Advanced Engineering Thermodynamics (The Constructal Law) - Adrian Bejan

13
The Constructal Law

of Configuration Generation
13.1
THE CONSTRUCTAL LAW
Why is geometry (shape, structure, similarity) a characteristic of natural flow

systems? What is the basis for the hierarchy, complexity, and rhythm of nat
ural structures? Is there a single physics principle from which form and
rhythm can be deduced without the use of empiricism?
I think that there is such a principle, and it is based on the common (uni
versal) observation that if a flow system (e.g., river basin, blood vessel) is
endowed with sufficient freedom to change its configuration, the system ex
hibits configurations that provide progressively better access routes for the
currents that flow. Observations of this kind come in the billions, and they
mean one thing: A time arrow is associated with the sequence of flow con
figurations that constitutes the existence of the system. Existing drawings are
replaced by easier-flowing drawings.
I formulated this principle in 1 996 as the constructal law of the generation
of flow configuration [ 1 -4]:
For a finite size flow system to persist in time (to survive) it must evolve in such
a way that it provides easier and easier access to the currents that flow through it.
This law is the basis for the constructal theOJy of organization in nature,
which was first summarized in the 1 997 edition of this book. The origin of
the term constructal is explained in Ref. 2 and Section 1 3 .2.3. Today, after
the new developments outlined in this third edition, this entire body of work
represents a new extension of thermodynamics: the thermodynamics of flow
systems with configuration (Section 1 3 .7.3).
705
706
THE CON STRUCTAL LAW O F CO NFI GU RATI O N GENERATI ON
To see why the constructal law is a law of physics, ask why the constructal
Jaw is different than (i.e., distinct from, or complementary to) the other laws
of thermodynamics. Think of an isolated thermodynamic system that is ini
tially in a state of internal nonuniformity (e.g., regions of higher and lower
pressures or temperature, separated by internal partitions that suddenly break).
The first and second laws account for billions of observations that describe a
tendency in time, a time arrow: If enough time passes, the isolated system
settles into a state of equilibrium (no internal flows, maximum entropy at
constant energy, etc.) (see Sections 2.4 and 6. 1 ). The first and second laws
speak of a black box. They say nothing about the configurations (the draw
ings) of the things that flow. Classical thermodynamics was not concerned
with the configurations of its nonequilibrium (flow) systems. It should have
been. After all, Sadi Carnot did not write about black boxes. He wrote about
machines (contrivances, configurations) and how they should change (i.e.,
improve) in the future.
This tendency, this time sequence of drawings that the flow system exhibits
as it evolves, is the phenomenon covered by the constructal law: not the
drawings per se, but the time direction in which they morph if given freedom.
No drawing in nature is "predetermined" or "destined" to be or to become
a particular image. The actual evolution or lack of evolution (rigidity) of the
drawing depends on many factors, which are mostly random, as we will see
in Fig. 1 3 . 1 8. One cannot count on having the freedom to morph in peace
(undisturbed).
Once again, the juxtaposition of this new law with the laws of classical
thermodynamics can be useful. No isolated system in nature is predetermined
or destined to end up in a state of mathematically uniform intensive properties
so that all future flows are ruled out. One cannot count on the removal of all
the internal constraints. One can count even less on anything being left in
peace, in isolation.
As a thought, the second law does proclaim the existence of a "final"
state: the concept of equilibrium in an isolated system, at sufficiently long
times. Similarly, the constructal law proclaims the existence of a concept: the
equilibrium flow architecture, when all possibilities of increasing morphing
freedom have been exhausted. We discuss this in greater detail in Table 1 3.5.
Constructal theory is now a fast-growing field with contributions from
many sources, which have been reviewed on several occasions [5- 12] . In this
chapter I keep the educational aspect of the theory [ 1 1 , 1 3] in plain view as I
try to strike a balance between the original ( 1 997) version of the chapter and
the newer developments [5 - 12]. Striking a balance is a welcome opportunity
to reflect, because during the 1 0 years that passed, I questioned my earliest
work and improved my thinking, results, drawings, and language. The bottom
line, however, is that constructal theory is the 1 996 law cited in the third
paragraph of this section.
How to deduce a class of flow configurations by invoking the constructal
law is an entirely different (separate, subsequent) thought, which in my teach-
THE CON S TRUCTAL LAW
707
ing effort is called the researcher's freedom to choose the problem and so
lution method [ 1 4] . The constructal law statement is general; it does not use
words such as tree, complex versus simple, or natural versus engineered.
There are several classes of flow configurations, and each class can be derived
from the constructal law in several ways: analytically (pencil and paper) or
numerically, approximately or more accurately, blindly (random search) or
using intelligence (strategy, shortcuts), and so on. Classes that my group
treated in detail, and by several methods, are the cross-sectional shapes of
ducts, the cross-sectional shapes of rivers, internal spacings, and tree-shaped
architectures [5, 1 0, 1 1 ] .
Regarding trees, our group treated them not as modelst (many have pub
lished and continue to publish models) but as fundamental access
maximization problems: volume to point, area to point, line to point, and the
respective reverse flow directions. Important is the geometric notion that the
"volume," the " area," and the "line" represent infinities of points. Our the
oretical discovery of trees stems from the decision to connect one point
(source, or sink) with an infinity of points (volume, area, line). It is the reality
of the continuum (the infinity of points) that is routinely discarded by mod
elers who approximate the space as a finite number of discrete points and
then cover the space with "sticks" drawings, which (of course) cover the
space incompletely (and from this, fractal geometry). Recognition of the con
tinuum requires a study of the interstitial spaces between the tree links. The
interstices can only be bathed by high-resistivity diffusion (an invisible, dis
organized flow), whereas the tree links serve as conduits for low-resistivity
organized flow (visible streams, ducts).
The two modes of flowing with thermodynamic imperfection, the intersti
ces and the links, must be balanced so that together they contribute minimum
imperfection to the global flow architecture. The flow architecture is the
graphical expression of the balance between links and their interstices. The
deduced architecture (tree, duct shape, spacing, etc.) is the optimal distribu
tion of impe1fection. Those who model natural trees and then draw them as
black lines on white paper (while not optimizing the layout of every black
line on its allocated white patch) miss half of the drawing (here I am being
kind-they miss the whole picture !). The white is as important as the black.
Our discovery of tree-shaped flow architectures was based on three ap
proaches. In 1 996, I started with an analytical shortcut [ 1 ,2,4] based on several
simplifying assumptions: 90 angles between stem and tributaries, a construc
tion sequence in which smaller optimized constructs are retained, constant
thickness branches, and so on. Months later, we published the same problem
[ 1 5] but we did it numerically by abandoning most of the simplifying as
sumptions (e.g., the construction sequence) used in the first papers. In 1 998
we revisited the problem numerically [ 1 6] in an area-point flow domain with
' The great conceptual difference between modeling and theory is spelled out in Section 13.7.4.
708
THE CON STRUCTAL LAW OF CONFIGU RATION GENERATI ON
random low-resistivity blocks embedded in a high-resistivity background by

using the language of Darcy flow (permeability instead of conductivity and
resistivity). Along the way, we found better performance and "more natural
looking" trees as we progress in time; that is, as we endowed the flow struc
ture with more freedom to morph.
So I end this introduction with the "click" that I felt as I ended my second
paper on constructal trees (for the full version, see pp. 8 1 3 -8 1 5 in Ref. 2):
The commonality of these phenomena is much too obvious to be overlooked. It
was noted in the past and most recently (empirically) in fractal geometry, where it
was simulated based on repeated fracturing that had to be assumed and truncated.
The origin of such algorithms was left to the explanation that the broken pieces (or
building blocks, from the point of view of this paper) are the fruits of a process of
self-optimization and self-organization. The present paper places a purely deter
ministic approach behind the word "self" : the search for the easiest path (least
resistance) when global constraints (cmTent, flow rate, size) are imposed.
If we limit the discussion to examples of living flow systems (lungs, circulatory
systems, nervous systems, trees, roots, leaves), it is quite acceptable to end with
the conclusion that such phenomena are common because they are the end result
of a long running process of "natural selection." A lot has been written about
natural selection and the impact that efficiency has on survival. In fact, to refer to
living systems as complex power plants has become routine. The tendency of living
systems to become optimized in every building block and to develop optimal as
sociations of such building blocks has not been explained: it has been abandoned
to the notion that it is imprinted in the genetic code of the organism.
If this is so, then what genetic code might be responsible for the development of
equivalent structures in such nonliving systems as rivers and lightning? What ge
netic code is responsible for man-made networks (such as the trees in this paper)?
Certainly not mine, because although highly educated, neither of my parents knew
heat transfer (by the way, thermodynamics was not needed in this paper). Indeed,
whose genetic code is responsible for the societal trees that connect us, for all the
electronic circuits, telephone lines, air lines, assembly lines, alleys, streets highways
and elevator shafts in multistory buildings?
There is no difference between the living and the nonliving when it comes to the
opportunity to find a more direct route subject to global constraints, for example,
the opportunity of getting from her to there in an easier (faster) manner. If living
systems can be viewed as engines in competition for better thermodynamic per
formance, then physical systems too can be viewed as living entities (animal s ! ) i n
competition for survival. This analogy is purely empirical: w e have a n immense
body of case-by-case observations indicating that flow configurations (living and
nonliving) evolve and persist in time, while others do not. Now we know the par
ticular feature (maximum flow access, minimum global flow) that sets each surviv
ing design apart, but we have no theoretical basis on which to expect that the design
that persists in time is the one that has this particular feature. This body of empirical
evidence forms the basis for a new law of nature that can be summarized as (the
THE AREA-POINT ACCESS PROBLEM
709
constructal law, at the start of this section). This "fourth law" b1ings life and time
explicitly into thermodynamics and creates a bridge between physics and biology.
In this chapter we review this natural phenomenon-this progression to

ward better flowing architectures in general-by looking at the work of our
group and that of many other contributors. All such work is based on invoking
the constructal law and supports constructal theory. This is how in 2003 we
ourselves aITived at the analytical formulation of the constructal law [ 1 7] ,
such that constructal theory is now the thermodynamics of flow systems with
confi g uration [ 1 1 , 1 8] (see Section 1 3 .7.3).
13.2
As I wrote in Section 3 .6.2, my interest in maximizing flow access between

one point and an infinity of points was triggered by an exam problem re
garding fluid networks which I wrote for this course in 1 995. Consequently,
I first worked on deducing trees for fluid flow (e.g., Section 3.6.2), but when
I saw how old, voluminous, and established the empirical (modeling) litera
ture of this field is (from physiology to river morphology), I remembered hard
lessons learned early in my career. No way. It is risky for an amateur to
submit something entirely new to an established group [ "risk" means rejec
tion and worse: loss of credit for the idea, because (in accordance with con
structal theory) good ideas travel fast! ] . The surer approach was to translate
the idea into the safer language of minimum travel time [ 1 ] and minimal
thermal-diffusion resistance [2] in freely morphing area-point flows, and to
submit the idea to peers who do not have an ax to grind. It worked, despite
the delay that this double translation caused in the dissemination of the con
structal law.
In this section we review briefly the area-point problem chronologically,
in the order in which it was published: street patterns, heat conduction, and
fluid flow. The emphasis this time is on the fact that tree-shaped flows can
and have been deduced from the constructal law in several ways, from the
simplest to the more accurate: simple construction sequence, global numerical
optimization, and numerical generation (by erosion) of the best position of
each grain of low-resistivity material on a high-resistivity background.
13.2.1
Street Patterns: A Simple Construction Sequence
Why are streets usually arranged in clusters (patterns, grids) that look almost
similar from block to block and from city to city? Why are streets and street
patterns a mark of civilization? Indeed, why do streets exist? In this section
I outline a purely deterministic theory that provides answers to these questions
710
THE CON S TRUCTA L LAW O F CONFIGU R ATION G ENERATION
in an astonishingly simple and direct way. In later sections I show that the
implications of this theory extend well beyond transportation, into physiology
and thermodynamics. The theory is the result of addressing the following
access optimization problem:
Consider a finite-size geographical area A and a point M situated inside A or on its
boundary (Fig. 1 3 . 1 ). Each member of the population living in A must travel be
tween his or her point of residence P(x, y) and point M. The latter serves as common
destination for all the people who live in A. The density of this traveling popula
tion-that is, the rate at which people must travel to M-is fixed and described by
1i" (people/m 2 s). This also means that the rate at which people are streaming into
Determine the optimal bouquet of paths that link the
M is constrained, 11 =
points P of area A with the common destination M such that the time of travel
required by the entire population is the shortest.
1i"A.
The problem is how to connect a finite area (A) to a single point (M).
Area A contains an infin ite number of points, and every one of these points
must be taken into account when optimizing the access from A to M and
back. Time has shown that this problem was a lot tougher than the empirical
game of connecting "many points" : that is, a finite number of points distrib
uted over an area. The many-points problem can be solved on the computer
using brute-force methods (random walk or Monte Carlo-more points on
better computers), which are not theory.
The fundamental problem of Fig. 1 3. 1 was stated above in the most general
and abstract terms, because its solution and its diverse manifestations benefit
from such a formulation. It helps, however, to see a real-life problem in this
statement before attempting a solution. The area A could be a flat piece of
farmland populated uniformly, with M as its central market or harbor. It also
helps to think in time by beginning with the most ancient type of community
that faced this access optimization problem. The oldest solution to this prob-
A rea A
!VI
P(x, y )
1i"
Figure
1 3 . 1 Finite-size area (A) covered by a uniformly distributed population (Ii'')

traveling to a common destination (M) (From Ref. 1 .)
THE AREA-PO lNT ACCE SS PRO BLEM
71 1
!em was also the simplest: Unite with a straight line each point P and the
common destination M, and you will minimize the total time spent by the
population en route to M.
The straight-line solution was probably, the prefeITed pattern as long as
humans (load and ox) had only one mode of locomotion: walking, with the
average speed V0. The farmer and the hunter would walk straight to the point
(farm, village, river) where the market was located. This radial pattern of
access paths can still be seen today, especially in perfectly flat and uniformly
rural areas such as the plain of the lower Danube in Romania. The once
ancient (Roman times) market is now a larger village, and the surrounding
farmers have become a constellation of almost equidistant tiny villages.
The radial pattern disappeared naturally in areas where settlements were
becoming too dense to permit straight-line access to everyone. Why the radial
pattern disappeared "naturally" is the present problem.
Another important development was the horse-driven carriage: With it,
people had two modes of locomotion, walking ( V0), and riding in a carriage
with an average velocity V 1 that was significantly greater than V0. It is as if
the area A became a composite material with two conductivities, V0 and V 1
Clearly, it would be faster for every inhabitant (P, in Fig. 1 3 . 1 ) to travel in
straight lines to M with the speed V1 This would be impossible, however,
because the area A would end up being covered by beaten tracks, leaving no
space for the inhabitants and their land properties.
The modern problem, then, is one of bringing the street near a small but
finite-size group of inhabitants; this group would have to walk first to reach
the street. The problem is one of allocating a finite length of street to each
finite patch of area A 1 , where A 1 A. The problem is also one of connecting
these street lengths in an optimal way such that the time of travel of the
population is minimum.
The first analytical approach that I chose for solving this problem is "at
omistic" from the smaller subsystem (detail) of area A, to the larger subsys
tem, and ultimately to area A itself. I had used this time direction in 1 98 1 ,
in the buckling theory of eddy formation and self-similar (compounded) de
velopment of turbulence (Section 1 3 .3 .6). This direction is also the direction
of time, history and growth: see the spreading of a flow architecture to its
limits, which are reached at "equilibrium" [ 1 8] (Section 1 3 .7.3).
The area subsystem to which a street length may be allocated cannot be
smaller than the size fixed by the living conditions (e.g., property) of the
people who will be using the street. This smallest area scale is labeled A 1 in
Fig. 1 3.2. For simplicity we assume that the A 1 element is rectangular. Al
though A 1 is fixed, its shape or aspect ratio H 1 I L 1 is not. Indeed, the first
objective is to anticipate optimal form: the area shape that maximizes the
access of the A 1 population to the street segment allocated to A 1
Symmetry suggests that the best position for the street segment is along
the longer of the axes of symmetry of A 1 This choice has been made in Fig.
1 3.2, where L1 > H 1 and the street has length L 1 and width D 1 The traveling
7 12
THE CON STRUCTAL LAW OF CONFIGUR ATI ON G ENER ATION
T
1
H1
r
0
1A1
. ,,
11
D1
Vo
-
P(x, y)
V1
. ,,
n
L1
Figure 1 3 .2
Smallest (innermost) elemental area, A 1 , and the street segment allocated

to it. (From Ref. 1 .)
population density 1i" is distributed uniformly on A 1 To get out of A 1, each

person must travel from a point of residence P(x, y) to the (0,0) end of the
street. The person can travel at two speeds: ( 1 ) a low speed V0 when off the
street, and (2) a higher speed V1 when on the street.
We assume that the rectangle H 1 x L 1 is sufficiently slender (L 1 > H i ) so
that the V0 travel is approximated well by a trajectory aligned with the y axis.
This makes sense from the point of view of the person who must travel from
P(x, y) to (0, 0) the fastest; since the street is a faster conduit, it is reasonable
to head straight toward it (i.e., vertically in Fig. 1 3 .2). We show later that
when V1 is not sensibly greater than V0, the optimal V0 travel is not straight
downward but at an angle such that the V 1 travel is somewhat shorter than in
Fig. 1 3 .2.
The time of travel between-P(x, y) and (0,0) is (x/ V1 ) + (y/ V0). The average
travel time of the A 1 population is given by
(13.1)
which yields
( 1 3 .2)
The elemental area is fixed (A1 = H 1 L 1 , constant); therefore, t 1 can be expressed as a function of H 1 , which represents the shape of A 1 :
THE AREA-POINT ACCE SS PROBLEM
) 1 12
713
The average travel time has a sharp minimum with respect to H , . Solving
at 1 I aH, = 0, we obtain
H , . opt =
and subsequently,
L , _ opt =
(1')
I
opt
v:
A,,
12
)1
(
( 1 3 .4)
V' A '
2 vo
( 1 3 .5)
:<1
( 1 3 .6)
2 0
I
Equation ( 1 3 .6) shows the optimal slenderness of the smallest area element
A 1 This result validates the initial assumption that H, IL , < 1 ; indeed, the
optimal smallest rectangular area should be slender when the street velocity
is sensibly greater than the lowest (walking) velocity.
According to eq. ( 1 3 .6), the rectangular area A , must become more slender
as V1 increases relative to V0-that is, as time passes and technology ad
vances. This trend is confirmed by a comparison between the streets built in
antiquity and those that are being built today. In antiquity the first streets were
short, typically with two or three houses on one side. In the housing devel
opments that are being built today, the first streets are sensibly longer, with
1 0 or more houses on one side. This contrast is illustrated beautifully by
modern Rome (Fig. 1 3.3), the actual birthplace of civility (city living, liter
ally). In the center of Fig. 1 3 .3, which is the ancient city, the streets are
considerably shorter than in the more recently built, pe1ipheral areas (e.g., the
upper corners in Fig. 1 3 .3).
The remaining analysis can be shortened based on the observation that
exactly the same optimum [eqs. ( 1 3.4)-( 1 3 .6)] is found by minimizing the
longest travel time (t, ) instead of minimizing the average time of eq. ( 1 3 . 1 ).
The longest time is required by those who travel from one of the distant
corners (x = L 1 , y = H, 12) to the origin (0,0) and is given by
( 1 3.7)
Eguations ( 1 3 .7) and ( 1 3.2) show that the geometric minimizations of t,

and t , are equivalent. It is both interesting and important that the optimization
71 4
THE CON STRUCTAL LAW OF CONFIG URATI ON G ENERATI ON
Figure 1 3 .3
Plan of modern Rome, showing that in the ancient city (the center) the
street length scales are considerably shorter than i n the newer outskirts. (Rep1inted by
permission of the Istituto Geografico de Agostini, Novara, Italy.)
of the A 1 element is of interest to every inhabitant: What is good for the most
disadvantaged person is good for every member of the community. This con
clusion has profound implications in the spatial organization of all living
groups, from bacterial colonies all the way to our own societies. The urge to
organize is an expression of selfish behavi01'.
The time obtained by minimizing t 1 or by substituting eqs. ( 1 3.4) and ( 1 3 .5)
into eq. ( 1 3.7), is
t i . min =
( )112
vOJI
2A
( 1 3 .8)
At this minimum, the two terms that make up t , in eq. ( 1 3 .7) are equal. This
equipartition of time principle means that the total travel time is minimum
when it is divided equally between traveling along the street and traveling
perpendicularly to the street.
Another observation concerns the width of the first street segment, D 1 The
total "flow rate" of travelers generated by the A 1 element, and taken out of
A 1 through the origin (0,0), is ri"A 1 The same quantity can be expressed as
p 1 D, V, , where p, is the instantaneous number of persons found per unit of
715
street area in the vicinity of the exit (0, 0). In conclusion, the first street width
is given by
( 1 3.9)
where both p1 and V1 are technological parameters. Equation ( 1 3.9) sheds
light on the time evolution of the smallest street, which is also the innermost
street in a more complex (more evolved, newer) grid. When the smallest street
was first built, D1 was dictated by the width of one carriage. But as the
traveling density (or population, and affluence) increased in time, the con
strained D 1 stimulated technological developments that led to increases in the
product p1 V1 that matched the increases in 1i". The technological aspect of
V1 is clear: The family car is faster than the best cmTiage, even on the smallest
street. Increases in p1 on the other hand, were registered as the number of
vehicles present on the road increased.
In Fig. 1 3.2 we see the smallest loop of the rectangulm grid that will
eventually cover the given area A. The next question is how to connect the
D 1 streets such that each innermost loop has access to the common destination
M. One answer-the simplest, albeit approximate-is obtained by repeating
the preceding geometric optimization several times, each time for a larger
area element, until the largest scale (A) is reached.
Consider then the rectangular area A 2 = H2L2 shown in Fig. 1 3.4. This
area consists of a certain number of the smallest patches A 1 The purpose of
this assembly of A 1 elements is to connect the D 1 streets so that the traveling
population (Ii'' A 2) can leave A 2 in the quickest manner. We invoke symmetry
as the reason for placing the new (second) street along the long axis of the
T
1
r I
i
t
Dz
V2
I
I
I
I
A , -1
1 D1 I+I -.
A,
I
I
Ai
1 3 .4 Area construct A 2 as an assembly of connected innermost elements A 1

(From Ref. I .)
Figure
71 6
THE CONS TRUCTAL LAW OF CONFIGURATION G ENERATIO N
A 2 rectangle. In Fig. 1 3 .4, the stream of travelers (ri"A 2) leaves A 2 through the
left end of the D2 street.
For the sake of consistency in notation, we write 2 and
for the speed
V p2
and traveler density associated with travel on the second street. These param
eters are generally not the same as the corresponding parameters of the first
street
most likely), because technology evolves in time. The

same time arrow is indicated by the growth process in which several A 1
neighborhoods are now incorporated into a larger community called A 2
With reference to Fig. 1 3 .4, the longest time of travel occurs between one
of the distant corners of A 2 and the left end of the D 2 street:
(p2 V2 p1 V1 ,
( 1 3. 1 0)
The first term accounts for the portion traveled along the D 2 street, and the
second term represents the minimum time required to travel across the A 1
element that occupies the most distant corner. Using eq. ( 1 3.8) for min' eq.
( 1 3 .5) for A 1 , the geometiic assembly relation L1 = H / 2 and the area con
straint A 2 = H2L2 , we can rewrite eq. ( 1 3. 1 0) to show explicitly the effect of
area shape (H2) on the total travel time:
2 ,
t1 .
( 1 3. 1 1 )
The optimal shape of the A 2 area element is determined by m1mm1zmg t2
with respect to H2 and again finding the equipartition of travel time,
1 12
V
1
H2. or1 = ( , A , _)
V_
12
V
L2. opt ( 2 k_)
V1
=
t2.
min
( Av7 ) 1 12
vi
( 1 3 . 1 2)
( 1 3 . 1 3)
( 1 3 . 1 4)
( 1 3. 15)
These results look similar to eqs. ( 1 3.4) -( 1 3 .6) and ( 1 3 .8); however, this
time the ratio
needs not be significantly smaller than 1. Note that in
this section we did not have to assume that the area element A 2 is slender; in
Fig. 1 3 .4 the travel through the A end-corner is always "downward" because
VJV2
71 7
of the D 1 street, not because of the assumption H2 < L2 , which we did not
make. Equation ( 1 3 . 14) shows that the optimal A 2 element is more slender
when the speed on the second street (the avenue) is greater than on the first
street.
The geometric relation H2 = 2L 1 and eqs. ( 1 3 .5 ) and ( 1 3 . 1 2) provide the
relation between the sizes of the first two area elements, or the optimal num
ber of elements A 1 that must be assembled into one element of size A 2 :
( 1 3 . 16)
As in eq. ( 1 3 .9), we find that the width of the second street is D2 = 1i"A 2/ p2
V2 , or that the street width enlargement factor is
( 1 3 . 1 7)
This ratio can be expected to be greater than 1 because of eq. ( 1 3.6) and the
fact that in a given technological age, p2 is of the same order as p 1 while V 1
is sensibly larger than V0 .
The atomistic construction started in Figs. 1 3 .2 and 1 3 .4 can be continued
toward larger assemblies as shown in Refs. 1 and 5 and Fig. 1 3.5. This is not
the best way to allocate streets to areas mathematically, but it is the simplest
and most transparent. Its value is that it shows the emergence of a tree net
work (the streets) from principle (the maximization of access), not by copying
from nature. In constructal theory, fl o w architectures such as trees are dis
covered. They are now known, observed, modeled, or copied from nature.
This sequence is shown in Fig. 1 3.5 (top) only for illustration, because it
is unlikely to be repeated beyond the third-generation street. The reason is
that as the community and the area inhabited by it grow, other common
destinations (e.g., church, hospital, bank, school, train station) emerge in A
in addition to the miginal M point (Fig. 1 3 . 1 ). Some of the streets that were
meant to provide access to only one end of the area element must be extended
all the way across the area to provide access to both ends of the street. As
the destinations multiply and/or shift around the city, the dead ends of the
streets of the first few generations disappear, and what replaces the growth
pattern of Figs. 1 3.2 to 1 3 .5 is a grid with access to both ends of each street.
The multiple scales of this grid, and the self-similar structure of certain areas
(neighborhoods) of the grid, however, are the fingerprints of the deterministic
organization principle uncovered in this section.
The optimal access problem solved in this section and Ref. 1 was stated
in two dimensions (Fig. 1 3. 1 ). The corresponding problem in three dimen
sions is this : Minimize the time of travel from all the points P of a volume
V to one common destination point M, subject to the constraint that the trav-
718
THE CON STRUCTAL LAW OF CONFIGURATION GENERATI ON
(b)
A 2 (Fig.
1 3.4)
le) v
-1 = 3 '
vI
(d)
Figure
13.5 Top: higher-order constructs in the sequence of Figs. 1 3 . 2 and 1 3 .4.

(From Ref. I .) Bottom: urban growth patterns in which each construct was optimized
for overall shape and angle of street confluence. (From Ref. 1 9 . )
TH E AREA-PO INT ACCE SS PROBLEM
719
eling population rate is fixed. One application is the sizing and shaping of
the floor plan in a multistory building, along with the selection and placement
of the optimal number of elevator shafts and staircases. The floor plan opti
mization proceeds according to the method illustrated based on Fig. 1 3.2,
where V0 corresponds to the travel through the rooms and V1 is the travel
along the corridor. The unknown in this first building block is the shape of
the floor area A 1 Instead of using Fig. 1 3.4 to determine the optimal assembly
of the A 1 elements, we stack a certain number (n 2 ) of A 1 elements on the
vertical, such that the next path ( V2 ) is vertical and accounts for the elevator
or the staircase. The construction sequence may be taken to a third assembly
(or even higher-order assemblies) if the towers optimized in the three
dimensional equivalent of Fig. 1 3.4 must be integrated into a larger building
with several wings. The construction in three dimensions was illustrated with
fluid-flow and heat-conduction analogs in Refs. 4 and 20.
The same organization theory can be extended generally to areas that are
populated unevenly, or specifically to highways, railroads, telecommunica
tions, and air routes (e.g., the organization of such connections into hubs, or
centrals). In the 1 997 edition of this book I noted that a clear application of
these concepts is in operations research and manufacturing, where the inven
tion of the first auto assembly line is analogous to the appearance of the first
street. This was demonstrated subsequently by the research thrust illustrated
in Refs. 2 1 to 23.
Another opportunity to generate the street pattern geometry arises in cases
where the vehicular speed (V;, i 2': 1 ) increases with the street width (D;), and
the paved surface of all the streets is constrained. This aspect is treated in
Ref. 1 and the 1 997 edition of this book.
The atomistic construction sequence presented until now is just an approx
imate and simple way to illustrate how a tree of organized (channeled) flow
emerges on a background covered by individual (disorganized) movement.
The "exact" way to generate the tree architecture from the same principle is
to endow the flow architecture with maximum freedom to morph [ 1 8] and to
use numerical simulations to morph the flow structure through all its eligible
configurations. Some of this more exact work is illustrated in the next two
subsections. Here we illustrate it by relaxing the assumption that in Figs. 1 3 .2
and 13 .4 the paths intersect at right angles.
For example, consider again the geometric optimization of the smallest
elemental area (Fig. 1 3.2) by assuming that the angle between the V0 and V 1
paths may vary. This general situation is shown in Fig. 1 3.6, which is set for
calculating the maximum travel time between the distant corner (P) and the
common destination (M). In place of eq. ( 1 3 .7), we obtain
( 1 3 . 1 8)
720
THE CON STRUCTAL LAW OF CONFIGURATION G ENERATION
Figure 1 3 .6
T
1
Smallest area (A 1 ) and the variable angle between the V0 and V1 paths.
V0
H1
where f3 is the angle between the

path and the
side. Now the minimi
zation of has two degrees of freedom: the geometric aspect ratio H1 I and
the angle f3 . The optimal angle for minimum
t1
f3opt
- -1 VoI
t1 ,
L1
S 111
( 1 3 . 1 9)
Va
confirms the statement made above eq. ( 1 3 . 1 ) that

should be perpendicular
to vi (i.e., that f3 = 0) when
v i . The minimization with respect to I
subject to A =
is the same as earlier in this section. The twice
minimized travel time is
Vo
H
L
1 11
L1
tI
.min =
( Vo I
2A
COS
HI
112
)
f3opt
( 1 3.20)
The lower part of Fig. 1 3 .5 shows four examples of optimal urban growth,
in which each area construct (A 1 , A 2 ,
) has been optimized in two ways:
overall shape and angle between each new street and its tributaries. The as
sumed changes in velocity are listed under each drawing. Comparing exam
ples a and d we see that when the velocity increase factor
is large
the street pattern spreads fast (in few steps) over the given area, and each
area assembly is slender. In the opposite limit, the spreading rate is lower,
the assembly steps are more numerous, and each area assembly is less slender.
These trends appear together in example d, where the velocity increase factor
decreases as the construction grows.
Comparing eq. ( 1 3.20) with eq. ( 1 3.8), we note that the second degree of
freedom (the optimized angle f3) plays only a minor role as soon as
is
greater than
In other words, the change from
to
does not have to
VJV;_ 1
Va.
V0 V1
V1
THE AREA-POI NT ACCESS PROBLEM
721
be dramatic for the f3 = 0 design (Fig. 1 3 .2) to perform nearly as well as the
optimal design. We reach the important conclusion that small internal varia
tions in the organization pattern have almost no effect on the global perform
ance U 1 .min' in this case) of the organized system. We will run into this
observation more than once as we progress through the chapter. The practical
aspect of this observation is that a certain degree of variability (imperfection,
if you will) is to be expected in the patterns that emerge naturally. These
patterns are not identical, or perfectly similar; this accounts for the historic
diffi c ulty of attaching a theory to naturally organized systems. Natural patterns
are quasi-similar, but only in the same sense in which no two human faces
are identical. Their performance, however, is practically the same as that of
the mathematically optimized pattern. We call these top performers equilib
rium flow structures [ 1 8] : see Section 1 3 .7.3. The contribution of constructal
theory is that the performance and the main geometric features (mechanism,
structure) of the organized system can be predicted in purely deterministic
fashion.
13.2.2
Heat Flow Trees
We now turn our attention to another field-another language-that is essen

tial to thermodynamicists: heat transfer. In the cooling of electronic compo
nents and packages, the objective is to install a maximum amount of
electronics (heat generation) in a fixed volume in such a way that the maxi
mum temperature does not exceed a certain level. Most of the cooling tech
niques that are in use today rely on convection or conjugate convection and
conduction, where the coolant is either a single phase fluid or one that boils.
The frontier is being pushed in the direction of smaller package dimensions.
There comes a point where miniaturization makes convection cooling im
practical, because the ducts through which the coolant must flow take too
much space. The only way to channel the generated heat out of the electronic
package is by conduction. And this conduction path will have to be very
effective (of high thermal conductivity, kP).
Conduction paths, too, take space: Designs with fewer and smaller paths
are better suited for the miniaturization evolution. The fundamental point
volume heat flow problem formulated in 1 997 edition is:
Consider a finite-size volume i n which heat is being generated at every point, and
which is cooled through a small patch (heat sink) located on its boundary. A finite
amount of high conductivity (k,,) material is available. Determine the optimal dis
tribution of k,, material through the given volume such that the highest temperature
is minimized.
The function of the electronic package or assembly of packages is the

generation of heat at a certain rate (q) in a fixed volume ( \!), so that the
rI
722
THE CON S TRUCTAL LAW OF CONFIGURATION GENERATION
temperature difference between the hot spot (the heart of the package) and
the heat sink (on the side of the package) will not exceed a certain value.
The heat generation rate per unit volume is q"' = q/ V; for simplicity, we
assume that q"' is constant (i.e., q is distributed uniformly), although this
assumption can easily be abandoned in follow-up studies of the fundamental
problem.
The fraction of the volume V that is occupied by all the high-conductivity
paths is VP . This fraction too is fixed, although as noted already, a smaller
ratio V,) V is better for miniaturization. For simplicity, we assume that "'1,
V. The thermal conductivity of the conducting paths is constant (k") and much
larger than the thermal conductivity of the electronic material (k0) that oc
cupies the rest of the volume.
The easiest and most visible way to present the emergence of geometry is
in a plane. For this we make the assumption that the heat generation and
temperature fields are two-dimensional.
The most basic function of any portion of the conducting path is to be in
touch with the electronic material that generates heat volumetrically. In this
way we arrive at a problem of optimal allocation of conducting path length
to volume of heat generating material, or vice versa. An extremely important
observation is this: The allocation cannot be made at infinitesimally small
scales throughout V, because the conducting paths must be of finite length so
that they can be interconnected to channel the total q to one side of V, where
the heat sink is located.
We illustrate the optimization of length to volume allocation at the smallest
volume scale. The elemental volume shown in Fig. 1 3 .7 is finite and fixed
(H0L0 W), where W is the dimension in the direction perpendicular to the plane
(x, y). It is a small part of the volume V of the actual device, and its size is
YI
Ho/2
:::J
"'
H0L0W
-H0!2
Figure 13.7
L kp
"'
ko, q
ko , q"'
Do
+
t
Lo
Slender elemental volume with uniform volumetric heat generation and

one high-conductivity path along its axis of symmetry. (From Ref. 2.)
THE AREA-PO INT ACCESS PROBLEM
723
dictated by manufacturing constraints. The dimensions H0 and L0 may vary

