48 Handbook on entropy, complexity and spatial dynamics
The most important variation of this would be the metaphorical one measuring informa-
tional entropy, formulated by Shannon (1948) and Shannon and Weaver (1949). This measures
the probability distribution of states’ informational uncertainty, H(pi) for message i, given by
H ( pi … pn ) = −k ∑ pi ln pi . (3)
While these two forms of entropy apply to very different situations and no ontological law
of entropy operates with regard to Shannon’s metaphorical informational entropy, they are
fundamentally related.5 In particular, as N goes to infinity, Shannon entropy will converge on
a proportion of Boltzmann entropy.
This fundamental unity extends to later variations and generalizations of the entropy concept
as developed by Rényi (1961), Tsallis (1988) and Thurner and Hanel (2012). This latter links
to a development in Russia of the ‘new entropy’, which links to ergodicity theory, where
entropy is seen as an isomorphism between Bernoulli states (Kolmogorov, 1958; Sinai, 1959;
Ornstein, 1970).
3. THE WILSON MODEL AND OTHER SPATIAL DISTRIBUTION
ENTROPIC MODELS
The most influential modeler of urban and regional systems to use the concept of entropy has
been Sir Alan G. Wilson (1967; 1969; 1970; 2000; 2010). His original main model was of the
spatial distribution of retail activity flows, based on a model by Reilly (1931). The space is
partitioned by origins i and destinations j (often central places) so that Sij is a matrix of money
flows from origins i to retails sites j. Then the entropy to be maximized, subject to budget
constraints of the flows, is given by
Max S = −∑ Sij ln Sij , (4)
where for benefits of a retail site given by Wj and costs of going from an origin to a retail site
given by cij, this will give a rent-maximizing spatial distribution
( )
S = ∑W j exp cij . (5)
This then can be further modified by specifying more activities with population levels and
types of retail outlets. In principle this is broadly consistent with the original von Thünen
(1826) model of ring-patterned rent around a central place, although Wilson rarely stressed
this point.
This basic model from Wilson has since gone through many modifications and extensions,
including many by Wilson himself with various coauthors. Thus while Wilson originally
assumed that transport costs grow linearly with the log of benefits, both may be logarithmic,
which might then be true for a model of long trips involved in interurban transport, with other
functional forms possible as constraints are adjusted accordingly (Haynes and Phillips, 1987).
The model has also been extended to various other applications. Thus Rees and Wilson
(1976) and Rogers (2008) placed it into models of migration flows. Straussfogel (1991)
Entropy and complexity in urban and regional systems 49
used it in studies of suburbanization. In models of trade flows, input–output relations can be
introduced into integrated models (Kim et al., 1983; Roy and Flood, 1992).
While the basic model assumed discrete zones, Angel and Hyman (1976) extended entropy-
maximizing to continuous space representations. Problems of empirical estimation arise in
connection with aggregation and spatial structure in models of spatial interaction (Batty and
Sikdar, 1982). Econometric models of spatial autocorrelation in this framework have been
developed (Berry et al., 2008), as have broader forms of spatial interaction (Fischer and
Griffith, 2008).
Greater emphasis on a metaphoric Shannon information entropy approach was due to
Snickars and Weibull (1977). Fotheringham (1983) applied this for the case of competing
destination zones. Smith and Hsieh (1997) introduced a Markov equivalent. Anas (1984)
linked utility maximization and entropy maximization in these models using a multinomial
logit model. Wilson (2010) argues that these approaches are consistent with the ‘disorganized
complexity’ interpretation of Shannon’s information entropy approach as posited by Weaver
(1948). This contrasts with Wilson’s (1967; 1970) initial entropy approach, which drew more
on Boltzmann.
A substantial expansion of this framework within the Boltzmann framework was formulated
by Harris and Wilson (1978), who introduced slow dynamics into the model. This took the
form of bringing in elements derived from Lotka (1925) and Volterra (1938), with Wilson
(2008) labeling the result of this combination of Boltzmann, Lotka and Volterra the BLV
approach. Slow dynamics allow for growth depending on the profitability of given locations,
with the related fast dynamics affecting shorter-term equilibrium adjustments. This setup
provided a basis for considering models of catastrophic bifurcations and cascades (Wilson,
1981; Batty, 2009) as well as chaotic dynamics (May, 1973; Rosser, 1991).6
This would eventually lead to a broader consideration of how the Wilson model fits into
a broader complexity framework, especially linking with Weaver’s (1948) model between
organized and disorganized forms of complexity. For this, entropy can be seen as providing a
key organizing principle (Phillips, 2004; Wilson, 2006) relying on the BLV approach. This has
even been proposed to provide an explanation of how entropic-based models of lower-level
flows can provide a foundation for scale-free power law distributions relating to the distribu-
tions of settlement area sizes (Dearden and Wilson, 2009), which we shall consider below as
associated with anti-entropic organizational principles.
