www.elsevier.com/locate/ijhmt
Abstract
In this paper we develop an analytical and graphical formulation of the constructal law of maximization of ow
access in systems with heat and uid ow irreversibilities and freedom to change conguration. The ow system has
global performance (e.g., minimization of global ow resistance) and global properties, or constraints (e.g., overall size,
and total duct volume). The innity of possible ow structures occupies a region of the two-dimensional domain of
global performance versus freedom to morph. This region of nonequilibrium ow structures is bounded by a line
representing the best ow structures that are possible when the freedom to morph is limited. The best of all such
structures are the equilibrium ow structures: here the performance level is the highest, and it does not change even
though the ow architecture can change with maximum freedom. The universality of this graphical and analytical
presentation is illustrated with examples of ow structures from three classes: ow between two points, ow between a
circle and its center, and ow between one point and an area. In sum, this paper presents an analytical and graphical
formulation of the constructal principle of generation of ow architecture. The place of this new and self-standing
principle in the greater framework of thermodynamics is outlined.
2004 Elsevier Ltd. All rights reserved.
Keywords: Constructal theory; Design as science; Thermodynamics; Nonequilibrium; Conguration; Survival; Evolution; Territory;
Growth
Nomenclature
The generation of ow architecture is the means by such as total volume, total volume occupied by all the
which the ow system achieves its global objective under ducts, etc. A ow system is also characterized by per-
the constraints. formance (function, objective) and ow structure
In recent years, this activity of thermodynamic opti- (conguration, layout, geometry, architecture). Unlike
mization through the selection of ow conguration has the black box of classical thermodynamics, which rep-
become more focused on the end result, which is the resents a system at equilibrium, a ow system has
generation of the architecture of the ow system. This is performance and especially conguration. Each ow
particularly evident in modern computational heat and system has a drawing. By means of examples, we show
uid ow, where large numbers of ow congurations that each ow system has a fundamental relation be-
can be simulated, compared and optimized. The gener- tween performance, properties and ow structure
ation of ow architecture is a phenomenon at work (drawing).
everywhere, not only in engineered ow systems but also We start with one of the simplest examples of how
in natural ow systems (animate and inanimate). The the collision between global objective and global con-
universality of this observation was expressed in a straints generates the complete architecture of the ow
compact statement (the constructal law [1,2]) that pro- system. Consider the ow between two points (Fig. 1),
claims a natural tendency in time: the maximization of where simple are only the optimal and near-optimal
access for the currents that ow through a morphing architectures. This makes the example easy to present
ow system. The thought that this principle can be used graphically. The rest of the design process is conceptu-
to rationalize the occurrence of optimized ow struc- ally as vast and complicated as in any other example.
tures in nature (e.g., tree networks, round tubes) was When the ow architecture is free to morph, the design
named constructal theory [14]. space is innite. There is an innity of ow architectures
This paper is a formulation of the constructal prin- that can be chosen to guide a uid stream (m) _ from one
ciple in analytical and graphical terms that are analo- point to another point.
gous to terms employed in thermodynamics [2]. This Constructal theory [1,2] begins with the global
formulation makes the universality of the constructal objective(s) and the global constraint(s) of the ow
law more evident. system, and the fact that in the beginning the geometry
of the ow is missing. Geometry is the unknown. In Fig.
1 the objective is to force the single-phase uid stream m_
2. Flow between two points to ow from one point to another, while using minimal
pumping power. When m_ is xed, this objective is the
A ow system, or nonequilibrium thermodynamic same as seeking ow architectures with minimal pressure
system, is characterized by properties (constraints), overall dierence (DP ), minimal overall ow resistance
A. Bejan, S. Lorente / International Journal of Heat and Mass Transfer 47 (2004) 32033214 3205
Fig. 1. General, undened ow architecture for guiding a stream from one point to another point.
