Professional Documents
Culture Documents
Optimal Control Problems Arising in Forest Management: Alexander J. Zaslavski
Optimal Control Problems Arising in Forest Management: Alexander J. Zaslavski
Alexander J. Zaslavski
Optimal Control
Problems
Arising in Forest
Management
123
SpringerBriefs in Optimization
Series Editors
Sergiy Butenko
Mirjam Dür
Panos M. Pardalos
János D. Pintér
Stephen M. Robinson
Tamás Terlaky
My T. Thai
123
Alexander J. Zaslavski
Department of Mathematics
The Technion – Israel Institute of Techn
Rishon LeZion, Israel
This Springer imprint is published by the registered company Springer Nature Switzerland AG.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
The growing importance of optimal control has been recognized in recent years.
This is due not only to impressive theoretical developments but also because of
numerous applications to engineering, economics, life sciences, etc. This book
is devoted to the study of a class of optimal control problems arising in forest
management. The forest management problem is an important and interesting topic
in mathematical economics that was studied by many researchers including Nobel
laureate P.A. Samuelson [68]. As usual, for this problem, the existence of optimal
solutions over infinite horizon and the structure of solutions on finite intervals are
under consideration. In our books [84, 86], we study a class of discrete-time optimal
control problems which describe many models of economic dynamics except for
the model of forest management. This happens because some assumptions posed in
[84, 86], which are true for many models of economic dynamics, do not hold for
the model of forest management. By this reason, the forest management problem
is not a particular case of general models of economic dynamics and is studied
separately in the literature. In this book, we study the forest management problem
using the approach introduced and employed in our research [80, 81, 83]. Namely,
we analyze a class of optimal control problems which contains, as a particular case,
the forest management problem. For this class of problems, we show the existence
of optimal solutions over infinite horizon and study the structure of approximate
solutions on finite intervals and their turnpike properties, the stability of the turnpike
phenomenon, and the structure of approximate solutions on finite intervals in the
regions close to the endpoints.
In Chap. 1, we provide some preliminary knowledge on turnpike properties. The
forest management problem is discussed in Chap. 2, which also contains existence
results for infinite horizon problems. In Chap. 3, we establish the turnpike properties
of approximate solutions. Chapter 4 contains generic turnpike results. We consider
a class of optimal control problems which is identified with a complete metric space
of objective functions and show the existence of a Gδ everywhere dense subset of
the metric space, which is a countable intersection of open everywhere dense sets,
such that the turnpike property holds for any of its element. Chapter 5 is devoted to
v
vi Preface
the study of the structure of approximate solutions on finite intervals in the regions
close to the endpoints. In Chap. 6, we again consider the forest management problem
and show that the results of Chaps. 3 and 5 are true for it.
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Convex Discrete-Time Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The Turnpike Phenomenon. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Infinite Horizon Optimal Control Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 The Forest Management Problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Infinite Horizon Problems Without Discounting . . . . . . . . . . . . . . . . . . . . . 16
2.3 Auxiliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Proofs of Theorems 2.6 and 2.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5 Proof of Theorem 2.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6 Proof of Theorem 2.10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7 Infinite Horizon Problems with Discounting . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.8 An Auxiliary Result for Theorem 2.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.9 Proof of Theorem 2.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.10 An Application to the Forest Management Problem . . . . . . . . . . . . . . . . . 30
3 Turnpike Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1 Preliminaries and Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Auxiliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3 Proof of Theorem 3.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Proof of Theorem 3.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.5 Proof of Theorem 3.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.6 Proof of Theorem 3.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7 Proof of Theorem 3.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.8 Stability Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.9 Agreeable Programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4 Generic Turnpike Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Equivalence of the Turnpike Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Generic Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
vii
viii Contents
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Chapter 1
Introduction
The study of optimal control problems and variational problems defined on infinite
intervals and on sufficiently large intervals has been a rapidly growing area of
research [9, 10, 16, 18, 24, 37, 43, 48, 57, 59, 66, 76, 84–86] which has various
applications in engineering [1, 44, 87], in models of economic growth [2, 5, 14, 20,
23, 28, 38–40, 47, 52–55, 60, 61, 63, 65, 67, 68, 70, 77, 80, 81, 83], in the game theory
[29, 32, 42, 74, 82], in infinite discrete models of solid-state physics related to dis-
locations in one-dimensional crystals [6, 71], and in the theory of thermodynamical
equilibrium for materials [19, 45, 50, 51]. Discrete-time problems were considered
in [7, 8, 13, 21, 26, 30, 33, 72, 73, 78, 79] while continuous-time problems were
studied in [3, 4, 11, 12, 15, 17, 22, 25, 27, 31, 41, 46, 49, 56, 58, 62, 69, 75].
In this chapter we discuss turnpike properties for a class of simple convex
dynamic optimization problems.
Let R n be the n-dimensional Euclidean space with the inner product ·, · which
induces the norm
n 1/2
|x| = xi2 , x = (x1 , . . . , xn ) ∈ R n .
i=1
−1
T
v(xi , xi+1 ) → min, (P0 )
i=0
We intend to study the behavior of solutions of the problem (P0 ) when the points
y, z and the real number T vary and T is sufficiently large. Namely, we are interested
to study a turnpike property of solutions of (P0 ) which is independent of the length
of the interval T , for all sufficiently large intervals. To have this property means,
roughly speaking, that solutions of the optimal control problems are determined
mainly by the objective function v and are essentially independent of T , y, and z.
Turnpike properties are well known in mathematical economics. The term was first
coined by Samuelson in 1948 (see [67]) where he showed that an efficient expanding
economy would spend most of the time in the vicinity of a balanced equilibrium path
(also called von Neumann path). This property was further investigated for optimal
trajectories of models of economic dynamics (see, for example, [47, 53, 65] and the
references mentioned there). Many turnpike results are collected in [76, 84, 86].
In order to meet our goal we consider the auxiliary optimization problem
It follows from the strict convexity of v and (1.1) that the problem (P1 ) has a unique
solution x̄. Let
where l1 , l2 ∈ R n . Since x̄ is a solution of (P1 ) it follows from (1.2) that for each
h ∈ Rn,
Thus
l2 = −l1 and
Since the functions v and L are both strictly convex it follows from (1.4) that
and
lim L(yi , zi ) = 0,
i→∞
then
holds and by (1.7) (y, z) = (x̄, x̄). This implies that (x̄, x̄) = limi→∞ (yi , zi ).
Thus the property (C) holds, as claimed.
4 1 Introduction
Relation (1.8) implies that the problems (P0 ) and (P2 ) are equivalent. Namely,
{xi }Ti=0 ⊂ R n is a solution of the problem (P0 ) if and only if it is a solution of
the problem (P2 ).
Let T be a natural number and Δ ≥ 0. A sequence {xi }Ti=0 ⊂ R n is called (Δ)-
optimal if for any sequence {xi
}Ti=0 ⊂ R n satisfying xi = xi
, i = 0, T the inequality
−1
T −1
T
v(xi , xi+1 ) ≤ v(xi
, xi+1
)+Δ
i=0 i=0
−1
T
M1 = L(xi
, xi+1
)
i=0
and
T −1
M2 = inf L(xi , xi+1 ) : {xi }Ti=0 ⊂ R , x0 = z, xT = y .
n
(1.9)
i=0
1.1 Convex Discrete-Time Problems 5
Clearly,
0 ≤ M2 ≤ M1 .
M2 < M1 . (1.10)
(k)
There exists a sequence {xi }Ti=0 ⊂ R n , k = 1, 2, . . . such that for any natural
number k,
(k) (k)
x0 = z, xT = y (1.11)
and
−1
T
(k) (k)
lim L(xi , xi+1 ) = M2 . (1.12)
k→∞
i=0
−1
T
(k) (k)
L(xi , xi+1 ) < M1 for all integers k ≥ 1. (1.13)
i=0
(k)
xi = lim xi . (1.15)
k→∞
x0 = z,
xT = y. (1.16)
−1
T
xi ,
L( xi+1 ) = M2 .
i=0
6 1 Introduction
Together with (1.16) and (1.9) this implies that { xi }Ti=0 is a solution of the
problem (P2 ). This completes the proof of Proposition 1.1.
Denote by Card (A) the cardinality of a set A.
The following result establishes a turnpike property for approximate solutions of
the problem (P0 ).
Proposition 1.2 Let M1 , M2 , be positive numbers. Then there exists a natural
number k0 such that for each integer T > 1 and each (M1 )-optimal sequence
{xi }Ti=0 ⊂ R n satisfying
Proof By condition (C) there is δ > 0 such that for each (y, z) ∈ R n ×R n satisfying
L(y, z) ≤ δ, (1.18)
we have
Set
Assume that an integer T > 1 and that an (M1 )-optimal sequence {xi }Ti=0 ⊂ R n
satisfies (1.17). Set
y0 = x0 , yT = xT , yi = x̄, i = 1, . . . , T − 1. (1.22)
Since the sequence {xi }Ti=0 is (M1 )-optimal it follows from (1.22) that
−1
T −1
T
v(xi , xi+1 ) ≤ v(yi , yi+1 ) + M1 .
i=0 i=0
1.1 Convex Discrete-Time Problems 7
−1
T −1
T
L(xi , xi+1 ) ≤ L(yi , yi+1 ) + M1 = L(x0 , x̄) + L(x̄, xT ) + M1 .
i=0 i=0
−1
T
L(xi , xi+1 ) ≤ M1 + 2M3 .
i=0
It follows from the choice of δ (see (1.18) and (1.19)), (1.21), and the inequality
above that
−1
T
≤ δ −1 L(xi , xi+1 ) ≤ δ −1 (M1 + 2M3 ) ≤ k0 .
i=0
It is easy now to see that the optimal solution {xi }Ti=0 of the problem (P0 ) spends
most of the time in an -neighborhood of x̄. By Proposition 1.3 the number of all
integers i ∈ {0, . . . , T − 1}, such that xi does not belong to this -neighborhood,
does not exceed the constant k0 which depends only on M, and does not depend
on T . Following the tradition, the point x̄ is called the turnpike. Moreover we can
show that the set
is contained in the union of two intervals [0, k1 ]∪[T −k1 , T ], where k1 is a constant
depending only on M, .
8 1 Introduction
In the previous section we proved the turnpike result for rather simple class of
discrete-time problems. The problems of this class are unconstrained and their
objective functions are convex and differentiable. It should be mentioned that,
using the methods of convex analysis [64], this result can be extended to the class
of objective functions which are merely convex. In the turnpike theory and its
applications to the mathematical economics our goal is to establish the turnpike
property for constrained optimal control problems. In particular, in this book and in
[84, 86] we study the structure of approximate solutions of an autonomous discrete-
time control system with a compact metric space of states X. This control system
is described by a nonempty closed set Ω ⊂ X × X which determines a class of
admissible trajectories (programs) and a bounded upper semicontinuous function
v : Ω → R 1 which determines an optimality criterion. We study the problem
−1
T
T −1
v(xi , xi+1 ) → max, {(xi , xi+1 )}i=0 ⊂ Ω, x0 = z, xT = y, (P )
i=0
In this book we study a class of optimal control problems arising in the forest
management. The forest management problem is an important and interesting topic
in mathematical economics studies by many researchers. We consider the discrete
time model for the optimal management of a forest of total area S occupied by k
species I = {1, . . . , k} with maturity ages of n1 , . . . , nk years respectively. This
model and its versions were studied in [20, 38–40, 54, 55, 61, 68, 80, 81, 83]. Here
the optimal harvesting of a mixed forest composed of multiple species is studied,
each one having a different maturity age, where only mature trees can be harvested.
As usual, for this problem the existence of optimal solutions over infinite horizon
and the structure of solutions on finite intervals are under consideration. In our books
[84, 86] we study a class of discrete-time optimal control problems which describe
many models of economic dynamics except of the model of the forest management.
This happens because some assumptions posed in [84, 86], which are true for many
models of economic dynamics, do not hold for the model of the forest management.
Namely, in [84, 86] we studied optimal control problems of type (P ) for which the
turnpike property holds with the turnpike x̄ such that (x̄, x̄) is an interior point of the
set Ω ⊂ X × X. This assumption, which plays an important role in [84, 86], holds
for many models of economic dynamics except of the forest management problem
considered in this book.
By this reason, the forest management problem is not a particular case of general
models of economic dynamics and is studied separately in the literature. In this
book we study the forest management problem using the approach introduced and
employed in our research [80, 81, 83]. Namely, we analyze a class of optimal control
problems which contains, as a particular case, the forest management problem. For
this class of problems we show the existence of optimal solutions over infinite
horizon, study the structure of approximate solutions on finite intervals and their
turnpike properties, the stability of the turnpike phenomenon, and the structure of
approximate solutions on finite intervals in the regions close to the end points.
Chapter 2
Infinite Horizon Optimal Control
Problems
In this chapter we present the forest management problem and study some properties
of its solutions. This problem is a particular case of a general optimal control
problem which is also introduced in this chapter. The corresponding infinite horizon
problems without discounting as well as with discounting are considered. The
results of the chapter were obtained from [80, 81].
We consider a discrete time model for the optimal management of a forest of total
area S occupied by k species I = {1, . . . , k} with maturity ages of n1 , . . . , nk years
respectively. This model and its versions were studied in [20, 38–40, 54, 55, 61,
68, 80, 81, 83]. Here the optimal harvesting of a mixed forest composed of multiple
species is studied, each one having a different maturity age, where only mature trees
can be harvested.
j
For each period t = 0, 1, . . . we denote xi (t) ≥ 0 the area covered by trees of
species i that are j years old with j = 1, . . . , ni and x̄i (t) ≥ 0 the area occupied
by over-mature trees (older than ni ). We must decide how much land vi (t) ≥ 0 to
harvest and how to reallocate this land to new seedlings.
Assuming that only mature trees can be harvested we must have
and then the area not harvested in that period will comprise the over-mature trees at
the next step, namely
(n
i∈I i + 1). Every vector x ∈ R is represented as x = (x1 , . . . , xk ), where
ni ni +1 +1
xi = (xi , . . . , xi , xi
1 )∈R n i for all integers i = 1, . . . , k.
Denote by Δ the set of all x ∈ R+ N such that
⎡ ⎤
n i +1
⎣ xi ⎦ = S.
j
(2.6)
i∈I j =1
where for i = 1, . . . , k,
Define
k
Δ0 = v ∈ k
R+ : vi ≤ S .
i=1
k
wt (V (x, y)) = W (i) (vi (x, y)), (x, y) ∈ Ω
i=1
n̄ = max{ni : i ∈ I }. (2.9)
Our model has an important property established by the following result. Its proof
is given in [80].
+n̄+1
Proposition 2.1 Let x, y ∈ Δ. Then there exists a program {x(t)}N
t=0 such that
x(0) = x and x(N + n̄ + 1) = y.
14 2 Infinite Horizon Optimal Control Problems
j +1 j
xi (t + 1) = xi (t), i ∈ I, j ∈ {1, . . . , ni } \ {ni },
xi1 (t + 1) = 0, i ∈ I,
j
xi (N ) = 0, i ∈ I, j = 1, . . . , ni ,
xini +1 (N ) = S. (2.11)
i∈I
Is = {i ∈ I : ni = s}. (2.12)
xi1 (N + 1) = 0, i ∈ I \ In̄ .
j
xi (N + 1) = 0. (2.14)
such that
ui = xi1 (N + 1). (2.16)
i∈I i∈In̄
Put
j
then xi (N + q) = 0 for all integers j satisfying 1 ≤ j ≤ ni ;
(P2) If an integer s satisfies n̄ ≥ s > n̄ − q and i ∈ Is , then
p n +1+p−(q+s−n̄)
xi (N + q) = yi i , p = 1, . . . , q + s − n̄, (2.18)
p
xi (N +q) = 0 for all integers p satisfying q+s−n̄ < p ≤ ni . (2.19)
xi1 (N + q + 1) = yini +1 , xi (N + q + 1) = 0
p
(2.21)
p+1 p n +1+p−(q+s−n̄)
xi (N + q + 1) = xi (N + q) = yi i , p = 1, . . . , q + s − n̄,
(2.22)
n +1−(q+s−n̄)
xi1 (N + q + 1) = yi i , (2.23)
p
xi (N + q + 1) = 0 for all integers p satisfying q + 1 + s − n̄ < p ≤ ni . (2.24)
such that
[xini +1 (N + q) − xini +1 (N + q + 1)] = xi1 (N + q + 1). (2.27)
i∈I i∈I
Clearly,
and the assumption made for q holds also for q + 1. Thus by induction we have
+n̄
constructed a program {x(t)}N
t=0 such that (P1) and (P2) hold for all q = 1, . . . , n̄.
Consider the state x(N + n̄). Let i ∈ Is where 1 ≤ s ≤ n̄. By (P1) and (P2),
p p+1
xi (N + n̄) = yi , p = 1, . . . , ni . (2.28)
Set
x(N + n̄ + 1) = y. (2.29)
Let (K, ρ) be a compact metric space and Ω be a nonempty closed subset of K ×K.
A sequence {xt }∞ t=0 ⊂ K is called an (Ω)-program (or a program if the set Ω is
understood) if (xt , xt+1 ) ∈ Ω for all integers t ≥ 0.
Let the integers T1 , T2 satisfy T1 < T2 . A sequence {xt }Tt=T
2
1
⊂ K is called an
(Ω)-program (or a program if the set Ω is understood) if (xt , xt+1 ) ∈ Ω for all
integers t satisfying T1 ≤ t < T2 .
For each integer t ≥ 0, let ut : Ω → R 1 be a bounded upper semicontinuous
function.
For each pair of integers T1 , T2 satisfying 0 ≤ T1 < T2 and each y, z ∈ K we
consider the optimization problems
2.2 Infinite Horizon Problems Without Discounting 17
2 −1
T
ui (xi , xi+1 ) → max,
i=T1
{xi }Ti=T
2
1
⊂ K, xT1 = y, xT2 = z
and
2 −1
T
ui (xi , xi+1 ) → max,
i=T1
{xi }Ti=T
2
1
⊂ K, xT1 = y.
2 −1
T
ut (xt , xt+1 ) = U (z, T1 , T2 ).
t=T1
U (y0 , ỹ0 , T1 , T2 )
xT1 = y0 , xT 2 = ỹ0 ,
2 −1
T
ut (xt , xt+1 ) = U (y0 , ỹ0 , T1 , T2 ).
t=T1
(T1 , T2 ) be finite.