while their product A0 = H0L0 remains fixed. The chief unknown in this
generic problem is the shape of the elemental volume (or the geometric aspect
ratio H0/ L0) that maximizes the access of the heat generated (q111H0L0 W) to
the lone conducting path that services the elemental volume. Symmetry and
the requirement to minimize the maximum temperatures are why the con
ducting path is placed on one of the axes of symmetry of the elemental
volume. The entire heat generation rate is channeled to the exterior through
one end of its allocated conducting path. With the exception of the migin (0,
0), the boundary of the rectangular area H0 x L0 is adiabatic. We assume that
the shape of the rectangular area is sufficiently slender,
( 1 3.2 1 )
that the conduction through the heat generating material (k0) can be assumed
oriented in the y direction. The validity domain of assumption ( 1 3.2 1 ) will
be established later in eq. ( 1 3.29). Integrating the equation for steady con
duction with uniform heat generation in the k0 material (e.g., Ref. 24, p. 47),
a 2 T q111
+ - = 0
ko
ay2
( 1 3 .22)
subject to the boundary conditions aT/ ay = 0 at y = H0/2 and T = T0(x) at

y = 0, we obtain
q/11
T(x, y) = lk (HoY - y2 ) + T0(x)
0
( 1 3.23)
The conduction problem along the x axis is governed by the fin-type equa
tion (e.g., Ref. 24, p. 63)
2
a
,ic" Do c1 T
qm Ho = 0
dx2 +
( 1 3.24)
where q111 H0 accounts for the rate at which the generated heat is being col
lected by the high-conductivity path. Integrating eq. ( 1 3.24) subject to dT0/
dx = 0 at x = L0 and T0
T(O, 0) at x = 0 and substituting T0(x) into eq.
( 1 3 .23), we obtain
T(x, y) - T(O, 0)
= 2k0qm (HoY - y-,) + qmk DHoo (Lox - x22 )

p
( 1 3 .25)
724
THE CONS TRUCTAL LAW OF CONFIGURATION GENERATION
This expression is strictly valid for y > 0. The corresponding solution for y
< 0 can be obtained by replacing H0 with -H0 in the first term on the right
hand side.
In conclusion, eq. ( 1 3.25) shows that the maximum temperature in Fig.
1 3 .7 occurs at the corners situated the farthest from the origin: at x = L0 and
y = H0/2. We call this maximum temperature difference J},, T0, and we non
dimensionalize it by noting that the area (A0 = H0 L0) and the volumetric
heat generation rate q"' are fixed:
( 1 3 .26)
On the right side, the ratio D0/ H0 is a manufacturing constant (a constraint)
accounting for the proportion in which high-conductivity material is sand
wiched with blades of the original material at the elemental level. Equation
( 1 3 .26) shows that J},, T0 can be minimized with respect to the shape of the
elemental system (H(/L0). The optimization results are
( )
Ho
Lo
opt
- 2
( )
ko Ho
kp Do
I /?
( 1 3 .27)
( 1 3.28)
They are valid when the elemental system is slender; in view of eq. ( 1 3.27),
this condition means that the conductivity ratio k" /k0 should be much greater
than 1 , such that
( 1 3.29)
Two additional features of this geometric optimum are worth stressing.
First, the maximum difference along the x axis (from x = L0 to x = 0) is
exactly the same as the temperature drop from the hot spot (x = L0, y =
H0/2) to its projection on the x axis,
1
T(L0, 0) - T(O, 0) = T(L0, H012) - T(L0, 0) = l !},, To.min
( 1 3 .30)
The second feature can be seen by combining eqs. ( 1 3.27) and ( 1 3.28):
( 1 3 .3 1 )
725
This means that at the elemental level the excess temperature decreases as
ffl-hence the incentive for manufacturing the smallest possible elemental
system.
The generation of geometry for heat conduction at larger scales has been
pursued in several ways. In the first and simplest approach [2], optimized
building blocks such as Fig. 1 3.7 were assembled into progressively larger
constructs. At the same time, we published a fully numerical approach [ 1 5]
in which we replaced the construction sequence with numerical simulations
of conduction in the composite domain (k0, kP) , with freedom to vary all the
geometric features of the emerging tree structure. In what follows, I rely on
key results from Ref. 1 5 .
First, w e abandoned the assumption that the high-conductivity insert i s a
blade of constant thickness. As shown in Fig. 1 3 .8b, the optimal profile of
the kP insert is such that D0 increases as x 1 1 2 , where x is measured away from
the tip. Relative to eq. ( 1 3 .28), the decrease in the global thermal resistance
of the elemental volume is 6 percent. We obtained even greater reductions in
global thermal resistance when we abandoned the assumption that the element
is rectangular. Figure 1 3 .8d shows that there is an optimal leaflike elemental
shape, as there is an optimal shape for the k" fiber. Increasing the freedom to
morph the structure leads to higher performance levels and to designs that
look more natural.
Second, we abandoned the assumption that the kP branches are perpendic
ular to their stems. For example, in a first construct with /( k" I k0 = SO and
cp 1
V" 1 I V 0. 1 , we found that the optimal angle is such that the branch
deviates from the perpendicular by 4, and that in this refined geometry the
global resistance is smaller by 5.8 percent than before. This finding is anal
ogous to what we saw in Fig. 1 3 .6 for travel on a rectangular area. Angled
branches increase the global performance and make the tree architecture look
more natural.
Third, we abandoned the sequential construction of larger assemblies, and
in a composite domain such as Fig. 1 3.9, we optimized the numbers and
positions of every high-conductivity insert. There are three kP-blade thick
nesses, D0 < D 1 < D2 Figure 1 3.9 was drawn for k = k" / k0 = 300, cp2 =
VP2 / V = 0. 1 , DJD0 = S, D 2 1D 1 = 2, and eight D0 blades on one D 1 blade.
The purpose of this figure is to illustrate the effect that the number (n2 ) of
D 1 blades has on the global performance. From (a) to (c), the dimensionless
global thermal resistance D.T)(0/ q"'A 2 takes the values 0.0379, 0.0354 and
0.0374, where A 2 is the total size of the rectangular domain, and D.T2 is the
temperature difference between the hot spot (the left corners) and the heat
sink (the midpoint of the right side).
The competing designs in Fig. 1 3 .9 show that the best is (b), where the
number of D 1 inserts is n 2 = 4. This result differs from the simplest approach
[2], in which the rule of assembly was pairing (dichotomy) (i.e., 112 = 2). The
result that n2 = 4 was verified five years later by Ghodoossi and Egrican [26],
who derived it analytically using the method of Ref. 2.
=
726
THE CO NS TRUCTAL LAW OF CO NFIG URATION GENERATION
Resistance
C
lnsulaied
(a)
Tn=
(b)
2
Tmin C)
[,
ko. q"'
21 12
1.50
1 .41
Tmin C)
3/2
2
Tmin C)
172
1 .06
(c)
Tnw
(d)
Tmin C)
Do - .r115
Figure
13 .8 Elemental conduction volume with progressively greater freedom to

morph and progressively higher performance. (From Ref. 25.)
Figure 1 3.9 also shows that the performance of designs (a) and (c) is not
too far from that of design (b). This means that tree-shaped flow architectures
that have been optimized partially or completely are robust. By using the
results tabulated in Ref. 2, one can show that the analytical construction
sequence produces a structure with a global resistance (l::i. T2k0 / q"'A2) of
the same order of magnitude as in Fig. 1 3 .9, but larger. This comparison
shows the approximate character of the simplest approach and the merits
of increasing the number of degrees of freedom of the structure simulated
numerically. Because of the increased freedom, the nume1ical formulation

1 12
11 2
<!> 2 = 0. 1
(a)
Figure 13.9
(b)
ll2
727
<b = 0. 1
(c)
Second construct optimized numerically, and the effect of changing the

number (112 ) of high-conductivity inserts of i ntermediate size (D 1 ). (From Ref. 1 5 . )
allowed us to omit the thin (D0) branches that would have crowded the
stem (D2 ).
13.2.3
Constructal Theory versus Fractal Algorithms
In drawings such as Figs. 1 3 .5 and 1 3 .9, we discover tree patterns that max
imize flow access. Every detail of the tree geometry is the result of invoking
the constructal law. The discovery of the tree as the flow architecture for
maximal access between one point and an infinity of points is general; it is
not restricted to trees of streets and trees of high-conductivity inserts. The
generality of the tree discovery is stressed by Table 1 3 . 1 , which shows that
"how" unites and "what" divides. How the tree is generated (through a bal
ance between high resistivity and low resistivity) is the same in many classes
of flow systems, regardless of the diversity of the cmTents that flow through
them. To see the unique principle that unites is a lot harder than to scream
the obvious observation that diversity characterizes the flow systems of nature
and engineering. The common pattern generated by the constructal law is
subtle, while the diversity observed is banal.
To appreciate how much is new in this theory, it is important to note that
one portion of the network pattern (i.e., only the portion formed by the higher
order assemblies; Fig. 1 3 .5) is not new. It was first proposed in physiology
728
THE CON S TRUCTAL LAW OF CONFI GURATION G ENERATION
TABLE 1 3. 1 How, Not What: Balancing High-Resistivity Flow with Low-Resistivity

Flow in a Wide Diversity of Flow Systems
How
Application
What
Electronics
packages
River basins
Heat
Lungs
Air
Circulatory
systems
Turbulent flow
Urban traffic
Economics
Blood
Water
Momentum
People
Goods
Interstices:
High Resistance at the
Smallest (Fixed) Scale
Channels:
Low Resistance
at Larger Scales
Low-conductivity
substrate
Darcy flow through
porous media
Diffusion in alveoli,
tissues
Diffusion in capillaries,
tissues
Laminar, viscous diffusion
Walking in urban structure
Hand delivery and
collection
High-conductivity inserts
(blades, needles)
Rivulets, rivers
Bronchial passages
Blood vessels, capillaries,
arteries, veins
Streams, eddies
Street traffic
Freight, rail, truck, air,
ship
Source: Private communication from S. Perin.
as a three-dimensional heuristic model for the vascular system [27], where it

was known empirically that each tube is followed by two smaller tubes; that
is, each tube undergoes bifurcation. It was also known that the tube diameter
must decrease by a constant factor c2- 1 1 3 ) during each bifurcation; this result
was derived based on flow resistance minimization [28] and is the only theory
based notion present in the algorithms used to simulate the pulmonary tree
or other tree-shaped patterns that appear in nature (trees, roots, leaves, river
basins, deltas lightning, lungs, nervous system, vascular system). The descrip
tion of these geometric constructions was made popular through the advent
of fractal geometry; in fact, a two-dimensional version of Cohn's branching
fluid network [similar to Fig. 1 3.5 (top)] appeared in the books of Mandelbrot
[29] and Prigogine [30] , where it is presented heuristically as a "model of
the lung." These authors do not say anything predictive or deterministic en
route to the algorithm assumed in making the drawing. Artists who design
bathroom floor tile patterns enjoy the same creative freedom.
The fractal description of images such as Fig. 1 3.5 is only partially correct,
because it misses the genesis of such phenomena, which lies in the smallest
(finite, predictable) length scale. What is wrong with the fractal description
and by wrong I mean totally upside down-is the time arrow of the descrip
tion. For example, contrary to the impression given by Richardson's [3 1 ] eddy
cascade, which was adopted by Mandelbrot [29], in a turbulent flow field the
eddies do not break down from the large to the small, all the way down to
THE A RE A-POI NT ACCESS PROBLEM
729
the viscous length scale. The eddy cascade proceeds in the opposite direction
in time. In Refs. 5 and 14 I showed that eddies coalesce and arrange them
selves into larger and larger structures, the sizes of which increases stepwise.
The oldest geometric feature is the smallest, and the youngest (most recent)
feature is the largest. Never mind that eddies of the smallest (elemental) size
are being formed all the time-intermittently yes, but without interruption.
So are the smallest branches on a tree. The point is that at the start of its
existence, the system consisted of one element of the smallest size; this is
also the smallest geometric feature that is found in the system's structure at
subsequent points in time. The largest eddy and the largest branch are geo
metric features of the present.
If fractal is an appropriate Latin-based wordr for breaking things [29]
that is, for the opposite of the direction in which natural systems evolve
then the appropriate word for the geometry and evolution of organized natural
phenomena is constructal. *
Some would argue that fractal geometry has nothing to do with time, and
they would be right as far as descriptive geometry goes. The geometrical
images produced by the repetitive algorithms are frozen in time. The assumed
algorithm can certainly be executed in both directions, from the largest scale
to the smallest and from the smallest to the largest. As a descriptive aid for
natural phenomena, however, the fractal description represents a clear choice:
from the largest scale all the way to size zero in an infinite number of steps.
The word fractal has the concept of time built in it: The act of breaking
something evolves in time from large pieces to smaller pieces (see also p .
745 footnote).
It is not that the mathematician is wrong to be an artist and to paint a
lifeless image that resembles the instantaneous image of a natural system. As
a way to think about predicting the morphology of natural systems, however,
the fractal paradigm is oriented backward. The theoretician must still predict
the algorithm postulated by the mathematician or created by the artist. To
paraphrase Bloom's [32] remark about quantitative tools in river morphology,
fractals are description, not explanation.
13.2.4
Fluid-Flow Trees
We now turn our attention to the application of constructal theory to fluid

flow patterns that occur naturally. This is the class of organized phenomena
that has been studied the most. The features of treelike patterns in fluid flow
t From the Latin verb .firn1gere (to break), which survives unchanged in both Italian and Romanian.
* From the Latin verb co11stnli'!re (to build), which survives as construire in French, Italian, and
Romanian.
The techniques of quantitative tluvial geomorphology give an excellent descriptio11 of drainage
networks but no exp/a11atio11" (32, p. 204].
730
THE CONS TRUCTA L LAW OF CONFIGURATION G ENERATI ON
have been described quantitatively in surprisingly sharp (reproducible) terms,

notably in the study of lungs [33,34] , vascularized tissues [27,35,36] , botan
ical trees [37,38], and rivers [32,39-43] . In recent years, these natural
phenomena have been visualized on the computer by means of repetitive
fracturing algo1ithms, which had to be postulated. The origin of the phenom
ena was left to the "explanation" that the broken pieces are the fruits of a
nondeterministic process of se(foptimization and self-organization.
In this section we continue to show that constructal theory places a purely
deterrninistic method behind the word self: the search for the easiest flow path
when global constraints (flow rate, volume size) are imposed. For example,
previous studies of the fluid access problem used as starting point the empir
ical observation that tubes and streams bifurcate. This property is known as
dichotomous branching, which is a term derived from the Greek noun dicho
tomia (division into two parts). To predict this geometric feature-the integer
2-is an essential objective of the theory.
To construct a theory that is both effective and transparent, it is advisable
to rely on the simplest possible model that still retains the most essential
feature of the network: its function. Figure 1 3 . 1 0 shows that the function of
the network is to distribute the stream 111 from the point M to every elemental
volume /1 V. In some cases two superimposed networks are needed. For ex
ample, in the bronchial tree [Fig. 1 3. 1 1 (left)] the flow is periodic (in and
out), and only one network is needed. In the circulatory system [Figs. 1 3 . 1 1
(right) and 1 3 . 1 2] two identical networks in counterflow are needed, such that
each elemental volume /1 V receives arterial blood at the rate dictated by local
metabolism, 11111. The elemental volume returns the blood at the same rate to
the network of veins (e.g., the dark network in Fig. 1 3 . 1 0). Similar pairs of
networks in counterflow are encountered in trees, roots, and leaves. A river
basin needs only one network because the volume V is flat (the basin surface)
and l11ii is proportional to the rainfall per unit surface.
In sum, the function of the network is well represented by the volumetric
mass flow rate density n1"' = 11111./ /1 V, which must be collected (integrated)
by the network over the volume V and then channeled as a single stream (iii)
111
-
------

Ill
Figure 13. 10
a
Trees of ducts connecting one point (M) with every point residing inside
finite volume ( V) .
731
Figure 13. 1 1
Dichotomous branching in the air passages of the human lung (left) and
the coronary arteries of the heart muscle (right). (Courtesy of Ewald Weibel, University
of Bern. Reprinted from Ref. 34 with permission from Birkhauser Verlag, Basel, Swit
zerland.)
Figure 1 3. 1 2
Dichotomous branching in the arteries of the pig kidney (left) and rabbit
kidney (right). (Reprinted from Ref. 36 with permission from World Scientific Pub
lishing Co., Singapore.)
732
THE CONS TRUCTAL LAW OF CONFIG URATION G ENERATION
to the point M. This operational characteristic (111'") is assumed given and

serves as constraint in the generation of the network. For simplicity we as
sume that iii"' is distributed uniformly over V, such that 111 = 1i1"'V.
Focus on only one of the networks of Fig. 1 3 . 1 0 and recognize that 6.1i1 is
driven from the elemental volume to the origin by the pressure difference P
- PM. This difference varies with the position of the 6. V element relative to
M: Of special interest is the maximum pressure difference, 6.P = Pmax - PM,
which is needed by the elemental volumes that are situated the farthest from
the end of the network. In the lungs and the circulatory system, for example,
6.P is the pressure level that must be maintained by the thorax and heart
muscles. The time-averaged mechanical power consumed by these pumps is
proportional to the product 111 6.P (e.g., Section 3 .5.2). The total flow rate 1i1
is fixed, because V and 1i1"' are fixed. In conclusion, the thermodynamic op
timization of the network is equivalent to minimizing the maximum pressure
difference.
We generated tree-shaped architectures for fluid flow by using three meth
ods: the analytical construction [4] illustrated in Section 1 3.2. 1 , numerical
simulations of the flow in many configurations and the selection of the best
configuration [44], which was illustrated in Section 1 3 .2.2; and a grain-by
grain erosion model (construction scenario) in which the emerging tree has
maximum freedom to morph [ 1 6]. In this section I illustrate the grain-by
grain construction for several reasons: It is the newest, it is based on maxi
mum freedom and freedom is good for performance (global flow access) [ 1 8],
and because this type of construction is reappearing in the heat transfer lit
erature (e.g., Ref. 45), where the Darcy flow model [here eq. ( 1 3.32)] is
replaced by a completely equivalent model (Fourier conduction, an alternative
language for the same idea). Instead of grains of high permeability defining
a tree on a background of low pe1meability, in heat conduction grains of high
conductivity are distributed through a medium with low conductivity. Not
surprisingly, the new conduction trees look like the original (constructal) trees
for fluid flow [ 1 6] , even though they are sometimes termed bionic and evo
lutionary by their authors. For more on this tendency to rename an idea, see
the discussion provided recently by eight distinguished colleagues [46] .
We saw that a tree-shaped flow structure is generated by two flow regimes
that work hand in glove toward minimizing the volume-point flow resistance.
In Fig. 1 3 . 1 3 the two regimes are seepage (Darcy flow) through regions of
low permeability (K) and seepage through high-permeability regions (KP) cre
ated by grains that have been removed (eroded) . The smface area A = HL
and its shape HI L are fixed. The area is coated with a homogeneous porous
layer of permeability K. The small thickness of the K layer (i.e., the dimension
perpendicular to the plane H x L) is W, where W < < (H, L).
An incompressible Newtonian fluid is pumped through one of the A faces
of the A x W parallelepiped such that the mass flow rate per unit area is
uniform, 111" (kg/m2 s). The other A face and most of the perimeter of the
H x L rectangle are impermeable. The collected stream (n1"A) escapes through
THE AREA -POINT ACCE SS PROB LEM
733
y
H/2
111
"A
1I i1"--,
l_
w r------
T
A
I I
Figure 13.13
<
-------<:J
l' L
-I,,'
I
?peak
II
u-1---+--+-------1--
o
Kp D,D
P(x, y)
.r
-H/2
-------
\.
--
Area-to-point flow in a porous medium with Darcy flow and grains that
can be dislodged and swept downstream. (From Ref. 1 6.)
a small port of size D x W placed over the origin of the (x, y) system. The
fluid is driven to that outlet by the pressure field P(x, y) that develops ovr
A. The pressure field accounts for the effect of slope and gravity in a real
river drainage basin, and the uniform flow rate n"i'' accounts for the rainfall.
We can determine the flow field after making an assumption about the flow
regime (i.e., the relation between flow rate and pressure difference). A good
assumption is that the flow through the K medium is in the Darcy regime
[47]:
u=
K aP
'
, ax
v =
K aP
, ay
( 1 3.32)
In these equations, (u, v) are the volume-averaged velocity components and

, is the viscosity. The conservation of mass in every infinitesimal volume
element Lix y W requires that [ 1 4]
( 1 3.33)
where p is the density of the fluid. Combining eqs. ( 1 3 .32) and ( 1 3 .33), we
obtain a Poisson equation for the pressure field,
734
THE CON STRUCTAL LAW OF CONFIG URATION GENERATI ON
a1P a 2P n1''v
- + -2 + - = 0
WK
ax1 ay
( 1 3 . 34 )
where v = Ip is the kinematic viscosity. The boundary conditions for solv

ing eq. ( 1 3 . 3 4 ) are impermeable boundaries (u = 0 or v = 0) all around the
domain except over the sink located at x = 0 and -D/2 < y < D/2, where
p = 0.
The global resistance to this area-to-point flow is the ratio between the
maximal pressure difference (Ppeak) and the total flow rate (nl'A ). The location
of the point of maximal pressure is not the issue, although in the model
described until now its identity is clear (see Fig. 1 3 . 1 3). It is important to
calculate Ppeak and to reduce it at every possible turn by making appropriate
changes in the internal structure of the A x W system. Determinism results
from invoking a single principle and using it consistently.
Changes are possible because finite-size portions (blocks, grains) of the
system can be dislodged and ejected through the sink. The removable blocks
are of the same size and shape (square, D x D x W). The critical force (in
the plane of A ) that is needed to dislodge one block is TD2 , where T is the
yield shear stress averaged over the base area D2 . The yield stress and the
length scale D are assumed known. They provide an erosion criterion and a
useful estimate for the order of magnitude of the pressure difference that can
be sustained by the block. At the moment when one block is dislodged, the
critical force TD2 is balanced by the net force induced by the local pressure
difference across the block l:::.P: namely, l:::.P DW. The balance TD2 l:::.P DW
suggests the pressure-difference scale l:::.P
TD I W, which along with D can
be used for the purpose of nondimensionalizing eq. ( 1 3 . 3 4 ) :
( 1 3.3 5 )
(
.,,
x,
y) -
(x, y)
P=
p
-
TD! W'
M=
1ii"
vD
TK
( 1 3.3 6 )
B ased on what criterion do we search for the blocks that are dislodged?
Let s be the direction of the resultant of all the pressure forces that act_n
the block perimeter D x D. The block does not break away as long as (aP/
as) W < T , Which in dimensionless terms is
- <
aP
as
( 1 3.37)
The pressure gradient aP I as is averaged over the square base area of one
block. When condition ( 1 3 . 37) is violated, the block is removed and its place
is taken by a channel that is considerably more permeable-more conductive
735
for fluid flow-than the original medium (K). A simple way to implement
this change is to assume that the space vacated by the block is also a porous
medium with Darcy flow except that the new permeability (K") of this medium
is sensibly greater, K" > K. This happens to be the correct assumption when
the flow is slow enough (and W is small enough) that the flow regime in the
vacated space is Hagen-Poiseuille between parallel plates. The equivalent K,,
value for such a flow is W2 / 1 2. The main reason for introducing the K"
assumption in this model is to simplify the calculation of the pressure distri
bution over the area occupied by the dislodged block. Over that area, the
pressure is governed by an equation that is the same as eq. ( 1 3 .35) except
that M is replaced with MK! KP . The pressure varies continuously between
the original material K and the vacated domain K,,. When we account for
mass continuity across the interface between the K and the K,, domains, the
ratio Kl K,, emerges as a dimensionless parameter of the system.
The pressure P and the block-averaged gradient [condition ( 1 3.37)] in
crease in proportion with the imposed mass flow rate (M), because eq. ( 1 3 .35)
is linear. When M exceeds a critical value Mc, condition ( 1 3.37) is violated
and the first block is dislodged. For example, the critical flow-rate parameter
is Mc = 0.00088932: This value does not depend on Kl KP because in the
beginning the entire system is occupied by K material. The first block that
breaks away is the one that has the outlet port as one of its four sides. The
peak pressure is located in the farthest corners of the A domain and experi
ences a drop when the first block is removed at constant flow rate (M = MJ.
This is, in fact, the purpose of the change in the internal structure of the area
to-point flow system, that is, the physics principle that we invoke: The resis
tance to fluid flow is decreased through geometric changes in the internal
architecture of the system.
To generate higher pressure gradients that may lead to the removal of a
second block, we must increase the flow-rate parameter M above the first Mc,
by a small amount. The removable block is one of the blocks that borders
the newly created K,, domain. The peak pressure rises as M increases, and
then drops partially as the second block is removed. This process can be
repeated in steps marked by the removal of each additional block. In each
step we restart the process by increasing M from zero to the new critical value
Mc. During this sequence the peak pressure decreases, and the overall area
to-point flow resistance (PpeaJ MJ decreases monotonically.
The key result is that the removal of certain blocks of K material and their
replacement with K,, material generate macroscopic internal structure. The
generalizing mechanism is the minimization of flow resistance, and the re
sulting structure is deterministic: Every time we repeat this process we obtain
exactly the same sequence of images.
For illustration, consider the case KIK,, = 0. 1 , shown in Fig. 1 3 . 14. The
number n represents the number of blocks that have been removed. The do
main A is square and contains a total of 260 1 building blocks of base size D
x D; in other words, H = L = S ID. The pressure field equations ( 1 3 .35)-
736
THE CONSTRUCTAL LAW OF CONFIGURATION GENERATION

}/
II = 50
11 = 200
1 00
II = 800
70
0.10
K!K = O. l
p
60
0.08
50
0.06
40
30
0.04
20
0.02
IO
Figure 13. 14
Ref. 1 6. )
200
400
II
600
0.00
800
Evolution of the tree structure of Fig. 1 3. 1 3 when KI K,,
0. 1 . (From
( 1 3 .36) were solved with the finite-element method described in Ref. 1 6. In

most cases the variation of P over the edges of the blocks located near the
interfaces between the two porous domains was quite smooth, almost linear.
The average pressure on each face was obtained by numerical integration,
which was followed by calculation of the resulting force and its projection
onto the direction in which the block would be free to move. The entire
numerical procedure was automated: The cycle formed by calculations and
by the necessary changes in the internal structure (K, KP ) was repeated without
any inte1ference from the outside. That is, the curves plotted in Fig. 1 3. 1 4
are unique (reproducible).
Figure 1 3 . 1 4 also shows the evolution of the critical flow rate and peak
pressure. The curves appear ragged because of an interesting feature of the
erosion model: Every time that a new block is removed, the pressure gradients
redistribute themselves and blocks that used to be "safe" are now ready to
be dislodged even without an increase in M. The fact that the Mc values that
737
are plotted drop from time to time is due to restarting the search for Mc from
0 at each step n.
The shape o f the high-permeability domain KP that expands into the low
permeability material K is that of a tree. New branches grow in order to
channel the flow collected by the low-permeability K portions. The growth
of the first branches is stunted by the fixed boundaries (size, shape) of the A
domain. The older branches become thicker; however, their early shape (slen
derness) is similar to the shape of the new branches.
The slenderness of the KP channels and the interstitial K regions is dictated
by the Kl KP ratio, that is, by the degree of dissimilarity between the two flow
paths. Highly dissimilar flow regimes (Kl KP < < 1 ) lead to slender channels
(and slender K interstices) when the overall area-to-point resistance is 1nini
mized. This behavior confirms what we predicted analytically in eqs. ( 1 3 .6)
and ( 1 3 .27) for elemental streets and heat conduction. In other words, the
optimization and generation of the grain-by-grain structure confirms numer
ically the optimal allocation of flow lengths to area elements.
Figure 1 3 . 1 4 stresses the observation that the availability of two dissi1nilar
flow regimes (KP =/= K) is a necessary precondition for the formation of de
terministic structures through flow-resistance minimization. The "glove" is
the high-resistance regime (K), and the "hand" is the low-resistance regime
(K,,): Both regimes w9rk toward minimizing the overall resistance. Even
though the M ( n ) and Ppeak(n) functions show fluctuations that can be related
to temporary changes in the internal structure, the overall flow resistance Ppeak/
M, decreases monotonically as each additional block is removed (Fig. 1 3 . 15).
The decrease is more accentuated when the vacated space is more permeable
to flow than the original material.
The raggedness of the Ppeak(n) curves disappears when the flow-rate pa
rameter M is increased monotonically from one step to the next (e.g., Fig.
5 x 10
0. 1
o -l-,...,_,_,...,,.,...,...,,..,.,,.....-;
0
100 200 300 400 500 600 700 800
11
Figure
13.15 Monotonic decrease of the global flow resistance during erosion proc
esses of the type shown in Fig. 1 3 . 14. (From Ref. 1 6.)
738
1 3 . 1 6). In this new sequence, each step begins with the removal of the first
block that can be dislodged by the flow rate M. Following the removal of the
first block, the M value is held fixed, the pressure field is recalculated, and
condition ( 1 3.37) is applied again to the blocks that border the newly shaped
KP domain. The additional blocks that violate condition ( 1 3 .37) are removed.
To start the next step, the M value is increased by a small amount D.M. The
M(n) curves shown in Fig. 1 3 . 1 6 are "stepped" because of the assumed size
of D.M and the finite number (/in) of blocks that are removed during each
step. Although the monotonic M(n) curves obtained in this manner are not
the same as the critical flow-rate curves Mc(n) plotted in Fig. 1 3 . 14, they too
are deterministic.
Figure 13 . 16 corresponds to a composite porous material with KI KP = 0. 1 ,
which is the same material from which the river basin of Fig. 1 3 . 1 4 was
constructed. Compare the shapes of the high-conductivity domains shown in
these figures. The hand-in-glove structure is visible in all three figures; how-
70
60
50
40
30
K/Kp
!J.M
II
II = 100
11 = 50
200
0. 1 5
0. 1
0.001
0. 1 0
p peak
-+-=--
--
20
0.05
10
1 00
Figure
200
300
400
ll
500
600
700
0.00
800
13.16 Evolution of the tree structure of Fig. 1 3 . 1 3 when the flow rate M is
increased in steps flM = 1 0- 3 (KIK , = 0. 1 ). (From Ref. 1 6. )
,
NATURAL FLOW PATTERNS
739
ever, the finer details of the KP domain depend on how the flow rate M is
varied in time. The main difference between the patterns of Fig. 13 . 1 4 and
those of Fig. 1 3 . 1 6 is visible relatively early in the erosion process: Diagonal
fingers form when the flow rate is increased monotonically. In conclusion,
the details of the internal structure of the system depend on the external
"forcing" that drives it, in our case the function M(n). This sensitivity is
illustrated by Figs. 1 3. 1 4 and 1 3 . 1 6, which are two responses exhibited by
the same square system (KIK" = 0. 1 ) . The structure is deterministic, because
it is known when the function M(n) is known.
There are still major differences between natural river drainage structures
and the deterministic structures produced in this subsection. One obvious
difference is the lack of symmetry in natural river trees. How do we reconcile
the lack of symmetry and unpredictability of the finer details of a natural
pattern with the deterministic resistance-minimization mechanism that led us
to the tree networks of Figs. 1 3 . 14 and 1 3 . 1 6? The answer is that the devel
oping structure depends on two entirely different concepts: the generating
mechanism, which is deterministic, and the properties of the natural flow
medium, which are not known accurately and at every point [ 1 6] . To illustrate
how the resistance-minimization mechanism can lead to irregular tree net
works, let us assume that the resistance ( aP I aS) that characterizes each re
movable block is distributed randomly over the basin area. This characteristic
of river beds is well known in the fi e ld of river morphology.
In Fig. 1 3 . 1 7 we assumed for (aP I aS) a normal distribution with the mean
equal to 1 , a standard deviation equal to 0.66, and a variance equal to 0.00425 .
The gray-scale code attached to the square mosaic shows that the darker
blocks are easier to dislodge than the lighter ones. For the erosion process
we chose the system (Kl KP = 0. 1 ) and the M(n) function of Fig. 1 3 . 1 6, in
which M increased monotonically in steps of 0.00 1 . The evolution of the
drainage system is shown in Fig. 1 3 . 1 8 . The emerging tree network is con
siderably less regular than in Fig. 1 3 . 1 6 and reminds us more of natural river
basins. The unpredictability of this pattern, however, is due to the unlmown
spatial distribution of system properties, not to the geometry generating prin
ciple (the constructal law), which is Imown.
13.3
Next to the botanical trees that surround us, rivers provide the most numerous
and stunning natural images that exhibit the features anticipated based on
constructal theory. Rivers are an extremely important subject even without
the large-scale similarities that make them a part of this chapter. First, rivers
are the primary mechanism responsible for shaping Earth's surface. River and
hill slope processes constitute the central theme of geomorphology [39] . Sec
ond, rivers are the most common high Reynolds number flows !mown. Fluid
mechanics in general, and turbulence research in particular, will never be
740

400
350
300
Blocks
250
200
150
100
50
)' t
0
0.7
0.8
0.9
dP!ds
I.I
1.2
1.3
dPtds
1 .20
-0
Figure 13.17
Basin with random resistance to erosion. (From Ref.
1 6.)
themy if they cannot predict the tightly correlated (reproducible) geometry of

river flow:
l . The river channel tends to acquire a sinusoidal shape (meanders, buck
les) with a wavelength that is proportional to the channel width.
2. The width of the channel is proportional to the maximum depth.
3. The drainage basin [Figs. 1 3 . 1 9 and 1 3 .20 (bottom)] and the delta [Fig.
1 3.20 (top)] have the plane tree features that are now anticipated based
on the area-to-point constructions shown in Section 1 3.2.

