This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

### Publishers

Scribd Selects Books

Hand-picked favorites from

our editors

our editors

Scribd Selects Audiobooks

Hand-picked favorites from

our editors

our editors

Scribd Selects Comics

Hand-picked favorites from

our editors

our editors

Scribd Selects Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

P. 1

Control Theory on Lie Groups|Views: 31|Likes: 0

Published by RodolfoReyes-Báez

See more

See less

https://www.scribd.com/doc/76372119/Control-Theory-on-Lie-Groups

12/23/2011

text

original

- 1 Motivation
- 1.1 Bilinear systems
- 1.2 Rotations of a rigid body
- 2 Lie groups and Lie algebras
- 2.1 Linear Lie groups
- 2.2 The Lie algebra of a Lie group
- 2.3 Matrix exponential
- 2.4 Left-invariant vector ﬁelds
- 3 Left-invariant control systems
- 3.1 Deﬁnitions
- 3.2 Right-invariant control systems
- 3.3 Basic properties of orbits and reachable sets
- 3.4 Normal attainability
- 3.5 General controllability conditions
- 4.1 Saturate
- 4.2 Lie saturate of invariant system
- 5 Induced systems on homogeneous spaces
- 6.1 Symmetric systems
- 6.2 Compact Lie groups
- 6.3 Semisimple Lie groups
- 6.4 Solvable Lie groups
- 6.5 Semi-direct products of Lie groups
- 7.1 Hamiltonian systems on T∗M
- 7.2 Pontryagin Maximum Principle on smooth manifolds
- 7.3 Hamiltonian systems on T∗G
- 7.4 Hamiltonian systems in the case of compact Lie group
- 8.1 Riemannian problem on compact Lie group
- 8.2 Sub-Riemannian problem on SO(3)
- 8.3 Sub-Riemannian problem on the Heisenberg group
- 8.4 Euler’s elastic problem
- 8.5 The plate-ball system

Yuri L. Sachkov

Program Systems Institute

Russian Academy of Sciences

Pereslavl-Zalessky, Russia

E-mail: sachkov@sys.botik.ru

Abstract

Lecture notes of an introductory course on control theory on Lie groups.

Controllability and optimal control for left-invariant problems on Lie groups

are addressed. A general theory is accompanied by concrete examples.

The course is intended for graduate students, no preliminary knowledge

of control theory or Lie groups is assumed.

SISSA 15/2006/M (March 2006)

Control theory on Lie groups

∗

Yu. L. Sachkov

Program Systems Institute

Pereslavl-Zalessky, Russia

e-mail: sachkov@sys.botik.ru

Abstract

Lecture notes of an introductory course on control theory on Lie groups.

Controllability and optimal control for left-invariant problems on Lie groups

are addressed. A general theory is accompanied by concrete examples.

The course is intended for graduate students, no preliminary knowledge

of control theory or Lie groups is assumed.

∗

Lectures given in SISSA, Trieste, Italy, January–February 2006. The author thanks SISSA

for support. Work partially supported by Russian Foundation for Basic Report, Project

No. 05-01-00703-a.

1

Contents

1 Motivation 4

1.1 Bilinear systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2 Rotations of a rigid body . . . . . . . . . . . . . . . . . . . . . . 4

2 Lie groups and Lie algebras 5

2.1 Linear Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.2 The Lie algebra of a Lie group . . . . . . . . . . . . . . . . . . . 9

2.3 Matrix exponential . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.4 Left-invariant vector ﬁelds . . . . . . . . . . . . . . . . . . . . . . 13

3 Left-invariant control systems 16

3.1 Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.2 Right-invariant control systems . . . . . . . . . . . . . . . . . . . 17

3.3 Basic properties of orbits and reachable sets . . . . . . . . . . . . 18

3.4 Normal attainability . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.5 General controllability conditions . . . . . . . . . . . . . . . . . . 24

4 Extension techniques for left-invariant systems 26

4.1 Saturate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.2 Lie saturate of invariant system . . . . . . . . . . . . . . . . . . . 29

5 Induced systems on homogeneous spaces 31

6 Controllability conditions for special classes of systems and Lie

groups 36

6.1 Symmetric systems . . . . . . . . . . . . . . . . . . . . . . . . . . 36

6.2 Compact Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . 37

6.3 Semisimple Lie groups . . . . . . . . . . . . . . . . . . . . . . . . 39

6.4 Solvable Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . 43

6.5 Semi-direct products of Lie groups . . . . . . . . . . . . . . . . . 45

7 Pontryagin Maximum Principle for invariant optimal control

problems on Lie groups 47

7.1 Hamiltonian systems on T

∗

M . . . . . . . . . . . . . . . . . . . . 47

7.2 Pontryagin Maximum Principle on smooth manifolds . . . . . . . 48

7.3 Hamiltonian systems on T

∗

G . . . . . . . . . . . . . . . . . . . . 49

7.4 Hamiltonian systems in the case of compact Lie group . . . . . . 51

8 Examples of invariant optimal control problems on Lie groups 53

8.1 Riemannian problem on compact Lie group . . . . . . . . . . . . 53

8.2 Sub-Riemannian problem on SO(3) . . . . . . . . . . . . . . . . . 55

8.3 Sub-Riemannian problem on the Heisenberg group . . . . . . . . 57

8.4 Euler’s elastic problem . . . . . . . . . . . . . . . . . . . . . . . . 63

8.5 The plate-ball system . . . . . . . . . . . . . . . . . . . . . . . . 69

2

References 74

Index 76

3

1 Motivation

1.1 Bilinear systems

In the study of controllability of a bilinear control system

˙ x = Ax +uBx, x ∈ R

n

, u ∈ R, (1)

where A and B are constant n n matrices, one naturally passes from the

system (1) for vectors to the similar system for matrices:

˙

X = AX + uBX, X n n matrix, u ∈ R. (2)

Such a passage is very natural: recall that in the study of the linear ODE

˙ x = Ax, we pass to the matrix ODE

˙

X = AX, here X is the Cauchy matrix for

the linear ODE.

There is a clear and important relation between controllability properties of

the bilinear system (1) and the matrix system (2):

“ system (2) is controllable ⇒ system (1) is controllable ”

In the sequel we make this statement precise and remove the quotation marks.

But this implication is clear: if we can control on matrices, the more so we can

control on vectors.

One may think that matrix systems (2) are more complicated then the bi-

linear ones (1), but this is not the case: the matrix systems evolve on matrix

groups (linear Lie groups), while the bilinear ones just on smooth submanifolds

of R

n

(homogeneous spaces of linear Lie groups). And the study of controlla-

bility for matrix systems is an easier problem since here the group structure

provides powerful additional techniques. We will clarify all these questions in

our course.

1.2 Rotations of a rigid body

Some important control systems in mechanics, physics, geometry etc. naturally

evolve on groups.

Consider rotations of a rigid body in R

3

around a ﬁxed point (e.g. rotations

of a space satellite around its center of mass). In order to describe motion of the

body, choose a ﬁxed orthonormal frame e

1

, e

2

, e

3

in the ambient space, and a

moving orthonormal frame f

1

, f

2

, f

3

attached to the body. Then the orientation

matrix

X : (e

1

, e

2

, e

3

) → (f

1

, f

2

, f

3

)

is a 3 3 orthogonal unimodular matrix. Moreover, it is easy to see that the

matrix

˙

XX

−1

= Ω is skew-symmetric (the angular velocity of the body). So we

obtain the equation of motion

˙

X = ΩX.

4

If we suppose that we can control the matrix Ω, than the previous system is

a matrix control system similar to (2). In the study of such systems, many

questions arise, and one of the ﬁrst ones is, what is the state space of such

systems:

X ∈ ?

We will answer this question in the next section.

2 Lie groups and Lie algebras

2.1 Linear Lie groups

The most important class of Lie groups is formed by linear Lie groups, i.e.,

groups of linear transformations of R

n

.

Let X : R

n

→ R

n

be a linear mapping. In a basis e

1

, . . . , e

n

of R

n

, the

operator X has a matrix X = (x

ij

), i, j = 1, . . . , n, which we identify with the

operator itself. So we are going to consider groups of matrices.

Denote the linear space of all n n matrices with real entries as

M(n, R) = ¦X = (x

ij

) [ x

ij

∈ R, i, j = 1, . . . , n¦.

For short, we will usually denote this space by M(n). The matrix entries x

ij

provide coordinates on M(n) = R

n

2

.

Example 2.1 (General Linear Group). The general linear group consists

of all n n invertible matrices:

GL(n, R) = GL(n) = ¦X ∈ M(n) [ det X ,= 0¦.

The following properties of GL(n) are easily established.

(1) By continuity of determinant, det : M(n) → R, the set GL(n) is an

open domain, thus a smooth submanifold in the linear space M(n).

(2) Further, GL(n) is a group with respect to matrix product. Indeed, if

X, Y ∈ GL(n), then the product XY ∈ GL(n). Further, the identity matrix

Id = (δ

ij

) (where δ

ij

= 1 if i = j and δ

ij

= 0 if i ,= j, the Kronecker symbol)

is contained in GL(n). Finally, for a nonsingular matrix X, its inverse X

−1

is

nonsingular as well.

(3) Moreover, the group operations in GL(n) are smooth:

(X, Y ) → XY (XY )

ij

are polynomials in X

ij

, Y

ij

,

X → X

−1

(X

−1

)

ij

are rational functions in X

ij

.

Deﬁnition 2.1. A set G is called a Lie group if:

(1) G is a smooth manifold,

(2) G is a group, and

(3) the group operations in G are smooth.

5

In the previous example we showed that GL(n) is a Lie group.

Deﬁnition 2.2. A Lie group G ⊂ M(n) is called a linear Lie group.

A convenient suﬃcient condition for a set of matrices to form a linear Lie

group is given in the following general proposition.

Theorem 2.1. If G is a closed subgroup of GL(n), then G is a linear Lie group.

Proof. See e.g. [45].

In other words, in order to verify that a set of matrices G ⊂ M(n) is a linear

Lie group, it suﬃces to show that the following three conditions hold:

(1) G ⊂ GL(n),

(2) G is a group w.r.t. matrix product, and

(3) G is topologically closed in GL(n) (i.e., G = GL(n) ∩ S, where S is a

closed subset in M(n)).

Now we consider several important examples of linear Lie groups in addition

to the largest one, GL(n). In all these cases the hypotheses of Theorem 2.1 are

easily veriﬁed.

Example 2.2 (Special Linear Group). The special linear group consists of

n n unimodular matrices:

SL(n, R) = SL(n) = ¦X ∈ M(n) [ det X = 1¦.

Such matrices correspond to linear operators v → Xv preserving the standard

volume in R

n

.

Example 2.3 (Orthogonal Group). The orthogonal group is formed by nn

orthogonal matrices:

O(n) = ¦X ∈ M(n) [ XX

T

= Id¦,

where X

T

denotes the transposed matrix of X. Orthogonal transformations

v → Xv preserve the Euclidean structure in R

n

.

Since 1 = det(XX

T

) = det

2

X, it follows that orthogonal matrices have

determinant det X = ±1.

Example 2.4 (Special Orthogonal Group). Orthogonal unimodular matri-

ces form the special orthogonal group:

SO(n) = ¦X ∈ M(n) [ XX

T

= Id, det X = 1¦.

Special orthogonal transformations v → Xv preserve both the Euclidean struc-

ture and orientation in R

n

.

6

Example 2.5 (Aﬃne Group). The aﬃne group is deﬁned as follows:

Aﬀ(n) =

_

X =

_

Y b

0 1

_

∈ M(n + 1) [ Y ∈ GL(n), b ∈ R

n

_

⊂ GL(n + 1).

Such matrices correspond to invertible aﬃne transformations in R

n

of the form

v → Y v +b, i.e., a linear mapping Y plus a translation b.

Example 2.6 (Euclidean Group). The Euclidean group is the following sub-

group of the aﬃne group:

E(n) =

_

X =

_

Y b

0 1

_

∈ M(n + 1) [ Y ∈ SO(n), b ∈ R

n

_

⊂ GL(n + 1).

Such matrices parametrize orientation-preserving aﬃne isometries v → Y v +b.

Example 2.7 (Triangular Group). And the last example of a group formed

by real matrices: the triangular group consists of all invertible triangular ma-

trices:

T(n) =

_

¸

¸

¸

_

¸

¸

¸

_

X =

_

_

_

_

_

∗ ∗ ∗

0 ∗ ∗

.

.

.

.

.

.

.

.

.

0 0 ∗

_

_

_

_

_

∈ GL(n)

_

¸

¸

¸

_

¸

¸

¸

_

= ¦X = (x

ij

) ∈ M(n) [ x

ij

= 0, i > j, x

ii

,= 0¦.

These are matrices of invertible linear operators v → Xv preserving the ﬂag of

subspaces Re

1

⊂ span(e

1

, e

2

) ⊂ ⊂ span(e

1

, . . . , e

n−1

) ⊂ R

n

.

Now we pass to complex matrices. Denote the space of all n n matrices

with complex entries as

M(n, C) = ¦Z = (z

jk

) [ z

jk

∈ C, j, k = 1, . . . , n¦.

Since any entry decomposes into the real and imaginary parts:

z

jk

= x

jk

+iy

jk

,

a complex matrix decomposes correspondingly:

Z = X +iY, X = (x

jk

), Y = (y

jk

).

The real coordinates x

jk

, y

jk

turn M(n, C) into R

2n

2

.

The realiﬁcation of a complex matrix is deﬁned as follows. To any complex

matrix of order n corresponds a real matrix of order 2n:

Z ∼

_

X −Y

Y X

_

∈ M(2n, R), Z = X +iY ∈ M(n, C).

7

The matrix

_

X −Y

Y X

_

is just the matrix of the real linear operator in R

2n

=

C

n

corresponding to Z in the basis over reals e

1

, . . . , e

1

, ie

1

, . . . , ie

n

. So it is

natural that realiﬁcation respects the matrix product: if

Z

1

= X

1

+iY

1

∼

_

X

1

−Y

1

Y

1

X

1

_

, Z

2

= X

2

+iY

2

∼

_

X

2

−Y

2

Y

2

X

2

_

,

then

Z

1

Z

2

= X

1

X

2

−Y

1

Y

2

+i(Y

1

X

2

+X

1

Y

2

)

∼

_

X

1

X

2

−Y

1

Y

2

−X

1

Y

2

−Y

1

X

2

X

1

Y

2

+Y

1

X

2

X

1

X

2

−Y

1

Y

2

_

=

_

X

1

−Y

1

Y

1

X

1

_

_

X

2

−Y

2

Y

2

X

2

_

.

The realiﬁcation provides the embedding M(n, C) ⊂ M(2n, R).

Example 2.8 (Complex General Linear Group). The complex general

linear group consists of all complex n n invertible matrices:

GL(n, C) = ¦Z ∈ M(n, C) [ det Z ,= 0¦

=

__

X −Y

Y X

_

∈ M(2n, R) [ det

2

X + det

2

Y ,= 0

_

.

There holds a proposition similar to Theorem 2.1.

Theorem 2.2. If G is a closed subgroup of GL(n, C), then G is a linear Lie

group.

Example 2.9 (Complex Special Linear Group). The complex special linear

group is formed by all complex n n unimodular matrices:

SL(n, C) = ¦Z ∈ M(n, C) [ det Z = 1¦

=

__

X −Y

Y X

_

∈ M(2n, R) [ det(X +iY ) = 1

_

.

Example 2.10 (Unitary Group). An important example of a linear Lie group

is the unitary group consisting of all n n unitary matrices:

U(n) = ¦Z ∈ M(n, C) [

¯

Z

T

Z = Id¦.

(Here

¯

Z denotes the complex conjugate matrix of Z.) Such matrices correspond

to linear transformations that preserve the unitary structure in C

n

. Compute

the realiﬁcation of a unitary matrix.

We have

Z = X +iY,

¯

Z

T

= X

T

−iY

T

,

8

thus

¯

Z

T

Z ∼

_

X

T

Y

T

−Y

T

X

T

_

_

X −Y

Y X

_

=

_

X

T

X +Y

T

Y −X

T

Y +Y

T

X

−Y

T

X +X

T

Y X

T

X +Y

T

Y

_

=

_

Id 0

0 Id

_

∼ Id +i 0.

That is why the realiﬁcation of the unitary group reads

U(n) =

__

X −Y

Y X

_

∈ M(2n, R) [ X

T

X +Y

T

Y = Id, X

T

Y −Y

T

X = 0

_

.

Compute determinant of a unitary matrix:

1 = det(

¯

Z

T

Z) = det Z det Z = [ det Z[

2

,

so

det Z = e

iϕ

, ϕ ∈ R.

Example 2.11 (Special Unitary Group). Another important example of a

group formed by complex matrices is the special unitary group:

SU(n) = U(n) ∩ SL(n, C) = ¦Z ∈ M(n, C) [

¯

Z

T

Z = Id, det Z = 1¦

=

__

X −Y

Y X

_

∈ M(2n, R) [ X

T

X +Y

T

Y = Id, X

T

Y −Y

T

X = 0,

det(X +iY ) = 1

_

.

2.2 The Lie algebra of a Lie group

Example 2.12 (T

Id

GL(n)). Consider the tangent space to the general linear

group at the identity:

T

Id

GL(n) =

_

˙

X(0) [ X(t) ∈ GL(n), X(0) = Id

_

.

We compute this space explicitly. Since the velocity vector

˙

X(t) = ( ˙ x

ij

(t)) is

an n n matrix, we obtain

˙

X(0) = ( ˙ x

ij

(0)) = A ∈ M(n).

Thus

T

Id

GL(n) ⊂ M(n).

In order to show that this inclusion is in fact an equality, choose an arbitrary

matrix A ∈ M(n). The curve X(t) = Id +tA belongs to GL(n) for [t[ < ε and

small ε > 0. Moreover, X(0) = Id and

˙

X(0) = A. Consequently,

T

Id

GL(n) = M(n).

9

The tangent space T

Id

GL(n) is a linear space. Moreover, it is endowed with

an additional operation, commutator of matrices:

[A, B] = AB −BA ∈ M(n), A, B ∈ M(n).

Notice that this operation satisﬁes the following properties:

(1) bilinearity,

(2) skew-symmetry: [B, A] = −[A, B],

(3) Jacobi identity:

[[A, B], C] + [[B, C], A] + [[C, A], B] = 0.

Deﬁnition 2.3. A linear space L endowed with a binary operation [ , ] which

is:

(1) bilinear,

(2) skew-symmetric, and

(3) satisﬁes Jacobi identity,

is called a Lie algebra.

The space M(n) with the matrix commutator is a Lie algebra. In order to

underline the relation of this Lie algebra with the Lie group GL(n), this Lie

algebra is denoted as gl(n). Summing up,

T

Id

GL(n) = gl(n).

Such a construction has a generalization of fundamental importance.

Deﬁnition 2.4. The tangent space to a Lie group G at the identity element is

called the Lie algebra of the Lie group G:

L = T

Id

G.

We compute the Lie algebras of the Lie groups considered above.

Example 2.13 (The Lie Algebra of SL(n)). The Lie algebra of the special

linear group is denoted by sl(n). We have

sl(n) = T

Id

SL(n) = ¦

˙

X(0) [ X(t) ∈ SL(n), X(0) = Id¦.

Take a curve X(t) = Id +t

˙

X(0) +o(t) ∈ SL(n, then

1 = det X(t) = det(Id +t

˙

X(0) +o(t)) = 1 +t tr

˙

X(0) +o(t), t → 0,

so tr

˙

X(0) = 0.

10

Thus

sl(n) = ¦A ∈ M(n) [ tr A = 0¦,

the traceless matrices. To be precise, we proved only the inclusion ⊂. The

reverse inclusion we do not prove here for the sake of time and leave it to the

reader as an exercise. This can be done by comparing dimensions of the linear

spaces. (The same remark holds for similar computations in examples below.)

Example 2.14 (The Lie Algebra of SO(n)). This Lie algebra is denoted as

so(n) = T

Id

SO(n) = ¦

˙

X(0) [ X(t) ∈ SO(n), X(0) = Id¦.

We have X(t)X

T

(t) ≡ Id, thus

0 =

˙

X(0) X

T

(0)

. ¸¸ .

Id

+X(0)

. ¸¸ .

Id

˙

X

T

(0) =

˙

X(0) +

˙

X

T

(0).

Denoting A =

˙

X(0), we obtain A+A

T

= 0 and

so(n) = ¦A ∈ M(n) [ A+A

T

= 0¦,

the skew-symmetric matrices.

In a similar way one computes the Lie algebras in the following three cases.

Example 2.15 (The Lie Algebra of Aﬀ(n)).

aﬀ(n) = T

Id

Aﬀ(n) =

__

A b

0 0

_

[ A ∈ gl(n), b ∈ R

n

_

.

Example 2.16 (The Lie Algebra of E(n)).

e(n) = T

Id E(n) =

__

A b

0 0

_

[ A ∈ so(n), b ∈ R

n

_

.

Example 2.17 (The Lie Algebra of T(n)).

t(n) = T

Id T(n) = ¦A = (a

ij

) ∈ M(n) [ a

ij

= 0, i > j¦,

the triangular matrices.

Finally compute the Lie algebras of the unitary and the special unitary

groups.

Example 2.18 (The Lie Algebra of U(n)).

u(n) = T

Id U(n),

and we proceed in the same way as for SO(n). For a curve Z(t) ∈ U(n),

Z(0) = Id, we have Z(t)

¯

Z

T

(t) ≡ Id. Thus

0 =

˙

Z(0)

¯

Z

T

(0)

. ¸¸ .

Id

+Z(0)

.¸¸.

Id

¯

˙

Z

T

(0).

11

Denoting A =

˙

Z(0), we obtain A +

¯

A

T

= 0, a skew-Hermitian matrix. Conse-

quently,

u(n) = ¦A ∈ M(n, C) [ A+

¯

A

T

= 0¦.

Example 2.19 (The Lie Algebra of SU(n)).

su(n) = T

Id

SU(n) = ¦A ∈ M(n, C) [ A+

¯

A

T

= 0, tr A = 0¦.

Summing up, we considered the passage from a Lie group G to the cor-

responding linear object — the Lie algebra L of the Lie group G. A natural

question on the possibility of the reverse passage is solved (for linear Lie groups)

via matrix exponential.

2.3 Matrix exponential

In order to approach matrix control systems, ﬁrst consider a matrix ODE:

˙

X = XA, (3)

where A ∈ M(n) is a given matrix. In the case n = 1, solutions to the ODE

˙ x = xa

are given by the exponential:

x(t) = x(0)e

at

,

e

a

= 1 +a +

a

2

2!

+ +

a

n

n!

+ .

For arbitrary natural n, we can proceed in a similar way and deﬁne for a matrix

A ∈ M(n) its exponential by the same series:

exp(A) = e

A

= Id +A+

A

2

2!

+ +

A

n

n!

+ .

This matrix series converges absolutely, thus it can be diﬀerentiated termwise:

_

e

At

_

=

_

Id +At +

A

2

t

2

2!

+ +

A

n

t

n

n!

+

_

= A+

A

2

t

1!

+ +

A

n

t

n−1

n!

+

= e

At

A.

Thus the matrix exponential X(t) = e

At

is the solution to the Cauchy problem

˙

X = XA, X(0) = Id,

and all solutions of the matrix equation (3) have the form

X(t) = X(0)e

At

.

12

Notice that for an arbitrary matrix A ∈ gl(n), its exponential exp(A) ∈

GL(n) since det exp(A) = exp(tr A) ,= 0. So we constructed a (smooth) mapping

exp : gl(n) → GL(n).

We generalize this construction in the following subsection.

2.4 Left-invariant vector ﬁelds

We saw that for an arbitrary matrix A ∈ gl(n), the Cauchy problem

˙

X = XA, X(0) = X

0

, X ∈ GL(n),

has a (unique) solution of the form

X(t) = X

0

exp(tA).

What can we say about a similar problem in any (linear) Lie group G:

˙

X = XA, X ∈ G ?

Example 2.20. Consider e.g. a Cauchy problem in the special linear group:

˙

X = XA, X(0) = Id, X ∈ SL(n). (4)

By uniqueness, solutions of this ODE must be given, as above, by the matrix

exponential, but the question is whether it is in the Lie group under considera-

tion:

X(t) = exp(tA) ∈ SL(n) ?

It is obvious that in general the answer is negative. Indeed, if X(t) = exp(tA) ∈

SL(n), then

A =

d

dt

¸

¸

¸

¸

t=0

exp(tA) ∈ T

Id

SL(n) = sl(n).

So if A / ∈ sl(n), then ODE (4) is not well-deﬁned, i.e., the vector ﬁeld XA is

not tangent to the Lie group SL(n).

What is the tangent space to a Lie group G at its point X ? This question

has a simple answer given in the following statement.

Proposition 2.1. Let G be a linear Lie group, L its Lie algebra, and let X ∈ G.

Then

T

X

G = XT

Id

G = XL = ¦XA [ A ∈ L¦.

Proof. Compute the tangent space

T

X

G = ¦

˙

X(0) [ X(t) ∈ G, X(0) = X¦.

For a smooth curve X(t) starting from X, one easily constructs a curve starting

from the identity:

Y (t) = X

−1

X(t), Y (0) = X

−1

X = Id .

13

Then

˙

Y (0) = X

−1

˙

X(0) ∈ L.

We denote A =

˙

Y (0) ∈ L and get

˙

X(0) = XA, A ∈ L.

Thus T

X

G ⊂ XL. Since these linear spaces have the same dimension, we obtain

T

X

G = XL.

So the left product by X translates the tangent space L at identity to the

tangent space XL at the point X.

Thus for any element

A ∈ L,

the vector

V (X) = XA ∈ T

X

G, X ∈ G,

i.e., the vector ﬁeld V (X) is tangent to the Lie group G. So the ODE

˙

X = XA, X ∈ G, (5)

is well-deﬁned and has the solutions

X(t) = X(0) exp(At) ∈ G.

Notice the following important property of ODE (5): if a curve X(t) is a

trajectory of the ﬁeld V (X) = XA, then its left translation Y X(t) is also a

trajectory of this ODE for any Y ∈ G. Indeed:

X(t) = X(0) exp(tA),

thus

Y (t) = Y X(t) = Y exp(tA)X(0) = Y (0) exp(tA).

Deﬁnition 2.5. Vector ﬁelds of the form

V (X) = XA, X ∈ G, A ∈ L,

are called left-invariant vector ﬁelds on the linear Lie group G.

Suppose we have two left-invariant vector ﬁelds on a Lie group G:

A, B ∈ L,

V (X) = XA, W(X) = XB, X ∈ G.

There arises a natural question: what is the Lie bracket of such vector ﬁelds?

Since the ﬁelds V and W are left-invariant, it is clear that the ﬁeld [V, W] is

left-invariant as well. In order to compute this ﬁeld, recall the deﬁnition of Lie

bracket of vector ﬁelds.

14

Deﬁnition 2.6. Let V and W be smooth vector ﬁelds on a smooth manifold M.

The Lie bracket (or commutator) of the ﬁelds V , W is the vector ﬁeld [V, W] ∈

Vec M such that

[V, W](X) =

d

d t

¸

¸

¸

¸

t=0

γ(

√

t), X ∈ M,

where the curve γ is deﬁned as follows:

γ(t) = e

−tW

◦ e

−tV

◦ e

tW

◦ e

tV

(X).

Here e

tV

denotes the ﬂow of the vector ﬁeld V :

d

d t

e

tV

(X) = V (e

tV

(X)), e

tV

¸

¸

t=0

(X) = X,

and Vec M denotes the space of all smooth vector ﬁelds on a smooth manifold M.

Now we compute the Lie bracket of left-invariant vector ﬁelds.

Proposition 2.2. Let G be a linear Lie group, L its Lie algebra, and let A, B ∈

L. Let V (X) = XA and W(X) = XB be left-invariant vector ﬁelds on G. Then

[V, W](X) = [XA, XB] = X[A, B] = X(AB −BA), X ∈ G.

Proof. The ﬂows of the left-invariant vector ﬁelds are given by the matrix ex-

ponential:

e

tV

(X) = X exp(tA), e

tW

(X) = X exp(tB).

Compute the low-order terms of the curve γ from Deﬁnition 2.6:

γ(t) = X exp(tA) exp(tB) exp(−tA) exp(−tB)

= X

_

Id +tA+

t

2

2

A

2

+

__

Id +tB +

t

2

2

B

2

+

_

_

Id −tA+

t

2

2

A

2

−

__

Id −tB +

t

2

2

B

2

−

_

= X

_

Id +t(A+B) +

t

2

2

(A

2

+ 2AB +B

2

) +

_

_

Id −t(A+B) +

t

2

2

(A

2

+ 2AB +B

2

) +

_

= X(Id+t

2

[A, B] + ),

thus

γ(

√

t) = X(Id+t[A, B] + ),

notice that it is a smooth curve at t = 0,

d

d t

¸

¸

¸

¸

t=0

γ(

√

t) = X[A, B].

By Deﬁnition 2.6, this is the Lie bracket [XA, XB].

15

Corollary 2.1. Left-invariant vector ﬁelds on a Lie group G form a Lie algebra

isomorphic to the Lie algebra L = T

Id

G. The isomorphism is deﬁned as follows:

left-invariant vector ﬁeld XA ∈ Vec G ↔ A ∈ L.

Thus in the sequel we identify these two representations of the Lie algebra

of a Lie group G:

(1) L = T

Id

G, and

(2) L = ¦left-invariant vector ﬁelds on G¦.

3 Left-invariant control systems

3.1 Deﬁnitions

Let G be a Lie group and L its Lie algebra.

Deﬁnition 3.1. A left-invariant control system Γ on a Lie group G is an arbi-

trary set of left-invariant vector ﬁelds on G, i.e., any subset

Γ ⊂ L. (6)

Example 3.1 (Control-aﬃne left-invariant systems). A particular class of

left-invariant systems, which is important for applications is formed by control-

aﬃne systems

Γ =

_

A+

m

i=1

u

i

B

i

[ u = (u

1

, . . . , u

m

) ∈ U ⊂ R

m

_

, (7)

where A, B

1

, . . . , B

m

are some elements of L. If the control set U coincides

with R

m

, then system (7) is an aﬃne subspace of L.

Remark. Throughout these notes, we write a left-invariant control system as (6)

or (7), i.e., as a set of vector ﬁelds, a polysystem. In the classical notation,

control-aﬃne systems (7) are written as follows:

˙

X = XA+

m

i=1

u

i

XB

i

, u = (u

1

, . . . , u

m

) ∈ U, X ∈ G. (8)

Polysystem (6) can also be written in such classical notation via a choice of a

parametrization of the set Γ.

Deﬁnition 3.2. A trajectory of a left-invariant system Γ on G is a continuous

curve X(t) in G deﬁned on an interval [t

0

, T] ⊂ R so that there exists a partition

t

0

< t

1

< < t

N

= T

16

and left-invariant vector ﬁelds

A

1

, . . . , A

N

∈ Γ

such that the restriction of X(t) to each open interval (t

i−1

, t

i

) is diﬀerentiable

and

˙

X(t) = X(t)A

i

for t ∈ (t

i−1

, t

i

), i = 1, . . . , N.

In the classical notation, this corresponds to piecewise-constant admissible

controls. In the study of global controllability for inﬁnite time we can restrict

ourselves by such a class of admissible controls.

Deﬁnition 3.3. For any T ≥ 0 and any X in G, the reachable set for time T of

a left-invariant system Γ ⊂ L from the point X is the set /

Γ

(X, T) of all points

that can be reached from X in exactly T units of time:

/

Γ

(X, T) = ¦X(T) [ X() a trajectory of Γ, X(0) = X¦.

The reachable set for time not greater than T ≥ 0 is deﬁned as

/

Γ

(X, ≤ T) =

_

0≤t≤T

/

Γ

(X, t).

The reachable (or attainable) set of a system Γ from a point X ∈ G is the set

/

Γ

(X) of all terminal points X(T), T ≥ 0, of all trajectories of Γ starting at X:

/

Γ

(X) = ¦X(T) [ X() a trajectory of Γ, X(0) = X, T ≥ 0¦ =

_

T≥0

/

Γ

(X, T).

If there is no ambiguity, in the sequel we denote the reachable sets /

Γ

(X, T)

and /

Γ

(X) by /(X, T) and /(X), respectively.

Deﬁnition 3.4. A system Γ ⊂ L is called controllable if, given any pair of

points X

0

and X

1

in G, the point X

1

can be reached from X

0

along a trajectory

of Γ for a nonnegative time:

X

1

∈ /(X

0

) for any X

0

, X

1

∈ G,

or in other words, if

/(X) = G for any X ∈ G.

In the control literature, this notion corresponds to global controllability, or

complete controllability. Although, for left-invariant systems these properties

are equivalent to local controllability at the identity, see Theorem 3.5 below.

3.2 Right-invariant control systems

Similarly to left-invariant vector ﬁelds

˙

X = XA, one can consider right-invari-

ant vector ﬁelds of the form

˙

Y = BY .

17

The inversion

i : G → G, i(X) = X

−1

= Y,

transforms left-invariant vector ﬁelds to right-invariant ones. Indeed, let X(t)

be a trajectory of a left-invariant ODE

˙

X = XA. Compute the ODE for

Y (t) = X

−1

(t). Since Y (t)X(t) = Id, we have

˙

Y (t)X(t) +Y (t)

˙

X(t) = 0, thus

˙

Y (t) = −Y (t)

˙

X(t)X

−1

(t) = −Y (t)X(t)AY (t) = −AY (t).

Consequently,

˙

X = XA ⇔

˙

Y = −AY, Y = X

−1

.

Since X(t) = X

0

e

tA

, then Y (t) = e

−tA

Y

0

.

Notice that similarly to Proposition 2.1, it is easy to show that T

X

G = LX,

and that the Lie algebra L = T

Id

G of a Lie group G can be identiﬁed with the

Lie algebra of right-invariant vector ﬁelds ¦AX [ A ∈ L¦ on G.

Deﬁnition 3.5. A right-invariant control system on a Lie group G is an arbi-

trary set of right-invariant vector ﬁelds on G.

Deﬁnition 3.6. A control-aﬃne right-invariant control system on a Lie group

G has the form

˙

Y = AY +

m

i=1

u

i

B

i

Y, u ∈ U ⊂ R

m

, Y ∈ G. (9)

The inversion X = Y

−1

transforms right-invariant system (9) to the left-in-

variant system

˙

X = −XA−

m

i=1

u

i

XB

i

, u ∈ U, X ∈ G.

Summing up, all problems for right-invariant control systems are reduced to

the study of left-invariant systems via inversion.

3.3 Basic properties of orbits and reachable sets

Let G be a linear Lie group, and let L be its Lie algebra, i.e., the space of

left-invariant vector ﬁelds on G.

Lemma 3.1. Let A ∈ L and X

0

∈ G. Then the Cauchy problem

˙

X = XA, X(t

0

) = X

0

,

has the solution X(t) = X

0

exp((t −t

0

)A).

Due to this obvious lemma we can obtain a description of an endpoint of a

trajectory via product of exponentials.

18

Lemma 3.2. Let X(t), t ∈ [0, T], be a trajectory of a left-invariant system

Γ ⊂ L with X(0) = X

0

. Then there exist N ∈ N and

τ

1

, . . . , τ

N

> 0, A

1

, . . . A

N

∈ Γ

such that

X(T) = X

0

exp(τ

1

A

1

) exp(τ

N

A

N

),

τ

1

+ +τ

N

= T.

Proof. By the deﬁnition of a trajectory, there exist N ∈ N and

0 = t

0

< t

1

< < t

N

= T, A

1

, . . . , A

N

∈ Γ

such that X(t) is continuous and

t ∈ (t

i−1

, t

i

) ⇒

˙

X(t) = X(t)A

i

.

Consider the ﬁrst interval:

t ∈ (0, t

1

) ⇒

˙

X = X(t)A

1

, X(0) = X

0

.

Thus

X(t) = X

0

exp(A

1

t), X(t

1

) = X

0

exp(A

1

t

1

).

Further,

t ∈ (t

1

, t

2

) ⇒

˙

X = X(t)A

2

, X(t

1

) = X

0

exp(A

1

t

1

).

So

X(t) = X

0

exp(t

1

A

1

) exp((t −t

1

)A

2

),

X(t

2

) = X

0

exp(A

1

t

1

) exp((t

2

−t

1

)A

2

)

= X

0

exp(A

1

τ

1

) exp(τ

2

A

2

), τ

1

= t

1

, τ

2

= t

2

−t

1

.

We go on in such a way and ﬁnally obtain the required representation:

X(t

N

) = X(T) = X

0

exp(τ

1

A

1

) exp(τ

N

A

N

),

τ

N

= t

N

−t

N−1

, . . . , τ

2

= t

2

−t

1

, τ

1

= t

1

,

τ

N

+ +τ

1

= t

N

= T.

Now we can obtain a description of attainable sets and derive their elemen-

tary properties.

Lemma 3.3. Let Γ ⊂ L be a left-invariant system, and let X be an arbitrary

point of G. Then

19

(1) /

Γ

(X) = ¦X exp(t

1

A

1

) exp(t

N

A

N

) [ A

i

∈ Γ, t

i

> 0, N ≥ 0¦;

(2) /

Γ

(X) = X/

Γ

(Id);

(3) /

Γ

(Id) is a subsemigroup of G;

(4) /

Γ

(X) is an arcwise-connected subset of G.

Proof. Items (1) and (2) follow immediately from Lemma 3.2.

(3) Since

/

Γ

(Id) = ¦exp(t

1

A

1

) exp(t

N

A

N

) [ A

i

∈ Γ, t

i

> 0, N ≥ 0¦,

then for any X

1

, X

2

∈ /

Γ

(Id), the product X

1

X

2

∈ /

Γ

(Id).

(4) Any point in /

Γ

(X) is connected with the initial point X by a trajec-

tory X(t).

Deﬁnition 3.7. The orbit of a system Γ through a point X ∈ G is the following

subset of the Lie group G:

O

Γ

(X) = ¦X exp(t

1

A

1

) exp(t

N

A

N

) [ A

i

∈ Γ, t

i

∈ R, N ≥ 0¦,

compare with the description of attainable set /

Γ

(X) given in item (1) of

Lemma 3.3.

Obviously,

/

Γ

(X) ⊂ O

Γ

(X).

In the orbit, one is allowed to move both forward and backward in time, while

in the attainable set only the forward motion is allowed. The structure of orbits

is simpler than that of attainable sets.

