<< Chapter < Page Chapter >> Page >
Description of vector spaces, subspaces, bases and spans.

Vector spaces

Definition 1 A linear vector space ( X , R , + , · ) is given by a signal space X (called vectors), a set of scalars R , an addition operation + : X × X X , and a multiplication operation · : R × X X , such that:

  1. X forms a group under addition:
    1. x , y X ! x + y X , (closed under addition)
    2. 0 X such that 0 + X = X + 0 = X . (additive identity)
    3. x X y X such that x + y = 0 , (additive inverse)
    4. x , y , z X x + ( y + z ) = ( x + y ) + z . (associative law)
  2. Multiplication has the following properties: for any x , y X and a , b R :
    1. a · x X , (closure in X under multiplication)
    2. a · ( b · x ) = ( a · b ) · x , (compatibility)
    3. ( a + b ) · x = a · x + b · x , (distributive law over R )
    4. a · ( x + y ) = a · x + a · y . (distributive law over X )
  3. The set R has the following properties:
    1. There exists 1 R s.t. 1 · x = x x X , (multiplication identity)
    2. There exists 0 R s.t. 0 · x = 0 x X . (multiplicative null element)

Example 1 Here are some examples of vector spaces:

  • X = R n (space of all vectors of length n ) over R = R is a vector space.
  • X = C n ( C is complex numbers) over R = C is a vector space.
  • X = R n over R = C is not a vector space, because closure in X under multiplication is not met.
  • X = C [ T ] (continuous functions in T ) over R = R is a vector space.

Subspaces

Definition 2 A subset M X is a linear subspace of X if M itself is a linear vector space. Note that, in particular, this implies that any subspace M must obey 0 M .

Example 2 Here are some examples of subspaces:

  • In X = R 2 over R = R , any line that passes through the origin is a subspace of X :
    M = ( x , y ) R 2 such that y x = c .
  • In X = C [ T ] over R = R , the followings are subspaces of X :
    M 1 = { f ( x ) = a x 2 + b x + c : a , b , c R } M 2 = { f ( x ) : f ( x 0 ) = 0 } .
    In contrast, the set M 3 = { f ( x ) : f ( x 0 ) = a 0 } is not a subspace.

Proposition 1 If M and N are subspaces of X , then M N is also a subspace.

Proof: We assume that M and N hold properties of linear vector space, and show that so does M N :

  1. x , y M N x , y M x + y M x , y N x + y N x + y M N
  2. M linear vector space 0 M N linear vector space 0 N 0 M N
  3. x M N x M y M s.t. x + y = 0 x N y N s.t. x + y = 0 y M N

The other properties are shown in a similar fashion.

Definition 3 A vector x X , where ( X , R , + , · ) is a vector space, is a linear combination of a set { x 1 , x 2 , ... , x n } X if it can be written as x = i = 1 n a i · x i , a i R . The set of all linear combinations of a set of points { x 1 , x 2 , ... , x n } builds a linear subspace of X .

Example 3 Q = i = 0 2 a i x i is a linear subspace of ( C [ T ] , R , + , · ) containing the set of all quadratic functions, as it corresponds to all linear combinations of the set of functions { x 2 , x , 1 } .

Bases and spans

Definition 4 For the set S = { x 1 , x 2 , ... , x n } X , the span of S is written as

[ S ] = span ( S ) = x : x = i = 1 n a i x i , a i R .

Example 4 The space of quadratic functions Q is written as Q = [ S 1 ] , with S 1 = { x 2 , x , 1 } . The space can also be written as [ S 2 ] with S 2 = { 1 , x , x 2 - 2 } ) , i.e. [ S 1 ] = [ S 2 ] . To prove this, we need to show [ S 2 ] [ S 1 ] and [ S 1 ] [ S 2 ] . For the former case we have

x = a 1 + a 2 x + a 3 ( x 2 - 2 ) = ( a 1 - 2 a 3 ) + a 2 x + a 3 x 2 ,

which means that every element that can be spanned by S 2 , can also be spanned by S 1 , and hence [ S 2 ] [ S 1 ] . The latter case can be shown in a similar manner.

Definition 5 A set S is a linearly independent set if

i = 1 n a i x i = 0 a i = 0 , i { 1 , 2 , ... , n } .

Otherwise, the set S is linearly dependent .

Definition 6 A finite set S of linearly independent vectors is a basis for the space X if [ S ] = X , i.e. if X is spanned by S .

