jyLEAgK WATEB POLLUTION CONTROL RESEARCH SERIES • 15040 EJ2 12/71
      Selective Withdrawal  From
         A Stratified  Reservoir
 ENVIRONMENTAL PROTECTION AGENCY • WATER QUALITY OFFICE

-------
       WATER POLLUTION CONTROL RESEARCH SERIES
The Water Pollution Control Research Reports describe
the results and progress in the control and abatement
of pollution in our Nation's waters.  They provide a
central source of information on the research, develop-
ment, and demonstration activities in the Environmental
Protection Agency, Water Quality Office, through inhouse
research and grants and contracts with Federal, State,
and local agencies, research institutions, and industrial
organizations.

A tripilicate abstract card sheet is included in the
report to facilitate information retrieval.  .Space is
provided on the card for the user's accession number
and for additional uniterms.

Inquiries pertaining to Water Pollution Control Research
Reports should be directed to the Head, Project Reports
System, Office of Research and Development, Environmental
Protection Agency, Hater Quality Office, Washington, D. C.
20242.

-------
Selective  Withdrawal From A Stratified  Reservoir
                         Jorg Imberger
                              and
                        Hugo B. Fischer
                    University of California
                   Berkeley, California  94720
                            for the


                Environmental Protection  Agency


                  Project Number 15040  E  J  Z

                         December 1970
           For sale by the Superintendent of Documents, U.S. Government Printing Office
                       Washington, D.C., 20402 - Price $1

-------
                        EPA Review Notice
This report has been reviewed by the Environmental Protection
Agency and approved for publication.  Approval does not signify
that the contents necessarily reflect the views and policies
of the Environmental Protection Agency, nor does mention of
trade names or commercial products consitute endorsement or
recommendation for use.

-------
                             ABSTRACT
This study describes both theoretically and experimentally the flow
into a line sink in a linearly density stratified reservoir.  The
geometry of the boundaries is simplified to a parallel walled duct with
the line sink at the center of the fluid.  The primary focus is on par-
titioning the flow into distinct flow regimes, and predicting the with-
drawal layer thickness as a function of the distance from the sink.

For fluids with a Schmidt number of order unity, the withdrawal layer is
shown to be composed of distinct regions in each of which a definite
force balance prevails.  The outer flow, where inertia forces are neg-
lected, changes from a parallel uniform flow upstream to a symmetric
self similar withdrawal layer near the sink.  Inertia becomes a force of
equal importance to buoyancy and viscosity for distances from the sink
smaller than  xc = (q/v)3'2 (vD/eg) '4 .  When viewed from the outer
coordinates, this is a small distance, but in physical terms it may be
equal to, or longer than, the length of the reservoir, depending on the
sink discharge.  The equations valid in this inner region, and which
have solutions capable of being matched to the outer flow, are derived.
Using the inner limit of the outer flow as the upstream boundary condi-
tion, these inner equations are approximately solved for the withdrawal
layer thickness by an integral method.  The inner and outer variations
of   6  , the withdrawal layer thickness, are combined to yield a com-
posite solution, and it  is seen that inclusion- of the inertia forces
yields layers thicker than a strict buoyancy-viscous force balance.
Also in terms of the inner variables, the equations, their solutions
and any experimental data are shown to be dependent only on the Schmidt
number.

Experimental laboratory experiments were carried out to verify the con-
clusions reached from the theoretical considerations.  The observed
withdrawal layer thicknesses were shown to be closely predicted by the
integral solution.  Furthermore, the data could be represented in terms
of the inner variables by a single curve verifying the independence of
any parameter other than the Schmidt number.

Prototype data is still sparse, but comparison with the measurements made
by the Tennessee Valley Authority indicates that in a reservoir the with-
drawal layers are approximately one and a half times as large as pre-
dicted.  The difference in thickness is probably explained by the
presence of a certain amount of turbulence in the reservoir, and inter-
ference by the rear boundary of the reservoir.

This report was submitted in fulfillment of Grant No. 15040 E.J.Z. be-
tween the Environmental Protection Agency and the University of Califor-
nia, Berkeley.

-------
                             CONTENTS
                                                                 Page

ABSTRACT       	      i
FIGURES        	     iv
TABLES         	      v
CONCLUSIONS    	     vl
RECOMMENDATIONS  	      x

1.  INTRODUCTION   	    1

2.  A REVIEW OF PREVIOUS WORK     	      5
    A Physical Discussion         	      5
    Unsteady Flow                 	      8
    Type  (b) Layers  F -> 0;  R =  1/G =0    	      9
    Type  (c) Layers  F2/Re -* 0 ;   F = 0 = 1/G    	     12
    Type  (c) Layer in a Diffusive Fluid
       F  = 0 ;  G = 0 (1) , R -» 0    	     16
    Numerical Studies  	     17
    Experimental Studies 	     18
    Summary      	

3.  ANALYTIC CONSIDERATIONS     	     21
    The Governing Equations     	     21
    The Initial Value Problem     	     24
    Structure of the Withdrawal Layer - Outer Solution    ...     27
    Structure of the Withdrawal Layer - Inner Equations   ...     33
    Structure of the Withdrawal Layer - Inner Solution    ...     38
    Summary of the Layer Structure - Composite  Solution
       for   6                   	     42

4.  EXPERIMENTAL OBJECTIVES AND APPARATUS	     49
    Review of Previous Experimental Data    	     49
    Experimental Objectives     	     49
    The Experimental Tank       	     49
    The Salinity Probe          	     52
    Establishment of a Linear Density Profile     	     55
    The Discharge Meter         	     58
    The Dye  Particles           	     58
    Photography and  Timing      	     60
    Experimental Procedure      	     60
    Experimental Program and Data    .	     62

5.  ANALYSIS AND DISCUSSION OF EXPERIMENTAL RESULTS   	     65

    Analysis of Data            	     65
    Effects  of the Finite  Size  of the Tank    	     67
    Comparison with  the Theoretical  Results and Previous
       Experiments              	     67
                                  11

-------
6.  COMPARISON WITH FIELD DATA     	73

    Unsteady Withdrawal  	 73
    Steady Withdrawal    	 73

ACKNOWLEDGEMENTS        	   77
REFERENCES              	   79
GLOSSARY OF TERMS       	   83
APPENDIX                	   85
                                111

-------
                             FIGURES







                                                                    Page






3.1   Theoretical Model       	     22




3.2   Outer Solution          	     31




3.3   Outer Solution          	     32




3.4   Integral Solution Velocity Profile    	     43




3.5   Theoretical Withdrawal Layer Thickness    	     44




3.6   The Structure of the Withdrawal Layer     	     45




3,7   The Composite Solution:  G = 2.34         	     46




3.8   The Composite Solution:  G = 29.0	     47




4.1   Experimental Data - after Koh (1966b)     	     50




4.2   General Layout of Experimental Tank       	     51




4.3   Photographs of Experimental Equipment     	     53




4.4   Conductance Probe and Float Circuit       	     54




4.5a  Example of the Probe Calibration          	     56




4.5b  Example of the Density Profile    	     57




4.6   Discharge Meter                   	     59




4.7   Example of the Dye Traces         	     61




5.1   Experimental Withdrawal Layer Thickness:  G = 29.0    ...     66




5.2   Experimental Velocity Profiles:  G = 29.0     	     68




5.3   Variation of the Integral Solution with  Kg   	     70




6.1   Prototype Experimental Data       	     75
                                 IV

-------
                             TABLES
                                                                    Page






4.1   Experimental Program     	     63




6.1   Field Data	     74

-------
                            CONCLUSIONS
The long term objective  of this research is to provide an operational
technique for managing the quality of reservoir outflow.  This objective
requires a complete understanding of the details of stratified flow and
mixing in a prototype reservoir.  In recent years a number of bulk man-
agement schemes have been devised and tested.  These schemes are en-
couraging, but their predictions are usually restricted to the reservoir
for which they have been calibrated.  This failing is predominantly due
to their primitive description of the reservoir mechanics.

The present report describes a detailed mathematical and experimental
investigation of one aspect of the management problem, the withdrawal
layer which forms at the level of the exit.

The following summarizes the conclusions drawn from a critical study of
previous work and the conclusions derived from the present study.

                          Prior Knowledge

Previous work on the steady flow problem may be divided into the follow-
ing two categories:

(i)  The fluid is inviscid and the upstream boundary condition is given
by  PT IT = constant.  The layers corresponding to this class have been
designated as type (b) in Chapter 2; or

(ii)  The inertia forces are completely neglected and the fluid is
assumed to be of infinite extent.  The layers corresponding to this
class have been designated as type  (c) in Chapter 2.

The weakness of the solutions of class  (i), when applied to reservoir
flow, is that in a reservoir the upstream condition is supplied by a
solution of type  (c) or by a solid vertical or sloping wall, and not
by the rather special constant head condition  pT IT = constant.  It
is therefore not clear where and how well this solution approximates the
flow.

The solutions of class  (ii) have generally been thought to be valid at
large distances from the sink.  In Section 2 it is shown, however, that
both the diffusive and non-diffusive solutions corresponding to this
class have a physically unrealistic behavior at infinity.  In both cases
the work done on the fluid at infinity by the pressure anomaly is in-
finite.  Furthermore, the validity of neglecting the inertia terms in
any part of the reservoir must be established.

Both classes of solution yield withdrawal layer thicknesses much smaller
than obtained from experimental or prototype observations.
                                 VI

-------
Koh's  (1966a) analysis may be used to estimate  the  time  taken  for  steady
withdrawal layer to form.  In Section 6 a comparison with  some prototype
data showed that this yields good results.


    Contributions of the Present Study to the Unsteady Problem

The solution by Koh (1966a) predicts, as mentioned  above,  times  for  the
development of a withdrawal layer which are  in  good agreement  with proto-
type and experimental observations.  However, the method of  solution  in-
volved a non-uniformity for small times and  for large distances  from  the
sink,  the effects of which were uncertain.   This inconsistency in  the
method of solution was removed by:

(a)  considering flow confined in a duct of  finite  depth,

(b)  by not making the Boussinesq assumption, but leaving  the  domain
infinite.

In both cases the time taken for the flow to reach  a thickness  6S  was
the same as predicted by Koh  (1966a) and is  given by


                           t =  J1  —
                               /eg" Sc
Contributions of  the Present Study for  the  Steady Withdrawal  Layer

The withdrawal  layer in a duct was analyzed and  partitioned  into  differ-
ent flow regions.  Each region is characterized  by  a  particular force
balance summarized in Fig. 3.6.  By matching the region where inertia
may be neglected  to the thin layer zone nearer the  sink,  a form of non-
dimensionalization was derived, which leaves the layer dynamics dependent
only on a single  parameter - the square root G of the Schmidt number.
The magnitude of  the withdrawal layer thickness, determined by an inte-
gral method applied to the dimensionless equations, and the observation
that G is the only parameter remaining  when the  flow  is viewed from the
dimensionless coordinates, was in good  agreement with experimental data.

(a)  Flow Regions:  In Section 3 a partial  answer has been found  to the
intricacies of the flow structure of the withdrawal layer for Schmidt
numbers of order unity.  The limit used to  derive the structure shown
in Fig. 3.6 was that the Froude number  and  the Rayleigh number tend to
zero,  but the Reynolds number  q/v  tends to  infinity.  The outer flow,
where inertia forces are neglected, changes  from a  parallel uniform flow
upstream to a symmetric withdrawal layer near the sink.   In the limit
of zero x,  the outer flow pattern asymptotes  to  the similarity solution
found by Koh (1966b)  for a withdrawal layer  in a fluid of infinite extent,
Thus,  even though the problem in an infinite domain, considered by Koh
(1966b),  is not well set,  its solution  approximates the correct solution
                                 vii

-------
in a region near the sink, but sufficiently far away to avoid the in-
fluence of inertia.

Inertia is shown to become a force of equal importance to buoyancy and
viscosity for distances from the sink, smaller than
This is referred to as the inner region.  When the flow reaches a dis-
tance of  Order (q//Cg) from the sink, an inviscid core of thickness of
0 (q/i/eg)  forms.  To be able to meet the edge boundary conditions it
is probable that a further zone, consisting of a buoyancy-viscous force
balance, forms at the perimeter of the inviscid core.

(b)  Integral Solution:  The equations valid in the inner region, and
which have solutions capable of being matched to the outer flow, are
derived in Section 3.  Using the inner limit of the outer flow as the up-
stream boundary condition, these inner equations were approximately solvec
for the withdrawal layer thickness by an integral method.  The inner and
outer variations of  6 , the withdrawal Tayer thickness, were then com-
bined to yield a composite solution.  This is shown graphically for dif-
ferent Froude and Schmidt numbers in Figs. 3.7 and 3.8.

(c)  G as a parameter:  When the outer region was matched to the inner
region it was noticed that the inner equations, in terms of the inner
stretched variables  (x/x0 , z/F2) , were only dependent on the parameter G.
                              m
Hence any experimental data, or solution of these equations, when repre-
sented in terms of the stretched variables will depend only on G .  So
far, G has been assumed to be of order unity.  The flow structure, de-
veloped for Schmidt number of order unity could, however, be quite dif-
ferent when the Schmidt number is large.  Extra regions in which vis-
cosity is important, but diffusion is not, would be introduced.  The
new domains would, however, be governed by equations which are special
cases of the present inner equations.  Hence even if G is not of order
unity, the conclusion that the flow, in terms of the inner variables,
is only dependent on the parameter G, is still valid.  The integral solu-
tion and any experimental withdrawal layer thicknesses may therefore
quite generally be represented non-dimensionally with G being the only
remaining parameter.

(d)  Physical properties of water:  If s (z) is the variation with depth
of the concentration of a dissolved substance, for example 02 , then the
outlet concentration will be given by the weighted mean over the depth
of the hypolimnion;
                         JL
                       I P
                       —   u(z) s (z) dz
                       q J
                                viii

-------
If the reservoir is long enough, so that the outer  region  is well  estab-
lished in the upper reaches of  the reservoir,  then  this  simplifies to
                      -^ J s (z) dz
                        -jfc
 (e)  Comparison with experimental data:  Fig. 5.1 shows  the experimental
variation of   6/F^j  with   (x/xc)^/^  .  The scatter  in  the data  is  small,
indicating the independence of the results from any parameter other  than
G.  Comparison with the integral curve is good everywhere except near
the sink.  The assumption that the velocity profile is self similar,
made in the integral method, is verified in Fig. 5.2.  Koh  (1966b) had
noticed this similarity for each individual discharge q, but this  is now
seen to be true more generally in terms of the inner coordinates.  The
data in Fig. 5.2 shows that the velocity profile remained self  similar,
except for small changes in the back flow, for a variation of  (x/xc) from
0.008 to 64.

 (f)  Comparison with prototype data:  The available prototype data are
reproduced in  Fig. 6.1.  Included are the composite integral solution,
the linear solution, and the withdrawal layer thickness  obtained from the
inviscid model.  It is seen that the integral solution is an improvement
over the other two, especially as  x/xc  increases, but  the increase in
the withdrawal layer thickness, gained by the inclusion  of the  inertia
terms, is still not sufficient to fully explain the observed prototype
values of  6 .

It is further  noteworthy that in the prototype studies the value of  xc  ,
when based on  molecular transport coefficients, is  usually greater than
the length of  the reservoir.  With eddy transport coefficients  which have
been adjusted  so that the T.V.A.  (1969 (a), (b)) data fits the integral
solution,  xc  becomes roughly equal to the length of a  typical reser-
voir.   This means that the flow in the reservoir, in either case,  is
predominantly characterized by the inner solution and that insufficient
length is available for the outer solution, to which Koh's (1966b) solu-
tion is an approximation, to form.  Of course, the inner solution  depends
on the existence of the outer solution,  so that the presence of the
reservoir end may be expected to invalidate the inner solution  to  some
extent, and may be responsible for the discrepancy with prototype  measure-
ments.   Nevertheless even in relatively short reservoirs the inner solu-
tion appears to be a reasonably correct description of the flow.   The
reasons for a larger layer in the prototype, than predicted by  the inte-
gral method,  have therefore been restricted to the presence of  small
amounts of effective eddy viscosity or the presence of a rear boundary;
probably a combination of both is needed for the full explanation.
                                 IX

-------
                          RECOMMENDATIONS

In the process of this research the following natural extensions became
apparent:

(a)  The analysis in Section 3 was carried out for the simplest geometry.
This work should be extended to include the following:

     (i)    end walls and a free surface
     (ii)   a more realistic bottom geometry
     (iii)  different sink positions
     (iv)   no slip walls.

(b)  The unsteady outer flow should be reanalyzed to include viscous and
diffusive effects.  This would make the patching suggested in Section 3.2
unnecessary.

(c)  The structure of the withdrawal layer flow should be examined for
large values of the Schmidt number.

(d)  The matching of the inviscid core IV near the sink, to the inner
boundary layer flow III, should be further studied.  This will, however,
require an understanding of the behavior of the solutions of the non-
linear Euler equation.

(e)  More experimental data are needed on the values of the transport
coefficients in actual reservoirs.

(f)  The flow changes caused by a sudden decrease in discharge are still
unknown and should be investigated.

(g)  The reverse problem, a jet moving into a stratified reservoir, appears
to have received little theoretical attention.  The river inflow at the
upper end of the reservoir is such a jet, and its solution is required
for a better understanding of the overall reservoir dynamics.

