Professional Documents
Culture Documents
Alexander J. Zaslavski
Structure of
Approximate
Solutions of
Optimal Control
Problems
123
SpringerBriefs in Optimization
Series Editors
Panos M. Pardalos
János D. Pintér
Stephen M. Robinson
Tamás Terlaky
My T. Thai
Structure of Approximate
Solutions of Optimal Control
Problems
123
Alexander J. Zaslavski
Department of Mathematics
Technion – Israel Institute of Technology
Haifa, Israel
v
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1 Infinite Horizon Variational Problems . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.2 The Turnpike Phenomenon . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 5
1.3 Structure of Solutions of Variational Problems .. . .. . . . . . . . . . . . . . . . . . . . 7
2 Turnpike Properties of Optimal Control Problems . .. . . . . . . . . . . . . . . . . . . . 11
2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
2.2 The Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 13
2.3 Uniform Boundedness of Trajectory-Control Pairs . . . . . . . . . . . . . . . . . . . 17
2.4 Auxiliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 22
2.5 Proofs of Theorems 2.1–2.3 and 2.5 . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 68
3 Infinite Horizon Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 77
3.1 Existence of Optimal Solutions . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 77
3.2 Auxiliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 82
3.3 Proof of Proposition 3.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 85
3.4 Proof of Proposition 3.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 89
3.5 Proof of Theorem 3.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 91
3.6 Proof of Theorem 3.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 97
3.7 Proof of Theorem 3.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 100
4 Linear Control Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 105
4.1 The Class of Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 105
4.2 Proof of Proposition 4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 107
4.3 A Continuity Property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 111
4.4 A Boundedness Property . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 116
4.5 The Existence and Structure of Solutions .. . . . . . . . .. . . . . . . . . . . . . . . . . . . . 118
References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 125
Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 127
vii
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 [4, 5, 8–13, 16, 18, 22–24, 27, 34–38, 40, 45–47, 50]. These problems arise in
engineering [1, 25, 53], in models of economic growth [2, 3, 12, 15, 17, 24, 28, 33, 39,
41–43,50], in infinite discrete models of solid-state physics related to dislocations in
one-dimensional crystals [7, 44], in the calculus of variations on time scales [29, 32]
and in the theory of thermodynamical equilibrium for materials [14, 26, 30, 31].
Consider the infinite horizon problem of minimizing the expression
T
f (t, x(t), x (t))dt
0
|f (t, x1 , u1 ) − f (t, x2 , u2 )| ≤
In Chap. 1 of [50] and in [48] we studied the set of (f )-good functions and proved
the following results.
Theorem 1.1. For each h ∈ M and each z ∈ Rn there exists an (h)-good function
Z h : [0, ∞) → Rn satisfying Z h (0) = z such that:
1. For each f ∈ M, each z ∈ Rn and each a. c. function y : [0, ∞) → Rn one
of the following properties holds:
(i) I f (0, T, y) − I f (0, T, Z f ) → ∞ as T → ∞;
(ii) sup{|I f (0, T, y) − I f (0, T, Z f )| : T ∈ (0, ∞)} < ∞,
I g (T1 , T2 , Z g ) ≤ I g (T1 , T2 , y) + Q.
Our next result, which was also proved in Chap. 1 of [50] and in [48], shows that
for every bounded set E ⊂ Rn the C([0, T ]) norms of approximate solutions x :
1.2 The Turnpike Phenomenon 5
[0, T ] → Rn for the minimization problem on an interval [0, T ] with x(0), x(T ) ∈
E, are bounded by some constant which does not depend on T .
Theorem 1.4. Let f ∈ M and M1 , M2 , c be positive numbers. Then there exist a
neighborhood U of f in M and a number S > 0 such that for each g ∈ U , each
T1 ∈ [0, ∞) and each T2 ∈ [T1 + c, ∞) the following properties hold:
(i) If x, y ∈ Rn satisfy |x|, |y| ≤ M1 and if an a. c. function v : [T1 , T2 ] → Rn
satisfies
then
|v(t)| ≤ S, t ∈ [T1 , T2 ].
then
|v(t)| ≤ S, t ∈ [T1 , T2 ].
The results presented in this section are important ingredients in the proofs of
turnpike results for variational problems [47, 48, 50].
T2 T2
f (t, u(t), u (t))dt ≤ f (t, z(t), z (t))dt + δ
T1 T1
T2 − T1 ≥ L, |u(Ti )| ≤ K, i = 1, 2,
then
ψ(t) → ∞ as t → ∞.
We use the notation and definitions introduced in the previous sections. We con-
sider the space of integrands M introduced in Sect. 1.1.
We equip the set M with two topologies where one is weaker than the other.
We refer to them as the weak and the strong topologies, respectively. For the set M
we consider the uniformity determined by the following base:
where > 0. It is not difficult to see that the uniform space M with this uniformity
is metrizable and complete. This uniformity generates in M the strong topology.
We also equip the set M with the uniformity which is determined by the
following base:
where N > 0, > 0, λ > 1. This uniformity which was introduced in Sect. 1.1
generates in M the weak topology.
In Chap. 2 of [50] and in [49] we established the existence of a set F ⊂ M
which is a countable intersection of open (in the weak topology) everywhere
dense (in the strong topology) subsets of M such that the following theorems are
valid.
8 1 Introduction
Theorem 1.5.
1. For each g ∈ F and each pair of (g)-good functions vi : [0, ∞) → Rn , i = 1, 2,
Theorems 1.5 and 1.6 establish the existence of (g)-overtaking optimal functions
and describe the asymptotic behavior of (g)-good functions for g ∈ F .
Theorem 1.7. Let g ∈ F , S1 , S2 , > 0 and let Y : [0, ∞) → Rn be a (g)-
overtaking optimal function. Then there exists a neighborhood U of g in M with
the weak topology, a number L > 0, and an integer Q ≥ 1 such that if h ∈ U,
T1 ∈ [0, ∞), T2 ∈ [T1 + LQ, ∞) and if an a. c. function v : [T1 , T2 ] → Rn satisfies
one of the following relations:
q ≤ Q, bi < di ≤ bi + L, i = 1, . . . , q,
|v(t) − Y (t)| ≤ for each t ∈ [T1 , T2 ] \ ∪qi=1 [bi , di ].
2.1 Preliminaries
where I is either R1 or [T1 , ∞) or [T1 , T2 ] (−∞ < T1 < T2 < ∞), and x : I → Rn
is an absolutely continuous (a. c.) function which satisfies
We suppose that M is a Borel measurable subset of Rn+m+1 and that the function
G : M → Rn is borelian.
For any t ∈ R1 set
ψ(t) → ∞ as t → ∞.
Denote by Mreg the set of all functions f ∈ M which satisfy the following
assumption:
(B) There exist a trajectory-control pair
xf : R1 → Rn , uf : R1 → Rm
(iii) For each S1 > 0 there exist S2 > 0 and an integer c > 0 such that
I f (T1 , T2 , xf , uf ) ≤ I f (T1 , T2 , x, u) + S2
|z − xf (T )| ≤ δ
there are
τ1 ∈ (T, T + bf ] and τ2 ∈ [T − bf , T ),
14 2 Turnpike Properties of Optimal Control Problems
x1 : [T, τ1 ] → Rn , u1 : [T, τ1 ] → Rm ,
x2 : [τ2 , T ] → Rn , u2 : [τ2 , T ] → Rm
which satisfy
x1 (T ) = x2 (T ) = z,
xi (τi ) = xf (τi ), i = 1, 2,
|x1 (t) − xf (t)| ≤ for all t ∈ [T, τ1 ],
|x2 (t) − xf (t)| ≤ for all t ∈ [τ2 , T ],
I f (T, τ1 , x1 , u1 ) ≤ I f (T, τ1 , xf , uf ) + ,
I f (τ2 , T, x2 , u2 ) ≤ I f (τ2 , T, xf , uf ) + .
Note that assumption (B) means that the trajectory-control pair
xf : R1 → Rn , uf : R1 → Rm
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
I f (T1 , T2 , xf , uf ) ≤ I f (T1 , T2 , x, u) + S.
x : [s, ∞) → Rn , u : [s, ∞) → Rm
I f (s, t, x, u) − I f (s, t, xf , uf ) → ∞ as t → ∞;
(b)
It is easy to see that fr ∈ Mreg for each f ∈ Mreg and each r > 0.
Let A be a subset of Mreg such that fr ∈ A for each f ∈ A and each r ∈ (0, 1).
Denote by Ā the closure of A in the uniform space M and consider the topological
subspace Ā ⊂ M with the relative topology.
In this chapter we will establish the existence of a set F ⊂ Ā which is a countable
intersection of open everywhere dense sets in Ā and such the following theorems
hold.
Theorem 2.2. For each f ∈ F and each S > 0 there exist a neighborhood U of f
in M and positive numbers δ, Q such that the following assertions hold:
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
and
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
and
x : R1 → Rn , u : R1 → Rm
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
and
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
inf{|x(t)| : t ∈ [T1 , T2 ]} ≤ M1 ,
I f (T1 , T2 , x, u) ≤ M0 (2.10)
18 2 Turnpike Properties of Optimal Control Problems
Proof. Fix
[recall a0 in assumption (A)]. By assumption (A) there exist h0 > M1 + 1 such that
f (t, x, u) ≥ 4(M0 + a0 τ0 )δ −1
for each (t, x, u) ∈ M satisfying |x| ≥ h0 (2.13)
Choose a number
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
By the choice of h0 , (2.10), (2.12), (2.13), and assumption (A), there exists t1 ∈
[T1 , T2 ] such that
There exists
t2 ∈ [min{t0 , t1 }, max{t0 , t1 }]
such that
We define
E1 = {t ∈ [min{t1 , t2 }, max{t1 , t2 }] :
|G(t, x(t), u(t))| ≥ a0 |x(t)| + γ0 },
E2 = [min{t1 , t2 }, max{t1 , t2 }] \ E1 . (2.19)
It follows from (2.1), (2.19), (2.17), (2.18), (2.14), (2.10), and assumption (A) that
t2
|x(t2 ) − x(t1 )| = | G(t, x(t), u(t))dt|
t1
t2
≤ a0 | |x(t)|dt| + (|G(t, x(t), u(t))| − a0 |x(t)|)+ dt
t1 E1
+ (|G(t, x(t), u(t))| − a0 |x(t)|)+ dt
E2
≤ a0 |x(t2 )|δ + γ0 δ + 8−1 f (t, x(t), u(t))dt
E1
Together with (2.12), (2.18), (2.16), and (2.17) this implies that
This contradicts (2.15). The contradiction we have reached proves Proposition 2.6.
Proposition 2.6 and assumption (A) imply the following result.
Proposition 2.7. Let M1 > 0 and 0 < τ0 < τ1 . Then there exists M2 > 0 such
that for each f ∈ M, each T1 ∈ R1 , each T2 ∈ [T1 + τ0 , T1 + τ1 ] and each
trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
I f (T1 , T2 , x, u) ≤ M1 (2.20)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
20 2 Turnpike Properties of Optimal Control Problems
which satisfies (2.20) and each t1 , t2 ∈ [T1 , T2 ] satisfying |t1 −t2 | ≤ δ the inequality
|x(t1 ) − x(t2 )| ≤
holds.
Proof. Let a number M2 > 0 be as guaranteed in Proposition 2.7. By assump-
tion (A) there exists h0 > 0 such that
Fix a number
Let f ∈ M, T1 ∈ R1 , T2 ∈ [T1 + τ0 , T1 + τ1 ],
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
Set
It follows from the choice of M2 , (2.20), (2.11), (2.23), (2.24), (2.21), (2.22), and
(2.1) that
t2 t2
|x(t2 ) − x(t1 )| ≤ a0 |x(t)|dt + (|G(t, x(t), u(t))| − a0 |x(t)|)+ dt
t1 t1
≤ a0 δM2 + δh0 + (|G(t, x(t), u(t))| − a0 |x(t)|)dt
E1
≤ a0 δM2 + δh0 + (4(M1 + a0 τ1 + 8))−1 f (t, x(t), u(t))dt
E1
Proposition 2.9. Let f ∈ M, 0 < c1 < c2 , and D, > 0. Then there exists
a neighborhood V of f in M such that for each g ∈ V , each T1 ∈ R1 , each
T2 ∈ [T1 + c1 , T1 + c2 ], and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
the inequality
holds.
