THE JORDAN-CHEVALLEY DECOMPOSITION JOO HEON YOO Abstract. This paper illustrates the Jordan-Chevalley decomposition through two related problems. Contents 1. Introduction 1 2. Linear Algebra Review 1 3. Chinese Remainder Theorem for Polynomials 2 4. A Special Case of the Jordan-Chevalley Decomposition 3 5. Lie Algebra Review 4 6. The Jordan-Chevalley Decomposition 7 7. Application of the Jordan-Chevalley Decomposition in Lie Algebra 9 Acknowledgements 9 References 10 1. Introduction In this paper, we start out by introducing basic concepts of linear algebra, polynomials, and Lie algebras, which will later be used for solving two related problems and thereby illustrating the Jordan-Chevalley decomposition. Afterward, some of the consequences of the Jordan-Chevalley decomposition will be discussed briefly. 2. Linear Algebra Review We first define a few important concept that will be directly involved with solving the first problem. The terms that are not defined here can be easily found in [1]. Definition 2.1. Let V be a vector space on the field F and let T be a linear operator on V. A characteristic value of T is a scalar c in F such that there is a non-zero vector α in V with Tα = cα. If c is a characteristic value of T, then (a) any α such that Tα = cα is called a characteristic vector of T associated with the characteristic value c; (b) the collection of all α such that Tα = cα is called the characteristic space associated with c. Date: September 8, 2014. 1 2 JOO HEON YOO Definition 2.2. Let A be an n×n matrix over some field F. A characteristic value of A is a scalar c in F such that det(A−c·I) = 0, i.e. (A−c·I) is a singular matrix. Consider the polynomial f(x) = det(x·I −A). Clearly, the characteristic value of A in F are just the scalars c in F such that f(c) = 0. f is called the characteristic polynomial of A. Remarks 2.3. The term characteristic value is interchangeable with the term eigen value. Definition 2.4. In an n×n matrix, if all the non-diagonal elements, a such that i (cid:54)= j, are ij 0, then it is called a diagonal matrix. If a = 0 for all i ≥ j, then it is called a strictly ij upper−triangular matrix. It is easy to see that for an n × n diagonal matrix, the diagonal elements a are the ii eigen values and Πn (a ) is its determinant. Consequently, Πn (x−a ) is its characteristic i=1 ii i=1 ii polynomial. Forann×nstrictlyupper-triangularmatrix, theeigenvaluesareall0andthecharacteristic polynomial is xn. A matrix is called nilpotent if for some positive integer k, the kth power of the matrix is 0. The smallest such k is called the degree of the matrix. In fact, it is easy to show that for an n×n strictly upper-triangular matrix A, we have An = 0. Lemma 2.5. For n × n strictly upper-triangular matrix A in a field kn, An = 0 i.e. A is nilpotent. Proof. Let A be a strictly upper-triangular matrix over a field kn which has an ordered bases {e ,e ,e ,...,e }. Let V be the span of {e ,e ,e ,...,e } for 0 ≤ i = 0 ≤ n (with e = 0). 1 2 3 n i 0 1 2 i 0 It is clear that 0 = V ⊆ V ⊆ ... ⊆ V = kn. Meanwhile, AV = V since A is a strictly 0 1 n i i−1 upper-triangular matrix. Hence, applying An to kn = V , we get AnV = V = 0. Thus, we n n 0 conclude An = 0, and hence by definition, A is nilpotent. (cid:3) 3. Chinese Remainder Theorem for Polynomials In this section, we introduce the application of the Chinese Remainder Theorem to k[x], the ring of polynomials with coefficients in k. This theorem will be later used extensively in solving the two problems. Definition 3.1. If a(x),b(x),f(x) ∈ k[x] are polynomials then we write: a(x) ≡ b(x) mod f(x) if f(x) divides the