VOL. 19, NO.

11 AIAA JOURNAL NOVEMBER 1981

History of Key Technologies AIAA 81-4291

The Boundary Layer and I

A. M. O. Smith
San Marino, Calif.

Nomenclature v = kinematic viscosity

amplitude of Tollmien-Schlichting waves or p = mass density
Gortler vortices; also Van Driest's damping- A = Pohlhausen'sA=(6 2 /*>)(dw <? /ch:)
length constant
A0 amplitude at neutral stability
c constant in velocity distributions of the form I. Early Motivation
ue = cxm Y story begins in June 1945, in a basement at the
profile drag coefficient
transformed value of du/dy at the wall in the
M Gottingen Aerodynamische Versuchsanstalt (AVA),
Germany, at the end of World War II. But first let us back up
Falkner-Skan equation slightly. In 1937 and 1938, I took a course, called AE 267 -
k height of a roughness Real Fluids, at Caltech from C. B. Millikan. This course dealt
m exponent in velocities of the form ue = cxm with laminar and ordinary viscous flows. AE 269 dealt with
n measure of amplification ratio, n-^i(A/A0) turbulent flows. AE 267 was rather similar to the laminar and
Pr Prandtl number viscous portions of Goldstein's Modern Developments in
r longitudinal radius of curvature for Gortler Fluid Dynamics and PrandtPs contribution in Durand's
vortices Aerodynamic Theory, Vol. III. It was a very good course and
Rc chord Reynolds number Millikan was probably the clearest and best organized teacher
Rd Reynolds number based on momentum thickness I ever had. I know that he always conscientiously reviewed his
u velocity in x direction next day's lectures the day before. However, I was mainly
ue edge velocity interested in becoming a preliminary designer, and at that
x,y coordinates, generally along and perpendicular to time, boundary layers and viscous effects seemed quite long-
the wall haired and abstruse.
Hartree's boundary-layer profile parameter, note Part of the course dealt with the Orr-Sommerfeld equation
governing the stability of laminar boundary layers and at-
y intermittency factor tempts at the solution of the equation. This was long before
6 thickness of a boundary layer Sehubauer and Skramstad verified its applicability to the
6* displacement thickness process of boundary-layer transition, so the analysis was a
e turbulent eddy viscosity; e, , inner, e0 , outer kind of academic exercise. At one time in some general
6 momentum thickness of a boundary layer discussion, Millikan mentioned that suction should change
von Kdrmdn's constant the shape of the boundary-layer profile and probably make it

A. M. O. Smith retired in 1975 as Chief Aerodynamics Engineer for Research at Douglas Aircraft
Company, Calif. He was born in Columbia, Mo., on July 2, 1911. He majored in Mechanical and
Aeronautical Engineering at the California Institute of Technology and received an M.S. degree in
both fields. After graduation in 1938, he joined Douglas Aircraft as Assistant Chief Aerodynamicist.
During this period, he worked on aerodynamic and preliminary design problems of the DC-5 trans-
port; SBD dive bomber, and the A-20, DB-7, and B-26 attack bombers. He had prime responsibility
for detailed aerodynamic design of the B-26. Because of earlier work with rockets at Caltech, he was
asked by Gen. H. H. Arnold to organize and head the Engineering Department of Aerojet as their
first Chief Engineer, on leave of absence from Douglas in 1942 to 1944. After expanding the
department from 6 to more than 400 personnel and seeing the company into production on JATO
units, he returned to Douglas and aerodynamics. There he handled aerodynamics of the D-558-1
Skystreak and the F4D-1 Skyray, both of which held world speed records. In 1948 he moved into the
research aspect of aerodynamics. Since then he has developed powerful methods of calculating
potential and boundary-layer flows; culminating in a book co-authored with T. Cebeci entitled,
Analysis of Turbulent Boundary Layers. He has published over 60 papers. For his work at Aerojet, he
received the Robert H. Goddard Award of the American Rocket Society. For his early rocket work at
Caltech, he is commemorated in bronze at the NASA Jet Propulsion Laboratory. In 1970, he
received—jointly—the F. W. (Casey) Baldwin Award of the Canadian Aeronautical Sciences In-
stitute. In 1974 he was awarded the Wright Brothers Lectureship of the AIAA and in 1975 he received
an honorary D. Sc. degree from the University of Colorado.

Received March 13, 1981; revision received June 1, 1981. Copyright © American Institute of Aeronautics and Astronautics, Inc., 1981. All rights
reserved.
EDITOR'S NOTE: This manuscript was invited as a History of Key Technologies paper as part of AIAA's 50th Anniversary celebration. It is not
meant to be a comprehensive survey of the field. It represents solely the author's own recollection of events at the time and is based upon his own
experiences.
1377
1378 A.M.O. SMITH AIAA JOURNAL

SUCTION
SLOTS.

Fig. 2 The DESA-2 airfoil and its pressure distribution, 6.6% thick,
Fig. 1 A typical thick symmetric Griffith airfoil, a = 0 deg. design CL =0.1.