11
70
60
K/Kp
11M
50
"'"
II
50
100
11
0. 1 0
40
30
20
IO
Figure 13.18
when
K I K,,
200
200
0. 1 5
0.1
0.001
100
741
300
400
II
500
600
0.05
700
0.00
800
Development of structure in the random-resistance system of Fig. 1 3 . l 7

0. 1 and M increases in steps of O.OO J . (From Ref. 1 6 . )
These simjlarity types have been documented extensively and very suc
cessfully in geophysics: The correlations can be found in all the modern
treatises [32,39-43] . The limited space of this chapter allows me to run only
briefly through statements l to 3 above to show that all these geometrical
features can now be anticipated based on pure theory.
13.3.1
River Meanders
The river meander puzzle of statement 1 is how I started to write about large
scale organization in nature [49) when I proposed a buckling theory for tur
bulent mixing regions. I was captivated by this geometric puzzle, and by
geometry in general, as I was being brought up in Galati, Romania [Fig. 1 3 .20
(top)] . The river channel of cross-sectional area A, mean longitudinal velocity
V, and density p can be viewed as a finite-size column in end-to-end com
pression. The compression force is pAV2. When the Reynolds number VA 1 1 2/
742
Figure
13.19 Development of an artificial 1iver basin over a 1 5 .2 x 9 . 1 m2 rainfall

erosion area. (From Ref. 48; courtesy of Stanley A. Schumm, Colorado State Univer
sity, Fort Collins, CO.)
v is large such that the bulk flow may be modeled as inviscid, the river column
is "elastic" and develops a resistive bending moment. The Euler-type infini
tesimal buckling analysis of the river column leads to the following conclu
sion [49] : The natural shape of the river must be sinusoidal, and the buckling
wavelength A8 must be approximately twice the river width W:
( 1 3.38)
The buckling theory is in very good agreement with field observations on
the incipient formation of river meanders [50] . It is useful that this geometric
feature (A8/ W
2) provides both a tape measure and a stopwatch for con
structing other turbulent flows, not just rivers. The method is reviewed brfofly
in Ref. 14. Most of the applications of the buckling theory of turbulent flow
have come in the field of convective heat transfer, where the list of successful
predictions continues to grow [ 1 4] . For example, the theory anticipates the
transition to turbulence in all flow configurations, the size of the smallest
eddy, the y+ - 10 thickness of the viscous sublayer, the vortex shedding
frequency, the linear growth of all turbulent mixing regions, the Colburn anal
ogy between turbulent heat and momentum transfer, and the frequency of
pulsating fire plumes.
-
13.3.2
River Basins and Deltas
The second puzzle (statement 2), dealing with the proportionality between
river width and maximum depth, is an astonishingly simple geometric feature
743
Figure 13.20
Top: Danube meanders and delta, from Galati to the Black Sea. (From
Landsat.) Bottom: drainage basin near the mouth of the Colorado River, San Felipe,
Mexico. (From U.S. Navy, after Ref. 32.)
that must be taken into account when treating the third puzzle (statment 3),
the dendritic t (treelike) construction of drainage basins and deltas. This con
struction is proposed as an exercise, because it is the two-dimensional (area
to-point) counterpart of the volume-to-point construction developed in Section
1 3 .2. The smallest feature of the flow structure is an elemental system of
known size, where the volumetric flow coexists with the first stream flow.
This observation is especially important in the field of river morphology,
because in that field the elemental system has not been recognized. It is hard
to see such small rivers in nature, or even in artificial rain-erosion simulations
such as Fig. 1 3 . 19. Furthermore, the elemental system is not a feature of the
drainage basin models that are being simulated and optimized nume1ically on
the computer [5 1-53), by assuming a network such as Fig. 1 3 . 1 9 and moving
1
From the Greek noun de11dro11 (tree).
744
the many links (ducts) around until an overall power dissipation figure is
minimjzed.
It is quite relevant that the computer optimizations of drainage basins [5 1 5 3 ] reveal patterns that are strikingly similar to those that occur naturally.
Relevant is also the authors' speculation that some form of global optimization
principle underlies the existence of natural "fractal" structures. The construc
tal law is that very principle.
To illustrate the coexistence of the two flow regimes (volumetric + duct)
at the elemental level, I experimented with coffee sediment as shown in Fig.
1 3 .2 1 . The sediment is easy to prepare, although if you drink unfiltered coffee
Figure
13.21 Coffee grounds visualization of the formation of the elemental areas

(smallest rivers) in the wet coffee sediment or concave surfaces facing upward. Bottom
right: smallest rivers after the rain, on a sloped sandy terrain.
745
you have a distinct advantage. You grind the coffee beans as finely as you
can, add water, and bring the mixture to a boil (a dangerous moment if you
lose yourself in admi1ing the Benard cells!). You want the finest sediment,
which is still emulsified in the coffee, not the coarse sediment that settles
early at the bottom of the pot. Therefore, you wait the first 3 minutes and
decant the liquid, which is still muddy. You wait another 30 minutes until the
second (fi n e) sediment settles at the bottom of your coffee. Pour out the liquid
but leave a small amount in so that the sediment has the consistency of soft
honey or paint. With this thick liquid you wet a concave surface (rotate the
smface to spread the liquid puddle around), hold the smface still (facing
upward), and watch "live" the birth of the elemental areas and the first rivers.
The water flows volumetrically through the layer of coffee grounds that
covers the smface. At the same time, each elemental area develops its first
river, along which the sacrificed grains of sediment are swept downstream.
The merit of this anti-establishment experimental method (anti-establishment
because it has zero cost) is that it works every time. In the first three frames
of Fig. 1 3.2 1 , I show only two of the many surfaces I painted with the same
(recycled) coffee grounds: one funnel (shown twice) and one shallow saucer,
from the center of which I drew the accumulating liquid through a straw. The
simple trees that fo1m at the elemental level are worth noting, especially the
smaller trees visible at mid-radius all around the saucer (on which I used a
runnier coating). Those who still believe in "fractal" rivers should look for
the " smaller and smaller rivers ad injinitum" t in Fig. 1 3.2 1 , or in their own
coffee cups. They will not find them.
The last frame of Fig. 13 . 2 1 shows the pattern formed on sloped sandy
soil right after a torrential rain on the Duke campus. The flow is from left to
right, down an incline of a few degrees. The distance covered by this flow in
the photograph is roughly 0.5 m. I looked for this flow pattern right after a
downpour because I was driven by theory: I wanted to show that the first
(finite-size) iivers exist and that their length scale is associated with the debris
and porous medium through which the volumetric flow (rainfall) flows.
Streams and diffusion, hand in glove. Those who believe in fractal rivers
should try to photograph rivers infinitely smaller than this.
The drainage basin flow in reverse is the river delta; in this case the flow
direction is from one point to a finite area. The water flow rate is fixed,
t
infinite
euclofThef (iadriabnitrdiarmilysense)quencainodn made(eHausof fdfiroranciffttue)r,iexinthgessttsienpscomplis theteinedefit(hreisad:nlinimgeudiistt(aatdteean)mentinfiniimtofyages).frWhenactpralingeomet

ttheedsonequencrypaper; theeisnonlocokut
Thetlhikeefpatcontracttearrnlaralsygweiors titrshueme: iEvermanin natyputulrheian.tgoThir,shssownucoipporntciotsdustencehebyvidoesenawtutrhenotat, andthmeane everrealthyiatmthageinnatg udone(rea.lg.pat, undeshteeetrnrstofhaerepatafpbrealerc,tbyaorl !
sotahnderwscisuleptweure)woulis eucld beidean.seeinThig nots ihsinwhyg butthble uimr ageed imcanagesbeanddistisnhguiadesshedof (gri.ea.y.seen)wr,otbece thausesee
montcommenths lastequesr, thetiosniameng tpoihe n"ft rwaacstamal geomet
de in rthyeofpagesnaturofe" in the by Avnieditrioent alof. this book. A few
strictly
1 997
Science
[54].
746
because it is dictated by the river than arrives near the shoreline. This stream
distributes itself through channels with relatively low resistance, and it finally
seeps volumetrically into the ground with high flow resistance. Superimposed
on this point-to-area flow is the related problem of distributing (again, with
least flow resistance) a fraction of the original stream to the coastal perimeter
of the delta area.
Figure 1 3 .22 shows a remarkably simple simulation of the delta flow. The
two-dimensional flow is created by injecting dyed water into the 75-m-thin
gap between two 1 0 x 10 cm2 glass plates. This apparatus is !mown as a
Hele-Shmv cell. The gap is initially filled with glycerine. The inner surfaces
of the glass plates are roughened artificially with random dots (dimples), each
with a diameter and dot-to-dot distance of order 250 m and a dot depth of
order 22 m. The Hele-Shaw cell is held horizontally, and the water is in
jected through a central hole in the top plate. The dyed water displaces the
more viscous liquid (glycerin) and creates the now familiar pattern predicted
by the solution to the point-to-area access problem (Section 1 3.2).
A river delta [Fig. 1 3 .20 (top)] looks like only a portion-a segment-of
the disks shown in Fig. 1 3 .22, because the smface of the delta is sloped
toward the sea. I have two reasons for exhibiting Fig. 1 3.22 in this section.
First, this simple experiment shows that the water chooses to organize itself
into streams while invading an area characterized volumetrically by high re
sistance (rough plates, viscous glycerin). This geometric arrangement is com
pletely analogous not only to delta flows, but to every other point-to-area or
point-to-volume dendrite found in nature. Streams (dendrites) never occur if
the roles are reversed: that is, if the injected fluid is more viscous than the
displaced fluid. Similar flows can be generated by pouring water from a cup
on a rough horizontal surface (e.g., cement, pavement, asphalt).
The second reason for using Fig. 1 3 .22 as an illustration of optimal point
to-area access is that in the physics literature this flow pattern is widely
recognized under a different name (another class): two-dimensional fingers.
Figure 1 3 .22
Delta dendrites obtained by injecting water into a thin gap occupied by

glycerine. (From Ref. 55; reprinted with permission from Springer-Verlag, Berlin.)
747
This class has been studied exhaustively in physics, where the focus has been
on stability aspects that may control the shape (tips) of the fingers of the
expanding flow. The fingers formed during the growth of bacterial colonies
[56] are an analogous phenomenon, for the reason given above eq. ( 1 3.8):
The urge to organize is an expression of selfish behavior. As with the other
naturally organized phenomena mentioned in this chapter, the contribution of
constructal theory is that it brings under the same tent another example (fin
gering) that has received a lot of attention as an isolated phenomenon.
13.3.3
Electric Discharges
Volumetric electric discharges such as lightning [57] and nerve impulses [58]
are additional examples of natural phenomena anticipated by constructal the
ory. Lightning is a three-dimensional analog of the river drainage basin, ex
cept that it is more obviously a one-shot process (more obviously than the
river, or the tmTent). The electric charge that is initially distributed volumet
rically in a finite air volume is conducted to one point on the volume boundary
(e.g., the roof of a house) : first, by volumetric electric diffusion from air to
the smallest branches of the lightning dendrite, and then as channelled flow
(unidirectional electric conduction) along ionized paths of much higher con
ductivity, which at high levels of assembly become large enough to be visible.
Figure 1 3.23 (top) shows two neural dendrites (dendritic arbors) in a trans
versal plane of the spinal chord of a rat. Similar images are observed during
branched lightning [Fig. 1 3.23 (bottom)] .
We used a heat conduction analog to simulate the generation of tree-shaped
structures during volume-to-point discharge [59]. The volume was a two
dimensional domain of low conductivity k0, as in Fig. 1 3 .9. Its size was fixed,
but its external shape could vary. A fixed fraction of this volume was occupied
by a material of higher conductivity kP . The domain was initially "charged"
(i.e., isothermal at a high temperature T;). At the time t = 0, the midpoint of
the right side of the domain was placed in contact with a heat sink of tem
perature T0. The temperature history of the domain T(x, y, t) was determined
numerically for a large number of configurations of blades distributed over
the k0 background. The global discharge time tc as defined as the time when
the average excess temperature of the domain (T - T0) had dropped to one
tenth of the initial value (T; - T0).
The discharge time tc is a function of the assumed configuration, the ratio
k,,I k0, and the kP volume fraction. We found numerically that t" can be min
imized by selecting the geomet1ic features of the structure of k,, inserts. The
best structures for fastest discharge are shaped as trees; in fact, the best trees
are nearly the same as those obtained for steady conduction in Fig. 1 3 .9. This
conclusion is important because it stresses the primacy of the maximization
of flow access as a mechanism that generates geometry. How the flow varies
748
Figure 13.23
Top: dendritic arbors in a transversal plane of the spinal chord of a rat.

(From Ref. 5 8 ; reprinted with permission from the Society for Neuroscience, Wash
ington, DC.) Bottom: branched lightning over Durham, North Carolina. (Photograph
by Chuck Liddy and Jim Sparks; reprinted with permission from The Herald-Sun,
Durham, NC.)
749
in time (transient, one-shot versus steady state) has little effect on the emerg
ing tree pattern. Drainage basins result from both sudden downpours and
steady drizzles.
13.3.4
Rivers of People
We presented the constructal-theory basis for the emergence of tree patterns

for fluid flow by simulating the generation of dendritic channels formed in
an erodable porous medium (Figs. I 3. I 3 to 13. I 8). Erosion creates the much
needed second flow regime-the path of lower resistance-that enters in a
marriage with the original high-resistance flow and generates flow structure.
River basins are natural structures. The way in which we deduced river
structures from constructal principle can be repeated when we contemplate
other natural flows [5] . An example from the animate world is the movement
of people in a crowd that pushes toward a single opening such as the entrance
to a stadium or the exit from a theater. We all know that in a tight crowd
each person's walk toward the gate is not in a straight line. The flow of people
is not in a quasiradial pattern, as in diffusion from an area to a point sink.
The walk is not straight because it is much easier to step into the space
just vacated by your neighbor, preferably the neighbor who is in front of you
or almost in front. You fi n d the path of least resistance by following the person
or persons who are already moving. Almost imperceptible rivers of people
appear through the crowd that aims for the same goal. The crowd is the river
basin. The space vacated by the person near or ahead of you is the eroded
river bed, the path of low resistance. The conga lines formed by those who
move faster through the crowd are the river branches.
Most of the people fill the spaces between these streams. They march in
place, shifting their weight from one foot to the other, waiting for an opening,
hoping to be carried away by a stream. They are unwitting analogs of the
disorganized molecular motion that accounts for diffusion: the high-resistance
mode that filled the interstices of the fluid trees discussed in this chapter.
What we do in a tight crowd in front of a theater entrance we do every
where in society. With so many of us pursuing the same goals, it is natural
to look for the best way, the shortest route, and the least costly methods to
reach our goals. In this process we form associations, if not rivers of people,
then railroads, highways, airways, and telecommunication networks. Our own
movement and the flow of goods and money describe patterns on Earth's
smface.
The economics of location or spatial (geographic) economics is a wide
arena in which the constructal law has been operating from the beginnings
of humankind and other forms of social animal behavior. These applications
of constructal theory are explored in Refs. 5 and 62.
750
13.3.5
Channel Cross Sections
In this section we examine from the point of view of constructal theory the
second similarity puzzle of river flow at the start of Section 1 3 .3. Why is the
width of a river a certain multiple of its depth? Why should such a simple
geometric proportionality exist?
These questions appear difficult when we think of the many complications
of river flow and channel development. Turbulent flow, secondary flow (cross
circulation), bed erosion, sediment transport, time, and the geological char
acteristics of the terrain all play important roles, which have been amply
recognized in the literature [32,39-43]. These complications are responsible
for significant variations in channel cross section from one river to another,
or along the same river. They can also be invoked when observing that the
ratio between the river width ( W) and the maximum depth (d) is not a uni
versal constant: There is considerable scatter in the WI d data.
The theoretical approach that has been offered as an explanation for the
shape of the channel cross section recognizes from the beginning the com
plicated, developing nature of the river bed (e.g., Refs. 39 and 40). An equi
lib1ium was envisaged between the forces acting on a ground particle at rest
on the bottom of the channel. As noted by Scheidegger [40] , this equilibrium
condition is insufficient and must be complemented by questionable assump
tions concerning the distribution of shear stress (or bottom velocity) with
depth. After numerical integration, this approach yields a bottom profile that
is roughly sinusoidal, in which the river width W and the maximum depth d
are two input (undetermined) constants. In sum, the river bed equilibrium
theories do not explain the scaling law between W and d.
Despite the many complicated phenomena that have been recognized as
diversity and radnomness, one geometric fact stands out: Large (wide) rivers
are deeper than tiny rivers. This begs for a theory that should be as simple
as the scaling law itself,
W-d
( 1 3.39)
Here is the construcal theory of why eq. ( 1 3 .39) should be expected. The
cross section of an open channel has an upper straight segment (the free
surface), which is shear-free, and the rest of the perimeter (the bottom), which
is labeled p and is characterized by shear (Table 1 3 .2). The channel is straight.
The cross-sectional area A and the mass flow rate ni are known constants.
The area A is the result of the slow development of the river bed, and, in this
sense, it describes the age of the river bed. By fixing A we focus our inquiry
on the present, or, in view of the slow development, on the channel as a
system in quasi-steady state.
The mass flow rate is fixed by the global constraint represented by the
rainfall on the drainage smface situated upstream of A. Several important
conclusions follow from the A and 1fi constraints. First the mean flow velocity
751
TABLE 13.2 Optimized Cross-Sectional Shapes of Open Channels

Optimal Shape
c:=J
Rectangle
Triangle
Parabola
Circle
9
0
1-
( W/ d)"P '
11
/Jmin / A 2
2.828
2.828
2.056
2.561
2.507
W ---j
OJ
p
is fixed, V = 1i1./ pA, where p is the water density. Next, the longitudinal shear
stress along the bottom is fixed, T = -1-CfpV-, because in turbulent flow over
a very rough surface the average skin friction coefficient
is a constant [ 1 4] .
Here we are using the nomenclature of duct fluid mechanics (e.g., heat ex
changers) because soon we will make reference to actual duct flows (e.g.,
blood vessels).
The total force per unit of channel length in the flow direction is p T. The
flow resistance pTln"1. decreases in proportion with the bottom perimeter p.
This conclusion can also be expressed in thermodynamic terms. The rate of
work destruction p TV at the longitudinal station where A is located is also
proportional to p. The same holds for the rate of entropy generation in A,
pTVI T0, where T0 is the ambient temperature.
We arrived at a very simple geometric problem that accounts for the ther
modynamic optimization (minimum flow resistance, work destruction, or en
tropy generation) of the river flow through the cross section A :
C1
Find the cross-sectional shape that has the minimum ground perimeter p, which,
along with a straight (free surface) segment, encloses the fixed area A .
Table 1 3.2 shows four steps in the search for the optimal cross-sectional
shape. In each step, the shape of the bottom had to be assumed. In the first
752
example, the bottom shape is rectangular, and its perimeter (p = W + 2d)

can be minimized subject to the area constraint (A = Wd). The result is the
optimal proportionality
w
( 1 3.40)
2d
This is the first instance in which the natural scaling law ( 1 3.39) is derived
based on pure theory. The next three shapes (triangular, parabolic, circular)
reveal the same scaling law. The minimized bottom perimeter decreases
slightly from one example to the next.
Table 1 3 .2 also shows the route to the mathematically optimal channel
cross section. That optimal image has two parts, or two degrees of freedom:
( 1 ) the shape, or bottom profile, which had tcr be assumed in each of the
examples of Table 1 3 .2, and (2) the slenderness ratio W/d, which was opti
mized analytically in each example.
I . The optimal shape of the bottom can be determined based on variational
y(x)
calculus. We assume an arbitrary function

for the bottom profile,
where is the depth corresponding to the coordinate x measured along
the free surface (in the plane of A: see the bottom sketch in Table 1 3 .2).
The objective is to minimize the total length of the bottom perimeter:
1/2
2
[
JW/1
2
)
(
]
dy
p =
-W/2 dx dx
l +
( 1 3.4 1 )
where, for the time being, the width W is considered fixed. The optimal
function
that minimizes the integral ( 1 3.41 ) subject to the cross
sectional area constraint
y(x)
A =
JW/2 y dx
-W/2
( 1 3 .42)
is readily obtained by solving the associated Euler equation (see the

Appendix). The conclusion is that the optimal
is the arc of a circle,
or that A is the segment of a circle such that W is the chord. There is
a family of such solutions, from the very thin slice to the full circle,
depending on how W compares with the length scale A
2. The optimal slenderness ratio WI d of the optimal shape is determined
next by minimizing the ground contact perimeter subject to fixed A.
The result of this operation is the half-circle [WI d = 2 and Pmi n
(27TA)
which is shown as the fourth case in Table 1 3.2.
y(x)
112
1 12] ,
753
In summary, the scaling law between river width and maximum depth can
be anticipated in nakedly simple terms by minimizing the mechanical power
destroyed in the entire cross-sectional area A. In this way, the optimal ge
ometry of the channel is associated with a global optimization subject to
imposed, global constraints (A , n'z) . In this sense, the present theory is in
agreement not only with all the examples covered in this chapter, but also
with the conclusions reached in the global optimization of assumed lattices
of channels in drainage basin models [52,53]. Furthermore, the proportionality
d is supported by recently optimized numerical models of drainage
W
basins [5 1 -53], where it was shown that in every optimized channel both W
and d are proportional to the square root of the flow rate.
The examples aligned in Table 1 3.2 reveal two additional features of the
optimal channel geometry. First, in the optimal shape (half circle) the river
banks extend vertically downward into the water and are likely to crumble
under the influence of erosion (drag on particles) and gravity. This will de
crease the slopes of the river bed near the free surface and, depending on the
bed material, will increase somewhat the slenderness ratio WI d. The important
point is that there remains plenty of room for the equilibrium theories pro
posed in the past [39,40] , in fact, their territory remains intact. The equilib
rium theories begin where constructal theory leaves off.
The second feature revealed by Table 1 3.2 is that in practical, engineering
terms the four designs have nearly the same performance (the same p111;,/
A 1 1 2 value). The optimized rectangular and triangular channels perform iden
tically, whereas the parabolic and semicircular channels agree within 2.2 per
cent. The two most extreme cases are separated by only 1 2 percent, even
though the rectangle is strikingly "rough" relative to the parabola or the
circle. This high level of agreement with regard to performance is very im
portant. It accounts for the significant scatter in the data on river bottom
profiles, if global thermodynamic performance is what matters, not local
shape. Again, this is in agreement with the new work on drainage basins,
where the computer-optimized (randomly generated) network looks like the
many, never identical networks seen in the field.
Yes, there is uncertainty in reproducing the many shapes that we see in
nature, but this is not important. What is important is that there is very little
uncertainty in anticipating global characteristics such as performance and ge
ometric scaling laws (the ratio W/d in this case).
This situation is completely analogous to the highly complex optimization
processes that are performed routinely in the field of engineering design.
Almost as a rule, it turns out that in the end the chosen configuration of the
optimized system performs nearly the same as the truly optimql configuration,
even though, visually, the nearly optimal and optimal configurations may be
different [60, p. 4).
Striking support for this view is provided by the equally wider class of
internal ducts found in organisms: for example, the blood vessels and the
754
pulmonary airways. The fact that every duct is round or nearly round is
correctly attributed to the minimization of flow resistance or power destruction
(e.g., Ref. 61 ) . This principle, however, deserves a second look in view of
the nearly-optimal shapes found for river cross sections.
Two features distinguish the internal duct from the open channel optimized
in this section. First, there is no shear-free segment on the perimeter (W =
0). If the flow is highly turbulent and in the fully rough regime, the solution
is the same as that of paragraph I in the WIA 1 1 2 --+ 0 limit: The optimal shape
is the complete circle. This is why the arms of natural caves traveled by
subterranean rivers tend to have round cross sections. The earthworm cross
section is round for the same reason.
The second feature that may be different in an internal duct is the flow
regime. In laminar flow the thermodynamic optimization is not as simple as
the minimization of p subject to fixed A. The pressure drop (/J.. P) per unit of
duct length (/J..L) is /J..P l/J..L = (2flD")pV2 , where D,, is the hydraulic diameter,
D,, = 4Alp , and f is the friction factor. When the laminar flow is fully de
veloped, the friction factor is inversely proportional to the Reynolds number
(Re = D,, Viv):
f=
Po
Re
( 1 3.43)
where the Poiseuille factor Po depends solely on the shape of the cross section
[ 1 4) . Table 1 3.3 shows several regular polygonal cross sections. The flow
resistance per unit of duct length can be expressed as
( 1 3.44)
TABLE 13.3 Laminm Flow Resistances of Ducts with Regular Polygonal Cross
Sections with Sides"
11
3
4
5
6
8
10
C/J
Po
p !A ' 1 1
2
p Po/A
401 3
1 4.23
14.74
1 5 .054
1 5 .4 1 2
1 5 .60
16
4.559
4
3.8 1 2
3.722
3 .64 1
3.605
2 7Tl / 2
277 . l
227.6
2 1 4. l
208.6
204.3
202.7
20 1 . l
"The Po data are from Ref. 1 4.
755
In conclusion, the flow resistance decreases in proportion with p 2 Po when

the only variable is the shape of the perimeter. Alternatively, p 2 Po is also
proportional to the total mechanical power destroyed in A, or the total rate of
entropy generation in A. The table shows that p decreases as the shape be
comes rounder; however, this effect is tempered by Po which increases in the
same direction.
Even though the round shape is the best, the nearly round shapes perform
almost as well. For example, the relative p 2 Po change from the hexagon to
the circle is only 3.7 percent. Yet the hexagonal ducts have the great advantage
that they can be packed in paiallel into bundles, to fill a volume. Square ducts
also have this packing advantage, and their flow resistance is only 9. 1 percent
greater than that of hexagonal ducts.
In conclusion, the nearly round cross sections are practically as effective
as the perfectly round cross section. Even if the duct cross section is imper
fect-that is, with features such as sharp corners, which concentrate fluid
friction-its performance is already as good as it can be. Furthermore, its
best performance (p2 Po) can be predicted quite accurately when the global
constraints (A, 111) are specified.
13.3.6
Turbulent Flow
In the tree flows of Section 1 3 .2 we saw that the flow has the property to
become organized when it can cover a finite volume by existing in two re
gimes, not one. The two regimes were volumetric diffusion (Darcy flow) and
stream flow. Each flow path contained a portion with high resistance (e.g.,
diffusion) at the elemental level and contained several portions with low re
sistance (streams) at lmger scales.
Since a turbulent flow field is notorious for combining the same two re
gimes-viscous diffusion and streams (eddies)-it should be another example
that is covered by constructal theory. Indeed, turbulent flow was the first
phenomenon in which the time direction of organization and the time equi
partition principle were used in a deterministic manner. In Refs. 14 and 49
the fl o w fi e ld was constructed in time, by starting with the smallest scale and
continuing in steps of geometrically similar building blocks that coalesced
toward larger scales, covering eventually the field.
(a) Buckling and Eddy Formation. The first such example, the two
dimensional shear layer, is reproduced in Figs. 1 3 .24 and 1 3 .25. To proceed
theoretically in time is to predict the flow from left to right in Fig. 1 3 .24,
which is the correct direction of flow development. The theory showed that
the shear layer must always begin with a laminar length, the transition to a
buckling (meandering) shape occurs when the viscous time becomes longer
than the rolling (eddy formation) time, and the flow evolves as a sequence of
756
r.,______

fuw
Laminar
_..,____ ____ Buckling and rolling

__
_
__
fuw
u_
,. -
.....
- - - - -- - - -
u_
-- - -- - -- D
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _
----- L o-----
(a)
Figure 13.24
- - -I
I
- --::::... I
:.::==
:
::
:
jr
-::=.- -
- - ::. -
_ _
- -
First_
roll
1
1
(b)
-- Etc.
- - - --:::.. -= - ::::: :..
- - -
L-
Seco
11nd- ro
Laminar length at the start (tip) of a shear layer. (From Ref. 49.)
geometrically similar assemblies (eddies) the sizes of which increase stepwise

(Fig. 1 3.25). Note that Fig. 1 3.24 is a close-up view of the starting section
(the tip) of Fig. 1 3.25.
The laminar tip of the shear layer is the "elemental system" that repeats
itself many times in the downstream direction. This is how the flow (shearing
effect) is able to cover every point of the volume called the mixing region.
We see this better if we view the time development of the shear layer in a
frame that rides at Uj2 from left to right in Fig. 1 3.24. What we see is
Half-angle o f visual growth: arctan
Lo ca I t h ick ness
(f--.!. D )
2 t..B
- - - - -,
---1-----i..cio,-=I
- -- - - - - , - - - +-- - "'"....::;:
...
J_
-
Virtual
orig i n
Local wavelength
--
Knee
T i m e-averaged
velocity profile
Figure
13.25 Constant-angle growth of a shear layer as the repeated coalescence of

geometrically similar building blocks. (From Ref. 1 4.)
757
sketched in Fig. 1 3 .26: The upper fluid moves to the right at U,)2, and the
lower fluid moves to the left with the same speed.
Immediately after t = 0, the horizontal midplane is occupied by a one
dimensional laminar shear layer. The instantaneous thickness of this layer (D)
increases as t 1 1 2 , and its order of magnitude is given by the scaling solution
to the momentum diffusion problem (cf. Ref. 24, p. 662, or Ref. 14, p. 3 1 6):
D/2
- 0( 1 )
2(vt) 1 1 2
( 1 3.45)
As in Section 1 3 .2, the question is one of optimal (fastest) access: How can
the flow cover the space in the shortest time possible? Spatial growth in the
frame of Fig. 1 3 .26 means only up and down, because the configuration is
one-dimensional.
The laminar regime is effective only in the beginning, when the time de
rivative dD I dt is large. As time increases, the laminar swelling of the flow
slows down monotonically, because dDI dt decreases as c 1 1 2 . The flow has
access to the vertical direction via a second regime, which is better known
as eddy formation, or buckling and roll-up ( 1 4,49] . The vertical velocity
.!. u
2
D
_
eq. ( 1 3.45)
0 "---
0
(a)
Figure 1 3.26 Stepwise expansion of the flow of Figs. 1 3 .24 and 1 3.25 when the
observer travels to the right with the velocity Ux/ 2 .
758
dD/dt along this second route is constant ( U,)2), because the peripheral speed
of the roll-up is set by the upper and lower fluid reservoirs.
How far can the shear flow extend by buckling and rolling up once? The
buckling wavelength As - 2D [cf. eq. ( 1 3.38)] dictates the spacing between
two adjacent eddies, hence the diameter of the roll formed out of the flow of
thickness D. The thickness increases from D to 2D during the roll-up time:
ts -
4D
As
-Uoc
Ux/2
--
(l 3.46)
The rate of lateral expansion through eddy formation is (2D - D) Its - Uj

4, whereas the viscous swelling rate indicated by eq. ( 1 3.45) is dD/dt 2( vlt) 1 1 2 - 8v/ D. The most rapid expansion occurs when the viscous regime
is followed by eddy formation, not the other way around. This happens when
the viscous growth rate is just outpaced by the eddy growth rate: that is, when
D U,,,/ v - 32, or
( 1 3.47)
Equation ( 1 3 .47) is the local Reynolds number criterion, which was found to
predict and correlate the transition to turbulence in all known configurations
(e.g., Refs. 14 and 24). As in the original presentation of this theory [49] , the
same criterion can be obtained by equating the time scale of viscous diffusion
over the distance D [cf. eq. ( 1 3 .45)] with the time scale of eddy travel over
the same distance [cf. eq. ( 1 3 .46)].
In Fig. 13 .26 the size of the first eddy is labeled D 1 In the preceding
analysis this length scale was called 2D (see Fig. 1 3.26a). The order of mag
nitude of D, is given by the local Reynolds number criterion ( 1 3 .47): The
smallest eddy in a turbulent flow field-the size of the elemental system-is
such that its Reynolds number is of order 1 02 . This result of the theory con
tradicts the established view in fluid mechanics, where the smallest eddy is
thought to have a Reynolds number of order 1 . The experimental record is
strongly in favor of eq. ( 1 3.47) ; just measure the photographs showing the
first eddies that occur in the wake behind a cylinder in cross-flow.
Beyond the first eddy formation event (DJ, the flow continues to expand
in steps, by rolling and forming eddies. Each step leads to the doubling of
the mixing region [cf. As - 2D above eq. ( 1 3 .46)] . Viscous diffusion does
not have time to act-to compete-over larger distances such as D 2 , D3 , and
so on. The outer boundaries of the mixing region are extended now by the
second flow regime: streams (eddies). The analogy with the other structures
deduced based on constructal theory in this chapter is now complete.
The eddy doubling sequence sketched in Figs. 1 3.25 and 1 3 .26 cannot
continue in two dimensions indefinitely. The reason is once again the optimal-
759
access principle that guided us up to this point. Two-dimensional eddies

would be curved two-dimensional jets; the infinitely flat cross sections of such
jets are far from having the most efficient shape, which was discovered in
Section 1 3.3.5. Consequently, what starts out as a two-dimensional curved jet
is destined to become a three-dimensional flow by organizing itself into sev
eral parallel jets with nearly round cross sections (for the same reason that
delta branches form in Fig. 1 3.22). The subsequent constructal evolution of
each round jet can be traced in the same way as the shear layer; this con
struction is reported for round jets and round plumes in Ref. 1 4.
(b) Laminar Lengths. A convincing manifestation of the maximization of

mixing [eq. ( 1 3.47)] is the universal proportionality between the lengths of
laminar tip regions and their cotTesponding buckling wavelengths. Figure
1 3 .27 reproduces Gore et al.'s [63] compilation of the laminar lengths of the
starting regions of all known laminar flows that undergo transition to turbu
lence. The transition length L, r is labeled L0 in Fig. 1 3 .24, and the buckling
first wavelength A8 would be a longitudinal wave of length 2D0 deforming
the layer of thickness D11 (this deformation becomes the "first roll"). Figure
1 3 .27 shows that the following proportionality is universal:
0.1 -
a buoyant plume
<> sudden expansion
o buoyant and nonbuoyant water jets
<J water jet issuing into air
I> cavity flow
11 flow over spinning projectile
O plane mixing layer
t::. confined coaxial jets
0 flow over flat plate
0.01 i-
. -:-
___,'--.<-L.....L...L
.L J..LI._-.L--,_.__,_.-'-':'"
..L..1.
t_
0.001 i___,___._.1.-L...ul
0.1
O.o1
0.001
0.0001
"'-a (m)
Figure
13.27 Proportionality between the length of the laminar section and the buck
ling wavelength. (After Ref. 63.)
760
( 1 3 .48)
In Ref. 14 I showed that this proportionality is a consequence of the construc
tal law [eq. ( 1 3.47)] and AB - 2D, where U and D are the local (near
transition) velocity and thickness scales. For example, if the flow is a laminar
boundary layer on a flat plate, the boundary layer thickness at transition is
D - cL".
( )-J/7
UX L,,
( 1 3.49)
--
where c is a factor of order 2 or 3, depending on how the boundary layer

thickness is defined. By eliminating Uoc and D between eqs. ( 1 3 .47) and
( 1 3 .49) and AB - 2D, we obtain
( 1 3 .50)
which anticipates the empirical correlation ( 1 3.48). The sources for the data
compiled in Fig. 1 3.27 are indicated in Ref. 63.
(c) Benard Convection. The most talked about example of pattern formation
in turbulent flow is Benard convection: the rolls formed in a fluid layer heated
from below, when the heating rate or the layer thickness is large enough. In
the second edition of this book I showed that the onset of Benard convection
is just another application of the coexistence of two heat transfer mechanisms
at the elemental-system level-that is, another example of the constructal law
of maximization of access. In a subsequent paper [64] we showed that all the
features of Benard convection, not just the onset, are anticipated by the con
structal law. We showed this for Benard convection in fluids and in fluid
saturated porous media. These developments are also detailed in Ref. 5 .
Consider the one-dimensional fluid layer shown i n Fig. 1 3 .28. The thick
ness is H, and the bottom excess temperature is 6.T = T,,
Tc. The time
needed by this heating effect to travel by thermal diffusion the distance is
-
t0
H2
4a
( 1 3 .5 1 )
where a i s the thermal diffusivity of the fluid. Equation ( 1 3.5 1 ) comes from
eq. ( 1 3 .45), in which we substituted H for D/2 and a for v. Thermal diffusion
continues to be the preferred heat transfer mechanism, and the fluid layer
remains motionless, as long as H is small enough such that t0 is the shortest
time of transporting heat across the layer.
The alternative to diffusion is convection: the channeling of energy trans
port as fluid streams, which act as conveyor belts (rolls, Fig. 1 3 .28, right
11
T
1
H
Quiescent,
thennally stratified
fluid
------+----
-----
Ra
Nu
<
761
Cellular
flow
pattern
Cooled
)000(
HeatedJ
Ra > 1 708
H
1 708
Nu
= I
> I
Figure 13.28 Horizontal fluid layer held between two parallel walls and heated from
below. (From Ref. 24.)
side). The question is whether the convection time (t,) across the layer H is
shorter than t0. The convection time is t1 - HIv, where v is the vertical
velocity of the fluid (the peripheral velocity of the roll). To evaluate the v
scale we rely on scale analysis (e.g., Ref. 1 4, p. 6 1 8). First, we note that the
effective diameter of each roll is of order H, but smaller (e.g., H12). When
the roll turns, an excess temperature of order t:iT/2 is created between the
moving stream and the average temperature of the fluid layer. This excess
temperature induces a buoyancy effect (modified gravitational acceleration)
of order gf3 t:iT/2 where f3 is the volumetric coefficient of thermal expansion.
The buoyancy force that drives the roll is of order (g/3 t:iT/2) p(H12) 2 . When
the Prandtl number is of order 1 or greater [ 1 4] . the driving force is balanced
by the viscous shearing force TH12, where the shear stress scale is T fLV I
(H/4). The force balance described until now (buoyancy - friction) yields
the velocity scale v - gf3 t:iTJfl/ 1 61/, and the coITesponding convection time
scale
-
t -
16v
g/3 t:iTH
( 1 3 .52)
---
As H increases, the thermal diffusion time t0 increases in accelerated fash

