(12) Ruckenstein, E., IXD. ENG. CHEM.

FUNDAMENTALS
3, 260 (18) Toomey, R. D., Johnstone, H. F., Chem. Eng. Progr. 48, 220
(1964). (1952).
( 1 3 ) Ruckenstein, E., Rets. Chim. (Bucharest) 11, 721 (1960). E. RUCKENSTEIIS
(14) Ruckenstein, E., Zh. Prikl. Khim. 35, 70 (1962). Polytechnical Institute
(15) Simpson, H. C., Rodger, B. \V., Chem. Eng. Sci.16, 153 (1962). Bucharest, Rumania
(16) Sutherland, K. S., Trans. Znst. Chem. Engrs. (London) 39, RECEIVED
for review June 8, 1964
188 (1961). RESVBMITTED March 25, 1965
(17) Teoreanu, I., Rei'. Chirn. (Bucharest) 11, 691 (1960). ACCEPTED November 5, 1965

PARAMETRIC PUMPING: A DYNAMIC PRINCIPLE

FOR SEPARATING FLUID MIXTURES
Alternating axial displacement of a fluid mixture in a column of adsorptive particles upon which an axial
temperature difference is imposed leads, through coupling of oscillatory thermal and mass fields with
alternating flow displacements, to a difference in limiting-condition, time-average compositions a t the column
ends. The separation takes place at the expense of thermal energy; continuous operation with a fixed-bed
adsorber also becomes feasible. The separation has been substantiated by experiment and b y analysis of
the mathematical formulation.

NTEREST in separation principles and techniques is common

I to many areas of application and research-for example, in
the saline water problem, in the intercellular transport of ions
in living cells, and in the separation of all manner of chemical
mixtures of industry.
This communication is restricted to the elements of separa-
tion by means of parametric pumping.

Parametric Pumping
T h e application of dynamic adsorption principles for sepa-
rating the components of a homogeneous binary fluid mixture
is illustrated through the experimental elements depicted in
Figure 1. A column containing a bed of porous, particulate
adsorptive material and a charge of a fluid mixture is equipped
a t its ends by driving and driven pistons acting in tandem.
The pistons cause relative position displacements to take place
bet\\een the column of particles and the column of ambient Figure 1 . Elementary arrangement for closed-system
fluid. As portrayed, the system is closed and its total volume is separation b y parametric pumping
constant. Initially let each phase be uniform in solute con-
centration, let the temperature be uniform throughout the HOT FEED HOT PRODUCT
column, and permit a solute concentration equilibrium to
become established betneen fluid and solid phases. There- I f
upon a t a starting time, to, displacement alternations are initi-
ated and a thermodynamic gradient is imposed on the column,
in this case, by bringing fluid temperatures a t the column
ends to different constant values through the use of heat sources
and sinks. I t has been shown experimentally and theoret-
ically that after the above nonsymmetrical process arrange-
ment has been initiated and continued to limiting conditions-
i.e., until time-averaged values of all properties have become
constant-an axial fluid-phase time-average solute composition
gradient will have developed. T h e resulting difference in I
interparticle solute concentration a t the column ends, which C O L D PRODUCT COLD FEED
arises from coupled heat and mass transfer processes within the
bed, is the separation in question. FLOW TRANSFER FLOW TRANSFER

I n common \z ith other separation processes, individual A B C 0

columns may be used as batch separators with total reflux,
as shown in Figure 1, as continuous-flow, open-system single
columns with various arrangements for feed introduction and
product withdrawal, and finally as elements in multicolumn
arrays; different system arrangemmts lead to a range
of separation potentials and thermal efficiencies. Figure 2
presents a n open-system arrangement.
Figure 2. Elementary arrangement for open-system
separation by parametric pumping
By way of further amplification consider the following cycle :
A fluid volume is displaced downward, raising the temperature
of the adjacent adsorbent. As a result of the temperature
change, the adsorbent transfers solute to the fluid. The

