Chemical Engineering Science 56 (2001) 6385–6391

www.elsevier.com/locate/ces

Studies of gas–liquid $ow through reactors internals using

VOF simulations
Ludovic Raynal ∗ , Isabelle Harter
Institut Francais du Petrole, BP 3, 69390 Vernaison, France

Abstract
Two-phase $ow through reactor internals have been experimentally and numerically studied. Experiments have been carried out
with a setup running under ambient pressure for two con3gurations. The 3rst con3guration consists of a mixing box ori3ce inlet
through which liquid $ows as a 3lm sheared by a gas $ow. The liquid height at ori3ce inlet is documented over a wide range
of liquid and gas $owrates. The second con3guration consists of the two-phase $ow through a downcomer of a distributing tray.
Two and three dimensional computational $uid dynamic (CFD) simulations using the volume of $uid approach have been used
to compute both $ows for similar $ow conditions as used in the experiments. It is shown that the agreement between experiments
and calculations is very good. Based on this good agreement, it is 3nally discussed how CFD can be used to achieve better design
rules for gas liquid reactor internals via simulations carried out for industrial process conditions. ? 2001 Elsevier Science Ltd. All
rights reserved.

Keywords: Hydrodynamics; Multiphase $ow; Modeling; Computational $uid dynamics (CFD); VOF method; Multiphase reactor

1. Introduction $uid dynamic (CFD) simulations using the volume of

$uid (VOF) approach can be used to 3ll the gap between
Many oil re3ning processes involve gas–liquid $ow cold $ow experiments and real process conditions.
under severe pressure and temperature conditions (e.g., In the last decade, CFD has been considered as a
hydrotreating or acid gas treating). Optimization of powerful tool to help chemical engineering development
such processes is possible through the combined im- (Trambouze, 1996; Kuipers & van Swaaij, 1997). CFD
provement of the catalyst e:ciency, the quench thermal models have been signi3cantly enhanced and calcula-
performance and the gas–liquid distribution device per- tion speeds have greatly increased, so that CFD has ac-
formance. Due to the tightening of the speci3cations, tually been more and more used to simulate hydrody-
concerning diesel sulfur content for example, the design namics in complex $ows such as two-phase $ow bubble
of the latter internals is becoming a crucial issue (Swain columns (Krishna, Urseanu, Van Baten, & Ellenberger,
& Zonnevylle, 2000). The liquid distribution problem 1999, Sanyal, Vasquez, Roy, & Dudukovic, 1999) or $u-
has thus already been considered through detailed in- idized beds (Neri & Gidaspow, 2000). This numerical
vestigations (e.g. Ouwerkerk, Bratland, Hagan, Kikkert, tool is now widely used and its advantages and limi-
& Zonnevylle, 1999). However, the design of reactor tations are more clearly identi3ed (Sundaresan, 2000).
internals (distributing and contacting devices) is com- However, most of the work concerning two-phase $ows
monly extrapolated from experiments carried out at small has been focused on the “heart” of the chemical reactor
scale with $uids at low pressure and low temperature, that is the reacting zone of the reactor, e.g., the hydrody-
usually air and water at ambient conditions, which do namics of the recirculating liquid phase in the particular
not correspond to industrial conditions and may not be case of bubble columns. Less work has been published
representative. This paper focuses on how computational on $ows within internals such as distributing trays, or
mixing boxes.
∗ Corresponding author. Tel.: +33-4-78-02-20-20; fax: +33-4-78-

02-20-09.
Van Baten and Krishna (2000) and Liu, Yuan, Yu,
E-mail address: [email protected] (L. Raynal). and Zhu (2000) have considered the $ow on sieve trays.

0009-2509/01/$ - see front matter ? 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 0 1 ) 0 0 2 2 0 - 2
6386 L. Raynal, I. Harter / Chemical Engineering Science 56 (2001) 6385–6391

Fig. 2. Schematic overview of the experimental setup.

Fig. 1. Sketch of a hydrotreating unit typical con3guration.