a(x)−b(x). Theorem 3.2 (ChineseRemainderTheorem). If a (x),a (x),b (x) and b (x) are polynomials 1 2 1 2 ∈ k[x] such that gcd(a (x),a (x)) = 1, then there exists a polynomial c(x) ∈ k[x] with 1 2 deg(c(x)) < deg(a (x))+deg(a (x)) such that 1 2 c(x) ≡ b (x) (mod a (x)). i i for i = 1,2. THE JORDAN-CHEVALLEY DECOMPOSITION 3 Remarks 3.3. Clearly we may expand this theorem to more than just two linear congruences by simple induction. Example 3.4. Consider the polynomial ring R[x]. Suppose f(x) = x+1, g(x) = x, a(x) = 3,b(x) = 0. Since, gcd(f(x),g(x)) = 1, we can apply the Chinese Remainder Theorem to showthatthereexistsc(x)inR[x]suchthatdeg(c(x)) < deg(f(x))+deg(g(x))whichsatisfies c(x) ≡ 3 (mod f(x)) and c(x) ≡ 2 (mod g(x)) In fact, for this example, it is easy to see that c(x) = −3x. We will give a short proof of the Chinese Remainder Theorem for polynomial ring. Note that the proof works in the same way for the integer ring, because the main property that we will be using is that both of them are Unique Factorization Domains. Proof of the Chinese Remainder Theorem. We will prove the theorem in full generality. Let A(x) = Πn a (x) and A (x) = A/a (x) for i = 0,1,...,n. Because each a (x) is coprime to i=0 i i i i a (x) for each i (cid:54)= j, we have gcd(A (x),a (x)) = 1. Hence, A (x)g (x) ≡ 1 (mod a (x)) has j i i i i i a unique solution. Then consider following polynomial: c(x) := b (x)A (x)g (x)+b (x)A (x)g (x)+...+b (x)A (x)g (x) 0 0 0 1 1 1 n n n It is easy to check that this polynomial c(x) satisfies the required linear congruences. Since a (x) | A (x) for i (cid:54)= j, we have A (x) ≡ 0 (mod a (x)) for i (cid:54)= j. Thus, i j j i c(x) ≡ b (x)A (x)g (x) ≡ b (x) (mod a (x)). i i i i i (cid:3) 4. A Special Case of the Jordan-Chevalley Decomposition Following proposition is a special case of the Jordan-Chevalley Decomposition for matrices. In the next section, we will shortly review a few concepts of Lie Algebra and move onto the generalization of the Jordan-Chevalley Decomposition over Lie algebras. Proposition 4.1. Let s,u ∈ M (k) be a pair of commuting matrices such that s is a diagonal m matrix and u is a strictly upper triangular matrix (with zeros at the diagonal). Put a = s+u. Then s can be written as f(a) for some polynomial f(x) ∈ k[x], such that f does not have any constant term. Proof. Suppose s has (cid:96) distinct eigenvalues {a ,a ,...a } such that a has multiplicity d and 1 2 (cid:96) i i let m = d +d +...+d . We denote the characteristic polynomial of s by c (x). Then 1 2 (cid:96) s c (x) = Π (x−a )di s i i Now we know that each of the (x − a )’s is coprime to each other since each a is dis- i i tinct. Hence by Chinese remainder theorem we can find f(x) which satisfies the following congruences: f(x) ≡ a (mod (x−a )m)) i i 4 JOO HEON YOO and f(x) ≡ 0 (mod x) We can decompose km into a direct sum of V ’s, where V := Ker(s−a ·I)di. Let v ∈ V . i i i i Since f(a)−a ·I is divisible by (a−a ·I)m we can find some g(x) ∈ k[x] such that i i f(a)·v −a ·v = (g(a)·(a−a ·I)m)·v = (g(s+u)·(s+u−a ·I)m)·v i i i But since s and u commute, we can expand the right hand side by the Binomial theorem to write m (cid:88) ((s+u−a ·I)m)·v = ((s−a ·I)(m−n) ·un)·v i i n=0 But, s·v = a ·v . Thus, the RHS simply becomes um ·v. But, um is 0 since u is strictly i i i upper triangular. Hence, for v ∈ V , f(a)·v −a ·I ·v = 0. Thus, f(a) = a ·I on each V , i i i i and hence f(a) = s. (cid:3) 5. Lie Algebra Review The notion of Lie Algebras arise ”in nature” as vector spaces of linear transformations en- dowed with a new operation, defined abstractly with a few axioms. This algebraic structures