A Review of Mathematical Models in Economic Environmental Problems

Annals of Operations Research 97 (2000) 165–201 165
A review of mathematical models in economic

environmental problems
Zbigniew Nahorski a and Hans F. Ravn b
a
Systems Research Institute, Polish Academy of Sciences, Newelska 6, 01-447 Warsaw, Poland
E-mail: nahorski@ibspan.waw.pl
b
Risø National Laboratory, Roskilde, Denmark
The paper presents a review of mathematical models used in economic analysis of en-
vironmental problems. This area of research combines macroeconomic models of growth,
as dependent on capital, labour, resources, etc., with environmental models describing such
phenomena like natural resources exhaustion or pollution accumulation and degradation. In
simpler cases the models can be treated analytically and the utility function can be optimized
using, e.g., such tools as the maximum principle. In more complicated cases calculation of
the optimal environmental policies requires a computer solution.
Keywords: dynamic models, capital, labour, resources, pollution, utility function, optimal
solution
1. Introduction
Usually there is a tradeoff between economic development and preservation of

the environment. The difficulty in economic tackling of the problem arises because of
difficulty in estimating the values of wildlands and its benefits to the human life, and
moreover in prediction of the future values and benefits, like, e.g., from preserving
natural biota or botanic specimen. This is also connected with irreversibility of many
development projects when their influence on the environment is considered. It was
probably Mill [61] who first emphasized the importance of the environment for the
quality of life. More elaborated views on the economics of the problem, influencing
further studies, were, however, formulated only in the 1960s, with the works of Kru-
tilla [45] and Boulding [8]. These issues have further been intensively discussed in
the economic literature and an extensive list of literature on this subjects can be found
in survey papers, like [16,26].
This paper is concerned with the use of mathematical models in the study of
economic environmental issues, and specifically, the dynamic models. There is quite a
tradition that the dynamic economic problems involving the dynamic models are coped
with by using the optimal control theory. This idea can be traced to early applications
of variational methods in [24,38,72], but mainly arose in the second half of the 1960s
where newly developed optimal control theory tools like the maximum principle and
dynamic programming went into use. In the 1970s many books presenting this ap-
 J.C. Baltzer AG, Science Publishers

166 Z. Nahorski, H.F. Ravn / Mathematical models in economic environmental problems
proach appeared, and particularly the very influential book by Arrow and Kurz [2]. See
also the historical remarks in [25]. For the discussion of static models and methods
of their analysis see, e.g., [6,83].
There is quite a handful of books dealing with the economics of environment,
like [6,12,14,18,20,25,41,83,97]. But the area discussed in this paper includes also
many vast subjects with independently developing literature. Some are cited in sec-
tion 2 where these subjects are shortly presented.
The aim of this paper is twofold. One is to review the area and to introduce the
reader, with models, methods and problems spotted there. This is mainly expected to
be achieved in section 2. The second is connected with a deeper mathematical analysis
of the models and methods. This subject is included in sections 3 and 4, where the one
and two state variable models are analyzed. In section 5 we discuss problems that are
more complicated in the sense that their analysis can only be provided by numerical
computation.
2. Models
When we talk on preservation of natural environments, we usually have to take

into account at least (i) the pollution connected with the human activity, and specifically
with production, (ii) the exploitation of natural resources, which may be connected
with pollution, but also with deteriorating its natural beauty and scenic wilderness, and
(iii) that part of the capital, which is connected with the above activities. Quality of the
environment is also connected with (iv) population growth which leads to congestion.
In this section we concentrate on reviewing some basic models describing these four
activities and notions connected with them. For an older review the reader is referred
to [85].
To allow for better comparison, all values below are expressed in monetary
equivalents.
2.1. A prototype basic model
We start with the following prototype optimization model
maximize W(C, Q, P ) (1)
constrained by
Ṗ (t) = J(K, L, Q, P , A), (2)

Ṙ(t) = G(R, K, L, Q), (3)
K̇(t) = W (K, C, A, Q, L, I), (4)
L̇(t) = S(L, C, P ), (5)
P (0) = P0 , R(0) = R0 , K(0) = K0 , L(0) = L0 , (6)
Z. Nahorski, H.F. Ravn / Mathematical models in economic environmental problems 167
where W is a welfare function, C is consumption, P is a pollution stock, R represents

remaining reserves of a resource and Q its extraction, K is the capital, I is the
investment, and L is the labour (as a constant part of population), all in time t. A are
the expenses connected with the abatement of pollution. J, G, S, W are some,
unspecified yet, functions. The triple (C, Q, P ) will be called an environmental policy.
Let us assume that
1
J(K, L, Q, P , A) = j(k, q, p, a), (7)
L
1
G(R, K, L) = g(r, k), (8)
L
1
W (K, C, A, Q, L) = w(k, c, a, q), (9)
L
1
S(L, C, P ) = s(c, p). (10)
L
Then the above model can be transformed to per capita values
P (t) R(t) K(t)

p(t) = , r(t) = , k(t) = , (11)
L(t) L(t) L(t)
C(t) I(t) A(t)
c(t) = , i(t) = , a(t) = . (12)
L(t) L(t) L(t)
As generally, for x(t) = X(t)/L(t)
Ẋ X L̇
ẋ = − (13)
L LL
then the equations (2)–(4), (6) take the following form
ṗ(t) = −s(c, p)p(t) + j(k, c, q, a), (14)

ṙ(t) = −s(c, p)r(t) + g(r, k), (15)
k̇(t) = −s(c, p)k(t) + w(k, c, a, q), (16)
p(0) = p0 , r(0) = r0 , k(0) = k0 . (17)
This transformation is used if growth of labour has to be considered, and is particularly

useful when s(c, p) = s is a constant (the exponential growth of labour). For the
constant labour rather the equations (2), (3) and (4) are used.
Let us notice that the assumptions (7)–(10) can be satisfied for some classes of
functions. Take for example functions that are linear in their arguments. Other classes
will be considered in the sequel.
In the next subsections we shall discuss the elements of the above model in more
details.
2.2. Criterion
The precise specification of the welfare function (1) obviously has decisive im-
plications for the characteristics of the solution structure of the model (1)–(6). In the
sequel we discuss its two aspects: structure of the welfare function (1) and methods
of defining the utility function.
2.2.1. Criterion function structure

The most common welfare function (1) is the integrated discounted utility crite-
rion
Z ∞
W(C, Q, P ) = e−δt U (C, Q, P ) dt, (18)
0
where U is a so-called utility function. The discount rate δ specifies how much weight
we give to the utility function in the future, as compared to the present value. Small δ
(e−δt decreasing slowly) is connected with a higher evaluation of the utility function
in the future. Big δ (e−δt decreasing quickly) is connected with neglect of the future
values of the utility function (the myopic point of view). The above criterion will be
frequently referred to in the sequel.
Ramsey’s welfare criterion [72] ranks the policies (C, Q, P ) taking into account
their proximity to the “bliss function”. Let us assume that the utility function is
normalized so that the normalized function U e satisfies 0 6 Ue 6 1. Then the Ramsey
welfare criterion has the form
Z ∞

W(C, Q, P ) = − 1−U e (C, Q, P ) dt, (19)
0
where 1 is the bliss function. A disadvantage of this criterion is connected with im-
possibility of comparison of all policies, as some of them may cause non-convergence
of the integral.
The overtaking criterion [95] ranks one policy (C1 , Q1 , P1 ) over another
(C2 , Q2 , P2 ) if after a sufficiently long time the integral of U (C1 , Q1 , P1 ) is greater
than the integral of U (C2 , Q2 , P2 ), i.e.,
Z T Z T
∃TK , ∀T > TK , U (C1 , Q1 , P1 ) dt > U (C2 , Q2 , P2 ) dt. (20)
0 0
Neither does this criterion allow for ordering of all policies, as there may not exist a
time TK satisfying the required condition. A simple example is when the domination
of one integral over another changes periodically.
In the long run average criterion the utility functions are compared in an average
term. Thus, a policy (C1 , Q1 , P1 ) is better that (C2 , Q2 , P2 ) if
∃TK , ∆K , ∀T > TK , ∆ > ∆K ,
Z Z
1 T +∆ 1 T +∆
U (C1 , Q1 , P1 ) dt > U (C2 , Q2 , P2 ) dt. (21)
∆ T ∆ T
The Rawlsian criterion [73] compares the infima of utility function. Thus
W(C, Q, P ) = inf U (C, Q, P ). (22)
t
Pointing to a drawback of this criterion let us assume that U (C1 , Q1 , P1 ) >

U (C2 , Q2 , P2 ) and that there exists a nonzero interval where U (C1 , Q1 , P1 ) >
U (C2 , Q2 , P2 ) but inf t U (C1 , Q1 , P1 ) = inf t U (C2 , Q2 , P2 ). Then both policies
(C1 , Q1 , P1 ) and (C2 , Q2 , P2 ) are ranked equal by the Rawlsian criterion while it seems
reasonable to rank the former over the latter.
The satisfaction of basic needs criterion [10] seeks the policy which attains a
given value of the utility function Ub (representing the basic needs) in the shorter
time, i.e.,

W(C, Q, P ) = − min t: U C(τ ), Q(τ ), P (τ ) > Ub , ∀τ > t . (23)
This criterion has the same drawback as the previous one. It can not distinguish two
policies with equal minimal time of attaining the basic needs Ub even when one of
them has otherwise higher values of the utility functions.
One of a greater concern in the recent international environment policy determi-
nation community is “sustainable development”. In 1987 the Brundtland Commission
defined it as developments which “satisfy the needs of the present without compro-
mising the needs of the future”. Starting with this a mathematical formulation of this
concept is proposed in [11] where notions of “a dictatorial role of the present” and “a
dictatorial role of the future” are introduced. The welfare function W gives a dicta-
torial role for the present if it is sensitive only to an initial part of the environmental
policy. And it gives a dictatorial role for the future if it is sensitive only to the asymp-
totic part of the environmental policy. Then a welfare function gives the sustainable
preferences if it gives neither a dictatorial role for the present nor for the future.
None of the earlier well defined criteria satisfies this requirement. The integrated
discounted utility criterion gives a dictatorial role for the present for any δ > 0. The
long run average criterion gives a dictatorial role for the future. Other are rejected
from the reasons mentioned before.
In [11] the discrete time is considered and the author proves that the sustainable
preference welfare function has to have the form which in the continuous time may
be written as
Z ∞
W(C, Q, P ) = e−δt U (C, Q, P ) dt + Φ(C, Q, P ), (24)
0
where
Φ(C, Q, P ) = lim U (C, P , T ) extended to all (C, Q, P ).
t→∞
The function Φ is extended from the set of those policies for which the limit exist to
the set of all policies by Hahn–Banach theorem. Precise mathematical treatment of
this subject can be found in [11].
2.2.2. Defining the utility function