more stable, but nothing was known about how much suction briefing on ways to extend the range of aircraft. One way was
it would take or whether there could be net gains in efficiency, by aerial refueling. This seemed very wild to me and if people
such as net gains in total required horsepower. After leaving were that desperate to extend the range, then laminar flow
Caltech, I did not remember much about this theory but did control should surely be looked into as an alternate. After
remember his conjecture regarding stabilization by suction that briefing I began active work in this area, mainly around
and the question about how much suction might be required. my idea. Then in attempting to design practical and successful
My work at Douglas Aircraft from 1938 to 1945 was on Griffith type airfoils I was forced into experimental and
general airplane design and ordinary aerodynamics, and I theoretical boundary-layer work rather extensively.
never thought about the subject of laminar-flow control,
although my Master's thesis had been on high-lift boundary-
layer control. However, in 1945 at the end of the war, I was
appointed a member of the U.S. Naval Technical Mission in II. DESA-2 Airfoil and Theoretical Developments
Europe (NAVTECMISEU) to help investigate the German This airfoil was the type I conceived at Goldstein's lecture.
aeronautical developments. One part of my work involved It was not successful, but it was a very basic stimulus to much
exploring the Gottingen AVA and Kaiser Wilhelm Institute of my work in the field, so I feel it deserves separate mention.
(KWI) complexes. I got into their reprint room in a basement We first designed and tested a preliminary model, the G00107
and browsed through many interesting new and old and airfoil. It was symmetrical and had a flap. We obtained 98%
famous classical papers. One that I came across was by laminar flow to the capacity of the wind tunnel,
Schlichting and Bussmann. 1 This paper solved the Falkner- /?c = 4.25x!0 6 . At that time this was the highest Reynolds
Skan equation for suction and blowing type of boundary number for full laminar flow that yet had been reached. The
conditions on the wall. The authors found that the stability tests were a complete success. Therefore, we began designing'
would be changed tremendously by very small amounts of a thin, cambered and more realistic shape for tests up to
suction. Thus, it answered the question posed by Millikan /?c = 40 or 50xl0 6 in the NACA two-dimensional pressure
seven years before and caused me to pay serious attention to tunnel. It was 6.6% thick. The shape and pressure
laminar-flow control as a possibility for drag reduction. distribution are shown in Fig. 2. Just behind each pressure
However, I did not immediately try to do any work in this jump is a short concave region. At the scale of the figure, the
area and nothing happened, except that I followed Pfen- waviness and concavities required to generate the sawtooth
ninger's Swiss work. Then it was my good luck to be in pressure distribution do not show. Moreover, the pressure
Washington in December 1947 at the time Sidney Goldstein jumps are so low that, in case of suction failure, the shape
gave his Wright Brothers lecture about the Griffith airfoils should act like an ordinary airfoil. In wind tunnel tests, both
developed in England. A typical Griffith airfoil is illustrated the G00107 and the DESA-2 did so behave.
in Fig. 1. It has long, continuously favorable pressure The model has two unique features: 1) very strong
gradients on each side. In order to close the airfoil, there must stabilization due to the very favorable pressure gradients and
be a pressure jump toward the rear, and after that, the flow 2) several short regions of substantial concave curvature
again accelerates. At the pressure jump, suction is applied to behind the slots. The concave curvature was substantial, but
prevent separation. One of the unique features is that, behind not so great as to invalidate boundary-layer theory. By this, I
the slots, the shape must be concave as seen in the illustration mean that, in the basic equations, b/r is still small. Here, 5 is a
in order to develop the higher pressures. While listening to the measure of the boundary-layer thickness, and r the
lecture, it occurred to me that an airfoil could be designed longitudinal concave radius of curvature. Moreover, the
with several small pressure jumps instead of one large one on regions were very short. In thinking about this situation as
each side—perhaps 10 per side. My design would lead to an well as the favorable gradient regions, it did not seem
airfoil that had a more conventional shape, and in case reasonable to me that transition would be determined by an
suction failed, its performance should not deteriorate instantaneous value of some parameter. To change the flow,
catastrophically. I wanted to find something better than some process had to act for a finite period, because physical
Pfenninger's approach of using a large number of slots (40 processes are never truly impulsive, and there is nothing to
per side or more) and its resultant mechanical complications. give an impulse. It seemed to me that any analysis of the
But as already noted, when I learned of the results of Ref. process should involve an integration, not just a limiting
1, I did not immediately try to start work on laminar flow value, unlike an allowable stress in the science of strength of
airfoils. In 1948, I believe, I attended a Rand Corporation materials.
NOVEMBER 1981 THE BOUNDARY LAYER AND I 1379

The concave regions would develop Gortler vortices, which started solving it on a desk calculator as a check. The process
if too strong, would cause transition. Gortler's work dealt required inversion of the matrix and repeated multiplication
only with neutral stability; however, in our problem, there by a column to get the eigenvalue. Also, the matrix was nearly
would be strong instability, but only for a very short time. singular, i.e., very difficult. The man who attacked the
Hence, we wanted to know rates of growth, and Gortler's problem with a desk calculator was D. A. Callin. The IBM
paper gave no indication of rates. Thus, this peculiarity of the operators ran into trouble with accuracy. In the meantime,
DESA-2 directly involved me deeply in boundary-layer theory Callin plugged along and got our first answer. So we started
and made me feel that an amplification ratio was more on the next case, and the IBM operators still had trouble, but
significant than the maximum value of some parameter, e.g., Callin got the second answer, and so it continued until we
Gortler's Re(6/r) 1/2. Here 0 is momentum thickness, and Re finally calculated 26 different cases, 25 by Callin. The IBM
is the usual momentum thickness Reynolds number. Fur- machines finally gave us two answers, which only checked the
thermore, it should be noted that the stability equations yield hand calculations already completed. Each case required just
rate parameters, not strength parameters. Strength of a over 100 h of continuous hand calculation, often involving
disturbance is some kind of integral of rate times distance, or double-precision arithmetic. You can see that the calculations
rate times time. took over a year of continuous work by one man. Callin could
During the design of the DESA-2 airfoil, I often stated that do all this effortlessly, often working two days without a
I was not sure the airfoil would work, but at least I was mistake. If we had not been so fortunate as to have his help, I
determined not to have to apologize for a poor job of doubt if we would have been able to finish the problem,
designing it. Therefore, I decided to try to calculate the certainly not by the time we needed results. Thus, whenever I
growth rates of Gortler vortices. For me this was a substantial think of him I think of him as the virtuoso of the Marchant.
effort, because, up to that time at Douglas, I had hardly used The calculations and work were successful, and by the
any mathematics beyond ordinary algebra. I studied various spring of 1952, we had most of the answers to the Gortler
methods of solving the eigenvalue problem, including the stability question. At the same time we applied our results and
basic mathematics. I remember studying Pauling and found that an amplification ratio correlated data much better
Wilson's Introduction to Quantum Mechanics and Margenau than Gortler's parameter Re(0/r) l/2 at transition. In fact we
and Murphy's book, The Mathematics of Physics and found that Gortler type transition occurred about when
Chemistry. Also, you might say Ince's Ordinary Differential &u4A40 = 10, A/A0 being the amplification ratio. This work
Equations was my bible. In the boundary-layer field, I studied was nearly complete in early 1952, but was all classified. The
both Meksyn's work, G. I. Taylor's, and Gortler's own. In internal Douglas report, ES 17110, was published in March
fact, we solved several cases by Gortler's own method for 1953. We later studied and correlated further data which
neutral stability, but it did not seem to have the power needed ended up as my paper on Gortler stability, including the e"
for solving the problem we wished to solve, which involved method of predicting this kind of transition. As noted,
amplification rates rather than just neutral stability. As will transition was found to occur when %*A/A0 was about 10.
be seen, we did find what we thought was a more powerful Thus, you see that the DESA-2 airfoil had much to do with
method. development of the amplification ratio method, first
However, one problem bothered me. I could not tie time stimulated by the Gortler problem. For the very short highly
and distance together, as can be done with Tollmien- concave regions, it did not stand to reason that a high value of
Schlichting waves, and of course, Gortler's and Meksyn's Gortler's parameter, acting for an exceedingly short time,
analyses were for wave growth in time. I pondered this should cause transition. Rate and total growth of the
problem for some months and one weekend, while laying disturbance somehow must be involved. For the DESA-2
flagstones in my backyard, saw the solution. Instead of using airfoil in the worst concave region, Gortler vortices grew
a disturbance expression that grew as exp^, the problem across it by a ratio V*iA/A0 of 5 at Rc - 60 x 106 so we thought
should be reformulated into one that grew as expx we were safe.
At about the same time, late 1949 I believe, I talked with Naturally we wondered just how rapidly the convex regions
Douglas Dill, head of our Douglas El Segundo Flutter Group. would damp out the Gortler vortices. We tried stability
I was looking for help on methods of calculation. He calculations several times for the convex regions, but never
suggested that I look into Galerkin's method. I did and found could get convergence to an eigenvalue.
it so powerful that it could nicely handle the kind of problem The other considerable effort was on the question of
we wished to solve. This method involves solution of a matrix. Tollmien Schlichting stability. It seemed to me that am-
Mechanical automatic calculators, which were good at plification rate and ratio for this kind of problem also gave
matrices, were just arriving, and this was more incentive for the most rational guide to transition, but at the time, the en
using Galerkin's method. method had not been fully developed. In fact, the thinking
Accordingly, we worked hard on deriving the extended was something of a carry over from the Gortler problem. The
equations and studying their structure, order of magnitude of prediction of transition was in a primitive state. As I
various terms, methods for calculation of eigenvalues, etc. remember, the best way of predicting transition was to use
Frazier, Duncan, and Collar's book, Elementary Matrices, some value of Re as the criterion. We needed our guidance
was my source book for matrix methods and theory. I was around 1950 or 1951, and Re was the only guide. Significant
forced to gain a great deal of knowledge or I would never have papers on the subject such as Gazley's and Granville's did not
solved our problem. The unusual nature of the DESA-2 and come out until 1953. Michel's paper appeared in 1951, but we
the high-performance stakes were the stimulus. At the design did not know about it until after the DESA-2 was tested.
Rc = 5Qx\Q6, the predicted effective drag coefficient was Re at transition was not then known as a function of the
CD =0.00064. For comparison, wind tunnel tests show the pressure gradient or Pohlhausen's A. From NACA airfoil
drag of the NACA 66,1-015 airfoil to be CD =0.0050 at this tests, we of course knew there was, indeed, an effect of
same Reynolds number. Therefore, as you see, the DESA-2 pressure gradient, and we wanted a method of analysis that at
airfoil and its accompanying stability problem caused me to least took into account some of the changes in stability caused
pay much more attention to boundary layer and applied by it. Pretsch's charts were a great help on this question and
mathematics literature. on the airfoil design. I do not remember all our thinking at the
As an aside, it should be noted that the calculations we time, but in the report, the airfoil is very nearly designed as if
ended up doing can only be called "heroic." While we studied the e9 criterion was applicable. That is, at Rc = 50x 106, the
accuracy with lesser matrices, for our final results, we set up a calculated amplification ratio for the entire airfoil is e9-5 on
12 x 12 matrix system, expecting the IBM machine to give us the upper surface and e7-2 on the lower surface. In fact, in the
the answers. When we gave the operator the first case, we report,2 the design conditions are stated to be as follows.
1380 A.M.O. SMITH AIAA JOURNAL