Proposition 2.4 Let the integers T1 , T2 satisfy T1 < T2 and let U
T2
Then there exists a program {xt }t=T1 such that
2 −1
T
(T1 , T2 ).
ut (xt , xt+1 ) = U
t=T1
(A) There exists a natural number L̄ such that for each y, z ∈ K there is a program
{xt }L̄
t=0 such that x0 = y and xL̄ = z.
Note that in [72, 73, 75] the case where L̄ = 1 was studied. In view of
Proposition 2.1, assumption (A) holds for the forest management problem.
Theorem 2.6 There exists M > 0 such that for each x0 ∈ K there exists a program
{x̄t }∞
t=0 such that x̄0 = x0 and that for each pair of integers T1 , T2 ≥ 0 satisfying
T1 < T2 ,
T2 −1
ut (x̄t , x̄t+1 ) − U (T1 , T2 ) ≤ M.
t=T1
−1
T
ut (xt , xt+1 ) ≥ U (x(0), x(T ), 0, T ) − M0 .
t=0
Then there exists M1 > 0 such that for each pair of integers T1 ≥ 0, T2 > T1 ,
T2 −1
(T1 , T2 ) ≤ M1 .
ut (xt , xt+1 ) − U
t=T1
is bounded or
−1
T −1
T
ut (xt , xt+1 ) − ut (x̄t , x̄t+1 ) → −∞ as T → ∞. (2.35)
t=0 t=0
0 −1
T 0 −1
T
ut (xt , xt+1 ) − ut (x̄t , x̄t+1 ) < −Q − M. (2.36)
t=0 t=0
−1
T −1
T 0 −1
T
ut (xt , xt+1 ) − ut (x̄t , x̄t+1 ) = ut (xt , xt+1 )
t=0 t=0 t=0
0 −1
T −1
T −1
T
− ut (x̄t , x̄t+1 ) + ut (xt , xt+1 ) − ut (x̄t , x̄t+1 )
t=0 t=T0 t=T0
−1
T
(T0 , T ) −
< −Q − M + U ut (x̄t , x̄t+1 ) ≤ −Q.
t=T0
Since Q is any positive number we conclude that (2.35) is true. Proposition 2.8 is
proved.
Now assume that ut = u0 , t = 0, 1, . . . . The following result will be proved in
Sect. 2.5.
Theorem 2.9 Let M > 0 be as guaranteed by Theorem 2.6. There exists μ =
(0, p)/p and
limp→∞ U
In view of Proposition 2.1, the results stated in this section are applied to the
forest management problem which is discussed in Sect. 2.1. These results allow us to
use new optimality criterions for the forest management problem when the functions
2.3 Auxiliary Results 21
Set
x0 = y, xT = z, (2.38)
2 −1
T
ut (xt , xt+1 ) ≥ U (y, z, T1 , T2 ) − M1 . (2.39)
t=T1
T1 ≤ τ1 < τ2 ≤ T2
2 −1
τ
(τ1 , τ2 ) − M1 − M.
ut (xt , xt+1 ) ≥ U (2.40)
t=τ1
T1 ≤ τ1 < τ2 ≤ T2 .
2 −1
τ
ut (xt
, xt+1
(τ1 , τ2 ).
)=U (2.41)
t=τ1
1 +L̄
By assumption (A) there exist programs {yt }τt=τ , {yt
}τt=τ
2
such that
1 2 −L̄
yτ
= xτ
, yτ
2 = xτ2 , yτ1 = xτ1 , yτ1 +L̄ = xτ
. (2.43)
2 −L̄ 2 −L̄ 1 +L̄
Set
zt = xt , t = T1 , . . . , τ1 , zt = yt , t = τ1 + 1, . . . , τ1 + L̄,
zt = xt
, t = τ1 + L̄ + 1, . . . , τ2 − L̄,
zt = yt
, t = τ2 − L̄ + 1, . . . , τ2 ,
2 −1
T 2 −1
T
−M1 ≤ ut (xt , xt+1 ) − ut (zt , zt+1 )
t=T1 t=T1
2 −1
τ 2 −1
τ
= ut (xt , xt+1 ) − ut (zt , zt+1 )
t=τ1 t=τ1
2 −1
τ 2 −1
τ
≤ ut (xt , xt+1 ) − ut (xt
, xt+1
)
t=τ1 t=τ1
+ 2 sup{ut : t = 0, 1, . . . }2(L̄ + 1)
2 −1
τ
≤ (τ1 , τ2 ) + M
ut (xt , xt+1 ) − U
t=τ1
and
2 −1
τ
(τ1 , τ2 ) − M1 − M.
ut (xt , xt+1 ) ≥ U
t=τ1
x0 = y,
2 −1
T
ut (xt , xt+1 ) ≥ U (y, T1 , T2 ) − M1 .
t=T1
T1 ≤ τ1 < τ2 ≤ T2 ,
2 −1
τ
(τ1 , τ2 ) − M1 − M.
ut (xt , xt+1 ) ≥ U
t=τ1
Proof of Theorem 2.6 Let x0 ∈ K. By Proposition 2.2, for each natural number k
there exists a program {xt(k) }kt=0 such that
(k)
k−1
(k) (k)
x0 = x0 , ut (xt , xt+1 ) = U (x0 , 0, k). (2.45)
t=0
2 −1
τ
ut (xt(k) , xt+1
(k) (T1 , T2 ) − M.
)≥U
t=τ1
(kj )
x̄t = lim xt . (2.46)
j →∞
x̄0 = x0 . (2.47)
24 2 Infinite Horizon Optimal Control Problems
2 −1
τ
(τ1 , τ2 ) ≥ −M0 − M.
ut (xt , xt+1 ) − U
t=τ1
Set u = u0 .
Let x0 ∈ K and let {x̄t }∞
t=0 be as guaranteed by Theorem 2.6. Then for each pair
of integers T1 , T2 ≥ 0 satisfying T1 < T2 ,
T2 −1
u(x̄t , x̄t+1 ) − U (T1 , T2 ) ≤ M.
(2.48)
t=T1
Let p be a natural number. We show that for all sufficiently large natural numbers
T,
−1
T
−1
p U (0, p) − T −1 u(x̄t , x̄t+1 ) ≤ 2M/p. (2.50)
t=0
Assume that T ≥ p is a natural number. Then there exist integers q, s such that
q ≥ 1, 0 ≤ s < p, T = pq + s. (2.51)
2.5 Proof of Theorem 2.9 25
= T −1 {u(x̄t , x̄t+1 ) : t is an integer such that pq ≤ t ≤ T − 1}
q−1 (i+1)p−1
+ (T −1
pq)(pq) −1 (0, p)
u(x̄t , x̄t+1 ) − p−1 U
t=0 t=ip
⎡ ⎛ ⎞ ⎤
q−1
(i+1)p−1
= (T −1 pq)(pq)−1 ⎣ ⎝ (0, p)⎠ + q U
u(x̄t , x̄t+1 ) − U (0, p)⎦
i=0 t=ip
− p−1 U(0, p)
+ T −1 u(x̄t , x̄t+1 ) : t is an integer such that pq ≤ t ≤ T − 1 .
(2.52)
−1
T
(0, p) − lim T −1
|p−1 U u(x̄t , x̄t+1 )| ≤ 2M/p. (2.54)
T →∞
t=0
−1
T
lim T −1 (0, p)/p.
u(x̄t , x̄t+1 ) = lim U (2.55)
T →∞ p→∞
t=0
Set
(0, p)/p.
μ = lim U (2.56)
p→∞
(0, p) − μ| ≤ 2M/p.
|p−1 U
Put u = u0 . By Proposition 2.4 and assumption (A), for each integer T > 0 there is
a program {xt(T ) }Tt=0 such that
−1
T
(T )
u(xt
(T ) (0, T ).
, xt+1 ) = U (2.57)
t=0
Set
−1
T −1
T
x (T ) = T −1 , y (T ) = T −1
(T ) (T )
xt xt+1 . (2.58)
t=0 t=0
Clearly,
(0, T ),
(x (T ) , y (T ) ) ∈ Ω, T u(x (T ) , y (T ) ) ≥ U
(0, T ),
u(x (T ) , y (T ) ) ≥ T −1 U
x (T ) − y (T ) ≤ 2T −1 sup{z : z ∈ K}.
2.7 Infinite Horizon Problems with Discounting 27
Clearly,
(0, T ) = μ.
u(x̄, x̄) ≥ lim T −1 U
T →∞
u(z, z) ≤ μ
We consider the optimal control problem of Sect. 2.2 and continue to use the
notation, definitions, and assumptions introduced there.
In this section we suppose that the following equality holds:
By (A) and Proposition 2.2, for each y ∈ K and each natural number T there is
(y,T ) T
a program {xt }t=0 such that
(y,T )
x0 = y, (2.60)
−1
T
(y,T ) (y,T )
ut (xt , xt+1 ) = U (y, 0, T ). (2.61)
t=0
T −1
(y) (y)
ut (xt , xt+1 ) − U (y, 0, T ) ≤ .
t=0
−1
T
(y) (y)
lim sup ut (xt , xt+1 ) − ut (xt , xt+1 ) ≤ 0.
T →∞ t=0
Note that the program {xt }∞t=0 which exists by Corollary 2.14 is called in the
literature as an overtaking optimal program [76, 84, 86].
Example 2.15 Let w : Ω → [0, ∞) be a bounded upper semicontinuous function,
{ρt }∞
t=0 ⊂ (0, 1) satisfy
lim ρt = 0, (2.62)
t→∞
Lemma 2.16 Let > 0. Then there exists a natural number τ such that for each
y ∈ K and each pair of integers T1 ≥ τ and T2 ≥ T1 + L̄,
1 −1
T
(y,T ) (y,T )
ut xt 2 , xt+1 2 ≥ U (y, 0, T1 ) − .
t=0
Proof By (2.59) and (2.63), there exists a natural number τ such that
T1 ≥ τ, T2 ≥ T1 + L̄. (2.65)
1 +L̄
In view of (A), there exists a program {xt }Tt=T 1
such that
(y,T1 ) (y,T2 )
xT1 = xT1 , xT1 +L̄ = xT . (2.66)
1 +L̄
Set
(y,T1 )
xt = xt , t = 0, . . . , T1 − 1, (2.67)
(y,T2 )
xt = xt for all integers t satisfying T1 + L̄ < t ≤ T2 .
x0 = y. (2.68)
2 −1
T 2 −1
T
(y,T2 ) (y,T )
0≤ ut (xt , xt+1 2 ) − ut (xt , xt+1 )
t=0 t=0
+L̄−1
T1 +L̄−1
T1
(y,T2 ) (y,T )
= ut (xt , xt+1 2 ) − ut (xt , xt+1 )
t=0 t=0
1 −1
T 1 −1
T +L̄−1
T1
(y,T ) (y,T ) (y,T ) (y,T )
≤ ut (xt 2 , xt+1 2 ) − ut (xt 1 , xt+1 1 ) + 2 ||ut ||
t=0 t=0 t=T1
1 −1
T
, xt+1 2 ) − U (y, 0, T1 ) + 2L̄((4L̄)−1 )
(y,T2 ) (y,T )
≤ ut (xt
t=0
and
1 −1
T
(y,T2 ) (y,T )
ut (xt , xt+1 2 ) ≥ U (y, 0, T1 ) − .
t=0
Clearly, {xt }∞
y
t=0 is a program for all y ∈ K.
Let > 0 and let a natural number τ be as guaranteed by Lemma 2.16. Assume
that an integer T ≥ τ and y ∈ K. Then for all sufficiently large natural numbers k,
−1
T
(y,Tk ) (y,T )
ut (xt , xt+1 k ) ≥ U (y, 0, T ) − .
t=0
−1
T
(y) (y)
ut (xt , xt+1 ) ≥ U (y, 0, T ) − .
t=0
We consider the forest management problem discussed in Sect. 2.1 using the
notation, definitions, and assumptions introduced there.
Recall that for each (x, y) ∈ Ω,
where for i = 1, . . . , k,
and that
k
Δ0 = u ∈ k
R+ : ui ≤ S .
i=1
2.10 An Application to the Forest Management Problem 31
It is easy now to see that all the results of Sect. 2.2 hold for our model with ut =
wt ◦ V , t = 0, 1, . . . .
If
sup{|wt (z)| : z ∈ Δ0 } → 0 as t → ∞,
then all the results of Sect. 2.7 hold for our model with ut = wt ◦ V , t = 0, 1, . . . .
Chapter 3
Turnpike Properties
We continued to study the class of optimal control problems introduced in Sect. 2.2
using the same notation, definitions, and assumptions.
Let (K, ρ) be a compact metric space and let Ω be a nonempty closed subset of
K × K.
A sequence {xt }∞ t=0 ⊂ K is called an (Ω)-program (or a program if Ω is
understood) if (xt , xt+1 ) ∈ Ω for all integers t ≥ 0.
Let the integers T1 , T2 satisfy T1 < T2 . A sequence {xt }Tt=T
2
1
⊂ K is called an
(Ω)-program (or a program if Ω is understood) if (xt , xt+1 ) ∈ Ω for all integers t
satisfying T1 ≤ t < T2 .
For each integer t ≥ 0, let ut : Ω → R 1 be a bounded upper semicontinuous
function.
For each pair of integers T1 , T2 satisfying 0 ≤ T1 < T2 and each y, z ∈ K we
consider the optimization problems
2 −1
T
ui (xi , xi+1 ) → max, (P1 )
i=T1
{xi }Ti=T
2
1
⊂ K is a program, xT1 = y, xT2 = z,
2 −1
T
ui (xi , xi+1 ) → max, (P2 )
i=T1
{xi }Ti=T
2
1
⊂ K is a program, xT1 = y
and
2 −1
T
ui (xi , xi+1 ) → max, (P3 )
i=T1
{xi }Ti=T
2
1
⊂ K is a program.
The interest in discrete-time optimal problems of these types stems from the
study of various optimization problems which can be reduced to it, e.g., continuous-
time control systems which are represented by ordinary differential equations
whose cost integrand contains a discounting factor [16], tracking problems in
engineering [44, 87], the study of Frenkel-Kontorova model related to dislocations
in one-dimensional crystals [6, 71], the analysis of a long slender bar of a
polymeric material under tension in [19, 45, 50, 51], and models of economic
growth [84, 86]. See also [72, 73] where these problems were studied with
Ω = K × K.
For each integer t ≥ 0 set
⎧
⎨T
2 −1
U (y, ỹ, T1 , T2 ) = sup ut (xt , xtt+1 ) :
⎩
t=T1
⎫
⎬
{xt }Tt=T
2
is a program and xT1 = y, xT2 = ỹ . (3.4)
1 ⎭
In this chapter we suppose that the following assumption, introduced in Sect. 2.2,
holds.
(A) There exists a natural number L̄ such that for every pair of points y, z ∈ K
there exists a program {xt }L̄
t=0 which satisfies x0 = y and xL̄ = z.
This assumption holds for our forest management problems. Note that all the
results stated in Sect. 2.2 (Theorems 2.6, 2.7, 2.9, and 2.10 and Propositions 2.5 and
2.8) are valid.
We suppose that ut = u0 for all integers t ≥ 0. Set
Let us consider the constant μ defined by Theorem 2.9 and suppose that
is bounded.
A program {xt }∞t=0 is called (u, Ω)-bad ((u)-bad if the set Ω is understood or
T −1
bad if the pair (u, Ω) is understood) if t=0 (u(xt , xt+1 ) − μ) → −∞ as T → ∞.
Theorem 2.9, Proposition 2.8, and Theorem 2.6 imply the following result.
Proposition 3.1 Let {xt }∞
t=0 be a program. Then it is either good or bad. For every
point x0 ∈ K there exists a good program {xt }∞
t=0 .
36 3 Turnpike Properties
, t = 0, . . . , L∗ .
We will prove the following results which describe the structure of approximate
solutions of problems (P1 )–(P3 ).
Theorem 3.2 Let be a positive number. Then there exist an integer τ > 0 and a
positive number δ such that for every natural number T > 2τ and every program
{xt }Tt=0 which satisfy
−1
T
u(xt , xt+1 ) ≥ U (x0 , xT , 0, T ) − δ
t=0
ρ(xt , x̄) ≤ , t = τ1 , . . . , τ2 .
the inequality
holds.
Theorem 3.3 is proved in Sect. 3.4.
Theorem 3.4 Let z ∈ K. Then there exists a program {xt }∞
t=0 such that x0 = z and
that for each program {yt }∞
t=0 satisfying y0 = z,
3.1 Preliminaries and Main Results 37
−1
T
lim sup [u(yt , yt+1 ) − u(xt , xt+1 )] ≤ 0. (3.9)
T →∞ t=0
−1
T
lim sup [u(yt , yt+1 ) − u(xt , xt+1 )] ≤ 0
T →∞ t=0
−1
S+τ
u(xt , xt+1 ) ≥ U (x(S), x(S + τ ), S, S + τ ) − δ
t=S
τ1 ∈ [0, τ ], τ2 ∈ [T − τ, T ],
ρ(xt , x̄) ≤ , t = τ1 , . . . τ2 .
Proof In view of (A) and (3.8), there exists a program {xt }Tt=0 such that
In view of (3.10), for every natural number k and every pair of integers S1 , S2 which
satisfy
0 ≤ S1 < S2 ≤ Tk
we have
2 −1
S
(k) (k) (k) (k)
u(xt , xt+1 ) ≥ U (xS1 , xS2 , 0, S2 − S1 ) − M. (3.12)
t=S1
3.2 Auxiliary Results 39
Extracting a subsequence and re-indexing if necessary we may assume that for every
nonnegative integer t ≥ 0 there is
It follows from Proposition 3.7 and (3.12) that the following property holds:
(P1) For every natural number k and every pair of integers S1 , S2 which satisfy
0 ≤ S1 and S1 + 2L̄ < S2 ≤ Tk , we have
2 −1
S
(k) (k)
u(xt , xt+1 ) ≥ (S2 − S1 )μ − 4(L̄ + 1)u − M.
t=S1
Property (P1) and (3.13) imply that for every pair of integers S1 , S2 ≥ 0 which
satisfy S2 − S1 > 2L̄ we have
2 −1
S
!
u(xt , xt+1 ) ≥ (S2 − S1 )μ − 4 L̄ + 1 u − M. (3.14)
t=S1
Proposition 3.1, (3.14), and property (B1) imply that the program {xt }∞
t=0 is good
and that
lim ρ(xt , x̄) = 0.
t→∞
In view of (3.13), there exists a natural number k such that Tk > S0 + 4 and that
(k)
ρ(xS0 , xS0 ) ≤ /4.