VOL. 5 NO. 1 F E B R U A R Y 1966 141

enriched (and cooled) fluid next is displaced axially upward
to come in contact with warmer adsorbent. The fluid cools
the adsorbent now adjacent and loses solute to it. The trans-
port of fluid back to the point of origin completes the cycle.
The difference between the fluxes of solute in the two flow
directions is the net flux due to parametric pumping. In
closed systems the net flux approaches zero as limiting condi-
tions are reached.
A central concept of parametric pumping thus comprises
the coupling of oscillatory temperature and composition fields
with alternating flow displacements. For a given system the
time-phase relations among the fields and displacements deter-
mine the magnitude of the separation.
Experimental
In the elementary open-system experimental arrangement
(illustrated in Figure 2) hot and cold feed of 0.2% NaCl
concentration was introduced alternately into the opposite
ends of a column of Rohm and Haas Amberlite resins (equal
volumes of IRC-50 and IR-45). The pulse length was equal
to the column length-i.e., enough feed was pumped into a
column end each half cycle to replace the interstitial liquid.
Temperatures at the column ends were 90' and 40" C . ; the
total cycle time was 40 minutes. The cycle was nonsymmetric,
with a waiting period of 6 minutes a t the higher temperature,
30 minutes at the lo\\er, and tivo 2-minute periods for fluid
displacement. At long-time, limiting conditions the concen-
tration ratio bet\veen product streams \\Vas 1.2.
S o attempt was made to optimize operating conditions-
for example, to achieve maximum separation or to diminish
cycle time.
Theoretical study of open and closed systems has supported
experimental findings in that finite, limiting-condition separa-
tions may be achieved and that the same separation is reached
whether the limit is approached from above or below.
Mathematical Model
For column operation a minimum-essential formulation for
parametric pumping separation is stated in these dimensionless
equations:
HEATBALANCES
af(t)-
ae, + be,
- + p be2 = 0
az dt at
30,
- $L y(0, - e,) = 0
at
MASSBALAXES
2 + A(+*
3
dt
- 4,) = 0
EQUILIBRIUM
Q* = B + Des + Q+s + R+ses + $42 + uBs2 f .. .
Equations 1 and 2 are dimensionless heat balances in time
and position for interparticle and intraparticle space, the
corresponding dimensionless temperatures being 0, and 8,
(see Figure 1). Dimensionless temperatures are defined to
vary between 0 and 1 within the temperature limits of the
system. Variables t and z are dimensionless time and axial
position.
Equations 3 and 4 are dimensionless mass balances, inter-
particle and intraparticle concentrations being $ and +,,
respectively. The fact that the mass balances are here written
for only one component assumes that the solute is present in
142 I&EC FUNDAMENTALS
dilute concentration. In general, there would be n mass
balances for n components.
The first term of Equation 1 is the axial fluid phase convec-
tive heat transfer rate, the second iq the fluid phase transient,
and the third is the solid phase transient. The two tempera-
ture variables are linked through Equation 2, which is a simpli-
fied rate equation for transfer of heat to and from particles.
The terms in Equations 3 and 4 have meanings analogous to
those in Equations 1 and 2.
Both thermal and compositional field equations are "driven"
by the periodic velocity, af(t). I t is essential that this driving
coefficient of the derivatives a(
)/atchange in sign periodically.
Anf(t) is chosen which is optimal for a given system.
The coupling between the set of thermal fields (Equations 1
and 2) and the set of compositional fields (Equations 3 and 4)
is accomplished through the equilibrium relationship, Equa-
tion 5 . The equilibrium fluid-phase composition, $*, is a
function of both Os and Os. the composition and temperature,
respectively, of the local solid phase. The "pump" is identi-