1) Airfoil thickness to be 6.6% in order to be usable on the difference methods for the time being. The computer was the
F4D-1 airplane, if desired. IBM Card Programmed Calculator—a semimechanical
2) Design lift coefficient =0.1. computer.
3) Cm 25 to be zero at CL = 0.1. In tests on the DESA-2, we could get 100% laminar flow
4) At Rc = 50x 106 for Schlichting waves the amplification only to a Reynolds number of about 6x 106, so the attempt
ratio at the trailing edge must be less than exp 10. and my airfoil idea were unsuccessful. I never understood why
5) At /?c = 50x!0 6 the amplification ratio for Gortler until 1978. Not knowing any better, I had assumed that the
vortices must be less than exp 10 and preferably as low as much longer convex regions between the concave regions
possible (for each concavity). would damp out the Gortler vortices so much that, at each
6) From the practical standpoint, the design should have as new concavity, the flow would essentially start out afresh;
few slots as possible. Tollmien-Schlichting waves behave this way. When they
Thus, it is seen that the stimulus for the e9 method lay in the encounter a sufficiently favorable pressure gradient, they
DESA-2 airfoil. By 1952, a large number of amplification damp out. Besides, since the airfoil has significant thickness,
ratios at transition had been calculated from the various test there was more convexity than concavity. In 1978, Professor
data available, and while I believed they formed the most Eli Reshotko pointed out to me the principle of exchange of
logical method of prediction, there was a large amount of stabilities. According to this principle, when the flow en-
scatter in the calculated values, so no rational method of counters a concavity, Gortler vortices amplify; but when they
prediction was evident. That was not so true for the Gortler reach the convex portion, they do not damp, they remain
problem. In it, for the available test data, the calculated neutral in stability. Mathematically speaking, it works out
amplification ratios at transition were fairly uniform, at a that, for concave flows, the exponential giving the eigenvalue
value &i/4M0 = 10. Once, probably in 1953, I showed our has both a real and imaginary part, the real part being a
studies of test data to a co-worker, Deane Morris, who later measure of the amplification rate. However, for convex
moved to the Rand Corporation. He saw the large scatter in flows, it is a pure imaginary, meaning an undamped fluc-
calculated amplification ratios at transition and suggested tuation. Hence, the behavior of Gortler vortices is con-
that, instead of worrying about it, we just take some average siderably different from that of Tollmien-Schlichting waves.
value and see how well we could actually predict the transition In my original concept of the entire amplification process, I
point. We did and found the point was predicted well and visualized it as a series of climbs and descents; but according
rationally. Thus was born the en method, at least my con- to the principle of exchange of stabilities, the process would
tribution. After further studies and rounding out the work, I be something like a staircase: a climb, a level region, a climb,
presented my paper about it in 1956. For test data at tran- a level region, etc. So the total amplification ratio for the
sition, the calculated amplification ratio is often a sensitive airfoil would be the product of the amplification ratios for
parameter. Conversely, the position of transition is relatively each individual concavity. Also, this neutral stability for
insensitive to the value of the amplification ratio. convex surfaces probably was the cause of our inability to
It should be noted that this development was not started converge to eigenvalues in our Gortler calculations when we
with the expectation of finding that an amplification factor tried such surf aces.
alone would be a suitable criterion, because it is only Therefore, the airfoil was not a practical success, but it was
something that multiplies an initial disturbance. Transition an indirect success because, with its peculiar sawtooth
must depend on disturbance strength, not just a ratio. pressure distribution, it was a great stimulus to development
However, calculating the amplification ratio was a well- of boundary-layer methods.
defined problem, and so it was done first. We then expected to Another study was of the size Of roughness that could trip a
study turbulence inputs at the neutral point. Some work was laminar boundary layer, done for the U. S. Navy. While our
done, such as multiplying percent turbulence times am- work on developing a laminar-flow airfoil had stopped, we
plification ratio to get a disturbance strength to see if we could still had our knowhow, and in 1954, the U. S. Navy asked us
predict the effect of turbulence on flat-plate transition. This to look into the practical problems of laminar boundary-layer
proved no good; the problem is far more complicated. control. It seemed to me that one of the most severe practical
However, two things happened, the DESA-2 contract ended, problems was that of roughness tripping the boundary layer.
and thanks to Deane Morris's suggestion, we found that We made a large wind tunnel model and very carefully and
A/A0 by itself was a good parameter. That is about where the extensively looked into this problem. Our interest was the
prediction of low speed transition stands to this very day. height of roughness that first affects a long laminar run.
A lesser development, but also stimulated by the peculiar There had been many tests by others, e.g., Dryden, Tani, and
pressure distribution of this airfoil, was my piecewise method Potter, on the roughness required to bring transition to the
of approximate laminar boundary-layer calculation.3 The roughness but not on the height of roughness that first affects
sawtooth pressure distribution gave me the idea that ficticious transition. About the time we had obtained all our data and
origins could be used for similar flows of the type ue = cxm, were studying various ways of correlating it, I talked to
from which approximate solutions quickly could be patched Professor Hans Liepmann. He commented that he and Frank
together. We especially wanted to use similar solutions Goddard were having very good luck on the problem of
because Pretsch's stability charts were for various values of turbulent skin friction of rough surfaces at high Mach
Hartree's /?. As I remember it, this idea came up in some numbers by correlating test data with really local conditions at
discussion one afternoon, and by the end of the day, we had the bottom of the boundary layer. We followed suit and
made our first general calculation. A preliminary form is in obtained very successful correlation by using the true
the DESA-2 report.2 We gradually assessed the method, Reynolds number of the roughness, ukk/vk where k is the
refined it, and published it in 1956. height of the roughness, uk the velocity in the boundary layer
Another event in connection with the DESA-2 airfoil was at the top of the roughness and vk the kinematic viscosity at
an attempt at finite-difference boundary-layer calculation. To the same point. Both on this problem and many others, a
answer stability questions, we wanted to know very accurately great many results were all explained by correlating with the
the boundary-layer profiles between slots, and integral true local flow at the roughness. Hence our advance was in the
methods are inadequate. I studied various methods of writing correlation method, and it was due a good deal to the con-
a finite-difference system and selected one that seemed the versation with Hans Liepmann.
best. I then gave it to the head of computing, W. C. Schlieser, The Navy did nothing more than ask us to look into the
and asked for an estimate of the machine time required to practical problems of laminar-flow control. Thus, the idea of
calculate 20 or so stations. He came back with an estimate of tests of roughness were entirely our own, in fact, my own as I
2000 machine hours! That stopped my interest in finite- remember. Also, the idea of the DESA-2 project was entirely
NOVEMBER 1981 THE BOUNDARY LAYER AND I 1381