While Wilson’s work inspired a large effort by many people, as seen in the previous section,
others also used various entropic measures to study spatial distributions in urban and regional
systems of various things. One line of research was inspired by applying the Theil (1972)
index, which is based on the Shannon information entropy measure. Among the first to pursue
this line was Batty (1974). The basic spatial version of the Theil index, where H is the index,
n is the number of zones and pi is the probability that variable x appears in zone i, is given by
⎡ ⎛ 1 ⎞⎤
⎢∑ pi log ⎜ ⎟⎥
⎢⎣ ⎝ pi ⎠⎥⎦
Hn = . (6)
log n
This entropy measure can vary from 0 to 1, with the latter indicating a fully equal distribution
across the spatial zones, at maximum entropy, and 0 indicating a total concentration in one
50 Handbook on entropy, complexity and spatial dynamics
zone, or a maximum degree of inequality and anti-entropy. This index has been widely applied
across many social and natural sciences.
Batty’s (1974) variation of this, which he called spatial entropy, involves considering what
happens as the size of the zones shrinks, also implying an increasing number of them. If Δxi is
zone size, then the Batty spatial entropy index is given by
⎛ p ⎞
H = ( lim Δxi → 0) − ∑ pi log ⎜⎜ i ⎟⎟. (7)
⎝ Δxi ⎠
This formulation is very similar to one proposed by Bailey (1990) for measuring social
entropy, with again the focus on degrees of similarity or equality across social groups or zones.
Among the more direct applications of this for urban systems has been in studying urban
sprawl (Cabral et al., 2013). One line has been to measure the degree of fragmentation of
ownership. Miceli and Sirmans (2007) argue that fragmentation discourages development,
as real estate developers prefer less dispersed patterns of ownership. Scattered patterns
associated with urban sprawl lead to a form of monopoly power that manifests itself
through the holdout problem. More broadly urban sprawl is seen as contributing to a
variety of social and environmental problems, with higher costs of infrastructure and even
greater public health problems (Brueckner, 2000; Nechyba and Walsh, 2004; Frenkel and
Ashkenazi, 2007).
While most observers see urban sprawl as posing major problems, it has its defenders.
Wassmer (2008) argues that sprawl increases satisfaction with housing and schools, lowers
crime rates and allows greater convenience of car travel, although the latter is a target of those
arguing sprawl exacerbates environmental problems. Cabral et al. (2013) see this as a matter
of tradeoffs. Higher spatial entropy levels are demanding for transport and infrastructure, while
lower levels increase inequality and social-economic fragmentation.
Unsurprisingly information entropy measures have been used to measure degrees of racial
segregation in urban areas for both residences and schools (Mora and Ruiz-Castillo, 2011).
Probably the most commonly used measure in these studies is the Theil index shown in equa-
tion (6) above and first proposed for studying school segregation by Theil and Finizza (1971),
with applications such as studying segregation in the San Francisco Bay area (Miller and
Quigley, 1990). However, Mora and Ruiz-Castillo argue for the superiority of the denormal-
ized form of this known as the mutual information index, also from Theil (1971), which may
be more useful for studying decomposability by schools.
Yet further spatial applications include measuring diversity of land use patterns (Walsh
and Webber, 1977) and spatial settlement distributions (Medvedkov, 1967) as well as spatial
patterns of population distribution (Chapman, 1970). Purvis et al. (2019) provide an overview
of many of these applications.
4. THERMODYNAMIC SUSTAINABILITY
Most of the models discussed in the previous two sections have stemmed from the metaphorical
information concept of entropy coming from Shannon and Weaver, with the possible excep-
tion of Wilson’s development of slow dynamics, which draws more directly on Boltzmann.
Entropy and complexity in urban and regional systems 51
However, another strand of entropic analysis of urban and regional systems relies more on the
original ontological approach, in which an urban or regional system is seen as being driven by
thermodynamics in its original physical sense involving energy transfers and transformations
following the second law of thermodynamics. Among those pursuing such an approach have
been Rees (1992), Balocco et al. (2004), Zhang et al. (2006), Marchettini et al. (2006) and
Purvis et al. (2019).