Proof. By Proposition 2.7 there exists S > 0 such that for each g ∈ M, each
T1 ∈ R1 , each T2 ∈ [T1 + c1 , T1 + c2 ], and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
I g (T1 , T2 , x, u) ≤ D + 1
Define
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
Set
It follows from (2.28), (2.29), (2.27), and the definition of S that (2.25) holds
and that
Define
for all t ∈ [T1 , T2 ]. By (2.25), (2.27), (2.29), (2.26), and assumption (A), for each
t ∈ E2 ,
and
It follows from (2.29), (2.30), assumption (A), (2.28), (2.26) and the inequality
above that
|I f (T1 , T2 , x, u) − I g (T1 , T2 , x, u)| ≤ δc2 + (Γ − 1) h(t)dt + (Γ − 1)c2
E2
Consider the control system described by (2.1)–(2.5). Suppose that f ∈ M and that
there exists a trajectory-control pair
x∗ : R1 → Rn , u∗ : R1 → Rm
such that:
(i)
(iii) For each S1 > 0 there exist S2 > 0 and an integer c > 0 such that
I f (T1 , T2 , x∗ , u∗ ) ≤ I f (T1 , T2 , x, u) + S2
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
|z − x∗ (T )| ≤ δ
there are
τ1 ∈ (T, T + b∗ ] and τ2 ∈ [T − b∗ , T ),
x1 : [T, τ1 ] → Rn , u1 : [T, τ1 ] → Rm ,
x2 : [τ2 , T ] → Rn , u2 : [τ2 , T ] → Rm
which satisfy
x1 (T ) = x2 (T ) = z,
xi (τi ) = x∗ (τi ), i = 1, 2,
|x1 (t) − x∗ (t)| ≤ for all t ∈ [T, τ1 ],
|x2 (t) − x∗ (t)| ≤ for all t ∈ [τ2 , T ],
I f (T, τ1 , x1 , u1 ) ≤ I f (T, τ1 , x∗ , u∗ ) + ,
I f (τ2 , T, x2 , u2 ) ≤ I f (τ2 , T, x∗ , u∗ ) + .
x∗ : R1 → Rn , u∗ : R1 → Rm
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm ,
S2 > S0 ,
ψ(S2 ) > a0 + sup{|I f (j, j + 1, x∗ , u∗ )| : j = 0, ±1, ±2, . . . }. (2.33)
By property (iii) there exists a number S1 > 0 and an integer c > 0 such that for
each T1 ∈ R1 , each T2 ≥ T1 + c, and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
I f (T1 , T2 , x∗ , u∗ ) ≤ I f (T1 , T2 , x, u) + S1
holds.
Fix a number
S > S1 + 2 + 2a0 (2 + c)
+(c + 4) sup{|I f (j, j + 1, x∗ , u∗ )| : j = 0, ±1, ±2, . . . }. (2.34)
Let T1 ∈ R1 , T2 ≥ T1 + c and
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
Set
It follows from the choice of S1 , (2.36), (2.33), and assumption (A) that
I f (T1 , T3 , x∗ , u∗ ) − I f (T1 , T3 , x, u)
≤ S1 + 2a0 (1 + c) + (c + 2) sup{|I f (j, j + 1, x∗ , u∗ )| : j = 0, ±1, ±2, . . . }.
(2.37)
I f (T3 , T2 , x∗ , u∗ ) − I f (T3 , T2 , x, u)
≤ (T2 − T3 + 2) sup{|I f (j, j + 1, x∗ , u∗ )| : j = 0, ±1, ±2, . . . }
+2a0 − (ψ(S2 ) − a0 )(T2 − T3 )
≤ 2 sup{|I f (j, j + 1, x∗ , u∗ )| : j = 0, ±1, ±2, . . . }.
Together with (2.34) and (2.37) this implies (2.32). Lemma 2.10 is proved.
1
Fix a number d0 > 0 and define a continuous function φ : A → R by
Lemma 2.12. Let b1 ≥ 1 be an integer and let b2 , Q, S > 0. Then there exists
S1 > S such that
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
and
such that
S0 > S + 1,
f
ψ(S0 ) > a0 + sup{|I (j, j + 1, x∗ , u∗ )| : j = 0, ±1, ±2, . . . } + 4. (2.43)
By property (iii), there exist Q1 > 0 and an integer c1 ≥ 1 such that for each
T1 ∈ R1 , each T2 ≥ T1 + c1 , and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti )| ≤ S0 + 1, i = 1, 2
Fix a number
By Proposition 2.6 there exists S1 > S0 +2 such that for each g ∈ M, each T1 ∈ R1 ,
each
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
inf{|x(t)| : t ∈ [T1 , T2 ]} ≤ S0 ,
I g (T1 , T2 , x, u) ≤ S̃ (2.46)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies conditions (a) and (b). Let τ1 and τ2 be as guaranteed in condition (b). We
show that (2.41) holds.
Assume the contrary. Then there exists T0 ∈ [T1 , T2 ] such that
x1 : [T1 , T2 ] → Rn , u1 : [T1 , T2 ] → Rm
such that
It follows from (2.49) conditions (a) and (b), and assumption (A) that
I fr (T1 , T2 , x, u) ≤ I fr (T1 , T2 , x1 , u1 ) + Q
≤ Q + I fr (T1 , T2 , x∗ , u∗ ) + 2a0 b1 + 2b2 + 2. (2.50)
28 2 Turnpike Properties of Optimal Control Problems
Set
It follows from the choice of Q1 and c1 , (2.44), (2.51), (2.40), condition (a), and
assumption (A) that
I fr (T1 , t1 , x∗ , u∗ ) − I fr (T1 , t1 , x, u)
≤ Q1 + a0 (c1 + 2) + (c1 + 2)Λ0 ,
I fr (t2 , T2 , x∗ , u∗ ) − I fr (t2 , T2 , x, u)
≤ Q1 + a0 (c1 + 2) + (c1 + 2)Λ0 . (2.52)
I fr (t1 , t2 , x∗ , u∗ ) − I fr (t1 , t2 , x, u)
≤ (t2 − t1 + 2)Λ0 + 2a0 − (t2 − t1 )(ψ(S0 ) − a0 ) ≤ 2Λ0 + 2a0 − 4(t2 − t1 ).
t2 − t1
≤ 2Λ0 + 2a0 + Q + 2a0 b1 + 2b2 + 2 + 2Q1 + 2a0 (c1 + 2) + 2(c1 + 2)Λ0 .
(2.53)
It follows from the choice of S1 , (2.53), (2.46), (2.47), (2.48), and (2.51) that
Together with (2.39), (2.50), (2.52), (2.40), and assumption (A) this implies that
−Q − 2a0 b1 − 2b2 − 2
≤ I fr (T1 , T2 , x∗ , u∗ ) − I fr (T1 , T2 , x, u)
≤ 2Q1 + 2a0 (c1 + 2) + 2(c1 + 2)Λ0
+I fr (t1 , t2 , x∗ , u∗ ) − I fr (t1 , t2 , x, u)
≤ 2Q1 + 2a0 (c1 + 2) + 2(c1 + 2)Λ0 + Λ0 (t2 − t1 + 2) + 2a0 − S̃.
2.4 Auxiliary Results 29
This contradicts (2.45). The contradiction we have reached proves Lemma 2.12.
Lemma 2.13. Let b1 ≥ 1 be an integer and let b2 , Q, S > 0. Then there exists
S0 > 0 such that for each T1 ∈ R1 , each T2 ≥ T1 + 2b1 , and each trajectory-
control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies conditions (a) and (b) of Lemma 2.12 the following inequality holds:
Proof. Let S1 > S be as guaranteed in Lemma 2.12. By property (iii), there exist
Q1 > 0 and an integer c1 ≥ 1 such that for each T1 ∈ R1 , each T2 ≥ T1 + c1 , and
each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti )| ≤ S1 + 1, i = 1, 2
Fix a number
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
x1 : [T1 , T2 ] → Rn , u1 : [T1 , T2 ] → Rm
which satisfies (2.49). By (2.49), conditions (a) and (b) of Lemma 2.12 and
assumption (A), relation (2.50) is true. It follows from the choice of S1 and
Lemma 2.12 that
It follows from (2.50), (2.59), (2.57), (2.40), and assumption (A) that
−Q − 2a0 b1 − 2b2 − 2
≤ I fr (T1 , T2 , x∗ , u∗ ) − I fr (T1 , T2 , x, u)
≤ 2Q1 + 2a0 (c1 + 2) + 2(c1 + 2)Λ0
+I fr (T, T + 1, x∗ , u∗ ) − I fr (T, T + 1, x, u)
≤ 2Q1 + 2a0 (c1 + 4) + 2(c1 + 4)Λ0 − S0 .
This contradicts (2.56). The contradiction we have reached proves Lemma 2.13.
Lemma 2.14. Let Δ > 0. Then there exists δ ∈ (0, Δ) such that for each T1 ∈ R1 ,
each T2 > T1 , and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
Proof. There exists δ ∈ (0, Δ) such that property (iv) holds with = 3−1 Δ (see
the definition of f ).
2.4 Auxiliary Results 31
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies (2.60). By (2.60), the choice of δ and property (iv), there exist
x1 : [τ1 , T1 ] → Rn , u1 : [τ1 , T1 ] → Rm
and
x2 : [T2 , τ2 ] → Rn , u2 : [T2 , τ2 ] → Rm
such that
xi (Ti ) = x(Ti ), i = 1, 2,
xi (τi ) = x∗ (τi ), i = 1, 2,
I f (τ1 , T1 , x1 , u1 ) ≤ I f (τ1 , T1 , x∗ , u∗ ) + 3−1 Δ,
I f (T2 , τ2 , x2 , u2 ) ≤ I f (T2 , τ2 , x∗ , u∗ ) + 3−1 Δ. (2.62)
It follows from the choice of (x1 , u1 ), (x2 , u2 ) and property (i) that
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti ) − x∗ (Ti )| ≤ δ, i = 1, 2,
I (T1 , T2 , x, u) ≤ U fr (T1 , T2 , x(T1 ), x(T2 )) + δ
fr
(2.63)
Proof. There exists δ0 ∈ (0, 8−1 ) such that property (iv) holds with = 1 and
δ = δ0 . There exists an integer b1 such that
b∗ + 1 < b1 ≤ b∗ + 2 (2.65)
Q = 1,
S = sup{|x∗ (t)| : t ∈ R1 } + 4,
b2 = 2 + b∗ + 2a0 + (b∗ + 3)Λ0 . (2.66)
x : [T, T + 1] → Rn , u : [T, T + 1] → Rm
which satisfies
holds.
Choose a number
There exists
such that property (iv) (see the definition of f ) holds with = 1 , δ = δ2 . There
exists
δ ∈ (0, δ2 ) (2.70)
such that Lemma 2.14 holds with Δ = 16−1 δ1 r. Fix a number
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
t1 ∈ [T1 , T2 ],
|x(t1 ) − x∗ (t1 )| > . (2.72)
x1 : [T1 , τ1 ] → Rn , u1 : [T1 , τ1 ] → Rm ,
x2 : [τ2 , T2 ] → Rn , u2 : [τ2 , T2 ] → Rm
which satisfy
xi (Ti ) = x(Ti ), i = 1, 2,
xi (τi ) = x∗ (τi ), i = 1, 2,
|x1 (t) − x∗ (t)| ≤ 1 for all t ∈ [T1 , τ1 ],
|x2 (t) − x∗ (t)| ≤ 1 for all t ∈ [τ2 , T2 ],
I f (T1 , τ1 , x1 , u1 ) ≤ I f (T1 , τ1 , x∗ , u∗ ) + 1 ,
I f (τ2 , T2 , x2 , u2 ) ≤ I f (τ2 , T2 , x∗ , u∗ ) + 1 . (2.74)
x3 : [T1 , T2 ] → Rn , u3 : [T1 , T2 ] → Rm
by
It follows from (2.75), (2.39), (2.74), (2.73), (2.65), and the definition of x3 that
I fr (T1 , T2 , x3 , u3 ) − I fr (T1 , T2 , x∗ , u∗ )
τ1
≤ φ(t, x1 (t))dt + I f (T1 , τ1 , x1 , u1 )
T1
T2
−I f (T1 , τ1 , x∗ , u∗ ) + φ(t, x2 (t))dt
τ2
+I f (τ2 , T2 , x2 , u2 ) − I f (τ2 , T2 , x∗ , u∗ )
≤ 21 + 21 b1 .