were introduced to study the concept of infinitesimal transformations and the term was first introduced by Hermann Weyl in the 1930s. Definition 5.1. A vector space L over a field F, with an operation L × L → L, denoted (x,y) (cid:55)→ [xy] and called the bracket or commutator of x and y, is called a Lie Algebra over F if the following axioms are satisfied: (L1) The bracket operation is bilinear. (L2) [xx] = 0 for all x in L. (L3) [x[yz]]+[y[zx]]+[z[xy]]+0 for all x,y,z ∈ L Remark 5.2. Bilinear simply means the function is linear in each argument separately i.e. for u,v,w ∈ L and λ ∈ F, we have • f(u+v,w) = f(u,w)+f(v,w) • f(u,v +w) = f(u,v)+f(u,w) • f(λu,v) = f(u,λv) = λf(u,v) Axiom (L3) is called theJacobi Identity. Note that (L1) and (L2) applied to [x+y,x+y] imply anti-commutativity: (L2’) [xy] = −[yx] Proof. It is easy to see that 0 = [x+y,x+y] = [xx]+[yx]+[xy]+[yy] by(L1) = [xy]+[yx] by(L2) (cid:3) THE JORDAN-CHEVALLEY DECOMPOSITION 5 Definition 5.3. We say that two Lie Algebras L and L(cid:48) are isomorphic if there exists a vector space isomorphism φ : L → L(cid:48) satisfying φ([xy]) = [φ(x)φ(y)] for all x,y in L. In that case, φ is called an isomorphism of Lie Algebras. Definition 5.4. A subspace K of L is called a subalgebra if [xy] ∈ K whenever x,y ∈ K. Thus, K is a Lie Algebra of its own right relative to the inherited operation. If V is a finite dimensional vector space over F, we denote by End(V) the set of linear transformation V → V. For x,y ∈ End(V), define a new operation [x,y] = xy −yx (where the product is the usual composition of linear operations), called the bracket of x and y. With this operation, we can view End(V) as a Lie Algebra over F. In order to distinguish this new algebra structure, we write gl(V) for End(V) viewed as a Lie Algebra, and call it the general linear algebra over V. We can easily check that the operation satisfies the three aforementioned axioms. (L1)−(L2) are obvious. For (L3), based on how the brackets were defined, we have [x,[y,z]]+[y,[z,x]]+[z,[x,y]] =(x[y,z]−[y,z]x)+(y[z,x]−[z,x]y)+(z[x,y]−[x,y]z) =xyz −xzy −yzx+zyx+yzx−yxz −zxy +zyx+zxy −zyx−xyz +yxz =0 Remark 5.5. Since a dimension n vector space V over a field F is isomorphic to Fn, we get that gl(V) ∼= gl(Fn), sometimes also written as gl(n,F). There are three important subalgebras of gl(n,F) that are worth mentioning; (1) t(n,F), the set of upper-triangular matrices a i.e. where a = 0 if i > j. ij (2) η(n,F), the set of strictly upper-triangular matrices a i.e. where a = 0 if i ≥ j. ij (3) δ(n,F), the set of diagonal matrices a i.e. where a = 0 if i (cid:54)= j. ij It is trivial to check that each of these sets is closed under bracket, satisfying conditions for ∼ being subalgebras of gl(n,F) = gl(V). Also, t(n,F) is, in fact, direct sum of η(n,F) and δ(n,F). Definition 5.6. By an F-algebra, we mean a vector space U over F endowed with a bilinear operation U×U → U, usually denoted by juxtaposition (unless U is a Lie algebra, in which case we use the bracket). By a derivation on U we mean a linear map δ : U → U satisfying the product rule: δ(ab) = δ(a)b+aδ(b). It is easy to check that Der U, the collection of all derivations on U, is a vector subspace of End(U). In fact, more generally under the Bracket operation, Lemma 5.7. Der U is a subalgebra of gl(U). 