Stable Economic Agglomeration Patterns in Two Dimensions: Beyond The Scope of Central Place Theory

JOURNAL OF REGIONAL SCIENCE, VOL. 57, NO. 1, 2017, pp.
132–172
STABLE ECONOMIC AGGLOMERATION PATTERNS IN TWO

DIMENSIONS: BEYOND THE SCOPE OF CENTRAL PLACE THEORY
Kiyohiro Ikeda
Department of Civil and Environmental Engineering, Tohoku University, Aoba, Sendai 980-8579,
Japan. E-mail: kiyohiro.ikeda.b4@tohoku.ac.jp
Kazuo Murota
School of Business Administration, Tokyo Metropolitan University, Tokyo 192-0397, Japan.
E-mail: murota@tmu.ac.jp
Yuki Takayama
Institute of Science and Engineering, Kanazawa University, Kakuma, Kanazawa 920-1192, Japan.
E-mail: ytakayama@se.kanazawa-u.ac.jp
ABSTRACT. This paper elucidates which agglomeration patterns exist in two-dimensional economic
space and how such patterns appear stably. Hexagonal lattices, that with and that without a bound-
ary, are advanced, respectively, as practical and theoretical spatial platforms of economic activities.
Agglomeration patterns on these lattices include hexagons in central place theory, but also encompass
megalopolis and racetrack-shaped decentralization. As the transport cost decreases, stable economic
agglomeration undergoes the formation of the smallest hexagon and transition to patterns with larger
market areas, often undergoing downtown decay but finally leading to a megalopolis. Formulas for
break points are provided in an economic geography model.
1. INTRODUCTION
Economic agglomeration displays various spatial patterns serving as a cradle of re-
gional development and prosperity. Cities and towns in southern Germany, which are
spread out with geometrical orderliness, led to the finding of hexagonal distributions in
central place theory (Christaller, 1933). A chain of cities extends from Boston to Washing-
ton DC in a closed long narrow corridor between the Atlantic Ocean and the Appalachian
Mountains. Some spatial agglomerations are unstable and transient, but several spatial
agglomerations have developed and prospered stably worldwide. Nowadays, downtown
areas are being revitalized through investment in transportation systems.
It is desirable to know what stable economic agglomeration patterns can exist in
two-dimensional economic space. Yet there might be widespread pessimism that such
stable equilibria cannot be grasped completely because they are literally infinite. To re-
buff this pessimism, the following question is considered in this paper: “What are stable
agglomeration patterns in two dimensions?” A key to answer this question is to distin-
guish model-independent spatial properties and microeconomic properties of individual
models.1
Received: October 2014; revised: July 2015, December 2015, March 2016; accepted: March 2016.
1 Anas (2004, p. 181) stated “Of course, when the number of cities or the geographic space itself is
limited or asymmetric, then agglomeration can arise as an artifact of the constraints imposed by geography
as demonstrated by numerous NEG models. This reveals that the central agglomeration force in the NEG
is space itself and not the underlying economic relations.”

C 2016 Wiley Periodicals, Inc. DOI: 10.1111/jors.12290
132
IKEDA, MUROTA, AND TAKAYAMA: STABLE ECONOMIC AGGLOMERATION PATTERNS 133
Long narrow Racetrack Lattice

Note: Circles represent places and lines denote roads.
FIGURE 1: Spatial Platforms for Economic Activities.
A preliminary and mandatory step for answering this question is to answer another
question: “What is a suitable spatial platform of spatial economic activities?” Several
spatial platforms have been developed, including the two-place economy, the long narrow
economy, the racetrack economy, and the lattice economy2 (Figure 1). The lattice economy
accommodates numerous locations displaying two-dimensional patterns of various kinds.
The two-place economy is too simple, despite its vital role in the development of new
economic geography (NEG) models. Other one-dimensional economies, such as the long
narrow and the racetrack economies, are believed to be capable of representing some
important agglomeration properties.
The long narrow economy has been used as a spatial platform of economic activities
displaying diverse agglomeration patterns ranging from continuous to discrete ones. A
continuous agglomeration has been observed and described as “formation of a megalopolis
which consists of large core cities that are connected by an industrial belt, i.e., a continuum
of cities” associated with lower transport costs by Mori (1997). A discretized highly regular
central place system à la Christaller was observed when population size increased (Fujita
and Mori, 1997). An atomic monocenter has been found (Fujita and Mori, 1997; Fujita,
Krugman, and Mori, 1999).
The racetrack economy has been used to show the evolution of a regular lattice
by Krugman (1993) and Fujita, Krugman, and Venables (1999). Krugman (1996, p. 91)
regarded the racetrack economy as one-dimensional and inferred its extendibility to a
two-dimensional economy to engender hexagonal distributions.3 Bifurcation produced a
chain of spatially repeated core-periphery patterns à la Christaller and Lösch, which
denotes a spatial alternation of a core place with a large population and a peripheral
place with a small population.4
A hexagonal lattice has been used as a spatial platform of hexagonal distributions in
central place theory. Although this theory is a powerful idea, it is based only on a normative
and geometrical approach and is not derived from market equilibrium conditions.5 As an
2 Several studies of spatial agglomeration have been conducted on a square lattice (Clarke and Wilson,
1983, 1985; Weidlich and Haag, 1987; Munz and Weidlich, 1990; Brakman et al., 1999). Moreover, Stelder
(2005) conducted a simulation of agglomeration for cities in Europe using a grid of points.
3 Krugman (1996, p. 91) stated “I have demonstrated the emergence of a regular lattice only for a
one-dimensional economy, but I have no doubt that a better mathematician could show that a system of
hexagonal market areas will emerge in two dimensions.”
4 Tabuchi and Thisse (2011) examined the racetrack economy for a multi-industry model to show the
emergence of central places. See also Picard and Tabuchi (2010), Ikeda, Akamatsu, and Kono (2012), and
Akamatsu, Takayama, and Ikeda (2012).
5 Fujita, Krugman, and Mori (1999, p. 212) stated “it [central place theory] is a powerful idea too good
for being left as an obscure theory.”

C 2016 Wiley Periodicals, Inc.
134 JOURNAL OF REGIONAL SCIENCE, VOL. 57, NO. 1, 2017
(a) Hexagon (b) Monocenter (c) Megalopolis (d) Racetrack

Note: The area of a circle denotes the population size.
FIGURE 2: Agglomeration Patterns of Economic Interest.
early attempt to provide central place theory with a microeconomic foundation, Eaton and
Lipsey (1975, 1982) showed the existence of a hexagonal distribution of mobile production
factors (e.g., firms and workers) by a partial equilibrium approach without referring to the
stability of the hexagonal agglomeration. Recently, theoretical studies on the existence
of equilibria of hexagonal distributions and numerical analyses of their stability have
been conducted (Ikeda, Murota, and Akamatsu, 2012; Ikeda and Murota, 2014; Ikeda
et al., 2014). Hexagonal distributions observed in the numerical analyses were clearer
in comparison with those on a square lattice (footnote 2). This theory resorts only to
spatial properties. As such, it is endowed with much-desired model independency. Not
much attention, however, has been devoted to nonhexagonal patterns.
Studies of all these spatial platforms have been conducted somewhat independently6
and several stable agglomeration patterns have been observed fragmentarily. It would be
desirable to have a synthetic view of stable spatial patterns on these platforms.
The aim of this paper is to clarify, through a synthetic study of stable patterns in
two dimensions, that hexagonal patterns are not the only stable agglomeration patterns
and that there are stable nonhexagonal patterns of great economic interest. These non-
hexagonal patterns include the megalopolis (Figure 2c), the racetrack (Figure 2d), the
long narrow, and the deformed hexagonal patterns. In particular, racetrack patterns ex-
press decentralization, in contrast to the centralization of hexagonal patterns. The major
contributions of this paper are twofold: (1) to demonstrate, by numerical simulations, tran-
sitions between hexagonal and nonhexagonal patterns under decreasing transport costs;
(2) to set forth the theoretical analysis procedure of break points,7 giving the transport
cost at which hexagonal and nonhexagonal patterns emerge.
Two kinds of hexagonal lattices, that with and that without a boundary, are con-
sidered in this paper. The former is heterogeneous due to the presence of the boundary,
whereas the latter is not. A trade-off exists by which the former is more realistic and
the latter is suitable for theoretical study. It is a basic strategy employed in this paper
to describe and understand agglomeration characteristics of the lattice with a boundary
based on theoretical information drawn from the lattice without a boundary.
Whereas real economic activities accommodate models of various kinds, to deepen
discussion of the stability, we refer to a specific economic geography model, i.e., that of
6A rare comparative study of the long narrow economy and the racetrack economy in a continuous
space was conducted by Mossay and Picard (2011).
7 For the two-place economy, the break point of the transport cost was highlighted as a key concept
(Fujita, Krugman, and Venables, 1999). Formulas to determine the break point and the bifurcation pattern
for a class of footloose-entrepreneur models have been presented by Pflüger and Südekum (2008).

Forslid and Ottaviano (2003), which is a version of modeling by Krugman (1991).8 By

numerical comparative static analysis using this model, the most likely stable progress
of agglomeration patterns is shown to be the formation of the smallest hexagonal pattern
and gradual transition to patterns with larger market areas, finally leading to an atomic
monocenter en route to a megalopolis (Figures 2(b) and 2(c)). Stable racetrack patterns
(Figure 2(d)) expressing the decay of the center of the domain (downtown) are shown
to exist.
The agglomeration out of uniformity is therefore observed based on numerical results
for specific parameter values. We propose formulas for the break point9 at which the
uniformity is broken under reduced transport costs. When investment in transportation
infrastructure is committed continuously to increase the downtown population, a break
point indexes the functioning of this investment.
This paper is organized as follows. Modeling of a spatial economy for an analytically
solvable core-periphery model is presented in section 2. Bifurcating agglomeration pat-
terns for a two-dimensional economy are predicted theoretically in section 3. Formulas for
break points are presented in section 4. Stable agglomeration patterns in the hexagonal
lattice without a boundary are investigated numerically in section 5. Stable patterns in
the lattice with a boundary are examined in section 6.
2. MODELING OF THE SPATIAL ECONOMY

Modeling of the spatial economy is presented in this section. As a representative of
spatial economy models, an analytically solvable core-periphery model by Forslid and Ot-
taviano (2003) is used.10 The fundamental logic and governing equation of a multiregional
version of the model are presented based on work of Akamatsu and Takayama (2009), as
well as that of Ikeda et al. (2014).
Basic assumptions are presented in the first subsection. Market equilibrium is in-
troduced in the second subsection, and spatial equilibrium is presented in the third
subsection.
Basic Assumptions
The economy of this model comprises K places (labeled i = 1, . . . , K), two factors of
production (skilled and unskilled labor), and two sectors (manufacturing, M, and agri-
culture, A). Both H skilled and L unskilled workers consume final goods of two types:
manufacturing sector goods and agricultural sector goods. Workers supply one unit of
each type of labor inelastically. Skilled workers
are mobile among places, and the number
of skilled workers in place i is denoted by λi ( Ki=1 λi = H). The total number H of skilled
workers is normalized as H = 1. Unskilled workers are immobile and distributed equally
across all places with unit density (i.e., L = 1 × K).
Preferences U over the M- and A-sector goods are identical across individuals. The
utility of an individual in place i is
8 Models of this kind emphasize the trade-off between the transport cost and scale economies (e.g.,
Baldwin et al., 2003). There are unskilled and skilled workers. The former are immobile and equally
distributed along places, whereas the latter (footloose entrepreneurs) are mobile and choose a place to
maximize wages. Immobile workers can be interpreted as a population attached to certain amenities.
9 This study is conducted by extension of the strategy for a racetrack economy described by Akamatsu
et al. (2012) to a hexagonal lattice.