The shape of the utility function is important in the theoretical analysis of the
economic problems in environment. The important characteristics for the analysis are
the second (partial) derivatives UXX of the utility function with respect to any of its
argument X. In the stochastic setting the following notions are often used:
• risk aversion, if UXX < 0 (U concave in X),
• risk preference, if UXX > 0 (U convex in X),
• risk neutrality, if UXX = 0 (U linear in X).
Another important characteristic may be separability of U , e.g., when U (C, P ) =
U1 (C) + U2 (P ).
Detailed definition of the utility function is basically related to the possibility of
measuring the benefits and costs of environmental policy. While it may be compara-
tively easier to do for many resource problems involving goods which can be bought
and sold on markets, and which thus can be easily expressed in monetary terms, this
is hardly the case in pollution problems. The direct estimation is sometimes possible
when the cost of abatement is considered (like expenditures on cleaner fuels, abatement
control equipment, etc.) or after damages caused by pollution have occurred (usually
in forestry, fishing, and agriculture). But even then the possibility of adjustment of
firms or individuals to the changes caused by pollution has to be considered. Such ad-
justment can, for example, include irrigation of the land or alterations in the cultivated
plants or number of acres planted, as well as passing on part of the cost increase to
consumers.
The direct estimation is, however, difficult for such goods as clean air or water.
To find values of this kind of goods some indirect methods of measurements have been
developed. The detailed discussion of these methods is out of the scope of this paper.
It can be found, for instance, in a recent survey [16], from where also the classification
given below has been taken.
Let us introduce the damage function S(P , Z) that links pollution P and another
factor Z with some, perhaps abstract, value S. S may be, for example, time spent on
recreation, swimming or boating on a lake, or the number of average days in a year
with some respiratory problems due to air pollution. Z may be connected with some
actions undertaken to mitigate the effect of pollution, like willingness to travel to a lake
being more far than a close, but polluted one. Or moving to another part of the city,
or another city. Or just buying some medicine to relieve the respiratory symptoms due
to air pollution. Now, with a known change in pollution it is possible to measure the
corresponding changes of the factor Z. Thus, if Z is expressed in monetary equivalent,
it allows us to calculate willingness of people to pay for improvement in environmental
standard. This approach was called the averting behaviour approach.
A closely related way, called the weak complementarity approach, is connected
with valuating factors which are complementary to environment quality, like for ex-
ample more visits to a recreational site when the source of pollution was removed and
the environment cleaned, or increase in the household prices there.
A third approach uses hedonic market methods. This approach exploits the con-
cept of hedonic prices. This is a notion that the price of a good can be decomposed
into the prices of the attributes that make up the good. For example, the price of a
house may include air quality, or wages for a job may include risk of death. Then
regressing the prices on the corresponding attributes the value of the clean environment
can be estimated.
Sometimes none of the indirect methods can be applied, for example because
there are no averting or mitigating behaviour or there is no variability in environmental
quality in a region. An example of another situation when none of the indirect methods
can be used may be preservation of an endangered species. In such cases direct
questioning can be used. This approach is called the contingent valuation method.
Although its idea is simple, some skepticism remains connected with the hypothetical
rather than actual valuation in responses to questions. Criticisms include strategic
answers of the questioned persons and their not sufficient familiarity with the respective
commodities.
Although sometimes imprecise, this valuation techniques have been applied in
many practical problems, see [16], and may help in developing the utility function for
a specific case.
2.3. Capital
The following mathematical model of the capital accumulation is due to Ram-

sey [72]. It has the form
K̇(t) = −αK(t) + I(t), (25)
where α is the rate of the capital depreciation. As before, K is the capital and I is the
investment. Let us assume that the output of the economy is given by a production
function
Y = eθt F (K, L, R), (26)
where θ is a rate of technological progress. The technological progress can be also
modelled by multiplying the labour by a factor X(t) growing in time, i.e., using the
equation
Y = F (K, XL, R).
The production is divided as follows
Y (t) = C(t) + I(t) + A(t). (27)
Inserting I(t) from above in (25) we get
K̇(t) = −αK(t) + eθt F (K, L, R) − C(t) − A(t). (28)

An often used form of the production function is the CES (constant elasticity of
substitution) production function
h nh σ1 −1 σ1 −1 i
σ1 o σ2 −1
σ2 −1 i
σ2
σ1 −1 σ2 σ2 −1
F (K, L, R) = β23 β12 K σ1
+ (1 − β12 )L σ1
+ (1 − β23 )R σ2
,
(29)
where σ1 and σ2 are elasticities of substitution of different factors. The expression
shown in (29) exemplifies a nested two level function. For σ1 = σ2 = σ this function
reduces to
σ/(σ−1)
F (K, L, R) = β1 K (σ−1)/σ + β2 L(σ−1)/σ + β3 R(σ−1)/σ , (30)
where βi > 0, i = 1, 2, 3, β1 +β2 +β3 = 1 and 0 < σ < ∞, σ 6= 1. The function (30)
is sometimes called a normal multifactor CES function.
A special case of (30), for σ → 1, is of importance. We have
(σ−1)/σ +β (σ−1)/σ +β R(σ−1)/σ ]
F (K, L, R) = eσ/(σ−1) ln[β1 K 2L 3
. (31)
Denote the exponent above by ζ(σ). Using the de L’Hospital rule we have
ln[β1 K (σ−1)/σ + β2 L(σ−1)/σ + β3 R(σ−1)/σ ]
lim ζ(σ) = lim
σ→1 σ→1 (σ − 1)/σ
1/σ 2 [β1 K (σ−1)/σ ln K + β2 L(σ−1)/σ ln L + β3 R(σ−1)/σ ln R]
= lim
σ→1 1/σ 2 [β1 K (σ−1)/σ + β2 L(σ−1)/σ + β3 R(σ−1)/σ ]
= β1 ln K + β2 ln L + β3 ln R. (32)
Then
lim F (K, L, R) = eβ1 ln K+β2 ln L+β3 ln R = K β1 Lβ2 Rβ3 . (33)
σ→1
This limit function is called the Cobb–Douglas production function

F (K, L, R) = K β1 Lβ2 Rβ3 . (34)
Observe that we could get the same result starting from the nested function (29) and
taking sequentially limits σ1 → 1, σ2 → 1.
As limσ→∞ (σ − 1)/σ = 1, then
lim F (K, L, R) = β1 K + β2 L + β3 R. (35)
σ→∞
Also transforming this in a similar way as in the previous paragraph we get for σ → 0
lim F (K, L, R) = min{K, L, R}. (36)
σ→0
Thus taking into account the above limits we can define the CES function for all values
of 0 6 σ 6 ∞.
Let us notice that for all the above production functions the transformations of
the kind (7)–(10) are valid.
Assume that in the Cobb–Douglas function each of the factors change p times.
Then we have
F (pK, pL, pR) = pβ1 +β2 +β3 K β1 Lβ2 Rβ3 = pβ1 +β2 +β3 F (pK, pL, pR).
So, when β1 + β2 + β3 = 1 the production function is linearly homogeneous, i.e.,
changes also to p times. Thus the production function exhibits constant returns to
scale. This case is called the neoclassical production function. Other possibilities
include increasing returns to scale, when β1 + β2 + β3 > 1, and decreasing returns to
scale, when β1 + β2 + β3 < 1.
The production function in the model (28) must be specified before its solution
is attempted. The solution is induced by the factors present in the production function
and therefore the model has an exogenous character. Recently some interest emerged
in specifying an endogenous capital model. It started with the works of Romer [76],
Lucas [48], Barro [3] and Rebelo [74], see also [4], and is sometimes referred to as
Rebelo or Barro–Rebelo model. The main idea of these models is to condition the
exogenous factors of the previous models on the endogenous ones, mainly on the
accumulated capital which may include both physical and human capital.
Let us assume that each worker (agent), numbered by the index i, can allocate
his/her activity between labour, li (t), leisure, λi (t), and production of skills, ai (t),
so that li (t) + λi (t) + ai (t) = 1. The rates of allocation of the activity result from
maximization of the individual utility function which in a simplified form may be
taken identical for all agents. Growth of the stock of the human capital depends on
the fraction φi (t) of the physical capital and efficient labour on accumulation of skills
ai (t)Hi (t), with the depreciation rate νi . Denoting by Fi the production function (e.g.,
Cobb–Douglas) of accumulation of the human capital we get

Ḣi (t) = −νi Hi (t) + Fi φi (t)K(t), ai (t)Hi (t) . (37)
P
Denote by φ(t) = i φi (t) the P fraction of the physical capital used for accumula-
tion of skills. We also have L(t) = i li (t)Hi (t). Then the production function takes
the form
X
Y = e F (1 − φ)K,
θt
li Hi , R .
i
The analysis can be simplified if it is assumed that all agents are identical.
In the simplest form it may be assumed that the production function is propor-
tional to the capital, that is
Y (t) = bK(t)
which can be formally obtained from (30) taking β1 = 1 and β2 = β3 = 0 or from (36)
under assumption on the limiting role of the capital on the production function. Here
K can be also interpreted as being a composite of the human and the physical capital.
Assuming further, for example, that the firms produce to maximize profits and
the agents maximize their (known) utility functions the problem can be reduced to
analysis of relatively small subset of factors deciding on development of the economy.

This permits to study such phenomena as endogenous growth, income convergence
across different regions and generations and so on, see, e.g., [7,56,74,91].
2.4. Resources
The problem of natural resources exhaustion was a subject of interest already in

the 19th century when Malthus discussed the food limitation for the growth of human
population, with exponentially growing population and limited agricultural land. The
mathematical treatment of this subject began, however, with the work of Hotelling [38]
in 1931. In 1950s Gordon [33] and Schaefer [77] discussed some economic questions
connected with fishery, treated as a common resource. The Hotelling ideas were also
continued by Gordon [34]. In 1970s, partially due to the Club of Rome activity
(see [58,60]) and the oil crises, the problem of resources exhaustion became quite a
popular subject in the economic literature. Many now classical papers were presented
at the Symposium on the Economics of Exhaustible Resources and printed in The
Review of Economic Study, [19,86–88]. This subject is also extensively discussed in
books, see, e.g., [12,14,20,25,35,82]. The detailed model depends very much on the
particular case considered, see, e.g., [67]. However, a rough classification of cases is
possible. This is what will be done in the sequel.
2.4.1. Nonrenewable resources

Nonrenewable resources are usually related to mining. The models of nonrenew-
able resources usually are classified into two groups: pure depletion or exhaustible
resources, and nonrenewable resources with possibility of discoveries.
Exhaustible resources. The model in this case takes the very simple form
Ṙ(t) = −Q(t), (38)

R(0) = R0 , (39)
where Q(t) is extraction of the resource. The extraction may equal the demand Q = D
which can then be a function of the rent π(t), which leads to the dependence in the
form

Ṙ(t) = −D π(t) . (40)
Under competition a so called Hotelling rule is often applied
π̇(t)
=δ (41)
π(t)
which connects the change of the rent with the interest rate δ. The Hotelling rule gives
the simple equation for the rents π(t) = π(0)eδt . Then denoting by T the time when
reserves will be exhausted we derive the following simple problem
Z T
Q(t) dt = R0 , (42)
0

Q(T ) = Q π(0)eδT = 0. (43)
Knowing Q, the equations may be solved for π(0) and T .