[d1 , d2 ] ⊂ [T1 , T2 ]
such that
d2 − d1 = 1,
t1 ∈ [d1 , d2 ]. (2.78)
It follows from (2.63), (2.68)–(2.70), (2.73), and (2.77) that the trajectory-control
pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies conditions (a) and (b) of Lemma 2.12 with b1 , b2 , Q, S defined by (2.65)
and (2.66). Therefore by the choice of S0 and Lemma 2.13,
I fr (d1 , d2 , x, u) ≤ S0 . (2.79)
2.4 Auxiliary Results 35
It follows from the choice of δ1 , (2.67), (2.78)–(2.80), and (2.72) that for every
t ∈ [d1 , d2 ] ∩ [t1 − δ1 , t1 + δ1 ]
we have
and
Therefore
T2
r φ(t, x(t))dt ≥ (3/4)rδ1
T1
On the other hand in view of the choice of δ [see (2.70)], (2.63) and Lemma 2.14,
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti ) − x∗ (Ti )| ≤ δ, i = 1, 2,
I (T1 , T2 , x, u) ≤ U g (T1 , T2 , x(T1 ), x(T2 )) + δ
g
(2.81)
36 2 Turnpike Properties of Optimal Control Problems
Proof. There exists δ0 ∈ (0, 8−1 ) such that property (iv) (see the definition of f )
holds with = 1 and δ = δ0 . By Lemma 2.15 there exist
Δ1 ≥ 2b∗ + 2 (2.83)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti ) − x∗ (Ti )| ≤ δ1 , i = 1, 2,
I fr (T1 , T2 , x, u) ≤ U fr (T1 , T2 , x(T1 ), x(T2 )) + δ1 (2.84)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
the inequality
holds.
Assume that
g ∈ U, T1 ∈ R1 , T2 ∈ [T1 + Δ1 , T1 + Δ2 ]
2.4 Auxiliary Results 37
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
x1 : [T1 , τ1 ] → Rn , u1 : [T1 , τ1 ] → Rm ,
x2 : [τ2 , T2 ] → Rn , u2 : [τ2 , T2 ] → Rm
such that
xi (Ti ) = x(Ti ), i = 1, 2,
xi (τi ) = x∗ (τi ), i = 1, 2,
|x1 (t) − x∗ (t)| ≤ 1 for all t ∈ [T1 , τ1 ],
|x2 (t) − x∗ (t)| ≤ 1 for all t ∈ [τ2 , T2 ],
I f (T1 , τ1 , x1 , u1 ) ≤ I f (T1 , τ1 , x∗ , u∗ ) + 1,
I f (τ2 , T2 , x2 , u2 ) ≤ I f (τ2 , T2 , x∗ , u∗ ) + 1. (2.89)
x3 : [T1 , T2 ] → Rn , u3 : [T1 , T2 ] → Rm
by
It follows from (2.91), (2.83), (2.81), and the choice of U [see (2.86), (2.87)] that
and
Now (2.82) follows from the relation above, (2.81), (2.85) and the choice of δ1 , Δ1
[see (2.83), (2.84)]. This completes the proof of Lemma 2.16.
Lemma 2.17. Let S > 0 and δ ∈ (0, min{1, d0 }). Then there exist Δ ≥ 1 and
a neighborhood U of fr in M such that for each g ∈ U, each T1 ∈ R1 , each
T2 ≥ T1 + Δ and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
By Proposition 2.7 there exists D2 > D1 such that for each g ∈ M, each T1 ∈ R1 ,
each T2 ∈ [T1 + 8−1 , T1 + 8], and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
we have
By Lemma 2.10 there exist D3 > D2 and an integer c1 ≥ 1 such that for each
T1 ∈ R1 , each T2 ≥ T1 + c1 , and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
holds.
Choose numbers
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
the inequality
holds.
Fix a number
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies (2.92). By assumption (A), (2.40), (2.97), and the choice of U [see (2.98),
(2.99)] for each t1 , t2 ∈ [T1 , T2 ] satisfying t2 − t1 ∈ [2−1 , 2]
p − 1 < T2 − T1 ≤ p. (2.102)
Set
b0 = T1 , bj = T1 + j, j = 1, . . . , p − 2, bp−1 = T2 , τ0 = T1 . (2.103)
then we set τq+1 = bj(q)+1 . If (2.104) does not hold and there exists an integer
k ∈ {j(q) + 1, . . . , p − 2} such that
We estimate
If (2.204) holds, then by (2.101) which holds with t1 = τq , t2 = τq+1 and (2.94),
Assume that (2.104) does not hold and there exists k ∈ {j(q) + 1, . . . , p − 2} which
satisfies (2.105). We show that
It follows from (2.105), the choice of τq+1 , (2.101), and (2.97) that
By the choice of U [see (2.98), (2.99)], (2.105), (2.101), and (2.97) for i =
j(q), . . . , k − 1,
Since (2.104) does not hold it follows from the definition of D2 [see (2.95)] that
k − j(q) ≥ S1
we obtain (2.108). If k − j(q) < S1 , then (2.108) follows from (2.109), (2.101),
(2.97), and assumption (A).
Assume that (2.104) does not hold and there is no k ∈ {j(q) + 1, . . . , p − 2}
satisfying (2.105). Then
τq+1 = bp−1 , q = Q − 1,
I g (bi , bi+1 , x, u) < D4 , i = j(q), . . . , p − 2. (2.114)
By the choice of U [see (2.98), (2.99)], (2.114), (2.101), and (2.97), relation (2.110)
holds for i = j(q), . . . , p − 2. Relation (2.110), which holds for i = j(q), . . . , p − 2,
(2.92) and (2.114) imply that
Since (2.104) does not hold it follows from the choice of D2 [see (2.95)] that
I f (τq , T2 , x, u) − I f (τq , T2 , x∗ , u∗ )
≥ −D3 − a0 c1 − Λ0 (c1 + 2) − 2a0 .
I g (τq , T2 , x, u) − I g (τq , T2 , x∗ , u∗ )
≥ 2−1 rδ(T2 − τq ) − D3 − (a0 + Λ0 )(c1 + 2) (2.116)
k ∈ {j(Q − 1) + 1, . . . , p − 2}
Q − 1 = τQ − T1 ,
I g (T1 , T2 , x, u) − I g (T1 , T2 , x∗ , u∗ )
≥ (3/4)(Q − 1)D1 − D3 − (c1 + 2)(Λ0 + a0 )
≥ 2−1 ΔD1 − D3 − (c1 + 2)(Λ0 + a0 ) ≥ 2S
Consider the case (ii). It is easy to see that (2.108) holds. Relation (2.93) follows
from (2.116), which holds with q = Q − 1, (2.108), (2.107), and (2.97). This
completes the proof of Lemma 2.17.
Lemma 2.18. There exists γ > 0 such that for each δ ∈ (0, γ) and each S > 0
there exist Δ ≥ 1 and a neighborhood U of fr in M such that for each g ∈ U, each
T1 ∈ R1 , each T2 ≥ T1 + Δ, and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti ) − x∗ (Ti )| ≤ γ, i = 1, 2,
|x(t) − x∗ (t)| > δ for all t ∈ [T1 , T2 ] (2.118)
the inequality
holds.
Proof. There exists
γ ∈ (0, min{8−1 , d0 })
such that property (iv) (see the definition of f ) holds with = 1 and δ = γ.
Let δ ∈ (0, γ) and S > 0. By Proposition 2.9 there exists a neighborhood U1 of
fr in M such that for each g ∈ U1 , each T1 ∈ R1 , each T2 ∈ [T1 + 8−1 , T1 + 8(b∗ +
2)], and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
the inequality
holds.
Fix a number
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
Fix a number
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
x1 : [T1 , τ1 ] → Rn , u1 : [T1 , τ1 ] → Rm ,
x2 : [τ2 , T2 ] → Rn , u2 : [τ2 , T2 ] → Rm
such that
xi (Ti ) = x(Ti ), i = 1, 2,
xi (τi ) = x∗ (τi ), i = 1, 2,
|x1 (t) − x∗ (t)| ≤ 1 for all t ∈ [T1 , τ1 ],
|x2 (t) − x∗ (t)| ≤ 1 for all t ∈ [τ2 , T2 ],
I f (T1 , τ1 , x1 , u1 ) ≤ I f (T1 , τ1 , x∗ , u∗ ) + 1,
I f (τ2 , T2 , x2 , u2 ) ≤ I f (τ2 , T2 , x∗ , u∗ ) + 1. (2.127)
2.4 Auxiliary Results 45
x3 : [T1 , T2 ] → Rn , u3 : [T1 , T2 ] → Rm
by
I fr (T1 , T1 + 1 + b∗ , x3 , u3 ) ≤ I f (T1 , T1 + 1 + b∗ , x3 , u3 ) + b∗
≤ I f (T1 , T1 + b∗ + 1, x∗ , u∗ ) + b∗ + 1
and
I fr (T2 − 1 − b∗ , T2 , x3 , u3 ) ≤ I f (T2 − 1 − b∗ , T2 , x3 , u3 ) + b∗
≤ I f (T2 − 1 − b∗ , T2 , x∗ , u∗ ) + b∗ + 1.
It follows from (2.118), (2.124), and the choice of U and Δ1 [see (2.122), (2.123)]
that
It follows from this relation, (2.128), (2.126), (2.130), (2.121), and assumption (A)
that
I g (T1 , T2 , x, u) − I g (T1 , T2 , x3 , u3 )
> S1 − 2[1 + Λ0 (4 + b∗ ) + 4a0 + b∗ + 1] − 2a0 (1 + b∗ ) > 2S + 1.
Together with (2.129) this implies (2.125). This completes the proof of Lemma 2.18.
46 2 Turnpike Properties of Optimal Control Problems
Lemma 2.19. Let ∈ (0, min{1, d0 }). Then there exist δ ∈ (0, ), Δ ≥ 1 and
a neighborhood U of fr in M such that for each g ∈ U, each T1 ∈ R1 , each
T2 ≥ T1 + Δ, and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti ) − x∗ (Ti )| ≤ δ, i = 1, 2,
I g (T1 , T2 , x, u) ≤ U g (T1 , T2 , x(T1 ), x(T2 )) + δ (2.131)
Proof. There exist Δ1 ≥ 1 and δ0 ∈ (0, ) such that Lemma 2.16 holds with δ = δ0 .
Let a number γ > 0 be as guaranteed in Lemma 2.18. Choose a number
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti ) − x∗ (Ti )| ≤ γ, i = 1, 2,
|x(t) − x∗ (t)| > δ for all t ∈ [T1 , T2 ] (2.133)
the inequality
holds.
Since Lemma 2.16 holds with δ = δ0 and Δ1 there exists a neighborhood U of
fr in M such that
U ⊂ U1
2.4 Auxiliary Results 47
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti ) − x∗ (Ti )| ≤ δ0 , i = 1, 2,
I (T1 , T2 , x, u) ≤ U g (T1 , T2 , x(T1 ), x(T2 )) + δ0
g
(2.135)
Choose a number
Assume that
g ∈ U, T1 ∈ R1 , T2 ≥ T1 + Δ
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
t2 ∈ [t1 + Δ1 , t1 + Δ1 + Δ2 ],
|x(t2 ) − x∗ (t2 )| ≤ δ. (2.139)
Then
There exist
such that
This contradicts (2.131). The contradiction we have reached proves that for each
number t1 satisfying (2.138) there exists a number t2 which satisfies (2.139).