10 This model replaces the production function of Krugman (1991) with that of Flam and Helpman
(1987).


(1) U CiM , CiA = μ ln CiM + (1 − μ) ln CiA (0 < μ < 1),
where μ is a constant parameter expressing the expenditure share of manufacturing

sector goods, CiA stands for the consumption of the A-sector product in place i and CiM
represents the manufacturing aggregate in place i, which is defined as
⎛ ⎞σ/(σ−1)
K nj
(2) Ci ≡ ⎝
M
qji ()(σ−1)/σ
d⎠ ,
j=1 0
where qji () is the consumption in place i of a variety ∈ [0, n j ] produced in place j, n j
is the number of produced varieties at place j, and σ > 1 is the constant elasticity of
substitution between any two varieties. The budget constraint is given as
K
nj
(3) pAi CiA + pji ()qji ()d = Yi ,
j=1 0
where pAi is the price of A-sector goods in place i, pji () is the price of a variety in place i
produced in place j, and Yi is the income of an individual in place i. The incomes (wages)
of skilled workers and unskilled workers are represented, respectively, by wi and wLi .
An individual in place i maximizes the utility in (1) subject to the budget constraint
in (3). This yields the following demand functions of
Yi Yi ρσ−1 Yi
(4) CiA = (1 − μ) , CiM = μ , qji () = μ i
,
pAi ρi pji ()σ
where ρi denotes the price index of the differentiated products in place i, which is
⎛ ⎞1/(1−σ)
K nj
(5) ρi = ⎝ pji ()1−σ d⎠ .
j=1 0
Because the total income in place i is wi λi + wLi , the total demand Qji () in place i for a
variety produced in place j is given as
ρσ−1
(6) Qji () = μ i
(wi λi + wLi ).
pji ()σ
The A-sector is perfectly competitive and produces homogeneous goods under
constant-returns-to-scale technology, which requires one unit of unskilled labor per unit
output. A-sector goods are transported without transportation cost and are chosen as the
numéraire. In equilibrium, we have pAi = wLi = 1 for each i.
The M-sector output is produced under increasing-returns-to-scale technology and
Dixit-Stiglitz monopolistic competition. A firm incurs a fixed input requirement of α units
of skilled labor and a marginal input requirement of β units of unskilled labor. An M-sector
firm located in place i chooses (pi j () | j = 1, . . . , K) that maximizes its profit

K

(7) Πi () = pi j ()Qi j () − αwi + βxi () ,
j=1
where xi () denotes the total supply of variety produced in place i and (αwi + βxi ())
signifies the cost function introduced by Flam and Helpman (1987).

The transportation costs for M-sector goods are assumed to take the iceberg form.
That is, for each unit of M-sector goods transported from place i to place j (= i), only a
fraction 1/Ti j < 1 actually arrives (Tii = 1). Consequently, we have

K
(8) xi () = Ti j Qi j ().
j=1
It is assumed that Ti j = Ti j (τ) is a function in a transport cost parameter τ.

Then the profit function of an M-sector firm in place i, given in (7) above, can be
rewritten as
⎛ ⎞
K
K
(9) Πi () = pi j ()Qi j () − ⎝αwi + β Ti j Qi j ()⎠ ,
j=1 j=1
which is maximized by the firm. The first-order condition for this profit maximization
yields
σβ
(10) pi j () = Ti j .
σ−1
This implies that pi j (), Qi j (), and xi () are independent of . Therefore, argument is
suppressed in the sequel.
Market Equilibrium
In the short run, skilled workers are immobile between places, i.e., their spatial
distribution λ = (λ1 , . . . , λK ) is assumed to be given. The market equilibrium conditions
consist of three conditions: the M-sector goods market clearing condition, the zero-profit
condition attributable to the free entry and exit of firms, and the skilled labor market
clearing condition. The first condition is written as (8) above. The second requires that
the operating profit of a firm, given in (7), be absorbed entirely by the wage bill of its
skilled workers. This gives
⎧ ⎫
1 ⎨ ⎬
K
(11) wi = pi j Qi j − βxi .
α⎩ ⎭
j=1
The third condition is expressed as

(12) αni = λi .
The price index ρi in (5) can be rewritten as
⎛ ⎞1/(1−σ)
σβ ⎝ 1
K
(13) ρi = λ j dji ⎠
σ−1 α
j=1
by (10) and (12). Here

(14) dji = Tji1−σ
is a spatial discounting factor between places j and i representing distance decaying
friction.
The market equilibrium wage wi in (11) can be represented as
μ di j
K
(15) wi = (w j λ j + 1)
σ Δj
j=1


using (6), (8), (10), (13), and (14). Here, Δ j = K k=1 dk j λk . Equation (15) is solvable for wi
as follows. With the notation

w = (wi ), D = (di j ), Δ = diag(Δ1 , . . . , ΔK ),
Λ = diag(λ1 , . . . , λK ), 1 = (1, . . . , 1) ,
Equation (15) can be written as
μ
(16) w= DΔ−1 (Λw + 1),
σ
which is solved for w as
μ μ −1
(17) w= I − DΔ−1 Λ DΔ−1 1.
σ σ
The indirect utility vi is obtained as
μ
(18) vi = ln Δi + ln wi .
σ−1
We define the indirect utility function vector v = v(λ, τ) = (v1 (λ, τ), . . . , vK (λ, τ)) .
Spatial Equilibrium Conditions

For the description of long-run dynamics for mobile workers, we assume an adjust-
ment dynamics
dλ(t)
(19) = F(λ(t), τ).
dt
The spatial (long-run) equilibrium of the economic state is defined as (λ, τ) which satisfies
(20) F(λ, τ) = 0.
The stability of a spatial equilibrium λ and the occurrence of bifurcation can be investi-
gated via eigenanalysis11 of the Jacobian matrix J(λ, τ) = ∂F/∂λ.
The skilled workers are assumed to be heterogeneous in their preferences for location
choice (e.g., Tabuchi and Thisse, 2002; Murata, 2003). Then we have a specific functional
form
(21) F(λ, τ) = HP(v(λ, τ)) − λ.
Here, H = 1 and P(v) = (P1 , . . . , PK ) is the choice function vector of the logit type of choice
functions
exp(θvi )
(22) Pi (v) = K ,
j=1 exp(θv j )
where θ is a positive parameter.12 The adjustment process described by (19) with (21) and
(22) is called the logit dynamics (e.g., Fudenberg and Levine, 1998).
11 A solution is designated as linearly stable if every eigenvalue of the Jacobian matrix J(λ, τ) has
a negative real part, and is designated as linearly unstable if at least one eigenvalue has a positive real
part. Bifurcation might take place when one or more eigenvalues become zero.
12 Parameter θ in (22) denotes the inverse of variance of the idiosyncratic taste, which is assumed to
follow the Gumbel distribution that is identical across places (e.g., Anderson, de Palma, and Thisse, 1992).
In the limit of θ → ∞, this reduces to the standard replicator dynamics.

y 9 9
1 2 1 2
9
9 9
2
1 2 x
1 1 2 1 2
(a) 3 × 3 Hexagonal lattice (b) Spatially repeated

with a periodic boundary 3 × 3 hexagonal lattices
FIGURE 3: System of Places on a 3 × 3 Hexagonal Lattice with a Periodic Boundary.
3. BIFURCATING HEXAGONAL AND NONHEXAGONAL PATTERNS

Spatial properties and microeconomic properties are highlighted in this paper as
major sources of spatial agglomeration, and the former properties, which are model in-
dependent, are studied in this section. In sections 4–6, these properties are meshed with
microeconomic properties in section 2 to enable the study of stable agglomeration in two
dimensions.
A hexagonal lattice without a boundary is introduced as a two-dimensional spatial
platform suited for theoretical treatment. This lattice serves as a discretized counterpart
of the isotropic infinite plain in central place theory. Possible bifurcating patterns are
classified as a summary and a reorganization of the theoretical studies of Lösch’s hexagons
and nonhexagonal patterns in Ikeda and Murota (2014) and Ikeda et al. (2014). These
nonhexagonal patterns are given new interpretations related to central place formation,
decentralization leading to decay of downtown, and formation of a megalopolis.
A hexagonal lattice without a boundary is introduced in first subsection and Lösch’s
hexagons are explained in second subsection. Procedures for group-theoretic bifurcation
analysis are presented in third subsection. The bifurcation of the smallest Lösch’s hexagon
is investigated in fourth subsection as a simple example, and bifurcating equilibria are
classified in fifth subsection.
A Hexagonal Lattice without a Boundary

A hexagonal lattice comprising uniformly spread n × n places without a boundary13 is
used as two-dimensional economic space (see Figure 3a for n = 3). Goods are transported
along the homogeneous transportation link of this lattice connecting neighboring places
by interplace roads of the same length.
13 To be strict, we consider a periodic boundary. By virtue of this boundary, this lattice can be repeated
spatially to cover infinite two-dimensional space. Every place is linked to six hexagonal neighboring places
(Figure 3b).

(a) D = 1 (flat earth) (b) D = 3 (c) D = 4 (d) D = 7

Note: These patterns are obtained by spatially repeating n × n hexagonal lattices and cutting out hexagonal
windows; circles represent first-level places.
FIGURE 4: Lösch’s Hexagons on a Hexagonal Lattice.
The transport cost Ti j between places i and j is defined as an iceberg form14

(23) Ti j = exp(τ m(i, j) L̃),
where τ is the transport cost parameter, m(i, j) is an integer proportional to the shortest
link between places i and j along the lattice, and L̃ is the distance unit, which is chosen
as 1/n for the n × n hexagonal lattice. (We define Tii = 1.) With the use of
(24) r = exp[− τ(σ − 1)L̃]
(0 < r < 1 for τ > 0) expressing trade freeness, the spatial discounting factor di j = Ti1−σ
j
in (14) is expressed as
(25) di j = r m(i, j) .
The flat earth equilibrium (uniform distribution),
1
(26) λ∗ = (1, . . . , 1)
K
on the n × n hexagonal lattice with K = n2 , is a solution to the spatial equilibrium equa-
tion F(λ, τ) = 0 in (20) for any value of the parameter τ. Consequently, (λ∗ , τ) forms an
equilibrium curve.
Lösch’s Hexagons
Lösch’s hexagons are advanced as geometrically feasible agglomeration patterns in
an infinite plain in central place theory (Lösch, 1940). The spatial period L between
spatially repeated hexagons is defined as the shortest Euclidean distance between the
first-level centers. This period takes some specific values, such as
√
(27) L/L̃ = D, D = 1, 3, 4, 7, 9, 12, 13, 16, 19, 21, 25, . . . ,
where L̃ is the distance unit between two neighboring places. Figure 4 depicts some
of these hexagons, where the area of a circle represents the population size. There are
first-level places with large populations denoted by circles, whereas the other places
have extremely small populations. The smallest value D = 1 corresponds to the flat earth
14 The
present discussion with minor modifications is applicable to models that employ a linear
transport cost (e.g., Beckmann, 1976; Ottaviano, Tabuchi, and Thisse, 2002; Thisse, 2010; Mossay and
Picard, 2011).

equilibrium. The next three smallest values of D = 3, 4, and 7 are, respectively, associated
with Christaller’s k = 3, 4, and 7 systems (Christaller, 1933). The emergence of these
hexagons on the hexagonal lattice has been studied by Ikeda and Murota (2014) using
group-theoretic bifurcation analysis (the following subsection).
Group-Theoretic Bifurcation Analysis Procedure

Restricting possible bifurcation modes emerging on the hexagonal lattice to a subset
of eigenvector directions using group-theoretic analysis is justified. Herein, we briefly
introduce the procedure of this analysis to identify possible bifurcation modes from the
flat earth equilibrium curve (Ikeda and Murota, 2014).
Symmetry of the n × n hexagonal lattice is characterized by invariance with respect
to geometric transformations of four kinds:
r: counterclockwise rotation about the origin at an angle of π/3,
s: reflection y → −y,
p1 : periodic translation along the 1 -axis (i.e., the x-axis), and
p2 : periodic translation along the 2 -axis.
Consequently, the lattice symmetry is described by the group generated by these
transformations as G = r, s, p1 , p2 , where · · · denotes a group generated by the elements
therein.
In investigation of the possible patterns of the bifurcated solutions, it is crucial to
formulate the symmetry of the governing equation for the system of n × n places on the
hexagonal lattice. This symmetry is formulated as the condition
(28) T(g)F(λ, τ) = F(T(g)λ, τ), g∈G
in terms of a K × K orthogonal matrix representation T of group G = r, s, p1 , p2 .
15
Consider a critical point (λ∗ , τc ) on the flat earth equilibrium curve, which is said to
have multiplicity M (≥ 1) if the Jacobian matrix J = ∂F/∂λ of F at (λ∗ , τc ) has M zero
eigenvalues. Let (qi | i = 1, . . . , K) be an orthonormal basis of RK such that
(29) Jqi = 0, i = 1, . . . , M.
We express the variable λ as

M
(30) λ = λ∗ + ξi qi
i=1
and τ as
(31) τ = τc + τ̃,
where τ̃ denotes an increment of τ.
The full system of equations F(λ, τ) = 0 in (20) is reduced,16 in a neighborhood of
∗
(λ , τc ), to a system of M equations (called bifurcation equations)
(32) F̃(ξ, τ̃) = 0
15 Matrix representation means that (i) for each element g ∈ G, T(g) is a K × K matrix with
T(g) T(g) = I (identity matrix), and (ii) T(g)T(h) = T(gh) for all g, h ∈ G.
16 This is a standard procedure called the “Liapunov–Schmidt reduction with symmetry” (Golubitsky,
Stewart, and Schaeffer, 1988; Ikeda and Murota, 2014).

