratory and bench-scale
ation theory. Theoretical relationships have been derived units should be as similar as possible to that expected in the
between growth rate and supersaturation for each of these full-scale unit. The generation of impurities in upstream
conditions but are considered beyond the scope of this process units can depend on the way those units are oper-
discussion. ated, and protocols of such units should follow a consistent
The screw-dislocation theory (sometimes referred to as practice. It is equally important to monitor the composi-
the BCF theory because of its development by Burton, tion of recycle streams so as to detect any accumulation of
Cabrera, and Frank) is based on a mechanism of continu- impurities that might lead to a reduction in growth rates.
ous movement in a spiral or screw of a step or ledge on the The solvent from which a material is crystallized influ-
crystal surface. The theory shows that the dependence of ences crystal morphology and growth rate. These effects
growth rate on supersaturation can vary from a parabolic have been attributed to two sets of factors. One has to do
relationship at low supersaturations to a linear relationship with the effects of solvent on viscosity, density, and dif-
at high supersaturations. In the BCF theory, growth rate is fusivity and, therefore, mass transfer. The second factor
given by: is concerned with the structure of the interface between
 2   crystal and solvent; a solute–solvent system that has a
σ b high solubility is likely to produce a rough interface and,
G = kG tanh (24)
b σ concomitantly, large crystal growth rates.
where is screw dislocation activity and b is a system-
dependent quantity that is inversely proportional to tem- E. Crystal Growth in Mixed Crystallizers
perature. It can be shown that the dependence of growth
Population balances on crystals in a crystallizer require a
rate on supersaturation is linear if the ratio b/ σ is large,
definition of growth rates in terms of the rate of change of
but the dependence becomes parabolic as the ratio be-
a characteristic dimension:
comes small. It is possible, then, to observe variations in
the dependence of growth rate on supersaturation for a dL
G= (26)
given crystal-solvent system. dt
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

102 Crystallization Processes

Furthermore, the solution of a differential population bal-

ance requires that the relationship between growth rate
and size of the growing crystals be known. When all crys-
tals in the magma grow at a constant and identical rate, the
crystal–solvent system is said to follow the McCabe L
law, while systems that do not are said to exhibit anoma-
lous growth.
Two theories have been used to explain growth-rate
anomalies: size-dependent growth and growth-rate disper-
sion. As with systems that follow the L law, anomalous
growth by crystals in a multicrystal magma produces crys-
tal populations with characteristic forms. Unfortunately,
it is difficult to determine the growth mechanism from
an analysis of these forms. This means that either size-
dependent growth or growth-rate dispersion may be used
to correlate population density data without a certainty that
the correct source of anomalous growth has been identi-
fied. Determining the actual source of anomalous growth
is not trivial, but it may be worthwhile since alignment
between a mathematical model and system behavior en-
hances the utility of the model.
Size-dependent crystal growth results when the rate of FIGURE 6 Transient population density plot showing growth-rate
dispersion.
growth depends on the size of the growing crystal. Cer-
tainly, this may be the case if bulk transport is the control-
ling resistance to crystal growth, and the literature abounds
with expressions for the appropriate mass-transfer coef- IV. PURITY, MORPHOLOGY, AND
ficients. In the more common situation in which surface SIZE DISTRIBUTIONS
integration controls growth rate, there are no mechanis-
tic relationships between growth rate and crystal size, Crystal properties can be divided into two categories: those
and simple empirical expressions are called upon for that based on the individual crystal and those involving all crys-
purpose. tals of a given population. The three characteristics of the
Growth-rate dispersion is the term used to describe the section title compose what is often called crystal qual-
behavior of similar sized crystals in the same population ity. They represent, along with yield, the most important
exhibiting different growth rates or growth rates that vary criteria in the operation of a crystallizer. In the following
with time. The consequences of growth-rate dispersion are discussion, some of the factors that influence purity and
illustrated in Fig. 6, which shows the growth of a crystal morphology are described and an introduction is given to
population that has been immersed in a supersaturated so- methods of quantifying crystal size distributions.
lution. The spread of the distribution increases as the crys-
tal population grows; the slower growing crystals form the
A. Crystal Purity
tail of the advancing distribution while the faster growing
ones form the leading edge. If all crystals in the popula- The purity of a crystalline product depends on the nature
tion grew at the same rate, the distribution would advance of the other species in the mother liquor from which the
uniformly along the size axis. Two causes of growth-rate crystals are produced, the physical properties of the mother
dispersion have been observed. In one, the growth rate of liquor, and the processing that occurs between crystalliza-
each crystal in a population is nearly constant, but crystals tion and the final product (downstream processing). Im-
in the population may grow at a different rate; in the other, purities can find their way into the final product through
the growth rate of an individual crystal fluctuates about a a number of mechanisms: the formation of occlusions,
mean value. trapping of mother liquor in physical imperfections of the
The consequences of anomalous growth depends upon crystals or agglomerates, adsorption of species onto crys-
the process involved, and this will be pointed out in the tal surfaces, as part of chemical complexes (hydrates or
discussion on population balances. solvates), or through lattice substitution.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 103

Occlusions find their way into the crystal structure when The characteristic macroscopic shape of a crystal re-
the supersaturation in close proximity to the crystal surface sults, in large measure, from the internal lattice structure;
is high enough to lead to an unstable surface. Such insta- surfaces are parallel to planes formed by the constituent
bility leads to the creation of dendrites, which then join units of the crystal. Moreover, although the Law of Con-
to trap mother liquor in pools of liquid within the crystal. stant Interfacial Angles is a recognition that angles be-
Occlusions are often visible and can be avoided through tween corresponding faces of all crystals of a given sub-
careful control of the supersaturation in the crystallizer. stance are constant, the faces of individual crystals of that
Mother liquor can be flushed from a cake of crystals substance may exhibit varying degrees of development.
on a filter or centrifuge by washing with a liquid that also As a result, the general shape or habit of a crystal may
may dissolve a small portion of the cake mass. To be ef- vary considerably.
fective, the wash liquid must be spread uniformly over Crystal morphology (i.e., both form and shape) affects
the cake and flow through the porous material without crystal appearance; solid–liquid separations such as filtra-
significant channeling. Such washing is hindered when tion and centrifugation; product-handling characteristics
the crystals themselves have significant cracks, crevices, such as dust formation, agglomeration, breakage, and
or other manifestations of breakage or the mother liquor washing; and product properties such as bulk density,
has a viscosity that is significantly greater than the wash dissolution kinetics, catalytic activity, dispersability, and
liquid. In the latter event, significant channeling (also caking.
called fingering) may reduce the effectiveness of the wash The shape of a crystal can vary because the relative rates
process. of growth of crystal faces can change with system con-
Lattice substitution requires that the incorporated im- ditions; faster growing faces become smaller than faces
purity be of similar size and function to the primary crys- that grow more slowly and in the extreme may disappear
tallizing species. In other words, the impurity must fit from the crystal altogether. For illustration, consider the
into the lattice without causing significant dislocations. two-dimensional crystal shown in Fig. 7a and the process
An example of such a system is found in the crystalliza- variables that would cause the habit to be modified to the
tion of L-isoleucine in the presence of trace quantities of forms shown in Figs. 7b and c. The shape of the crystal
L-leucine. The two species have similar molecular struc- depends on the ratio of the growth rate of the horizontal
tures, differing only by one carbon atom in the position faces, G h , to the growth rate of the vertical faces, G v . For
of a methyl side group. In this system, the incorporation the shapes shown in Fig. 7,
of L-leucine in L-isoleucine crystals is proportional to the      
Gh Gh Gh
concentration of L-leucine in the mother liquor. Moreover, < < (27)
the shape of the recovered crystals changes as the content Gv b Gv a Gv c
of L-leucine in recovered crystal increases. Growth rates depend on the presence of impurities,
system temperature, solvent, mixing, and supersaturation,
and the importance of each may vary from one crystal face
B. Crystal Morphology
to another. Consequently, an alteration in any or all of these
Both molecular and macroscopic concepts are important variables can result in a change of the crystal shape.
in crystal morphology. Molecular structures (i.e., the ar- Modeling intermolecular and intramolecular inter-
rangements of molecules in specific lattices) can greatly actions through molecular mechanics calculations has
influence the properties of a crystalline species and varia- advanced significantly in the past decade, and it has pro-
tions from a single structure lead to the prospect of poly- vided the basis for prediction of the equilibrium shape of
morphic systems. In such systems, the molecular species
of the crystal can occupy different locations depending
on the conditions at which the crystal is formed, and both
microscopic and macroscopic properties of the crystal can
vary depending on the polymorph formed. There is, in gen-
eral, a single stable polymorph for prevailing conditions,
but that polymorph may not have been formed during the
crystallization process. In such cases, system thermody-
namics will tend to force transformation from the unstable
polymorph to the stable one at rates that may vary from
being nearly instantaneous to infinitely slow. Additional
discussion of the molecular structures of crystalline ma-
terials has been provided elsewhere. FIGURE 7 Effect of facial growth rates on crystal shape.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

