MIT 18.701 — Lecture 3

Kernel and Image, Cosets, Normal Subgroups, Conjugation, The Correspondence Theorem

§ Kernel and Image

Definition (Kernel and Image). Consider a homomorphism f:GHf: G \to H between groups GG and HH. Then the kernel of ff is ker(f):={gG:f(g)=1H}\ker(f) := \{g \in G : f(g) = 1_H\}, and the image of ff is im(f):={hH:h=f(g) for some gG}\im (f) := \{h \in H: h = f(g) \text{ for some } g \in G\}.

Note that ker(f)\ker(f) is a subgroup of GG, and im(f)\im(f) is a subgroup of HH. (Check this!)

Example. (Application of the Kernel) Take a homomorphism f:GHf: G \to H, and suppose that f(g1)=f(g2)f(g_1) = f(g_2) for g1,g2Gg_1, g_2 \in G. What can we say about g1g_1 and g2g_2?

Solution: Observe the following:

f(g1)=f(g2)      f(g1g21)=1      g1g21ker(f)      g1=kg2  for some kker(f).f(g_1) = f(g_2) ~ \iff ~ f(g_1g_2^{-1}) = 1 ~ \iff ~ g_1g_2^{-1} \in \ker(f) ~ \iff ~ g_1 = kg_2 ~ \text{ for some } k \in \ker(f).

If we multiply by the left inverse instead of the right inverse in the first step, we also find:

f(g1)=f(g2)      f(g21g1)=1      g21g1ker(f)      g1=g2k for some kker(f).f(g_1) = f(g_2) ~ \iff ~ f(g_2^{-1}g_1) = 1 ~ \iff ~ g_2^{-1}g_1 \in \ker(f) ~ \iff ~ g_1 = g_2k \text{ for some } k \in \ker(f).

Consolidating these results together, we have:

{f(g1)=f(g2)      g1=kg2 for some kker(f)f(g1)=f(g2)      g1=g2k for some kker(f)\boxed{\begin{cases}f(g_1) = f(g_2) ~ \iff ~ g_1 = kg_2 \text{ for some } k \in \ker(f) \\ f(g_1) = f(g_2) ~ \iff ~ g_1 = g_2k \text{ for some } k \in \ker(f)\end{cases}}

Example. Take the homomorphism f:(C,+)(C×,)f: (\mathbb{C}, +) \to (\mathbb{C}^{\times}, \cdot) defined by f(z):=ezf(z) := e^z. Then the image of ff is im(f)=C×\im(f) = \mathbb{C}^{\times} (i.e. ff is surjective), and the kernel of ff is ker(f)=2πiZ\ker(f) = 2\pi i \cdot \mathbb{Z}.

Example. Take the homomorphism f:GLn(R)R×f: GL_n(\mathbb{R}) \to \mathbb{R}^{\times} defined by f(M)=detMf(M) = \det M. Then the image of ff is im(f)=R×\im(f) = \mathbb{R}^{\times} (i.e. ff is surjective), and the kernel of ff is ker(f)=SLn(R)\ker(f) = SL_n(\mathbb{R}).

Example. Take the homomorphism f:Sn{1,1}f: S_n \to \{-1, 1\} sending permutations πSn\pi \in S_n to the sign of π\pi. The kernel of ff must be a subgroup of SnS_n; this subgroup is the alternating group AnA_n, consisting of all even permutations in SnS_n.

Example. Let GG be a group, and fix an element gGg \in G. Then we may take the homomorphism f:(Z,+)Gf: (\mathbb{Z}, +) \to G defined by f(n):=gnf(n) := g^n. In this case,

§ Cosets

Example. Recall that 3Z3\mathbb{Z} is a subgroup of Z\mathbb{Z}. This subgroup can be “copied-and-translated” twice to form the subsets 3Z+13\mathbb{Z} + 1 and 3Z+23\mathbb{Z} + 2 in Z\mathbb{Z}. (Note that these subsets are not subgroups!)

Let's give a name to these “copied-and-translated” subsets associated with a subgroup: cosets.

Definition (Cosets). Let HH be a subgroup of GG. Then a left coset of HH is a subset gH:={gh:hH}GgH := \{gh : h \in H\} \subseteq G for some gGg \in G. Similarly, a right coset of HH is a subset Hg:={hg:hH}GHg := \{hg : h \in H\} \subseteq G for some gGg \in G.

The phrase “coset” (without further specification) should be assumed to mean “left coset”.

Theorem. (Coset Properties) Let HH be a subgroup of GG. The cosets of HH in GG satisfy the following properties:

  1. Any two cosets of HH have the same size.

  2. Any two distinct cosets are disjoint.

Proof: To prove the first fact, simply note that the size of any coset gHgH is equal to H|H|. This is because the natural map T:HgHT: H \to gH defined by T(h)=ghT(h) = gh is a bijection.   \blacksquare

To prove the second fact, consider two distinct cosets C1=g1HC_1 = g_1H and C2=g2HC_2 = g_2H. Suppose, for the sake of contradiction, that these two cosets are not disjoint. Then there exists h1,h2Hh_1, h_2 \in H such that g1h1C1g_1h_1 \in C_1 and g2h2C2g_2h_2 \in C_2 are the same; that is, g1h1=g2h2g_1h_1 = g_2h_2. Equivalently, g1=g2h2h11g_1 = g_2h_2h_1^{-1}. However, this implies

C1=g1H=(g2h2h11)H=g2(h2h11H)=g2H=C2.C_1 = g_1H = (g_2h_2h_1^{-1})H = g_2(h_2h_1^{-1}H) = g_2H = C_2.