(a) q(3)
1 (hexagon with D = 3) (b) q(3)
2 (c) − q(3)
1 (racetrack)
Note: A black circle denotes a positive component, a white circle indicates a negative one, and the area of a
circle expresses the magnitude of the component.
FIGURE 5: Spatial Patterns Expressed by Vectors q(3) (3) (3)

1 , q2 , and −q1 on the 6 × 6
Hexagonal Lattice.
for some function F̃ in ξ = (ξ1 , . . . , ξM ) ∈ RM in (30) and τ̃ ∈ R in (31). In this reduction

process, the symmetry condition (28) of the full system is inherited by the reduced system
(32).
The reduced equation (32) is to be solved for ξ as ξ = ξ(τ̃), which is often possible by
virtue of the symmetry inherited by F̃. Because (ξ, τ̃) = (0, 0) is a singular (critical) point of
(32), there can be many solutions ξ = ξ(τ̃) with ξ(0) = 0, which give rise to bifurcation. By
(30), each ξ uniquely
determines a solution λ of the full system (20). Among the (critical)
eigenvectors M ξ
i=1 i i , only those vectors which satisfy (32) are related to bifurcating
q
equilibria, whereas those which do not satisfy (32) are not. In this way, possible bifurcating
modes can be predicted using group-theoretic bifurcation analysis. It is noteworthy that
several bifurcating modes satisfying (32) possibly coexist at the same bifurcation point.
Lösch’s Hexagon with D = 3: Simple Example

Bifurcation producing Lösch’s hexagons is explained using the smallest hexagon with
D = 3 (Ikeda and Murota, 2014), which plays the most important role in this study. This
hexagon is associated with a bifurcation point with twice repeated zero eigenvalues (M =
2) of the Jacobian matrix J(λ, τ) = ∂F/∂λ. For the 6 × 6 hexagonal lattice, for example,
the two eigenvectors are given explicitly as
1
q(3)
1 = √ (cos(2π(n1 − 2n2 )/3) | n1 , n2 = 0, 1, . . . , 5)
3 2
1
= √ (2 −1 −1 2 −1 −1 −1 −1 2 −1 −1 2 −1 2 −1 −1 2 −1
6 2
2 −1 −1 2 −1 −1 −1 −1 2 −1 −1 2 −1 2 −1 −1 2 −1) ,
1
q(3)
2 = √ (sin(2π(n1 − 2n2 )/3) | n1 , n2 = 0, 1, . . . , 5)
3 2
1 √ √ √ √ √ √ √ √ √ √ √ √
= √ (0 3 − 3 0 3 − 3 3 − 3 0 3 − 3 0 − 3 0 3 − 3 0 3
6 2
√ √ √ √ √ √ √ √ √ √ √ √
0 3 − 3 0 3 − 3 3 − 3 0 3 − 3 0 − 3 0 3 − 3 0 3) .
Therein, the components are defined in accordance with Figure 3. These eigenvectors are
depicted in Figures 5(a) and 5(b), in which positive components of the eigenvectors are

expressed by (•) and negative ones by (◦). Their magnitudes are expressed by the areas of
the circles. √
Bifurcating solutions in the six directions expressed by ±q(3)
1 and ± 1 (3)
q
2 1
± 2
3 (3)
q2 can
(3)
be found by solving the bifurcation equation (32). Vector q1 represents Lösch’s hexagon
with D = 3, as shown by the dashed lines in Figure 5(a), in which the first-level place (•)
with an increasing population is surrounded by six second-level places (◦) with decreasing
populations. Vector −q(3)1 represents a spatially repeated racetrack pattern as depicted
by the dashed circles in Figure 5(c), in which the second-level place with a decreasing
population shown by () is surrounded by six first-level places with increasing popu-
lations shown by (•). Consequently, the hexagon with D = 3 for q(3) and the racetrack
for −q(3) √emerge simultaneously at the same bifurcation point. The other four vectors
± 12 q(3) 3 (3)
1 ± 2 q2 represent spatially shifted patterns of ±q1 .
(3)
Classification of Bifurcating Equilibria

Bifurcating equilibria of the hexagonal lattice with an arbitrary lattice size n have
been described in earlier reports by Ikeda and Murota (2014) and Ikeda et al. (2014).
The bifurcating equilibria for the lattice size n = 6, which encompass hexagonal and
nonhexagonal patterns of various kinds, are described below as a summary of those
reports.
It is necessary to introduce an index k that characterizes bifurcation points by im-
plying the size D of possible bifurcating hexagons (see Remark A1 in Appendix A for
additional details). Consider a bifurcation point associated with k and let M be its multi-
plicity. Then the Jacobian matrix J(λ, τ) = ∂F/∂λ at λ = λ∗ has M (mutually orthogonal)
eigenvectors corresponding to zero eigenvalues, which are denoted by
(33) q(k) (k)

1 , . . . , qM .
The possible values of k and the correspondence between k and M are given as
k 1 3 4 9 12 36(I) 36(II)
(34) M 1 2 3 6 6 6 12
Here, two k values, k = 36(I) and k = 36(II), are associated with D = 36. The concrete
forms of the eigenvectors in (33) are given by discrete cosine and sine series in (A2)–(A8)
in Appendix A.
The eigenvectors for the hexagonal patterns are obtained as
⎧
⎪ (3)
⎨ q(3) = q1 ,
(35) q(4) = q(4) (4)
1 + q2 + q3 ,
(4)
⎪
⎩ q(k) = q(k) + q(k) + q(k) ,
1 3 5 k = 9, 12.
The hexagonal patterns represented by these eigenvectors are illustrated in Figures 5(a)
and 6(a)–6(d); the hexagon with D = 7 on the 7 × 7 hexagonal lattice17 in Figure 6(b) is
included for comparison.
At these bifurcation points, hexagonal and nonhexagonal patterns branch simultane-
ously. In the following, nonhexagonal patterns, which are not considered in central place
theory, are given new interpretations. These patterns are termed the megalopolis, the
17 The hexagon with D = 7 exists when the lattice size n is a multiple of 7 (but not when n = 6).

(a) Hexagon with D = 4 (q (4) ) (b) Hexagon with D = 7 (q (7) )
(c) Hexagon with D = 9 (q (9) ) (d) Hexagon with D = 12 (q(12) )
(e) Megalopolis with D = 36 (q(36(I)) ) (f) Megalopolis with D = 36 (q(36(II)) )

Notes: The lattice size n is chosen as n = 6 except for n = 7 in (b); these patterns are obtained by spatially re-
peating n × n hexagonal lattices and cutting out hexagonal windows. A black circle denotes a positive component,
a white circle indicates a negative one, and the area of a circle expresses the magnitude of the component.
FIGURE 6: Hexagonal and Megalopolis Patterns on a Hexagonal Window Expressed by

Eigenvectors q(k) .
racetrack, the long narrow, and the deformed hexagonal patterns.18 Megalopolis patterns,
shown in Figures 6(e) and 6(f), are associated with eigenvectors:

q(36(I)) = q(36(I))
1 + q(36(I))
3 + q(36(I))
5 ,
(36) (36(II)) (36(II))
q(36(II))
= q1 + q3 + q(36(II))
5 + q(36(II))
7 + q(36(II))
9 + q(36(II))
11 .
18 The existence of the deformed hexagonal pattern is proved in the preprint (Ikeda, Murota, and
Takayama, 2014) of this paper, while the existence of other patterns was proved in Ikeda and Murota
(2014).

Satellite places with small population are scattered around the center of the hexagonal
window (downtown) to form a megalopolis.
Racetrack patterns of several kinds are associated with eigenvectors −q(k) (k =
3, 4, 9, 12, 36(I), and 36(II)) with the reversed sign. These patterns are interpreted as rep-
resenting the decay of a downtown area through decentralization. For −q(3) in Figure 5(c)
and −q(4) in Figure 7(a), one place decaying into a second-level center is surrounded by
six places developing into first-level centers. Semicircular zones of growing places are
observed for −q(12) , −q(36(I)) , and −q(36(II)) (Figures 7d–7f).
Long narrow patterns are given by the eigenvectors q(4) (9) (12)
2 , q3 , q5 , and q3
(36(I))
(36(II))
and q1 . First-level places are located along spatially repeated narrow stripes
(Figure 8). They represent a chain of cities forming an industrial belt in a two-dimensional
infinite space. In particular, q(36(II))
1 in Figure 9 displays a characteristic pattern termed
deformed hexagonal patterns herein, for which the first-level places form spatially
repeated deformed hexagons.
Possible agglomeration patterns in two dimensions have been clarified in preparation
for the study of the stability and economic implications of these patterns in sections 5 and
6.
4. BREAK POINT TRIGGERING SPATIAL AGGLOMERATION

Possible bifurcating patterns are obtained in section 3 based on the spatial properties
of the hexagonal lattice. Formulas for break points are developed in this section by exploit-
ing spatial and microeconomic properties. A break point is defined as the value of τ for the
occurrence of a bifurcation that breaks uniformity (footnote 7). When investment in trans-
portation infrastructure is committed continuously to increase the downtown population,
a break point indexes the functioning of this investment. In particular, it is shown that
the hexagon with D = 3 emerges first and the megalopolis emerges last under decreasing
transport costs. These formulas are derived for the analytically solvable core–periphery
model (section 2) in the hexagonal lattice with size n = 6. Nonetheless, the methodology
used herein is general and is extendable to other spatial economy models (see Remark 1)
and to the hexagonal lattice of any size n.
Theoretical formulas for the break points are given in the first subsection. The order
of emerging hexagons is presented in the second subsection.
Laws for Break Point
Bifurcation points of six kinds associated with k = 3, 4, 9, 12, 36(I), and 36(II) exist
on the flat earth equilibrium curve (λ∗ , τ) with λ∗ = 36 1
(1, . . . , 1) of the 6 × 6 hexago-
nal lattice (subsection “Classification of Bifurcating Equilibria”). The formulas for break
points are presented below with details given in Appendix B. It must be recalled that both
hexagonal and nonhexagonal patterns branch simultaneously at the same bifurcation
points.
A bifurcation point on the flat earth equilibrium λ = λ∗ is determined as the break
point τ = τbreak at which the Jacobian matrix J(λ∗ ) = ∂F ∂λ
(λ∗ ) of the equilibrium equation
has one or more zero eigenvalues (footnote 11).
From the governing equation F in (21) with H = 1, the Jacobian matrix J(λ) is related
to the Jacobian matrix V (λ) = ∂v/∂λ of the indirect utility as (see (B12))
⎡ ⎤ ⎡ ⎤
P1 P1
⎢ ⎥ ⎢ ⎥
(37) J(λ) = −θ⎣ ... ⎦ P1 · · · PK V (λ) + θ⎣ ..
. ⎦V (λ) − I,
PK PK

(a) −q (4) (racetrack) (b) −q (7) (racetrack)
(c) −q (9) (racetrack) (d) −q (12) (racetrack)
(e) −q (36(I)) (racetrack) (f) −q (36(II)) (racetrack)

Notes: The lattice size n is chosen as n = 6 except for n = 7 in (b). A black circle denotes a positive component,
a white circle indicates a negative one, and the area of a circle expresses the magnitude of the component.
FIGURE 7: Racetrack Patterns on Hexagonal Windows Expressed by Eigenvectors

−q(k) .

(a) q(4)
2 (b) q(9)
3
(c) q(12)
5 (d) q(36(I))
3
FIGURE 8: Long Narrow Patterns on the Hexagonal Lattice Expressed by Eigenvectors

on the 6 × 6 Hexagonal Lattice.
q(36(II))
1
FIGURE 9: Deformed Hexagonal Pattern on the Hexagonal Window Expressed by an

Eigenvector on the 6 × 6 Hexagonal Lattice.
where I is the identity matrix. At the flat earth equilibrium λ∗ , (37) yields
θ θ
(38) J(λ∗ ) = − 11V (λ∗ ) + V (λ∗ ) − I,
K2 K

where 1 = (1, . . . , 1) . As worked out in (B14)–(B21) in Appendix B, we have

!
∗ D D −1 D D
(39) V (λ ) = K κ + I − κ · κI − ,
d d d d
where D = (di j ) is the spatial discounting matrix given by (25),

μ μ
(40) κ= , κ =
σ σ−1
and d = d(r) = 1 + 6r + 12r2 + 15r3 + 2r4 with r being the trade freeness parameter in-
troduced in (24).
The eigenvector η of J(λ∗ ) is also an eigenvector of D, i.e.,
(41) Dη = ˜ (k) η, k = 3, 4, 9, 12, 36(I), 36(II)
with the eigenvalue ˜ (k) given (see (B6)) by

⎧
⎨ 1 − 3r + 3r2 − 3r3 + 2r4 for k = 3,
2 3 4
⎪
(42) ˜ = ˜ (r) =
(k) (k) 1 − 2r + 4r − 5r + 2r for k = 4,
⎪
⎩ . ..
.. .
From (38) with (39) and (41), we obtain (see (B22)–(B24))

(κ − ) 1
(43) J(λ∗ ) · η = θ κ + − η
1 − κ θ
with
˜ (k) (r)
(44) = ,
d(r)
which is solved for r by the implicit function theorem as
(45) r = Φ(k) ()
for some function Φ(k) .