In these two studies, gas and liquid are considered as

Fig. 3. Ori3ce con3guration. Top view.
interpenetrating $uids, where gas $ow rises as dispersed
bubbles through a liquid 3lm. This two-$uid model, or
Euler=Euler approach, gives macroscopic information transparent Plexiglas column inside which diMerent kind
since a computational cell is assumed to contain several of internals can be placed. Gas (air) is injected at the top
bubbles, the interface between gas and liquid not being of the column and liquid (industrial water) is injected
resolved. In the present paper, we shall focus on $ows on both lateral walls of the column, making sure that
through internals where liquid and gas strongly interact iso-distribution of liquid $ux is achieved between the two
but without interpenetrating, keeping an interface clearly inlets. Liquid is recirculating while air is provided by
identi3ed. In the case of a hydroprocessing reactor compressors. Flowrates are controlled via precision $ow
(see Fig. 1), internals of concern are mostly distributing meters, liquid $owrate ranges from 2:77 × 10−5 to 5:55 ×
devices located at the reactor inlet and between two con- 10−4 m3 =s, while air $owrate ranges from 1:38 × 10−3 to
secutive catalytic beds, quenching and mixing devices 22:2 × 10−3 m3 =s.
located in interbed zones and reactor outlet devices such Two con3gurations have been used in this study. The
as a bottom basket, where vapor and gas are collected 3rst one consists of a single ori3ce corresponding to a
before exiting the reactor. geometry used for a mixing device inlet. The mixing de-
The experimental setup with the two kind of internals vice, not shown here, is made of a complex geometry
considered are described in Section 2. Section 3 deals to ensure that good mixing between the two phases is
with the numerical approach; the models used and the achieved i.e. that thermal homogeneity and thermody-
computational procedure are presented. Experimental re- namic equilibrium are reached. The geometry is so com-
sults and the comparison with numerical results are dis- plex that a corresponding VOF simulation would require
cussed in Section 4. In Section 5, a tentative strategy is a too long calculation time; our study is consequently re-
proposed to integrate CFD calculations in an internal de- stricted to the $ow at the device inlet only. The ori3ce
sign procedure to provide secure rules for operation under used is shown in Fig. 3. With this con3guration, liquid
real industrial conditions. $ows along the periphery of the ori3ce as a liquid 3lm
while air $ows at the center. Fig. 4 shows a sketch of
the hydraulic con3guration at ori3ce inlet, UL and UG
2. Experimental setup being, respectively, the local liquid and gas velocities, e,
being the thickness of the liquid 3lm in the plane of the
The experimental setup (shown in Fig. 2) consists ori3ce. The measurements were performed through high
of a 1 m high, 700 mm in span and 100 mm in width, magni3cation CCD camera visualization. It consisted of
 L. Raynal, I. Harter / Chemical Engineering Science 56 (2001) 6385–6391 6387

Fig. 4. Hydraulic con3guration at ori3ce inlet (side view), de3nition Fig. 6. Sketch of the interface calculation: (a) real shape of the inter-
of the local variables. face, (b) shape calculated by VOF geometric reconstruction scheme.

adapted for calculating the breakup of liquid jets or 3lms

sheared by a gas $ow (Keller, Li, Vallet, Vandromme,
& Zaleski, 1994) or bubble dynamics (Fan, Yang, Lee,
Tsuchiya, & Luo, 1999).
In the VOF model used, the 3elds for all variables
(pressure, velocity, etc) are shared by the two phases and
correspond to volume-averaged values. It is thus neces-
sary to know the volume fraction, q , of each phase, q,
in all the computational domain. This is possible through
the resolution of the volume fraction equation for the
phase q,
@q @q
+ ui = 0; (1)
@t @xi
where ui and xi denote, respectively, the velocity com-
ponent and the coordinate in the direction i (i = 1; 2 or
3), t, being the time; and through the resolution of the
momentum equation shared by the two considered $uids,
Fig. 5. Downcomer con3guration and corresponding computational
domain. @ @
uj + ui uj
@t @xi
measuring the height of liquid 3lm at two diMerent loca-