104 Crystallization Processes

a known crystalline species. While not yet uniformly suc- 1. Intrinsic growth rates
cessful on a quantitative basis, the definition and modeling a. Temperature: The growth rates of individual
of crystal lattice potential energy equations has provided crystal faces depend on temperature, typically
an understanding of crystal growth and morphology on the following an Arrhenius rate law:
molecular level. Derivation of external crystal morphol-  
ogy from internal lattice structures via simulation has been E G
G = G 0 exp − (28)
proven possible for several organic compounds. Numeri- RT
cal minimization techniques, coupled with the appropriate If different crystal faces have different activation
valence and nonbonded energy expressions, have enabled energies, variation of the temperature at which
accurate determination of favorable molecular arrange- crystallization takes place modifies individual
ments within a wide variety of molecular crystals. growth rates to varying degrees and results in a
The shape of crystals obtained as a result of following modified crystal shape.
a specific crystallization protocol may be unsatisfactory b. Mixing: The intensity of mixing may determine
and, as a result, methods for modifying the habit of con- the degree to which bulk mass transfer is involved
siderable interest. The predictive capabilities cited in the in growth kinetics, and this can influence the
preceding paragraph are of great utility in such an instance resulting crystal shape.
as they may be used to determine factors leading to the c. Supersaturation: The dependence of growth
unsatisfactory shape and guide subsequent experiments in kinetics on supersaturation may vary from one
which a more desirable shape is sought. Inevitably, such a crystal face to another. Accordingly, different
search involves extensive laboratory or bench-scale exper- prevailing supersaturations can lead to different
iments to determine processing variations that will lead to crystal shapes.
a desired crystal shape.
2. Interfacial behavior
As an example of the variations in shape that can be ex-
hibited by a single crystalline material, consider the forms a. Solvent: Different solvents exhibit different
of potassium sulfate shown in Fig. 8. Clearly, the process- interactions with crystal faces and can alter crystal
ing characteristics and particulate properties of the differ- shape. A change in solvent also can alter the
ently shaped potassium sulfate crystals will vary. stoichiometry of the crystal (e.g., from a hydrated
The mechanisms and variables affecting crystal shape to an anhydrate stoichiometry), which can produce
can be categorized as follows: crystals with quite different morphology.
b. Surfactants: Addition of a surfactant to a
crystallizing system can influence the crystal
shape in a manner illustrated schematically in
Fig. 9. Here, surfactant molecules are shown being
attracted to crystal faces in varying ways; the
hydrophilic head groups favor the horizontal faces,
while the hydrophobic tail groups are
preferentially attracted to the vertical faces. A
growth unit must displace the surfactant to
approach a growing crystal face. As hydrophilic
interactions are typically much stronger than
hydrophobic ones, the growth unit preferentially
enters the vertical faces and growth in the
horizontal direction is favored.
3. Access to growth site
a. Blockage by species attracted to growth site:
Impurities may preferentially locate at a kink or
other favored growth site and block growth at that
site. A difference in the character of the kink or
growth site from one face to another could result
in modification of the crystal shape.
FIGURE 8 Shapes of K2 SO4 crystals. [From Mullin, J. W. (1993). b. Species partially fitting into crystal lattice: In
“Crystallization,” 3rd ed. Butterworth-Heinemann, London. With these instances, an impurity molecule is comprised
permission.] of two parts, one that fits into the crystal lattice
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 105

FIGURE 9 Attachment of surfactant molecules to crystal surfaces.

and a second that does not. The part that does not lation distribution relates the number of crystals at each
fit repulses incoming growth units or causes size to the size, while the mass distribution expresses how
significant interatomic stress because of its mass is distributed over the size range. In the following
position in the crystal lattice. If such species are paragraphs, methods for describing and using distribution
purposely added to the crystallizing system to functions will be outlined.
modify crystal morphology, they are referred to as Size distribution is a major determinant of the proper-
tailor-made additives. ties of crystalline products, especially appearance, and to
downstream processing and handling of crystalline mate-
rials. Solid–liquid separation by filtration or centrifugation
C. Crystal Size Distributions
can be straightforward with a desired CSD, but it can be
Most crystallization processes produce particles whose disastrous when an inappropriate one increases resistance
sizes cover a range of varying breadth. If the particles to liquor flow through a filter or centrifuge cake. Like-
consist of single crystals, the resulting distribution is a wise, CSD affects other downstream processing such as
crystal size distribution (CSD); on the other hand, if the the removal of impurities and mother liquor by washing,
particles consist of agglomerates or other combination of dissolution or reaction of the crystals, and transporting or
multiple crystals, the distribution is a particle size dis- storing crystals.
tribution. In either case, the distribution is expressed in Crystal size distributions may be expressed by: (1) his-
terms of either population (number) or mass. The popu- tograms, which are the amount or fraction of mass or
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

106 Crystallization Processes

TABLE II Sieved KNO3 Crystals from Hypothetical 1-Liter Sample

Sieve L ∆Mi M(L) F(L) L̄i ∆Ni N(L) m n
no., i (µm) (g/L) (g/L) (frac) (µm) (no./L) (frac) (g/µm·L) (no./µm·L)

1 707 0 32.974 1.000 1611

2 500 7.296 25.678 0.779 603.5 16 1595 0.0352 0.076
3 354 11.512 14.166 0.430 427.0 70 1525 0.0789 0.480
4 240 9.011 5.154 0.156 297.0 163 1362 0.0790 1.430
5 177 3.145 2.009 0.061 208.5 164 1198 0.0499 2.610
6 125 1.322 0.687 0.021 151.0 182 1016 0.0254 3.500
7 88 0.462 0.225 0.007 106.5 181 834 0.0125 4.900
8 63 0.159 0.066 0.002 75.5 175 659 0.0064 7.000
9 44 0.055 0.011 0.000 53.5 171 488 0.0029 9.000
10 0 0.011 0.000 0.000 22.0 488 0 0.0002 11.100
Total 32.974 1611

number over each increment in size; (2) cumulative distri- The method by which crystals are sized gives either
butions, which are the total or fraction of mass or number number or mass of crystals in a given size range. The sieve
below (or above) a given size; and (3) density functions, analysis in the above example gives mass distributions, so
which are the derivatives (with respect to size) of cumu- that the histogram is constructed in terms of crystal mass,
lative distributions. These definitions will be illustrated and a cumulative mass distribution, M(L), can be defined
by considering a hypothetical potassium nitrate system as the mass of crystals in the sample passing through the
from which a 1-liter slurry sample has been withdrawn, sieve of size L. In other words,
filtered, washed, dried, and sieved to give the results shown
in Table II. 
L(i)
M(L) = Mi (29)
The first three and the sixth columns give the sieve data
L=0
and should be read as follows: All of the sieved matter
passed through the 707-µm sieve, and 7.296 g remained Such calculations give the mass of crystals below size L,
on the 500-µm sieve and had an arithmetic average size of and the results are shown in column 4 of Table II. Column
603.5 µm. Similar descriptions can be given for crystals 5 gives the cumulative mass fraction distribution:
that remained on the other sieves and pan (L = 0). The
total crystal mass recovered was 39.974 g. A histogram of M(L)
F(L) = (30)
the mass distribution from these data is shown in Fig. 10. Mtotal

FIGURE 10 Histogram of size distribution from example.