The break point τbreak is determined from the condition that the eigenvalue of the
Jacobian matrix J(λ∗ ) given in (43) for the eigenvector η becomes zero, where r and
τ are related by (24). As worked out in (B25)–(B27), this condition gives a quadratic
equation θ(b − a 2 ) − 1 = 0 in , where a = κκ + 1 > 0 and b = κ + κ + θ−1 κ > 0. Of the
two solutions of this equation, the larger,
√
b + b2 − 4aθ−1
(46) +∗ =
2a
is related to the first bifurcation when τ is reduced from a large value, and the smaller,
√
∗ b − b2 − 4aθ−1
(47) − =
2a
is related to the last bifurcation. We have 0 < −∗ < +∗ .
When the transport cost parameter τ is reduced continuously from +∞ to 0, two
break points are encountered for each hexagon under certain conditions of the values of
μ, σ, and θ. Recall that n = 6.

PROPOSITION 1. For each Lösch’s hexagon and associated nonhexagonal distribution(s),

two break points, τ+ and τ− , exist with τ+ > τ− > 0 when the following conditions19 are
satisfied:
μ
(48) < 1 + θ−1 ,
σ−1
" #2
1 1 μ2
(49) μ2 (1 + θ−1 ) + − 4θ−1 + 1 > 0.
σ σ−1 σ(σ − 1)
Because these conditions for the existence of break points are common for the
hexagons with k = 3, 4, 9, 12, 36(I), and 36(II), all these hexagons disappear if either of
the conditions is not satisfied. This actually presents the worst-case scenario in downtown
development through social investment in that no agglomeration emerges at whatever
cost.
(k) ∗
The value of r = r(k) (k) (k)
± corresponding to = ± is given as r± = Φ (± ) by (45). Then,
from r = exp[− τ(σ − 1)L̃] in (24) with L̃ = 1/n = 1/6, the break point τ(k)± is given in the
following proposition.
PROPOSITION 2. Break points τ(k) (k)
+ and τ− for Lösch’s hexagon and associated nonhexag-
onal distribution(s) are given as
6 6
τ(k)
+ = − ln(Φ(k) (+∗ )), τ(k)
− =− ln(Φ(k) (−∗ )),
σ−1 σ−1
(50) k = 3, 4, 9, 12, 36(I), 36(II).
Although formula (50) is rigorous, the following approximate formula is convenient

for discussing parameter dependence of the break point for Lösch’s hexagon with D = 3.
PROPOSITION 3. Under the condition
(51) θ (σ/μ)2 1,
the larger break point τ(3)
+ for Lösch’s hexagon with D = 3 is given approximately as
μ1/3
(52) + = 18 · 2
τ(3) 1/3
.
(σ − 1)4/3
Formula (52) indicates that the onset of agglomeration is hastened by a lower sub-
stitution σ between any two varieties and a higher expenditure share μ of manufac-
tured goods. This fact concurs with economic intuition and the numerical examples to be
presented in subsections “Progress of Stable Equilibria” and “Parameter Dependence of
Progress of Stable Equilibria.”
REMARK 1. The proposed procedure for the analysis of break points might be adapted to
many spatial economy models for which the indirect utility (or profit) vector v takes the form
v = v(λ, D). Here, D = (di j ) is the matrix of spatial discounting factors di j = di j (r) that are
functions in r. In social interaction models, r is given as a monotonically increasing function
of the parameter τ expressing accessibility between places (see, e.g., Fujita and Ogawa, 1982;
the limit of θ → +∞ (footnote 12), the second condition (49) is always satisfied and the first
19 In
condition (48) reduces to the no-black-hole condition μ/(σ − 1) < 1 described by Forslid and Ottaviano
(2003).

Tabuchi, 1986). In contrast, in NEG models (see, e.g., Oyama, 2009; Akamatsu et al., 2012),
r is given as a monotonically decreasing function.
Order of Emerging Hexagons

There are several bifurcations that engender hexagonal patterns of various kinds
(section 3). It is possible to predetermine the order of the emergence of hexagonal patterns
as expounded in Proposition 4 (see appendix subsection “Order of Emerging Hexagons”
for the proof).
PROPOSITION 4. The flat earth equilibrium is stable for a large τ (> τ(3) + ). When τ is
reduced continuously from +∞ to 0 [or when r is increased continuously from 0 to 1] and
bifurcations take place, these bifurcations occur in the following order:
(i) Bifurcation producing Lösch’s hexagon with D = 3 occurs first at τ = τ(3)
+ .
(ii) Bifurcation producing the megalopolis occurs last at τ = τ(36(I))

− .
The smallest hexagon with D = 3 is the most important one that breaks uniformity.
It is no wonder that this hexagon was highlighted as Christaller’s k = 3 system. Another
hexagon with k = 36(I), which is not considered in central place theory, is also important
because this hexagon expresses centralization leading to a megalopolis.20
5. STABLE AGGLOMERATION PATTERNS: A HEXAGONAL LATTICE WITHOUT A

BOUNDARY
Based on the theoretically possible agglomeration patterns presented in subsec-
tion “Classification of Bifurcating Equilibria,” this section tackles the main objective of
this paper, namely to elucidate which patterns are stable. Equilibria of the 6 × 6 hexag-
onal lattice are studied using comparative static analysis with respect to the transport
cost of the core-periphery model of Forslid and Ottaviano (2003) (section 2). Stable equi-
libria related to central place formation, decentralization leading to decay of downtown,
and formation of a megalopolis are shown to exist in the first subsection. The progress of
stable equilibria under decreasing transport cost is examined in the second subsection.
Parameter values are chosen as (σ, μ) = (5.0, 0.4), following Fujita, Krugman, and
Venables (1999). The parameter θ in (22) is chosen as θ =1,000; the fixed input requirement
is chosen as α = 1.0. This set of parameter values satisfies the conditions (48) and (49) for
the existence of break points presented in Proposition 1. The distance unit is chosen as
L̃ = 1/6.
Bifurcating Stable Equilibria

The equilibrium curves obtained by comparative static analysis with respect to the
transport cost parameter τ are shown in Figure 10, which plots the relations between τ
and the population λcenter of the place at the center of the hexagonal window. The flat
earth equilibrium λ∗ = 36 1
(1, . . . , 1) in (26), which exists for any value of τ, forms the
equilibrium curve on the horizontal line λcenter = 1/36. Bifurcation points A through L on
this curve were obtained using formula (50). In agreement with the theoretical prediction
of the order of the occurrence of break points (Proposition 4), a bifurcation producing
Lösch’s hexagon with D = 3 at point A occurred first and that producing the megalopolis
at point L occurred last.
20 This megalopolis formation is inherent for the logit dynamics used herein, but it is absent for the
replicator dynamics.

D = 36 M=6
M=2
M=3 M = 12
1
a a
0.9
0.8
0.7
D = 12 D=9
0.6
center
0.5
0.4
Deformed hexagon D=4
b
0.3 c
0.2 d
a e f
0.1
1/36 D=3
0 L KJ I H G F ED C B A
0 0.5 1 1.5 2 2.5
Transport cost parameter 2.42
Note: Solid curves represent stable equilibria, dashed ones represent unstable ones, and M is the multiplicity
of critical points on the flat earth equilibrium curve and on bifurcating equilibrium curves.
FIGURE 10: Equilibrium Curves Related to Hexagonal Patterns and Associated

Population Distributions Displayed in the Hexagonal Windows Containing 36 Places.
Bifurcating equilibria21 branching from these bifurcation points were found in ac-
cordance with the theoretical prediction in the subsection “Classification of Bifurcating
Equilibria.” Equilibrium curves for bifurcating hexagonal patterns are shown in Figure 10
as a slight modification of Ikeda, Murota, and Akamatsu (2012). As new results presented
in this paper, the patterns on the curves a–a –a for very small values of τ are investi-
gated in detail. In addition, unstable equilibrium curves were computed and are shown
by dashed curves in Figure 10. Stable population distributions found in this manner are
depicted in Figure 11 using hexagonal windows containing 36 places. Therein, the area of
a circle represents the population size of first-level places with large populations, whereas
other locations have very small populations.
In addition to the hexagons in central place theory, nonhexagonal patterns are pre-
dicted by bifurcation theory (subsection “Classification of Bifurcating Equilibria”) and
have been observed fragmentarily in one-dimensional economies, as reported in the lit-
erature. Stable nonhexagonal patterns found here are listed below. The megalopolis
pattern (curve a –a in Figure 10) and the atomic monocenter pattern (curve a–a )
21 Most bifurcating equilibrium curves display the so-called catastrophic agglomeration with hystere-
sis (Pflüger and Südekum, 2008).

¨
(a) Losch hexagon (D = 3) ¨
(b) Losch hexagon (D = 4) ¨
(c) Losch hexagon (D = 7)
¨
(d) Losch hexagon (D = 9) (e) Losch
¨ hexagon (D = 12) (f) Deformed hexagon
or
(g) Racetrack (− q(12) ) (h) Racetrack or triangle
(i) Two-places (j) Semisquare (k) Long narrow

Note: Lösch’s hexagon with D = 7 on the 7 × 7 hexagonal lattice in (c) does not exist on the 6 × 6 hexagonal
lattice, but it is included for comparison.
FIGURE 11: Market Areas of the First-Level Centers for Numerically Computed Stable
Equilibria Observed in Figures 10, 12, and 13.

associated with the hexagon with D = 36 (k = 36(I)) emerge stably for small τ. These pat-
terns have also been observed in the long narrow economy.22 The deformed hexagonal
pattern (point d in Figures 10 and 11(f)) was overlooked in the previous studies but is
found to be stable for a wide range of τ here.
Other stable nonhexagonal patterns were found, as shown in Figures 12 and 13.
Racetrack patterns in Figures 11(g) and 11(h) are stable for very short ranges of the
transport cost parameter τ (1.71 < τ < 1.74 and 0.66 < τ < 0.76, respectively), as shown
at points k and i in the equilibrium curves in Figure 12. These patterns are interpreted as
decentralization leading to downtown decay or extinction. They are a major topic in the
discussion of economic implications. For example, after World War II, rising automobile
use contributed to the decentralization of American downtown areas (Robertson, 1995).
The two-place and the semisquare patterns at points l and n in Figure 13 (Figures 11(i) and
11(j)) are quite close to those found in the racetrack economy, in which four identical first-
level places are transformed into two identical places (Ikeda, Akamatsu, and Kono, 2012;
Akamatsu et al., 2012). The long narrow patterns are unstable except for that at point m
in Figure 13 (Figure 11(k)). Although these patterns resemble an industrial belt, such as
the Atlantic seaboard of the United States, they play a limited role in agglomeration in
wide two-dimensional space, such as southern Germany.
The shapes of market areas in Figure 11, which display hexagons, deformed hexagons,
rectangles, diamonds, and trapezoids, influence the stability of the associated agglomer-
ation patterns. Lösch’s hexagons have superior stability as they remain stable in wide
ranges of τ, and have geometrically superior shapes of market areas highlighted in cen-
tral place theory. The deformed hexagonal pattern with semihexagonal market areas also
has superior stability. The racetrack pattern associated with −q(12) in Figure 11(g) has reg-
ular triangular market areas with spatial orderliness, which presumably has contributed
to making this pattern stable even in a very short range of τ.
Progress of Stable Equilibria

We now shift our attention to an issue of great economic interest, i.e., the progress
of stable equilibria under decreasing transport cost τ. For this purpose, we refer to the
ranges of τ for stable equilibria depicted in Figure 14, which are obtained from the equilib-
rium curves for bifurcating hexagonal patterns in Figure 10 and those for nonhexagonal
patterns in Figures 12 and 13. When τ is reduced, the number N1st of first-level centers
tends to be reduced and, in turn, the market area is enlarged. This tendency arises from
a trade-off between transportation cost and scale economies.23 Hexagons with D = 3, 4, 9,
and 12 become stable in this order as τ decreases from a large value, thereby showing the
foresight of central place theory, which predicted these hexagons by geometrical consider-
ation. Moreover, we can find several stable nonhexagonal patterns, such as the racetrack,
the semisquare, the long narrow, the two-place, and the deformed hexagonal patterns.
The place at the center of this window can be interpreted as the downtown area
of a city. Accordingly, the progress of agglomeration at this place presents rich economic
implications. The existence of stable racetrack patterns indicates the possibility of decay or
extinction of the downtown area. If only hexagonal patterns were considered, progress of
22 In the long narrow economy, a megalopolis consisting of a continuous industrial zone around the
center was observed by Mori (1997), and a monocenter was observed by Fujita and Mori (1997) and Fujita,
Krugman, and Mori (1999).
23 Firms operating in a small market area enjoy the benefit of low transportation costs at the expense
of small-scale economies. In contrast, firms in large market areas enjoy the benefit of scale economies at
the expense of high transportation costs.