6 JOO HEON YOO Proof. We simply need to check Der U is closed under the bracket operation. let δ,δ(cid:48) ∈ Der U and a,b ∈ U. Then [δ,δ(cid:48)](ab) = (δδ(cid:48) −δ(cid:48)δ)(ab) = δδ(cid:48)(ab)−δ(cid:48)δ(ab) = δ(aδ(cid:48)(b)+δ(cid:48)(a)b)−δ(cid:48)(aδ(b)+δ(a)b) = aδ(δ(cid:48)(b))+δ(a)δ(cid:48)(b)+δ(cid:48)(a)δ(b)+δ(δ(cid:48)(a))b−aδ(cid:48)(δ(b)) −δ(cid:48)(a)δ(b)−δ(a)δ(cid:48)(b)−δ(cid:48)(δ(a))b = aδ(δ(cid:48)(b))+δ(δ(cid:48)(a))b−a(δ(cid:48)(δ(b))−δ(cid:48)(δ(a))b = a(δδ(cid:48)(b)−δ(cid:48)δ(b))+(δδ(cid:48)(a)−δ(cid:48)δ(a))b = a[δδ(cid:48)](b)+[δδ(cid:48)](a)b showing [δ,δ(cid:48)] ∈ Der U, by definition. (cid:3) Definition 5.8. Since a Lie algebra L is an F-algebra in the above sense, Der L is well- defined. For x ∈ L, the map ad(x) defined by ad(x)(y) = [xy] is an endomorphism of L. In fact, ad(x) ∈ Der L and the map L → Der L sending x to ad(x) is called the adjoint representation of L. Definition 5.9. A subspace of I of a Lie algebra L is called an ideal of L if [xy] ∈ I for all x ∈ L,y ∈ I. Obviously, 0 and L are ideals of L. A less trivial example is the center of L, denoted Z(L) := {z ∈ L|[xz] = 0 ∀x ∈ L}. Another example is the derived algebra of L, denoted [LL], which consists of all linear combinations of the commutators [xy]. Definition 5.10. L is called simple if L has no ideal except itself and 0 and if [LL] (cid:54)= 0. The condition [LL] (cid:54)= 0 is imposed in order to avoid giving undue prominence to the one dimensional algebra. If L is not simple it is possible to factor out a nonzero proper ideal I and thereby obtain a Lie Algebra of smaller dimension, a quotient algebra, denoted by L/I. Definition 5.11. A homomorphism between two Lie algebras L,L(cid:48) (over the same base field F) is a linear map φ : L → L(cid:48) that is compatible with the respective commutators i.e. f([x,y]) = [f(x),f(y)]. It is interesting to observe that Ker φ is an ideal of L. If φ(x) = 0, and if y ∈ L is arbitrary, then φ([xy]) = [φ(x)φ(y)] = [0φ(y)] = 0. Hence, [xy] ∈ Ker φ, and thus, Ker φ is an ideal of L by definition. Similarly, it is easy to check that Im φ is a subalgebra of L(cid:48). Definition 5.12. A representation of a Lie algebra L on V is a Lie algebra homomorphism φ : L → gl(V). Let us find out what the kernel of ad : L → gl(L) is. It consists of all x ∈ L for which ad(x) = 0, i.e. for which [xy] = 0 for all y ∈ L. Thus, Ker(ad) = Z(L). This has the following interesting consequence. If L is simple, then Z(L) = 0. Hence ad : L → Der L ⊂ gl(L) is a monomorphism. This means that any simple Lie algebra is isomorphic to a linear Lie algebra. THE JORDAN-CHEVALLEY DECOMPOSITION 7 Definition 5.13. For a Lie algebra L, its Derived series is a sequence of ideals of L defined as following: L(0) = L,L(1) = [LL],...,L(i) = [L(i−1)L(i−1)]. L is called solvable if L(n) = 0 for some n. For example, if L is abelian then L is solvable, since abelian is defined as having trivial bracket operation, i.e. [xy] = 0 for all x,y ∈ L, implying L(1) = [LL] = 0. For a Lie algebra L, the following three properties hold regarding solvability. (a) If L is solvable, then so are all subalgebras and homomorphic images of L. (b) If I is a solvable ideal of L such that L/I is solvable, then L, itself is solvable. (c) If I,J are solvable ideals of L, then so is I +J. Property (c) forces the existence of a unique maximal solvable ideal for for a finite dimen- sional Lie algebra L, called the radical of L, and denoted, Rad L. Definition 5.14. A Lie algebra is called semi−simple if its Radical is zero. Equivalently, L is semisimple if it does not contain any non-zero abelian ideal. Thus, in particular, a simple algebra L is semisimple since it has no ideals except itself and 0. Remark 5.15. Note that for an arbitrary Lie algebra L, L/Rad L is semisimple by property (b) from before. Definition 5.16. The Descending central series of L is a sequence of ideals of L defined as follows: L0 = L,L1 = [LL],...Li = [LLi−1]. L is called nilpotent if Ln = 0 for some n. For a Lie algebra L, there are, like before, the following three properties related to nilpo- tency, that hold: (a) If L is nilpotent, then so are all subalgebras and homomorphic images of L (b) If L/Z(L) is nilpotent, then so is L. (c) If L is nilpotent and nonzero, then Z(L) (cid:54)= 0. By Engel’s theorem, a Lie algebra is nilpotent if and only if for every u in L the adjoint endomorphism ad(u) : L → L defined by ad(u)v = [u,v] is nilpotent. 