@P @ @ui @uj
tions, far from the ori3ce where the liquid 3lm is of con- =− +  + + gj + Fj ; (2)
@xj @xi @xj @xi
stant height, h, and just at the ori3ce weir, h (see Fig. 4).
The second con3guration consists of a downcomer where P, g and F denote, respectively, the pressure, the
used on a hydrotreating reactor distributing tray. Since gravitational acceleration and the external force per unit
the experimental geometry and the one used in the CFD volume, the physical properties (density, ; and viscos-
computations are exactly the same, Fig. 5 shows the ity, ) being determined by volume-weighted averages.
downcomer with the corresponding computational do- The tracking of the interface is done in the cells where
main (unstructured mesh with 69 037 cells). Gas $ows the volume fraction is diMerent from 0 or 1. As it is il-
through the top aperture of the downcomer while liquid lustrated in Fig. 6, the reconstruction of the interface is
$ows through the three levels of three holes distributed at accomplished via the use of the geometric reconstruc-
◦
every 120 . Considering the existing two planes of sym- tion scheme. This calculation scheme represents the ac-
metry, only one third of the domain was simulated. tual interface (Fig. 6a) as a piecewise-linear geometry
(Fig. 6b). Since, in the present cases, the Reynolds num-
bers of the considered $ows are high, the standard k–
3. Numerical approach approach is used for turbulence modeling (Launder &
Spalding, 1974). The commercial code, Fluent 5.4 has
Since gas and liquid $ows do not interpenetrate in the been utilized for the computations.
considered cases, a calculation with the VOF approach Two and three dimensional simulations have been car-
has been chosen. The VOF model enables the computa- ried out for the ori3ce con3guration while all simulations
tion of two-phase $ows where the phases do not mix, i.e. carried out for the downcomer con3guration were three
the gas–liquid interface is clearly identi3ed. It is thus well dimensional. Fig. 7 shows the computational grid used
6388 L. Raynal, I. Harter / Chemical Engineering Science 56 (2001) 6385–6391

Fig. 8. Mass $owrate at outlet versus iteration number for a ori3ce

2D simulation. QG = 8:33 × 10−3 m3 =s, QL = 4:72 × 10−4 m3 =s.
Fig. 7. Computational domain, mesh and boundary conditions for the
ori3ce 2D simulations.

for the two dimensional computations, where the bound-

ary conditions are listed. The grid consists of a combi-
nation of structured and unstructured meshes, the region
close to the wall being meshed with rectangular cells of
0:5 mm in height. Considering that the geometry shown
in Fig. 3 has two planes of symmetry, the grid used in the
2D calculations corresponds to half of the real geometry
while the grid used in the 3D calculations corresponds
to one-fourth of the real geometry. The mesh used in the
3D calculations is based on the 2D mesh. The grids cor-
responding to the 2D and 3D ori3ce calculations contain
5135 cells and 69 995 cells, respectively. Computations
are time dependent; they are carried out until steady state Fig. 9. Liquid heights of the 3lm at far 3eld, h, and at ori3ce,
h , versus liquid $owrate, QL , case with QG = 0 m3 =s. Comparison
conditions are reached. These conditions are character- between experiments (open symbols), VOF 3D simulations (closed
ized by a constant liquid height on the plate and most symbols), and the Blevins (1984) correlation for h (solid line).
importantly, by a constant mass $owrate at exit (ori3ce
or central portion of the downcomer). Fig. 8 shows a
typical example of the computed mass $owrate at outlet computational time is within a few hours for the two di-
versus iteration number, N , for a 2D computation. Until mensional simulations and is about 3–5 days for the three
liquid reaches the outlet, which is characterized by a very dimensional simulations. Calculations have been carried
sharp peak at about N = 800, the time step is very low out on a R12000 300 MHz processor Octane Silicon
(about 10−7 s). The time step is increased little by little workstation.
from 10−7 s to a 3nal value of 10−4 s or 10−3 s, the latter
value depending mostly on the density ratio between the
two $uids. The time-stepping strategy depends on the 4. Results
number of iterations by time step needed to ensure very
low residuals values (all less than 10−3 ). If one obtains 4.1. Results on the ori1ce
10 consecutive time calculations with less than three
iterations by time step, the time step value is increased Fig. 9 shows, for zero gas $owrate, the liquid 3lm
by a factor of three. In the representative case shown height measured far from the ori3ce, h, and at the ori3ce,
in Fig. 8, the time step was increased up to 5 × 10−4 s, h versus the liquid $owrate, QL . Since the liquid sheet
and the calculation was stopped after 500 iterations with is sligtly oscillating, averaged values for h are reported.
this time step without any signi3cant change in the out- The experimental results are compared to the correlation
let mass $ux value. The corresponding real time is of given by Blevins (1984) for liquid height at weirs in the
the order of a few seconds, while the corresponding case of turbulent free surface $ows. This correlation is
 L. Raynal, I. Harter / Chemical Engineering Science 56 (2001) 6385–6391 6389