1.0 M=2 M=6

M=3 M = 12
0.8
Racetrak
or triangle g
h
0.6
D=9
center
0.4
c
0.2 D=3
f
1/36 L J
0.0 h F g A
i
0 0.5 1 1.5 2 2.5
Transport cost parameter
(a)
1.0
0.8
k j
0.6
center
D = 12
0.4
b
0.2
e D=4
1/36
0.0 L j C A
K k
0 0.5 1 1.5 2 2.5
(b)
Note: Solid curves represent stable equilibria, dashed ones represent unstable ones, and M is the multiplicity
of critical points on the flat earth equilibrium curve and on bifurcating equilibrium curves.
FIGURE 12: Equilibrium Curves Related to Racetrack Patterns and Associated

Population Distributions Displayed in the Hexagonal Windows.

l m
0.5
M=2 M=6
M=3 M=1
0.4
D = 12 D=9
b
0.3
n
max D=4
c
0.2
e
0.1 f D=3
1/36
0.0 L KJ F D C A
0 0.5 1 1.5 2 2.5
l m n
Notes: The ordinate λmax stands for the maximum population among 36 places on the hexagonal lattice. Solid
curves represent stable equilibria, dashed ones represent unstable ones, and M is the multiplicity of critical points
on the flat earth equilibrium curve and on bifurcating equilibrium curves.
FIGURE 13: Equilibrium Curves Related to the Two-Place, the Long Narrow, and the
Semisquare Patterns and Associated Population Distributions Displayed in the
Hexagonal Windows.
stable equilibria with decreasing τ would engender continuous progress of centralization

through the increase of the size of hexagonal patterns. This, however, is not a true scenario
because competition exists between centralization by hexagons and decentralization by
racetrack patterns. The importance of the scope of the study, which extends beyond central
place theory and which encompasses diverse patterns other than hexagonal patterns, has
been demonstrated in this manner.
It is noteworthy that multiple stable equilibria exist for intermediate values of τ,
while a unique stable equilibrium exists for a large τ or a small τ. In view of this, the
progress of stable equilibria with decreasing τ is classifiable into three stages: dawn,
intermediate, and mature stages, as presented in Table 1.
The dawn stage was predicted by central place theory. When τ is reduced from a large
value, the flat earth equilibrium becomes unstable at the break point τ = τbreak ≈ 2.42 and

Mature stage Intermediate stage Dawn stage
36 Flat earth
Flat earth
12
N1st
Deformed hexagon
6
Racetrack Racetrack
or triangle
4
3
Two -places Semisquare
2
1 Long narrow pattern
Atomic monocenter and megalopolis
0 0.5 1.0 1.5 2.0 2.5
Notes: The ordinate N1st stands for the number of the first-level places with the largest population. A first-level
place at the corner of the hexagonal window is counted as 1/3, and that at the midpoint of the two neighboring
corners is counted as 1/2 in N1st .
FIGURE 14: Ranges of the Transport Cost Parameter τ for Stable Equilibria.
TABLE 1: Three Stages of Progress of Stable Agglomerations for the Hexagonal Lattice
without a Boundary
Dawn Stage Flat Earth ⇒ Hexagon (D = 3) ⇒
⎧ ⎫
⎪ Hexagon (D = 4, 9, 12) ⎪
⎪
⎪ ⎪
⎪
⎪ Deformed hexagon
⎪ ⎪
⎪
⎨ ⎬
Intermediate stage Racetrack ⇒
⎪
⎪ ⎪
⎪
⎪ Triangle
⎪ ⎪
⎪
⎪
⎩ .. ⎪
⎭
.
Mature stage: Atomic monocenter ⇒ Megalopolis ⇒ Flat earth
renders the role of the unique stable equilibrium to the hexagon with D = 3 (τ ≈ 2.0). Then
a state of dual stable equilibria of the hexagons with D = 3 and D = 4 comes into existence
(1.7 < τ < 2.0). At this stage, the underlying predominance of the market-crowding effect
is weakened by an increase in the market-access effect that enlarges the agglomeration
force. This reorganizes firms into locations with greater competition, thereby engendering
hexagonal patterns.
The intermediate stage is beyond the scope of central place theory. After the coex-
istence of stable equilibria for hexagons with D = 3 and D = 4, we encounter various
kinds of stable equilibria, such as hexagons with D = 4, 9, and 12, the racetrack pat-
terns, and the deformed hexagonal pattern. The market-crowding effect gradually de-
creases, whereas the market-access effect increases. The balance between an increasing

agglomeration force and a decreasing dispersion force determines the emerging agglom-
eration pattern.
The mature stage for small τ is the final stage of economic agglomeration, in which the
atomic monocenter, the megalopolis pattern, and the flat earth equilibrium become stable
in this order. Because the transport cost is extremely low at this stage, the market-access
effect greatly decreases and the dispersion force arising from the taste heterogeneity of
workers prevails. Therefore, the spatial agglomeration pattern appears to be continuous,
unlike at the previous two stages.
At the end of this section, it is emphasized that the existence of stable equilibria is
robust against the change of the elasticity σ of substitution between any two varieties, as
explained in Remark 2. This robust character demonstrates the robustness of the present
discussion of stable equilibria.
REMARK 2. Numerical simulations for σ = 4 and σ = 10 provide similar results to those
obtained for the standard value σ = 5.0, as shown by the durations of stable equilibria for
σ = 4.0 and 10 depicted in Figure 15. The hexagonal (D = 3, 4, 9, and 12), the deformed
hexagonal, the atomic monocenter, and the megalopolis patterns all exist as stable equilib-
ria. Moreover, when the transport cost decreases from a large value, the hexagon with D = 3
is formed first, followed by several stable equilibria, en route to the atomic monocenter, the
megalopolis pattern, and the flat earth equilibrium.
6. STABLE AGGLOMERATION PATTERNS: A HEXAGONAL LATTICE WITH

A BOUNDARY
The hexagonal lattice without a boundary studied in sections 3–5 is a discretized
model of an infinite plain (homogeneous space) in central place theory. This lattice allows
theoretical prediction of agglomeration patterns. Yet a criticism arises that a realistic eco-
nomic space has a boundary that makes the space heterogeneous: places at the boundary
are not as competitive as places inside the boundary.
In search of realistic agglomeration patterns, the hexagonal lattice with a boundary
in Figure 16 is used in this section. This lattice lacks geometric-translational symmetry;
therefore, spatially repeated patterns, such as hexagonal patterns, are absent in a strict
sense for this lattice. To compensate for this absence, its agglomeration is observed with
reference to theoretically predicted spatial patterns of the lattice without a boundary.
Progress of stable equilibria under decreasing transport cost is examined in the
first subsection. Parameter dependence of progress of stable equilibria is addressed in
the second subsection. Parameter dependence of break points is studied in the third
subsection.
Progress of Stable Equilibria

For a lattice with a boundary, stable equilibrium curves and associated population
distributions at points a–f in Figure 17 were obtained for the same parameter values
(σ, μ) = (5.0, 0.4) and L̃ = 1/6 as those used in section 5. There appeared several hexagon-
like patterns that are quite similar to the hexagonal patterns observed on the lattice
without a boundary. It may be understood that this similarity stems from the hexagonal
shape of the lattices and dissipating boundary effects away from the boundaries.
The stable equilibria start with the flat earth equilibrium24 and progress as listed
in Table 2. As in the hexagonal lattice without a boundary (section 5), there are three
24 Because of the boundary effect, the flat earth cannot be an equilibrium in the strict sense of the
word. “Flat earth” in this section denotes a pattern that looks very close to a flat earth.

Flat earth
36
Flat earth
12
N1st 6 Deformed hexagon

Racetrack Racetrack
or triangle
4
3
Two-places Long narrow pattern
2
1
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
(a) σ = 4.0
Flat earth Flat earth

36
12
N1st 6 Deformed hexagon
4
3 Two-places
2 Semisquare
1
0 0.2 0.4 0.6 0.8 1.0

(b) σ = 10.0
Notes: N1st is the number of first-level places in the hexagonal window. A first-level place at the corner of the
hexagonal window is counted as 1/3; and that at the midpoint of two neighboring corners is counted as 1/2 in N1st .
FIGURE 15: Ranges of the Transport Cost Parameter τ for Stable Equilibria for Several
Values of σ.

FIGURE 16: Hexagonal Lattice with a Boundary with 91 Places.
TABLE 2: Progress of Stable Agglomeration Patterns for the Hexagonal Lattice with a
Boundary
Dawn Stage Flat Earth ⇒ Hexagon (D = 3) ⇒ Hexagon (D = 4) ⇒
Intermediate stage: Racetrack (D = 36) ⇒ Hexagon (D = 9) ⇒
Mature stage: Atomic monocenter ⇒ Megalopolis ⇒ Flat earth
stages: dawn, intermediate, and mature stages. At the dawn stage, a hexagon with D = 3
is formed. At the mature stage, a megalopolis pattern emerges in agreement with the
theoretical prediction (Proposition 4). At the dawn stage (τ > 2.0), after the flat earth
equilibrium at point a, a hexagon with D = 3 is formed near the center of the lattice at
point b.25 A state of dual stable equilibria of the hexagon with D = 3 and that with D = 4
(point d) exists at τ ≈ 2. At the intermediate stage (1.1 < τ < 2.0), the racetrack pattern
(point e) and a hexagon with D = 9 (point f) emerge stably. At the mature stage (0 < τ <
1.1), the atomic monocenter, the megalopolis pattern, and the flat earth equilibrium occur
stably in this order.
The durations of stable equilibria for the hexagon with D = 3 for the lattices of
two kinds shown by the solid and dashed lines in Figure 18 display an amazing degree
of quantitative agreement. The racetrack pattern appears in both lattices, but remains
stable for a longer duration in the lattice with a boundary. Other patterns exhibit a
fair degree of qualitative agreement. This suffices to demonstrate the validity of the
basic strategy of this paper to extract theoretical information from the lattice without
a boundary and to describe agglomeration on the lattice with a boundary based on this
information.
The racetrack pattern (point e), which is interpreted as decentralization leading to
downtown decay, was observed in this study but has not been considered in central place
theory. Figure 17 shows recurrences of decentralization (points c and e) and centralization
(points f and g), i.e., downtown decay and revitalization. At point c, the population at the
center largely disappears to form inner and outer racetracks. At point e, almost the entire
population is concentrated to a racetrack. At point f, some population migrates to the
center to produce the hexagon with D = 9. A megalopolis pattern emerges at point h.
25 Thisformation of the hexagon with D = 3 is primarily the result of uniformity, but this hexagon on
the lattice with a boundary is blurred away from the center by spatial heterogeneity attributable to the
existence of the boundary (see footnote 1).

Intermediate
Mature stage Dawn stage
stage
1
g
0.8
0.6
center
0.4
0.2
h f
d b
0 a
e c
0 0.5 1 1.5 2 2.5
Transport cost parameter 2.30
= 5.000 = 3.830 = 2.815 = 3.300

point a (τ = 2.600) point b (τ = 2.298) point c (τ = 1.689) point d (τ = 1.980)
(Flat earth) (Hexagon (D = 3)) (Hexagon (D = 4))
Dawn stage
= 2.000 = 1.673 = 1.000 = 0.0164

point e (τ = 1.200) point f (τ = 1.004) point g (τ = 0.600) point h (τ = 0.098)
(Racetrack) (Hexagon (D = 9)) (Atomic (Megalopolis)
monocenter)
Intermediate stage Mature stage

FIGURE 17: Stable Equilibrium Curves and Associated Population Distributions for the
Standard Case with (σ, μ) = (5.0, 0.4).

c b
Hexagon D = 3
d
Hexagon D = 4
e
Racetrack or triangle
f
Hexagon D = 9
h g
Atomic monocenter & megalopolis
0.0 0.5 1 1.5 2 2.5
FIGURE 18: Comparison of Progress of Stable Equilibria for the Lattice with a
Boundary (Dashed Lines) and the Lattice without a Boundary (Solid Lines).
TABLE 3: Parameter Dependence of Progress of Stable Agglomerations on σ

Strong agglomeration (3 ≤ σ ≤ 4):
⎧ earth ⇒ Hexagon
Flat ⎫ (D = 3) ⇒ Hexagon (D = 4) ⇒
⎨ Hexagon (D = 12) ⎬ $
Hexagon (D = 9)
Racetrack ⇒ ⇒
⎩ ⎭ Triangle
Triangle
Atomic monocenter ⇒ Megalopolis ⇒ Flat earth
Moderate agglomeration (5 ≤ σ ≤ 8):
Flat earth ⇒ Hexagon (D = 3) ⇒ Hexagon (D = 4) ⇒
Racetrack ⇒ Hexagon (D = 9) ⇒
Atomic monocenter ⇒ Megalopolis ⇒ Flat earth
Weak agglomeration (σ = 10):
Flat earth ⇒ Racetrack ⇒ Triangle ⇒ Racetrack ⇒
Atomic monocenter ⇒ Megalopolis ⇒ Flat earth.
Parameter Dependence of Progress of Stable Equilibria

Agglomeration is known to be parameter dependent.26 Its dependence is investigated
here for two parameters: (1) the elasticity σ of substitution between any two varieties and
(2) the expenditure share μ of manufactured goods.
The influence of parameter σ on the progress of stable equilibria is investigated for
μ = 0.4 and σ = 3, 4, 5, 6, 8, and 10.27 The agglomeration characteristics are classifiable
into strong, moderate, and weak agglomerations, as listed in Table 3. In all cases, the
26 For instance, Berliant and Yu (2014) demonstrated the dependence of agglomeration on the cost of
living.
27 See Figure 17 and the preprint (Ikeda, Murota, and Takayama 2014) of this paper for numerically
obtained equilibrium curves for these parameter values.