P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 107

Transforming a mass distribution to a number distri- From Eq. (37), it can be demonstrated that the total num-
bution, or vice versa, requires a relationship between the ber of crystals, the total length, the total area, and the
measured and desired quantities. The mass of a single total volume of crystals, all in a unit of sample volume,
crystal, m crys , is related to crystal size by the volume shape can be evaluated from the zeroth, first, second, and third
factor, kvol (see Eq. (19)): moments of the population density function. Moments of
the population density function also can be used to esti-
m crys = ρkvol L 3 (31)
mate number-weighted, length-weighted, area-weighted,
Consequently, the number of crystals on a sieve in the and volume- or mass-weighted quantities. These averages
example, Ni , can be estimated by dividing the total mass are calculated from the general expression:
on sieve i by the mass of an average crystal on that sieve. If m j+1
the crystals on that sieve are assumed to have a size equal L̄ j+1, j = (38)
mj
to the average of the sieve through which they have passed
and the one on which they are held, L̄ i = (L i−1 + L i )/2, where j = 0 for a number-weighted average, 1 for a length-
then: weighted average, 2 for an area-weighted average, and
Mi 3 for a volume- or mass-weighted average.
Ni = (32) Crystal size distributions may be characterized usefully
ρkvol L̄ i3
(though only partially) by a single crystal size and the
Potassium nitrate crystals have a density of 2.11 × 10−12 spread of the distribution about that size. For example,
g/µm3 , which allows for the determination of the esti- the dominant crystal size represents the size about which
mated crystal numbers on each sieve in Table II. A cumu- the mass in the distribution is clustered. It is defined as
lative number distribution, N (L), and a cumulative num- the size, L D , at which a unimodal mass density function
ber fraction distribution, F(L), can be calculated using is a maximum, as shown in Fig. 11; in other words, the
methods similar to those for calculating M(L) and W (L). dominant crystal size L D is found where dm/d L is zero.
Mass and population densities are estimated from (The data used to construct Fig. 11 are from Table II.) As
the respective cumulative number and cumulative mass the mass density is related to the population density by:
distributions:
m = ρkv n L 3 (39)
Mi
m( L̄) = (33) the dominant crystal size can be evaluated from the pop-
L i
ulation density by:
Ni
n( L̄) = (34) d(n L 3 )
L i =0 at L = LD (40)
So that if L i → 0, dL
∞ The spread of the mass-density function about the dom-
d M(L)
m(L) = ⇒ M(L) = m dL (35) inant size is the coefficient of variation (c.v.) of the CSD.
dL 0
and
∞
d N (L)
n(L) = ⇒ N (L) = n dL (36)
dL 0
Equations (33) and (34) are used to obtain the last two
columns of Table II.
In the example, all of the results are for the given sam-
ple size of 1 liter and the quantities estimated have units
reflecting that basis. This basis volume is arbitrary, but use
of the calculated quantities requires care in defining this
basis consistently in corresponding mass and population
balances. The volume of clear liquor in the sample is an
alternative, and sometimes more convenient, basis.
Moments of a distribution provide information that can
be used to characterize particulate matter. The jth moment
of the population density function n(L) is defined as:
∞
mj = L j n(L) d L (37) FIGURE 11 Mass–density function with single mode showing
0 dominant size.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

108 Crystallization Processes

It is defined by: tion obtained from a continuous crystallizer. The effects

σ of operating variables on crystal size distributions will be
c.v. = (41) discussed in Section VI.
LD
The throughput per unit crystallizer volume is greater
and estimated from the moments of the distribution: for a continuous system. Batch units have several operat-
1/2
m3m5 ing steps in a cycle—charging, heating or cooling, crystal-
c.v. = −1 (42) lizing, discharging, and cleaning—and the unit production
m 24
rate is based on the total cycle time, even though the for-
This is especially useful for systems that cannot be de- mation of crystals may occur only during a small portion
scribed by an analytical distribution function. of the cycle.
It may be easier to operate a continuous system so that
it reproduces a particular crystal size distribution than it
V. CRYSTALLIZER CONFIGURATION is do reproduce crystal characteristics from a batch unit.
AND OPERATION Moreover, the coupling of several transient variables and
nucleation make it difficult to model and control the op-
Crystallization equipment can vary in sophistication from eration of a batch crystallizer.
a simple stirred tank to a complicated multiphase column,
and the protocol can range in complexity from simply al-
A. Relationship of Solubility
lowing a vat of liquor to cool to the careful manipulation
to Mode of Operation
required of batch cyclic operations. In principle, the objec-
tives of these systems are the same: to produce a product The driving force for crystal formation can be generated
meeting specifications on quality at an economical yield. through a variety of means, including cooling or heating
This section will examine some of the considerations that to reduce or increase the system temperature, evaporating
go into the selection of a crystallizer so as to meet these solvent, evaporative (flash) cooling, inducing a chemical
objectives. reaction (when the reaction product is sparingly soluble,
One of the first decisions that must be made is whether the process is called precipitation), adjusting pH, salting
the crystallizer operation is to be batch or continuous. In out through the addition of a nonsolvent, direct-contact
general, the advantages of each type of operation should cooling with a refrigerant, or some other means. All of
be weighed in choosing one over the other, but more often these modes of operation can be implemented in either a
the decision rests on whether the other parts of the process batch or a continuous process. In addition, two or more of
are batch or continuous. If they are batch, then it is likely the modes may be combined to enhance the product yield.
that the crystallizer also should be batch. As discussed in an earlier section, solubility is intrinsic
The equipment required for batch crystallization can be to the solute-solvent system, and the relationship of solu-
very simple. For example, some crystalline materials are bility to temperature often determines the mode by which a
produced by simply allowing a charge of hot liquor to cool. crystallizer is operated. Recall for example that the solubil-
After the crystals have formed, the magma is discharged ity of NaCl (see Fig. 2) is essentially independent of tem-
through a filter or the liquor may be decanted and the set- perature, while Na2 SO4 · 10H2 O has a solubility that ex-
tled slurry filtered. Very large crystals can be obtained by hibits a strong dependence on temperature. Consequently,
allowing encrustations formed on the walls of these crys- cooling a sodium chloride solution cannot generate sig-
tallizers to grow undisturbed; after the system has come nificant product yield; solvent evaporation is the primary
to equilibrium, the liquor is drained and the crystals are mode of NaCl production. On the other hand, reducing
removed by scraping them from the surface. the temperature of a saturated solution of sodium sulfate
Batch crystallizers can be used in a campaign to produce generates substantial product and may be used alone or in
a particular product and in a second campaign to produce combination with evaporation.
another product. Generally, it is not possible to operate Cooling crystallizers utilize a heat sink to remove both
continuous processes in this way. Batch crystallizers can the sensible heat from the feed stream and the heat of crys-
handle viscous or toxic systems more easily than can con- tallization released or, in some cases absorbed, as crystals
tinuous systems, and interruption of batch operations for are formed. The heat sink may be no more than the am-
periodic maintenence is less difficult than dealing with in- bient surroundings of a batch crystallizer, or (as is more
terruptions in continuous processes. The latter factor may likely) it may be cooling water or another process stream.
be especially important in biological processes that require Evaporative crystallizers generate supersaturation by
frequent sterilization of equipment. Batch crystallizers can removing solvent from the mixture, thereby increasing
produce a narrow crystal size distribution, whereas special the solute concentration. They may be operated under vac-
processing features are required to narrow the distribu- uum, and in those circumstances it is necessary to have a
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 109