fied as the oscillatory thermal field. The larger the change in
equilibrium composition produced by a change in tempera-
ture, the larger will be the separation achieved by the pumping
effect. The fact of separation does not depend upon non-
linearity in the equilibrium relationship.
Separation as well as thermal efficiencies depend upon
these variables:
a f ( t ) . Velocity of fluid. For a given f ( t ) , a determines
the displacement amplitude, which may be larger or smaller
than the column length.
6. Ratio of the average volumetric heat capacity of the
adsorbent phase to that of the fluid phase. This constant
has a value close to 1.O for a liquid in a densely packed column
but is larger \$hen a gaseous mixture is the fluid phase.
Gaseous separators are more difficult to design because of this
greater imbalance betxveen fluid and solid volumetric heat
capacities.
K. Ratio of volume of the adsorbent phase to that of the
fluid phase.
l/y. Time constant for thermal response to adsorptive
particles relative to the characteristic time of displacement.
1 / X . Time constant for mass transfer response of adsorptive
particles compared to the characteristic time of displacement.
B, D, Q, R, S, U. Empirical constants in equilibrium
Equation 5.
T o complete the mathematical formulation, initial and
boundary conditions are required. These conditions for vari-
ous column and cascade arrangements differ in detail. Gener-
ally, fluxes of mass entering a column during one-half cycle will
depend upon the output during the previous half cycle from
(3) the same or adjacent column. The boundary and initial
conditions in the open system depicted in Figure 2 are:
(4) BOUNDARY
CONDITIONS
Whenf(t) < 0: 8, = 1.0, z = 1.0
4, = 1.0, 2 = 1.0
(5)
Whenf(t) 2 0: O f = 0,z = 0
$f = 1.0, z = 0
INITIAL
CONDITIONS
(for example)
F o r t = 0 , 0 < z < 1.0
0, = 8, = 0
+, = 1.0
$s is chosen to be in equilibrium with $I = 1.0 at
temperature Os = 0.
Two refinements of the above model deserve mention.
T h e first is the use of more detailed rate equations than Equa-
tions 2 and 4 . For ease of initial analysis these are now written discussions with Murray McAndrew and Roger Rolke are
as simplified rate equations. When necessary, the complete acknowledged.
intraparticle diffusion a:nd interphase transport equations for Current work is supported through Grant GP-2286 of the
heat and mass should be used. The second refinement is the National Science Foundation, Engineering Division. An
inclusion of diffusional and dispersional transport of heat and initiation grant from the Research Board of Princeton Uni-
mass in the axial direction, governed by terms of the form versity is acknowledged, as is also the use of a digital computer
6 [a2(),‘bt2]to be placed in Equations 1 and 3. in the Princeton University Computer Center which is sup-
ported in part by K’ational Science Foundation Grant NSF-
Discussion GP579. Alan Rice held a National Science Foundation
Postgraduate Fellowship during part of the study.
Separation of mixture:s by parametric pumping in principle
has \Tide applicability. Thus the class of appropriate fluids
might include a variety of adsorbable homogeneous liquid and
gaseous mixtures. T h e list of known useful adsorbents is long;
all of them presumably have a finite temperature coefficient of Nomenclature
adsorption. Examples of adsorbents are silica gel, adsorbent
A = characteristic constant of interparticle fluid
carbons, ion exchange resins, molecular sieves, and chroma- velocity-for example, if u = urnaxsin ut’, A =
tographic solids, such a’s partitioning liquids carried in a fixed Z’mx, LllT
matrix of kieselguhr. 0
B = a constant in Equation 5, dimensionless
Parametric processes in general involve two or more alter- c* = concentration in fluid that would be in equilib-
nating fields that arc time-coupled in some defined fashion. rium with adsorbent, M / L f 3