10 4
5 2
0 0
1 5 10 0 0.5 1
(a) Dependence on σ (μ = 0.4) (b) Dependence on μ (σ = 5.0)

Notes: Here, (•) denotes the break point obtained by the numerical analysis for the lattice without a boundary
in section 5, (+) denotes that for the lattice with a boundary in subsection“Parameter Dependence of Progress of
Stable Equilibria,” the dashed curve means that for the theoretical law in (50), and the solid curve means that
for the approximate law in (52).
FIGURE 19: Dependence of Break Point τ(3)

+ on the Parameters σ and μ (θ = 1, 000).
agglomeration starts from the flat earth equilibrium and ends up with the formation of
an atomic monocenter, en route to a megalopolis pattern and the flat earth equilibrium.
More hexagonal patterns tend to emerge for a larger agglomeration force. A racetrack
pattern emerges in almost all cases and often transforms into a triangular pattern.28
The influence of parameter μ is investigated for σ = 5.0 and μ = 0.1, 0.2, 0.4, 0.5, 0.6,
and 0.8. Similarly, the agglomeration characteristics are classifiable as strong agglomer-
ation (0.5 ≤ μ ≤ 0.8), moderate agglomeration (0.2 ≤ μ ≤ 0.4), and weak agglomeration
(μ = 0.1).
Parameter Dependence of Break Points

We specifically examine the break point τ = τ(3)+ for the first bifurcation producing
the hexagon with D = 3 (Proposition 4(i)). The dependence of τ(3) + on the parameters
σ and μ expressed by the theoretical law in (50) is depicted by the dashed curve in
Figures 19(a) and 19(b), respectively. As σ increases, the economic balance shifts in favor
of dispersion and τ(3)
+ decreases. As μ increases, the economic balance shifts in favor of
agglomeration and τ(3) + increases. The solid curve of the approximate law in (52) is in
fair agreement with the dashed curve of the theoretical law. Such agreement ensures the
validity of the approximate law.
In the numerical analyses in section 5 and in this section, the hexagon with D = 3
emerge first both for the hexagonal lattices with and without a boundary. The values of
τ(3)
+ for the lattice without a boundary shown by (•) in Figure 19 show perfect agreement
with the theoretical curve. Those values shown by (+) for that with a boundary display
good agreement with the curve. Such agreement ensures the validity of the basic strategy
used in this paper to interpret and describe agglomeration characteristics of a lattice with
a boundary based on the theoretical information for a lattice without a boundary.
28 Such transformation into a triangular pattern was found for a racetrack economy in Ikeda, Aka-
matsu, and Kono (2012).

As clarified in Proposition 1, the existence of the break point is conditional on the

values of parameters. For σ = 5.0 and θ = 1, 000 in Figure 19(b), formula (49) gives a
condition μ > 0.141 for the existence of the break point. Although this condition is vio-
lated only for exceptional cases, due regard should be devoted to the existence of such
cases, in which investment in transportation is wasted without leading to economic
agglomeration.
7. CONCLUSION
Characteristics of spatial economic agglomeration are elucidated in this paper based
on the basic strategy of distinguishing spatial properties and microeconomic properties.
The former properties are model-independent and are endowed with much-desired gen-
erality, although the latter properties are model-dependent.
By the study of spatial properties of a hexagonal lattice without a boundary, ag-
glomeration patterns beyond the scope of central place theory were predicted theoreti-
cally. Possible agglomeration patterns on this lattice were found to be hexagonal patterns
à la Christaller and Lösch, racetrack patterns, long narrow patterns, and so on. In par-
ticular, racetrack patterns were advanced as a source of decentralization leading to the
decay of downtown areas. It must be emphasized that this theoretical prediction is model-
independent and applicable to economic models of various kinds.
There might be widespread pessimism that stable equilibria in two dimensions can-
not be grasped to a full extent because they are literally infinite. Nonetheless, stable
equilibria are endowed with geometrically rational forms with rich economic implications
and the variety of these forms is quite limited. To rebuff that pessimism, stable equilibria
for a specific spatial economy model were traced. Hexagonal patterns associated with cen-
tral place formation are superior in stability, thereby demonstrating the insight of central
place theory. We found several stable nonhexagonal patterns, which are beyond the scope
of this theory. The atomic monocenter and megalopolis pattern are stable for small trans-
port costs, whereas racetrack patterns representing decentralization are stable for short
durations. Other patterns are mostly unstable.
An amazing resemblance was observed for the progress of stable agglomerations on
the lattice with a boundary and that without a boundary. This resemblance shows the
validity and usefulness of the strategy used in this paper to extract theoretical informa-
tion from the lattice without a boundary and to interpret and describe agglomeration
characteristics of the lattice with a boundary based on this information.
If only hexagonal patterns in central place theory were considered, then prospects
for the progress of stable equilibria would be optimistic: a continuous increase of the size
of hexagonal patterns (Figure 10), leading to continuous growth of downtown areas. Nev-
ertheless, this is not a true scenario because competition exists between centralization
by hexagonal patterns and decentralization by racetrack patterns. The downtown area
would recurrently undergo setbacks during short periods of the dominance of racetrack
patterns. For development of downtown areas through investment in transportation, a
possible scenario implied by this study is a bumpy course undergoing several short pe-
riods of downtown decay (stable racetrack patterns). Nonetheless, one should not be too
pessimistic about such decay because it is merely transient. The downtown area is des-
tined to be revitalized en route to development of a megalopolis if continuous investment
is maintained.
We proposed formulas for the break point, at which the uniformity is broken under
declining transport costs. Results show that the smallest hexagon with D = 3 is the first
nonuniform agglomeration pattern that breaks uniformity, irrespective of the parameter
values.

The search for stable economic equilibria in two dimensions is a difficult task. Such
a search in this paper was conducted using a spatial economy model that admittedly em-
ploys bold assumptions related to microeconomy. Although some agglomeration properties
are model-dependent (Pflüger and Südekum, 2008), possible spatial patterns presented
in this paper are general and expected to exist universally in various models. Therefore,
knowledge of these patterns would be most useful for the search for stable economic equi-
libria. A future task will be to investigate stable equilibria for microeconomic models of
various kinds.
APPENDIX A: EIGENVECTORS OF THE JACOBIAN MATRIX

Eigenvectors of the Jacobian matrix J of governing equation (20) for the 6 × 6 hexag-
onal lattice are presented as a summary of Ikeda and Murota (2014) (see Remark A1 for
the notation).
The coordinate of a place on the n × n hexagonal lattice is given by
x = n1 1 + n2 2 ,
n1 , n2 = 0, 1, . . . , n − 1
√
with n = 6 for the present case, 1 = L̃(1, 0) , and 2 = L̃(−1/2, 3/2) . Consequently, the
K = n2 places are indexed by (n1 , n2 ). For a vector
(g(n1 , n2 ) | n1 , n2 = 0, 1, . . . , n − 1)
on this lattice with g(n1 , n2 ) being the (n1 , n2 )-component, the following notation g(n1 , n2 )
is used for its normalization:
⎛ ⎞1/2

n−1
n−1
(A1) g(n1 , n2 ) = (g(n1 , n2 )/ ⎝ g(i, j) ⎠ | n1 , n2 = 0, 1, . . . , n − 1).
2
i=0 j=0
The concrete forms of the eigenvectors q(k) (k)

1 , . . . , qM in (33) for k = 3, 4, 9, 12, 36(I),
29
and 36(II) are as follows:
1
(A2) q(1)
1 = (1, . . . , 1) ,
6
% &
(A3) q(3) (3)
1 , q2 = cos(2π(n1 − 2n2 )/3) , sin(2π(n1 − 2n2 )/3) ,
% &
(A4) q(4) (4) (4)
1 , q2 , q3 = cos(πn1 ) , cos(πn2 ) , cos(π(n1 − n2 )) ,
% &
q(9)
1 , . . . , q(9)
6 = cos(2π n1 /3) , sin(2π n1 /3) ,
cos(2π(−n2 )/3) , sin(2π(−n2 )/3) ,

(A5) cos(2π(−n1 + n2 )/3) , sin(2π(−n1 + n2 )/3) ,
% &
q(12)
1 , . . . , q(12)
6 = cos(π(n1 + n2 )/3) , sin(π(n1 + n2 )/3) ,
cos(π(n1 − 2n2 )/3) , sin(π(n1 − 2n2 )/3) ,

(A6) cos(π(−2n1 + n2 )/3) , sin(π(−2n1 + n2 )/3) ,
29 These
expressions of the eigenvectors were obtained by adapting the original expressions on pages
189–190 in Ikeda and Murota (2014) related to the correspondence of notations in Remark A1.

% &
q(36(I))
1 , . . . , q(36(I))
6 = cos(π n1 /3) , sin(π n1 /3) ,
cos(π(−n2 )/3) , sin(π(−n2 )/3) ,

(A7) cos(π(−n1 + n2 )/3) , sin(π(−n1 + n2 )/3) ,
% &
q(36(II))
1 , . . . , q (36(II))
12 = cos(π(2n1 + n2 )/3) , sin(π(2n1 + n2 )/3) ,
cos(π(n1 − 3n2 )/3) , sin(π(n1 − 3n2 )/3) ,

cos(π(−3n1 + 2n2 )/3) , sin(π(−3n1 + 2n2 )/3) ,
cos(π(2n1 − 3n2 )/3) , sin(π(2n1 − 3n2 )/3) ,
cos(π(n1 + 2n2 )/3) , sin(π(n1 + 2n2 )/3) ,

(A8) cos(π(−3n1 + 2n2 )/3) , sin(π(−3n1 + 2n2 )/3) .
REMARK A1. The index k used in this paper is, in fact, a short-hand notation of a
systematic notation to distinguish the “irreducible representation” associated with a critical
point. In consulting Ikeda and Murota (2014), note the correspondence:
k = 1, 3, 4, 9, 12, 36(I), 36(II)

⇐⇒ μ = (1; +, +), (2; +), (3; +, +), (6; 2, 0, +), (6; 1, 1, +), (6; 1, 0, +), (12; 2, 1)
between the notations in the present study and Ikeda and Murota (2014).
APPENDIX B: DERIVATION OF FORMULAS FOR BREAK POINTS

Formulas for break points presented in Section 4 are derived. The method in Aka-
matsu et al. (2012) for the racetrack economy is adapted to the 6 × 6 hexagonal lattice
without a boundary with the spatial discounting matrix D = (di j ) in (25).
Eigenanalysis of Spatial Discounting Matrix. Recall the spatial discounting matrix
D = (di j ) in (25) with
(B1) di j = r m(i, j)
and
(B2) r = exp[− τ(σ − 1)L̃]
in (24). The distance unit L̂ is chosen as L̂ = 1/n = 1/6.

The spatial discounting matrix D for the 6 × 6 hexagonal lattice takes the following
form:
⎡ ⎤
D0 D1 D2 D3 D4 D5
⎢ D5 D0 D1 D2 D3 D4 ⎥
⎢ ⎥
⎢ D4 D5 D0 D1 D2 D3 ⎥
(B3) D=⎢ ⎢ D3 D4 D5 D0 D1 D2 ⎥ ,
⎥
⎢ ⎥
⎣ D2 D3 D4 D5 D0 D1 ⎦
D1 D2 D3 D4 D5 D0

which is a block-circulant matrix made up of circulant matrices.