vacuum pump or ejector as a part of the unit. If the boiling

point elevation—the increase in boiling temperature due
to the presence of the solute—is low, mechanical recom-
pression of the vapor obtained from solvent evaporation
may be used in some cases to produce a heat source to
drive the operation.
Evaporative-cooling crystallizers are fed with a liquor
whose temperature is such that solvent flashes upon feed
entry to the crystallizer. They typically are operated under
vacuum, and flashing of solvent increases the solute con-
centration in the remaining liquor while simultaneously
reducing the temperature of the magma. The mode of this FIGURE 12 Schematic diagram of a forced-circulation evapora-
operation can be reduced to that of a simple cooling crys- tive crystallizer.
tallizer by returning condensed solvent to the crystallizer
body. schematic diagram of an evaporative forced-circulation
Salting-out crystallization operates through the addition crystallizer that withdraws a slurry from the crystallizer
of a nonsolvent to the magma in a crystallizer. The selec- body and pumps it through a heat exchanger. Heat trans-
tion of the nonsolvent is based on the effect of the solvent ferred to the circulating magma causes evaporation of sol-
on solubility, cost, properties that affect handling, inter- vent as the magma is returned to the crystallizer. This
action with product requirements, and ease of recovery. type of unit is used to control circulation rates and veloc-
Adding a nonsolvent to the system increases the complex- ities past the heat transfer surfaces, and the configuration
ity of the process; it increases the volume required for a shown is especially useful in applications requiring high
given residence time and produces a highly nonideal mix- rates of evaporation. A calandria that provides heat trans-
ture of solvent, nonsolvent, and solute. fer through natural convection is an alternative to forced-
Melt crystallization operates with heat as a separating circulation systems.
agent, but a crystalline product is not generated in the Scale formation on the heat exchanger surfaces or at the
process. Instead, crystals formed during the operation are vapor–liquid surface in the crystallizer can cause opera-
remelted and the melt is removed as the product. Such op- tional problems with evaporative crystallizers. These can
erations are often used to perform the final purification of be overcome by avoiding vaporization or excessive tem-
products after prior separation units; for example, the pu- peratures within the heat exchanger and by properly in-
rity of an acrylic acid feed may be increased from 99.5 to troducing the circulating magma into the crystallizer. For
99.9%. Melt crystallizers do not require solids handling example, introducing the circulating magma a sufficient
units nor do they utilize solid–liquid separation equip- distance below the surface of the magma in the crystallizer
ment. Finally, in some instances the use of melt crystal- prevents vaporization upon re-entry and forces it to occur
lization can eliminate the use of solvents, thereby reducing at a well-mixed zone above the point of re-entry. Alter-
the environmental impact of the process. natively, the magma may be introduced so as to induce a
swirling motion that dislodges encrustations from the wall
B. Crystallizers of the crystallizer at the vapor-liquid interface.
Figure 13 shows a schematic diagram illustrating the
The basic requirements of a crystallization system are (1) a
configuration of a surface cooling (indirect heat transfer)
vessel to provide sufficient residence time for crystals to
crystallizer. Heat can be transferred to a coolant in an
grow to a desired size, (2) mixing to provide a uniform
external heat exchanger, as shown, or in coils or a jacket
environment for crystal growth, and (3) a means of gener-
ating supersaturation. Crystallization equipment is manu-
factured and sold by several vendors, but some chemical
companies design their own crystallizers based on exper-
tise developed within their organizations. Rather than at-
tempt to describe the variety of special crystallizers that
can be found in the marketplace, this section will provide
a brief general survey of types of crystallizers that utilize
the modes outlined above.
The forced-circulation crystallizer is a simple unit de-
signed to provide high heat-transfer coefficients in either FIGURE 13 Schematic diagram of a forced-circulation, surface-
an evaporative or a cooling mode. Figure 12 shows a cooling crystallizer.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

110 Crystallization Processes

within the crystallizer. An external cooling surface re-

quires the use of a circulation pump, but this expense may
be mitigated by obtaining a higher heat-transfer coeffi-
cient than would result with the use of coils or a jacketed
vessel. The rate of heat transfer Q from the circulation
loop of a cooling crystallizer must be sufficient to reduce
the temperature of the feed and to remove the heat of
crystallization of the solute. Assuming that no substantial
crystallization occurs in the heat exchanger and limiting
the difference between entering and leaving temperatures
of the circulating magma (Tin −Tout ), so as to minimize for-
mation of encrustations, the required magma circulation
rate ṁ circ can be determined from the equation:
Q
ṁ circ = (43)
[C P (Tin − Tout )]circ
where C P is the heat capacity of the circulating magma.
The methods by which Q can be evaluated were discussed
in Section II. It is not uncommon to limit the decrease in
magma temperature to about 3 to 5◦ C; therefore, both the FIGURE 14 Draft-tube-baffle crystallizer. [Courtesy of Signal
circulation rate and heat-transfer surface must be large. Swenson Division.]
The feed to cooling crystallizers should be rapidly
mixed with the magma so as to minimize the occurrence baffle extending into the chamber and the outside wall of
of regions of high supersaturation. Such regions lead to the crystallizer. Flow through the quiescent zone can be
excessive nucleation, which is detrimental to the crystal adjusted so that crystals below a certain size (determined
size distribution. The type of pump used in the circulation by settling velocity) are removed in the fines-dissolution
loop also can lead to degradation of the crystal size dis- circuit. In the elutriation leg, crystals below a certain size
tribution; an inappropriate pump causes crystal attrition are preferentially swept back into the crystallizer by the
through abrasion, fracture, or shear, and most commer- flow of recycled mother liquor; accordingly, larger crys-
cial systems use specially designed axial-flow pumps that tals, which have a higher settling velocity, are removed
provide high flow rates and low shear. preferentially from the system.
Direct-contact refrigeration can be used if either the A second major type of crystallizer with special chan-
operating temperature of the crystallizer is low in com- neling devices is comprised of those having configurations
parison to the temperature of available cooling water or like the Oslo crystallizer shown in Fig. 15. The objective
there are severe problems with encrustations. In such an of this unit is to form a supersaturated solution by evapora-
operation, a refrigerant is mixed with the crystallizer con- tion in the upper chamber and to have crystal growth in the
tents and vaporized at the magma surface. On vaporizing, lower (growth) chamber. The use of the downflow pipe in
the refrigerant removes sufficient heat from the magma to the crystallizer provides good mixing in the growth cham-
cool the feed and to remove the heat of crystallization. The ber. As shown, the lower chamber has a varying diameter,
refrigerant vapor must be compressed, condensed, and re- which can provide some internal classification of crystals.
cycled for the process to be economical. Moreover, the The lowest portion of the chamber has the smallest diame-
refrigerant must be insoluble in the liquor to minimize ter and can be considered perfectly mixed; as the chamber
losses and product contamination. diameter increases, the upward velocity of the slurry de-
Special devices for classification of crystals may be creases and larger crystals tend to settle. In principle, only
used in some applications. Figure 14 shows a draft-tube- small crystals are supposed to leave the chamber in the
baffle (DTB) crystallizer that is designed to provide pref- circulating slurry, to flow through the circulation pump,
erential removal of both fines and classified product. As and to enter the upper chamber. As the probability of a
shown, feed is introduced to the fines circulation line so crystal colliding with the impeller decreases with decreas-
that any nuclei formed upon introduction of the feed can be ing crystal size, the internal classification provided by the
dissolved as the stream flows through the fines-dissolution Oslo crystallizer could provide some control of contact
heat exchanger. The contents of the crystallizer are mixed nucleation.
by the impeller, which forces the slurry to flow in the in- Melt crystallizers can be operated in a variety of ways.
dicated direction. A quiescent zone is formed between the In one, feed enters the crystallizer and contacts a slurry
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 111