Lemma 3.4. Let Γ ⊂ L be a left-invariant system, and let X be an arbitrary

point of G. Then

(1) O

Γ

(X) = XO

Γ

(Id);

(2) O

Γ

(Id) is the connected Lie subgroup of G with the Lie algebra Lie(Γ).

Here and below we denote by Lie(Γ) the Lie algebra generated by Γ, i.e., the

smallest Lie subalgebra of L containing Γ.

Proof. Item (1) is obvious.

(2) First of all, the orbit O

Γ

(Id) is connected since any point in it is connected

with the identity by a continuous curve provided by the deﬁnition of an orbit.

Further, it is easy to see that O

Γ

(Id) is a subgroup of G. If X, Y ∈ O

Γ

(Id),

then XY ∈ O

Γ

(Id) as a product of exponentials. If

X = exp(t

1

A

1

) exp(t

N

A

N

) ∈ O

Γ

(Id),

then

X

−1

= exp(−t

N

A

N

) exp(−t

1

A

1

) ∈ O

Γ

(Id).

20

Finally, Id ∈ O

Γ

(Id).

It follows from the general Orbit Theorem (see [14], [1]) that O

Γ

(Id) ⊂ G is

a smooth submanifold with the tangent space T

Id

O

Γ

(Id) = Lie(Γ).

Then the orbit O

Γ

(Id) is a Lie subgroup of G with the Lie algebra Lie(Γ),

see [45].

Proposition 3.1. A left-invariant system Γ is controllable iﬀ /

Γ

(Id) = G.

Proof. By deﬁnition, Γ is controllable iﬀ /

Γ

(X) = G for any X ∈ G. Since

/

Γ

(X) = X/

Γ

(Id), controllability is equivalent to the identity /

Γ

(Id) = G.

That is why in the sequel we use the following short notation for the attain-

able set and orbit from the identity:

/

Γ

(Id) = /

Γ

= /, O

Γ

(Id) = O

Γ

= O.

Given any subset l of a vector space V , we denote by span(l) the vector

subspace of V generated by l and by co(l) the positive convex cone generated

by the set l.

We denote the topological closure and the interior of a set S by cl S and

int S, respectively.

3.4 Normal attainability

If a point Y ∈ Gis reachable (or attainable) from a point X ∈ G, then there exist

elements A

1

, . . . , A

N

∈ Γ and t = (t

1

, . . . , t

N

) ∈ R

N

with positive coordinates

such that

Y = X exp(t

1

A

1

) exp(t

N

A

N

).

That is, the point Y is in the image of the mapping

F : (s

1

, . . . , s

N

) → X exp(s

1

A

1

) exp(s

N

A

N

), s

i

> 0.

The following stronger notion turns out to be important in the study of

topological properties of reachable sets and controllability.

Deﬁnition 3.8. A point Y ∈ G is called normally attainable from a point

X ∈ G by Γ if there exist elements A

1

, . . . , A

N

in Γ and t ∈ R

N

with positive

coordinates t

1

, . . . , t

N

such that the mapping

F : R

N

→ G, F(t

1

, . . . , t

N

) = X exp(t

1

A

1

) exp(t

N

A

N

)

satisﬁes the following conditions:

(i) F(t) = Y .

(ii) rank D

t

F = dimG.

The point Y is said to be normally accessible from X by A

1

, . . . , A

N

.

21

Lemma 3.5. If a point Y ∈ G is normally attainable from X ∈ G by Γ, then

Y ∈ int /

Γ

(X).

Proof. The mapping F is a local diﬀeomorphism near t. That is, there exists a

neighborhood V ÷ t such that the restriction F[

V

is a diﬀeomorphism. Then

the set F(V ) is open. On the other hand, for any s = (s

1

, . . . , s

N

), s

i

> 0, the

point F(s) is in /

Γ

(X). Thus F(V ) ⊂ /

Γ

(X) is a neighborhood of the point

Y = F(t).

Theorem 3.1 (Krener). Let Lie(Γ) = L. Then:

(1) In any neighborhood V of the identity Id ∈ G, there are points normally

attainable from Id by Γ;

(2) Consequently, for any neighborhood V ÷ Id, the intersection int /∩ V is

nonempty;

(3) In particular, the interior / is nonempty.

Proof. We prove item (1) since items (2) and (3) follow from it.

Denote n = dimL = dimLie(Γ). If n = 0, everything is clear. Assume that

n ≥ 1 and ﬁx a neighborhood V of the identity Id.

There exists a nonzero element A

1

∈ Γ, otherwise dimLie(Γ) = 0. Consider

the mapping

F

1

: s

1

→ exp(s

1

A

1

), 0 < s

1

< ε

1

,

for suﬃciently small positive ε

1

. We have

d F

1

d s

1

¸

¸

¸

¸

s1=0

= A

1

,= 0, consequently,

rank D

s1

F

1

= 1 for small s

1

. The curve

M

1

= ¦F

1

(s

1

) [ 0 < s

1

< ε

1

¦

is a smooth one-dimensional manifold contained in the neighborhood V for

suﬃciently small positive ε

1

. If n = 1, then any point X

1

∈ M

1

is normally

attainable from Id by A

1

.

If n > 1, there exist an element A

2

∈ Γ and a point X

1

∈ M

1

as close to

identity as we wish such that

X

1

A

2

/ ∈ T

X1

M

1

.

Otherwise Lie(Γ)(X

1

) ⊂ T

X1

M

1

for any X

1

∈ M

1

and dimLie(Γ) ≤ dimM

1

=

1, a contradiction. We have

X

1

= exp(t

1

1

A

1

) for some t

1

1

> 0.

Consider the mapping

F

2

: (s

1

, s

2

) → exp((t

1

1

+s

1

)A

1

) exp(s

2

A

2

), 0 < s

i

< ε

i

.

For small s > 0 we have rank D

s

F

2

= 2, thus the set

M

2

= ¦F

2

(s

1

, s

2

) [ 0 < s

i

< ε

i

, i = 1, 2¦

22

is a smooth two-dimensional manifold that belongs to V for suﬃciently small

positive ε

1

and ε

2

. If n = 2, the theorem is proved, since in this case any point

of M

2

is normally attainable from Id by A

1

and A

2

.

If n > 2, we proceed in a similar manner. There exist A

3

∈ Γ and X

2

∈ M

2

close to Id such that

X

2

A

3

/ ∈ T

X2

M

2

.

Otherwise Lie(Γ)(X

2

) ⊂ T

X2

M

2

for any X

2

∈ M

2

and dimLie(Γ) ≤ 2, a con-

tradiction. We have

X

2

= exp(t

2

1

A

1

) exp(t

2

2

A

2

) for some t

2

i

> 0.

Consider the mapping

F

3

: (s

1

, s

2

, s

3

) → exp((t

2

1

+s

1

)A

1

) exp((t

2

2

+s

2

)A

2

) exp(s

3

A

3

), 0 < s

i

< ε

i

.

Since the vector ﬁeld A

3

is not tangent to the manifold M

2

at the point X

2

, the

diﬀerential D

s

F

3

has rank 3 for small s > 0. Thus

M

3

= ¦F

3

(s

1

, s

2

, s

3

) [ 0 < s

i

< ε

i

, i = 1, 2, 3¦

is a smooth three-dimensional manifold belonging to V for suﬃciently small

positive ε

i

. In the case n = 3, the theorem is proved, otherwise we proceed by

induction.

As a result of the inductive construction, we ﬁnd an element A

n

∈ Γ and a

point

X

n−1

= exp(t

n−1

1

A

1

) exp(t

n−1

n−1

A

n−1

) ∈ M

n−1

suﬃciently close to Id such that

X

n−1

A

n

/ ∈ T

Xn−1

M

n−1

.

Then the mapping

F

n

: (s

1

, . . . , s

n

)

→ exp((t

n−1

1

+s

1

)A

1

) exp((t

n−1

n−1

+s

n−1

)A

n−1

) exp(s

n

A

n

),

0 < s

i

< ε

i

,

is an immersion for small s > 0. Consequently, any point X

n

∈ M

n

= ImF

n

is

normally attainable from Id. Moreover, X

n

can be chosen as close to Id as we

wish.

Deﬁnition 3.9. A system Γ ⊂ L is said to have a full rank (or to satisfy the

Lie Algebra Rank Condition) if

Lie(Γ) = L.

Proposition 3.2. Let Γ ⊂ L. Then:

23

(1) int

O

/ ,= ∅;

(2) Moreover, / ⊂ cl int

O

/.

We denote by int

O

S the interior of a subset S of the orbit O in the topology

of O.

Proof. (a) Assume ﬁrst that the system Γ has full rank: Lie(Γ) = L, then the

orbit O ⊂ G is an open subset, and the relative interior w.r.t. O coincides

with the interior in G. By Krener’s theorem, int / ,= ∅, and item (1) of this

proposition follows.

We prove item (2). Take any element X ∈ / and any its neighborhood

V ÷ X. Then the open set V X

−1

is a neighborhood of the identity. By Krener’s

theorem, there exists a point Y ∈ V X

−1

∩ int /, thus Y X ∈ V . Further, since

Y ∈ int /, there exists a neighborhood W ÷ Y , W ⊂ /. Then the open

set WX is a neighborhood of the point Y X, moreover, WX ⊂ /. Finally,

Y X ∈ int / ∩ V . Consequently, any neighborhood V of the point X contains

points from int /, thus X ∈ cl int /.

(b) If Lie(Γ) ,= L, we consider the restriction of the system Γ to the orbit O,

a Lie subgroup of G with the Lie algebra Lie(Γ). The system Γ is full-rank

on O, thus the statement in the case (b) follows from the case (a).

3.5 General controllability conditions

Let G be a linear Lie group, L its Lie algebra, and Γ ⊂ L a left-invariant system

on G. In this subsection we prove some basic controllability conditions for Γ

on G.

Theorem 3.2 (Connectedness Condition). If Γ ⊂ L is controllable on G,

then the Lie group G is connected.

Proof. The attainable set / is a connected subset of G.

Example 3.2. The Lie group GL(n) is not connected since it consists of two

connected components GL

+

(n) and GL

−

(n), where

GL

±

(n) = ¦X ∈ M(n) [ sign(det X) = ±1¦.

Thus there are no controllable systems on GL(n), but a reasonable question to

study is controllability on its connected component of identity GL

+

(n).

Example 3.3. Similarly, the orthogonal group O(n) = SO(n) ∪ O

−

(n) is dis-

connected, where O

−

(n) = ¦X ∈ O(n) [ det X = −1¦. So there no controllable

systems on O(n); instead, one can study controllability on SO(n).

Theorem 3.3 (Rank Condition). Let Γ ⊂ L.

(1) If Γ is controllable, then Lie(Γ) = L.

(2) int / ,= ∅ if and only if Lie(Γ) = L.

24

Proof. (1) If Γ is controllable, then / = G, the more so O = G, thus Lie(Γ) = L.

(2) By Krener’s theorem, if Lie(Γ) = L, then int / ,= ∅.

Conversely, let Lie(Γ) ,= L. Then dimO = dimLie(Γ) < dimL = dimG.

Thus int O = ∅, the more so int / = ∅.

Theorem 3.4 (Group Test). A system Γ ⊂ L is controllable on a Lie group G

iﬀ the following conditions hold:

(1) G is connected,

(2) Lie(Γ) = L,

(3) the attainable set / is a subgroup of G.

Proof. The necessity is obvious, we prove suﬃciency. If / ⊂ G is a subgroup,

then for any element X ∈ /, its inverse X

−1

belongs to / as well. Recall the

descriptions of the attainable set and orbit through identity:

/ = ¦exp(t

1

A

1

) exp(t

N

A

N

) [ t

i

≥ 0, A

i

∈ Γ¦,

O = ¦exp(±t

1

A

1

) exp(±t

N

A

N

) [ t

i

≥ 0, A

i

∈ Γ¦.

For any exponential exp(t

i

A

i

) ∈ /, the inverse

(exp(t

i

A

i

))

−1

= exp(−t

i

A

i

) ∈ /,

thus the attainable set / coincides with the orbit O. But O ⊂ G is a connected

Lie subgroup with Lie algebra Lie(Γ) = L. Then it follows that O = G, see [45].

Thus / = O = G.

Deﬁnition 3.10. A control system is called locally controllable at a point X if

X ∈ int /(X).

Theorem 3.5 (Local Controllability Test). A system Γ ⊂ L is controllable

on a Lie group G iﬀ the following conditions hold:

(1) G is connected,

(2) Γ is locally controllable at the identity.

Notice that identity element is always contained in the attainable set, and

there may be two cases: either Id ∈ int /, or Id ∈ ∂/. In the ﬁrst case the

system is controllable, while in the second case not.

Now we prove Theorem 3.5.

Proof. The necessity is obvious, we prove suﬃciency. There exists a neighbor-

hood V ÷ Id such that V ⊂ /. Consider the powers of this neighborhood:

V

n

⊂ / for any n ∈ N. But the Lie group G is connected, thus it is generated

by any neighborhood of identity [45]:

_

n∈N

V

n

= G.

Then / ⊃

n∈N

V

n

= G, thus / = G.

25

Theorem 3.6 (Closure Test). A system Γ ⊂ L is controllable on a Lie

group G iﬀ the following conditions hold:

(1) Lie(Γ) = L,

(2) cl / = G.

Proof. The necessity is straightforward. Let us prove the suﬃciency. Consider

the time-reversed system

−Γ = ¦−A [ A ∈ Γ¦.

Trajectories of the system −Γ are trajectories of the initial system Γ passed in

the opposite direction, thus

/

−Γ

= ¦exp(−t

1

A

1

) exp(−t

N

A

N

) [ t

i

≥ 0, A

i

∈ Γ¦

=

_

(exp(t

N

A

N

) exp(t

1

A

1

))

−1

[ t

i

≥ 0, A

i

∈ Γ

_

= /

−1

Γ

.

Since Lie(−Γ) = Lie(Γ) = L, it follows that int /

−Γ

,= ∅, thus there exists an

open subset V ⊂ /

−Γ

. Further, by the hypothesis of this theorem, cl /

Γ

= G,

thus there exists a point X ∈ /

Γ

∩V ,= ∅. We have X ∈ V ⊂ /

−Γ

= /

−1

Γ

, thus

the open set V

−1

⊂ /

Γ

is a neighborhood of the inverse X

−1

. Consequently,

the open set V

−1

X ⊂ /

Γ

. But Id = X

−1

X ∈ V

−1

X ⊂ /

Γ

, thus Id ∈ int /

Γ

,

and the system Γ is controllable by Theorem 3.5.

The previous theorem has important far-reaching consequences. It means

that in the study of controllability of full-rank systems one can replace the

attainable set /by its closure cl /. This idea gives rise to the powerful extension

techniques described in the following section.

4 Extension techniques

for left-invariant systems

4.1 Saturate

Deﬁnition 4.1. Let Γ

1

, Γ

2

⊂ L. The system Γ

1

is called equivalent to the

system Γ

2

: Γ

1

∼ Γ

2

if

cl /

Γ1

= cl /

Γ2

.

It is easy to show that not only the attainable set /, but also its closure is

a semigroup.

Lemma 4.1. Let Γ ⊂ L. Then cl /

Γ

is a subsemigroup of G.

Proof. Let X, Y ∈ cl /

Γ

. Then there exist sequences

¦X

n

¦, ¦Y

n

¦ ⊂ /

Γ

such that X

n

→ X, Y

n

→ Y as n → ∞.

26

Then

¦X

n

Y

n

¦ ⊂ /

Γ

and X

n

Y

n

→ XY as n → ∞.

Lemma 4.2. If Γ

1

∼ Γ and Γ

2

∼ Γ, then Γ

1

∪ Γ

2

∼ Γ.

Proof. We have cl /

Γ1

= cl /

Γ2

= cl /

Γ

. The inclusion

cl /

Γ

⊂ cl /

Γ1∪Γ2

(10)

is obvious in view of the the chain cl /

Γ

= cl /

Γ1

⊂ cl /

Γ1∪Γ2

.

Now we prove the inclusion

/

Γ1∪Γ2

⊂ cl /

Γ

. (11)

Take an arbitrary element

X = exp(t

1

A

1

) exp(t

N

A

N

) ∈ /

Γ1∪Γ2

, t

i

≥ 0, A

i

∈ Γ

1

∪ Γ

2

.

We have

exp(t

i

A

i

) ∈ /

Γ1

∪ /

Γ2

⊂ cl /

Γ

,

thus by Lemma 4.1 it follows that X ∈ cl /

Γ

. So inclusion (11) is proved, and

cl /

Γ1∪Γ2

⊂ cl /

Γ

. In view of inclusion (10), it follows that Γ

1

∪ Γ

2

∼ Γ.

The previous lemma allows one to unite equivalent systems. It is then natural

to consider the union of all systems equivalent to a given one.

Deﬁnition 4.2. The saturate of a right-invariant system Γ ⊂ L is the following

system:

Sat(Γ) = ∪¦Γ

⊂ L [ Γ

∼ Γ¦.

Proposition 4.1. (1) Sat(Γ) ∼ Γ.

(2) Sat(Γ) = ¦A ∈ L [ exp(R

+

A) ⊂ cl /

Γ

¦.

Item (1) means that the saturate of Γ is the largest right-invariant system

on G equivalent to Γ, while item (2) describes Sat(Γ) as a kind of a tangent

object to cl /

Γ

at the identity.

Proof. (1) Obviously, Γ ∼ Γ, thus Γ ⊂ Sat(Γ), so cl /

Γ

⊂ cl /

Sat(Γ)

. In order to

prove the inclusion

/

Sat(Γ)

⊂ cl /

Γ

, (12)

take any element

X = exp(t

1

A

1

) exp(t

N

A

N

) ∈ /

Sat(Γ)

, t

i

> 0, A

i

∈ Sat(Γ).

Each element A

i

is contained in a system Γ

i

∼ Γ, thus exp(t

i

A

i

) ∈ /

Γi

⊂ cl /

Γ

.

By the semigroup property, cl /

Γ

÷ X. Inclusion (12) and item (1) follow.

27

(2) Denote the system

´

Γ = ¦A ∈ L [ exp(R

+

A) ⊂ cl /

Γ

¦.

First we prove the inclusion

´

Γ ⊂ Sat(Γ). (13)

We show that

´

Γ ∼ Γ. Consider the representation

/

Γ

= ¦exp(t

1

A

1

) exp(t

N

A

N

) [ t

i

> 0, A

i

∈

´

Γ¦.

Since all exp(t

i

A

i

) ∈ cl /

Γ

, it follows that /

Γ

⊂ cl /

Γ

. Moreover, since Γ ⊂

´

Γ,

then /

Γ

⊂ /

Γ

. Thus cl /

Γ

= cl /

Γ

, hence

´

Γ ∼ Γ. Inclusion (13) is proved.

In order to prove the reverse inclusion

Sat(Γ) ⊂

´

Γ, (14)

take any element A ∈ Sat(Γ). Then A ∈ Γ

∼ Γ. Thus exp(tA) ∈ cl /

Γ

, i.e.,

A ∈

´

Γ. Inclusion (14) follows. Taking into account inclusion (13), we obtain the

required equality: Sat(Γ) =

´

Γ.

Remark. Unfortunately, the saturate is not the appropriate tangent object to

cl / responsible for controllability: it is possible that Sat(Γ) = L, and never-

theless Γ is not controllable.

Example 4.1 (Irrational winding of the torus). The torus is a two-

dimensional Abelian Lie group:

G = T

2

= S

1

S

1

= ¦(x mod 1, y mod 1)¦.

Its Lie algebra is

L = T

Id

T

2

= R

2

.

Consider the following right-invariant system on G:

Γ = ¦A¦, A = (1, r), r ∈ R ¸ Q.

The attainable set is the irrational winding of the torus:

/ = exp(R

+

A) = ¦(x mod 1, kx mod 1) [ x ≥ 0¦ ,= T

2

,

cl / = T

2

.

Thus

Γ ∼ L = Sat(Γ),

although Γ is not controllable on T

2

. The reason is clear — the rank condition

is violated:

Lie(Γ) = RA ,= L.

28

4.2 Lie saturate of invariant system

It is the following Lie-generated tangent object to cl /

Γ

that is responsible for

controllability of Γ.

Deﬁnition 4.3. Lie saturate of a left-invariant system is deﬁned as follows:

LS(Γ) = Lie(Γ) ∩ Sat(Γ).

The following description of Lie Saturate follows immediately from Proposi-

tion 4.1.

Corollary 4.1. LS(Γ) = ¦A ∈ Lie(Γ) [ exp(R

+

A) ⊂ cl /

Γ

¦.

Theorem 4.1 (Lie Saturate Test). A left-invariant system Γ ⊂ L is control-

lable on a connected Lie group G if and only if LS(Γ) = L.

Proof. Necessity follows from the deﬁnition of the Lie saturate.

Suﬃciency. Assume that LS(Γ) = L. The connected Lie group G is gener-

ated by the one-parameter semigroups ¦exp(tA) [ A ∈ L, t ≥ 0¦ as a semigroup;

thus the equality Sat(Γ) = L implies that cl(/) = G. Since, in addition, the

rank condition Lie(Γ) = L holds, then Γ is controllable by Theorem 3.6.

The basic properties of Lie saturate are collected in the following proposition.

Theorem 4.2. (1) LS(Γ) is a closed convex positive cone in L, i.e.,

(1a) LS(Γ) is topologically closed:

cl(LS(Γ)) = LS(Γ),

(1b) LS(Γ) is convex:

A, B ∈ LS(Γ) ⇒ αA + (1 −α)B ∈ LS(Γ) ∀ α ∈ [0, 1],

(1c) LS(Γ) is a positive cone:

A ∈ LS(Γ) ⇒ αA ∈ LS(Γ) ∀ α ≥ 0.

Thus,

A, B ∈ LS(Γ) ⇒ αA +βB ∈ LS(Γ) ∀ α, β ≥ 0.

(2) For any ±A, B ∈ LS(Γ) and any s ∈ R,

exp(s ad A)B = B + (s adA)B +

(s ad A)

2

2!

B +. . . +

(s adA)

n

n!

B +

∈ LS(Γ).

(3) If ±A, ±B ∈ LS(Γ), then ±[A, B] ∈ LS(Γ).

29

(4) If A ∈ LS(Γ) and if the one-parameter subgroup ¦exp(tA) [ t ∈ R¦ is

periodic (i.e., compact), then −A ∈ LS(Γ).

(5) Moreover, if A ∈ LS(Γ) and if the one-parameter subgroup ¦exp(tA) [ t ∈

R¦ is quasi-periodic:

exp(R

−

A) ⊂ cl exp(R

+

A), (15)

then −A ∈ LS(Γ).

We denote by ad A the adjoint operator corresponding to A ∈ L:

adA : L → L, adA : B → [A, B].

Proof. (1a) Take a converging sequence LS(Γ) ÷ A

n

→ A ∈ L. Since the linear

space Lie(Γ) is closed, we have

A

n

∈ Lie(Γ) ⇒ A ∈ Lie(Γ).

Further, it follows that Sat(Γ) is closed as well: since A

n

∈ Sat(Γ), we have

cl / ÷ exp(tA

n

) → exp(tA) ∈ cl /, t ≥ 0,

and A ∈ Sat(Γ). Consequently, LS(Γ) = Lie(Γ) ∩ Sat(Γ) is topologically closed.

(1b) Take any A, B ∈ LS(Γ), α ∈ [0, 1], and consider the convex combination

C = αA+βB, β = 1 −α. It is easy to see (exercise) that

exp(tC) = lim

n→∞

exp

_

α

n

tA

_

exp

_

β

n

tB

_

exp

_

α

n

tA

_

exp

_

β

n

tB

_

. ¸¸ .

n pairs

,

thus C ∈ Sat(Γ), and Sat(Γ) is convex. Since the linear space Lie(Γ) is convex,

it follows that LS(Γ) is convex as well.

(1c) It is easy to show that LS(Γ) is a cone. Take any A ∈ LS(Γ), α > 0.

Then exp(tαA) ∈ cl /, t ≥ 0, i.e, αA ∈ LS(Γ).

To prove (2), assume that ±A, B ∈ LS(Γ). Denote the element

B

s

= exp(s adA)B, s ∈ R. (16)

It is easy to see that this element admits the following representation:

B

s

= exp(sA)B exp(−sA). (17)

Indeed, the both curves (16) and (17) are solutions to the Cauchy problem

B

0

= B,

d

d s

B

s

= [A, B

s

].

Further, it is obvious from (16) that B

s

∈ Lie(Γ). Representation (17) implies

that

exp(tB

s

) = exp(tA) exp(sB) exp(−tA) ∈ cl(/

Γ

)

30

for any t ≥ 0, s ∈ R; thus B

s

∈ LS(Γ) for all s ∈ R.

Now (3) easily follows: if ±A, ±B ∈ LS(Γ), then ±e

t ad A

B, ±B ∈ LS(Γ),

that is why

±[A, B] = ±lim

t→0

e

t ad A

B −B

t

∈ LS(Γ).

(4) follows from the chain

¦exp(tA) [ t ≥ 0¦ = ¦exp(tA) [ t ∈ R¦ ⊂ cl /

Γ

,

which is valid for all A ∈ LS(Γ) with a periodic one-parameter group.

Finally, we prove a more strong property (5). It follows from the quasi-

periodic property (15) that

exp(−tA) = exp(t(−A)) ∈ cl exp(R

+

A) ⊂ cl /

Γ

for any t ≥ 0, thus −A ∈ LS(Γ).

Usually, it is diﬃcult to construct the Lie saturate of a right-invariant system

explicitly. That is why Theorems 4.1 and 4.2 are applied as suﬃcient conditions

of controllability via the following procedure. Starting from a given system Γ,

one constructs a completely ordered ascending family of extensions ¦Γ

α

¦ of Γ,

i.e.,

Γ

0

= Γ, Γ

α

⊂ Γ

β

if α < β.

The extension rules are provided by Theorem 4.2:

(1) given Γ

α

, one constructs Γ

β

= cl(co(Γ

α

));

(2) for ±A, B ∈ Γ

α

, one constructs Γ

β

= Γ

α

∪ e

Rad A

B;

(3) for ±A, ±B ∈ Γ

α

, one constructs Γ

β

= Γ

α

∪ R[A, B];

(4, 5) given A ∈ Γ

α

with periodic or quasi-periodic one-parameter group, one

constructs Γ

β

= Γ

α

∪ RA.

Theorem 4.2 guarantees that all extensions Γ

α

belong to LS(Γ). If one obtains

the relation Γ

α

= L at some step α, then LS(Γ) = L, and the system Γ is

controllable by Theorem 4.1.

5 Induced systems on homogeneous spaces

Example 5.1 (Bilinear systems). Consider the following right-invariant sys-

tem on GL

+

(n):

Γ = A+RB ⊂ gl(n).

In the classical notation, this system reads

˙

X = AX +uBX, X ∈ GL

+

(n), u ∈ R. (18)

31

Introduce also the following bilinear system:

˙ x = Ax +uBx, x ∈ R

n

¸ ¦0¦, u ∈ R. (19)

We exclude the origin from R

n

since linear vector ﬁelds vanish at the origin,

thus it is an equilibrium for bilinear systems.

If X(t) is a trajectory of the right-invariant system (18) with X(0) = Id,

then the curve x(t) = X(t)x

0

is a trajectory of the bilinear system (19) with

x(0) = x

0

.

Assume that the right-invariant system (18) is controllable on GL

+

(n). Then

it is easy to see that the bilinear system (19) is controllable on R

n

¸¦0¦. Indeed,

take any two points x

0

, x

1

∈ R

n

¸ ¦0¦. There exists a matrix X

1

∈ GL

+

(n)

such that X

1

x

0

= x

1

. By virtue of controllability of Γ, there exists a trajectory

X(t) of the right-invariant system such that X(0) = Id, X(T) = X

1

for some

T ≥ 0. Then the trajectory x(t) = X(t)x

0

of the bilinear system steers x

0

to

x

1

:

x(0) = X(0)x

0

= Id x

0

= x

0

, x(T) = X(T)x

0

= X

1

x

0

= x

1

.

We showed that if the right-invariant system (18) is controllable on GL

+

(n),

then the bilinear system (19) is controllable on R

n

¸ ¦0¦.

There were three key points in the preceding argument.

(1) The Lie group G = GL

+

(n) acts on the manifold M = R

n

¸ ¦0¦, that is,

any X ∈ G deﬁnes a mapping

X : M → M, X : x → Xx.

(2) G acts transitively on M:

∀ x

0

, x

1

∈ M ∃ X ∈ G such that Xx

0

= x

1

.

(3) The bilinear system (19) is induced by the right-invariant system (18):

if X(t) is a trajectory of (18), then X(t)x is a trajectory of (19).

This construction generalizes as follows.

Deﬁnition 5.1. A Lie group G is said to act on a smooth manifold M if there

exists a smooth mapping

θ : GM → M

that satisﬁes the following conditions:

(1) θ(Y X, x) = θ(Y, θ(X, x)) for any X, Y ∈ G and any x ∈ M;

(2) θ(Id, x) = x for any x ∈ M.

Deﬁnition 5.2. A Lie group G acts transitively on M if for any x

0

, x

1

∈ M

there exists X ∈ G such that θ(X, x

0

) = x

1

. A manifold that admits a transitive

action of a Lie group is called the homogeneous space of this Lie group.

32

Deﬁnition 5.3. Let A ∈ L. The vector ﬁeld θ

∗

A ∈ Vec M induced by the

action θ is deﬁned as follows:

(θ

∗

A)(x) =

d

dt

¸

¸

¸

¸

t=0

θ(exp(tA), x), x ∈ M.

Example 5.2. The Lie group GL

+

(n) acts transitively on R

n

¸ ¦0¦ as follows:

θ(X, x) = Xx, X ∈ GL

+

(n), x ∈ R

n

¸ ¦0¦.

For a right-invariant vector ﬁeld V (X) = AX, its ﬂow through the identity is

e

V t

(Id) = exp(At), thus

(θ

∗

V )(x) =

d

dt

¸

¸

¸

¸

t=0

θ(e

V t

(Id), x) =

d

dt

¸

¸

¸

¸

t=0

exp(At)x = Ax.

Deﬁnition 5.4. Let Γ ⊂ L be a right-invariant system. The system

θ

∗

Γ ⊂ Vec M,

(θ

∗

Γ)(x) = ¦(θ

∗

A)(x) [ A ∈ Γ¦, x ∈ M,

is called the induced system on M.

Example 5.3. Let Γ = ¦A + uB [ u ∈ R¦ ⊂ L be a right-invariant system on

a linear Lie group G ⊂ GL(n). In the classical notation, Γ reads as

˙

X = AX +uBX, X ∈ G, u ∈ R.

We have θ

∗

(AX) = Ax, θ

∗

(BX) = Bx, thus θ

∗

(AX + uBX) = Ax + uBx. So

the induced system θ

∗

Γ is bilinear:

˙ x = Ax +uBx, x ∈ R

n

¸ ¦0¦, u ∈ R.

Lemma 5.1. If X(t) is a trajectory of a right-invariant system Γ, then x(t) =

θ(X(t), x

0

) is a trajectory of the induced system θ

∗

Γ for any x

0

∈ M.

Proof. We can consider the case where the whole trajectory X(t) satisﬁes a

single ODE

˙

X = AX(t), A ∈ Γ, since an arbitrary trajectory of Γ is a concate-

nation of such pieces. Then X(t) = exp(At)X

0

and x(t) = θ(exp(At)X

0

, x

0

).

Then the required ODE is veriﬁed by diﬀerentiation:

˙ x(t) =

d

d t

θ(exp(At)X

0

, x

0

) =

d

d ε

¸

¸

¸

¸

ε=0

θ(exp(A(t +ε))X

0

, x

0

)

=

d

d ε

¸

¸

¸

¸

ε=0

θ(exp(Aε), θ(exp(At)X

0

, x

0

)

. ¸¸ .

x(t)

)

= (θ

∗

A)(x(t)).

33

Theorem 5.1. Let θ be a transitive action of a Lie group G on a manifold M,

let Γ ⊂ L be a right-invariant system on G, and let θ

∗

Γ ⊂ Vec M be the induced

system on M.

(1) If Γ is controllable on G, then θ

∗

Γ is controllable on M.

(2) Moreover, if the semigroup /

Γ

acts transitively on M, then θ

∗

Γ is con-

trollable on M.

Proof. Item (1) follows from (2), so we prove (2). Take any points x

0

, x

1

∈ M.

The transitivity of action of /

Γ

on M means that there exists X ∈ /

Γ

such

that θ(X, x

0

) = x

1

. Further, the inclusion X ∈ /

Γ

means that some trajectory

X(t) of Γ steers Id to X: X(0) = Id, X(T) = X, T ≥ 0. Then the curve

x(t) = θ(X(t), x

0

) is a trajectory of θ

∗

Γ that steers x

0

to x

1

:

x(0) = θ(Id, x

0

) = x

0

, x(T) = θ(X, x

0

) = x

1

.

Important applications of Theorem 5.1 are related to the linear action of

linear groups G ⊂ GL(n; R) on the vector space R

n

. In this case, the induced

systems are bilinear, or more generally, aﬃne systems.

Example 5.4 (G = GL

+

(R), M = R

n

¸ ¦0¦). We have

θ(X, x) = Xx,

Γ =

_

A+

m

i=1

u

i

B

i

_

⊂ gl(n),

θ

∗

Γ : ˙ x = Ax +

m

i=1

u

i

B

i

x, x ∈ R

n

¸ ¦0¦.

If / = GL

+

(n) or / = SL(n), then the bilinear system θ

∗

Γ is controllable

on R

n

¸ ¦0¦. The attainable set may be even less, for example, in the case

/ = SO(n) R

+

Id the bilinear system θ

∗

Γ remains controllable.

Remark. Linear groups acting transitively on R

n

¸ ¦0¦ or S

n

are described, see

[6, 7, 38, 8, 9, 27].

Example 5.5 (G = SL(n), M = R

n

¸ ¦0¦). Similarly,

θ(X, x) = Xx,

Γ =

_

A+

m

i=1

u

i

B

i

_

⊂ sl(n),

θ

∗

Γ : ˙ x = Ax +

m

i=1

u

i

B

i

x, x ∈ R

n

¸ ¦0¦.

If / is transitive on R

n

¸¦0¦, then the bilinear system θ

∗

Γ is controllable on R

n

¸

¦0¦.

34

Example 5.6 (G = SO(n), M = S

n−1

).

θ(X, x) = Xx,

Γ =

_

A+

m

i=1

u

i

B

i

_

⊂ so(n),

θ

∗

Γ : ˙ x = Ax +

m

i=1

u

i

B

i

x, x ∈ S

n−1

.

Example 5.7 (G = U(n) or SU(n), M = S

2n−1

).

θ(Z, z) = Zz,

Γ =

_

A+

m

i=1

u

i

B

i

_

⊂ u(n) or su(n),

θ

∗

Γ : ˙ z = Az +

m

i=1

u

i

B

i

z, z ∈ S

2n−1

.

Example 5.8 (G = Aﬀ

+

(n), M = R

n

). The connected component of identity

in the aﬃne group

Aﬀ

+

(n) =

__

X y

0 1

__

⊂ GL(n + 1)

acts transitively on the space

M = R

n

=

__

x

1

__

⊂ R

n+1

as follows:

θ

__

X y

0 1

_

,

_

x

1

__

=

_

X y

0 1

__

x

1

_

= Xx +y.

Consider a right-invariant system on G:

Γ =

_

C

0

+

m

i=1

u

i

C

i

_

, C

i

=

_

A

i

b

i

0 0

_

∈ aﬀ(n).

The induced vector ﬁelds are aﬃne:

θ

∗

_

A b

0 0

_ _

x

1

_

= Ax +b,

and the induced system reads

θ

∗

Γ : ˙ x = A

0

x +b

0

+

m

i=1

u

i

(A

i

x +b

i

), x ∈ R

n

.

35

In particular, for b

0

= 0, A

1

= = A

m

= 0, we obtain the linear system

˙ x = A

0

x +

m

i=1

u

i

b

i

, x ∈ R

n

, u

i

∈ R. (20)

Exercise 5.1. Show that if the Kalman condition holds:

span(b

1

, . . . , b

m

; A

0

b

1

, . . . , A

0

b

m

; . . . ; A

n−1

0

b

1

, . . . , A

n−1

0

b

m

) = R

n

,

then the linear system (20) is controllable on R

n

.

Example 5.9 (G = E(n), M = R

n

). This case is completely similar to the

case of Aﬀ

+

(n).

In this section we developed a theory of induced systems for right-invariant

systems because of the important class of bilinear systems ˙ x = Ax+uBx, where

x ∈ R

n

¸ ¦0¦ is a column vector. Obviously, the theory of induced systems for

left-invariant systems is quite the same; in this case the induced systems read

˙ y = yA+uyB, where y ∈ R

n

¸ ¦0¦ is a row vector.

6 Controllability conditions for special classes of

systems and Lie groups

6.1 Symmetric systems

We return to the exposition for left-invariant systems Γ ⊂ L on a Lie group G.

Deﬁnition 6.1. A system Γ ⊂ L is called symmetric if

Γ = −Γ,

i.e., together with any element A, this system contains also the sign-opposite

element −A.

Given a symmetric system, for any admissible direction of motion A, the

motion in the opposite direction −A is also admissible.

Lemma 6.1. Let Γ = −Γ. Then / = O.

Proof. We have

O = ¦exp(±t

1

A

1

) exp(±t

N

A

N

) [ t

i

> 0, A

i

∈ Γ¦ .

But all −A

i

∈ Γ, thus / = O.

Thus the study of controllability for symmetric Γ is reduced to the veriﬁca-

tion of the rank condition.

Theorem 6.1. A symmetric left-invariant system Γ ⊂ L is controllable on a

connected Lie group G if and only if Lie(Γ) = L.

36

Proof. The necessity is a general fact. Suﬃciency follows since for a full-rank

system on a connected Lie group the orbit coincides with the whole Lie group.

Example 6.1 (Control-linear systems). A control-linear system

Γ =

_

m

i=1

u

i

A

i

[ u = (u

1

, . . . , u

m

) ∈ U ⊂ R

m

_

is symmetric if the set of control parameters U is symmetric with respect to the

origin: U = −U; in particular, if U = R

m

:

Γ = span(A

1

, . . . , A

m

) ⊂ L.

Such a system is controllable on a connected Lie group G iﬀ Lie(A

1

, . . . , A

m

) =

L.