⎡ ⎤ ⎡ ⎤
1 r r2 r 3 r 2 r r r r 2 r3 r3 r2
⎢ r 1 r r2 r3 r2 ⎥ ⎢ r2 r r r2 r3 r3 ⎥
⎢ 2 ⎥ ⎢ 3 2 ⎥
⎢ r r 1 r r 2 r3 ⎥ ⎢r r r r r2 r3 ⎥
⎢
D0 = ⎢ 3 2 ⎥ D1 = D5 = ⎢
⎥,
2⎥, ⎢ r3 r3 r2 r
⎢ r r r 1 r r ⎥ ⎢ r r2 ⎥
⎥
⎣ r2 r3 r2 r 1 r ⎦ ⎣ r2 r3 r3 r2 r r⎦
2 3 2
r r r r r 1 r r2 r3 r3 r2 r
⎡ 2 2 2 3 4 3⎤ ⎡ ⎤
r r r r r r 1 1 1 1 1 1
⎢ r3 r2 r2 r2 r3 r4 ⎥ ⎢1 1 1 1 1 1⎥
⎢ 4 3 2 2 2 3⎥ ⎢ ⎥
⎢r r r r r r ⎥ ⎢ 1⎥
D2 = D4 = ⎢ ⎥ , D3 = r3 ⎢ 1 1 1 1 1 ⎥.
⎢ r3 r4 r3 r2 r2 r2 ⎥ ⎢1 1 1 1 1 1⎥
⎢ ⎥ ⎢ ⎥
⎣ r2 r3 r4 r3 r2 r2 ⎦ ⎣1 1 1 1 1 1⎦
r2 r2 r3 r4 r3 r2 1 1 1 1 1 1
The direct bifurcation from the flat earth equilibrium λ∗ = 1

K
(1, . . . , 1) (K = n2 = 62 )
in the direction of the eigenvector
(B4) η = q(k) , k = 3, 4, 9, 12, 36(I), 36(II)
of J(λ∗ ) is investigated, where q(k) is given in (35) and (36).
It is easy to verify, using (B3), that the vector η is also an eigenvector of the spatial
discounting matrix D, i.e.,
(B5) Dη = ˜ (k) η, k = 3, 4, 9, 12, 36(I), 36(II)
with the eigenvalue ˜ (k)
given by
⎧
⎪
⎪ 1 − 3r + 3r2 − 3r3 + 2r4 for k = 3,
⎪
⎪
⎪
⎪ 1 − 2r + 4r2 − 5r3 + 2r4 for k = 4,
⎨
1 − 3r2 + 3r3 − r4 for k = 9,
(B6) ˜ (k) = ˜ (k) (r) =
⎪
⎪ 1 + r − 5r2 + r3 + 2r4 for k = 12,
⎪
⎪
⎪
⎪ 1 + 4r + r2 − 5r3 − r4 for k = 36(I),
⎩
1 − 2r + r2 + r3 − r4 for k = 36(II).
Denote by d the sum of the entries of a column of D, which is given as

K
(B7) d= r m(i, j) = 1 + 6r + 12r2 + 15r3 + 2r4 .
i=1
Then, for the vector η in (B5), we have

D
(B8) η = η
d
with
˜ (k)
(B9) = .
d
Because 0 < r < 1, we have
(B10) <1
by (B6) and (B7), as shown in Figure B1. By the implicit function theorem, the relation
(B9) between and r is solvable for r as
(B11) r = Φ(k) ()

0.5
k=4
k=9
k = 12
k = 36(I)
k = 36(II)
k=3
0
0 0.5 1
r
FIGURE B1: Curves of Plotted against r (0 < r < 1).
with some function Φ(k) .

Break Point for Hexagons. From the governing equation F in (21) with H = 1, we have
∂Fi ∂Fi ∂vk

K K
∂vk ∂vi
(B12) = − δi j = −θ Pi Pk + θPi − δi j ,
∂λ j ∂vk ∂λ j ∂λ j ∂λ j
k=1 k=1
where δi j is the Kronecker delta. Equation (B12) shows that the Jacobian matrices

∂F ∂Fi ∂v ∂vi
J(λ) = = , V (λ) = =
∂λ ∂λ j ∂λ ∂λ j
are related as
⎤ ⎡ ⎡ ⎤
P1 P1
⎢ ⎥ ⎢ ⎥
(B13) J(λ) = −θ⎣ ... ⎦ P1 · · · PK V (λ) + θ⎣ ..
. ⎦V (λ) − I,
PK PK
where I is the identity matrix.
In regard to V (λ) we recall (18):
μ
(B14) vi = ln Δi + ln wi
σ−1
as well as (15):
μ dik
(B15) wi = (wk λk + 1),
σ Δk
k

where

K
Δk = Δk (λ, τ) = d jk λ j .
j=1
The differentiations of (B14) and (B15) with respect to λ j yield, respectively,

∂vi dji 1 ∂wi
(B16) = κ + ,
∂λ j Δi wi ∂λ j
K " #
∂wi dik ∂wk
(B17) =κ λk + wk δ k j Δ k − (wk λk + 1)d jk ,
∂λ j Δ 2 ∂λ j
k=1 k
where
μ μ
(B18) κ= , κ = .
σ σ−1
We have 0 < κ < 1 and κ > 0 because σ > 1, 0 < μ < 1.
At the flat earth equilibrium λ∗ = K1 (1, . . . , 1) , (B13) yields
θ θ
(B19) J(λ∗ ) = − 2
11V (λ∗ ) + V (λ∗ ) − I,
K K
where 1 = (1, . . . , 1) . The matrix V (λ∗ ) in (B19) can be evaluated as shown below. At
λ = λ∗ , we have

K
d
Δ j = Δ j (λ∗ , τ) = dk j λk = .
K
k=1
Because w j is independent of j, we may put w j = w; then (B15) becomes

K
K w
w=κ di j + 1 = κ (w + K ) ,
d K
j=1
which yields
κK
(B20) w= .
1−κ
At λ = λ∗ , (B17) becomes
" #
K2 d w
K
∂wi 1 ∂wk
=κ dik + wδk j − + 1 d jk ,
∂λ j d2 K ∂λ j K K
k=1
which in matrix form reads as

" #
K2 d 1 w+K
W =κ 2 D W + wI − D
d K K K
with W = (∂wi /∂λ j ). With the use of (B20), this equation can be rewritten as

D D D
I−κ W = Kw κI − ,
d d d
which can be further rewritten as
−1
D D D
W = Kw I − κ · κI − .
d d d

Then the partial derivatives in (B16) can be evaluated in matrix form as

!
∗ D D −1 D D
(B21) V (λ ) = K κ + I − κ · κI − .
d d d d
Then (B21) with (B8) shows that
(B22) V (λ∗ ) · η = γη
with
(B23) γ = K[κ + (1 − κ)−1 · (κ − )].
Multiplying (B19) by the vector η from the right and using
1V (λ∗ ) · η = γ1 η = 0,
we obtain

(κ − ) 1
(B24) J(λ∗ ) · η = θ κ + − η.
1 − κ θ
Then the eigenvalue β of the Jacobian matrix J(λ∗ ) for the eigenvector η is expressed in
terms of as
(B25) β = Ψ()
with a function Ψ defined as

x(κ − x) 1
(B26) Ψ(x) = θ κ x + − .
1 − κx θ
Break point τbreak is determined from the condition that the eigenvalue β for τ = τbreak
vanishes. Recall the dependence of the variables:
(B.25) (B.9) (B.2)
β ← ← r ← τ.
The value ∗ satisfying Ψ( ∗ ) = 0 is a solution x = ∗ of the quadratic equation
(B27) θ(bx − ax2 ) − 1 = 0,
where
(B28) a = κκ + 1 > 0, b = κ + κ + θ−1 κ > 0,
which are constants. Of the two solutions of (B27), the larger

√
∗ b + b2 − 4aθ−1
(B29) + =
2a
is related to the first bifurcation when τ is reduced from a large value. The smaller
√
b − b2 − 4aθ−1
(B30) −∗ =
2a
is related to the last bifurcation. Therefore, 0 < −∗ < +∗ .
By < 1 in (B10), we have ±∗ < 1, which gives the condition (48). Another (discrimi-
nant) condition b2 − 4aθ−1 > 0 with (B18) and (B28) gives (49), thereby proving Proposi-
tion 1.

(k) ∗
The value of r = r(k) (k) (k)
± corresponding to = ± is given as r± = Φ (± ) by (B11). Then,
(k)
from r = exp[− τ(σ − 1)L̃] in (B2) with L̃ = 1/n = 1/6, τ± is given as
6 6
(B31) τ(k)
± =− ± =−
ln r(k) ln(Φ(k) (±∗ )).
σ−1 σ−1
This proves (50).
Approximate Formula for Hexagon with D = 3. We search for an approximate formula
of τ(3)
+ for the hexagon with D = 3 under the conditions
(B32) θ (σ/μ)2 1,
which yield
2μ b 2μ
(B33) a ≈ 1, b ≈ κ + κ ≈ , +∗ ≈ ≈ 1.
σ−1 a σ−1
Because +∗ > 0 and the numerator ˜ (3) of (B9) is equal to (1 − r)(1 − 2r)(1 + r2 ), we
have
1
(B34) 0 < r(3)
+ < .
2
By (B34), it is possible to introduce a fairly accurate assumption
+ ) 1.
(r(3) 3
(B35)
From (B9) for k = 3, we have
(1 − 3r + 3r2 − r3 ) − 2r3 + 2r4 (1 − r)3 − 2r3 (1 − r)
= = .
(1 + 6r + 12r2 + 8r3 ) + 7r3 + 2r4 (1 + 2r)3 + r3 (7 + 2r)
Then for r = r(3)
+ satisfying (B35), we have
' (3
1 − r(3)
+
+∗ ≈ ,
1 + 2r(3)
+
which yields
(3) ∗ 1 − (+∗ )1/3

+ = Φ (+ ) ≈
r(3) .
1 + 2(+∗ )1/3
Then from (B31) with (B33), we obtain
' (
n 1 − (+∗ )1/3
τ(3)
+ ≈− ln
L̂(σ − 1) 1 + 2(+∗ )1/3
μ1/3
(B36) ≈ 18 · 21/3 ,
(σ − 1)4/3
which proves Proposition 3. This formula (B36) is fairly accurate as shown in Table B1,
which lists the relative error of τ(3)
+ :
Error = |[(Approximate value) − (Exact value)]/(Exact value)| × 100 (%).

Order of Emerging Hexagons. When r (or τ) is changed continuously, the first and the
last bifurcations are the most important bifurcations. It is possible to predetermine the
hexagonal distributions associated with these bifurcations as expounded below.

TABLE B1: Relative Error of the Approximation (B36) for τ(3)

+
r(3)
+ 0.0 0.1 0.2 0.3 0.4 0.5
True value of τ(3)
+ (σ = 5.0) 0.00 0.687 0.726 0.704 0.664 0.620
Approximate value of τ(3)
+ (σ = 5.0) 0.00 0.692 0.748 0.749 0.733 0.710
Error (%) 0.0 0.75 3.01 6.40 10.4 14.4
Under the condition (B10), the flat earth equilibrium is stable for a large τ (= +∞)
because τ = +∞ entails = 1 via (B2) and (B9) and then the eigenvalue β in (B25) with
(B26) becomes negative under condition (48).
The functions (r) = ˜ (k) (r)/d for k = 3, 4, 9, 12, 36(I), 36(II) in the range 0 < r < 1 are
shown in Figure B1. Then, for a = ±∗ , the associated r = r(k) (k) ∗
± = Φ (± ) of (B11) satisfies
inequalities
+ < r+ < r+
r(3) , − < r− < r− , k = 4, 9, 12, 36(II).
(k) (36(I))
r(3) (k) (36(I))
Then from (B2), for the associated transport cost parameter τ(k)
± , we have
+ > τ+ > τ+
τ(3) (k) (36(I))
, τ(3)
− > τ− > τ−
(k) (36(I))
, k = 4, 9, 12, 36(II).
Therefore, when τ is reduced from a large value, the first bifurcation is associated with
τ(3) (3)
+ ( > τ− ) for D = 3 and the last one with τ−
(36(I))
( < τ(36(I))
+ ) for D = 36. This proves
Proposition 4.
REFERENCES
Akamatsu, Takashi and Yuki Takayama. 2009. “A Simplified Approach to Analyzing Multi-Regional Core–
Periphery Models,” MPRA Paper 21739, University Library of Munich. Munich: Ludwig Maximilian University
of Munich.
Akamatsu, Takashi, Yuki Takayama, and Kiyohiro Ikeda. 2012. “Spatial Discounting, Fourier, and Racetrack
Economy: A Recipe for the Analysis of Spatial Agglomeration Models,” Journal of Economic Dynamics and
Control, 36(11), 1729–1759.
Anas, Alex. 2004. “Vanishing Cities: What Does the New Economic Geography Imply about the Efficiency of
Urbanization?” Journal of Economic Geography, 4(2), 181–199.
Anderson, Simon P., André de Palma, and Jacques-Franco̧is Thisse. 1992. Discrete Choice Theory of Product
Differentiation. Cambridge: MIT Press.
Baldwin, Richard E., Rikard Forslid, Philippe Martin, Gianmarco I. P. Ottaviano, and Frederic Robert-Nicoud.
2003. Economic Geography and Public Policy. Princeton: Princeton University Press.
Beckmann, Martin J. 1976. “Spatial Equilibrium and the Dispersed City,” in Y. Y. Papageorgiou (ed.), Mathemat-
ical Land Use Theory. Lexington: Lexington Books, pp. 117–125.
Berliant, Marcus and Chia-Ming Yu. 2014. “Locational Signaling and Agglomeration,” MPRA Paper, 55410,
University Library of Munich. Munich: Ludwig Maximilian University of Munich.
Brakman, Steven, Harry Garretsen, Charles van Marrewijk, and Marianne van den Berg. 1999. “The Return
of Zipf: Towards a Further Understanding of the Rank-Size Distribution,” Journal of Regional Science, 39(1),
183–213.
Christaller, Walter. 1933. Die zentralen Orte in Süddeutschland. Gustav Fischer, Jena. English translation: 1966.
Central Places in Southern Germany. Englewood Cliffs: Prentice Hall.
Clarke, Martin and Alan G. Wilson. 1983. “The Dynamics of Urban Spatial Structure: Progress and Problems,”
Journal of Regional Science, 23(1), 1–18.
———. 1985. “The Dynamics of Urban Spatial Structure: The Progress of a Research Programme,” Transactions
of the Institute of British Geographers, New Series 10(4), 427–451.
Eaton, B. Curtis and Richard G. Lipsey. 1975. “The Principle of Minimum Differentiation Reconsidered: Some
New Developments in the Theory of Spatial Competition,” Review of Economic Studies, 42(1), 27–49.
———. 1982. “An Economic Theory of Central Places,” Economic Journal, 92(365), 56–72.
Flam, Harry and Elhanan Helpman. 1987. “Industrial Policy under Monopolistic Competition,” Journal of Inter-
national Economics, 22, 79–102.