This solution is usually compared either with the monopolistic one, i.e., when the
rent is determined by a monopolist, who would wish to maximize
Z T

e−δt π Q(t) Q(t) − Cs Q(t) dt (44)
0
subject to (38), where Cs is the cost of extraction; or with the social optimum one,
where the problem is to maximize
Z T

e−δt U Q(t) dt (45)
0
subject to (38), where U is a suitably defined utility function. Both the above are
optimal control problems which can be further analyzed using, e.g., maximum principle
conditions and phase plane analysis, see also the literature cited above.
Exploration and discoveries. The problem can be slightly more complicated if ex-
ploration and discoveries may augment reserves. Let then X(t) represent cumulative
discoveries. We may now write a model for discoveries

Ẋ(t) = d E(t), X(t) , X(0) = X0 , (46)
where d is a discovery-rate function depending on E(t) – the exploratory effect, which

usually will be a function of the capital allocated for exploration. Including the effect
of discoveries in (38) we get

Ṙ(t) = d E(t), X(t) − Q(t), R(0) = R0 . (47)
This case then leads to a two-state ((46) and (47)) optimization problem. Note that
the objective function should now include the cost of exploration.
Research and development. Another possibility to enhance the time of using an

exhaustible resource relates to change of technology. This is usually connected with
research and development in the area. Constant progress can be described by the
rate θ connected with the production function (26). Here we consider the case of
one change in technology. Assume that a breakthrough occurs at time T . This will
influence consumption C and extraction of the resource Q, and possibly also the utility
function U . Then the objective function takes the form
Z T Z ∞
−δt
e U1 (C1 , Q1 ) dt + e−δt U2 (C2 , Q2 ) dt. (48)
0 T
Time T depends on the level of knowledge X(t). We assume that the level necessary
for the breakthrough is Xs , that is the time T is given by
T : X(T ) = Xs . (49)
Accumulation of knowledge may be described by a differential equation dependent on
the capital KB (t) allocated to the research and development

Ẋ(t) = B KB (t) , X(0) = 0. (50)
If this equation is a deterministic one, then T can be found from the relation
Z T

B KB (t) dt = Xs (51)
0
and the problem reduces to an optimal control problems with a slightly more compli-
cated objective function (48).
However, we may consider that actually the dependence (50) is a stochastic one,
and therefore T is a stochastic variable, see [19,21,40]. We assume then that it is
desired to optimize the expected value E{·} of the objective function (48)
Z T Z ∞
−δt −δt
E e U1 (C1 , Q1 ) dt + e U2 (C2 , Q2 ) dt (52)
0 T
with respect to the stochastic variable T . Moreover, we assume that the probability
of discovery of the new technology is known and described by the probability density
function ω(T ), such that
Z ∞
ω(T ) dT = pB 6 1. (53)
0
Thus, we allow for the nonnegative probability pN = 1 − pB that the new technology
will not be discovered at all. Now we can write the objective function as
Z ∞
pN e−δt U1 (C1 , Q1 ) dt
0
Z ∞ Z T Z ∞
−δt −δt
+ ω(T ) e U1 (C1 , Q1 ) dt + e U2 (C2 , Q2 ) dt dT. (54)
0 0 T
Denoting
Z ∞
e−δt U2 (C2 , Q2 ) dt = W C2 (T ), Q2 (T ) (55)
T
and integrating the second (nested) integral in (54) by parts we get

Z ∞ Z ∞ Z ∞
pN e−δt U1 (C1 , Q1 ) dt + e−δt U1 (C1 , Q1 ) dt ω(t) dt
0 0 0
Z ∞ Z t
−δt
− e U1 (C1 , Q1 ) ω(τ ) dτ dt
Z ∞
0 0

+ ω(T )W C2 (T ), Q2 (T ) dT
Z ∞0 Z ∞
−δt

= e U (C1 , Q1 ) pN + Ω(t) dt + ω(t)W (C2 , Q2 ) dt, (56)
0 0
R∞
where Ω(t) = t ω(τ ) dτ . This way the objective function (52) is reduced to the
deterministic one. This again allows us to apply the optimal control theory approach.
Let us, however, observe that the requirements of knowledge of the probability char-
acteristics, necessary for determining pN and ω(τ ), may be limiting its use.
To learn more on research and development problems see, e.g., [1,75].
2.4.2. Renewable resources

Renewable resources consist of biological populations which can reproduce and
grow, or some inanimate resources which are subject to supplementary flux, like water
or wind.
Growth function. The models connected with this kind of resources need to take
into account the nature of growth or renewal of the resource. This may require rather
thorough knowledge of the underlying phenomena, like biology of the species consid-
ered or climate and geology of the terrain. As usual in the economic literature, we
consider that under lack of the supplementary flux the amount of the resource stock
R(t) changes according to a differential equation

Ṙ(t) = G R(t) , R(0) = R0 , (57)
where G is called the growth function. In the case of biological populations examples
of such models may, for example, comprise:
• the Malthus [54] model Ṙ(t) = γR(t), with a constant γ,
• the Verhulst [92] model Ṙ(t) = γR(t)[1 − R(t)
W ], with constants γ, W ,
• the Gompertz [32] model Ṙ(t) = γR(t) ln( R(t)
W
), with constants γ, W ,
• the Monod [63,64] model Ṙ(t) = γR(t) WS(t) +S(t) , with constants γ, W , where S(t)
is the amount of substance which limits the growth (like, e.g., food),
• the Lotka–Volterra [47,93] predator-prey model Ṙ1 (t) = γR1 (t) − αR1 (t)R2 (t),
Ṙ2 (t) = βR1 (t)R2 (t) − κR2 (t), with constants γ, α, β, κ (in this case two species
are taken into account),
and many other, see also section 2.6 Labour.
In the case of nonbiological resource often stochasticity of the supplementary

flux has to be considered. This will not be considered here.
Harvesting. The renewable resource may be harvested. This notion, typically used
for biological populations, can also mean exploitation of an inanimate resource, e.g.,
ground water, with supplementary flow. The rate of harvest (yield) Y per unit time
will usually depend on “effort” E(t) and the available stock, such that

Y (t) = M E(t), R(t) . (58)
Then the relation (57) takes the form

Ṙ(t) = G R(t) − Y (t) = G R(t) − M E(t), R(t) , R(0) = R0 . (59)
A possible concept here is to keep the resource on a constant level. Then Ṙ(t) = 0
and we get the set of algebraic equations
G(R) − Y = 0, Y = M (E, R) (60)
which, after having eliminated R, allow us to find the sustained yield function Y =
Y (E), giving the level of harvesting compensated by growth of the resource, and the
respective level of the resource stock R(E). With the above, given the stock R, the
yield can be optimized.
More generally we can introduce an objective function (usually a benefit) to find
an optimal effort. Here also the monopolist and social optimum can be considered.
An important issue, however, is to avoid overexploitation of the resource, which may
result in stock extinction. This in particular may easily happen when no regulation
is imposed (common-property resource). When connected with the capital equation a
question of optimal investment strategy may arise. These problems were extensively
developed in 1970s by Clark and a good source of their discussion is [14].
2.5. Pollution
Pollution models are often even more difficult to build than those for resources.
First of all, the harmful influence of a pollutant is usually related to its concentration,
which may be difficult to calculate knowing the emissions. Moreover, the emission
of some pollutants may be difficult to estimate, as it may be distributed on bigger
regions or, even more troublesome, related with some processes of not fully known
nature. Secondly, the flows of the pollutant may be of a rather complicated nature,
including dispersion in air, surface and/or underground water, or soil, with possibility
of interactions. It can travel with the receiving media to remote sites and influence
environment far from the place of their disposal. The site of the pollutant’s deposit
may depend on random phenomena, like meteorological conditions. Some pollutants
can also accumulate in living organisms, giving rise to its propagation in food chains.
Finally, the pollutants are usually subject to different chemical and/or biological trans-
formations, often little known. Fundamentals of modelling and many examples of
models can be found in [39].
The early papers on economic analysis of dynamic models of pollutants and waste
accumulations appeared at the beginning of 1970s, [17,42,51,71,78,84,94]. Probably
the best exposition of the early models is in [30], see also the review paper [85].
The early papers on economic dynamic pollution problems took a rather simplistic
models of pure accumulation of the pollutant or accumulation with a possibility of its
degradation
Ṗ (t) = E(t) − βP (t), P (0) = P0 , (61)
where P is the pollutant stock, β > 0 is the rate of its degradation and E is here
the emission. The emission is often connected with the production, as it is frequently
understood that the pollutants are mostly emitted as wastes in the production processes.
Unlike in the resource sector, here abatement of the pollution is possible (which
may also include recycling of the waste). This is mainly connected with capital KA
allocated to the abatement technology. Then the model takes the form
Ṗ (t) = E(t) − βP (t) − A(KA ), P (0) = P0 . (62)
Important issues in the pollution modelling are connected with the technical
progress and international dimension of polluting activity.
2.6. Labour
When longer times and bigger territories, like countries, continents or the globe,
are considered, the change in labour due to growth of population has to be considered.
The models here are usually of the exogeneous kind – they do not depend on other
model variables. For deeper description of population models see [43].
The most popular model used in the economic considerations is the Malthus
model [54] of exponential growth. It has the form
L̇(t) = γL(t), L(0) = L0 , (63)
where γ is a constant, called the intrinsic rate of growth. It has a simple solution, very
suitable for calculation of per capita values
L(t) = L0 eγt . (64)
This model describes the constant growth of the population and therefore has only
limited applications, mainly for shorter time horizons. There are other models which
take into account some saturation effects. For example the Verhulst model [92]

L(t)
L̇(t) = γL(t) 1 − , L(0) = L0 , (65)
W
where γ and W are constants, and W is called the carrying capacity of the environment.
Its solution is a famous logistic curve
W
L(t) = W −L0
(66)
1+ L0 e−γt
with the limit limt→∞ L(t) = W . Another famous model of this kind is the Gompertz
model [32]

W
L̇(t) = γL(t) ln , L(0) = L0 , (67)
L(t)
where γ and W are constants. Its solution takes the form

L0 −γt
L(t) = W e (68)
W
and again we have limt→∞ L(t) = W .
More complicated models take into account the age distribution in the population.
The models, mainly developed in the first half of this century, include integral, matrix,
difference and partial differential equations ones. This subject is extensively studied
in demography, see, e.g., [43,44].
2.7. Environment regulation