ion [eq. ( 1 3.5 1 )] while the convection time t1 (a property of the H system
even if quiescent) decreases monotonically [cf. eq. ( 1 3.52)]. A critical size H
distinguishes the system in which t, drops below t0. Setting t1 ::s t0 and using
eqs. ( 1 3.5 1 ) -( 1 3.52), we find that the first streams occur when Ra
0( 1 02)
where Ra = g/3 !:iTH3/ a v is known as the Rayleigh number. The exact so
lution for this critical condition is Ra = 1 708 [ 1 4]; in other words, Ra =
-
762
0( 1 03). The one order of magnitude error in the scale-analysis prediction can
be attributed to the imprecise factors of order 1 (geometric ratios) introduced
along the argument made above eq. ( 1 3 .52). It is important, however, that the
critical Ra is a constant considerably greater than 1 ; this constant is a con
glomerate of all the geometric ratios of the roll-between-plates configuration.
The same agglomeration of geometric information is responsible for the con
stant of order 1 02 in the local Reynolds number criterion ( 1 3 .47). Had we
neglected the geometric reality of how the rolls fit, or the geometric fact that
4 belongs in the denominator of eq. ( 1 3 .5 1 ), we would have obtained only
Ra - I -that is, the correct dimensionless group but not its geometric critical
value.
Each roll characterized by t0 - t 1 is an elemental system in the same sense
that this concept was used in Section 1 3.2. Thermal diffusion is present (and
does its job) at every point inside the elemental volume (e.g., H X H in Fig.
1 3 .28). An optimal distribution of convection "streets" guides this diffusion
faster across H. The more cmrunon language for "faster" in the field of heat
transfer is to say that the onset of convection is followed by an increase in
the overall Nusselt number Nu11 = q" HI k b,.T, where q" is the wall-averaged
heat flux. So far, in constructal theory we have been fixing the current: If we
fix q" in Benard convection, we see once again that the optimization of the
heat flow_Qattern at the elemental level leads to a smaller overall b,.T-hence
a larger Nu11
As the system size H increases above the critical value, a time imbalance
(t0 > t 1 ) would develop in the elemental system if its shape does not change.
This is why the rolls become "tall" as their number increases. Narrower rolls
means narrower streams, and shorter thermal diffusion times between streams
and walls and between adjacent streams. The proliferation and narrowing of
Benard cells are well known, and can be predicted with the methods of con
vective heat transfer (e.g., the intersection of asymptotes method [ 1 4,64]). So
can all the heat and fluid flow patterns of Sections 1 3.2 to 1 3 .6. The important
point-the reason why we discuss Benard convection in this thermodynamics
book-is to show that yet another natural pattern (well known, in isolation)
finds a place under the wide deterministic umbrella of constructal theory.
13.3.7
Cracks in Shrinking Solids
Wet soil exposed to the sun and the wind becomes drier, shrinks superficially,
and develops a network of cracks (Fig. 1 3.29). The loop in the network has
a characteristic length scale. The loop is round, more like a hexagon or a
square, not slender. The loop is smaller when the wind blows harder: that is,
when the drying rate is higher.
These unexplained characteristics of mud cracks are major hints that their
pattern is another natural occmTence of access optimization: (a) the maxi
mization of the mass transfer rate from the system (wet soil) to the ambient
Figure 13.29
763
Patterns of cracks on the surface of a Duke baseball field.
or (b) the minimization of the overall drying time. In view of the analogy
between mass transfer and heat transfer, we can explore this theoretical route
by considering the thermal analog sketched in Fig. 1 3 .30. Consider a one
dimensional solid slab of thickness L, which is initially at the high tempera
ture TH and has the property to shrink upon cooling. The coolant is a
single-phase fluid of temperature Tc The question is how to maximize the
thermal contact between the solid and the fluid or how to minimize the overall
cooling time.
In Fig. 1 3 .30 the cracks are spaced uniformly, but their spacing (R) is
arbitrary at this point. The channel width (D) increases in time, as each solid
piece (R) shrinks. The fluid is driven by the pressure difference /J..P, which is
Solid
T11, p,
l
Figure 13.30
c,
T( t)
Coolant ril ' , Tv
!J.P , Pr, cp, kr
Channels in a solid cooled by a single-phase fluid.
764
maintained across the solid thickness L. The imposed !1P is an essential aspect
of the channel spacing selection mechanism. For example, in the air cooling
of a hot solid layer the scale of !1P is set at +_pf U, where Pt and U'" are the
density and free-stream velocity of the external airflow (the wind), which
sweeps the cracked surface from above (i.e., parallel to the plane of Fig.
1 3 .30.)
To examine the effect of the channel spacing R on the time needed for
cooling the solid, we consider the two extremes R __. 0 and R __. 00, and then
use the intersection of asymptotes [ 14] . The approach is the same as in the
geometric optimization of the cooling arrangement of electronic packages and
assemblies [5, 1 4] . In other words, computers emerge as patterns of heat
generating blocks separated by optimal cooling channels for the same reason
that optimal patterns of cracks occur in Fig. 1 3 .29. This observation reinforces
the commonality of natural and man-made patterns under the constructal the
ory umbrella.
When the number of channels per unit length is large, the spacing R is
small and so is the shrinkage experienced by each R element. This means
that in the limit R __. 0 we can expect D __. 0. In this limit, the flow through
each D-thin channel is laminar (Hagen-Poiseuille), such that the channel mass
flowrate is given by [ 14]
( 1 3.53)
In the same limit, R is small enough so that the solid conduction is described
by the lumped thermal capacitance model [24] . The solid piece R is charac
terized by a single temperature T, which decreases in time from the initial
level TH to the inlet temperature of the fluid, TL . This cooling effect is gov
erned by the energy balance
pcRL
dT
= - q'
dt
( 1 3 .54)
where p and c are the density and specific heat of the solid. The cooling effect
(q ' ) provided by the flow through the channel is represented well by
( 1 3.55)
where cP is the specific heat of the coolant. Equation ( 1 3 .55) means that in
the D __. 0 limit the fluid becomes as warm as the surrounding solid before
it reaches the end of the channel. If we combine eqs. ( 1 3 .54) and ( 1 3 .55), we
obtain
!J.T
pcRL - - n1 ' cp !J.T
t
765
( 1 3.56)
where !J.T is the scale of the instantaneous solid excess temperature T - TL .

Finally, by using the ni ' scale provided by eq. ( 1 3 .53), we conclude that the
cooling time scale is
pc .RU
t - -- --PrCp D 3 !J.P
(R __, 0)
( 1 3 .57)
In the opposite limit, R is large and the shrinkage (D) is potentially very
large-in proportion to R. The fluid present at one time in the channel is
mainly isothermal at the inlet temperature TL . The cooling of each solid side
of the crack is ruled by one-dimensional thermal diffusion into a semi-infinite
medium [24] . The cooling time is the same as the time of thermal diffusion
(penetration) over the distance R,
(R __, oo )
( 1 3.58)
where a = kl pc, and k is the thermal conductivity of the solid.

To summarize, eq. ( 1 3 .57) showed that in the limit R
0 the cooling time
is proportional to RI D 3 or R -2 , because we expect a proportionality between
D and R:
__,
D
- - f3 !J.T 1
R
( 1 3.59)
where !J.T - TH - TL and f3 is the coefficient of thermal contraction of the

solid. In the opposite limit [eq. ( 1 3 .5 8)] the cooling time is proportional to
R2 Put together, these proportionalities suggest that the cooling time pos
sesses a sharp minimum with respect to R or the channel density. Intersecting
eqs. ( 1 3 .57) and ( 1 3 .58) we find that the optimal crack distance (Rapt) for
fastest cooling is
( 1 3.60)
This result is promising for two fundamental reasons. One reason is that
the optimal crack distance decreases as the external pressure (or flow) is
intensified. This effect is in accord with observations (unexplained before
constructal theory) that mud cracks become denser when the wind speed
increases. The second reason is that eq. ( 1 3.60) predicts a higher density of
766
cracks (a smaller Rapt) as the solid excess temperature D.T increases. This
trend, too, is in agreement with observations.
An important geometric implication of eq. ( 1 3 .60) is that the optimal dis
tance between consecutive cracks must increase as L 1 1 2 This result is relevant
to predicting the length scale of the lattice of vertical cracks formed in a
horizontal two-dimensional surface cooled (or dried) from above, under the
influence of external forced convection (e.g., wind). As the airflow direction
changes locally from time to time, and since the material (its graininess) is
such that cracks may propagate in more than one direction, we arrive at the
problem of cooling a two-dimensional te1rnin (area A, when seen from above)
with cracks of length L and associated area elements of width Rapt "
Figure 1 3 .3 1 shows the two extremes in which L may find itself in relation
to R0P" When L is considerably shorter than Rapt (Fig. 1 3.3 l a) it is impossible
to cover the area A exclusively with patches of size L X Rapt " When two
cracks of length L are joined at an angle, the elemental area ( -L2) trapped
between them is too small to accommodate the amount of ideally cooled solid
material.
When L is considerably longer than Rapt (Fig. 1 3.3 1 b ) , any lattice of cracks
will fail to cover the area A completely. The trapped elemental area (-L2 ) is
considerably larger than the amount of ideally cooled solid ( -LRapt) . This
means that most of the interior of the area element of size L2 would require
a cooling time that is considerably longer than the minimum time determined
in eq. ( 1 3.60).
In conclusion, to cool the entire solid (A) in the fastest way possible is to
cover the A cross section with L x Rapt elements in which L - Rapt " In other
Figure 13.31
(a)
L << Rapt
L >> Rap t
(b)
The two extremes in covering a two-dimensional solid (A) with cracks

(L) and optimally cooled solid elements (Rapt) .
767
words, the optimal pattern is one with "round" loops, not slender loops.
Combining L - R0", with eq. ( 1 3 .60), we find the optimal length scale of the
loop in the network of cracks that will minimize the cooldown time:
Ro"' -
(o:, v klk ,) 1 1 2
U,,(/3 IJ.. T) 31 2
( 1 3 .6 1 )
Once again, in agreement with observations, we see that the lattice length
scale R0"1 must decrease as the wind speed and the initial excess temperature
increase. The constructal theory of crack patterns was pursued further (nu
merically) in Ref. 65 .
13.3.8
Dendritic Crystals
In 1 6 1 1 , Kepler [66] drew attention to the shapes, numbers, and geometric

similarities exhibited by snowflakes. The study of dendritic crystals has grown
into a major field that can be reviewed by reading Ref. 67. This work deals
mainly with two aspects: the shape and the growth of dendritic crystals. In
the 1 997 edition of this book I provided a constructal-law explanation for the
origin of the dendrite. I showed that treelike needles are a necessary archi
tecture so that the rate of solid formation is maximized. Dendrites accelerate
the approach of the subcooled liquid to equilibrium.
Consider the solidification of a single-component substance with the solid
ification point T,, and assume that the liquid and solid phases have the same
density p. The solid phase is isothermal at T,. Sufficiently far from the solid
ification front the liquid is metastable (subcooled) at the temperature T00 ( <TJ
The specific heat of the liquid is constant, c. The solid is symmetric about
the axis x.
The uniform density assumption means that the solidification process does
not induce liquid motion in the vicinity of the solidification front. Addition
ally, we assume that the shape of the solid part is sufficiently slender (e.g., R
L in Fig. 1 3.32) such that longitudinal conduction in the x direction through
the liquid can be neglected. This slenderness assumption is the basis for the
following boundary layer type of argument [ 1 4] . The phase-change heat trans
fer process is described approximately by the classical solution for the infi
nitely long solid cylinder [68, p. 295] :
R( t)
T(r, t) - T,,,
T, - T,,,
= 2A( o:t) 1 1 2
Ei( - r2 I 4o:t)
Ei(- A2)
( 1 3 . 62 )
(r
2::
R)
( 1 3 .63)
where R(t) is the radius of the solidified cylinder, t = 0 is the start of the
solidification process, and T(r, t) is the temperature in the l iquid of thermal
768
I;
r
R;
.r
xi
__J
Figure
13.32 Axially symmetric needle generated by the conduction solution for a

solidifying cylinder that moves to the left.
conductivity k and thermal diffusivity a = kl pc. The cylindrical geometry to

which eqs. ( 1 3 .62) and ( 1 3 .63) refer is visualized in the x = L cut through
the solid shown in Fig. 1 3.32. Ei is the exponential integral function
EI(w) =
Ix
II'
exp(- y)
dy
y
( 1 3.64)
while in eq. ( 1 3 .62), A is a monotonic function of the liquid subcooling Stefan

number [Ste = c(T, - T.,,) I h,1] :
2
2
A exp(A )Ei( - A2) + Ste = 0
( 1 3 .65)
The body of revolution geometry shown in Fig. 1 3 .32 can be constructed

by sandwiching together a large number of cylindrical solutions of type
( 1 3.62)-( 1 3 .65) in such a way that the slice thickness L'.lx is proportional to
the time increment. This is equivalent to imagining that the cylindrical solu
tion moves along the x axis with a constant velocity ( U), so that
L = Ut
( 1 3 .66)
The temperature distribution in any intermediate plane x = X; is obtained by

replacing t with (t - t;) in eqs. ( 1 3 .62) and ( 1 3.63), where according to the
time line drawn parallel to x in Fig. 1 3.32,
- X;
= Ut;
( 1 3.67)
The temperature field in the x = X; plane is approximated by the cylindrical

temperature field of a solidification process that started at t = t;. Solidification
at the tip (x = 0) starts at the time t.
'
769
In the (x, r) frame attached to the tip of the solid, the liquid and the solid
move to the left with the velocity U. Combining eqs. ( 1 3 .62), ( 1 3 .66), and
( 1 3 .67), we find that the solid has a parabolic profile,
R;(x;) = 2A
( a J) 1 /2
x.
( 1 3 .68)
in which the generating plane is represented by R = R;(L). The slenderness

assumption that must be satisfied by this boundary layer solution depends on
the assumed (unspecified) velocity U:
( a ) 1 12
R
- = 2A
L
LU
-
( a) 1 1 2
2A
=
U t
-
( 1 3 .69)
The validity of the slenderness assumption improves in the downstream di

rection (as t increases), and as the propagation speed U increases.
Figure 1 3 .33 is a geometric summary of the conclusions reached up to this
point. The two needles are drawn perpendicularly for clarity. We regard U as
an undetermined constant. The radial thickness of the liquid heated around
the solid scales as r1 - 2( at)
[cf. the argument of the function Ei in eq.
( 1 3 .63)] . The thickness of the solid, R(t), has the same time dependence [cf.
eq. ( 1 3 .62)] . In conclusion, the two thicknesses are proportional, and the
proportionality factor is constant when the Stefan number is specified:
1 12
( 1 3 .70)
/I\
.'.l
L
Figure
time.
13.33
1------
1
Time -
.'.l
L
Region occupied by the solidification process, and its evolution
' l
770
Because of the time independence of the ratio

r 1 , in Fig. 1 3.33 we can
represent the transversal dimension of the temperature field using just one
time-dependent length scale for both solid and warmed liquid. The geometric
aspect ratio of the temperature field
L) decreases as the time increases [cf.
eq. ( 1 3 .69)] . In other words, although both R and L increase in time, the
region penetrated by solidification becomes more slender: This shape fills the
plane Jess well than in the early stages of the solidification process.
RI
(RI
1 . Why plane ? Kepler asked the first fundamental question about the ge
ometry of the snowflake: Why six? This question was answered after the
geometry of the molecular arrangement of ice became known. An even more
important question is this: Why is the snowflake plane? This question cuts to
the heart of the meaning of rapid solidification.
The classical solutions for unidirectional solidification in plane, cylindrical,
and spherical geometries [68] show that in all cases the liquid and solid
thicknesses increase as (at) 1 1 2 , and that as in eq. ( 1 3 .70), their ratio is a
constant. This point is made in Fig. 1 3 . 34, where the horizontal alignment
should be noted. The three geometries are quite different when it comes to
their ability to fill the space with solid:
solid volume
total volume
B._ <
()2
r1
(R ) '
plane
<<
cylinder
<<<
( 1 3.7 1 )
sphere
The most effective arrangement for solidifying a volume e n masse i s the

planar one. That the snowflake is plane is overwhelming evidence in support
of the constructal law: the maximization of the speed of volumetric solidifi
cation is an integral part of the nature of the process.
Plants (vegetation) are completely analogous to snowflakes. They occur
and survive because they facilitate "rapid" ground-air mass transfer.
----------1--
-{ _ J
t
/'
I
I
I
/ -.... --------------,
B +'1 --- - 3
'
\
I
___ _ _ _ _ _ _ _ _ _ _ _ _ _ __,
Plane
Figure 13.34
Configuration
/
I
I
Cylinder
for
I
I
I
\I
'
A \
\jj
,,.. ,,,. ..... -
- ..... ...... ,
' .....
,,,. ,,,
'
I
I
Sphere
the volumetric solidification of a subcooled liquid.
771
2. Why Branches ? The most important problem is that of predicting the

dendrite architecture. Let us follow, in time, the growth of the snowflake
beginning with its birth (t = 0) at a point-size nucleation site (Fig. 1 3.35).
Molecular structure and surface tension-based stability arguments indicate that
ice needles will begin to grow with the velocity U (constant) in the six di
rections shown. But this is only one aspect of the solidification process that
just started. At the same time (t = 0), the temperature of the nucleation site
jumped to T,(> TJ and triggered a spherical wave of thermal diffusion
2(at) 1 1 2 . The initial speed of prop(warmth) whose radius increases as r 1
agation of the liquid heating effect is infinite: that is, larger than any constant
speed (U) that the tip of the needle might have. In time, the speed drJdt
decreases at t - 1 1 2 and is eventually overtaken by U.
Figure 1 3 .36 shows how the warmed liquid sphere and the needle length
grow in time. A critical time tc is reached when the needle length L overtakes
the radial length scale of the warmed liquid:
( 1 3.72)
This equation yields
( 1 3 .73)
Note that Fig. 1 3 .36a is analogous to Fig. 1 3 .26a. At times slightly greater
than t,, the needle of length Vt, just sticks its tip out of the warmed liquid
sphere. The tip is once again surrounded by isothermal subcooled liquid from
Figure
13.35 Formation of new needles after each time interval t,. as the repeated
manifestation of the mechanism shown in Fig. 1 3 .36.
772
Radial
L = Vt
Needle tip
growth
/
/- - - - -
1 '
0
Figure 13.36
Warmed 1Iqu1d sphere

,. , 2(an1 "
,'
,'
,'
fc
(a)
(b)
Simultaneous growth of the needle and the warmed liquid sphere trig
gered by its tip, and the time interval t,. after which the process is repeated.
the half-space that lies in front of it. Therefore, the situation at t - t, is the
same as that at t = 0, except that the new nucleation site (the needle tip) can
send new needles only forward (note that the trailing half space is already
warm and/or solidified). In conclusion, at t - t, each tip serves as nucleation
site for three forward-leaning fresh needles. There is no difference between
the ages of these fresh needles. Because of the 60 angular symmetry, how
ever, the middle needle (b) in Fig. 1 3 .35 looks l ike a continuation of the
original needle, and consequently, the other two needles (a, c) look like
branches.
From t - t, until t - 2tc, the new generation of needles experiences the
"growth inside the warm liquid sphere" process that we saw between t = 0
and t = tc. One difference is that the liquid spheres of the side needles (a, c)
interfere eventually with the spheres of the side needles of the adjacent orig
inal directions. At t - 2tc all the side needles become "suffocated" by the
warm (saturated) liquid environment, and their needlelike growth ceases. Each
middle needle (b) however, pierces its warmed liquid sphere and generates
another group of three forward-leaning fresh needles. This new (third) gen
eration and its fully grown version at t - 3tc are i llustrated in Fig. 1 3.35
which is based on repeating the argument of Fig. 1 3 .35a three times as shown
in Fig. 1 3 .36b.
With this, the task of predicting the existence of dendrites has been ac
complished. Left undetermined is the length scale of the fully grown needle
(age tJ-undetermined because U is undetermined: L, - U( - 4a/ U.
Imagine that you are someone who knows biology but not thermophysics.
You may desc1ibe Figs. 1 3.35 and 1 3.36 as follows. The organism is born at
773
the time t = 0, and its morphology (e.g., the number 6) is a reflection of

information stored at the molecular level. The organism grows: that is, ex
pands into its surroundings. The growth is very fast when the organism is
young, and it slows down with age. There comes a time-the time of death,
(.-when the organism becomes disposable. Continuity is assured in the form
of several offspring, which repeat the life cycle of the original organism. The
offspring may or may not be attached to their ancestor. In sum, the periodic
regeneration and multiplication of the organism is an expression of the natural
tendency toward optimal (or fastest) flow access, which drives all the organ
ized phenomena covered in this chapter.
"Can tree-shaped structures exist without flow through them?" [69]. This
is a most basic question that we must consider as we bring both fluid and
solid structures under the big tent of constructal theory. In the case of the
dendritic crystal, the solid is not moving through the needles. What flows is
the heat generated by the solidification process, and this "river delta" of heat
(cf. Fig. 1 3 .22) flows along two paths: diffusion through the liquid, perpen
dicular to the needle, and organized movement (convection), as the needle
grows longitudinally. When thermal equilibrium is reached, the flow of heat
ends, the once living tree of heat dies, and the solid dendrite remains. The
solid is a fossil, a reminder of the living flow system that once existed, like
the skeleton of an animal, and like a dry river basin in the desert.
13.3.9
Solid Bodies in Flow
As an invitation in yet another new direction, let us ask why solid matter
travels organized through a flowing medium. Why does it not travel as a
swarm of small granules? Why do celestial bodies form in time?
Consider two such granules: for example, two balls of diameter D 1 and
density p, such that their total mass is 111 = 2p(7T!6) Df. The drag force felt
by each ball is FD = CD (7T/4) DTtP/P, where Pr is the density of the flowing
medium that fills the space, and U is the relative velocity between ball and
medium. For simplicity, assume that the Reynolds number is sufficiently large
such that the drag coefficient CD is a constant of order I . The drag force
experienced by the total mass 111 is
( 1 3 .74)
Can the two masses reduce their resistance to travel? In other words, can
the solid spread faster and farther through space? Yes; two balls fused into
one larger ball encounter a smaller resistance than when they travel separately.
Mass conservation dictates that the diameter of the larger ball is D2
2 1 1 3D 1 The drag force on this larger ball,
774
( 1 3 .75)
is sensibly smaller than in the original configuration, F2 1 F 1 = 2- 13 = 0.794.

Given enough time, two masses will coalesce into a larger mass, in a process
that repeats itself many times. We see it everywhere. In outer space we see
it in astronomically slow motion. This process is also visible during large
scale explosions. The debris does not expand uniformly (spherically) outward;
rather, it comes together into several radial streamers.
The tendency to coalesce and travel in formation is particularly visible in
living systems. Migratory birds fly in large groups organized in precise pat
terns. The same external organization of movement is visible on the ground,
in team bicycle racing, and in the slipstream shed by the lead race car. The
principle is the same: Aerodynamic drag decreases when individual masses
coalesce, and in this way mass travels faster and farther. Strings of racers on
the ground and flying carpets (birds, people) in the air-they are all driven
by the same principle.
Processions are visible everywhere. Schools of fi s h are beautiful because
they display precise patterns of organization. The synchronized swimming
and jumping of dolphins are also due to spatial organization. Ducklings pad
dle behind their mother arranged in a pattern, like pieces on a chessboard.
Biologists and fluid mechanicists have long recognized the geometric re
lation between a patterned procession and the regular occurrence of vortices
in the wake shed by the body that leads the procession. They are right, of
course, but not in the general sense that is meant in constructal theory. It is
the inanimate fluid in the wake of the leading body that organize itself. It
does so using no brain power whatsoever, so that it may travel and spread
itself the fastest through the stationary fluid.
The fish and the ducklings, for their own power-minimization (constructal)
reasons, act as markers in the flow. They feel the flow, and they follow it. In
this way they visualize the constructal law that accounts not only for the
eddies and turbulence in the fluid (Section 1 3 .3 .6), but also for the animals'
urge (tendency) to advance bodily and think most easily (Section 1 3 .9). We
return to the oneness of eddy and animal movement in Section 1 3.8.
13.4
CONSTRUCTAL THEORY OF DISTRIBUTION OF CITY SIZES
A. Bejan, S. Lorentet, A. F. Miguel,t and A. H. Reist

Every sector of society is a conglomerate of mating flows that morph in time
in order to flow more easily: people, goods, money, information, and so on.
t Laboratory of Materials and Durability of Constructions, Department of Civil Engineering, Na
tional Institute of Applied Sciences (INSA), I 35 Avenue de Rangueil, 3 I 077 Toulouse, France.
* Department of Physics, University of Evora, Coiegio Luis Verney, 7000-67 1 Evora, Portugal.
775
Like all the flows of nature and engineering, social constructs are "alive."
The view that society is a flow system with intertwined morphing (improving)
architectures was part of the original disclosure of constructal theory (Ref. 1
and the 1 997 edition of this book). The generation of flow configuration in
society was illustrated for traffic, transportation of goods, distribution of wa
ter, and other aspects of urban design [5] .
This deterministic physics principle i s in sharp contrast with the empirical
(descriptive, modeling) approaches that have been tried to explain social or
ganization. Society is viewed like the photograph of a turbulent flow. Even
though the existence of structure is obvious, the image is so complicated, and
so much the result of individual behavior, that description is the norm, not
prediction. For a review, see Bretagnolle et al. [70] , who argue in favor of
introducing a spatial dimension (geography) in modeling, toward the devel
opment of an evolutionary theory of settlement systems. We agree. Such a
theory would provide insights for better policy in the future, and will predict
the future evolution of towns, cities, and their heterogeneous distribution on
land.
Society may be complicated, but pattern is not. Indeed, pattern is "pattern"
because it is not complicated. If it were not simple enough for us to grasp,
it would be noise, chaos, turbulence, and randomness. Strikingly clear images
such as Fig. 1 3 .37 remain unexplained: The size of a city in Europe is in
versely proportional to its rank [70-72] , throughout history. Why?
In this section we show that social patterns of organizations are consistent
with, and can be deduced from, the constructal law. Figure 1 3 .37 is derivable
1 800 1 850
Figure 13.37
Ref. 70.)
City rank
City size (population) versus city rank, in 1 600- 1 980 Europe. (After
776
from the constructal tree-shaped structures deduced for traffic (Section

1 3.2. 1 ), which we now review as an introduction. Consider the minimization
of travel time for traffic between an infinity of points (an area) and one point
(Figs. 1 3 .4 and 1 3.5). Two distinct modes of movement are available: slow
(e.g., walking) of speed V0, and fast (e.g., riding on a vehicle) of speed V 1
The construction of the tree-shaped architecture of the river basin of people
starts with the smallest elemental area A 1 , which is fixed by the culture of
those who live on A 1 For example, A 1 is the farmland surrounding the small
est road ( V1 , L 1 ) that leads to a single marketplace, M 1 The slow movement
covers A 1 and touches every point: the slow movement attaches every single
inhabitant to the traffic architecture. In eq. ( 1 3.6) we saw that there is an
optimal A 1 shape such that the travel time is minimized for everybody who
resides on A 1
The existence of two modes of movement implies a certain level of civi
lization: person living not alone (i.e., person with horse) and person with
vehicle. Civilization, or "city living" is also the name for the coexistence of
farmland (A 1 ) with markets (M 1 ). Those who live on A 1 exchange farm prod
ucts with those who manufacture products and deliver services in compact
places such as M 1 It is this balanced counterflow between A 1 and M 1 that
justifies this key idea, which will explain patterns such as Fig. 1 3 .37:
The number of those who live on A 1 must be proportional to the number N1 of
inhabitants living at M1, and both numbers must be proportional to A 1 Both groups
are sustained by the agriculture and the "environment" that A 1 provides; therefore,
N1 = cA 1 , where c is the average number of inhabitants per unit area.
The "culture" factor c accounts for the age and history of the civilization
(e.g., technology, commerce, neighbors, natural disasters, plagues, war,
peace).
The next larger area that is civilized (A 2) is covered by an assembly of n 1
optimal A 1 rectangles (A 2 = n 1 A 1 ). A central road (speed V2 , such that V2 >
V 1 ) collects or distributes the traffic associated with the elements. This first
construct (A2 ) can be optimized to provide minimal travel time between A 2
and the new boundary point M2
The counterflow of goods between the 11 1 small markets (M 1 ) and the larg
est market (M2 ) requires a proportionality between the number (N2 ) of inhab
itants at M2 and the total number of inhabitants at the M 1 points. This means
that N2 = cA 2 We see here two directions in which hierarchy develops: Areas
coalesce, from elemental to first construct, and at the same time, the popu
lation develops concentrations, from farmers on several A 1 plots, to traders at
several points M 1 , and finally, to one trading point M2 , which is perhaps a
small town.
At the next level of assembly, a number (n2 ) of first constructs coalesce
into a second construct (A 3 = n 2A 2) such that a new central road (speed V3 ,
where V3 2:: V2 ) links the M1 markets with one new market on the boundary,
777
M3 In a society, the counterflow of goods and people between M3 and M2

markets means that the number (N3 ) of people living at M3 must be propor
tional to n 2N2 Furthermore, because 11 2 = A 3 /A 2 , we conclude that N3 = cA 3
The traffic analysis [ l ] showed that when V2 is not much greater than V1 ,
the optimization of A3 for minimal travel time yields 112 = 2; this feature
(pairing, dichotomy) prevails at higher orders of assembly. Dichotomy is a
constructal-theory result, not an assumption. The optimjzed constructs alter
nate between squares (for i = even) and rectangles with 2 : 1 aspect ratio
(for i = odd), where i is the order of the construct A ;. The balance between
the flow of goods from each new construct A; to its concentration point (M;)
on the boundary requires that N; = cA ;, where c is the history-dependent
culture factor defined earlier.
The i ::::: 3 constructs cover the territory with multiscale cities, as shown in
the upper part of Fig. 1 3.5. Because constructs double in size from one con
struct level to the next (A ; = 2A ;_ 1 , i ::::: 3), population sizes also double (N;
= 2N;_ 1 ) and the number of concentration points decreases to half during
each step (n; = 11; _ 1 /2). In Fig. 1 3 .38 this is illustrated by disk-shaped cities
the radii of which increase by the factor 2 1 1 2 The end of this sequence (i =
m, odd or even) occurs when the construct area (A111 ) matches the size of the
available tetTitory (e.g., country, or continent). At this uppermost level of
assembly, the number of largest cities is one or two. In conclusion, macro
cephaly, which is evident in Fig. 1 3 .37 and throughout demography, is a
constructal-theory result.
The slope of the distribution of city sizes in Fig. 1 3 .37 is also a constructal
feature. The concentration points discovered in Fig. 1 3.38 have population
sizes that decrease by a factor of 2 when the total number of cities increases
by a factor of 2. The stepped line in Fig. 1 3 .37 is from constructal theory,
and its slope agrees with the trend exhibited by the data. The stepped line is
rough in comparison with the data because the construction deduced in Fig.
1 3.38 does not take into account the complex and the uncertain: geography,
geology, and human history. This discrepancy is not the issue. It is the agree
ment between theory ad reality that counts, because it shows the origin of
the striking pattern of organization that exists in even the most complex of
all flow structures: human society.
The theoretical curve tells us more than the empirical data. In time, the
entire curve must shift upward while remaining parallel to itself. This trend
is due to the culture factor c (or average population density), which increases
in time and has the same value throughout the country (A111). The constructal
pattern of interlinked cities (Fig. 1 3 .38), and all of the tree architectures de
duced earlier from the constructal law are not fractal objects. In constructal
theory the construction algorithm is deduced, not assumed.
A new theory is a new language, which performs new services. One is to
summarize into a simple and predictive statement a large volume of empirical
information (e.g., Fig. 1 3.37). Another is to predict future observations: fea
tures that may be present but not evident in the existing volume of empirical
778
-
.
Figure 13.38
During the spreading of interlinked city flows, total areas double in size,
and the size of the largest city doubles.
data. For example, because dichotomy is replaced by numerous elements

when size is small (e.g., A in Fig. 1 3 .5), the curves plotted in Fig. 1 3.37
should become less steep as the city rank increases (i.e., as towns with less
than 5000 inhabitants are counted).
Another theoretical eye opener is provided by Fig. 1 3 .38: If we imagine
all cities growing in time, we see that the larger will "eat" the smaller. Ac
cording to constructal theory, then, urban growth must be stepwise, uneven
in direction and time. This prediction is supported by the strange (uneven)
fingering exhibited by all large urban areas.
The most special and, potentially, most lasting cont1ibution of this new
theory is the accidentt that it explains the origin of huge volumes of obser
vations in fields far removed from that of the theory and its authors. For
example, the constructal distribution of city sizes in Fig. 1 3 .37 happens to
have the same trend as a Zipfian distribution of the occurrence of words in
language (cf. www.nist.gov/dads/HTML/zipfslaw.html). According to the
empirical formula known as Zipf's law, the probability of occtmence of words
or other items starts high and tapers off: P11 1 I n, where P11 is the frequency
of the n'h-ranked item and a is close to 1 . In English, for example, the words
and and the occur very frequently, whereas a word such as juxtaposition is
t J. S. Adams, private communication to A. Bejan, August 2 1 , 2005, Sun Valley Writers' Confer
ence, S u n Valley, Idaho.
CONSTRUCTAL THEORY OF DISTRIBUTION OF RIVER SIZES
779
very rare. This empirical law characterizes the occurrence of words in human
and computer languages, operating systems calls, and colors in images, and
is the basis of many compression approaches. The fact that in Fig. 1 3 .37 such
a trend is predicted from the analysis of a tree-shaped flow system on opti
mized geography suggests that all the domains in which Zipfian distributions
are observed (information, news) are homes to tree-shaped flow systems
(point-area, point-volume). These systems inhabit flow architectures (draw
ings) optimized subject to fixed global size. Information and news are con
structal flows: river basins and deltas akin to those predicted in the next
section. In this way, information and news are brought under the great tent
of constructal theory: Geometry, geography, tree architectures, freedom to
morph, and optimized finite complexity (hierarchy) are constructal properties
of the flow of information.
13.5
A. Bejan, S. Lorentet , A. F. Miguel/' and A. H. Reis*

Nature abounds with extremely complex flow systems whose organization is
so clear that it calls for a principle of the generation of flow configuration.
To see that Fig. 1 3 .37 is not the only "highly complex" puzzle that is solved
by the constructal law, consider the data on river basin morphology. Mea
surements all over the globe show overwhelmingly that relatively tight cor
relations exist between the numbers and length scales of channels in a river
basin [73 ,74] . Here we deduce these relations from the constructal law, and
then we compare them with the river basin data.
According to constructal theory, the river basin is a flow architecture that
emerges to facilitate the flow of the area-integrated rainwater to one point
(river mouth, estuary, or delta). The water flow is driven by a difference in
altitude from the mountain peak to the sea. Every river channel erodes its
bed and tries to decrease its slope. This must be done in harmony with the
other channels, because the overall drop in altitude is fixed. What results is
an optimal distribution of drops in altitude from one tributary to the next
larger stream.
The optimal step change and all the other architectural features of the river
basin follow from the constructal Jaw. Consider the two constructs shown in
Fig. 1 3 .39, where four square areas A 0 collect rainwater and channel it away
as a stream of length L0 /2, width D0 and depth d0 . Constructal theory showed
that for minimal flow resistance along the channel with free surface, there
must be a universal proportionality between D; and d; (Section 1 3.3.5). This
t Laboratory of Materials and Durability of Constructions, Department of Civil Engineering, Na
tional Institute of Applied Sciences (INSA), 1 35 Avenue de Rangueil, Toulouse 3 1 077, France.
* Department of Physics, University of Evora, Colegio Luis Verney, 7000-67 1 Evora, Portugal.
780
Ao
Do
D,
(a)
Dz
111
-
(b)
Square river basins with four A0 area elements: (a) square obtained by
two successive pairings (A 0 -+ 2A0, 2A0 -+ 4A0); (b) square obtained by one quadru
pling (A 0 -+ 4A 0).
Figure 13.39
means that the optimized river channel has only one transversal dimension,
which is called D;. The age of the construct is fixed, and is represented by
the volume of eroded riverbed (i.e., the volume of all the channels).
The sizes D; of the river channels in Fig. 1 3 .39a and b can be optimized
such that the ensuing flow structure has minimal flow resistance. The flow
along each channel is turbulent, which means that the drop in altitude along
the channel (z;) is proportional to ni?LJ D, where L; is the channel length
and n1; is the mass flow rate through the channel. The construct of Fig. 1 3 .39a
is obtained by two successive pairings (as in Fig. 1 3.5, top), and the result is
a square. In Ref. 75 we showed that in the case of the first pairing (A0 with
A0), the optimal size ratio of channels with turbulent flow is Dr/D 1 = 2- 3 17.
In addition, it can be shown that the minimization of the overall z for the
4A0 construct of Fig. 1 3.39 yields D J D2 = 2-317 .
Performed for Fig. 1 3 .39b, the same optimization analysis yields D0/D 1 =
2-417, and D J D2 = 2- 2 17. The overall flow resistances of the two 4A0 con
structs is such that C:.z)z" = 1 .964, which shows that Fig. 13 .39b is signif
icantly better. In other words, to assemble four elements as in Fig. l 3.39b is
better than to pair two A0 elements and, later, pair two 2A0 constructs, as done
in Fig. 1 3 .39a.
In Fig. 1 3.40 we use two river basins of size 8A0 to see whether quadru
pling (Fig. 1 3 .40a) is also better than assembling eight A0 elements in one
move (Fig. 1 3 .40b). The optimization of the distribution of channel volume
in Fig. 1 3.40a yields Dr/ D 1 = 2-417, DJ D 2 = 2- 2 17, and D 2 / D = 2- 3 1 7 . In
3
the case of Fig. l 3.40b, the optimal distribution of channel sizes is D0/ D , =
2 - 517, DJD2 = 2- 2 17, D 2 /D3 = (2/3) 2 17, and D3 /D4 = (3 /4)2 17. The mini
mized flow resistances (z/1172) are proportional to each other, but not equal,
z) z" = 0.8 1 5 .
781
111
Figure 13.40
(a)
(Ii)
River basin with eight A0 area elements: (a) construct obtained by pair
ing two square constructs of size 4A0; (b) eight A0 elements assembled in one move.
In conclusion, quadrupling is best, because it fits between pairing (Fig.