distribution of the product or intermediate material. Such

balances are not independent of those on mass and energy,
and their solution requires an independent expression for
nucleation kinetics.
In formulating a population balance, crystals are as-
sumed sufficiently numerous for the population distribu-
tion to be treated as a continuous function. One of the key
assumptions in the development of a simple population
balance is that all crystal properties, including mass (or
volume), surface area, and so forth are defined in terms of
a single crystal dimension referred to as the characteris-
tic length. For example, Eq. (19) relates the surface area
and volume of a single crystal to a characteristic length L.
In the simple treatment provided here, shape factors are
taken to be constants. These can be determined by simple
measurements or estimated if the crystal shape is simple
and known—for example, for a cube karea = 6 and kvol = 1.
The beginning point for any balance is the following
statement:

input + generation − output − consumption

= accumulation (44)
FIGURE 15 Oslo crystallizer.
where each of the terms may be expressed as a rate or an
of mother liquor and crystals of the desired product. The amount. In a population balance, the number of entities
crystals are forced to move in a specific direction by grav- (such as crystals) is the balanced quantity and each of
ity or rotating blades. As they flow towards the appropriate the terms has dimensions of number of crystals per unit
end of the crystallizer, the crystals encounter a heated re- time for a differential balance or number of crystals for an
gion and are melted. A portion of the melt is removed integral balance. The principles involved in formulating
as product, while the remainder flows countercurrently to a balance are outlined in the following sections, and they
the crystals, thereby providing some refining and remov- provide guidance in developing corresponding balances
ing impure adhering liquid. for systems whose configurations do not conform to those
In a second method of operation, the feed material is cir- described here.
culated through a bank of tubes, each of which has a diam-
eter of up to about 8 cm. The walls of the tubes are cooled,
and material crystallizes on them throughout a fixed op- A. Perfectly Mixed, Continuous Crystallizers
erating period. At the end of that period, the remaining The balance equation must be constructed for a control
liquid is sent to a holding tank for further processing, and volume, which for a perfectly mixed crystallizer may be
then the tubes are heated slowly to cause partial melting assumed to be the total volume of the crystallizer VT .
of the adhering solids. This step is known as “sweating,” Then, a balance on the number of crystals in any size
and the impure “sweated” liquid produced is removed range (say, L 1 to L 2 = L 1 + L) must account for crys-
from the crystallizer and held for further processing. Fi- tals that enter and leave that size range by: (1) convective
nally, the product is obtained by adding additional heat to flow, (2) crystal growth, (3) crystal agglomeration, and
the tubes and melting the remaining adhering solids. The (4) crystal breakage. Agglomeration and breakage can
actual sequencing of these steps and the reprocessing of be detected through careful inspection of product parti-
residual and sweated liquids may be quite complicated. cles, and they can be quite significant in some processes.
For simplicity, however, they will be assumed negligible
in the present analysis. The rate of crystal growth G will
VI. POPULATION BALANCES AND be defined as in Eq. (26); i.e., the rate of change of the
CRYSTAL SIZE DISTRIBUTIONS characteristic crystal dimension L:

A balance on the population of crystals in a crystallizer dL

G=
can be used to relate process variables to the crystal size dt
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

112 Crystallization Processes

Then, If the crystallizer has a clear feed, growth is invariant, but

if the magma volume VT is allowed to vary, the population
growth rate into the size range = VT (Gn) L 1 (45)
balance gives:
growth rate out of the size range = VT (Gn) L 2 (46) ∂n ∂(nG) ∂(ln VT ) Vout n
+ +n + =0 (55)
∂t ∂L ∂t VT
L2
removal rate of crystals in the size range = Vout n dL The system model that led to the development of the
L1 last two equations is referred to as the mixed-suspension,
(47) mixed-product removal (MSMPR) crystallizer.
L2 Under steady-state conditions, the rate at which crystals
feed rate of crystals in the size range = Vin n in d L are produced by nucleation must be equal to the difference
L1
in rates at which crystals leave and enter the crystallizer.
(48) Accordingly, for a clear feed,
∂ L2 ∞
1 ∞
accumulation rate in the crystallizer = nVT d L VT B ◦ = Vout n d L ⇒ B◦ = n dL (56)
∂t L1 0 τ 0
(49) For crystallizers following the constraints given above,
Substituting the terms from Eqs. (46) through (49) into B ◦ = n◦ G (57)
Eq. (44) gives: For a given set of crystallizer operating conditions, nu-
L2 cleation and growth rates can be determined by measuring
VT (Gn) L 1 + Vin n in d L the population density of crystals in a sample taken from
L1
either the well-mixed zone of a crystallizer or the prod-
L2
∂ L2
uct stream flowing from that zone. Sample analyses are
= VT (nG) L 2 + Vout n dL + nVT d L (50)
L1 ∂t L1 correlated with Eqs. (54) and (57), and nucleation and
Manipulation of this equation leads to growth rates are determined from those correlations. The
sample must be representative of the crystal population
∂(nG) Vout n Vin n in ∂n in the crystallizer (or leaving the well-mixed unit), and
+ − =− (51)
∂L VT VT ∂t experience with such measurements is invaluable in per-
Equation (51) may be used as a starting point for the forming this analysis properly. Figure 16 shows a plot of
analysis of any crystallizer that has a well-mixed active
volume and for which crystal breakage and agglomera-
tion can be ignored. As an illustration of how the equation
can be simplified to fit specific system behavior, suppose
the feed to the crystallizer is free of crystals and that it is
operating at steady state. Then, n in = 0 and ∂n/∂t = 0.
Now suppose that the crystal growth is invariant with
size and time; in other words, assume the system follows
the McCabe L law and therefore exhibits neither size-
dependent growth nor growth-rate dispersion. Then,
∂(nG) ∂n
=G (52)
∂L ∂L
Defining a mean residence time τ = VT /Vout and applying
the aforementioned restrictions leads to
dn n
G + =0 (53)
dL τ
(τ is often referred to as the drawdown time to reflect the
fact that it is the time required to empty the contents from
the crystallizer.) Integrating Eq. (53) with the boundary
condition n = n ◦ at L = 0:
 