The focus of most of this research is particularly on the ecological sustainability of urban
and regional systems, viewing them as open dissipative systems experiencing inflows and
outflows of energy and materials (Georgescu-Roegen, 1971; Prigogine, 1980). While in
closed systems entropy increases, with open systems entropy can either increase or decrease
if energy and materials flow into the system. This was indeed the Schrödinger (1945) concept
of life, that it involves an anti-entropic process whereby living things draw in energy and
create order and structure as long as they live. A specific term for anti-entropy is exergy
(Rant, 1956).
Let us then distinguish three concepts: total entropy or Stotal, inside entropy or Si and outside
entropy or So. These are related dynamically according to
dStotal dSi dS0 dS
= + ,with i >0. (8)
dt dt dt dt
dS
However, dt0 can be either positive or negative, so if it is negative and has an absolute value
exceeding that of the absolute value of Si, then total entropy may decline as the system gener-
ates order as it draws in energy and materials, only to export them as waste and disorder, with
entropy increasing outside the system. As Wackernagel and Rees (1996) put it, ‘Cities are
entropic black holes,’ raising serious questions about their sustainability as they generate large
ecological footprints.
Exergy is often defined as the maximum amount of useful work possible to reach a maxi-
mum entropy state, which means it must be zero if a maximum entropy state is achieved. Rant’s
(1956) original formulation was in the context of chemical engineering. If B is exergy, U is
internal energy, P is pressure, V is volume, T is temperature, S is entropy, μi is the chemical
potential of component i and Ni is the moles of component i, then Rant’s formulation is given
by
B = U + PV − TS − ∑ui N i . (9)
This implies, ceteris paribus, that
dB dS
≤0↔ ≥ 0, (10)
dt dt
which highlights the interpretation of exergy as being anti-entropy.7
An application of this using a modification of Rant’s equation from Moran and Sciubba
(1994) has been done by Balocco et al. (2004). They study the exergy involved in building
construction and real depreciation in the town of Castelnuovo Berardenga near Siena, Italy.
This involves also using input–output relations associated with the construction industry.
52 Handbook on entropy, complexity and spatial dynamics
They conclude that more recent buildings are not as efficient as older ones, with those built in
1946–60 providing the highest sustainability.
Following Wackernagel and Rees and Balocco et al., as well as Haken (1988) and Svirezhev
(2000), Zhang et al. (2006) engage in an ambitious effort to apply entropy concepts to the study
of sustainable development in Ningbo, China, a city of nearly six million somewhat south of
Shanghai in Zhejiang province. Their effort combines ontological measures of entropy with
metaphoric information ones as they break their analysis into four parts. The first two, which
are tied to development, are sustaining input entropy and imposed output energy, which are
basically determined by production. The second two are considered to be part of the metabolism
of the urban system, regenerative metabolism and destructive metabolism, which are tied to
the generation of pollution and its cleanup. This becomes a measure of harmony with the
environment. The outcome of the first gives the developmental degree, while the second gives
the harmony degree. They estimate these for the 1996–2003 period and find that these two
measures were generally going in opposite directions, with the developmental degree rising
(associated with declining entropy) as the harmony degree was declining (associated with rising
entropy). This poses the problem of sustainability of urban development in China quite sharply.
Marchettini et al. (2006) consider this approach from a more general level,8 drawing on ideas
from Morin (1995) regarding autonomy versus dependence of systems on their environment,
while using the dissipative structures approach of open systems associated with Prigogine
(1980). They see urban systems evolving between extremes of autarchy and globalization.
However, they argue that in the end neither of these extremes is sustainable. In their advocacy
of a balanced path they emphasize how urban and regional systems are ecosystems that operate
on the basis of energy flows (Odum, 1969) within a set of complex wholes emerging from a
set of interacting micro-level components (Ulanowicz, 2012).
5. THE ANTI-ENTROPIC ALTERNATIVE, COMPLEXITY
AND SELF-ORGANIZATION IN URBAN SYSTEMS
Pushing against this entropic version of the structure of urban and regional systems is a power
law version of such structuring, at least for certain cases and situations. Arguably this is dealt
with in the entropy framework, given the matter of the balance between exergy and entropy
in urban and regional systems. Most of the systems and measures up until now have involved
essentially internal relations or distributions within urban or regional systems. But when one
considers higher-level distributional systems, the entropy relation may break down or even
become completely irrelevant.
One way that anti-entropic forces can manifest is by the appearance of power law distribu-
tions (Rosser, 2016),9 with substantial evidence that city sizes may follow such distributions
(Gabaix, 1999). Pareto (1897) identified the concept of power law distributions. When P is
population, r is rank and A and α are constants, then
rPr α = A, (11)
which can be put into log-log form, which is linear, as
ln r = ln A − α (lnPr). (12)
