ConceptComplete

Kernel and Image

The kernel (null space) and image (range) of a linear transformation are the two fundamental subspaces associated with it. They measure how much information the map "destroys" and "covers."


Definitions

Definition2.2Kernel (null space)

Let T:V→WT : V \to W be a linear transformation. The kernel (or null space) of TT is

ker⁑(T)=N(T)={v∈V∣T(v)=0}.\ker(T) = N(T) = \{v \in V \mid T(v) = 0\}.

Definition2.3Image (range)

The image (or range) of TT is

im(T)=R(T)={T(v)∣v∈V}={w∈W∣w=T(v) for some v∈V}.\text{im}(T) = R(T) = \{T(v) \mid v \in V\} = \{w \in W \mid w = T(v) \text{ for some } v \in V\}.

Definition2.4Rank and nullity

The rank of TT is rank(T)=dim⁑(im(T))\text{rank}(T) = \dim(\text{im}(T)).

The nullity of TT is nullity(T)=dim⁑(ker⁑(T))\text{nullity}(T) = \dim(\ker(T)).

Theorem2.2Kernel and image are subspaces

For T:V→WT : V \to W linear:

  1. ker⁑(T)\ker(T) is a subspace of VV.
  2. im(T)\text{im}(T) is a subspace of WW.
Proof

For (1): T(0)=0T(0) = 0, so 0∈ker⁑(T)0 \in \ker(T). If T(u)=0T(u) = 0 and T(v)=0T(v) = 0, then T(au+bv)=aT(u)+bT(v)=0T(au + bv) = aT(u) + bT(v) = 0, so au+bv∈ker⁑(T)au + bv \in \ker(T).

For (2): T(0)=0∈im(T)T(0) = 0 \in \text{im}(T). If w1=T(v1)w_1 = T(v_1) and w2=T(v2)w_2 = T(v_2), then aw1+bw2=T(av1+bv2)∈im(T)aw_1 + bw_2 = T(av_1 + bv_2) \in \text{im}(T).

β– 

Examples

ExampleKernel of a matrix transformation

For A=(123456)A = \begin{pmatrix} 1 & 2 & 3 \\ 4 & 5 & 6 \end{pmatrix}, TA:R3β†’R2T_A : \mathbb{R}^3 \to \mathbb{R}^2 with TA(x)=AxT_A(x) = Ax.

Row reducing: (1230βˆ’3βˆ’6)β†’(10βˆ’1012)\begin{pmatrix} 1 & 2 & 3 \\ 0 & -3 & -6 \end{pmatrix} \to \begin{pmatrix} 1 & 0 & -1 \\ 0 & 1 & 2 \end{pmatrix}.

So ker⁑(TA)={t(1,βˆ’2,1)∣t∈R}\ker(T_A) = \{t(1, -2, 1) \mid t \in \mathbb{R}\}. Nullity =1= 1, rank =2= 2.

ExampleKernel of differentiation

D:Pn(R)β†’Pnβˆ’1(R)D : P_n(\mathbb{R}) \to P_{n-1}(\mathbb{R}), D(p)=pβ€²D(p) = p'.

ker⁑(D)={p∈Pn(R)∣pβ€²=0}={c∣c∈R}\ker(D) = \{p \in P_n(\mathbb{R}) \mid p' = 0\} = \{c \mid c \in \mathbb{R}\} (the constant polynomials).

So nullity(D)=1\text{nullity}(D) = 1 and im(D)=Pnβˆ’1(R)\text{im}(D) = P_{n-1}(\mathbb{R}), giving rank(D)=n\text{rank}(D) = n.

ExampleKernel and image of a projection

Ο€:R3β†’R2\pi : \mathbb{R}^3 \to \mathbb{R}^2, Ο€(x,y,z)=(x,y)\pi(x, y, z) = (x, y).

ker⁑(Ο€)={(0,0,z)∣z∈R}\ker(\pi) = \{(0, 0, z) \mid z \in \mathbb{R}\} (the zz-axis), nullity =1= 1.

im(Ο€)=R2\text{im}(\pi) = \mathbb{R}^2, rank =2= 2.

ExampleKernel of the trace

tr:M2Γ—2(F)β†’F\text{tr} : M_{2 \times 2}(F) \to F, tr(abcd)=a+d\text{tr}\begin{pmatrix} a & b \\ c & d \end{pmatrix} = a + d.

ker⁑(tr)={A∣a+d=0}\ker(\text{tr}) = \{A \mid a + d = 0\}, the trace-zero matrices. This has dimension 33 (nullity =3= 3).

im(tr)=F\text{im}(\text{tr}) = F, rank =1= 1.

Check: 3+1=4=dim⁑M2Γ—2(F)3 + 1 = 4 = \dim M_{2 \times 2}(F).