my own. There was no spirit of competition on this problem around, watching the tab sheet being printed, making plots to
or trying to out-do others, except very indirectly as an im- estimate the new value of Hartree's F'^, etc. I hope this story
provement over Griffith airfoils. On both the transition and gives you an idea of what automatic computing was like in its
piecewise method of boundary-layer calculation, one might early days. The tables were published by the Institute of
say we were working in a vacuum. The procedures just seemed Aeronautical Sciences as a Sherman Fairchild Fund paper in
quite logical, so we explored and implemented them. 1954 but all the work was done in 1951.
One other aspect should be mentioned. When the program
III. Some Comments on the Practical Side was transferred to El Segundo from Santa Monica, I naturally
of Getting Work Done duplicated some runs. The printout was eight-decimal places,
The two wind tunnel models were rather expensive, and a I believe. For a number of steps the new and old tab sheets
great deal of theoretical work was involved that took man- would check exactly. But then after a while there would be a
hours. Experienced readers may wonder how I managed to get gradual drift; first the eighth place would not check, then the
the work authorized. I do not really know because I did not seventh, and finally down to the sixth or fifth, so that I could
have any problems like those in later years. For exploratory not be certain I met my goal of five place accuracy. I talked
theoretical work there was a substantial budget of indirect and worked with the computer staff on the problem for about
time (called 9603 time at Douglas) that one was relatively free two weeks and no explanation was found. The discrepancy
to charge to in the late 1940's. Gradually, due to tighter could not be tolerated, so in desperation I sat down at a
budgetary controls, this fund became smaller and smaller over Marchant and, using exactly the same mathematics and
the years. roundoff procedures, set out to duplicate a run by hand. After
I do not remember any specific battles to sell authorization about a day and a half of steady calculating I found my first
of funds for the G00107 or DESA-2 airfoils although I do disagreement. This gave some guidance as to where to look
remember talking about them and their principles. The work for the trouble. The deck of cards containing the program for
was authorized as amendments to the XF4D-1 tailless airplane this problem was about 0.9-in. thick. In order to reduce the
contract as peripheral studies with the idea that results might frequency of reloading the deck in the hopper, eleven
be applicable to the airplane in the future. So you see that duplicate decks were made, making a dozen in all. What we
funds did not have to be contributed by the company, they found when we traced through all the decks was that one card
came from the Navy. I have always found the Navy alert to had been misplaced so that eleven times instructions were
developments that might lead to better airplanes. Because the correct but were wrong the twelfth time. The erroneous card
work was under amendments to a classified military contract, determined the value of one of the last terms in the Taylor
it too was classified until the end of 1955. series extrapolation so the error was small. This experience
The computer mainly used at this time was the IBM Card has made me extremely cautious about trusting the output of
Programmed Calculator (CPC). It was an electromechanical a large scale computer on a complicated problem, because
relay type and a great advance over a Marchant-type desk there are so many possibilities for error.
calculator, but almost infinitesimally slow compared with I was not sure of the propagation and growth of roundoff
modern calculators. A set of IBM cards was properly pun- errors, so after finding all the solutions, I checked by
ched, using machine language, and put in a hopper to instruct rerunning them again with two more terms in the Taylor series
the machine on what to do. It was a considerable advance and longer steps. Everything checked. I hope this chronicle
over earlier ones for which a plugboard had to be wired. gives you the flavor of automatic computers in the early days.
Storage was limited and mechanical. Computing rate was
governed by rate of feed of the card deck. I did not observe IV. Developing a General Method
the operation of the CPC closely when we tried to solve the for Calculating Laminar Boundary Layers
Go'rtler problem but I did on another problem. After my success with the panel method of solving the
The original table of solutions to the Falkner-Skan Neumann potential flow problem, I chose as a home study
equation were calculated by D. R. Hartree, using a Bush-type project an attempt to solve the inviscid transonic flow
mechanical differential analyzer which yielded between three- problem. I studied at home for about a year without success. I
and four-place accuracy. In our airfoil and other work we considered all kinds of methods, but did not consider pure
used his solutions so extensively that it seemed to be a good finite-difference calculations because the computing machines
idea to work out more complete and more accurate solutions. were too slow.
Now that a much better and digital type of computer was Thus you see that my sole interest was not boundary layers,
available, it did not seem to be a large job. but after about a year of frustration, I turned back to them.
Accordingly, we started working, charging to the general Now my goal was not to control the boundary layer but just to
9603 indirect charge number. (At Douglas, computer use was try to tell what it would do on its own. I wanted a method that
free just like a desk calculator, to foster their use; so in the would be accurate enough for stability calculations and that
good old days the only problem was access.) In a moment of would be quite general. At the time, some methods had ac-
optimism I decided to make tables of five-place accuracy. curacy but were not sufficiently general, and other methods
Only later did I gradually realize that developing that ac- had generality but were not sufficiently accurate. In fact,
curacy for certain on a nonlinear equation is a formidable Table 1, taken from report ES 40446,4 is an excellent sum-
undertaking. We solved the equation by a Taylor series ex- mary of the methods available at the time and their
trapolation, beginning the work at Santa Monica because El limitations. All, except the last method which was our
Segundo's computer was not yet operable. After a while the development, had some sort of limitation. It should be said
work was transferred to El Segundo, where I often ran the that Oscar Seidmann of the U. S. Navy supported our work,
machine myself, using the shooting method to meet boundary but he questioned whether our proposed work was any ad-
conditions. vance, and for this reason, the rather elaborate chart was
If about 100 steps were taken and with the proper number prepared.
of terms in the series to assure five-place accuracy, a run with In trying to find a truly general and accurate method, I had
given initial conditions would require about 15 min. One no idea what to do. I read the literature extensively, both at
looked at the tab sheet to see whether the boundary-layer Douglas and at home. I studied various forms of equations. I
profile went above or below the asymptote. Then a new initial studied various methods of solution. There are many types of
condition was punched into a card and a new try made. If one transformations, Crocco's, von Mises', Falkner and Skan's,
had uninterrupted use of the machine, a proper solution for a Gortler's, Cope and Hartree's, and many others, not to
particular value of Hartree's /3 could be made in about half a mention all sorts of integral methods. For most trans-
day. So when I was doing this work I was mostly standing formations, I found some drawback. After looking at these
 u>
CO