given by
 2=3
0:715 gQL
h0 = × ; (3)
g 2R( + 2)
the subscript ( )0 corresponding to zero gas velocity.
The experimental results are also compared to the re-
sults obtained with three dimensional simulations. It is
known that the transition from laminar to turbulent $ow
is achieved for liquid Reynolds number, Re, more than
2000, where
L UL 4h
Re = ; (4)
L
with 4h corresponding to the hydraulic diameter of such Fig. 10. Instantaneous picture of the liquid volume fraction contours
an open channel $ow. Since, this condition is achieved for the two-phase $ow at the ori3ce inlet, 3D VOF simulation for
for this $ow for QL = 10−4 m3 =s, computations, which QG = 8:33 × 10−3 m3 =s, QL = 4:72 × 10−4 m3 =s ( 14 of the real domain
is shown). Red corresponds to the liquid bulk and blue corresponds
assume turbulent $ow via the use of the k– model,
to the interface. Arrows correspond to gas velocity vectors, the length
have been carried out for QL ¿ 10−4 m3 =s only. One ob- of the arrows being proportional to the intensity of the velocity.
serves, from Fig. 9, that for both h and h , the agreement
between simulations and experiments is very good, h
being well predicted by the Blevins correlation as
well. The disagreement between the experiments and
the Blevins correlation observed for the lowest liquid
$owrate values is due to mainly two reasons. First, the
experimental uncertainty and the computational grid size
are both equal to 0:5 mm which is important in compar-
ison to the measured height of 2 mm. Second, for such
low liquid $owrate values, the $ow is not fully turbulent
and the correlation no longer applies.
In the case where gas $owrate is not zero, it was not
possible to get a real steady state with 3D computations.
A pseudo-steady state, oscillating around a mean value,
was reached instead. In order to get a precise mean value, Fig. 11. Liquid height at ori3ce, h , versus gas $owrate, QG ,
it was necessary to run the simulations for a few sec- for various liquid $owrates, QL . (♦) QL = 1:11 × 10−4 m3 =s; ( )
onds ($ow physical time) which corresponds to more than QL = 2:58 × 10−4 m3 =s; (P) QL = 4:72 × 10−4 m3 =s. Comparison be-
two weeks of CPU time. This required computation time tween experiments (open symbols), VOF 2D simulations (black
closed symbols) and VOF 3D simulations (grey closed symbols).
was so long that only one calculation has been carried
out. An instantaneous picture, corresponding to this case
(QG = 8:33×10−3 m3 =s, QL = 4:72×10−4 m3 =s) is shown
in Fig. 10. The picture shows the contours of the liquid
volume fraction (bulk in red and interface in blue), some increase of gas $owrate results in an increase of the shear
gas velocity vectors, colored by their intensity, are also at the interface, accelerating the $ow within the liquid
shown. One observes that the liquid 3lm is not of constant 3lm, which 3nally results in a liquid height decrease. One
height all along the periphery of the ori3ce (only a 1=4 of observes a very good agreement between experiments
the domain is shown). The determination of h was conse- and 2D simulations in particular for h values lower than
quently done by averaging the liquid 3lm height all along 5 mm which correspond to moderate liquid velocities or
the periphery and by averaging the latter value in time. high gas velocities. The slight disagreement observed
Further computations have been carried out with a two di- for QL = 4:72 × 10−4 m3 =s and for low or moderate gas
mensional approach reaching a steady state regime within $owrates between experiments and the 2D simulation, is
short computation times. The 2D equivalent $owrates no longer observed when considering the corresponding
were determined such that local gas and liquid velocities three dimensional simulation. This means that 3D VOF
at the ori3ce, as shown in Fig. 4, were conserved in accor- simulations are very well capturing the physics of the
dance with the real 3D geometry. Fig. 11 shows the liquid two-phase $ow at the ori3ce; the slight disagreement with
height at the ori3ce versus gas $owrate for diMerent liquid the 2D simulation being due to the fact that the 2D as-
velocities; experimental points are compared to 2D simu- sumption is valid only if h is small compared to the ori3ce
lations and 3D simulations. For a given liquid $owrate, an perimeter.
6390 L. Raynal, I. Harter / Chemical Engineering Science 56 (2001) 6385–6391