Example 6.2 (Symmetric bilinear system). Let A

1

, . . . , A

m

∈ gl(n). Con-

sider the corresponding symmetric bilinear system:

˙ x =

m

i=1

u

i

A

i

x, x ∈ R

n

¸ ¦0¦, u

i

∈ R. (21)

Denote Lie(A

1

, . . . , A

m

) = L, and let G ⊂ GL(n) be the connected Lie subgroup

corresponding to the Lie algebra L. If G acts transitively on R

n

¸¦0¦ (or S

n−1

),

then the bilinear system (21) is controllable on R

n

¸ ¦0¦ (respectively on S

n−1

).

6.2 Compact Lie groups

In this section, we consider the case of a Lie group that is compact as a topolog-

ical space. For example, the Lie groups SO(n), U(n), SU(n) are compact and

connected.

The following simple fact is crucial for the controllability problemon compact

Lie groups.

Lemma 6.2. Let a Lie group G be compact, and let A belong to the Lie alge-

bra L. Then the one-parameter subgroup exp(RA) is quasi-periodic:

exp(R

−

A) ⊂ cl exp(R

+

A).

Proof. Denote X = exp(tA) for an arbitrary ﬁxed t > 0. We have to prove that

exp(−tA) = X

−1

∈ cl exp(R

+

A).

The sequence ¦X

n

¦, n ∈ N, has a converging subsequence in the compact

Lie group G:

X

n

k

→ Y ∈ G as k → ∞, n

k+1

> n

k

.

Then

X

n

k+1

−n

k

−1

= X

n

k+1

X

−n

k

X

−1

→ Y Y

−1

X

−1

= X

−1

as k → ∞.

But n

k+1

−n

k

−1 ≥ 0, thus X

−1

∈ cl exp(R

+

A).

37

Corollary 6.1. Let G be compact, and let Γ ⊂ L. Then LS(Γ) = Lie(Γ).

Proof. We show that LS(Γ) is a Lie algebra. If A, B ∈ LS(Γ), then ±A, ±B ∈

LS(Γ) by Lemma 6.2. Thus αA+βB ∈ LS(Γ), α, β ∈ R, since LS(Γ) is a cone.

Moreover ±[A, B] ∈ LS(Γ). It follows that LS(Γ) is a Lie subalgebra of L.

Taking into account the chain Γ ⊂ LS(Γ) ⊂ Lie(Γ), we conclude that

LS(Γ) = Lie(Γ).

Theorem 6.2. A right-invariant system Γ ⊂ L is controllable on a compact

connected Lie group G if and only if Lie(Γ) = L.

Proof. Apply Corollary 6.1.

Example 6.3 (SO(3)). Let G = SO(3), the set of all 3 3 real orthogonal

matrices with positive determinant. The Lie group G is compact and connected.

Its Lie algebra L = so(3) is the set of all 3 3 real skew-symmetric matrices.

Take any linearly independent matrices A

1

, A

2

∈ so(3) and consider the

right-invariant system Γ = ¦A

1

, A

2

¦. Notice that the matrices A

1

, A

2

, and

[A

1

, A

2

] span the whole Lie algebra so(3). By Theorem 6.2, the system Γ is

controllable. That is, any rotation in SO(3) can be written as the product of

exponentials

exp(t

1

A

i1

) exp(t

N

A

iN

), t

j

≥ 0, i

j

∈ ¦1, 2¦, N ∈ N. (22)

The single-input right-invariant aﬃne in control system

˙

X = (A

1

+uA

2

)X, u ∈ U ⊂ R, X ∈ SO(3) (23)

is also controllable (for any control set U containing more than one element).

Consequently, the induced bilinear system

˙ x = A

1

x +uA

2

x, x ∈ S

2

, u ∈ U

is controllable on the sphere S

2

.

Example 6.4 (SO(n)). The previous considerations are generalized to the

group G = SO(n) of rotations of R

n

. In this case, the Lie algebra L of G is the

set of all n n skew-symmetric matrices so(n).

Take the matrices A

1

=

n−2

i=1

(E

i,i+1

−E

i+1,i

) and A

2

= E

n−1,n

− E

n,n−1

.

We denote by E

ij

the nn matrix with the only identity in row i and column j,

and all other zero entries.

It is easy to show that Lie(A

1

, A

2

) = so(n). Thus, even though the group

SO(n) is

1

2

n(n −1)-dimensional, the system

˙

X = (A

1

+uA

2

)X, X ∈ SO(n), u ∈ U ⊂ R,

in which only one control is involved, is controllable (if the set of control pa-

rameters U contains at least two distinct points).

Notice that the set of pairs (A

1

, A

2

) such that Lie(A

1

, A

2

) = L is open and

dense in L L (this is valid for any semisimple Lie algebra L; see [42]). Thus,

we can replace the matrices A

1

and A

2

by an “almost arbitrary” pair in LL.

38

Example 6.5 (SU(2)). For the Lie group G = SU(2), its Lie algebra can be

represented as follows:

L = su(2) = span

__

i 0

0 −i

_

,

_

0 1

−1 0

_

,

_

0 i

i 0

__

.

For any linearly independent A

1

, A

2

∈ L, we have [A

1

, A

2

] / ∈ span(A

1

, A

2

),

thus Lie(A

1

, A

2

) = L. So the system Γ = ¦A

1

+ uA

2

[ u ∈ U¦ (where U

contains more than one element) is controllable on G = SU(2). Consequently,

the induced bilinear system

˙ z = A

1

z +uA

2

z, z ∈ S

3

, u ∈ R

is controllable on the sphere S

3

.

6.3 Semisimple Lie groups

Deﬁnition 6.2. A subspace I ⊂ L is called an ideal of a Lie algebra L if

[I, L] ⊂ I.

Deﬁnition 6.3. A Lie algebra L is called simple if it is not Abelian and contains

no proper (i.e., distinct from ¦0¦ and L) ideals.

Deﬁnition 6.4. A Lie algebra L is called semisimple if it contains no nonzero

Abelian ideals.

A semisimple Lie algebra is a direct sum of its simple ideals.

Deﬁnition 6.5. A Lie group G is called simple (resp., semisimple) if its Lie

algebra L is simple (resp., semisimple).

The Lie groups SL(n) and SU(n) are simple; the Lie groups SO(n), n ,= 4,

are simple, while SO(4) is semisimple.

For the controllability problem, we are interested in the case of SL(n) since

the other two groups are compact and for them controllability is equivalent to

the rank condition.

We start from an example of a control system that has the full rank and is

not controllable.

Example 6.6. Let G = SL(2) and Γ = A + RB ⊂ sl(2). Here A and B are

traceless matrices of the form

A = (a

ij

), a

12

> 0, a

21

> 0,

B =

_

−b 0

0 b

_

, b ,= 0.

Let us show ﬁrst that

Lie(A, B) = L = sl(2). (24)

39

Since dimsl(2) = 3, we have to obtain just one element in Lie(A, B) linearly

independent of A and B. Compute the commutator:

[A, B] = 2b

_

0 a

12

−a

21

0

_

,

now it is obvious that

span(A, B, [A, B]) = sl(2).

Equality (24) follows, i.e., the system Γ is full-rank.

In order to show that Γ is noncontrollable on SL(2), we prove that the

bilinear system θ

∗

Γ is noncontrollable on R

2

¸ ¦0¦. The induced system reads

˙ x = Ax +uBx, x ∈ R

2

¸ ¦0¦, u ∈ R. (25)

It is easy to see that the ﬁeld Bx is tangent to the axes of coordinates ¦x

1

= 0¦

and ¦x

2

= 0¦. On the other hand, the ﬁeld Ax is directed inside the ﬁrst

quadrant R

2

+

= ¦x

1

≥ 0, x

2

≥ 0¦ on its boundary. Consequently, R

2

+

is

an invariant set of the bilinear system (25). Thus the induced system θ

∗

Γ is

not controllable on the homogeneous space R

2

¸ ¦0¦, hence the right-invariant

system Γ is not controllable on the Lie group SL(2).

The controllability problem on SL(n) is much harder than the one on com-

pact Lie groups. In fact, the whole machinery of the Lie saturation on Lie

groups was developed primarily for the study of controllability on SL(n). There

are no controllability tests in this case, but there are good suﬃcient conditions

for controllability on SL(n).

Theorem 6.3. Let G = SL(n) and Γ = A + RB ⊂ sl(n). Suppose that the

matrices A = (a

ij

) and B satisfy the conditions:

(1) a

1n

a

n1

< 0;

(2) the matrix A is permutation-irreducible;

(3) B = diag(b

1

, . . . , b

n

);

(4) b

1

< b

2

< < b

n

;

(5) b

i

−b

j

,= b

k

−b

m

for (i, j) ,= (k, m).

Then the system Γ is controllable on the group SL(n).

An n n matrix A is called permutation-reducible if there exists a permuta-

tion matrix P such that

P

−1

AP =

_

A

1

A

2

0 A

3

_

,

where A

3

is a kk matrix with 0 < k < n. An nn matrix is called permutation-

irreducible if it is not permutation-reducible. Permutation-irreducible matrices

are matrices having no nontrivial invariant coordinate subspaces.

Now we prove Theorem 6.3 (in the case n = 2 only: in the general case the

proof is longer but uses essentially the same ideas [12]).

40

Proof. In the case n = 2 we have:

A =

_

a

11

a

12

a

21

a

22

_

, a

12

a

21

< 0,

B =

_

−b 0

0 b

_

, b > 0.

Without loss of generality, we can assume that

a

12

> 0, a

21

< 0,

in the case of opposite signs the proof is the same.

We show that

LS(Γ) = sl(2) = span(E

22

−E

11

, E

12

, E

21

). (26)

First of all,

LS(Γ) ÷

A+uB

[u[

→

u→±∞

±B ∈ LS(Γ),

thus

A, ±B ∈ LS(Γ).

That is why

A

t

= exp(t ad B)A ∈ LS(Γ), t ∈ R.

Compute the matrix of the adjoint operator

adB : sl(2) → sl(2), B = −b(E

11

−E

22

),

in the basis (26). We have

(ad B)(E

11

−E

22

) = 0,

(ad B)E

12

= −2bE

12

,

(ad B)E

21

= 2bE

21

.

Thus the adjoint operator has the diagonal matrix

ad B =

_

_

0 0 0

0 −2b 0

0 0 2b

_

_

,

and its exponential is easily computed:

exp(t ad B) =

_

_

1 0 0

0 exp(−2bt) 0

0 0 exp(2bt)

_

_

.

Further, in the basis (26)

A =

_

_

a

11

a

12

a

21

_

_

, A

t

= exp(t ad B)A =

_

_

a

11

exp(−2bt)a

12

exp(2bt)a

21

_

_

∈ LS(Γ).

41

Since ±B = ∓b(E

11

− E

22

) ∈ LS(Γ), it follows that ±(E

11

− E

22

) ∈ LS(Γ),

and we can kill the ﬁrst coordinate of A

t

:

A

1

t

= A

t

−a

11

(E

11

−E

22

) =

_

_

0

exp(−2bt)a

12

exp(2bt)a

21

_

_

∈ LS(Γ).

We go on:

LS(Γ) ÷ exp(2bt)A

1

t

=

_

_

0

a

12

exp(4bt)a

21

_

_

→

t→−∞

_

_

0

a

12

0

_

_

∈ LS(Γ).

Consequently,

1

a

12

_

_

0

a

12

0

_

_

= E

12

∈ LS(Γ).

Similarly,

LS(Γ) ÷ exp(−2bt)A

1

t

=

_

_

0

exp(−4bt)a

12

a

21

_

_

→

t→+∞

_

_

0

0

a

21

_

_

∈ LS(Γ).

Then

1

[a

21

[

_

_

0

0

a

21

_

_

= −E

21

∈ LS(Γ).

Summing up,

E

12

−E

21

∈ LS(Γ).

But this element generates a periodic one-parameter group:

exp(t(E

12

− E

21

)) =

_

cos t sin t

−sint cos t

_

.

That is why

±(E

12

−E

21

) ∈ LS(Γ).

Recall that ±(E

11

−E

22

) ∈ LS(Γ) as well. Thus

±[E

12

−E

21

, E

11

−E

22

] = ∓2(E

12

+E

21

) ∈ LS(Γ).

It follows that

±E

12

, ±E

21

, ±(E

11

−E

22

) ∈ LS(Γ),

thus LS(Γ) = sl(2), and the system Γ is controllable on SL(2).

There exist generalizations of the previous theorem for the case of complex

spectrum of the matrix B and for general semisimple Lie groups G [15, 16, 2].

42

6.4 Solvable Lie groups

For a Lie algebra L, its derived series is the following descending chain of sub-

algebras:

L ⊃ L

(1)

= [L, L] ⊃ L

(2)

= [L

(1)

, L

(1)

] ⊃ .

Deﬁnition 6.6. A Lie algebra L is called solvable if its derived series stabilizes

at zero:

L ⊃ L

(1)

⊃ L

(2)

⊃ ⊃ L

(N)

= ¦0¦

for some N ∈ N. A Lie group with a solvable Lie algebra is called solvable.

Example 6.7. The groups T(n) and E(2) are solvable.

There is a general controllability test for right-invariant systems on con-

nected, simply connected solvable Lie groups (recall that a topological space M

is called simply connected if any closed loop in M can be contracted to a point).

Theorem 6.4. Let a Lie group G be connected, simply connected, and solvable.

A right-invariant system Γ ⊂ L is controllable iﬀ the following two properties

hold:

(1) Lie(Γ) = L and

(2) Γ is not contained in a half-space in L bounded by a subalgebra.

If G is not simply connected, conditions (1), (2) remain suﬃcient for controlla-

bility of Γ.

We will prove only the easy part of this test — necessity. Suﬃciency is highly

nontrivial, its proof may be found in [19].

And necessity in Theorem 6.4 is a consequence of the following necessary

controllability condition for general (not necessarily solvable) simply connected

Lie groups.

Theorem 6.5. Let G be a connected, simply connected Lie group, and let Γ ⊂ L.

If Γ is contained in a half-space in L bounded by a subalgebra, then Γ is not

controllable on G.

Proof. Suppose that Γ is contained in a half-space Π ⊂ L bounded by a sub-

algebra l ⊂ L, diml = dimL − 1. We have Π = R

+

A + l for some A ∈ L.

There exists a Lie subgroup H ⊂ G with the Lie algebra l. Since G is simply

connected and dimH = dimG − 1, the subgroup H is closed in G. Then the

coset space

G/H = ¦XH [ X ∈ G¦

is a smooth manifold. Moreover,

dimG/H = dimG−dimH = 1.

Further, since G is simply connected, its quotient G/H is simply connected as

well. Summing up,

G/H = R.

43

The quotient G/H is a homogeneous space of G: the transitive action is

θ : GG/H → G/H, θ(Y, XH) = Y XH.

In order to show that Γ is noncontrollable on G, we prove that the induced

system θ

∗

Γ is noncontrollable on the homogeneous space G/H.

Denote the projection

π : G → G/H, π(X) = XH.

For any C ∈ l we have

θ

∗

C[

π(Id)

=

d

d t

¸

¸

¸

¸

t=0

θ(exp(tC), H) =

d

d t

¸

¸

¸

¸

t=0

exp(tC)

. ¸¸ .

∈H

H

=

d

d t

¸

¸

¸

¸

t=0

π(H) =

d

d t

¸

¸

¸

¸

t=0

π(Id)

= 0.

That is,

θ

∗

l[

π(Id)

= 0.

Since Γ ⊂ Π = R

+

A+l, then

θ

∗

Γ[

π(Id)

⊂ θ

∗

(R

+

A+l)[

π(Id)

= θ

∗

(R

+

A)[

π(Id)

= R

+

θ

∗

A[

π(Id)

.

So admissible velocities of the induced system θ

∗

Γ at π(Id) ∈ R belong to a

half-line. Thus θ

∗

Γ is not controllable on R = G/H and Γ is not controllable

on G.

In addition to Theorem 6.4, it would be desirable to have a controllability

condition with easy to verify hypotheses. We give such a condition for a subclass

of solvable Lie groups.

Deﬁnition 6.7. A solvable Lie algebra is called completely solvable if all adjoint

operators ad A, A ∈ L, have only real eigenvalues.

Example 6.8. The triangular algebra t(n) is completely solvable.

Deﬁnition 6.8. A Lie algebra is called nilpotent if all adjoint operators ad A,

A ∈ L, have only zero eigenvalues.

Example 6.9. The Lie group

T

0

(n) = ¦X = (x

ij

) [ x

ij

= 0 ∀i > j, x

ii

= 1 ∀i¦

is nilpotent.

Any nilpotent Lie algebra is completely solvable. An example of a solvable

but not completely solvable Lie algebra is provided by the Lie algebra e(2) of

the Euclidean group of the plane.

44

Theorem 6.6. Let G be a completely solvable, connected, simply connected Lie

group, and let

Γ =

_

A+

m

i=1

u

i

B

i

[ u

i

∈ R

_

⊂ L.

The system Γ is controllable iﬀ Lie(B

1

, . . . , B

m

) = L.

Proof. Suﬃciency. We have

LS(Γ) ÷

A+u

i

B

[u

i

[

→

u→±∞

±B

i

∈ LS(Γ),

thus

Lie(B

1

, . . . , B

m

) ⊂ LS(Γ).

If Lie(B

1

, . . . , B

m

) = L, then LS(Γ) = L, and Γ is controllable.

Necessity is based upon the following general fact: in a completely solvable

Lie algebra L, any subalgebra l

1

⊂ L, l

1

,= L, is contained in a subalgebra

l

2

⊃ l

1

such that diml

2

= diml

1

+ 1, see [31].

Let Lie(B

1

, . . . , B

m

) = l

1

,= L. Then there exists a codimension one subal-

gebra l

2

in L containing l

1

:

l

1

⊂ l

2

⊂ L, diml

2

= dimL −1.

The system Γ is contained in a larger system:

Γ =

_

A+

m

i=1

u

i

B

i

_

⊂ A+ Lie(B

1

, . . . , B

m

) = A+l

1

⊂ R

+

A+l

2

.

(1) If A / ∈ l

2

, then Π = R

+

A+l

2

is a half-space bounded by the subalgebra l

2

and containing Γ. Thus Γ is not controllable.

(2) And if A ∈ l

2

, then R

+

A+l

2

= l

2

is a subalgebra containing Γ. Thus Γ

is not full-rank, thus it is not controllable.

6.5 Semi-direct products of Lie groups

Deﬁnition 6.9. Let a Lie group K act linearly on a vector space V . The

semi-direct product of V and K is the Lie group deﬁned as the set

G = V K = ¦(v, k) [ v ∈ V, k ∈ K¦

endowed with the product smooth structure, and the group operation

(v

1

, k

1

) (v

2

, k

2

) = (v

1

+k

1

v

2

, k

1

k

2

).

Example 6.10. The Euclidean group E(n) is the semi-direct product R

n

**SO(n), this is obvious since
**

E(n) =

__

X y

0 1

_

∈ M(n + 1) [ X ∈ SO(n), y ∈ R

n

_

45

and

_

X

1

y

1

0 1

__

X

2

y

2

0 1

_

=

_

X

1

X

2

X

1

y

2

+y

1

0 1

_

.

The following controllability test for semi-direct products can be seen as a

generalization of the controllability test for compact Lie groups given in Th. 6.2.

Theorem 6.7. Let K be a compact connected Lie group acting linearly on a

vector space V , and let G = V K. Assume that the action of K has no

nonzero ﬁxed points in V . An invariant system Γ ⊂ L is controllable on G iﬀ

Lie(Γ) = L.

Example 6.11. The group SO(n) has no nonzero ﬁxed points in R

n

, thus an

invariant system Γ ⊂ e(n) is controllable on E(n) = R

n

SO(n) iﬀ Γ is full-rank.

We prove Theorem 6.7 in the simplest case G = E(2), Γ = A + RB. The

proof in the general case, as well as a generalization for the case where K has

ﬁxed points in V , may be found in [5].

Let G = E(2), Γ = A+RB ⊂ e(2). We have

e(2) = span(e

1

, e

2

, e

3

), e

1

= E

12

−E

21

, e

2

= E

13

, e

3

= E

23

.

The multiplication table in L = e(2) is as follows:

[e

1

, e

2

] = −e

3

, [e

1

, e

3

] = e

2

, [e

2

, e

3

] = 0, (27)

thus the derived series is

L = span(e

1

, e

2

, e

3

) ⊃ L

(1)

= span(e

2

, e

3

) ⊃ L

(2)

= ¦0¦,

so e(2) is solvable. Further, Sp(ad e

1

) = ¦0, ±i¦ ,⊂ R, so e(2) is not completely

solvable. It easily follows from multiplication table (27) that span(e

2

, e

3

) is the

only 2-dimensional subalgebra in e(2).

Now we give a controllability test on E(2).

Theorem 6.8. A system Γ = A + RB ⊂ e(2) is controllable on G = E(2) iﬀ

the following conditions hold:

(1) A, B are linearly independent and

(2) ¦A, B¦ ,⊂ span(e

2

, e

3

).

Proof. Necessity. If A, B are linearly dependent or ¦A, B¦ ⊂ span(e

2

, e

3

), then

Lie(Γ) = Lie(A, B) ,= L, thus Γ is not controllable.

Suﬃciency. Let A, B are linearly independent and ¦A, B¦ ,⊂ span(e

2

, e

3

).

Then there exist linearly independent A

u

= A+uB and A

v

= A+vB such that

A

u

, A

v

/ ∈ span(e

2

, e

3

). For the element

A

u

= α

1

e

1

+α

2

e

2

+α

3

e

3

, α

1

,= 0,

46

the one-parameter subgroup

exp(sA

u

) =

_

_

cos(α

1

s) sin(α

1

s)

α2

α1

sin(α

1

s) +

α3

α1

(1 −cos(α

1

s))

−sin(α

1

s) cos(α

1

s)

α2

α1

(cos(α

1

s) −1) +

α3

α1

sin(α

1

s)

0 0 1

_

_

is periodic. Since A

u

∈ Γ, then ±A

u

∈ LS(Γ). Similarly, ±A

v

∈ LS(Γ). Thus

the subspace l = Lie(A

u

, A

v

) ⊂ LS(Γ). But l is not contained in span(e

2

, e

3

) —

the only 2-dimensional subalgebra in e(2). Thus l = e(2), LS(Γ) = e(2) = L,

and Γ is controllable on E(2).

Now we are able to prove Theorem 6.7 in the simplest case.

Corollary 6.2. A system Γ = A+RB ⊂ e(2) is controllable on E(2) iﬀ Lie(Γ) =

e(2).

Proof. Necessity of the rank condition is a general fact. On the other hand, if

Lie(Γ) = e(2), then conditions (1), (2) of Theorem 6.8 are satisﬁed, thus Γ is

controllable on E(2).

7 Pontryagin Maximum Principle for invariant

optimal control problems on Lie groups

Now we turn to optimal control problems of the form

˙ q = f(q, u), q ∈ M, u ∈ U ⊂ R

m

,

q(0) = q

0

, q(t

1

) = q

1

,

J(u) =

_

t1

0

ϕ(q(t), u(t)) dt → min .

Here M is a smooth manifold, f(q, u) and ϕ(q, u) are smooth, and admissible

controls u(t) are measurable locally bounded.

In order to state the fundamental necessary optimality condition — Pontrya-

gin Maximum Principle [30] — we recall some basic notions of the Hamiltonian

formalism on the cotangent bundle.

7.1 Hamiltonian systems on T

∗

M

Let M be a smooth n-dimensional manifold. At any point q ∈ M, the tangent

space T

q

M has the dual space — the cotangent space T

∗

q

M = (T

q

M)

∗

. The dis-

joint union of all cotangent spaces is the cotangent bundle T

∗

M =

q∈M

T

∗

q

M,

it is a smooth manifold of dimension 2n. In order to construct local coordinates

on T

∗

M, take any local coordinates (x

1

, . . . , x

n

) on M. Then dx

1q

, . . . , dx

nq

are basis linear forms in T

∗

q

M, and any covector λ ∈ T

∗

q

M is decomposed as

λ =

n

i=1

p

i

dx

iq

. The 2n-tuple (p

1

, . . . , p

n

; x

1

, . . . , x

n

) provides local coordi-

nates called canonical coordinates on the cotangent bundle T

∗

M.

47

The canonical projection π : T

∗

M → M maps a covector λ ∈ T

∗

q

M to the

base point q ∈ M.

The tautological 1-form s ∈ Λ

1

(T

∗

M) is deﬁned as follows. Take any point

λ ∈ T

∗

M, π(λ) = q, and any tangent vector ξ ∈ T

λ

(T

∗

M). Then

¸s

λ

, ξ) = ¸λ, π

∗

ξ).

The symplectic form σ ∈ Λ

2

(T

∗

M) is deﬁned as the diﬀerential

σ = ds.

Any smooth function h ∈ C

∞

(T

∗

M) is called a Hamiltonian. The correspond-

ing Hamiltonian vector ﬁeld

h ∈ Vec(T

∗

M) is introduced in the following way.

The diﬀerential dh is a 1-form on T

∗

M. On the other hand, for any vector ﬁeld

V ∈ Vec(T

∗

M), one can deﬁne the 1-form σ(V, ) = i

V

σ ∈ Λ

1

(T

∗

M). The

Hamiltonian vector ﬁeld corresponding to a Hamiltonian function h is deﬁned

as such vector ﬁeld

h ∈ Vec(T

∗

M) that

dh = −i

h

σ.

Example 7.1. In canonical coordinates (p

1

, . . . , p

n

; x

1

, . . . , x

n

) on T

∗

M, we

have:

s = p dx =

n

i=1

p

i

dx

i

,

σ = dp ∧ dx =

n

i=1

dp

i

∧ dx

i

.

For a Hamiltonian h = h(p, x) ∈ C

∞

(T

∗

M), the Hamiltonian system of ODEs

˙

λ =

**h(λ) reads in canonical coordinates as
**

˙ p = −

∂ h

∂ x

,

˙ x =

∂ h

∂ p

.

7.2 Pontryagin Maximum Principle on smooth manifolds

Consider optimal control problem of the form

˙ q = f(q, u), q ∈ M, u ∈ U ⊂ R

m

, (28)

q(0) = q

0

, q(t

1

) = q

1

, t

1

ﬁxed or free, (29)

J(u) =

_

t1

0

ϕ(q(t), u(t)) dt → min . (30)

Let λ ∈ T

∗

M be a covector, ν ∈ R a parameter, and u ∈ U a control

parameter. Introduce the family of Hamiltonians

h

ν

u

(λ) = ¸λ, f(q, u)) +νϕ(q, u).

48

Theorem 7.1 (PMP on smooth manifolds). Let ˜ u(t), t ∈ [0, t

1

], be an

optimal control in the problem (28)–(30) with ﬁxed time t

1

. Then there exists a

Lipschitzian curve λ

t

∈ T

∗

˜ q(t)

M, t ∈ [0, t

1

], and a number ν ∈ R such that:

˙

λ

t

=

−→

h

ν

˜ u(t)

(λ

t

), (31)

h

ν

˜ u(t)

(λ

t

) = max

u∈U

h

ν

u

(λ

t

), (32)

(λ

t

, ν) ,≡ (0, 0), t ∈ [0, t

1

], (33)

ν ≤ 0. (34)

Remark. For the problem (28)–(30) with free terminal time t

1

, necessary opti-

mality conditions read as (31)–(34) plus the additional equality h

ν

˜ u(t)

(λ(t)) ≡ 0.

The proof of Pontryagin Maximum Principle on smooth manifolds may be

found in [1].

7.3 Hamiltonian systems on T

∗

G

Notice that in general the cotangent bundle T

∗

M of a smooth manifold M is not

trivial, i.e., cannot be represented as the direct product EM of a vector space

E with M. Although, the cotangent bundle T

∗

G of a Lie group G has a natural

trivialization. We will apply this trivialization in order to write Hamiltonian

system of PMP for optimal control problems on Lie groups.

Let E be a vector space of dimension dimE = dimM = n.

Deﬁnition 7.1. A trivialization of the cotangent bundle T

∗

M is a diﬀeomor-

phism Φ : E M → T

∗

M such that:

(1) Φ(e, q) ∈ T

∗

q

M, e ∈ E, q ∈ M,

(2) Φ( , q) : E → T

∗

q

M is a linear isomorphism for any q ∈ M.

At any point (e, q) of the trivialized cotangent bundle E M

∼

= T

∗

M, we

have the following identiﬁcations of the tangent and cotangent bundles:

T

(e,q)

(E M)

∼

= T

e

E ⊕T

q

M

∼

= E T

q

M,

T

∗

(e,q)

(E M)

∼

= T

∗

e

E ⊕T

∗

q

M

∼

= E

∗

T

∗

q

M.

Respectively, any tangent and cotangent vector are decomposed into the vertical

and horizontal parts:

V = V

v

+V

h

, V ∈ T

(e,q)

(E M), V

v

∈ E, V

h

∈ T

q

M,

ω = ω

v

+ω

h

, ω ∈ T

∗

(e,q)

(E M), ω

v

∈ E

∗

, ω

h

∈ T

∗

q

M.

For a Lie group G, the cotangent bundle T

∗

G has a natural trivialization as

follows:

Φ : L

∗

G → T

∗

G, (a, X) → ¯ a

X

, a ∈ L

∗

, X ∈ G.

49

Here L

∗

is the dual space of the Lie algebra L = T

Id

G, and ¯ a ∈ Λ

1

(G) is

the left-invariant 1-form on G obtained by left translations from the covector

a = ¯ a

Id

∈ L

∗

:

¸¯ a

X

, XA) = ¸a, A), a ∈ L

∗

, A ∈ L, X ∈ G.

Now we compute the pull-back of the tautological 1-form s, the symplectic

2-form σ, and a Hamiltonian vector ﬁeld

h to the trivialized cotangent bundle

L

∗

G

∼

= T

∗

G.

We start from the tautological 1-form

´

Φs ∈ Λ

1

(L

∗

G). Take any point

(a, X) ∈ L

∗

G and a tangent vector (ξ, XA) ∈ L

∗

⊕T

X

G. Then

_

(

´

Φs)

(a,X)

, (ξ, XA)

_

=

¸

s

¯ aX

, Φ

∗(a,X)

(ξ, XA)

_

=

¸

¯ a

X

, π

∗

Φ

∗(a,X)

(ξ, XA)

_

= ¸¯ a

X

, XA) = ¸a, A) . (35)

Further, compute the symplectic 2-form

´

Φσ ∈ Λ

2

(L

∗

G). For any tangent

vectors (ξ, XA), (η, XB) ∈ L

∗

⊕T

X

G we have

(

´

Φσ)

(a,X)

((ξ, XA), (η, XB))

since

´

Φσ =

´

Φds = d

´

Φs

= (d

´

Φs)

(a,X)

((ξ, XA), (η, XB))

since dω(V, W) = V ¸ω, W) −W¸ω, V ) −¸ω, [V, W])

= (ξ, XA)¸

´

Φs

(a,X)

, (η, XB)) −(η, XB)¸

´

Φs

(a,X)

, (ξ, XA))

−¸

´

Φs

(a,X)

, [(ξ, XA), (η, XB)])

taking into account formula (35) for

´

Φs

= (ξ, A)¸a, B) −(η, B)¸a, A) −¸a, [A, B])

= ¸ξ, B) −¸η, A) −¸a, [A, B]).

Finally, take a Hamiltonian h = h(a) not depending on X ∈ G, this is the

form of the Hamiltonian of PMP for a left-invariant optimal control problem

on the Lie group G. Decompose the required Hamiltonian vector ﬁeld

h ∈

Vec(L

∗

G) into the vertical and horizontal parts:

h(a, X) = (ξ, XA) ∈ L

∗

⊕T

X

G, a ∈ L

∗

, X ∈ G.

Apply the identity dh = −

´

Φσ(

**h, ) to an arbitrary tangent vector (η, XB) ∈
**

L

∗

⊕T

X

G. Since the Hamiltonian h does not depend on X, we denote

dh =

∂ h

∂ a

∈ (L

∗

)

∗

= L.

50

Taking into account formula (35) for

´

Φσ, we obtain:

_

∂ h

∂ a

, (η, XB)

_

= ¸dh, (η, XB)) = −

´

Φσ

(a,X)

((ξ, XA), (η, XB))

= −¸ξ, B) +¸η, A) +¸a, [A, B]). (36)

Setting B = 0 in (36), we compute the vertical part of

h:

_

∂ h

∂ a

, (η, 0)

_

=

_

η,

∂ h

∂ a

_

= ¸η, A) ∀η ∈ L

∗

,

thus A =

∂ h

∂ a

.

Now we set η = 0 in (36) and ﬁnd the horizontal part of

h:

0 = ¸dh, (0, XB)) = −¸ξ, B) +¸a, [A, B]),

thus

¸ξ, B) = ¸a, [A, B]) = ¸(adA)

∗

a, B) ∀B ∈ L.

So ξ = (ad A)

∗

a =

_

ad

∂ h

∂ a

_

∗

a.

Summing up, the Hamiltonian system on T

∗

G

∼

= L

∗

G for a left-invariant

Hamiltonian h = h(a), a ∈ L

∗

, reads as follows:

_

¸

¸

¸

¸

_

¸

¸

¸

¸

_

˙ a =

_

ad

∂ h

∂ a

_

∗

a, a ∈ L

∗

,

˙

X = X

∂ h

∂ a

, X ∈ G.

(37)

7.4 Hamiltonian systems in the case of compact Lie group

The Hamiltonian system (37) simpliﬁes in the case of a compact Lie group G.

Let G ⊂ GL(N) be a compact linear Lie group. Then it is easy to show that

in fact G ⊂ O(N). That is, there exists an inner product g( , ) on R

N

such

that

g(Xu, Xv) = g(u, v) ∀X ∈ G, ∀u, v ∈ R

N

.

Indeed, start from an arbitrary inner product ˜ g( , ) on R

N

, and choose any

left-invariant 1-forms ω

1

, . . . , ω

n

∈ Λ

1

(G) linearly independent at each point of

G. Then the required inner product g can be constructed as follows:

g(u, v) =

_

G

˜ g(Xu, Xv) ω

1

∧ ∧ ω

n

.

So in the sequel we assume that G ⊂ O(N), thus L ⊂ so(N). But the Lie

algebra so(N) has an invariant inner product ¸, ):

¸A, B) = −tr(AB).

51

Writing skew-symmetric matrices as

A = (A

ij

), B = (B

ij

), A

ij

= −A

ji

, B

ij

= −B

ji

,

we have

¸A, B) =

N

i,j=1

A

ij

B

ij

.

The product ¸, ) is invariant in the sense of the following identity:

¸

e

t ad C

A, e

t ad C

B

_

= ¸A, B) ∀A, B, C ∈ so(N), ∀t ∈ R. (38)

In other words, the operator e

t ad C

: so(N) → so(N) is orthogonal. This

identity easily follows since e

t ad C

A = e

tC

Ae

−tC

and

¸

e

t ad C

A, e

t ad C

B

_

= −tr(e

tC

Ae

−tC

e

tC

Be

−tC

) = −tr(e

tC

ABe

−tC

)

= −tr(AB) = ¸A, B)

by invariance of trace.

Diﬀerentiating identity (38) w.r.t. t at t = 0, we obtain the inﬁnitesimal

version of the invariance identity:

¸ad C(A), B) +¸A, ad C(B)) = 0 ∀A, B, C ∈ so(N),

i.e., the operator ad C : so(N) → so(N) is skew-symmetric.

Consequently, the Lie algebra L ⊂ so(N) is endowed with an invariant scalar

product. This allows us to identify the Lie algebra L with its dual space L

∗

:

A ↔

˜

A = ¸A, ), A ∈ L,

˜

A ∈ L

∗

.

Via this identiﬁcation, the operator

_

ad

∂ h

∂ a

_

∗

: L

∗

→ L

∗

becomes deﬁned in

L. Let A ∈ L, we compute the action of the operator (ad A)

∗

: L → L. For

any B, C ∈ L, we have

¸(ad A)

∗

˜

B, C) = ¸

˜

B, (adA)C) = ¸B, (ad A)C) = −¸(ad A)B, C)

= −¸

(ad A)B, C).

Thus (ad A)

∗

˜

B = −

adA(B), so the operator (adA)

∗

: L → L coincides with

−adA.

In particular, the operator

_

ad

∂ h

∂ a

_

∗

: L

∗

→ L

∗

is identiﬁed with the

operator −ad

∂ h

∂ a

: L → L. So for a compact Lie group G, the vertical part of

the Hamiltonian system is deﬁned on the Lie algebra L:

_

¸

¸

¸

_

¸

¸

¸

_

˙ a = −

_

ad

∂ h

∂ a

_

a =

_

a,

∂ h

∂ a

_

, a ∈ L,

˙

X = X

∂ h

∂ a

, X ∈ G.

(39)

52

Now we apply expressions (37), (39) for Hamiltonian systems in order to

study invariant optimal control problems on Lie groups.

8 Examples of invariant optimal control

problems on Lie groups

8.1 Riemannian problem on compact Lie group

Let G be a compact connected Lie group. The invariant scalar product ¸ , )

in the Lie algebra L deﬁnes a left-invariant Riemannian structure on G:

¸XA, XB)

X

= ¸A, B), A, B ∈ L, X ∈ G, XA, XB ∈ T

X

G.

So in every tangent space T

X

G there is a scalar product ¸ , )

X

. For any

Lipschitzian curve

X : [0, t

1

] → M

its Riemannian length is deﬁned as integral of velocity:

l =

_

t1

0

[

˙

X(t)[ dt, [

˙

X[ =

_

¸

˙

X,

˙

X).

The problem is stated as follows: given any pair of points X

0

, X

1

∈ G, ﬁnd the

shortest curve in G that connects X

0

and X

1

.

The corresponding optimal control problem is as follows:

˙

X = Xu, X ∈ G, u ∈ L, (40)

X(0) = X

0

, X(t

1

) = X

1

, (41)

X

0

, X

1

∈ G ﬁxed, (42)

l(u) =

_

t1

0

[u(t)[ dt → min . (43)

First of all, notice that invariant system (40) is controllable since Γ = L is

full-rank and symmetric, while G is connected.

By Cauchy-Schwartz inequality,

(l(u))

2

=

__

t1

0

[u(t)[ dt

_

2

≤

_

t1

0

[u(t)[

2

dt t

1

,

moreover, the equality occurs only if [u(t)[ ≡ const. Consequently, the Rieman-

nian problem l → min is equivalent to the problem

J(u) =

1

2

_

t1

0

[u(t)[

2

dt → min . (44)

The functional J is more convenient than l since J is smooth and its extremals

are automatically curves with constant velocity. In the sequel we consider the

problem with the functional J: (40)–(42), (44).