Therefore there exists a sequence of numbers {tj }Q
j=1 such that
t1 = T 1 , tQ = T 2 ,
tj+1 − tj ∈ [Δ1 , 2Δ1 + 2Δ2 + 4], j = 1, . . . , Q − 1,
|x(tj ) − x∗ (tj )| ≤ δ, j = 1, . . . , Q. (2.144)
In view of (2.131),
It follows from this inequality, (2.132), (2.144) and the choice of U [see (2.133),
(2.136)] that for all j = 1, . . . , Q − 1,
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
2.4 Auxiliary Results 49
which satisfies
Moreover, if
|x(T1 ) − x∗ (T1 )| ≤ δ,
then
and if
|x(T2 ) − x∗ (T2 )| ≤ δ,
then
Proof. By property (ii) (see the definition of f ), assumption (A) and Proposition 2.9,
there exists a neighborhood U0 of fr in M such that for each g ∈ U0 , each T1 ∈ R1 ,
each T2 ≥ T1 + 8−1 ,
There exists δ1 > 0 such that property (iv) (see the definition of f ) holds with = 1,
δ = δ1 . By Proposition 2.9 there exists a neighborhood U1 of fr in M such that
U1 ⊂ U0
and for each g ∈ U1 , each T1 ∈ R1 , each T2 ∈ [T1 + 4−1 b∗ , T1 + 4b∗ ], and each
trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
50 2 Turnpike Properties of Optimal Control Problems
which satisfies
the inequality
holds.
By Lemma 2.19 there exist
U2 ⊂ U1
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
|x(Ti ) − x∗ (Ti )| ≤ δ, i = 1, 2,
I g (T1 , T2 , x, u) ≤ U g (T1 , T2 , x(T1 ), x(T2 )) + δ (2.151)
U ⊂ U2
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
Fix a number
Assume that
g ∈ U, T1 ∈ R1 , T2 ≥ T1 + 2Δ
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies (2.145).
By the definition of Δ1 , U [see (2.153)], (2.145) and (2.154) there exists t̃ ∈
[T1 , T2 ] which satisfies
|x(t̃) − x∗ (t̃)| ≤ δ.
Set
It follows from (2.155), property (iv), and the choice of δ1 that there exist
x1 : [τ1 , t1 ] → Rn , u1 : [τ1 , t1 ] → Rm ,
x2 : [t2 , τ2 ] → Rn , u2 : [t2 , τ2 ] → Rm
such that
xi (ti ) = x(ti ), i = 1, 2,
xi (τi ) = x∗ (τi ), i = 1, 2,
|x1 (t) − x∗ (t)| ≤ 1 for all t ∈ [τ1 , t1 ],
52 2 Turnpike Properties of Optimal Control Problems
τ2 = t2 + b∗ , τ1 = t1 − b∗ . (2.158)
I g (τ1 , t1 , x1 , u1 ) ≤ I g (τ1 , t1 , x∗ , u∗ ) + b∗ + 3,
I g (t2 , τ2 , x2 , u2 ) ≤ I g (t2 , τ2 , x∗ , u∗ ) + b∗ + 3. (2.160)
We show that
t1 − T1 ≤ 8−1 Δ.
x3 : [T1 , T2 ] → Rn , u3 : [T1 , T2 ] → Rm
by
τ1 − T1 ≥ 2Δ0 + 2Δ1 + 2,
|x(t) − x∗ (t)| ≥ δ for all t ∈ [T1 , τ1 ].
It follows from this inequality, (2.162), (2.160), (2.158), and assumption (A) that
t1 − T1 ≤ 8−1 Δ. (2.163)
T2 − t2 ≤ 8−1 Δ. (2.164)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
54 2 Turnpike Properties of Optimal Control Problems
which satisfies
such that property (iv) (see the definition of f ) holds with = 1, δ = δ0 . Fix
S1 > S + 1. (2.169)
There exist
Δ0 ≥ 16 + 8b∗ , δ1 ∈ (0, δ0 )
S = 4S1 + 8, = δ0 , Δ = Δ0 ,
δ = δ1 , U = U1 . (2.170)
U ⊂ U1 ∩ U2
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
the inequality
holds.
By Proposition 2.7 there exists
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
Choose numbers
Assume that
g ∈ U, T1 ∈ R1 , T2 ≥ T1 + 1
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies (2.166). We show that (2.167) holds. There are two cases:
(1) T2 − T1 ≥ 4Δ0 + 4;
(2) T2 − T1 < 4Δ0 + 4.
Consider the case (1). It follows from Lemma 2.20 and the definition of Δ0 ≥ 1,
δ1 ∈ (0, δ0 ) [see (2.170)], (2.166), (2.176), and (2.169) that
x1 : [τ1 , T1 + Δ0 ] → Rn , u1 : [τ1 , T1 + Δ0 ] → Rm ,
x2 : [T2 − Δ0 , τ2 ] → Rn , u2 : [T2 − Δ0 , τ2 ] → Rm
such that
xi (τi ) = x∗ (τi ), i = 1, 2,
x1 (T1 + Δ0 ) = x(T1 + Δ0 ),
x2 (T2 − Δ0 ) = x(T2 − Δ0 ),
I f (τ1 , T1 + Δ0 , x1 , u1 ) ≤ I f (τ1 , T1 + Δ0 , x∗ , u∗ ) + 1,
I f (T2 − Δ0 , τ2 , x2 , u2 ) ≤ I f (T2 − Δ0 , τ2 , x∗ , u∗ ) + 1,
|x1 (t) − x∗ (t)| ≤ 1 for all t ∈ [τ1 , T1 + Δ0 ],
|x2 (t) − x∗ (t)| ≤ 1 for all t ∈ [T2 − Δ0 , τ2 ]. (2.179)
τ1 = T1 + Δ0 − b∗ , τ2 = T2 − Δ0 + b∗ . (2.180)
x3 : [T1 , T2 ] → Rn , u3 : [T1 , T2 ] → Rm
and
x4 : [T1 , T2 ] → Rn , u4 : [T1 , T2 ] → Rm
by
I fr (T1 , T1 + Δ0 , x3 , u3 ) ≤ I f (T1 , T1 + Δ0 , x3 , u3 ) + b∗
≤ I f (T1 , T1 + Δ0 , x∗ , u∗ ) + b∗ + 1,
I fr (T2 − Δ0 , T2 , x4 , u4 ) ≤ I f (T2 − Δ0 , , T2 , x4 , u4 ) + b∗
≤ I f (T2 − Δ0 , T2 , x∗ , u∗ ) + b∗ + 1. (2.182)
I g (T1 , T1 + Δ0 , x3 , u3 ) ≤ I f (T1 , T1 + Δ0 , x∗ , u∗ ) + b∗ + 3,
I g (T2 − Δ0 , T2 , x4 , u4 ) ≤ I f (T2 − Δ0 , T2 , x∗ , u∗ ) + b∗ + 3. (2.183)
S ≥ I g (T1 , T2 , x, u) − I g (T1 , T2 , x3 , u3 )
= I g (T1 , T1 + Δ0 , x, u) − I g (T1 , T1 + Δ0 , x3 , u3 )
≥ I g (T1 , T1 + Δ0 , x, u) − b∗ − 3 − Λ0 (Δ0 + 2) − 2a0
and
Analogously,
Together with (2.177), (2.173), (2.168), and (2.176) this implies (2.167).
58 2 Turnpike Properties of Optimal Control Problems
Consider the case (2). By (2.166), the choice of U [see (2.171), (2.172)], and
assumption (A),
I g (i, i + 1, x∗ , u∗ ) ≤ Λ0 + 1, i = 0, 1, . . .
and
I g (T1 , T2 , x, u) ≤ I g (T1 , T2 , x∗ , u∗ ) + S
≤ S + 2a0 + (Λ0 + 1)(4Δ0 + 6).
Relation (2.167) follows from the relation above, the choice of D1 , (2.175) and
(2.176). This completes the proof of Lemma 2.22.
Lemma 2.23. Let 0 , S > 0. Then there exist Δ > 0 and a neighborhood U of
fr in M such that for each g ∈ U, each T1 ∈ R1 , each T2 ≥ T1 + Δ, and each
trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
We may also assume without loss of generality that property (iv) holds with = 1,
δ = 0 .
By property (ii) and assumption (A) there is Λ̃ > 0 such that
x : [S1 , S2 ] → Rn , u : [S1 , S2 ] → Rm
which satisfies
the inequality
holds.
By (2.188) and Lemma 2.17 there exist Δ1 ≥ 1 and a neighborhood U of fr in
M such that
U ⊂ U1
y : [S1 , S2 ] → Rn , v : [S1 , S2 ] → Rm
which satisfies
Choose a number
Δ ≥ Δ1 + 2b∗ + 2. (2.194)
Assume that
a trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
60 2 Turnpike Properties of Optimal Control Problems
τ ∈ [T1 , T2 − Δ].
I g (T1 , T2 , x, u) < ∞.
In order to complete the proof of the lemma it is sufficient to show that (2.187)
holds.
Assume the contrary. Then
It is not difficult to see that there exist real numbers τ1 , τ2 such that
T1 ≤ τ1 < τ2 ≤ T2 ,
[τ, τ + Δ] ⊂ [τ1 , τ2 ], (2.197)
|x(t) − x∗ (t)| ≥ 0 for all t ∈ [τ1 , τ2 ] (2.198)
Define ξ1 , ξ2 ∈ Rn as follows:
y : [τ1 , τ2 ] → Rn , v : [τ1 , τ2 ] → Rm .
Since property (iv) holds with = 1 and δ = 0 it follows from (2.199)–(2.205) that
there exist trajectory-control pairs
y1 : [τ1 , τ1 + b∗ ] → Rn , v1 : [τ1 , τ1 + b∗ ] → Rm
2.4 Auxiliary Results 61
and
y2 : [τ2 − b∗ , τ2 ] → Rn , v2 : [τ2 − b∗ , τ2 ] → Rm
such that
y1 (τ1 ) = ξ1 ,
y1 (τ1 + b∗ ) = x∗ (τ1 + b∗ ), (2.207)
|y1 (t) − x∗ (t)| ≤ 1 for all t ∈ [τ1 , τ1 + b∗ ], (2.208)
I f (τ1 , τ1 + b∗ , y1 , v1 ) ≤ I f (τ1 , τ1 + b∗ , x∗ , u∗ ) + 1, (2.209)
y2 (τ2 − b∗ ) = x∗ (τ2 − b∗ ),
y2 (τ2 ) = ξ2 , (2.210)
|y2 (t) − x∗ (t)| ≤ 1 for all t ∈ [τ2 − b∗ , τ2 ], (2.211)
I f (τ2 − b∗ , τ2 , y2 , v2 ) ≤ I f (τ2 − b∗ , τ2 , x∗ , u∗ ) + 1. (2.212)
I fr (τ1 , τ1 + b∗ , y1 , v1 ) ≤ I f (τ1 , τ1 + b∗ , y1 , v1 ) + b∗
≤ I f (τ1 , τ1 + b∗ , x∗ , u∗ ) + 1 + b∗ (2.213)
and
I fr (τ2 − b∗ , τ2 , y2 , v2 ) ≤ I f (τ2 − b∗ , τ2 , y2 , v2 ) + b∗
≤ I f (τ2 − b∗ , τ2 , x∗ , u∗ ) + 1 + b∗ . (2.214)
y : [τ1 , τ2 ] → Rn , v : [τ1 , τ2 ] → Rm
such that
I g (τ1 + b∗ , τ2 − b∗ , y, v)
is finite.
62 2 Turnpike Properties of Optimal Control Problems
I fr (τ1 , τ1 + b∗ , y, v) ≤ I f (τ1 , τ1 + b∗ , x∗ , u∗ ) + 1 + b∗ ≤ Λ̃ + 1 + b∗ ,
I fr (τ2 − b∗ , τ2 , y, v) ≤ I f (τ2 − b∗ , τ2 , x∗ , u∗ ) + 1 + b∗ ≤ Λ̃ + 1 + b∗ .