2.7.1. Basic ideas
A central idea in pollution control is connected with the fact that the firms,
looking for optimal production conditions in a competitive environment, may discharge
excessive wastes, engaging this way in polluting activities. Let us consider an example
which for simplicity is a static case, that is the variables are constant in time. We
assume that all the functions below are sufficiently smooth. Assume now that a
firm discharges a waste with emission E. The waste contains a pollutant causing
its accumulation in the environment P (E), which can be also effectively observed as
the ambient concentration of the pollutant. The firm production function F depends
on some input K (which may be the capital, the labour etc.), the emission E and the
pollution P , i.e., its form is F (K, E, P (E)). We adopt some conditions
∂F ∂F dP
> 0, 6 0, > 0. (69)
∂E ∂P dE
Moreover we assume that the functions above are concave. The optimal emission for
the firm can be found from the first order necessary optimality conditions
dF ∂F ∂F dP
= + = 0. (70)
dE ∂E ∂P dE
The social optimum will involve some utility function U (F , P ) and we assume
that
∂U ∂U
> 0, <0 (71)
∂F ∂P
and that the function U is concave. Now, the first order necessary optimality conditions
are

dU ∂U ∂F ∂F dP ∂U dP
= + + =0 (72)
dE ∂F ∂E ∂P dE ∂P dE
or
∂F ∂F dP ∂U dP . ∂U
+ + = 0. (73)
∂E ∂P dE ∂P dE ∂F
Let us denote the social optimum by E S . Due to the assumptions we have

dF ∂F ∂F dP ∂U dP . ∂U
= + = − > 0. (74)
dE E S ∂E ∂P dE E S ∂P dE ∂F E S
We see that the social optimum does not give the optimal solution for the firm. More-
over, to get the optimal solution it would be necessary for the firm to increase the
emission of wastes. Thus the firm optimizing its production discharges excessive
waste. To reach the social optimum some additional restrictions on emission should
be imposed.
The problem of externalities, like the one in the above example, was considered
by Pigou [68] who proposed to use taxes as a regulator. In the above example, denoting
the tax for the excessive emission by λ, we get the new function for the firm to be
optimized

f (E) = F K, E, P (E) + λ E − E S (75)
with the optimality condition
df ∂F ∂F dP
= + + λ = 0. (76)
dE ∂E ∂P dE
This condition is equal to the social optimum condition if

∂U dP . ∂U dF
λ= =− . (77)
∂P dE ∂F dE S E
Likewise, it is possible to tax the firm for pollution it causes, rather than for emission.
This case can be handled mathematically in a similar way. The pollution tax may be
more convenient for the central planner, as the pollution enters directly in the utility
function U (F , P ). On the other hand, it requires the firm to estimate the effects of
emission to efficiently manage the case.
It is easy to see that the assumption on concavity of the function F is crucial for
the effectiveness of the regulation. Assume that the central agency settles the tax λ
given by (77). Now, the problem of optimization by the firm can be geometrically
interpreted as in figure 1. Let us consider the left drawing. The tangent line to the
graph of the function y1 = F at the point E S has the form y2 = −λ(E − E S ) + α,
with α = y2 (E S ). We have

F + λ E − E S − α = y1 − y2 6 0 (78)
with the maximum equal to 0 at E = E S . As the constant term −λE S −α is inessential
for the maximization, the firm which maximizes the function
F + λE (79)
Figure 1. Concave and nonconcave production functions.
will end, under assumption of concavity of F , with the emission E = E S required by

the central agency.
Under lack of concavity, the situation changes, see the right drawing in figure 1.
After imposing the tax λ = −dF /dE|E S the firm seeking to maximize the function
F + λE will find now as optimal either the emission E 1 or E 2 . Both of them are
incorrect from the regulation point of view.
2.7.2. Regulation policies

Consider now the problem with N parties involved in the pollution, under as-
sumption of concavity of the functions. We have
XN

F K, E, P (E) = Fi Ki , Ei , P (E) (80)
i=1
P PN
with K = N i=1 Ki and E = i=1 Ei . In general we assume that the production
function of each firm depends on pollution caused by emissions of all firms. Water
from a lake polluted by all firms and at the same time used by them for the productions
may be an example of such situation.
To settle the tax the central planner calculates the derivatives of the utility function
U (F , P )

dU ∂U ∂F ∂F dP ∂U dP
= + + =0 (81)
dE ∂F ∂E ∂P dE ∂P dE
to find E S satisfying the equality and sets, as in the case of one firm,

dF ∂U dP . ∂U
λ=− = . (82)
dE E S ∂P dE ∂F E S
Notice, however, that
dU dU dE dU
= = (83)
dEi dE dEi dE
and similarly
∂Fi ∂F ∂F ∂P ∂P
= = , = . (84)
∂Ei ∂Ei ∂E ∂Ei ∂E
Thus we have

∂U dP . ∂U ∂Fi ∂F dP
λ= =− + . (85)
∂P dEi ∂F ∂Ei ∂P dEi
From another side, the derivative of the ith firm production function is

dFi ∂Fi ∂Fi dP ∂(F − Fi ) dP
= + = −λ − (86)
dEi E S ∂Ei ∂P dEi E S ∂P dEi
and it differs from −λ unless ∂(F − Fi )/∂P = 0 which is true, for example, when each
production function Fi depends only on the pollution P caused by the respective firm,
i.e., the function F has the form F (K, E, P (E)) = N i=1 Fi (Ki , Ei , Pi (Ei )), P (E) =
PN
i=1 Pi (Ei ), or it does not depend on pollution at all, i.e., each production function
is of the form Fi (Ki , Ei ). In these simple cases the tax settled by the central agency
imposes on the firms the conditions inducing the social optimum emissions. In general
case the tax λ may not induce them. Moreover, to find the optimal emission, each
firm has to know production functions of the other firms as well.
Thus, as more parties are involved the problem becomes more difficult.
Complications connected with imperfect information and defensive activities of
the victims may arise. That is why the Pigouvian taxes have been repeatedly attacked.
An early criticism, in the case of two parties, is due to Coase [13], whose argument
was that the distortions associated with the externalities would be resolved through
bargaining among interested parties. This may be, however, difficult to implement in
some conflicts involving pollution. And when more than two parties are involved,
the problem of externalities may never be resolved through bargaining. Some other
actions, like subsidies for lowering the emission, marketable permits, effluent charge,
etc., were discussed. There is also a possibility to use legal liabilities after pollution
effects have been found.
Analysis of some of those approaches is similar to the tax cases investigated
earlier. This is, e.g., true for the marketable permits, proposed for the first time by
Crocker [15], where the central agency issues emission permits for E S units of the
emittant. Also the case of issuing the ambient concentration permits is analogous to
taxing for the ambient concentration of a polluter. To discuss the latter case in more
details let us assume that the central agency considers the problem of maximizing the
utility function U (F , P ) with respect to P solving the equation
∂F ∂U . ∂U
+ = 0. (87)
∂P ∂P ∂F
Let us call the social optimum solution P S . The agency issues the permits for the
ambient concentration of P S units. A price q for the unit permit is then set on a
market and an individual firm has to solve the following optimization problem

max Fi Ki , Ei , P S − qpi , (88)
Ei ,pi
where pi is the number of permit units bought by the firm. To solve this problem the
dependence Ei = g(pi ) would be needed. This dependence can be easily found when
P is a linear function, see [65]. Finding this dependence for a nonlinear function will
be, however, in general impossible. Nonlinear dependences, for example, may arise
when different wastes contain components such that their common effect is reinforced,
as discussed in [98] where the linear case is extended and effectively solved for a class
of nonlinear problems.
Good sources of discussion of different actions and their impacts are [83] and [6].
For review of problems related with the policy instruments and reaction of the involved
actors as well as of the literature see also [16].
As the direct approach involving calculation of damages caused by some parties to
other constitute big obstacles, other solutions seem more practically relevant. A popular
approach is to determine first some standards for environmental quality which enables
then to design a regulatory system to achieve these standards. This often leads to
so called command-and-control policy, as opposed to economic incentives mentioned
above.
The Pigouvian-type environmental taxes discussed above disregard their inter-
action with the pre-existing ordinary taxes. Suggestions appeared in the economic
literature that the revenues from the pollution taxes could be used to decrease other
taxes, yielding this way a “double dividend”, i.e., not only forcing a cleaner environ-
ment but also giving opportunity to a less distortionary tax system. Some recent works
show results contrary to that hope. In an early paper Bovenberg and de Mooij [9] pre-
sented a model in which introduction of a tax on consumption of “dirty” commodities,
with constant government budget, gives a rise of the taxes on labour, when empirical
estimates of a parameter in the model are used. In the survey paper [36] impact of
additional factors missing in the Bovenberg–Mooij model, such as capital, subsidies,
taxes on “dirty” intermediate goods, emission taxes, unemployment are analyzed. This
analysis together with review of numerical findings from different models tend to leed
to the conclusion that the “double dividend” effect is rather unlikely unless the pre-
existing ordinary tax system is inefficient. The introduction of the environmental tax
may, however, help in obtaining other “dividends”, like higher employment or better
terms of trades.
A conclusion of the “double dividend” works is that the optimal environmental tax
can be generally less than the Pigouvian one, see the discussion in [36] and references
cited therein. To better investigate it the benefits from pollution reductions should be
taken into account. The gross benefits associated with improvement of the environment
may vastly outweigh the gross costs of introducing the environmental taxes. The policy
influences also the technological progress [62].
2.7.3. International aspects of pollution control

Pollution can travel long distances and cross the intercountry boundaries. This
can drastically change the results of applying environment regulation policy. Let us
imagine a simple case of two countries interchanging the pollutants by their boundary
through air or water. It is obvious that if each country maximizes its welfare functions
on the basis of its own emission and disregarding the exchange, the result of their
environmental regulation will depend on the net difference of exchange. If this dif-
ference is big, one of the countries will overregulate while another will underregulate
the pollution.
If emissions are known and the flow of pollutions through boundaries can be
estimated, each country can undertake their own environment actions to maximize
their welfare functions. The benefit from emission reduction may there be included
independently for each country, given the emission of each country involved, or de-
pending on emission reduction in all countries. Both these cases can be modelled as
noncooperative dynamic games, see, e.g., [5,46,59], which lead to the noncooperative
Nash equilibrium.
If all countries decide to cooperate, the cooperative solution of the game may
be found which always gives better or at least equal values of welfare functions pro-
vided that adequate compensations are paid to those countries which loose from the
cooperation when compared to the noncooperative solution.
The detailed presentation of the problems and solutions connected with environ-
mental policies in transboundary pollutions is out of scope of this paper. This and
related problems are discussed, e.g., in [27,28,52,53,69,70].
3. One state variable pollution problems
Although theoretically possible, the detailed analysis of the problem (1)–(6) seems
rather cumbersome. Various simpler models have been therefore proposed and ana-
lyzed. The simplest one with the one state variable will be reviewed in this section.
If the labour growth is constant or given exogeneously (and then eliminated),
then there are three possible variations of the problem (1)–(6) giving rise to one state
variable problems. These are: with the pollution accumulation model, with the capital
accumulation model, and with the model describing dynamics of a resource extraction.
We discuss here only the first of them, with the integrated discounted utility criterion
(dut). The early formulations of this problem are due to Plourde [71] and Keeler
et al. [42] but its main analysis was done by Forster [30]. Besides the discussion of the
standard equilibrium model we give here also full discussion of the nonstandard models
with no equilibrium points but only steady state stationary points on the boundaries.
The discussion of other models can be found elsewhere. The book [25] is a good
source where also discussion of the Ramsey model of capital accumulation can be
found, together with pollution treated as a static variable. Various resource extraction
models are also extensively discussed in [12,14].
3.1. Assumptions and problem formulation
The utility function is here expressed as a smooth (twice continuously differ-

entiable, including continuities at the boundaries of the feasibility set) finite value
function which depends on only two variables: consumption and pollution, i.e., it has
the form U (C, P ). It is also assumed that in the interior and on the boundary (except
possibly in few places where it will be clearly stated) of the set of feasible solutions
UC > 0, UCC < 0, UP < 0, UP P < 0, UCP 6 0. (89)
For some technical proofs we assume sometimes that the higher cross derivatives of
the above function are zero, i.e.,
UC k P l = 0 for k > 0, l > 0, k + l > 2. (90)
Notice that this assumption is always true when the function U is separable in C and P .
We call the functions satisfying the above condition weakly connected in arguments. It
is further assumed that a constant stream of output K < ∞ is produced. It is allocated
to consumption C and pollution control (abatement) KA so that
K = C(t) + KA (t). (91)
The stock of pollution increases according to the equation (62) discussed in the sub-
section 2.5, which in this case has the form