Table 1 Summary of principal methods for solving the laminar boundary-layer equations

NO. METHOD WORKING FORM OF EQUATION IMPORTANT ACCURACY AS MEASURED ADVANTAGES CRITICISMS PRINCIPAL REASONS FOR REJECTION AS
CONTRIBUTORS BT ERRORS IN T, THE BASIC METHOD FOR PRESENT
SKIN FRICTION INVESTIGATION
1. Momentum Von KaVman's momentum Von Karman (Ref. l) 3 figure accuracy possible Very fast and convenient. In well Accuracy entirely unsatisfac- Completely unable to handle many
Integral integral equation Pohlhausen (Ref.l) in the most favorable conditioned problems accuracy is tory for problems of boundary problems of importance. Poor,
Walz (Ref.l) cases. Greater than 100 good*. layer stability. Method in- even on well conditioned problems
percent error in adverse sensitive to shape of upstream in compressible flow.
cases. boundary layer. It considers
only thickness.
2. Correlation None, it is based on Thwaites (Ref. 13) About the same as above. About the same as above. About the same as above. Completely unable to handle many
empirical correlation of problems of importance. Extensive
available exact solution. compressible solutions essential
for correlation are not now
available.

3. Local Based on similar solu- Smith (Ref. 16) About the same as Nos. About the same as Nos. 1 and 2. About the same as Nos. 1 and Method has promise as an approxi-
Similarity tions, as of Falkner- 1 and 2. 2. mate procedure, but little work
Skan equation. has been done. At best it will be
completely unable to handle many
problems of importance.

k. Infinite Various transformations Blasius (Ref. l) 5 to 8 figure accuracy Exact solutions can be obtained Cannot handle ordinary Unable to handle the most common
Series of momentum equation Howarth (Ref. l) possible in the most easily for certain fortuitous pressure distributions such type of pressure distribution. In
where x, y remain essen- Tifford (Ref. Ik) favorable cases, as near edge velocity distributions. as those of airfoils. high speed flow there are so many
tially as the independent Gortler (Ref. l) stagnation point. Several variables, that calculation of
variables . Cope and Hartree hundred percent error in universal functions is impractical.
(Ref. 15) adverse cases.
5- Dorodnitzin Transformed momentum Dorodnitzin (Ref. 17) Theoretically exact, but Principal advantage is ability For high accuracy a high Computing and setup time appears
equation in which x and Pallone (Ref. 1?) insufficient work has been to give much better answers order system of simultaneous unreasonably long when high accuracy
y remain essentially as done to establish accuracy- than No. 1 with reasonable first -order differential is desired as for B.L. Stability
the independent variables computing time relation. labor. Method is essentially equations must be solved. investigations .
High accuracy will probably a generalized Pohlhausen method. Starting requires special
require lengthy computation procedure.
6. Finite Unmodified momentum and Wu (Ref. 18) Theoretically exact, but Simple and sufficiently fast Boundary layer thickness in Accuracy-computing time relation
Difference, energy equations. Gortler (Ref. l) insufficient work has been for low accuracy. Theoreti- this system varies greatly, is poor, if high accuracy demanded.
Explicit done to establish the cally can handle any boundary causing computing complica- Starting the solution is a problem.
accuracy, computing time layer problem exactly in limit. tions. Computing time exces- Numerical stability question is a
relation. sive where high accuracy de- disadvantage.
manded. Has numerical stabil-
ity problem.
7. Finite Crocco's transformed Baxter and Flugge- Theoretically exact. Accu- Theoretically exact and con- Computing time great where Computing time too long, ' 'overshoot1'
Difference, equation, in which x, u Lotz (Ref. 19) racy high (3 to 5 figure) venient formulation, because high accuracy demanded. difficulty eliminates Crocco's equa-
Explicit are independent vari- Raetz (Ref. 20) Computing time is long. boundaries are at u = 0, Crocco's equation not suit- tion as basis for a completely
ables . u = 1, and Crocco's equation is able for problems involving general method.
second order. ' 'overshoot' ' . Has numerical
stability problem.
8. Finite Crocco's transformed Kramer and Lieber- Theoretically exact, but Same as No. 7 above plus rapid Crocco's equation not suit- ' 'Overshoot' ' difficulty eliminates
Difference, equation, in which x, u stein (Ref. 21) insufficient work has been computing time. Computing is able for problems involving Crocco's equation from consideration.
Implicit are independent vari- done to establish accuracy - fast because a ' 'trick' * good "overshoot".
ables. computing time relation. for second order equations has
Computing time should be been used. No numerical stabil-
short. ity problem.
9- Ear tree- Simple transformation Hartree (Refs. k, Theoretically exact. Accu- Theoretically can handle any Probably slower than No. 8
Womersley of momentum equation in 5, 6) racy high (3 to 5 figure). boundary -layer problem, exactly in cases where No. 8 can
Finite which x, y remain essen- Smith and Computing time how medium, in limit. Computing time is good. handle the problem.
Difference, tially as independent Clutter but could be reduced Has best starting procedure of
Implicit variables . greatly. any exact method. No numerical
stability problem.