Most previously described c,oupled parametric processes-such c, = concentration in fluid, M / L j 3
as the parametric voltage amplifier ( 7 , p. 92), for ex- CS = concentration in adsorbent, M / L,3
CO = reference concentration, fluid (initial concentra-
ample-are formulated through equations of motion and
tion in closed system, feed concentration in open
energy and commonly involve one independent variable, time system), M / L I 3
or position. I n the present instance the equations are mass G f = specific heat of interparticle fluid, H / M T
and enthalpy balances and the corresponding fields are CP, = specific heat of adsorbent, H / M T
“driven” simultaneously in time by means of the alternating D = a constant in Equation 5, dimensionless
fluid displacements. T h e independent variables, however, f 0) = see a f ( t )
hh = volumetric heat transfer coefficient, H/L,3 T T
are both time and position, leading to the sets of partial differen- h, = volumetric mass transfer coefficient, L,3/L,37
tial equations given above. Q = a constant in Equation 5, dimensionless
A general definition of parametric pumping will not be R = a constant in Equation 5, dimensionless
S = a constant in Equation 5, dimensionless
attempted, but certain characteristics are suggested.
T/ = temperature of fluid, T
The system is defined \vith respect to a fixed set of coordi- Ts = temperature of adsorbent, T
nates. T h e driving coefficient changes sign periodically and as i. = temperature of heat source, T
a net result the dependent variables vary periodically. T h e
system as a whole proceeds from a n initial condition to a final, T = temperature of heat sink, T
limiting condition a t \vhich the dependent variables continue t’ = time, T
to vary- periodically thereafter. t = ut’, time, dimensionless
Most commonly in electrical and mechanical application u = a constant in Equation 5, dimensionless
(7), as \\-ell as in the present \vork, there are a t least two coupled
fields defined by two sets of differential equations for the de-
V = V, + V,. total volume of bed, L t
pendent variables.
v, =
=
interparticle volume in bed, L,3
particle volume in bed, L,3
vs
U = velocity of interparticle fluid, L J T
T h e thermal parametric pumping principle here presented 2 = z‘/Z, distance from bottom of bed, dimensionless
suggests that, by way of generalization, analogous separation 2‘ = distance from bottom of bed, L 1
arrangements using other thermodynamic gradients than Z = bed length, L 1
temperature are possible. T h e driving coefficient need not be
velocity as in the present case but might, for example, be a n UNITS
alternating voltage; and coefficients now indicated as con-
L = length
stants may be caused to be periodic. M = mass
I t is instructive to compare thermal parametric pumping H = quantity of heat
with the classical thermal diffusion of Clusius and Dickel, since T = temperature
both processes require a n over-all temperature gradient. T = time
Separation in thermal diffusion is rooted in couplings a t the
molecular level in single-phase fluids. The mechanisms seem SUBSCRIPTS
not to be clearly identified. By contrast, separation in thermal
parametric pumping depends upon multiphase, macroscopic f = fluid phase (interparticle space)
S = adsorbent phase (intraparticle space)
heat and mass transfer processes which are relatively well t = both phases together, fluid and adsorbent
understood.
GREEK
Acknowledgment
01 = 2A/uZ, displacement amplitude, dimensionless
\$’e thank Donald Jost and Alan Bridge for assistance in af(t) = sign-changing fluid velocity function, dimension-
experimental work. Peter Halbert and Leslie Mitchell, less
Electronic Associates, Inc., Princeton, N. J., helped a t a n P = psCpse,/pjCp,ef, dimensionless
Y = hh/upsCps,dimensionless
early stage to elucidate properties of the mathematical formu- 6 = coefficient of axial diffusion or dispersion, dimen-
lation through the use of a n analog computer. Valuable sionless