-------
                        1.  INTRODUCTION


For centuries surface water reservoirs have been constructed as a
means of storing water for flood control and water supply.  As such
their function is well understood and their design is no longer a
field of new research.  Only in the past few decades, however, have
engineers become aware of a major, sometimes helpful and sometimes
deleterious side effect of reservoirs, namely their effect on water
quality.  A surface reservoir represents not only a given number of
cubic feet of stored water, but also an inventory of dissolved oxygen,
heat, nutrients, minerals, turbidity, plants, and animals, all dis-
tributed in an uneven manner throughout the reservoir in a delicate
balance whose maintenance is vital to the health of the reservoir
environment and the river downstream.  Engineers know a great deal
about managing the amount of water in a reservoir, but very little
about managing its quality.  Properly managed, reservoirs can provide
a useful means of regulating the quality of water in a stream, in-
cluding its temperature, dissolved oxygen, and other constituents, in
whatever way will be most beneficial to the downstream users.  Im-
properly managed a reservoir can contribute to the destruction of a
river environment by permanently changing its temperature, oxygen
content, and turbidity, with disastrous effects on the downstream
ecology.  Much remains to be learned about how the flow in a  reservoir
can  be  controlled to produce the desired quality of outflow.  This
report  presents a contribution to our understanding of one aspect of
the  problem, namely prediction of the thickness of the withdrawal
layer when the reservoir  is stratified.

Most large surface reservoirs are density stratified throughout most
or all  of the year.  The  surface, or epilimnion water, consisting of
the  upper 15 feet or .so,  is heated by the sun and becomes lighter than
the water beneath.  Below this there is a sharp drop in temperature,
called  the thermocline.   The deep water, called the hypolimnion, has a
slight, nearly linear stable temperature gradient, meaning that the
density of the water increases continuously as one goes down  in the
reservoir.  The effects of this slight stratification are profound,
because the flow patterns found in stratified water are totally dif-
ferent  from those found in water of constant density.  In stratified
systems flow in the vertical direction requires the expenditure of
energy, whereas flow in the horizontal direction does not; hence flows
in stratified waters tend to extend over great horizontal distances,
while being confined to narrow vertical layers.  When water is taken
out  of  a stratified reservoir the discharge tends to come only from
the level of the exit, flowing in a layer which extends over  the en-
tire length of the reservoir but only over a fraction of the depth.
The  temperature of the outflow, as well as the concentration  of
dissolved constituents such as oxygen and nutrients, is that  of the
water in the reservoir at the exit level.  Management of the tempera-
ture and quality of water downstream of the reservoir can be  achieved
by varying the level of the exit to select the desired quality.

-------
The long-term objective of this research is to provide an opera-
tional technique for managing the quality of reservoir outflow.
This objective, a substantial undertaking, will require a complete
understanding of the details of stratified flow and mixing in an
arbitrarily shaped reservoir.  The complexities are enormous, and
may never be understood in complete detail.  Nevertheless, in recent
years several bulk management schemes have been put into use, for
instance those described by Orlob and Selna (1970) and Beard and
Wiley (1970).  In these schemes a computer is told as much as is
presently known about reservoir mechanics, that the incoming river
water sinks till it reaches its level of buoyancy, that the outflow
comes from the level of the exit, that solar energy is supplied at
the water surface, and that some type of vertical mixing takes place.
The computer is given an historical sequence of in and outflows and
is programmed to keep track of temperature and water quality as a
function of depth in the reservoir.  These schemes have considerable
promise, their only failing being in the primitiveness of their
description of the reservoir mechanics.  Thus Beard and Wiley (1970)
find that the coefficients required for their program vary from
reservoir to reservoir by a substantial factor, and that the varia-
tion cannot be predicted within the present state of knowledge.
Improvement in the accuracy of schemes such as this one will come
only by improving our understanding of the basic mechanics of the
processes involved.

The present report describes a mathematical and experimental investi-
gation of one aspect of the management problem, the withdrawal layer
which forms at the level of the exit.  Even in this detail the
mathematical difficulties are so formidable that the reservoir must be
idealized to obtain a solution.  Mathematically the reservoir is con-
sidered to be infinitely long and of constant width and depth, and the
density profile is taken to be linear.  The exit is assumed to be
located halfway from the surface to the bottom, and all walls are
assumed to be stress-free.  Within these limitations a mathematical
solution can be obtained; the details of the derivation are described
in Section 3.  Sections 4 and 5 describe a laboratory experiment and
compare the results of the experiment to the analysis of Section 3;
the comparison is extremely good.  In Section 6 the theory is compared
with the few available measurements in prototype reservoirs; the
thickness of the predicted layer is less than is observed, but not by
a large amount, and the result is much closer than those of previous
theories.  The most important reason for the error is believed to be
the mathematical assertion of an infinitely long reservoir;  indeed,
one significant result of the mathematics  is that prototype  reservoirs
are in general short compared to the length required for complete
development of the withdrawal layer.  A major recommendation for fur-
ther research is study of the influence of the sloping bottom of the
reservoir on the formation of the layers.

The major analytical technique used in this report is the method of
matched asymptotic expansions, a method developed by applied

-------
mathematicians only within the past approximately 20 years for the
solution of problems in boundary layer theory.  The method has proved
a powerful tool in many branches of fluid mechanics, and appears to
be of significant usefulness in many civil engineering problems, even
though it obviously will not be familiar to many practicing engineers.
The authors hope that a detailed understanding of the mathematics will
not be essential to understanding the results of the study; those
less interested in the mathematics may skip directly to the descrip-
tion of the discussion of the experiments, and their results on the
comparison of the theoretical predictions with field observations.

-------
                  2.  A REVIEW OF PREVIOUS WORK

The following is a review of previous work on selective withdrawal
by a line sink.  The physical limitations of the various solutions
are investigated and their applicability to experimental work  is
discussed.  The comments here complement the recent survey  article
by Brooks and Koh (1969).

                      A Physical Discussion

A sink placed in a homogeneous fluid of infinite extent  induces a
radial motion.  When the same sink is placed in a stably stratified
fluid a thin layer of fluid is observed to move, from  both  the left
and the right, towards the sink.  It is this layer which is of
interest here.  Before proceeding, however, to discuss specific known
solutions of the above phenomenon it is perhaps helpful  to  look at
the general implications of a variable density.
Let
where
                    PT =  PQ + Pe(z) +  p(x,z,t),
        p        = total density

        p        = mean over the depth  of  the density

        p  (z)    = density  above p  when motion  is  absent

        p(x,z,t) = density  changes  induced by the motion

then a measure  of the  scale of  the  density gradients  is given  by
                                p   dz   '
                                o

The principal flow modification a variable density induces, concerns
the creation of a non-trivial body force.  The main force, gravity
acting on p , leads only to a passive hydrostatic pressure which may
be removed  from the equations of motion.  However, if say a particle
of volume 6v moves a vertical distance of order L, then its density
anomaly with the surroundings will be of order ELp .   This induces a
body force, in the direction of the particle's original level, of
order £L p  g&v.  This force, which now actively enters any vertical
momentum balance, tends to inhibit vertical motion.  If the stratifi-
cation is strong this force will dominate the other prevailing forces,
and the flow will be layered or jet like.

-------
Moment of momentum considerations may be used to show that the above
force not only inhibits vertical motion, but also induces a vorticity
in the fluid motion.   Consider again a sphere 6v of a stratified
fluid.  The pressure  force acts, to first order, through the center of
volume, but the body  force vector passes through the center of gravity.
In general, as the sphere has more of its mass concentrated in the
lower half, the two centers will not coincide and the sphere will tend
to rotate.  The production of vorticity by this applied moment yields
baroclinic motion and the usual notions about inviscide fluids exe-
cuting irrotational motion must be discarded.

If this buoyancy force is important in an unsteady fluid motion, the
time scale will be fixed by the period of the internal waves present.
The natural period of the fundamental oscillation may be found by
analyzing the motion of the model sphere after  it is displaced a ver-
tical distance z from its original position, in a linear density
field.  The net force experienced by this particle is of the order 6v
e Po gz towards its original position.  Disregarding other  forces, the
equation  of motion governing the particle is,

                       .            .  d z
                      g&vep z  = p   6v —-
                           °     °    dt;
or
                           d2z   f
                           —5- = egz  i
                           dt2

and the solution  is given by

                         -i/eg  t    o+i/eg  *
                   z = Ae        +  Be

Hence the time  scale T of the  internal  oscillations  is  l//eg .   The
frequency /Eg is  generally known as the Brunt-Vaisalla  frequency.   If
the  fluid is  forced  to respond  at  a much faster rate,  then the  buoyancy
forces may  be overwhelmed  and  the  fluid will respond as a homogeneous
mass.  The  equations  of  motion are  then elliptic.

The  other modification a variable  density  causes is  that the inertial
 resistance  of the fluid  is  now a function of the vertical height.   Two
particles separated  by a vertical  distance L will have a density
difference  of the order  £L.   If both of these particles are placed
under a  similar pressure gradient,  their inertial resistance will be
 correspondingly different;  the lighter particle will accelerate faster
 than its heavier counterpart.   The importance of this effect is related
 directly to the magnitude of eL.  In a reservoir L is bounded by 2i,
 the depth of hypolimnion,  making 2t,e of 0(1CT3).  The smallness of EL
 implies  that the percentage difference in the^inertial resistance is
 small, which in turn means that the term PT — in the equations of

motion may be approximated by p  — .   This is what is generally known

-------
as the Boussinesq approximation and it will be used throughout this
report.

The above discussion shows that the pressure gradient induced by
switching on a sink, meets only a small viscous resistance in the
horizontal direction, but it is counteracted by viscosity and
buoyancy in the vertical direction.  As the horizontal velocity at
the level of the sink increases, the velocity everywhere else will
decrease, since the discharge q remains constant.  This results in
a thin horizontal velocity jet, with an associated vorticity con-
centration at the level of the sink.  As the layer narrows, the
velocity and velocity gradients increase and inertia and viscous
forces become important.

In contrast to boundary layer theory in a homogeneous fluid, the thin
vorticity concentrations in a stratified fluid may be sustained by
three distinct mechanisms:

(a)  Cross stream diffusion of vorticity is balanced by longitudi-
     nal convection  (ordinary boundary layer).

(b)  Production of vorticity by the baroclinicity  is balanced by
     downstream convection (inviscide buoyancy layer).

(c)  Cross stream diffusion of vorticity is balanced by the creation
     of vorticity of opposite sign by the  baroclinicity (creeping
     flow layer).

A layer of type  (a)  is usually relevant only when  a wall is present
and the velocity of  the fluid is high enough to  overcome any
buoyancy effects.  Redekopp (1970) has shown how this type of layer
may, however, form an obstruction with its displacement thickness,
for flow in which buoyancy is important.   The  other two layers both
involve an active participation of the buoyancy  forces; they are the
ones which constitute the withdrawal layer considered here.

The existence of these layers requires, as already mentioned, that
the buoyancy forces  be strong enough to make vertical displacements
within the layer small compared to the reservoir depth.  Dimen-
sionally this means  that when assessed with respect to the reservoir
depth, the buoyancy  forces should be larger than either the inertial
or viscous forces, but that also there be  a scale present, smaller
than the depth, based on which the forces  balance.  The quotient of
the relevant forces to the buoyancy forces, when all are assessed
with respect to the  reservoir depth, then  form the small parameters
usually associated with a thin layer theory.   For layers of type (b),
this yields the quotient
                                  ->
                                 Du
                           o v n  ——      9       9
        inertia forces	o Dt  _ _U	q   _  2
        buoyancy forces ~ 6V e I (>n g ~ firl2 ~    fi

-------
where F is called the densimetric Froude number.  Similarly for
layers of type (c) the parameter becomes
                                                   2
          viscous forces  _ (J,U6v    1 ___ VU  _ F
          buoyancy forces ~  .2  SveAp g      .3   R   '
                             Xj        o    G g*     e
           £u
where R  = — = Reynolds number.
       e   v
When diffusion of the stratifying species is non-zero, then the speed
of propagation of the species is of order D/A .  If U  is of the same
order or smaller than this, the adjustment in density by diffusion is
important and the relevant number becomes the Rayleight number R,
where

                           2        ' I •
In general there are thus three independent parameters F, R  and R
which are important in the study of withdrawal layers.  The internal
Froude number F is the relevant parameter for layers  of type (b) ,
while F /Rg and R are the parameters governing the  non-diffusive and
diffusive layers of type (c) .  Sometimes it is more convenient  to
deal with the alternative triplet of parameters F,  R, and G, where
G = (v/D)2.  Known solutions may be categorized into  one of the
following classes:

                   F -» 0 ;  R = 1/G = 0

                   F = 0 ;  F2/Re -» 0 ;  1/G = 0

                   F = 0 ;  R -» 0 ;  G = 0(1) ,

where the arrow -» 0 indicates that the parameter is small, but  non-
zero, and the solution is the limiting solution as  the parameter
approaches zero.

                          Unsteady Flow

The unsteady flow development appears to have received very little
attention.  The analysis by Koh (1966a) seems to be the only one on
the subject.  Koh (1966a) deals with the unsteady flow due to a
variable sink strength q(t) in a fluid of infinite  extent and linear
stratification.  The analysis neglects all forces except the unsteady
                  ~<~^
inertia force, p  -5— , pressure and buoyancy, and from above it
may be anticipated that the time taken for the flow to reach steady
state is k//Fg , where k is a constant to be determined.

For the case where the sink is suddenly brought to  full strength q,
the solution for small time is near radial flow, while for large time

-------
                                ^ ______          c
             ,     .      cos 6  ;n  Q2  sin Bt
            u(x,z,t)  ~ q — ^-Vl-P   - 5 —  ~ <*

where 6 = artan -  , f3 - sin 8 and t* = t/e"g  .  The solution,  for
large time, thus collapses to a thin line moving with  infinite  speed
towards the sink.  This may be expected as all the forces which could
counteract buoyancy have been neglected.  Koh  (1966a)  evaluated the
constant k by defining the layer thickness to  be given by the 1st
velocity reversal  and patching the above solution onto a steady layer
thickness 6 .  The velocity u is zero when

                             Pt* = TT

hence let

                            6 t*
                             s
                            -  = Tt
                             X
or
                  ou
In assuming  that  -^-   is multiplied by  PQ,  rather  than  pT ,
(1966a) assumed a Boussinesq  approximation.   However  the  solution is
radial for small  times and thus  the  scale  of  the  motion L becomes
infinite a large  distance from the sink.   This  means  that £L is  not
small  and the  Boussinesq approximation should not have  been  made.  To
see the effect of this non-uniformity,  the full problem was  analyzed
by Imberger  (1970) .   Although the flow pattern  obtained by him was
different, the time  estimate  for the withdrawal layer formation  was
insensitive  to the non-uniformity and  is the  same as  that found  by
Koh (1966a) .  Another way to  overcome  the  non-uniformity  is  to put
the sink between  parallel walls  of finite  spacing - this  is  done in
Section 3.

              Type (b) Layers:   F -»  0;  R = 1/G  =  0

The fluid is inviscid and non diffusive and the depth of  the
fluid  is large enough so that F  is small, causing  the  flowing layer
to separate  from  the walls.   With R  =  1/G  = 0,  the equations of  mo-
tion reduce  to the Euler equations for a stratified fluid and
    = 0.0; the density  is conserved  along  streamlines.   The  equations
may be rearranged to form a  single equation  for  the  streamfunction
(Yih  (1965) , p. 76) .
                                                ,                 (2.D
                           0 d\f    d
            / PT
where ty  =  / —  \|r and H  is the total head.  The  flow  is  inviscid  and
              o

-------
the only generation of vorticity is by the baroclinicity effect.
The equation really states:

     V  ill' - Pseudo    = Pseudo vorticity  +  Pseudo vorticity in-
             vorticity   generated in the     herited by the flow
                         particles flow       from any upstream
                         history              shear.

              dp
For the case  -TT— = 0,  Fraenkel (1961) and Childress (1966) have
deduced some general results regarding the formation of vorticities
in corners, but otherwise little is known about a general solution
of this equation.
                                                   T"       H fT
Yin (1965), pp. 75-140, discusses the case when  —T-  and  —r-  are
linear functions of l|f.  For a sink between parallel walls, the follow-
ing two upstream conditions meet this requirement.
                                                        2
     (a)  Upstream density has a linear variation and pU  = constant.
                                                        oo

     (b)  Upstream flow velocity and density are parabolic.

In case (a), the solution depends on the Froude number
F = u//₯g t  .   For F larger than 1/TT the solution indicates a motion
towards the sink over the complete depth of the channel, the only
exception to this being a large eddy in the corner above the sink.
This eddy grows as F decreases and when F = 1/rr it extends to infinity
thus violating the assumed upstream boundary condition.  List (1970)
has investigated the corresponding problem in three dimensions and
found that no solution satisfying the above upstream boundary condi-
tion exists for any value of F = q//eg~ J&2 .  This may be explained
simply by noting that for a finite q, u -» 0 as r -» °°, hence
F = u//£g J&2 -» 0, and by the two dimensional result, no solution
exists.  Case (b) always falls into the category F > I/TT and there-
fore is only relevant for large discharges where no separate layer
forms and the fluid dynamics is only slightly dependent on the
buoyancy forces.