*By "well conditioned" is meant that the problem to be solved is similar to those for which the method is at its best. N.B. References are from original report.
NOVEMBER 1981 THE BOUNDARY LAYER AND I 1383

various forms for some time hoping to find something that handle the species equation for such things as nonequilibrium
would give an easy solution but not finding it, common sense flows. A summary of this work is in a Pergamon Press
got the better of me, and I realized that what I wanted was in publication. 5 This method could truly produce high-accuracy
the full partial-differential Falkner-Skan equation. It did not answers for any two-dimensional or axisymmetric problem
distort the x axis and had no pathological problems such as that arose. Any Pr(x,y) (Prandtl number), any edge velocity,
with overshoot profiles, or starting. (An overshoot profile is any suction distribution, any wall temperature distribution, or
one in which the velocity ratio u/ue may exceed unity before any heat transfer distribution could be analyzed. The only
decaying to its final asymptotic value. Accelerating flows past restriction we encountered was a sensitivity to short x steps
hot walls can generate them.) We gradually focused on this toward the rear of the body, but that was overcome. Our
equation, but for a considerable time, we considered original Douglas report was issued in July 1961.4 While the
Howarth- or Gortler-type universal solutions for a series method has been largely superseded, it still remains as one
representation of the edge velocity. We studied this at that is both accurate and simple to write out and program.
Douglas. Both velocity and temperature distribution were
represented by series expansion but I do not remember all the
conditions being represented. Hajimu Ogawa, now a V. Other Laminar Work
professor of mathematics, worked on this problem, and he About 1965, Professor Marten Landahl and R. E. Kaplan
obtained universal solutions that were far more general than came out with their purification scheme of solving the Orr-
any in existence, such as those of Tifford. It was a major Sommerfeld equation. Because this broke a bottleneck in the
accomplishment and easily could have been an important numerical solution, we decided to get active in this area and I
paper. talked it over with R. S. Shevell, Director of Aerodynamics,
However, I got to thinking. To have an adequate supply of pointing out the gains in knowledge and the possibility of
universal solutions for the many variables of a general getting research-type contracts. He approved getting into the
problem, a very extensive library of universal solutions would field; then I proposed work on it to Professor A. R. Wazzan,
be needed. I recalled the old saying, a table of one variable a consultant. He studied and learned the problem and the field
requires a page, two variables a book, three an encyclopedia, and began calculations. Shortly after we began our work, a
and four a large library. I realized that the number of dif- new book by Bellman and Kalaba, entitled Quasilinearization
ferential equations that must be solved to provide the and Nonlinear Boundary Value Problems, came out in 1965.
necessary universal functions would be far more than if I Bellman and Kalaba had very good reputations for con-
solved the basic Falkner-Skan partial differential equation tributing useful methods, so I bought the book and gave it to
directly for each individual problem that came up—at least in Tom Okamura in my group to see if it had anything useful in
my lifetime. Besides, the universal-series-type solutions are it. He discovered Gram-Schmidt orthogonalization which was
poor for airfoil-type pressure distributions. Therefore, after a more rigorous and powerful generalization of Landahl and
accomplishing all this good work, we dropped it cold. Kaplan's purification technique. I had nothing to do directly
Because I knew Galerkin's method very well from the with its application but do claim enough vision to buy
Gortler stability problem, we tried this form of solution, as Bellman and Kalaba's book. I do not know how we stand
far as writing out the system of simultaneous equations; but it timewise with respect to the orthogonalization procedure and
was nonlinear, and we did not like it. Another that we studied other workers. For instance, Radbill et al. came out with the
was Picard's method, the method of successive substitutions. method at about the same time, but at least we discovered it
We tried it on Hartree's ordinary differential equation, and it independently. Eigenvalues are basically found numerically
was fast and converged beautifully for all values of /3. I have by calculating independent solutions of the Orr-Sommerfeld
already mentioned that I solved this same equation some years equation and then combining them to meet boundary con-
before by the shooting method, to construct the tables of ditions. One of the solutions used is a weakly growing ex-
solutions of the Falkner-Skan equation for various values of ponential and the other is strongly growing. Due to roundoff
0. the strongly growing solution can find its way into the weakly
With our success using Picard's method, we obtained Navy growing solution. Then, at high Reynolds numbers after
support to try to solve the full partial-differential equation. calculating for some distance in y, one does not have the
To do that, we planned to represent the x-derivative terms by necessary two independent solutions needed for meeting
finite-difference approximations which then gave us an or- boundary conditions. They are proportional. Gram-Schmidt
dinary differential equation at such x station to solve in the y orthogonalization keeps them independent.
direction. Hartree and Hartree and Womersley had used this After learning and exploring the Gram-Schmidt or-
method successfully with the old Bush mechanical differential thogonalization procedure which now made it possible to
analyzer to solve a number of problems. reach almost any Reynolds number, I thought a very useful
Hence, in getting under contract with the Navy to solve the contribution would be to modernize Pretsch's old Falkner-
full equations, we had confidence because of our success with Skan stability charts and do both spatial and temporal am-
Hartree's ordinary differential equation. The additional x- plification calculations. We did this, and the charts were first
derivative terms looked small, and we did not see them as any published in 1968. 9
problem. In fact, we considered the flows more or less as The original e method used Pretsch's charts, hand
locally similar, and the x term would merely be a slight calculation, and interpolation between charts. We now had
correction. the capability of solving the stability equation at each x
However, that is not the way it worked. Picard's method station for the exact profile at that station, supplied by the
now would not converge, and try as we might with various exact solution of the partial-differential boundary-layer
alterations, it still would not converge. The contract time was equations as described earlier. Okamura, Jaffe, and Wazzan
running out, and we were getting desperate. So one day, we carried out this work which became a general and highly
dropped Picard's method and switched to the old shooting accurate method of calculating amplification ratios,
method that had been used without trouble on the Hartree j8 frequencies, etc. The method then became rather dormant at
profiles. It worked fine. That is how our original method was Douglas Aircraft, because transports have no laminar flow.
born. It was hardly a cold, deliberate, and logical study that Around 1974, the Navy first found out about this work and
then gave birth to a successful method. that it could analyze both two-dimensional and axisymmetric
The original work was on the momentum equation alone. It flows, the effect of heating and cooling in both water and air,
is the worst problem, because the equation is nonlinear. After suction, etc. Because such calculations were of little direct
success with it, it was fairly easy to extend the same method to interest to Douglas, I proposed to the Navy that we pull
include the energy equation. Further extension was made to together our potential flow calculations that gave pressures,
1384 A.M.O. SMITH AIAA JOURNAL