6. The Jordan-Chevalley Decomposition This section illustrates the crux of the Jordan- Chevalley Decomposition. The proof of the theorem is similar to the first one for the algebraically closed field k. This can be extended further to the non-algebraically closed field k using either Galois Theory or via the method outlined at the end of this section. Proposition 6.1. let k be a field of characteristic zero and V a finite-dimensional k vector space. We fix a linear map a : V → V and write A ⊂ End (V) for the k-subalgebra a k generated by a. Then (i) there exists a semisimple element s ∈ A and a nilpotent element u ∈ A such that a a a = s+u, and we have su = us. 8 JOO HEON YOO (ii) Let s+u = s(cid:48) +u(cid:48), where s,s(cid:48) are semisimple, resp., u,u(cid:48) are nilpotent, and such that we have su = us and s(cid:48)u(cid:48) = u(cid:48)s(cid:48). Then s = s(cid:48) and u = u(cid:48). The above unique decomposition of a into s and u is called the Jordan-Chevalley De- composition. Proof. (i) Let a ,...,a (with multiplicities m ,...,m ) be the distinct eigenvalues of a, so 1 k 1 k that the characteristic polynomial is Π(x−a )mi. If V = Ker(a−a ·I)mi, then V is i i i the direct sum of the subspaces V ,...,V , each stable under x. On V , a clearly has the 1 k i characteristic polynomial (x−a )mi. i Now applying the Chinese Remainder Theorem (for the ring k[x]) we can find a polynomial f(x) satisfying the congruences, with pariwise coprime moduli: f(x) ≡ a (mod (x−a )mi) i i and f(x) ≡ 0 (mod x). Note that each (x−a ) is necessarily coprime to the other since each a is distinct. i i Also, note that the last congruence is redundant if 0 is one the eigenvalue of a, and if not, then x is relatively prime to other moduli (x−a ). i Now set g(x) = x−f(x). Evidently each of f(x) and g(x) has zero constant term, since f(x) ≡ 0 (mod x). Setting s = f(a) and u = g(a), we will show that they are semisimple and nilpotent, respectively. Since f(a),g(a) are polynomials of a, it is easy to see that they commute with each other i.e. su = su. The congruence f(x) ≡ a (mod (x−a )mi) i i shows that the restriction of s − a · I to V is zero for all i. Hence, it shows that s i i acts diagonally on V with single eigenvalue a . Since k is an algebraically closed field, i i being semisimple is equivalent to being diagonalizable. Hence we have prove tat s is semisimple. On the other hand, by definition, u = a − s, which makes it quite clear that u is nilpotent. Thus, we have decomposed a into the form a = s+u in which s is semisimple and u is nilpotent. (ii) We need to show the uniqueness of s and u. Let a = s(cid:48) + u(cid:48) be some other such decomposition, so we have s+u = s(cid:48)+u(cid:48) i.e. s−s(cid:48) = u(cid:48)−u. such that we have su = us and s(cid:48)u(cid:48) = u(cid:48)s(cid:48). But a sum of commuting semisimple algebras is again semisimple and a sum of commuting nilpotent algebras is again nilpotent. Hence, s − s(cid:48) = u(cid:48) − u is both semisimple and nilpotent which means it is necessarily 0. Thus, we have shown that s = s(cid:48) and u = u(cid:48). (cid:3) We finish this section by outlining an approach on how to generalize this result to the non-algebraically closed field case. The main reult to use is the following. Let f ∈ k[x] be such that gcd (f, ∂f) = 1. Then we can use the Chinese remainder theorem to construct ∂x inductively polynomials g ∈ k[x],r = 1,2,..., such that, setting r p := x+f ·g +f2 ·g +...