In an effort to extend the above work to layered flow, that is where
F ^ — , Trustrum (1964) linearized the convective acceleration term
about the upstream uniform velocity.  This may be compared with the
Oseen approximation in slow viscous flow.  She then investigated the
time dependent problem for suddenly switching on the sink.  For  F > —
the solution converges in the limit t -» °°, to flows which are very
similar to those found by Yih (1965).  However, as F -» I/TT or F < I/TT,
the eddies in the upstream corner extend to x = °° as t -» » and the
original assumption involved in the linearization is violated every-
where.  Thus, the analysis suggests uniqueness for F > I/TT, but
nothing is learned for the case F ^ I/TT.
                                10

-------
The fact that the corner eddy extends in both Yih' s  (1965)  and
Trustrum's (1964) solutions to infinity, has led Kao  (1965  and 1970)
to postulate a slip line within the flow field, separating  the
flowing fluid from a stagnant upper layer.  The results  of  Kao' s
work, especially the later paper, suggests that the Froude  number,
based on the flowing depth, remains constant and equal to 1/n,
independent of the discharge q.

Two serious questions remain to be answered.  Firstly, what happens
to the corner eddies as F decreases below I/TT?  To say that these
are statically unstable is rather vague at best.  They are  the re-
sult of a dynamic force balance in which the centrifugal force
plays an important part.  To suddenly neglect this in the eddy
cannot be justified.  What is required is an analysis of their
stability to small (or large) disturbances.  Segur (1970) has re-
cently carried out such an analysis and he found the  flow to be
stable if
                                     < n
                         gz sT - a? < °  '

where r is the local  radius of curvature-.  By  this  criterion  the
slow moving eddies are  in  fact stable and  it  is  not  clear  how they
degenerate into a stagnant fluid as the Froude number decreases
below I/TF.  It could  be  that viscous forces predominate within the
eddy, but this does not  appear to have received  any  attention.

The second question concerns the non-uniformity  of  the flow as
x -» oo when viscosity  is  included.  The hope in solving any inviscid
problem is that the solution so obtained  is in fact  the solution  of
the full Navier-Stokes  equations in the limit  as v -» 0.  However,
in the slipline solution,  viscosity would  be  important around the
vortex sheets and the two  viscous layers  (top  and bottom of the in-
viscid layer) would coalesce at some x.   For x greater than this,
viscosity would be important throughout the whole layer.   This in
turn could lead to a  different upstream boundary condition for the
inviscid core, which  if  Fraenkel's (1961)  work is a  guide  could
have first order effects on the slipline  solution.   What is there-
fore needed is that these  three layers be  matched in the sense of
matched asymptotic expansions.  The difficulty here  is that this
requires not only a general solution of the equation governing the
inviscid core, but also  a  solution to the  coalesced  flow.  So far,
neither is known.   In Section 3, an approximate  procedure  is
suggested.

The experiments of Debler  (1959) are tailored to the slipline solu-
tion, in that the upstream boundary condition was not allowed to
develop naturally but was  forced in the manner suggested by Yih
(1965),  to be of the  form  p U2 = constant.  Hence, although the ex-
periments were in excellent check of Kao1 s (1970) solution, they do
not throw any light on whether the slip line solution with p_,U2 =
                                11

-------
constant upstream,  is the correct  limiting solution close to the sink
for a long withdrawal layer.  Long here  means one in which viscous
forces are important upstream.

                                F2
              Type  (c) Layer  :  — -  -» 0  ;  F = 0 = l/G.
                                R
The fluid is non diffusive  and  the  discharge  is small enough so that
the inertia terms may be neglected  completely.   This results in a
                                      p2
linear equation, which  for  the  case  —  -* 0,  possesses a thin layer
like solution.                         e

Gelhar and Mascolo  (1966) investigated such  a layer imbedded in a
fluid of infinite extent, by making a boundary  layer assumption.
                                         DpT
Further, as the fluid is non diffusive,  - 1  = 0 and p  is not deter-
                                         Dt           T
mined until it is specified somewhere in the  flow field.  The authors
assumed
                                                                 (2-2a)
where f is a constant.  This  leads  to a single equation for l|f,  viz.

                         o~  ill    f  S\li
                         i;	3L  .  i_  UT _ n                         ^o 9H^
                            rr  +    -v   — u .                        \£..6\»
                         ~  4    v  ox
                         Sz
                                          2
The thickness of the layer, in  terms of F /R ,  may be deduced from
the following argument.  When z -» «, the vertical velocity w -> 0,
which means that for conservation of mass
                   I
udz = l|r(oo) - ty(-oo) = - q                    (2.3)
Hence let

                  lit* = -  ; z*  =  r  and
                                                ,
where L is taken, in the absence  of  a  duct  width,  as the distance of
the point of consideration  from the  sink.   Then
                         3   *4  r    *  =  °                        (2>4)
                       fL  Bz      Sx

      Q       —Q               ^Ap       AP
Also  ^24^  ^ which implies  -^- s p < -^- and f is constant.
                                 12

-------
                               «„
where Ap is the density jump across the layer.   Hence  Eq .  (2.4)
becomes
                       R    ,. *4   -*
                        e   cz     ox

where F  = p q /g&pL   and R  = —  .  Now  rescaling  z* ,  such that
            o               e   v

                            z** -  **
                            z   -  6
and  6 = \— /  leads to  a  single parameter  free  equation
             /
           e
                                                                 (2.5)
Gelhar and Mascolo  (1966)  found  the  solution to this equation by
postulating  a  similarity  solution
                                     x*n
                                                                 (2,6)
The condition  §*(<*>)  -  ^(*(-<»)  = -  1  implies m = 0 and to satisfy
Eq. (2.5) n must  be  equal  to  1/4  and  the  function f must satisfy
the equation

                         4f(lV)  -  Tjf'  = 0  ,                      (2.7)

where
                                   **
The solution was evaluated  numerically,  with the  help of basic power
series and the  reader  is  referred  to  the author' s paper for further
details.  The solution showed  a  large center line jet moving towards
the sink, with  a small amount  of backflow at the  edges.

In a recent paper, Gill and  Smith  (1970)  have given analytic solu-
tions for the more general  equation
                      4f(tv)  - Tjf'  + mf  =  0  .                    (2.8)
They found that
                                 13

-------
                    f    =[  p-*-1  e^  dp ,                  (2.9)
                     m.k   «)
where C  are four different paths  in the complex plane, yielding four
independent solutions.  For the  sink flow m = 0 and the appropriate
real solution reduce  to

                           f = A J (Tl)                           (2.10)

where J  is defined in the paper,  and A is a constant adjusted so
that ^(oo) - $(~°°) = 1.  For the  asymptotic properties and a plot of
JQ, the reader is referred to the  original paper.

It is important to note that as  well as   dT^ .                      (2.12)
                                —oo
                            CD
Hence for J to be constant   1 p(T))dT) must be equal to zero.  That
                            _oo
this is indeed so,  is not difficult  to show.  From the momentum
equation, with p ~  x* p(T|)
                                 14

-------
                         p - Tip' = 4f'"                          (2.13)

Integrating this between !)=-<» and * yields
               ao
            2  !  pdTl - Tip    = 4(f  <-«)  -  f  (-»))  -  0
               J             l
i.e.
                           DO
                             pdTl = 0  .
                          —CD
[Conversely, in the problem of a wake  in  front  of  a  plate,  Long
(1959) and others assume J^O, i.e. m =  - 5.   This  means that
q = Ot which again is evident from  an  examination  of their solutions,]
Although    pdz = 0 implies that   !  pgdz  = potential  energy = 0
          O                       tj
         -00                       -03
i.e. the potential energy  of  the  flow  is  zero,  the energy balance of
the flow reads; the work done by  the pressure is equal to the loss

in unit potential energy    pwdzdx plus the work done  against viscous
                         v
dissipation.   The pressure work  is porportional to
                   r               r
                    up       dz  -  !  up I      dz
                  J    'x=x       <;   H'x=x
                          L.               £
                  — OO
                                CD

                  = (x*  - x|)   f  f'(Tl)  P(T,)  dT\
                      1     2.   u
(2.14)
If x^ =0 and x^ ~* ^   then an infinite amount of work must be done on
the flow at  infinity  rather than  at  the sink.   In other words,  the
flow is being pushed  from  infinity  rather than being sucked towards
the sink.

Another difficulty with  the solution is the matter of how this sort
of flow gets started.  Gelhar and Mascolo (1966)  assume that


                                = get                            (2.15)
where t is the width of the  opening.   This  has  no physical basis,
Further, assuming that the flow  started  from an abrupt density
variation, with the sink at  the  density  discontinuity is not valid
either, as with no diffusion the density profile could not adjust
                                15

-------
to a smooth variation.

These didfficulties are crystalized somewhat by  noting  that  as  the
layer grows, since q is constant, the velocity u will decrease.   At
some point u will be small enough to be of 0(D/L)  and then diffusion
will modify the solution.  Hence, the above non  diffusive solution
may be regarded as a solution for intermediate x.   But  again
matching is absent.

     Type (c) Layer in a Diffusive Fluid: F - 0; G =  0(1) , R -»  0

The flow is slow enough so that  inertia may be neglected, and the
diffusion of the species is important.  From the previous discussion
this may be seen to be a limit of the full equations  very far from
the sink.  In Section 3 a rigorous matching is carried  out.

Koh (1966b) has investigated the sink flow in this limit, in an
infinite domain with linear stratification.   (He also considers the
axi-symmetric case.)  In a similar way to the  above  the
equations of motion may be reduced to a single equation for

the stream function ty  =
                         q
                                                                 (2.16)
                        -v **t>    -x  *2
                        §z**     dx
              *                  i /o
where z** = -=-75-  .  Hence 6 = 0(R    ).   Koh  (1966b)  investigated the
            R1/3
solution of the above equation  by  postulating a similarity solution

                                           **
                    t* = x*m f(Tl)  ;  Tl = -^-TT-                   (2.17)
                                          Jc Jl / O
                                         x*
                            00

and again m = 0 to satisfy  I  u*dz** = - 1.   The resulting ordinary
                            J
                           -co
differential equation is


                   f(vi) + I T2f"  + - r\f'  =  0                   (2.18)
    • f     f'
or  if g = f
                     g(V)  +|T\2B/  +|% =0                   (2.19)
Koh  (1966b)  obtained  the  solution by a numerical integration from
T| =  T10 » 1  to  T)  =  0.  He obtained the value of g,  at the point T|0,
by noting that  for  large  T\ the  differential Eq. (2.19) reduces to

                          TjV  + 4% = 0 .                        (2.20)
                                16

-------
The solution to this equation  is g  = —— .   However,  this  variation

                                     ^        1
for the velocity  implies  that  the density  p ~ —  -» oo     This means
                                               z
that an infinite  amount of  salt would have to be transferred from
below the level of  the sink to above it.   List (1968)  in  a similar
problem for sink  flow in  a  porous media  obtained:

                    p ~ -  sgn z     as z  -»  ± »

- an even worse singularity.   He reasoned  that the flow could there-
fore not develop  unless this singularity is already  present in the
fluid before the  motion is  started.

In an effort to resolve these  difficulties, the  five solutions of
Eq. (2.19) were determined  analytically.   A detailed discussion of
the properties of these solutions is given by Imberger (1970).  It
is found that of  the five solutions two  are exponentially large,
two exponentially small,  and one decays  as 1/T] , as  T| -» « .   By
symmetry, the stream function  must  be an odd  function of  z,  and it
is shown that this  is only  possible if  g contains the algebraically
decaying solution - as well as the  two  exponentially small functions.
The physical interpretation of the  1/T]   decaying solution is that
the pressure associated with it grows logarithmically as  x -» »    But
as J = 0, this is not acceptable.   This  unpleasant behavior is not
due to assuming the Boussinesq approximation,  which  is quite valid as
it makes little sense to  keep  the variation of inertia, when inertia
itself is completely neglected, but rather it is due to posing the
infinite problem  with a linear density  profile.

In addition, when the solution is substituted into the neglected in-
ertia terms, it is  seen that near the sink they  dominate  the re-
tained viscous terms.  Hence the solution  has non uniformities both
far and near the  sink.  These  shortcomings are investigated  in
Section 3.

                        Numerical Studies

In an effort to obtain a  solution which  includes both  the effects  of
inertia and viscosity, Koh  (1964) applied  an  integral  technique to
the boundary layer  form of  the Navier-Stokes  equations.   The fluid
was assumed to be non-diffusive and of  infinite  extent.   Assuming
the fluid to be non-diffusive  means that the  density distribution
must be specified somewhere.   Koh (1964) postulated  a  relation simi-
lar to Eq. (2.2a):

                          PT =  po(l  - £11)  ,                       (2.21)

where T) = z/6 (x) .   This leads,  however,   to the same  difficulty in
the interpretation of e,   encountered by  Gelhar and Mascolo (1966).
                                17

-------
In recent years a number of numerical algorithms have been written
for solving the full Navier-Stokes equation when diffusion is absent.
Slotta et al.  (1969) have used what is generally known as the Marker-
Cell technique to study withdrawal from a stratified fluid.  They
show only results for a two layer stratification - a case for which
analytic solutions are available for inviscid fluids.  They claim
that the height of the withdrawing layer d  yields a Froude number

                             u
                     F = ,.•  •  °° 	 = 0.293 .                   (2.22)




This is then compared with Kao's (1965) result F  - — .   However, the
latter is derived for the blocked flow in a linearly stratified fluid
and is really not relevant to a two layered flow.  It is surprising
that dg is not an independent parameter as in the analytic solutions.
Until these discrepancies are explained, comparison with other work
is difficult.

                      Experimental Studies

Koh (1966b) appears to be the only one who has studied selective
withdrawal experimentally in a continuously stratified reservoir,
A discussion of his results is left to Section 4, as his data was
replotted to conform with the theory developed in Section 3.  The
Tennessee Valley Authority have over the last few years begun inves-
tigating withdrawal in actual reservoirs.  The data is still sparse,
but a full discussion and comparison is given in Section 6.

                             Summary

A withdrawal layer forms when a sink is placed into a density strati-
fied fluid, because the buoyancy force inhibits vertical fluid
motion.  The solution, obtained by Koh (1966a), for the transient
switch on problem, clearly shows how a density gradient causes a
withdrawal layer to form.  Since there is no resistance to horizontal
motion, the horizontal velocity at the level of the sink increases
at the expense of the motion elsewhere, and a layer forms the thick-
ness of which decreases with time.  The non-uniformity associated
with Koh1s (1966a) solution at large distances from the sink, due to
the use of Boussinesq approximation, was shown not to be serious, but
for completeness it is remedied in Section 3, by a slight modifica-
tion of the boundary conditions.

If viscous and convective inertia forces are included, then layers
of finite thickness form for which the available theories fall into
two classes.  Firstly, there is the work of Kao (1965, 1970), in
which the fluid is assumed to be inviscid and non-diffusive and
secondly, similarity solutions have been found by Gelhar and Mascolo
                                18

-------
(1965) and Koh (1966b) when the fluids inertia is neglected.  For
selective withdrawal in a long reservoir the flow changes,  as the
fluid moves towards the sink, from one in which inertia may be
neglected to one in which viscosity and diffusion can be  assumed
zero.  Hence, the known solutions appear to cover the two extremes
of the withdrawal layer; the flow very close and the flow very  far
from the sink.  For this to be so the solutions must, however,
correspond in the language of matched asymptotic expansions to  the
inner and outer expansions of the problem with respect to the
variable x.  In the above review it is shown that this is not so,
and further that the inertia free solutions have physically un-
realistic pressure distributions at large distances from  the sink.

What is therefore required is an analysis of the overall  structure
of the layer with proper matching of the different regions.  This
subject is treated in detail in Section 3.
                                19

-------
                    3.  ANALYTIC CONSIDERATIONS

In Section 2 it was discussed how a density gradient inhibits vertical
motion and causes the fluid to flow in a thin horizontal layer towards
the sink.  This layer is, within the Boussinesq approximation, symmetric
about the  x  and the  z  axis and is present not because the Reynolds
number is high, but because the overall Froude number  is small.

In actual reservoirs the stratification is usually due to a temperature
variation with depth.  Hence G is of order one.  Also  the parameters F
and R, when the length is based on the half depth  i,  (Fig. 3.1), are
always very small.  Typical values are of the order 10~2 and 10~5, re-
spectively.  However, the value of the Reynolds number is usually very
high.  Hence in what follows, the symmetric problem shown in Fig. 3.1 is
analyzed in the limit  F -> 0, R -> 0, Rg -» °° .

First a simplified unsteady problem, similar to the one considered by
Koh  (1966a) is analyzed for the present geometry.  The results here are
a substantiation of Koh's (1966a) solution and they have been included
mainly for completeness.

Following this the structure of a steady withdrawal layer is analyzed.
For simplicity the horizontal walls are assumed to be  stress free and the
density along them is assumed to be constant  (p =0).  This approximates
a well mixed free surface.  More complex boundary conditions could be
treated in a similar way, but boundary layers at a no-slip wall would
further complicate the analysis.  Flow in a duct was chosen primarily to
overcome the logarithmic rise in pressure at infinity  in Koh's (1966b)
solution.

The analysis assumes that at large distances from the  sink, the flow is
uniform and the density varies linearly with depth.  The method of matched
asymptotic expansions shows that the structure of the  flow changes pro-
gressively from a slow uniform flow at infinity, to a  layer of type (c)
closer in, which in turn develops into a jet flow in which inertia is
important.  There is some evidence that very close to  the sink a type (b)
layer forms.

The outside flow, where the influence of inertia is neglected, is solved
in closed-form and in the limit as  x  and  z  tend to zero the solution
approaches a similarity solution.  The inner equations are solved approx-
imately by Galerkin's method and composite solution for the withdrawal
layer thickness is derived.