ExampleKernel of rotation

RΞΈ:R2β†’R2R_\theta : \mathbb{R}^2 \to \mathbb{R}^2 (rotation by ΞΈβ‰ 0\theta \neq 0).

ker⁑(Rθ)={0}\ker(R_\theta) = \{0\} (only the origin is fixed), nullity =0= 0.

im(RΞΈ)=R2\text{im}(R_\theta) = \mathbb{R}^2 (rotation is surjective), rank =2= 2.

ExampleKernel of the zero map

T0:V→WT_0 : V \to W, T0(v)=0T_0(v) = 0 for all vv.

ker⁑(T0)=V\ker(T_0) = V, nullity =dim⁑V= \dim V.

im(T0)={0}\text{im}(T_0) = \{0\}, rank =0= 0.

ExampleKernel of the identity

idV:V→V\text{id}_V : V \to V, idV(v)=v\text{id}_V(v) = v.

ker⁑(idV)={0}\ker(\text{id}_V) = \{0\}, nullity =0= 0.

im(idV)=V\text{im}(\text{id}_V) = V, rank =dim⁑V= \dim V.

ExampleKernel and image of transpose

T:M2Γ—2(F)β†’M2Γ—2(F)T : M_{2 \times 2}(F) \to M_{2 \times 2}(F), T(A)=ATT(A) = A^T.

ker⁑(T)={0}\ker(T) = \{0\} (since AT=0A^T = 0 implies A=0A = 0), nullity =0= 0.

im(T)=M2Γ—2(F)\text{im}(T) = M_{2 \times 2}(F) (since (BT)T=B(B^T)^T = B), rank =4= 4.

ExampleNull space of a matrix

For any A∈MmΓ—n(F)A \in M_{m \times n}(F), N(A)=ker⁑(TA)N(A) = \ker(T_A) is the solution set of Ax=0Ax = 0. The column space C(A)=im(TA)C(A) = \text{im}(T_A) is spanned by the columns of AA.

ExampleKernel of definite integration

T:C[βˆ’1,1]β†’RT : C[-1, 1] \to \mathbb{R}, T(f)=βˆ«βˆ’11f(x) dxT(f) = \int_{-1}^1 f(x)\,dx.

ker⁑(T)\ker(T) is the set of continuous functions with zero integral over [βˆ’1,1][-1, 1]. For instance, all odd functions belong to ker⁑(T)\ker(T). This kernel is infinite-dimensional.

ExampleKernel of evaluation

evc:Pn(F)β†’F\text{ev}_c : P_n(F) \to F, evc(p)=p(c)\text{ev}_c(p) = p(c).

ker⁑(evc)={p∈Pn(F)∣p(c)=0}\ker(\text{ev}_c) = \{p \in P_n(F) \mid p(c) = 0\}, i.e., polynomials with root at cc. These are of the form (xβˆ’c)q(x)(x - c)q(x) with deg⁑q≀nβˆ’1\deg q \leq n - 1, so nullity =n= n.

ExampleKernel of the left shift

L:Fβˆžβ†’F∞L : F^\infty \to F^\infty, L(a1,a2,a3,…)=(a2,a3,a4,…)L(a_1, a_2, a_3, \ldots) = (a_2, a_3, a_4, \ldots).

ker⁑(L)={(a1,0,0,…)∣a1∈F}\ker(L) = \{(a_1, 0, 0, \ldots) \mid a_1 \in F\}, which is 1-dimensional.

im(L)=F∞\text{im}(L) = F^\infty (surjective: for any (b1,b2,…)(b_1, b_2, \ldots), take (0,b1,b2,…)(0, b_1, b_2, \ldots)).


Injectivity and surjectivity

Theorem2.3Kernel characterizes injectivity

A linear map T:Vβ†’WT : V \to W is injective (one-to-one) if and only if ker⁑(T)={0}\ker(T) = \{0\}.

Proof

(β‡’\Rightarrow) If TT is injective and T(v)=0=T(0)T(v) = 0 = T(0), then v=0v = 0.

(⇐\Leftarrow) If ker⁑(T)={0}\ker(T) = \{0\} and T(u)=T(v)T(u) = T(v), then T(uβˆ’v)=0T(u - v) = 0, so uβˆ’v=0u - v = 0, i.e., u=vu = v.

β– 
Theorem2.4Image characterizes surjectivity

T:V→WT : V \to W is surjective (onto) if and only if im(T)=W\text{im}(T) = W.

RemarkLooking ahead

The precise relationship between kernel and image is given by the Rank-Nullity Theorem: dim⁑V=rank(T)+nullity(T)\dim V = \text{rank}(T) + \text{nullity}(T). A linear map that is both injective and surjective is an isomorphism.