◦ L
n = n exp − (54) FIGURE 16 Typical population density plot from perfectly mixed,
Gτ continuous crystallizer.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 113

typical population density data obtained from a crystal- are fixed, the crystal size distribution is determined in its
lizer meeting the stated assumptions. The slope of the plot entirety. In addition, such distributions have the following
of such data may be used to obtain the growth rate while characteristics:
the intercept can be used to estimate nucleation rate.
Many industrial crystallizers operate in a well-mixed or r Mass density function (from Eq. (39)):
near well-mixed manner, and the equations derived above  
L ◦
can be used to describe their performance. Also, the sim- m = ρkvol n L exp − 3
(59)
plicity of the equations describing an MSMPR crystallizer Gτ
make experimental equipment configured so as to meet r Dominant crystal size (from Eq. (40)):
the assumptions leading to Eq. (54) useful in determining
nucleation and growth kinetics. From a series of runs at L D = 3Gτ (60)
different operating conditions, correlations of nucleation r Moments of n (from Eq. (37)):
and growth kinetics with appropriate process variables can
be obtained (see, for example, the discussions of Eqs. (18) m i = i!n ◦ (Gτ )i+1 (61)
and (25)). The resulting correlations can then be used to r Total number of crystals per unit volume:
guide either crystallizer scale-up or the development of an ∞
operating strategy for an existing crystallizer. NT = n d L = m 0 = n ◦ Gτ (62)
It is often very difficult to measure supersaturation, 0
especially in systems that have high growth rates. Even r Total length of crystals per unit volume:
though the supersaturation in such systems is so small ∞
that it can be neglected in writing a solute mass balance, LT = n L d L = m 1 = n ◦ (Gτ )2 (63)
it is important in setting nucleation and growth rates. In 0
such instances it is convenient to substitute growth rate r Total surface area of crystals per unit volume:
for supersaturation by combining Eqs. (18) and (25). This ∞
gives: AT = karea n L 2 d L = karea m 2 = 2karea n ◦ (Gτ )3
0
B ◦ = knuc G i MT N k
j
(58) (64)
r Total solids volume per unit volume:
The constant knuc depends on process variables other than
∞
supersaturation, magma density, and intensity of mixing;
VTS = kvol n L 3 d L = kvol m 3 = 6kvol n ◦ (Gτ )4
these include temperature and presence of impurities. If 0
sufficient data are available, these variables may be sepa- (65)
r The coefficient of variation of the mass density
rated from the constant by adding more terms in a power-
law correlation. knuc is specific to the operating equipment function (from Eq. (42)) is 50%.
r The magma density M (mass of crystals per unit
and not transferable from one equipment scale to another. T
The system-specific constants i and j are obtainable from volume of slurry or liquor) is the product of the crystal
experimental data and may be used in scale-up, although density, the volumetric shape factor, and the third
j may vary considerably with mixing conditions. moment of the population density function:
As shown by Eq. (54), growth rate G can be obtained MT = 6ρkvol n ◦ (Gτ )4 (66)
from the slope of a plot of the log of population density
against crystal size; nucleation rate B ◦ can be obtained System conditions often allow for the measurement of
from the same data by using the relationship given by magma density, and in such cases is should be used as
Eq. (57), with n ◦ being the intercept of the population a constraint in evaluating nucleation and growth
density plot. Nucleation rates obtained by these proce- kinetics from measured population densities. This
dures should be checked by comparison with values ob- approach is especially useful in instances of
tained from a mass balance (see the later discussion of uncertainty in the determination of population
Eq. (66)). densities from sieving or other particle sizing
The perfectly mixed crystallizer of the type described techniques.
in the preceding discussion is highly constrained. Alter-
ation of the characteristics of crystal size distributions
B. Preferential Removal of Crystals
produced by such systems can be accomplished only
by modifications of the nucleation and growth kinetics As indicated above, crystal size distributions produced in
of the system being crystallized. Indeed, examination of a perfectly mixed crystallizer are constrained by the na-
Eq. (54) shows that once nucleation and growth kinetics ture of the system. This is because both liquor and solids
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

114 Crystallization Processes

have the same residence time distributions, and it is the Clear-liquor advance is used for two purposes: (1) to
crystal residence time distribution that gives the popula- reduce the quantity of liquor that must be processed by
tion density function the characteristic exponential form the solid–liquid separation equipment (e.g., filter or cen-
in Eq. (54). Nucleation and growth kinetics can influence trifuge) that follows the crystallizer, and (2) to separate
the population density function, but they cannot alter the the residence time distributions of crystals and liquor. The
form of the functional dependence of n on L. reduction in liquor flow through the separation equipment
Crystallizers are made more flexible by the introduc- can allow the use of smaller equipment for a fixed produc-
tion of selective removal devices that alter the residence tion rate or increased production through fixed equipment.
time distributions of materials flowing from the crys- Separating the residence time distributions of crystals and
tallizer. Three removal functions—clear-liquor advance, liquor means that crystals will have an average residence
classified-fines removal, and classified-product removal— time longer than that of the liquor. This should, in princi-
and their idealized removal devices will be used here to ple, lead to the production of larger crystals, but because
illustrate how design and operating variables can be ma- the crystallizer is otherwise well mixed, the crystal pop-
nipulated to alter crystal size distributions. Idealized repre- ulation density will have the same form as that for the
sentations of the three classification devices are illustrated MSMPR crystallizer (Eq. (54)).
in Fig. 17. The analysis goes as follows: Let Vin , VCL , and Vout
Clear-liquor advance from what is called a double draw- represent volumetric flow rates of clear liquor fed to the
off crystallizer is simply the removal of mother liquor crystallizer, of clear-liquor advance, and of output slurry
without simultaneous removal of crystals. The primary ac- respectively. The population density function is given by
tion in classified-fines removal is preferential withdrawal the expression:
from the crystallizer of crystals of a size below some spec-  
ified value; this may be coupled with the dissolution of the ◦ L
n = n exp − (67)
crystals removed as fines and the return of the resulting Gτprod
solution to the crystallizer. Classified-product removal is
carried out to remove preferentially those crystals of a where τprod = VT /Vout . Increasing VCL decreases Vout and
size larger than some specified value. In the following thereby increases the residence time of the crystals in the
discussion, the effects of each of these selective removal crystallizer. Unless the increase in magma density results
functions on crystal size distributions will be described in in significant increases in nucleation, the utilization of
terms of the population density function n. Only the ideal clear-liquor advance will produce an increase in the dom-
solid–liquid classification devices will be examined. It is inant crystal size. Often the increase is much greater than
convenient in the analyses to define flow rates in terms that predicted from theory, and it is suspected that this is
of clear liquor. Necessarily, then, the population density because the stream being removed as clear liquor actu-
function is defined on a clear-liquor basis. ally contains varying amounts of fines. If this is the case,