                      The Governing Equations

Changes of density in the reservoir water are either caused by tempera-
ture differences or by a variable concentration of a denser substance
such as salt.   In the first  case, the basic equations governing the mo-
tion will be the Navier-Stokes equation with the energy equation de-
scribing the influence of the temperature gradients on the motion.  When

                                 21

-------
r
   INFLOW = q.
to
to
        SINK
                                                   WALL
                                 ZERO STRESS

                                   />=0
                                WITHDRAWAL LAYER
                                                      INFLOW=q.
                        FIG. 3.1  THEORETICAL MODEL

-------
the background stratification  is  the result  of  a varying concentration
of a second species the governing equations  are those of a mixture (Green
and Naghdi  (1969)).  The water in a reservoir can be assumed to be
incompressible and if a second species  such  as  salt is present, its
concentration will be small.   Under these  circumstances, Mu'ller (1968)
has shown that the equations of motion  reduce to the Navier Stokes equa-
tions for a single compound species, the density of which is governed
by Pick's law.  Since the vertical motion  is bounded by the duct width
2H , and as   &SL « 1  , the Boussinesq approximation can be used to fur-
ther simplify the equations.   The derivation of the simplified equations
under this  assumption is given by Redekopp (1969) and the equations are
as follows:


                        V  • u  = 0                                 (3.1)
                    Du                   2                            nx
                  P  7TT  +  ^P  ~ ~  PgR  +  M-V u                       (3.2)
                  o Dt
                                    2
                              w  = DV  p                           (3.3)
where
                     Dt   dz
        T(x, ±i) = 0                              (3.4)

        p(x, ±j&) = 0                              (3.5)


    u(oo)Z)  = - JL i H(t),                         (3.6)


   u(o,z) =- q6(z) H(t),                         (3.7)



 PT =  P0 +  Pe(z) +  p(x,z,t)


  D = diffusion coefficient

 p  = p (z) + p(x,z,t)



dT = ~ (PO + Pe(z)) g

  ->
  T = wall  stress


                23

-------
               H(t)  = unit step function

               6(z)  = Dirac delta function.


 Finally,  within  the  Boussinesq assumption,  the flow will be symmetric
 about  the line   z  = 0 .


                     The  Initial Value Problem

 The  aim of  this  section  is to  investigate the error induced by the
 Boussinesq  assumption in  Koh's (1966a)  estimate of the time taken for a
 steady withdrawal  layer  to establish itself.   It is shown below,  within
 the  framework of linear  inviscid theory,  that the finite duct problem is
 really only  a slight  extension of Koh's (1966a)  work and that to  this
 accuracy, the time  taken  to achieve steady  conditions is independent of
 the wall  spacing and  insensitive to the Boussinesq assumption.

 Within the  framework  of Eqs.  (3.1-3.7), when  the sink is opened,  the
 pressure  field will be equivalent to one in a homogeneous fluid.   Only
 when the  fluid begins to  move  will it experience the buoyancy force.
 Assumed here of course,  is that the pressure  at the sink is not so low,
 discharge so high,  that  inertia completely  dominates the buoyancy forces,
 and the flow is only  a slight  modification  of motion in a homogeneous
 fluid.  Once the fluid begins  to move,  buoyancy tends to restore  any
 vertical displacement and  the  resulting motion is an oscillatory  adjust-
 ment towards the final flow.   In this phase of the flow,  viscous  forces
 play only a minor role and  the  flow is  essentially governed by  a  bal-
 ance of pressure, buoyancy  and  inertia.   The  time scale is thus l//Sg
 as discussed in the introduction.
Let
                          0    *    u     *
                          A  ; U   =          =
                        t* - /eg t  and  p*
where

                            P = Ag  i
and
                        A = tt PQ   q//£g JL2
Equations (3.1) to (3.7) then become:
                          V • u* = 0                              (3.8)
                                24

-------
where
              -     + F u*vi*  + Vp*  = -  p*k + R.G.   u*             (3.9)
              ot
                     -|*  + F u*Vp*  -  w* =| ^p*                  (3.10)
                        1  <, z*  ^1  and  0 < x* < °°
                                R ~      —
                                  -~
ana
                                c -  -
                                G - I-
In a  reservoir  F « 1   and   R « 1  and thus to  the  zeroth order .



                                 y . u = 0                         (3.11)
                               £ + Vp - -  pk                       (3.12)





                                |C - w = 0                         (3.13)
                                ot



                     x,z e (x,z|-l < z < 1  and   0  < x < »}



                           u(», z, t) = iH(t)


 and

                           u(0, z, t) = 6(z) H(t)  .



 (The  constant  density boundary condition  becomes  redundant as  DVp = 0.)
 t  The  asterisks  have been dropped for convenience.
                                 25

-------
 Equations  (3.11)  - (3.13)  are identical to the ones solved  by  Koh  (1966a)
 for the  infinite  domain,  and his method of solution is  adopted here.
 The solution  in terms of  the Laplace Transform of the streamfunction   i|t
 is  shown by  Imberger (1970)  to be given by:
    fy  = —  artan  (coth
                                      —*—
                                             tan
                                                  _nTb
                                                 2/1+p2
                                    (3.14)
The complete  solution  is now given by the Laplace inverse of Eq.  (3.14)
This inversion  is  simplified by noting that there exist a positive
number M such that
                                1+P
                                       < M
for all  p  with  Rel(p)  = c > 1 .

Hence for small   x  and   z
                               — artan -
                               Tip       X
                                    (3.15)
which is uniformly valid  in   p .   This is the solution for  the  infinite
domain and Koh  (1966a) has given  the solution for  u  , viz.
  u(x,z,t) =
cos 6
  nr
J (ID) sin P(t-u))duu|  ,
 o                 /
                                    (3.16)
where
                          = sin  6 ,   6 = tan   z/x,
and  J   is the Bessel  function of zero order.

Consider  x  and  z   fixed,  then as  t -» »
u(x,zt)
                       COS
                          TTT
   2
ilf-  I
  p  J
                                      J <">>  sin
                                       O
                               cos   6  sin
                                                6(z) .
                                    (3.17)
                                26

-------
The layer thickness, defined to be the distance to where  u = 0  , is
therefore given for large  t  by the equation


                           3t = TT .


In terms of the layer thickness this means


                                                                  (3.18)
 (2n  is obtained instead of the graphical value of 6.4  in  Koh's  1966
work. )

The finite duct problem has thus the same time response  as the  infinite
domain.  The solution is, however, seen to be non-uniform  in  both  time
and space.  The space non-uniformity is dealt with in the  next  section
for the steady problem and is due to the large velocities  near  the sink
for all times.  But also as  t -+ 
-------
where
                         P =


                         A = F e I n  -
                                   Ho

In  terms of  these  new variables Eqs. (3.1)  - (3.5) become


F(u*u** + w*u**) + p**  = R.G.  u**^ + R G uj^*                   (3.20)


F(u*w* + w*w* )  +  p*  = -  p*  + R G w*  * + R G  w*  *             (3.21)
     A       £j      &                 £ £j          A A.


F(u*p% + w* p*,)  - w*  = \ R p**z* + \ R p*^                    (3.22)

where
                 F  = -—  ;  R =.     -     ;
                                    sg
The outer region is defined  as  that  part of the flow field in which the
inertia forces may be neglected.   By an order of magnitude analysis
Imberger (1970) has shown  that  the inertia forces may be neglected every-
where except near the origin.   The equations governing this outer flow
are given by


                         p**  =  R.G.  u**z*                         (3.23)


                         P**  =  -  p*                                (3.24)
                                                                  (3'25)
which must now be solved in the  finite  duct.   The "inner region" is then
that part of the flow where the  solution to Eqs. (3. 23) -( 3 . 25) no  longer
describes the flow.  This will be near the  sink.

The solution is most easily found by taking a  finite Fourier Transform
in the  z*  direction.  But first define   i|r*   such that

                          u* =  fz*                               (3.26)
and
                          w* = -  ** * .                           (3.27)
                                   A

Then Eqs. (3.23) -  (3.25) yield, on the elimination of  the  pressure,
                                28

-------
                     *        *       -  °
                         dz      Sx

                              2
The presence of the factor   R   implies  a  layer thickness of the order
      and the appropriate scaling  is given by
                  z**  =-^  z*                                    (3.29)
                         R17

and

                  x**  =  n3 x* .                                    (3.30)


Although this  scaling  is used in  the  next subsection, the solution  to
the outer problem is more conveniently represented in terms of  the
scaling
                      Z  = TT Z

and
                             *                                      (3.31)
                      x  = TT  x*  R .                                 (3.32)
This  also  accomplishes  the  removal of the dependence of   R  of  Eq.(3.27),
with  the additional  advantage that the wall spacing remains  independent
of  R .  In  terms  of these  variables the boundary value  problem becomes,
                     5 T1   + ° ^  = 0                               (3.33)
                      **^Q      i^^j
                     9e      Sx
                   -  TT ^ z ^ TT  and  0 <, x <  °°


with  boundary  conditions given by

                   ty*     = =F |    z' = ± TT      (a)
                                     = ± TT      (b)

                                                                   (3.34)
                   .4 *
                          = 0      I = ± TT     (C)
                                sgn z           (d)
                                 29

-------
The solution of Eq.  (3.33) satisfying Eq.  (3.34) and which  is bounded
at  x" = + °°  was found by Imberger  (1970)  to be given by,


        ir.~   i r  /  i  -n3^   .   ~    (-i)n   .   ~\
       w(x.z) = —  )  I - — e     sin nz - 	  sin nz I
                rr L  \   n                  n          I
                  n=lX                               I
                1 V     1  -n°^   .        z
                — )   - — e     sin nz - 7—
                rr L.     n                2rr
                  n=l
)   -  — e  " " sin nz - 7^                      (3.35)
Convergence of the series is very rapid everywhere,  except  at  the  origin
where it diverges.  This is the singularity corresponding to the  sink.
The solution given by Eq. (3.35) is shown in Fig.  3.2.

Figure 3.3 is a log-log plot of two streamlines within  the  forward
flowing layer.  It is seen that as  "X -» 0  the slope  tends  to  1/3, which
may be compared with Koh's (1966b) solution for the  infinite domain.

To study the non-uniformitv at the sink, it is necessary to know  the
limiting solution of  \ji(x,z)  as  'Sf -» 0  and  *z -» 0  .   The  above  indicates
that this is given by Koh's (1966b) solution.  A  more formal proof is
obtained by noting that the same limit is achieved by fixing   x   and  "z
and letting the duct walls move apart.  This process  yields
                        -;
          1  -p x      ~  ~
          — e     sin pz dz
          P
                                                                  (3.36)
                          00
                                 3
                            1  -q       ~
                            — e    sin qT| dq

                          o

       ~   — ~l/3
where  Tj = z/x    .  Imberger  (1970) has  shown  that  the  above  is  the
solution to the corresponding problem in  the  infinite  domain.

The outer flow therefore changes from a uniform channel  flow far  from
the sink to a self similar motion as the  sink is  approached.

Before turning to the inner flow, it should be  noted that  although  the
simplified mass flux condition  (3.1) has  been used,  the  solution  does
not imply an accumulation of mass in the  duct.  The  rate of mass  leaving
the duct at the sink is, by symmetry, equal to
                                 30

-------
w
        0
         0           0.5            1.0           1.5           2.0
            NON-DIMENSIONAL HORIZONTAL DISTANCE FROM THE SINK  7
f
                               FIG. 3.2  OUTER SOLUTION

-------
w
t\5
           NON-DIMENSIONAL VERTICAL DISTANCE FROM THE SINK X = j-773-R
                           FIG. 3.3  OUTER  SOLUTION

-------
                           (p  (o)  +  p  ) q  .
The rate of which mass enters the duct at  infinity  is  given by
I
                             PT u dz

                         -I
                                SL

                                  n~ dz
the same as  that  leaving at the sink.  Similarly  if   s(z)   is the  con-
centration distribution of a passive substance, such  as  C>2 , then the
concentration   SQ at the sink is given by  the mass balance:
                     3  q =   s(z) u(z) dz
                     o     J
                          -£
                                 A

                         = q TTJ  J s(z) dz                        (3.37)
                                -Jt
and
                         =       s(z) dz
        Structure of the Withdrawal Layer - Inner Equations

The inner region is that part of the flow in which Eq.  (3.35)  fails  to
yield the proper description of the flow.

In the above subsection it was shown that the governing equations
within the layer in the outer region are obtained by  scaling  the   z*
coordinate with respect to  E^^ .  Carrying this out for  the  full
                                 33

-------
Eqs.  (3.20) -  (3.22) yields
                                    ^•xr 'n      Y_'1c'r_*p'r_'p'fr
 o /« V Y jk*  f **„*    *„*  Y_. **_*=!"    *Y*      T™^T_T.T-_
 ^/o   'Z    ZX     XZZ,        A       ^-  ^  z.
                                                                   <3'38>
                 ***  ***z**>  + P***  = - P**
                 A   A fj        £-*
                                                                   <3-39)
                                     2/3
R
                               1/3
                                    *
where the new density   p**  = R    p* .   The outer layer  flow  equations
are obtained as the  limit   R -» 0  and  F/R2/3 -» 0 .

One method  (Koh  (1966b))  to compute the influence of  the nonlinear con-
vection near the  sink would be to expand the stream function   \|r*   in a
regular series in  the small parameter  F/R2/3 .  However,  an  inspection
of the second term of this  series shows that the series  diverges  when
x  becomes  smaller than
where


                           * = (F/R2/3)3/2 .                       (3.41)
X
 C
 In other words,  for distances smaller than  x*   the  influence of
 inertia is  not merely a correction to the outer  flow,  but it is the
 prime  force.  This means that near the sink   6 = 0(p2)   and  new
 stretched variables are needed.
 Let
                        **
                       x
                                   3/2
                                  34

-------
and
                                   **
                        *** =
                                                                   (3.42)
The outer solution   ty* (x* ,  z**>  when  rewritten in terms of the inner


variables is
                                       *   *
Now as  F/R2/3   -» 0  ,   this  becomes   ty (i) ,   where
hence  the  inner stream-function remains unsealed.  Also  to  be  able  to


match   p** ,  it must  be  rescaled
                           _ _

                            1/3. 1/2
                           R   /F
 as
 The equations in terms of these inner variables may  thus  be written:
it***   V*
 2      Z
                                        = G
                                                                   (3.43)
                  _ r llr*
                   2 " Vz***x**x**



                   e

                                               = ~  P***
                                                                   (3.44}
                  R
                                  R
                                 35

-------
              ,*    ***   ,  ,|.*    _ -1   ***     j. i  ***              /•:
            -  ? **  P #**  +  *„**	9  P.,**-.**  T r ?„***„***         v"
              xz       x      R/xx      Gzz
                                   e


where  Rg = q/v  .

In the limit  as  Rg -> °°  Eqs.  (3.43)  -  (3.45) reduce to: t


                 till    -  ill   ill    + p   = G  ill                        (3.46)
                 TZ  ZX    TX  ZZ   X     ZZZ


                         p  = -  p                                 (3.47)
                          Z
                  A_A._^=0                         (3.48)
                  Yz  px    Yx  pz    Yx    G pzz


which constitute  the required  inner  equations for the limit
R-»0, F-»0, Re-*" and  F/R2/3  -» 0 .   Equations (3.46) -  (3.48)
must now be solved with  the  solution (3.35) as the upstream boundary
condition.

It is interesting to note  that  Eqs.  (3.46) -  (3.48) include  the outer
equations, and hence for those  regions in which the wall influence  is
absent, the flow  only depends  on the parameter G.  Any experimental
flow at a given value of G is  therefore self similar with respect to
the coordinates   (x**, z***).   The experiments described in  Section 5
verify this result very  well  (Fig. 5.1).

The reason why  Rg   should be  large  for the above equations  to be valid
is perhaps best seen by  examining the width-to-length ratio of the  region
of nonuniformity.  The physical width of  the region  6  is of order
       and the physical  length   xc  is of order  F3/2VR.  Hence
and hence   6/x   « 1   only if  Rg  is large.  It should be noted  that
with respect  to  the outside scale  x*,  z**  this region is very small
as  F/R2-^3  ->  0  .   However, in an actual reservoir


                              3/2  ,_,l/4
                                          = 0(10 ) ft.
t Asterisks have  been  dropped for convenience.
                                 36

-------
 and  the inertial region may occupy the whole of the reservoir.  When
 Re   is small,  Eqs. (3.46)  - (3.48) are nowhere valid, and  the  flow
 near the sink is governed  by the full Navier Stokes equations.

 As  in the case of flow at  the leading edge of a flat plate, Eqs.  (3.46)
 -  (3.48) cannot be expected to hold very near the sink.  For large  IU
 this non-uniformity may be examined without actually solving Eqs.
 (3.46)  - (3.48).  Consider distances closer than 0(F1//2).  Let
                                       **
                                                                   (3.49)
 then  Eqs.  (3.43)  - (3.45)  become
                      _G_

                       e
                                                     ***
                                                                   (3.50)
                  *****  t****x***  + ty*
                           ***
                                        ***
                                                                   tr> =o\
                                                                   (3.52)
Hence the flow at the origin is inviscide to 1st order  in  1/Rg  and  the
euations are  iv     ^
equations are given
                                                                  (3'53)
                                                                  (3.54)
t Asterisks have been omitted.
                                37

-------
The solution of Eqs. (3.53) - (3.55) depends on the solution of Eqs.
 (3.46) - (3.48) and vice versa.   Neither admits a similarity solution,
nor are they tractable by linear methods.  For the very special upstream
condition of uniform flow Kao (1970) has solved Eqs.  (3.53) -  (3.55).
Here a slip line is required at  the edge of the layer^  The^elocity
along this slip line remains constant and of order  q2/(eg)^  which
means that around this slip line there is a backward  growing viscous-
buoyancy layer of thickness  0(1^/3).  However, at the moment  it  is a
matter of speculation whether these two layers could  be matched onto a
solution of Eqs. (3.46) -  (3.48).