So in fact, we have shown that if there is just a single pair of elements g1h1g_1h_1 and g2h2g_2h_2 in C1C_1 and C2C_2 that match, then we can show the entirety of C1C_1 and C2C_2 must match completely!   \blacksquare

Here's an equivalent, more intuitive rephrasing of the above theorem:

Theorem. (Cosets form Partitions) Let HH be a subgroup of GG. Then the set of all left cosets of HH partitions the elements of GG into disjoint subsets of size H|H|.

Note that all of the above theorems hold equally true if every instance of “left” is replaced with “right”. However, the partitioning of GG formed by left cosets usually does not equal the partitioning formed by right cosets!

Example. Consider the group G=(R2,+)G = (\mathbb{R}^2, +) and a subgroup H:={(a,2a):aR}H := \{(a, 2a): a \in \mathbb{R}\}. Then cosets of HH look like translated copies of lines parallel to y=2xy = 2x. These lines tile the plane R2\mathbb{R}^2.

Definition (Index). The number of cosets of HH in GG is the index of HH in GG, denoted [G:H][G:H]. Note that the theorem above implies G=[G:H]H|G| = [G:H] \cdot |H|.

Theorem. (Lagrange's Theorem) Let HH be a (finite) subgroup of a (finite) group GG. Then H|H| divides G|G|.

Proof: Follows from the equation G=[G:H]H|G| = [G:H] \cdot |H|, along with the fact that [G:H][G:H] is an integer.   \blacksquare

Lagrange's Theorem is very powerful! Here's an interesting corollary of the above theorem.

Theorem. (Groups of Prime Order) Any group of prime order must be cyclic.

Proof: Suppose GG is a group of prime order pp. Pick some gGg \in G that is not the identity. Then H=gH = \langle g \rangle is a subgroup of GG. Importantly, H>1|H| > 1 (since gg is not the identity), and H|H| divides pp. So it must be that H=p|H| = p, meaning that GG and HH are the same group. And since HH was a cyclic group, GG must be cyclic, too.

§ Normal Subgroups

How do cosets and homomorphisms relate?

Example. (Application of the Coset) Take a homomorphism f:GHf: G \to H, and suppose that f(g1)=f(g2)f(g_1) = f(g_2) for g1,g2Gg_1, g_2 \in G. What can we say about g1g_1 and g2g_2?

Solution: Recall the following result we derived earlier:

{f(g1)=f(g2)      g1=kg2 for some kker(f)f(g1)=f(g2)      g1=g2k for some kker(f)\boxed{\begin{cases}f(g_1) = f(g_2) ~ \iff ~ g_1 = kg_2 \text{ for some } k \in \ker(f) \\ f(g_1) = f(g_2) ~ \iff ~ g_1 = g_2k \text{ for some } k \in \ker(f)\end{cases}}

We can now use cosets to rephrase the above statement as follows:

{f(g1)=f(g2)      g1 and g2 are in the same right coset of ker(f).f(g1)=f(g2)      g1 and g2 are in the same left coset of ker(f).     ()\boxed{\begin{cases}f(g_1) = f(g_2) ~ \iff ~ g_1 \text{ and } g_2 \text{ are in the same right coset of } \ker(f). \\ f(g_1) = f(g_2) ~ \iff ~ g_1 \text{ and } g_2 \text{ are in the same left coset of } \ker(f). \end{cases}} ~~~~~ (\bigstar)

The above motivates the following definition:

Definition (Normal). A subgroup NGN \subseteq G is normal if gN=NggN = Ng for all gGg \in G.

This is because the above result ()(\bigstar) immediately implies the following theorem:

Theorem. (Normal Kernel) For a homomorphism f:GHf: G \to H, its kernel K=ker(f)K = \ker(f) is a normal subgroup of GG.

Example. Consider G=S3G = S_3 and a subgroup N={e,(1 2)}N = \{e, (1 \ 2)\}. Then:

Thus, NN is not normal, since (e.g.) N(1 3)(1 3)NN(1 \ 3) \neq (1 \ 3)N. In other words, the partitioning of GG formed by the left cosets of NN is different from the partitioning of GG formed by the right cosets of NN.

Remark. By corollary, there is no homomorphism f:GHf: G \to H to any other group HH such that ker(f)=N\ker(f) = N.

Example. If GG is abelian, then every subgroup NGN \subseteq G is normal. This is because:

gN={gn:nN}={ng:nN}=Ng.gN = \{gn : n \in N\} = \{ng : n \in N\} = Ng.