53

Further, Filippov’s theorem [1] implies existence of optimal controls in prob-

lem (40)–(42), (44), thus in the initial problem (40)–(43) as well.

The Hamiltonian of PMP for the problem J → min has the form:

h

ν

u

(a, X) = ¸¯ a

X

, Xu) +

ν

2

[u[

2

= ¸a, u) +

ν

2

[u[

2

= h

ν

u

(a).

We apply PMP. If a pair (u(t), X(t)) is optimal, t ∈ [0, t

1

], then there exist

a curve a(t) ∈ L and ν ≤ 0 such that:

(1) (a(t), ν) ,= 0,

(2)

_

¸

_

¸

_

˙ a =

_

a,

∂ h

∂ a

_

= [a, u],

˙

X = X

∂ h

∂ a

= Xu.

(3) h

ν

u(t)

(a(t)) = max

u∈L

h

ν

u

(a(t)).

Since the group G is compact, we write Hamiltonian system (2) in the form (39).

Consider ﬁrst the abnormal case: ν = 0. The maximality condition

h

0

u

(a) = ¸a, u) → max

u∈L

implies that a(t) ≡ 0. This contradicts PMP since the pair (ν, a) should be

nonzero. So there are no abnormal extremal trajectories.

Now consider the normal case: ν < 0. Notice that conditions of PMP (1)–

(3) are preserved under multiplications of (a, ν) by positive constants, so we can

assume that ν = −1. The maximality condition

h

−1

u

(a) = ¸a, u) −

1

2

[u[

2

→ max

u∈L

gives u(t) ≡ a(t). The Hamiltonian system (2) for such a control has the form:

_

˙ a = [a, a] = 0,

˙

X = Xa.

Thus optimal trajectories are left translations of one-parameter subgroups in M:

X(t) = X

0

e

ta

, a ∈ L.

We showed that for any X

0

, X

1

∈ G and any t

1

> 0 there exists a ∈ L such

that

X

1

= X

0

e

at1

.

In particular, for the case X

0

= Id, t

1

= 1, we obtain that any point X

1

∈ G

can be represented in the form

X

1

= e

a

, a ∈ L.

That is, any element X

1

in a connected compact Lie group G has a logarithm

a in the Lie algebra L.

54

8.2 Sub-Riemannian problem on SO(3)

Consider the case G = SO(3), and modify the previous problem. As before, we

should ﬁnd the shortest path between ﬁxed points X

0

, X

1

in the Lie group G.

But now admissible velocities

˙

X are not free: they should be tangent to a left-

invariant distribution (of corank 1) on X. That is, we deﬁne a left-invariant

ﬁeld of tangent hyperplanes on X, and

˙

X(t) should belong to the hyperplane

attached at the point X(t). A problem of ﬁnding shortest curves tangent to a

given distribution ∆

X

⊂ T

X

G is called a sub-Riemannian problem:

˙

X ∈ ∆

X(t)

,

X(0) = X

0

, X(t

1

) = X

1

,

l(X()) → min .

To state the problem as an optimal control one, choose an element b ∈ L,

[b[ = 1, such that ∆

Id

= b

⊥

= ¦u ∈ L [ ¸u, b) = 0¦. Denote U = b

⊥

. Then

∆

X

= XU, and the restriction

˙

X ∈ ∆

X

can be written as

˙

X = Xu, u ∈ U.

For a rigid body rotating in R

3

with orientation matrix X ∈ SO(3), this

restriction on velocities means that we ﬁx an axis b in the rigid body and allow

only rotations of the body around any axis u orthogonal to b.

The optimal control problem is stated as follows.

˙

X = Xu, X ∈ G, u ∈ U,

X(0) = X

0

, X(1) = X

1

,

X

0

, X

1

∈ G ﬁxed,

l(u) =

_

t1

0

[u(t)[ dt → min .

Controllability: we have Γ = b

⊥

= span(a

1

, a

2

) for some linearly indepen-

dent a

1

, a

2

∈ so(3). Since [a

1

, a

2

] / ∈ span(a

1

, a

2

), the system Γ has the full rank,

thus it is controllable on SO(3).

Similarly to the Riemannian problem, the length minimization problem is

equivalent to the problem

J(u) =

1

2

_

t1

0

[u(t)[

2

dt → min,

and Filippov’s theorem guarantees existence of optimal controls.

The Hamiltonian of PMP is the same as in the previous problem:

h

ν

u

(a) = ¸a, u) +

ν

2

[u[

2

.

Consider ﬁrst the abnormal case: ν = 0. The maximality condition of PMP

reads

h

0

u

(a) = ¸a, u) → max

u⊥b

. (45)

55

Consider the decomposition

a = a

+a

⊥

, a

| b, a

⊥

⊥ b. (46)

Then maximality condition (45) is rewritten as

h

0

u

(a) = ¸a

⊥

, u) → max

u⊥b

,

which yields a

⊥

= 0, i.e.,

a(t) = α(t)b, α(t) ,= 0.

The vertical part of Hamiltonian system (39) for our problem yields

˙ αb = ˙ a =

_

a,

∂ h

∂ a

_

= [a, u] = α[b, u]. (47)

Further, by invariance of the scalar product in so(3),

¸b, [b, u]) = −¸[b, b], u) = 0.

Thus

[b, u] ⊥ b ⇒ ˙ αb ⊥ b ⇒ ˙ α = 0.

Then equality (47) implies α[b, u] = 0, so [b, u] = 0. But such an equality

in so(3) means that u | b. Since u ⊥ b, we obtain u ≡ 0 for an abnormal

optimal control. Then the horizontal part of Hamiltonian system (39) reads

˙

X = X

∂ h

∂ a

= Xu = 0. That is, X ≡ const, abnormal optimal trajectories are

constant and give only trivial solutions to our problem.

Now consider the normal case: ν = −1. Via decomposition (46), the maxi-

mality condition of PMP reads

h

−1

u

(a) = ¸a

⊥

, u) −

1

2

[u[

2

→ max

u⊥b

,

thus normal optimal controls are

u = a

⊥

= a −¸b, a)b.

The vertical part of the Hamiltonian system of PMP takes the form

˙ a = [a, u] = [a, a −¸b, a)b] = ¸b, a)[b, a]. (48)

It is easy to see that this ODE has the integral ¸b, a) ≡ const:

¸b, a)

·

= ¸b, ˙ a) = ¸b, [b, a])

. ¸¸ .

=0

¸b, a) = 0.

So equation (48) can be rewritten as

˙ a = ¸b, a

0

)[b, a] = ad(¸b, a

0

)b)a a

0

= a(0),

56

which is immediately solved:

a(t) = e

t ad(b,a0b)

a

0

.

Now consider the horizontal part of the Hamiltonian system of PMP:

˙

X = Xu = X(a −¸b, a

0

)b) = X

_

e

t ad(b,a0b)

a

0

−¸b, a

0

)b

_

= Xe

t ad(b,a0b)

(a

0

−¸b, a

0

)b) .

In the notation

c = ¸b, a

0

)b, d = a

0

−¸b, a

0

)b,

we obtain the ODE

˙

X = Xe

t ad c

d = Xe

tc

d e

−tc

,

that is,

˙

Xe

tc

= Xe

tc

d.

After the change of variable Y = Xe

tc

, we come to the equation

˙

Y =

˙

Xe

tc

+Xe

tc

c = Xe

tc

(d +c) = Y (d +c),

which is solved as

Y (t) = Y (0)e

t(d+c)

.

Finally,

X(t) = Y (t)e

−tc

= Y (0)e

t(d+c)

e

−tc

= X(0)e

ta0

e

−tb,a0b

.

Summing up, we showed that all optimal trajectories in the sub-Riemannian

problem on SO(3) are products of two one-parameter subgroups.

8.3 Sub-Riemannian problem on the Heisenberg group

The Heisenberg group is the deﬁned as

G =

_

_

_

_

_

1 x z

0 1 y

0 0 1

_

_

[ x, y, z ∈ R

_

_

_

.

This group is diﬀeomorphic to R

3

x,y,z

, thus it is not compact.

Its Lie group is

L =

_

_

_

_

_

0 α γ

0 0 β

0 0 0

_

_

[ α, β, γ ∈ R

_

_

_

= span(e

1

, e

2

, e

3

),

where we denote

e

1

= E

12

, e

2

= E

23

, e

3

= E

13

. (49)

57

The multiplication table in this basis looks like

[e

1

, e

2

] = e

3

, [e

1

, e

3

] = [e

2

, e

3

] = 0,

thus

ade

1

=

_

_

0 0 0

0 0 0

0 1 0

_

_

, ad e

2

=

_

_

0 0 0

0 0 0

−1 0 0

_

_

, ade

3

= 0.

So any adjoint operator

adA =

_

_

0 0 0

0 0 0

−A

2

A

1

0

_

_

, A =

3

i=1

A

i

e

i

∈ L, (50)

has the zero spectrum. Consequently, the Heisenberg group G is nilpotent.

In the dual of the Heisenberg Lie algebra L one can choose the basis dual to

basis (49):

L

∗

= span(ω

1

, ω

2

, ω

3

), ¸ω

i

, e

j

) = δ

ij

, i, j = 1, 2, 3.

We write elements of the Lie algebra as column vectors

L ÷ A =

3

i=1

A

i

e

i

=

_

_

A

1

A

2

A

3

_

_

,

and elements of its dual as row vectors:

L

∗

÷ a =

3

i=1

a

i

ω

i

=

_

a

1

a

2

a

3

_

.

For a linear operator C : L → L, its dual C

∗

: L

∗

→ L

∗

acts as

¸C

∗

a, A) = ¸a, CA) =

_

a

_

_

_

C

_

_

_

_

A

_

_

,

the product of a row vector, a square matrix, and a column vector. Thus

C

∗

a =

_

a

_

_

_

C

_

_

.

Consider the left-invariant sub-Riemannian problem on the Heisenberg group

determined by the orthonormal frame (e

1

, e

2

). The plane

∆

Id

= span(e

1

, e

2

) ⊂ L

58

generates the left-invariant distribution

∆

X

= span(Xe

1

, Xe

2

) ⊂ T

X

G.

Further, the scalar product ¸, )

Id

in ∆

Id

deﬁned by

¸e

i

, e

j

)

Id

= δ

ij

, i, j = 1, 2,

generates the left-invariant scalar product ¸, )

X

in ∆

X

as follows:

¸Xe

i

, Xe

j

)

X

= δ

ij

, i, j = 1, 2.

The distribution ∆

X

⊂ T

X

G with the scalar product ¸, )

X

in ∆

X

determine a

left-invariant sub-Riemannian structure on the Lie group G.

Consider the corresponding sub-Riemannian problem:

˙

X ∈ ∆

X

,

X(0) = X

0

, X(t

1

) = X

1

,

l(X()) =

_

t1

0

[

˙

X[ dt =

_

t1

0

_

¸

˙

X,

˙

X) dt → min .

The corresponding control system reads

˙

X = u

1

Xe

1

+u

2

Xe

2

, (u

1

, u

2

) ∈ R

2

. (51)

Since

[

˙

X[ = [u

1

Xe

1

+u

2

Xe

2

[ = [u

1

e

1

+u

2

e

2

[ =

_

u

2

1

+u

2

2

,

the sub-Riemannian length functional takes the form

l =

_

t1

0

_

u

2

1

+u

2

2

dt → min . (52)

We solve optimal control problem (51), (52).

Controllability: we have Γ = span(e

1

, e

2

) ⊂ L. Since [e

1

, e

2

] = e

3

, the

system Γ has full rank. Moreover, Γ is symmetric and G is connected, thus Γ is

controllable.

As before, we pass to the functional

J =

1

2

_

t1

0

(u

2

1

+u

2

2

) dt → min .

Filippov’s theorem implies existence of optimal controls.

The Hamiltonian of PMP reads

h

ν

u

(a, X) = ¸¯ a

X

, u

1

Xe

1

+u

2

Xe

2

) +

ν

2

(u

2

1

+u

2

2

)

= ¸a, u

1

e

1

+u

2

e

2

) +

ν

2

(u

2

1

+u

2

2

) = u

1

a

1

+u

2

a

2

+

ν

2

(u

2

1

+u

2

2

)

= h

ν

u

(a).

59

The Heisenberg group is noncompact, thus a ∈ L

∗

, and we will write the Hamil-

tonian system of PMP in the form (37). First we consider the vertical part

˙ a =

_

ad

∂ h

∂ a

_

∗

a, a ∈ L

∗

. (53)

We have

∂ h

∂ a

= u

1

e

1

+u

2

e

2

=

_

_

u

1

u

2

0

_

_

∈ L.

Taking into account equality (50), we obtain

ad

∂ h

∂ a

=

_

_

0 0 0

0 0 0

−u

2

u

1

0

_

_

.

Thus the vertical part (53) of the Hamiltonian system of PMP takes the form

_

˙ a

1

˙ a

2

˙ a

3

_

=

_

a

1

a

2

a

3

_

_

_

0 0 0

0 0 0

−u

2

u

1

0

_

_

=

_

−a

3

u

2

a

3

u

1

0

_

,

that is,

˙ a

1

= −a

3

u

2

,

˙ a

2

= a

3

u

1

,

˙ a

3

= 0.

Consider ﬁrst the abnormal case: ν = 0. Then the maximality condition of

PMP

h

0

u

(a) = u

1

a

1

+u

2

a

2

→ max

(u1,u2)∈R

2

yields a

1

= a

2

= 0, thus a

3

,= 0. Then the Hamiltonian system implies

˙ a

1

= −a

3

u

2

≡ 0,

˙ a

2

= a

3

u

1

≡ 0,

whence the abnormal optimal controls are u

1

= u

2

≡ 0. Then the horizontal

part of the Hamiltonian system

˙

X = X

∂ h

∂ a

= X(u

1

e

1

+u

2

e

2

)

gives X ≡ 0. Thus there are no nonconstant abnormal optimal trajectories.

In the normal case ν = −1 the maximality condition

h

−1

u

(a) = u

1

a

1

+u

2

a

2

−

1

2

(u

2

1

+u

2

2

) → max

(u1,u2)∈R

2

60

implies u

1

= a

1

, u

2

= a

2

. Consequently, the normal Hamiltonian system of

PMP reads as follows:

˙ a

1

= −a

3

a

2

,

˙ a

2

= a

3

a

1

,

˙ a

3

= 0,

˙

X = X(a

1

e

1

+a

2

e

2

).

It is easy to see that this system has an integral a

2

1

+a

2

2

≡ const since

(a

2

1

+a

2

2

)

·

= 2a

1

(−a

3

a

2

) + 2a

2

a

3

a

1

= 0.

So it is convenient to pass to the polar coordinates

a

1

= r cos θ, a

2

= r sin θ,

in which the vertical part of the Hamiltonian system reads

˙ r = 0,

˙

θ = a

3

,

˙ a

3

= 0.

Now the vertical subsystem is immediately integrated:

θ = θ

0

+a

3

t,

a

1

= r cos(θ

0

+a

3

t),

a

2

= r sin(θ

0

+a

3

t).

We rewrite the horizontal subsystem as

_

_

0 ˙ x ˙ z

0 0 ˙ y

0 0 0

_

_

=

_

_

1 x z

0 1 y

0 0 1

_

_

_

_

0 a

1

0

0 0 a

2

0 0 0

_

_

=

_

_

0 a

1

xa

2

0 0 a

2

0 0 0

_

_

,

that is,

˙ x = a

1

,

˙ y = a

2

,

˙ z = xa

2

.

In view of the left invariance of the problem, we can restrict ourselves by tra-

jectories starting from the identity: X(0) = X

0

= Id, i.e.,

x(0) = y(0) = z(0) = 0.

61

Consider ﬁrst the case a

3

= 0. Then

x =

_

t

0

r cos θ

0

dt = tr cos θ

0

,

y =

_

t

0

r sinθ

0

dt = tr sin θ

0

,

z =

_

t

0

tr

2

cos θ

0

sin θ

0

dt =

t

2

2

r

2

cos θ

0

sin θ

0

.

And if a

3

,= 0, then

x =

_

t

0

r cos(θ

0

+a

3

t) dt =

r

a

3

(sin(θ

0

+a

3

t) −sin θ

0

),

y =

_

t

0

r sin(θ

0

+a

3

t) dt =

r

a

3

(cos θ

0

−cos(θ

0

+a

3

t)),

z =

_

t

0

r

a

3

(sin(θ

0

+a

3

t) −sin θ

0

)r sin(θ

0

+a

3

t) =

=

r

2

a

3

_

t

2

−

sin(2(θ

0

+a

3

t)) −sin 2θ

0

4a

3

+

sin θ

0

a

3

(cos(θ

0

+a

3

t) −cos θ

0

)

_

.

If a

3

= 0, then projections of extremal trajectories X(t) to the plane (x, y)

are straight lines, thus the whole trajectories X(t), t ∈ [0, +∞) are optimal.

And if a

3

,= 0, then such projections are arcs of circles. One can show that

such arcs are optimal up to the ﬁrst complete circle: X(t), t ∈ [0, 2π/[a

3

[].

We found solutions of the minimization problem

_

t1

0

_

˙ x + ˙ y dt → min

along Lipshchitzian plane curves (x(t), y(t)) under the boundary conditions

(x, y, z)(0) = (x

0

, y

0

, z

0

), (x, y, z)(t

1

) = (x

1

, y

1

, z

1

),

where

z(t) =

_

xdy

is the algebraic area of the domain in the plane (x, y) bounded by the curve

(x(t), y(t)), the axis y, and the straight line perpendicular to this axis.

Geometrically, this problem can be stated as follows. Given two points

(x

0

, y

0

), (x

1

, y

1

), a plane curve γ

0

connecting (x

1

, y

1

) to (x

0

, y

0

), and a number

S = z

1

−z

0

, one should ﬁnd a plane curve γ connecting (x

0

, y

0

) to (x

1

, y

1

) such

that the domain bounded by γ and γ

0

has the algebraic area S, and γ is the

shortest possible curve. Solutions to this problem are straight lines and arcs

of circles. This is one of the ancient optimization problems known as Dido’s

problem, it goes back to IX B.C [28].

62

8.4 Euler’s elastic problem

Now we consider a problem studied ﬁrst by L. Euler in 1744 [20].

Suppose that we have two points (x

0

, y

0

), (x

1

, y

1

) in the plane and two unit

vectors v

0

, v

1

, [v

0

[ = [v

1

[ = 1, attached respectively at these points. We should

ﬁnd the proﬁle of the elastic rod with ﬁxed endpoints (x

0

, y

0

), (x

1

, y

1

) and ﬁxed

tangents v

0

, v

1

at these endpoints.

Let (x(t), y(t)), t ∈ [0, t

1

], be the arc-length parametrization of the elastic

rod, t

1

being its length assumed ﬁxed. Let θ(t) be the angle between the velocity

vector ( ˙ x(t), ˙ y(t)) and the positive direction of the axis x. Then the elastic

problem can be stated as follows:

˙ x = cos θ,

˙ y = sinθ,

˙

θ = u,

(x, y, θ)(0) = (x

0

, y

0

, θ

0

), (x, y, θ)(t

1

) = (x

1

, y

1

, θ

1

),

where v

0

= (cos θ

0

, sin θ

0

), v

1

= (cos θ

1

, sin θ

1

). The elastic energy of the rod is

measured by the integral

J =

1

2

_

t1

0

k

2

dt → min,

where k is the curvature of the rod. For an arc-length parametrized curve, the

curvature is, up to sign, equal to the angular velocity, thus k

2

=

˙

θ

2

= u

2

, and

we obtain the cost functional for the optimal control problem:

J =

1

2

_

t1

0

u

2

dt → min .

This problem has obvious symmetries — translations and rotations in the

plane (x, y). So it is natural to expect that it can be stated as an invariant

problem on the Euclidean group E(2). Indeed, the state space of the control

system is

G = E(2) =

_

_

_

_

_

cos θ −sin θ x

sinθ cos θ y

0 0 1

_

_

[ (x, y) ∈ R

2

, θ ∈ S

1

_

_

_

.

Further, the dynamics of the system reads

˙

X =

d

d t

_

_

cos θ −sinθ x

sin θ cos θ y

0 0 1

_

_

=

_

_

−sinθ u −cos θ u cos θ

cos θ u −sinθ u sin θ

0 0 0

_

_

=

_

_

cos θ −sinθ x

sinθ cos θ y

0 0 1

_

_

_

_

0 −u 1

u 0 0

0 0 0

_

_

.

63

The Lie algebra of the Euclidean group is

L = e(2) = span(E

21

−E

12

. ¸¸ .

e1

, E

13

.¸¸.

e2

, E

23

.¸¸.

e3

).

So the elastic problem is left-invariant:

˙

X = X(e

2

+ue

1

), u ∈ R, X ∈ G,

X(0) = X

0

, X(t

1

) = X

1

,

J =

1

2

_

t1

0

u

2

dt → min .

We already computed multiplication table in e(2), see (27):

[e

1

, e

2

] = e

3

, [e

1

, e

3

] = −e

2

, [e

2

, e

3

] = 0,

whence

ad e

1

=

_

_

0 0 0

0 0 −1

0 1 0

_

_

, ade

2

=

_

_

0 0 0

0 0 0

−1 0 0

_

_

, (54)

ad e

3

=

_

_

0 0 0

1 0 0

0 0 0

_

_

. (55)

We choose the dual basis in the dual space to the Lie algebra:

L

∗

= span(ω

1

, ω

2

, ω

3

), ¸ω

i

, e

j

) = δ

ij

,

and write elements of the Lie algebra as column vectors

L ÷ A =

3

i=1

A

i

e

i

=

_

_

A

1

A

2

A

3

_

_

,

and elements of the dual space as row vectors:

L

∗

÷ a =

3

i=1

a

i

ω

i

=

_

a

1

a

2

a

3

_

.

Controllability. The system Γ = e

2

+ Re

1

⊂ L is controllable on G = E(2)

by Theorem 6.8.

Now we ﬁnd extremal trajectories. The Hamiltonian of PMP reads

h

ν

u

(a) = ¸a, e

2

+ue

1

) +

ν

2

u

2

, a ∈ L

∗

, u, ν ∈ R.

Thus

∂ h

∂ a

= e

2

+ue

1

, and in view of (54), (55)

ad

∂ h

∂ a

=

_

_

0 0 0

0 0 −u

−1 u 0

_

_

.

64

Consequently, the Hamiltonian system (37) reads as follows:

˙ a

1

= −a

3

, ˙ x = cos θ,

˙ a

2

= ua

3

, ˙ y = sin θ,

˙ a

3

= −ua

2

,

˙

θ = u.

In the abnormal case ν = 0, and the maximality condition

h

0

u

(a) = a

2

+ua

1

→ max

u∈R

yields a

1

(t) ≡ 0. Then the vertical subsystem takes the form

˙ a

1

= 0 = −a

3

,

˙ a

2

= ua

3

= 0,

˙ a

3

= 0 = −ua

2

.

We have a

1

= a

3

≡ 0, so a

2

≡ const ,= 0 and u ≡ 0. Notice that this is a singular

control , i.e., it is not determined immediately by the maximality condition of

PMP. Now we integrate the horizontal subsystem:

θ = θ

0

,

x = t cos θ

0

,

y = t sin θ

0

.

Consider the normal case: ν = −1,

h

−1

u

(a) = a

2

+ua

1

−

1

2

u

2

→ max

u∈R

,

whence u = a

1

. So the vertical subsystem reads

˙ a

1

= −a

3

,

˙ a

2

= a

1

a

3

,

˙ a

3

= −a

1

a

2

.

In view of the integral a

2

2

+a

2

3

≡ const, we pass to the polar coordinates:

a

2

= r cos α, a

3

= r sin α.

The vertical subsystem simpliﬁes:

˙ r = 0,

˙ α = −a

1

,

˙ a

1

= −r sin α.

The angle α satisﬁes the equation of mathematical pendulum ¨ α = r sin α. Fur-

ther,

˙

θ = u = a

1

= −˙ α, thus θ = β −α, β = const. Finally, the angle θ satisﬁes

65

the equation

¨

θ = −r sin(θ − β), and we obtain the following closed system for

optimal trajectories:

˙ x = cos θ,

˙ y = sin θ,

¨

θ = −r sin(θ −β).

If r = 0, then θ = θ

0

+t

˙

θ

0

, and Euler elastica, i.e., optimal curves (x(t), y(t)),

are the same as in the sub-Riemannian problem on the Heisenberg group, i.e.,

lines and circles.

Let r > 0. Then we can apply homotheties in the plane (x, y) in order to

obtain r = 1, and further apply rotations in this plane in order to have β = 0.

Then the angle θ satisﬁes the standard equation of the mathematical pendulum

¨

θ = −sinθ, i.e.,

˙

θ = c,

˙ c = −sinθ.

Here c is the curvature of Euler elastica. The diﬀerent qualitative types of

solutions to the equation of pendulum depend on values of the energy of the

pendulum

E =

c

2

2

−cos θ ∈ [−1, +∞).

The following cases are possible:

(1) E = −1,

(2) E ∈ (−1, 1),

(3a) E = 1, θ ,= ±π,

(3b) E = 1, θ = ±π,

(4) E ∈ (1, +∞).

It is known that the equation of mathematical pendulum is integrable in

elliptic functions [18]. One can show that equations for elastica are integrable

in elliptic functions as well, and the following qualitative types of elastica are

possible:

(1) straight line, Fig. 1,

(2) inﬂectional elastica, Fig. 2–5,

(3a) critical elastica, Fig. 6,

(3b) straight line, Fig. 7,

(4) non-inﬂectional elastica, Fig. 8–10,

(5) r = 0 ⇒ circles, Fig. 11, and straight lines.

66

0 1 2 3 4

-1

-0.5

0

0.5

1

x

Figure 1: E = −1

0 1 2 3 4 5

0

0.5

1

1.5

2

2.5

3

0 1 2 3 4

0

0.5

1

1.5

2

2.5

3

3.5

Figure 2: E ∈ (−1, 1) Figure 3: E ∈ (−1, 1)

-0.6 -0.4 -0.2 0 0.2 0.4 0.6

0

0.5

1

1.5

2

2.5

3

3.5

y

-20 -15 -10 -5 0

0

1

2

3

4

Figure 4: E ∈ (−1, 1) Figure 5: E ∈ (−1, 1)

67

-3 -2 -1 0 1 2 3

0

0.5

1

1.5

2

y

0 1 2 3 4

-1

-0.5

0

0.5

1

x

Figure 6: E = 1, θ ,= π Figure 7: E = 1, θ = π

-10 -8 -6 -4 -2 0

0

0.25

0.5

0.75

1

1.25

1.5

1.75

-1.25 -1 -0.75 -0.5 -0.25 0 0.25

0

0.2

0.4

0.6

0.8

y

Figure 8: E ∈ (1, +∞) Figure 9: E ∈ (1, +∞)

-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15

0

0.05

0.1

0.15

0.2

0.25

0.3

y

-1 -0.5 0 0.5 1

0

0.5

1

1.5

2

y

Figure 10: E ∈ (1, +∞) Figure 11: r = 0

68

8.5 The plate-ball system

Consider a unit 2-dimensional sphere rolling on a horizontal 2-dimensional plane

without slipping and twisting. Given an initial and a terminal contact conﬁg-

uration of the sphere and the plane, the problem is to roll the sphere from the

ﬁrst conﬁguration to the second one in such a way that the curve in the plane

traced by the contact point be the shortest possible.

Fix an orthonormal frame (e

1

, e

2

, e

3

) in R

3

such that the plane is spanned

by e

1

, e

2

, and the vector e

3

is directed upwards (to the half-space containing

the sphere). In addition, choose an orthonormal frame (f

1

, f

2

, f

3

) attached to

the sphere. Then orientation of the sphere in the space is determined by the

orientation matrix

R : (e

1

, e

2

, e

3

) → (f

1

, f

2

, f

3

), R ∈ SO(3),

and position of the contact point of the sphere with the plane is given by its

coordinates (x, y) in the plane corresponding to the frame (e

1

, e

2

). Then the

state of the system is described by the tuple

X = (R, x, y) ∈ SO(3) R

2

.

We have initial and terminal states ﬁxed:

X(0) = X

0

, X(t

1

) = X

1

,

and the cost functional is

l =

_

t1

0

_

˙ x

2

+ ˙ y

2

dt → min .

Moreover, it is easy to see that the dynamics of the system is described by the

following ODEs:

˙ x = u

1

,

˙ y = u

2

,

˙

R = R

_

_

0 0 −u

1

0 0 −u

2

u

1

u

2

0

_

_

.

The ﬁrst two equations mean that the contact point (x, y) moves in the plane

with an arbitrary velocity (u

1

, u

2

), while the third equation means that the

angular velocity of the rolling sphere is horizontal and perpendicular to (u

1

, u

2

),

see [14] for details.

We can assemble the state X to a single 6 6 matrix

X =

_

_

_

_

_

_

_

_

R 0

1 0 x

0 0 1 y

0 0 1

_

_

_

_

_

_

_

_

,

69

denote by G the Lie group of all such matrices for all R ∈ SO(3), (x, y) ∈ R

2

.

Then the dynamics of the system takes the left-invariant form as follows:

˙

X =

_

_

_

_

_

_

_

_

˙

R 0

0 0 ˙ x

0 0 0 ˙ y

0 0 0

_

_

_

_

_

_

_

_

=

_

_

_

_

_

_

_

_

R 0

1 0 x

0 0 1 y

0 0 1

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

0 0 −u

1

0 0 −u

2

0

u

1

u

2

0

0 0 u

1

0 0 0 u

2

0 0 0

_

_

_

_

_

_

_

_

,

that is,

˙

X = X(u

1

(E

31

−E

13

+E

46

) +u

2

(E

32

−E

23

+E

56

)), X ∈ G, (u

1

, u

2

) ∈ R

2

.

The Lie algebra of the Lie group G is

L = span(E

32

−E

23

. ¸¸ .

e1

, E

13

−E

31

. ¸¸ .

e2

, E

21

−E

12

. ¸¸ .

e3

, E

46

.¸¸.

e4

, E

56

.¸¸.

e5

),

with the multiplication rules inherited from so(3):

[e

1

, e

2

] = e

3

, [e

2

, e

3

] = e

1

, [e

3

, e

1

] = e

2

, ade

4

= ade

5

= 0. (56)

The nonzero adjoint operators read as follows:

ad e

1

=

_

_

_

_

_

_

_

_

0 0 0

0 0 −1 0

0 1 0

0 0

_

_

_

_

_

_

_

_

, ade

2

=

_

_

_

_

_

_

_

_

0 0 1

0 0 0 0

−1 0 0

0 0

_

_

_

_

_

_

_

_

, (57)

ad e

3

=

_

_

_

_

_

_

_

_

0 −1 0

1 0 0 0

0 0 0

0 0

_

_

_

_

_

_

_

_

. (58)

As usual, we choose the dual basis in the space dual to the Lie algebra:

L

∗

= span(ω

1

, . . . , ω

5

), ¸ω

i

, e

j

) = δ

ij

, i, j = 1, . . . , 5,

70

write elements of the Lie algebra as column vectors:

L ÷ A =

5

i=1

A

i

e

i

=

_

_

_

A

1

.

.

.

A

5

_

_

_,

and elements of the dual space as column vectors:

L

∗

÷ a =

5

i=1

a

i

ω

i

=

_

_

a

1

. . .

a

5

_

_

.

Now we study the plate-ball optimal control problem:

˙

X = X(u

1

(e

4

−e

2

) +u

2

(e

5

+e

1

)), X ∈ G, (u

1

, u

2

) ∈ R

2

,

X(0) = X

0

, X(t

1

) = X

1

,

J =

1

2

_

t1

0

(u

2

1

+u

2

2

) → min,

notice that we replace the functional l by J as always.

Controllability: multiplication rules (56) imply that the control system has

full rank. Since it is symmetric and G is connected, controllability follows.

Existence of optimal controls follows from Filippov’s theorem.

The Hamiltonian of PMP has the form

h

ν

u

(a) = ¸a, u

1

(e

4

−e

2

) +u

2

(e

5

+e

1

)) +

ν

2

(u

2

1

+u

2

2

),

then

∂ h

∂ a

= u

1

(e

4

−e

2

) +u

2

(e

5

+e

1

),

and it follows from (56), (57), (58) that

ad

∂ h

∂ a

=

_

_

_

_

_

_

_

_

0 0 −u

1

0 0 −u

2

0

u

1

u

2

0

0 0

_

_

_

_

_

_

_

_

.

So the vertical subsystem of the Hamiltonian system of PMP reads

_

˙ a

1

˙ a

2

˙ a

3

˙ a

4

˙ a

5

_

=

_

a

1

a

2

a

3

a

4

a

5

_

_

_

_

_

_

_

_

_

0 0 −u

1

0 0 −u

2

0

u

1

u

2

0

0 0

_

_

_

_

_

_

_

_

.

71

So the whole Hamiltonian system of PMP takes the form:

˙ a

1

= u

1

a

3

, ˙ x = u

1

,

˙ a

2

= u

2

a

3

, ˙ y = u

2

,

˙ a

3

= −u

1

a

1

−u

2

a

2

,

˙ a

4

= ˙ a

5

= 0,

˙

R = R

_

_

0 0 −u

1

0 0 −u

2

u

1

u

2

0

_

_

.

Consider ﬁrst the abnormal case, ν = 0. Then

h

0

u

(a) = u

1

(a

4

−a

2

) +u

2

(a

5

+a

1

) → max

(u1,u2)

∈ R

2

,

whence a

4

−a

2

≡ 0, a

5

+a

1

≡ 0. Thus

a

1

= −a

5

≡ const,

a

2

= a

4

≡ const,

˙ a

1

= 0 = u

1

a

3

,

˙ a

2

= 0 = u

2

a

3

.

But non-constant extremal curves of the functional J satisfy the identity u

2

1

+

u

2

2

≡ const ,= 0, so a

3

= 0. Finally,

˙ a

3

= 0 = −u

1

a

1

−u

2

a

2

.

Then optimal abnormal controls (u

1

, u

2

) are constant, the corresponding curve

(x, y) is a straight line, and the orientation matrix is

R(t) = R

0

exp

_

_

t

_

_

0 0 −u

1

0 0 −u

2

u

1

u

2

0

_

_

_

_

.

Now we pass to the normal case, ν = −1. Then

h

−1

u

(a) = u

1

(a

4

−a

2

) +u

2

(a

5

+a

1

) −

1

2

(u

2

1

+u

2

2

) → max

(u1,u2)∈R

2

,

whence

u

1

= a

4

−a

2

, u

2

= a

5

+a

1

.

For these controls, the vertical subsystem of the Hamiltonian system of PMP

takes the form

˙ a

1

= (a

4

−a

2

)a

3

,

˙ a

2

= (a

5

+a

1

)a

3

,

˙ a

3

= −(a

4

−a

2

)a

1

−(a

5

+a

1

)a

2

,

˙ a

4

= ˙ a

5

= 0.

72

Introduce the variables

b

1

= a

4

−a

2

= u

1

,

b

2

= a

5

+a

1

= u

2

,

b

3

= a

3

,

then the above system reduces to the following one:

˙

b

1

= −b

2

b

3

,

˙

b

2

= b

1

b

3

,

˙

b

3

= a

5

b

1

−a

4

b

2

.

This system has an integral b

2

1

+ b

2

2

≡ const, which can be set equal to 1 by

homogeneity of the system. We pass to the polar coordinates

b

1

= cos θ, a

4

= r cos ϕ,

b

2

= sin θ, a

5

= r sin ϕ,

in which

˙

θ = b

3

,

˙

b

3

= r sin(ϕ −θ),

that is, the angle θ satisﬁes the equation of pendulum

¨

θ = −r sin(θ −ϕ).

The coordinates of the contact point satisfy the ODEs

˙ x = u

1

= b

1

= cos θ,

˙ y = u

2

= b

2

= sin θ.

Thus we obtain a remarkable result: the contact point of the sphere rolling

optimally traces Euler elastica!

A description of the corresponding orientation matrix R(t) can be found

in [14].

Remarks on bibliography

The bibliography contains references of several kinds:

(1) textbooks on control theory [1, 14, 30], sub-Riemannian geometry [28],

nonholonomic dynamics [44], and diﬀerential geometry and Lie groups [45],

(2) works on controllability of invariant systems on Lie groups [2, 3, 4, 5, 11,

12, 13, 15, 16, 17, 19, 21, 22, 31, 32, 33, 34, 39, 40, 41, 42, 43], including

a survey on the subject [35],

(3) papers on optimal control for invariant problems on Lie groups [10, 23,

24, 25, 26, 29, 36, 37],

(4) other works referred to in these notes.

73

References

[1] A.A. Agrachev, Yu. L. Sachkov, Control Theory from the Geometric View-

point, Springer Verlag, 2004.

[2] R. El Assoudi, J. P. Gauthier, and I. Kupka, “On subsemigroups of semisim-

ple Lie groups,” Ann. Inst. Henri Poincar´e, 13, No. 1, 117–133 (1996).

[3] V. Ayala Bravo, Controllability of nilpotent systems. In: Geometry in

nonlinear control and diﬀerential inclusions, Banach Center Publications,

Warszawa, 32, 35–46 (1995).

[4] B. Bonnard, Contrˆ ollabilit´e des syst`emes bilin´eaires, Math. Syst. Theory,

15, 79–92 (1981).

[5] B. Bonnard, V. Jurdjevic, I. Kupka, and G. Sallet, “Transitivity of families

of invariant vector ﬁelds on the semidirect products of Lie groups,” Trans.

Amer. Math. Soc., 271, No. 2, 525–535 (1982).

[6] W. Boothby, “A transitivity problem from control theory,” J. Diﬀ. Equat.,

17, 296–307 (1975).

[7] W. Boothby and E. N. Wilson, “Determination of the transitivity of bilinear

systems,” SIAM J. Control, 17, 212–221 (1979).

[8] A. Borel, “Some remarks about transformation groups transitive on spheres

and tori,” Bull. Amer. Math. Soc., 55, 580–586 (1949).

[9] A. Borel, “Le plan projectif des octaves et les sph`eres comme espaces ho-

mog`enes,” C. R. Acad. Sci. Paris, 230, 1378–1380 (1950).

[10] U. Boscain, T. Chambrion, J.-P. Gauthier, On the K + P problem for

a three-level quantum system: Optimality implies resonance, J. Dynam.

Control Systems, 8, 547–572 (2002).

[11] R. W. Brockett, “System theory on group manifolds and coset spaces,”

SIAM J. Control, 10, 265–284 (1972).