(R), or minimal rate of entropy generation by uid structure, which are fundamental and universally appli-
friction. cable. They are geometric relatives of truths such as the
There are two global constraints, one external and universal observation that all things ow naturally from
the other internal. The external constraint is the system high to low (the second law of thermodynamics).
size, which is represented by the distance between the Here are the classical lessons that abbreviate the
two points, L. The internal global constraint is the search through the broad categories listed as (i)(vi).
amount invested in making the ow architecture. In Again, for simplicity assume that ducts are slender, and
Fig. 1 that amount is the total volume (V ) of all the the ows are slow so that in each cross-section the re-
ducts of the ow structure. Without such an investment gime is laminar and fully developed. Each lesson is
there is no ownot even a drawing that would show identied by the symbol of the geometric feature that it
the ow. A ow must be guided. Flow means direction, addresses:
geometry and architecture, in addition to ow rate.
Why is there an innity of eligible ow architectures (i)(iii) A single duct with large cross-section oers a
that meet the global objective and global constraints smaller ow resistance than two ducts with
recognized above? There are many reasons, i.e., many smaller cross-sections connected in parallel.
thoughts in the direction of which the number of pos- (iv) The lowest resistance belongs to the shortest
sible architectures increases without bounds: duct, in this case the straight duct between the
two points.
ii(i) the ow pattern may be two-dimensional (in the (v) The duct with cross-sectional geometry that does
plane of Fig. 1), or three-dimensional, not vary longitudinally has a lower resistance
i(ii) any number of ducts may be connected in parallel than the duct with variable cross-section. In heat
between the two points, conduction, for example, this principle is in-
(iii) a duct may have any number of branches or tribu- voked to deduce the optimal n shape [5]. If
taries at any location between the two points, the cross-section is the throttle that represents
(iv) a single duct may have any length, the imperfection (resistance) of the ow path,
i(v) the cross-sectional shape may vary along the duct, then, in line with constructal theory, the duct
(vi) a duct may have any cross-sectional shape. with constant cross-section is the duct with
optimal distribution of imperfection [1].
We will see that the best conguration (the straight
round tube) is far from being alone on the podium of Summing up, out of the innity of designs repre-
maximal performance. This podium and the congura- sented by (i)(v) we have selected a single straight duct
tion world under it are the new physics domain charted with a cross-sectional shape that does not vary from one
by this paper, and made a part of thermodynamics. end of the duct to the other. According to (vi), however,
How do we identify the geometric features that bring there is still an innite number of possible cross-section
a ow architecture to the highest level of global per- shapes: symmetric versus asymmetric, smooth versus
formance? There are many lessons of this type polygonal, etc. Which impedes the ow the least?
throughout engineering, and, if remembered, they con- The answer becomes visible if we assume cross-sec-
stitute strategythey shorten dramatically the search tions with polygonal shapes. Start with an arbitrary
for the geometry in which all the features are useful in cross-section shaped as a triangle. The area of the cross-
serving the global objective. Constructal theory is about section A is xed because the total duct volume V and
strategy, about compact lessons of optimal shape and the duct length L are xed, namely A V =L. Triangular
3206 A. Bejan, S. Lorente / International Journal of Heat and Mass Transfer 47 (2004) 32033214
cross-sections constrict the ow when one of the angles ow through a straight duct with polygonal cross-sec-
is much smaller than the other two. tion can be written as (Ref. [1, pp. 127128])
The least resistance is oered by the most open
DP mL Cp2
triangular cross-section, which is shaped as an equilat- 1
m_ 8V 2 A
eral triangle. Once again, if one very small angle and two
larger ones represent a nonuniform distribution of geo- where p is the perimeter of the cross-section. As shown
metric features of imperfection (i.e., features that impede in Table 1, the dimensionless perimeter p=A1=2 is only a
the ow), then the equilateral triangle represents the function of n. The same is true about C, which appears
architecture with optimal distribution of imperfec- in the solution for friction factor in HagenPoiseuille
tion. ow,
The same holds for any other polygonal shape. The
C
least resistance is oered by a cross-section shaped as a f 2
Re
regular polygon. In conclusion, out of the innity of
ow architectures recognized in class (vi) we have se- where Re U Dh =m, Dh 4A=p and U m=qA.
_
lected an innite number of candidates. They are or- In conclusion, the group Cp2 =A depends only on n,
dered according to the number of sides (n) of the regular and accounts for how this last geometric degree of
polygon, from the equilateral triangle (n 3) to the freedom inuences global performance. The group
circle (n 1). The ow resistance for HagenPoiseuille Cp2 =A is the dimensionless global ow resistance of the
ow system. The smallest Cp2 =A value is the best, and
the best is the round cross-section.