Ṗ (t) = E C(t) − A KA (t) − βP (t), P (0) = P0 > 0. (92)
A function Z is again defined as a smooth (as in the case of U ) finite function
Z(C) = E(C) − A(K − C) (93)
and the following assumptions are taken in the interior and on the boundary (with
some exceptions as described before) of the set of feasible controls
Z 0 (C) > 0, Z 00 (C) > 0 (94)
and we assume that Z is bounded on [0, K]. We also assume that there exists a C0
such that Z(C0 ) = 0 and that 0 < C0 < K. Then

 < 0 if 0 6 C < C0 (net abatement),
Z(C) = = 0 if C = C0 (net sustainability), (95)

> 0 if C0 < C 6 K (net pollution).
Thus the problem can be formulated as follows

Z ∞
max e−δt U (C, P ) dt, δ > 0, (96)
C 0
Ṗ = Z(C) − βP , P (0) = P0 > 0, (97)
C > 0, K − C > 0, P > 0. (98)
Notice that because of the assumptions taken

Z ∞ Z ∞
−δt U (K, 0)
e U (C, P ) dt 6 e−δt U (K, 0) dt 6 < ∞. (99)
0 0 δ
Thus the integral exists.
Different simpler formulations of the above problem have been considered, for
instance as in illustrative examples in [14,79]. The simplifications mainly include
separation in parameters of the utility function and often linearity of the equation
of motion (97). An extension of this problem, for the case when the utility function
depends on the derivative (rate) of the pollution accumulation Ṗ was discussed in [90].
However, in that paper the utility function is separable and the equation of motion is
linear.
Let us now define the (current value) Hamiltonian and Lagrangian functions

e
H(C, P , λ) = λ0 U (C, P ) + λ Z(C) − βP , (100)
e
L(C, P , λ, µ1, µ2 , µ3 ) = H(C, P , λ) + µ1 C + µ2 (K − C) + µ3 P. (101)
Then the necessary (maximum principle) optimality conditions are

e C, P ∗ , λ∗ ,
C ∗ = arg max H (102)
C

LC = λ0 UC + λ∗ Z 0 C ∗ + µ1 − µ2 = 0, (103)
∗ ∗
λ̇ = (δ + β)λ − UP − µ3 , (104)

µ1 C ∗ = µ2 K − C ∗ = µ3 P ∗ = 0, µi > 0, i = 1, 2, 3, (105)

λ0 , λ∗ 6= (0, 0), λ0 = 0 or 1 (106)
and additionally (97), (98).
3.2. Some solution properties
3.2.1. Discussion of λ0
Let us consider first the case λ0 = 0. In this case the solution does not depend
on the utility function but is determined by constraints. Indeed, the Hamiltonian (100)
now is

e
H(C, P , λ) = λ Z(C) − βP . (107)
From (106) we have λ∗ 6= 0. When λ∗ > 0, then the maximum of H(C, e P ∗ , λ∗ )

is at the value maximizing Z(C), that is at C = K (no abatement). Solving the
equation (97) we get in this case
Z(K)
P (t) = P0 e−βt + 1 − e−βt > 0. (108)
β
Then from (105) µ3 = 0 and from (104) we get at the steady state
UP
λ∗ = < 0. (109)
δ+β
As λ∗ is continuous in t, this contradicts our assumption λ∗ > 0.
Now assume λ∗ < 0. Then the value maximizing the Hamiltonian is C = 0 (full
abatement). In this case, as Z(0) < 0
βP0
P0 e−βt + Z(0)
β (1 − e−βt ) for t < β1 ln 1 − Z(0) ,
P (t) = (110)
0 otherwise.
But then, for t > β1 ln(1 − βP0 /Z(0)), the abatement K − C0 is sufficient which means
that the sustainable value C = C0 > 0 can be shifted to consumption, increasing
this way the utility function. Then C = 0 does not maximize the Hamiltonian. This
contradiction finally eliminates the case λ0 = 0 and therefore we may assume λ0 = 1.
3.2.2. Solution at P = 0
We start this case again with discussion of solutions on the boundaries. Solution
of the equation (97) is
Z t
−βt

P (t) = P0 e + e−β(t−τ ) Z C(τ ) dτ. (111)
0
Then, if C > C0 (i.e., Z > 0), then P (t) > 0 for all finite t. If not, then there
exists at least some interval at the beginning where P (t) > 0 (because P0 > 0 and
Z is bounded). Therefore µ3 = 0, at least for a sufficiently small t. Let us also
notice as a byproduct of this analysis that as by assumption Z(C) < Z(K), then
P (t) < P0 e−βt + β1 Z(K)(1 − e−βt ). Then solution of the above differential equation
is bounded and the asymptotical solution is bounded by Z(K)/β.
Let us, however, notice that the steady state (equilibrium) solution may or may
not be on a boundary P = 0, dependent on some function characteristics [29]. For
this let us consider the costate-state (λ-P ) plane and assume that µi = 0, i = 1, 2, 3.
For the curve λ̇∗ = 0 we have from (104) and (89)
UP
λ∗ = < 0 for P > 0. (112)
δ+β
Now, differentiating (103) with respect to P we get
UCP
C 0 (P ) = − > 0. (113)
UCC + λ∗ Z 00 (C)
Finally, differentiating (112) with respect to P yields

UP P + UCP C 0 (P )
λ0∗ (P ) = < 0. (114)
δ+β
So the curve λ∗ (P ) is strictly decreasing. Its value for P → 0+ is given by
limP →0+ UP (C, P )
lim λ∗ (P ) = ≡ λλ . (115)
P →0+ δ+β
Let us now consider the curve Ṗ = 0. From (97) we have

Z(C)
P = (116)
β
and then
Z 0 (C)C 0 (λ∗ )
P 0 (λ∗ ) = . (117)
β
Again from (103), differentiating now with respect to λ∗
UCC C 0 (λ∗ ) + UCP P 0 (λ∗ ) + Z 0 (C) + λ∗ Z 00 (C)C 0 (λ∗ ) = 0 (118)
which, after some small algebraic manipulations, gives
[Z 0 (C)]2
P 0 (λ∗ ) = − > 0. (119)
βUCC + βλ∗ Z 00 (C) + UCP Z 0 (C)
So the curve P (λ∗ ) is strictly increasing.
We can find the value of λ∗P giving P (λ∗P ) = 0. At P = 0 we get Z(C) = 0
which is satisfied for C = C0 (net sustainability assumption). Then from (103) we
have
UC (C0 , 0)
λP = − 0 . (120)
Z (C0 )
Now (see also figure 2), if λP < λλ , then the curves for λ̇∗ = 0 and Ṗ = 0 cut
for P > 0 and the steady state value of pollution is positive. Otherwise the steady state
value of pollution is 0 and the asymptotic stationary point1 is (λ∗P , 0), with λ̇∗ = 0 for
µ3 = (δ + β)λP − UP (C0 , 0). A condition to assure the positive steady state pollution
may be λλ = 0 which is for example satisfied for
lim UP (C, P ) = 0 for any C. (121)

P →0+
1
Notice that the stationary point in [29] is misplaced due to not considering the constraint P > 0 there.
There may be an asymptotic solution on a boundary, not violating the constraint, but this may happen
only when the curves for λ̇∗ = 0 and Ṗ ∗ = 0 cut at the boundary.
Figure 2. The equilibrium (case 1) and the stationary point (case 2) at the costate-state phase plane.
3.2.3. Solutions at C = 0 or C = K
From the equation (103) we get
UC + µ1 − µ2
λ∗ = − . (122)
Z 0 (C ∗ )
Thus, if C ∗ is in the interior of the interval [0, K] (where µ1 = µ2 = 0), then

λ∗ < 0 and the Hamiltonian is concave there. Then from the maximization in (102)
we can get a solution in the interior or possibly at a boundary point.
Solutions at the boundaries 0 or K can be eliminated by making appropriate
assumptions on the problem functions. As we have
Z 0 (C) = E 0 (C) + A0 (K − C) (123)
then
lim A0 (C) = +∞ ⇒ lim Z 0 (C) = +∞. (124)
KA →0+ C→K −
Now, UC must be bounded at K because the function U is increasing and concave

with respect to C. If we confine ourselves to finite µ2 (or finite λ∗ ), then we get
from (122) λ∗ = 0. But this cannot be true all the time as from (104) we have in this
case (notice that µ3 = 0)
λ̇∗ = −UP > 0. (125)
Thus with the assumption (124) the solution cannot be at C = K for finite λ∗ .
However, we cannot take a similar assumption at C = 0, because Z is increasing

and convex. Therefore Z 0 (0) must be finite. Instead, another assumption may be taken
lim UC (C, P ) = ∞ for any P > 0. (126)
C→0+
With the above from (103) either λ∗ or µ2 must be infinite. So it is impossible to

satisfy the condition (103) for finite values of the involved variables.
3.2.4. Additional remarks

Summarizing, the following assumptions are sufficient to keep the solution out
of the boundaries (for finite values of variables):
lim UP (C, P ) = 0 for any C (to have P > 0), (127)
P →0+
lim Z 0 (C) = +∞ (to have C < K), (128)
C→K −
lim UC (C, P ) = ∞ for any P > 0 (to have C > 0). (129)
C→0+
A possible shape of the Z(C) curve is depicted at figure 3.

Let us, finally, check the constraint qualification. Calculating the derivatives of
the constraints with respect to C in the first column and adding constraint on the
diagonal elements of the second part we construct the matrix
 
1 C 0 0
 −1 0 K − C 0  . (130)
PC 0 0 P
For the maximum principle to hold this matrix should be of full rank. The troubles
with the full rank may occur when PC = 0 for P = 0 and either C = 0 or C = K.
From (103) we have for P = 0
ṖC = Z 0 (C). (131)
Figure 3. A possible shape of the function Z(C).