Forslid, Rikard and Gianmarco I. P. Ottaviano. 2003. “An Analytically Solvable Core-Periphery model,” Journal
of Economic Geography, 3, 229–240.
Fudenberg, Drew and David K. Levine. 1998. The Theory of Learning in Games. Cambridge: MIT Press.
Fujita, Masahisa, Paul Krugman, and Tomoya Mori. 1999. “On the Evolution of Hierarchical Urban Systems,”
European Economic Review, 43(2), 209–251.
Fujita, Masahisa, Paul Krugman, and Anthony J. Venables. 1999. The Spatial Economy: Cities, Regions, and
International Trade. Cambridge: MIT Press.
Fujita, Masahisa and Tomoya Mori, 1997. “Structural Stability and the Evolution of Urban Systems,” Regional
Science and Urban Economics, 42, 399–442.
Fujita, Masahisa and Hideaki Ogawa. 1982. “Multiple Equilibria and Structural Transition of Non-Monocentric
Urban Configurations,” Regional Science and Urban Economics, 12(2), 161–196.
Golubitsky, Martin, Ian Stewart, and David G. Schaeffer. 1988. Singularities and Groups in Bifurcation Theory,
Volume 2. New York: Springer-Verlag.
Ikeda, Kiyohiro, Takashi Akamatsu, and Tatsuhito Kono. 2012. “Spatial Period-Doubling Agglomeration of
a Core–Periphery Model with a System of Cities,” Journal of Economic Dynamics and Control, 36(5),
754–778.
Ikeda, Kiyohiro and Kazuo Murota. 2014. Bifurcation Theory for Hexagonal Agglomeration in Economic Geogra-
phy. Tokyo: Springer-Verlag.
Ikeda, Kiyohiro, Kazuo Murota, and Takashi Akamatsu. 2012. “Self-Organization of Lösch’s Hexagons in Eco-
nomic Agglomeration for Core–Periphery Models,” International Journal of Bifurcation and Chaos, 22(8),
1230026-1–1230026-29.
Ikeda, Kiyohiro, Kazuo Murota, Takashi Akamatsu, Tatsuhito Kono, and Yuki Takayama. 2014. “Self-
Organization of Hexagonal Agglomeration Patterns in New Economic Geography Models,” Journal of Economic
Behavior & Organization, 99, 32–52.
Ikeda, Kiyohiro, Kazuo Murota, and Yuki Takayama. 2014. “Stable Economic Agglomeration Patterns in Two
Dimensions: Beyond the Scope of Central Place Theory,” METR 2014-24, University of Tokyo, Tokyo.
Krugman, Paul. 1991. “Increasing Returns and Economic Geography,” Journal of Political Economy, 99,
483–499.
———. 1993. “On the Number and Location of Cities,” European Economic Review, 37, 293–298.
———. 1996. The Self-Organizing Economy. Malden, MA: Wiley-Blackwell.
Lösch, August. 1940. Die räumliche Ordnung der Wirtschaft. Gustav Fischer, Jena. English translation: 1954.
The Economics of Location. New Haven: Yale University Press.
Mori, Tomoya. 1997. “A Modeling of Megalopolis Formation: The Maturing of City Systems,” Journal of Urban
Economics, 42, 133–157.
Mossay, Pascal and Pierre M. Picard. 2011. “On Spatial Equilibria in a Social Interaction Model,” Journal of
Economic Theory, 146(6), 2455–2477.
Munz, Martin and Wolfgang Weidlich. 1990. “Settlement Formation, Part II: Numerical Simulation,” Annals of
Regional Science, 24, 177–196.
Murata, Yasusada. 2003. “Product Diversity, Taste Heterogeneity, and Geographic Distribution of Economic
Activities: Market vs. Non-Market Interactions,” Journal of Urban Economics, 53(1), 126–144.
Ottaviano, Gianmarco I. P., Takatoshi Tabuchi, and Jacques-Franco̧is Thisse. 2002. “Agglomeration and Trade
Revisited,” International Economic Review, 43(2), 409–435.
Oyama, Daisuke. 2009. “Agglomeration under Forward-Looking Expectations: Potentials and Global Stability,”
Regional Science and Urban Economics, 39(6), 696–713.
Pflüger, Michael and Jens Südekum. 2008. “A Synthesis of Footloose-Entrepreneur New Economic Geogra-
phy Models: When is Agglomeration Smooth and Easily Reversible?” Journal of Economic Geography, 8,
39–54.
Picard, Pierre M. and Takatoshi Tabuchi. 2010. “Self-Organized Agglomerations and Transport Costs,” Economic
Theory, 42, 565–589.
Robertson, Kent A. 1995. “Downtown Redevelopment Strategies in the United States: An End-of-the-Century
Assessment,” Journal of the American Planning Association, 61(4), 429–437.
Stelder, Dirk. 2005. “Where do Cities Form? A Geographical Agglomeration Model for Europe,” Journal of Regional
Science, 45, 657–679.
Tabuchi, Takatoshi. 1986. “Urban Agglomeration Economies in a Linear City,” Regional Science and Urban
Economics, 16(3), 421–436.
Tabuchi, Takatoshi and Jacques-Franco̧is Thisse. 2002. “Taste Heterogeneity, Labor Mobility and Economic Ge-
ography,” Journal of Development Economics 69, 155–177.
———. 2011. “A New Economic Geography Model of Central Places,” Journal of Urban Economics, 69, 240–252.
Thisse, Jacques-Franco̧is. 2010. “Toward a Unified Theory of Economic Geography and Urban Economics,” Journal
of Regional Science, 50(1), 281–296.
Weidlich, Wolfgang and Günter Haag. 1987. “A Dynamic Phase Transition Model for Spatial Agglomeration
Processes,” Journal of Regional Science, 27(4), 529–569.

Copyright of Journal of Regional Science is the property of Wiley-Blackwell and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for
individual use.

Stable Economic Agglomeration Patterns in Two Dimensions: Beyond The Scope of Central Place Theory

Uploaded by

Document Information

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

Stable Economic Agglomeration Patterns in Two Dimensions: Beyond The Scope of Central Place Theory

Uploaded by

Copyright:

Available Formats

JOURNAL OF REGIONAL SCIENCE, VOL. 57, NO. 1, 2017, pp.

STABLE ECONOMIC AGGLOMERATION PATTERNS IN TWO

Long narrow Racetrack Lattice

FIGURE 1: Spatial Platforms for Economic Activities.

for being left as an obscure theory.”

(a) Hexagon (b) Monocenter (c) Megalopolis (d) Racetrack

FIGURE 2: Agglomeration Patterns of Economic Interest.

Forslid and Ottaviano (2003), which is a version of modeling by Krugman (1991).8 By

2. MODELING OF THE SPATIAL ECONOMY

et al. (2012) to a hexagonal lattice.

where μ is a constant parameter expressing the expenditure share of manufacturing

It is assumed that Ti j = Ti j (τ) is a function in a transport cost parameter τ.

The third condition is expressed as

by (10) and (12). Here

Spatial Equilibrium Conditions

(a) 3 × 3 Hexagonal lattice (b) Spatially repeated

3. BIFURCATING HEXAGONAL AND NONHEXAGONAL PATTERNS

A Hexagonal Lattice without a Boundary

(a) D = 1 (flat earth) (b) D = 3 (c) D = 4 (d) D = 7

FIGURE 4: Lösch’s Hexagons on a Hexagonal Lattice.

The transport cost Ti j between places i and j is defined as an iceberg form14

Group-Theoretic Bifurcation Analysis Procedure

Stewart, and Schaeffer, 1988; Ikeda and Murota, 2014).

FIGURE 5: Spatial Patterns Expressed by Vectors q(3) (3) (3)

for some function F̃ in ξ = (ξ1 , . . . , ξM ) ∈ RM in (30) and τ̃ ∈ R in (31). In this reduction

Lösch’s Hexagon with D = 3: Simple Example

Classification of Bifurcating Equilibria

(33) q(k) (k)

(a) Hexagon with D = 4 (q (4) ) (b) Hexagon with D = 7 (q (7) )

(c) Hexagon with D = 9 (q (9) ) (d) Hexagon with D = 12 (q(12) )

(e) Megalopolis with D = 36 (q(36(I)) ) (f) Megalopolis with D = 36 (q(36(II)) )

FIGURE 6: Hexagonal and Megalopolis Patterns on a Hexagonal Window Expressed by

4. BREAK POINT TRIGGERING SPATIAL AGGLOMERATION

(a) −q (4) (racetrack) (b) −q (7) (racetrack)

(c) −q (9) (racetrack) (d) −q (12) (racetrack)

(e) −q (36(I)) (racetrack) (f) −q (36(II)) (racetrack)

FIGURE 7: Racetrack Patterns on Hexagonal Windows Expressed by Eigenvectors

FIGURE 8: Long Narrow Patterns on the Hexagonal Lattice Expressed by Eigenvectors

FIGURE 9: Deformed Hexagonal Pattern on the Hexagonal Window Expressed by an

where 1 = (1, . . . , 1) . As worked out in (B14)–(B21) in Appendix B, we have

where D = (di j ) is the spatial discounting matrix given by (25),

(41) Dη = ˜ (k) η, k = 3, 4, 9, 12, 36(I), 36(II)

with the eigenvalue ˜ (k) given (see (B6)) by

From (38) with (39) and (41), we obtain (see (B22)–(B24))

(45) r = Φ(k) ()

for some function Φ(k) .

PROPOSITION 1. For each Lösch’s hexagon and associated nonhexagonal distribution(s),

Although formula (50) is rigorous, the following approximate formula is convenient

Order of Emerging Hexagons

(ii) Bifurcation producing the megalopolis occurs last at τ = τ(36(I))

5. STABLE AGGLOMERATION PATTERNS: A HEXAGONAL LATTICE WITHOUT A

Bifurcating Stable Equilibria

FIGURE 10: Equilibrium Curves Related to Hexagonal Patterns and Associated

sis (Pflüger and Südekum, 2008).

(g) Racetrack (− q(12) ) (h) Racetrack or triangle

(i) Two-places (j) Semisquare (k) Long narrow

Progress of Stable Equilibria

1.0 M=2 M=6

FIGURE 12: Equilibrium Curves Related to Racetrack Patterns and Associated

stable equilibria with decreasing τ would engender continuous progress of centralization

Mature stage Intermediate stage Dawn stage

Mature stage: Atomic monocenter ⇒ Megalopolis ⇒ Flat earth

6. STABLE AGGLOMERATION PATTERNS: A HEXAGONAL LATTICE WITH

where 1 = (1, . . . , 1) . As worked out in (B14)–(B21) in Appendix B, we have

∂Fi ∂Fi ∂vk

(B23) γ = K[κ + (1 − κ)−1 · (κ − )].

1V (λ∗ ) · η = γ1 η = 0,

(B28) a = κκ + 1 > 0, b = κ + κ + θ−1 κ > 0,