Table 1
The laminar ow resistances of straight ducts with regular- 3. Equilibrium ow structures
polygonal cross-sections with n sides
n C p=A1=2 Cp2 =A Fig. 2 is a plot of the ow resistance data of Table 1.
3 40/3 4.559 277.1 The ow structure with minimal global resistance is
4 14.23 4 227.6 approached gradually (with diminishing decrements) as
5 14.74 3.812 214.1 n increases. The polygonal cross-section with n 10
6 15.054 3.722 208.6 performs nearly as well as the round cross-section
8 15.412 3.641 204.3 (n 1). The evolution of the cross-sectional shape
10 15.60 3.605 202.7
stops when the number of features (n) has become in-
1 16 2p1=2 201.1
nite, i.e., when the structure has become the most free.
Fig. 2. The approach to the minimal global ow resistance between two points when the number of sides of the regular-polygon cross-
section (n) increases (data from Table 1).
A. Bejan, S. Lorente / International Journal of Heat and Mass Transfer 47 (2004) 32033214 3207
This conguration where changes in global performance large but nite number of points. One example is the
have stopped is the equilibrium ow architecture. ow that bathes with minimal global resistance a disc-
The curve plotted in Fig. 2 was generated by calcu- shaped area [6]. The stream (m) _ emerges through the
lations for regular-polygon cross-sections. The curve is center of the disc, and exits through a large number of
in reality a sequence of discrete points, one point for outlets positioned equidistantly along the disc perimeter
each n value. We drew a continuous line through these (Fig. 3). The objective is to minimize the global ow
points in order to stress an additional idea. Regardless resistance DP =m,_ where DP is the pressure dierence
of n, the regular polygon and straight duct with con- between the center and the rim. The global constraints
stant cross-section is already the winner from an in- are the system size (the disc radius L) and the total
nitely larger group of competing architectures. See volume occupied by the ducts (V ). Note that the disc
features (i)(vi) and Fig. 1. Qualitatively, this means that radius L plays the same role as L in Section 2: L is the
the global ow resistances of all the designs that are not external global constraint of the system.
covered by Table 1 fall to the right of the curve plotted Even if the search for optimal conguration starts
in Fig. 2. with an assumed tree-shaped ow structure, there are
In sum, the immensely large world of possible designs still several geometric features that can be adjusted. We
occupies only a portion of the two-dimensional domain use the conclusions of Sections 2 and 3, and assume that
illustrated in Fig. 2. This domain can be described every tube is straight with round cross-section and fully
qualitatively as performance versus freedom, when developed laminar ow. We also assume that local
global properties such as L and V are specied. The pressure drops due to junctions are negligible. The ow
boundary of the domain is formed by a collection of the geometry is described by the lengths and diameters of all
better ow structures. The best is achieved by putting the ducts, the number of radial ducts (n0 ) that touch the
more freedom in the geometry of the ow structure (e.g., center of the disc, the distance between outlets (d, or
a larger n). The best performance belongs to the structure the number of outlets on the rim, N 2pL=d), and the
that was most free to morphthe equilibrium congu- number of pairing (or bifurcation) levels, which in Fig. 3
ration. In its immediate vicinity, however, we nd many are indicated with dashed circles. In general, the number
congurations that are dierent (they have nite n val- of tributaries that form a larger stream is free to vary. It
ues), but have practically the same global performance was shown that the number two (dichotomy, pairing,
level. These are near-equilibrium ow structures. bifurcation), which was chosen in Fig. 3, is the best such
number [6,7]. Although the tree structure chosen in Fig.
3 has multiple scales and is free to change, it is already
better than a huge number of other possible ow
4. Flow between one point and a large number of points structures. In this respect, a general dichotomous tree
structure is a near-equilibrium congurationthe number of central tubes (n0 ) decreases as the number of
equivalent of the straight duct with constant cross-sec- pairing levels (k) increases, namely n0 48=2k . This is
tion shaped as a regular polygon. the same as reading Fig. 4 in the vertical cut made at
What changes in the point-to-circle tree geometry? N 48. The results are shown in Fig. 5. The perfor-
Because L is xed, the architecture has ve degrees of mance improves as k increases. The equilibrium ow
freedom: the number of central tubes n0 , the length ra- structure is the one with four levels of pairing, k 4.
tios L0 =L and L1 =L, and the tube diameter ratios D1 =D0 This conguration is the most free to morph, and sat-
and D2 =D1 . Another classical lesson of ow-system ises the additional constraint (N 48). Its complexity
optimization is Murrays rule [1,8], according to which (k 4) is a result of optimization. Optimized complexity
the optimal diameter ratios are all equal (D1 =D0 should not be confused with maximized complexity.