Then ṖC > 0 except possibly at C = 0, if Z 0 (0) = 0. The latter case is,
however, not feasible with our assumptions on Z(C) because then P = 0 and C = 0
(full abatement under no pollution). Thus, even when PC = 0 at some t0 , then it could
only be at this specific time, with PC > 0 for t > t0 . We can then conclude that the
constraint qualifications are satisfied.
4. Two- or more-dimensional state
Leaving apart the labour model there are three possible combinations of the pairs
of different one state variable models among those listed in section 2. Two of them
include the pollution model. These pairs are: (1) pollution accumulation–capital accu-
mulation models, and (2) pollution accumulation–resource extraction models. Pair (1)
is a direct extension of the Forster one dimensional model discussed so far, with the dy-
namic description of the capital accumulation (growth) replacing the static description
of the Forster model.
Both pairs were discussed in the literature. Examples for the pair (1) may in-
clude [79, chapter 5, section 2, example 3] (growth that pollutes), [49,50] (which is
also presented in [25, section 15.2; 69,96] and examples for the pair (2) [14, sec-
tion 4.33], (resource depletion and residual accumulation), [79, chapter 3, section 8,
example 12] (resource extraction with waste) or [89].
Actually, many concepts of the previous section extend even to more than two
differential equations, and some also to the case when functions depend on t. Let x
be a n-dimensional vector. Let us again assume the twice continuous differentiability
of the functions and consider the set of nonlinear differential equations

ẋ(t) = F x(t), t .
Let us assume that a steady state solution x∗ (t) satisfying F (x∗ (t), t) = 0 exists.
Denote z(t) = x(t) − x∗ (t). Then we have

ż(t) = F x∗ (t) + z(t), t − F x∗ (t), t = Fx x∗ (t), t z(t) + f (z, t),
where f (z, t) = o(|z|). For the known x∗ (t) the derivative Fx (x∗ (t), t) is a function
of t only and we can always write it in a form Fx (x∗ (t), t) = A + B(t), where A does
not depend on t. This way we reduce our original problem to the following equation
ż(t) = Az(t) + B(t)z(t) + f (z, t) = Az(t) + g(z, t). (132)
In the important autonomous problem case F (x(t), t) = F (x(t)) does not depend
explicitly on t. In this case also a steady state solution of the equation F (x∗ ) = 0 does
not depend on t and forms a critical point (or an equilibrium). Then from the Taylor
expansion

ż(t) = ẋ(t) = Fx (x∗ ) x(t) − x∗ + O r 2 = Az(t) + g(z),
where r = |x(t) − x∗ | (| · | is an Euclidean norm) and we easily identify A = Fx (x∗ )
and g(z) = O(r 2 ).
Although the definitions of A and B(t) in (132) are to some extend arbitrary,
only some of them may be significant in further analysis. Namely, we require now,
that g be continuous, g(0, t) = 0 and that for a given ε > 0 there exist δ and tε
that |g(z, t) − g(z̃, t)| 6 ε|z − z̃| for |z| 6 δ, |z̃| 6 δ and t > tε (observe that these
conditions are obviously met for the autonomous system considered above). Let all
the characteristic roots of A have negative real parts. Then any solution starting close
enough to the origin converges to it, i.e., z̃(t) → 0 as t → ∞ (in the autonomous case
we can also say that the critical point x∗ is stable, of the node type).
Assume now, that k characteristic roots of A have negative real parts and n − k
characteristic roots have positive real parts. Then for any sufficiently large t there
exist in the z space a real k-dimensional manifold S containing the origin such that
any solution z ∗ (t) starting at the manifold S satisfies z ∗ (t) → 0 as t → 0. Moreover,
any solution near the origin but not on S does not converge to it (in the autonomous
case the critical point x∗ is of the saddle type).
As before, for the autonomous system the solution of a linearized system
ż(t) = Az(t) is of the form
z(t) = a1 W1 es1 t + · · · + ak Wk esk t + ak+1 Wk+1esk+1 t + · · · + an Wn esn t ,
where si , i = 1, 2, . . . n, are the eigenvalues and Wi , i = 1, 2, . . . n, are the eigen-
vectors of the matrix A. So also in this case it is possible to find directions of the
convergent manifolds.
However, analysis of the multidimensional equation is much more difficult. Let
us consider the case of two-dimensional equation of motion, that is x(t) is now a two
dimensional vector x(t) = [y(t) v(t)]T and take for ẋ(t) = F (y, v, u) also the following
notation

ẏ(t) = f y(t), v(t), u(t) , v̇(t) = g y(t), v(t), u(t) .
The maximum principle equations now are
T
η fu
Uu + = Uu + ηfu + λgu = 0, (133)
λ gu
" # T
η̇ η Uy η fy fv
−δ =− − .
λ̇ λ Uv λ gy gv
Then considering the state-costate case we have the following set of four equations to
be analyzed (for simplicity dependence of functions on time was dropped)

ẏ = f y, v, û(y, v) ,

v̇ = g y, v, û(y, v) ,

η̇ = δη − Uy y, v, û(y, v) − ηfy y, v, û(y, v) − λgy y, v, û(y, v) ,

λ̇ = δλ − Uv y, v, û(y, v) − ηfv y, v, û(y, v) − λgv y, v, û(y, v) ,
where û(y, v) is an implicit function of the equation (133).
Some special cases simplifying the task can be considered. For example, in [25]
two such classes of problems are identified. One is when one of the equations, for in-
stance the second, is linear in v, with a coefficient δ, i.e., g(y, v, u) = δv−h(y, u) and U
and f do not depenl on v. Then the second equation becomes λ̇(t) = δλ(t) − δλ(t) = 0
and therefore
Rt λ is a constant. Moreover, the equation for v can be solved as
v(t) = 0 e−δτ h(y(τ ), u(τ )) dτ . This case can then be reduked to the analysis of
the one state variable model, see [25].
A second class considered in [25] consists of the models with separable states, in
which all second derivatives with respect to states, or states and control, are equal to
zero, i.e., Uxx = 0, Uxu = 0, Fxx = 0, Fxu = 0. This means that first order partial
derivatives with respect to states must be some constants independent on neither states
nor control. In such a case the costate equations are linear with constant parameters
and can be solved. This argument is, of course, true also for more dimensional cases.
The case of full two state variables is analyzed in [25]. It is shown that the
eigenvalues ξi , i = 1, 2, 3, 4, of the linearized system are given by a formula
s
tr J tr J 2 K 1p 2
ξ1,2,3,4 = ± − ± K − 4 det J,
4 4 2 2
where tr stays for the trace of the matrix
 
ẏy ẏv ẏη ẏλ
 v̇y v̇v v̇η v̇λ 
J =  η̇y η̇v
 (134)
η̇η η̇λ 
λ̇y λ̇v λ̇η λ̇λ
and

ẏ ẏη v̇v v̇λ ẏ ẏλ
K = y + + 2 v .
η̇y η̇η λ̇v λ̇λ η̇v η̇λ
Obviously, under the conditions
K2
K < 0, 0 < det J 6
4
all eigenvalues are real and two of them (connected with y and v, see [25]) are negative,
and the two other positive. Then under the above conditions the problem has a saddle
point. In [25] also formulae to calculate the directions of the stable orbits on the plane
(y, v) are given.
With full Jacobian matrix, calculations for the two state variables are rather
cumbersome. But in special cases (as for matrices with many zero entries) calculations
may be easier.
The models with two dimensional state analyzed analytically in the literature are
usually even of a simpler form than those discussed in the previous subsection. Very
often the utility function is separable in its arguments and the equations of motion are
linear or at least separable and partly linear. Cases of more complicated formulation
are either treated for the finite horizon ([79] where the products of arguments is present
in the functions) or numerically (like in the Luptacik and Schubert model, see [25]).
A more general model problem was considered in [89] where Jacobian matrix analysis
was used. These examples show increasing difficulty when a general theory and
methods of analytical analysis are being developed for the multistate variable models.
5. More detailed modelling
Many real cases involve much more than one or two state variables. Also the
functions involved may be rather complicated and not easy to analyze analytically.
Their analysis can now only be provided by numerical computations. With this respect
the discrete time models are quite often considered, as more suitable for computer
applications. There are, however, differences in application of continuous and discrete
time models. For instance, the maximum principle does not always hold true for the
discrete time problems, even when similar assumptions on the problem functions are
taken, see [37]. The solution of the nonlinear difference equations may be chaotic, even
for quite simple functions, see, e.g., [23,57], which is not a case for the continuous time
analogues. Thus simplification in computations are to some extent counterweighted
by more difficult analysis and uncertainty in the form of produced results.
As the models are usually formulated with a maximization or minimization cri-
terion, cf. (1) and subsection 2.2.2, advantage is usually taken of numerical software
suited for solution of large scale dynamic optimization problems. This permits relative
freedom in the modelling of the problem to be analyzed.
As examples we may mention the models Global 2100 [55] and DICE [66] which
are multidimensional in the states, employ functions which are not easily analytically
tractable, which have stochastic elements and which are formulated as discrete time
optimal control models. The models focus on the cost of controlling the greenhouse
gases emissions, notably CO2 , produced by burning fossil fuels.
In the sequel we give a brief account of the structure of the basic version of
the DICE model. The model is an extension of the Ramsey model to environmental
policy, with emissions reductions in the extended model playing the role of investment
in the mainstream model. Thus, the model contains both a traditional economic sector
and a climate sector.
The equations of the model are the following:
X
max U c(t), L(t) (1 + ρ)−t , (135)
{c(t)}
t
1−α
U c(t), L(t) = L(t) c(t) − 1 /(1 − α), (136)
Q(t) = Ω(t)A(t)K(t)γ L(t)1−γ , (137)
Q(t) = C(t) + I(t), (138)
c(t) = C(t)/L(t), (139)

K(t) = 1 − δK K(t − 1) + I(t − 1), (140)

E(t) = 1 − µ(t) σ(t)Q(t), (141)

M (t) = βE(t − 1) + (1 − δM ) M (t − 1) − 590 + 590, (142)

F (t) = 4.1 log M (t)/590 / log(2) + O(t), (143)

T (t) = T (t − 1) + (1/R1 ) F (t) − λT (t − 1) (144)

− (R2 /τ12 ) T (t − 1) − T ∗ (t − 1) ,

T ∗ (t) = T ∗ (t − 1) + (1/R2 ) (R2 /τ12 ) T (t − 1) − T ∗ (t − 1) , (145)
D(t) = Q(t)Θ1 T (t)Θ2 , (146)
TC(t) = Q(t)b1 µ(t)b2 , (147)