1 3.39a) and assemblies of eight constituents (Fig. 1 3.40b). Constructal river
basins are area constructs that grow by a factor of 4 at each new level of
assembly. This conclusion coincides with what we found in Fig. 1 3.9 for area
point conduction with trees of high-conductivity inserts. If we neglect the size
and fixed shape of the smallest area element, where there is a balance between
slow flow (seepage through wet ground on the hill slope) and fast flow (the
smallest rivulet), we can use the quadrupling rule to predict the morphological
features of increasingly complex river basins.
Table 1 3.4 shows the start of the construction, where i is the order of the
construct, A0 is the smallest square construct (the equivalent of A2 in Fig.
1 3 .5), N; is the number of streams of all sizes present on A;, Lr; is the total
length of all the streams present on A;, Ru is the ratio of the length of the
largest stream on A; divided by the length of the largest stream on A;_ 1 , R8;
is the number of the largest streams of A;_ 1 divided by the number of the
largest streams of A;, D wi is called drainage density and is equal to Lr/A;, F_,;
is the stream frequency (Fs; = NjA;), and Ns; is the number of all the streams
found on A;.
The well-known measurements and correlations of river basins are antici
pated very well and completely by the constructal structure displayed in Table
782
THE CONSTRUCTAL LAW OF CONFIGURATION GENERATlON
TABLE 13.4 Theoretical Structure of Constructal River Basins

A/A0 lr/A1\' 2
0
I
2
7
2
35
2
N;
Ru
Ru;
21
42
43
76
85
44
316
341
River
basins
Dw;A//2 F.,;Ao F,/D; LA.,/A) 1 1
2
7
8
35
22
76
64
316
256
4
5
4
21
16
85
64
341
256
-
1 .63
I . IQ
0.94
0.87
2
3
4
3
4
3
4
3
4
-
1 .5 -3.5
3 -5
0.7
- 1 .4
Horton
Horton
Melton
Hack
1 3 .4. The empirical correlations do not account for the size of the basin, but
the theory does. The agreement improves as the size and complexity of the
basin (i) increase. Horton's empirical correlation of stream lengths indicates
that RL is a constant with a value between 1 .5 and 3.5. Horton's empirical
correlation of stream numbers shows that R8 is a constant with a value be
tween 3 and 5. Melton's correlation states that F.JD is a constant approxi
mately equal to 0.7. Finally, Hack's correlation indicates that the length of
the mainstream LM; is proportional to Af, where A; is the basin area and b is
an exponent between 0.5 and 0.56. In the last column of Table 1 3 .4, we used
b = 0.5 to show in dimensionless terms that the constructal architecture also
anticipates the trend correlated by Hack.
13.6
CONSTRUCTAL THEORY OF EGYPTIAN PYRAMIDS AND

FLOW FOSSILS IN GENERAL
A. Bej an and S. Perin t

Pyramids in Egypt and other ancient sites intrigue us with their size and
geometric form. Even by today's standards, their size is immense and their
form is perfect. So impressive are these images that our culture tends to
attribute them to an ancient scientific base that was lost, and to presumed
links between ancient peoples living on opposite sides of the globe (Fig.
1 3 .4 1 ) .
I
I
t NeoSYS,
Air Traffic Flow Management Engineering, 7 rue d u Theatre, 9 1 300 Massy, France.
The Bnmenez Cairn

- 4700 - 3500 B.C.
The Step
The Great
Pyramid
A
-2630 B. C.
Pyramid
-2600 B.C.
--:i
00
..,,
of the
/ "',"
Figure 13.41
"\
Menkaure's
!')"umid
2558 - 2532 B.C.
of the
Temple
IV
-750 A.D.
2490 - 2472 B.C.
Temple
;\ /\
-790 A.D.
Pepi
!l's
Pyramid
Pyramid
2575 - 2 5 5 1 B.C.
Pyramid
2575 - 255 I B.C.
2589 - 2566 B.C.
The Bent
Pyramid
-100 B.C .
The
Pyramid
Maidum
Pyramid
-2250 B.C.
a f the
Pyramid
Z\ ji_
-850 AD.
-900- 1000 A.D.
Pyramids of Gaul, Egypt, and Central America, chronologically. (After Ref. 76; photograph courtesy of S. Perin) .
784
We propose a considerably more direct explanation: pyramids as results of

a universal natural phenomenon that is so prevalent and permanent that it
governs the movement of all material on Earth. This view does not take
anything away from the known or presumed achievements of the ancients;
rather, it is a physics argument that what the ancients chose to do is natural,
that to engineer is natural, and that the geometry of all material flows (ani
mate, inanimate) can be reasoned based on a single principle. Pyramids and
ant hills are like the dried beds of rivers, cracked mud, and dendritic crystals
(snowflakes) (Section 1 3 .3.8): They are traces (fossils) of the optimized flow
configurations that once existed.
The universal phenomenon is the generation of flow architecture, and the
principle is the constructal law (Section 1 3 . 1 ): For a flow system to persist
in time (to survive), its configuration must change such that it provides easier
and easier access to its currents. In a flow system, easier access means less
thermodynamic imperfection (friction, flow resistances, drops, shocks) for
what flows through the river basin or the animal. The optimal distribution of
these numerous and highly diverse imperfections is the flow architecture itself
(lung, river basin, blood vascularization, atmospheric circulation, etc.). At the
global level, the system flows with minimum but finite imperfection, which
means that it destroys minimum useful energy (fuel, food, exergy).
In the making of a pyramid, the constructal law calls for the expenditure
of minimum work. This principle delivers the location and shape of the ed
ifice. The optimal location is in the middle of the quarry, because minimum
work means minimum sliding distance between mining site and construction
site (Fig. 1 3 .42). The pyramid and the quarry grow at the same time. If the
pyramid is a positive architecture (y > 0, Fig. 1 3.42), the quarry is its neg
ative. Such positive-negative pairs are everywhere in history and geography,
even though modern advances in transportation technology tend to obscure
them.
The shape of the pyramid can also be reasoned on the basis of the con
structal law. In a pile of stones held together by gravity ( "dry stones" ), shape
Ground level
Figure
13.42 Constructal growth of a pyramid. Every stone follows an optimal path,

which is a refracted ray. Sliding on the horizontal ( 1 -2) is less dissipative than hopping
up the incline (2-3). The pyramid shape (a) is the optimal distribution of losses, such
that the global movement of stones requires minimal effort.
CONSTRUCTAL THEORY OF EGYPTIAN PYRAMIDS
785
means the base angle a (Fig. 1 3 .42). For simplicity, assume a conical pile. If
is the thickness of the stone cut from the quarry, the conservation of stone
volume requires that
( 1 3 .76)
r R
where and are the radii of the base and the quarry at a time during the
construction. The pyramid and the quarry grow together. The newest stones
added to the edifice come from the active wall (the perimeter) of the quany
The work required to move one stone from the quarry to its place in the
pyramid is comparable with the work of moving the farthest stone, from 1 to
2 and 3 in Fig. 1 3 .42. Let
be the weight of one stone, and model the
movement from 1 to 2 and from 2 to 3 as Coulomb friction with constant
friction coefficients
and
The total work spent is
N
,1 ,2 .
W W1 2 W23 , 1 N(R - r) (,2N
=
cos a +
sin a)(fI'- +
r2) 11 2
( 1 3 .77)
HIR,
r/
R
r/R
D ) 112 (H) - 112 + H ( 1 3 .78)
(
2
3
11
(
)
2
/1- 1 11- /1- 1 R R R
HI R,
W
(RH) p = (43 ) 113 (D) 11 3 (11-2 - /1-1 )_ ; 3
( 1 3.79)
R
ot
() op = 6 11 3 (D) 113 (11-2 - /1- 1 ) - 113
( 1 3 .80)
t
R
(r)
(HI r)opt
The edifice has two aspect ratios,

and
which are related through
eq. ( 1 3.76). Eliminating
between eqs. ( 1 3.76) and ( 1 3.77), we obtain
RID
W
NR
where
is a dimensionless parameter that increases with the age of the
construction. The total work
can be minimized by selecting
7
Remarkable is that the optimal height (H) and base radius

are both pro
portional to
which means that they depend on time in the same way.
Consequent! y, the optimal shape
is time independent: It is the same
for all constructs, early (small) and late (large),
R2 13,
( 1 3 .8 1 )
,2 ,1
>
(i.e., when hopping up on the
An optimal shape exists when
incline is more dissipative than sliding horizontally). This is the case of the
pyramids of Egypt; according to Crozat's theory [77 -79] , the incline used for
moving the stones upward was the pyramid itself, and the movement from 2
786
to 3 was effected step by step along the pyramid by using wood levers and
ropes. Each stone was ( 1 ) pushed horizontally, (2) lifted, and (3) dropped at
the next, higher level. Steps (2) and (3) and the system of levers and ropes
distinguish the movement on the stepped incline from the more common
(older, and presumably more perfected) method used on the horizontal ; hence
f-L2 > /L i . Had this not been the case, the transportation of stones on the
horizontal would have been in hops (according to scenario 1 -2-3), not by
steady sliding on wet sand or by using a wooden sled.
The exact numerical value of the base angle a0P1 is not the issue, because
even if we could estimate the orders of magnitude of /L i and f-L2 today, this
is not the knowledge that generated the location and shape of the pyramids.
That knowledge was the "culture" of the time period-the tendency exhibited
by many generations of large numbers of movers and moved (people, objects).
Large numbers, time, freedom, and memory are essential in the generation of
better and better flow configurations. For example, in the construction of a
river basin the number of water packets that move is immense: Time and
freedom allow the configuration to change, and memory is provided by old
riverbeds and seismic faults. The relationship between freedom to morph and
performance is explored further in Section 1 3.7.2.
The contribution made by this theoretical argument is that an optimal base
angle exists and that it is independent of the size of the edifice. The optimized
movement ( 1 -2-3 in Fig. 1 3.42) is a refracted path in the sense of Ferma.t.
There are two "media" through which the stream of stones flows, one with
low resistivity (f-L 1 ) and the other with high resistivity (t-L2 ). When the two
media are highly dissimilar ( t-L2 > > f-L J , the angle of refraction approaches
90.
The law of refraction governs the movement of goods in economics, where
it is known as the law of parsimony [80-82] . The history of the development
of trade routes reveals the same tendency. We often hear that a city grew
because "it found itself" at the crossroads-at the intersection of trade routes.
We believe that it was the other way around: The optimally refracted routes
defined their intersection, the city, the port, the loading and unloading site,
and so on.
More complicated flows are bundles of paths, optimally refracted such that
the global flow access is maximized. A river basin under the falling rain is
like an area inhabited by people: Every point of the area must have maximum
access to a common point on the perimeter. There are two media, one with
low resistivity (channel flow, vehicles on streets) and the other with high
resistivity (Darcy seepage through wet riverbanks, walking). The shape of the
basin comes from the maximization of flow access.
For example, if in the rectangular teITitory of Fig. 1 3.43 the objective is
to have access to point M, and if every inhabitant (Q) has two modes of
transportation, walking with speed V0 and riding on a faster vehicle with speed
V 1 , the average travel time between all the points of the area and M is min
imum when the area shape is HI L = 2 V0/ V1 This optimally shaped rectangle
is a bundle of an infinite number of optimally refracted paths of type QRM.
CONSTRUCTAL THEORY OF EGYPTIAN PYRAMIDS
787
Ai
+ +
+
+
+ +
+
+
+ +
+
+
+ +
+
+
+ +
+
+
+
+
+
Figure
+
+
+ +
+ +
+ +
+ +
+ +
13.43 Optimally shaped rectangle-point access and the Atlanta airport. Be

cause the train speed (V1) is higher than the walking speed ( V0), the shape of the
rectangular territory H X L can be selected such that the i nhabitants of the area have
maximum access to one point (M). When the shape is optimal, the time of walking is
the same as the time of riding on the train. Because of this, the rectangular shape of
the Atlanta airport is optimal. (After Ref. 83.)
Why is the constructal rectangle of Fig. 1 3 .43 relevant to the discovery

that the shape of the pyramid can be optimized? Because optimally shaped
rectangular elements are everywhere in city living, even though the builders
of these elemental structures did not rely on the constructal law to design
them. They behaved in the same way as the builders of the pyramids and
their ancestors: They balanced two dissimilar efforts ( V0 versus V1 , or , 1
versus ,) to minimize the global effort.
As shown in Section 1 3 .2. 1 , when the shape HI L is optimal, the travel
time from P to N is the same as the travel time from N to M: namely, H12 V0
= LI V1 In constructal theory this is known as the equipartition of time (re
sistance) or the optimal distribution of impe1fection. This is illustrated by
modern edifices such as the Atlanta airport [83] (Fig. 1 3 .43). Several objec
tives were pursued in the development of this tree-shaped flow structure: the
minimization of travel time for pedestrians, the minimization of time and
788
transportation cost for the goods flowing between the terminal and each gate.
The black line is the high-conductivity stem serviced by a two-way train. The
dark bars are the concourses along which the travel is much slower (walking,
carts). In agreement with constructal theory, the time to walk on a concourse
is the same (-5 minutes) as the time to ride on the train.
Science is not the search for the "designer" of the flow architecture; huge
numbers worked and work on it, and they used time, freedom, and memory
(culture) to construct it. The dry riverbed and the seismic fault through which
the new river chooses to flow are examples of "memory." The new scientific
aspect of the flow system is that it possesses architecture (shape, structure)
and that the generation of architecture is anticipated by the constructal law.
What holds for the pyramids and the ant hills also holds for all our logistics
and manufacturing operations.
13.7
THE BROAD VIEW: BIOLOGY, PHYSICS, AND ENGINEERING
The occurrence of flow configuration (shape, structure) is a phenomenon so

universal that it unites the natural with the engineered and the animate with
the inanimate. From 1 996, constructal theory has shown that flow architec
tures such as trees and round tubes can be deduced from a single law of
maximization of access for ctments (Section 1 3 . 1 ). If this law is correct, how
do we account for the occurrence of configurations that obstruct the flow of
currents? Hair and fur prevent the flow of heat from the body of the animal
to the ambient surroundings. Pairs of blood vessels in counterflow serve a
similar insulation function in the tissue under the skin. Insulation technologies
are everywhere in engineering. Much more obvious examples come from the
flow of fluids; impermeable walls are everywhere, in fact, without them there
would be no "streams." How can we reconcile such obvious contradictions
with the maximization of flow access?
Even if the configurations that maximize flow access could be put with
those that obstruct flow under the same theoretical tent, the theory would
cover only flow systems. So how can we account for the occurrence of refined
mechanical (nonflow) structures such as the skeletons of animals and the
frames of airplanes? Principles of maximum stress uniformity and minimum
weight have been invoked to account for the generation of architecture in
mechanical support structures [5, Chap. 2] . How can such "static" principles
be reconciled with the principle that unifies flow systems?
In this section I answer these questions by reviewing a series of recent
developments based on constructal theory, where geometry is the big un
known-the mechanism by which the global performance of the flow system
is maximized. The presentation is based on Ref. 84. The argument goes as
follows: "What" flows (fluid, heat, electricity, goods, people) is not as im
portant as "how" the flow derives its architecture from the competition be
tween objectives and constraints (cf. Table 1 3 . 1 ) . The maximization of
THE BROAD VIEW: B IOLOGY, PHYSICS, AND ENGINEERING
789
" access" means many things, depending on what flows; for example, the
minimization of flow resistance and pumping power in fluid flow (blood ves
sels, atmospheric circulation), the minimization of electrical resistance and
Joule heating in all electrical networks (computers, power grids), and the
minimization of travel time and cost in transportation and business (the Fer
mat principle of urban design and economics). In isolated thermodynamic
systems, it means the acceleration of mixing en route to classical equilibrium
(no flow).
In engineering, where the heat engine was the stimulus for the discovery
of thermodynamics, the constructal law delivers precisely what Sadi Carnot
called for: the minimization of friction and shocks in fluid flow and the avoid
ance of large temperature differences in heat flow. Such thermodynamic " im
perfection" cannot be avoided, because of size and time constraints.
Resistances will always be present. As Poirier wrote recently [6], the only
way up on the "staircase to heaven" envisioned by Sadi Carnot is by mnng
ing and balancing the resistances against each other. To arrange and to dis
tribute is to make the drawing, to deduce what was missing-the mchitecture.
Natural flow systems exhibit the same tendency. The largest engine on
Emth-the wheels of atmospheric and oceanic circulation-achieves the
same objective as birds, airplanes, and many other blobs of organized material
movement ( "streams" ) such as eddies of turbulence: They exist to facilitate
the movement of matter all over the globe (i.e., to maximize the mixing of
the matter that is the globe). River basins are trees, in accordance with the
constructal law of maximization of flow access. River cross sections have a
universal proportionality between width and depth, which can be deduced
from the constructal law.
If animal design proceeds in accordance with the constructal law, the an
imal destroys less exergy (i.e., useful energy, fuel) and requires less food.
"Animal design" means cmTents that flow along maximum-access paths be
tween organs, currents that are guided by walls, not currents that leak directly
to the ambient. Fuel or food management means that the engine and the
animal must carry the optimal weight that makes the whole animal efficient,
not the individual organ. From this comes the need to spread maximum
stresses uniformly and to support the flow structure with a mechanical struc
ture having minimal weight. There is no contradiction between the constructal
law of flow architecture and weight (size) as a constraint for the mechanical
structure that supports the flow structure.
Along this theoretical route the constructal law provides the physics that
is missing from the Darwinian principle of the fittest animal being the one
that survives. It provides the physics definition of what is meant by "the
fittest," or by its equivalent "the survivor," not only in biology but also in
engineering, geophysics, and economics, where selection and evolution are
also evident. According to constructal theory, therefore, animals, river basins,
and all of us-the "human + machine species" -are the same.
790
13.7.1
Heat Loss versus Body Size
Constructal theory predicted the proportionality between metabolic rate and

body mass raised to the power i, by invoking the constructal law twice, in
the minimization of body heat loss and the minimization of blood pumping
power [5,85]. The minimization of pumping power yields the constructal fluid
tree [4,5]; this can be derived more succinctly by optimizing a plane construct
consisting of a plane T-shaped junction [75 ] (Fig. 1 3 .44a) . For simplicity,
assume right angles and Hagen-Poiseuille flow with constant properties in
every tube. The stream n1; encounters the flow resistance of two L;+ 1 tubes in
parallel, which are connected in series with one L; tube. When the resistance
1
is minimized by fixing the total tube volume, we find D;+ 1 1D; = 2- 13, which
is independent of the tube lengths (L;, L;+ 1 ) and the relative position of the
three tubes. Next, we optimize the lengths when the space allocated to the
construct is fixed, 2L;+ 1 L; = constant. This yields the optimal ratio L;+ 1 I
L; = f = 2- 1 13, where the smallest length scale is labeled i = n, and largest
i = 0.
The trees of blood vessels are an architectural feature under the skin, but
not the only one. The other is the superposition of the arterial and venous
trees so closely and regularly that tube i of one tree is in counterflow with
tube i of the other (Fig. 1 3.44c and d ) . This is a thermal insulation feature
[86] .
The arterial stream is warmer than the venous stream: Heat flows trans
versally, from stream to stream. Because the enthalpy of the warmer stream
is greater than that of the colder stream, the counterflow convects longitudi
nally the energy current q; = n1;cp l::i. T,, ;, where Cp is the specific heat of blood
and l::i. T,,; is the stream-to-stream temperature difference at level i. It was first
shown in heat transfer [86] and later in bioengineering [87] that such a coun
terflow sustains a longitudinal temperature gradient, l::i. TJ L;, and that the con
vective energy current is proportional to this gradient:
( 1 3.82)
where h; is the overall stream-to-stream heat transfer coefficient and P; is the
perimeter of contact between the two streams. The stream-to-stream thermal
resistance h; 1 is the sum of two resistances: the resistance through the fluid
in the duct (-D/ kf, where kf is the fluid thermal conductivity), plus the
resistance through the solid tissue that separates two tubes ( -t/ k, where D;
is diameter and k is the tissue thermal conductivity; t; is defined in Fig. 1 3.44d:
t; is the average thickness of the tissue that separates two adjacent D; tubes).
Even when the tubes touch, t; is of the same order as D;. In addition, because
k r - k , we conclude that h; - k l D;, and eq. ( 1 3.82) becomes
q 'L
-'-k
l::i. T - ' 1h?c
( 1 3 .83)
r - - - - - - - -
1
I
I
I
I
(a)
I,.
I
I
,_
_
t_L;
I
L
..,
_
_
_
D;+1
_ _
(b)
1 iz0
(c)
791
Ln , D n
/ / -
-- -- --
Nn
i=n
Iii;
Warm
-:
Cold
liz; ---- D.T; ----
(d)
Figure
-:
13.44 Construction of the tree of convective heat currents: (a) constrained

optimization of the geometry of a T-shaped construct; (b) stretched tree of optimized
constructs; (c) superposition of two identical trees oriented in counterflow; (d) con
vective heat flow along a pair of tubes in counterflow. (From Ref. 5 and 85.)
792
The double-tree fluid structure is a single tree of convective heat leakage

with zero net mass flow. The convective tree stretches from the core temper
ature of the animal (at i = 0) to the skin temperature. The latter is registered
in many of the elemental volumes (i
n) that are near the skin. The many
counterflow pairs of the two fluid trees sustain the overall temperature dif
ference !J.T:
=
( 1 3 .84)
In going from eq. ( 1 3 .83) to eq. ( 1 3. 84), we used the continuity relations for
fluid flow
= constant), and heat flow
=
constant).
=
Recalling the
constant .f, we substitute
=
and
= into eq. ( 1 3. 84):
2;
q
,
0
L
L
L; L(N;q;
f,
,
0
0.f", N;
(N;Ln1;i , IL; 1 10
+1
( 1 3 .85)
The right side has quantities that are constant and quantities that depend on
n (the number of construction steps). The ratio
is independent of body
size (11) because both
and
are proportional to the metabolic rate.
The volume inhabited by the tree is estimated by considering the stretched
tree as a cone in Fig. 1 3 .44b. The base of the cone (at i = n) has an area of
size
The height of the cone is on the same order as the sum of
all the tube lengths,
+ +
= (1
.f), and the
+
volume scale is
%11110
q0 1h0
N,,L 2"L. L0 L
1
L, L0 - .f" + 1 )/(l (2) 2 - .f"+

L,3,
1 - .f
f
]
( 1 3 .86)
The relationship between metabolic rate and total volume is obtained by

eliminating n between eqs. ( 1 3 .85) and ( 1 3 . 86). The result is visible in closed
> > 1 in eq. ( 1 3.85) and
form if n is sufficiently large so that
< < 1 in eq. ( 1 3.86). In this limit
is proportional to
and V to
From this follows
1
2.
f
)"+
(
q0
q() = (constant) V3 14
2"
f" + 1
.
(2/.f)".
( 1 3 .87)
1i.
It can be verified numerically that eq. ( 1 3 .87) also holds for small
In
conclusion, the proportionality between metabolic rate and body size raised
to the power f is predictable from pure theory.
Constructal theory also anticipates the proportionality between breathing
(or heartbeating) time and body size raised to the power t [5] . In one of my
first constructal papers [3], I showed that the pumping power required by the
THE B ROAD VIEW: B IOLOGY, PHYSICS, AND ENGINEERING
793
heart for blood circulation and the thorax for breathing is minimal if ( 1 ) the
flow is intermittent (in and out, on and off), and (2) the "in" time interval
(t 1 ) is on the same order of magnitude as the "out" time interval (t2). Both
ideas are very important because unlike in all physiological models, they
come from pure theory (the constructal law), not from observations. The
t) is
optimal time scale (t 1 .2
AD 1 1 1J,.C
m
2
( 1 3 .88)
where A is the total internal contact area of all the tubes of the tree, D is the
mass diffusivity, /J,.C is the concentration difference that drives the mass trans
fer process, and 1i1 is the total mass flow rate of the tree (blood, air). The flow
rate n"i is proportional to the metabolic rate of the animal.
Equation ( 1 3 .88) shows that to predict t as a function of body mass (M")
we need expressions for n"i(M,,) and A(M,,). From the optimized tree of con
vective currents we obtained eq. ( 1 3.87), or
TiJ
Aft 1 4
( 1 3.89)
To predict the relationship A(M"), we argue that the thickness of the tissue
penetrated by mass diffusion during the breathing or heartbeating time t is
proportional to t 1 1 2 The body volume (or mass) of the tissue penetrated by
mass diffusion during this time obeys the proportionality relationship M"
At 1 1 2 . Eliminating t between M" At 1 1 2 and t (A l n"i)2 , and using eq. ( 1 3 .89),
we conclude that the contact area should be almost proportional to the body
mass:
( 1 3 .90)
Finally, the proportionalities ( 1 3.88)-( 1 3 .90) mean that
M] / 4
( 1 3 .9 1 )
This allometric law i s supported convincingly by the large volume of obser

vations accumulated in the physiology literature [88].
The constructal law for body heat loss versus size [eq. ( 1 3.87)] stands out
on the background provided by the history of the theoretical attempts to pre
dict this relationship. The earliest was Rubner's heat transfer model: Because
the convective heat loss is proportional to the body surface, the metabolic
rate must be proportional to the length scale ( V 1 1 3) squared (i.e., the body
mass or volume raised to power 1-). The heat transfer model was discredited
by more recent observations of birds and mammals, which suggest an expo-
794
nent closer to t than f. Because of such observations, heat transfer was not
included as a feature in recent fluid mechanics tree network models (e.g.,
West et al. [89]), which, by the way, is a good illustration of why modeling
is empiricism. We return to these models in Section 1 3.7.4.
Why should anyone question the ctmently accepted models by resurrecting
heat transfer? First, and this is key, modeling is not theory. Models are sim
plified descriptions (facsimiles) of objects observed in nature. Second, the
minimization of pumping power, which is invoked by modelers, is the con
structal law, because less pumping power means less exergy destruction and
less food for the animal to survive. But a lower heat leak also means less
food and less exergy destruction. This is why the minimization of pumping
power goes hand in hand with the old heat-loss doctrine, not against it. Min
imum pumping power consumption and minimum loss of body heat are parts
of the same constructal law: how to be constructed to be the fittest (the
survivor), how to perform best, how to flow best.
Additional support for the constructal theory of body heat loss comes from
the allometric laws of the design of the hair coats of animals, such as the
proportionality between the hair strand diameter and the animal body length
scale raised to the power + (Fig. 1 3 .45). This allometric law was predicted
[90,9 1 ] by minimizing the body heat loss through the hair coat. The + ex
ponent was predicted for both natural and forced convection.
Another common feature of animal hair coats is the porosity, which is high
and nearly constant (between 0.95 and 0.99) for all animal sizes [24] . This
feature was predicted by minimizing the combined heat loss by conduction
and radiation through the hair air coat. This analysis and the most recent
design applications of the constructal law are reviewed in Ref. 1 0.
Constructal theory predicts not only the t exponent for eq. ( 1 3 .87) but also
the gradual decrease of this exponent as the body size decreases. The t ex-
L (m)
Figure 13.45
Allometric law for animal hair strand diameter and body length scale.
795
ponent is valid at the limit where the body beat loss is impeded primarily by
the convective resistance posed by the blood counterflow of perfectly matched
tube pairs (Fig. 1 3 .44b). As shown in Bejan [5,85] , heat-loss paths are in
general more complicated. The convective thermal resistance posed by the
trees in counterflow (R 1 ) resides inside the animal. This resistance runs in
parallel with a second internal resistance (R2) associated with the conductive
beat leak through the tissue. On the outside of the animal the beat current
flows through the convective resistance (R3 ) associated with the body surface
exposed to the ambient (air, water). The conductive resistance R2 is propor
tional to the body length scale V 1 1 3 divided by the body surface scale V2 1 3 ;
hence R1
v- 1 13. The tree resistance R I is proportional to v- 3 14 The ratio
2
1
R2 / R I
V5 1 shows that R1 becomes progressively weaker (i.e., the prefeITed
path) as the body size decreases. At that limit the exponent in the power law
between heat loss and body size becomes +. In other words, from constructal
theory we should expect a gradual decrease in the power-law exponent as the
body size decreases.
The generality of the constructal deduction of the allometric law of me
tabolism ( 1 3 .87) is due to the view that the ft.ow structure results from the
clash between two objectives: the need to carry certain substances from the
core to the periphery of the organism (e.g., nutrients, water, ions, waste prod
ucts) and the need to avoid the direct leakage of these substances and energy
(heat) into the ambient surroundings. All biological ft.ow architectures are
results of this clash, from microbes to plants and animals, including warm
and cold-blooded vertebrates. The regulated temperature difference between
the body core and the ambient surroundings (large or small) is not the issue.
According to constructal theory, the ft.ow system (the animal and its move
ment) must evolve like any other engine-propelled body on the surface of the
Earth (e.g., Figs. 1 3 .46 and 1 3 .47), and this means that the exergy derived
from the food must be channeled optimally through the motor (muscles), not
dumped straight into the ambient smrnundings.
13.7.2
Flight and Organ Sizes
The design principle and results reviewed so far are relevant across the board,
from biology to engineering. This point is pressed with vigor by the aircraft
sketched in Fig. 1 3 .46: If the word fuel is replaced by food, the same drawing
is valid for a bird and reveals how the exergy liberated by food is destroyed
by all the currents that ft.ow around and through the animal. The food or fuel
exergy is destroyed completely by cmTents that overcome resistances.
The flying system becomes "more fit" when the total destruction of exergy
is minimized: more body mass ft.own, to longer distances. This is just like
the Gulf stream: more ocean mass carried with less resistance (i.e., faster and
for longer distances). The mechanisms that destroy food exergy (e.g., air
friction) cannot be minimized individually and eliminated, because each such
796
Fluid ducts,
All other flow
blood vessels,
systems on
lung airways
board
To support
the weight
Food,
fuel
----""-----
Motor, muscle
exergy lo be
inefficiency
destroyed
(power
for flight)
Figure 13.46
Distributed destruction of food or fuel exergy during flight.
mechanism serves the flying body as a whole. This was illustrated in most
general terms [5] by minimizing the sum (W) of the food exergies required
by air friction and lifting (raising the falling body):
( 1 3 .92)
where D is the body linear dimension, V is cruising speed, g is gravitational
acceleration, Pa is air density, and p,, is body density (based on the total body
mass scale M,, p,,D3 ). The terms on the right side of eq. ( 1 3 .92) cannot be
eliminated. They can be minimized together, thus:
Wmin
..-._.
1 3 g3 1 2
1
1 / 2 M /J 1 6
P
I
Pb
when the cruising speed vopt has the scale
( 1 3.93)
_ ()
THE B ROAD VIEW: BIOLOGY, PHYSICS, AND ENGINEERING
Vopt
1 13 !L
Pb
Pa
I / ?-
M1 1 6
b
Flying animals and machines are represented by the scales p"

and Pa
l kg/m3 , such that eq. ( 1 3 .94) becomes
797
( 1 3 .94)
1 03 kg/m3
( 1 3 .95)
where V0P1 is expressed in m/s and Mb is in kilograms.
Figure 1 3.47 shows this theoretical line next to flying speed data taken
from extensive compilations (Tennekes [92] and references therein). The
agreement between the line and the data is remarkable in view of the sim
plicity of the body model with one length scale (D). Insects, birds, and air
planes have multiple length scales, and this may explain why some of the
data fall above or below the line.
Agreement over such a wide diversity of sizes and types of flying systems
shows that the constructal law unites the designs of all the flying systems,
the animate with the engineered. This is stressed by the additional features
of Fig. 1 3 .47. Small animals (insects, hummingbirds) flap their wings all the
time, and their engine propellers (the wings) also provide the lift. In this limit
of small mass, the motor and the lift functions are pe1formed by a single
structure: the wings. At the other end of the body mass scale, large masses
(aircraft) fly with separate motor and lift structures. The lift is provided by
the wings proper and the motor (thrust) by a different set of wings, the blades
of the turbofan engine.
Between the "fully integrated" and "separate" motor and lift we find the
"almost separate" distribution of motor and lift functions. We see this in the
V-shaped flocks of migratory birds. The goose is the motor when it flies at
or near the tip of the V formation; when it is not, the goose surfs on the
waves generated by the geese working in front. Pterosaurs are also in between.
Their motor and lift functions were almost separate: they flapped their wings
rarely, and glided most of the time under the hot sun [93].
More recently we showed that the minimized flood consumption dictates
the physical sizes of the various flow components of a complex flying system
[94]. This development is detailed in Section 1 1 .6.2. In brief, the total food
required for flying over a distance is proportional to the total food exergy
that is destroyed, W W111i 0L/ V0r" which according to eqs. ( 1 3 .93) -( 1 3 .94),
is proportional to body weight and distance: W M"gL. In the aircraft in
dustry this proportionality is known as the takeoff gross weight criterion: The
fuel penalty associated with placing a new component on board is propor
tional to the mass of that component. Smaller flow components are attractive.
But flow components function less efficiently when their sizes decrease: they
86L
"'l
i:iei"
c:
'"I
"'
(.;J
....
--1
:,,.
:TI
5
1. 1
'<
(IQ
'"O
(1l
(1l
en
0.
"'
0
.....,
(1l
"'
F
O"
::;
0.
.?'
I"
:::i
0.
i:: .
..a
p;"
:::i
(1l
(")
I"
:::i
.?'
0.
::r
-
::r
(1l
0
'1_ 1
Houe
o;
Hornet
S kylark
Blue jay
Sparrow hawk
'!.':,
Blue heron
""
0 ..,
White stork
Golden eagle
n
v;
"
Wandering albatross
Pheasant
Canada goose
.....
:::.
(")
i::..
0
:::i
(")
c:
en
::r
'.
'"O
(1l
(1l
"'
p.
'.41
0
q_
,...._
3
;;z.::I
(1l
0
'-"
F- 1 6
Mig-23
Fokker F-28
F- 1 4
Boeing 737
Airbus A3 10
c -------...J
'
"
Beech King Air