the liquid jet=surrounding gas interaction is well captured

by the VOF model. Many other cases, not shown here,
have been carried out for other liquid and gas $owrate
conditions giving always the same good agreement be-
tween calculations and experiments. One can thus assume
that the two-phase $ow dynamics inside the downcomer
is well predicted by VOF calculations whatever the $uid
properties (density, viscosity and surface tension). This
means that similar calculations can be carried out for $u-
ids with any physical properties. Fig. 13b shows the re-
sults obtained from calculations with exactly the same
liquid and gas $owrates as in Fig. 13a, but with physical
properties corresponding to a standard diesel under hy-
drotreating conditions (hydrogen and hydrocarbons un-
◦
der high pressure and temperature conditions T = 340 C,
P = 80 bar). One observes a dramatic change in the $ow
con3guration. Jets issuing from holes do not impact each
other anymore, the jet curvature in the downward direc-
tion being very strong. Instead of forming a liquid sheet
that will further breakup and be dispersed by the gas $ow,
liquid jets will strongly impact the downstream catalytic
bed modifying the resulting liquid distribution.
Fig. 12. Flow visualization inside the downcomer. This shows that a distributing tray, developed under
QL = 6:76 × 10−5 m3 =s and QG = 3:98 × 10−3 m3 =s, case of air and common laboratory conditions with conserving $uid
water. $uxes only, may act in a complete diMerent manner under
process conditions.
4.2. Results on downcomer

The second con3guration consists of the two-phase 5. Conclusion and discussion

$ow within a downcomer commonly used to distribute
$ows on trays (see geometry in Fig. 5). Experiments and This study shows, 3rst that VOF simulations are well
computations have been carried out for $owrates cor- adapted to capture two-phase $ow with an interface
responding to hydrotreating conditions. Liquid and gas clearly identi3ed as it is the case in many reactor inter-
are injected with $owrates of 6:76 × 10−5 and 3:98 × nals, and second that it is of great importance to correctly
10−3 m3 =s; respectively. account for process $ow properties during development
Fig. 12 shows a picture taken in the transparent ex- studies. The design of an internal optimized within lab-
perimental setup with air and water. It is experimentally oratory standard experimental ranges may not match
observed that water exits the 3rst level of holes to farther the design requirements for industrial $ow conditions.
$ow through the downcomer. The liquid jets issuing the A good strategy consists in using CFD calculations to
holes join at the central axis to form a single jet evolv- see how the two-phase $ow of interest behaves under
ing into a liquid 3lm occupying only about half of the laboratory conditions and real process conditions. If one
downcomer cross-section. It is clear that the shape of the gets similar $ows for both conditions, the developing
liquid jets is mainly given by the liquid velocity at the task is straightforward. If important diMerences are ob-
hole of the downcomer, by the interaction between both served, as it is shown in this paper with the particular
liquid jets and also by the interaction between the liquid case of the $ow through a hydrotreating downcomer,
jets and the surrounding gas. Fig. 13a shows the contours one should carry on numerical calculations with com-
of the liquid volume fraction obtained from a VOF 3D mon $uids (easy to use in laboratories or R&D centers
simulation carried out for identical conditions (same $u- and that would be used a priori) to de3ne suitable $ow
ids and same liquid and gas $owrates), which means in conditions matching process $ow conditions. It is indeed
particular, from mass conservation, for identical velocity easy to use VOF simulations to check what are the im-
conditions inside the downcomer. From this 3gure, one portant parameters that explain the numerically observed
clearly observes the same feature as obtained experimen- diMerences between standard experimental conditions
tally. This geometrical agreement between experiments and process $ow conditions. This can be done via the
and calculations implies quantitative agreement since the change of surface tension, liquid viscosity, pressure, etc.
shape of the jet is directly linked to the local liquid veloc- Once the main parameter responsible for the observed
ity. This means that the liquid jet=liquid jet interaction and diMerences has been clearly identi3ed, the experimental
 L. Raynal, I. Harter / Chemical Engineering Science 56 (2001) 6385–6391 6391