Definition 7 The dimension of X is the number of elements of its basis | S | . A vector space for which a finite basis does not exist is called an infinite-dimensional space .

Theorem 1 Any two bases of a subspace have the same number of elements.

Proof: We prove by contradiction: assume that S 1 = { x 1 , ... , x n } and S 2 = { y 1 , ... , y m } , m > n , are two bases for a subspace X with different numbers of elements. We have that since y 1 X it can be written as a linear combination of S 1 :

y 1 = i = 1 n a i x i .

Order the elements of S 1 above so that a 1 is nonzero; since y 1 must be nonzero then at least one such a i must exist. Solving the above equation for x 1 yields

x 1 = 1 a 1 y 1 - i = 2 n a i x i .

Thus { y 1 , x 2 , x 3 , ... , x n } is a basis, in terms of which we can write any vector of the space X , including y 2 :

y 2 = b 1 y 1 + i = 2 n b i x i .

Since y 1 , y 2 are linearly independent, at least one of the values of b i , i > 2 , must be nonzero. Sort the remaining x i so that b 2 is nonzero. Solving for x 2 results in

x 2 = 1 b 2 y 2 - b 1 b 2 y 1 + i = 3 n b i b 2 x i .

Therefore, { y 1 , y 2 , x 3 , ... , x n } is a basis for X . Continuing in this way, we can eliminate each x i , showing that { y 1 , y 2 , ... , y n } is a basis for X . Thus, we have y n + 1 = i = 1 n c i y i , or equivalently:

c n + 1 y n + 1 + i = 1 n c i y i = 0 with c n + 1 = - 1 .

As a result, S 2 is linearly dependent and is not a basis. Therefore, all bases of X must have the same number of elements.

Basis representations

Having a basis in hand for a given subspace allows us to express the points in the subspace in more than one way. For each point x [ S ] in the span of a basis S = { S 1 , S 2 ... S n } , that is,

x = i = 1 n a i S i ,

there is a one-to-one map (i.e., an equivalence) between x [ S ] and a = { a 1 , ... , a n } R n , that is, both x and a uniquely identify the point in S . This is stated more formally as a theorem.

Theorem 2 If S is a linearly independent set, then

i = 1 n a i S i = i = 1 n b i S i

if and only if a i = b i for i = 1 , 2 ... n .

Proof: Theorem 1 states that the scalars { a 1 , ... , a n } are unique for x . We begin by assuming that indeed

i = 1 n a i S i = i = 1 n b i S i .

This implies

i = 1 n a i S i - i = 1 n b i S i = 0 , i = 1 n ( a i - b i ) S i = 0 .

Since the elements of S are linearly independent, each one of the scalars of the sum must be zero, that is, a i - b i = 0 and so a i = b i for each i = 1 , ... , n .

Example 5 (Digital Communications) A transmitter sends two waveforms:

S 1 ( t ) = 2 / T cos ( 2 π f c t ) t [ 0 , T ] if bit 1 is transmitted,
S 0 ( t ) = 2 / T sin ( 2 π f c t ) t [ 0 , T ] if bit 0 is transmitted.

The signal r ( t ) recorded by the receiver is continuous, that is, r ( t ) C [ T ] . Assuming that the propagation delay is known and corrected at the receiver, we will have they the received signal must be in the span of the possible transmitted signals, i.e., r ( t ) span ( S 1 ( t ) , S 0 ( t ) ) . One can check that S 1 ( t ) and S 2 ( t ) are linearly independent. Thus, one can use a unique choice of coefficients a 0 and a 1 that denote whether bit 0 or bit 1 is transmitted and contain the amount of attenuation caused by the transmission:

r ( t ) = a 1 S 1 ( t ) + a 0 S 0 ( t ) .

The uniqueness of this representation can only be obtained if the transmitted signals S 0 ( t ) and S 1 ( t ) are linearly independent. The waveforms above are used in in phase shift keying (PSK); other similar examples include frequency shift keying (FSK) and quadrature amplitude modulation (QAM).

Get Jobilize Job Search Mobile App in your pocket Now!

Get it on Google Play Download on the App Store Now




Source:  OpenStax, Introduction to compressive sensing. OpenStax CNX. Mar 12, 2015 Download for free at http://legacy.cnx.org/content/col11355/1.4
Google Play and the Google Play logo are trademarks of Google Inc.

Notification Switch

Would you like to follow the 'Introduction to compressive sensing' conversation and receive update notifications?

Ask