Ω(t) = 1 − b1 µb2 / 1 + Θ1 T (t)Θ2 . (148)
The economy is endowed with an initial stock of capital C and labour L and an
initial level A of technology. All industries behave competitively. Each country maxi-
mizes an intertemporal objective function, (135), identical in each region, which is the
sum over the relevant time periods of discounted utilities of per capita consumption c
times population L; the utility function is logarithmic in per capita consumption (ob-
tained in the limit in (136) for α → 1) and ρ is the pure rate of social time preference.
Output Q is produced by a Cobb–Douglas production function in capital, labour and
technology, (137); γ is the elasticity of output with respect to capital, and Ω is de-
fined in (148). Output is distributed between consumption C and gross investment I,
(138). Per capita consumption is defined in (139), and the balance equation for capital
stock K, (140), reflects the lifetime of capital through the rate of depreciation δK . Pop-
ulation growth and technological changes are exogenous, while capital accumulation
is determined by optimizing the flow of consumption over time, (135).
The climate part of the model contains a number of geophysical relationships
that link together the different forces affecting climate change. Emissions include all
greenhouse gases emissions, although they are most easily viewed as CO2 . Greenhouse
gases emissions can be controlled by increasing the prices of factor inputs that are
greenhouse gases intensive. Atmospheric concentrations of greenhouse gases influence
global mean surface temperature, and the economic impacts of climate change are
assumed to be increasing in the realized temperature increase. This part is constituted
by an emission equation (141), a concentrations equation (142), a climate-change
equation (143), and a climate-damage relationship (144). In (141), µ represents the
fractional reduction of emissions relative to uncontrolled emissions, i.e., the emissions
control rate; σ represents the uncontrolled rate of greenhouse gases emissions. In (142),
M is the amount of greenhouse gases in the atmosphere, β and δM being constants.
In (143), F is the radiative forcing from CO2 , and O represents other greenhouse gases,
such as CH4 and N2 O. Radiative forcing and climate change are linked in (144), (145),
where T is the increase in the average temperatures in the atmosphere and upper level
of the ocean, and T ∗ is the increase in the average temperatures in the deep oceans;
parameters R1 , R2 , λ and τ12 express various physical phenomena.
The economic impact of climate change may now be specified. In (146) the
economic impact of climate change is expressed, such that D is the loss of global
output, with Θ1 and Θ2 being parameters. (147) expresses the cost TC of reduction
of emissions, with b1 and b2 being parameters. Combining the cost and damage
relationships we have the definition (148) of Ω, used in (137).
As seen, the concepts and ideas of relationships have already been met in the
previous discussion in this paper. In this respect therefore the DICE and similar mod-
els are clearly within the tradition of the works that have constituted the subject of
the present review. They distinguish themselves by having far more details, in terms
of variables, relationships between variables, complication of functions, data and pa-
rameters, etc. Thus, they may represent a closer approximation to the quantitative
side of the reality, which they aim at representing. This is attained at the cost of
higher demands on selection of functional forms and relationships, higher data acqui-
sition efforts, required validation efforts, higher solution efforts (relying on numerical
algorithms) and less transparency.
6. Conclusions
We have reviewed the literature on mathematical models in economic environ-

mental problems, specifically dynamic models. As seen, there is an extensive amount
of material dealing with the intertemporal analysis of the interplay between capital,
resources and pollution.
The emphasis in the presentations has been on the mathematical structure of the
models and on the solutions of them.
As seen, the models with one state variable (pollution, resources or capital) have
been extensively analyzed in the literature. Many of the discussed models permit
analytical solution and neat interpretation of the optimality conditions and solution
structure. Thus, these models may be considered well understood. To some extent this
also holds for models with two-dimensional states.
However, many real world problems, although suitable for analysis by mathe-
matical models, are so extensive that it will be not possible to adequately reflect them
in a model with few state variables and easily tractable functions. Thus, analysis
can only be performed by application of numerical methods. While this permits a
larger freedom in the modelling of the problem it also to some extent obscures any
simple solution structures and interpretations that are potentially present. Moreover,
with increasing model structure choices, data acquisition and validation become more
critical.
Therefore in this field as in any application of mathematical modelling there is a
trade-off between model details and transparency of model analysis and conclusions.
In line with the changing perspectives on environmental problems – conditioned
by changes in environmental conditions, resource depletion, economic development,
social and organizational structures, technological development and other things –
the need for analysis will change similarly. Thus, since mathematical models can
contribute to this, there will be a constant need for refinement of existing models and
development of new ones.
Such characteristics are well known from other fields of research. As just one
related example we may mention that tradition of economic modelling which deals
with computable general equilibrium (CGE) or applied general equilibrium (AGE)
models. Such models are not specifically aimed at environmental problems but deal
more broadly with economic problems. CGE models may be seen as datawise enriched
versions of models that are traditionally solved analytically, and they are therefore also
sometimes characterized as “theory with numbers”. However, with the introduction
of more data and parameters there arises the need to apply computationally more
advanced methods to solve the models. See, e.g., [22,31,80,81] for an introduction to
CGE.
The above reflections on the trade off between transparent and complex models
obviously have validity not only for the CGE models, but are of a more general nature
with applicability also to environmental models.
Acknowledgements
We would like to thank the anonymous reviewers for the work which they put in
preparing the reviews and the comments which helped us to improve an earlier version
of the paper.
References
[1] T.J. Allen, Managing the Flow of Technology. Technology Transfer and the Dissemination of Tech-
nological Information within R & D Organization (MIT Press, Cambridge, 1977).
[2] K.J. Arrow and M. Kurtz, Public Investment, the Rate of Return, and Optimal Fiscal Policy (John
Hopkins Press, Baltimore, 1970).
[3] J. Barro, Goverment spending in a simple model of endogenous growth, J. Political Economy 98
(1990) 103–125.
[4] R.J. Barro and X. Sala-i-Martin, Economic Growth (McGraw-Hill, New York, 1995).
[5] T. Basar and G. Olsder, Dynamic Noncooperrative Game Theory (Academic Press, New York,
1982).
[6] W.J. Baumol and W.E. Oates, The Theory of Environmental Policy (Cambridge University Press,
Cambridge, MA, 1988).
[7] G.S. Becker, K.M. Murphy and R. Tamura, Human capital, fertility, and economic growth, J. Po-
litical Economy 98(5) (1990) 12–37.
[8] K.E. Boulding, The economics of the coming spaceship earth, in: Environmental Quality in a
Growing Economy, ed. H. Jarret (John Hopkins Press, Baltimore, 1966) pp. 3–14.
[9] A.L. Bovenberg and R.A. de Mooij, Environmental levies and distortionary taxation, Amer. Econ.
Review 84(4) (1994) 1085–1089.
[10] G. Chichilnisky, Economic development and efficiency criteria in the satisfaction of basic needs,
Appl. Math. Modeling 1(6) (1977) 290–297.
[11] G. Chichilnisky, An axiomatic approach to sustainable development, Social Choice and Welfare 13
(1996) 231–257.
[12] C.W. Clark, Mathematical Bioeconomics: The Optimal Management of Renewable Resources
(Wiley, New York, 1976).
[13] R.H. Coase, The problem of social cost, J. Law Econ. 3 (1960) 1–44.
[14] J.M. Conrad and C.W. Clark, Natural Resource Economics (Cambridge University Press, New York,
1987).
[15] T.D. Crocker, Structuring of atmosferic pollution control systems, in: The Economics of Air Pollu-
tion, ed. H. Wolozin (Norton, New York, 1966).
[16] M.L. Cropper and W.E. Oates, Environmental economics: a survey, Journal of Economic Literature
30 (1992) 675–740.
[17] R.C. D’Arge and K.C. Kogiku, Economic growth and the environment, Review of Economic Studies
40 (1973) 61–77.
[18] P.S. Dasgupta, The Control of Resources (Basil Blackwell, Oxford, 1982).
[19] P.S. Dasgupta and G.M. Heal, The optimal depletion of exhaustible resources, in: Review of Eco-
nomic Studies (1974) pp. 3–28.
[20] P.S. Dasgupta and G.M. Heal, Natural Resource Economics (Cambridge University Press, Cam-
bridge, 1979).
[21] R. Davison, Optimal depletion of an exhaustible resource with research and development towards
an alternative technology, Review of Economic Studies 45 (1978) 355–367.
[22] P.B. Dixon, B.R. Parmenter, J. Sutton and D.P. Vincent, Notes and Problems in Applied General
Equilibrium Economics (North-Holland, 1992).
[23] C.C. Duarte, Renewable resource market obeying difference equations: stable points, stable cycles,
and chaos, Environmental and Resource Economics 4 (1994) 353–381.
[24] G.C. Evans, The dynamics of monopoly, American Math. Monthly 31 (1924) 77–83.
[25] G. Feichtinger and R.F. Hartl, Optimale Kontrolle ökonomischer Prozesse (De Gruyter, Berlin,
1986).
[26] A.C. Fischer and F.M. Peterson, The environment in economics: a survey, Economic Literature 14
(1976) 1–31.
[27] H. Folmer and C. Howe, Environmental problems and policy in the single european market, Envi-
ronmental and Resource Economics 1(1) (1991) 17–41.
[28] H. Folmer and I. Musu, eds., Environmental cooperation and policy in the single European market,
Environmental and Resource Economics 2(2) (1992).
[29] B.A. Forster, A note on the optimal control of pollution, Journal of Economic Theory 5 (1972)
537–539.
[30] B.A. Forster, On a one state variable optimal control problem – consumption-pollution trade-off,
in: Applications of Control Theory to Economic Analysis, eds. J.D. Pitchford and S.J. Turnovsky
(North-Holland, 1977) pp. 35–56.
[31] V. Ginsburgh and M. Keyzer, Structure of Applied General Equilibrium Models (MIT Press, 1997).
[32] B. Gompertz, On the nature of the function expressive of the law of human mortality, Philos. Trans.
R. Soc. 115 (1825) 513–585.
[33] H. Gordon, Economic theory of a common-property resource: the fishery, Journal of Political
Economy 62 (1954) 124–142.
[34] H. Gordon, A reinterpretation of the pure theory of exhaustion, Journal of Political Economy 75
(1967) 274–286.
[35] D. Gottwald, Die dynamische Theorie der Allokation erschöpfbarer Ressourcen (Vandenhoek and
Ruprecht, Göttingen, 1981).
[36] L.H. Goulder, Environmental taxation in a second-best world, in: The International Yearbook of
Environmental and Resource Economics 1997/1998, eds. H. Folmer and T. Tietenberg (Edward
Elgar, Chelthenham, 1997) pp. 28–54.
[37] H. Halkin, Optimal control for systems described by difference equations, in: Advances in Control
Systems, Vol. I, ed. C.T. Leondes (Academic Press, 1964) pp. 173–196.
[38] H. Hotelling, The economics of exhaustible resources, Journal of Political Economy 39 (1931)
137–175.
[39] S.E. Jørgensen, Fundamentals of Ecological Modelling (Elsevier, Amsterdam, 1986).
[40] M.I. Kamien and N.L. Schwartz, Optimal exhaustible resource depletion with endogenous technical
changes, Review of Economic Studies 45 (1978) 179–196.
[41] M.I. Kamien and N.L. Schwartz, Dynamic Optimization: The Calculus of Variations and Optimal
Control in Economics and Management (North-Holland, New York, 1981).
[42] E. Keeler, M. Spence and R. Zeckhauser, The optimal control of pollution, J. Economic Theory 4
(1972) 19–34.
[43] N. Keyfitz, Introduction to Mathematics of Population (Addison-Wesley, Reading, MA, 1968).
[44] N. Keyfitz, Applied Mathematical Demography (Springer, New York, 1985).
[45] J.V. Krutilla, Conservation reconsidered, American Economic Review 57 (1967) 777–786.
[46] G. Leitmann, Cooperative and Non-Cooperative Many Players Differential Games (Springer,
Vienna, 1974).
[47] A.J. Lotka, Elements of Physical Biology (Williams and Wilkins, Baltimore, 1925).
[48] R.E. Lucas, On the mechanics of economic development, J. Monetary Development 22 (1988) 3–42.
[49] M. Luptacik and U. Schubert, Optimale investitionspolitik unter berücksichtigung der umwelt:
Eine anwendung der kontrolltheorie, in: Operations Research Verfahren, Vol. 35, eds. W. Orttli and
F. Steffens (Athenäum/Hain, Meisenheim, 1979) pp. 271–282.
[50] M. Luptacik and U. Schubert, Optimal investment policy in productive capacity and pollution
abatement processes in a growing economy, in: Optimal Control Theory and Economic Analysis,
ed. G. Feichtinger (North-Holland, Amsterdam, 1982) pp. 231–243.
[51] K.G. Mäler, Environmental Economics (John Hopkins Press, Baltimore, 1974).
[52] K.G. Mäler, The acid rain game, in: Valuation Methods and Policy Making in Environmental
Economics, eds. H. Folmer and E. van Ierland (Elsevier, Amsterdam, 1989) pp. 221–252.
[53] K.G. Mäler, International environmental problems, Oxford Review of Economic Policy 6 (1990)
80–108.
[54] T.R. Malthus, Principle of Population as it affects the Future Improvement of Society, with Remarks
on the Speculations of Mr. Godwin, M. Condorcet and other Writers. Reprinted as: Population:
The first Essay (The University of Michigan Press, Ann Arbor, 1959). Original edition 1798.
[55] A.S. Manne and R.G. Richels, Buying Greenhouse Insurance (MIT Press, Cambridge, MA, 1992).
[56] A.P. Martins, Human capital and economic growth, Ph.D. thesis, Columbia University (1987).
[57] R.M. May, Biological populations obeying difference equations: stable points, stable cycles, and
chaos, J. Theor. Biol. 51 (1975) 511–524.
[58] D.H. Meadows, D.L. Meadows, J. Randers and W.W. Behrens III, The Limit to Growth (Universe
Books, New York, 1972).
[59] A. Mehlmann, Applied Differential Games (Plenum Press, New York, 1988).
[60] M. Mesarovic and E. Pestel, Mankind at the Turning Point (Dutton, New York, 1974).
[61] J.S. Mill, Principles of Political Economy (Augustus M. Kelly, New York) reprint of sixth edition,
1965 edition (1848).
[62] S.R. Milliman and R. Prince, Firm incentives to promote technological change in polution control,
Journal of Environmental Economics and Management 77 (1989) 247–265.
[63] J. Monod, Recherches sur la Crossance de Culture Bactériennes (Hermann, Paris, 1942).
[64] J. Monod, La technique de culture continue: theorie et applications, Annals Institute Pasteur 79
(1950) 390–410.
[65] W.D. Montgomery, Markets in licenses and efficient pollution control programs, Journal of Eco-
nomic Theory 5 (1972) 395–418.
[66] W.D. Nordhaus, Managing the Global Commons. The Economics of Climate Change (MIT Press,
Oxford, MA, 1994).
[67] E.C. Pielou, Mathematical Ecology (Wiley, New York, 1977).
[68] A.C. Pigou, The Economics of Welfare (Macmillan, London, 1932).
[69] F. van der Ploeg and C. Withagen, Pollution control and the Ramsey problem, Environmental and
Resource Economics 1 (1991) 97–118.
[70] F. van der Ploeg and A.J. de Zeeuw, International aspects of pollution control, Environmental and
Resource Economics 2(2) (1992) 117–139.
[71] C. Plourde, A model of waste accumulation and disposal, Canadian J. Econ. 5 (1972) 119–125.
[72] F.P. Ramsey, A mathematical theory of saving, Economic J. 38 (1928) 543–559.
[73] J. Rawls, A Theory of Justice (Bellknap Press, Cambridge, MA, 1971).
[74] S. Rebelo, Long run policy analysis and long run growth, J. Political Economy 99 (1991) 500–521.
[75] E.G. Roberts, The Dynamics of Research & Development (Harper & Row, 1964).
[76] P.M. Romer, Increasing returns and long run growth, J. Political Economy 94 (1986) 1002–1037.
[77] M.B. Schaefer, Some considerations of population dynamics and economics in relation to the man-
agement of marine fisheries, J. Fisheries Res. Board of Canada 14 (1957) 669–681.
[78] W.D. Schultze and R.C. D’Arge, The Coase proposition, information constraints, and long-run
equilibrium, American Economic Review 64 (1974) 763–772.
[79] A. Seierstad and K. Sydsæter, Optimal Control Theory with Economic Applications (North-Holland,
Amsterdam, 1987).
[80] J. Showen and J. Whalley, Applied general-equilibrium models of taxation and international trade:
an introduction and survey, Journal of Economic Literature 22 (1984) 1007–1051.
[81] J. Showen and J. Whalley, Applying General Equilibrium (Cambridge University Press, 1992).
[82] H. Siebert, Ökonomische Theorie natürlicher Ressourcen (Mohr, Tübingen, 1983).
[83] H. Siebert, Economics of the Environment (Springer, Heidelberg, 1998).
[84] V.L. Smith, Dynamic of waste accumulation: disposal versus recycling, Quart. J. of Econ. 76 (1972)
600–616.
[85] V.L. Smith, Control theory applied to natural and environmental resources, Journal of Environmental
Economics and Management 4 (1977) 1–24.
[86] R.M. Solow, Intergenerational equity and exhaustible resources, in: Review of Economic Studies
(1974) pp. 29–45.
[87] J. Stiglitz, Growth with exhaustible resources: efficient and optimal growth paths, in: Review of
Economic Studies (1974) pp. 123–137.
[88] J. Stiglitz, Growth with exhaustible resources: the competitive economy, in: Review of Economic
Studies (1974) pp. 139–152.
[89] O. Tahvonen, On the dynamics of renewable resource harvesting and pollution control, Environ-
mental and Resource Economics 1 (1991) 97–117.
[90] O. Tahvonen, Dynamics of pollution control when damage is sensitive to the rate of pollution
accumulation, Environmental and Resource Economics 5 (1995) 9–27.
[91] R. Tamura, Income convergence in an endogenous growth model, J. Political Economy 99(3) (1991)
522–540.
[92] P.F. Verhulst, Notice sur la loi que la population suit dans son accroissement, Correspondance
Mathématique et Physique 10 (1838) 113–121.
[93] V. Volterra, Leçons sur la Théorie Mathématique de la Lutte pour la Vie (Gauthier-Villars, Paris,
1931).
[94] M. Weinstein and R. Zeckhauser, Use patterns for depletable and recyclable resources, in: Review
of Economic Studies (1974) pp. 67–88.
[95] von Weizacker, Lemmas for a theory of approximate optimal growth, Rev. Econ. Studies 34 (1967)
143–151.
[96] C. Withagen, Pollution, abatement and balanced growth, Environmental and Resource Economics
5 (1995) 1–8.
[97] A. Xepapadeas, Advanced Principles in Environmental Policy (Edward Edgar, Cheltenham, 1997).
[98] T. Żylicz, Improving the environment through permit trading: the limits to a market approach,
in: International Environmental Economics Theories, Models and Applications to Climate Change,
International Trade and Acidification, ed. E.C. van Ierland (Elsevier, 1994) pp. 283–306.