[12] J. P. Gauthier and G. Bornard, Contrˆ olabilit´e des syst`emes bilin`eaires,

SIAM J. Control Optim. 20 (1982), 3, 377–384.

[13] J. Hilgert, K.H. Hofmann, J.D. Lawson, Controllability of systems on a

nilpotent Lie group, Beitr¨ age Algebra Geometrie, 20, 185–190 (1985).

[14] V. Jurdjevic, Geometric control theory, Cambridge University Press, 1997.

[15] V. Jurdjevic and I. Kupka, “Control systems subordinated to a group ac-

tion: Accessibility,” J. Diﬀer. Equat., 39, 186–211 (1981).

[16] V. Jurdjevic and I. Kupka, “Control systems on semi-simple Lie groups

and their homogeneous spaces,” Ann. Inst. Fourier, Grenoble 31, No. 4,

151–179 (1981).

74

[17] V. Jurdjevic and H. Sussmann, “Control systems on Lie groups,” J. Diﬀ.

Equat., 12, 313–329 (1972).

[18] D.F. Lawden, Elliptic functions and applications, Springer-Verlag, 1980.

[19] J. D. Lawson, “Maximal subsemigroups of Lie groups that are total,” Proc.

Edinburgh Math. Soc., 30, 479–501 (1985).

[20] A. E. H. Love, A treatise on the mathematical theory of elasticity, New York:

Dover, 1927.

[21] D. Mittenhuber, “Controllability of solvable Lie algebras”, J. Dynam. Con-

trol Systems, 6, No. 3, 453–459 (2000).

[22] D. Mittenhuber, “Controllability of systems on solvable Lie groups: the

generic case”, J. Dynam. Control Systems, 7, No. 1, 61–75 (2001).

[23] F. Monroy-P´erez, Non-Euclidean Dubins problem. J. Dynam. Control Sys-

tems, 4, no. 2, 249–272 (1998).

[24] F. Monroy-P´erez, A.Anzaldo-Meneses, “Optimal Control on the Heisenberg

Group”, J. Dynam. Control Systems, 5, no. 4, 473–499 (1999).

[25] F. Monroy-P´erez, A.Anzaldo-Meneses, “Optimal Control on Nilpotent Lie

Groups”, J. Dynam. Control Systems, 8, no. 4, 487–504 (2002).

[26] F. Monroy-P´erez, A. Anzaldo-Meneses, The step-2 nilpotent

(n, n(n + 1)/2) sub-Riemannian geometry, J. Dynam. Control Systems,

12, No. 2, 185–216 (2006).

[27] D. Montgomery and H. Samelson, “Transformation groups of spheres,”

Ann. Math., 44, 454–470 (1943).

[28] R. Montgomery, A tour of subriemannian geometries, their geodesics and

applications, Amer. Math. Soc., 2002.

[29] O. Myasnichenko, Nilpotent (3, 6) Sub-Riemannian Problem. J. Dynam.

Control Systems, 8, No. 4, 573–597 (2002).

[30] L.S. Pontryagin, V.G. Boltyanskij, R.V. Gamkrelidze, E.F. Mishchenko,

The Mathematical Theory of Optimal Processes, Pergamon Press, Oxford

(1964)

[31] Yu. L. Sachkov, “Controllability of hypersurface and solvable invariant sys-

tems,” J. Dyn. Control Syst., 2, No. 1, 55–67 (1996).

[32] Yu. L. Sachkov, “Controllability of right-invariant systems on solvable Lie

groups,” J. Dyn. Control Syst., 3, No. 4, 531–564 (1997).

[33] Yu.L. Sachkov, On invariant orthants of bilinear systems, J. Dyn. Control

Syst., 4, No. 1, 137–147 (1998).

75

[34] Yu. L. Sachkov, Classiﬁcation of controllable systems on low-dimensional

solvable Lie groups, Journal of Dynamical and Control Systems, 6 (2000),

2: 159–217.

[35] Yu. L. Sachkov, Controllability of Invariant Systems on Lie Groups and

Homogeneous Spaces, in: Progress in Science and Technology, Series on

Contemporary Mathematics and Applications, Thematical Surveys, Vol. 59,

Dynamical Systems-8, All-Russian Institute for Scientiﬁc and Technical In-

formation (VINITI), Ross. Akad. Nauk, Moscow, 1998 (to appear); English

transl: J. Math. Sci., v. 100, No. 4, 2000, 2355–2427.

http://www.botik.ru/PSI/CPRC/sachkov/public.html

[36] Yu.L. Sachkov, Exponential mapping in the generalized Dido problem (in

Russian), Matem. Sbornik, 194, 9: 63–90 (2003).

[37] Yu.L. Sachkov, Symmetries of Flat Rank Two Distributions and Sub-

Riemannian Structures. Transactions of the American Mathematical So-

ciety, 356, 2: 457–494 (2004).

[38] H. Samelson, “Topology of Lie groups,” Bull. Amer. Math. Soc., 58, 2–37

(1952).

[39] L. A. B. San Martin, “Invariant control sets on ﬂag manifolds,” Math. Con-

trol Signals Systems, 6, 41–61 (1993).

[40] L. A. B. San Martin, O. G. do Rocio, A.J. Santana, Invariant cones and

convex sets for bilinear control systems and parabolic type of semigroups,

Journal of Dynamical and Control Systems, 12 (2006), to appear.

[41] L. A. B. San Martin and P. A. Tonelli, “Semigroup actions on homogeneous

spaces,” Semigroup Forum, 14, 1–30 (1994).

[42] F. Silva Leite and P. Crouch, “Controllability on classical Lie groups,”

Math. Control Signals Systems, 1, 31–42 (1988).

[43] R.M.M. Troncoso, “Regular Elements and Global Controllability in

SL(d, R)”, Journal of Dynamical and Control Systems, 10 (2004) No. 1,

29–54.

[44] A.M. Vershik, V.Ya. Gershkovich, Nonholonomic dynamic systems. Geom-

etry of distributions and variational problems. Springer Verlag, EMS, 16,

5–85 (1987).

[45] F. W. Warner, Foundations of diﬀerentiable manifolds and Lie groups,

Glenview, Ill. : Scott, Foresman, 1971.

76

Index

E

ij

, 38

M(n), 5

M(n, C), 7

M(n, R), 5

V K, 45

[A, B], 10

[V, W], 15

Γ, 16

Γ

1

∼ Γ

2

, 26

¯

Z, 8

Aﬀ(n), 7

Aﬀ

+

(n), 35

E(n), 7, 45

GL(n, C), 8

GL(n, R), 5

GL

+

(n), 24

LS(Γ), 29

Lie(Γ), 20

SL(n, C), 8

SL(n, R), 6

SO(n), 6

SU(n), 9

T(n), 7

U(n), 8

cl S, 21

int S, 21

int

O

S, 24

span(l), 21

adA, 30

O(n), 6

Sat(Γ), 27

Vec M, 15

aﬀ(n), 11

co(l), 21

e(n), 11

gl(n), 10

sl(n), 11

so(n), 11

su(n), 12

t(n), 11

u(n), 12

exp(A), 12

θ, 32

θ

∗

A, 33

θ

∗

Γ, 33

e

A

, 12

X

T

, 6

/, 21

/(X), 17

/(X, T), 17

/

Γ

, 21

/

Γ

(X), 17

/

Γ

(X, T), 17

/

Γ

(X, ≤ T), 17

O, 21

O

Γ

, 21

O

Γ

(X), 20

abnormal case, 54

action of a Lie group on a manifold,

32

aﬃne group, 7

attainable set, 17

canonical coordinates, 47

canonical projection, 48

classical notation for left-invariant

systems, 16

commutator, 10, 15

compact Lie group, 37

complete controllability, 17

completely solvable Lie algebra, 44

complex general linear group, 8

complex special linear group, 8

control-aﬃne system, 16

control-linear system, 37

controllable system, 17

derived series, 43

equivalent systems, 26

Euclidean group, 7

Euler elastica, 66

exponential, 12

full rank, 23

77

general linear group, 5

global controllability, 17

Hamiltonian, 48

Hamiltonian vector ﬁeld, 48

Heisenberg group, 57

homogeneous space, 32

ideal, 39

induced system, 33

induced vector ﬁeld, 33

Jacobi identity, 10

Kalman condition, 36

left-invariant control system, 16

left-invariant vector ﬁeld, 14

Lie algebra, 10

Lie algebra of Lie group, 10

Lie Algebra Rank Condition, 23

Lie bracket, 15

Lie group, 5

Lie saturate, 29

linear Lie group, 6

linear system, 36

local controllability, 17, 25

nilpotent Lie algebra, 44

normal case, 54

normally attainable point, 21

optimal control problem, 47

orbit, 20

orthogonal group, 6

permutation-irreducible matrix, 40

permutation-reducible matrix, 40

polysystem, 16

Pontryagin Maximum Principle, 49

reachable set, 17

reachable set for time T, 17

reachable set for time not greater

than T ≥ 0, 17

realiﬁcation, 7

right-invariant control system, 18

right-invariant vector ﬁeld, 17

saturate, 27

semi-direct product, 45

semisimple Lie algebra, 39

semisimple Lie group, 39

simple Lie algebra, 39

simple Lie group, 39

simply connected space, 43

singular control, 65

solvable Lie algebra, 43

solvable Lie group, 43

special linear group, 6

special orthogonal group, 6

special unitary group, 9

sub-Riemannian problem, 55

symmetric system, 36

symplectic form, 48

tautological 1-form, 48

trajectory, 16

transitive action, 32

triangular group, 7

trivialization, 49

unitary group, 8

78

**Control theory on Lie groups
**

Yu. L. Sachkov Program Systems Institute Pereslavl-Zalessky, Russia e-mail: sachkov@sys.botik.ru

∗

Abstract Lecture notes of an introductory course on control theory on Lie groups. Controllability and optimal control for left-invariant problems on Lie groups are addressed. A general theory is accompanied by concrete examples. The course is intended for graduate students, no preliminary knowledge of control theory or Lie groups is assumed.

∗ Lectures given in SISSA, Trieste, Italy, January–February 2006. The author thanks SISSA for support. Work partially supported by Russian Foundation for Basic Report, Project No. 05-01-00703-a.

1

Contents

1 Motivation 1.1 Bilinear systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Rotations of a rigid body . . . . . . . . . . . . . . . . . . . . . . 2 Lie 2.1 2.2 2.3 2.4 groups and Lie algebras Linear Lie groups . . . . . . . . The Lie algebra of a Lie group Matrix exponential . . . . . . . Left-invariant vector ﬁelds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 4 5 5 9 12 13 16 16 17 18 21 24

3 Left-invariant control systems 3.1 Deﬁnitions . . . . . . . . . . . . . . . . . . . 3.2 Right-invariant control systems . . . . . . . 3.3 Basic properties of orbits and reachable sets 3.4 Normal attainability . . . . . . . . . . . . . 3.5 General controllability conditions . . . . . .

4 Extension techniques for left-invariant systems 26 4.1 Saturate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 4.2 Lie saturate of invariant system . . . . . . . . . . . . . . . . . . . 29 5 Induced systems on homogeneous spaces 6 Controllability conditions for special classes of groups 6.1 Symmetric systems . . . . . . . . . . . . . . . . 6.2 Compact Lie groups . . . . . . . . . . . . . . . 6.3 Semisimple Lie groups . . . . . . . . . . . . . . 6.4 Solvable Lie groups . . . . . . . . . . . . . . . . 6.5 Semi-direct products of Lie groups . . . . . . . systems and Lie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 36 37 39 43 45 47 47 48 49 51 53 53 55 57 63 69 31

7 Pontryagin Maximum Principle for invariant optimal control problems on Lie groups 7.1 Hamiltonian systems on T ∗ M . . . . . . . . . . . . . . . . . . . . 7.2 Pontryagin Maximum Principle on smooth manifolds . . . . . . . 7.3 Hamiltonian systems on T ∗ G . . . . . . . . . . . . . . . . . . . . 7.4 Hamiltonian systems in the case of compact Lie group . . . . . . 8 Examples of invariant optimal control problems on 8.1 Riemannian problem on compact Lie group . . . . . 8.2 Sub-Riemannian problem on SO(3) . . . . . . . . . . 8.3 Sub-Riemannian problem on the Heisenberg group . 8.4 Euler’s elastic problem . . . . . . . . . . . . . . . . . 8.5 The plate-ball system . . . . . . . . . . . . . . . . . Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

References Index 74 76 3 .

In the study of controllability of a bilinear control system x = Ax + uBx. while the bilinear ones just on smooth submanifolds of Rn (homogeneous spaces of linear Lie groups). e3 in the ambient space. And the study of controllability for matrix systems is an easier problem since here the group structure provides powerful additional techniques. f2 . f2 . geometry etc. rotations of a space satellite around its center of mass). but this is not the case: the matrix systems evolve on matrix groups (linear Lie groups). Then the orientation matrix X : (e1 . There is a clear and important relation between controllability properties of the bilinear system (1) and the matrix system (2): “ system (2) is controllable ⇒ system (1) is controllable ” In the sequel we make this statement precise and remove the quotation marks. But this implication is clear: if we can control on matrices. X n × n matrix. naturally evolve on groups. Consider rotations of a rigid body in R3 around a ﬁxed point (e. f3 attached to the body. e3 ) → (f1 . and a moving orthonormal frame f1 . We will clarify all these questions in our course. So we obtain the equation of motion ˙ X = ΩX. choose a ﬁxed orthonormal frame e1 . it is easy to see that the ˙ matrix XX −1 = Ω is skew-symmetric (the angular velocity of the body). (2) Such a passage is very natural: recall that in the study of the linear ODE ˙ x = Ax. u ∈ R. Moreover. One may think that matrix systems (2) are more complicated then the bilinear ones (1). 4 . physics. here X is the Cauchy matrix for ˙ the linear ODE. e2 .1 Motivation Bilinear systems x ∈ Rn .g. e2 . 1. f3 ) is a 3 × 3 orthogonal unimodular matrix. u ∈ R. ˙ (1) where A and B are constant n × n matrices.1 1.2 Rotations of a rigid body Some important control systems in mechanics. the more so we can control on vectors. one naturally passes from the system (1) for vectors to the similar system for matrices: ˙ X = AX + uBX. In order to describe motion of the body. we pass to the matrix ODE X = AX.

e. Finally. . Denote the linear space of all n × n matrices with real entries as M (n. (X −1 )ij are rational functions in Xij . i. . . then the product XY ∈ GL(n). GL(n) is a group with respect to matrix product. Y ∈ GL(n).1. 5 . i. Let X : Rn → Rn be a linear mapping. For short. . .1 (General Linear Group). and (3) the group operations in G are smooth. Further. The general linear group consists of all n × n invertible matrices: GL(n. j = 1. en of Rn . The following properties of GL(n) are easily established. . Y ) → XY X →X −1 (XY )ij are polynomials in Xij . Deﬁnition 2. i. Indeed. n. the set GL(n) is an open domain. det : M (n) → R. its inverse X −1 is nonsingular as well. 2 2. R) = GL(n) = {X ∈ M (n) | det X = 0}. (2) G is a group. many questions arise. than the previous system is a matrix control system similar to (2). what is the state space of such systems: X∈? We will answer this question in the next section. the group operations in GL(n) are smooth: (X.. . the Kronecker symbol) is contained in GL(n). which we identify with the operator itself. for a nonsingular matrix X. . (1) By continuity of determinant. . Example 2. In a basis e1 . j = 1. . if X. (2) Further. So we are going to consider groups of matrices. and one of the ﬁrst ones is.If we suppose that we can control the matrix Ω. n}. The matrix entries xij 2 provide coordinates on M (n) = Rn . we will usually denote this space by M (n). In the study of such systems. the identity matrix Id = (δij ) (where δij = 1 if i = j and δij = 0 if i = j. Yij . R) = {X = (xij ) | xij ∈ R. A set G is called a Lie group if: (1) G is a smooth manifold. .1 Lie groups and Lie algebras Linear Lie groups The most important class of Lie groups is formed by linear Lie groups. thus a smooth submanifold in the linear space M (n). (3) Moreover. . the operator X has a matrix X = (xij ). groups of linear transformations of Rn .

and (3) G is topologically closed in GL(n) (i. Proof. Example 2. Example 2.r. If G is a closed subgroup of GL(n).4 (Special Orthogonal Group). See e.1 are easily veriﬁed. then G is a linear Lie group. R) = SL(n) = {X ∈ M (n) | det X = 1}. Such matrices correspond to linear operators v → Xv preserving the standard volume in Rn .e. A convenient suﬃcient condition for a set of matrices to form a linear Lie group is given in the following general proposition. it suﬃces to show that the following three conditions hold: (1) G ⊂ GL(n).3 (Orthogonal Group). The special linear group consists of n × n unimodular matrices: SL(n.. where X T denotes the transposed matrix of X. it follows that orthogonal matrices have determinant det X = ±1. where S is a closed subset in M (n)). Deﬁnition 2.In the previous example we showed that GL(n) is a Lie group.t. Orthogonal transformations v → Xv preserve the Euclidean structure in Rn . Theorem 2. (2) G is a group w. In other words. matrix product. A Lie group G ⊂ M (n) is called a linear Lie group.g.2 (Special Linear Group). GL(n). [45]. Special orthogonal transformations v → Xv preserve both the Euclidean structure and orientation in Rn . det X = 1}. In all these cases the hypotheses of Theorem 2.2. G = GL(n) ∩ S. Since 1 = det(XX T ) = det2 X. Orthogonal unimodular matrices form the special orthogonal group: SO(n) = {X ∈ M (n) | XX T = Id. Example 2. The orthogonal group is formed by n×n orthogonal matrices: O(n) = {X ∈ M (n) | XX T = Id}. in order to verify that a set of matrices G ⊂ M (n) is a linear Lie group.1. Now we consider several important examples of linear Lie groups in addition to the largest one. 6 .

Example 2. These are matrices of invertible linear operators v → Xv preserving the ﬂag of subspaces Re1 ⊂ span(e1 . Denote the space of all n × n matrices with complex entries as M (n. i > j. . en−1 ) ⊂ Rn . R). C). . The Euclidean group is the following subgroup of the aﬃne group: E(n) = X= Y 0 b 1 ∈ M (n + 1) | Y ∈ SO(n). n}. b ∈ Rn ⊂ GL(n + 1). And the last example of a group formed by real matrices: the triangular group consists of all invertible triangular matrices: ∗ ∗ ··· ∗ 0 ∗ ··· ∗ T(n) = X = . 7 Z = X + iY ∈ M (n. The aﬃne group is deﬁned as follows: Aﬀ(n) = X= Y 0 b 1 ∈ M (n + 1) | Y ∈ GL(n). Such matrices correspond to invertible aﬃne transformations in Rn of the form v → Y v + b. . .e. k = 1. Y = (yjk ). b ∈ Rn ⊂ GL(n + 1). . a complex matrix decomposes correspondingly: Z = X + iY. yjk turn M (n. The realiﬁcation of a complex matrix is deﬁned as follows. . 2 The real coordinates xjk . C) = {Z = (zjk ) | zjk ∈ C.6 (Euclidean Group). X = (xjk ). Since any entry decomposes into the real and imaginary parts: zjk = xjk + iyjk . j. i. ∈ GL(n) . 0 0 ··· ∗ = {X = (xij ) ∈ M (n) | xij = 0. Now we pass to complex matrices. a linear mapping Y plus a translation b. ..7 (Triangular Group).5 (Aﬃne Group). . . . . .Example 2. Example 2. e2 ) ⊂ · · · ⊂ span(e1 . xii = 0}. . C) into R2n . Such matrices parametrize orientation-preserving aﬃne isometries v → Y v + b. To any complex matrix of order n corresponds a real matrix of order 2n: Z∼ X Y −Y X ∈ M (2n.. . .

Example 2. . We have ¯ Z = X + iY. −X1 Y2 − Y1 X2 X1 X2 − Y 1 Y 2 X2 Y2 −Y2 X2 . Compute the realiﬁcation of a unitary matrix. R) | det(X + iY ) = 1 .) Such matrices correspond to linear transformations that preserve the unitary structure in Cn . The complex general linear group consists of all complex n × n invertible matrices: GL(n. Theorem 2.9 (Complex Special Linear Group). Z2 = X2 + iY2 ∼ X2 Y2 −Y2 X2 . C) | det Z = 1} = X Y −Y X ∈ M (2n. . R).X −Y is just the matrix of the real linear operator in R2n = Y X Cn corresponding to Z in the basis over reals e1 . R) | det2 X + det2 Y = 0 . An important example of a linear Lie group is the unitary group consisting of all n × n unitary matrices: ¯T U(n) = {Z ∈ M (n. . So it is natural that realiﬁcation respects the matrix product: if The matrix Z1 = X1 + iY1 ∼ then Z1 Z2 = X1 X2 − Y1 Y2 + i(Y1 X2 + X1 Y2 ) ∼ = X1 X2 − Y 1 Y 2 X1 Y 2 + Y 1 X2 X1 Y1 −Y1 X1 X1 Y1 −Y1 X1 . C) | Z Z = Id}. 8 .2. .1. . Example 2. e1 .10 (Unitary Group). then G is a linear Lie group. · The realiﬁcation provides the embedding M (n. C) = {Z ∈ M (n. C) = {Z ∈ M (n. There holds a proposition similar to Theorem 2.8 (Complex General Linear Group). ien . C). . . If G is a closed subgroup of GL(n. Z T = X T − iY T . ie1 . ¯ (Here Z denotes the complex conjugate matrix of Z. C) | det Z = 0} = X Y −Y X ∈ M (2n. Example 2. C) ⊂ M (2n. . The complex special linear group is formed by all complex n × n unimodular matrices: SL(n.

TId GL(n) = M (n). we obtain ˙ X(0) = (xij (0)) = A ∈ M (n). det Z = 1} = X Y −Y X ∈ M (2n. Consequently. X T Y − Y T X = 0. X(0) = Id . 9 . Compute determinant of a unitary matrix: ¯ 1 = det(Z T Z) = det Z · det Z = | det Z|2 . det(X + iY ) = 1 . ϕ ∈ R. so det Z = eiϕ . The curve X(t) = Id +tA belongs to GL(n) for |t| < ε and ˙ small ε > 0. Example 2. C) | Z T Z = Id. X(0) = Id and X(0) = A.thus ¯ Z TZ ∼ = XT −Y T YT XT · X Y −Y X = Id 0 0 Id ∼ Id +i · 0. Moreover. X T Y − Y T X = 0 . R) | X T X + Y T Y = Id. X TX + Y T Y −Y T X + X T Y −X T Y + Y T X X TX + Y T Y That is why the realiﬁcation of the unitary group reads U(n) = X Y −Y X ∈ M (2n.12 (TId GL(n)). choose an arbitrary matrix A ∈ M (n). 2. ˙ Thus TId GL(n) ⊂ M (n). In order to show that this inclusion is in fact an equality. Another important example of a group formed by complex matrices is the special unitary group: ¯ SU(n) = U(n) ∩ SL(n. Consider the tangent space to the general linear group at the identity: ˙ TId GL(n) = X(0) | X(t) ∈ GL(n). R) | X T X + Y T Y = Id.11 (Special Unitary Group). C) = {Z ∈ M (n.2 The Lie algebra of a Lie group Example 2. ˙ We compute this space explicitly. Since the velocity vector X(t) = (xij (t)) is ˙ an n × n matrix.

We have ˙ sl(n) = TId SL(n) = {X(0) | X(t) ∈ SL(n). then ˙ ˙ 1 = det X(t) = det(Id +tX(0) + o(t)) = 1 + t tr X(0) + o(t). B] = 0.The tangent space TId GL(n) is a linear space. is called a Lie algebra. B] = AB − BA ∈ M (n). The tangent space to a Lie group G at the identity element is called the Lie algebra of the Lie group G: L = TId G. B]. A] + [[C. · ] which is: (1) bilinear. commutator of matrices: [A. (2) skew-symmetric. Moreover. (3) Jacobi identity: [[A. Notice that this operation satisﬁes the following properties: . (1) bilinearity. A linear space L endowed with a binary operation [ · . C]. Example 2. A]. this Lie algebra is denoted as gl(n). B]. ˙ Take a curve X(t) = Id +tX(0) + o(t) ∈ SL(n. A. (2) skew-symmetry: [B. We compute the Lie algebras of the Lie groups considered above. In order to underline the relation of this Lie algebra with the Lie group GL(n). Deﬁnition 2. A] = −[A. and (3) satisﬁes Jacobi identity. The Lie algebra of the special linear group is denoted by sl(n). X(0) = Id}. Such a construction has a generalization of fundamental importance. The space M (n) with the matrix commutator is a Lie algebra.13 (The Lie Algebra of SL(n)).4. Deﬁnition 2. 10 t → 0. C] + [[B. ˙ so tr X(0) = 0. Summing up.3. B ∈ M (n). it is endowed with an additional operation. TId GL(n) = gl(n).

we obtain A + AT = 0 and so(n) = {A ∈ M (n) | A + AT = 0}. Example 2. we proved only the inclusion ⊂ .17 (The Lie Algebra of T(n)). aﬀ(n) = TId Aﬀ(n) = A b 0 0 | A ∈ gl(n). and we proceed in the same way as for SO(n). Id Id 11 . X(0) = Id}. ¯ Z(0) = Id. the triangular matrices. t(n) = TId T(n) = {A = (aij ) ∈ M (n) | aij = 0. This Lie algebra is denoted as ˙ so(n) = TId SO(n) = {X(0) | X(t) ∈ SO(n). We have X(t)X T (t) ≡ Id. i > j}.15 (The Lie Algebra of Aﬀ(n)). e(n) = TId E(n) = A b 0 0 | A ∈ so(n). b ∈ Rn . Example 2. Finally compute the Lie algebras of the unitary and the special unitary groups.) Example 2. In a similar way one computes the Lie algebras in the following three cases.18 (The Lie Algebra of U(n)). For a curve Z(t) ∈ U(n).Thus the traceless matrices.14 (The Lie Algebra of SO(n)).16 (The Lie Algebra of E(n)). b ∈ Rn . (The same remark holds for similar computations in examples below. Example 2. thus ˙ ˙ ˙ ˙ 0 = X(0) X T (0) + X(0) X T (0) = X(0) + X T (0). This can be done by comparing dimensions of the linear spaces. we have Z(t)Z T (t) ≡ Id. u(n) = TId U(n). Thus ¯T ˙ ¯ ˙ 0 = Z(0) Z T (0) + Z(0) Z (0). the skew-symmetric matrices. The reverse inclusion we do not prove here for the sake of time and leave it to the reader as an exercise. To be precise. ˙ Denoting A = X(0). Example 2. Id Id sl(n) = {A ∈ M (n) | tr A = 0}.

2! n! This matrix series converges absolutely. In the case n = 1. A natural question on the possibility of the reverse passage is solved (for linear Lie groups) via matrix exponential. we obtain A + AT = 0. a skew-Hermitian matrix. C) | A + AT = 0}. 12 . ¯ su(n) = TId SU(n) = {A ∈ M (n. X(0) = Id. ea = 1 + a + a2 an +···+ +··· . 2! n! For arbitrary natural n. thus it can be diﬀerentiated termwise: eAt = A 2 t2 A n tn +···+ +··· 2! n! A2 t An tn−1 =A+ +···+ +··· 1! n! = eAt A. we considered the passage from a Lie group G to the corresponding linear object — the Lie algebra L of the Lie group G. C) | A + AT = 0. tr A = 0}. Consequently. solutions to the ODE x = xa ˙ are given by the exponential: x(t) = x(0)eat . Id +At + Thus the matrix exponential X(t) = eAt is the solution to the Cauchy problem ˙ X = XA. 2. and all solutions of the matrix equation (3) have the form X(t) = X(0)eAt .19 (The Lie Algebra of SU(n)). ¯ u(n) = {A ∈ M (n. Example 2. (3) In order to approach matrix control systems. Summing up.3 Matrix exponential ˙ X = XA. ﬁrst consider a matrix ODE: where A ∈ M (n) is a given matrix.˙ ¯ Denoting A = Z(0). we can proceed in a similar way and deﬁne for a matrix A ∈ M (n) its exponential by the same series: exp(A) = eA = Id +A + A2 An +···+ +··· .

e. What is the tangent space to a Lie group G at its point X ? This question has a simple answer given in the following statement. Y (0) = X −1 X = Id . as above. Let G be a linear Lie group. the vector ﬁeld XA is / not tangent to the Lie group SL(n). dt t=0 By uniqueness. if X(t) = exp(tA) ∈ SL(n). the Cauchy problem has a (unique) solution of the form X(t) = X0 exp(tA).4 Left-invariant vector ﬁelds ˙ X = XA. Indeed. So we constructed a (smooth) mapping exp : gl(n) → GL(n). Compute the tangent space ˙ TX G = {X(0) | X(t) ∈ G. Proof. then ODE (4) is not well-deﬁned. (4) Example 2. X(0) = X0 . Consider e. solutions of this ODE must be given.20. X ∈ GL(n). Proposition 2. i. It is obvious that in general the answer is negative. then d A= exp(tA) ∈ TId SL(n) = sl(n). one easily constructs a curve starting from the identity: Y (t) = X −1 X(t)..g. L its Lie algebra.1. its exponential exp(A) ∈ GL(n) since det exp(A) = exp(tr A) = 0. but the question is whether it is in the Lie group under consideration: X(t) = exp(tA) ∈ SL(n) ? So if A ∈ sl(n). X∈G? X ∈ SL(n). 2. 13 . For a smooth curve X(t) starting from X. What can we say about a similar problem in any (linear) Lie group G: ˙ X = XA. X(0) = X}. a Cauchy problem in the special linear group: ˙ X = XA. and let X ∈ G. X(0) = Id.Notice that for an arbitrary matrix A ∈ gl(n). We generalize this construction in the following subsection. We saw that for an arbitrary matrix A ∈ gl(n). by the matrix exponential. Then TX G = XTIdG = XL = {XA | A ∈ L}.

˙ X(0) = XA. 14 . Since these linear spaces have the same dimension. B ∈ L. V (X) = XA. A ∈ L.Then ˙ We denote A = Y (0) ∈ L and get ˙ ˙ Y (0) = X −1 X(0) ∈ L. Deﬁnition 2. ˙ X = XA. we obtain TX G = XL. then its left translation Y X(t) is also a trajectory of this ODE for any Y ∈ G. recall the deﬁnition of Lie bracket of vector ﬁelds. W ] is left-invariant as well. X ∈ G. X ∈ G. i. Vector ﬁelds of the form V (X) = XA. X ∈ G. In order to compute this ﬁeld. Thus TX G ⊂ XL. is well-deﬁned and has the solutions X(t) = X(0) exp(At) ∈ G. A ∈ L.. (5) are called left-invariant vector ﬁelds on the linear Lie group G. W (X) = XB. the vector V (X) = XA ∈ TX G. Indeed: X(t) = X(0) exp(tA). the vector ﬁeld V (X) is tangent to the Lie group G.e. So the ODE X ∈ G. thus Y (t) = Y X(t) = Y exp(tA)X(0) = Y (0) exp(tA). Notice the following important property of ODE (5): if a curve X(t) is a trajectory of the ﬁeld V (X) = XA. it is clear that the ﬁeld [V. There arises a natural question: what is the Lie bracket of such vector ﬁelds? Since the ﬁelds V and W are left-invariant. So the left product by X translates the tangent space L at identity to the tangent space XL at the point X. Suppose we have two left-invariant vector ﬁelds on a Lie group G: A.5. Thus for any element A ∈ L.

B] = X(AB − BA). etV t=0 (X) = X. W ](X) = [XA.6. Compute the low-order terms of the curve γ from Deﬁnition 2.6. t=0 By Deﬁnition 2.6: γ(t) = X exp(tA) exp(tB) exp(−tA) exp(−tB) t2 t2 2 A +··· Id +tB + B 2 + · · · 2 2 t2 2 t2 Id −tA + A − · · · Id −tB + B 2 − · · · 2 2 t2 2 = X Id +t(A + B) + (A + 2AB + B 2 ) + · · · 2 t2 Id −t(A + B) + (A2 + 2AB + B 2 ) + · · · 2 = X Id +tA + = X(Id +t2 [A. L its Lie algebra. dt and Vec M denotes the space of all smooth vector ﬁelds on a smooth manifold M . W ] ∈ Vec M such that √ d γ( t). Then [V. this is the Lie bracket [XA. notice that it is a smooth curve at t = 0. X ∈ G. The ﬂows of the left-invariant vector ﬁelds are given by the matrix exponential: etV (X) = X exp(tA). B] + · · · ). 15 . B] + · · · ). etW (X) = X exp(tB). XB]. The Lie bracket (or commutator ) of the ﬁelds V . W ](X) = X ∈ M. Now we compute the Lie bracket of left-invariant vector ﬁelds.Deﬁnition 2.2. W is the vector ﬁeld [V. B]. Proposition 2. Let V and W be smooth vector ﬁelds on a smooth manifold M . B ∈ L. Let V (X) = XA and W (X) = XB be left-invariant vector ﬁelds on G. and let A. thus √ γ( t) = X(Id +t[A. Proof. d dt √ γ( t) = X[A. d t t=0 where the curve γ is deﬁned as follows: γ(t) = e−tW ◦ e−tV ◦ etW ◦ etV (X). Here etV denotes the ﬂow of the vector ﬁeld V : d tV e (X) = V (etV (X)). Let G be a linear Lie group. XB] = X[A. [V.

2. Bm are some elements of L. and (2) L = {left-invariant vector ﬁelds on G}. X ∈ G. . If the control set U coincides with Rm . um ) ∈ U ⊂ Rm ..1 Left-invariant control systems Deﬁnitions Let G be a Lie group and L its Lie algebra. Remark. B1 . . . . . . 3 3.Corollary 2. Deﬁnition 3.e. . The isomorphism is deﬁned as follows: left-invariant vector ﬁeld XA ∈ Vec G ↔ A ∈ L. (8) Polysystem (6) can also be written in such classical notation via a choice of a parametrization of the set Γ. Left-invariant vector ﬁelds on a Lie group G form a Lie algebra isomorphic to the Lie algebra L = TId G. u = (u1 . . then system (7) is an aﬃne subspace of L. Throughout these notes.1. Deﬁnition 3.1 (Control-aﬃne left-invariant systems). . . control-aﬃne systems (7) are written as follows: m ˙ X = XA + i=1 ui XBi .e. i. as a set of vector ﬁelds. . which is important for applications is formed by controlaﬃne systems m Γ= A+ i=1 ui Bi | u = (u1 . . any subset Γ ⊂ L.. A trajectory of a left-invariant system Γ on G is a continuous curve X(t) in G deﬁned on an interval [t0 . i.1. um ) ∈ U. A left-invariant control system Γ on a Lie group G is an arbitrary set of left-invariant vector ﬁelds on G. In the classical notation. we write a left-invariant control system as (6) or (7). (7) where A. Thus in the sequel we identify these two representations of the Lie algebra of a Lie group G: (1) L = TId G. a polysystem. T ] ⊂ R so that there exists a partition t0 < t 1 < · · · < t N = T 16 . (6) Example 3. A particular class of left-invariant systems.

In the study of global controllability for inﬁnite time we can restrict ourselves by such a class of admissible controls. If there is no ambiguity. . T ) and A(X). X1 ∈ G. A system Γ ⊂ L is called controllable if. T ≥ 0} = T ≥0 AΓ (X.2 Right-invariant control systems ˙ Similarly to left-invariant vector ﬁelds X = XA. t). AN ∈ Γ such that the restriction of X(t) to each open interval (ti−1 . Although. X(0) = X. of all trajectories of Γ starting at X: AΓ (X) = {X(T ) | X(·) a trajectory of Γ. For any T ≥ 0 and any X in G.and left-invariant vector ﬁelds A1 . if A(X) = G for any X ∈ G. this notion corresponds to global controllability. . The reachable (or attainable) set of a system Γ from a point X ∈ G is the set AΓ (X) of all terminal points X(T ). . In the classical notation. ant vector ﬁelds of the form Y 17 . In the control literature.5 below. this corresponds to piecewise-constant admissible controls. one can consider right-invari˙ = BY . ti ) is diﬀerentiable and ˙ X(t) = X(t)Ai for t ∈ (ti−1 . T ) = {X(T ) | X(·) a trajectory of Γ. given any pair of points X0 and X1 in G. respectively.3. ≤ T ) = 0≤t≤T AΓ (X. T ). . or in other words.4. T ) and AΓ (X) by A(X. . Deﬁnition 3. T ) of all points that can be reached from X in exactly T units of time: AΓ (X. T ≥ 0. the reachable set for time T of a left-invariant system Γ ⊂ L from the point X is the set AΓ (X. X(0) = X}. for left-invariant systems these properties are equivalent to local controllability at the identity. i = 1. the point X1 can be reached from X0 along a trajectory of Γ for a nonnegative time: X1 ∈ A(X0 ) for any X0 . N. . in the sequel we denote the reachable sets AΓ (X. Deﬁnition 3. see Theorem 3. or complete controllability. The reachable set for time not greater than T ≥ 0 is deﬁned as AΓ (X. 3. . . ti ).