our boundary-layer calculations that gave boundary-layer mainly as y2, but as y4 at the very wall, which is theoretically
profiles, and our stability calculations that gave amplification correct. Then when it reached the value indicated by the outer
rates. They bought the proposal, and it was pulled together by formula, we would switch to it and finally fade everything out
A. E. Gentry with A. R. Wazzan handling the stability aspect. as the edge was reached. This formulation is quite simple but
Gentry named the program TAPS for transition analysis the results were in reasonable agreement with the ex-
program system. It is now in widespread use by various Navy perimental data that was available. While numerous ex-
organizations doing preliminary design studies of torpedoes tensions of the method have been made, it is interesting that
and other bodies. my original formulation has never been changed, not even the
One of the reasons the stability program was of interest to constants.
the Navy was that it had been extended to handle the problem It is interesting to note that I often received a lot of "static"
of water flow past warm walls. This modification began in the from O. R. Dunn, a later Director of Aerodynamics, about
middle sixties while we were still developing our boundary- our laminar efforts. He would ask, "Why are you wasting
layer and stability methods. I was aware that accurate your time on laminar boundary layers when the DC-8 is all
solutions of boundary-layer flows in water with heat transfer turbulent?" One day in connection with this turbulent ex-
did not exist so I suggested them as a kind of homework tension when he again asked me this question, I went to his
exercise for developing our methods, Douglas having no basic blackboard and wrote out the full incompressible equation
interest in water flows. These were worked out and Professor including the turbulent viscosity term e. He could see that the
Wazzan made a simple but accurate modification of the Orr- structure was the same for both the turbulent and laminar
Sommerfeld equation to analyze their stability. Both the cases, and that work on the laminar was a kind of training
boundary-layer calculations and the stability analysis were exercise for work on the turbulent case. After that, he became
presented in 1967 at an ASME meeting. The stability analysis a big booster for all of it.
revealed a profound effect, a very small amount of warming I did not have a high degree of confidence that in-
the walls increases the stability very, very much, almost as corporating accepted basic turbulent properties in the full
much as suction. At about the same time (1968), Hauptmann equations of motion would give good answers for all kinds of
discovered the effect, using approximate methods, whereas boundary-layer flows because an algebraic eddy-viscosity
our methods were highly accurate. We had no knowledge model has obvious flaws. Instead, initially I regarded the
beforehand of Hauptmann's work. effort as one merely to find an answer to the question, "If we
use the best available algebraic transport properties plus the
VI. Turbulent Boundary Layers exact equations of motion, how good are the answers?" Our
At least for airplanes, laminar boundary layers are not the first efforts were just to use the described eddy-viscosity
common case, turbulent layers are. Recognizing this fact and formulation and introduce it into the equations of our
now having success with the laminar case, I began to think shooting method. We then solved a number of cases by this
about the turbulent problem, not in general, but only for method to see what kind of accuracy was indicated and found
boundary layers. It seemed to me there were certain basic it to be very good for a considerable variety of problems. Our
relations that applied regardless of the pressure first paper about this method was given in 1966.6
distribution—the law of the wall, PrandtPs mixing length The studies seemed to answer my question favorably, so we
relation, the velocity defect law, intermittency at the edge of went ahead with development. The shooting method was very
the boundary layer, and Van Driest's modification of the law slow; the eddy viscosity was a variable function of the
of the wall. Previous attempts at calculating turbulent boundary layer itself, so a double iteration was needed. Hence
boundary layers had all used approximate forms of PrandtPs I started out to find a better method. After reading Fox's
equations or even none at all. I wondered what kind of an- book, The Numerical Solution of Two-Point Boundary
swers we might get if we incorporated these rather universal Problems, I settled on a five-point finite-difference method.
laws into the exact equations now being solved regularly for We were developing and applying this method at the time T.
laminar flows. Incorporation as we did it just meant adding Cebeci joined my group at Douglas. He soon got into this
an algebraic turbulent eddy-viscosity term to the equations. work and made the new finite-difference method operational.
Of course it was not a constant. For several years this appeared to be the leading all-around
More specifically, after examining the available literature method for boundary layers. Dr. Cebeci devised several
and information on the subject, such as Townsend's and ingenious extensions to the eddy-viscosity formulations to
Hinze's books, I used an inner eddy-viscosity formula, an account better for suction and blowing, low Reynolds
outer eddy-viscosity formula, and a decay law. For the inner number, heat transfer, etc. This extended version did very
eddy-viscosity formula, PrandtPs mixing length formula well at the Stanford Symposium in 1968.
basically was satisfactory, but I liked Van Driest's im- The method is still used throughout the country but, it has
provement of it, which had a certain amount of physical been considerably superseded by Cebeci's extension to
justification. The formula used was Keller's Box Method, a powerful and more universal
numerical method than my five-point method. This is only a
=pK2y2(1_e-y/A)2 ^L
mathematical method of solution; the equations are just the
e