Together with (2.215) and the choice of U1 [see (2.190), (2.191)] this implies that
I g (τ1 , τ1 + b∗ , y, v) ≤ Λ̃ + 2 + b∗ ,
I g (τ2 − b∗ , τ2 , y, v) ≤ Λ̃ + 2 + b∗ . (2.216)
S ≥ I g (τ1 + b∗ , τ2 − b∗ , x, u) − I g (τ1 + b∗ , τ2 − b∗ , x∗ , u∗ )
+I g (τ1 , τ1 + b∗ , x, u) + I g (τ2 − b∗ , τ2 , x, u)
−I g (τ1 , τ1 + b∗ , y, v) − I g (τ2 − b∗ , τ2 , y, v)
≥ I g (τ1 + b∗ , τ2 − b∗ , x, u) − I g (τ1 + b∗ , τ2 − b∗ , x∗ , u∗ )
−2a0 b∗ − 2Λ̃ − 4 − 2b∗
and
I g (τ1 + b∗ , τ2 − b∗ , x, u) − I g (τ1 + b∗ , τ2 − b∗ , x∗ , u∗ )
≤ S + 2b∗ (1 + a0 ) + 4 + 2Λ̃. (2.219)
On the other hand it follows from (2.195), (2.194), (2.197), (2.198) and the choice
of U [see (2.192), (2.193)] that
This contradicts (2.219). The contradiction we have reached completes the proof of
Lemma 2.23.
We suppose that the sum over empty set is zero.
Lemma 2.24. Let S, > 0. Then there exist real numbers l > 0, Δ > 0, a natural
number Q, and a neighborhood U of fr in M such that for each g ∈ U, each
T1 ∈ R1 , each T2 ≥ T1 + Δ, and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
and
δ0 ∈ (0, ) and Δ1 ≥ 1,
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
T1 ∈ R1 , T2 ≥ T1 + Δ2 ,
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
τ ∈ [T1 , T2 − Δ2 ],
we have
Set
U = U1 ∩ U2 , (2.226)
l = 2Δ2 + 2Δ1 (2.227)
and choose
Assume that
g ∈ U, T1 ∈ R1 , T2 ≥ T1 + Δ (2.230)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
2.4 Auxiliary Results 65
satisfies (2.220). By (2.230), (2.226), (2.228), (2.220), and the choice of U2 [see
(2.225)], there exists a strictly increasing sequence of real numbers {ti }pi=1 , where
p is a natural number, such that
Set
S 1 = t1 . (2.233)
such that
Si = tr(i) , i = 1, . . . , k, (2.234)
r(k) = p and Sk = tp (2.235)
then
such that
Si = tr(i) , i = 1, . . . , j,
S j < tp ,
for each integer i satisfying 1 ≤ i < j relation (2.236) holds, and if an integer i
satisfies 1 ≤ i < j and
If
then we set
r(j + 1) = p, Sj+1 = tp ,
complete our construction and it is easy to see that properties (C1) and (C2) hold.
Assume that
Set
and
k ≤ 2 + δ0−1 S. (2.240)
Set
Let
i ∈ A. (2.242)
By (2.246), (2.245), (2.231), and (2.230), (2.226) and the choice of U1 , δ0 , Δ1 [see
(2.222)–(2.224)],
The right-hand side of (2.249) is a finite union of closed intervals. By (2.240) and
(2.229) their number does not exceed
2k + 2 ≤ 6 + 2δ0−1 S ≤ Q
and in view of (2.241), ( 2.232), and (2.227) their maximal length does not exceed
2Δ2 + 2Δ1 = l.
Proof of Theorem 2.1. The validity of assertion 1 follows from Lemma 2.10 and
assumptions (A) and (B).
We will prove assertion 2. Let f ∈ Mreg , s ∈ R1 and
x : [s, ∞) → Rn , u : [s, ∞) → Rm
tk → ∞ as k → ∞, (2.250)
I f (s, tk , x, u) − I f (s, tk , xf , uf ) → ∞ as k → ∞. (2.251)
We will show that relation (a) holds. It follows from assumption (A) and (B) that
there exists a number S0 > 0 such that
S0 > 2|x(s)| + 8,
ψ(S0 − 4) − a0 − 8 ≥ 8 sup{|I f (j, j + 1, xf , uf )| : j = 0, ±1, ±2, . . . }.
(2.252)
I f (s, t, x, u) − I f (s, t, xf , uf )
≥ I f (s, tk , x, u) − I f (s, tk , xf , uf )
+I f (tk , τk , x, u) − I f (tk , τk , xf , uf ) − S
≥ I f (s, tk , x, u) − I f (s, tk , xf , uf ) − S − 4a0
−2 sup{|I f (j, j + 1, xf , uf )| : j = 0, ±1, ±2, . . . } → ∞ as k → ∞.
Therefore relation (a) holds. Together with assertion 1 and Proposition 2.7 this
implies the validity of assertion 2.
Construction of the set F :
Let A be a subset of Mreg such that fr ∈ A for each f ∈ A each r ∈ (0, 1).
Denote by Ā the closure of A in the space M.
It is easy to see that for each f ∈ A and each r ∈ (0, 1) Lemmas 2.12–2.24 hold
with
x∗ = xf , u∗ = uf , b∗ = bf , d0 = 1.
Set
S = k, U = V (f, r, k),
δ = δ(f, r, k), Q = Q(f, r, k);
We define
F = Ā ∩ [∩∞
k=1 ∪ {V (f, r, k) : f ∈ A, r ∈ (0, 1)}]. (2.256)
fk ∈ A and rk ∈ (0, 1)
such that
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
such that
f = fk , r = rk , S = k,
= δ(fk , rk , k), U = V (fk , rk , k),
Δ = Δ(fk , rk , k), δ = γ(fk , rk , k),
x∗ = xfk , u∗ = ufk .
Together with (2.260), (2.261), (2.259), and (2.257) this implies that
and
Since this relation holds for any integer N ≥ 1 we conclude that (2.258) holds.
It follows from (2.258) and (2.255) that there exists a bounded continuous function
Xf : R1 → Rn which satisfies
|Xf (t) − xfp (t)| ≤ δ(fp , rp , p) for all t ∈ R1 and all integers p ≥ 1. (2.262)
72 2 Turnpike Properties of Optimal Control Problems
and set
U = V (fp , rp , p),
Δ = Δ(fp , rp , p), δ = δ(fp , rp , p). (2.264)
Assume that
g ∈ U, T1 ∈ R1 , T2 ≥ T1 + 2Δ (2.265)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
such that
It follows from (2.264), (2.265), (2.266), (2.263), and Lemma 4.20 which holds with
f = fp , r = rp , S = p,
= (2p)−1 , U = V (fp , rp , p),
Δ = Δ(fp , rp , p), δ = 2δ(fp , rp , p),
x∗ = xfp , u∗ = ufp
that
Together with (2.262) and (2.263) this implies the validity of Theorem 2.3.
2.5 Proofs of Theorems 2.1–2.3 and 2.5 73
fk ∈ A, rk ∈ (0, 1)
such that
Set
Assume that
g ∈ U, T1 ∈ R1 , T2 ≥ T1 + L (2.270)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies
By (2.270), (2.271), (2.269), (2.267), condition (e), and Lemma 2.24 which holds
with
= (4k)−1 , S = 4k,
f = fk , r = rk ,
U = V (fk , rk , k),
x∗ = xfk , u∗ = ufk ,
l = l(fk , rk , k), Δ = L(fk , rk , k), Q = p(fk , rk , k)
q ≤ p(fk , rk , k) = p, (2.272)
ai ≤ bi ≤ ai + l for all integers i = 1, . . . , q (2.273)
and
y : [−τ, τ ] → Rn , v : [−τ, τ ] → Rm
satisfying
we have
Let
y : [−T, T ] → Rn , v : [−T, T ] → Rm
satisfy
By (2.277), (2.278), (2.268), condition (a), Lemma 2.20 which holds for fk and rk
with
we have
|y(t) − xfk (t)| ≤ (2k)−1 for all t ∈ [−T + Δ(fk , rk , k), T − Δ(fk , rk , k)].
Together with (2.274) and (2.267) this implies that for all
In this chapter we continue to use the notation and the definitions of Chap. 2.
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
y : [τ, ∞) → Rn , v : [τ, ∞) → Rm
satisfying
y(τ ) = x(τ )
This notion, known as the overtaking optimality criterion, was introduced in the
economics literature [17,43] and has been used in optimal control theory [12,24,49,
50].
A trajectory-control pair
x : I → Rn , u : I → Rm ,
where I is either R1 or [T, ∞) (with T ∈ R1 ) is (f )-good [12, 17, 50] if there exists
a number S > 0 such that for each T1 , T2 ∈ I satisfying T2 > T1 ,
Remark 3.1. It follows from Theorem 2.1 and assumptions (A) and (B) that
xf : R1 → Rn , uf : R1 → Rm
T1 ∈ R1 , T2 ≥ T1 + 2Δ
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
and
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
x(t) − Xf (t) → 0 as t → ∞.
The following optimality criterion for infinite horizon problems was introduced
by Aubry and Le Daeron [7] in their study of the discrete Frenkel–Kontorova model
related to dislocations in one-dimensional crystals.
Let f ∈ M. A trajectory-control pair
x : I → Rn , u : I → Rm ,
3.1 Existence of Optimal Solutions 79
xj : [T1 , T2 ] → Rn , uj : [T1 , T2 ] → Rm , j = 1, 2, . . .
which satisfies
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
such that
and
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
x∗ : [τ, ∞) → Rn , u∗ : [τ, ∞) → Rm
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
80 3 Infinite Horizon Problems
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
x∗ : [τ, ∞) → Rn , u∗ : [τ, ∞) → Rm
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
is called (f )-agreeable [19–21, 52] if for each T0 > τ and each > 0 there exists
T > T0 such that for each T > T there exists a trajectory-control pair
y : [τ, T ] → Rn , v : [τ, T ] → Rm
such that
and
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
3.1 Existence of Optimal Solutions 81
such that
x(τ ) = ξ
The following two turnpike results are proved in Sects. 3.6 and 3.7
respectively.
Theorem 3.7. Assume that f ∈ Mreg has the turnpike property, M > 0 and > 0.
Then there exists L0 > 0 such that for each τ ∈ R1 and each (f )-good and (f )-
minimal trajectory-control pair
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
for which x(τ ) is an (f, τ, M )-good point the following inequality holds:
Theorem 3.8. Assume that f ∈ Mreg has the turnpike property, M > 0 and > 0.
Then there exists l > 0, L > 0 and a natural number Q such that for each T1 ∈ R1 ,
each T2 ≥ T1 + L and each trajectory-control pair
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
and
T2 − T1 ≥ 2bf ,
|ξi − xf (Ti )| ≤ δ, i = 1, 2
Proof. Let
= 0 /2 (3.1)
T1 ∈ R1 , T2 ≥ T1 + 2bf (3.2)
(ξi , Ti ) ∈ A, i = 1, 2 (3.3)
and
By (3.2)–(3.4) and B(iv) which holds with and δ there exist trajectory-control pairs
x1 : [T1 , T1 + bf ] → Rn , u1 : [T1 , T1 + bf ] → Rm ,
x2 : [T2 − bf , T2 ] → Rn , u2 : [T2 − bf , T2 ] → Rm
such that
x1 (T1 ) = ξ1 ,
x1 (T1 + bf ) = xf (T1 + bf ),
x2 (T2 ) = ξ2 ,
x2 (T2 − bf ) = xf (T2 − bf ),
|x1 (t) − xf (t)| ≤ for all t ∈ [T1 , T1 + bf ],
3.2 Auxiliary Results 83
x̃ : [T1 , T2 ] → Rn , ũ : [T1 , T2 ] → Rm
such that
x̃(Ti ) = ξi , i = 1, 2. (3.7)
|ξi − xf (Ti )| ≤ δ, i = 1, 2
U f (T1 , T2 , ξ1 , ξ2 ) ≥ I f (T1 , T2 , xf , uf ) − 0 .