Let A ∈ L and X0 ∈ G. Consequently. i(X) = X −1 = Y. Notice that similarly to Proposition 2. Deﬁnition 3. (9) The inversion X = Y −1 transforms right-invariant system (9) to the left-invariant system m ˙ X = −XA − ui XBi .. 3. thus ˙ ˙ Y (t) = −Y (t)X(t)X −1 (t) = −Y (t)X(t)AY (t) = −AY (t). let X(t) ˙ be a trajectory of a left-invariant ODE X = XA. Deﬁnition 3. transforms left-invariant vector ﬁelds to right-invariant ones. then Y (t) = e−tA Y0 . we have Y (t)X(t) + Y (t)X(t) = 0.3 Basic properties of orbits and reachable sets Let G be a linear Lie group.5. Indeed. i=1 u ∈ U. Since X(t) = X0 etA . i. X(t0 ) = X0 . Then the Cauchy problem ˙ X = XA.e. Y = X −1 . all problems for right-invariant control systems are reduced to the study of left-invariant systems via inversion. and that the Lie algebra L = TId G of a Lie group G can be identiﬁed with the Lie algebra of right-invariant vector ﬁelds {AX | A ∈ L} on G. Lemma 3.1.1.6. ˙ X = XA ⇔ ˙ Y = −AY. u ∈ U ⊂ Rm . has the solution X(t) = X0 exp((t − t0 )A). Compute the ODE for −1 ˙ ˙ Y (t) = X (t). Summing up. Due to this obvious lemma we can obtain a description of an endpoint of a trajectory via product of exponentials. the space of left-invariant vector ﬁelds on G. X ∈ G. 18 . Y ∈ G. and let L be its Lie algebra. Since Y (t)X(t) = Id. it is easy to show that TX G = LX. A control-aﬃne right-invariant control system on a Lie group G has the form m ˙ Y = AY + i=1 ui Bi Y. A right-invariant control system on a Lie group G is an arbitrary set of right-invariant vector ﬁelds on G.The inversion i : G → G.

t ∈ (t1 . t1 ) Thus X(t) = X0 exp(A1 t). t ∈ [0. ⇒ ˙ X = X(t)A1 . such that X(T ) = X0 exp(τ1 A1 ) · · · exp(τN AN ). . Now we can obtain a description of attainable sets and derive their elementary properties. τ2 = t 2 − t 1 .3. We go on in such a way and ﬁnally obtain the required representation: X(tN ) = X(T ) = X0 exp(τ1 A1 ) · · · exp(τN AN ). T ]. AN ∈ Γ τ1 + · · · + τN = T. ⇒ ˙ X = X(t)A2 . τN = tN − tN −1 . A 1 . ti ) Consider the ﬁrst interval: t ∈ (0. τ1 = t 1 . . such that X(t) is continuous and t ∈ (ti−1 .Lemma 3. . and let X be an arbitrary point of G. be a trajectory of a left-invariant system Γ ⊂ L with X(0) = X0 . . . X(t1 ) = X0 exp(A1 t1 ). ⇒ ˙ X(t) = X(t)Ai . . t2 ) So X(t) = X0 exp(t1 A1 ) exp((t − t1 )A2 ). . . By the deﬁnition of a trajectory. there exist N ∈ N and 0 = t0 < t1 < · · · < tN = T. Further. Proof. . Then 19 . . . τN + · · · + τ1 = tN = T. X(0) = X0 . τN > 0.2. AN ∈ Γ X(t2 ) = X0 exp(A1 t1 ) exp((t2 − t1 )A2 ) = X0 exp(A1 τ1 ) exp(τ2 A2 ). Let Γ ⊂ L be a left-invariant system. X(t1 ) = X0 exp(A1 t1 ). . . Lemma 3. τ1 = t1 . . . Let X(t). A 1 . Then there exist N ∈ N and τ1 . τ2 = t2 − t1 .

Proof. 20 . ti ∈ R. (2) OΓ (Id) is the connected Lie subgroup of G with the Lie algebra Lie(Γ). ti > 0. one is allowed to move both forward and backward in time. Let Γ ⊂ L be a left-invariant system. Obviously. Items (1) and (2) follow immediately from Lemma 3. then XY ∈ OΓ (Id) as a product of exponentials. (2) First of all. Further. then for any X1 . The structure of orbits is simpler than that of attainable sets. In the orbit. the smallest Lie subalgebra of L containing Γ. (4) AΓ (X) is an arcwise-connected subset of G.(1) AΓ (X) = {X exp(t1 A1 ) · · · exp(tN AN ) | Ai ∈ Γ. Deﬁnition 3. and let X be an arbitrary point of G. the product X1 X2 ∈ AΓ (Id). the orbit OΓ (Id) is connected since any point in it is connected with the identity by a continuous curve provided by the deﬁnition of an orbit. Item (1) is obvious. Lemma 3. Then (1) OΓ (X) = XOΓ (Id). X2 ∈ AΓ (Id). Y ∈ OΓ (Id). i..4. it is easy to see that OΓ (Id) is a subgroup of G. (2) AΓ (X) = XAΓ (Id). then X −1 = exp(−tN AN ) · · · exp(−t1 A1 ) ∈ OΓ (Id). (4) Any point in AΓ (X) is connected with the initial point X by a trajectory X(t).e. If X = exp(t1 A1 ) · · · exp(tN AN ) ∈ OΓ (Id). AΓ (X) ⊂ OΓ (X). compare with the description of attainable set AΓ (X) given in item (1) of Lemma 3. N ≥ 0}. (3) Since AΓ (Id) = {exp(t1 A1 ) · · · exp(tN AN ) | Ai ∈ Γ. The orbit of a system Γ through a point X ∈ G is the following subset of the Lie group G: OΓ (X) = {X exp(t1 A1 ) · · · exp(tN AN ) | Ai ∈ Γ. ti > 0. Proof.3.2. N ≥ 0}. (3) AΓ (Id) is a subsemigroup of G. N ≥ 0}. while in the attainable set only the forward motion is allowed. Here and below we denote by Lie(Γ) the Lie algebra generated by Γ.7. If X.

[1]) that OΓ (Id) ⊂ G is a smooth submanifold with the tangent space TId OΓ (Id) = Lie(Γ). . Γ is controllable iﬀ AΓ (X) = G for any X ∈ G. AN in Γ and t ∈ RN with positive coordinates t1 . .4 Normal attainability If a point Y ∈ G is reachable (or attainable) from a point X ∈ G. A left-invariant system Γ is controllable iﬀ AΓ (Id) = G. . That is. we denote by span(l) the vector subspace of V generated by l and by co(l) the positive convex cone generated by the set l. tN such that the mapping F : RN → G. . Id ∈ OΓ (Id). . 3.Finally. . . Deﬁnition 3. tN ) = X exp(t1 A1 ) · · · exp(tN AN ) satisﬁes the following conditions: 21 . .8. Given any subset l of a vector space V . tN ) ∈ RN with positive coordinates such that Y = X exp(t1 A1 ) · · · exp(tN AN ). AN . Proposition 3. . That is why in the sequel we use the following short notation for the attainable set and orbit from the identity: AΓ (Id) = AΓ = A. . the point Y is in the image of the mapping F : (s1 . sN ) → X exp(s1 A1 ) · · · exp(sN AN ). We denote the topological closure and the interior of a set S by cl S and int S. . AN ∈ Γ and t = (t1 . si > 0. F (t1 . then there exist elements A1 . . It follows from the general Orbit Theorem (see [14]. Since AΓ (X) = XAΓ (Id). . The following stronger notion turns out to be important in the study of topological properties of reachable sets and controllability.1. . . . see [45]. (i) F (t) = Y . . Then the orbit OΓ (Id) is a Lie subgroup of G with the Lie algebra Lie(Γ). (ii) rank Dt F = dim G. . The point Y is said to be normally accessible from X by A1 . A point Y ∈ G is called normally attainable from a point X ∈ G by Γ if there exist elements A1 . respectively. . . . . controllability is equivalent to the identity AΓ (Id) = G. . . . By deﬁnition. . Proof. OΓ (Id) = OΓ = O. . .

We prove item (1) since items (2) and (3) follow from it. Then: (1) In any neighborhood V of the identity Id ∈ G. there are points normally attainable from Id by Γ. We have X1 = exp(t1 A1 ) for some t1 > 0. a contradiction. The mapping F is a local diﬀeomorphism near t. thus the set M2 = {F2 (s1 . s2 ) | 0 < si < εi . si > 0. 1 For small s > 0 we have rank Ds F2 = 2. 1 1 Consider the mapping F2 : (s1 . (2) Consequently. Id.5. s2 ) → exp((t1 + s1 )A1 ) exp(s2 A2 ). s1 =0 M1 = {F1 (s1 ) | 0 < s1 < ε1 } is a smooth one-dimensional manifold contained in the neighborhood V for suﬃciently small positive ε1 . sN ). the intersection int A ∩ V is (3) In particular. Proof. Denote n = dim L = dim Lie(Γ). the interior A is nonempty. Then the set F (V ) is open. for any s = (s1 . .Lemma 3. for any neighborhood V nonempty. Proof. Assume that n ≥ 1 and ﬁx a neighborhood V of the identity Id. Theorem 3. On the other hand. there exist an element A2 ∈ Γ and a point X1 ∈ M1 as close to identity as we wish such that X1 A 2 ∈ T X1 M 1 . for suﬃciently small positive ε1 . If n = 1. . i = 1. There exists a nonzero element A1 ∈ Γ.1 (Krener). 2} 22 0 < s i < εi . then Y ∈ int AΓ (X). . there exists a neighborhood V t such that the restriction F |V is a diﬀeomorphism. If n = 0. . We have rank Ds1 F1 = 1 for small s1 . the point F (s) is in AΓ (X). consequently. then any point X1 ∈ M1 is normally attainable from Id by A1 . otherwise dim Lie(Γ) = 0. If n > 1. 0 < s 1 < ε1 . / Otherwise Lie(Γ)(X1 ) ⊂ TX1 M1 for any X1 ∈ M1 and dim Lie(Γ) ≤ dim M1 = 1. If a point Y ∈ G is normally attainable from X ∈ G by Γ. . The curve d F1 d s1 = A1 = 0. everything is clear. That is. Consider the mapping F1 : s1 → exp(s1 A1 ). Let Lie(Γ) = L. Thus F (V ) ⊂ AΓ (X) is a neighborhood of the point Y = F (t).

any point Xn ∈ Mn = Im Fn is normally attainable from Id. Deﬁnition 3. In the case n = 3. 1 2 0 < s i < εi . 1 2 i Consider the mapping F3 : (s1 . the diﬀerential Ds F3 has rank 3 for small s > 0. we ﬁnd an element An ∈ Γ and a point n−1 n−1 Xn−1 = exp(t1 A1 ) · · · exp(tn−1 An−1 ) ∈ Mn−1 suﬃciently close to Id such that Xn−1 An ∈ TXn−1 Mn−1 . / Otherwise Lie(Γ)(X2 ) ⊂ TX2 M2 for any X2 ∈ M2 and dim Lie(Γ) ≤ 2.2. s3 ) | 0 < si < εi . Let Γ ⊂ L.9. / Then the mapping Fn : (s1 .is a smooth two-dimensional manifold that belongs to V for suﬃciently small positive ε1 and ε2 . since in this case any point of M2 is normally attainable from Id by A1 and A2 . Thus M3 = {F3 (s1 . s2 . 0 < s i < εi . we proceed in a similar manner. sn ) n−1 n−1 → exp((t1 + s1 )A1 ) · · · exp((tn−1 + sn−1 )An−1 ) exp(sn An ). There exist A3 ∈ Γ and X2 ∈ M2 close to Id such that X2 A 3 ∈ T X2 M 2 . 3} is a smooth three-dimensional manifold belonging to V for suﬃciently small positive εi . . Xn can be chosen as close to Id as we wish. s2 . Then: 23 . the theorem is proved. As a result of the inductive construction. s3 ) → exp((t2 +s1 )A1 ) exp((t2 +s2 )A2 ) exp(s3 A3 ). 2. Moreover. . If n > 2. . A system Γ ⊂ L is said to have a full rank (or to satisfy the Lie Algebra Rank Condition) if Lie(Γ) = L. otherwise we proceed by induction. Since the vector ﬁeld A3 is not tangent to the manifold M2 at the point X2 . the theorem is proved. Consequently. a contradiction. We have X2 = exp(t2 A1 ) exp(t2 A2 ) for some t2 > 0. i = 1. . is an immersion for small s > 0. If n = 2. Proposition 3.

Further.r. then Lie(Γ) = L. A ⊂ cl intO A. Theorem 3. (1) If Γ is controllable. and the relative interior w. Finally. then the Lie group G is connected. one can study controllability on SO(n). By Krener’s theorem. We prove item (2). Example 3. and Γ ⊂ L a left-invariant system on G. instead. then the orbit O ⊂ G is an open subset. where GL± (n) = {X ∈ M (n) | sign(det X) = ±1}. we consider the restriction of the system Γ to the orbit O.5 General controllability conditions Let G be a linear Lie group. (2) int A = ∅ if and only if Lie(Γ) = L. We denote by intO S the interior of a subset S of the orbit O in the topology of O. Proof. thus X ∈ cl int A. The Lie group GL(n) is not connected since it consists of two connected components GL+ (n) and GL− (n). Take any element X ∈ A and any its neighborhood V X.t. Theorem 3. (2) Moreover. Consequently.3 (Rank Condition). Then the open set V X −1 is a neighborhood of the identity. a Lie subgroup of G with the Lie algebra Lie(Γ). and item (1) of this proposition follows. Then the open set W X is a neighborhood of the point Y X.2.(1) intO A = ∅. W ⊂ A. 3. Thus there are no controllable systems on GL(n).2 (Connectedness Condition). Example 3. W X ⊂ A. By Krener’s theorem. The system Γ is full-rank on O. the orthogonal group O(n) = SO(n) ∪ O− (n) is disconnected. since Y ∈ int A. So there no controllable systems on O(n). thus the statement in the case (b) follows from the case (a). Let Γ ⊂ L. O coincides with the interior in G. Proof. moreover. Y X ∈ int A ∩ V . The attainable set A is a connected subset of G. 24 . Similarly. (a) Assume ﬁrst that the system Γ has full rank: Lie(Γ) = L. but a reasonable question to study is controllability on its connected component of identity GL+ (n). there exists a neighborhood W Y . In this subsection we prove some basic controllability conditions for Γ on G. int A = ∅. thus Y X ∈ V . where O− (n) = {X ∈ O(n) | det X = −1}. any neighborhood V of the point X contains points from int A. (b) If Lie(Γ) = L.3. there exists a point Y ∈ V X −1 ∩ int A. If Γ ⊂ L is controllable on G. L its Lie algebra.

then int A = ∅. and there may be two cases: either Id ∈ int A. A system Γ ⊂ L is controllable on a Lie group G iﬀ the following conditions hold: (1) G is connected. let Lie(Γ) = L. Thus int O = ∅. Theorem 3. the more so int A = ∅. But O ⊂ G is a connected Lie subgroup with Lie algebra Lie(Γ) = L. Ai ∈ Γ}. A system Γ ⊂ L is controllable on a Lie group G iﬀ the following conditions hold: (1) G is connected. Consider the powers of this neighborhood: V n ⊂ A for any n ∈ N. Recall the descriptions of the attainable set and orbit through identity: A = {exp(t1 A1 ) · · · exp(tN AN ) | ti ≥ 0. if Lie(Γ) = L. If A ⊂ G is a subgroup. Proof. Theorem 3. Deﬁnition 3. The necessity is obvious. thus it is generated by any neighborhood of identity [45]: V n = G. thus A = G. (3) the attainable set A is a subgroup of G. Now we prove Theorem 3. thus Lie(Γ) = L.5. Ai ∈ Γ}. A control system is called locally controllable at a point X if X ∈ int A(X). then A = G. then for any element X ∈ A. we prove suﬃciency. (2) By Krener’s theorem. There exists a neighborhood V Id such that V ⊂ A. Notice that identity element is always contained in the attainable set. Then dim O = dim Lie(Γ) < dim L = dim G.10. n∈N Then A ⊃ n∈N V n = G.4 (Group Test). or Id ∈ ∂A. The necessity is obvious. In the ﬁrst case the system is controllable. see [45]. (2) Lie(Γ) = L. while in the second case not. Thus A = O = G. Conversely. we prove suﬃciency. the more so O = G. (2) Γ is locally controllable at the identity. Then it follows that O = G. 25 .Proof. But the Lie group G is connected. Proof.5 (Local Controllability Test). the inverse (exp(ti Ai ))−1 = exp(−ti Ai ) ∈ A. O = {exp(±t1 A1 ) · · · exp(±tN AN ) | ti ≥ 0. its inverse X −1 belongs to A as well. For any exponential exp(ti Ai ) ∈ A. (1) If Γ is controllable. thus the attainable set A coincides with the orbit O.

it follows that int A−Γ = ∅. We have X ∈ V ⊂ A−Γ = A−1 . Then there exist sequences {Xn }. Then cl AΓ is a subsemigroup of G. by the hypothesis of this theorem. thus Γ the open set V −1 ⊂ AΓ is a neighborhood of the inverse X −1 .5. and the system Γ is controllable by Theorem 3. Proof. 4 4. Lemma 4. Let us prove the suﬃciency. It is easy to show that not only the attainable set A.Theorem 3. The previous theorem has important far-reaching consequences. Yn → Y as n → ∞. Trajectories of the system −Γ are trajectories of the initial system Γ passed in the opposite direction. A system Γ ⊂ L is controllable on a Lie group G iﬀ the following conditions hold: (1) Lie(Γ) = L. Let Γ1 . cl AΓ = G. Γ2 ⊂ L. but also its closure is a semigroup. thus A−Γ = {exp(−t1 A1 ) · · · exp(−tN AN ) | ti ≥ 0. Y ∈ cl AΓ . Ai ∈ Γ} = A−1 .1. the open set V −1 X ⊂ AΓ . Let X.6 (Closure Test). (2) cl A = G.1 Extension techniques for left-invariant systems Saturate Deﬁnition 4. Consider the time-reversed system −Γ = {−A | A ∈ Γ}. thus Id ∈ int AΓ . 26 . Ai ∈ Γ Since Lie(−Γ) = Lie(Γ) = L. thus there exists a point X ∈ AΓ ∩ V = ∅. The system Γ1 is called equivalent to the system Γ2 : Γ1 ∼ Γ2 if cl AΓ1 = cl AΓ2 . Consequently. Let Γ ⊂ L. But Id = X −1 X ∈ V −1 X ⊂ AΓ . {Yn } ⊂ AΓ such that Xn → X. Further. thus there exists an open subset V ⊂ A−Γ . The necessity is straightforward. It means that in the study of controllability of full-rank systems one can replace the attainable set A by its closure cl A. Proof. This idea gives rise to the powerful extension techniques described in the following section.1. Γ = (exp(tN AN ) · · · exp(t1 A1 ))−1 | ti ≥ 0.

If Γ1 ∼ Γ and Γ2 ∼ Γ. The saturate of a right-invariant system Γ ⊂ L is the following system: Sat(Γ) = ∪{Γ ⊂ L | Γ ∼ Γ}. (2) Sat(Γ) = {A ∈ L | exp(R+ A) ⊂ cl AΓ }. Proposition 4. We have exp(ti Ai ) ∈ AΓ1 ∪ AΓ2 ⊂ cl AΓ . (1) Sat(Γ) ∼ Γ. Take an arbitrary element X = exp(t1 A1 ) · · · exp(tN AN ) ∈ AΓ1 ∪Γ2 . In order to prove the inclusion ASat(Γ) ⊂ cl AΓ . The previous lemma allows one to unite equivalent systems. It is then natural to consider the union of all systems equivalent to a given one. Ai ∈ Sat(Γ). The inclusion cl AΓ ⊂ cl AΓ1 ∪Γ2 is obvious in view of the the chain cl AΓ = cl AΓ1 ⊂ cl AΓ1 ∪Γ2 . (12) take any element X = exp(t1 A1 ) · · · exp(tN AN ) ∈ ASat(Γ) . (11) (10) Each element Ai is contained in a system Γi ∼ Γ. Inclusion (12) and item (1) follow. and cl AΓ1 ∪Γ2 ⊂ cl AΓ . Proof.1 it follows that X ∈ cl AΓ . so cl AΓ ⊂ cl ASat(Γ) . then Γ1 ∪ Γ2 ∼ Γ. Item (1) means that the saturate of Γ is the largest right-invariant system on G equivalent to Γ. thus Γ ⊂ Sat(Γ). Proof. cl AΓ X. while item (2) describes Sat(Γ) as a kind of a tangent object to cl AΓ at the identity. thus exp(ti Ai ) ∈ AΓi ⊂ cl AΓ . So inclusion (11) is proved.2. (1) Obviously.Then {Xn Yn } ⊂ AΓ and Xn Yn → XY as n → ∞.1. A i ∈ Γ1 ∪ Γ2 . By the semigroup property. Γ ∼ Γ. Lemma 4.2. it follows that Γ1 ∪ Γ2 ∼ Γ. In view of inclusion (10). We have cl AΓ1 = cl AΓ2 = cl AΓ . Now we prove the inclusion AΓ1 ∪Γ2 ⊂ cl AΓ . 27 . ti > 0. thus by Lemma 4. ti ≥ 0. Deﬁnition 4.

Remark. cl A = T2 . and nevertheless Γ is not controllable. A = (1. it follows that AΓ ⊂ cl AΓ . mod 1)}. The reason is clear — the rank condition is violated: Lie(Γ) = RA = L. r ∈ R \ Q. Inclusion (13) is proved. Consider the following right-invariant system on G: Γ = {A}. 28 Γ ∼ L = Sat(Γ). Then A ∈ Γ ∼ Γ. i.(2) Denote the system Γ = {A ∈ L | exp(R+ A) ⊂ cl AΓ }. The torus is a twodimensional Abelian Lie group: G = T2 = S 1 × S 1 = {(x mod 1. Moreover. y Its Lie algebra is L = TId T2 = R2 .1 (Irrational winding of the torus). the saturate is not the appropriate tangent object to cl A responsible for controllability: it is possible that Sat(Γ) = L. We show that Γ ∼ Γ. Consider the representation AΓ = {exp(t1 A1 ) · · · exp(tN AN ) | ti > 0. The attainable set is the irrational winding of the torus: A = exp(R+ A) = {(x mod 1. Taking into account inclusion (13). . Example 4. we obtain the required equality: Sat(Γ) = Γ. First we prove the inclusion Γ ⊂ Sat(Γ). Ai ∈ Γ}..e. Since all exp(ti Ai ) ∈ cl AΓ . Inclusion (14) follows. since Γ ⊂ Γ. kx mod 1) | x ≥ 0} = T2 . Thus although Γ is not controllable on T2 . (14) (13) take any element A ∈ Sat(Γ). Thus exp(tA) ∈ cl AΓ . Unfortunately. r). hence Γ ∼ Γ. then AΓ ⊂ AΓ . In order to prove the reverse inclusion Sat(Γ) ⊂ Γ. A ∈ Γ. Thus cl AΓ = cl AΓ .

Corollary 4. B] ∈ LS(Γ). ∀ α ∈ [0. ±B ∈ LS(Γ).2 Lie saturate of invariant system It is the following Lie-generated tangent object to cl AΓ that is responsible for controllability of Γ. ∀ α ≥ 0. Theorem 4.1 (Lie Saturate Test). (1b) LS(Γ) is convex: A. B ∈ LS(Γ) and any s ∈ R. (1) LS(Γ) is a closed convex positive cone in L. Necessity follows from the deﬁnition of the Lie saturate. t ≥ 0} as a semigroup..2.e. i. Since. Assume that LS(Γ) = L. β ≥ 0. LS(Γ) = {A ∈ Lie(Γ) | exp(R+ A) ⊂ cl AΓ }. Theorem 4. Proof. Deﬁnition 4. exp(s ad A)B = B + (s ad A)B + ∈ LS(Γ).6. 1].1. Suﬃciency. The following description of Lie Saturate follows immediately from Proposition 4.3. thus the equality Sat(Γ) = L implies that cl(A) = G. then ±[A. B ∈ LS(Γ) ⇒ αA + βB ∈ LS(Γ) (2) For any ±A. Lie saturate of a left-invariant system is deﬁned as follows: LS(Γ) = Lie(Γ) ∩ Sat(Γ). The basic properties of Lie saturate are collected in the following proposition. (3) If ±A. in addition. A left-invariant system Γ ⊂ L is controllable on a connected Lie group G if and only if LS(Γ) = L. The connected Lie group G is generated by the one-parameter semigroups {exp(tA) | A ∈ L. A... . + B + ··· 2! n! ∀ α. the rank condition Lie(Γ) = L holds. then Γ is controllable by Theorem 3. B ∈ LS(Γ) ⇒ αA + (1 − α)B ∈ LS(Γ) (1c) LS(Γ) is a positive cone: A ∈ LS(Γ) ⇒ αA ∈ LS(Γ) Thus. 29 (s ad A)2 (s ad A)n B +. (1a) LS(Γ) is topologically closed: cl(LS(Γ)) = LS(Γ).1.4.

Then exp(tαA) ∈ cl A. Consequently. we have An ∈ Lie(Γ) cl A ⇒ (15) A ∈ Lie(Γ). d Bs = [A. We denote by ad A the adjoint operator corresponding to A ∈ L: ad A : L → L. Since the linear Proof. Further. (1a) Take a converging sequence LS(Γ) space Lie(Γ) is closed. B ∈ LS(Γ). (5) Moreover. β = 1 − α. B]. It is easy to see (exercise) that exp(tC) = lim exp n→∞ α tA exp n β tB n · · · exp n pairs α tA exp n β tB . (1c) It is easy to show that LS(Γ) is a cone. the both curves (16) and (17) are solutions to the Cauchy problem B0 = B. and Sat(Γ) is convex. (16) It is easy to see that this element admits the following representation: Bs = exp(sA)B exp(−sA). ds Further. s ∈ R. it is obvious from (16) that Bs ∈ Lie(Γ). t ≥ 0.. 1].e. i. then −A ∈ LS(Γ). An → A ∈ L. Denote the element Bs = exp(s ad A)B.(4) If A ∈ LS(Γ) and if the one-parameter subgroup {exp(tA) | t ∈ R} is periodic (i. LS(Γ) = Lie(Γ) ∩ Sat(Γ) is topologically closed. Take any A ∈ LS(Γ). ad A : B → [A. αA ∈ LS(Γ). Since the linear space Lie(Γ) is convex. t ≥ 0. it follows that Sat(Γ) is closed as well: since An ∈ Sat(Γ). and A ∈ Sat(Γ). Representation (17) implies that exp(tBs ) = exp(tA) exp(sB) exp(−tA) ∈ cl(AΓ ) 30 . Bs ]. (17) Indeed. then −A ∈ LS(Γ). we have exp(tAn ) → exp(tA) ∈ cl A. assume that ±A. and consider the convex combination C = αA + βB. (1b) Take any A. n thus C ∈ Sat(Γ). α ∈ [0. it follows that LS(Γ) is convex as well. To prove (2).e. if A ∈ LS(Γ) and if the one-parameter subgroup {exp(tA) | t ∈ R} is quasi-periodic: exp(R− A) ⊂ cl exp(R+ A). α > 0. B ∈ LS(Γ). compact).

5) given A ∈ Γα with periodic or quasi-periodic one-parameter group. 31 u ∈ R. s ∈ R. Γα ⊂ Γβ if α < β. thus −A ∈ LS(Γ). which is valid for all A ∈ LS(Γ) with a periodic one-parameter group.2 are applied as suﬃcient conditions of controllability via the following procedure.e. then LS(Γ) = L. Starting from a given system Γ. (18) . one constructs Γβ = Γα ∪ eR ad A B. and the system Γ is controllable by Theorem 4. If one obtains the relation Γα = L at some step α.2: (1) given Γα . Now (3) easily follows: if ±A. one constructs Γβ = cl(co(Γα )). That is why Theorems 4. ±B ∈ LS(Γ). B]. it is diﬃcult to construct the Lie saturate of a right-invariant system explicitly. then ±et ad A B. It follows from the quasiperiodic property (15) that exp(−tA) = exp(t(−A)) ∈ cl exp(R+ A) ⊂ cl AΓ for any t ≥ 0. Consider the following right-invariant system on GL+ (n): Γ = A + RB ⊂ gl(n). 5 Induced systems on homogeneous spaces Example 5. The extension rules are provided by Theorem 4. Finally.1 (Bilinear systems). (3) for ±A. (4.for any t ≥ 0.2 guarantees that all extensions Γα belong to LS(Γ). (2) for ±A..1. Γ0 = Γ. we prove a more strong property (5). one constructs a completely ordered ascending family of extensions {Γα } of Γ. B] = ± lim ∈ LS(Γ). X ∈ GL+ (n). i. In the classical notation. t→0 t (4) follows from the chain {exp(tA) | t ≥ 0} = {exp(tA) | t ∈ R} ⊂ cl AΓ . B ∈ Γα . that is why et ad A B − B ±[A. ±B ∈ LS(Γ). thus Bs ∈ LS(Γ) for all s ∈ R.1 and 4. Theorem 4. Usually. one constructs Γβ = Γα ∪ R[A. ±B ∈ Γα . one constructs Γβ = Γα ∪ RA. this system reads ˙ X = AX + uBX.

If X(t) is a trajectory of the right-invariant system (18) with X(0) = Id.2.1. Then the trajectory x(t) = X(t)x0 of the bilinear system steers x0 to x1 : x(0) = X(0)x0 = Id x0 = x0 . There exists a matrix X1 ∈ GL+ (n) such that X1 x0 = x1 . Y ∈ G and any x ∈ M . Then it is easy to see that the bilinear system (19) is controllable on Rn \ {0}. (1) The Lie group G = GL+ (n) acts on the manifold M = Rn \ {0}. then X(t)x is a trajectory of (19). (2) G acts transitively on M : ∀ x0 . θ(X. x0 ) = x1 . X : x → Xx. ˙ x ∈ Rn \ {0}. We showed that if the right-invariant system (18) is controllable on GL+ (n). x1 ∈ M there exists X ∈ G such that θ(X. Deﬁnition 5. any X ∈ G deﬁnes a mapping X : M → M. (3) The bilinear system (19) is induced by the right-invariant system (18): if X(t) is a trajectory of (18). (19) We exclude the origin from Rn since linear vector ﬁelds vanish at the origin. x1 ∈ Rn \ {0}. X(T ) = X1 for some T ≥ 0. A Lie group G is said to act on a smooth manifold M if there exists a smooth mapping θ : G×M →M that satisﬁes the following conditions: (1) θ(Y X. This construction generalizes as follows. x) = x for any x ∈ M . then the bilinear system (19) is controllable on Rn \ {0}. Indeed. (2) θ(Id. x) = θ(Y. 32 . u ∈ R. then the curve x(t) = X(t)x0 is a trajectory of the bilinear system (19) with x(0) = x0 . By virtue of controllability of Γ. take any two points x0 . x(T ) = X(T )x0 = X1 x0 = x1 . x1 ∈ M ∃ X ∈ G such that Xx0 = x1 . there exists a trajectory X(t) of the right-invariant system such that X(0) = Id. A manifold that admits a transitive action of a Lie group is called the homogeneous space of this Lie group. A Lie group G acts transitively on M if for any x0 . Deﬁnition 5.Introduce also the following bilinear system: x = Ax + uBx. There were three key points in the preceding argument. x)) for any X. thus it is an equilibrium for bilinear systems. that is. Assume that the right-invariant system (18) is controllable on GL+ (n).

x ∈ Rn \ {0}. X ∈ G.3. then x(t) = θ(X(t). (θ∗ Γ)(x) = {(θ∗ A)(x) | A ∈ Γ}. We have θ∗ (AX) = Ax. So the induced system θ∗ Γ is bilinear: x = Ax + uBx. In the classical notation. ˙ x ∈ Rn \ {0}. thus (θ∗ V )(x) = d dt θ(eV t (Id). Lemma 5. u ∈ R.1.Deﬁnition 5. x) = t=0 d dt exp(At)x = Ax. x0 ) is a trajectory of the induced system θ∗ Γ for any x0 ∈ M . its ﬂow through the identity is eV t (Id) = exp(At). A ∈ Γ. The system θ∗ Γ ⊂ Vec M.3. u ∈ R. The Lie group GL+ (n) acts transitively on Rn \ {0} as follows: θ(X. Γ reads as ˙ X = AX + uBX. Example 5. x). θ(exp(At)X0 . θ∗ (BX) = Bx. If X(t) is a trajectory of a right-invariant system Γ. x0 ). Let A ∈ L. x) = Xx. is called the induced system on M . Then X(t) = exp(At)X0 and x(t) = θ(exp(At)X0 . x ∈ M. x0 ) dt d ε ε=0 d = θ(exp(Aε). since an arbitrary trajectory of Γ is a concatenation of such pieces. Example 5. Let Γ ⊂ L be a right-invariant system. We can consider the case where the whole trajectory X(t) satisﬁes a ˙ single ODE X = AX(t). t=0 Deﬁnition 5. The vector ﬁeld θ∗ A ∈ Vec M induced by the action θ is deﬁned as follows: (θ∗ A)(x) = d dt θ(exp(tA). For a right-invariant vector ﬁeld V (X) = AX.2. x0 ) = θ(exp(A(t + ε))X0 . Then the required ODE is veriﬁed by diﬀerentiation: x(t) = ˙ d d θ(exp(At)X0 . thus θ∗ (AX + uBX) = Ax + uBx. Proof. X ∈ GL+ (n). x0 )) d ε ε=0 x(t) = (θ∗ A)(x(t)). 33 .4. Let Γ = {A + uB | u ∈ R} ⊂ L be a right-invariant system on a linear Lie group G ⊂ GL(n). t=0 x ∈ M.

and let θ∗ Γ ⊂ Vec M be the induced system on M . then θ∗ Γ is controllable on M . The transitivity of action of AΓ on M means that there exists X ∈ AΓ such that θ(X. x) = Xx. 9. x(T ) = θ(X. (1) If Γ is controllable on G. aﬃne systems. 34 . the induced systems are bilinear. then the bilinear system θ∗ Γ is controllable on Rn \ {0}. Similarly. m Γ= A+ i=1 ui B i m ⊂ sl(n). see [6. in the case A = SO(n) × R+ Id the bilinear system θ∗ Γ remains controllable. x ∈ Rn \ {0}. Example 5. Then the curve x(t) = θ(X(t). M = Rn \ {0}). Item (1) follows from (2). x) = Xx. Example 5. In this case. the inclusion X ∈ AΓ means that some trajectory X(t) of Γ steers Id to X: X(0) = Id. x ∈ Rn \ {0}. R) on the vector space Rn . x0 ) = x0 . x0 ) = x1 . ui Bi x. Let θ be a transitive action of a Lie group G on a manifold M . 38. T ≥ 0. X(T ) = X. x0 ) = x1 . x0 ) is a trajectory of θ∗ Γ that steers x0 to x1 : x(0) = θ(Id. 7. 8. We have θ(X. Important applications of Theorem 5. m Γ= A+ i=1 ui B i m ⊂ gl(n). θ(X. Proof.Theorem 5. M = Rn \ {0}). (2) Moreover. Further.4 (G = GL+ (R). let Γ ⊂ L be a right-invariant system on G.5 (G = SL(n). then the bilinear system θ∗ Γ is controllable on Rn \ {0}. ui Bi x.1 are related to the linear action of linear groups G ⊂ GL(n. The attainable set may be even less. if the semigroup AΓ acts transitively on M . Take any points x0 . for example. x1 ∈ M .1. Remark. so we prove (2). or more generally. θ∗ Γ : x = Ax + ˙ i=1 If A = GL+ (n) or A = SL(n). θ∗ Γ : x = Ax + ˙ i=1 If A is transitive on Rn \{0}. Linear groups acting transitively on Rn \ {0} or S n are described. then θ∗ Γ is controllable on M . 27].

The connected component of identity in the aﬃne group Aﬀ + (n) = acts transitively on the space M = Rn = as follows: θ X 0 y 1 .7 (G = U(n) or SU(n). x) = Xx. M = S 2n−1 ). ui Bi x. θ∗ Γ : x = Ax + ˙ i=1 Example 5. M = Rn ). x ∈ Rn . θ∗ Γ : z = Az + ˙ i=1 Example 5.8 (G = Aﬀ + (n). M = S n−1 ). z ∈ S 2n−1 . x 1 ⊂ Rn+1 X 0 y 1 ⊂ GL(n + 1) Consider a right-invariant system on G: m Γ= C0 + i=1 ui C i . and the induced system reads m θ∗ Γ : x = A 0 x + b 0 + ˙ i=1 ui (Ai x + bi ).6 (G = SO(n). x 1 = X 0 y 1 x 1 = Xx + y. θ(X. Ci = Ai 0 bi 0 ∈ aﬀ(n). ui Bi z. m Γ= A+ i=1 ui B i m ⊂ so(n).Example 5. 35 . x ∈ S n−1 . m Γ= A+ i=1 ui B i m ⊂ u(n) or su(n). The induced vector ﬁelds are aﬃne: θ∗ A b 0 0 x 1 = Ax + b. θ(Z. z) = Zz.

. Proof. bm . .In particular.1 Controllability conditions for special classes of systems and Lie groups Symmetric systems We return to the exposition for left-invariant systems Γ ⊂ L on a Lie group G. this system contains also the sign-opposite element −A. .e. . Let Γ = −Γ. But all −Ai ∈ Γ. thus A = O. we obtain the linear system m x = A0 x + ˙ i=1 ui b i . A0 b1 . the motion in the opposite direction −A is also admissible. A0 b1 . Example 5. (20) Exercise 5. then the linear system (20) is controllable on Rn . . together with any element A. In this section we developed a theory of induced systems for right-invariant systems because of the important class of bilinear systems x = Ax+uBx. M = Rn ). 36 . A system Γ ⊂ L is called symmetric if Γ = −Γ. A0 bm ) = Rn . where y ∈ Rn \ {0} is a row vector. Lemma 6. . x ∈ Rn . . for b0 = 0. ˙ 6 6. . Given a symmetric system. ui ∈ R. where ˙ x ∈ Rn \ {0} is a column vector. Thus the study of controllability for symmetric Γ is reduced to the veriﬁcation of the rank condition. . in this case the induced systems read y = yA + uyB. .1. A symmetric left-invariant system Γ ⊂ L is controllable on a connected Lie group G if and only if Lie(Γ) = L. Ai ∈ Γ} . Deﬁnition 6. . A0 bm .1. A1 = · · · = Am = 0. . the theory of induced systems for left-invariant systems is quite the same. Obviously.1. . i. This case is completely similar to the case of Aﬀ + (n).1.9 (G = E(n). Show that if the Kalman condition holds: n−1 n−1 span(b1 . . . Then A = O.. for any admissible direction of motion A. We have O = {exp(±t1 A1 ) · · · exp(±tN AN ) | ti > 0. . Theorem 6.