(k)
Combined with (3.15) this implies that ρ(xS0 , x̄) ≤ /2. This contradicts (3.11).
The contradiction we have reached proves Lemma 3.8.
Lemma 3.9 Let be a positive number. Then there exits a positive number δ such
that for every natural number T > 2L∗ and every pair of points y, z ∈ K which
satisfies
ρ(y, x̄), ρ(z, x̄) ≤ δ, (3.16)
the inequality
U (y, z, 0, T ) ≥ T μ −
holds.
40 3 Turnpike Properties
Proof Since the function u is continuous at the point (x̄, x̄) (see assumption (B2))
there exists a number 0 ∈ (0, ) such that for every point (y, z) ∈ Ω which satisfies
the inequality
holds. In view of assumption (B3), there exists a number δ ∈ (0, 0 ) such that for
every pair of points y, z ∈ K which satisfies
Assume that a natural number T > 2L∗ and that a pair of points y, z ∈ K
satisfies (3.16). It follows from (3.16) and the choice of the number δ (see (3.18))
that there exists a program {xt }Tt=0 such that
x0 = y, ρ(xt , x̄) ≤ 0 , t = 0, . . . , L∗ ,
xt = x̄, t = L∗ , . . . , T − L∗ ,
xT = z, ρ(xt , x̄) ≤ 0 , t = T − L∗ , . . . , T . (3.19)
−1
T ∗ −1
L
U (y, z, 0, T ) ≥ u(xt , xt+1 ) = T μ + [u(xt , xt+1 ) − u(x̄, x̄)]
t=0 t=0
−1
T
+ [u(xt , xt+1 ) − u(x̄, x̄)] ≥ T μ
t=T −L∗
−1
T
ρ(x0 , x̄), ρ(xT , x̄) ≤ δ, u(xt , xt+1 ) ≥ U (x0 , xT , 0, T ) − δ,
t=0
3.2 Auxiliary Results 41
the inequality
ρ(xt , x̄) ≤ , t = 0, 1, . . . , T
holds.
Proof Let k be a natural number. In view of assumption (B2), there exists a number
the inequality
holds.
Lemma 3.9 implies that there exists a number δk,1 ∈ (0, k ) such that the
following property holds:
(P2) For every natural number T > 2L∗ and every pair of points y, z ∈ K which
satisfies
we have
U (y, z, 0, T ) ≥ T μ − 2−k .
In view of assumption (B3), there exists a number δk,2 ∈ (0, k ) such that the
following property holds:
(P3) For every pair of points y, z ∈ K which satisfy
x0 = y, xL∗ = z
and
ρ(xt , x̄) ≤ k , t = 0, . . . , L∗ .
42 3 Turnpike Properties
We may assume without loss of generality that the sequence {δi }∞ i=1 is monotone
decreasing.
Assume that the assertion of the lemma does not hold. Then for every integer
k ≥ 1 there exist an integer Tk > 2L∗ and a program {xt(k) }Tt=0
k
which satisfies
(k) (k)
ρ(x0 , x̄), ρ(xTk , x̄) ≤ δk , (3.23)
k −1
T
(k) (k) (k) (k)
u(xt , xt+1 ) ≥ U (x0 , xTk , 0, Tk ) − δk , (3.24)
t=0
(k)
max{ρ(xt , x̄) : t = 0, . . . , Tk } > . (3.25)
Set
k
S0 = 0, Sk = (Ti + L∗ ) (3.26)
i=1
xt∗ = xt , t = 0, 1, . . . , T1 .
(1)
(3.27)
It follows from (3.22), (3.23), and property (P3) that there exists a program
1 +L∗
{xt∗ }Tt=T 1
such that
xS∗i +t = xt
(i+1)
, t = 0, . . . , Ti+1 ,
ρ(xS∗i +Ti+1 +t , x̄) ≤ i+1 , t = 0, . . . , L∗ . (3.30)
Set
xS∗k +t = xt
(k+1)
, t = 1, . . . , Tk+1 . (3.31)
In view of (3.31), (3.23), property (P3), (3.26), and (3.22), there exists a program
{xt∗ }t=S
Sk+1
k +Tk+1
such that
Sk+1 −1 Tk+1 −1
" #
∗ ∗
u(xt , xt+1 ) − u(x̄, x̄) = u(xt(k+1) , xt+1
(k+1)
) − u(x̄, x̄)
t=Sk t=0
Sk+1 −1
" ∗ ∗ #
+ u(xt , xt+1 ) − u(x̄, x̄)
t=Sk +Tk+1
(k+1) (k+1)
≥ U x0 , xTk+1 , 0, Tk+1 − δk+1
k +1
S
k+1
" ∗ ∗ #
u(xt , xt+1 ) − μ ≥ −(L∗ + 2) 2−i+1
t=0 i=0
lim x ∗ = x̄.
t→∞
Combined with (3.30) this contradicts (3.25). The contradiction we have reached
proves Lemma 3.10.
44 3 Turnpike Properties
We may assume that < 1/2. Lemma 3.10 implies that there exists a number
δ ∈ (0, ) such that the following property holds:
(P4) For every natural number T > 2L∗ and every program {xt }Tt=0 satisfying
and
−1
T
u(xt , xt+1 ) ≥ U (x0 , xT , 0, T ) − δ
t=0
Lemma 3.8 implies that there exists an integer T∗ > 2L̄ such that the following
property holds:
∗ −1
T
u(xt , xt+1 ) ≥ U (x0 , xT∗ , 0, T∗ ) − 4,
t=0
Set
τ = T∗ + L ∗ . (3.33)
Assume that a natural number T > 2τ and that a program {xt }Tt=0 satisfies
−1
T
u(xt , xt+1 ) > U (x0 , xT , 0, T ) − δ. (3.34)
t=0
In view of (3.34),
∗ −1
T
u(xt , xt+1 ) ≥ U (x0 , xT∗ , 0, T∗ ) − δ, (3.35)
t=0
−1
T
u(xt , xt+1 ) ≥ U (xT −T∗ , xT , 0, T∗ ) − δ. (3.36)
t=T −T∗
3.4 Proof of Theorem 3.3 45
Property (P5), (3.35), and (3.36) imply that there exist integers τ1 ∈ [0, T∗ ], τ2 ∈
[T − T∗ , T ] such that
It is clear that if ρ(x0 , x̄) ≤ δ, then we may set τ1 = 0, and if ρ(xT , x̄) ≤ δ, then
we may set τ2 = T .
In view of (3.34),
2 −1
τ
u(xt , xt+1 ) ≥ U (xτ1 , xτ2 , τ1 , τ2 ) − δ.
t=τ1
Combined with (3.37), property (P4), (3.33), and the inequality T > 2τ this implies
that
ρ(xt , x̄) ≤ , t = τ1 , . . . , τ2 .
Theorem 3.2 implies that there exist a positive number δ and an integer τ0 > 0 such
that the following property holds:
(P6) For every natural number T > 2τ0 and every program {xt }Tt=0 which satisfies
−1
T
u(xt , xt+1 ) ≥ U (x0 , xT , 0, T ) − δ,
t=0
ρ(xt , x̄) ≤ , t = τ1 , . . . , τ2 .
Set M1 = τ .
Assume that an integer T > τ and that a program {xt }Tt=0 satisfies
−1
T
u(xt , xt+1 ) ≥ U (x0 , xT , 0, T ) − M. (3.39)
t=0
Then there exists an integer Tk+1 > Tk such that Tk+1 ≤ T and that
Tk+1 −1
u(xt , xt+1 ) < U (xTk , xTk +1 , 0, Tk+1 − Tk ) − δ, (3.40)
t=Tk
Tk+1 −2
u(xt , xt+1 ) ≥ U (xTk , xTk+1 −1 , 0, Tk+1 − Tk − 1) − δ. (3.41)
t=Tk
⎫
⎬
k is an integer such that 0 ≤ k < q − 1 ≥ δ|q − 1|,
⎭
|q − 1| ≤ δ −1 M and q ≤ δ −1 M + 1. (3.42)
Proposition 2.2 implies that for every natural number T there exists a program
(T )
{xt }Tt=0 which satisfies
−1
T
x0(T ) = z, u xt(T ) , xt+1
(T )
= U (z, 0, T ). (3.43)
t=0
It follows from Proposition 3.7 and (3.43) that for every integer T > 0 and every
pair of integers S1 , S2 satisfying 0 ≤ S1 < S2 ≤ T and S2 − S1 > 2L̄, we have
2 −1
S
(T ) (T )
u xt , xt+1 ≥ (S2 − S1 )μ − 4(L̄ + 1)(u). (3.44)
t=S1
Theorem 3.2 and (3.43) imply that the following property holds:
(P7) For every positive number there exists an integer τ > 0 such that for every
natural number T > 2τ , we have
(T )
ρ xt , x̄ ≤ , t = τ , . . . , T − τ .
It is clear that there exists a strictly increasing sequence of natural numbers {Tj }∞
j =1
such that for all integers t ≥ 0 there exists
(Tj )
xt = lim xt . (3.45)
j →∞
Evidently, {xt }∞
t=0 is a program satisfying
x0 = z. (3.46)
48 3 Turnpike Properties
In view of (3.44) and (3.45), for all pairs of integers S1 , S2 satisfying S1 < S2 and
S2 − S1 > 2L̄ the inequality
2 −1
S
!
u(xt , xt+1 ) ≥ (S2 − S1 )μ − 4 L̄ + 1 (u) (3.47)
t=S1
is valid. Proposition 3.1, assumption (B1), and (3.47) imply that {xt }∞
t=0 is a good
program and that
lim xt = x̄. (3.48)
t→∞
−1
T
y0 = z, γ := lim sup [u(yt , yt+1 ) − u(xt , xt+1 )] > 0. (3.49)
T →∞ t=0
Assumption (B2) implies that there exists a number ∈ (0, 1) such that for every
point (y1 , y2 ) ∈ Ω which satisfy
the inequality
holds.
Assumption (B3) implies that there exists a number δ ∈ (0, ) such that the
following property holds:
(P8) For every pair of points y1 , y2 ∈ K which satisfies ρ(yi , x̄) ≤ δ, i = 1, 2
there exists a program {zt }L∗
t=0 such that
It follows from (3.48) and (3.50) that there exists an integer S0 > 0 such that
1 −1
S
[u(yt , yt+1 ) − u(xt , xt+1 )] > γ /2. (3.52)
t=0
It follows from (3.45) and upper semicontinuity of the function u that there exists
an integer number k > 0 such that
Tk > S1 + 4, (3.53)
(T )
ρ xt k , xt ≤ δ/4, t = 0, . . . , S1 + 4, (3.54)
u xt(Tk ) , xt+1
(Tk )
≤ u(xt , xt+1 ) + γ (16(S1 + 4))−1 , t = 0, . . . , S1 + 4. (3.55)
ρ(x̄t , x̄) ≤ , t = S1 − L∗ , . . . , S1 .
It follows from (3.57), (3.49), (3.43), (3.55), the choice of , (3.51), and (3.52) that
k −1
T k −1
T
(Tk ) (T )
0≤ u(xt , xt+1k ) − u(x̄t , x̄t+1 )
t=0 t=0
1 −1
S 1 −1
S 1 −1
S
u(xt , xt+1 ) + γ (16)−1
(T ) (T )
= u(xt k , xt+1k ) − u(x̄t , x̄t+1 ) ≤
t=0 t=0 t=0
S1 −L
∗ −1
− u(yt , yt+1 ) − (u(x̄, x̄) − γ (16L∗ )−1 )(2L∗ )
t=0
1 −1
S
≤ γ (16)−1 + γ /8 + u(xt , xt+1 )
t=0
1 −1
S
− u(yt , yt+1 ) + γ /8 ≤ γ /4 + γ 16−1 − γ /2 < 0.
t=0
Theorem 3.4 implies that there exists an overtaking optimal program {yt }∞
t=0
satisfying
y0 = x0 . (3.59)
Assumption (B2) implies that there exists a number ∈ (0, γ /2) such that for every
point (z1 , z2 ) ∈ Ω which satisfies
the inequality
holds.
In view of assumption (B3), there exists a number δ ∈ (0, ) such that the
following property holds:
It follows from (3.58) and (3.60) that there exists an integer S0 > 4L∗ + 4
such that
In view of (3.61), there exists a natural number S1 > S0 + 2 + 2L∗ such that
1 −1
S
(u(yt , yt+1 ) − u(xt , xt+1 )) > γ /2. (3.64)
t=0
Property (P9) and (3.63) imply that there exists a program {x̃t }St=0
1
such that
x̃S1 = xS1 .
1 −1
S 1 −1
S
0≤ u(xt , xt+1 ) − u(x̃t , x̃t+1 )
t=0 t=0
1 −1
S 1 −1
S
= u(xt , xt+1 ) − u(yt , yt+1 )
t=0 t=0
1 −1
S 1 −1
S
+ u(yt , yt+1 ) − u(x̃t , x̃t+1 )
t=S1 −L∗ t=S1 −L∗
≤ −γ /2 + γ 16−1 L−1
∗ (2L∗ + 2) < 0,
In view of Theorem 3.2, there exist an integer τ0 > 0 and a positive number δ such
that the following property holds:
(P10) For every natural number T > 2τ0 and every program {xt }Tt=0 which
satisfies
−1
T
u(xt , xt+1 ) ≥ U (x0 , xT , 0, T ) − δ,
t=0
52 3 Turnpike Properties
ρ(xt , x̄) ≤ , t = τ1 , . . . , τ2 ,
Set
τ = 3τ0 + 1. (3.66)
Assume that an integer T > 2τ and that a program {xt }Tt=0 for every integer S ∈
{0, . . . , T − τ } satisfies
−1
S+τ
u(xt , xt+1 ) ≥ U (x(S), x(S + τ ), S, S + τ ) − δ. (3.67)
t=S
ρ(xt , x̄) ≤ , t = τ0 , . . . , τ − τ0 ,
ρ(xt , x̄) ≤ , t = T − τ + τ0 , . . . , T − τ0 ,
ρ(xt , x̄) ≤ , t = T − τ + τ0 , . . . , T .
ρ(xt , x̄) ≤ , t = τ − τ0 , . . . , T − τ0 .
S ∈ [τ − τ0 , T − τ0 ]. (3.68)
It follows from (3.66) and (3.68) that there exist integers S1 , S2 ∈ [0, T ] such that
S1 + τ0 ≤ S ≤ S2 − τ0 , S2 − S1 = τ. (3.69)
ρ(xS , x̄) ≤ .
and
⎧
⎨T
2 −1
T2 −1
U {ut }t=T , z 0 = sup ut (xt , xt+1 ) : {xt }Tt=T
2
is a program
1 ⎩ 1
t=T1
⎫
⎬
such that xT1 = z0 .
⎭
(Here we assume that supremum over an empty set is −∞.) It is easy to see that the
following result holds.
Proposition 3.11 Let y0 , ỹ0 ∈ K, T1 < T2 be integers, ut , t = T1 , . . . , T2 − 1 be
T2 −1
bounded and upper semicontinuous functions and let U ({ut }t=T 1
, y0 , ỹ0 ) be finite.
Then there exists a program {xt }Tt=T
2
1
such that
2 −1
T
T2 −1
ut (xt , xt+1 ) = U {ut }t=T 1
, y0 , ỹ0 , xT1 = y0 , xT2 = ỹ0 .
t=T1
Theorem 3.12 Let be a positive number. Then there exist an integer τ > 0 and
a positive number γ such that for each natural number T > 2τ , each sequence
of bounded functions ut : Ω → R 1 , t = 0, . . . , T − 1 satisfying ut − u ≤ γ ,
t = 0, 1, . . . , T − 1 and each program {xt }Tt=0 which satisfies
−1
S+τ
S+τ −1
ut (xt , xt+1 ) ≥ U {ut }t=S xS , xS+τ − γ
t=S
54 3 Turnpike Properties
τ1 ∈ [0, τ ], τ2 ∈ [T − τ, T ],
ρ(xt , x̄) ≤ , t = τ1 , . . . τ2 .
γ = δ(4(τ + 1))−1 .
Now it is not difficult to see that the assertion of Theorem 3.12 holds.
Theorem 3.12 implies the following result.
Theorem 3.13 Let > 0. Then there exist a natural number τ , λ > 1, and γ > 0
such that for each integer T > 2τ , each sequence of bounded functions ut : Ω →
R 1 , t = 0, . . . , T − 1 satisfying ut − u ≤ γ , t = 0, 1, . . . , T − 1, each sequence
T −1
{αi }i=0 ⊂ (0, 1] such that for each i, j ∈ {0, . . . , T − 1} satisfying |i − j | ≤ τ the
inequality αi αj−1 ≤ λ holds and each program {xt }Tt=0 which satisfies
−1
S+τ
S+τ −1
αt ut (xt , xt+1 ) ≥ U {αt ut }t=S , xS , xS+τ − γ αS
t=S
τ1 ∈ [0, τ ], τ2 ∈ [T − τ, T ], (3.70)
ρ(xt , x̄) ≤ , t = τ1 , . . . τ2 .
Moreover,
and
T −1
1, each sequence {αi }i=0 ⊂ (0, 1] such that for each i, j ∈ {0, . . . , T −1} satisfying
|i −j | ≤ τ the inequality αi αj−1 ≤ λ holds and each program {xt }Tt=0 which satisfies
−1
T
T −1
αt ut (xt , xt+1 ) = U {αt ut }t=0 , x0 , xT
t=0
−1
T
T −1
ut (xt , xt+1 ) ≥ U ({ut }t=0 , x0 , xT ) − M
t=0
Proof Theorem 3.12 implies that there exist an integer τ ≥ 1 and a positive number
γ such that the following property holds:
(i) for each integer S > 2τ , each finite sequence of bounded functions ut : Ω →
R 1 , t = 0, . . . , S − 1 satisfying
ut − u ≤ γ , t = 0, . . . , S − 1
S−1
ut (xt , xt+1 ) ≥ U ({ut }S−1
t=0 , x0 , xS ) − γ
t=0
we have
ρ(xt , x̄) ≤ , t = τ, . . . , S − τ.
ut − u ≤ γ , t = 0, . . . , T − 1 (3.74)
−1
T
S−1
ut (xt , xt+1 ) ≥ U {ut }t=0 , x0 , xT − M. (3.75)
t=0
Tk < T − 2τ.