VOL. 5 NO. 1 FEBRUARY 1 9 6 6 143

 = V f / V ,L13/L13 Literature Cited
= V J V , L,3/L,3
= € $ / E / , L,3/Lf3
(1) Louisell, W. H., “Coupled Mode and Parametric Electronics,”
Wiley, New York, 1960.
= h,/w, L/3/L,3
0 1 0
= ( T , - T ) / ( T - T ) , dimensionless R . H. W I L H E L M
0 1 0 A. W. R I C E
= ( T , - , T ) / ( T - T ) , adsorbent temperature, di- A. R. BENDELIUS‘
mensionless
= C,/C,, concentration in fluid, dimensionless Princeton Unioersity
= C,/COj concentration in adsorbent, LJ3/L,3 Princeton, N . J .
= C*/Co, concentration in fluid that would be in
equilibrium with &, dimensionless RECEIVED
for review August 16, 1965
ACCEPTED December 6, 1965
P/ = density of interparticle fluid, M:/Lj3
P. = density of adsorbent particles, M I L S 3 1 Present address, Pfister Chemical LVorks, Inc., Ridgefield,
w = time scale, 1/7 N. .r.

CORRESPONDENCE
BINARY DIFFUSION COEFFICIENTS IN
NONPOLAR GASES

SIR: The paper by Holsen and Strunk (2) presents new
diffusion data for some binary systems near room temperature
as well as an interpretation of the difficulties encountered in
attempting to predict diffusivities from Lennard-Jones corre-
lations of viscosity measurements. We find the paper mis-
leading in several respects, and we would like to try to clarify
these points.
In one step of their procedure Holsen and Strunk attempt,
without success, to find “unique” Lennard- Jones parameters
for their data. Because the Lennard-Jones potential is itself
not physically unique and because experimental values are not
unique numbers, one cannot expect to find a unique set of
parameters. There is, instead, a region of values of u and
e / k which will reproduce data within the experimental error.
The regions peculiar to diffusivity data and to viscosity data
may or may not intersect. When they do not intersect, there
is indeed difficulty in reconciling correlations of the two proper-
ties.
The regions of correlation are not ordinarily revealed
accurately by graphical procedures such as the Buckingham
plot illustrated in Holsen and Strunk’s Figure 1. Rather one
needs to use a careful numerical search procedure, based on a
least-error criterion. M’e have carried out such a procedure
(3) using Holsen and Strunk’s data as well as some others,
with the results listed in Table I. I t is evident from the mean
deviations shown that effective correlation has been achieved.
Holsen and Strunk‘s data appear to be somewhat more precise
than most results for diffusion, which commonly exhibit cor-
relative deviations of 1 to 374.
We have also shown in Table I our correlation of the results
of Saxena and Mason (4)for the helium-argon system. T h e
two sets of data are in good agreement. There is apparently
no reason for discarding any of their own He-Ar data, as
Holsen and Strunk suggest may be necessary.
The parameters e l k = 35.6’ K. and u = 2.987 A. provide
a n excellent fit of available viscosity measurements with
helium-argon. Table I illustrates that if uncertainties in the
measurement of diffusivities are of the order of 1.7 to 3.0%,
a realistic range, viscosity, and diffusivity can be successfully
correlated by the same set of L-J parameters for the tempera-
tures a t which Holsen and Strunk worked.
M’hen there is a t least one polyatomic species in the mixture,
predictions of diffusivities from viscosities will be considerably
in error. This is illustrated for the COrargon system in
Table I. For the self-diffusion coefficient of C O P even larger
discrepancies, of the order of 3074, are observed (7). Con-
trary to the assumptions of Holsen and Strunk, however, these
discrepancies increase monotonically with temperature above
about 300’ K. Walker and b’estenberg ( 5 ) observe that this
is also true for the helium-argon system. They conclude their

Table 1. Force Constants for Lennard-Jones ( 1 2-6) Potential

N o . of Points Mean
System and Reference Temp. Range, ’ K. elk, ’ K. u, A . Deviation, %
He-Ar 3 (2) 276-346 25 3.100 0.7
14 ( 4 ) 251-418 69 2.774 1.9
276-346 69 2.774 2.6
276-346 35.65 2,987a 1.7
251-418 35. ba 2.987a
~ 3.0
He-air 276-346 40 3.060 1.1
He-COZ 276-346 45 3.240 0.6
Air-COz 276-317 25 4. . 4411
. ._ 0.4
COz-Ar 276-317 21 4. . 448
. ._ 0.5
276-317 277 3.208 2.5
276-1798 277 3.208 2.8
276-1 798 158a 3. 665a 13.6
a Parameters f r o m aiscosity data.

144 l&EC FUNDAMENTALS