Let a,b∈En. The vectors a,b are orthonormal if the following conditions are satisfied:
The vectors are normalised, i.e.: ∥a∥=1and∥b∥=1.
The vectors are orthogonal, i.e.: ⟨a,b⟩={1,0,a=ba=b
Remark
For an orthogonal matrix A∈En×n of the form A=(a1∣...∣an), the column vectors a1,...,an are pairwise orthonormal.
Linear Maps
Definition
A map f:X→Y is called injective if
∀x1,x2∈X:x1=x2⇒f(x1)=f(x2).
(In words: different elements of X are mapped to different images in Y.)
Definition
A map f:X→Y is called surjective if
∀y∈Y∃x∈X:f(x)=y.
(In words: every element of Y is "hit" by f.)
Definition
A map f:X→Y is called bijective if
∀y∈Y∃!x∈X:f(x)=y.
(In words: every element of Y is "hit" by fexactly once.)
Definition
A map F:V→W between E-vector spaces V and W is called linear (more precisely a homomorphism of E-vector spaces) if ∀v,w∈V and ∀λ∈E:
L1F(v+w)=F(v)+F(w) L2F(λv)=λF(v)
These two conditions can be combined into one:
LF(v+λw)=F(v)+λF(w).
Remarknotation
For F:V→W linear, F∈Hom(V,W).
Remark
It is common to sharpen the notion of homomorphism:
F∈Hom(V,W) and bijective ⇔isomorphism (notation: V=W)
F∈Hom(V,W) and V=W⇔endomorphism (notation: F∈End(V))
F∈End(V) and bijective ⇔automorphism
Moreover: (i) ⇔∃G:W→V linear such that F∘G=idW, G∘F=idV, i.e.
∀w∈W:∀v∈V:F(G(w))=wG(F(v))=v
Remark
Let M(F),M(G) be the representation matrices of F:V→W isomorphic and its inverse map G=F−1:W→V and let V,W be finite-dimensional vector spaces, i.e. dim(V)<∞ and dim(W)<∞. Then bijectivity of F means that
dim(V)=dim(W)
M(F)⋅M(G)=M(G)⋅M(F)=𝟙dim(V)⇔M(F)=(M(G))−1
Definition
For every basis B={v1,...,vn} of V there is exactly one isomorphism:
ϕB:En→V,(x1,...,xn)↦ϕB(x1,...,xn)=k=1∑nxkvk=x1v1+...+xnvn with ϕB(ei)=vi.
(In words: ϕB assigns to x its coordinates with respect to the basis B.)
Definition
Let F∈Hom(V,W).
Im(F):=F(V)={F(v)∣v∈V}⊂W is a subspace of W and is called image(F) or Im(F).
ker(F):={v∈V∣F(v)=0}⊂V is a subspace of V and is called ker(F).
Satz
Let F:V→W be linear and V, W vector spaces. Then:
F(0)=0, zero is always mapped to zero
F surjective ⇔Im(F)=W⇔dim(Im(F))=dim(W)
F injective ⇔ker(F)={0}⇔dim(ker(F))=0
F is an isomorphism ⇔dim(V)=dim(W)=rank(F)
Satz
If f,g are linear maps ⇒f∘g is a linear map.
Satz
If f,g are linear maps, then the function F:=f±g arising from the linear combination of f,g is again a linear map.
Definition
The rank of the linear map F is defined as:
rank(F)=dim(Im(F)).
Remark
The rank of the linear map F equals the rank of its representation matrix M(F). We have: rank(F)=dim(Im(F))=rank(M(F))=rank(M(F)T)
Remark
Row rank = column rank: rank(M(F))=rank(M(F)T) Caution: in general: column space = row space
Satz
Let V and W be two finite-dimensional subspaces of a larger vector space (finite-dimensional ⇔dim(V)=n<∞ and dim(W)=k<∞) and let f:V→W be linear; then the following dimension formulas hold:
dim(V+W)=dim(V)+dim(W)−dim(V∩W)
n=dim(V)=dim(ker(f))+dim(Im(f))
Remarkproperties of linear maps
Let V,W be E-vector spaces and B={v1,...,vn} a basis of V. Let F:V→W be linear.
Im(F)=span(F(v1),...,F(vn)), i.e. F is uniquely defined by its values on the basis vectors
If F is injective and v1,...,vn∈V are linearly independent, then F(v1),...,F(vn)∈Im(F) are linearly independent
dim(V)=dim(W)<∞ and F injective ⇒F is bijective!
Example 2
Let F:R→R,x↦5x−1.
F is not linear, since F(0)=−1=0.
Remark
Let f∈Hom(V,W) and let the vector spaces V,W be finite-dimensional. The image of f is spanned by the columns of M(f), i.e. Im(f)=span{columns of M(f)}.
Remark
Let F∈Hom(V,W) with corresponding representation matrix M(F). Then:
We can read off dim(Im(F))=rank(F)=rank(M(F))=2 directly from M(F). Thanks to the dimension formula we know that the basis of the kernel of F contains one basis vector.
Below we compute the solution set of M(F) (or M(F)), which is at the same time ker(F)=ker(M(F))=ker(M(F)), since we are solving the linear system M(F)x=0.
(As you saw in the exercise session, we can also "guess" the vectors in the kernel with the help of M(F).)