D2 =D1 21=3 ). Murrays rule was adopted in [6], The continuous curve that unites the ve points in
where the tree geometry was optimized for many n0 Fig. 5 has the same symbolic meaning as in Fig. 2. Such
values, and for various levels of pairing. For example, a curve cannot be plotted. What exists is a conceptual
the conguration selected in Fig. 3 is the optimal ow demarcation line in the performance versus freedom
layout for two levels of pairing and n0 3 (or N 12). domain. To the right of this line falls the cloud of all
It was shown that the minimized global ow resistance is the nonequilibrium ow structures that are possible. For
represented by a formula similar to Eq. (1), example, the square plotted at n0 3 and f^ 17:6
indicates the sub-optimal ow structure obtained more
DP L3
8pm 2 f^ 3 expediently by minimizing locally the ow path lengths
m_ V
and assembling them as building blocks in a hierarchical
The dimensionless factor f^ is a function of the number sequence [9], instead of minimizing globally the ow
of pairing levels and N (or n0 ), and is summarized in Fig. resistance [6]. The edge of the cloud is concave, and the
4. The curves represent the f^ values minimized with tangent to it approaches verticality near the point of
respect to all the length ratios and diameter ratios. equilibrium. This is another way of saying that a near-
In conclusion, the multitude of designs that depart equilibrium ow structure such as the point labeled
from the optimal conguration (e.g., Fig. 3) have larger k 3 performs at practically the same level as the
f^ values, and occupy the space above the respective equilibrium ow structure (k 4).
f^ N curve. This feature is similar to what we saw in Fig. The construction of Fig. 5 can be repeated for larger
2 for pointpoint ow structures. In fact, we arrive at N values, and every new gure will be qualitatively the
the equivalent of Fig. 2 if we focus on only one class of same as Figs. 5 and 2. The only change relative to Fig. 5
ow structures: the structures with the same number of will be the increasingly larger k values of the designs that
outlets on the rim, for example, N 48. These struc- mark the edge of the cloud. The number k plays the
tures have the same global properties (L; V ). Their same role as n in Fig. 2. The equilibrium ow structure
Fig. 4. The minimized global ow resistance of the tree-shaped ow structures that connect the center and the rim of a circular area [6].
A. Bejan, S. Lorente / International Journal of Heat and Mass Transfer 47 (2004) 32033214 3209
Fig. 5. The approach to the minimal global ow resistance between the center and the rim of a circular area when N 48 in Fig. 4.
will be at the bottom end of the curve. It will represent have an innite number of pairing levels [2,10]. This tree
the optimized architecture that results when the con- will connect the center of the disc with every point of the
stant-N class is endowed with most freedom as it mor- disc perimeter.
phs. Many examples of optimized tree-shaped ow struc-
tures have been published based on constructal theory
(e.g., Refs. [1,4,1114]). They all exhibit the organization
5. Flow between one point and an innite number of points illustrated here in Figs. 2 and 5. Consider just one
example of ow between an entire area and one point:
In the limit N ! 1, Fig. 5 will have an innite conduction in a rectangular domain of area A and
number of points on the edge of the cloud, and in the conductivity k0 , with uniform heat generation rate per
vicinity of the equilibrium structure the edge curve will unit area q000 (Fig. 6). The A domain is rectangular, and
appear continuous. The equilibrium structure will be a its boundaries are insulated except the mid point of one
fractal tree, because only in this hypothetical limit it will of the sides, which serves as heat sink (Tmin ). The hot
Fig. 6. Optimized high-conductivity trees embedded in low-conductivity media with uniform volumetric heat generation (Ap =A 0:1,
kp =k0 300) [14].
3210 A. Bejan, S. Lorente / International Journal of Heat and Mass Transfer 47 (2004) 32033214
Tmax Tmin
Rt 4
q000 A=k0
dV 6 0 constant L; R 8
According to Eqs. (12), the constructal law states that equilibrium ow architectures. For them the solid
the ultimate ow structure with specied global resis- curves shown in Figs. 7b and 8b proclaim the existence
tance (R) and internal size (V ) is the largest. A ow of the fundamental relation RL; V , the dierential of
architecture with specied R and V has a maximum size, which is
and this global size belongs to the equilibrium archi-
dR YL dL YV dV equilibrium 13
tecture. A ow structure larger than this does not exist.
This formulation of the constructal law has implications Physics requires that the rst partial derivatives of R
in natural design, e.g., the spreading of species and river have opposite signs, YL > 0 and YV < 0, as noted in Eqs.
deltas without access to the sea. (10) and (7). Analytical expressions for these derivatives
are available in simple cases, e.g., Eqs. (1) and (3). For
example, Eq. (1) states that R
L=V 2 , and from this
expression one can obtain YL oR=oLV and YV
7. Survival by increasing eciency, territory and com-
oR=oV L . Another example is the global ow resistance
pactness
of a T-shaped construct of round tubes with fully
developed laminar ow [7], where R
A3=2 =V 2 , and A
It is worth examining the ground covered so far. The
(or
L2 in this paper) is the area occupied by the con-
original statement of the constructal law was about the
struct. If the ow regime is turbulent, R is proportional
maximization of ow access under global size con-
to A7=4 =V 5=2 [7]. The global resistance of an equilibrium
straints (external L, internal V ). This behavior is illus-
ow structure can be decreased (dR < 0) through
trated by the structural changes 12e in Figs. 7b and
changes II and III, i.e., by making the structure occupy a
8b, and by Eqs. (5) and (6). This means survival by
smaller territory (dL < 0), and/or by endowing the
increasing eciencysurvival of the ttest. This is the
structure with a larger internal ow volume (dV > 0).
physics principle behind Darwins observations, the
In general, when the ow architecture has not
principle that rules not only the animate natural ow
reached equilibrium, R can be decreased by means I, II
systems, but also the inanimate natural ow systems and
and III. Then the general version of Eq. (13) is
the engineered ow systems. The engineered systems are
diverse species of man + machine beings. dR 6 YL dL YV dV 14
The alternative shown by the changes 1200 e00 in Fig.
8b is survival by spreading: growth as the mechanism for where the inequality sign refers to the time arrow of
being able to persist in time. The limit to growth is set by structural changes in a ow conguration that, at least
the specied constraints, in this case the xed global ow initially, was not of the equilibrium type. It is instruc-
resistance R and the global internal size V . A given living tive to review Eqs. (5), (8) and (11), to see that Eq. (14)
species (river delta, animal population) will spread over is a concise statement of the three analytical for-
a certain, maximal territory. mulations of the constructal law that we discussed so far:
An equivalent interpretation of the constructal prin-
ciple is based on processes of type 120 e0 , Fig. 7b. Flow R minimum at constant L and V ,
architectures with the same performance (R) and size (L) V minimum at constant R and L,
evolve toward compactnesssmaller volumes dedicated L maximum at constant V and R.
to the internal ducts, i.e., larger volumes dedicated to the
working volume elements, which are the interstices. This Another way to summarize the analytical formula-
is survival based on the maximization of the use of the tion that we have just constructed is by recognizing the
available space. analogy between the analytical constructal law and the
analytical formulation of classical thermodynamics [2].
The analogy is presented in Table 2. It is stressed further
8. The constructal law as an addition to thermodynamics by Figs. 9 and 10, which are from present-day thermo-
dynamics (see Ref. [2, pp. 241 and 244]). Fig. 9 expresses
Changes in performance (R) can be achieved through the energy minimum principle, which states that as the
changes of three types: internal constraints of a closed system are removed at
constant volume and entropy, the energy approaches a
I. Flow conguration. minimal value. Fig. 9 is analogous to Fig. 8a.
II. Global external size, or covered territory, L. Fig. 10 is a restatement of the energy minimum
III. Global internal size, or duct volume, V . principle in Helmholtz free-energy (F ) representation: as
the constraints of the closed system are removed at
The examples discussed so far showed that changes constant volume and temperature, F decreases towards
may occur in one category, or simultaneously in two or a minimal value. Fig. 10 is the thermodynamics equiv-
three. The simplest illustration is possible for the case of alent of the constructal-theory Fig. 7a.
A. Bejan, S. Lorente / International Journal of Heat and Mass Transfer 47 (2004) 32033214 3213
Table 2
The concepts and principles of classical thermodynamics and constructal theory
Thermodynamics Constructal theory
State Flow architecture (geometry, conguration, structure)
Process, removal of internal constraints Morphing, change in ow conguration
Properties (U ; S; Vol; . . .) Global objective and global constraints (R; L; V ; . . .)
Equilibrium state Equilibrium ow architecture
Fundamental relation, US; Vol; . . . Fundamental relation, RL; V ; . . .
Constrained equilibrium states Nonequilibrium ow architectures
Removal of constraints Increased freedom to morph
Energy minimum principle: Constructal principle:
U minimum at constant S and Vol R minimum at constant L and V
Vol minimum at constant F and T V minimum at constant R and L
S maximum at constant U and Vol L maximum at constant V and R
9. Concluding remarks
possible ow architectures will ll a cloud with the same [6] W. Wechsatol, S. Lorente, A. Bejan, Optimal tree-shaped
features as in Fig. 5. networks for uid ow in a disc-shaped body, Int. J. Heat
At equilibrium the ow conguration achieves the Mass Transfer 45 (2002) 49114924.
most that its freedom to morph has to oer. Equilibrium [7] A. Bejan, L.A.O. Rocha, S. Lorente, Thermodynamic
does not mean that the ow architecture (structure, optimization of geometry: T- and Y-shaped constructs of
uid streams, Int. J. Therm. Sci. 39 (2000) 949960.
geometry, conguration) stops changing. On the con-
[8] C.D. Murray, The physiological principle of minimal
trary, it is here at equilibrium that the ow geometry work, in the vascular system, and the cost of blood-
enjoys most freedom to change. Equilibrium means that volume, Proc. Acad. Nat. Sci. 12 (1926) 207214.
the global performance does not change when changes [9] S. Lorente, W. Wechsatol, A. Bejan, Tree-shaped ow
occur in the ow architecture. structures designed by minimizing path lengths, Int. J. Heat
Mass Transfer 45 (2002) 32993312.
[10] D. Avnir, O. Biham, D. Lidar, O. Malcai, Is the geometry
of nature fractal?, Science 279 (1998) 3940.
References [11] Y. Chen, P. Cheng, Heat transfer and pressure drop in
fractal tree-like microchannel nets, Int. J. Heat Mass
[1] A. Bejan, Shape and Structure, from Engineering to Transfer 45 (2002) 26432648.
Nature, Cambridge University Press, Cambridge, UK, [12] D.V. Pence, Improved thermal eciency and temperature
2000. uniformity using fractal-like branching channel networks,
[2] A. Bejan, Advanced Engineering Thermodynamics, 2nd in: G.P. Celata, V.P. Carey, M. Groll, I. Tanasawa, G.
ed., Wiley, New York, 1997 (Chapter 13). Zummo (Eds.), Heat Transfer and Transport Phenomena,
[3] H. Poirier, A theory explains the intelligence of nature, Sci. Begell House, New York, 2000, pp. 142148.
Vie 1034 (November) (2003) 4463. [13] Z.-Z. Xia, Z.-X. Li, Z.-Y. Guo, Heat conduction optimi-
[4] A. Bejan, I. Dincer, S. Lorente, A.F. Miguel, A.H. Reis, zation: high conductivity constructs based on the principle
Porous and Complex Flow Structures in Modern Tech- of biological evolution, in: Twelfth Int. Heat Transfer
nologies, Springer-Verlag, New York, 2004. Conf., Grenoble, France, 1823 August.
[5] A.D. Kraus, A. Aziz, J. Welty, Extended Surface Heat [14] G.A. Ledezma, A. Bejan, M.R. Errera, Constructal tree
Transfer, Wiley, New York, 2001. networks for heat transfer, J. Appl. Phys. 82 (1997) 89100.