Below, Eqs. (3.46)  - (3.48) are solved approximately  by an  integral
method.  For  x** = °°  , the withdrawal layer thickness  6   is  matched
to the inner limit  of the outer solution, and at  x** = 0   the boundary
condition chosen is  5*** = i, consistent with the above speculations.
This is further discussed in Section 5, but as long as the  solution is
not very sensitive  to the exact numerical value at  x** = 0  ,  this pro-
cedure should lead  to satisfactory results.


        Structure of the Withdrawal Layer - Inner Solution

In the previous section it was shown that for the flow near the sink,
Eqs. (3.46) -  (3.48) must be solved using Koh's  (1966b) similarity
solution as the upstream boundary condition at  x** = °°  .   For conveni-
ence the equations are reproduced below, omitting the asterisks.

                     u  + w  = 0                                  (3.56)
                      x    z

                 uu   +wu  + p  =Gu                           (3.57)
                    x      z    x      zz

                      PZ = -  p                                    (3.58)


                  u p  + w n  - w = i  p                           (3.59)
                    Kx     yz       G  Kzz
                     I
udz = - 1                                  (3.60)
where  G = (v/D)2 .

This set of equations does not appear  to  possess  a  similarity  solution
satisfying the required boundary conditions,  so to  obtain  at least  an
approximation of the layer growth,  the  above  equations  are solved using
an integral method.
                                 38

-------
Examination of the experimental data obtained by  Koh  (1966b)  shows that
for a particular discharge, the velocity profiles are  self  similar even
in the inner region.  In Section 5 this is confirmed  in  the more
general coordinates  (x**, z***).  This suggests postulating a velocity
variation given by
as is done in the Polhausen technique, and obtaining  f  from the
upstream flow.  However, in that solution   p ~ 1/z***  for large  z***,
which is non-integrable.  To be able to apply  an  integral method the
equations must therefore be premultiplied by a weighting function.
This is more easily done if  f  is an analytic function expressed in
terms of elementary functions.  It is therefore convenient to let

                                         2 _2
                     ,       -a(x)  (1-k z  &  )
                   u(x,z) = — ~ —  - 5 — 5-5-                   (3.61)
                              TT    ti , " n~ \^
                                   (1+z  6  >

                                        ~ 1
                             -b(x)    z6                              0.
                    p(x,z) = — -—  - 5 — 5-                     (3.62)
and by continuity


                          e f    -i .   * . —
                    w - "6 za  1"kz  6
                    w —
                          — :       on  o                          >
                          no    ,n , 2 ,-23
                                {1 +z  6   )


The value of  k  is then chosen  to yield the  best fit  to Koh's (1966b)
solution.  Hie unknown functions  are   a (x) , b(x)  and  6(x) .

Equations  (3.61) -  (3.63) are now substituted into the following
integral form of Eqs.  (3.5?) -  (3.60):
          j - L. — __   {(UU   +wu)  ~n   -  G  «    }dz=0        (3.64)
         J  .-  2 -2.      x     z z    vx      zzz
            (1 +z  o   )
         o
                                                                  (3.65)

-------
                       udz  = -  1                                  (3.66)
                     J
                 p 	O
The factor  1/1+z  6    is included  in  Eq.  (3.64)  to make  px
integrable.  Carrying out the  integration  yields  two ordinary differen-
tial equations and one algebraic  equation  for  a  ,   b  and  6 ,  viz.
          -aa' /£   _k	k_ \   !_  & /    -Ga
            TT  b +21- 105 J+  2      -^2

                        a5 = --                                  (3.69)
                             o — K


The prime interest is to determine  the  growth of the layer, that is,
to find  6  as a function of  x  .   Elimination of  a  and  b  from
Eq. (3.67)  -  (3.69) yields one second order equation for   6  viz.


   (a  65 -  o^ 6) 6" + (o^ 64 + 3a2)  (6'')2  + o/3 6' + a  = 0       (3.70)
where
          a
1   (3-k)


    128   1    (2-k)
2    3   n2   (3-k)2
                                /i + JL _  JE?\
                                I 7  21   105 /
               64    1     /6    k     k2\  1    16  (2-k)
             _ 8  Ilk + 32
          Q4 ~ 5    3-k


The boundary condition is that   6=1   when   x = 0 ,  and that  6
asymptotes to the solution of Eq.  (3.70)  with the inertia terms set to
                                40

-------
zero.  The method of solution was to reduce this boundary  value problem
to an initial value problem with the initial values of   6  ,  specified
at some large  x  , being evaluated from an asymptotic solutions of
Eq.  (3.70).  The equation was then integrated with a standard  fifth
order Runga Kutta algorithm available in the computer center library
under the call name  RK5.

To shorten the numerical integration the asymptotic series for large
x  of  6(x)  was computed by Imberger (1970) to the fifth  term.   The
series he found is given by


           1/3       -1/3       -2/3       -1       -4/3
   6 = K^ x    + K2 x     + K3 x     + K4 x   + K& x     + ---   (3.71)
where
                     K2 =
                      _ = arbitrary
                            2
                     K, =
                      4   „   5
                          14K" K0
                             JL  ^


The similarity of the series (3.71) with that proposed by Koh  (1966b)
should be noted.  However, the above contains extra terms which  are
all dependent on the arbitrary constant  K3 .  This free parameter is
necessary as nothing has been said about the behavior of the solution
at  x = 0 .   Similarly Koh's (1966b) proposed series should have  an
equivalent arbitrary coefficient.  An analogous problem is encountered
in rotating flow problems (Smith  (1969)).

That this arbitrary coefficient must be present can also be inferred
by noting that the independent variable  x  is absent from Eq.  (3.70).
Hence if  6 = g(x)  is a solution then so is  6 = g(x + c).  A formal
                                41

-------
 substitution of  x* = x + c  into  the  series  (3.71)  yields  the  same
 series with  K3  replaced by  K3 - cK-^/3  , but  K^   and   K2   remain
 unchanged.

 The arbitrary constant  K3  was used to convert the  boundary  value
 problem into an initial value problem.  In other words,   Ko   was
 fixed,  6  and  6'  were computed  from the series  (3.71)  for  a  suffi-
 ciently large value of  x , and the Eq. (3.70) was integrated numeri-
 cally to zero.  The constant  K3  was  adjusted until  6=1   at  x = 0.

 Inspection of Eq.  (3.70) shows that the equation has a singular point
when
                    6=6  =   -       .
                        c     cy


The slope of  6  becomes infinite at  this value of   6   and the  assump-
tion of a similar profile can no longer adequately describe  the flow
here.

The value of  k  was chosen to make   6  =1   and  K-, =3.2   the value
obtained by Koh  (1966b) for the present definition of   6 .   With
k = 1/3  the values  K,  and  6_  are 3.15 and 1.17.
                      i        «-

Figure 3.4 shows a comparison of the  velocity profile  given  by  the
integral curve (3.61) and that obtained by Koh  (1966b) .   The  variation
of  6  with  x  for  G = 29.0  and  G = 2.34  is  shown  in Fig.  3.5.
This corresponds to a salt and heat stratification,  respectively.


     Summary of the Layer Structure - Composite Solution  for   6

The overall layer structure is summarized in  Fig. 3.6.   The  solutions
to region I, II and VI have been found explicitly.   In  region  III an
integral method was used to obtain the layer  thickness.   The  solutions
to regions IV and V are not known and further work is  required.

The composite expansion  6  was obtained by adding the  inner  and outer
solution and subtracting  K^ x1/3 , the part  which is  common  to both.
This is shown in Fig. 3.7 and Fig. 3.8 for heat and  salt  stratification.
Unfortunately the wall spacing in the outer problem  does  not  scale with
F1/2 , so that the wall "moves," as F changes in  an  inner (x** ,  z***)
coordinate system.  The advantage of  this coordinate system, however,
is that near the sink the solution only depends on   G  .
                                42

-------
00
                         VELOCITY PROFILE
                         (after Koh  1966 b)
                                  ASSUMED INTEGRAL PROFILE
         0            0.5            1.0            1.5            2.0
           NON-DIMENSIONAL VERTICAL  DISTANCE FROM THE CENTERLINE
Z Umax
                      FIG. 3.4  INTEGRAL SOLUTION VELOCITY PROFILE

-------
                     INTEGRAL SOLUTION
                                           LINEAR THEORY
O
     0
      01234
         CUBE ROOTOF THE NON-DIMENSIONAL DISTANCE FROM THE SINK (
               FIG. 3.5  THEORETICAL WITHDRAWAL LAYER THICKNESS

-------
        'A/ / ////////////
                                    SLIP WALL
SLIP LAYER>
                  m   \
               BOUNDARY  t
               LAYER FLOW \
VISCOUS-BUOYANCY BALANCE

     (Eq. 3.41-3)
                                                                       UNIFORM

                                                                       PARALLEL
                                                                       FLOW
SINIT


^V 	 " X=xc «-
NNVISCID
CORE
(Eq. 3.70-2)
<— x


                 FIG. 3.6  STRUCTURE OF THE WITHDRAWAL LAYER

-------
05
       60
   
-------
5 uj
ox30
^£20


Q -J
 i
O
       /////// / / / / / / / / / / / / / / /s? / / / / / / / / / / / / / / /
              INNER SOLUTION
             I27T
                                        //////////// / / / / /.
                                                              67T
      0
                                  10
15
20
        CUBE ROOT OF THE NON-DIMENSIONAL DISTANCE FROM THE SINK (x/xc V*


                 FIG 3.8 THE COMPOSITE SOLUTION   G - 2 9.0

-------
             4.  EXPERIMENTAL OBJECTIVES AND APPARATUS

The analytic considerations of the previous chapter showed that the
dynamics of the flow due to a withdrawal from a stratified fluid may
be represented in a scaled coordinate system.  In this coordinate system
the only parameters remaining are the square root of the Schmidt number
and the depth of the tank.  Hence, when viewed from this reference axes
all flows in a fluid of constant Schmidt number are dynamically similar.


               Review of Previous Experimental Data

The experimental results obtained by Koh  (1966b) are shown re-plotted
in Fig. 4.1.  Each line represents the line of best fit to the experi-
mental points of a single run.  The withdrawal layer thickness has been
reduced by 0.9, from that used by Koh  (1966b) , to correspond  to the
definition used in this report.  It is seen that the two cases, salt
and heat stratification, fall roughly  into two separate lines on the
graph.  For the heat stratified mixture,  the results show little scatter
and good agreement with the theoretical predictions.  However, when salt
is used, although the agreement is reasonable for small values of
x/x   , the data diverges somewhat for  larger values.  The explanation
for this is probably twofold.
 (a)   The  data  for  large   (x/Xc)     are from  the  rear  of  the  tank,
where end effects  are probably affecting the  data.   This  suspicion  is
reinforced by  the  fact that  qf  , the forward discharge,  varied  con-
siderably from one station to the next for  the  same  run.

 (b)   The  definition of   6  as that point where  the velocity  is zero is
very  difficult to  measure  (Section 5.1).  This  is especially  so  for
larger  values  of   (x/x) •*•'•*  where the velocities are  very slow.   This
could account  for  some of  the scatter,

                      Experimental Objectives

In  view of the above, an experimental program was carried out in a
facility  much  larger than  that used by Koh  (1966b) to  repeat  and extend
his data. In  this way the transient effects  due  to  a  finite  sized  tank
were  kept to a minimum.  All the experiments  were conducted  in a salt
stratified solution, rather than one with a temperature  gradient,  to
avoid the side walls induced convection boundary  layers.   A  perfectly
still reservoir is essential, since the velocities in  the withdrawal
layers  are extremely small.  All the experiments  were  carried out in
the same  42' long  glass  tank.  For each stratification about six to nine
runs  at different  discharges could be made  before the  dye contamination
affected  visibility.  Four different stratifications were used.

                       The Experimental Tank

  The  experiments were performed  in a 42 foot  long by 4.0 foot deep by
  lj foot  wide  tank of the  University of California  Richmond Field

                                 49

-------
                                                   INTEGRAL CURVES
                                                   FOR  G =  29.0
                                                   AND  G =   2.34
                                      G = 2.34
                              LINEAR THEORY
0
 0              I              2              3               4
    CUBE ROOT OF THE NON-DIMENSIONAL DISTANCE  FROM THE SINK (Vxc)'3

               FIG 4.1  EXPERIMENTAL DATA AFTER  KOH I966b.

-------
                               PROBE STAND
      DEPTH SCALE
    CRANK
(to adjust depth
 of  probe )
                                                                                         -TROLLEY
                                                         ,-PLATES TO
                                                         \STRATIFY TANK
CALIBRATION
 MIXTURE
                                         TROLLEY
                                              SALINITY
                                              PROBE
 DISCHARGE SLOTS
                     FIG.  4.2  GENERAL  LAYOUT OF EXPERIMENTAL TANK

-------
Station.  The walls and bottom were constructed of tempered plate
glass.  The tank is shown schematically in Fig. 4.2 and a photograph
of the general layout is given in Fig. 4.3.

Two trolleys were placed on rails on top of the tank.  One served  as
a mounting platform for the stratifier while the  other held the
salinity probe and the associated vacuum pump.  The discharge  facil-
ity was located at one end.  Mounting the salinity probe on the
trolley allowed the same probe to be used to measure the density
gradients throughout the channel, and the salinity of the discharging
water.

                       The Salinity Probe

To measure the concentration of salinity a conductivity probe  was
built and mounted on one of the trolleys.  The design of the probe
was similar to those used successfully at the Delf Hydraulics  Labora-
tory, as observed by the second author during a visit in
1968.  The probe was used to measure the background near linear
salinity profile.  The salt used was an ordinary  kiln dried commercial
product.  No attempt was made to measure the very small deviations of
salinity due to the motion.

The probe and its associated circuit are shown in Fig. 4.4.  A photo-
graph of its mounting on the trolley is given in  Fig. 4.3(a).  The
probe was constructed from a thin plastic tube, which was closed to
a fine hole of a diameter 0.040" at one end, after the electrodes  had
been inserted.  The two platinum electrodes, one  on the inside, one
around the outside of the tube, created an electric field which had all
its field lines passing through the fine hole at  the tip of the probe
(Fig. 4.3(b)).  The other end of the 4' tube was  connected to  a small
vacuum pump which sucked fluid through the small  opening at the tip.
The actual volume sucked in amounted to a few drops per second, but
the velocity induced within the contraction was quite high.  In this
way the fluid is replaced almost instantaneously  in the contracted
opening as the density fluctuates in front of the probe.  The  fact
that the electric field is concentrated in the opening, as well means
that the probe measures the salinity of the water in the opening
only, and the rapid change of fluid there makes the probe respond  very
quickly to any density change.  The profiles could therefore be ob-
tained quite rapidly.

To calibrate the probe, a large bottle was filled with water and salt
was added to make a mixture of a concentration corresponding to the
maximum salinity anticipated.  This served as the stock solution.   The
specific gravity of the mixture was then measured with a hydrometer.
This method proved to be much more accurate than  weighing out  a cer-
tain amount of salt, adding it to water and then  neglecting the
volume of the salt.  After the tank had been stratified a series of
5 dilutions of the stock solution were prepared so that their  specific
gravity covered the range of the tank stratification.  These solutions
                               52

-------
   (a)
(b)
                         (c)
Fig.  4.3  Photographs of Experimental  Equipment
                      53

-------
     BRIDGE
 CARRIER
 AMPLIFIER
BRUSH INST.
R.V.4622-00
CHART
RECORDER
BRUSH INST
 M.K. 220
icon
IOOA
                   GRID
 CONDUCTANCE
 PROBE
POSITION
TRANSDUCER
LOG HEED
ELECT.
                                                             POWER SUPPLY
                                                                SRC 3566
UT
k-

Ch. 1
Ch2
i
4
frl
1 W \
\ \
\ U
•
6v D. C.
0.02 °/0
REG
                                                     FLOAT
                  FIG 4-4  CONDUCTANCE PROBE
                             AND FLOAT CIRCUIT

-------
were then used to calibrate the probe after they had reached the
temperature of the water in the tank.

The difference of the temperature of tank water and the temperature
at which the specific gravity was measured was always small, and
since only the density gradient of the tank water is required, no
compensation was made for this temperature difference.  The cali-
bration curves were nearly linear, with a slight decrease  in
sensitivity for the larger concentrations.  Figure 4.5(a)  shows a
typical calibration curve.  Over a day no drift was noticed.  The
probe was, however, recalibrated directly before a density profile
was taken.

              Establishment of a Linear Density Profile

The aim was to stratify the tank water so that the density  of the
salt water mixture varied linearly with depth.  The technique used
is essentially that reported by Clark et al. , (1967).

The whole tank was filled with fresh water, which was allowed to
come to room temperature.  A little detergent was added to reduce
the formation of bubbles on the side of the glass which would other-
wise hamper the visibility.  A gated partition was then introduced
vertically  into the center of the tank.  The partition consisted of
a piece of plywood with 9" hinged flaps at the bottom and  top (Fig.
4.2).  The partition was sealed to the glass walls with a  soft  rubber
spacer.   This was sufficient to stop any transfer of water from one
side to the other.

