Differential Geometric Approach • Introduction Vector field : A mapping f : D R n where D R n is a domain, is said to be a vector field. (n - dim column ve ctor) Covector field : A transpo se of a vector field is said to be a covector field. (n - dim row vector) n Inner Product : w, f w( x) f ( x) wi ( x) fi ( x) i 1 where w is a covector field and f is a vector field. Differential(gradient) : Let h : D R. The differenti al of h is a covector field, i.e., dh hx [ xh1 (h) h x n ] 10-27 Lie Derivative Lie derivative: Let h : D R and f : D R n . The Lie derivative of h with respect f or along f , is defined by L f h( x) hx f ( x) the directional derivative of h along the direction of f Let : L0f h( x) h( x) Lif h( x) L f ( Lif1h( x)) ( Lif1h( x)) f ( x) Ex: x f ( x) g ( x)u y h( x ) scalar function If Lg Lif h( x) 0 for 0 i r 2 r 1 g f L L h( x ) 0 x D0 10-28 Lie Derivative (Continued) then the system has relative degree r in D0 y hx h( f gu ) L f h Lg hu y r ( Lrf1h) x ( Lrf1h)( f gu ) Lrf h Lg Lrf1hu 0 Lie bracket (Lie product) : Let f and g be two vector fields on D R n . Then Lie bracket of f and g , [ f , g ] is a vector field g f defined by [ f , g ]( x) x f ( x) x g ( x) gf fg L f g Lg f Let ad g ( x) g ( x) 0 f where g f x x , are Jacobian matrix ad f g ( x) [ f , g ]( x) ad kf g ( x) [ f , ad kf 1 g ]( x) 10-29 Example Ex: x2 0 Let f ( x) , g x sin x1 x2 1 x2 1 0 0 0 0 Then [ f , g ] 1 0 sin x1 x2 cos x1 1 x1 x1 ad f g x x 1 2 ad 2f g [ f , ad f g ] x2 1 x1 0 1 0 1 1 sin x1 x2 cos x1 1 x1 x2 x1 2 x2 x1 x2 sin x1 x1 cos x1 If f and g are constant [ f , g ] 0 10-30 Lemma Lemma: Lie brackets have the following properties (i) Bilinearity : Let f1 , f 2 , g1 , g 2 be vector fields and 1 , 2 be real numbers. Then [ 1 f1 2 f 2. , g1 ] 1[ f1 , g1 ] 2 [ f 2 , g1 ] [ f1 , 1 g1 2 g 2 ] 1[ f1 , g1 ] 2 [ f1 , g 2 ] (ii) Skew commutativity : [ f , g ] [ g , f ] (iii) Jacobi identity : If f and g are vector fields and h is a real valued function. Then L[ f , g ]h( x) L f Lg H ( x) Lg L f H ( x) i.e., h [ f , g ] ( Lg h) f ( L f h) g Proof: See Chapter 6.2 in Applied Nonlinear Control. 10-31 Diffeomorphism Diffeomorphism : A mapping T : D R n is a diffeomorp hism if it is invertible on D, i.e., T 1 ( x) such that T 1 (T ( x)) x for all x D, and T ( x), T 1 ( x) are continuous ly differenti able. Lemma: If the Jacobian matrix T x is nonsingula r at a point x0 D, then T ( x) defines a local diffeomorp hism in a subregion of D. T is a global diffeomorp hism if it is a diffeomorp hism on R n and T (Rn ) Rn. 10-32 Example Lemma: T is a global diffeomorp hism iff (i) Tx is nonsingula r for all x R n (ii) lim T ( x) x Ex: e x1 Let f ( x) x 1 x2e Then f x e x1 x 1 e x2 0 f det 1 0, x1 x e x R 2 However f ( R 2 ) R 2 (Note that f1 e x1 0, x1 ) 10-33 Distribution Distribution : Let f1 ,, f k be vector fields on D R n . At any fixed point x D, f1 ( x), f 2 ( x),, f k ( x) are vectors in R n and ( x) span{ f1 ( x), f 2 ( x),, f k ( x)} is a subspace of R n . To each point x R n , we assign a subspace ( x). We refer to this assignment by span{ f1 , f 2 ,, f k } which we call a distributi on. In other wor ds, is the collection of all vector spaces ( x) for x D. Note that dim( ( x)) rank { f1 ( x), f 2 ( x),, f k ( x)} may vary with x. If span{ f1 ,, f r } where { f1 ,, f r } are linearly independen t for all x D, then dim( ( x)) r for all x D. Then every g can be expressed as r g ( x) ci ( x) f i ( x), ci ( x) : smooth function i 1 10-34 Involutive distribution Involutive distribution : A distributi on is involutive if g1 and g 2 [ g1 , g 2 ] Ex: Let D R 3 , span{ f1 , f 2 } where 2 x2 1 f1 1 , f 2 0 0 x2 dim ( ( x)) 2 0 f f [ f1 , f 2 ] 2 f1 1 f 2 0 x x 1 [ f1 , f 2 ] iff rank{ f1 ( x), f 2 ( x), [ f1 , f 2 ]( x)} 2, x D. 2 x2 However rank{ f1 , f 2 ,[ f1 , f 2 ]} rank 1 0 Hence is not involutive 1 0 x2 0 0 3, x D. 1 10-35 Codistribution & Complete Integrability Codistribution : ( x) {w ( R n )* : w, v 0, v ( x)} where ( R n )* is the n - dimensiona l space of row vectors. Complete Integrability : Let be a nonsingula r distributi on on D, generated by f1 , , f r . Then is said to be completely integrable if for each x0 D, there exists a neighborho od N of x0 and n r real valued smooth functions h1 ( x), , hn r ( x) such that h1 ( x), , hn r ( x) satisfy t he PDE h j f i ( x) 0, 1 i r , 1 j n r x and the covector fields h j ( x) are linearly independen t for all x D, i.e., span{h1 , , hn r } A key result from differenti al geometry is Frobenius theorem which states that a nonsingula r distributi on is completell y integrable iff it is involutive . 10-36 Input-state linearization • Input-state linearization Consider the SI system x f ( x) g ( x)u 1 where f , g are smooth ve ctor fields Note that 1 are said to be linear in control. Definition: A single input nonlinear system in the form above, with f ( x), g ( x) being smooth ve ctor fields in R n , is said to be input state linearizab le if there exists a region in R n , a diffeomorp hism T : R n and a nonlinear control law u ( x) ( x)v 2 such that the new state variable s z T ( x) and the new input v satisfy a linear ti me invariant relation z Az bv 10-37 Input-state linearization (Continued) 0 0 where A 0 0 The new state 1 0 0 0 0 0 1 0 0 , b 0 0 1 0 1 0 0 0 z is called the linearizin g state, 2 the linearizin g control law. is called Question: Can all nonlinear state eq. of 1 be input - state linearizab le? If not, when do such linearizat ions exist? Theorem: The nonlinear system 1 with f ( x), g ( x) being smooth ve ctor fields, is input - state linearizab le iff there exists a region such that the following conditions hold : 10-38 Input-state linearization (Continued) the vector fields {g , ad f g , , ad nf 1 g} are linearly independent in . the distribution D span{g , ad f g ,, ad nf 2 g} is involutive in . P roof: See Ch.13.3 T he first condition can be interprete d as a controllability condition for nonlinear system. For linear system [ g , ad f g , , ad nf 1 g ] becomes [b, Ab, , An 1b] . It is easy to show that if a system's linear approximation in a closed connected region in R n are all controllable, then under mild smoothness assumption, the system can be driven from any point in to any point in . However, a nonlinear system can be controllable while its linear approximation is not. T he involutivity condition is trivially satisfied for linear systems which have constant vector fields. 10-39