Proof. Let
= 0 /8 (3.8)
(Ti , ξi ) ∈ A, i = 1, 2 (3.9)
84 3 Infinite Horizon Problems
and
We show that
U f (T1 , T2 , ξ1 , ξ2 ) ≥ I f (T1 , T2 , xf , uf ) − 0 .
U f (T1 , T2 , ξ1 , ξ2 ) < ∞.
By (3.9), (3.10) and B(iv) which holds with and δ there exist trajectory-control
pairs
x1 : [T1 − bf , T1 ] → Rn , u1 : [T1 − bf , T1 ] → Rm ,
x2 : [T2 , T2 + bf ] → Rn , u2 : [T2 , T2 + bf ] → Rm
such that
x1 (T1 − bf ) = xf (T1 − bf ),
x1 (T1 ) = ξ1 ,
x2 (T2 ) = ξ2 ,
x2 (T2 + bf ) = xf (T2 + bf ),
|x1 (t) − xf (t)| ≤ for all t ∈ [T1 − bf , T1 ],
|x2 (t) − xf (t)| ≤ for all t ∈ [T2 , T2 + bf ],
I f (T1 − bf , T1 , x1 , u1 ) ≤ I f (T1 − bf , T1 , xf , uf ) + ,
I f (T2 , T2 + bf , x2 , u2 ) ≤ I f (T2 , T2 + bf , xf , uf ) + . (3.11)
x : [T1 − bf , T2 + bf ] → Rn , u : [T1 − bf , T2 + bf ] → Rm
such that
x(T1 − bf ) = xf (T1 − bf ),
x(T2 + bf ) = xf (T2 + bf ). (3.14)
I f (T1 − bf , T2 + bf , xf , uf ) ≤ I f (T1 − bf , T2 + bf , x, u)
= I f (T1 − bf , T1 , x1 , u1 ) + I f (T1 , T2 , x, u) + I f (T2 , T2 + bf , x2 , u2 )
≤ I f (T1 − bf , T1 , xf , uf ) + + I f (T1 , T2 , x, u) + I f (T2 , T2 + bf , xf , uf ) +
and
I f (T1 , T2 , xf , uf ) ≤ U f (T1 , T2 , ξ1 , ξ2 ) + 3
≤ U f (T1 , T2 , ξ1 , ξ2 ) + 0 .
T2 ≥ T1 + 2bf
and
|ξi − xf (Ti )| ≤ δ, i = 1, 2
|U f (T1 , T2 , ξ1 , ξ2 ) − I f (T1 , T2 , xf , uf )| ≤ .
By (3.16) and (LSC) property, for each integer N ≥ 1 there exists a trajectory-
control pair
xN : [τ, τ + N ] → Rn , uN : [τ, τ + N ] → Rm
such that
xN (τ ) = x(τ ),
I f (τ, τ + k, xN , uN ) = I f (τ, τ + N, xN , uN ) − I f (τ + k, τ + N, xN , uN )
≤ I f (τ, τ + N, x, u) − I f (τ + k, τ + N, x, u) + S
= I f (τ, τ + k, x, u) + S. (3.19)
{I f (τ + k, τ + k + 1, xN , uN )}∞
N =k+1
x∗ : [τ, ∞) → Rn , u∗ : [τ, ∞) → Rm
such that
Clearly,
x∗ (τ ) = x(τ ). (3.22)
3.3 Proof of Proposition 3.3 87
We show that
x∗ : [τ, ∞) → Rn , u∗ : [τ, ∞) → Rm
is an (f )-good trajectory-control pair. It follows from (3.20), and (3.19) that for each
integer j ≥ 1
Let
T1 ≥ τ and T2 > T1 .
I f (T1 , T2 , x∗ , u∗ )
= I f (τ, τ + q, x∗ , u∗ ) − I f (τ, T1 , x∗ , u∗ ) − I f (T2 , τ + q, x∗ , u∗ )
≤ I f (τ, τ + q, x, u) + S − I f (τ, T1 , x, u) + S
−I f (T2 , τ + q, x, u) + S = I f (T1 , T2 , x, u) + 3S.
Together with (3.16) this implies that (x∗ , u∗ ) is an (f )-good trajectory-control pair.
We show that (x∗ , u∗ ) is an (f )-minimal trajectory-control pair. Assume the
contrary. Then there exists an integer Q ≥ 1 such that
Set
y : [τ, τ + Q] → Rn , v : [τ, τ + Q] → Rm
such that
y(τ ) = x∗ (τ ),
y(τ + Q) = x∗ (τ + Q),
I f (τ, τ + Q, x∗ , u∗ ) − I f (τ, τ + Q, y, v) ≥ 2Δ. (3.26)
88 3 Infinite Horizon Problems
There exists δ ∈ (0, Δ) such that Lemma 3.11 holds with = 32−1 Δ(bf + 1)−1 .
Since (x∗ , u∗ ) is an (f )-good trajectory-control pair, it follows from (3.17) and
Proposition 3.2 that
x∗ (t) − xf (t) → 0 as t → ∞.
Fix an integer
j ≥ 4 + 4bf . (3.28)
k ≥ Q + j + T0 ,
I f (τ, τ + T0 + Q, x∗ , u∗ ) ≤ I f (τ, τ + T0 + Q, xNk , uNk ) + 8−1 δ,
|x∗ (τ + i) − xNk (τ + i)| < 8−1 δ, i = T0 + Q, T0 + Q + j. (3.29)
x̃ : [τ, τ + Nk ] → Rn , ũ : [τ, τ + Nk ] → Rm
such that
It follows from (3.27)–(3.30), (3.18) B(i), the choice of δ and Lemma 3.11 that
|I f (τ + T0 + Q, τ + Q + T0 + j, x̃, ũ)
−I f (τ + T0 + Q, τ + Q + T0 + j, xf , uf )| ≤ 32−1 Δ,
|I f (τ + T0 + Q, τ + Q + T0 + j, xNk , uNk )
−I f (τ + T0 + Q, τ + Q + T0 + j, xf , uf )| ≤ 32−1 Δ. (3.32)
This contradicts (2.31). The obtained contradiction we have reached proves that
(x∗ , u∗ ) is an (f )-minimal trajectory-control pair. Proposition 3.3 is proved.
By Proposition 3.2
Let
y : [τ, ∞) → Rn , v : [τ, ∞) → Rm
y(τ ) = x(τ ).
90 3 Infinite Horizon Problems
Assume the contrary. Then there exists a number > 0 for which
By Theorem 2.1 and Remark 3.1, we may assume without loss of generality that
(y, v) is an (f )-good trajectory-control pair. Proposition 3.2 and (3.33) imply that
Set
0 = 4−1 . (3.38)
There exists δ ∈ (0, 0 ) such that Lemma 3.11 holds with = 0 and δ. By (3.33),
(3.34), and (3.37) there exists an integer Q ≥ 1 such that
T0 > Q + 4,
I f (τ, T0 , x, u) − I f (τ, T0 , y, v) > . (3.40)
Fix an integer j ≥ 2bf + 8. By (3.39) and (3.40), the choice of δ, (LSC) property
and Lemma 3.11, there exists a trajectory-control pair
x̃ : [τ, ∞) → Rn , ũ : [τ, ∞) → Rm
such that
On the other hand it follows from (3.41), (3.42), (3.40), and (3.38) that
This contradicts (3.43). The contradiction we have reached proves Proposition 3.4.
This inequality implies that there exist a constant Δ0 > 0 and a trajectory-control
pair
x1 : [τ, S0 ] → Rn , u1 : [τ, S0 ] → Rm
such that
y : [τ, T ] → Rn , v : [τ, T ] → Rm
92 3 Infinite Horizon Problems
such that
y1 : [τ, T ] → Rn , v1 : [τ, T ] → Rm
such that
By (3.49), (3.46)–(3.48),
This contradicts (3.49) and (3.45). The contradiction we have reached proves that
(x, u) is an (f )-minimal trajectory-control pair.
Now assume that the trajectory-control pair (x, u) is (f )-minimal. We claim that
(x, u) is an (f )-agreeable trajectory-control pair. By Remark 3.1 (xf , uf ) is an (f )-
good trajectory-control pair. Therefore there is Q1 > 0 such that for each T1 ∈ R1
and each T2 > T1 ,
Let
By Lemma 3.11 there exists a number δ ∈ (0, 1) such that for each (Ti , ξi ) ∈
A, i = 1, 2 which satisfy
T2 ≥ T1 + 2bf
3.5 Proof of Theorem 3.6 93
and
It follows from the turnpike property and (3.44) that there exist
S1 ∈ R1 , S2 ≥ S1 + 2Δ0
y : [S1 , S2 ] → Rn , v : [S1 , S2 ] → Rm
which satisfies
I f (S1 , S2 , y, v)
and
Choose a number
y1 : [τ, T ] → Rn , v1 : [τ, T ] → Rm
94 3 Infinite Horizon Problems
such that
y1 (τ ) = x(τ ), (3.60)
I f (τ, T, y1 , v1 ) ≤ σ f (τ, T, x(τ )) + δ0 < ∞. (3.61)
I f (τ, T, y1 , v1 ) ≤ I f (τ, T, x, u) + 1
≤ inf{U f (τ, T, ξ1 , ξ2 ) : (τ, ξ1 ), (T, ξ2 ) ∈ A} + Q2 + 1. (3.62)
By (3.61), (3.62), (3.59), and the choice of Δ0 , δ0 [see (3.55), (3.57) and (3.56)],
By (3.59),
T − Δ0 − 2bf ≥ S0 . (3.64)
By (3.59),
By (3.65), (3.66), (3.68), (3.69), and the choice of δ [see (3.52) and (3.53)],
|I f (T − Δ0 − 2bf , T − Δ0 , xf , uf )
−U f (T − Δ0 − 2bf , T − Δ0 , η1 , η2 )| ≤ /8 (3.70)
3.5 Proof of Theorem 3.6 95
for all
and all
η2 ∈ {x(T − Δ0 ), y1 (T − Δ0 )}.
y2 : [T − Δ0 − 2bf , T − Δ0 ] → Rn , v2 : [T − Δ0 − 2bf , T − Δ0 ] → Rm
such that
y2 (T − Δ0 − 2bf ) = y1 (T − Δ0 − 2bf ),
y2 (T − Δ0 ) = x(T − Δ0 ), (3.71)
I f (T − Δ0 − 2bf , T − Δ0 , y2 , v2 )
≤ U f (T − Δ0 − 2bf , T − Δ0 , y2 (T − Δ0 − 2bf ), y2 (T − Δ0 )) + /16. (3.72)
By (3.71) and (3.60), the (f )-minimality of (x, u), (3.67), (3.72) and (3.70),
I f (τ, T − Δ0 , x, u)
≤ I f (τ, T − Δ0 − 2bf , y1 , v1 ) + I f (T − Δ0 − 2bf , T − Δ0 , y2 , v2 )
≤ I f (τ, T − Δ0 − 2bf , y1 , v1 )
+U f (T − Δ0 − 2bf , T − Δ0 , y2 (T − Δ0 − 2bf ), y2 (T − Δ0 )) + /16
≤ I f (τ, T − Δ0 − 2bf , y1 , v1 )
+I f (T − Δ0 − 2bf , T − Δ0 , xf , uf ) + /16 + /16. (3.73)
I f (T − Δ0 − 2bf , T − Δ0 , y1 , v1 )
≥ U f (T − Δ0 − 2bf , T − Δ0 , y1 (T − Δ0 − 2bf ), y1 (T − Δ0 ))
≥ I f (T − Δ0 − 2bf , T − Δ0 , xf , uf ) − /8. (3.74)
y : [τ, T ] → Rn , v : [τ, T ] → Rm
such that
I f (T − Δ0 − 2bf , T − Δ0 , y, v)
≤ I f (T − Δ0 , T, y1 , v1 ) + I f (τ, T − Δ0 − 2bf , x, u)
= I f (T − Δ0 , T, y1 , v1 ) + I f (τ, T − Δ0 − 2bf , x, u)
+I f (T − Δ0 − 2bf , T − Δ0 , x, u) + 3/8
= I f (T − Δ0 , T, y1 , v1 ) + I f (τ, T − Δ0 , x, u) + 3/8
y : [τ, T ] → Rn , v : [τ, T ] → Rm
Let f ∈ Mreg possess the turnpike property with the turnpike Xf , M > 0 and
> 0. By Remark 3.1,
Set
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
I f (τ, T, x, u)
≤ inf{U f (τ, T, ξ1 , ξ2 ) : (τ, ξ1 ), (T, ξ2 ) ∈ A} + M0 .
y : [τ, ∞) → Rn , v : [τ, ∞) → Rm
such that
y(τ ) = x(τ ),
I f (S1 , S2 , y, v)
≤ inf{U f (S1 , S2 , ξ1 , ξ2 ) : (Si , ξi ) ∈ A, i = 1, 2} + M (3.82)
for each S1 ≥ τ and each S2 > S1 . By Proposition 3.2, (3.80), and (3.82),
There exists δ ∈ (0, 1) such that B(iv) holds with = 1. By (3.83) and (3.84) there
exists
such that
Let
T ≥ T̃ . (3.88)
By (3.88), (3.87), (3.86), the choice of δ and B(iv) (which holds with = 1) there
exist trajectory-control pairs
y1 : [T − 2bf , T − bf ] → Rn , v1 : [T − 2bf , T − bf ] → Rm ,
y2 : [T − bf , T ] → Rn , v2 : [T − bf , T ] → Rm
such that
y2 (T − bf ) = xf (T − bf ),
y2 (T ) = x(T ),
I f (T − 2bf , T − bf , y1 , v1 ) ≤ I f (T − 2bf , T − bf , xf , uf ) + 1,
I f (T − bf , T, y2 , v2 ) ≤ I f (T − bf , T, xf , uf ) + 1. (3.89)
ỹ : [τ, T ] → Rn , ṽ : [τ, T ] → Rm
such that
T1 ∈ R1 , T2 ≥ T1 + 2Δ
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
which satisfies
and
Let τ ∈ R1 ,
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
I f (τ, T, x, u)
Let
T ≥ T̃ + 2Δ.
Since the inequality above holds for each T ≥ T̃ + 2Δ, we conclude that
We suppose that sum over empty set is zero. Since f ∈ Mreg possesses the turnpike
property (with the turnpike Xf ) it follows from Remark 3.1 and assumption B(i)
that
xf (t) = Xf (t) for all t ∈ R1 . (3.97)
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
Then
I f (S1 , S2 , x, u)
= I f (S1 , S2 , xf , uf ) + M + 3Q0
S1 ∈ R1 , S2 ≥ S1 + 2Δ
x : [S1 , S2 ] → Rn , u : [S1 , S2 ] → Rm
which satisfies
and
Fix
a natural number
Q > 3M δ −1 + 4 (3.104)
and
Assume that
T1 ∈ R 1 , T2 ≥ T1 + L
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies
T 1 ≤ S1 < S2 ≤ T 2
we have
t0 = T 1 . (3.108)
and
k ≤ M δ −1 . (3.111)
Set
t̃k+1 = inf{τ ∈ (tk , T2 ] : I f (tk , τ, x, u) > U f (tk , τ, x(tk ), x(τ )) + δ}. (3.113)
T2 ≥ t̃k+1 > tk .
such that
and in this case the assumption we made for k also holds for k + 1.
By (3.111) our sequence {ti } is finite. Let k ≥ 0 be an integer such that tk+1 is
its last element. By (3.111),
k + 1 ≤ M δ −1 + 1. (3.114)
104 3 Infinite Horizon Problems
and that for each integer i ∈ {1, . . . , k + 1} there a number si ∈ (ti−1 , ti ] such that
s i ≥ ti − 1 (3.115)
Let
i ∈ A. (3.117)
By (3.117), (3.116), and (3.115),
si − ti−1 ≥ 2Δ (3.118)
By (3.110), (3.119), (2.118), and the choice of Δ, δ [see (3.100) and (3.101)],
for all i ∈ A.
By (3.120),
3(k + 1) ≤ 3M δ −1 + 3 ≤ Q.
In this chapter we continue to use the notation and the definitions of Chaps. 2 and 3.
is controllable.
Denote by Mc the set of all continuous functions f ∈ M which satisfy the
following assumptions:
D(i) for each (t, x) ∈ Rn+1 the function f (t, x, ·) : Rn → R1 is convex;
D(ii) for each K > 0 there exists a constant af,K > 0 and an increasing function
such that
ψf,K (t) → ∞ as t → ∞
and
|f (t, x1 , u1 ) − f (t, x2 , u2 )| ≤
ψf : [0, ∞) → [0, ∞)
such that
ψf (t) → ∞ as t → ∞
and
f (t, x, u) ≥ ψf (|u|)|u| − af
We can easily deduce the following result (for the proof see Proposition 2.1
of [48]).
Proposition 4.1. Let f ∈ Mc and M, be positive numbers. Then there exist Γ > 0
and δ > 0 such that
xj : [T1 , T2 ] → Rn , uj : [T1 , T2 ] → Rm , j = 1, 2, . . .
It follows from (4.6), assumption D(ii) and the properties of the functions ψf,M0
that the sequence of functions {|ui (·)|}∞
i=1 is equiabsolutely integrable on [T1 , T2 ].
Therefore there exists subsequence {(xik , uik )∞k=1 and
such that
Clearly,
h1 = x.
By Proposition 2.8,
We show that
T2
I f (T1 , T2 , xik , uik ) − f (t, x(t), uik (t))dt → 0 as k → ∞. (4.12)
T1
Set
δ ∈ (0, δ0 ) (4.18)
such that
Clearly,
T2
I f (T1 , T2 , xik , uik ) − f (t, x(t), uik (t))dt = σ1 + σ2 , (4.23)
T1
where
σj = [f (t, xik (t), uik (t)) − f (t, x(t), uik (t))]dt, j = 1, 2. (4.24)
Ej
We estimate σ1 , σ2 separately.
Let t ∈ E2 . It follows from (4.22), (4.7), (4.14), (4.21), and the choice of δ (see
(4.19)) that
|f (t, xik (t), uik (t)) − f (t, x(t), uik (t))| ≤ [8(T2 − T1 + 1)]−1 .
110 4 Linear Control Systems
Let t ∈ E1 . It follows from (4.22), (4.7), (4.14), (4.18), (4.21), and the choice of
Γ, δ0 (see (4.17)) that
Together with assumption (A), (4.13) and (4.15) this implies that
|σ1 | ≤ γ f (t, xik (t), uik (t))dt
E1
f
≤ γ(I (T1 , T2 , xik , uik ) + a0 (T2 − T1 ))
≤ γ(M1 + a0 (T2 − T1 )) ≤ 4−1 . (4.26)
I f (T1 , T2 , x, h2 ) ≤ M2 . (4.27)
It follows from assumption D(i) that the set F is convex. By (4.12) and (4.28),
Since the set F is convex it follows (4.29) and (4.8) that there exists a sequence
{vi }∞
i=1 ⊂ F which satisfies
T2
|vi (t) − h2 (t)|dt → 0 as i → ∞.
T1
4.3 A Continuity Property 111
It follows from this relation, the continuity of f , assumption (A), the definition of
F , and Fatau’s lemma that
T2
f (t, x(t), h2 (t))dt
T1
T2
≤ sup f (t, x(t), vi (t))dt : i = 1, 2, . . . ≤ M2 + δ
T1
and
T2
f (t, x(t), h2 (t))dt ≤ M2 + δ.
T1
x = Ax + BB t y,
y = x − At y
Let τ be a positive number. It follows from Proposition 4.3 that for each ỹ, z̃ ∈
Rn there exists a unique solution x(·), y(·) of the following system
For any initial value (x0 , y0 ) ∈ Rn × Rn there exists a unique solution of (4.30)
(x(s), y(s)) = (esC ((Dτ,1 (ỹ, z̃), Dτ,2 (ỹ, z̃))t )t , s ∈ R1 (4.32)
Dτ,j : Rn × Rn → Rn , j = 1, 2
Proposition 4.6. Let f ∈ Mc , M, τ, > 0. Then there exists a number δ > 0 such
that:
1. for each T ∈ R1 , each y1 , y2 , z1 , z2 ∈ Rn satisfying
x1 : [T, T + τ ] → Rn , u1 : [T, T + τ ] → Rm
which satisfies
x1 (T1 ) = y1 , x1 (T2 ) = z1 ,
I f (T1 , T2 , x1 , u1 ) = U f (T1 , T2 , y1 , z1 )
x2 : [T, T + τ ] → Rn , u2 : [T, T + τ ] → Rm
such that
x2 (T1 ) = y2 , x2 (T2 ) = z2 ,
|I f (T1 , T2 , x2 , u2 ) − I f (T1 , T2 , x1 , u1 )| ≤ ,
|x1 (t) − x2 (t)| ≤ , t ∈ [T1 , T2 ].
Proof. Set
x : [T, T + τ ] → Rn , u : [T, T + τ ] → Rm
satisfying
I f (T, T + τ, x, u) ≤ M0 + 4
By Proposition 4.1 there exist Γ0 > 8 and δ2 ∈ (0, 8−1 ) such that
|xi | ≤ M1 + 8, i = 1, 2, |ui | ≥ Γ0 − 4, i = 1, 2,
|x1 − x2 |, |u1 − u2 | ≤ δ2 . (4.39)
such that
|xi |, |ui | ≤ Γ0 + M1 + 4, i = 1, 2,
|x1 − x2 |, |u1 − u2 | ≤ δ3 . (4.42)
such that
x1 : [T, T + τ ] → Rn , u1 : [T, T + τ ] → Rm
such that
x1 (T ) = y1 , x1 (T + τ ) = z1 ,
I (T, T + τ, x1 , u1 ) = U f (T, T + τ, y1 , z1 ).
f
(4.45)
4.3 A Continuity Property 115
It follows from (4.47), (4.48), (4.45), (4.30)–(4.32), and the definition of Dτ,1 , Dτ,2
that
Assumption (A), (4.45), (4.35), and (4.33) imply that for any measurable set E ⊂
[T, T + τ ],
f (t, x1 (t), u1 (t))dt ≤ M0 + a0 τ. (4.51)
E
Set
It follows from (4.46), (4.50), (4.40), and the choice of Γ0 , δ2 and (4.51) that
|f (t, x1 (t), u1 (t)) − f (t, x2 (t), u2 (t))|dt
E1
≤ δ1 f (t, x1 (t), u1 (t))dt ≤ δ1 (M0 + a0 τ ). (4.52)
E1
|I f (T1 , T2 , x2 , u2 ) − I f (T1 , T2 , x1 , u1 )| ≤ ,
U f (T, T + τ, y2 , z2 ) ≤ U f (T, T + τ, y1 , z1 ) + .
q−1
U f (T +kτ, T +(k+1)τ, xk , xk+1 )−U f (T +kτ, T +(k+1)τ, yk , yk+1 ) ≥M2 ,
k=0
(4.54)
where
yk = xk , k = 0, q, yk = 0, k = 1, . . . , q − 1.
and it is easy to see that (4.54) holds when the constant M3 is large enough.
We will prove assertion 2. Assume that T ∈ R1 , and integer q ≥ 1 and a
sequence {xk }qk=0 ⊂ Rn satisfies (4.56). Then there is j ∈ {1, . . . , q} such that
|xj | > M3 . Set
yi = xi , i = 0, . . . , i1 , yi = 0, i = i1 + 1, . . . , q.
Otherwise we set
and define {yk }qk=0 by (4.57). It is easy to verify that in both cases (4.54) holds.
We will prove assertion 3. Assume that T ∈ R1 , an integer q ≥ 1, and a sequence
{xk }qk=0 ⊂ Rn satisfies
yi = xi , i = i2 , . . . , q, yi = 0, i = 0, . . . , i2 − 1.
Otherwise we set
and define {yk }qk=0 by (4.57). It is easy to verify that in both cases (4.54) holds.