The appropriate amount of salt was then stirred into one half of the
partition tank.  To facilitate mixing the water was pumped  from the
bottom to the top.  Once the mixture was uniform it was allowed to
come to rest.  The flaps of the partition were then opened  very
slightly  to allow the salty water to flow beneath the fresh water and
the fresh water to flow over the  salty water.  About 1/2 hour elapsed
before the whole tank had a two layer stratification in it.  The
partition was then carefully removed.

To obtain a linear profile, a series of three plates, mounted on the
trolley, were then briskly moved  through the  interface with a crank
and pulley arrangement at the end of the tank.  The largest plate
moved directly through the interface while the two smaller ones
stirred up the top and bottom layers.  The wakes behind the plates
created sufficient turbulence to  allow mixing.  The mixing smoothed
out the interface, and after some adjustment  of the position of the
plates a very linear profile was  obtained throughout the whole depth.
Figure 4.2 shows the general layout  of the pulley, trolley and  the
mounted plates.  A typical density profile is given in Fig. 4.5(b),
showing how the linearity holds right to the  bottom and the free
surface.
                                55

-------
en
O)
                     PROBE CALIBRATION
                                       1.02          1.03

                                 SPECIFIC GRAVITY
                   FIG. 4.5a EXAMPLE OF THE  PROBE CALIBRATION

-------
Ol
                  WATER LEV
        CO
        LU
        Q_
        LU
        Q
            0
           10
           20
           30
           40
             1.00
                                           DENSITY  PROFILE
1.01           1.02
       SPECIFIC  GRAVITY
1.03
1.04
                  FIG.4.5b  EXAMPLE OF THE DENSITY PROFILE

-------
The linear profile was very stable and would last unaffected for up
to two weeks.  Only at the surface and bottom was there a slight
degeneration of the profile.   At these extremes a layer of uniform
density formed, the thickness of which increased with time.   Most
probably this was a temperature induced thermocline formation, as
the room air conditioning did not accurately regulate the air
temperature.

                       The Discharge Meter

To ensure constant monitoring of the discharge flowing out through
the sink, a float type discharge meter was installed.  This provided
a means of adjusting the flow rate at a desired value and also to
check its constancy as the experiment proceeded.  The actual dis-
charge used in the reduction of the data was, however, obtained by
integrating the velocity profiles.  The procedure used is explained
in Section 5.

The flow from the slotted discharge tube was drained off from both
sides of the tank into two plastic tubes.  The two tubes were joined
with a T section and the flow entered the discharge meter through a
gate valve behind the T.  The general layout is shown in Fig. 4.6
and the circuit used is depicted in Fig. 4.4.  On entering the meter
the water spilled over a cup down a pipe into the float chamber.  The
float motion activated a pully arrangement which in turn was moni-
tored by a potentiometer.

                        The Dye Particles

The velocities induced by the fluid withdrawal are very small.  The
magnitudes are of the order of a few centimeters per minute.  For
such low velocities the only suitable method of measurement appears
to be that of dyeing parts of the fluid and taking photographs of the
deformation of the dye lines.  The method used was to drop crystals
of dye into the center of the tank at incremented distances from the
sink.   The purple streaks left in the wake of the falling crystals
were then photographed repeatedly.

Of the dyes available, it was found that potassium permanganate was
the most suitable for the present experiment.  In contrast to organic
dyes,  this has the great advantage that it forms very fine intense
streaks, the lifetime of which are only about 1/4 hour.   Their rapid
decay, which is due to oxidation, allows experiments to be repeated
quite quickly.

The crystals used were about 1/4 - 1/2 mm. in diameter and care was
taken to choose the most spherical ones only.  Elongated or plate-like
crystals form  spiral paths which makes the velocity measurement more
tedious.  A razor blade was used to split larger crystals down to
the required size and shape.   The cutting of larger crystals also
exposed fresh permanganate and much more intense streaks could be
                               58

-------
T
       'POTENTI METER
        • WIRE
         •ROD
          •PLASTIC
           FLOAT
           WATER
                           CONSTANT HEAD
                           OVERFLOW

^

'? ,

\
                                     TANK
                  OUTLET
FIG 4.6 DISCHARGE METER
            59

-------
obtained.  The crystals were placed,  after cutting,  into a small re-
cess in a wooden dowling at a spacing of about 10 cm.   This wooden
bar was then placed on top of the tank.   To introduce  the crystals
into the water, the bar was turned through 180°.

                     Photography and Timing

The motion of the dye streaks was recorded photographically.  The
time between photographs was fixed by tripping the camera shutter
and the event marker on the recorder, simultaneously.

A Rolleflex S.L. 66 with an 80 mm lens was used throughout the ex-
periments.  The camera was fixed at one of two predetermined posi-
tions on a sturdy bench.  The positions were so chosen that two
photographs covered dye lines ranging in distance from 0 to 160 cm
from the sink.  The film used was Plus x professional and this was
developed in D 76 to a characteristic value of  .65.  The negatives
were then enlarged to 1/2 full scale on ll" x 14" Resisto-Rapid
photographic paper.  The plastic base of this paper makes it dimen-
sionally stable.  Successive photographs were aligned over each
other, by aligning the images of a series of marks on the glass walls
of the tank.  The profiles were then pricked through with a sharp
needle.  A typical dye profile is shown in Fig. 4.7.

To fix the scale of the photographs a 200 cm measuring scale was
held into the water at the center of the tank, where the dye lines
normally are.  This was then photographed.  As  the position of  the
bench and the  tank remained fixed, this calibration had to be done
only once.  Also in this way parallax errors do not occur.  The
scale negative was used to set the enlarger to  1/2 full scale.  The
alignment marks on the side of the tank offered spot checks on  the
scale of each  photograph.

The camera shutter was released with an electric  relay, which also
activated the  event marker on the Brush recorder.  The time between
successive photographs was thus displayed by the  distance between
marks on the  chart.  Times from 10 to 200 sec.  were typical.  The
accuracy, estimated by comparison with measurements with  a  stop watch,
was near 0.3  sec.

                     Experimental Procedure

After the tank had been filled with  tap water  and allowed  to stand
for a day or  so, the procedure  for each experimental run was identi-
cal and  may  be summarized  as  follows:

(a)  A small  amount of Photoflo detergent was  added and the
      tank water was stirred.

(b)  The  temperature  of the  room  and the water  was measured.
     The difference was kept  less than  2°C.
                               60

-------
Fig.  4.7  Example of Dye Traces
           '

-------
 (c)  The wooden partition was inserted at the center of the tank.

 (d)  The required amount of salt was dumped into one side.

 (e)  The salt was mixed by first agitating the water with a broom
     and then circulating the mixture with a small pump, mounted
     on the trolley.  After the mixture was uniform it was allowed
     to come to rest.

 (f)  The flaps on the partition were opened about l".  After about
     1/2 hour the interface was continuous past the partition and
     the latter was removed.

 (g)  The stratifying plates were pulled through the interface.

 (h)  Calibration solutions (dilutions of the stock solution)
     were prepared and their temperature was checked against that
     of the tank water.

 (i)  The probe was calibrated and then used to obtain a density
     profile of the tank water.

 (j)  The gate valve at the discharge end was set to yield the
     required flow.

 (k)  Crystals of permanganate were cut to the right size and placed
     on the wooden trough on the top of the tank.

 (1)  The camera was loaded and mounted in the rear position.
     The lights were adjusted.

 (m)  The rear crystals were dropped and the dye lines were photo-
     graphed.   The time between (j) and (m) was at least 1/2 an hour.

 (n)  The camera and lights were moved to the front window and the
     procedure listed in (m) was repeated.

 (o)  The salinity of the discharge water was measured.

 (p)  The gate valve was reset to a new flow rate and the procedure
     was begun at (j) again.

                  Experimental Program and Data

The experimental program is summarized in Table 4.1.  All the data,
giving the withdrawal layer thicknesses at different distances from
the sink, are reproduced in Appendix I.  The aim was to vary G and q
to yield a good variation of xc.  The effect of different viscosities
or diffusivities was not investigated.  The specific gravity of the
discharge water was always equal to that corresponding to the specific
gravity of the tank water at the level of the sink and it is therefore
not included in the  table.

                               62

-------
                           TABLE 4.1.  EXPERIMENTAL PROGRAM  (Key  to  table  on following page)
Exp.
No.
1-1
1-2
1-3
2-1
2-2
2-3
3-1
3-2
3-3
3-4
3-5
3-6
4-1
4-2
4-3
4-5
4-6
4-7
4-8
4-9
T °r
*W
14°C
14°
14°
20°
20°
20°
14°
15°
15°
18°
18°
18°
14°
14°
16°
16°
16°
14°
14°
14°
V
in-5 2 -1
10 cm sec
1120
1120
1120
1000
1000
1000
1120
1100
1100
1040
1040
1040
1120
1120
1080
1080
1080
1120
1120
1120
D
nn-5 2 -1
10 cm sec
1.25
1.25
1. 25
1.25
1.25
1. 25
1.25
1. 25
1.25
1.25
1.25
1,25
1.25
1.25
1.75
1.25
1.25
1.25
1.75
1. 25
€
I0"5cm"1
2.98
2.98
2.98
25.7
25.7
25.7
29.9
29.9
29.9
29.9
29.9
29.9
35.3
35.3
35.3
35.3
35.3
35.3
35.3
35.3
q
2 -1
cm sec
.210
.320
.454
.006
9
?
.005
.013
.031
.064
.118
.185
.006
.005
.007
.009
.012
.017
.022
.037
%
2 -1
cm sec
.225
.366
.496
.012
.062
.272
.007
.014
.033
.073
.143
.233
.007
.007
.010
.011
.013
.020
.024
.044
G
approx.
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
F
m
2
cm
1.31
2.14
2.90
.024
.124
.543
.013
.026
,062
.136
.264
.429
.012
.012
.017
.019
.022
.034
.041
.075
a3
o
,~3 -2
10 cm
.458
.458
.458
1.46
1.46
1.46
1.45
1.46
1.46
1.50
1.50
1.50
1.57
1.57
1.60
1.60
1.60
1.57
1.57
1.57
X
c
cm
688
1439
2260
5.4
64.3
584.0
2.18
6.43
22.8
25.0
204.0
420.0
2.14
2.22
3.81
4.38
5.85
9.90
13.40
32.2
w

-------
Column 1:  The first number corresponds to the density stratification,
           while the second designates the run number for that
           stratification.

Column 2:  T  is the temperature of the water in degrees centigrade.

Column 3:  v is the kinematic viscosity.  The values are those of
           pure water at T °C.
                          w

Column 4:  D is the molecular diffusivity of salt in water.

               i  dP
               1    e
Column 5:  e = --   computed by the formula.
               p   dz
                o
               Sp.  Gravity at z  - Sp. Gravity at z
Column 6:   q is the discharge measured with the discharge meter
           divided by the width of the channel.

Column 7:   q  is the forward discharge obtained by  integration of
           the velocity profiles (Section 4.2).

                                       1/2
Column 8:   G is the mean value of (v/D)    for all  the runs.

Column 9:   F  is the Froude number q //eg .
            m                       m
            3
Column 10:  a  is the inverse of the Rayleigh number.  This is used
           to facilitate comparison with Koh' s 1966b data (Sg/vD)5

Column 11:  x  is the critical distance below which  inertial effects
           become important.  (p^/2 Q>3)
                                m    o
                               64

-------
       5.  ANALYSIS AND DISCUSSION OF EXPERIMENTAL RESULTS

                        Analysis of Data

The half scale photographs of the dye lines at consecutive times
formed the unprocessed velocity data.  The time between photo-
graphs was obtained by measuring the length of chart between
notches at the event marker trace.   The reduction of this data may
be divided into three parts,

(a)  Discharge calculations: The no slip condition on the side walls
forced the layer to have a parabolic velocity variation across the
width of the tank, rather than a uniform distribution as assumed
in the theory.  For this reason the local discharge q , obtained by
integrating the velocity profile, was used in the reduction of the
data.  To obtain qm, successive photographs were aligned and  the
top profile was pricked through onto the other photograph.  The
area between the two curves of 6 roughly equally spaced profiles was
then measured with a planitneter.  The total net area was used rather
than just the forward flowing part.  To within 10% the areas  were
the same and the mean of the 6 areas was used to calculate the
discharge by the formula

           q  = mean area T time between photographs.

(b)  Withdrawal layer thickness 6:   In the integral solution  6 was
defined by the velocity profile

                                     2 —2
                           a(x)  1-kz 6
when z = 6
                            TT       2s;-2,
                                (1+z 6  )
                                   1
                            u      12
                             max
hence a convenient definition for 6 in the experimental data is that
width at which the velocity has dropped to 1/12 the center line
value.   This value of 5 was also much better defined on the photo-
graphs than the point where the velocity is zero.   The profile had
a large non-zero slope here, and it formed a definite intersection
with the constant line of u = — u   .   The values of 6, measured as
                              J_ ^  TrlclX
described, and the above value of qm were then used to set up a table
as shown in Appendix I.   The actual quantities plotted on Fig. 5.1 are:


                      f^l/3               3/2
               versus  I—)    ,   where x  = F
                      \x /    '          cm
                        c
                               65

-------
03
CT)
        CO
        CO
        or
        LJ
                        INTEGRAL SOLUTION
          10
       CE
       Q

       X
       o
       CO
       o

       z
       o
          0
                                                      i  1
              0.2
 t
0.67
                           1.0
1.62
      2.0
3.0
4.0 f
  4.13
                  CUBE ROOT OF THE NON-DIMENSIONAL DISTANCE FROM THE SINK (*/x )/3
                                                                                    v

                        FIG.5.1 EXPERIMENTAL WITHDRAWAL LAYER THICKNESS!  G =29.0

-------
                                                 1/3
(c)  Velocity Profiles:   At four values of (x/x )    the actual
velocity profiles were reduced.  To obtain the velocity profile, the
image of the dye lines was transferred onto a transparent graph
paper.  From this the ordinates were read off and divided by the
time between photographs to yield velocities.  The above velocity
profiles were all plotted on a single graph by normalizing the
maximum velocity and the discharge to 1.   This is shown in Fig.
5.2.  The solid line again corresponds to the integral velocity
distribution.  Rewritten to have u/u    as the ordinant and zu
    .,    ,        ., .       ,           max                       max
as the abscissa this reads:

                            1 - .367(zu   )2
                     u                 max
                     max   (l+l.l(zu   ))
                                      max

                Effects of the Finite Size of the Tank

Figures 5.1 and 5. 2 show little scatter, indicating that  random
errors due to parallax, enlarging, and aligning of photographs are
quite small.  Systematic errors are, however, possible as the tank
has a finite width and a finite length.   The possibility  of  a
systematic error of this sort is now assessed.

The finite width introduces a parabolic profile in the y  direction.
Koh (1966b) has made a detailed analysis of the effects of the para-
bolic profile on the withdrawal layer thickness.  As long as the
scale in the vertical direction, 26, is smaller than the  width, and
provided the local centerline discharge qm is used, the error intro-
duced will be small.  In the present experiments the ratio of thick-
ness to width of the layer was kept to less than 0.1 by taking data
only in the first 6 feet of the tank.

The tank was long enough so that the measured discharges  in  the test
section were independent of the distance from the sink.   This means
that the layer spread sufficiently to prevent any influence  of the
back wall.  To ensure that the layer far upstream was inertia free,
the discharge q was limited to make x  less than the length  of the
tank.  In the experiments xc was kept, with one exception, to below
1/6 of the tank length.

With the above precautions, it is reasonable to assume that  the data
is a true representation of flow in a long reservoir.

    Comparison With The Theoretical Results And Previous  Experiments

The present experimental results have been summarized on  Figs. 5.1
and 5.2.  The data from Koh (1966b) was replotted to conform with
theoretical scaling as shown in Fig. 4.1.  Comparison shows  excellent
agreement up to (x/xc) '   of approximately 2.  For larger values of
this parameter Koh (I966b) found values of 5 slightly larger than
                               67

-------
en
CO
     X
     O
     E
       -1.0
    o
    O
     >
     _l
     <
o
M
cc
O
     O
     c/)
     Z
     LU   0
     O
                         _____	!	


                         INTEGRAL  PROFILE
                                   I-0.367 (-
          0
                                              •>']   J
          KEY  SYMBOL
            I
            2      o
            3     A
            4      n
                                       4.13
                                       1.62
                                       0.67
                                       0.20
                                                                        * ff
                     0.5
.0
1.5
2.0
                NON-DIMENSIONAL VERTICAL DISTANCE FROM THE CENTERLINE

                           FIG 5.2 EXPERIMENTAL VELOCITY  PROFILES! G = 29.0
                                                                                Z U
2.5
                                                                                   max

-------
those measured by the author.  However, as already pointed out, this
difference is not significant as it is most probably due to end wall
effects in the former's experiments.

Theory showed that the flow  should be independent of any parameter
except G, when the x and z coordinates are scaled correctly.   This
is verified in Figs. 5.1 and 4.1 where the data is seen to fall all
onto one curve for a particular G.  The above observation is  equally
true for .the  salt stratification where G is quite large, even though
the theory was developed for G of order one.  This may be explained
by noting that the inner Eqs. (3.46) - (3.48) are merely a boundary
layer  type simplification of the full equations and therefore still
contain a representation of  all the forces.  Thus even if there are
for large G,  regions in which diffusion is absent, the governing
equations there would be special cases of the more general Eqs.
(3.46) -  (3.48).