+fr ·g ∈ k[x], r 1 2 r THE JORDAN-CHEVALLEY DECOMPOSITION 9 we have f(p (x)) ∈ (f(x))r · k[x]. In particular we can deduce that for any irreducible r polynomial f ∈ k[x] and any (cid:96) ≥ 1, there is a k-algebra imbedding k[x]/(f) (cid:44)→ k[x]/(f(cid:96)). 7. Application of the Jordan-Chevalley Decomposition in Lie Algebra We will finish this exposition to the Jordan-Chevalley decomposition by mentioning one of its applications to Lie algebras. If A is any finite-dimensional F-algebra (for example associative or Lie), then recall that End (A) contains the Lie algebra of derivations Der A. F We can generalize the Jordan-Chevalley decomposition to End (A). We claim that F Proposition 7.1. If δ ∈ Der A then so are its semisimple part σ and its nilpotent part ν. Proof. Clearly, it is enough to show that σ ∈ Der A. Just like we decomposed V in the proof of the Jordan-Chevalley Decomposition, we can break A down into eigenspaces of δ - or, equivalently, of σ. But, this time, we will index them by the eigenvalue. Thus, let A a consists of those x ∈ A such that [δ −a·I]k(x) = 0 for sufficiently large k. Now, we have the identity: n (cid:18) (cid:19) (cid:88) n [δ −(a+b)·I]n(xy) = [δ −a·I]n−i(x)[δ −b·I]i(y) i i=0 If a sufficiently large power of δ − a · I is applied to x and a sufficiently large power of δ − b · I is applied to y, then both of them are zero. Thus for sufficiently large n either one of the factors in each term will be zero, and so the entire sum is zero. Thus we have proved that A A ⊆ A . Hence, if we take x ∈ A and y ∈ A then xy ∈ A , and thus a b a+b a b a+b σ(xy) = (a+b)(xy). On the other hand, σ(x)y + xσ(y) = axy + bxy = (a + b)xy. And thus σ satisfiies the derivation property: σ(xy) = σ(x)y +xσ(y). So, σ and ν are both in Der A. (cid:3) Let us look at what we can say about the Jordan-Chevalley decomposition of the adjoint representation which is an element of Der A. We note that just as the adjoint of a nilpotent endomorphism is nilpotent, the adjoint of a semisimple endomorphism is semisimple. Indeed, if {v }n is an ordered basis of V such that matrix of x is diagonal with eigenvalues a , then i i=0 i we let e be the standard basis element of gl(n,F), which is isomorphic to gl(V) using the ij bases v . It’s easy to verify that i [ad(x)](e ) = (a −a )e ij i j ij and thus ad(x) is diagonal with respect to these bases. Sonowifx = x +x istheJordan-Chevalleydecompositionofx, thenad(x )issemisimple s n s and ad(x ) is nilpotent. Clearly they commute, since n [ad(x ),ad(x )] = ad([x ,x ]) = ad(0) = 0 s n s n Since ad(x) = ad(x )+ad(x ) is the decomposition of ad(x) into a semisimple and a nilpotent s n part which commute with each other, it is the Jordan-Chevalley decomposition of ad(x). Acknowledgements. It is a pleasure to thank my mentor, Subhadip Chowdhury, for his guidance in writing this paper and for overall help throughout the REU program. I would also like to thank Peter May and the University of Chicago Math Department for running the REU program and giving me this wonderful opportunity to participate in it. 10 JOO HEON YOO References [1] Hoffman, Kenneth, and Ray Alden Kunze. Linear Algebra. Prentice-Hall, 1961. [2] James Humphreys. Introduction to Lie Algebras and Representation Theory. Springer, 1997.