If
−1
T
T −1
ut (xt , xt+1 ) ≥ U {ut }t=Tk
, xTk , xT − γ,
t=Tk
Tk+1 ≥ Tk + 2,
Tk+1 −1
Tk+1 −1
ut (xt , xt+1 ) < U {ut }t=Tk
, xTk , xTk+1 −γ (3.76)
t=Tk
and
Tk+1 −2
Tk+1 −2
ut (xt , xt+1 ) ≥ U {ut }t=Tk
, xTk , xTk+1 −1 − γ . (3.77)
t=Tk
3.8 Stability Results 57
T 0 = 0 < · · · , Tq = T ,
q
where q > 0 is an integer. It follows from (3.75) and the construction of {Ti }i=0
(see (3.76) and (3.77)) that
−1
T
T −1
M ≥ U ({ut }t=0 , x0 , xT ) − ut (xt , xt+1 )
t=0
Tk+1 −1
T −1
≥ {U ({ut }t=T
k+1
k
, xTk , xTk+1 ) − ut (xt , xt+1 ) :
t=Tk
and
q ≤ γ −1 M + 1.
0 ≤ k ≤ q − 1, Tk+1 − Tk > 2τ + 1,
q ≤ γ −1 M + 1,
A program {xt∗ }∞t=0 is called agreeable if for any natural number T0 and any > 0
there exists an integer T > T0 such that for any integer T ≥ T there exists a
program {xt }Tt=0 which satisfies
xt = xt∗ , t = 0, . . . , T0
and
−1
T
u(xt , xt+1 ) ≥ U (x0∗ , 0, T ) − .
t=0
The notion of agreeable programs is well known in the economic literature [34–
36]. In this section we employ a strong version of it.
A program {xt∗ }∞t=0 is called strongly agreeable if for any natural number T0 and
any > 0 there exist an integer T > T0 and δ > 0 such that for each integer
T ≥ T and each finite sequence of bounded functions ut : Ω → R 1 , t = 0, . . . , T
satisfying
ut − u ≤ δ, t = 0 . . . , T − 1
xt = xt∗ , t = 0, . . . , T0
and
−1
T
T −1 ∗
ut (xt , xt+1 ) ≥ U {ut }t=0 , x0 − .
t=0
we have
By (B3) there exists δ2 ∈ (0, δ1 ) such that the following property holds:
(ii) for each z1 , z2 ∈ K satisfying
ρ(zi , x̄) ≤ δ2 , i = 1, 2
ρ(ξt , x̄) ≤ δ1 , t = 0, . . . , L∗ ,
ξ0 = z1 , ξL∗ = z2 .
By Theorem 3.12, there exist a natural number L1 and δ3 ∈ (0, δ2 ) such that
the following property holds:
(iii) for each integer T > 2L1 , each finite sequence of bounded functions ut : Ω →
R 1 , t = 0, . . . , T − 1 satisfying
ut − u ≤ δ3 , t = 0 . . . , T − 1
−1
T
T −1
ut (zt , zt+1 ) ≥ U ({ut }t=0 , z0 , zT ) − δ3
t=0
we have
ρ(zt , x̄) ≤ δ2 , t = L1 , . . . , T − L1 .
Choose an integer
and set
ut − u ≤ δ, t = 0 . . . , T − 1. (3.81)
y0 = x0 , (3.82)
−1
T
−1
ut (yt , yt+1 ) ≥ U {ut }Tt=0 , x0 − δ. (3.83)
t=0
Property (ii), (3.78), (3.79), and (3.84) imply that there exists a program {zt }Tt=0
such that
zt = yt , t = 0, . . . , T − L∗ − L1 , (3.85)
zt = xt , t = T − L1 , . . . , T . (3.88)
T −L
1 −1 T −L
1 −1
u(xt , xt+1 ) ≥ u(zt , zt+1 )
t=0 t=0
T −L
∗ −L1 −1
≥ u(yt , yt+1 ) + L∗ (u(x̄, x̄)
t=0
− (/8)(L∗ + 1)−1 )
T −L
∗ −L1 −1
≥ u(yt , yt+1 ) + L∗ u(x̄, x̄) − /8. (3.89)
t=0
3.9 Agreeable Programs 61
T −L
∗ −L1 −1
≥ u(yt , yt+1 ) + L∗ u(x̄, x̄) − /8 − L1 u(x̄, x̄) − /8
t=0
T −L
∗ −L1 −1
= u(yt , yt+1 ) − /4. (3.90)
t=0
T −L
∗ −L1 −1 T −L
∗ −L1 −1
ut (xt , xt+1 ) ≥ ut (yt , yt+1 ) − /4 − /8. (3.91)
t=0 t=0
Property (ii), (3.78), (3.79), and (3.84) imply that there exists a program {ξt }Tt=0 such
that
ξt = xt , t = 0, . . . , T − L∗ − L1 , (3.92)
ρ(ξt , x̄) ≤ δ1 , t = T − L1 − L∗ + 1, . . . , T − L1 , (3.93)
ξt = yt , t = T − L1 , . . . , T . (3.94)
−1
T −1
T
ut (ξt , ξt+1 ) ≥ ut (yt , yt+1 ) − 5/8
t=0 t=0
T −1
≥ U ({ut }t=0 , x0 ) − 5/8 − δ
T −1
≥ U ({ut }t=0 , x0 ) − .
4.1 Preliminaries
We continue to study the class of optimal control problems introduced in Sect. 2.2
using the same notation, definitions, and assumptions.
Let (K, ρ) be a compact metric space and let Ω be a nonempty closed subset of
K × K.
We recall that a sequence {xt }∞ t=0 ⊂ K is called an (Ω)-program (or a program
if Ω is understood) if (xt , xt+1 ) ∈ Ω for all integers t ≥ 0.
Let the integers T1 , T2 satisfy T1 < T2 . A sequence {xt }Tt=T
2
1
⊂ K is called an
(Ω)-program (or a program if Ω is understood) if (xt , xt+1 ) ∈ Ω for all integers t
satisfying T1 ≤ t < T2 .
Let u : Ω → R 1 be a bounded upper semicontinuous function. Set
⎧
⎨T
2 −1
U (u, y, ỹ, T1 , T2 ) = sup u(xt , xtt+1 ) :
⎩
t=T1
⎫
⎬
{xt }Tt=T
2
is a program and xT1 = y, xT2 = ỹ ,
1 ⎭
⎧ ⎫
⎨T
2 −1 ⎬
(u, T1 , T2 ) = sup
U u(xt , xtt+1 ) : {xt }Tt=T
2
is a program .
⎩ 1 ⎭
t=T1
In this chapter we suppose that the following assumption, introduced in Sect. 2.2,
holds.
(A) There exists a natural number L̄ such that for each y, z ∈ K there is a program
{xt }L̄
t=0 such that x0 = y and xL̄ = z.
This assumption holds for the important class of forest management problems.
Note that in Chaps. 2 and 3 the function u was fixed and for each pair of integers
T1 , T2 satisfying 0 ≤ T1 < T2 and each y, ỹ ∈ K we used the notation
U (y, T1 , T2 ) = U (u, y, T1 , T2 ),
(T1 , T2 ) = U (u, T1 , T2 ).
U
Note that all the results stated in Sect. 2.2 (Theorems 2.6, 2.7, 2.9, and 2.10 and
Propositions 2.5 and 2.8) hold in our case with ut = u, t = 0, 1, . . . .
The results of this chapter are new.
T −1 ∞
{u(xt , xt+1 ) − μ
t=0 T =1
is bounded.
A program {xt }∞ t=0 is called (u)-bad (or bad if the function u is understood) if
T −1
t=0 (u(xt , xt+1 ) − μ) → −∞ as T → ∞.
By Proposition 3.1, any program is either good or bad and for any x0 ∈ K there
is a good program {xt }∞ t=0 .
We suppose that assumptions (B2) and (B3) introduced in Sect. 3.1 hold.
0, . . . , L∗ .
Property (B1) is known in the literature and is called as the asymptotic turnpike
property. Clearly, if (B1) holds, then all the results of Chap. 3 are true. In particular,
Theorem 3.2, a turnpike result, is true. The next result shows that if the turnpike
property established by Theorem 3.2 holds, then (B1) holds too.
Theorem 4.1 Assume that for each > 0 there exist a natural number τ and δ > 0
such that the following property holds:
(i) for each integer T > 2τ and each program {zt }Tt=0 satisfying
−1
T
u(zt , zt+1 ) ≥ U (z0 , zT , 0, T ) − δ
t=0
the inequality
ρ(zt , x̄) ≤
lim xt = x̄.
t→∞
66 4 Generic Turnpike Properties
Let > 0 and δ > 0 and a natural number τ be such that property (i) holds. We
show that there exists an integer Tδ ≥ 0 such that for each integer T > Tδ ,
−1
T
u(xt , xt+1 ) ≥ U (xTδ , xT , Tδ , T ) − δ. (4.4)
t=Tδ
Assume the contrary. Then there exists a strictly increasing sequence of integers
{Ti }∞
i=0 such that T0 = 0 and for each integer i ≥ 1,
Ti+1 −1
u(xt , xt+1 ) < U (xTi , xTi+1 , Ti , Ti+1 ) − δ. (4.5)
t=Ti
yTi = xTi ,
Ti+1 −1 Ti+1 −1
u(xt , xt+1 ) < u(yt , yt+1 ) − δ.
t=Ti t=Ti
In view of the relation above and (4.3) for each natural number q,
Tq −1
u(yt , yt+1 ) − Tq μ
t=0
Tq −1 Tq −1 Tq −1
= u(yt , yt+1 ) − u(xt , xt+1 ) + u(xt , xt+1 ) − Tq μ
t=0 t=0 t=0
≥ qδ − M → ∞ as q → ∞.
This contradicts Proposition 3.1. The contradiction we have reached proves that
there exists an integer Tδ ≥ 0 such that the following property holds:
(ii) for each integer T > Tδ , (4.4) is true.
4.3 Generic Results 67
Properties (i) and (ii), (4.9), and the choice of δ, τ imply that for each integer
T > Tδ + 2τ , we have
ρ(xt , x̄) ≤ , t = τ + Tδ , . . . , T − τ.
We use the notation, definitions, and assumptions introduced in Sect. 4.1. Denote by
C(Ω) the set of all continuous functions on Ω. Clearly, (C(Ω), · ) is a Banach
space. For each u1 , u2 ∈ C(Ω) set
d(u1 , u2 ) = u1 − u2 .
(v, 0, p)p−1
μ(v) = lim U (4.6)
p→∞
(z, z) ∈ Ω.
It is not difficult to see that M is a closed subset of C(Ω). We consider the complete
metric space (M, d). Suppose that the following assumption holds.
(B4) There exists a natural number L∗ such that for each > 0 there exists δ > 0
such that for each y1 , y2 ∈ K satisfying ρ(y1 , y2 ) ≤ δ there is a program
{xt }L∗
t=0 such that
x0 = y1 , xL∗ = y2
68 4 Generic Turnpike Properties
and
ρ(xt , x̄) ≤ , t = 0, . . . , L∗ .
T −1 ∞
{v(xt , xt+1 ) − μ(v)
t=0 T =1
is bounded.
Theorem 4.2 There exists a set F ⊂ M which is a countable intersection of open
everywhere dense subsets of M such that for each v ∈ F the following properties
hold:
1. there exists a unique point xv ∈ K such that
μ(v) = v(xv , xv ).
lim ρ(xt , xv ) = 0.
t→∞
4.3 Generic Results 69
Proposition 3.1, (4.10), and (4.12) imply that for all natural numbers T ,
−1
T −1
T
−M ≤ v(xt , xt+1 ) − T v(xv , xv ) − r ρ(xt , xv ). (4.13)
t=0 t=0
−1
T
lim ρ(xt , xv ) < ∞.
T →∞
t=0
lim ρ(xt , xv ).
t→∞
−1
T
T −1
ut (xt , xt+1 ) ≥ U {ut }t=0 x0 , xT − γ (v, r, n)
t=0
we have
Define
F = ∩∞
p=1 ∪ {U (v, r, n) : v ∈ M, r ∈ (0, 1], n ≥ p is an integer}. (4.14)
Let
p > 4 −1 . (4.16)
It follows from (4.14) and (4.15) that there exist v ∈ M, r ∈ (0, 1], and an integer
n ≥ p such that
Assume that
ξ1 , ξ2 ∈ K,
It is not difficult to see that for each i = 1, 2 and each integer T > 0,
U (w, ξi , ξi , 0, T ) = T w(ξi , ξi ).
a singleton.
Assume that an integer T > 2τ (v, r, n) and that a program {xt }Tt=0 satisfies
−1
T
w(xt , xt+1 ) ≥ U (w, x0 , xT , 0, T ) − γ (v, r, n).
t=0
4.3 Generic Results 71
Together with property (i), (4.16), (4.17), and (4.19) this implies that for all integers
t = τ (v, r, n), . . . , T − τ (v, r, n),
ρ(xt , xv ) ≤ 1/n
and
Combined with Theorem 4.1 this implies that every (w)-good program converges
to ξ1 . Theorem 4.2 is proved.
Chapter 5
Structure of Solutions in the Regions
Close to the Endpoints
5.1 Preliminaries
We continued to study the class of optimal control problems studied in Chap. 3 (see
Sects. 3.1 and 3.8) using the same notation, definitions, and assumptions.
Let (K, ρ) be a compact metric space and let Ω be a nonempty closed subset of
K × K.
Let u : Ω → R 1 be a bounded upper semicontinuous function.
In this chapter we suppose that the following assumption, introduced in Sect. 2.2,
holds.
(A) There exists a natural number L̄ such that for each y, z ∈ K there is a (Ω)-
program {xt }L̄
t=0 such that x0 = y and xL̄ = z.
We also assume that there is x̄ ∈ K such that
We also assume that the following assumptions (B1)–(B3) introduced in Sect. 3.1
hold.
(B1) For each (u, Ω)-good program {xt }∞ t=0 , limt→∞ xt = x̄.
(B2) u is continuous at (x̄, x̄) with respect to Ω.
(B3) There exists a natural number L∗ such that for each > 0 there exists δ > 0
such that for each y1 , y2 ∈ K satisfying ρ(y1 , x̄), ρ(x̄, y2 ) ≤ δ there is an
(Ω)-program {xt }L t=0 such that x0 = y1 , xL∗ = y2 and ρ(xt , x̄) ≤ , t =
∗
0, . . . , L∗ .
We study the structure of approximate solutions of problems (P1 )–(P3 ) (see
Sect. 3.1).
We define a function π u (x), x ∈ K which plays an important role in our study.
Let x ∈ K. Set
−1
T
π (x) = sup lim sup
u
(u(xt , xt+1 ) − v(x̄, x̄)) :
T →∞ t=0
{xt }∞
t=0 is an (Ω) − program such that x0 = x . (5.2)
In view of Theorems 2.6 and 2.9, there exists Mu > 0 such that
Let x ∈ K. By (5.3),
−1
T
π u (x) = sup lim sup (u(xt , xt+1 ) − u(x̄, x̄)) :
T →∞ t=0
{xt }∞
t=0 is an (u, Ω) − good program such that x0 = x . (5.4)
Denote by P(u, x) the set of all (u, Ω)-overtaking optimal programs {xt }∞ t=0
satisfying x0 = x. By Theorem 3.4, the set P(u, x) is nonempty. Definition (5.2)
implies the following result.
Proposition 5.1 Let T ≥ 1 be an integer and {xt }Tt=0 be a (Ω)-program. Then for
each integer t = 0, . . . , T − 1,
The next result follows from the definition of (u, Ω)-overtaking optimal pro-
grams.
5.1 Preliminaries 75
−1
T
π u (x0 ) = lim sup (u(xt , xt+1 ) − u(x̄, x̄)).
T →∞ t=0
Set
By (B3), there exists δ ∈ (0, 1 ) such that the following property holds:
(i) for each z1 , z2 ∈ K satisfying
ρ(zi , x̄) ≤ δ, i = 1, 2
x0 = z1 , xL∗ = z2
and
ρ(xt , x̄) ≤ 1 , t = 0, . . . , L∗ .
Assume that
Property (i) and (5.8) imply that there exists an (Ω)-program {xt }L∗
t=0 such that
x0 = z1 , xL∗ = z2 , (5.9)
and
π u (z1 ) = π u (x0 )
∗ −1
L
≥ (u(xt , xt+1 ) − u(x̄, x̄)) + π u (xL∗ )
t=0
−1
T
π (x0 ) = lim
u
(u(xt , xt+1 ) − u(x̄, x̄)).
T →∞
t=0
Proof It follows from (B1), Propositions 5.4, 5.5, and Corollary 5.3 that
We show that
π u (x) ≥ lim sup π u x (i) .
i→∞
By Theorem 3.4 and Proposition 5.6, for each integer i ≥ 1, there exists a (u, Ω)-
overtaking optimal program {xt }∞
(i)
t=0 such that
(i)
x0 = x (i) , (5.13)
−1
T
(i) (i)
π u x (i) = lim u(xt , xt+1 ) − u(x̄, x̄) . (5.14)
T →∞
t=0
(i)
xt = lim xt . (5.15)
i→∞
Since {xt(i) }∞
t=0 , i = 1, 2, . . . are (u, Ω)-overtaking optimal programs it follows
from Theorem 3.2 that there exists a natural number L1 such that for all sufficiently
large natural numbers i,
(i)
ρ xt , x̄ ≤ δ for all integers t ≥ L1 . (5.17)
Let T ≥ L1 be an integer. By Corollary 5.3, (5.16), and (5.17), for all integers i ≥ 1,
−1
T
(i) (i)
(u(xt , xt+1 ) − u(x̄, x̄))
t=0
(i) (i) (i)
= π u (x0 ) − π u (xT ) ≥ π u (x0 ) − /2.
78 5 Structure of Solutions in the Regions Close to the Endpoints
In view of the relation above and the upper semicontinuity of u, for all integers
T ≥ L1 ,
−1
T
(u(xt , xt+1 ) − u(x̄, x̄))
t=0
−1
T
≥ lim sup u(xt(i) , xt+1
(i)
) − u(x̄, x̄)
i→∞ t=0
≥ lim sup π u x0(i) − /2.
i→∞
−1
T
π u (x) ≥ lim sup (u(xt , xt+1 ) − u(x̄, x̄)) ≥ lim sup π u x (i) − /2.
T →∞ t=0 i→∞
Then {xt }∞
t=0 is a (u, Ω)-overtaking optimal program.
Proof Proposition 5.1 and (5.18) imply that {xt }∞
t=0 is a (u, Ω)-locally optimal
program. By Theorem 3.5, {xt }∞
t=0 is a (u, Ω)-overtaking optimal program. Propo-
sition 5.8 is proved.
In order to study the structure of solutions of the problems in the regions close to
the end points we introduce the following notation and definitions.
Set
2 −1
T
ūT2 −t+T1 −1 (x̄t , x̄t+1 )
t=T1
2 −1
T
= uT2 −t+T1 −1 (xT2 −t+T1 −1 , xT2 −t+T1 )
t=T1
2 −1
T
= ut (xt , xt+1 ). (5.22)
t=T1
−1
T T
−1
(1) (1) (2) (2)
ut xt , xt+1 ≥ ut xt , xt+1 − M
t=0 t=0
if and only if
−1
T T
−1
(1) (1) (2) (2)
ūT −t−1 x̄t , x̄t+1 ≥ ūT −t−1 xt , xt+1 − M.
t=0 t=0
⎧
⎨T
2 −1
T2 −1
U {ut }t=T , Ω, z0 = sup ut (xt , xt+1 ) : {xt }Tt=T
2
is an (Ω)-program
1 ⎩ 1
t=T1
⎫
⎬
such that xT1 = z0 , (5.24)
⎭
⎧
⎨T
2 −1
{ut }T2 −1 , Ω, z1 = sup
U ut (xt , xt+1 ) : {xt }Tt=T
2
is an (Ω)-program
t=T1 ⎩ 1
t=T1
⎫
⎬
such that xT2 = z1 , (5.25)
⎭
⎧
⎨T
2 −1
T2 −1
U {ut }t=T , Ω, z 0 , z 1 = sup ut (xt , xt+1 ) : {xt }Tt=T
2
is an (Ω)-program
1 ⎩ 1
t=T1
⎫
⎬
such that xT1 = z0 , xT2 = z1 . (5.26)
⎭
⎧
⎨T
2 −1
T2 −1
U {wt }t=T1 , Ω̄, z0 , z1 = sup wt (xt , xt+1 ) : {xt }Tt=T
2
is an (Ω̄)-program
⎩ 1
t=T1
⎫
⎬
such that xT1 = z0 , xT2 = z1 , (5.28)
⎭
5.2 Lagrange Problems 81
⎧
⎨T
2 −1
{wt }T2 −1 , Ω̄, z1 = sup
U wt (xt , xt+1 ) : {xt }Tt=T
2
is an (Ω̄)-program
t=T1 ⎩ 1
t=T1
$
such that xT2 = z1 , (5.29)
⎧ ⎫
⎨T
2 −1 ⎬
T2 −1 T2
U {wt }t=T , Ω̄ = sup w (x , x
t t t+1 ) : {x }
t t=T1 is an ( Ω̄)-program .
1 ⎩ ⎭
t=T1
(5.30)
Proposition 5.9 implies the following result.
Proposition 5.10 Let T ≥ 1 be an integer, M ≥ 0, ut : Ω → R 1 , t = 0, . . . T − 1
be bounded functions and {xt }Tt=0 be an (Ω)-program. Then {x̄t }Tt=0 is an (Ω̄)-
program and the following assertions
hold:
T −1 T −1
if t=0 ut (xt , xt+1 ) ≥ U {ut }t=0 , Ω − M, then
−1
T
T −1
ūT −t−1 (x̄t , x̄t+1 ) ≥ U {ūT −t−1 }t=0 , Ω̄ − M;
t=0
T −1
T −1
if t=0 ut (xt , xt+1 ) ≥ U {ut }t=0 , Ω, x0 , xT − M, then
−1
T
−1
ūT −t−1 (x̄t , x̄t+1 ) ≥ U {ūT −t−1 }Tt=0 , Ω̄, x̄0 , x̄T − M;
t=0
T −1
T −1
if t=0 ut (xt , xt+1 ) ≥ U {ut }t=0 , Ω, x0 − M, then
−1
T
{ūT −t−1 }T −1 , Ω̄, x̄T − M;
ūT −t−1 (x̄t , x̄t+1 ) ≥ U t=0
t=0
T −1
if {ut }T −1 , Ω, xT − M, then
ut (xt , xt+1 ) ≥ U
t=0 t=0
−1
T
T −1
ūT −t−1 (x̄t , x̄t+1 ) ≥ U {ūT −t−1 }t=0 , Ω̄, x̄0 − M.
t=0
Note that the pair (ū, Ω̄) satisfies assumptions (A), (B2), and (B3), and that
μ(ū) = μ(u) = μ.
82 5 Structure of Solutions in the Regions Close to the Endpoints
It follows from Theorems 3.2 and 4.1 and Proposition 5.10 that (B1) also holds for
the pair (ū, Ω̄). Therefore all the results presented above for the pair (u, Ω) are also
true for the pair (ū, Ω̄).
We prove the following results which describe the structure of approximate
solutions of our optimal control problems in the regions close to the endpoints.
Theorem 5.11 Let τ0 ≥ 1 be an integer and > 0. Then there exist δ > 0 and an
integer T0 ≥ τ0 such that for each integer T ≥ T0 , each finite sequence of bounded
functions ut : Ω → R 1 , t = 0, . . . T − 1 satisfying
ut − u ≤ δ, t = 0 . . . , T − 1
−1
T
T −1
ut (xt , xt+1 ) ≥ U {ut }t=0 , Ω, x0 − δ
t=0
π ū (x0∗ ) = sup(π ū ),
ρ(xT −t , xt∗ ) ≤ , t = 0, . . . , τ0 .
Theorem 5.12 Let τ0 ≥ 1 be an integer and > 0. Then there exist δ > 0 and an
integer T0 ≥ τ0 such that for each integer T ≥ T0 , each finite sequence of bounded
functions ut : Ω → R 1 , t = 0, . . . T − 1 satisfying
ut − u ≤ δ, t = 0 . . . , T − 1
−1
T
T −1
ut (xt , xt+1 ) ≥ U {ut }t=0 ,Ω − δ
t=0
Theorem 5.13 Let τ0 ≥ 1 be an integer and > 0. Then there exist δ > 0 and an
integer T0 ≥ τ0 such that for each integer T ≥ T0 , each finite sequence of bounded
functions ut : Ω → R 1 , t = 0, . . . T − 1 satisfying
ut − u ≤ δ, t = 0 . . . , T − 1
−1
T
{ut }T −1 , Ω, xT − δ
ut (xt , xt+1 ) ≥ U t=0
t=0
π u (x0∗ ) = sup(π u ),
ρ(xt , xt∗ ) ≤ , t = 0, . . . , τ0 .
Theorem 5.13 is proved in Sect. 5.4 while Sect. 5.3 contains an auxiliary result.
Theorem 5.11 follows from Theorem 5.13 applied with (ū, Ω̄) and Proposition 5.10.
Theorem 5.12 easily follows from Theorems 5.11 and 5.13.
Lemma 5.14 Let T0 ≥ 1 be an integer and ∈ (0, 1). Then there exists δ ∈ (0, )
such that for each (Ω)-program {xt }Tt=0
0
which satisfies
π u (x0 ) ≥ sup(π u ) − δ,
0 −1
T
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT ) ≥ −δ,
t=0
π u (z0 ) = sup(π u ),
ρ(zt , xt ) ≤ , t = 0, . . . , T0 . (5.31)
Proof Assume that the lemma does not hold. Then there exist a sequence {δk }∞
k=1 ⊂
(k) T0
(0, 1) and a sequence of (Ω)-programs {xt }t=0 , k = 1, 2, . . . such that
lim δk = 0 (5.32)
k→∞
84 5 Structure of Solutions in the Regions Close to the Endpoints
and that for each integer k ≥ 1 and each (u, Ω)-overtaking optimal program {zt }∞
t=0
satisfying (5.31),
(k)
π u x0 ≥ sup(π u ) − δk , (5.33)
0 −1
T
(k) (k) (k) (k)
u xt , xt+1 − u(x̄, x̄) − π u x0 + π u xT0 ≥ −δk (5.34)
t=0
we have
(k)
max ρ zt , xt : t = 0, . . . , T0 > . (5.35)
(k)
xt = lim xt . (5.36)
k→∞
0 −1
T
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT0 )
t=0
⎛ ⎞
0 −1
T
≥ lim sup ⎝ (u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT0 )⎠
(k) (k) (k) (k)
k→∞ t=0
≥ lim sup(−δk ) = 0.
k→∞
Thus
0 −1
T
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT0 ) ≥ 0.
t=0
Together with Proposition 5.1 this implies that for all integers t = 0, . . . , T0 − 1,
Theorem 3.4 implies that there is a (u, Ω)-overtaking optimal program {x̃t }∞
t=0
satisfying
xt = x̃t−T0 . (5.40)
π u (x0 ) = sup(π u ).
(k)
By (5.36), for all sufficiently large natural numbers k, ρ(xt , xt ) ≤ /4, t =
0, . . . , T0 . This contradicts (5.35). The contradiction we have reached proves
Lemma 5.14.
We may assume that < 1. By Lemma 5.14, there exists δ1 ∈ (0, /4) such that the
following property holds:
(i) for each (Ω)-program {yt }τt=0
0
which satisfies
π u (y0 ) ≥ sup(π u ) − δ1 ,
0 −1
τ
(u(yt , yt+1 ) − u(x̄, x̄)) − π u (y0 ) + π u (yτ0 ) ≥ −δ1
t=0
π u (z0 ) = sup(π u ),
ρ(zt , yt ) ≤ , t = 0, . . . , τ0 .
By Propositions 5.4 and 5.5 and (B2), there exists δ2 ∈ (0, δ1 ) such that the
following properties hold:
(ii) for each z ∈ K satisfying ρ(z, x̄) ≤ 2δ2 , we have
we have
|u(x, y) − u(x̄, x̄)| ≤ (δ1 /8)(L∗ + 1)−1 .
By (B3), there exists δ3 ∈ (0, δ2 ) such that the following property holds:
(iv) for each x, y ∈ K satisfying
ξ0 = x, ξL∗ = y
and
ρ(ξt , x̄) ≤ δ2 , t = 0, . . . , L∗ .
ut − u ≤ δ4 , t = 0 . . . , T − 1
−1
T
{ut }T −1 , Ω, zT − δ4
ut (zt , zt+1 ) ≥ U t=0
t=0
we have
ρ(zt , x̄) ≤ δ3 , t = L0 , . . . , T − L0 .
ut − u ≤ δ, t = 0 . . . , T − 1 (5.45)
−1
T
T −1
ut (xt , xt+1 ) ≥ U {ut }t=0 , Ω, xT − δ. (5.46)
t=0
Property (iv), (5.42), (5.44), and (5.47) imply that there exist an (Ω)-program
{x̃t }Tt=0 such that
x̃t = zt , t = 0, . . . , L0 + τ0 + τ1 + 3, (5.48)
x̃t = xt , t = L0 + L∗ + τ0 + τ1 + 3, . . . , T . (5.50)
−1
T
δ≥ (ut (x̃t , x̃t+1 ) − ut (xt , xt+1 ))
t=0
L0 +L∗
+τ0 +τ1 +2
= (ut (x̃t , x̃t+1 ) − ut (xt , xt+1 ))
t=0
L0 +L∗
+τ0 +τ1 +2
≥ (u(x̃t , x̃t+1 ) − u(xt , xt+1 ))
t=0
Properties (ii) and (iii), (5.42), (5.48), and (5.49) imply that
t = L0 + τ0 + τ1 + 3, . . . , L0 + L∗ + τ0 + τ1 + 2.
Corollary 5.3, property (ii), (5.42), (5.43), (5.44), (5.47), (5.48), (5.51), and (5.52)
imply that
L0 +L∗
+τ0 +τ1 +2
δ≥ u(zt , zt+1 ) + L∗ u(x̄, x̄) − L∗ (δ1 /8)(L∗ + 1)−1
t=0
L0 +L∗
+τ0 +τ1 +2
− u(xt , xt+1 ) − 2δ(L0 + L∗ + τ0 + τ1 + 3)
t=0
≥ sup(π u ) − δ1 /8 − δ1 /4 − π u (x0 ) − δ1 /8
⎛ ⎞
L0 +L
∗ +τ1 +2
−⎝ (u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xL0 +L∗ +τ1 +3 )⎠ .
t=0
(5.53)
0 −1
τ
π u (x0 ) + (u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xτ0 )
t=0
≥ sup(π u ) − δ1 ,
π u (x0 ) ≥ sup(π u ) − δ1 , (5.54)
0 −1
τ
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xτ0 ) ≥ −δ1 . (5.55)
t=0
5.5 The First Class of Bolza Problems 89
By (5.54), (5.55), and the property (i), there exists a (u, Ω)-overtaking optimal
program {yt }∞
t=0 such that
π u (y0 ) = sup(π u ),
ρ(yt , xt ) ≤ , t = 0, . . . , τ0 .
sup(h) = sup{h(y) : y ∈ Y }.
dK (h1 , h2 ) = h1 − h2 .
2 −1
T
T −1
ut (xt , xt+1 ) + h(xT2 ) → max, {(xt , xt+1 )}t=0 ⊂ Ω, xT1 = x (PB,1 )
t=T1
and set
⎧
⎨T
2 −1
T2 −1
U h, {ut }t=T , Ω, x = sup ut (xt , xt+1 ) + h(xT2 ) :
1 ⎩
t=T1
⎫
⎬
{xt }Tt=T
2
is an (Ω) − program and xT1 = x .
1 ⎭
In Sect. 5.7 we prove the following result which describes the structure of
approximate solutions of the problems of the type (PB,1 ) in the regions close to
the right endpoints.
Theorem 5.15 Let g ∈ M(K) be upper semicontinuous function, τ0 ≥ 1 be an
integer, and > 0. Then there exist δ > 0 and an integer T0 ≥ τ0 such that for each
integer T ≥ T0 , each h ∈ M(K) satisfying
h − g ≤ δ,
ut − u ≤ δ, t = 0 . . . , T − 1
τ +T
0 −1
τ +T0 −1
ut (xt , xt+1 ) ≥ U {ut }t=τ , Ω, xτ , xτ +T0 − δ
t=τ
−1
T
T −1
h(xT ) + ut (xt , xt+1 ) ≥ U h, {ut }t=T −T0 , Ω, xT −T0 − δ
t=T −T0
ρ(xT −t , xt∗ ) ≤ , t = 0, . . . , τ0 .
(π u + g)(x0 ) ≥ sup(π u + g) − δ,
0 −1
T
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT0 ) ≥ −δ
t=0
ρ(zt , xt ) ≤ , t = 0, . . . , T0 .
Proof Assume that the lemma does not hold. Then there exist a sequence of real
numbers {δk }∞ (k) T0
k=1 ⊂ (0, 1] and a sequence of (Ω)-programs {xt }t=0 , k = 1, 2, . . .
such that
lim δk = 0 (5.57)
k→∞
and that for each natural number k and each (u, Ω)-overtaking optimal program
{zt }∞
t=0 satisfying (5.56),
(k)
(π u + g) x0 ≥ sup(π u + g) − δk , (5.58)
0 −1
T
u xt(k) , xt+1
(k)
− u(x̄, x̄) − π u x0(k) + π u xT(k)
0
≥ −δk , (5.59)
t=0
we have
max ρ(zt , xt(k) ) : t = 0, . . . , T0 > . (5.60)
(k)
xt = lim xt . (5.61)
k→∞
(k)
π u (x0 ) ≥ lim sup π u (x0 ),
k→∞
(k)
g(x0 ) ≥ lim sup g(x0 ). (5.62)
k→∞
92 5 Structure of Solutions in the Regions Close to the Endpoints
(k)
g(x0 ) = lim g(x0 ). (5.64)
k→∞
It follows from upper semicontinuity of u and π u , (5.57), (5.59), (5.61), and (5.64)
that
0 −1
T
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT0 )
t=0
⎛ ⎞
0 −1
T
≥ lim sup ⎝ (u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT0 )⎠
(k) (k) (k) (k)
k→∞ t=0
≥ lim sup(−δk ) = 0.
k→∞
Combined with Proposition 5.1 this implies that for all integers t = 0, . . . , T0 − 1,
It follows from Theorem 3.4 that there exists a (u, Ω)-overtaking optimal program
{x̃t }∞
t=0 satisfying
x̃0 = xT0 . (5.66)
xt = x̃t−T0 . (5.67)
Evidently, {xt }∞
t=0 is an (Ω)-program. It follows from Corollary 5.3 that (5.65) holds
for all integers t ≥ 0. Proposition 5.8 and (5.65) imply that {xt }∞ t=0 is a (u, Ω)-
overtaking optimal program satisfying (5.63). By (5.61), for all sufficiently large
(k)
natural numbers k, ρ(xt , xt ) ≤ /4, t = 0, . . . , T0 . This contradicts (5.60). The
contradiction we have reached proves Lemma 5.16.
5.7 Proof of Theorem 5.15 93
By Lemma 5.16 applied to the pair (ū, Ω̄) there exists a real number
δ1 ∈ (0, /2)
(π ū + g)(y0 ) ≥ sup(π ū + g) − δ1 ,
0 −1
τ
(ū(yt , yt+1 ) − ū(x̄, x̄)) − π ū (y0 ) + π ū (yτ0 ) ≥ −δ1
t=0
ρ(zt , yt ) ≤ , t = 0, . . . , τ0 .
Propositions 5.4 and 5.5 applied to the pair (ū, Ω̄) and assumption (B2) imply
that there exists a real number δ2 ∈ (0, δ1 ) such that the following properties
hold:
(ii) for each point z ∈ K satisfying ρ(z, x̄) ≤ 2δ2 ,
By (B3), there exists δ3 ∈ (0, δ2 ) such that the following property holds:
(iv) for each x, y ∈ K satisfying
ξ0 = x, ξL∗ = y
and
ρ(ξt , x̄) ≤ δ2 , t = 0, . . . , L∗ .
94 5 Structure of Solutions in the Regions Close to the Endpoints
ut − u ≤ δ4 , t = 0 . . . , T − 1
τ +L
1 −1
τ +L1 −1
ut (zt , zt+1 ) ≥ U {ut }t=τ , Ω, zτ , zτ +L1 − δ4
t=τ
ρ(zt , x̄) ≤ δ3 , t = L1 , . . . , T − L1 .
h − g ≤ δ, (5.72)
ut − u ≤ δ, t = 0 . . . , T − 1 (5.73)
τ +T
0 −1
τ +T0 −1
ut (xt , xt+1 ) ≥ U {ut }t=τ , Ω, xτ , xτ +T0 − δ (5.74)
t=τ
5.7 Proof of Theorem 5.15 95
Property (iv), (5.69), (5.71), and (5.76) imply that there exists an (Ω)-program
{x̃t }Tt=0 such that
x̃t = xt , t = 0, . . . , T − L1 − L∗ − τ0 − τ1 − 4, (5.77)
x̃t = zT −t , t = T − L1 − τ0 − τ1 − 4, . . . , T , (5.78)
ρ(x̃t , x̄) ≤ δ2 , t = T − L1 − L∗ − τ0 − τ1 − 4, . . . , T − L1 − τ0 − τ1 − 4.
(5.79)
−1
T −1
T
≥ h(x̃T ) + ut (x̃t , x̃t+1 ) − h(xT ) − ut (xt , xt+1 )
t=T −T0 t=T −T0
−1
T
= h(x̃T ) − h(xT ) + (ut (x̃t , x̃t+1 ) − ut (xt , xt+1 ))
t=T −L1 −L∗ −τ0 −τ1 −4
−1
T
≥ g(x̃T ) − g(xT ) + u(x̃t , x̃t+1 )
t=T −L1 −L∗ −τ0 −τ1 −4
−1
T
− u(xt , xt+1 ) − 2δ(L1 + L∗ + τ0 + τ1 + 5)
t=T −L1 −L∗ −τ0 −τ1 −4
≥ g(x̃T ) − g(xT )
−1
T
+ u(x̃t , x̃t+1 ) + L∗ u(x̄, x̄) − L∗ (δ1 /8)(L∗ + 1)−1
t=T −L1 −τ0 −τ1 −4
−1
T
− u(xt , xt+1 ) − 2δ(L1 + L∗ + τ0 + τ1 + 5).
t=T −L1 −L∗ −τ0 −τ1 −4
96 5 Structure of Solutions in the Regions Close to the Endpoints
−1
T
g(xT ) + u(xt , xt+1 )
t=T −L1 −L∗ −τ0 −τ1 −4
−1
T
≥ g(x̃T ) + u(x̃t , x̃t+1 ) + L∗ u(x̄, x̄) − δ1 /8 − δ1 /8
t=T −L1 −τ0 −τ1 −4
L1 +τ
0 +τ1 +2
= g(z0 ) + ū(zt , zt+1 ) + u(x̃T −L1 −τ0 −τ1 −4 , x̃T −L1 −τ0 −τ1 −3 )
t=0
|u(x̃T −L1 −τ0 −τ1 −4 , x̃T −L1 −τ0 −τ1 −3 ) − u(x̄, x̄)| ≤ δ1 /8. (5.81)
−1
T
g(xT ) + u(xt , xt+1 )
t=T −L1 −L∗ −τ0 −τ1 −4
L1 +τ
0 +τ1 +2
≥ g(z0 ) + ū(zt , zt+1 ) + (L∗ + 1)u(x̄, x̄) − 3δ1 /8. (5.82)
t=0
Set
yt = xT −t , t = 0, . . . , T . (5.83)
L1 +L∗
+τ0 +τ1 +3
g(y0 ) + ū(yt , yt+1 )
t=0
−1
T
= g(xT ) + u(xt , xt+1 )
T −L1 −L∗ −τ0 −τ1 −4
L1 +τ
0 +τ1 +2
≥ g(z0 ) + ū(zt , zt+1 ) + (L∗ + 1)u(x̄, x̄) − 3δ1 /8. (5.84)
t=0
5.7 Proof of Theorem 5.15 97
(π ū + g)(y0 ) − sup(π ū + g)
0 −1
τ
+ (ū(yt , yt+1 ) − ū(x̄, x̄)) − π ū (y0 ) + π ū (yτ0 )
t=0
≥ (π ū + g)(y0 ) − (π ū + g)(z0 )
L1 +L∗
+τ0 +τ1 +3
+ (ū(yt , yt+1 ) − u(x̄, x̄)) − π ū (y0 ) + π ū (yL1 +L∗ +τ0 +τ1 +4 )
t=0
L1 +τ
0 +τ1 +2
≥ π ū (y0 ) − π ū (z0 ) + (ū(zt , zt+1 ) − ū(x̄, x̄))
t=0
(π ū + g)(y0 ) − sup(π ū + g)
0 −1
τ
+ (ū(yt , yt+1 ) − ū(x̄, x̄))
t=0
0 −1
τ
(ū(yt , yt+1 ) − ū(x̄, x̄))
t=0
It follows from the relations above, (5.83) and the property (i) that there exists an
(ū, Ω̄)-overtaking optimal program {ξt }∞
t=0 such that
2 −1
T
T2 −1
ut (xt , xt+1 ) + h(xT1 , xT2 ) → max, {(xt , xt+1 )}t=T 1
⊂Ω
t=T1
and define
⎧
⎨T
2 −1
T2 −1
U h, {ut }t=T1 , Ω = sup ut (xt , xt+1 ) + h(xT1 , xT2 ) :
⎩
t=T1
⎫
⎬
{xt }Tt=T
2
is an (Ω) − program .
1 ⎭
T2 −1
U (h, v, Ω, T1 , T2 ) = U (h, {vt }t=T 1
, Ω) where vt = v, t = T1 , . . . , T2 − 1.
Our next result describes the structure of approximate solutions in the regions
close to the endpoints.
Theorem 5.17 Suppose that g ∈ M(K × K) is an upper semicontinuous function,
τ0 is a natural number, and ∈ (0, 1). Then there exist a positive number δ and a
natural number T0 ≥ τ0 such that for every natural number T ≥ T0 , every function
h ∈ M(K × K) satisfying h − g ≤ δ, every finite sequence of bounded functions
ut : Ω → R 1 , t = 0, . . . , T − 1 which satisfies
ut − u ≤ δ, t = 0 . . . , T − 1,
−1
T
T −1
h(x0 , xT ) + ut (xt , xt+1 ) ≥ U h, {ut }t=0 , Ω − δ,
t=0
and that for all integers t = 0, . . . , τ0 , ρ(xt , xt∗ ) ≤ and ρ(xT −t , x̄t∗ ) ≤ .
ψg (x0 , y0 ) + δ ≥ sup(ψg ),
0 −1
T
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT0 ) ≥ −δ,
t=0
0 −1
T
(ū(yt , yt+1 ) − u(x̄, x̄)) − π ū (y0 ) + π ū (yT0 ) ≥ −δ,
t=0
100 5 Structure of Solutions in the Regions Close to the Endpoints
Proof Assume that the lemma does not hold. Then there exist a sequence {δk }∞ k=1 ⊂
(k) T0
(0, 1], a sequence of (Ω)-programs {xt }t=0 , k = 1, 2, . . . and a sequence of (Ω̄)-
programs {yt }Tt=0
(k) 0
, k = 1, 2, . . . such that
lim δk = 0, (5.89)
k→∞
0 −1
T
(k) (k) (k) (k)
u(xt , xt+1 ) − u(x̄, x̄) − π u x0 + π u xT0 ≥ −δk , (5.91)
t=0
0 −1
T
(k) (k) (k) (k)
ū(yt , yt+1 ) − u(x̄, x̄) − π ū y0 + π ū yT0 ≥ −δk (5.92)
t=0
ψg (z0 , ξ0 ) = sup(ψg ),
we have
(k) (k)
max ρ zt , xt + ρ ξt , yt : t = 0, . . . , T0 > .
By (5.95)–(5.98),
(k)
π u (x0 ) = lim π u x0 , (5.99)
k→∞
(k)
π ū (y0 ) = lim π ū y0 , (5.100)
k→∞
(k) (k)
g(x0 , y0 ) = lim g x0 , y0 . (5.101)
k→∞
0 −1
T
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xT0 )
t=0
⎛ ⎞
0 −1
T
≥ lim sup ⎝ u(xt(k) , xt+1
(k)
− u(x̄, x̄)) − π u x0(k) + πu xT(k)
0
⎠
k→∞ t=0
≥ lim sup(−δk ) = 0,
k→∞
0 −1
T
(ū(yt , yt+1 ) − u(x̄, x̄)) − π ū (y0 ) + π ū (yT0 )
t=0
102 5 Structure of Solutions in the Regions Close to the Endpoints
⎛ ⎞
0 −1
T
≥ lim sup ⎝ − u(x̄, x̄) − π ū y0 + π ū yT0 ⎠
(k) (k) (k) (k)
u yt , yt+1
k→∞ t=0
≥ lim sup(−δk ) = 0.
k→∞
Together with Proposition 5.1 this implies that for all integers t = 0, . . . , T0 − 1,
we have
It follows from Theorem 3.4 that there exist a (u, Ω)-overtaking optimal
program {x̃t }∞
t=0 satisfying
Clearly, {xt }∞ ∞
t=0 is an (Ω)-program and {yt }t=0 is an (Ω̄)-program. Corollary 5.3
implies that (5.102) and (5.103) hold for all integers t ≥ 0. It follows from
(5.102), (5.103), and Proposition 5.8 that {xt }∞t=0 is a (u, Ω)-overtaking optimal
∞
program and {yt }t=0 is a (ū, Ω̄)-overtaking optimal program satisfying (5.98).
In view of (5.93), for all sufficiently large natural numbers k,
(k) (k)
ρ(xt , xt ) ≤ /4, ρ(yt , yt ) ≤ /4, t = 0, . . . , T0 .
Together with (5.98) this contradicts property (i). The contradiction we have
reached proves Lemma 5.18.
Lemma 5.18 implies that there exists a number δ1 ∈ (0, /2) such that the following
property holds:
5.10 Proof of Theorem 5.17 103
0 −1
τ
(u(ξt , ξt+1 ) − u(x̄, x̄)) − π u (ξ0 ) + π u (ξτ0 ) ≥ −2δ1 ,
t=0
0 −1
τ
(ū(ηt , ηt+1 ) − u(x̄, x̄)) − π ū (η0 ) + π ū (ητ0 ) ≥ −2δ1
t=0
Propositions 5.4, 5.5, and assumption (B2) imply that there exists a number
δ2 ∈ (0, δ1 ) such that the following properties hold:
(ii) for every point z ∈ X satisfying ρ(z, x̄) ≤ 2δ2 , we have
we have
|u(x, y) − u(x̄, x̄)| ≤ (δ1 /16)(L∗ + 1)−1
By (B3), there exists δ3 ∈ (0, δ2 ) such that the following property holds:
(iv) for each z1 , z2 ∈ K satisfying
ρ(zi , x̄) ≤ δ3 , i = 1, 2
ξ0 = z1 , ξL∗ = z2
and
ρ(ξt , x̄) ≤ δ2 , t = 0, . . . , L∗ .
104 5 Structure of Solutions in the Regions Close to the Endpoints
Theorem 3.12 implies that there exist a natural number L1 and a positive
number δ4 ∈ (0, δ3 ) such that the following property holds:
(v) for every natural number T > 2L1 , every finite sequence of bounded functions
ut : Ω → R 1 , t = 0, . . . T − 1 satisfying
ut − u ≤ δ3 , t = 0 . . . , T − 1
−1
T
T −1
ut (xt , xt+1 ) ≥ U {ut }t=0 , Ω, x0 , xT − δ4
t=0
h − g ≤ δ, (5.112)
ut − u ≤ δ, t = 0 . . . , T − 1 (5.113)
−1
T
T −1
h(x0 , xT ) + ut (xt , xt+1 ) ≥ U h, {ut }t=0 , Ω − δ. (5.114)
t=0
5.10 Proof of Theorem 5.17 105
Property (iv), (5.108), (5.109), (5.111), and (5.115) imply that there exists an (Ω)-
program {x̃t }Tt=0 such that
−1
T
T −1
δ ≥ U (h, {ut }t=0 , Ω) − h(x0 , xT ) − ut (xt , xt+1 )
t=0
−1
T −1
T
≥ h(x̃0 , x̃T ) + ut (x̃t , x̃t+1 ) − h(x0 , xT ) − ut (xt , xt+1 )
t=0 t=0
2L1 +τ
0 +2+L∗
= h(ξ0 , η0 ) − h(x0 , xT ) + (ut (x̃t , x̃t+1 ) − ut (xt , xt+1 ))
t=0
−1
T
+ (ut (x̃t , x̃t+1 ) − ut (xt , xt+1 ))
t=T −2L1 −τ0 −3−L∗
≥ g(ξ0 , η0 ) − g(x0 , xT )
2L1 +τ
0 +2+L∗
+ (u(x̃t , x̃t+1 ) − u(xt , xt+1 ))
t=0
−1
T
+ (u(x̃t , x̃t+1 ) − u(xt , xt+1 )) − 2δ(1 + 2L1 + τ0 + 4 + L∗ )
t=T −2L1 −τ0 −3−L∗
≥ g(ξ0 , η0 ) − g(x0 , xT )
+τ0 +2
2L1
+ u(ξt , ξt+1 ) + L∗ (u(x̄, x̄) − (δ1 /16)(L∗ + 1)−1 )
t=0
106 5 Structure of Solutions in the Regions Close to the Endpoints
2L1 +τ
0 +2+L∗ −1
T
− ut (xt , xt+1 ) + u(ηT −t , ηT −t−1 )
t=0 t=T −2L1 −τ0 −3
≥ g(ξ0 , η0 ) − g(x0 , xT )
+τ0 +2
2L1 +τ0 +2
2L1
+ u(ξt , ξt+1 ) + ū(ηt , ηt+1 ) + 2L∗ u(x̄, x̄)
t=0 t=0
2L1 +τ
0 +2+L∗ −1
T
− u(xt , xt+1 ) − u(xt , xt+1 ) − 3δ1 /16.
t=0 t=T −2L1 −τ0 −3−L∗
(5.121)
Define
yt = xT −t , t = 0, . . . , T . (5.122)
2L1 +τ
0 +L∗ +2 2L1 +τ
0 +L∗ +2
g(x0 , y0 ) + u(xt , xt+1 ) + ū(yt , yt+1 )
t=0 t=0
+τ0 +2
2L1 +τ0 +2
2L1
≥ g(ξ0 , η0 ) + u(ξt , ξt+1 ) + ū(ηt , ηt+1 )
t=0 t=0
It follows from (5.108), (5.109), (5.111), (5.112), (5.115), (5.123), Proposition 5.1,
Corollary 5.3, and properties (ii) and (iii) that
g(x0 , y0 ) − g(ξ0 , η0 )
0 −1
τ
+ (u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xτ0 )
t=0
0 −1
τ
+ (ū(yt , yt+1 ) − u(x̄, x̄)) − π ū (y0 ) + π ū (yτ0 )
t=0
5.10 Proof of Theorem 5.17 107
≥ g(x0 , y0 ) − g(ξ0 , η0 )
2L1 +τ
0 +2+L∗
+ (u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (x2L1 +τ0 +L∗ +3 )
t=0
2L1 +τ
0 +L∗ +2
+ (ū(yt , yt+1 ) − ū(x̄, x̄)) − π ū (y0 ) + π ū (y2L1 +τ0 +3+L∗ )
t=0
+τ0 +2
2L1
≥ (u(ξt , ξt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (x2L1 +τ0 +3+L∗ )
t=0
+τ0 +2
2L1
+ (ū(ηt , ηt+1 ) − ū(x̄, x̄)) − π ū (y0 ) + π ū (y2L1 +τ0 +3+L∗ ) − δ1 /4
t=0
0 −1
τ
+ (ū(yt , yt+1 ) − u(x̄, x̄)) − π ū (y0 ) + π ū (yτ0 ) ≥ −δ1 . (5.125)
t=0
0 −1
τ
(u(xt , xt+1 ) − u(x̄, x̄)) − π u (x0 ) + π u (xτ0 ) ≥ −δ1 ,
t=0
0 −1
τ
(ū(yt , yt+1 ) − u(x̄, x̄)) − π ū (y0 ) + π ū (yτ0 ) ≥ −δ1 .
t=0
108 5 Structure of Solutions in the Regions Close to the Endpoints
It follows from the relations above and property (i) that there exists a (u, Ω)-
overtaking optimal program {xt∗ }∞
t=0 and a (ū, Ω̄)-overtaking optimal program
{yt∗ }∞
t=0 such that
In this chapter we continue the discussion of the forest management problem and
show that for this problem the results of Chaps. 3 and 5 hold.
6.1 Preliminaries
We consider a discrete time model for the optimal management of a forest of total
area S occupied by k species I = {1, . . . , k} with maturity ages of n1 , . . . , nk years
respectively. This model was introduced in Sect. 2.1. For the reader’s convenience
we recall here the notation and definitions.
j
For each period t = 0, 1, . . . we denote xi (t) ≥ 0 the area covered by trees of
species i that are j years old with j = 1, . . . , ni and x̄i (t) ≥ 0 the area occupied by
over-mature trees (older than ni ).
Assuming that only mature trees can be harvested we must have
and then the area not harvested in that period will comprise the over-mature trees at
the next step, namely
j +1 j
xi (t + 1) = xi (t), j = 1, . . . , ni − 1. (6.3)
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2019 109
A. J. Zaslavski, Optimal Control Problems Arising in Forest Management,
SpringerBriefs in Optimization, https://doi.org/10.1007/978-3-030-23587-1_6
110 6 Applications to the Forest Management Problem
The total harvested area i∈I ui (t) is allocated to new seedlings which is
expressed by the equation
xi1 (t + 1) = ui (t). (6.4)
i∈I i∈I
where xi (t) = (xi1 (t), . . . , xini (t), xini +1 (t)) describes the areas occupied in year t
by trees of species i with ages 1, 2, . . . , ni and over ni . The first and last components
of each vector xi (t) are controlled by the sowing and harvesting policies.
Let R+m = {x = (x , . . . , x ) ∈ R m : x ≥ 0, i = 1, . . . , m}.
1 m i
Let N = i∈I (ni +1). Every vector x ∈ R N is represented as x = (x1 , . . . , xk ),
where xi = (xi1 , . . . , xini , xini +1 ) ∈ R ni +1 for all integers i = 1, . . . , k.
Denote by Δ the set of all x ∈ R+ N such that
⎡ ⎤
i +1
n
⎣ xi ⎦ = S.
j
(6.6)
i∈I j =1
A sequence {x(t)}∞ t=0 ⊂ Δ is called a program if for all integers t ≥ 0 and all
i ∈ I (6.1)–(6.4) hold (see (6.5)) with some u(t) = (u1 (t), . . . , uk (t)) ∈ R+
k.
T2
Let the integers T1 , T2 satisfy 0 ≤ T1 < T2 . A sequence {x(t)}t=T1 ⊂ Δ is called
a program if (6.1)–(6.4) hold for all i ∈ I and for all integers t = T1 , . . . , T2 − 1
(see (6.5)) with some u(t) = (u1 (t), . . . , uk (t)) ∈ R+
k.
where for i = 1, . . . , k,
Set
k
Δ0 = x ∈ k
R+ : xi ≤ S .
i=1
n̄ = max{ni : i ∈ I }. (6.9)
This model is a particular case of the general optimal control system studied in
Chaps. 2, 3, and 5. Proposition 2.1 implies that assumption (A) holds with
L̄ = N + n̄ + 1.
All the results of Chap. 2 hold for the particular case considered here with u = w◦V .
We use the notation and definitions introduced and used in Chaps. 2 and 3.
Let y, z ∈ Δ and let the integers T1 , T2 satisfy T1 < T2 . Set
⎧
⎨T
2 −1
U (y, T1 , T2 ) = sup w(V (xt , xtt+1 )) :
⎩
t=T1
⎫
⎬
{xt }Tt=T
2
is a program and xT1 = y ,
1 ⎭
112 6 Applications to the Forest Management Problem
⎧
⎨T
2 −1
U (y, ỹ, T1 , T2 ) = sup w(V (xt , xtt+1 )) :
⎩
t=T1
⎫
⎬
{xt }Tt=T
2
is a program and xT1 = y, xT2 = ỹ ,
1 ⎭
⎧ ⎫
⎨T
2 −1 ⎬
(T1 , T2 ) = sup
U w(V (xt , xtt+1 )) : {xt }Tt=T
2
is a program .
⎩ 1 ⎭
t=T1
Let
μ := μ(w ◦ V )
for each
x = (x1 , . . . , xk ), y = (y1 , . . . , yk ) ∈ Δ0
satisfying
xi ≤ yi , i = 1, . . . , k
and that
x1 , . . . ,
x = ( xk ) ∈ Δ (6.12)
6.2 Auxiliary Results 113
such that
xi = xini +1 , i = 1, . . . , k,
xi1 . . . ,
x,
( x ) ∈ Ω, (6.13)
x,
w(V ( x )) ≥ w(V (x, x)) for all x ∈ Δ such that (x, x) ∈ Ω, (6.14)
xini +1 = 0.
j
xi =
xi1 , j = 1, . . . , ni ,
j
xi =
xi1 , j = 1, . . . , ni .
If
xini +1 = 0, i = 1, . . . , k,
Then
x,
V ( x ) = (
x11 , . . . ,
xk1 )
Let
x = (
x1 , . . . ,
xk ) ∈ Δ, (
x,
x ) ∈ Ω, (6.18)
xi = (
xini +1 ), i = 1, . . . , k,
xi1 . . . ,
for i = 1, . . . , k,
xini +1 = 0,
j
xi =
xi1 , j = 1, . . . , ni , (6.19)
xi1 : i ∈ I˜}).
∈ (0, min{ (6.21)
x(0) = x, x(N + n̄ + 1) = y,
x(t) −
x ≤ , t = 0, . . . , N + n̄ + 1. (6.22)
k
k
vi (t) ≤ xini (t) + xini +1 (t), xi1 (t + 1) = vi (t) (6.24)
i=1 i=1
6.2 Auxiliary Results 115
and then the state x(t + 1) is defined as follows for all integers i = 1, . . . , k:
j j −1
xi (t + 1) = xi (t), j is an integer satisfying 2 ≤ j ≤ ni , (6.25)
Thus the construction of the program is done as follows: if the states are defined till
moment t ≤ N + n̄, then we choose V (t) ∈ R+ k and (x 1 (t +1), . . . , x 1 (t +1)) ∈ R k
1 k +
such that (6.24) holds and define x(t + 1) by (6.25) and (6.26).
For each t = 0, . . . , N − 1 set
xi1 − (2N)−1 , i = 1, . . . , k0 ,
xi1 (t + 1) = vi (t) =
xi1 (t + 1) = vi (t) = 0 for all integers i satisfying k0 < i ≤ k (6.27)
j
and define xi (t + 1), i = 1, . . . , k, j = 2, . . . , ni + 1 by (6.25) and (6.26).
In order to show that {x(t)}N t=0 is a program it is sufficient to show that the first
part of (6.24) is true for all integers i = 1, . . . , k0 .
It follows from (6.23), (6.27), and (6.25) that for all integers t = 0, . . . , N, all
integers i = 1, . . . , k0 , and all integers j = 1, . . . , ni we have
Equations (6.27) and (6.28) imply (6.24) for all integers i = 1, . . . , k0 . Then
{x(t)}N
t=0 is a program. In view of (6.23), (6.27), and (6.28), for all integers
t = 0, . . . , N ,
k
k0
0≤ xini +1 ≤ S − xi1 + N(2N )−1 ≤ 2−1 .
ni (6.29)
i=1 i=1
x ≤ 2−1 , t = 0, . . . , N.
x(t) − (6.30)
If at the moment t the state x(t) has already been defined, then we define V (t) ∈
k and (x 1 (t + 1), . . . , x 1 (t + 1)) ∈ R k such that (6.24) holds and then define
R+ 1 k +
x(t + 1) by (6.25) and (6.26).
For every integer s = 1, . . . , n̄ set
j
We have defined xi1 (t + N), i ∈ I, t = 1, . . . , n̄. Then we define xi (t + 1) by
(6.25) for all t = N, . . . , N + n̄ − 1, all integers j satisfying 2 ≤ j ≤ ni , and all
j
i ∈ I . Thus we have defined xi (t + N), t = 1, . . . , n̄, i ∈ I, j = 1, . . . , n(i).
Set
ni
j
φ(t) = xi (N + t) − xini (N + t)
i∈I j =1 i∈I
+ xi1 (N + t + 1), t = 0, . . . , n̄ − 1. (6.37)
i∈I
ni
j
φ(t) = xi (N + t + 1). (6.38)
i∈I j =1
6.2 Auxiliary Results 117
Hence
φ(t) ≤ S, t = 0, . . . , n̄ − 1. (6.40)
+n̄
In order to complete the construction of the program {x(t)}N t=0 we need to
n +1
determine V (t) ∈ R+ k t = N, . . . , N + n̄ − 1 and x i (t + 1) ≥ 0, t =
i
N, . . . , N + n̄ − 1] such that the second part of (6.24) is valid for all integers
t = N, . . . , N + n̄ − 1 and
xi1 (N + 1) = vi (N ), vi (N ) ≤ xini (N ) + xini +1 (N ), i ∈ I. (6.43)
i∈I i∈I
Vi (N + τ ) ≤ xini +1 (N + τ ) + xini (N + τ ), i ∈ I.
118 6 Applications to the Forest Management Problem
j j +1
xi (N + n̄) = yi for all integers j satisfying 1 ≤ j < n(i),
Evidently,
ni i +1
n
j j
xi = yi = S.
i∈I j =1 i∈I j =1
+n̄+1
Set x(N + n̄ + 1) = y. It is clear that (x(N + n̄), y) ∈ Ω. Hence {x(t)}N
t=0 is a
program such that x(0) = x, x(N + n̄ + 1) = y. In view of (6.33)–(6.36),
x(t) − x̄ ≤ , t = 0, . . . , N + n̄ + 1.
Note that w(V (x, y)) is the benefit obtained today if the forest today is x and the
forest tomorrow is y where (x, y) ∈ Ω.
In the sequel we assume that supremum over empty set is −∞.
Theorem 2.9 implies the following result.
Theorem 6.3 There exists M > 0 and the limit
(0, p)/p
μ := μ(w ◦ V ) := lim U (6.45)
p→∞
6.3 Turnpike Results 119
such that
x = (
x1 , . . . ,
xk ) ∈ Δ
such that
x,
( x ) ∈ Ω,
for i = 1, . . . , k,
xi =
xini +1 ,
xi1 . . . ,
xini +1 = 0,
j
xi =
xi1 , j = 1, . . . , ni ,
μ = w(V (
x,
x )) ≥ w(V (x, x)) for all x ∈ Δ satisfying (x, x) ∈ Ω.
Proposition 6.2 implies that (B3) holds. It is clear that (B2) holds too.
Theorem 6.5 Assume that a program {x(t)}∞
t=0 is good. Then
lim x(t) =
x.
t→∞
Evidently, {z(t)}∞
t=0 is a program and
It follows from (6.48) and the concavity of the function w that {z(t)}∞
t=0 is a good
program and the sequence
T −1 ∞
w(V (z(t), z(t + 1))) − T μ is bounded. (6.49)
t=0 T =1
Assume that (6.46) does not hold. There exist a positive number and a strictly
increasing sequence of natural numbers {tj }∞
j =1 such that
V (
x,
x ) − V (x(tj ), x(tj + 1)) ≥ . (6.50)
Since the function w is continuous and strictly concave there exists a positive
number δ such that for every pair of points y1 , y2 ∈ Δ0 which satisfies y1 − y2 ≥
, we have
w 2−1 (y1 + y2 ) − 2−1 w(y1 ) − 2−1 w(y2 ) ≥ δ. (6.51)
In view of (6.48), the strict concavity of the function w, (6.50) and the choice of the
number δ (see (6.51)), (6.52), for every integer k > 0,
k +1
t
w(V (z(t), z(t + 1))) − tk μ − 2μ
t=0
k +1
t
= w(2−1 V (x(t), x(t + 1)) + 2−1 V (
x,
x )) − tk μ − 2μ
t=0
k +1
t
= [w(2−1 V (x(t), x(t + 1))
t=0
−1
+2 V ( x )) − 2−1 w(V (x(t), x(t + 1))) − 2−1 w(V (
x, x,
x ))]
k +1
t
− tk μ − 2μ + 2−1 w(V (x(t), x(t + 1))) + 2−1 (tk + 2)w(V (
x,
x ))
t=0
k +1
t
≥ δk + 2−1 [w(V (x(t), x(t + 1))) − μ]
t=0
≥ δk − M0 → ∞ as k → ∞,
Set
= V (
V x, x11 ,
x ) = ( x21 , . . . ,
xk1 ) (6.53)
where
V (t) = (v1 (t), . . . , vk (t)). (6.55)
Let t ≥ 0 be an integer. In view of (6.2), (6.5), (6.26), and (6.54), for all integers
i = 1, . . . , k,
xini (t) = xini +1 (t + 1) − xini +1 (t) + vi (t),
and
j +1 j
xi (t + 1) = xi (t), j = 1, . . . , ni − 1 (6.57)
for all integers i = 1, . . . , k if ni ≥ 2. It follows from (6.56) and (6.57) that for all
integers T ≥ 1 and for all integers i = 1, . . . , k,
T +n
i −1 T +n
i −1
vi (t) = xini +1 (t) + xini (t) − xini +1 (t + 1)
t=T t=T
T +n
i −1
= xini +1 (T ) − xini +1 (T + ni ) + xini (t)
t=T
ni
= xini +1 (T ) − xini +1 (T + ni ) +
j
xi (T ). (6.58)
j =1
By (6.58)
⎡ ⎤ ⎡ ⎤
T +n i −1 n +1 ni
⎣ vi (t)⎦ =
n 1
xi i (T ) − xi + (T + ni ) + ⎣ xi (T )⎦
j
lim x ni +1 (T + ni ) = 0, i = 1, . . . , k. (6.61)
T →∞ i
lim x(t) =
x.
t→∞
k
Δ0 = {v ∈ R+
k
: vi ≤ S},
i=1
f = sup{|f (x)| : x ∈ Δ0 }.
6.4 Generic Results 123
where for i = 1, . . . , k,
Denote by Cm (Δ0 ) the set of all w ∈ C(Δ0 ) such that w : Δ0 → [0, ∞),
and that
It is not difficult to see that Cm (Δ0 ) is a closed subset of C(Δ0 ). The space Cm (Δ0 )
is equipped with the metric dC (f1 , f2 ) = f1 − f2 , f1 , f2 ∈ Cm (Δ0 ).
Let w ∈ Cm (Δ0 ). Lemma 6.1 implies that there exists
x = (
x1 , . . . ,
xk ) ∈ Δ
such that
x,
( x ) ∈ Ω, (6.67)
x,
w(V ( x )) ≥ w(V (x, x)) for all x ∈ Δ satisfying (x, x) ∈ Ω, (6.68)
for i = 1, . . . , k,
xi = (
xini +1 ),
xi1 . . . ,
xini +1 = 0.
j
xi =
xi1 , j = 1, . . . , ni , (6.69)
Assume that
μ(w ◦ V ) = w(V (
x,
x )) > 0. (6.70)
I0 = {i ∈ {1, . . . , k} :
xi1 > 0}. (6.71)
124 6 Applications to the Forest Management Problem
φi (x) = w(V (
x,
x )) for all x ∈ [
xi1 , ∞), (6.72)
lim x(t) =
x.
t→∞
Proof Theorem 2.9 and (6.70) imply that there exists M0 > 0 such that for each
(Ω)-program {x(t)}∞
t=0 and each integer T ≥ 1,
−1
T
w(V (x(t), x(t + 1))) ≤ T w(V (
x,
x )) + M0 . (6.75)
t=0
Let {x(t)}∞
t=0 be a program. By (6.72)–(6.75), for each integer T > 0,
−1
T
wγ (V (x(t), x(t + 1)))
t=0
−1
T −1
T
= w(V (x(t), x(t + 1))) + γ φi (vi (x(t), x(t + 1)))
t=0 t=0 i∈I0
≤ T w(V ( x )) + M0 + γ T
x, (w ◦ V )(
x,
x)
i∈I0
= M0 + T w(V (
x,
x ))(1 + γ Card(I0 )) = M0 + (wγ ◦ V )(
x,
x )T .
lim x(t) =
x.
t→∞
6.4 Generic Results 125
In view of (6.76), there exists M1 > 0 such that for each integer T ≥ 1,
T −1
wγ (V (x(t), x(t + 1))) − T wγ (V ( x )) ≤ M1 .
x, (6.77)
t=0
− M1 + T w(V (
x,
x ))(1 + γ Card(I0 ))
= −M1 + T (wγ ◦ V )(
x,
x)
−1
T
≤ wγ (V (x(t), x(t + 1)))
t=0
−1
T −1
T
= w(V (x(t), x(t + 1))) + γ φi (vi (x(t), x(t + 1)))
t=0 t=0 i∈I0
−1
T
≤ T (w ◦ V )( x ) + M0 + γ
x, φi (vi (x(t), x(t + 1))). (6.78)
t=0 i∈I0
−1
T
=γ ((w ◦ V )(
x,
x ) − φi (vi (x(t), x(t + 1)))). (6.79)
t=0 i∈I0
Let i ∈ I0 . It follows from (6.72), (6.73), and (6.79) that for every integer T > 0,
−1
T
γ −1 (M0 + M1 ) ≥ ((w ◦ V )(
x,
x ) − φi (vi (x(t), x(t + 1)))). (6.80)
t=0
and there exists an integer t > 0 such that for all integers t ≥ t ,
(w ◦ V )(
x,
x ) − φi (vi (x(t), x(t + 1))) < .
126 6 Applications to the Forest Management Problem
Since an arbitrary positive number it follows from (6.72) and (6.73) that
j +1 j
xi (t + 1) = xi (t), j = 1, . . . , ni − 1 (6.85)
T +n
i −1 T +n
i −1
vi (t) = xini +1 (t) + xini (t) − xini +1 (t + 1)
t=T t=T
T +n
i −1
= xini +1 (T ) − xini +1 (T + ni ) + xini (t)
t=T
ni
= xini +1 (T ) − xini +1 (T + ni ) +
j
xi (T ). (6.86)
j =1
k
=S− xini +1 (T + ni ). (6.87)
i=1
6.4 Generic Results 127
k
= S − lim sup xini +1 (T + ni ). (6.88)
T →∞ i=1
lim x ni +1 (T + ni ) = 0, i = 1, . . . , k. (6.89)
T →∞ i
lim vi (t) =
xi1 , i = 1, . . . , k,
t→∞
lim x ni (t) =
xi1 , i = 1, . . . , k,
t→∞ i
j
lim x =
xi1 , i = 1, . . . , k, j = 1, . . . , ni .
t→∞ i
x w = (x1w , . . . , xkw ) ∈ Δ
such that
(x w , x w ) ∈ Ω,
for i = 1, . . . , k,
= xiw,1 , j = 1, . . . , ni , xiw,ni +1 = 0
w,j
xi
lim x(t) = x w .
t→∞
μ(w ◦ V ) > 0,
γ ∈ (0, k −1 ) and p ≥ 1 be an integer. By Lemmas 6.1 and 6.6 and Theorem 3.12,
there exist an open neighborhood U (w, γ , p) of wγ in Cm (Δ0 ), δ(w, γ , p) > 0, a
natural number τ (w, γ , p) and
x w = (x1w , . . . , xkw ) ∈ Δ
(x w , x w ) ∈ Ω,
for i = 1, . . . , k,
= xiw,1 , j = 1, . . . , ni , xiw,ni +1 = 0,
w,j
xi
μ(w ◦ V ) = w(V (x w , x w ));
lim x(t) = x w ;
t→∞
6.4 Generic Results 129
(c) for each integer T > 2τ (w, γ , p), each ut ∈ U (w, γ , p), t = 0, . . . , T − 1 and
each program {x(t)}Tt=0 which satisfies
−1
T
T −1
(ut ◦ V )(xt , xt+1 ) ≥ U {ut ◦ V }t=0 , x0 , xT − δ(w, γ , p)
t=0
we have
Define
F = ∩∞
q=1 ∪ {U (w, γ , p) :
q > 4 −1 . (6.95)
It follows from (6.93) and (6.94) that there exist w ∈ Cm (Δ0 ), γ ∈ (0, k −1 ), and an
integer p ≥ q such that
μ(w ◦ V ) > 0,
Assume that
ξ1 , ξ2 ∈ Δ,
(ξi , ξi ) ∈ Ω, i = 1, 2,
(u ◦ V )(ξi , ξi ) = μ(u ◦ V ), i = 1, 2.
It is not difficult to see that for each i = 1, 2 and each integer T > 0,
U (u ◦ V , ξi , ξi , 0, T ) = T (u ◦ V )(ξi , ξi ).
130 6 Applications to the Forest Management Problem
ξ1 − ξ2 < .
is a singleton.
Denote its unique element by x u . In view of (6.97),
x u = (x1u , . . . , xku ) ∈ Δ,
(x u , x u ) ∈ Ω,
for i = 1, . . . , k,
= xiu,1 , j = 1, . . . , ni , xiu,ni +1 = 0.
u,j
xi
Assume that an integer T > 2τ (w, γ , p) and that a program {xt }Tt=0 satisfies
−1
T
(u ◦ V )(xt , xt+1 ) ≥ U (u ◦ V , x0 , xT , 0, T ) − δ(w, γ , p).
t=0
Together with property (c), (6.96), and (6.98) this implies that for all integers t =
τ (w, γ , p), . . . , T − τ (w, γ , p),
x(t) − x w ≤ 1/p
and
Combined with Theorem 4.1 this implies that every (u◦V )-good program converges
to x u . Theorem 6.7 is proved.
It is clear that for every w ∈ F (B1), (B2), and (B3) hold for w ◦ V and all the
results of Chaps. 3 and 5 hold.
References
1. Anderson BDO, Moore JB (1971) Linear optimal control. Prentice-Hall, Englewood Cliffs, NJ.
2. Arkin VI, Evstigneev IV (1987) Stochastic models of control and economic dynamics.
Academic Press, London.
3. Aseev SM, Kryazhimskiy AV (2004) The Pontryagin Maximum principle and transversality
conditions for a class of optimal control problems with infinite time horizons, SIAM J. Control
Optim.: 43, 1094–1119.
4. Aseev SM, Veliov VM (2012) Maximum principle for infinite-horizon optimal control
problems with dominating discount, Dynamics Contin. Discrete Impuls. Syst., Series B: 19,
43–63.
5. Atsumi H (1965) Neoclassical growth and the efficient program of capital accumulation,
Review of Economic Studies: 32, 127–136.
6. Aubry S, Le Daeron PY (1983) The discrete Frenkel-Kontorova model and its extensions I,
Physica D: 8, 381–422.
7. Bachir M, Blot J (2015) Infinite dimensional infinite-horizon Pontryagin principles for discrete-
time problems, Set-Valued and Variational Analysis: 23, 43–54.
8. Bachir M, Blot J (2017) Infinite dimensional multipliers and Pontryagin principles for discrete-
time problems, Pure and Applied Functional Analysis: 2, 411–426.
9. Baumeister J, Leitao A, Silva GN (2007) On the value function for nonautonomous optimal
control problem with infinite horizon, Syst. Control Lett.: 56, 188–196.
10. Blot J (2009) Infinite-horizon Pontryagin principles without invertibility, J. Nonlinear Convex
Anal.: 10, 177–189.
11. Blot J, Cartigny P (2000) Optimality in infinite-horizon variational problems under sign
conditions, J. Optim. Theory Appl.: 106, 411–419.
12. Blot J, Hayek N (2000) Sufficient conditions for infinite-horizon calculus of variations
problems, ESAIM Control Optim. Calc. Var.: 5, 279–292.
13. Blot J, Hayek N (2014) Infinite-horizon optimal control in the discrete-time framework.
SpringerBriefs in Optimization, New York.
14. Brock WA (1970) On existence of weakly maximal programmes in a multi-sector economy,
Review of Economic Studies: 37, 275–280.
15. Carlson DA (1990) The existence of catching-up optimal solutions for a class of infinite horizon
optimal control problems with time delay, SIAM Journal on Control and Optimization: 28,
402–422.
16. Carlson DA, Haurie A, Leizarowitz A (1991) Infinite horizon optimal control. Berlin, Springer-
Verlag.
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2019 131
A. J. Zaslavski, Optimal Control Problems Arising in Forest Management,
SpringerBriefs in Optimization, https://doi.org/10.1007/978-3-030-23587-1
132 References
17. Carlson DA, Jabrane A, Haurie A (1987) Existence of overtaking solutions to infinite
dimensional control problems on unbounded time intervals, SIAM Journal on Control and
Optimizaton: 25, 517–1541.
18. Cartigny P, Michel P (2003) On a sufficient transversality condition for infinite horizon optimal
control problems, Automatica J. IFAC: 39, 1007–1010.
19. Coleman BD, Marcus M, Mizel VJ (1992) On the thermodynamics of periodic phases, Arch.
Rational Mech. Anal.: 117, 321–347.
20. Cominetti R, Piazza A (2009) Asymptotic convergence of optimal harvesting policies for a
multiple species forest, Mathematics of Operations Research: 34, 576–593.
21. Damm T, Grune L, Stieler M, Worthmann K (2014) An exponential turnpike theorem for
dissipative discrete time optimal control problems, SIAM Journal on Control and Optimization:
52, 1935–1957.
22. De Oliveira VA, Silva GN (2009) Optimality conditions for infinite horizon control problems
with state constraints, Nonlinear Analysis: 71, 1788–1795.
23. Evstigneev IV, Flam SD (1998) Rapid growth paths in multivalued dynamical systems
generated by homogeneous convex stochastic operators, Set-Valued Anal.: 6, 61–81.
24. Gaitsgory V, Grune L, Thatcher N (2015) Stabilization with discounted optimal control,
Systems and Control Letters: 82, 91–98.
25. Gaitsgory V, Mammadov M, Manic L (2017) On stability under perturbations of long-run
average optimal control problems, Pure and Applied Functional Analysis: 2, 461–476.
26. Gaitsgory V, Parkinson, A, Shvartsman, I (2017) Linear programming formulations of
deterministic infinite horizon optimal control problems in discrete time, Discrete Contin. Dyn.
Syst. Ser. B: 22, 3821–3838.
27. Gaitsgory V, Rossomakhine S, Thatcher N (2012) Approximate solution of the HJB inequality
related to the infinite horizon optimal control problem with discounting, Dynamics Contin.
Discrete Impuls. Syst., Series B: 19, 65–92.
28. Gale D (1967) On optimal development in a multi-sector economy, Rev. Econ. Stud.: 34, 1–18.
29. Glizer VY, Kelis O (2017) Upper value of a singular infinite horizon zero-sum linear-quadratic
differential game, Pure and Applied Functional Analysis: 2, 511–534.
30. Grune L, Guglielmi R (2018) Turnpike properties and strict dissipativity for discrete time linear
quadratic optimal control problems, SIAM J. Control Optim.: 56, 1282–1302.
31. Gugat M, Trelat E, Zuazua, E (2016) Optimal Neumann control for the 1D wave equation:
finite horizon, infinite horizon, boundary tracking terms and the turnpike property, Systems
Control Lett. 90, 61–70.
32. Guo X, Hernandez-Lerma O (2005) Zero-sum continuous-time Markov games with unbounded
transition and discounted payoff rates, Bernoulli: 11, 1009–1029.
33. Hayek (2011) Infinite horizon multiobjective optimal control problems in the discrete time
case, Optimization: 60, 509–529.
34. Hammond PJ (1974) Consistent planning and intertemporal welfare economics. University of
Cambridge, Cambridge.
35. Hammond PJ (1975) Agreeable plans with many capital goods, Rev. Econ. Stud.: 42, 1–14.
36. Hammond PJ, Mirrlees JA (1973) Agreeable plans, Models of economic growth. Wiley, New
York, 283–299.
37. Jasso-Fuentes H, Hernandez-Lerma O (2008) Characterizations of overtaking optimality for
controlled diffusion processes, Appl. Math. Optim.: 57, 349–369.
38. Jerry M, Rapaport A, Cartigny P (2010) Can protected areas potentially enlarge viability
domains for harvesting management? Nonlinear Anal. Real World Appl.: 11 720–734.
39. Khan MA, Piazza A (2011) The economics of forestry and a set-valued turnpike of the classical
type, Nonlinear Anal.: 74, 171–181.
40. Khan MA, Piazza A (2012) On the Mitra-Wan forestry model: a unified analysis, J. Econom.
Theory: 147, 230–260.
41. Khlopin DV (2017) On Lipschitz continuity of value functions for infinite horizon problem,
Pure and Applied Functional Analysis: 2, 535–552.
References 133
42. Kolokoltsov V, and Yang W (2012) The turnpike theorems for Markov games, Dynamic Games
and Applications: 2, 294–312.
43. Leizarowitz A (1985) Infinite horizon autonomous systems with unbounded cost, Appl. Math.
and Opt.: 13, 19–43.
44. Leizarowitz A (1986) Tracking nonperiodic trajectories with the overtaking criterion, Appl.
Math. and Opt.: 14, 155–171.
45. Leizarowitz A, Mizel VJ (1989) One dimensional infinite horizon variational problems arising
in continuum mechanics, Arch. Rational Mech. Anal.:106, 161–194.
46. Lykina V, Pickenhain S, Wagner M (2008) Different interpretations of the improper integral
objective in an infinite horizon control problem, J. Math. Anal. Appl.: 340, 498–510.
47. Makarov VL, Rubinov AM (1977) Mathematical theory of economic dynamics and equilibria.
Springer-Verlag, New York.
48. Malinowska AB, Martins N, Torres DFM (2011) Transversality conditions for infinite horizon
variational problems on time scales, Optimization Lett.: 5, 41–53.
49. Mammadov M (2014) Turnpike theorem for an infinite horizon optimal control problem with
time delay, SIAM Journal on Control and Optimization: 52, 420–438.
50. Marcus M, Zaslavski AJ (1999) On a class of second order variational problems with
constraints, Israel J. Math.: 111, 1–28.
51. Marcus M, Zaslavski AJ (1999) The structure of extremals of a class of second order variational
problems, Ann. Inst. H. Poincaré, Anal. non linéaire: 16, 593–629.
52. Maruyama T (1981) Optimal economic growth with infinite planning time horizon, Proc.
Japan Acad. Ser. A Math. Sci.: 57, 469–472.
53. McKenzie LW (1976) Turnpike theory, Econometrica: 44, 841–866.
54. Mitra T, Wan HW (1985) Some theoretical results on the economics of forestry, Review of
Economics Studies: 52, 263–282.
55. Mitra T, Wan HW (1986) On the Faustmann solution to the forest management problem,
Journal of Economic Theory: 40, 229–249.
56. Mordukhovich BS (1990) Minimax design for a class of distributed parameter systems,
Automat. Remote Control: 50, 1333–1340.
57. Mordukhovich BS (2011) Optimal control and feedback design of state-constrained parabolic
systems in uncertainly conditions, Appl. Analysis: 90, 1075–1109.
58. Mordukhovich BS, Shvartsman I (2004). Optimization and feedback control of constrained
parabolic systems under uncertain perturbations, Optimal control, stabilization and nonsmooth
analysis. Lecture Notes Control Inform. Sci., Springer, 121–132.
59. Moser J (1986) Minimal solutions of variational problems on a torus, Ann. Inst. H. Poincaré,
Anal. non linéaire: 3, 229–272.
60. Ocana Anaya E, Cartigny P, Loisel P (2009) Singular infinite horizon calculus of variations.
Applications to fisheries management, J. Nonlinear Convex Anal.: 10, 157–176.
61. Piazza A (2009) The optimal harvesting problem with a land market: a characterization of the
asymptotic convergence, Economic Theory: 40, 113–138.
62. Pickenhain S, Lykina V, Wagner M (2008) On the lower semicontinuity of functionals involv-
ing Lebesgue or improper Riemann integrals in infinite horizon optimal control problems,
Control Cybernet.: 37, 451–468.
63. Rapaport A, Sraidi S, Terreaux JP (2003) Optimality of greedy and sustainable policies in the
management of renewable resources, Optimal Control Application and Methods: 24, 23–44.
64. Rockafellar RT (1970) Convex analysis, Princeton University Press, Princeton, NJ.
65. Rubinov AM (1984) Economic dynamics, J. Soviet Math.: 26, 1975–2012.
66. Sagara N (2018) Recursive variational problems in nonreflexive Banach spaces with an infinite
horizon: an existence result, Discrete Contin. Dyn. Syst. Ser. S: 11, 1219–1232.
67. Samuelson PA (1965) A catenary turnpike theorem involving consumption and the golden rule,
Amer. Econom. Rev.: 55, 486–496.
68. Samuelson PA (1976) Economics of forestry in an evolving society, Economic Inquiry: XIV,
466–492.
134 References
69. Trelat E, Zhang C, Zuazua E (2018) Optimal shape design for 2D heat equations in large time,
Pure and Applied Functional Analysis: 3, 255–269.
70. von Weizsacker CC (1965) Existence of optimal programs of accumulation for an infinite
horizon, Rev. Econ. Studies: 32, 85–104.
71. Zaslavski AJ (1987) Ground states in Frenkel-Kontorova model, Math. USSR Izvestiya: 29,
323–354.
72. Zaslavski AJ (1995) Optimal programs on infinite horizon 1, SIAM J. Control Optim.: 33,
1643–1660.
73. Zaslavski AJ (1995) Optimal programs on infinite horizon 2, SIAM J. Control and Optim.: 33,
1661–1686.
74. Zaslavski AJ (1999) Turnpike property for dynamic discrete time zero-sum games, Abstract
and Applied Analysis: 4, 21–48.
75. Zaslavski AJ (2005) The turnpike property of discrete-time control problems arising in
economic dynamics, Discrete and Continuous Dynamical Systems, B: 5, 2005, 861–880.
76. Zaslavski AJ (2006) Turnpike properties in the calculus of variations and optimal control.
Springer, New York.
77. Zaslavski AJ (2007) Turnpike results for a discrete-time optimal control systems arising in
economic dynamics, Nonlinear Anal.: 67, 2024–2049.
78. Zaslavski AJ (2009) Two turnpike results for a discrete-time optimal control systems,
Nonlinear Anal.: 71, 902–909.
79. Zaslavski AJ (2010) Stability of a turnpike phenomenon for a discrete-time optimal control
systems, Journal of Optimization theory and Applications: 145, 597–612.
80. Zaslavski AJ (2010) On a class of infinite horizon optimal control problems, Communications
in Mathematical Analysis: 9, 1–11.
81. Zaslavski AJ (2010) Good programs for a class of infinite horizon optimal control problems,
Communications on Applied Nonlinear Analysis: 17, 1–14.
82. Zaslavski AJ (2011) The existence and structure of approximate solutions of dynamic discrete
time zero-sum games, Journal of Nonlinear and Convex Analysis: 12, 49–68.
83. Zaslavski AJ (2011) Turnpike properties of solutions for a class of optimal control problems
with applications to a forest management problem, Dynamics of Continuous, Discrete and
Impulsive Systems, Ser. B Appl. Algorithms: 18, 399–434.
84. Zaslavski AJ (2014) Turnpike phenomenon and infinite horizon optimal control. Springer
Optimization and Its Applications, New York.
85. Zaslavski AJ (2016) Structure of solutions of optimal control problems on large intervals: a
survey of recent results. Pure and Applied Functional Analysis: 1, 123–158.
86. Zaslavski AJ (2017) Discrete-Time optimal control and games on large intervals. Springer
Optimization and Its Applications, Springer, 2017.
87. Zaslavski AJ, Leizarowitz A (1997) Optimal solutions of linear control systems with nonperi-
odic integrands, Math. Op. Res.: 22, 726–746.
Index
A E
Agreeable program, 58–62 Euclidean norm, 1
Approximate solution, 6, 8, 9, 33, 35, 36, 73, Euclidean space, 1
74, 82, 90, 99
Asymptotic turnpike property, 65
Autonomous discrete-time control system, 8 F
Forest management problem, 9, 11–16, 18, 20,
21, 30–31, 33, 35, 64, 109–130
B
Bad program, 35, 65 G
Baire category approach, 63 Generic turnpike property, 2, 6, 8, 9, 63–71
Banach space, 67, 89, 98 Good program, 35, 36, 65, 66, 68, 71, 74, 124,
Bolza problem, 89–90, 98–99 128, 130
C I
Cardinality of a set, 6, 35 Increasing function, 124
Compact metric space, 8, 16, 33, 63, 73 Infinite horizon, 9, 11–31, 33
Compact set, 13, 111 Infinite horizon optimal control problem,
Complete metric space, 63, 67 11–31
Concave function, 13, 20, 118, 119 Inner product, 1
Constrained problem, 8 Interior point, 9
Continuous-time problem, 1
Control system, 2, 8, 34, 73, 111
Convex discrete-time problems, 1–7 L
Convex problems, 1 Lagrange problem, 78–83
Convex set, 119 Locally optimal program, 37, 50, 58, 78
D N
Differentiable function, 2, 3, 8 Neumann path, 2
Discrete-time problem, 1–8 Normed space, 20
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2019 135
A. J. Zaslavski, Optimal Control Problems Arising in Forest Management,
SpringerBriefs in Optimization, https://doi.org/10.1007/978-3-030-23587-1
136 Index
O S
Objective function, 2, 8, 63, 73 Smooth function, 13
Optimal control problem, 1, 2, 8, 9, 11–31, 33, Strictly convex function, 2, 3, 13
63, 73, 82, 123 Strongly agreeable program, 58
Optimality criterion, 8, 20, 28, 33, 37, 73
Overtaking optimal program, 28, 50, 74, T
75, 77, 78, 82–86, 90–94, 98–100, Turnpike, 7–9, 73
102–104, 108 Turnpike phenomenon, 8–9, 33
Turnpike property, 1, 2, 6, 8, 9, 33–71
Turnpike result, 2, 7, 8, 33, 55, 65, 118–122
P
Program, 8, 12–19, 21–23, 26–30, 33–45, U
47–56, 58–61, 63–71, 73–87, 89–95, Unconstrained problem, 8
98–100, 102–105, 108, 110–112, 114, Upper semicontinuous function, 8, 13, 28, 31,
115, 117–120, 124, 128–130 53, 63, 73, 90, 98, 99, 111