M(F)M(F)⇒row 3t:=x3,t∈Rfree parameter⇒row 2−2x2+5x3=0⇔t=x3−2x2+5t=0⇔2x2=5t⇔x2=25tM(F)⇒row 12x1+3x3=0⇔t=x32x1+3t=0⇔2x1=−3t⇔x1=−23t⟹L=⎩⎨⎧x1x2x3−23251⋅t,t∈R⎭⎬⎫⇒ker(F)=ker(M(F))=⎩⎨⎧x1x2x3−23251⋅t,t∈R⎭⎬⎫=span⎩⎨⎧x1x2x3−23251⎭⎬⎫=span⎩⎨⎧x1x2x3−352⎭⎬⎫⇒A basis of ker(F) is for example:Bker(F)=⎩⎨⎧−352⎭⎬⎫
For the image Im(F) we know from the remark above:
Im(F)=span⎩⎨⎧132,110,−123⎭⎬⎫
From the columns of M(F) (a spanning set of Im(F)), choose two vectors as a basis of Im(F), namely those that have pivot elements in M(F):
We can read off dim(Im(F))=rank(F)=rank(M(F))=2 directly from M(F). Thanks to the dimension formula we know that the basis of the kernel of F contains one basis vector.
⇒ker(F)=ker(M(F))=⎩⎨⎧x1x2x31−21⋅t,t∈R⎭⎬⎫=span⎩⎨⎧x1x2x31−21⎭⎬⎫⇒A basis of ker(F) is for example:Bker(F)=⎩⎨⎧1−21⎭⎬⎫
For the image Im(F) we know from the remark above:
Im(F)=span{(14),(25),(36)}
From the columns of M(F) (a spanning set of Im(F)), choose two vectors as a basis of Im(F), namely those that have pivot elements in M(F):
⇒BIm(F)={(14),(25)}
Algorithmmatrix of a linear map (representation matrix; special case with standard basis)
Given:V,W vector spaces, F:(V,A)→(W,B),v↦F(v) and a basis of V with A={a1,...,an} and the standard basis of W with B={b1,...,bm}
Let A=(e1,...,en) be the canonical basis of the vector space V and B=(b1,...,bn) another basis of V, described in terms of the canonical basis. Then there exists a change-of-basis matrix with:
TAB=(b1∣⋯∣bn) with bi=TABei,i∈{1,...,n}.VATBAVB
Remark
The following calculation rule holds: TBA=(TAB)−1
With the definition above we therefore obtain: TBAbi=ei,i∈{1,...,n}.
Satz
Let E be a field, V an E-vector space with dim(V)=n<∞. Let v∈V and A={a1,...,an},B={b1,...,bn} bases of V. Then there exist unique λ1,...,λn∈E and unique μ1,...,μn∈E such that
v=k=1∑nλkak=k=1∑nμkbk.
This raises the question of how to switch between the bases A and B; concretely, you are for example given the matrix of a linear map with respect to A and now want to express the matrix with respect to B.
Definition
For every basis B={v1,...,vn} of V there is exactly one isomorphism:
ϕB:En→V,(x1,...,xn)↦ϕB(x1,...,xn)=k=1∑nxkvk=x1v1+...+xnvn with ϕB(ei)=vi.
(In words: ϕB assigns to x its coordinates with respect to the basis B.)
Definition
Let V with basis A={v1,...,vm} and W with basis B={w1,...,wn} be vector spaces over E. Then for every linear map f:V→W there is exactly one matrix MBA(f) such that MBA(f)j=f(vj)=a1jw1+...+anjwn for j=1,...,m.
Remark
The j-th column of the matrix MBA(f) above is the coordinate vector of f(vj) with respect to the basis B.
Remarkimportant
The columns of a matrix of a linear map contain the images of the basis vectors, i.e. MBA(f)=(f(v1)∣⋯∣f(vm)).
Remark
Using the commutative diagram, the matrix MBA(f) can also be described as follows:
The regular change-of-basis matrixTBA with bases A={v1,...,vn},B={w1,...,wn} of the vector space V looks as follows:
TBA=ϕB−1ϕA=t11⋮tn1⋯⋯t1n⋮tnn,
EnEnVTBAϕAϕB
This lets you now describe wi=ti1v1+...+tinvn=(TBAvA)i,i∈{1,...,n}, where wi is expressed with respect to B and vi with respect to A:
TBAvA=wB,where vA=v1⋮vn w.r.t. A,wB=w1⋮wn w.r.t. B.
Remarkcalculation rules
TAA=𝟙
TBA=(TAB)−1
λA∈Kn a coordinate vector with respect to A μB∈Kn a coordinate vector with respect to B ⇒TBAλA=μB
f:V→V linear with matrix MAA(f), where the domain and the codomain are given with respect to A. Analogously, the matrix MBB(f) is given with respect to B in both the domain and the codomain. We achieve a change of basis from A to B of the matrix MAA(f) with the change-of-basis matrices TAB,TBA:
MBB(f)=TBAMAA(f)TAB
EnEnVVEnEnMBB(f)fMAA(f)ϕBϕBϕAϕATABTBA
f:V→V linear with matrix MB2B1(f), where the domain is given with respect to B1 and the codomain with respect to B2. Analogously, the matrix MB2′B1′(f) is given with respect to B1′ in the domain and B2′ in the codomain. We achieve a change of basis from B1 to B1′ (domain) and from B2 to B2′ (codomain) of the matrix MB2B1(f) with the change-of-basis matrices TB2′B2,TB1B1′: