Manifolds, Vector Fields & Flows

Harry G. Kwatny
Department of Mechanical Engineering & Mechanics Drexel University

Outline

From Flat to Curved State Space Why? Manifolds Maximum rank condition Regular manifolds Tangent space & tangent bundle Differential map Vector fields & flows Lie bracket

From Flat to Curved State Space

Global models may require it:

Mechanical systems spatial rotation Electric power systems DAE description

Correct local approximations at least require acknowledging it Even if not required it may have conceptual benefits Computation often involves flat local approximations but this may not be necessary (e.g. quaternions vs. Euler angles)

Representations of Surfaces

Explicit Implicit Parametric

y = g ( x)
f ( x, y ) = 0

x = h1 ( s ), y = h2 ( s ), s U R
y

y = 1 x2 x2 + y 2 1 = 0 x = cos s, y = sin s, s [0, 2 )

x

Examples: Parametrically defined manifolds

cos t sin u t [0, 2 ), u [0, ) sin sin f = t u cos u

cos t (3 + cos u ) t [0, 2 ), u [0, 2 ) f = sin t (3 cos u ) + sin u

Definition - Manifold
An m-dimensional manifold is a set M together with a countable collection of subsets U i M and one-to-one mappings onto open subsets of R m , i : U i Vi , with the following properties: the pair (U i , i ) is called a coordinate chart the coordinate chartes cover M , on the overlap of any pair of charts the composite map is a smooth function f = j i1 : i (U i U j ) j (U i U j ) if p U i and q U j are distinct points of M , then there are neighborhoods W of i ( p ) Vi and U of j ( q ) V j such that
1 i1 (W ) j (U ) =

Definition: Manifold
Ui

Uj

i
Vi

Vj

f = j i1

Example: Planet Earth

Example: Circle
The unit circle S1 = {(x,y) | x2+y2=1} can be viewed as a one dimensional manifold with two coordinate charts. Define the charts U1=S1-{(-1,0)} and U2=S1-{(1,0)}. Now we define the coordinate maps by projection as shown in the figure.

{(x,y) | x2 +y2 =1}

(x,y)

(x,y)

{(x,y) | x2 +y2 =1}

(-1,0)

(1,0)

Submanifold & Immersion

Definition: Let F: Rm Rn be a smooth map The rank of F at x0 Rm is the rank of the Jacobian DxF at x0. F is of maximal rank on S Rm if the rank of F is maximal for each x0S. Definition: A (smooth) submanifold of Rn is a subset MRn, together with a smooth one-to-one map :RmM which satisfies the maximal rank condition everywhere, where the parameter space is and M = () is the image of . If the maximal rank condition holds but the mapping is not one-to-one, then M is an immersion. Note that a submanifold of the space Rn is a parametrically defined surface.

Pathologies

Regular Submanifold
Definition: A regular submanifold N of Rn is a submanifold parameterized by a smooth mapping such that maps homeomorphically onto its image, i.e., for each xN there exists neighborhoods U of x in Rn such that -1[UN] is a connected open subset of the parameter space
M= R
2

N=R

p = lim t (t)

immersion

submanifold

regular submanifold (imbedding)

Implicitly Defined Regular Manifolds

Proposition: Consider a smooth mapping F: Rm Rn, nm. If F is of maximal rank on the set S = {x: F(x)=0}, then S is a regular, (m-n)-dimensional submanifold of Rm. Example
f ( x, y ) = ( x 2 + y 2 1) y
2 2 Df = xy + x + y 2 1 3
-1.5 -1 -0.5 -0.5 y 1

0.5

x 0.5 1 1.5

singular points: ( 1,0), (0, 1/ 3)

-1

The Tangent Space

Definition: Let p: RM be a Ck k 1 map so that p(t) is a curve in M. The tangent vector v to the curve p(t) at the point p0 = p(t0) is defined by

p(t ) p(t 0 ) v = p(t 0 ) = lim t t 0 t t0 The set of tangent vectors to all curves in M passing through p0 is the tangent space to M at p0, denoted TM p0.
R v

p0
M

p (t )

Tangent Space / Implicit Manifold

If M is an implicit submanifold of dimension m in Rm+k, i.e., F: Rm+kRk, M = {xRm+k | F(x)=0} and DF satisfies the maximum rank condition on M, Then TMx is the ker Dx F ( x) (translated, of course to the point x). That is TMp is the tangent hyperplane to M at p.

Im

LMFG F IJ OP NH x K Q
T

TM x = Ker

LM F OP N x Q

Rm+k

Tangent Vectors
Definition : The components of the tangent vector v to the curve p ( t ) in M in local coordinates (U , ) are the m numbers v1 , , vm where vi = d ( p ( t ) ) / dt. Consider the map F : M R. Let y = f ( x ) , x (U ) R m denote the realization of F in the local coordinates (U , ) . Again, p ( t ) denotes a curve in M with x ( t ) its image in R m . Then the rate of change of F at a point p on this curve is df f = v1 + dt x1 + vm f xm

Tangent Vectors as Derivations

x (t )

p(t )

Rm

vm ] is uniquely determined by the The tangent vector v = [ v1 action of the directional derivative operator (called a derivation)

v = v1 + x1

vm x m

Natural Basis
0 v = 1 i th v = xi 0

Definition: The set of partial derivative operators constitute a basis for the tangent space TMp for all points pUM which is called the natural basis. The natural coordinate system on TMp induced by (U,) has basis vectors that are tangent vectors to the coordinate lines on M passing through p.

Tangent Bundle
Definition: The union of all the tangent spaces to M is called the tangent bundle and is denoted TM,

TM = pM TM p
Remark: The tangent bundle is a manifold with dim TM = 2 dim M. A point in TM is a pair (x,v) with xM, vTMx. If (x1,..,xm) are local coordinates on M and (v1,..,vm) components of the tangent vector in the natural coordinate system on TMx, then natural local coordinates on TM are ( x1 , , x m , v1 , , v m ) = ( x1 , , x m , x1 , , x m ) Recall the natural unit vectors on TMx are v1 = x1 ,, v m = x m

Summary
z2

1 1, continous (homeomorphic)
x3

x = ( z1 , z2 )

Regular Manifold

Parametrically defined Implicitly defined

z1 x1

x2

Tangent Space, Tangent Vector, Tangent Bundle

f M = {x R 3 f1 ( x1 , x2 , x3 ) = 0} rank 1 x1

f1 =1 on M x2

Mechanical System State Space

A mechanical system is a collection of mass particles which interact through physical constraints or forces. A configuration is a specification of the position for each of its constituent particles. The configuration space is a set M of elements such that any configuration of the system corresponds to a unique point in the set M and each point in M corresponds to a unique configuration of the system. The configuration space of a mechanical system is a differentiable manifold called the configuration manifold. Any system of local coordinates q on the configuration manifold are called generalized coordinates. The generalized velocities q are elements of the tangent spaces to M, TMq. The state space is the tangent bundle TM which has local coordinates (q, q ) .

Example: Pendulum
TM M=S 1 TM

Differential Map
F
d dt
F(p ) p

d dt

an arbitray curve ( t ) on M passing through point p

(t )

(t ) = F ( (t ))

maps into ( t ) on N passing through point F ( p )

R

Given the map F : M N, the differential map is the induced mapping F* : TM p TN F ( p ) that takes tangent vectors into tangent vectors.

Differential Map ~ local coordinates

In local coordinates, the chain rule yields

d F d F = v = v x dt x dt
The map F* is also denoted dF The Jacobian is the representation of the differential map in local coordinates

Vector Fields
Definition: A vector field v on M is a map which assigns to each point pM, a tangent vector v(p)TMp. It is a Ck-vector field if for each pM there exist local coordinates (U,) such that each component vi(x), i=1,..,m is a Ck function for each x(U). Definition: An integral curve of a vector field v on M is a parameterized curve p=(t), t(t1,t2)R whose tangent vector at any point coincides with v at that point.

Integral Curves
In local coordinates (U,), the image of an integral curve
x(t ) = (t ) satisfies the ode

dx = v( x) dt
v
V
x (t )

p(t )

Rm

Flow
Definition: Let v be a smooth vector field on M and denote the parameterized maximal integral curve through pM by (t,p) and (0,p)=p. (t,p) is called the flow generated by v. Properties of flows: satisfies ode
d ( t , p ) = v ( ( t , p ) ) , ( 0, p ) = p dt

semigroup property t2 , ( t1 , p ) = ( t1 + t2 , p )

Exponential Map
We will adopt the notation

etv p := ( t , p )
The motivation for this is that the flow satisfies the three basic properties ordinarily associated with exponentiation from properties of (t,p).