Leerjet 31
Douglas DC- J O
Boeing 747
"'
'-'
Mute swan
Piper Warrior
Beech Bonanza
Beech Baron
g
"E.
"'
0
...,
oO
;J>
(")
0 0
0
0
-.
(1l
s 0
'!,
c;'-"
Black vulture
Gannet
..
Partridge
c;
c;
'!.':,
House sparrow
V (m/s )
"
CT
Barn swallow
Ruby-throated
hummingbird
House wren
'-" 1
Bumblebee
Magnolia
warbler
Sand martin
c;I
Honeybee
.\
Dung beetle
Goldcrest
w
0
0
Meat fly
Peregrine falcon
c;
V (m/s)
.
..
.
.
..
799
pose greater resistance to flows and destroy more exergy, so the flying animal
or machine requires more food or fuel to carry such components.
Constructal theory accounts for the existence of characteristic (proportion
ate) organ sizes in animals and engineering installations. The fundamental
trade-off in body size is illustrated in Fig. 1 1 .28. The total food exergy re
quired by a flow component is the sum of the food exergy destroyed by the
component and the food exergy required (and destroyed) by the flying system
to cany the component on board. There is an optimal component size such
that its impact (penalty) on the total food required by the bird is minimum.
This trade-off is fundamental: It rules the optimization of organ sizes in every
flow system, animals and vehicles alike.
13.7.3
Survival by Increasing Freedom, Performance, Svelteness,

and Territory
In constructal theory, body size, architecture, and complexity are results, not
assumptions. They are intrinsic parts of the drawing: the optimal configuration
to which the flow system tends in time, in accordance with the constructal
law. This tendency was recently put on an analytical basis, such that the
constructal law becomes a new extension of thermodynamics: the thermo
dynamics of nonequilibrium (flow) systems with configuration [ 1 1 , 1 7 , 1 8] .
This formulation i s condensed in Fig. 1 3.48. A flow system (e.g., a tree) has
"properties" that distinguish it from a static (nonflow) system. The properties
of a flow system are:
1 . Global external size (e.g., the length scale of the body bathed by the
tree flow, L)
2. Global internal size (e.g., the total volume of the ducts, V)
3 . At least one global objective, or performance (e.g., the global flow re
sistance of the tree, R)
4. Configuration, drawing, architecture
5 . Freedom to morph (i.e., freedom to change the configuration)
The global external and internal sizes (L, V) mean that a flow system has two
length scales, L and V ' 1 3 These form a dimensionless ratio-the svelteness
Sv-which is a new global property of the flow configuration [95 ]:
external length scale
Sv = -------
internal length scale
1 13
( 1 3 .96)
The flow systems covered by the constructal law stated in Section 1 3 . l

populate and move i n the V = constant plane shown in Fig. 1 3.48. This plane
houses a performance versus freedom diagram: In time, and if the architecture
800
constant I
..<:::
&
L=
fInlotwernalsize,
Figure 13.48 Performance versus freedom to change configuration, at fixed global

external size. (After Refs. 1 7 and 1 8.)
is free to change, R decreases (i.e., performance increases) at constant L and
V. The configuration with the smallest R value represents the equilibrium flow
structure. The configurations that preceded it are nonequilibrium flow struc
tures.
An illustration of the "movement" of the flow architecture in time (at
constant L and V) is illustrated in Fig. 1 3.49. The abscissa parameter f is the
global dimensionless flow resistance for Hagen-Poiseuille flow between the
center of a circle and N = 1 92 points distributed equidistantly on the circle.
The svelteness Sv is large enough so that the overall pressure difference f),,p
is due entirely to fluid friction along the straight tubes that form the tree
networks. The external size (disk radius L) and internal size (tube volume V)
are fixed. The parameter f corresponds to the performance parameter R in
Fig. 1 3.48.
The ordinate of Fig. 1 3 .49 shows the number of tubes (n0) that come out
of the center. The parameter p represents the number of bifurcation or pairing
levels of the tree structure. To assume a constant-p structure is to curtail the
freedom of the morphing tree structure. To allow p to increase (downward in
Fig. 1 3 .49) is to endow the flow architecture with more freedom to morph.
When p is fixed (e.g., p = 4), all the possible flow configurations lie to
the tight of point c. Here the network is dichotomous (an optimization result),
the tube diameter ratios are optimized in accordance with Murray's law, eq.
(3.87), and every tube length is optimized. Point a represents a tree structure
801

200
N = 192
p = number of
pairing levels
100
p=0
p=2
110
b
-"
fr
0
s
B
s
1 <z7
10
0
-0
"
&:
Figure
10
p=5
"
.),.,>. .
_/
I
o a
Nonequilibrium
flow structures
Equilibrium
flow structure
Performance
1 00
f = llPY.____
1i1 87tL3
200
13.49 Performance versus freedom domain of flows that connect the center
with N equidistant points on a circle. (From Ref. 95.)
almost as "perfect" as c, except that the tube lengths of a are assumed

(guessed), not optimized. In time, the flow structure morphs and migrates
from points a to c. If the number of pairing levels (p) is also free to vary,
the migration is from the entire right plane (e.g., points a, b, and c) to d,
which is the equilibrium flow structure for this class of flows. Configuration
d is the best; it has only six levels of pairing. The number six represents finite
complexity, not "maximum" complexity. The number p = 6 is an optimiza
tion result. The optimized complexity comes from the increase in freedom to
morph (i.e., from the abandonment of architectural features that curtail the
freedom to morph).
Returning to the general Fig. 1 3 .48, we conclude that at equilibrium the
flow configuration achieves the most that its freedom to morph has to offer.
Equilibrium does not mean that the flow architecture stops changing. On the
contrary, it is here at equilibrium that the flow geometry enjoys most freedom
to change. Equilibrium means that the global performance does not change
when changes occur in the flow architecture.
The evolution of configurations in the constant-V cut (also at constant L,
Fig. 1 3 .48) represents survival through increasing performance-survival of
the fittest. This is the physics principle that now underpins Darwin's argument,
the law that rules not only the animate flow systems but also the natural
inanimate flow systems and all the human + machine species. The constructal
802
law defines the meaning of "the survivor" or of the equivalent concept of

"the more fit." The constructal-law idea that freedom to morph is good for
pe1formance (Fig. 1 3.48) also accounts for Darwin's argument that the sur
vivor is the one most capable to adapt.
In the bottom plane of Fig. 1 3 .48, the locus of equilibrium structures is a
curve with negative slope, (aR/aV) L < 0, because of flow physics; the resis
tance decreases when the size of the internal space inhabited by the flow
increases. This slope means that the nonequilibrium flow structures occupy
the hypersurface suggested by the three-dimensional surface sketched in Fig.
1 3 .48. The time evolution of nonequilibrium flow structures toward the bot
tom edge of the smface (the equilibrium structures) is the action of the con
structal law.
The same time arrow can be described alternatively with reference to the
constant-R cut through the three-dimensional space of Fig. 1 3.48. Flow ar
chitectures with the same global performance (R) and global size (L) evolve
toward compactness and svelteness: smaller volumes dedicated to internal
ducts [i.e., larger volumes reserved for the working "tissue" (the interstices)] .
Paraphrasing the original statement of the constructal law, we may describe
the evolution at constant L and R as follows:
For a system with fixed global size and global performance to persist in time (to
live), it must evolve in such a way that its flow structure occupies a smaller fraction
of the available space.
This is survival based on the maximization of the use of the available space.
Survival via increasing svelteness (compactness) is equivalent to survival via
increasing performance; both statements are the constructal law.
A third equivalent statement of the constructal law becomes evident if we
recast the constant-L design world of Fig. 1 3 .48 in the constant-V design space
of Fig. 1 3.50. In this new figure, the constant-L cut is the same performance
versus freedom diagram as in Fig. 1 3.48, and the constructal law means sur
vival by increasing performance. The new aspect of Fig. 1 3.50 is the shape
and orientation of the hypersmface of nonequilibrium flow structures: The
slope of the curve in the bottom plane (aR/ aL)\! is positive because of flow
physics (i.e., because the flow resistance increases when the distance traveled
by the stream increases).
The world of possible designs can be viewed in the constant-R cut made
in Fig. 1 3 .50, to see that flow structures of a certain performance level (R)
and internal flow volume ( V) morph into new flow structures that cover pro
gressively larger territories. Again, flow configurations evolve toward greater
svelteness Sv. The constructal law statement becomes:
For a flow system with fixed global performance (R) and internal size (V) to persist
in time, the architecture must evolve in such a way that it covers a progressively
larger teJTitory.
..c
e-
803
V= constant
Figure
flow size, L
13.50 Performance versus freedom to change configuration, at fixed global

internal size. (After Refs. 1 7 and 1 8.)
There is a limit to the spreading of a flow structure, and it is set by global

properties such as performance (technology) and internal flow volume, R and
V. River deltas in the desert, animal species on the plain, and the Roman
empire spread to their limits. Such is the constructal law of survival by spread
ing, by increasing territory for flow and movement. Now we know why this
should be so [95].
Another way to smrunarize this analytical formulation of the constructal
law is by recognizing the analogy [ 1 7, 1 8] between the analytical constructal
law and the analytical formulation of classical thermodynamics (Chapter 4).
The analogy is presented in Table 13 .5.
13.7.4 Modeling Is Not Theory

This chapter provides a broad view of how the constructal law accounts for
the generation of configuration everywhere, from flow architectures to me
chanical structures, and from animate systems (biology), to inanimate systems
(physics) and human + machine species (engineering, economics, business).
The constructal literature reviewed in this chapter covers all these domains;
this literature is recommended because, although known in engineering, it is
largely unknown in biology and physics.
The applications of constructal theory are not limited to the examples given
in this chapter. In fact, the connections between constructal theory and the
804
TABLE 13.5 Concepts and Principles of Classical Thermodynamics and

Constructal Theory
Thermodynamics
Constructal Theory
State
Flow architecture (geometry, configuration,

structure)
Morphing, change in flow configuration
Global objective and global constraints (R,
L, V, . . . )
Equilibrium flow architecture
Fundamental relation, R(L, V, . . )
Nonequilibrium flow architectures
Increased freedom to morph
Constructal principle:
R minimum at constant L and V
V minimum at constant R and L
L maximum at constant V and R
Process, removal of internal constraints

Properties ( U, S, Vol , . . . )
Equilibrium state
Fundamental relation, U(S, Vol, . . . )
Constrained equilibrium states
Removal of constraints
Energy minimum principle:
U minimum at constant S and Vol
Vol minimum at constant F and T
S maximum at constant U and Vol
Source: Refs. 1 7 and 1 8.
large volume of observations available in established fields such as biology

are yet to be made. They deserve to be explored to depths much greater than
in engineering [5] . In particular, because Section 1 3. 7 was mainly about al
lometric laws and body size, it is important to see the position of constructal
theory relative to the model of West et al. [89].
The constructal theory [ 1 -3] reviewed in the 1 997 edition of this book
predates the model of West et al. [89] and covers a physics that is much
greater than the allometric behavior correlated by the model. As stated in the
title of their paper, the work of West et al. is the 1nodeling of certain biological
systems, which they observe and then describe analytically. Modeling is em
piricism, not theory.
The model of West et al. is based on at least three ad hoc assumptions:
I . There is a "space-filling fractal-like branching pattern" (read: tree).

2. The final branch of the network is a size-invariant unit.
3. The energy required to distribute resources is minimized.
These three features were already present in 1 996 constructal theory, not
as convenient assumptions to polish a model and make it work but as invo
cations of a single principle: the constructal law. West and Brown [96] ac
knowledged the overlap. Specifically, feature 3 is covered by the constructal
law, feature 1 is the tree-flow architecture that in constructal theory is deduced
from the constructal law, and feature 2 is the smallest-element scale that is
fixed in all the constructal tree architectures. To repeat, in constructal theory
the tree-shaped flow is a discovery, not an observation and not an assumption.
CONSTRUCTAL THEORY OF RUNNING, SWIMMING, AND FLYING
805
Because features 1 to 3 are shared by constructal theory and by the model

of West et al., every single allometric law that West et al. connect to their
model is an affirmation of the validity of constructal theory. Every success of
constructal theory in domains well beyond the reach of their model (e.g., river
basins, flight, dendritic solidification, global circulation, mud cracks [5] ) is
an indication that animal design is an integral part of a general theoretical
framework-a new thermodynamics-that unites biology with physics and
engineering. Biology and "natural" selection have just been made a part of
physics.
13.8
CONSTRUCTAL THEORY OF RUNNING, SWIMMING,

AND FLYINGt
A. Bejan and J. H. Marden*

Biologists have treated as evident the view that fundamental differences exist
among running, flying, and swimming, because the forms of locomotion and
the animals are so different: limbs and wings versus body undulations, neu
trally buoyant versus weighted bodies, and so on. In this section we show
that all forms of locomotion can be described by a pure physics theory. The
theory is an invocation of the constructal law, as we showed in Section 1 3 .7.2
for flying. This principle delivers in surprisingly direct fashion the relations
observed between speed and body mass (M) raised to 1;, and between fre
quency (stride, flapping) and M- 1 1 6, and shows why these relations hold for
running, flying, and swimming. Animal locomotion is an optimized two-step
intermittency: An optimal balance is achieved between the vertical loss of
useful energy (lifting the body weight, which later drops) and the horizontal
loss caused by friction against the surrounding medium. The theory predicts
additional features of animal design: the Strauhal number constant, which
holds for running as well as flying and swimming, the proportionality between
maximum force and body mass in animal motors, and the fact that undulating
swimming and flapping flight occur only if the body Reynolds number ex
ceeds approximately 30. This theory and the general body of work known as
constructal theory together now show that animal movement (running, flying,
swimming) and fluid eddy movement (turbulent structure) are both forms of
optimized intermittent movement.
Running, flying, and swimming occur in very different physical environ
ments. Not surprisingly, then, the mechanics of moving a body on legs that
contact solid ground is vastly different from what is required to achieve
weight support in air or to move a neutrally buoyant body through liquid [97].
t The theory summarized i n this section is explored further and compared with data o n animal
locomotion i n the paper: A. B ejan and J. H . Marden, Unifying constructal theory for scale effects
in running, swimming and flying, J. Exp. Biol., Vol . 209, 2006, pp. 238-248.
' Pennsylvania State University, Department of B iology, University Park, PA 1 6802.
806
Despite these differences there are strong convergences in certain functional

characteristics of runners, swimmers, and fliers. The stride frequency of run
ning vertebrates [98 - 1 0 1 ] scales with approximately the same mass exponent
(M- 0 ) as swimming fish [ 1 02] . Velocity of running animals [ 103, 104] scales
with approximately the same mass exponent (M 1 7) as the observed and the
oretically predicted speed of flying birds [5, 1 05 - 1 07]. Force output of the
musculoskeletal motors of runners, swimmers, and fliers conforms with sur
prisingly little variation to a universal mass specific value of about 60 N /kg
[ 1 08]. What explains these consistent features of animal design?
In the absence of a theory that unifies design features across different forms
of locomotion, biologists have concentrated on potentially common con
straints. For example, Drucker and Jensen [ 1 02] hypothesized that the scaling
of muscle shortening velocity for maximal power output during oscillatory
contraction (M- 0 ) [ 1 09] might explain the common scaling of stride fre
quency. Numerous authors have hypothesized that scale effects in locomotion
are caused by constraints related to biomechanical safety factors and the need
to avoid premature structural failure [ 1 1 0- 1 1 4], or to maintain dynamic sim
ilarity [97, 1 15 ]. Here we take the different approach of starting not with
constraints but with general and presumably universal design goals that can
be used to deduce principles for optimized locomotion systems. Our approach
is approximate (order of magnitude accuracy) and is not intended to account
for all forms of biological variation. Rather, it predicts central tendencies that,
if valid, provide starting points for more detailed analyses and baselines from
which deviations can be quantified and therefore studied in a more rigorous
fashion. Furthermore, such a theory is not mutually exclusive of other hy
potheses such as common constraints, because constraints have evolved within
a design framework (i.e., perhaps theory can provide explanations for the
nature of constraints).
How can constructal theory be applied to the streams of mass flow on
Earth that are called running, flying, and swimming? The answer is: in the
same way that it has been applied to design features of flow systems such as
the morphing of river basins, lung design, turbulent structure, vascularization,
snowflakes, mud cracks, and atmospheric circulation. Better routes for the
huge number of small and large cun-ents that flow through and around the
animal means the minimization of all the thermodynamic imperfections (fric
tion, flow resistances) together, such that at the global level the animal de
stroys minimum useful energy [or food, or "exergy" in contemporary
thermodynamics (Chapter 3)] . We show that constructal theory delivers in
surprisingly simple and direct fashion the body-mass scaling relations for
running, flying, and swimming-the complete relations, the slopes and the
intercepts, not just the exponents of the body mass.
There is a long and productive history of optimality models in analyses of
animal locomotion [97, 1 05, 1 1 6, 1 1 7] , so much so that the term maximum
range speed is part of the common vocabulary of the field and instantly brings
to mind a U-shaped curve of cost versus speed on which there is one speed
17
17
807
that maximizes the ratio of distance traveled to energy expended. Our ap

proach is similar in that it predicts maximum range speeds, but differs from
previous efforts by being general across different forms of locomotion rather
than attempting to provide a detailed explanation for a single type of loco
motion. Like other optimality models, our theory does not maintain that an
imals must act or be designed in the predicted fashion, only that over large
size ranges and diverse taxa predictable central tendencies should emerge (cf.
the end of Section 1 3 .6). Ecological considerations will often favor species
that move in ways other than that which optimizes distance per cost: for
example, where energy is abundant and the risk of being captured by active
predators is high. Evolutionary history and the chance nature of mutation can
also restrict the range of trait variation that has been available for selection.
These and other factors should act primarily to increase the variation around
predicted central tendencies.
13.8.1
Running
Consider the cyclical motion of running (Fig. 1 3 .5 l a). To maintain a constant

horizontal speed V, the animal must perform work to account for its two
mechanisms of work destruction. One is the vertical loss: the destruction of
y
v
--
M
Vt
2Vt
(a)
)'
H
v
1-- s--j
(b)
'1
(c)
F=O
+ I
13.51 (a) Periodic trajectory of a running animal, (b) model of foot contact
with soft ground, and (c) history of vertical movement during one cycle, where F =
0 refers to the time t 1 when the foot no longer exerts a force on the ground.
Figure
808
the gravitational potential energy accumulated at the peak of each jump, W,

= MgH, which is destroyed during each landing (for simplicity, we neglect
elastic storage during landing). The other is the horizontal loss, W2 , which
occurs when the foot (moving with the speed V horizontally) comes to rest
on the ground during a short contact time. The constructal law calls for the
minimization of the total destruction of work per distance traveled L = 2tV,
( 1 3 .97)
where t is the Galilean time of frictionless fall from the height H, t = (2HI
g) 1 1 2 .
The horizontal loss W2 depends on the nature of the terrain. In the follow
ing analysis we consider the two extremes, highly deformable ground and
hard flat ground and show that the condition of the ground surface does not
affect significantly the optimal running speed predicted.
Assume first a deformable ground surface such as sand, mud, or snow of
density p, and use the method of scale analysis [ 1 4] (i.e., neglect dimension
less factors of order 1 ). Assume further that the speed of foot impact is high
enough that the terrain deformation consists of an amount of solid (powder,
particles) that is hit as if with a paddle and is accelerated in the direction of
the paddle. A 'high enough ' speed means that the accelerated terrain material
does not have time to interact by friction with and entrain its neighboring
terrain material. This model is analogous to the behavior of a pool of fluid
that is hit by a blunt body at a sufficiently high speed. In summary, we assume
that the sand behaves as an inviscid liquid when it is suddenly struck by a
blunt body (the foot) (Fig. 1 3.S l b) . The foot contact surface is A . The foot
hits the ground with the vertical Galilean velocity Vv - (gH) 1 1 2 . By analogy
with drag in high-Reynolds flow, the vertical force felt by the foot during
impact scales as F" - pVAC0, where C0 is a constant of order 1 . The work
done by the foot to defoiw the sand vertically is F)5, where 8 is the depth
of the sand indentation. This work is the same as W-1 ; hence, F" 8 - MgH.
The foot also moves horizontally to the distance g through the sand. The
horizontal drag force is Fx - pV2A 1 1 2 8C0, where C0 - 1 and A 1 1 2 8 is the
frontal area of the foot as it slides horizontally through the sand. The work
destroyed by horizontal deformation of the sand is W2 - F, g, where g = Vtc,
and the time of contact with the ground is tc - 81 VY. Putting these formulas
together, we find that W2 - V3M 2 / pA 3 1 2 (gH) 1 1 2 , and eq. ( 1 3.97) becomes
w
Mg3 1 2H 1 1 2
+
v
'------v--'
Vertical loss
v2M 2
pA 3 1 2H
---
'-------,-J
( 1 3.98)
Horizontal loss
The optimal running speed for minimal work per unit of length traveled is
809
( 13 .99)
The simplest reading of this result makes use of the rough approximation
that animal bodies are geometrically similar (especially when compared over
large size ranges). In this case (pA 3 1 2 IM)u 3 is a factor of order 1 that does
not depend on M. Since the types of animal motion that we are considering
are cyclical, with motion of body parts along a roughly circular or oblong
path and reestablishment of starting positions at the beginning of each cycle,
it follows that height deviations, in this case the height of the run H, scale
with the body length scale Lb = (Ml pb) u 3 , where Pb is the body density.
Given these scaling features, the optimal running velocity scale is V0p, g 1 1 2p-b u6M 1 1 6 The correspondinab relative optimal speed is Vopt ILb g 1 1 2pb1 16M- 1 16' which yields Vvpt ILb - 1 0 s- 1 M- u5 when M is expressed in
kilograms. These formulas agree in both trend and magnitude with the numerous speed data compiled by empirical studies (Fig. 1 3.52a).
The corresponding stride frequency scale is t-;;P1, - g 1 1 2pl16M 1 16, which
agrees well with the proportionality observed between stride frequency and
body mass raised to -0. 14 in the most rigorously defined study of scale
effects on stride frequency of runners [98]. Quantitatively, the predicted fre
quency is approximately 10 s- 1 M- 1 1 6 when M is expressed in kilograms (Fig.
1 3.52b).
These predictions do not change much if the model of the running surface
changes. If the surface is flat and hard, we may use a Coulomb friction model
and write that the horizontal loss is W2 - Fs, where is a constant coef
ficient of friction, F is the normal force during the foot contact time t, and s
is the foot sliding distance, s = Vt. The contact time scale is dictated by the
impact that the body experiences in the vertical direction, such that when the
body makes contact with the ground it is decelerated from its free-fall velocity
(gH) 1 1 2 to zero. Writing Newton's second law of motion, F - M(gH) 1 1 2 lt,
and using the preceding formulas we find that W2 - VM(gH) u 2 , such that
eq. ( 1 3 .97) becomes
w
Mg
( 1 3. 100)
Vertical loss
Horizontal loss
This monotonic function of V shows that there is a lower bound for the
optimal running speed. The minimum work per distance requirement is
1
achieved when V exceeds the scale - 1 g u 2p& u6M 16, which is essentially the
same as the scale predicted in eq. ( 1 3 .99) and tested against animal data.
The recommended speed cannot increase above - 1 g 1 1 2p& 1 1 6M1 16 indefi
nitely, because of air drag. The air friction force is on the order of F0 p<,V2LtC0, where C0 - 1 and Pa is the air density. If air drag dominates ground
810
( pb )t/3
Pa
112 - 1 !6 ll6
M
Pb
Running with air drag; flying
10
1 12 - 1 !6 l l6
M
Pb
Running on soft ground ; swimming
u
0
II
o_
0. 1
0.01
1 07
106
1 05
1000
1 0'
10'
1 0'
[J - -
o o
10'
10"
(a)
b.
Running mammals
\J
Running arthropods
Flying bats
Swimmin!!. mammals
Running lizards
Flying birds
Fying insects
Swimming fish
Swimmin
g cruswccans
Swimming human
( pb )l /3
100
'N
>-.
u
c
"
"
JO
Pa
1 12 l /6 - l /6
M
Pb
Running with air drag; flying
g:
1 12 1!6 - l !6
M
Pb
Running on soft ground; swimming

O. l
(b )
2 gM
Force. output
10'
z
"
0
w..
10
'
2 10 '
10"'
10
\7 Flying birds, bats, insects
'
Running mammals, reptiles, insects
Swimming fish, crayfish
10"7
10"6
10"5
10-'
10"3
1 0"'
10"1
1 0
1 01
10'
103
!04
Body mass (kg)
Figure 13.52
(c)
Comparison of theoretical predictions with the speeds, stroke frequen

cies, and force outputs of a wide variety of animals. The theoretical predictions are
based on scale analysis, which neglects factors of order I and therefore should be
accurate i n an order-of-magnitude sense. Note, however, that in nearly all cases the
theory comes closer than an order-of-magnitude agreement with empirical data. (Data
from Refs. 92, 98, 1 02- 1 04, and 1 1 8- 1 32.)
811
friction, W2 - FD L, and the last tenn in eq. ( 1 3 . 1 00) is replaced by FD In

this case the total loss WIL is minimum when V0P, - (pblp,,) 1 1 3g 1 1 2p"b 1 1 6
M 1 1 6 , where (p1/ p,) 1 1 3 = 1 0, because Pb = 1 03 kg/m3 and Pa = 1 kg/m3 . A
compilation of velocity data (Fig l 3 .52a) for animals running over a variety
of surfaces shows that they tend to use speeds that fall in between the two
extreme cases of running on soft ground and running against primarily air
drag. In sum, the effect of the horizontal friction model is felt through a factor
that is a dimensionless constant in the range 1 to 1 0: namely, (pA 3 1 2 /M) 1 1 3
for deformable ground, , - i for hard ground, and (p1, / p,,) 1 1 3 for air drag. The
optimal running speed is a remarkably robust result, always on the order of
g 1 1 2p"b 1 1 6M 1 1 6 . We propose that it is this robustness that accounts for the al
lometric law connecting V with M 1 1 6 in animals of so many different sizes
and habitats (Fig. 1 3 .52a).
Additional evidence that this running theory is valid comes from the cal
culation of the vertical force F that propels M to the height H during each
cycle (Fig. 1 3.5 lc). The force F acts during a short time t 1 when the leg
makes contact with the ground, and the moverrient of M upward is governed
by Newton's second law, Md 2y/dt2 = F - Mg. Integrating this from the initial
conditions y = dy/ dt = 0 at t = 0, we find that the body altitude and vertical
speed at t = t 1 are y 1 = (FIM - g)tT/2 and (dyldt)v, = (FIM - g) t 1 When
t exceeds t 1 , the body continues to move upward, to y = H at t = t2 , where
dy/dt = 0. Integrating Newton's second law with F = 0, and satisfying the
continuity conditions at y = y 1 [namely, y = y 1 and dy/dt = (dy/dt)" ' ' we
find that (FIM)t 1 = gt2 . The body falls to the ground during the time i nterval
t3 = (2H/g) 1 1 2 _
Next, we note that t 1 < t2 , so that in an order-of-magnitude sense, t2 = t3
Eliminating t1 and t2 from the equations derived above yields the result (FI
Mg) 2 = (Hly 1 ) (F/Mg - 1 ). Because y 1 scales with H, and H > y 1 , the final
result is F = (H/y 1 )gM. In conclusion, the force produced by the leg while
running at optimal speed is a multiple of the body weight. In the next two
sections, we show that the same theoretical force characterizes flying and
swimming. Figure 1 3.52c shows that these predictions are supported by the
large volume of data on the maximal force produced by animal motors, which
are closely c01Telated by F = 2gM over animal body sizes ranging from small
insects to large mammals [ 1 08].
13.8.2
Flying
The proportionality between speed and M 1 1 6 has been examined previously

using constructal theory, to predict speeds of animal and machine flight [5]
(cf. Section 1 3 . 7.2). That the same proportionality rules optimal running is
not a coincidence; rather, it is an illustration of the unifying power of con
structal theory. Running requires least food when during each cycle a certain
amount of work is destroyed by vertical impact and a certain amount by
812
horizontal f1iction. The same balancing act is responsible for optimal flight:
W1 is the work (MgH) required to lift the body that had fallen to the vertical
distance H during the cyclic time interval t - (Hlg) 1 1 2 During the same
period, the work spent on overcoming drag is W2 - F0L, where and C0 1 . Cycles in which the vertical and horizontal losses (W1 , W2) alternate to
maintain cruising at constant altitude are sketched in Fig. 1 3 .53a. The total
work spent per distance traveled is
V(;1 1 2
'--v-.1
PcY2Lt
Vertical loss
( 1 3. 1 0 1 )
Horizontal loss
The altitude increment (H) achieved during each stroke of the wing is dictated
by the wing length scale, which is the length scale of the flying body, H L1;. From eq. ( 1 3 . 1 0 1 ) we learn that the spent work is minimal when (Fig.
1 3.52a)
- FD
1
i-- L -I
(a)
..
'
----------
M
water
..
Figure
13.53
..
'
.. - - - - - - - - - -
- FD
f
jw,
- - - -
------
M
water
(a) Periodic trajectory of a flying animal; (b) cyclical progress of a
swimming animal.
()
CONSTRUCTAL THEORY OF RUNNING, S WI M M ING, AND FLYING
Vopt
Pb
Pa
1 /3
g 1 1 2p-b 1 1 6M 1 1 6
813
( 1 3 . 1 02)
The wing flapping frequency that corresponds to this optimal flying speed is
t- 1 - (g/Lb) 1 1 2 , which is the formula shown in Fig. 1 3 .52b. The dimensionless
frequency is the Strauhal number,
( 1 3. 103)
which for optimal flight becomes a constant: St - (p)p1,) ' 1 3 - 1 0- 1 This
agrees with the large volume on St data of animal flight [ 1 1 8] .
13.8.3
Swimming
Swimming exhibits the same body-mass scaling as running and flying, not
because of a coincidence, but because swimming is thermodynamically anal
ogous to running and flying. Swimming is another example of optimal dis
tribution of imperfections in time, or the optimization of intermittency (cf.
Ref. 5, Chap. 1 0) . The analogy with flying is shown in Fig. 1 3 .53. The vertical
loss, W1 - MgLb, is the work spent by the fish to l ift above itself the body
of water (M, the same as the fish mass) that it displaces during one cycle.
The duration of the cycle, t - (L1)g) 1 1 2 , is the time in which the lifted water
mass falls, to occupy the space just vacated by the fish. During this time, the
fish (M) and its water-body partner (M) can be thought of as a large-scale
eddy (Section 1 3 .3.6) that will dissipate W1 in time and space, in the wake.
The fish mass M is as much a part of the eddy as is the water mass M.
During the same time interval, the fish also overcomes drag by performing
the work W2 - FD L, where - Vt, FD - pb V2LCD and CD - 1 . The fish
body density Pb is the same as the water density. In sum, the total work spent
is
p,,V7-L0t,
MgLb
+
V(Lb/g) 1 1 2
'--v----'
Vertical loss
( 1 3 . 1 04)
Horizontal loss
1
1 2 1
op
The optimal swi1runing speed is V0r, - g 1 p/; 1 6M 1 6 (Fig. 1 1 3 . 52a) . The
2pb 1 6M- 1 1 6 (Fig.
vopJ
c
1
L
g
timal undulating frequency of the body is
l 3.52b), with the corresponding Strauhal number St = c 1 L" I V0r, - 1 , which
agrees in an order-of-magnitude sense with all known observations.
The corresponding minimum work per distance traveled is 2gM. The force
2gM plotted in Fig. 1 3.52c is the order of magnitude of the average force
814
exerted by the fish, which is remarkably close to the maximum force indicated
by the empirical data in the figure. The average force scale 2gM also holds
for flying [5, p. 239] and is comparable with the maximum force estimated
in this article for running, (Hly 1 )gM. This is why in Fig. 1 3 .52c the line F
= 2gM is compared with the force data for all forms of animal locomotion.
(Note that this differs from the -6gM figure for motor force output [ 1 08 , 1 1 4]
because in the present case M refers to total body mass rather than motor
mass, and animal motors average about 25 to 33 percent of body mass).
To put swimming in the same theory with flying and running (Fig. 1 3 .52)
may seem counterintuitive, because everybody knows that fish are neutrally
buoyant and birds are not. This intuition has delayed the emergence of a
theory that unifies swimming with the rest of locomotion. There is a "ground
effect" in swinm1ing and running as well as in flying. The water mass M can
only be displaced upward because the lake bottom is rigid. Said another way,
the only sp1ing in which the fish can store (temporarily) its stroke work W1
is the gravitational spring of constant force Mg and vertical displacement L1,.
The swimming cycle sketched in Fig. 1 3 .53b is identical to that of a
shallow-water gravitational wave of depth L1;, wavelength 2L" , and horizontal
speed V - (gL,,) 1 1 2 , which is also the speed of a tsunami and a hydraulic
jump. This is not a coincidence. The fact that this wave speed is the same as
the optimized swinm1ing speed (g 1 1 2p/; 1 1 6M1 1 6) and the observed swimming
speeds of fish [ 1 02] and marine mammals [ 1 1 9] (Fig. 1 3 .52a) provides ad
ditional support for the hypothesis that fundamentally, swinuning is an opti
mized intermittent movement in the gravitational .field, like flying and
running.
13.8.4
Locomotion and Turbulent Structure
A new theory predicts, explains, and organizes a body of knowledge that was
growing empirically. This we have done in this section by bringing under the
constructal law the cruising speeds and frequencies of running, flying, and
swimming. Animal locomotion is no different than other flows, animate and
inanimate: They all develop (morph, evolve) architecture in space and time
(self-organization, self-optimization), so that they optimize the flow of matter
in nature.
In the past it made sense to describe the flapping frequencies of swimmers
and flyers in terms of the Strauhal number (St). It made sense because such
animals generate eddies, and because St is part of the language of turbulent
fluid mechanics. After constructal theory, one can also talk about the Strauhal
number of runners, St = c 1 L1;! V0pl' which in view of the first part of the
analysis turns out to be a constant in the range 0. 1 to 10, just as for swimmers
and flyers. The St constant is an optimization result of constructal theory, and
it belongs to all flow systems with optimized intermittency, animate and in
animate.
SCIENCE AND CIVILIZATION AS CONSTRUCTAL FLOW SYSTEMS
815
All animals, regardless of their habitat (land, sea, air) mix air, water, and
soil much more efficiently than in the absence of flow structure. It sounds
crude, but when all is said and done, this is what living flow systems accom
plish and why their legacy is the same as that of the rivers and the winds (cf.
Section 1 3 .6). Constructal theory has already predicted the emergence of tur
bulence by showing that an eddy of length scale L1,, peripheral speed V, and
kinematic viscosity v transports momentum across its body faster than laminar
shear flow when the Reynolds number Lb Vi v exceeds approximately 30 (cf.
Section 1 3 .3.6) . This agrees very well with the zoology literature, which
shows that undulating swimming and flapping flight (i.e., locomotion with
eddies of size L1,) is possible only if L,yI v is greater than approximately 30
[ I 20] .
So we conclude with the most unexpected link that this simple physics
theory reveals : The generation of optimal distribution of imperfection (optimal
intermittency) in running, swimming, and flying is governed by the same
principle as the generation of turbulent flow structure. The eddy and the an
imal that produces it are the optimized "construct" that travels through the
medium the easiest and mixes Earth's crust most effectively.
In sum, a thermodynamic approach can predict complex features of animal
design. This differs markedly from one of the major viewpoints of evolution
ary biology, which maintains that evolution involves a large degree of chance
and historical contingency, to the extent that if the evolutionary process was
reset to its beginning, a very different set of creatures would evolve. What
our work shows is that an evolutionary process would produce runners, swim
mers and fliers with these speeds, stroke/ stride frequencies and force outputs.
The theory even predicts how these features would evolve on another planet
that has a different gravitational force and density of the gaseous and liquid
environment.
13.9
SCIENCE AND CIVILIZATION AS CONSTRUCTAL