FIGURE 17 Schematic representations of idealized removal functions. (a) Clear-liquor advance, (b) classified-fines removal, (c) classified-
product removal.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 115

both clear-liquor advance and fines-removal are enhancing

crystal size.
As an idealization of classified-fines removal, assume
that two streams are withdrawn from the crystallizer, one
corresponding to the product stream and the other a fines-
removal stream. Designate the flow rate of the clear so-
lution in the product stream to be Vout and set the flow
rate of the clear solution in the fines-removal stream to be
(R − 1)Vout . Also, assume that the device used to separate
fines from the larger crystals functions so that only crystals
below an arbitrary size L F are in the fines-removal stream
and that all crystals below size L F have an equal probabil-
ity of being withdrawn as fines. Under these conditions,
the crystal size distribution is characterized by two mean
residence times, one for the fines and the other for crystals
FIGURE 18 Population density plot for product from crystallizer
larger than L F , that are related by the equations: with idealized classified-fines removal.
VT
τ = (for L > L F ) (68)
Vout scale systems, by simply inserting a length of pipe or
tubing of appropriate diameter into the well-mixed crys-
VT τ tallizer chamber. The separation of crystals in the settling
τF = = (for L ≤ L F ) (69) zone is based on the dependence of settling velocity on
RVout R
crystal size. Crystals entering the settling zone and hav-
where VT is the total volume of clear solution in the crys-
ing a settling velocity greater than the upward velocity of
tallizer.
the slurry remain in the crystallizer. As the cross-sectional
For systems following invariant growth, the crystal pop-
area of a settling zone is invariant, the flow rate of slurry
ulation density in each size range will decay exponentially
through the zone determines the cut-size L F , and it also
with the inverse of the product of growth rate and residence
determines the parameter R used in Eqs. (69) through (71).
time. For a continuous distribution, the population densi-
In a crystallizer equipped with classified-product re-
ties of the classified fines and the product crystals must be
moval, crystals above some coarse size L C are removed
the same at L = L F . Accordingly, the population density
at a rate Z times the removal rate of smaller crystals. This
for a crystallizer operating with classified-fines removal
can be accomplished by using an elutriation leg, a hydro-
is given by:
cyclone, or a screen to separate larger crystals for removal
RL from the system. Using the analysis of classified-fines re-
n = n ◦ exp − (for L ≤ L F ) (70) moval as a guide, it can be shown that the crystal popula-
Gτ
tion density is given by the equations:
(R − 1)L F L L
n = n ◦ exp − exp − (for L > L F ) n = n ◦ exp − (for L ≤ L C ) (72)
Gτ Gτ Gτ
(71)
(Z − 1)L C ZL
n = n ◦ exp exp − (for L > L C )
Figure 18 shows how the population density function Gτ Gτ
changes with the addition of classified-fines removal. The
(73)
lines drawn are for a hypothetical system, but they illus-
trate qualitatively what can be demonstrated analytically; where τ is defined as the residence time VT /Vout . Figure 19
that is, fines removal increases the dominant crystal size, shows the effects of classified-product removal on crys-
but it also increases the spread of the distribution. tal size distribution; the dominant crystal size is reduced
A simple method for implementation of classified-fines and the spread of the distribution becomes narrower. Note
removal is to remove slurry from a settling zone in the that it is impossible for crystals smaller than L C to leave
crystallizer. The settling zone can be created by construct- the idealized classified-product crystallizer illustrated in
ing a baffle that separates the zone from the well-mixed Fig. 17c. Accordingly, the population densities shown on
portion of the vessel—recall, for example, the draft-tube- Fig. 19 for the classified-product crystallizer represent
baffle crystallizer described in Section V—or, in small- conditions inside the perfectly mixed region of the unit.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

116 Crystallization Processes

surfaces or blinding of screens. In addition, classified-

product removal can lead to cycling of the crystal size dis-
tribution. Often such behavior can be minimized or even
eliminated by increasing the fines-removal rate.
Moments of the population density function given by
Eqs. (74) through (76) can be evaluated in piecewise
fashion:
LF LC ∞
mi = Li n d L + Li n d L + L i n d L (77)
0 LF LC

Equation (77) is used to estimate the moments of the

population density function within the crystallizer, not
of the product distribution. (Recall that moments of the
distribution within the crystallizer are often required for
kinetic equations.) Assuming perfect classification, mo-
FIGURE 19 Population density plot for crystals in crystallizer with
ments of the product distribution can be obtained from the
idealized classified-product removal.
expression:
∞
If both fines and product are removed on a classified ba- m i,prod = Li n d L (78)
LC
sis, the population density will be given by the equations:
Moments can be used to characterize the material pro-
◦ RL duced from or contained in a crystallizer with classified-
n = n exp − (for L ≤ L F ) (74)
Gτ fines or classified-product removal or to evaluate the effect
(R − 1)L F L of these selective removal functions on product character-
n = n ◦ exp − exp − istics. All that is required is the use of the equations derived
Gτ Gτ
earlier to relate special properties, such as coefficient of
(for L F < L < L C ) (75) variation to the operational parameters R and Z .
(R − 1)L F (Z − 1)L C
n = n ◦ exp − exp
Gτ Gτ C. Batch Crystallization
ZL As with continuous crystallizers, the mode by which su-
× exp − (for L ≥ L C ) (76)
Gτ persaturation is generated affects the crystal yield and size
Selection of a crystallizer that has both classified-fines distribution; however, it is the rate at which such super-
and classified-product removal is done to combine the saturation is generated that is most important in determin-
best features of each: increased dominant size and nar- ing product characteristics. Furthermore, there are infinite
rower distribution. Figure 20 illustrates the effects of
both removal functions on population density. Note that
this plot of population density results from sampling the
magma within a crystallizer, not from sampling the prod-
uct stream, which for the ideal classification devices con-
sidered here can only have crystals larger than L C . As
discussed earlier for the classified-product crystallizer,
the population densities shown in Fig. 20 represent those
found in the crystallizer.
The model of the crystallizer and selective removal de-
vices that led to Eqs. (74) through (76) is referred to as
the R-Z crystallizer. It is an obvious idealization of actual
crystallizers because of the perfect cuts assumed at L F and
L C . However, it is a useful approximation to many systems
and it allows qualitative analyses of complex operations.
Although many commercial crystallizers operate with
some form of selective crystal removal, such devices can FIGURE 20 Population density plot for crystals in crystallizer with
be difficult to operate because of fouling of heat-exchanger idealized classifiedfines and classified-product removal.
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 117

possibilities in selecting cooling profiles, T (t), or vapor The identification of an initial condition associated with
generation profiles, V (t), or time dependencies of precip- the crystal size distribution is very difficult. If the system
itant or nonsolvent addition rates. is seeded, the initial condition becomes:
For illustrative purposes, consider that the protocol for
a cooling crystallizer can involve either natural cooling— n̄(L , 0) = n̄ seed (L) (82)
cooling resulting from exposure of the crystallizer con- where n̄ seed is the population density function of the seed
tents to a heat sink without intervention of a control crystals. If the system is unseeded, the nuclei often are
system—or manipulation of cooling to reduce the sys- assumed to form at size zero.
tem temperature in a specific manner. In both cases, the The rate of cooling, or evaporation, or addition of dilu-
instantaneous heat-transfer rate is given by: ent required to maintain specified conditions in a batch
Q = U A(T − Tsink ) (79) crystallizer often can be determined from a population-
balance model. Moments of the population density func-
where U is a heat-transfer coefficient, A is the area avail- tion are used in the development of equations relating the
able for heat transfer, T is the temperature of the magma, control variable to time. As defined earlier, the moments
and Tsink is the temperature of the cooling fluid. If Tsink is are
a constant, the maximum heat-transfer rate and, therefore, ∞
the highest rate at which supersaturation is generated are mi = L i n̄ d L (83)
0
at the beginning of the process. This protocol can lead to
excessive primary nucleation and the formation of encrus- Recognizing that the zeroth moment is the total number
tations on the heat-transfer surfaces. of crystals in the system, it can be shown that:
The objective of programmed cooling is to control the dm 0 d NT
rate at which the magma temperature is reduced so that su- = n̄ ◦ G = B ◦ = (84)
dt dt
persaturation remains constant at some prescribed value,
usually below the metastable limit associated with pri- Moment transformation of Eq. (80) leads to the following
mary nucleation. Typically the batch is cooled slowly relationship:
at the beginning of the cycle and more rapidly at the ∂m j
end. An analysis that supports this approach is presented = j Gm j−1 (85)
∂t
later. In size-optimal cooling, the objective is to vary
the cooling rate so that the supersaturation in the crys- Combining Eq. (85) with the relationships of moments
tallizer is adjusted to produce an optimal crystal size to distribution properties developed in Section VI.A for
distribution. j = 1, 2, 3 gives:
Protocols similar to those described above for cool- dm 1 m 0 =NT d L T
ing crystallizers exist for crystallization modes involving = Gm 0 −→ = G NT (86)
dt dt
evaporation of solvent and the rate at which a non solvent
dm 2 m 1 =L T d AT
or a reactant is added to a crystallizer. = 2Gm 1 −→ = 2Gkarea L T (87)
A population balance can be used to follow the devel- dt dt
 