A Review of Mathematical Models in Economic Environmental Problems

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

A Review of Mathematical Models in Economic Environmental Problems

Hochgeladen von

Copyright:

Verfügbare Formate

Annals of Operations Research 97 (2000) 165–201 165

A review of mathematical models in economic

Usually there is a tradeoff between economic development and preservation of

 J.C. Baltzer AG, Science Publishers

When we talk on preservation of natural environments, we usually have to take

2.1. A prototype basic model

We start with the following prototype optimization model

maximize W(C, Q, P ) (1)

Ṗ (t) = J(K, L, Q, P , A), (2)

where W is a welfare function, C is consumption, P is a pollution stock, R represents

P (t) R(t) K(t)

As generally, for x(t) = X(t)/L(t)

ṗ(t) = −s(c, p)p(t) + j(k, c, q, a), (14)

This transformation is used if growth of labour has to be considered, and is particularly

2.2.1. Criterion function structure

Pointing to a drawback of this criterion let us assume that U (C1 , Q1 , P1 ) >

2.2.2. Defining the utility function

The following mathematical model of the capital accumulation is due to Ram-

Y (t) = C(t) + I(t) + A(t). (27)

Inserting I(t) from above in (25) we get

K̇(t) = −αK(t) + eθt F (K, L, R) − C(t) − A(t). (28)

This limit function is called the Cobb–Douglas production function

analysis of relatively small subset of factors deciding on development of the economy.

The problem of natural resources exhaustion was a subject of interest already in

2.4.1. Nonrenewable resources

Ṙ(t) = −Q(t), (38)

Under competition a so called Hotelling rule is often applied

Knowing Q, the equations may be solved for π(0) and T .

where d is a discovery-rate function depending on E(t) – the exploratory effect, which

Research and development. Another possibility to enhance the time of using an

and integrating the second (nested) integral in (54) by parts we get

2.4.2. Renewable resources

In the case of nonbiological resource often stochasticity of the supplementary

2.7. Environment regulation

Figure 1. Concave and nonconcave production functions.

will end, under assumption of concavity of F , with the emission E = E S required by

2.7.2. Regulation policies

2.7.3. International aspects of pollution control

3. One state variable pollution problems

3.1. Assumptions and problem formulation

The utility function is here expressed as a smooth (twice continuously differ-

Thus the problem can be formulated as follows

Notice that because of the assumptions taken

Then the necessary (maximum principle) optimality conditions are

and additionally (97), (98).

3.2. Some solution properties

From (106) we have λ∗ 6= 0. When λ∗ > 0, then the maximum of H(C, e P ∗ , λ∗ )

Finally, differentiating (112) with respect to P yields

Let us now consider the curve Ṗ = 0. From (97) we have

UCC C 0 (λ∗ ) + UCP P 0 (λ∗ ) + Z 0 (C) + λ∗ Z 00 (C)C 0 (λ∗ ) = 0 (118)

which, after some small algebraic manipulations, gives

lim UP (C, P ) = 0 for any C. (121)

Thus, if C ∗ is in the interior of the interval [0, K] (where µ1 = µ2 = 0), then

Z 0 (C) = E 0 (C) + A0 (K − C) (123)

Now, UC must be bounded at K because the function U is increasing and concave

However, we cannot take a similar assumption at C = 0, because Z is increasing

With the above from (103) either λ∗ or µ2 must be infinite. So it is impossible to

3.2.4. Additional remarks

A possible shape of the Z(C) curve is depicted at figure 3.

Figure 3. A possible shape of the function Z(C).

4. Two- or more-dimensional state

5. More detailed modelling