Example. If NN is a subgroup of GG such that [G:N]=2[G:N] = 2, then NN must be normal. This is because any partitioning of GG into cosets—whether they be left cosets or right cosets—must look like G=N  (GN)G = N \ \sqcup \ (G \setminus N). So the (two) left cosets of NN must match the (two) right cosets of NN.

§ Conjugation

Definition (Conjugation). Given elements a,bGa, b \in G, the conjugation of aa by bb is defined to be bab1bab^{-1}. More generally, if HH is a subgroup of GG, then the conjugation of HH by aa is aHa1aHa^{-1}.

It turns out (easy to check) that if HH is a subgroup, then aHa1aHa^{-1} is also always a subgroup. Thus, we can think of conjugation by gGg \in G as an action that sends subgroups of GG to subgroups of GG.

Theorem. (Normality by Conjugation Condition) A subgroup NGN \subseteq G is normal if and only if NN does not change under conjugation by any element gGg \in G.

Proof: Recall that NN is normal if and only if gN=NggN = Ng for all gGg \in G. Equivalently, gNg1=NgNg^{-1} = N for all gGg \in G, which may directly be read as: the conjugation of NN by gg equals NN for any gGg \in G.   \blacksquare

Theorem. (Intersecting Normal Subgroups) If N1N_1 and N2N_2 are normal subgroups of GG, then their intersection N1N2N_1 \cap N_2 is also normal.

Proof: We argue using the formulation of normality in terms of conjugation described above. For any gGg \in G,

g(N1N2)g1= {gng1:nN1N2}= {gng1:nN1}{gng1:nN2}= (gN1g1)(gN2g1)= N1N2 since N1 and N2 are normal.\begin{align*}g(N_1 \cap N_2)g^{-1} = \ & \{gng^{-1} : n \in N_1 \cap N_2\} \\ = \ & \{gng^{-1} : n \in N_1\} \cap \{gng^{-1}: n \in N_2\} \\ = \ & (gN_1g^{-1}) \cap (gN_2g^{-1}) \\ = \ & N_1 \cap N_2 \text{ since } N_1 \text{ and } N_2 \text{ are normal.}\end{align*}

Thus, N1N2N_1 \cap N_2 remains unchanged by conjugation by gg, so N1N2N_1 \cap N_2 is normal.   \blacksquare

§ Correspondence Theorem

Here's a useful theorem that we won't prove today.

Theorem. (Correspondence Theorem) Let f:GGf: G \to G' be a surjective homomorphism, and let K=ker(f)K = \ker(f) be its kernel. Consider the following collections of subgroups:

A:={subgroups of G containing K}  and  B:={subgroups of G}.A := \{\text{subgroups of } G \text{ containing } K\} ~ \text{ and } ~ B := \{\text{subgroups of } G'\}.

Then there is a bijection from AA to BB that sends subgroups HAH \in A to subgroups f(H):={f(h):hH}Bf(H) := \{f(h): h \in H\} \in B.

Example. (Correspondence Theorem Application #1) Consider the determinant homomorphism f=det:GGf = \det: G \to G', with G=GLn(R)G = GL_n(\mathbb{R}) and G=R×G' = \mathbb{R}^{\times}. Then K=ker(f)=SLn(R)K = \ker(f) = SL_n(\mathbb{R}). What does the Correspondence Theorem say?

Solution: The set B:={subgroups of R×}B := \{\text{subgroups of } \mathbb{R}^{\times}\} is easier to analyze; for example, Q×\mathbb{Q}^{\times} is a subgroup in BB.

By the Correspondence Theorem, the subgroup Q×\mathbb{Q}^{\times} must correspond to a new, quirky subgroup of GLn(R)GL_n(\mathbb{R}) containing K=SLn(R)K = SL_n(\mathbb{R}) that we didn't know of before. This subgroup is:

f1(Q×)={MGLn(R):detMQ×}A.f^{-1}(\mathbb{Q}^{\times}) = \{M \in GL_n(\mathbb{R}): \det M \in \mathbb{Q}^{\times}\} \in A.

There are many, many other subgroups of GLn(R)GL_n(\mathbb{R}) that live between KK and GG.

Example. (Correspondence Theorem Application #2) Take G=ZG = \mathbb{Z} and G=Cn=gG' = C_n = \langle g \rangle, and define the homomorphism f:GGf: G \to G' by f(k)=gkf(k) = g^k. What does the Correspondence Theorem say?

Solution: Observe that ker(f)=nZ\ker(f) = n \mathbb{Z}. Then:

A:={subgroups of G containing K}={dZ:dn}.A := \{\text{subgroups of } G \text{ containing } K\} = \{d\mathbb{Z} : d \mid n\}.

It then follows that we can classify the set of all subgroups of G=CnG' = C_n:

B:={subgroups of G}=f(A)={f(dZ):dn}={gd:dn}.B := \{ \text{subgroups of } G'\} = f(A) = \{f(d\mathbb{Z}) : d \mid n\} = \{ \langle g^d \rangle : d \mid n \}.

So in fact, this tells us every subgroup of CnC_n is of the form gd\langle g^d \rangle for some dnd \mid n.