dy same. There now are numerous turbulent methods, some far

more sophisticated than mine. Sometimes their predictions
Here /c is von K£rmdn's constant. The pK2y2 quantity is just are better and sometimes worse, so on the average our method
PrandtPs mixing-length formula. Van Driest's modification is still holding up for ordinary boundary layers. When Cebeci
brings solutions properly to the very wall. For the outer and I wrote our book about turbulent boundary layers, I
region, we used a form suggested by F. Clauser, but with a feared it might become obsolete rather fast because there was
slightly different constant, namely, great activity in developing various transport equation
methods. They definitely are more logical, but still lack a
e0 = 0.0168pued*y basic ingredient—a sounder description of turbulence—and
so seem rather bogged down. For ordinary boundary layers,
As you can see, at any particular station e0 is a constant our simple algebraic eddy-viscosity formulation seems as
viscosity except for 7, the intermittency factor. Finally, as one good as any and much easier to calculate. Ordinary boundary
gets to the outer edge, intermittency sets in and the turbulent layers are the important kind on aircraft.
transport processes fade away. We used Klebanoff's ex- When I began to try to calculate the turbulent boundary
pression for 7 to describe this fadeout. So you see that, at the layer, I was not aware of any similar effort. More accurate
very wall, the turbulent viscosity was zero, but then it grew calculation of turbulent boundary layers was just a desirable
NOVEMBER 1981 THE BOUNDARY LAYER AND I 1385

goal. After a while, I became aware of Mellor's work which the design engineer who probably needed an answer the next
used a similar approach, and of Bradshaw and Ferriss's work, day.
but not at first, when the course of development was being set. In general, at least for the most significant or important
problems, I had little trouble getting the support of
VII. Closure management. Of course I had to show clearly the goals and
A final contribution in the field of turbulent boundary- gains that would occur if a project was successful. We were
layer theory was my extension to the axisymmetric case of rather selective about seeking government support, going
Stratford's method of predicting separation. Stratford's after it only when we thought we could accomplish the job
method is an approximate procedure that is easy to use and and at the same time benefit the company. This is why we
reasonably accurate and so is very useful. At the time I did solicited considerable work from the Navy in hydrodynamics,
this work, the main methods of predicting this kind of because much hydrodynamics is just incompressible
separation were by Presz and Pitkin or else by some method aerodynamics. Also, I carefully watched another aspect—to
like the Cebeci-Smith that involves the full partial-differential be sure our work was useful to the company. I have seen a
equations. These are rather long, and I wondered if one could number of industrial research organizations founder,
not find a shorter method—an extension of Stratford's. I was probably because their work was too far out or too unrelated
at UCLA at that time and had some free time before a to the company's more immediate problems. So we only
research contract on torpedo-like bodies came in. So I decided worked on problems for which we could make a strong
to use this time by seeing if I could work out the extension. I defense if attacked, the Neumann program, for example. I
managed to. Stratford's method involves a transformation of had a simple motto "there are far more problems than we can
boundary layers from one position to another. It is interesting possibly solve, so we might as well pick on ones that are
that the old DESA-2 airfoil report materially helped me on useful."
this problem because, in it, I had worked out a related One final comment is in order about why I gradually drifted
transformation for the boundary layer on each side of a into research. I suppose the ambition of most young engineers
suction slot. A paper about the Stratford extension was is to be a chief engineer. It was one of mine and I attained it
published in ZAMP and dedicated to Professor Nicholas Rott very early—at Aerojet. However, I soon found it was not
on his 60th birthday. what I wanted. I found myself just interpreting other people's
In thinking back about my contributions, I find a very high input which at times I believed was biased. Gradually I
percentage of my really basic ideas were conceived at home, realized I preferred creating information rather than in-
not at work. Generally I was not seriously thinking about a terpreting something given to me by others. This is just a way
problem, but still it was in the back of my mind, and suddenly of saying I preferred the technical side or more specifically
I saw a solution. Another process that is very common is as creating information, which is just research. So after going
follows. I would work on some problem rather intensively all back to Douglas I stuck to the strictly technical.
day long, without satisfaction. I would usually be trying to I hope this confession is of interest and helpful to others.
implement some line of action but without full success. Also, I would like to make it clear that if I had not had a
During the day, the problem was studied thoroughly and group of very capable people at Douglas to help me im-
learned in intimate detail. Then frequently, when I had quit plement the methods, much less would have been done. To
for the day, given up thinking, and freed my mind from the them, who are too many to name, I feel much gratitude; I was
course followed all day, I would suddenly see the satisfactory lucky to have such good help.
solution I had been searching for. I think my subconscious Acknowledgment is made, with thanks to Richard S. Lee
brain is smarter than my conscious one. Also, I am more or for suggesting the title.
less a loner. I have never been especially good at theoretical
work and have not been very good at theoretical exchanges References
1
and mutual creation of ideas. The only way that has been Schlichting, H. and Bussmann, K., "Exakte Losungen fur die
successful for me has been continuous and persistent study by Laminare Grenzschicht mit Absaugung und Ausblasen," Schriften
myself in some area. I am rather slow to understand a subject der Deutschen Akademie der Luftfahrtforschung, May 7, 1943.
2
Smith, A. M. O., "Design of the DESA-2 Airfoil," Douglas
and, consequently, have to work on it very had. However, Aircraft Co., Rept. ES17117, AD 143008, Nov. 1952.
when I finally understand it, I understand it well. 3
Smith, A. M. O., "Rapid Laminar Boundary Layer Calculations
While working on research within the engineering by Piecewise Application of Similar Solutions," Journal of the
department of an aircraft manufacturing company (Douglas Aerospace Sciences, Vol. 23, Oct. 1956, pp. 901-912.
4
Aircraft Co.) has its disadvantages, there are certain ad- Smith, A. M. O. and Clutter, D. W., "Solution of the In-
vantages. One is that a person is more intimately exposed day compressible Laminar Boundary Layer Equations," Douglas Aircraft
to day to real life problems and directly feels the great Co., Rept. ES 40446, AD 266271, July 1961.
5
pressure for better and practical answers. This fact was often Jaffe, N. A. and Smith, A. M. O., "Calculation of Laminar
in the background of my contributions. I would see some area Boundary Layers by Means of a Differential Difference Method,"
Progress in Aerospace Sciences, Pergamon Press, 1972, pp. 49-211.
where there were problems, then gradually, I would see the 6
Smith, A. M. O., Jaffe, N. A., and Lind, R. C., "Progress in
essence of the problem, e.g., general turbulent boundary- Solving the Full Equations of Motion of a Compressible Turbulent
layer calculations. Then my group and I would figuratively go Boundary Layer," presented at the Seventh BOWACA Symposium
off to the side and work on a fundamental solution. Being a on Aeroballistics, U. S. Naval Missile Center, Point Mugu, Calif.,
research group, we could take a longer range approach than AD634988, June 1966.