The necessity is a general fact. Denote X = exp(tA) for an arbitrary ﬁxed t > 0. We have to prove that exp(−tA) = X −1 ∈ cl exp(R+ A). . Proof. . For example. um ) ∈ U ⊂ Rm is symmetric if the set of control parameters U is symmetric with respect to the origin: U = −U . ui ∈ R. thus X −1 ∈ cl exp(R+ A). U(n). then the bilinear system (21) is controllable on Rn \ {0} (respectively on S n−1 ). . Am ) ⊂ L. Am ) = L. . A control-linear system m Γ= i=1 ui Ai | u = (u1 . Example 6. Then X nk+1 −nk −1 = X nk+1 X −nk X −1 → Y Y −1 X −1 = X −1 as k → ∞. and let G ⊂ GL(n) be the connected Lie subgroup corresponding to the Lie algebra L. .2 (Symmetric bilinear system). if U = Rm : Γ = span(A1 . The following simple fact is crucial for the controllability problem on compact Lie groups. the Lie groups SO(n). Suﬃciency follows since for a full-rank system on a connected Lie group the orbit coincides with the whole Lie group. Am ∈ gl(n). and let A belong to the Lie algebra L. Then the one-parameter subgroup exp(RA) is quasi-periodic: exp(R− A) ⊂ cl exp(R+ A). . in particular. . we consider the case of a Lie group that is compact as a topological space. SU(n) are compact and connected. . Am ) = L. Consider the corresponding symmetric bilinear system: m x= ˙ i=1 ui Ai x. Such a system is controllable on a connected Lie group G iﬀ Lie(A1 . If G acts transitively on Rn \ {0} (or S n−1 ). . . . . . 37 .Proof. . x ∈ Rn \ {0}. The sequence {X n }. Let A1 . n ∈ N. But nk+1 − nk − 1 ≥ 0. . 6.2 Compact Lie groups In this section. Let a Lie group G be compact. (21) Denote Lie(A1 . . . has a converging subsequence in the compact Lie group G: X nk → Y ∈ G as k → ∞. . nk+1 > nk . Example 6. .1 (Control-linear systems). . Lemma 6.2.

1. Thus. α. n−2 Take the matrices A1 = i=1 (Ei. ˙ X = (A1 + uA2 )X. tj ≥ 0. A2 ] span the whole Lie algebra so(3). A2 ) = L is open and dense in L × L (this is valid for any semisimple Lie algebra L. Proof. the system Γ is controllable. in which only one control is involved. we can replace the matrices A1 and A2 by an “almost arbitrary” pair in L × L.i+1 − Ei+1. since LS(Γ) is a cone. It is easy to show that Lie(A1 . 38 x ∈ S 2.2. u ∈ U ⊂ R.1. If A. u∈U . Notice that the set of pairs (A1 . and [A1 . X ∈ SO(n). Consequently. then ±A.Corollary 6. Thus.4 (SO(n)). ±B ∈ LS(Γ) by Lemma 6. Proof. is controllable (if the set of control parameters U contains at least two distinct points). Take any linearly independent matrices A1 . We show that LS(Γ) is a Lie algebra.n−1 . u ∈ U ⊂ R.n − En. Apply Corollary 6.2. we conclude that LS(Γ) = Lie(Γ). Moreover ±[A. ij ∈ {1.3 (SO(3)). Theorem 6. Its Lie algebra L = so(3) is the set of all 3 × 3 real skew-symmetric matrices.2. A2 ∈ so(3) and consider the right-invariant system Γ = {A1 . A2 ) such that Lie(A1 . A2 }. In this case. and let Γ ⊂ L. A2 ) = so(n). A2 . The Lie group G is compact and connected. We denote by Eij the n × n matrix with the only identity in row i and column j. ˙ is controllable on the sphere S 2 . the Lie algebra L of G is the set of all n × n skew-symmetric matrices so(n). Let G = SO(3). Then LS(Γ) = Lie(Γ). Taking into account the chain Γ ⊂ LS(Γ) ⊂ Lie(Γ). Thus αA + βB ∈ LS(Γ). even though the group 1 SO(n) is 2 n(n − 1)-dimensional. and all other zero entries.i ) and A2 = En−1. Let G be compact. Example 6. β ∈ R. It follows that LS(Γ) is a Lie subalgebra of L. any rotation in SO(3) can be written as the product of exponentials exp(t1 Ai1 ) · · · exp(tN AiN ). A right-invariant system Γ ⊂ L is controllable on a compact connected Lie group G if and only if Lie(Γ) = L. 2}. the set of all 3 × 3 real orthogonal matrices with positive determinant. That is. By Theorem 6. the induced bilinear system x = A1 x + uA2 x. see [42]). The previous considerations are generalized to the group G = SO(n) of rotations of Rn . N ∈ N. the system ˙ X = (A1 + uA2 )X. B] ∈ LS(Γ). B ∈ LS(Γ). Notice that the matrices A1 . X ∈ SO(3) (22) The single-input right-invariant aﬃne in control system (23) is also controllable (for any control set U containing more than one element). Example 6.

. z ∈ S3. the induced bilinear system z = A1 z + uA2 z. Deﬁnition 6. Let G = SL(2) and Γ = A + RB ⊂ sl(2). For any linearly independent A1 . So the system Γ = {A1 + uA2 | u ∈ U } (where U contains more than one element) is controllable on G = SU(2). A2 ). A Lie group G is called simple (resp. We start from an example of a control system that has the full rank and is not controllable. B= Let us show ﬁrst that Lie(A. A semisimple Lie algebra is a direct sum of its simple ideals.4.3.. For the Lie group G = SU(2). / thus Lie(A1 .Example 6.. .6. Consequently. B) = L = sl(2). A subspace I ⊂ L is called an ideal of a Lie algebra L if Deﬁnition 6. Deﬁnition 6. . are simple. u∈R 6.5. semisimple) if its Lie algebra L is simple (resp. b = 0. L] ⊂ I. n = 4. Deﬁnition 6. A2 ) = L. A Lie algebra L is called simple if it is not Abelian and contains no proper (i. Here A and B are traceless matrices of the form A = (aij ). For the controllability problem.e. ˙ is controllable on the sphere S 3 . The Lie groups SL(n) and SU(n) are simple. distinct from {0} and L) ideals. 0 1 −1 0 . we have [A1 . its Lie algebra can be represented as follows: L = su(2) = span i 0 0 −i . the Lie groups SO(n). A2 ∈ L.5 (SU(2)). A2 ] ∈ span(A1 .3 Semisimple Lie groups [I.2. A Lie algebra L is called semisimple if it contains no nonzero Abelian ideals. semisimple). 0 i i 0 . Example 6. we are interested in the case of SL(n) since the other two groups are compact and for them controllability is equivalent to the rank condition. a21 > 0. 39 (24) −b 0 0 b a12 > 0. while SO(4) is semisimple.

Now we prove Theorem 6. where A3 is a k×k matrix with 0 < k < n. Then the system Γ is controllable on the group SL(n). hence the right-invariant system Γ is not controllable on the Lie group SL(2).e. . (3) B = diag(b1 . B]) = sl(2). There are no controllability tests in this case. [A. Consequently. we have to obtain just one element in Lie(A. . the whole machinery of the Lie saturation on Lie groups was developed primarily for the study of controllability on SL(n).3. (25) It is easy to see that the ﬁeld Bx is tangent to the axes of coordinates {x1 = 0} and {x2 = 0}. but there are good suﬃcient conditions for controllability on SL(n). Thus the induced system θ∗ Γ is not controllable on the homogeneous space R2 \ {0}. Equality (24) follows. In fact. 40 . Let G = SL(n) and Γ = A + RB ⊂ sl(n). (5) bi − bj = bk − bm for (i. Compute the commutator: [A. In order to show that Γ is noncontrollable on SL(2). B. Permutation-irreducible matrices are matrices having no nontrivial invariant coordinate subspaces. we prove that the bilinear system θ∗ Γ is noncontrollable on R2 \ {0}. Theorem 6. u ∈ R. m). the system Γ is full-rank. . B] = 2b now it is obvious that span(A.3 (in the case n = 2 only: in the general case the proof is longer but uses essentially the same ideas [12]). Suppose that the matrices A = (aij ) and B satisfy the conditions: (1) a1n an1 < 0. On the other hand.Since dim sl(2) = 3. The controllability problem on SL(n) is much harder than the one on compact Lie groups. An n × n matrix A is called permutation-reducible if there exists a permutation matrix P such that P −1 AP = A1 0 A2 A3 . ˙ x ∈ R2 \ {0}. (2) the matrix A is permutation-irreducible. B) linearly independent of A and B. 0 −a21 a12 0 . . the ﬁeld Ax is directed inside the ﬁrst quadrant R2 = {x1 ≥ 0. (4) b1 < b2 < · · · < bn . The induced system reads x = Ax + uBx. bn ).. x2 ≥ 0} on its boundary. j) = (k. An n×n matrix is called permutationirreducible if it is not permutation-reducible. R2 is + + an invariant set of the bilinear system (25). i.

0 0 2b 0 . E12 . |u| A. First of all. a21 < 0. 0 exp(2bt) . in the basis (26). t ∈ R. (ad B)(E11 − E22 ) = 0. −b 0 0 b Without loss of generality. LS(Γ) thus That is why A + uB →u→±∞ ±B ∈ LS(Γ). a12 a21 < 0. (26) Compute the matrix of the adjoint operator ad B : sl(2) → sl(2).Proof. in the case of opposite signs the proof is the same. a21 exp(2bt)a21 41 and its exponential is easily computed: 1 0 exp(t ad B) = 0 exp(−2bt) 0 0 Thus the adjoint operator has the diagonal matrix 0 0 0 ad B = 0 −2b 0 . We have B = −b(E11 − E22 ). (ad B)E21 = 2bE21 . At = exp(t ad B)A ∈ LS(Γ). we can assume that a12 > 0. At = exp(t ad B)A = exp(−2bt)a12 ∈ LS(Γ). Further. E21 ). In the case n = 2 we have: A= B= a11 a21 a12 a22 . We show that LS(Γ) = sl(2) = span(E22 − E11 . . ±B ∈ LS(Γ). (ad B)E12 = −2bE12 . in the basis (26) a11 a11 A = a12 . b > 0.

it follows that ±(E11 − E22 ) ∈ LS(Γ). ±(E11 − E22 ) ∈ LS(Γ). LS(Γ) Then 0 0 exp(−2bt)A1 = exp(−4bt)a12 →t→+∞ 0 ∈ LS(Γ). 2]. t exp(2bt)a21 We go on: LS(Γ) 0 0 →t→−∞ a12 ∈ LS(Γ). There exist generalizations of the previous theorem for the case of complex spectrum of the matrix B and for general semisimple Lie groups G [15.Since ±B = b(E11 − E22 ) ∈ LS(Γ). Summing up. thus LS(Γ) = sl(2). a12 0 exp(2bt)A1 t Consequently. 42 2(E12 + E21 ) ∈ LS(Γ). and we can kill the ﬁrst coordinate of At : 0 A1 = At − a11 (E11 − E22 ) = exp(−2bt)a12 ∈ LS(Γ). cos t sin t − sin t cos t . t a21 a21 0 1 0 = −E21 ∈ LS(Γ). Similarly. . Recall that ±(E11 − E22 ) ∈ LS(Γ) as well. E11 − E22 ] = It follows that ±E12 . But this element generates a periodic one-parameter group: exp(t(E12 − E21 )) = That is why ±(E12 − E21 ) ∈ LS(Γ). 16. |a21 | a21 E12 − E21 ∈ LS(Γ). Thus ±[E12 − E21 . and the system Γ is controllable on SL(2). ±E21 . a12 = exp(4bt)a21 0 0 1 a12 = E12 ∈ LS(Γ).

And necessity in Theorem 6. (2) remain suﬃcient for controllability of Γ. the subgroup H is closed in G. simply connected Lie group. The groups T(n) and E(2) are solvable.4. If G is not simply connected.7. dim l = dim L − 1.4 is a consequence of the following necessary controllability condition for general (not necessarily solvable) simply connected Lie groups. A right-invariant system Γ ⊂ L is controllable iﬀ the following two properties hold: (1) Lie(Γ) = L and (2) Γ is not contained in a half-space in L bounded by a subalgebra. There exists a Lie subgroup H ⊂ G with the Lie algebra l. 43 . Let a Lie group G be connected.6. dim G/H = dim G − dim H = 1. Suﬃciency is highly nontrivial. We will prove only the easy part of this test — necessity. its derived series is the following descending chain of subalgebras: L ⊃ L(1) = [L. Moreover. Deﬁnition 6. conditions (1). Then the coset space G/H = {XH | X ∈ G} is a smooth manifold. There is a general controllability test for right-invariant systems on connected. and let Γ ⊂ L. simply connected.5. G/H = R. then Γ is not controllable on G. its proof may be found in [19]. Since G is simply connected and dim H = dim G − 1. Let G be a connected.4 Solvable Lie groups For a Lie algebra L. Proof. Theorem 6. and solvable. Summing up. L] ⊃ L(2) = [L(1) . Theorem 6. A Lie algebra L is called solvable if its derived series stabilizes at zero: L ⊃ L(1) ⊃ L(2) ⊃ · · · ⊃ L(N ) = {0} for some N ∈ N. If Γ is contained in a half-space in L bounded by a subalgebra. since G is simply connected. Suppose that Γ is contained in a half-space Π ⊂ L bounded by a subalgebra l ⊂ L. simply connected solvable Lie groups (recall that a topological space M is called simply connected if any closed loop in M can be contracted to a point). its quotient G/H is simply connected as well. A Lie group with a solvable Lie algebra is called solvable. L(1) ] ⊃ · · · . We have Π = R+ A + l for some A ∈ L. Further.6. Example 6.

The triangular algebra t(n) is completely solvable.The quotient G/H is a homogeneous space of G: the transitive action is θ : G × G/H → G/H. We give such a condition for a subclass of solvable Lie groups. An example of a solvable but not completely solvable Lie algebra is provided by the Lie algebra e(2) of the Euclidean group of the plane. Example 6. Deﬁnition 6. Since Γ ⊂ Π = R+ A + l. A Lie algebra is called nilpotent if all adjoint operators ad A. A ∈ L.8. then θ∗ Γ|π(Id) ⊂ θ∗ (R+ A + l)|π(Id) = θ∗ (R+ A)|π(Id) = R+ θ∗ A|π(Id) . it would be desirable to have a controllability condition with easy to verify hypotheses. XH) = Y XH. d π(H) = dt t=0 π(Id) t=0 θ∗ l|π(Id) = 0. Example 6. The Lie group T0 (n) = {X = (xij ) | xij = 0 ∀i > j. Thus θ∗ Γ is not controllable on R = G/H and Γ is not controllable on G.9. For any C ∈ l we have θ∗ C|π(Id) = d dt θ(exp(tC). d dt t=0 exp(tC) ·H ∈H d = dt = 0. H) = t=0 π(X) = XH. have only zero eigenvalues. 44 . A ∈ L. we prove that the induced system θ∗ Γ is noncontrollable on the homogeneous space G/H. Any nilpotent Lie algebra is completely solvable. That is.8.7. So admissible velocities of the induced system θ∗ Γ at π(Id) ∈ R belong to a half-line. In order to show that Γ is noncontrollable on G. In addition to Theorem 6. xii = 1 ∀i} is nilpotent. θ(Y.4. Denote the projection π : G → G/H. have only real eigenvalues. A solvable Lie algebra is called completely solvable if all adjoint operators ad A. Deﬁnition 6.

Let G be a completely solvable. . . is contained in a subalgebra l2 ⊃ l1 such that dim l2 = dim l1 + 1. . then R+ A + l2 = l2 is a subalgebra containing Γ. |ui | dim l2 = dim L − 1. . . Proof. then LS(Γ) = L. Thus Γ is not full-rank.9. k2 ) = (v1 + k1 v2 . Thus Γ is not controllable. connected. thus it is not controllable. Bm ) = l1 = L. then Π = R+ A+l2 is a half-space bounded by the subalgebra l2 / and containing Γ. . Necessity is based upon the following general fact: in a completely solvable Lie algebra L. k) | v ∈ V. Bm ) = L. . Then there exists a codimension one subalgebra l2 in L containing l1 : l1 ⊂ l2 ⊂ L. this is obvious since E(n) = X 0 y 1 ∈ M (n + 1) | X ∈ SO(n). The system Γ is controllable iﬀ Lie(B1 . . Let Lie(B1 . y ∈ Rn 45 . . Example 6. simply connected Lie group. . . Bm ) = L. k ∈ K} endowed with the product smooth structure.5 Semi-direct products of Lie groups Deﬁnition 6. and Γ is controllable. and let m Γ= A+ i=1 ui B i | ui ∈ R ⊂ L. .6. (1) If A ∈ l2 .Theorem 6. k1 k2 ). . We have LS(Γ) thus Lie(B1 . If Lie(B1 .10. . Suﬃciency. The Euclidean group E(n) is the semi-direct product Rn SO(n). . Let a Lie group K act linearly on a vector space V . . m A + ui B →u→±∞ ±Bi ∈ LS(Γ). see [31]. . Bm ) = A + l1 ⊂ R+ A + l2 . The semi-direct product of V and K is the Lie group deﬁned as the set G=V K = {(v. . . k1 ) · (v2 . (2) And if A ∈ l2 . and the group operation (v1 . 6. any subalgebra l1 ⊂ L. The system Γ is contained in a larger system: Γ= A+ i=1 ui B i ⊂ A + Lie(B1 . Bm ) ⊂ LS(Γ). l1 = L. .

The multiplication table in L = e(2) is as follows: [e1 . e2 = E13 . Theorem 6. e3 ). [e1 . Assume that the action of K has no nonzero ﬁxed points in V . as well as a generalization for the case where K has ﬁxed points in V . Proof. Then there exist linearly independent Au = A + uB and Av = A + vB such that Au . Necessity. 6. We have e(2) = span(e1 . B are linearly dependent or {A. We prove Theorem 6. then Lie(Γ) = Lie(A. The group SO(n) has no nonzero ﬁxed points in Rn . thus the derived series is L = span(e1 . Let K be a compact connected Lie group acting linearly on a vector space V . (27) 46 . e3 ] = 0.2. An invariant system Γ ⊂ L is controllable on G iﬀ Lie(Γ) = L. e3 ) is the only 2-dimensional subalgebra in e(2). B} ⊂ span(e2 . Suﬃciency. [e2 . e3 ). e2 . e2 . α1 = 0. e2 ] = −e3 . e3 ) ⊃ L(2) = {0}. Theorem 6. If A.11.8. e1 = E12 − E21 . e3 ). and let G = V K. so e(2) is solvable. B} ⊂ span(e2 . e3 ). B are linearly independent and (2) {A. ±i} ⊂ R. The proof in the general case. Let A.7. Further.7 in the simplest case G = E(2). so e(2) is not completely solvable. may be found in [5]. Example 6. thus an invariant system Γ ⊂ e(n) is controllable on E(n) = Rn SO(n) iﬀ Γ is full-rank. Av ∈ span(e2 . Now we give a controllability test on E(2). B are linearly independent and {A. The following controllability test for semi-direct products can be seen as a generalization of the controllability test for compact Lie groups given in Th. e3 ). e3 ] = e2 . For the element / Au = α 1 e1 + α 2 e2 + α 3 e3 . B} ⊂ span(e2 . It easily follows from multiplication table (27) that span(e2 . A system Γ = A + RB ⊂ e(2) is controllable on G = E(2) iﬀ the following conditions hold: (1) A. Sp(ad e1 ) = {0. e3 ) ⊃ L(1) = span(e2 . e3 = E23 . thus Γ is not controllable.and X1 0 y1 1 X2 0 y2 1 = X1 X2 0 X1 y 2 + y 1 1 . Γ = A + RB ⊂ e(2). Let G = E(2). B) = L. Γ = A + RB.

Thus l = e(2). . . . Now we are able to prove Theorem 6. LS(Γ) = e(2) = L. Proof. ˙ q(0) = q0 . u(t)) dt → min . Since Au ∈ Γ. dxnq ∗ ∗ are basis linear forms in Tq M . Necessity of the rank condition is a general fact. thus Γ is controllable on E(2). . and any covector λ ∈ Tq M is decomposed as λ = n pi dxiq . At any point q ∈ M . then conditions (1). . u ∈ U ⊂ Rm . . But l is not contained in span(e2 . On the other hand. xn ) on M . 47 . and admissible controls u(t) are measurable locally bounded. Similarly. . f (q. (2) of Theorem 6. . then ±Au ∈ LS(Γ).is periodic. 7. if Lie(Γ) = e(2). take any local coordinates (x1 . Here M is a smooth manifold. . it is a smooth manifold of dimension 2n. the one-parameter subgroup cos(α1 s) sin(α1 s) exp(sAu ) = − sin(α1 s) cos(α1 s) 0 0 α2 α3 α1 sin(α1 s) + α1 (1 − cos(α1 s)) α2 α3 α1 (cos(α1 s) − 1) + α1 sin(α1 s) 1 7 Pontryagin Maximum Principle for invariant optimal control problems on Lie groups q = f (q. .1 Hamiltonian systems on T ∗ M Let M be a smooth n-dimensional manifold. the tangent ∗ space Tq M has the dual space — the cotangent space Tq M = (Tq M )∗ . Corollary 6.8 are satisﬁed. xn ) provides local coordii=1 nates called canonical coordinates on the cotangent bundle T ∗ M . t1 Now we turn to optimal control problems of the form q ∈ M. A system Γ = A+RB ⊂ e(2) is controllable on E(2) iﬀ Lie(Γ) = e(2). In order to state the fundamental necessary optimality condition — Pontryagin Maximum Principle [30] — we recall some basic notions of the Hamiltonian formalism on the cotangent bundle. . In order to construct local coordinates on T ∗ M . u) and ϕ(q. pn .7 in the simplest case. x1 . The 2n-tuple (p1 . Then dx1q . .2. . q(t1 ) = q1 . . u) are smooth. . ±Av ∈ LS(Γ). The dis∗ joint union of all cotangent spaces is the cotangent bundle T ∗ M = q∈M Tq M . and Γ is controllable on E(2). u). Av ) ⊂ LS(Γ). J(u) = 0 ϕ(q(t). . e3 ) — the only 2-dimensional subalgebra in e(2). Thus the subspace l = Lie(Au .

Take any point λ ∈ T ∗ M . x) ∈ C ∞ (T ∗ M ). The symplectic form σ ∈ Λ2 (T ∗ M ) is deﬁned as the diﬀerential σ = ds. ∂x ∂h x= ˙ . the Hamiltonian system of ODEs ˙ λ = h(λ) reads in canonical coordinates as ∂h . for any vector ﬁeld V ∈ Vec(T ∗ M ). xn ) on T ∗ M . n σ = dp ∧ dx = i=1 dpi ∧ dxi . u). and any tangent vector ξ ∈ Tλ (T ∗ M ). . The Hamiltonian vector ﬁeld corresponding to a Hamiltonian function h is deﬁned as such vector ﬁeld h ∈ Vec(T ∗ M ) that dh = −ih σ. . ˙ q(0) = q0 . The tautological 1-form s ∈ Λ1 (T ∗ M ) is deﬁned as follows. u 48 . ∂p p=− ˙ 7. we have: n s = p dx = i=1 pi dxi .2 Pontryagin Maximum Principle on smooth manifolds q = f (q. one can deﬁne the 1-form σ(V. π∗ ξ . (28) (29) (30) J(u) = 0 Let λ ∈ T ∗ M be a covector.∗ The canonical projection π : T ∗ M → M maps a covector λ ∈ Tq M to the base point q ∈ M . On the other hand. π(λ) = q. t1 ﬁxed or free. . pn . f (q. In canonical coordinates (p1 . and u ∈ U a control parameter. ϕ(q(t). The corresponding Hamiltonian vector ﬁeld h ∈ Vec(T ∗ M ) is introduced in the following way. . ξ = λ. q(t1 ) = q1 . For a Hamiltonian h = h(p. Then sλ . u) + νϕ(q. Introduce the family of Hamiltonians hν (λ) = λ. x1 . Any smooth function h ∈ C ∞ (T ∗ M ) is called a Hamiltonian. u(t)) dt → min . . ν ∈ R a parameter.1. · ) = iV σ ∈ Λ1 (T ∗ M ). . . u ∈ U ⊂ Rm . . u). t1 Consider optimal control problem of the form q ∈ M. Example 7. The diﬀerential dh is a 1-form on T ∗ M .

Let E be a vector space of dimension dim E = dim M = n. Although.q) (E × M ). and a number ν ∈ R such that: ˜ ν ˙ λt =hu(t) (λt ). V ∈ T(e. Let u(t). t ∈ [0. q ∈ M. ∗ ω ∈ T(e. the cotangent bundle T ∗ G has a natural trivialization as follows: Φ : L∗ × G → T ∗ G.q) (E × M ) = e = q q Respectively. q) : E → Tq M is a linear isomorphism for any q ∈ M . ω = ωv + ωh . 0). Vv ∈ E. t1 ]. ¯ 49 a ∈ L∗ . q) ∈ Tq M. cannot be represented as the direct product E × M of a vector space E with M . Deﬁnition 7.Theorem 7. We will apply this trivialization in order to write Hamiltonian system of PMP for optimal control problems on Lie groups. ∗ (2) Φ( · . q) of the trivialized cotangent bundle E × M ∼ T ∗ M . X ∈ G. = ∗ ∼ T ∗ E ⊕ T ∗ M ∼ E ∗ × T ∗ M. 7. For a Lie group G. ν) ≡ (0.. At any point (e.q) (E × M ) = Te E ⊕ Tq M ∼ E × Tq M. X) → aX . . ˜ ν hu(t) (λt ) ˜ − → (31) (32) (33) (34) t ∈ [0. (a. For the problem (28)–(30) with free terminal time t1 . t ∈ [0. ν ≤ 0. necessary optiν mality conditions read as (31)–(34) plus the additional equality hu(t) (λ(t)) ≡ 0. be an ˜ optimal control in the problem (28)–(30) with ﬁxed time t1 . Vh ∈ Tq M. ˜ The proof of Pontryagin Maximum Principle on smooth manifolds may be found in [1]. u u∈U (λt . T(e. t1 ].1.3 Hamiltonian systems on T ∗ G Notice that in general the cotangent bundle T ∗ M of a smooth manifold M is not trivial.e.1 (PMP on smooth manifolds). i. Remark. A trivialization of the cotangent bundle T ∗ M is a diﬀeomorphism Φ : E × M → T ∗ M such that: ∗ (1) Φ(e. ∗ ωh ∈ Tq M. any tangent and cotangent vector are decomposed into the vertical and horizontal parts: V = Vv + Vh . we = have the following identiﬁcations of the tangent and cotangent bundles: ∼ T(e. e ∈ E. the cotangent bundle T ∗ G of a Lie group G has a natural trivialization. = max hν (λt ). Then there exists a ∗ Lipschitzian curve λt ∈ Tq (t) M .q) (E × M ). ωv ∈ E ∗ . t1 ].

XB)) since dω(V. A − a. W − W ω. XB)] taking into account formula (35) for Φs = (ξ. XA). XB) ∈ L∗ ⊕ TX G we have (Φσ)(a. XA = a. (ξ. [V. A . compute the symplectic 2-form Φσ ∈ Λ2 (L∗ × G). W ) = V ω. Φ∗(a.X) . (ξ. B] = ξ. Now we compute the pull-back of the tautological 1-form s. W ] = (ξ. π∗ Φ∗(a. X) = (ξ.X) . this is the form of the Hamiltonian of PMP for a left-invariant optimal control problem on the Lie group G. Then (Φs)(a. and a Hamiltonian vector ﬁeld h to the trivialized cotangent bundle L∗ × G ∼ T ∗ G. Take any point (a. Decompose the required Hamiltonian vector ﬁeld h ∈ Vec(L∗ × G) into the vertical and horizontal parts: h(a. (η. XA = a. B) a. and a ∈ Λ1 (G) is ¯ the left-invariant 1-form on G obtained by left translations from the covector a = aId ∈ L∗ : ¯ aX . A − a. XA) ¯ ¯ ¯ = aX . ∂a 50 . the symplectic 2-form σ. XA) ∈ L∗ ⊕ TX G. XA) = saX .X) (ξ. A . XA) Φs(a. take a Hamiltonian h = h(a) not depending on X ∈ G. For any tangent vectors (ξ. (35) Further. XA) ∈ L∗ ⊕ TX G. (η. X ∈ G.X) ((ξ. (η. (η. XA) = aX . XA). [(ξ.Here L∗ is the dual space of the Lie algebra L = TId G.X) . (η. Apply the identity dh = −Φσ(h. XB) Φs(a. XB) − (η. A) a. B − (η. XA). V − ω. · ) to an arbitrary tangent vector (η. XA). ¯ a ∈ L∗ . X ∈ G. = We start from the tautological 1-form Φs ∈ Λ1 (L∗ × G). Finally. B] . Since the Hamiltonian h does not depend on X. [A.X) ((ξ. A ∈ L.X) (ξ. B − η.X) . XB) ∈ L∗ ⊕ TX G. XB)) since Φσ = Φds = dΦs = (dΦs)(a. a ∈ L∗ . we denote dh = ∂h ∈ (L∗ )∗ = L. [A. X) ∈ L∗ × G and a tangent vector (ξ. XA) − Φs(a.

ωn ∈ Λ1 (G) linearly independent at each point of G. ∂a So ξ = (ad A)∗ a = ad ∀B ∈ L. [A. [A. . v ∈ RN . g(Xu. ∂a Summing up. XB) ∂a = dh. ∀u. Then the required inner product g can be constructed as follows: g(u. B = − tr(AB). XB) = − ξ. there exists an inner product g( · . B] = (ad A)∗ a. XA). B = a. · ) on RN . 0) ∂a thus A = = η. thus L ⊂ so(N ). That is. (36) ∂h . . ∂a Now we set η = 0 in (36) and ﬁnd the horizontal part of h: 0 = dh. XB)) = − ξ. (η.X) ((ξ. B + η. . Setting B = 0 in (36). thus ξ. Xv) = g(u. a ∈ L∗ . ˙ X ∈ G. we obtain: ∂h . we compute the vertical part of h: ∂h . the Hamiltonian system on T ∗ G ∼ L∗ × G for a left-invariant = Hamiltonian h = h(a). a ∈ L∗ . ∂h ∂a = η.4 ∂h a. Hamiltonian systems in the case of compact Lie group Indeed. . · ) on RN such that g(Xu. ˜ So in the sequel we assume that G ⊂ O(N ). reads as follows: ∗ a = ad ∂ h a. (η. v) ∀X ∈ G.Taking into account formula (35) for Φσ. A + a. A ∀η ∈ L∗ . (η. (η. [A. · : A. B] . v) = G The Hamiltonian system (37) simpliﬁes in the case of a compact Lie group G. XB) = −Φσ(a. and choose any ˜ left-invariant 1-forms ω1 . start from an arbitrary inner product g ( · . 51 . B + a. B ∗ 7. Let G ⊂ GL(N ) be a compact linear Lie group. B] . Xv) ω1 ∧ · · · ∧ ωn . (0. Then it is easy to show that in fact G ⊂ O(N ). ˙ ∂a (37) X = X ∂ h . But the Lie algebra so(N ) has an invariant inner product ·.

we obtain the inﬁnitesimal version of the invariance identity: ad C(A). ∂ h . we have A. the operator e : so(N ) → so(N ) is orthogonal. ∂a 52 . t ad C i. the operator ad C : so(N ) → so(N ) is skew-symmetric. Diﬀerentiating identity (38) w. Consequently. ∗ ˜ Thus (ad A)∗ B = −ad A(B). C = B. we compute the action of the operator (ad A) : L → L. ˙ ∂a ∂a (39) X = X ∂ h . · is invariant in the sense of the following identity: et ad C A. ∂h : L∗ → L∗ becomes deﬁned in ∂a ∗ L. This identity easily follows since et ad C A = etC Ae−tC and et ad C A. Aij Bij . the operator ad : L∗ → L∗ is identiﬁed with the ∂a ∂h operator − ad : L → L. B. et ad C B = A.j=1 B = (Bij ). the Lie algebra L ⊂ so(N ) is endowed with an invariant scalar product. ∀t ∈ R. ∗ ∂h In particular. For any B. ˙ X ∈ G.. the operator ad A ∈ L. so the operator (ad A) : L → L coincides with − ad A. C ∈ so(N ). (ad A)C = B.Writing skew-symmetric matrices as A = (Aij ). ∗ ˜ A ∈ L∗ . we have ˜ ˜ (ad A)∗ B. This allows us to identify the Lie algebra L with its dual space L∗ : ˜ A ↔ A = A. the vertical part of ∂a the Hamiltonian system is deﬁned on the Lie algebra L: a = − ad ∂ h a = a. B + A. ad C(B) = 0 ∀A. (ad A)C = − (ad A)B. Let A ∈ L. Via this identiﬁcation. The product ·.r. C = − (ad A)B. C . B. C ∈ L. Aij = −Aji . a ∈ L. et ad C B = − tr(etC Ae−tC etC Be−tC ) = − tr(etC ABe−tC ) = − tr(AB) = A. B = i. C ∈ so(N ). N Bij = −Bji .t. t at t = 0. So for a compact Lie group G. · . B ∀A. B by invariance of trace. (38) In other words.e.

X1 ∈ G.Now we apply expressions (37). XA. So in every tangent space TX G there is a scalar product Lipschitzian curve X : [0. moreover. (l(u))2 = 0 t1 2 t1 |u(t)| dt ≤ 0 |u(t)|2 dt · t1 . · in the Lie algebra L deﬁnes a left-invariant Riemannian structure on G: XA. while G is connected. XB X = A. 53 . the Riemannian problem l → min is equivalent to the problem J(u) = 1 2 t1 0 |u(t)|2 dt → min . notice that invariant system (40) is controllable since Γ = L is full-rank and symmetric. 8 8. the equality occurs only if |u(t)| ≡ const. B . (44). The problem is stated as follows: given any pair of points X0 . u ∈ L. X ∈ G. ﬁnd the shortest curve in G that connects X0 and X1 . The invariant scalar product · .· X.1 Examples of invariant optimal control problems on Lie groups Riemannian problem on compact Lie group Let G be a compact connected Lie group. XB ∈ TX G. (44) The functional J is more convenient than l since J is smooth and its extremals are automatically curves with constant velocity. X 1 ∈ G t1 X ∈ G. A. X0 . In the sequel we consider the problem with the functional J: (40)–(42). (39) for Hamiltonian systems in order to study invariant optimal control problems on Lie groups. First of all. By Cauchy-Schwartz inequality. ˙ |X| = ˙ ˙ X. X(0) = X0 . The corresponding optimal control problem is as follows: ˙ X = Xu. X(t1 ) = X1 . (40) (41) (42) (43) l(u) = 0 |u(t)| dt → min . ·. Consequently. ﬁxed. t1 ] → M its Riemannian length is deﬁned as integral of velocity: t1 For any l= 0 ˙ |X(t)| dt. B ∈ L. X .

So there are no abnormal extremal trajectories. we write Hamiltonian system (2) in the form (39). t1 = 1. then there exist a curve a(t) ∈ L and ν ≤ 0 such that: (1) (a(t). a = a. Consider ﬁrst the abnormal case: ν = 0. thus in the initial problem (40)–(43) as well. ∂a u∈L (3) hν (a(t)) = max hν (a(t)). That is. ∂ h = [a. X(t)) is optimal. Notice that conditions of PMP (1)– (3) are preserved under multiplications of (a. (44). This contradicts PMP since the pair (ν. The Hamiltonian system (2) for such a control has the form: a = [a. a) should be nonzero. ˙ ˙ X = Xa. we obtain that any point X1 ∈ G can be represented in the form X1 = e a . t1 ]. Now consider the normal case: ν < 0. If a pair (u(t). ν) = 0. for the case X0 = Id. 54 . Xu + |u|2 = a. In particular. The Hamiltonian of PMP for the problem J → min has the form: ν ν hν (a. ¯ u u 2 2 We apply PMP. a ∈ L. t ∈ [0. a] = 0.Further. The maximality condition 1 h−1 (a) = a. X1 ∈ G and any t1 > 0 there exists a ∈ L such that X1 = X0 eat1 . u u(t) Since the group G is compact. u]. u − |u|2 → max u u∈L 2 gives u(t) ≡ a(t). a ∈ L. ν) by positive constants. ˙ ∂a (2) ˙ X = X ∂ h = Xu. Filippov’s theorem [1] implies existence of optimal controls in problem (40)–(42). The maximality condition h0 (a) = a. We showed that for any X0 . X) = aX . Thus optimal trajectories are left translations of one-parameter subgroups in M : X(t) = X0 eta . u + |u|2 = hν (a). so we can assume that ν = −1. u → max u u∈L implies that a(t) ≡ 0. any element X1 in a connected compact Lie group G has a logarithm a in the Lie algebra L.