For the same  reason  a large  value of G does not invalidate the inte-
gral solution results.  A comparison between the theory  and  experi-
ment is shown in  Fig. 5.1 and it can be considered good, everywhere
except near  the  sink.  The difficulty here results from  the  still
unknown behavior  of  the flow very near the sink.  There  is  a tempta-
tion to adjust K3, the floating upstream constant, to  improve the fit
near the  origin.  Figure 5.3 shows  curves  for different  values of K3
and  it is  seen that  the integral  curve for K3 = -82.2  fits  the data
better than  the  originally adopted  curve, which has K3 = -81.8.   This
is  an  encouraging flexibility of  the method, but the  original choice
of  K3  remains the only theoretically justifiable one until  a more
definite  solution for the flow very close  to the sink  is obtained,  to
which  the  integral  solution  may be  matched.

The depth  of the  experimental tank  was 4  feet, which was large com-
pared  with  the values of F1/2 used, so that  the top  and  bottom
boundaries  did not  influence the  flow  for  the experimental  values of
 (x/xc).   This meant  that the linear or outer solution,  as  derived
 in  Section 3.3 and  included  in  Fig. 5.1, was  identical with Koh1s
 (1966b)  solution.   Inspection of  Fig.  5.1  shows that  the integral
 solution,  which  reflects the effects  of  the  inertial  terms  in the
equations  of motion, predicts layer thicknesses more  in  agreement
with the  experimental data  than either the solution  of Koh  (1966b)  or
 the solution 6/F1/2  =1.35  found  by Kao  (1970).
                 m
 By  examining Koh1s  (1966b)  data Brooks and Koh  (1969)  anticipated to
 some extent  the  theoretical  scaling.   In Fig.  9 of that  reference
Koh's  (1966b) data  is  replotted in terms of

                         a             q
                        —  versus
                                      „  2/3
                                      X x
                                       o
                                69

-------
01234
CUBE ROOT OF THE NON-DIMENSIONAL DISTANCE FROM THE SINK (Mcc)'/3
  FIG.5.3  VARIATION OF THE INTEGRAL SOLUTION WITH  K3
                                                                oo

-------
A little rearranging shows that this is in the present  notation
              3.57 f
   1/3
X i
 c
        versus
                                          x .2/3
PN<
\x 1
The only real difference between the above and that established  in
Section 3 is the factor (x/xc)1//3 in the ordinant.

It was hoped that the velocity profile measurements would  indicate
some of the structure changes very close to the sink,  and  clarified
the nature of the inviscid core.  Figure 5.2 shows, however,  that
even for x of the order of F1'2 the velocity profiles  are  self simi-
lar to those for large (x/xc).  The only change is that  the  back
flow velocity decreases, in relation to the center line  velocity, as
x is decreased or, equivalently, as q is increased.  Further work is
needed both experimentally and theoretically to describe the flow
close to the sink.
                               71

-------
                6.   COMPARISON WITH FIELD DATA

The tests conducted  by  the  Tennessee Valley Authority on some of their
reservoirs  appears to be  the  only  source of actual reservoir data.
The work summarized  in  the  T.V.A.  reports (1969(a),(b))  is concerned
with the measurement of the velocity,  temperature, and  water oxygen
content distribution in a reservoir from which water is being drained.
The investigations involved the Fontana and Cherokee Reservoirs with
further work  in progress  in the Douglas and Barron-Nolin Reservoirs.

The difficulty of measuring velocities  is compounded by the unsteady
nature of the operating discharge  and  the irregular geometry of most
reservoirs. Over the first  20  miles the Cherokee Reservoir is in the
form of a 1 mile wide trough making it  the most suitable example for
a comparison with the simplified geometry of the theory.  Altogether,
there are data from 5 steady runs,  4 from the Cherokee  and one from
the Fontana reservoir.  The complete data is summarized in Table 5.1.
The investigations dealing  with the transient response  are still very
limited;  only one complete  run has  been published.

                        Unsteady Withdrawal

In Section  3  it was shown that the  time scale of the reservoir res-
ponse , to an  increase in  turbine discharge,  was of the  order of
l//£g.  The actual time it  will take to form a withdrawal  layer of
thickness 6   at a distance  x from  the sink was found to be given by
            S
                                 1    TTX
                            t = /^F  6   •
The value of 6g is obtained  from the composite  solution given in Fig.
3.7.  The turbine response for the two partial  tests  given was very
close to a step function, and the time taken  to reach steady  state,
at the station corresponding to the data 8-16-67, was of the  order of
12 hours.  By comparison the theoretical value  yields

                     1            8980      1     , „  „  ,
            t =          ==  TT                 = 13.6  hrs.
                /•? QvTn~4    lo.bXl.bU dbUU
                i*J O . O A 1 U

The agreement is rather encouraging.

                        Steady Withdrawal

Figure 6.1 shows the available prototype data,  plotted  non-dimension-
ally with respect  to molecular transport properties,  as described  in
Section 3.  The half layer thickness chosen was that  which corres-
ponded to a velocity 1/12 the maximum velocity.   The  theoretical
curve is shown for comparison.  The prototype values  are seen to be
50-100% larger than those predicted.  This  is especially noticeable
since the experimental results in Fig. 5.1  are  always a little smaller
                               73

-------
TABLE 6.1.  PROTOTYPE DATA FROM T.V.A.  (1969(a) &  (b)

Date




T°C



V
io"5
ft*
-1
sec
D
TO'5
ft*
-1
sec
G
TO'5
ft'1


eg
10"1

-i
sec
3
ao
TO3
ft'2


g

ft2
-1
sec
F
m

ft2


rJ
m
ft.


X
c
IO6
ft


6

ft


V
F^
m


X

ft


(x/xc)




2*

ft


2V
p2
m


Cherokee
8-17-57
9-20-67
8-16-67
8-24-67
9-14-67
25
20
20
Ex
20
0.900
0.980
0.980
traneou
0.980
0.144
0.144
0.144
s curre
0.144
12
0.36
1.2
nts.
1.65
3.8
1.1
3.8

5.3
5.2
2.8
5.2

6.1
5.6
5.2
6.75

3.8
288
495
346

165
17.0
22.2
18.6

12.8
25.5
30.8
33.3

12.8
40
40
40

30
2.4
1.8
2.2

2.3
2,100
2,100
8,980

52,800
.049
.041
.065

.16
150
140
140

120
8.9
6.3
7.5

9.4
Font ana
9-9-66
20
.980
.144
1.11
3.6
5.1
7.6
400
20
30.6
45
2.3
3,430
.022
330
16.5

-------
Ol
U. \ o
0 ao £

ten
2E
tn i-
LU a:  |
-5 LU  '
Q
 i
         0
            x =i
                           (0.065,1.6)
                                             DATE  RESERVOIR SOURCE
                       LAYER THICKNESS
                       IN AN INVISCID FLUID
                       (BROOKS 8 KOH 1969)
                                      V
                                      0
                                      o
                                      D
9-9-66
9-10-67
8-17-67
8-16-67
FONTANA ^jT.V.A. WATER
CHEROKEE  RESOURCES
CHEROKEE\RESEARCH
CHEROKEE! REPORT N0.4
                                      A 9-14-67 CHEROKEEJ AND 13
           0             0.05            O.I             0.15            0.2
            CUBE ROOT OF THE NON-DIMENSIONAL DISTANCE FROM THE SINK (Vx,
                            FIG. 6.1  PROTOTYPE  EXPERIMENTAL DATA

-------
than the integral  curve  for  such small values of

Column 11 of Table 6.1  shows that for molecular transport coeffi-
cients, the corresponding values of xc for both the  Cherokee and
Fontana data are  about  10-100 times larger than the  length of the
reservoirs.  This  means  that the inertial region III of  Fig. 3.6
extends over the whole  reservoir.  Consequently the  outer solution
is valid nowhere  in the  reservoir.   Brooks and Koh (1969), in an
effort to explain  the difference between the observed withdrawal
layer thickness  and that predicted by Koh (I966b) , postulated eddy
transport coefficients  much  larger than their molecular  counter-
parts.  The use  of Koh's (1966b) solution is now seen to be
incorrect, but  a  similar procedure can be used for the present
theory.  An increase in v and D, decreases xc and this moves the
data points to  the right in Fig. 6.1.  With the assumption that
v = D, a typical  value  of the eddy transport values, sufficient to
move the data onto the  theoretical curve is 10~3 ft2 sec" .  This
is an  increase  over the molecular values of 102, considerably less
than the 103 postulated by Brooks and Koh (1969) on  the  basis of Koh's
(1966b) theory.   The value of xc, corresponding to such  transport
coefficients and the mean of the data for the first  three runs listed
in Table 6.1,  is approximately 23 miles, which is roughly the length
of the  reservoir.   This means therefore that the use of  the higher
transport  coefficients still requires the inclusion  of inertial
effects.

The  weakness  of the above procedure  is that as yet,  there is no ex-
planation  of  a  mechanism which could  sustain these larger eddy
 transport  coefficients.  Measurements  (T.V.A. 1969b) of the local
 Richardson number with depth, for a  typical run  indicate that this
 varies between 10 and 10,000; values  much higher than required for
 local  shear instability.

 The  large  values  of x  encountered  in reservoirs with peak dis-
 charges means that the end walls cannot be neglected and these could
have 1st  order effects on the thickness of the layer.  An end wall
 stops  the  withdrawal layer  and much  larger reverse  flows are induced.
A  large discharge could  cause a  vigorous layer to extend to the  rear
wall  and  this would  induce  a reverse  layer above  and below the center
layer.   Schiff  (1966) has found, in  a related problem, that this  in-
duced  snaking of the flow with possible mixing at the boundaries  leads
to increased  diffusion of the species.

In summary  it can be said that the  theory, although yielding good
agreement with  laboratory experiments, is still  not  able to completely
explain prototype behavior.   However,  it does yield  a rational  basis
from which to proceed to include the  effects  of  larger eddy transport
coefficients and  end wall effects.
                                 76

-------
                            ACKNOWLEDGEMENTS

The authors would Like to thank Dr. G. Corcos and Dr. A, Lewy  for
reading the manuscript and offering their valuable  suggestions.
Appreciation is also due to Mr. K. Bremel who designed  and  superx^ised
the construction of the experimental  tank; Mr.  E. Parscale,  who
built most of the experimental equipment; Mr. W. Kot, who prepared
the drawings; and Mrs. D. Aoki and Mrs.  D. Haasarud, who typed the
manuscript.

The first writer wishes to thank  the  Robert and Maude Gledden  Fund  of
the University of Western Australia for  making  it possible  for him  to
come to Berkeley by providing a fellowship for  two  years.   Furthermore,
he would like to thank the University of California for providing him
with a Science Fellowship in the  third year.

The initial portion of the research was  supported by funds  from  the
Water Resources Center of the University of California  under project
W-204.   The support of the project by the Environmental Protection
Agency and the help provided by Mr. Phillip C.  Woods, Grant Project
Officer, is acknowledged -with sincere thanks.
                                 77

-------
                           REFERENCES

Beard,  L.,  and Wiley, 1970.  An approach to reservoir temperature
     analysis, Water Resources Research (in press).

Bellman, R.,  1953.  Stability theory of differential equations.
     McGraw-Hill, New York.

Brooks, N.  H., and Koh, R. C. Y. , 1969.  Selective withdrawal  from
     density-stratified reservoirs.  Journal of the Hydraulics Div.,
     Proc.  ASCE, Vol. 95, HY4, pp. 1369-1400.

Childress,  S., 1966.  Solutions of Euler's equations containing
     finite eddies.  The Physics  of Fluids, Vol. 9, No. 5,
     pp. 860-872.

Clark,  C.  B., Stockhausen, P. J., and Kennedy, J. F., 1967.  A method
     for generating linear density profiles in laboratory tanks.
     Journal  of Geophysical  Research, Vol. 72, No. 4, pp. 1393-1395.

Debler, W.  R. , 1959.  Stratified  flow  into a line sink.  Journal Eng.
     Mech.  Div., Proc. ASCE, Vol. 85, pp: 673-695.

Fraenkel,  L.  E., 1961.  On corner eddies in plane inviscid shear
     flow.   Journal of Fluid Mechanics, Vol. 11, Pt. 3, pp. 400-406.

Gelhar, L.  W. , and Mascolo,  D. M., 1966.  Non-diffusive characteris-
     tics of slow viscous stratified flow towards a line sink.
     M.I.T. Hydrodynamics Laboratory Report No. 88, pp. 1-39.

Gill, A. E.,  and Smith, R. K., 1970.  On similarity solutions  of
     the differential equation ^zzzz + tyx = 0.  Proc. Camb. Phil.
     Soc.,  Vol. 67, pp. 163-171.

Green,  A.  E., and Naghdi, P. M.,  1969.  On basic equations for mix-
     tures.  Univ. of Calif. College of Eng. Report No. AM-69-4,
     pp. 1-19.

Imberger, J., 1970.  Selective withdrawal from a stratified reser-
     voir.   Ph.D. Thesis, Univ. of Calif., Berkeley, Dec. 1970.

Kao, T. W., 1965.  A free-streamline solution  for stratified  flow
     into a line sink.  Journal of Fluid Mechanics, Vol. 21,  Pt. 3,
     pp. 535-543.

Kao, T. W., 1970.  A free-streamline theory for  inviscid stratified
     flow.   Physics of Fluids, Vol. 13, No. 3, pp.  558-564.

Koh, R. C.  Y., 1964.  Karman integral  solution to the  problem of
     selective withdrawal.   California  Institute  of Tech., W.  M.
     Keck Laboratory Tech. Memo.  64-14, pp. 1-16.
                                79

-------
Koh,  R. C.  Y. ,  1966a.   Unsteady stratified flow into a line sink.
      Journal  of Hydraulic Research, Vol. 4, No. 2, pp. 21-34.

Koh,  R. C.  Y. ,  1966b.   Viscous stratified  flow towards a sink.
      Journal  of Fluid Mechanics, Vol. 24,  Ft. 3, pp. 555-575.

Krenkel,  P. A., Thackston, E. L. ,  and Parker, F. L. , 1969.
      Impoundment and temperature effect on waste assimilation,
      Journal  of the Sanitary Eng.  Div., Proc. A.S.C.E., Vol.  95,
      No.  SA1,  pp.  37-64.

List,  E.  J. ,  1968.   A two-dimensional sink in a density-stratified
      porous medium.  Journal of Fluid Mechanics, Vol. 33, Pt.  3,
      pp.  529-543.

List,  E.  J.,  1970.   Private communication.

Long,  R.  R. ,  1959.   The motion of fluids with density stratification.
      Journal  Geophys. Res., Vol. 64, pp. 2151-2163.

Lorence,  K. Z., 1952.  King Solomon's Ring.  Thomas Y. Crowell
      Company,  New York.

Miles, J. W. ,  1961.  On the stability of heterogeneous shear flows.
      Journal  of Fluid Mechanics, Vol. 10,  Pt. 4, pp. 496-508.

Mliller,  I. , 1968.  A thermodynamic theory  of mixtures of fluids
      Arch. Rational Mech. Anal., Vol. 28,  pp. 1-39.

Orlob, G. T. ,  and  Selna, L. G., 1970.   Temperature variations  in
      deep reservoirs.  Journal of the Hydraulics Div., Proc.  ASCE,
      Vol. 96,  HY2, pp. 391-410.

Redekopp, L.  G. , 1969.  Theoretical  analysis of boundary layer
      regions  in stratified flow.  Univ. of Calif. Los Angeles,
      School of Eng. and App. Sc. Report No. 69-23, pp. 1-156.

Redekopp, L.  G., 1970.  The development of horizontal boundary layers
      in stratified flow.  Journal Fluid Mechanics, Vol. 42, Pt.  3,
      pp.  513-527.

Schiff, L. I.,  1966.  Lateral boundary mixing  in  a  simple model  of
      ocean convection.  Deep Sea Research, Vol. 13,  pp. 621-626.

Segur, H.,  1970.   Private communication.

Stotta, L.  S. ,  Elwin, E. H. , Mercier, H. T. , and  Terry, M.D. ,  1969.
     Stratified Reservoir Currents.  Oregon State Univ. Eng.  Exp.
     Station,  Bull.  No.  44.
                                 80

-------
Smith, R. K., 1969.  On the effects of vorticity entrainment  in
     zonal jet flows.  Journal of Atmospheric Sciences, V:>1.  26,
     No. 6, pp. 1233-1237.

Spiegel, E. A. and Veronis, G., 1960.  On  the Boussinesq  approxi-
     mation for a compressible fluid.  Astrophysics! Journal,
     Vol. 131, pp. 442-447.

Tennessee Valley Authority, 1969.  The prediction  of withdrawal
     layer thickness in density stratified  reservoirs.  Water
     Resources Research Lab.  Report, No. 4, pp.  1-8.

Tennessee Valley Authority, 1969.  Cherokee Reservoir  selective
     withdrawal.  Water Resources Research  Lab,  Report  No.  13,
     pp. 1-30.

Trustrum, K., 1964.  Rotating and stratified  flow.  Journal of
     Fluid Mechanics, Vol.  19, Pt. 2, pp.  415-432.