FLOW SYSTEMS
Contemporary fields that deal with flow systems (turbulence, hydrology, bi

ology, etc.) are proving every day that science has hit a wall. Principles such
as Newton's Second Law of Motion (the Navier-Stokes equations) are not
enough. Because of progressively more powerful computational and infor
mation gathering tools, models are becoming more numerous and more com
plex, with more empirical features to be fitted to more measurements. They
provide better description, not explanation. They do not provide a mental
viewing of how things should be. They are many trees, but no forest.
Needed are principles with the same universal reach as that of Newton's
Second Law of Motion and the First and Second Laws of Thermodynamics .
Needed are new laws of physics. A prerequisite for success o n this path is a
816
new attitude, a new way of thinking: Physics is not and never will be com
plete. There is no such thing as "the end of physics."
Physics is our knowledge of how nature works. Physics (or nature) is
everything, including engineering: the biology and medicine of human +
machine species. Our knowledge is condensed in simple statements (thoughts,
connections), which evolve in time by being replaced by simpler statements.
We "know more" because of this evolution in time, not because brains get
bigger and neurons smaller and more numerous. Our finite-size brains keep
up with the steady inflow of new information through a process of simplifi
cation by replacement: in time, and stepwise, bulky catalogs of empirical
information (e.g., measurements, data, complex empirical models) are re
placed by much simpler summarizing statements (e.g., concepts, formulas,
constitutive relations, principles, laws). A hierarchy of statements emerges
along the way: It emerges naturally, because it is better (cf. the constructal
law).
The simplest and most universal are the laws. The bulky and the laborious
are being replaced by the compact and the fast. In time, science optimizes
and organizes itself in the same way that a river basin evolves: toward con
figurations (links, connections) that provide faster access, or easier flowing.
The bulky measurements of pressure drop versus flow rate through round
pipes and saturated porous media were rendered unnecessary by the formulas
of Poiseuille and Darcy. The measurements of how things fall (faster and
faster, and always from high to low) were rendered unnecessary by Galilei's
principle and the Second Law of Thermodynamics.
The hierarchy (specialization) that science exhibited at every stage in the
history of its development is an expression of its never-ending struggle to
optimize and redesign itself. Hierarchy means that measurements, ad hoc
assumptions, and empirical models come in huge numbers, a "continuum"
above which the compact statements (the laws) rise as needle-shaped peaks.
Both are needed, the numerous and the singular. One class of flows (infor
mation links) sustains the other. The many and unrelated heat engine builders
of Britain fed the imagination of one Sadi Carnot. In turn, Sadi Carnot's
mental viewing (thermodynamics today) feeds the minds of contemporary and
future builders of all sorts of machines throughout the world.
Civilization with all its constructs (science, religion, language, and writing,
etc.) is this never-ending physics of generation of new configurations, from
the flow of mass, energy, and knowledge to the world migration of the special
persons to whom ideas occur (the creative). Good ideas travel. Better flowing
configurations replace existing configurations (the constructal law). Empirical
facts (observations) are extremely numerous, like the hill slopes of a river
basin. The laws are the extremely few big rivers, the Seine and the Danube.
13.10
FREEDOM IS GOOD FOR DESIGN
In Refs. 1 33 to 1 37 we pursued in great detail the morphing and generating

of progressively better tree-shaped flow networks. We did this by focusing on
817
trees that connect a circle with its center. The summary given in Fig. 1 3 .49
shows that when there are 1 92 equidistant points on the circle, the equilibrium
flow architecture has a certain (optimized, finite) degree of complexity, which
is represented by six levels of bifurcation or pairing.
More recently we discovered [ 1 38] that the circle-point tree architectures
can be morphed and improved further, so that their points migrate even farther
to the left on the performance versus freedom domain of Fig. 1 3 .49. The true
equilibrium flow architecture lies slightly to the left of point d indicated in
Fig. 1 3 .49. The path to this higher level of performance is made possible by
increasing the freedom of the morphing architecture. How this is done is
explained in Fig. 1 3 .54.
The upper-left frame of Fig. 1 3 .54 il!i.1strates the type of circle-point trees
that has been studied by a vast literature until now. It is based on the reason
able assumption that at each level of pairing or bifurcation the mother tube
splits into two daughter tubes that have the same length. This assumption is
so "reasonable" and so popular (no doubt, because of fractal geometry) that
no one questioned it. In truth, however, any simplifying assumption that the
designer makes is a straight jacket-it curtails the freedom of the morphing
architecture.
*
I
'
Figure 13.54
More freedom to morph leads to higher performance and asymmetry in

the circle-point tree networks of Fig. 1 3 .49. (After Ref. 1 3 8 . )
818
The other frames of Fig. 1 3.54 show what happens to the tree architecture
when the assumption of daughter tubes of equal length is not made. We have
discovered in this way the emergence of several types of asymmetry in equi
librium tree architectures: different tube lengths at the same level of branch
ing, different mass flow rates at junctions, and different main branches [ 1 38].
The emergence of asymmetry in the best tree networks (human-made or nat
ural) is the fingerprint of the constructal law in action
In constructal theory, we have seen tree asymmetry earlier during the com
pletely free search for a tree layout (A 3 ) that consists of triangular and hex
agonal area elements (A0) (Fig. 1 3.55). The fact that some branches appear
unlike the others (e.g., single branch attached laterally to a stem) is a state
ment of perfection, not imperfection. And so, now we understand why Mau
roy et al. [1 39] wrote that optimal bronchial trees may be "dangerous" (these
authors found that the human lung is not built exactly like the "optimal"
one). The explanation is that what Mauroy et al. took as optimal is a sym
metrically bifurcated structure, as in the upper-left frame of Fig. 13.54. The
natural tree does not look and perform in exactly the same way because it is
more free to morph (i.e., because it is better). What appears to be a slight
deviation from the constructal law is a loud and finely tuned (strident ! ) en
dorsement of the constructal law.
Freedom is good for the performance and survival of a flow structure, be
that natural or engineered, animate or inanimate, human or animal society,
and so on. Design improvement without freedom to change the structure is
nonsense. Rigid flow structures are brittle: dictatorial schemes and straight
river channels are short lived. I thank my son William for this passage from
Sophocles, which proves that this idea is as old as the Greeks (i.e., as old as
science itself) [ 1 40] :
is
A man who thinks that he alone is right, or what he says, or what he

himself,
unique, such men, when opened up, are seen to be quite empty. For a man, though
he be wise, it is no shame to learn-learn many things, and not maintain his views
Figure 13.55 Emergence of asymmetry in the optimal layout of trees consisting of

triangular and hexagonal area elements. (From Ref. 1 34.)
819
too rigidly. You notice how by streams in wintertime the trees that yield preserve
their branches safely, but those that fight the tempest perish utterly.
Although space limitations do not permit a review of how the field has
grown lately, it is worth showing how free thinking has taken the field in new
directions. Upham and Wolo [ 1 4 1 ] propose the use of constructal theory to
improve the delivery of medical care to the most impoverished regions of the
world. Hernandez et al. [ 1 42] used the maximization of flow access to design
platforms for industrial production (see also Refs. 2 1 to 23). Brod [ 1 43] used
an optimal distribution of imperfection (the wall shear rate) approach to de
termine the geometrical details of a tree-shaped network that distributes pol
ymer melts such that the polymer residence time is minimized throughout the
network. Tondeur and Luo [ 144] used the same philosophy to determine the
optimal hierarchy of scales for tree-shaped fluid distributors.
Constructal trees of all scales are invading rapidly the research journals
that deal with water distribution and collection in urban design [ 145- 1 48],
cooling and miniaturization of electronics [ 1 33- 1 38, 1 49- 1 52], cooling trees
at nanoscales [ 1 53], tree-shaped fins and heat spreaders [ 1 54- 1 57], and tree
shaped heat exchangers [ 1 5 8, 1 59]. Constructal trees are now actively pursued
in the conceptual design of new configurations for fuel cells (e.g., Refs. 1 05
to 1 09 in Chapter 1 1 ). Azoumah et al. [ 1 60] broke new ground by applying
constructal theory to the design of tree architectures for chemically reactive
porous media with combined heat and mass transfer. Muzychka [ 1 6 1 ] devel
oped the constructal geometry of parallel microchannels with forced convec
tion and various cross-sectional shapes. The constructal-theory concept of
multiple-length scales arranged optimally (nonuniformly) in a package bathed
by a stream (Ref. 1 1 0 of Chapter 1 1 ) was applied to packages of multidi
ameter cylinders in cross-flow [ 1 62].
Considerable attention is given to the mating-trees concept proposed in
Ref. 1 63 (see also Ref. 5, Chap. 8), where a space is invaded by one stream
configured as a tree, which later is reconstituted as a single stream of used
fluid, which leaves the space. Seen from the outside, the space has one inlet
and one outlet. Inside this space, the inflowing tree and the outflowing tree
are matched canopy to canopy tlu-oughout the available space. This concept
has been applied to two-phase cooling of a large surface [ 1 64] , micro heat
exchangers [ 1 65 , 1 66] and cold storage [ 1 67].
Most rewarding is to see papers that take constructal theory in directions
that are entirely new. Reis et al. [ 1 68] used constructal theory to predict for
the first time that the human lung should have 23 levels of bifurcation. Pra
manick and Das developed the constructal design of thermal insulation sys
tems [ 1 69] and thermoelectric devices [ 1 70] . Teresa Mia Bejan [ 1 7 1 ] made
constructal law a part of the philosophy literature, and showed the relation
between it and natural law and natural design. The place of constructal theory
in the philosophy of science was described by Rosa et al. [7] and Kremer
Marietti (see the Foreword to Ref. 1 1 ). The relationship between the construc
tal law, optimal distribution of imperfection, and the equipartition principle
820
in design was described by Pramanick and Das [ 1 72] (see also Lewins [9]).
The hand of the constructal law in the natural shaping of bodies that move
through air (already noted in the constructal theory of flight, Section 1 3 .7.2)
is visible to everyone who thinks about the shape of a falling droplet [ 173].
REFERENCES
1 . A. Bejan, Street network theory of organization in Nature, J. Adv. Tra11Sp. Vol.
30, No. 1 , June 1 996, pp. 85- 1 07.
.
2. A. Bejan, Constructal-theory network of conducting paths for cooling a heat

generating volume, Int. J. Heat Mass Transfe1; Vol. 40, 1 997, pp. 799-8 1 6, issue
published on Nov. 1 , 1 996.
3. A. Bejan, Theory of organization in Nature : pulsating physiological processes,
Int. J. Heat Mass Tra11f"e 1; Vol. 40, 1 997, pp. 2097-2 1 04, issue published Feb.
4, 1 997.
4. A. Bejan, Constructal tree network for fluid flow between a finite-size volume
and one source or sink, Int. J. Therm. Sci., Vol. 36, 1 997, pp. 592-604.
5. A. Bejan, Shape and Structure, .fimn Engineering to Nalure, Camb1idge Univer

sity Press, Cambridge, 2000.
6. H. Poirier, Une theorie expligue ] ' intelligence de la nature, Sci. Vie, No. 1 034,
Nov. 2003, pp. 44-63.
7. R. N. Rosa, A. H. Reis, and A. F. Miguel, Bejan 's Constructal TheOIJ' of Shape
and Structure, Evora Geophysics Center, University of Evora, Portugal, 2004.
8. N. Torre, La Natura, vi svelo le formule della perfezione, Macch. Tempo, Vol .
5 , N o . 1 -2, 2004, pp. 36-46.
9. J. Lewins, Bejan's constructal theory of equal potential distribution, Int. J. Heat

Mass Tra11fe1; Vol. 46, 2003, pp. 1 54 1 - 1 543.
1 0. A. Bejan, I. Dincer, S. Lorente, A. F. Miguel, and A. H. Reis, Porous and Com

plex Flow Structures in Modern Technologies, Springer-Verlag, New York, 2004.
1 1 . A. Bejan and S . Lorente, La Loi constructale, translated by A. Kremer-Marietti,

L'Harmattan, Paris, 2005.
1 2. A. H. Reis, Constructal theory: from engineering to physics, and how flow sys
tems develop shape and structure, Appl. Mech. Rev. , Vol. 59, 2006, No. 5.
1 3 . A. Bejan and S. Lorente, A course on flow-system configuration and multi-scale
design, Paper IMECE2004-59203, presented at the International Mechanical En
gineering Congress and Exposition, Anaheim, CA, Nov. 1 4- 1 9, 2004.
1 4. A. Bejan, Convection Heal Tra11f"e1; 3rd ed., Wiley, New York, 2004, p. 58.
15. G. A. Ledezma, A. Bejan, and M. R. Errera, Constructal tree networks for heat
transfer, 1. Appl. Phys. , Vol. 82, 1 997, pp. 89- 1 00.
1 6. M . R. Errera and A. Bejan, Deterministic tree networks for river drainage basins,
Fractals, Vol . 6, 1 998, pp. 245-26 1 .
1 7. A. Bejan and S. Lorente, Thermodynamic formulation of the constructal law,
ASME Paper IMECE2003-4 1 1 67, presented at the International Mechanical En
gineering Congress and Exposition, Washington, DC, Nov. 1 6-2 1 , 2003.
REFERENCES
821
1 8. A. Bejan and S. Lorente, The constructal law and the thermodynamics of flow
systems with configuration, Int. J. Heat Mass Transfe1; Vol. 47, 2004, pp. 320332 14.
1 9. G. A. Ledezma and A. Bejan, Streets tree networks and urban growth: optimal
geometry for quickest access between a finite-size volume and one point, Physica
A, Vol . 255, 1 998, pp. 2 1 1 -2 1 7.
20. G. A. Ledezma and A. Bejan, Constructal three-dimensional trees for conduction
between a volume and one point, J. Heat Transfe1; Vol . 1 20, 1 998, pp. 977-984.
2 1 . G. Hernandez, Design of platforms for customizable products as a problem of
access in a geomet1ic space, Ph.D. dissertation, Georgia Institute of Technology,
Atlanta, GA, 200 1 .
22. M . J . Carone, Applying constructal theory for product platform design i n the
context of group decision-making and uncertainty, M.S. thesis, Georgia Institute
of Technology, Atlanta, GA, 2003.
23. M. J. Carone, C. B. Williams, J. K. Allen, and F. Mistree, An application of
constructal theory in the multi-objective design of product platforms, ASME
Paper DETC2003 / DTM-48667, Proceedings of DETC 'OJ, ASME 2003 Design
Engineering Technical Co1iferences and Computers and h\/ormation in Engi

neering Co1f'erence, Chicago, Sept. 2-6, 2003.
24. A. Bejan, Heat Transj'e1; Wiley, New York, 1 993.
25. M. Neagu and A. Bejan, Constructal-theory tree networks of "constant" thermal
resistance, ]. Appl. Phys. , Vol. 86, 1 999, pp. 1 1 36- 1 1 44.
26. L. Ghodoossi and N. Egrican, Exact solution for cooling of electronics using
constructal theory, J. Appl. Phys. , Vol. 93, 2003, pp. 4922-4929.
27. D. L. Cohn, Optimal systems; I: The vascular system. Bull. Math. Biophys. , Vol.
1 6, 1 954, pp. 59-74.
28. D' A. W. Thompson, On Growth and Form, Cambridge University Press, Cam
bridge, 1 942.
29. B. B . Mandelbrot, The Fractal Geometry of Nature, W.H. Freeman, New York,
1 982.
30. I. Prigogine, From Being to Becoming, W.H. Freeman, S an Francisco, 1 980.
3 1 . L. F. Richardson, Atmospheric diffusion shown on a distance-neighbor graph,
Proc. R. Soc. London, Vol. A 1 1 0, No. 756, 1 926, pp. 709-737.
32. A. L. Bloom, Geomo1phology, Prentice-Hall, Englewood Cliffs, NJ, 1 978, p. 204.
33. E. R. Weibel, Mo1phometry of the Human Lung, Academic Press, New York,
1 963.
34. E. R. Weibel, Design of biological organisms and fractal geometry, Fractals in
Biology and Medicine, in T. F. Nonnenmacher, G. A. Losa, and E. R. Weibel,
eds., B irkhi.iuser Verlag, Basel, Switzerland, 1 994, pp. 69-85.
35. G. S. Krenz, J. H. Linehan, and C. A. Dawson, A fractal continuum model of
the pulmonary arterial tree, J. Appl. Plzysiol. , Vol. 72, 1 992, pp. 2225-2237.
36. M. Sernetz, M. Justen, and F. Jestczemski, Dispersive fractal characterization of
kidney arteries by three-dimensional mass-radius-analysis, Fractals, Vol . 3, 1 995,
pp. 879-89 1 .
37. N. MacDonald, Trees and Networks in Biological Models, Wiley, Chichester,
West Sussex, England, 1 983.
822
38. T. A . McMahon, Tree structures: deducing the principle of mechanical design,

J. The01: Biol., Vol. 59, 1 976, pp. 443-466.
39. L. B . Leopold, M. G. Wolman, and J. P. Miller, Fluvial Processes in Geomor
phology, W.H. Freeman, San Francisco, 1 964.
40. A. E. Scheidegger, Theoretical Geomorphology, 2nd ed., Springer-Verlag, Berlin,
1 970.
4 1 . R. J. Chorley, S. A. Schumm, and D. E. Sugden, Geomorphology, Methuen,
London, 1 984.
42. M. Morisawa, Rivers: Form and Process, Longman, London, 1 985.
43. S. A. Schumm, M. P. Mosley, and W. E. Weaver, Experimental Fluvial Geo
morphology, Wiley, New York, 1 987.
44. M. R. Errera and A. Bejan, Tree networks for flows in composite porous media,
J. Porous Media, Vol . 2, 1 999, pp. 1 - 1 8 .
45. Z.-Z. Xia, Z.-X, Li, and Z.-Y. Guo, Heat conduction optimization: high conduc
tivity constructs based on the principle of biological evolution, presented at the
1 2th International Heat Transfer Conference, Grenoble, France, Aug. 1 8-23,
2002.
46. E. Yantovski, A. D. Kraus, J. V. C. Vargas, G. Guerreri, T. Basak, S. Kjelstrup,

D. Bedeaux, and J. Cervantes de Gortari, Discussion of the paper " Self
organization and self-similarity in boiling systems" [by L. H. Chai and M. Shoji,
J. Heat Tran,f(n; Vol . 1 24, 2002, pp. 507-5 1 5] , J. Heat Transfe1; Vol . 1 25, 2004,
pp. 660-663.
47. D. A. Nield and A. Bejan, Convection in Porous Media, 3rd ed., Springer-Verlag,
New York, 2006.
48. R. S. Parker, Experimental study of drainage basin evolution and its hydrologic
implications, Hydrology Paper 90, Colorado State University, Fort Collins, CO,
1 977.
49. A. Bejan, Entropy Generation through Heat and Fluid Flow, Wiley, New York,
1 982, Chap. 4.
50. A. Bejan, Theoretical explanation for the incipient formation of meanders in
straight rivers, Geophys. Res. Lett., Vol. 9, 1 982, pp. 83 1 -834.
5 1 . I. Rodriguez-Iturbe, A. Rinaldo, R. Rigon, R. L. Bras, A . Marani, and E. Ijj asz

Vasquez, Energy dissipation, runoff production, and the three-dimensional struc
ture of river basins, Water Resources Res., Vol. 28, 1 992, pp. 1 095- 1 1 03.
52. T. Sun, P. Meakin, and T. Jossang, Minimum energy dissipation model for river
basin geometry, Phys. Rev. E., Vol . 49, 1 994, pp. 4865-4872.
53. A. Maritan, F. Colaiori, A. Flammini, M. Cieplak, and J. R. Banavar, Universality
classes of optimal channel networks, Science, Vol . 272, 1 996, pp. 984-986.
54. D. Avnir, 0. B iham, D. Lidar, and 0. Malcai, Is the geometry of nature fractal?
Science, Vol. 279, 1 998, pp. 39-40.
55. J.-D. Chen, Radial viscous fingering patterns in Hele-Shaw cells, Exp. Fluids,
Vol . 5, 1 987, pp. 363-37 1 .
56. E. B en-Jacob, 0. Shochet, I . Cohen, and A. Tenenbaum, Cooperative strategies
i n formation of complex bacterial patterns, Fractals, Vol . 3, 1 995, pp. 849-868.
57. M. A. Uman, lightning, Dover, New York, 1 984.
REFERENCES
823
5 8 . M. N. Rand and S. M. Breedlove, Androgen alters the dendritic arbors of SNB

motoneurons by acting upon their target muscles, J. Neurosci. , Vol . 1 5, 1 995,
pp. 4408-44 1 6.
59. N. Dan and A. Bejan, Constructal-theory networks for the time-dependent dis
charge of a finite-size volume to one point, J. Appl. Phys. , Vol. 84, 1 998, pp.
3042-3050.
60. A. Bejan, G. Tsatsaronis, and M. Moran, Thermal Design an.d Optimization,
Wiley, New York, 1 996.
6 1 . S. Vogel, Life 's Devices, Princeton University Press, Princeton, NJ, 1 988.
62. A. B ejan, V. Badescu, and A. De Vos, Constructal theory of economics structure

generation in space and time, Energy Convers. Manage., Vol. 4 1 , 2000, pp. 1 4291 45 1 .
63. R. A. Gore, C. T. Crowe, and A. Bejan, The geometric similarity of the laminar
sections of boundary layer-type flows, Int. Commun. Heat Mass Transfe1; Vol.
1 7, 1 990, pp. 465-475.
64. R . A. Nelson, Jr. and A. Bejan, Constructal optimization of internal flow ge
ometry in convection, J. Heat Transfe1; Vol. 1 20, 1 998, pp. 357-364.
65. A. Bejan, Y. Ikegami, and G. A. Ledezma, Constructal theory of natural crack
pattern formation for fastest cooling, Int. J. Heat Mass Transfe 1; Vol. 4 1 , 1 998,
pp. 1 945- 1 954.
66. J . Kepler, The Six-Cornered S11m1'.f/ake, Oxford University Press, Oxford, 1 966
(original in Latin, 1 6 1 1 ).
67. D. A. Kessler, J. Koplik, and H. Levine, Pattern selection in fingered growth
phenomena, Adv. Phys. , Vol. 37, 1 988, pp. 255 -339.
68. H. S. Carlslaw and J. C. Jaeger, Conduction of Heat in Solids, Oxford University
Press, Oxford, 1 959.
69. S . Lorente, Private communication, July 2005.
70. A. Bretagnolle, H. Mathian, D. Pumain, and C. Rozenblat, Long-term dynamics
of European towns and cities: towards a spatial model of urban growth, Cybergeo,
Vol. 1 3 1 , March 29, 2000.
7 1. P. Bairoch, J. Batou, and A. Chevre, The Population of European Cities .fimn
800 to 1 850, Droz, Geneva, 1 988.
72. F. Moriconi-Ebrard, Geopolis, Anthropos-Economica, Paris, 1 994.
73. R. N. Rosa, River basins: geomorphology and dynamics, in Bejan 's Constructal
Theory of Shape and Structure, R. N. Rosa, A. H. Reis and A. F. Miguel, eds.,
E vora Geophysics Center, University of Evora, Portugal, 2004.
74. A. H. Reis, Constructal view of the global circulation of the atmosphere and
flow architectures of river basins and lung tree, in A long with Constructal Theo1y,
A. Bejan, S . Lorente, A. F. Miguel and A. H. Reis, eds., University of Lausanne
Press, Lausanne, Switzerland, 2006.
75. A. B ejan, L. A. 0. Rocha, and S. Lorente, Thermodynamic optimization of ge

ometry: T- and Y-shaped constructs of fluid streams, Int. J. Therm. Sci., Vol. 39,
2000, pp. 949-960.
76. C. D. Linden, The evolution of the pyramid, Term paper, Course ME 1 66, Con
structal Theory and Design, Duke University, Durham, NC, Spring 2005.
824
77. P. Crozat, Le Genie des Pyramides, Doctoral dissertation, National Polytechnical

Institute of Lo1rnine, Nancy, France, 2003.
78. P. Crozat and T. Verdel, Systeme constructif des pyramides: de la geologie a

! ' edification, presented at Journees Nationales de Geotechnique et de Geologie
de l ' lngenieur, JNGG 2002, Nancy, France, Oct. 8-9, 2002.
79. G. Goyon, Le secret des batisseurs des grandes pyramides, Pygmalion, Paris,
July 1 997.
80. A. Losch, The Economics of Location, Yale University Press, New Haven, CT,
1 954.
8 1 . P. Haggett, Locational Analysis in Human Geography, Edward Arnold, London,
1 965.
82. P. Haggett and R. J. Chorley, Network Analysis in Geography, St. Martin's Press,
New York, 1 969.
83. A. Bejan and S. Lorente, Thermodynamic optimization of flow geometry in me

chanical and civil engineering, J. Non-Equilib. Thermodyn. , Vol. 26, 200 1 , pp.
305-354.
84. A. Bejan, The constructal law of organization in nature: tree-shaped flows and
body-size, J. Exp. Biol., Vol. 208, 2005, pp. 1 677- 1 686.
85. A. Bejan, The tree of convective heat streams: its thermal insulation function and
the predicted 3 I 4-power relation between body heat loss and body size, Int. J.
Heat Mass Transfe1; Vol . 44, 200 1 , pp. 699-704.
86. A. Bej an, A general variational principle for thermal i nsulation design, Int. J.
Heat Mass Tran4e1; Vol. 22, 1 979, pp. 2 1 9-228.
87. S. Weinbaum and L. J. Jiji, A new simplified bioheat equation for the effect of
blood flow on the local average tissue temperature, J. Biomech. Eng. , Vol. 1 07,
1 985, pp. 1 3 1 - 1 39.
88. K. Schmidt-Nielsen, Scaling: Why Is Animal Size So Important? Cambridge Uni

versity Press, Cambridge, 1 984.
89. G. B. West, J. H. Brown, and B. J. Enquist, A general model for the origin of
allometric scaling laws in biology, Science, Vol . 276, 1 997, pp. 1 22- 1 26.
90. A. Bej an, Theory of heat transfer from a surface covered with hair, .!. Heat
Trans.fe1; Vol . 1 1 2, 1 990, pp. 662-667.
9 1 . A. Bejan, Optimum hair strand diameter for minimum free-convection heat trans
fer from a surface covered with hair, Int. J. Heat Mass Trans.fe1; Vol. 33, 1 990,
pp. 206-209.
92. H. Tennekes, The Simple Science of Flight, MIT Press, Cambridge, MA, 1 996.
93. E. Frey, H. Tischlinger, M.-C. Buchy, and D. M. Martill, New specimens of

Pterosauria (Reptilia) with soft parts with implications for pterosaurian anatomy
and locomotion, in Evolution and Palaeobiology of Pterosaurs, E. Buffetaut and
J.-M. Mazin, eds., Geological Society, Special Publications, London, 2003, pp.
233-266.
94. A. Bejan and S. Lorente, Thermodynamic optimization of flow architecture: den
dritic structures and optimal sizes of components, ASME Paper IMECE-20023 3 1 58, presented at the International Mechanical Engineering Congress and
Exposition, New Orleans, LA, Nov. 1 7-22, 2002.
REFERENCES
825
95. S . Lorente and A. Bejan, Svelteness, freedom to morph, and constructal multi
scale flow structures, //If. J. Therm. Sci., Vol . 44, 2005, pp. 1 1 23 - 1 1 30.
96. A. Bejan, A. M. Morega, G. B. West, and J. H . Brown, Constructing a theory
for scaling and more, Phys. Today, July 2005, pp. 20-2 1 .
97. R . M . Alexander, Principles of Animal Locomotion. Princeton University Press,

Princeton, NJ, 2003.
98. N. C. Heglund, C. R. Taylor, and T. A. McMahon, Scaling stride frequency and
gait to animal size: mice to horses, Science, Vol . 1 86, 1 974, pp. 1 1 1 2- 1 1 1 3.
99. R. M. Alexander and G. M. 0. Maloiy, St1ide lengths and stride frequencies of
primates, J. Zoo/. London, Vol. 202, 1 984, pp. 577-582.
1 00. N. C. Heglund and C. R. Taylor, Speed, stride frequency and energy cost per
stride: how do they change with body size and gait? J. Exp. Biol., Vol . 1 3 8, 1 988,
pp. 3 0 1 -3 1 8.
1 0 1 . S . M. Gatesy and A. A. Biewener, Bipedal locomotion: effects of speed, size
and limb posture in birds and humans, J. Zoo!. London, Vol. 224, 1 99 1 , pp. 1 271 47 .
1 02. E. G. Drucker, and J. S. Jensen, Pectoral fin locomotion i n striped surfperch, I I :