e0v p = p d tv e p = v ( etv p ) dt e( t1 +t2 ) v p = et1v et2 v p

boundary condition differential equation semi-group property

Series Expansion Along Trajectory

Suppose x ( t ) satisfies x = v ( x ) , x ( 0 ) = x0 . Let f : R m R p . 2 1 d2 d f ( x ( t ) ) = f ( x0 ) + f ( x ( t ) ) t + 2 f ( x ( t ) ) t + 2 dt dt t =0 t =0 d f f + f ( x (t )) = v ( x ) = v1 ( x ) x x1 dt f + vm ( x ) = v ( f )( x ) xm

f d2 2 = = v f x t v x v x ( ) ( ) ( ) ( f )( x ) ( ) 2 x x dt 1 2 f ( x ( t ) ) = f ( x0 ) + v ( f )( x0 ) t + v ( f )( x0 ) t 2 + 2

Series Representation of Exp Map

For f a scalar or vector, we can derive the Taylor expansion of f(x(t)) about t=0
k t f ( etv x ) = v k ( f )( x ) k =0 k !

Choose f(x)=x, to obtain

k t etv x = v k ( x )( x ) k =0 k !

Example: scalar linear fields

dx = 1 (t , x ) = x + t v =1 dt 2 t 1 (t , x ) = e x x = 1 + t + t 2 2 + x 2 x

x = x+t

dx v=x = x (t , x ) = et x dt k k tx t t (t , x ) = e x x = v k ( x )( x ) = x = et x k =0 k ! k =0 k !

Example: general linear field

a1 a + + an x v ( x ) = Ax = 2 x v = a1 x x1 xn an 1 0 0 v ( x ) = a1 x + + an x = Ax 0 0 1 v 2 ( x ) = v ( Ax ) = Av ( x ) = A2 x v k ( x ) = v ( Ak 1 x ) = Ak 1 v ( x ) = Ak x

tk k tk k e x = v ( x )( x ) = A x = e At x k =0 k ! k =0 k !
tv

Example: Affine Field

a1 b1 a b v ( x ) = Ax + b = 2 x + 2 v = ( a1 x + b1 ) + + ( an x + bn ) x1 xn an bn 1 0 0 v ( x ) = ( a1 x + b1 ) + + ( an x + bn ) = Ax + b 0 0 1 v k ( x ) = v ( Ak 1 x + Ak 2 b ) = Ak 1 v ( x ) = Ak x + Ak 1b
k k k t t t etv x = v k ( x )( x ) = Ak x + Ak 1b = e At x + e At A1b k =0 k ! k =0 k ! k =0 k !

Examples, Contd
v ( x ) = x3 x = x3 Exact solution: x0 2 4 2 20 6 3 35 8 4 63 10 5 = (1 x0 + x (t ) = t + 12 x t x t x t 8 0 8 0 8 0 8 x0 t + 2 1 + 2 x0 t via exponential map: x (t ) = e
x3 x 2 4 2 20 6 3 35 8 4 63 10 5 + x0 = (1 x0 t + 12 x t x t x t 0 0 0 8 8 8 8 x0 t +

)x

)x

Lie Derivative
Definition: Let v(x) denote a vector field on M and F(x) a mapping from M to Rn, both in local coordinates. Then the Lie derivative of order 0,,k is
k 1 L k v = = L0 F F , L ( F ) v ( ) v v x

With this notation we can write

( F ) ( x) = L ( F ) ( x)
k v

Example: Exponential Map of a Nonlinear Field

v1 ( x ) v x ( ) 2 v = v1 ( x ) + + vn ( x ) v ( x) = x1 xn vn ( x ) v ( x ) = Lv ( x ) = v ( x ) v k ( x ) = v ( Ak 1 x ) = Ak 1 v ( x ) = Lk v ( x)
k k t t tv k k e x = v ( x )( x ) = Lv ( x ) k =0 k ! k =0 k !

v 2 ( x ) = v ( v ( x ) ) = L2 v ( x)

Lie Bracket
Definition: If v,w are vector fields on M, then their Lie bracket [v,w] is the unique vector field defined in local coordinates by the formula

Property:

w v v w [v, w] = x x
dw ( ( t , x ) ) dt
t =0

= [ v, w] x

The rate of change of w along the flow of v

Lie Bracket Interpretation

Let us consider the Lie bracket as a commutator of flows. Beginning at point x in M follow the flow generated by v for an infinitesimal time which we take as for convenience. This takes us to point

y = exp( v) x
Then follow w for the same length of time, then -v, then -w. This brings us to a point given by

( , x ) = e

Lie Bracket Interpretation Continued

u -v -w w y z

(, x)

[v,w] v x

d 0 + , x = [ v, w] x d

Summary

Definition of regular manifold

Implicitly defined & parametrically defined

Local coordinates Tangent space, vector field, integral curve Differential map, exponential map Lie derivative Lie bracket