Fig. 13. VOF 3D calculations of the contours of liquid volume fraction inside the downcomer for identical gas and liquid $owrates as in Fig.
12 ( 23 of the real domain is shown). Red corresponds to the liquid bulk and blue corresponds to the interface: (a) computations for air and
water, (b) computations for $uids with physical properties corresponding to a diesel under hydrotreating conditions.

development can then be carried out for the latter adapted Fan, L. S., Yang, G. G., Lee, D. J., Tsuchiya, K., & Luo, X. (1999).
conditions to be representative of real process conditions. Some aspects of high-pressure phenomena of bubbles in liquids
In the case of the gas–liquid mixing box, an interesting and liquid–solid suspensions. Chemical Engineering Science, 54,
4681–4709.
characteristic of the $ow at ori3ce is the $ow regime. Keller, F. X., Li, J., Vallet, A., Vandromme, D., & Zaleski, S.
Mainly three regimes can be obtained. The 3rst one (1994). Direct numerical simulation of interface breakup and
corresponds to free falling 3lm corresponding to low in- atomization. Proceedings, ICLASS, Paper I-8 (pp. 56 – 63).
teraction between gas and liquid. The second one corre- Krishna, R., Urseanu, M. I., Van Baten, J. M., & Ellenberger, J.
sponds to slight atomization with part of the liquid being (1999). In$uence of scale on the hydrodynamics of bubble columns
operating in churn-turbulent regime: Experiments vs. Eulerian
teared oM the liquid sheet, forming droplets. The third simulations. Chemical Engineering Science, 54, 4903–4911.
regime corresponds to complete atomization, the liquid Kuipers, J. A. M., & van Swaaij, W. P. M. (1997). Application of
3lm being fully transformed into droplets. According to computational $uid dynamics to chemical reaction engineering.
Lefebvre (1989), in complement to the liquid Reynolds Review in Chemical Engineering, 13(3), 1–118.
number, another parameter is important to quantify Launder, B. E., & Spalding, D. B. (1974). The numerical computation
of turbulent $ows. Applied Mechanics and Engineering, 3, 269–
atomization at ori3ce, namely the Weber number, We, 289.
given by, Lefebvre, A. H. (1989). Atomization and sprays. Washington, DC:
G (UG − UL )2 e Hemisphere.
We = (5) Liu, C. J., Yuan, X. G., Yu, K. T., & Zhu, X. J. (2000). A $uid
 dynamic model for $ow pattern on a distillation tray. Chemical
with e being the thickness of the liquid 3lm in the ori3ce Engineering Journal, 55, 2287–2294.
plane (see Fig. 4). For any geometry, one can determine Neri, A., & Gidaspow, D. (2000). Riser hydrodynamics: simulation
using kinetic theory. A.I.Ch.E. Journal, 46(1), 52–67.
from VOF calculations the $ow characteristics of inter- Ouwerkerk, E. S., Bratland, A. P., Hagan, B. L., Kikkert, J. P., &
est, namely the liquid 3lm height at inlet and the gas and Zonnevylle, M. C. (1999). Performance optimisation of 1xed bed
liquid local velocities; from the latter, one can deduce processes. ERTC Berlin, PTQ Spring (pp. 21–30).
the pertinent non-dimensional numbers Re and We. With Sanyal, J., Vasquez, S., Roy, S., & Dudukovic, M. P. (1999).
this information, one can check that the expected regime Numerical simulation of gas–liquid dynamics in cylindrical
bubble column reactors. Chemical Engineering Science, 54,
(3lm or spray depending on the 3xed goals) is attainable 5071–5083.
or not in a quick and low cost way. Sundaresan, S. (2000). Modeling the hydrodynamics of multiphase
$ow reactors: current status and challenges. A.I.Ch.E. Journal,
46(6), 1102–1105.
References Swain, J., & Zonnevylle, M. (2000). Are you getting the most
from your hydroprocessing reactors? European re1ning technology
Van Baten, J. M., & Krishna, R. (2000). Modelling sieve conference, Rome, November 15 (pp. 1–26).
tray hydraulics using computational $uid dynamics. Chemical Trambouze, P. (1996). CFD applied to process engineering. Revue
Engineering Journal, 77, 143–151. de l’Institut Francais du Petrole, 51(2), 199–203.
Blevins, R. D. (1984). Applied 7uid dynamics handbook. New York:
Krieger.