Scaling swimming kinematics and performance at gait transition, J. Exp. Biol.,
Vol . 1 99, 1 996, pp. 2243-2252.
1 03. C. J. Pennycuick, On the running of the Gnu (Connochaetes taurinus) and other
animals, J. Exp. Biol. , Vol. 63, 1 975, pp. 775-799.
1 04. J. Iriarte-Diaz, Differential scaling of locomotor pe1formance in small and large
ten-estrial mammals, J. Exp. Biol. , Vol . 205, 2002, pp. 2897-2908.
1 05. V. Tucker, Bird metabolism during fl ight: evaluation of a theory, J. Exp. Biol. ,
Vol . 58, 1 973, pp. 689-709.
1 06. J. Lighthill, Aerodynamic aspects of animal flight, 5th Fluid Science Lecture,
British Hydrodynamics Research Association, June 1 974, 30 pp.
1 07. C. H. Greenewalt, The flight of birds: the significant dimensions, their departure
from the requirements of geometrical similarity, and the effect on flight aero
dynamics of that departure, Trans. Am. Philos. Soc., Vol. 65, No. 4, 1 975, pp.
1 -67.
1 08. J. H. Mar-den and L R. Allen, Molecules, muscles, and machines: universal
char-acteristics of motors, Proc. Natl. Acad. Sci., Vol. 99, 2002, pp. 4 1 6 1 -4 1 66.
1 09. M. E. Anderson and J. A. Johnston, Scaling of power output in fast muscle fibers
of the Atlantic cod during cyclical contractions, J. Exp. Biol., Vol . 1 70, 1 992,
pp. 1 43 - 1 54.
1 1 0. T. A. McMahon, Size and shape in biology, Science, Vol . l 79, 1 973, pp. 1 20 1 1 204.
l l l . T. A. McMahon, Using body size to understand the structural design of animals:
quadrupedal locomotion, J. Appl. Physiol. , Vol . 39, 1 975, pp. 6 1 9-627.
l 1 2 . A. A. Biewener and C. R. Taylor, Bone strain: a determinant of gait and speed?
J. Exp. Biol., Vol . 1 23, 1 986, pp. 3 83-400.
1 1 3. A. A. Biewener, Biomechanical consequences of scaling, J. Exp. Biol., Vol . 208,
2005, pp. 1 665- 1 676.
1 1 4. J. H . Marden, Scaling of maximum net force output by motors locomotion, J.
Exp. Biol., Vol . 208, 2005, pp. 1 65 3 - 1 664.
826
1 1 5 . R. M. Alexander and A. J. Jayes, A dynamic similarity hypothesis for the gaits

of quadrupedal mammals, J. Zool. , Vol . 20 1 , 1 983, pp. 1 35 - 1 52.
1 1 6. R. M. Alexander, Optima for Animals, rev. ed., Princeton University Press,
Princeton, NJ, 1 996.
1 1 7. A. Ruina, J. E. Bertram, and M. Srinivasan, A collisional model of the energetic
cost of support work qualitatively explains leg sequencing in walking and gal
loping, pseudo-elastic leg behavior in running and the walk-to-run transition,
TlzeOJ: Biol., June 1 3, 2005 [Epub ahead of print].
1 1 8. G. K. Taylor, R. L. Nudds, and A. L. R. Thomas, Flying and swimming animals
cruise at a Strouhal number tuned for high power efficiency, Nature, Vol. 425,
2003, pp. 707-7 1 1 .
1 1 9. J. J . Roln and F. E. Fish, Strouhal numbers and optimization of swimming by
odontocete cetaceans, J. Exp. Biol., Vol. 207, 2004, pp. 1 633- 1 642.
1 20. S. Childress and R. Dudley, Transition from ciliary to flapping mode in swim
ming mollusc: flapping fight as a bifurcation in Re,,,, J. Fluid Mech., Vol. 498,
2004, pp. 257-28 8.
1 2 1 . G. A. B artholomew and T. M. Casey, Oxygen consumption of moths during rest,
pre-flight warm-up, and flight in relation to body size and wing morphology, J.
Exp. Biol., Vol. 76, 1 978, pp. 1 1 -25.
1 22. R. L. Marsh, Ontogenesis of contractile properties of skeletal muscle and sprint
performance in the lizard Dipsosaurus dorsalis, J. Exp. Biol., Vol . 1 37, 1 988,
pp. 1 1 9- 1 39.
1 23. J. Wakelling and C. Ellington, Dragonfly flight, II: Velocities, accelerations and
kinematics of flapping flight, J. Exp. Biol., Vol . 200, 1 997, pp. 557-5 82.
1 24. J. H. Marden, M. R. Wolf, and K. E. Weber, Aerial performance of Drosophila
melanogaster from populations selected for upwind flight ability, J. Exp. Biol.,
Vol . 200, 1 997, pp. 2747-2755.
1 25 . J. W. Kiceniuk and D. R. Jones, The oxygen transport system in trout (Salmo
gairdneri) during sustained exercise, J. Exp. Biol., Vol . 69, 1 977, pp. 247-260.
1 26. M. L. May, Dependence of flight behavior and heat production on air temperature
in the green darner dragonfly, Anax Junius (Odonata: Aeshnidae), J. Exp. Biol.,
Vol. 1 98, 1 995, pp. 2385-2392.
1 27 . S. A. Arnott, D. M. Neil, and A. D. Ansell, Tail-flip mechanism and size

dependent kinematics of escape swimming in the brown shrimp Cangon crangon,
J. Exp. Biol., Vol. 20 1 , 1 998, pp. 1 7 7 1 - 1 784.
1 28. S . J. Peake and A . P. Farrell, Locomotory behaviour and post-exercise physiology

in relation to swimming speed, gait transition and metabolism in free-swimming
smallmouth bass (Micropterus dolomieu), J. Exp. Biol., Vol . 207, 2004, pp. 1 5 631 575.
1 29. U. K. Muller and J. L. van Leeuwen, Swimming of larval zebrafi sh: ontogeny
of body waves and implications for locomotory development, J. Exp. Biol., Vol .
207, Pt. 5, 2004, pp. 853-868.
1 30. G. A. Bartholomew, J. R. B . Lighton, and G. N. Louw, Energetics of locomotion

and patterns of respiration in tenebrionid beetles from the Namib desert, J. Comp.
Physiol., Vol . 155, 1 985, pp. 1 55- 1 62.
1 3 1 . R . Blickhan and R. J. Full, Locomotion energetics of the ghost crab, II: Me
chanics of the center of mass during walking and running, J. Exp. Biol., Vol.
1 30, 1 987, pp. 1 55- 1 74.
REFERENCES
827
1 32. R. J. Full and M. S. Tu, Mechanics of rapid running insects: two-, four- and six
legged locomotion, J. Exp. Biol., Vol. 1 56, 1 99 1 , pp. 2 1 5-23 1 .
1 33. W. Wechsatol, S . Lorente, and A. Bejan, Optimal tree-shaped networks for fluid
flow in a disc-shaped body, Int. J. Heat Mass Transfe1; Vol. 45, 2002, pp. 49 1 1 4924.
1 34. S. Lorente, W. Wechsatol, and A. Bejan, Tree-shaped flow structures designed

by minimizing path lengths, Int. J. Heat Mass Tra11sfe1; Vol. 45, 2002, pp. 329933 1 2.
1 35 . W. Wechsatol, S. Lorente, and A. Bejan, Dendritic convection on a disc, Int. J.
Heat Mass Transfe1; Vol. 46, 2003, pp. 43 8 1 -439 1 .
1 36. W. Wechsatol, S. Lorente, and A. Bejan, Tree-shaped networks with loops, Int.
J. Heat Mass Tra11sfe1; Vol. 48, 2005, pp. 573-5 83.
1 37. W. Wechsatol, S. Lorente, and A. Bej an, Tree-shaped flow architectures: strate
gies for increasing optimization speed and accuracy, Num. Heat Transfer A, Vol.
48, 2005, pp. 73 1 -744.
1 38. L. Gosselin and A. Bejan, Emergence of asymmetry in constructal tree flow

networks, J. Appl. Phys. , Vol. 98, 2005, 1 04903.
I 39. B. Mauroy, M. Filoche, E. R. Weibel, and B. Sapoval, An optimal bronchial tree
may be dangerous, Nature, Vol . 427, Feb. 1 2, 2004, pp. 633-636.
1 40. Sophocles, Sophocles /: Antigone, translated by D. Grene, 2nd ed., University of
Chicago Press, Chicago, 1 99 1 , p. 690.
1 4 1. G. Upham and J. Wolo, Coordination of dynamic rehabilitation experiments of
people living with post-polio syndrome and call for research into PPS as a for
gotten disease, presented at the Global Forum for Health Research, Forum 8,
Mexico, D.F., Nov. 2004.
1 42. G. Hernandez, J. K. Allen, and F. Mistree, Platform design for customizable
products as a problem of access in a geometric space, Eng. Opt., Vol. 35, No.
3, 2003, pp. 229-254.
1 43. H. Brod, Residence times optimized choice of tube diameters and slit heights in
distribution systems for non-Newtonian liquids, J. Non-Newtonian Fluid Mech.,
Vol. 1 1 1 , 2003, pp. 1 07- 1 25 .
1 44. D. Tondem and L. Luo, Design and scaling laws o f ramified fluid distributors
by the constructal approach, Chem. Eng. Sci., Vol. 59, 2004, pp. 1 799- 1 8 1 3 .
1 45. W. Wechsatol, S . Lorente and A. Bejan, Tree-shaped insulated designs for the
uniform distribution of hot water over an area, Int. J. Heat Mass Tranfe1; Vol.
44, 200 1 , pp. 3 1 1 1 -3 1 23 .
1 46. W. Wechsatol, S. Lorente, and A. Bejan, Development o f tree-shaped flows by
adding new users to existing networks of hot water pipes, Int. J. Heat Mass
Transfe1; Vol . 45 , 2002, pp. 723-733.
1 47 . L. Gosselin, Minimum pumping power fl uid tree networks without a priori flow
regime assumption, Jnr. J. Heat Mass Transje1; Vol. 48, 2005, pp. 2 1 59-2 1 7 1 .
1 48. L. Gosselin and A. Bejan, Tree networks for minimal pumping power, Int. J.
Therm. Sci., Vol. 44, 2005, pp. 5 3-63.
1 49. Y. Chen and P. Cheng, Heat transfer and pressure drop in fractal tree-like mi
crochannel nets, Int. J. Heat Mass Tra11.fe1; Vol. 45, 2002, pp. 2643-2648.
1 50. Y. Chen and P. Cheng, An experimental investigation on the thermal efficiency
of fractal tree-like microchannel nets, Int. Commun. Heat Mass Transfe1; Vol . 32,
2005, pp. 93 1 -938.
828
1 5 1 . D. V. Pence, Reduced pumping power and wall temperature in microchannel heat

sinks with fractal-like branching channel networks, Microscale Thermophys.
Eng. , Vol . 6, 2002, pp. 3 1 9-330.
1 52. A. Y. Alharbi, D. V. Pence, and R. N. Cullion, Fluid flow through microscale
fractal-like branching channel networks, J. Fluids Eng. , Vol. 1 25, 2003, pp.
1 05 1 - 1 057.
1 53. L. Gosselin and A. Bejan, Constructal heat trees at micro and nanoscales, J.
Appl. Phys. , Vol. 96, 2004, pp. 5 852-5859.
1 54. A. D. Kraus, Constructal theory and the optimization of fi n arrays, 2003, pre
sented at the ASME International Mechanical Engineering Congress and Expo
sition, Washington, DC. 1 6-2 1 Nov.
1 55 . A. K. da Silva, C. Vasile, and A. Bejan, Disc cooled with high-conductivity
inserts that extend inward from the perimeter, Int. J. Heat Mass Transfe1; Vol .
47, 2004, pp. 4257-4263.
1 56. M. A. Almogbel, Constructal tree-shaped fins, Int. J. Therm. Sci. Vol. 44, 2005,
pp. 342-348.
.
1 57. G. F. Jones and S. Ghassemi, Thermal optimization of a composite heat spreader,

presented at the ASME Heat Transfer/Fluids Engineering Summer Conference,
Charlotte, NC, July 1 1 - 1 5 , 2004.
1 5 8. A. K. da Silva, S. Lorente, and A. Bej an, Constructal multi-scale tree-shaped
heat exchangers, J. Appl. Phys. , Vol. 96, 2004, pp. 1 709- 1 7 1 8.
1 59. S . M. Senn and D. Poulikakos, Laminar mixing, heat transfer and pressure drop
in tree-like microchannel nets and their application for thermal management in
polymer electrolyte fuel cells, J. Power Sources, Vol . 1 30, 2004, pp. 1 78-1 9 1 .
1 60. Y. Azoumah, N. Mazet, and P. Neveu, Constructal network for heat and mass
transfer in a solid-gas reactive porous medium, Int. J. Heat Mass Transfe1; Vol.
47, 2004, pp. 296 1 -2970.
1 6 1 . Y. S. Muzychka, Constructal design of forced convection cooled microchannel
heat sinks and heat exchangers, Int. J. Heat Mass Transfe1; Vol. 48, 2005, pp.
3 1 1 9-3 1 27.
1 62. T. Bello-Ochende and A. Bejan, Constructal multi-scale cylinders in crossftow,
Int. J. Heat Mass Transfe1; Vol. 48, 2005, pp. 1 373- 1 383.
1 63. A. Bejan and M. R . En-era, Convective trees of fluid channels for volumetric
cooling, Int. J. Heat Mass Transfe1; Vol . 43, 2000, pp. 3 1 05-3 1 1 8 .
1 64. C. Zamfirescu and A. Bejan, Constructal tree-shaped two-phase fl ow for cooling

a surface, Int. J. Heat Mass Transfe1; Vol. 46, 2003, pp. 2785-2797.
1 65. F. Lundell, B . Thonon and J. A. Gruss, Constructal networks for efficient cooling/
heating, presented at the 2nd Conference on Microchannels and Minichannels,
Rochester, NY, 2004.
1 66. M. Lallemand, F. Ayela, M. Favre-Marinet, A. Gruss, D. Maillet, P. Marty, H.
Peerhossaini, and L. Tadrist, Transferts thenniques dans des microcanaux: applications au microechangeurs, Congres Frarn;:ais de Thermique, SFT 2005,
Reims, France, May 30-June 2, 2005.
1 67. C. Zamfirescu and A. Bejan, Tree shaped structures for cold storage, Int. J.
Reji-ig., Vol . 28, 2005, pp. 23 1 -24 1 .
PROBLEMS
829
1 68. A. H. Reis, A. F. Miguel, and M. Aydin, Constructal theory of flow architecture

in the lungs, Med. Phys., Vol. 3 1 , No. 5, 2004, pp. 1 1 35- 1 1 40.
1 69. A. K. Pramanick and P. K. Das, Heuristics as an alternative to variational calculus
for the optimization of a class of thermal insulation systems, Int. J. Heat Mass
Tra11sfe1; Vol. 48, 2005, pp. 1 85 1 - 1 857.
1 70. A . K. Pramanick and P. K. Das, Constructal design of a thermoelectric device,
Int. J. Heat Mass Transfe1; Vol. 49, 2006, pp. 1 420- 1 429.
1 7 1 . T. M. Bejan, Natural law and natural design, Dogma, http: I I dogma.free.fr/txt/

TB-Natura!Design.htm, Feb. 25, 2005.
1 72. A. K. Pramanick and P. K. Das, Note on constructal theory of organization in
nature, Int. J. Heat Mass Transfe1; Vol . 48, 2005, pp. 1 974- 1 98 1 .
1 73. Editorial, Constructal theory for obtaining optimal shapes: the example of a water
droplet, le G/UON, No. 2 1 , Dec. 2004, pp. 1 -4, University Joseph Fourier,
Grenoble, France.
1 74. D. S. Miklosovic, M. M. MuITay, L. E. Howle, and F. E. Fish, Leading-edge
tubercles delay stall on humpback whale (Megaptera 11ovaea11gliae), Phys. Flu
ids, Vol . 1 6, No. 5, 2004, pp. L39-L42.
1 75 . J. P. Den Hartog, Strength of Materials, Dover, New York, 1 96 1 .
1 76. J . J . McDonnell, private communication, July 2005.
PROBLEMS
13.1
One tube of length L0 and diameter D0 splits symmetrically into two

tubes of length L 1 and diameter D 1 The free ends of the three tubes
define an isosceles triangle with the base B and side R (Fig. P 1 3 . 1 ) .
The aspect ratio B I R i s a known parameter. An incompressible new
tonian fluid flows from the L0 tube into the L 1 tubes, or in the opposite
/,
RI
I
Lo
Do
\R
\
\
\
I
B
Figure P13.1
830
direction. The flow is laminar and fully developed, and the pressure
drop due to the Y junction is negligible. According to Murray's law
[eq. (3 .87)] the global flow resistance is minimum when D0 1 D 1 =
2 1 1 3 . Minimize the global flow resistance further by selecting the tube
lengths. Report the optimal L01 R and L, IR as functions of B l R. Show
that the optimal bifurcation angle f3 formed between the L, tubes is
74.94 [75], which is independent of the triangle aspect ratio Bl R
[ 1 33].
13.2
The corners of a cube are the outlets of a tree flow that originates
from the center of the cube. There are four central tubes connected
to the center, and each undergoes a symmetric bifurcation (Fig.
P l 3 .2). The flow is laminar and fully developed with negligible junc
tion losses. Determine the optimal tube lengths L0 and L 1 in relation
to the cube side B, so that you may draw to scale the optimal tree
network. Note that this is a special case of the more general solution
described in Problem 1 3. 1 .
13.3
The minimization of global flow resistance in Problem 1 3.2 leads to

the conclusion that L0I R = 0.063 1 9 and L, IR = 0.949 1 . Calculate
the overall flow resistance of this design. Next, note that the optimal
L0 is so small that perhaps this complication may be omitted. The
simpler design is to have eight straight tubes of length R connecting
the center of the cube with its corners. Calculate the flow resistance
of this simpler design, and make sure that the total tube volume is
the same as in the design with bifurcations. How much better is the
design with bifurcations relative to the design with eight radial tubes?
B1
Figure P13.2
PROBLEMS
831
13.4
A vertical slender structure of variable cross-sectional area A(z) must

be shaped such that the compressive stress at every level is the same.
The base area A0 and the height H are specified. Assume that the
structure is solid with the density p. Determine the shape A(z), the
top cross-sectional area A H, the volume of the structure, and the
weight of the superstructure that is supported by A H .
13.5
Recent and very interesting research on the flippers of humpback

whales [ 1 74] leads to the idea to replace the two-dimensional design
of an airplane wing with a three-dimensional design that has bulbous
features on the front of the wing (Fig. P 1 3 .5). If the frontal portion
of the wing is approximated by a long cylinder of diameter De, the
proposal is qualitatively the same as replacing this material with sev
eral equidistant spheres of diameter D., and spacing S. The approach
air velocity is V. In the Reynolds number range 1 OJ < VDc..J v < 1 05 ,
the sphere and cylinder drag coefficients are C.1 == 0.4 and C, == I
(see Ref. 1 4, p. 365). Evaluate the merit of this proposal by comparing
the drag forces experienced by the sphere and cylinder segment, alone
(i.e., by neglecting the rest of the wing).
13.6
A tree can be described approximately as a body of revolution with

two main regions, a canopy of diameter DJx) and a trunk of diameter
D,(x) (Fig. P 1 3 .6). The canopy feels the drag force exerted by a hor
izontal wind of speed V. Expressed per unit length along the trunk,
this force is F' = C0D, 1 pV2 , where C0 == 1 . Assume that the canopy
I
l
s
1 (1
'.
T
[/)
Bulbous
front
Two-dimensional
front
Figure P13.5
832
l v
Figure P13.6
shape is De = ax", where a is constant and the shape parameter n is
unknown. Determine the corresponding shape of the trunk D1(x) such
that the maximum bending stress along the trunk is independent of
x. Show that when the canopy is conical (n = 1 ) the trunk is also
conical.
13.7
The temperature of gas-turbine blades is controlled by circulating cool

air through channels made longitudinally in the body of the blade.
Figure P. 1 3 .7 shows a two-dimensional model of the technique. The
q"
Tm
L
Tm
- -
2
. fl
m
-
"
ax
T
T I
ED
1
I
.
Figure P13.7
PROBLEMS
13.8
833
heat input received from the hot gases of combustion can be modeled
as a curtain of uniform heat flux q". The channels are round with
diameter D and center-to-center spacing S. The blade is a two
dimensional conducting slab of thickness H (fixed) and thermal con
ductivity k. The hole centers are on the axis of symmetry of the slab
cross section. The hole surface is isothermal at temperature Tmin .The
hot spots ( T111a,) occur on the external surface at the points situated
the farthest from the holes. The objective is to minimize T111ax by
selecting the best flow configuration (S, D) subject to the constraint
that the volume fraction ( </>) occupied by all the channels is fixed.
Assume that to minimize T111..x - T111in is approximately the same as
minimizing the length L of the straight path from Tnrnx to Tmin This
approach is known as the minimal length method [ 1 34] and represents
the near-optimal allocation of one flow path length (L) to elemental
area H x S/2. Determine analytically the optimal channel diameter
and spacing in the range <p < < 1 .
The hot-spot temperature of the configuration optimized in Problem
1 3 . 7 can be reduced further by installing two smaller D2 channels
halfway between two of the original D 1 channels. The new configu
ration is shown in Figure P l 3.8. In every elemental volume of cross
section H x S there are two D2 channels and one D 1 channel. The
volume fraction (</>) occupied by all the channels is fixed. The new
hot spots occur at a distance x from the transversal plane in which
the D2 centers are located. The heat flux that enters the hot spot splits
into two equal currents as it flows toward the two heat sinks, D 1 and
D2 Approximate the respective flow paths as the straight segments
r--x--j f -x r-
"
q
\I
I
J_ . -
"
\. _ .
.L
r- f
Figure P13.8
T
1
DI
834
L 1 and L2 [ 1 34] : consequently, the equipartition of the heat flux means

that L, = L2 ( = L). The minimization of (T111ax - T111i,,) is the same as
minimizing L. Nondimensionalize all the dimensions with respect to
H (fixed), and determine numerically the optimal configuration in
dimensionless terms (D 1 I H, D2 1 H, SI H, xi H, yl H).
13.9
Lorente et al. [ 1 34] showed that near-optimal fluid-flow trees can be

constructed by minimizing the length of every tube on its allocated
area. The rectangular area A shown in Fig. P l 3.9 is crossed by a
round tube of diameter D. The tube connects one corner with the
middle of the opposing side. The tube volume V is fixed. The area A
is constant, but its shape (xly) may vary.
(a) Assume fully developed laminar flow and show that the flow re
sistance of the tube is governed by the geometric group R =
LID'\ where L is the tube length. Find the rectangle shape for
which R is minimal.
(b) Show that if the flow is turbulent in the fully rough regime (f
constant), the flow resistance is proportional to R = LI D5 . Find
the rectangle shape for which R is minimal.
=
13.10
The rectangular territory shown in Fig. P 1 3 . 10 has the size fixed (S

= AB), but its shape Al B may vary. The rectangular area S houses a
Y-shaped construct of three round ducts, one of length L 1 and diam
eter D 1 and two of length L2 and diameter D2 The Y connects one
corner of the rectangle with two points on the side B. The distance
between the two points is B 12, and the distance between one of these
points and the nearest corner is BI 4. The j unction between the L2
ducts and the L1 duct is located at the distance H from the side B.
The total tube volume is fixed.
I
l
\ 111
Figure P13.9
PROBLEMS
835
B
2
i-- s
Figure P13.10
The objective is to detennine the Y configuration for minimum

flow resistance. The flow through each pipe is laminar and fully de
veloped. Junction pressure drops are assumed negligible. Use results
from the text and earlier problems. For example, the optimal ratio of
diameters is D2 /D 1 = 2 - 1 1 3 , and the minimization of global flow
resistance is equivalent to minimizing the expression (L 1 + 2 1 1 3L2).
Use S 1 1 2 as the length scale in the nondimensionalization of all the
lengths. Determine the optimal Y geometry, and report L 1 , L2 , H, A l
B , and angles a1 and a2 . Draw the optimal configuration.
13. 11
In this problem we learn that by optimizing a simpler configuration

we can come close to the performance of the equilibrium flow struc
ture. The equilibrium structure is the one optimized in Problem 1 3 . 1 0.
A simple version of the original morphing Y is when we assume that
L 1 and L2 are always collinear (a 1 = a2 = a) [ 1 34] . In this case the
only degree of freedom is the aspect ratio of the area S or the angle
a. As in Problem 1 3. l 0, begin with the observation that for Hagen
Poiseuille flow through round tubes the optimal ratio of successive
tube diameters is 2- 1 1 3 , and the global flow resistance of the Y-shaped
construct is proportional to R = L , + 2 1 1 3L2 . Show that the optimal
geometry is such that a = 45, and the minimized global flow resis
tance is R = 1 .3 3 1 5 1 1 2 The performance of this nonequilibrium flow
structure is only 0.5 percent inferior relative to the performance of
the optimized structure that was the most free to morph.
836
13 . 12
In eq. ( 1 3 .28) we saw that the minimal thermal resistance of a rec

tangular element can be further reduced by increasing the relative
amount of high-conductivity material, <Po = D01 H0. This cannot go
unquestioned, because when <Po increases the fixed heat current (q')
generated b y A0 ( = H0L0) must come from a vanishingly small space,
( 1 - </J0)A0. A city consisting of paved streets and no living space is
not a good city. To show that there is an optimal trade-off ( c/J0) be
tween the amounts of k,, material and k0 material, recognize that q"'
of eq. ( 1 3.28) refers to the heat generation rate in the k0 material, and
that the total current q' = q"' (A0 - A,,) is fixed. Find the optimal
area fraction <Po such that the dimensionless global resistance D.TJ<c/
q' is minimized.
13 . 13
Everyone knows that the time required to inflate a tire or a basketball

is shorter than the time needed by the compressed air to leak out if
the valve is left open. Think about this as you model the inhaling
exhaling cycle executed by the lungs, and try to explain why the
exhaling time is longer than the inhaling time even though the two
times are on the same order of magnitude. A simple one-dimensional
model is shown in Fig. P l 3. 1 3. The lung volume expands and con
tracts as its contact surface with the thorax travels the distance L. The
thorax muscles pull this surface with the force F. The pressure inside
the lung is P, and outside is P"1111 The flow of air into and out of the
lung is impeded by the flow resistance r, such that Patm - P = ni7i
during inhaling and P - P,11111 = nhu' during exhaling. The exponent
n is a number between n == 2 (turbulent flow) and n = I (laminar
flow). For simplicity, assume that the density of air is constant and
that the volume expansion rate during inhaling (dx! dt) is constant.
The lung tissue is elastic and can be modeled as a linear spring that
places the restraining force kx on smface A. Determine analytically
the inhaling time (t1 ) and the exhaling time (t2 ), and show that t2 /t 1
;::::: 2.
0
I
L
I
I
x .
Patm
,.
Thoax A
I
Inhaling
Patm
--
,;,in
L
I
Patm
0
I
k
Exhaling
Figure P13.13
Patm
Lungs
.--.-.or,.
til out
PROBLEMS
837
13.14
The modern impulse turbine developed based on the Pelton wheel

invention has buckets in which water jets are forced to make 1 80
turns. According to the simple model shown in Fig. P l 3. 14, nozzles
bring to the wheel a stream of water of mass flow rate 111. , room
temperature, room pressure, and inlet velocity V;n (fixed). The bucket
peripheral speed is V. The relative water velocity parallel to V, in the
frame attached to the bucket, is V,.. The outlet stream velocity in the
stationary frame is V0ui Determine the optimal wheel speed V such
that the shaft power generated by the wheel is maximum.
13.15
In the constructal generation of solid structure [5] it was shown that

the material situated in the vicinity of the equator of the beam cross
section is not stressed. By minimizing the amount of such material,
one arrives at cross-sectional shapes where most of the stressed ma
terial is concentrated at distances far from the center (e.g., the I-beam
and rail profile shown in Fig. P l 3 . 1 5 ) . The ultimate design is obtained
by combining the I profile with the optimally tapered design devel
oped in Ref. 5 : The result is the beam shown in Fig. P l 3 . 1 5 , where
one dorsal fiber of cross section A(x) is attached to a ventral fiber of
cross section A(x), by using a rigid web of vanishing thickness and
height H(x). The stress level is specified and uniform ( s111J in both
fibers, tension in the top fiber and compression in the bottom fiber.
The end load P and the end deflection o are specified.
(a) Is the beam geometry unique? Can A(x) and H(x) be specified
independently?
11112 Ill
=::
1hl2 -
v,.
A , viewed i n the radial direcLion
Figure Pl3.14
838
H.
A.
x =L
Figure P13.15
( )I/I
(b) Assume the power-law shapes
A(x) = A:1:
13.16
13.17
x
L
where A,. and H,. are the dimensions in the plane of implantation
(x = L). Show that the required volume of beam material is in
dependent of the assumed shapes and is equal to V = EP8/sw '
where E is the modulus of elasticity.
The leaning tower of Pisa may be approximated as a slender cylin
drical beam implanted into the ground. The tower aspect ratio is
H0 / D0 == 3 .4, where H0 and D0 are the tower height and diameter
[ 1 75 ]. In the Middle Ages and in the modern era until the invention
of portlant cement ( 1 830), tall structures were built as piles of large
stones held together by gravity (dry-stone construction). Such struc
tures were not able to withstand tension. The tower of Pisa is one
structure of this kind. It is now in danger because the stones on the
dorsal (high) side of the leaning tower may be pulled apart by tensile
bending stresses. This would happen in places where the bending
stresses are greater than the compressive stresses that are due to the
weight of the stones of the tower. Determine the maximum angle of
inclination that the stone structure of the tower can withstand before
its dorsal stones separate. Compare this angle with the angle calcu
lated by treating the tower as a solid block that rests on a table (in
this case, the block threatens to tip over when the diagonal of its H0
X D0 longitudinal cross section becomes aligned with the vertical).
Which disaster scenario is more threatening, the bending of a pile of
stones or the tipping of a solid cylinder?
Reexamine Problem 1 3 . 1 6 and think whether by shaping the tower
one could have enhanced its survivability. The answer to Problem
PROBLEMS
839
1 3.6 is that the critical angle for incipient stone separation in a cylin
drical tower is sin a0 = D</ 4H0, and that the fi r st stones to be pulled
apart are at the base. The way to improve the tower design is to
distribute its imperfection (the separation of stones) optimally, that is,
along the entire height. Assume a new tower shape, which is conical ,
D = c (H - y), where c is a constant, D and H are the diameter and
the height, and y is the altitude. The conical tower contains the same
volume of stones as the cylindrical tower. Show that the condition for
stone separation is sin a = c/6, where a is the small angle of incli
nation of the cone axis. This condition holds everywhere, from y =
0 to y = H. Show further that if a = a0, the conical tower can be
made taller than the cylindrical tower. Conversely, show that if the
two towers have the same height, the conical one can lean more be
fore its dorsal stones come apart.
13.18
The side of a container is heated by a ring-shaped electrical resistance

of radius R,,,. The ring is embedded in a concentric disk-shaped "heat
spreader" of radius R and conductivity k. The disk is attached with
very good thermal contact to the container wall. The objective is to
determine the ring radius Rm so that the temperature distribution over
the disk is the least nonuniform distribution possible. This is a con
structal problem of deriving geometry from the optimal distribution
of imperfection. To solve the problem analytically, analyze the two
dimensional model shown in Fig. P 1 3 . 1 8, which represents a disk of
unit thickness. The heat generation effect is represented by a ring that
generates the heat flux q"(W /m1). The heat transmission from the disk
Wall
I nsulated
perimeter
Heating
ring
Disk heat
spreader
Container
I-Ieating
ring
Figure P13.18
840
to the container wall is represented by the heat sink effect q'" (WI
m3 ) that is distributed uniformly over the disk. Determine the tem
perature distribution over the disk, where R,,, may vary. If T,,,, Tc, and
T,, are, in order, the temperature of the hot spot (the ring), the tem
perature in the disk center, and the temperature on the disk perimeter,
determine the optimal ring radius such that the larger of T,,, - Tc and
T,,,
T,, is the smallest possible. Show that in this configuration Tc
is equal to , .
-
13.19
Show that the scale of the residence time of water in a river basin is
proportional to the scale of the flow path length divided by the gra
dient (slope) of the flow path [ 1 76] . A simple and effective way to
demonstrate this is to rely on constructal theory, according to which
"optimal distribution of imperfection" in river basins means a balance
between the high-resistance seepage along the hill slope and the low
resistance channel flow. Consequently, the global scales of the 1iver
basin are comparable with the scales of Darcy-flow seepage along the
hill slope of length L, slope z/ L, and water travel time t. Show that
the proportionality between t and L / (z /L) is t
( vlgK)L l (z /L),
where v is the water kinematic viscosity and g is the gravitational
acceleration. The permeability of soil K has values in the range 2.9
x 1 0-9 to 1 .4 x 1 0-7 cm2 [ 1 0] .
-
13.20
A round tube of diameter D1 and length L1 splits into two tubes of

diameters D 2 and D3 and lengths L2 and L3. The tube length L2 is not
equal to the tube length L3 The three tubes form a Y-shaped con
struct: The layout of this Y on a plane or in the three-dimensional
space is not an issue in this problem. The tube lengths are given. The
tube diameters may vary, but the total volume occupied by the tubes
is fixed. The flow is in the Poiseuille regime in every duct. Each tube
is svelte enough [eq. ( 1 3 .96)] so that the pressure drop across the
entire Y construct, from the free end of L1 to the free ends of L2 and
L3, is dominated by fluid friction along the straight portions of the
ducts. In other words, the pressure loss at the Y junction is negligible.
Minimize the overall flow resistance and show that the optimal dis
tribution of tube diameters is
Show that when L2 = L3, these results confirm the Hess-Murray law
[eq. (3.87)]. In other words, the results obtained in this problem rep
resent a generalization of Murray's law for the entire range where L2
13.21
=I=
L3 .
Two newtonian liquids flow coaxially in the Poiseuille regime through

a tube of radius R. The liquids have the same density p, and the
PROBLEMS
841
Mg
Figure P13.22
longitudinal pressure gradient ( - dP I dx) is specified. The liquids have
different viscosities, ,1 > ,2 One liquid flows through a central core
of radius R;, and the other flows through an annular cross-section
extending from r = R; to the wall (r = R). Show that the total flow
rate is greater when the ,2 liquid coats the wall. This phenomenon
of self-lubrication is observed in nature and is demanded by the con
structal law. Self-lubrication is achieved through the generation of
flow configuration in accordance with the constructal law.
13.22 The constructal theory of all forms of locomotion (Section 1 3 .8)

showed that in all animals there must be a proportionality between
animal motor force and body mass (Fig. 1 3.52c). Marden and Allen
[ 1 08, 1 1 4] have found that the force-mass data of engineered motors
exhibit the same porportionality. This problem is about predicting the
force-mass proportionality for an engineered motor such as a car en
gine and an electric motor. The shaft turns while exerting the torque
T on the external system that receives the power produced by the
motor. The weight of the motor is Mg. The motor force F is Tia,
where the arm a is the same as the motor length scale. From the
rotational equilibrium of the motor body with respect to the edge as
pivot (Fig. P l 3.22), derive the proportionality FI M - g, where g is
the gravitational acceleration. Note that this is the same scaling law
as for animal motors (Fig. 1 3.S2c).

Advanced Engineering Thermodynamics (The Constructal Law) - Adrian Bejan

Uploaded by

Copyright:

Available Formats

Advanced Engineering Thermodynamics (The Constructal Law) - Adrian Bejan

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Advanced Engineering Thermodynamics (The Constructal Law) - Adrian Bejan

Uploaded by

Copyright:

Available Formats

13

The Constructal Law

THE CONSTRUCTAL LAW

Why is geometry (shape, structure, similarity) a characteristic of natural flow

THE CON STRUCTAL LAW O F CO NFI GU RATI O N GENERATI ON

THE CON S TRUCTAL LAW

THE CON STRUCTAL LAW OF CONFIGU RATION GENERATI ON

random low-resistivity blocks embedded in a high-resistivity background by

THE AREA-POINT ACCESS PROBLEM

In this chapter we review this natural phenomenon-this progression to

THE AREA-POINT ACCESS PROBLEM

As I wrote in Section 3 .6.2, my interest in maximizing flow access between

Street Patterns: A Simple Construction Sequence

THE CON S TRUCTA L LAW O F CONFIGU R ATION G ENERATION

1 3 . 1 Finite-size area (A) covered by a uniformly distributed population (Ii'')

THE AREA-PO lNT ACCE SS PRO BLEM

THE CON STRUCTAL LAW OF CONFIGUR ATI ON G ENER ATION

Smallest (innermost) elemental area, A 1 , and the street segment allocated

population density 1i" is distributed uniformly on A 1 To get out of A 1, each

THE AREA-POINT ACCE SS PROBLEM

Eguations ( 1 3 .7) and ( 1 3.2) show that the geometric minimizations of t,

THE CON STRUCTAL LAW OF CONFIG URATI ON G ENERATI ON

THE AREA-POINT ACCESS PROBLEM

1 3 .4 Area construct A 2 as an assembly of connected innermost elements A 1

THE CONS TRUCTAL LAW OF CONFIGURATION G ENERATIO N

most likely), because technology evolves in time. The

THE AREA-POINT ACCESS PROBLEM

THE CON STRUCTAL LAW OF CONFIGURATION GENERATI ON

13.5 Top: higher-order constructs in the sequence of Figs. 1 3 . 2 and 1 3 .4.

TH E AREA-PO INT ACCE SS PROBLEM

THE CON STRUCTAL LAW OF CONFIGURATION G ENERATION

where f3 is the angle between the

confirms the statement made above eq. ( 1 3 . 1 ) that

THE AREA-POI NT ACCESS PROBLEM

Heat Flow Trees

We now turn our attention to another field-another language-that is essen

The function of the electronic package or assembly of packages is the

THE CON S TRUCTAL LAW OF CONFIGURATION GENERATION

Slender elemental volume with uniform volumetric heat generation and

THE AREA-PO INT ACCESS PROBLEM

dictated by manufacturing constraints. The dimensions H0 and L0 may vary

subject to the boundary conditions aT/ ay = 0 at y = H0/2 and T = T0(x) at

= 2k0qm (HoY - y-,) + qmk DHoo (Lox - x22 )

THE CONS TRUCTAL LAW OF CONFIGURATION GENERATION

THE AREA-POINT ACCE SS PROBLEM

THE CO NS TRUCTAL LAW OF CO NFIG URATION GENERATION

13 .8 Elemental conduction volume with progressively greater freedom to

THE AREA-POINT ACCE SS PROBLEM

Second construct optimized numerically, and the effect of changing the

Constructal Theory versus Fractal Algorithms

THE CON S TRUCTAL LAW OF CONFI GURATION G ENERATION

TABLE 1 3. 1 How, Not What: Balancing High-Resistivity Flow with Low-Resistivity

Source: Private communication from S. Perin.

as a three-dimensional heuristic model for the vascular system [27], where it

THE A RE A-POI NT ACCESS PROBLEM

We now turn our attention to the application of constructal theory to fluid

THE CONS TRUCTA L LAW OF CONFIGURATION G ENERATI ON

have been described quantitatively in surprisingly sharp (reproducible) terms,

_.., Buckling and rolling