Yin, C.  S., 1965.  Dynamics of Nonhomogeneous Fluids.   The  MacMillan
     Co., New York.
                                 81

-------
                        GLOSSARY OF  TERMS
D    - Molecular diffusion  coefficient  of  salt  or heat


                                                q

F,F  = Densimetric Froude number    — - -  -
                                        H
G    = Square  root  of  the  Schmidt  number I (v/D)2 1



g    = Acceleration due  to gravity



J    = Momentum  flux



K.   = Coefficients in the asymptotic  expansion  of  the withdrawal

       layer thickness



L    = A typical length  scale  of the motion



9j    = Half depth of the reservoir



P    = Scale of  the pressure perturbation



p    = Total pressure



p    = Hydrostatic  pressure



p    = Pressure due to the motion



q,q  = Discharge into the  sink  from the  right; measured q


                           \2    T        /,,r>\2





R    = Reynolds number (— )
 e                     \v/
       D  n  •  v.    K
R    = Rayleigh number I — 1  • — —  or
                        2    2 2N^
       Radius vector ( (x  + z )  j
T    = Brunt Vaisala period ( l/(£g)2 1



t    = Time
u
     = Velocity vector 1  (u , v, w) j



x    = Horizontal distance from the sink

                                                   (\

                                                 — ;
                                                 I I
 5k 3k                                              (X
x    = Scaled horizontal distance from the  sink  I— T-
                                                 \x*
                                                   c
                                83

-------
                                                  I  ^jjc \

 ***  = Scaled horizontal  distance  from the sink y- x   j
                                       F3/2\   f 3/2
x
x* x  = Length  of  the  non-uniformity  I—
 c  c                                 SK


z     = Vertical distance  from the  sink



z*    = Scaled  vertical  distance from the sink



                                               (  z* >
z**   = Scaled  vertical  distance from the sink I-  ,„ j
                                                *D*-/ *•* *
z***   =  Scaled vertical distance from the  sink f



a      =  (€g/vD)1/6
 o


6      =  Withdrawal layer half thickness

                                               fl
£      =  Normalized background density gradient (-

                                                  o

A      =  Scale of the density perturbation


p,      =  Molecular viscosity
 v     = Kinematic molecular viscosity I—j

                                         o


 p     = Total value of the density of the water
  T

 p     = Mean of the density of the water over the depth
  o


 p     = Background density
  e

 p     = Density perturbation due to the motion
                        / • \

 p*    = Scaled density (-T- i
 p**   = Scaled density (R    p*)


                        /Fl/2    \

 p***   = Scaled density  •—- p**
T      =  Surface stress


l|f      -  Stream function


l|f      =  Scaled stream function I—
                                 84

-------
                 APPENDIX I
EXPERIMENTAL WITHDRAWAL LAYER THICKNESS DATA
                      85

-------
CO
e a F x 6
„ o m c
Exp . No .
10"5 cm"1 103 cm"2 cm cm cm
1-1 2.98 0.458 1.145 688 1.0
1.3
1.5
1.7
2.0
2.1
2.3
2.7
2.8
2.9
3.3
3.5
1-2 2.98 0.458 1.46 1439 1.6
1.8
2.3
2.4
6/F 2
m
0.87
1.14
1.31
1.49
1.75
1.83
2.00
2.36
2.44
2.54
2.88
3.06
1.10
1.23
1.57
1.64
X
cm
1.0
3.4
6.0
8.3
13.6
15.2
20.8
29.0
37.4
43.4
52.0
60.8
2.6
7.0
10.8
12.6
<*/*/ 3
0.113
0.170
0.206
0.229
0.270
0.280
0.315
0.348
0.378
0.398
0.422
0.445
0.106
0.169
0.196
0.206

-------
3 *
e a F x 6
Exp. No. o m c
-5 -1 3 -2
10 cm 10 cm cm cm cm
2.4
2.6
2.9
3.0
3.3
3.3
00
3.5
1-3 2.98 0.458 1.70 2260 1.6
1.9
2.3
2.8
3.4
3.5
3.5
3.7
3.8
6/F *
m
1.64
1.78
1.99
2.06
2.26
2.26
2.40
0.94
1.12
1.35
1.65
2.00
2.06
2.06
2.18
2.24
x
cm
13.6
17.4
24.4
30.8
38.0
47.0
53.6
2.6
4.0
8.0
16.4
22.6
30.2
39.4
47.0
57.1
(x/xc)1 3
0.215
0.230
0.255
0.278
0.298
0.319
0.333
0.105
0.121
0.152
0.194
0.215
0.237
0.259
0.275
0.294

-------
e a F 2 x 6
o m c
Exp. No. ' * , ^ 2
10~° cm 10 cm cm cm cm
2-1 25.7 1.46 0.16 5.4 1.10
1.2
1.25
1.47
1.50
1.45
oo
00 2-2 25.7 1.46 0.35 64.3 1.02
1.20
1.28
1.34
1.58
1.80
2-3 25.7 1.46 0.74 584.0 1.53
1.64
1.77
1.93
6/F 2
m
7.10
7.75
8.07
9.49
9.68
9.37
2.90
3.41
3.64
3.80
4.48
5.11
2.08
2.22
2.40
2.69
X
cm
16.8
22.0
26.1
32.9
37.6
43.5
4.6
16.8
21.4
24.0
44.6
47.3
4.6
9.4
13.0
17.6
/ / x1/3
(x/xc)
1.46
1.59
1.69
1.82
1.91
2.00
0.424
0.644
0.693
0.720
0.886
0.902
0.200
0.252
0.280
0.310

-------
oo
3 \
e a F x 6
,-, o m c
Exp . No .
— 5—1 *•? —2
10 cm 10 cm cm cm cm
2.07
2.17
2.20
2.21
2.46
2.46
2.70
2.80
3-1 29.9 1.45 0.11 2.18 0.83
0.80
0.90
1.05
1.20
1.35
1.40
1.50
6/F 2
m
2.82
2.94
2.98
3.00
3.34
3.34
3.66
3.79
7.30
7.01
7.92
9.22
10.56
11.80
12.21
13.11
X
cm
20.4
25.2
27.0
30.2
37.0
39.6
44.4
46.0
6.8
14.2
19.4
28.4
33.8
44.4
52.2
111.4
ii \l/3
(x/xc)
0.327
0.350
0.353
0.373
0.398
0.408
0.424
0.429
1.46
1.87
2.07
2.36
2.49
2.73
2.88
3.74

-------
e a F 2 x 6
o m c
Exp . No .
— ^ — 1 ^ — 2
10 cm" 10 cm cm cm cm
1.50
1.60
1.70
1.75
1.79
1.82
>o
o
1.80
3-2 29.9 1.46 0.16 6.43 0.65
0.77
0.92
1.12
1.20
1.30
1.38
1.40
1.45
6/F 2
m
13.11
14.00
14.92
15.31
15.64
15.89
15.80
3.99
4.72
5.64
6.87
7.36
7.98
8.46
8.60
8.87
x
cm
118.4
125.4
132.4
141.0
148.2
156.0
163.0
2.8
8.6
14.6
21.4
28.4
33.0
43.4
47.0
53.2
(x/xc)1/3
3.78
3.86
3.93
4.00
4.09
4.15
4.21
0.76
1.10
1.31
1.49
1.64
1.72
1.89
1.94
2.02

-------
3 \
e a F 2 x 6
Exp. No. o m c
— 5 —1 3 —2
10 cm 10 cm cm cm cm
1.50
1.75
1.86
1.82
1.82
1.85
1.85
1.85
1.80
2.00
2.00
3-3 29.9 1.46 0.25 22.8 1.00
1.15
1.20
1.35
1.40
6/F 2
m
9.20
10.71
11.41
11.12
11.12
11.31
11.31
11.31
11.00
12.22
12.22
4.00
4.62
4.80
5.44
5.62
x
cm
60.6
108.2
115.1
119.4
125.3
130.0
136.2
142.4
148.0
156.0
160.0
7.0
14.6
19.8
25.6
29.0
(x/x//3
2.11
2.56
2.62
2.64
2.69
2.72
2.76
2.79
2.84
2.89
2.92
0.67
0.86
0.95
1.04
1.08

-------
3 2 c
e a F x 6
o m c
Exp. No. ,00
«^~O "•*• - ,-,«5 — <3
10 cm 10 cm cm cm cm
1.50
1.50
1.61
3-4 29.9 1.50 0.37 75.0 0.98
1.20
1.46
co
NJ
1.62
1.70
1.92
1.90
2.12
2.10
2.30
2.35
2.38
2.40
6/F 2
m
6.00
6.00
6.42
2.66
3.25
3.95
4.39
4.61
5.20
5.15
5.75
5.69
6.24
6.36
6.45
6.50
X
cm
38.0
41.8
50.8
2.4
8.8
15.8
21.6
29.1
38.4
42.4
52.4
59.6
106.9
116.0
119.0
124.6
(x/xc)1/3
1.19
1.23
1.31
0.32
0.49
0.60
0.66
0.73
0.80
0.83
0.89
0.93
1.12
1.16
1.17
1.18

-------
•D 1.
O o c
e a F x o
TH o m c
Exp. No. _5 _1 _2
10 cm 10 cm cm cm cm
2.45
2.40
2.40
2.55
2.65
2.65
co
W 3-5 29.9 1.50 0.51 204.0 0.65
1.00
1.20
1.35
1.40
1.70
1.80
2.00
2.20
2.25
6/F 2
m
6.64
6.50
6.50
6.90
7.16
7.16
1.27
1.95
2.34
2.64
2.73
3.32
3.51
3.90
4.29
4.39
X
cm
130.4
132.0
141.6
149.0
155.0
165.0
0.6
3.8
7.0
9.0
13.2
21.8
31.8
36.4
45.8
52.0
t / \1/3
(x/xc)
1.20
1.20
1.24
1.26
1.27
1.30
0.14
0.27
0.32
0.35
0.40
0.47
0.54
0.56
0.61
0.63

-------
ID
e a F 2 x 6
o m c
Exp. No. 3-2
10 cm" 10 cm cm cm cm
2.25
2.45
2.45
2.55
2.70
2.75
2.80
2.90
3.20
3.10
3-6 29.9 1.50 0.65 420.0 1.00
1.00
1.20
1.30
1.35
1.40
6/F 2
m
4.39
4.78
4.78
4.97
5.26
5.36
5.47
5.65
6.24
6.04
1.53
1.53
1.84
1.99
2.06
2.14
X
cm
61.0
111.6
117.4
124.0
130.4
133.6
140.4
147.8
156.2
160.5
3.2
4.8
10.0
12.6
16.4
20.6
(*/*/ 3
0.67
0.82
0.83
0.85
0.86
0.87
0.82
0.90
0.91
0.92
0.20
0.23
0.29
0.31
0.34
0.37

-------
tn
3 I
e a F x 6
,-, .. o m c
Exp. No. _<1 3 _2
10 cm 10 cm cm cm cm
1.65
1.65
1.75
1.80
2.00
2.70
2.70
2.80
2.90
3.00
3.00
3.20
4-1 35.3 1.57 0.11 2.14 0.55
0.68
0.75
0.90
5/F 2
m
2.52
2.52
2.68
2.75
3.06
4.13
4.13
4.29
4.44
4.59
4.59
4.89
4.96
6.15
6.78
8.13
x
cm
28.6
33.6
40.4
48.4
60.0
109.4
115.0
123.2
132.8
141.4
148.3
158.0
2.3
4.9
9.0
17.0
<*/v1/3
0.41
0.43
0.46
0.49
0.52
0.64
0.65
0.66
0.68
0.70
0.71
0.72
1.04
1.32
1.61
1.99

-------
e a F x 6
Exp . No . °
10~5 cm"1 103 cm"2 cm cm cm
1.00
1.15
1.22
1.30
1.52
1.61
CO
1.68
1.65
1.70
1.72
4-2 35.3 1.57 0.11 2.22 0.40
0.65
0.78
0.84
0.93
1.05
6/F *
m
9.03
10.41
11.00
11.72
13.72
14.53
15.18
14.90
15.35
15.50
3.57
5.80
6.96
7.50
8.30
9.40
X
cm
24.8
32.8
44.2
52.8
116.0
124.8
135.1
143.6
153.6
163.4
0.4
4.4
6.1
9.0
14.4
23.6
(x/xc)
2.26
2.48
2.74
2.91
3.79
3.90
3.98
4.06
4.15
4.24
0.56
1.25
1.40
1.59
1.86
2.20

-------
(0
3 -
e a F 2 x 6
Exp. No. o m c
10~5 cm"1 103 cm"2 cm cm cm
1.14
1.19
1.25
1.55
1.60
1.64
1.60
1.65
1.66
1.68
1.67
4-3 35.3 1.60 0.13 3.81 0.75
0.85
0.85
0.90
1.00
6/F 2
m
10.20
10.61
11.22
13.83
14.31
14.61
14.30
14.70
14.82
15.00
14.93
5.63
6.36
6.36
6.75
7.50
x
cm
31.4
37.0
48.2
112.4
119.0
124.8
131.3
137.6
143.7
154.0
165.0
4.3
5.3
6.9
8.2
17.1
, i \ J. / »J
(x/x )
c
2.42
2.55
2.79
3.70
3.77
3.83
3.90
3.96
4.01
4.11
4.20
1.04
1.12
1.22
1.29
1.64

-------
<£>
oo
e a F * x 6
Exp. No. ° _2
10 cm 10 cm cm cm cm
1.10
1.15
1.20
1.30
1.55
1.55
1.60
1.65
1.65
1.62
1.70
1.65
1.80
1.70
1.85
4-4 35.3 1.60 0.14 4.38 0.60
6/F *
m
8.25
8.60
9.00
9.75
11.60
11.60
12.00
12.41
12.41
12.20
12.75
12.41
13.55
12.75
13.81
4.20
X
cm
23.8
31.1
39.2
48.0
110.7
115.4
119.5
126.2
130.5
137.3
143.7
151.2
154.8
161.2
166.9
2.6
(X/XC}1 3
1.84
2.01
2.17
2.32
3.07
3.12
3.15
3.21
3.24
3.30
3.35
3.41
3.44
3.48
3.52
0.84

-------

-------
e c£ Fm* xc 6
Exp. No. _ _ 3 ° _2
10 cm 10 cm cm cm cm
4-5 35.3 1.60 0.15 5.85 0.68
0.70
0.80
0.85
0.95
1.05
r-1
o
0 1.10
1.15
1.20
1.25
1.35
1.60
1.60
1.65
1.75
1.75
6/Fn,2
4.60
4.70
5.30
5.70
6.35
7.00
7.38
7.70
8.04
8.36
9.00
10.71
10.71
11.00
11.72
11.72
X
cm
1.8
2.8
3.4
7.0
10.8
17.2
21.6
32.3
39.4
47.1
51.4
108.1
114.0
128.9
133.0
145.0
(x/xc)
0.68
0.78
0.83
1.06
1.22
1.43
1.54
1.77
1.89
2.00
2.05
2.64
2.69
2.80
2.84
2. 92

-------
3 *
e a F 2 x 6
EXP' N°- -5 -1 3 ° -2
10 cm 10 cm cm cm cm
1.65
1.70
1.80
4-6 35.3 1.57 0.18 9.90 0.60
0.78
0.92
1.00
1.10
1.22
1.35
1.40
1.42
1.88
1.88
2.00
1.95
6/F 2
m
11.00
11.40
12.00
3.30
4.20
5.00
5.40
6.00
6.60
7.30
7.60
7.70
10.21
10.21
10.81
10.62
x
cm
148.9
155.1
159.1
1.6
4.8
8.4
14.1
20.8
29.8
41.6
49.6
57.4
111.1
116.2
121.4
127.2
(x/x )
c
2.94
2.98
3.00
0.55
0.78
0.95
1.13
1.28
1.44
1.61
1.71
1.80
2.24
2.27
2.31
2.34

-------
e ot F x 6
EXP' N°- -5 -1 3 ° -2
10 cm 10 cm cm cm cm
1.98
2.00
2.02
2.05
2.05
M 4-7 35.3 1.57 0.20 13.40 0.60
o
to
0.78
0.87
0.90
1.25
1.20
1.30
1.38
1.42
1.40
1.50
6/F *
m
10.71
10.83
10.90
11.10
11.10
2.94
3.83
4.27
4.41
6.14
5.89
6.38
6.77
6.97
6.90
7.35
X
cm
136.0
142.3
148.6
161.8
164.8
1.8
4.4
7.0
9.4
15.6
20.3
26.8
33.4
40.9
46.0
51.6
(x/x )1/3
c
2.39
2.43
2.46
2.54
2.56
0.51
0.69
0.81
0.89
1.05
1.15
1.26
1.35
1.45
1.51
1.57

-------
e a F xc 6
Exp. No. 3 ° -2
10 cm 10 cm cm cm cm
1.50
4-8 35.3 1.57 0.27 32.20 0.70
1.10
1.15
1.20
1.30
o
1.38
1.40
1.45
1.70
1.85
1.85
1.90
1.95
1.95
2.00
6/F *
m
7.35
2.56
4.02
4.20
4.38
4.75
5.04
5.10
5.28
6.20
6.75
6.75
6.84
7.12
7.12
7.30
X
cm
57.0
2.3
8.8
14.3
16.6
25.0
31.0
36.3
45.4
58.3
106.8
112.7
119.5
127.4
130.9
136.7
(x/xc)1/3
1.62
0.41
0.65
0.76
0.80
0.92
0.99
1.04
1.12
1.22
1.50
1.52
1.55
1.58
1.59
1.62

-------
e a F x 6
_ o m c
Exp. No. _2
10 cm" 10 cm~ cm cm cm
2.05
2.05
2.05
2.35
6/F *
m
7.50
7.50
7.50
8.57
X
cm
144.8
150.7
157.3
166.8
<*/v1/3
1.65
1.67
1.70
1.73

-------