This completes the proof of Proposition 4.8.
xN : [−N, N ] → Rn , uN : [−N, N ] → Rm
such that
Then there exists a number M1 > 0 which satisfies (4.53) with τ = 1. Let a number
M3 > M1 + 1 be as guaranteed in Proposition 4.8 with τ = 1, M2 = 1. We will
show that
Let N ≥ 1 be an integer. By (4.58) and (4.53) which holds with τ = 1 there exists
an integer q ∈ [−N, N ] such that
|xN (q)| ≤ M1 .
Relation (4.59) follows from this relation, (4.58), Proposition 4.8 which holds with
τ = 1, M2 = 1 and the choice of M3 .
By (4.58), (4.59) and Propositions 2.7 and 4.5,
By (4.58), (4.60), and Proposition 4.5 there exists a number M4 > 0 such that
for each pair of integers i ≥ 1 and p ≥ i. Together with assumption (A) this implies
that for each integer j the sequence
{I f (j, j + 1, xp , up ) : p ≥ |j| + 1}
xf : R1 → Rn , uf : R1 → Rm
such that
and
and
≤ lim inf U f (i, j, xNk (i), xNk (j)) = U f (i, j, xf (i), xf (j)). (4.65)
k→∞
Therefore f satisfies assumption B(i). It follows from (4.65), (4.64), and Proposi-
tion 4.5 that f satisfies assumption B(ii).
We show that f satisfies assumption B(iii). Let S1 > 0. Set c = 1 and
Choose a number
x : [T1 , T2 ] → Rn , u : [T1 , T2 ] → Rm
satisfies
|x(Ti )| ≤ S1 , i = 1, 2. (4.69)
Choose an integer
y : [−N, N ] → Rn , u : [−N, N ] → Rm
4.5 The Existence and Structure of Solutions 121
such that
I f (−N, N, xf , uf ) − I f (−N, N, y, v)
I f (−N, N, xf , uf ) − I f (−N, N, y, v)
≥ I f (T1 , T2 , xf , uf ) − I f (T1 , T2 , x, u) − 2a0 − 2c1 − 2.
I f (T1 , T2 , xf , uf ) ≤ I f (T1 , T2 , x, u) + S2 .
|f (t, x, u)| + 1 ≥ 2−1 |fj (t, x, u)| + 2−1 , (t, x, u) ∈ Rn+m+1 , |x| ≤ K. (4.73)
Since fj ∈ Mc satisfies assumption D(ii) there exists a constant ak,j > 0 and an
increasing function
such that
ψk,j (t) → ∞ as t → ∞,
fj (t, x, u) ≥ ψk,j (|u|)|u| − ak,j , (t, x, u) ∈ Rn+m+1 , |x| ≤ K. (4.74)
|f (t, x, u)| ≥ 2−1 ψk,j (|u|)|u| − 2−1 ak,j − 2−1 , (t, x, u) ∈ Rn+m+1 , |x| ≤ K.
(4.75)
There exists K0 ≥ 1 such that
By (4.75), (4.76), and assumption (A) for each (t, x, u) ∈ Rn+m+1 satisfying |x| ≤
K, |u| ≥ K0 ,
Since f, fj satisfy assumption D(ii) there exist a number aM > 0 and an increasing
function
ψM : [0, ∞) → [0, ∞)
such that
ψM (t) → ∞ as t → ∞,
fj (t, x, u), f (t, x, u) ≥ ψM (|u|)|u| − aM , (t, x, u) ∈ Rn+m+1 , |x| ≤ M.
(4.79)
4.5 The Existence and Structure of Solutions 123
Γ0 > 1,
ψM (Γ0 ) ≥ 2a0 + 2aM ,
λ2 (1 + 2ψM (Γ0 )−1 )2 − 1 < 8−1 . (4.80)
Γ > Γ0 ,
|fj (t, x1 , u1 ) − fj (t, x2 , u2 )| ≤ 1 min{fj (t, x1 , u1 ), fj (t, x2 , u2 )} (4.82)
f (t, x1 , u1 ) − f (t, x2 , u2 )
−1
≤ λ(1 + 2ψM (Γ0 ) )fj (t, x1 , u1 ) − (λ(1 + 2ψM (Γ0 )−1 ))−1 fj (t, x2 , u2 )
= λ(1 + 2ψM (Γ0 )−1 )[fj (t, x1 , u1 ) − fj (t, x2 , u2 )]
124 4 Linear Control Systems
+fj (t, x2 , u2 )[λ(1 + 2ψM (Γ0 )−1 ) − (λ(1 + 2ψM (Γ0 )−1 ))−1 ]
≤ 1 λ(1 + 2ψM (Γ0 )−1 )fj (t, x2 , u2 )
+fj (t, x2 , u2 )[λ(1 + 2ψM (Γ0 )−1 ) − (λ(1 + 2ψM (Γ0 )−1 ))−1 ]
≤ 1 [λ(1 + 2ψM (Γ0 )−1 )]2 fj (t, x2 , u2 )
+f (t, x2 , u2 )[λ2 (1 + 2ψM (Γ0 )−1 )2 − 1] ≤ f (t, x2 , u2 ).
Therefore the function f satisfies assumption D(iii). This completes the proof of
Proposition 4.10.
Assume that f ∈ Mc . By Proposition 4.9 f satisfies assumption (B). Let a
trajectory-control pair
xf : R1 → Rn , uf : R1 → Rm
x : [τ, ∞) → Rn , u : [τ, ∞) → Rm
x∗ : [τ, ∞) → Rn , u∗ : [τ, ∞) → Rm
1. Anderson, B.D.O., Moore, J.B.: Linear Optimal Control. Prentice-Hall, Englewood Cliffs
(1971)
2. Arkin, V.I., Evstigneev, I.V.: Stochastic Models of Control and Economic Dynamics.
Academic, London (1987)
3. Aseev, S.M., Besov, K.O., Kryazhimskii, A.V.: Russ. Math. Surv. 67, 195–253 (2012)
4. Aseev, S.M., Kryazhimskiy, A.V.: SIAM J. Control Optim. 43, 1094–1119 (2004)
5. Aseev, S.M., Veliov, V.M.: Dyn. Contin., Discret. Impulsive Syst. Ser. B: Appl. Algorithms 19,
43–63 (2012)
6. Aubin, J.P., Ekeland, I.: Applied Nonlinear Analysis. Wiley Interscience, New York (1984)
7. Aubry, S., Le Daeron, P.Y.: Physica D 8, 381–422 (1983)
8. Baumeister, J., Leitao, A., Silva, G.N.: Syst. Control Lett. 56, 188–196 (2007)
9. Blot, J., Cartigny, P.: J. Optim. Theory Appl. 106, 411–419 (2000)
10. Blot, J., Hayek, N.: ESAIM Control Optim. Calc. Var. 5, 279–292 (2000)
11. Blot, J., Michel, P.: Appl. Math. Lett. 16, 71–78 (2003)
12. Carlson, D.A., Haurie, A., Leizarowitz, A.: Infinite Horizon Optimal Control. Springer, Berlin
(1991)
13. Cartigny, P., Michel, P.: Automatica J. IFAC 39, 1007–1010 (2003)
14. Coleman, B.D., Marcus, M., Mizel, V.J.: Arch. Rational Mech. Anal. 117, 321–347 (1992)
15. Evstigneev, I.V., Flam, S.D.: Set-Valued Anal. 6, 61–81 (1998)
16. Gaitsgory, V., Rossomakhine, S., Thatcher, N.: Dyn. Contin., Discret. Impulsive Syst. Ser. B:
Appl. Algorithms 19, 43–63 (2012)
17. Gale, D.: Rev. Econ. Stud. 34, 1–18 (1967)
18. Guo, X., Hernandez-Lerma, O.: Bernoulli 11, 1009–1029 (2005)
19. Hammond, P.J.: Consistent Planning and Intertemporal Welfare Economics. University of
Cambridge, Cambridge (1974)
20. Hammond, P.J.: Rev. Econ. Stud. 42, 1–14 (1975)
21. Hammond, P.J., Mirrlees, J.A.: Models of Economic Growth, pp. 283–299. Wiley, New York
(1973)
22. Hayek, N.: Optimization 60, 509–529 (2011)
23. Jasso-Fuentes, H., Hernandez-Lerma, O.: Appl. Math. Optim. 57, 349–369 (2008)
24. Leizarowitz, A.: Appl. Math. Optim. 13, 19–43 (1985)
25. Leizarowitz, A.: Appl. Math. Optim. 14, 155–171 (1986)
26. Leizarowitz, A., Mizel, V.J.: Arch. Rational Mech. Anal. 106, 161–194 (1989)
27. Lykina, V., Pickenhain, S., Wagner, M.: J. Math. Anal. Appl. 340, 498–510 (2008)
28. Makarov, V.L., Rubinov, A.M.: Mathematical Theory of Economic Dynamics and Equilibria.
Springer, New York (1977)
29. Malinowska, A.B., Martins, N., Torres, D.F.M.: Optim. Lett. 5, 41–53 (2011)
30. Marcus, M., Zaslavski, A.J.: Ann. Inst. H. Poincare, Anal. Non Lineare 16, 593–629 (1999)
31. Marcus, M., Zaslavski, A.J.: Ann. Inst. H. Poincare, Anal. Non Lineare 19, 343–370 (2002)
32. Martins, N., Torres, D.F.M.: J. Optim. Theory Appl. 155(2), 453–476 (2012)
33. McKenzie, L.W.: Econometrica 44, 841–866 (1976)
34. Mordukhovich, B.S.: Automat. Remote Control 50, 1333–1340 (1990)
35. Mordukhovich, B.S.: Appl. Anal. 90, 1075–1109 (2011)
36. Mordukhovich, B.S., Shvartsman, I.: Optimal Control, Stabilization and Nonsmooth Analysis.
Lecture Notes in Control and Information Sciences, pp. 121–132. Springer, New York (2004)
37. Moser, J.: Ann. Inst. H. Poincare, Anal. Nonlineare 3, 229–272 (1986)
38. Ocana, E., Cartigny, P.: SIAM J. Control Optim. 50, 2573–2587 (2012)
39. Ocana, E., Cartigny, P., Loisel, P.: J. Nonlinear Convex Anal. 10, 157–176 (2009)
40. Pickenhain, S., Lykina, V., Wagner, M.: Control Cybernet 37, 451–468 (2008)
41. Rubinov, A.M.: J. Soviet Math. 26, 1975–2012 (1984)
42. Samuelson, P.A.: Am. Econ. Rev. 55, 486–496 (1965)
43. von Weizsacker, C.C.: Rev. Econ. Stud. 32, 85–104 (1965)
44. Zaslavski, A.J.: Math. USSR Izvestiya 29, 323–354 (1987)
45. Zaslavski, A.J.: SIAM J. Control Optim. 33, 1643–1660 (1995)
46. Zaslavski, A.J.: SIAM J. Control Optim. 33, 1661–1686 (1995)
47. Zaslavski, A.J.: Nonlinear Anal. 27, 895–931 (1996)
48. Zaslavski, A.J.: Abstr. Appl. Anal. 3, 265–292 (1998)
49. Zaslavski, A.J.: Nonlinear Anal. 42, 1465–1498 (2000)
50. Zaslavski, A.J.: Turnpike Properties in the Calculus of Variations and Optimal Control.
Springer, New York (2006)
51. Zaslavski, A.J.: Optimization on Metric and Normed Spaces. Springer, New York (2010)
52. Zaslavski, A.J.: Portugal. Math. 68, 239–257 (2011)
53. Zaslavski, A.J., Leizarowitz, A.: Math. Oper. Res. 22, 726–746 (1997)
Index
L
B
Lebesgue integrable function, 12
Borelian function, 12
Lebesgue measurable function, 12
Borel measurable set, 11
(LSC) property, 79
C
Complete metric space, 1 M
Complete uniform space, 3 Minimal solution, 2
Control constraint, 11
Control function, 11
Control system, 9 N
Neumann path, 6
D
Differential equation, 11
O
Overtaking optimality criterion, 1
E
Euclidean norm, 2
Euclidean space, 2 T
Topological subspace, 15
Trajectory-control pair, 12
G
Turnpike property, 6
Good function, 4
Good point, 81
U
I Uniformity, 3
Increasing function, 2 Uniform space, 3