opment of a crystal size distribution in batch crystallizer, dm 3 karea m 2 =AT d MT kvol
= 3Gm 2 −→ = 3Gρ AT (88)
but both the mathematics and physical phenomena being dt dt karea
modeled are more complex than for continuous systems
at steady state. The balance often utilizes the population where NT is the total number of crystals, L T is total crystal
density defined in terms of the total crystallizer volume, length, A T is total surface area of the crystals, and MT is
rather than on a specific basis: n̄ = nVT . Accordingly, the the total mass of crystals in the crystallizer. In addition to a
general population balance given by Eq. (51) can be mod- population balance, a solute balance must also be satisfied:
ified for a batch crystallizer to give: d(VT C) d MT
+ =0 (89)
∂(nVT ) ∂(GnVT ) ∂ n̄ ∂(G n̄) dt dt
+ = + =0 (80)
∂t ∂L ∂t ∂L where VT is the total volume of the system, and C is solute
concentration in the solution.
The solution to this equation requires both an initial con-
The above equations can be applied to any batch crys-
dition (n̄ at t = 0) and a boundary condition (usually ob-
tallization process, regardless of the mode by which su-
tained by assuming that crystals are formed at zero size):
persaturation is generated. For example, suppose a model
B ◦ (t) is needed to guide the operation of a seeded batch crys-
n̄(0, t) = n̄ ◦ (t) = (81) tallizer so that solvent is evaporated at a rate that gives
G(0, t)
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

118 Crystallization Processes

a constant crystal growth rate G and no nucleation; in an appropriate dependence of system temperature on time.
other words, supersaturation is to be held constant and The result depends upon the relationship of solubility to
only those crystals added at the beginning of the run are in temperature. If that relationship is linear, the cooling rate
the crystallizer. Model development proceeds as follows: varies with time in a parabolic manner; i.e.,
combining the solute balance, Eq. (89), with Eq. (88), dT
− = C1 t 2 + C2 t + C3 (100)
d(VT C) 3ρ AT kvol G dt
+ =0 (90)
dt karea An approximation to the temperature–time relationship
Recognizing that the process specification requires C to that serves as a good starting point for establishing a fixed
be a constant and taking the derivative of Eq. (90): protocol is given by:
   3
d 2 VT kvol d AT t
C + 3ρ G =0 (91) T = T0 − (T0 − Tfinal ) (101)
dt 2 karea dt τ
where τ is the overall batch run time.
⇓ Eq. (87)
It is clear that stringent control of batch crystallizers
is critical to obtaining a desired crystal size distribution.
d 2 VT
C + 6ρkvol G 2 L T = 0 (92) It is also obvious that the development of a strategy for
dt 2 generating supersaturation can be aided by the types of
Taking the derivative of the last equation: modeling illustrated above. However, the initial conditions
in the models were based on properties of seed crystals
d 3 VT d LT
C 3
+ 6ρkvol G 2 =0 (93) added to the crystallizer. In operations without seeding,
dt dt initial conditions are determined from a model of primary
⇓ Eq. (86) nucleation.

d 3 VT D. Effects of Anomalous Growth

C + 6ρkvol G 3 NT = 0 (94)
dt 3
Throughout this section, crystals have been assumed to
Suppose that the batch crystallizer is seeded with a mass grow according to the McCabe L law. This has simpli-
of crystals with a uniform size of L̄ seed . The number of seed fied the analyses of both continuous and batch crystal-
crystals is Nseed , and, as the operation is to be free from lizers and, indeed, crystal growth often follows the L
nucleation, the number of crystals in the system remains law. However, as outlined in Section III, size-dependent
the same as the number of seed crystals. The initial val- growth and growth-rate dispersion contribute to devia-
ues of total crystal length, total crystal surface area, total tions from the models developed here. Both of these phe-
crystal mass, and system volume are nomena lead to similar results: In continuous, perfectly
L T (0) = Nseed L̄ seed (95) mixed crystallizers, the simple expression for population
density given by Eq. (54) is no longer valid. Both size-
AT (0) = karea Nseed L̄ 2seed (96) dependent growth and growth-rate dispersion due to the
MT (0) = ρkvol Nseed L̄ 3seed (97) existence of a random distribution of growth rates among
crystals in a magma lead to curvature in plots of ln n vs.
VT (0) = VT0 (98) L. Models for both causes of this behavior exist but are
On integrating Eq. (94), the following dependence of sys- considered beyond the scope of the present discussion. In
tem volume on time can be obtained: batch crystallization, the effects of anomalous growth lead
to a broadening of the distribution, as was illustrated in
C(VT0 − VT ) = kvol ρ Nseed (Gt)3 + 3(Gt)2 L̄ seed Fig. 6.
+ 3(Gt) L̄ 2seed (99)
E. Summary
Therefore, for the specified conditions, the evaporation
rate (−d VT /dt) is a parabolic (second-order) function of The discussion presented here has focused on the princi-
time, and the rate of heat input to the crystallizer must be ples associated with formulating a population balance and
controlled to match the conditions called for by Eq. (99). applying simplifying conditions associated with specific
If a cooling mode is used to generate supersaturation, an crystallizer configurations. The continuous and batch sys-
analysis similar to that given above can be used to derive tems used as examples were idealized so that the principles
P1: GAE/LSK P2: FLV Final Pages
Encyclopedia of Physical Science and Technology EN004D-ID159 June 8, 2001 15:47

Crystallization Processes 119

could be illustrated, but the concepts can be applied to BIBLIOGRAPHY

more complicated configurations. Additionally, there has
been a growing body of work on aspects of population Moyers, G. C., and Rousseau, R. W. (1986). In “Handbook of Separation
balance formulation that greatly extends the ability to de- Process Technology” (R. W. Rousseau, ed.), Wiley, New York.
scribe complex systems. Such work has involved anoma- Mullin, J. W. (1993). “Industrial Crystallization,” 3rd ed. Butterworth-
Heinemann, London.
lous crystal growth, crystal agglomeration, and crystal
Myerson, A. S. (1993). “Handbook of Industrial Crystallization,”
breakage and necessarily results in substantially more Butterworth-Heinemann, London.
complex models. Randolph, A. D., and Larson, M. A. (1988). “Theory of Particulate
Processes,” 2nd ed. Academic Press, San Diego, CA.
Rousseau, R. W. (1993). In “Kirk-Othmer Encyclopedia of Chemical
SEE ALSO THE FOLLOWING ARTICLES Technology,” Vol. 7, 4th ed., pp. 683–730, John Wiley & Sons, New
York.
CRYSTAL GROWTH • CRYSTALLOGRAPHY • PRECIPITA- Rousseau, R. W. (1997). In “Encyclopedia of Separation Technology,”
Vol. 1 (D. M. Ruthven, ed.), pp. 393–439, Wiley Interscience, New
TION REACTIONS • SEPARATION AND PURIFICATION OF
York.
BIOCHEMICALS • SOLID-STATE CHEMISTRY • X-RAY Tavare, N. S. (1995). “Industrial Crystallization: Process Simulation,
ANALYSIS Analysis and Design,” Plenum, New York.