That is. The Hamiltonian of PMP is the same as in the previous problem: hν (a) = a. (45) u u⊥b 55 . a2 ) for some linearly independent a1 . X(0) = X0 . u ∈ U . 2 Consider ﬁrst the abnormal case: ν = 0. l(u) = 0 |u(t)| dt → min . u ∈ U. Similarly to the Riemannian problem. The maximality condition of PMP reads h0 (a) = a. X(0) = X0 . this restriction on velocities means that we ﬁx an axis b in the rigid body and allow only rotations of the body around any axis u orthogonal to b. we should ﬁnd the shortest path between ﬁxed points X0 . |b| = 1.8. / thus it is controllable on SO(3). X1 in the Lie group G. t1 X(t1 ) = X1 . A problem of ﬁnding shortest curves tangent to a given distribution ∆X ⊂ TX G is called a sub-Riemannian problem: ˙ X ∈ ∆X(t) . X1 ∈ G ﬁxed. the length minimization problem is equivalent to the problem J(u) = 1 2 t1 0 |u(t)|2 dt → min. Then ˙ ˙ ∆X = XU . u → max . a2 ] ∈ span(a1 . and the restriction X ∈ ∆X can be written as X = Xu. As before. we deﬁne a left-invariant ˙ ﬁeld of tangent hyperplanes on X. the system Γ has the full rank. The optimal control problem is stated as follows. b = 0}. such that ∆Id = b⊥ = {u ∈ L | u.2 Sub-Riemannian problem on SO(3) Consider the case G = SO(3). a2 ). X0 . X ∈ G. and modify the previous problem. u + u ν 2 |u| . Controllability: we have Γ = b⊥ = span(a1 . and Filippov’s theorem guarantees existence of optimal controls. Denote U = b⊥ . choose an element b ∈ L. a2 ∈ so(3). X(1) = X1 . and X(t) should belong to the hyperplane attached at the point X(t). Since [a1 . To state the problem as an optimal control one. ˙ But now admissible velocities X are not free: they should be tangent to a leftinvariant distribution (of corank 1) on X. l(X(·)) → min . ˙ X = Xu. For a rigid body rotating in R3 with orientation matrix X ∈ SO(3).

e. [b. a [b. a b. a − b. b]. But such an equality in so(3) means that u b. α(t) = 0. a · = b. u] = 0. . X ≡ const. a(t) = α(t)b. a] b. ˙ =0 (48) So equation (48) can be rewritten as a = b. Thus [b. [b. u] ⊥ b ⇒ αb ⊥ b ˙ ⇒ α = 0. a] = ad( b. u] = 0. (46) Then maximality condition (45) is rewritten as h0 (a) = a⊥ . so [b.Consider the decomposition a = a + a⊥ . abnormal optimal trajectories are ∂a constant and give only trivial solutions to our problem. u u⊥b which yields a⊥ = 0. Via decomposition (46).. a0 [b. u] = α[b. Now consider the normal case: ν = −1. a = b. u − |u|2 → max. b. we obtain u ≡ 0 for an abnormal optimal control. u] = [a. Since u ⊥ b. a]. by invariance of the scalar product in so(3). a0 b)a ˙ 56 a0 = a(0). The vertical part of the Hamiltonian system of PMP takes the form a = [a. Then the horizontal part of Hamiltonian system (39) reads ∂h ˙ X =X = Xu = 0. ˙ It is easy to see that this ODE has the integral b. u]. ∂a (47) Further. a⊥ ⊥ b. ˙ Then equality (47) implies α[b. i. That is. ˙ ˙ ∂h = [a. a ≡ const: b. a b] = b. a = 0. u u⊥b 2 thus normal optimal controls are u = a⊥ = a − b. u = 0. The vertical part of Hamiltonian system (39) for our problem yields αb = a = a. u] = − [b. u → max. a b. the maximality condition of PMP reads 1 h−1 (a) = a⊥ .

a0 b. 0 0 0 The Heisenberg group is the deﬁned 1 x G= 0 1 0 0 e1 = E12 . γ ∈ R = span(e1 . X(t) = Y (t)e−tc = Y (0)et(d+c) e−tc = X(0)eta0 e−t b. d = a0 − b. b. After the change of variable Y = Xetc.which is immediately solved: a(t) = et ad( b. a0 b) = X et ad( = Xet ad( In the notation c = b. 8.a0 b) b. which is solved as Y (t) = Y (0)et(d+c) . Its Lie group is 0 α γ L = 0 0 β | α.z . Summing up. e2 = E23 . we showed that all optimal trajectories in the sub-Riemannian problem on SO(3) are products of two one-parameter subgroups. 1 where we denote 3 This group is diﬀeomorphic to Rx. Now consider the horizontal part of the Hamiltonian system of PMP: ˙ X = Xu = X(a − b. ˙ X = Xet ad c d = Xetc d e−tc . 57 e3 = E13 . e2 . e3 ).a0 b) a0 − b. ˙ Xetc = Xetc d.a0 b) a0 . Finally. β. we obtain the ODE that is. (49) .a0 b . thus it is not compact. a0 b (a0 − b. we come to the equation ˙ ˙ Y = Xetc + Xetc c = Xetc(d + c) = Y (d + c).y. a0 b) . z ∈ R .3 Sub-Riemannian problem on the Heisenberg group as z y | x. a0 b. y.

e3 ] = [e2 . thus 0 0 0 ad e1 = 0 0 0 . −1 0 0 0 0 A1 0 0 . e2 ).The multiplication table in this basis looks like [e1 . A . In the dual of the Heisenberg Lie algebra L one can choose the basis dual to basis (49): L∗ = span(ω1 . e2 ] = e3 . ej = δij . 3. and a column vector. a square matrix. 0 3 [e1 . and elements of its dual as row vectors: 3 We write elements of the Lie algebra as column vectors 3 A1 L A= Ai ei = A2 . ω3 ). A = a. . ω2 . The plane ∆Id = span(e1 . its dual C ∗ : L∗ → L∗ acts as C ∗ a. CA = a C the product of a row vector. e3 ] = 0. Consequently. Thus C∗a = a Consider the left-invariant sub-Riemannian problem on the Heisenberg group determined by the orthonormal frame (e1 . (50) has the zero spectrum. So any adjoint operator A= i=1 Ai ei ∈ L. i=1 A3 L∗ a= i=1 a i ωi = a1 a2 a3 . the Heisenberg group G is nilpotent. e2 ) ⊂ L 58 C . i. ad e3 = 0. 0 1 0 0 ad A = 0 −A2 0 0 0 ad e2 = 0 0 0 . 2. For a linear operator C : L → L. ωi . j = 1.

e2 ) ⊂ L. ej Id in ∆Id deﬁned by i. u1 e1 + u2 e2 + (u1 + u2 ) = u1 a1 + u2 a2 + (u2 + u2 ) 2 2 2 2 1 ν = hu (a). t1 X(t1 ) = X1 . The corresponding control system reads ˙ X = u1 Xe1 + u2 Xe2 . X dt → min . Xej X in ∆X as follows: = δij . Since ˙ |X| = |u1 Xe1 + u2 Xe2 | = |u1 e1 + u2 e2 | = the sub-Riemannian length functional takes the form t1 (u1 . u2 + u2 . 59 . 2. thus Γ is controllable. generates the left-invariant scalar product ·. Moreover. i. As before. · e i . 1 2 Filippov’s theorem implies existence of optimal controls. we pass to the functional J= 1 2 t1 0 (u2 + u2 ) dt → min . the scalar product ·. Consider the corresponding sub-Riemannian problem: ˙ X ∈ ∆X . Further. · X in ∆X determine a left-invariant sub-Riemannian structure on the Lie group G. The distribution ∆X ⊂ TX G with the scalar product ·. e2 ] = e3 . Γ is symmetric and G is connected.generates the left-invariant distribution ∆X = span(Xe1 . · Xei . The Hamiltonian of PMP reads ν hν (a. ˙ |X| dt = t1 0 l(X(·)) = 0 ˙ ˙ X. j = 1. X(0) = X0 . 2. 1 2 (52) We solve optimal control problem (51). (52). 1 2 (51) l= 0 u2 + u2 dt → min . u2 ) ∈ R2 . the system Γ has full rank. j = 1. X) = aX . Controllability: we have Γ = span(e1 . Since [e1 . Xe2 ) ⊂ TX G. X Id = δij . u1 Xe1 + u2 Xe2 + (u2 + u2 ) ¯ u 2 2 1 ν 2 ν = a.

˙ a2 = a3 u1 ≡ 0. a1 = −a3 u2 . ˙ a3 = 0. In the normal case ν = −1 the maximality condition 1 h−1 (a) = u1 a1 + u2 a2 − (u2 + u2 ) → max 2 u 2 2 1 (u1 . thus a3 = 0. 0 Consider ﬁrst the abnormal case: ν = 0.The Heisenberg group is noncompact. First we consider the vertical part a= ˙ We have ad ∂h ∂a ∗ a. thus a ∈ L∗ . we obtain 0 0 ∂h 0 0 = ad ∂a −u2 u1 u1 ∂h = u1 e1 + u2 e2 = u2 ∈ L.u2 )∈R 60 . ˙ a 2 = a 3 u1 . ˙ whence the abnormal optimal controls are u1 = u2 ≡ 0. Then the maximality condition of PMP h0 (a) = u1 a1 + u2 a2 → max 2 u (u1 . Then the horizontal part of the Hamiltonian system ∂h ˙ X =X = X(u1 e1 + u2 e2 ) ∂a gives X ≡ 0. −u2 u1 0 that is. Thus there are no nonconstant abnormal optimal trajectories. and we will write the Hamiltonian system of PMP in the form (37).u2 )∈R yields a1 = a2 = 0. ˙ Taking into account equality (50). ∂a 0 0 0 . (53) Thus the vertical part (53) of the Hamiltonian system of PMP takes the form 0 0 0 a1 a2 a3 = a1 a2 a3 0 ˙ ˙ ˙ 0 0 = −a3 u2 a3 u1 0 . Then the Hamiltonian system implies a1 = −a3 u2 ≡ 0. a ∈ L∗ .

u2 = a2 . a2 = r sin θ. 1 2 So it is convenient to pass to the polar coordinates a1 = r cos θ. ˙ Now the vertical subsystem is immediately integrated: θ = θ0 + a3 t. ˙ ˙ X = X(a1 e1 + a2 e2 ). a2 = r sin(θ0 + a3 t). ˙ a3 = 0. ˙ In view of the left invariance of the problem.e. the normal Hamiltonian system of PMP reads as follows: a1 = −a3 a2 . Consequently. ˙ a2 = a 3 a1 . 0 that is. ˙ ˙ θ = a3 . we can restrict ourselves by trajectories starting from the identity: X(0) = X0 = Id. ˙ z = xa2 . ˙ y = a2 . We rewrite the horizontal subsystem as 0 x z ˙ ˙ 1 x z 0 a1 0 0 y = 0 1 y 0 0 ˙ 0 0 0 0 0 1 0 0 x = a1 . i. 0 0 a1 a2 = 0 0 0 0 0 xa2 a2 . x(0) = y(0) = z(0) = 0.. in which the vertical part of the Hamiltonian system reads r = 0. a1 = r cos(θ0 + a3 t). It is easy to see that this system has an integral a2 + a2 ≡ const since 1 2 (a2 + a2 )· = 2a1 (−a3 a2 ) + 2a2 a3 a1 = 0. 61 . a3 = 0.implies u1 = a1 .

And if a3 = 0. y) bounded by the curve (x(t). then projections of extremal trajectories X(t) to the plane (x. and the straight line perpendicular to this axis. y1 ) such that the domain bounded by γ and γ0 has the algebraic area S. y1 ). y0 ) to (x1 . y1 . r sin θ0 dt = tr sin θ0 . is the algebraic area of the domain in the plane (x.Consider ﬁrst the case a3 = 0. We found solutions of the minimization problem t1 0 x + y dt → min ˙ ˙ along Lipshchitzian plane curves (x(t). 2π/|a3 |]. one should ﬁnd a plane curve γ connecting (x0 . t ∈ [0. One can show that such arcs are optimal up to the ﬁrst complete circle: X(t). y(t)). z)(t1 ) = (x1 . This is one of the ancient optimization problems known as Dido’s problem. y. Solutions to this problem are straight lines and arcs of circles. a3 t r y= r sin(θ0 + a3 t) dt = (cos θ0 − cos(θ0 + a3 t)). and a number S = z1 − z0 .C [28]. y0 ). (x1 . 0 t y= z= 0 tr2 cos θ0 sin θ0 dt = t2 2 r cos θ0 sin θ0 . 2 And if a3 = 0. y(t)) under the boundary conditions (x. then such projections are arcs of circles. and γ is the shortest possible curve. a plane curve γ0 connecting (x1 . it goes back to IX B. z0 ). the axis y. a3 0 t r (sin(θ0 + a3 t) − sin θ0 )r sin(θ0 + a3 t) = z= a3 0 r2 t sin(2(θ0 + a3 t)) − sin 2θ0 sin θ0 = − + (cos(θ0 + a3 t) − cos θ0 ) . Given two points (x0 . y) are straight lines. +∞) are optimal. y0 . a3 2 4a3 a3 r cos(θ0 + a3 t) dt = 0 t If a3 = 0. y1 ) to (x0 . z1 ). z)(0) = (x0 . y0 ). y. this problem can be stated as follows. t ∈ [0. Geometrically. 62 . where z(t) = x dy (x. Then t x= 0 t r cos θ0 dt = tr cos θ0 . thus the whole trajectories X(t). then x= r (sin(θ0 + a3 t) − sin θ0 ).

be the arc-length parametrization of the elastic rod. This problem has obvious symmetries — translations and rotations in the plane (x. the state space of the control system is cos θ − sin θ x cos θ y | (x. Euler in 1744 [20]. y. equal to the angular velocity. y(t)). θ)(0) = (x0 . t1 being its length assumed ﬁxed. thus k 2 = θ2 = u2 . θ)(t1 ) = (x1 . Indeed. the ˙ curvature is. G = E(2) = sin θ 0 0 1 Further. ˙ y = sin θ. where k is the curvature of the rod. We should ﬁnd the proﬁle of the elastic rod with ﬁxed endpoints (x0 . So it is natural to expect that it can be stated as an invariant problem on the Euclidean group E(2). v1 . y1 . y0 . y1 ) and ﬁxed tangents v0 . where v0 = (cos θ0 . sin θ0 ). (x. t ∈ [0. y) ∈ R2 . |v0 | = |v1 | = 1. (x.4 Euler’s elastic problem Now we consider a problem studied ﬁrst by L. y1 ) in the plane and two unit vectors v0 . θ0 ). y. v1 = (cos θ1 . attached respectively at these points. θ ∈ S 1 . t1 ]. = sin θ 0 0 1 0 0 0 63 . y). Suppose that we have two points (x0 . v1 at these endpoints. θ1 ). (x1 . up to sign. (x1 . Let θ(t) be the angle between the velocity vector (x(t). Let (x(t).8. the dynamics of the system reads cos θ − sin θ x − sin θ u − cos θ u cos θ d ˙ X= sin θ cos θ y = cos θ u − sin θ u sin θ dt 0 0 1 0 0 0 cos θ − sin θ x 0 −u 1 cos θ y u 0 0 . y0 ). sin θ1 ). and we obtain the cost functional for the optimal control problem: J= 1 2 t1 0 u2 dt → min . y0 ). Then the elastic ˙ ˙ problem can be stated as follows: x = cos θ. For an arc-length parametrized curve. The elastic energy of the rod is measured by the integral J= 1 2 t1 0 k 2 dt → min. ˙ ˙ θ = u. y(t)) and the positive direction of the axis x.

**The Lie algebra of the Euclidean group is L = e(2) = span(E21 − E12 , E13 , E23 ).
**

e1 e2 e3

**So the elastic problem is left-invariant: ˙ X = X(e2 + ue1 ), u ∈ R, X(0) = X0 , X(t1 ) = X1 , J= 1 2
**

t1 0

X ∈ G,

u2 dt → min .

We already computed multiplication table in e(2), see (27): [e1 , e2 ] = e3 , whence [e1 , e3 ] = −e2 , [e2 , e3 ] = 0,

We choose the dual basis in the dual space to the Lie algebra: L∗ = span(ω1 , ω2 , ω3 ), ωi , ej = δij ,

0 0 0 ad e1 = 0 0 −1 , 0 1 0 0 0 0 ad e3 = 1 0 0 . 0 0 0

0 0 0 ad e2 = 0 0 0 , −1 0 0

(54)

(55)

and write elements of the Lie algebra as column vectors 3 A1 L A= Ai ei = A2 , i=1 A3 and elements of the dual space as row vectors:

3

L∗

a=

i=1

a i ωi =

a1

a2

a3

.

Controllability. The system Γ = e2 + Re1 ⊂ L is controllable on G = E(2) by Theorem 6.8. Now we ﬁnd extremal trajectories. The Hamiltonian of PMP reads ν hν (a) = a, e2 + ue1 + u2 , a ∈ L∗ , u, ν ∈ R. u 2 ∂h Thus = e2 + ue1 , and in view of (54), (55) ∂a 0 0 0 ∂h 0 0 −u . ad = ∂a −1 u 0 64

Consequently, the Hamiltonian system (37) reads as follows: a1 = −a3 , ˙ a2 = ua3 , ˙ a3 = −ua2 , ˙ x = cos θ, ˙ y = sin θ, ˙ ˙ θ = u.

**In the abnormal case ν = 0, and the maximality condition h0 (a) = a2 + ua1 → max u
**

u∈R

yields a1 (t) ≡ 0. Then the vertical subsystem takes the form a1 = 0 = −a3 , ˙ a2 = ua3 = 0, ˙ a3 = 0 = −ua2 . ˙

We have a1 = a3 ≡ 0, so a2 ≡ const = 0 and u ≡ 0. Notice that this is a singular control , i.e., it is not determined immediately by the maximality condition of PMP. Now we integrate the horizontal subsystem: θ = θ0 , x = t cos θ0 , y = t sin θ0 . Consider the normal case: ν = −1, 1 h−1 (a) = a2 + ua1 − u2 → max, u u∈R 2 whence u = a1 . So the vertical subsystem reads a1 = −a3 , ˙ a2 = a 1 a3 , ˙ a3 = −a1 a2 . ˙ In view of the integral a2 + a2 ≡ const, we pass to the polar coordinates: 2 3 a2 = r cos α, The vertical subsystem simpliﬁes: r = 0, ˙ α = −a1 , ˙ a1 = −r sin α. ˙ The angle α satisﬁes the equation of mathematical pendulum α = r sin α. Fur¨ ˙ ther, θ = u = a1 = −α, thus θ = β − α, β = const. Finally, the angle θ satisﬁes ˙ 65 a3 = r sin α.

¨ the equation θ = −r sin(θ − β), and we obtain the following closed system for optimal trajectories: x = cos θ, ˙ y = sin θ, ˙ ¨ = −r sin(θ − β). θ ˙ If r = 0, then θ = θ0 +tθ0 , and Euler elastica, i.e., optimal curves (x(t), y(t)), are the same as in the sub-Riemannian problem on the Heisenberg group, i.e., lines and circles. Let r > 0. Then we can apply homotheties in the plane (x, y) in order to obtain r = 1, and further apply rotations in this plane in order to have β = 0. Then the angle θ satisﬁes the standard equation of the mathematical pendulum ¨ θ = − sin θ, i.e., ˙ θ = c, c = − sin θ. ˙

Here c is the curvature of Euler elastica. The diﬀerent qualitative types of solutions to the equation of pendulum depend on values of the energy of the pendulum c2 − cos θ ∈ [−1, +∞). E= 2 The following cases are possible: (1) E = −1, (2) E ∈ (−1, 1), (3a) E = 1, θ = ±π, (3b) E = 1, θ = ±π, (4) E ∈ (1, +∞). It is known that the equation of mathematical pendulum is integrable in elliptic functions [18]. One can show that equations for elastica are integrable in elliptic functions as well, and the following qualitative types of elastica are possible: (1) straight line, Fig. 1, (2) inﬂectional elastica, Fig. 2–5, (3a) critical elastica, Fig. 6, (3b) straight line, Fig. 7, (4) non-inﬂectional elastica, Fig. 8–10, (5) r = 0 ⇒ circles, Fig. 11, and straight lines.

66

5 3 2.4 0. 1) Figure 5: E ∈ (−1. 1) Figure 3: E ∈ (−1.5 1 0.5 -1 0 1 2 3 4 Figure 1: E = −1 3.5 1 0.4 -0.5 2 1.2 0.5 0 x -0.5 1 0.5 0 -0.6 4 3 2 1 0 -0.6 -20 -15 -10 -5 0 Figure 4: E ∈ (−1.2 0 0. 1) y 3.5 2 1. 1) 67 .5 3 2.5 0 0 1 2 3 4 5 3 2.1 0.5 0 0 1 2 3 4 Figure 2: E ∈ (−1.5 2 1.

1 0.1 0.3 1 0. +∞) Figure 9: E ∈ (1.5 1 0 x 0.25 0 -10 -8 -6 -4 -2 0 0 -1.2 -0.y 2 1 1.25 Figure 8: E ∈ (1.2 0.75 -0.75 1.5 0 -3 -2 -1 0 1 2 3 -1 0 1 2 3 4 Figure 6: E = 1.15 -1 -0.4 0.25 1 0.5 y 0. +∞) Figure 11: r = 0 68 .5 -0.5 Figure 10: E ∈ (1.75 0.8 y 0.5 1 0.05 0 -0.2 0.5 1.25 0 0.5 -0.6 -1 -0.15 -0.1 -0.5 0.05 0.15 0. +∞) y 2 1.5 0 0.25 0. θ = π 1.25 0.05 0 0 0. θ = π Figure 7: E = 1.5 0.

it is easy to see that the dynamics of the system is described by the following ODEs: x = u1 . e3 ) → (f1 . f2 . In addition. u2 ). R ∈ SO(3). u2 ). see [14] for details. y) in the plane corresponding to the frame (e1 . choose an orthonormal frame (f1 . Then orientation of the sphere in the space is determined by the orientation matrix R : (e1 . f2 . Then the state of the system is described by the tuple X = (R. f3 ) attached to the sphere. and the cost functional is t1 X(t1 ) = X1 . f3 ). the problem is to roll the sphere from the ﬁrst conﬁguration to the second one in such a way that the curve in the plane traced by the contact point be the shortest possible.8. e2 . e2 . ˙ The ﬁrst two equations mean that the contact point (x. 0 69 . ˙ ˙ Moreover. We have initial and terminal states ﬁxed: X(0) = X0 . e3 ) in R3 such that the plane is spanned by e1 . while the third equation means that the angular velocity of the rolling sphere is horizontal and perpendicular to (u1 . y) ∈ SO(3) × R2 .5 The plate-ball system Consider a unit 2-dimensional sphere rolling on a horizontal 2-dimensional plane without slipping and twisting. Fix an orthonormal frame (e1 . l= 0 x2 + y 2 dt → min . y) moves in the plane with an arbitrary velocity (u1 . ˙ y = u2 . and the vector e3 is directed upwards (to the half-space containing the sphere). e2 . e2 ). We can assemble the state X to a single 6 × 6 matrix X= R 0 . 1 0 x 0 1 y 0 0 1 0 0 ˙ R = R 0 u1 0 0 u2 −u1 −u2 . Given an initial and a terminal contact conﬁguration of the sphere and the plane. and position of the contact point of the sphere with the plane is given by its coordinates (x. x.

(x. . e1 ] = e2 . E13 − E31 . [e3 . . e3 ] = e1 . 0 1 0 0 0 0 0 0 0 . (56) 0 −1 0 1 0 0 0 0 0 0 operators read as follows: 0 0 0 −1 0 0 . . 5. u2 ) ∈ R2 . ad e4 = ad e5 = 0. (u1 . ωi . . . . ad e2 = −1 0 0 . . ej = δij .denote by G the Lie group of all such matrices for all R ∈ SO(3). we choose the dual basis in the space dual to the Lie algebra: L∗ = span(ω1 . The nonzero adjoint 0 0 0 0 0 1 ad e1 = 0 ad e3 = [e2 . 70 i. 1 0 x 0 1 y 0 0 1 0 0 u1 0 0 u2 0 −u1 −u2 0 . . e2 ] = e3 . . j = 1. ω5 ). E21 − E12 . E56 ). Then the dynamics of the system takes the left-invariant form as follows: ˙ X= = ˙ R 0 0 0 x ˙ 0 0 y ˙ 0 0 0 0 0 R 0 that is. (57) (58) 0 As usual. e1 e2 e3 e4 e5 with the multiplication rules inherited from so(3): [e1 . 0 0 u1 0 0 u2 0 0 0 0 X ∈ G. E46 . The Lie algebra of the Lie group G is L = span(E32 − E23 . y) ∈ R2 . ˙ X = X(u1 (E31 − E13 + E46 ) + u2 (E32 − E23 + E56 )).

2 u 2 1 then ∂h = u1 (e4 − e2 ) + u2 (e5 + e1 ). J= 1 2 t1 0 write elements of the Lie algebra as column vectors: A1 5 . (u2 + u2 ) → min. controllability follows. (58) that 0 0 −u1 0 0 −u2 0 ∂ h u1 u2 0 . ad = ∂a 0 0 a1 ˙ a2 ˙ a3 ˙ a4 ˙ a5 ˙ So the vertical subsystem of the Hamiltonian system of PMP reads a2 a3 a4 a5 = a1 0 0 u1 0 0 u2 0 −u1 −u2 0 0 0 . ∂a and it follows from (56). The Hamiltonian of PMP has the form ν hν (a) = a. (u1 . Since it is symmetric and G is connected. u2 ) ∈ R2 . . X(t1 ) = X1 . 71 . .and elements of the dual space as column vectors: 5 a1 L∗ a = a i ωi = . u1 (e4 − e2 ) + u2 (e5 + e1 ) + (u2 + u2 ). 1 2 notice that we replace the functional l by J as always. . (57). . L A= Ai ei = . X(0) = X0 . i=1 a5 ˙ X = X(u1 (e4 − e2 ) + u2 (e5 + e1 )). . Controllability: multiplication rules (56) imply that the control system has full rank. Existence of optimal controls follows from Filippov’s theorem. i=1 A5 Now we study the plate-ball optimal control problem: X ∈ G.

Then 1 h−1 (a) = u1 (a4 − a2 ) + u2 (a5 + a1 ) − (u2 + u2 ) → max 2 . For these controls. ˙ ˙ 72 . u 2 2 1 (u1 . 2 a3 = 0 = −u1 a1 − u2 a2 . ˙ a3 = −u1 a1 − u2 a2 .u2 )∈R whence u1 = a 4 − a 2 . ˙ a4 = a5 = 0. so a3 = 0. ˙ y = u2 . ˙ Then optimal abnormal controls (u1 . ˙ a4 = a5 = 0.So the whole Hamiltonian system of PMP takes the form: a1 = u1 a3 . y) is a straight line. the corresponding curve (x. ˙ ˙ x = u1 . ˙ But non-constant extremal curves of the functional J satisfy the identity u2 + 1 u2 ≡ const = 0. a1 = 0 = u 1 a3 . ν = −1. ˙ a2 = (a5 + a1 )a3 . Then h0 (a) = u1 (a4 − a2 ) + u2 (a5 + a1 ) → max ∈ R2 . u2 ) are constant. a5 + a1 ≡ 0. the vertical subsystem of the Hamiltonian system of PMP takes the form a1 = (a4 − a2 )a3 . ˙ a2 = u2 a3 . ν = 0. R(t) = R0 exp t 0 u1 u2 0 Now we pass to the normal case. Thus a1 = −a5 ≡ const. ˙ 0 ˙ R = R 0 u1 0 0 u2 −u1 −u2 . 0 Consider ﬁrst the abnormal case. a2 = a4 ≡ const. ˙ a2 = 0 = u 2 a3 .u2 ) whence a4 − a2 ≡ 0. ˙ a3 = −(a4 − a2 )a1 − (a5 + a1 )a2 . and the orientation matrix is 0 0 −u1 0 −u2 . u2 = a 5 + a 1 . u (u1 . Finally.

12. b2 = a 5 + a 1 = u 2 . (4) other works referred to in these notes. in which a4 = r cos ϕ. b2 = sin θ. 41. 33. (3) papers on optimal control for invariant problems on Lie groups [10. 34. 25. Remarks on bibliography The bibliography contains references of several kinds: (1) textbooks on control theory [1. 11. 21. 19. 42. 26. including a survey on the subject [35]. 4. ˙ b3 = a 5 b1 − a 4 b2 . 29. 32. ˙ θ = b3 . that is. ˙ b3 = r sin(ϕ − θ). 31. 14. 36. 3. sub-Riemannian geometry [28]. nonholonomic dynamics [44]. 30]. 40. and diﬀerential geometry and Lie groups [45]. 17. ˙ y = u2 = b2 = sin θ. 16. ˙ Thus we obtain a remarkable result: the contact point of the sphere rolling optimally traces Euler elastica! A description of the corresponding orientation matrix R(t) can be found in [14]. a5 = r sin ϕ. 2 This system has an integral b1 + b2 ≡ const. 24. 5. 39.Introduce the variables b1 = a 4 − a 2 = u 1 . 13. 73 . b3 = a 3 . (2) works on controllability of invariant systems on Lie groups [2. The coordinates of the contact point satisfy the ODEs x = u1 = b1 = cos θ. 37]. 22. then the above system reduces to the following one: ˙ b1 = −b2 b3 . which can be set equal to 1 by 2 homogeneity of the system. 23. 15. 43]. We pass to the polar coordinates b1 = cos θ. ˙ b2 = b 1 b3 . the angle θ satisﬁes the equation of pendulum ¨ θ = −r sin(θ − ϕ).

o e e e 15. Boothby. Jurdjevic and I. In: Geometry in nonlinear control and diﬀerential inclusions.” J. 10. Boothby and E. Hilgert. 525–535 (1982). e [10] U. Acad. Controllability of nilpotent systems.” SIAM J. Geometric control theory. Cambridge University Press. Control Systems. [5] B. Gauthier.H. 2. Yu. Warszawa. [7] W. “Determination of the transitivity of bilinear systems.A. Jurdjevic and I. e [3] V. Hofmann. No. [2] R. 1378–1380 (1950). [8] A. Kupka. Sachkov. Kupka.” SIAM J. Math. J. 20 (1982). Chambrion. Bornard. Bonnard. N.” Ann. Theory. and I.” C. Fourier. Inst. 4. 230. Bonnard. Grenoble 31. “Some remarks about transformation groups transitive on spheres and tori. 8. [6] W. K. Equat. 13. [13] J. No. Soc. 296–307 (1975). No. Equat. Math. T. J. P. Control. Gauthier. Inst. Boscain. Control. 185–190 (1985). “Transitivity of families of invariant vector ﬁelds on the semidirect products of Lie groups. 32. 55. Brockett. Jurdjevic. Amer. Paris. 117–133 (1996). El Assoudi. Lawson. 74 . Henri Poincar´. 1997.” Ann. 151–179 (1981). a [14] V. “On subsemigroups of semisimple Lie groups. and G.” Bull. “Le plan projectif des octaves et les sph`res comme espaces hoe mog`nes. Borel. 547–572 (2002). Ayala Bravo.” J. 265–284 (1972). o e e e SIAM J. Contrˆllabilit´ des syst`mes bilin´aires. 39. 580–586 (1949). [11] R. 1. L. Soc. Borel. 20. Kupka. 79–92 (1981). [15] V.. [16] V. “Control systems on semi-simple Lie groups and their homogeneous spaces. V. [12] J. Banach Center Publications.References [1] A. Sallet.. P. Math. 3. 377–384. R. Springer Verlag. Diﬀer.. Diﬀ. Kupka. Control Theory from the Geometric Viewpoint. Dynam.D. 17. Sci. 212–221 (1979). “A transitivity problem from control theory. Amer. 271. Wilson. Gauthier and G.” Trans. 2004. I. Agrachev.-P. [4] B. “Control systems subordinated to a group action: Accessibility. W. Syst. J. Beitr¨ge Algebra Geometrie. J. 35–46 (1995). 17. Controllability of systems on a nilpotent Lie group. [9] A. Control Optim. 186–211 (1981). “System theory on group manifolds and coset spaces. Jurdjevic. Contrˆlabilit´ des syst`mes bilin`aires.. On the K + P problem for a three-level quantum system: Optimality implies resonance.

Elliptic functions and applications. V. A tour of subriemannian geometries. A. 3. Mittenhuber. Pontryagin. 4. Pergamon Press. A. “Controllability of systems on solvable Lie groups: the generic case”. [29] O..Anzaldo-Meneses. Control Syse tems. Mittenhuber. 61–75 (2001). 8.” Ann. [21] D. 2002.” J. no. Equat. 6) Sub-Riemannian Problem. 1980. 30. 479–501 (1985). New York: Dover.L. Amer. Montgomery and H.V. R.S. 75 . Montgomery. [28] R. No. The Mathematical Theory of Optimal Processes. H. 487–504 (2002). 5. 44. Monroy-P´rez. Dynam. 531–564 (1997).” Proc. On invariant orthants of bilinear systems. 249–272 (1998). [18] D. Control Syst. 2. L. Boltyanskij. [26] F. [27] D. 6. “Controllability of hypersurface and solvable invariant systems. their geodesics and applications. “Controllability of right-invariant systems on solvable Lie groups.. A. 4. Springer-Verlag. 313–329 (1972). [19] J. 1. Sachkov. “Optimal Control on Nilpotent Lie e Groups”. no. Control Syst. Gamkrelidze. [23] F. 1927. D. 1. 473–499 (1999). 1. Math. Jurdjevic and H.. E. J. No. 7.” J. Sussmann. Dynam. Sachkov. 185–216 (2006).” J. Non-Euclidean Dubins problem. n(n + 1)/2) sub-Riemannian geometry. Diﬀ. Love. Dynam. [33] Yu. J. “Maximal subsemigroups of Lie groups that are total. Control Systems. Lawson. No. [30] L. Dynam.[17] V. J.Anzaldo-Meneses. 573–597 (2002). Control Systems. Math. Soc. L. Dyn. The step-2 nilpotent e (n. Dyn. Samelson. Control Systems.F.F. No. No. 8. E. Soc. Anzaldo-Meneses. 4. e [24] F. Control Systems. 454–470 (1943). Edinburgh Math. Dyn. No. 2. 12. 4. 4. J. Control Systems.G. Lawden.. 3. Monroy-P´rez. Dynam. J. J.. 55–67 (1996). 4. Sachkov. 12. “Control systems on Lie groups. Monroy-P´rez. [22] D. Dynam. [32] Yu. “Transformation groups of spheres. A treatise on the mathematical theory of elasticity. Myasnichenko. Nilpotent (3. Mishchenko. No. Control Systems. 2.. no. “Controllability of solvable Lie algebras”. Monroy-P´rez.. “Optimal Control on the Heisenberg Group”. Dynam. J. 137–147 (1998). [25] F. Control Syst. J. Oxford (1964) [31] Yu. [20] A. 453–459 (2000).

Matem. W. “Topology of Lie groups.L. 4. Vershik. 6. Santana. [37] Yu. [39] L. Springer Verlag. 2: 159–217. B. Sachkov.ru/PSI/CPRC/sachkov/public. San Martin. 100. Series on Contemporary Mathematics and Applications. EMS.Ya. R)”. Foresman. Sachkov. Thematical Surveys. Journal of Dynamical and Control Systems. 31–42 (1988). [44] A. Ross. 58. [45] F. 2–37 (1952). “Semigroup actions on homogeneous spaces. 2355–2427.” Bull. Sbornik. L. 1971. Soc. [42] F. Journal of Dynamical and Control Systems. to appear. Exponential mapping in the generalized Dido problem (in Russian). Classiﬁcation of controllable systems on low-dimensional solvable Lie groups. 356. Dynamical Systems-8. Symmetries of Flat Rank Two Distributions and SubRiemannian Structures. 2: 457–494 (2004). 9: 63–90 (2003).” Math.M. 29–54. L. Tonelli. 1–30 (1994). O. Sci. Nauk. Math. San Martin. A. “Regular Elements and Global Controllability in SL(d. G.” Math. 12 (2006). in: Progress in Science and Technology. 2000. A. 5–85 (1987). A. Control Signals Systems. [41] L.. “Invariant control sets on ﬂag manifolds. 41–61 (1993). do Rocio. B. No. 14. : Scott. [43] R. 6 (2000).L.[34] Yu. Moscow. 10 (2004) No. Nonholonomic dynamic systems. Math. Controllability of Invariant Systems on Lie Groups and Homogeneous Spaces. Ill. [35] Yu. Troncoso. 1. 59. Silva Leite and P. http://www. Crouch. [40] L.M. Sachkov. Transactions of the American Mathematical Society. 1. V.M. Gershkovich. Amer. Foundations of diﬀerentiable manifolds and Lie groups. Geometry of distributions and variational problems. English transl: J. San Martin and P.J. v. 1998 (to appear). A. B. A.html [36] Yu. “Controllability on classical Lie groups. All-Russian Institute for Scientiﬁc and Technical Information (VINITI). [38] H. 16. Vol. Warner.” Semigroup Forum.botik. Glenview. Sachkov. Journal of Dynamical and Control Systems. 194. Akad. Control Signals Systems. 76 . Samelson.. Invariant cones and convex sets for bilinear control systems and parabolic type of semigroups.

R). 6 SO(n). 8 GL(n. 11 u(n). C). 17 AΓ . 35 E(n). 5 V K. 26 ¯ Z. 32 77 θ∗ A. 21 OΓ (X). 17 A(X. 7 U(n). 7 Aﬀ + (n). 6 A. 21 int S. 21 A(X). 23 .Index Eij . R). 17 AΓ (X. 20 SL(n. 24 LS(Γ). T ). 8 SL(n. R). 10 sl(n). 45 GL(n. 17 derived series. 10 [V. 38 M (n). 7 M (n. 48 classical notation for left-invariant systems. 12 X T. 17 AΓ (X. 21 intO S. 43 equivalent systems. 32 aﬃne group. 8 complex special linear group. 30 O(n). 54 action of a Lie group on a manifold. 17 completely solvable Lie algebra. W ]. 9 T(n). 21 OΓ . 8 control-aﬃne system. 47 canonical projection. 5 M (n. 11 gl(n). 11 su(n). 16 commutator. ≤ T ). 15 aﬀ(n). 37 complete controllability. 33 θ∗ Γ. 7 attainable set. 6 Sat(Γ). 8 Aﬀ(n). 45 [A. 12 full rank. 6 SU(n). 17 O. 27 Vec M . 20 abnormal case. 16 control-linear system. 8 cl S. 15 compact Lie group. C). 11 so(n). 5 GL+ (n). 37 controllable system. 12 t(n). 11 co(l). T ). 33 eA . 7. 15 Γ. 12 θ. 66 exponential. 26 Euclidean group. 10. 21 e(n). 7 Euler elastica. 24 span(l). 17 canonical coordinates. 12 exp(A). B]. 16 Γ1 ∼ Γ2 . 21 AΓ (X). 44 complex general linear group. 21 ad A. C). 29 Lie(Γ).

17 reachable set for time not greater than T ≥ 0. 6 permutation-irreducible matrix. 16 left-invariant vector ﬁeld. 16 Pontryagin Maximum Principle. 7 trivialization. 43 singular control. 39 simple Lie group. 17 reachable set for time T . 32 ideal. 40 polysystem. 45 semisimple Lie algebra. 14 Lie algebra. 49 reachable set. 29 linear Lie group. 27 semi-direct product. 6 linear system. 16 transitive action. 20 orthogonal group. 47 orbit. 39 induced system. 17 Hamiltonian. 9 sub-Riemannian problem. 39 simple Lie algebra. 39 semisimple Lie group. 10 Lie algebra of Lie group. 57 homogeneous space. 44 normal case.general linear group. 8 78 . 48 Hamiltonian vector ﬁeld. 18 right-invariant vector ﬁeld. 43 solvable Lie group. 65 solvable Lie algebra. 48 trajectory. 43 special linear group. 5 global controllability. 36 local controllability. 25 nilpotent Lie algebra. 10 Lie Algebra Rank Condition. 23 Lie bracket. 6 special unitary group. 32 triangular group. 15 Lie group. 21 optimal control problem. 5 Lie saturate. 36 left-invariant control system. 6 special orthogonal group. 40 permutation-reducible matrix. 39 simply connected space. 54 normally attainable point. 48 Heisenberg group. 17 realiﬁcation. 48 tautological 1-form. 33 induced vector ﬁeld. 36 symplectic form. 7 right-invariant control system. 17. 49 unitary group. 10 Kalman condition. 33 Jacobi identity. 55 symmetric system. 17 saturate.

CursoNL2-UNAM2009V5

Control Lineal y No Lineal de Un Levitador Magnetico

Tarea-Continuo

péndulo invertido

calculo de variaciones

Classical Mechanics Solutions 2nd Edition Goldstein

Lyapunovfinish

- Read and print without ads
- Download to keep your version
- Edit, email or read offline

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

CANCEL

OK

You've been reading!

NO, THANKS

OK

scribd

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->