KEY

(+)  Direct sum
~>  My attempt at an arrow (mapping to symbol)

1/21/99

Lie Algebras
1)    Definitions
2)    Examples
3)    Classification
4)    Representations
5)    Applications

Note:  In this class we will mainly study Lie algebras over the field C.

Def:  A Lie algebra L is a vector space over C with a bilinear product [ . , . ] from L X L to L satisfying
1)  [x, y] = - [y, x] (antisymmetry)
2)  [x, [y, z]] + [y, [z, x]] + [z, [x, y]] = 0 (Jacobi identity)

Ex:  Let L = {all n by n matrices over C} = gl(n, C).  Define [A, B] = AB - BA where AB is the usual matrix product.  The product here is frequently known as the commutator.  We must check our properties.  Antisymmetry is obvious because -(AB - BA) = BA - AB.  For the Jacobi identity, [A, [B, C]] + [B, [A, C]] + [C, [A, B]] = [A, (BC - CB)] + [B, (CA - AC)] + [C, (AB - BA)] = [A, BC] - [A, CB] + [B, CA] - [B, AC] + [C, AB] - [C, BA] = ABC - BCA - ACB + CBA - CAB - BAC + ACB + CAB - ABC - CBA + BAC = 0.

Ex:  (A) Let L = {all n by n matrices with trace zero over C} = sl(n, C).  We just need to check that L is closed under the commutator, because the properties hold from being a subalgebra of gl(n, C).  However trace [AB] = trace(AB - BA) = trace(AB) - trace(BA) = trace(AB) - trace(AB) = 0 for any A, B.  Thus [AB] is in sl(n, C).  We sometimes refer to Lie algebras of the from sl(n, C) as type A.

Ex:  (B and D) Let L = {all n by n skew matrices over C} = O(n, C).  These are matrices where XT = -X.  Again we check that L is closed under the commutator.  Suppose A and B are in O(n, C), so AT = -A and BT = -B.  Then [AB]T = (AB - BA)T = (AB)T - (BA)T = BTAT - ATBT = BA - AB = - (AB - BA) = - [AB] and hence [AB] is in O(n, C).  We call Lie algebras of type O(n, C) for n odd, type B.  For n even we call them type D.

Ex:  (C) Let J be the following 2n by 2n (nondegenerate skew) matrix
0
I
-I
0
Define L = {all n by n matrices over C so that XTJ = -JX} = sp(n, C) or sp(2n, C) depending on the author.  We must again check that L is closed under the commutator.  Suppose A and B are in L, thus ATJ = -JA and BTJ = -JB.  Then [AB]TJ = (AB - BA)TJ =  (BTAT - ATBT)J = -BTJA + ATJB = JBA - JAB = -J(AB - BA) = -J[AB].

Ex:  Consider L = sl(2, C) which has dimension 3.  Let x be the matrix with 1 in the upper right corner, y be the matrix with 1 in the lower left corner and h be the matrix with 1, -1 down the diagonal.  Then x, y, h is a basis for L.  We can write a multiplication table for the bracket operation as follows:
 
x
y
h
x
0
h
-2x
y
-h
0
2y
h
2x
-2y
0
Note:  Sometimes we write gl(V) or End(V) where V is some finite dimensional vector space.  Then gl(n,C) = gl(Cn).

Def:  Suppose L is a Lie algebra.  A representation of L on a vector space V is a linear map p: L ~> End(V) satisfying p([xy]) = p(y) - p(y)p(x) = [p(x), p(y)].

Def:  Suppose W is a subspace of V such that p(x)w is in W for all w in W, x in L.  Then by restriction we get a map p: L ~> End(W).  This gives a representation of L on W called a subrepresentation of p.  We call W a p(L) invariant subspace.

Def:  A representation of p on V is called irreducible if there are no non-trivial (not equal to 0 or V) invariant subspaces.

Question:  What are all the irreducible finite dimensional representations of sl(2, C)?  Well if p: sl(2, C) ~> End(V) is a representation, then we have linear transformations p(x), p(y),and p(h).  We will examine the eigenvectors of p(h) to answer this question next time.
 
1/25/99

Ex:  What are all the irreducible finite dimensional representations of sl(2, C)?  We choose a basis as before x,y,z with [xy] = h, [hx] = 2x, [hy] = -2y.  Rember that an irreducible representation V means that there does not exist a W subspace of V (W not equal to 0 or V) so that p(x)w is in W for all w in W, x in L (there are no nontrivial L-invariant subspaces).  Suppose (p  ,V) is an irreducible representation. Consider p(h) in End(V), a linear transformation over C.  As C is algebraically closed p(h) has at least one eigenvalue and corresponding eigenvector v.  Let W = span{eigenvectors of p(h)}.  We will show W to be an L invariant subspace thus forcing it to be V.
            As p is linear we only need check that p(x) and p(y) take p(h) eigenvectors to other eigenvectors.  Suppose p(h)v = lv (l in C).  Then p(h)(p(x)v) = p(h)p(x)v = (p([hx])+p(x)p(h))v = p(2x)v+p(x)lv = 2p(x)v+lp(x)v = (2 + l)(p(x)v) thus p(x)v is an eigenvector of p(h) with eigenvalue l + 2.  We have shown that if v is in W then p(x)v is in W.  We can do the same for p(y)v and conclude that if v is in W then p(y)v is in W.  Thus we have that p(L)W is in W.
            Next notice that p(h) is diagonalizable thus we can write V = (+)l in S Vl where our Vls are nonzero l-eigenspaces for p(h) and S is a finite subset of C.  Now there is some l in S such that l + 2 is not in S.  We can pick a nonzero eigenvector vl in Vl.  Then we know p(h)vl = lvl and p(x)vl = 0.  We define vl-2 = p(y)vl, vl-4 = p(y)vl,... as long as vl-2i is nonzero.  We get in this way vl, vl-2, vl-4, ..., vl-2k nonzero vectors so that p(y)vl-2k = 0.  The terminology is to call the p(h) eigenvectors weight vectors.  If p(x)v = 0 then v is a highest weight vector.  If p(y)v = 0 then v is a lowest weight vector.
            Let W = span{vl, vl-2, vl-4, ..., vl-2k}.  We want to show that W is L-invariant.  Now by definition W is p(y) invariant, and being eigenvectors of p(h), W is p(h) invariant as well.  We only need to show W is p(x) invariant.  We must compute p(x)vl-2i.  Well we can look at the simple cases first.  p(x)vl = 0.  p(x)vl-2 = p(x)p(y)vl = (p([xy]) + p(y)p(x))vl = (p(h) + p(y)p(x))vl = p(h)vl + p(y)0 = lvl.  To find the pattern we try the next case.  p(x)vl-4 = p(x)p(y)vl-2 = (p([xy]) + p(y)p(x))vl-2 = (p(h) + p(y)p(x))vl-2 = p(h)vl-2 + p(y)lvl = (l - 2)vl-2 + p(y)lvl = (l - 2)vl-2 + lvl-2 = (l + (l - 2))vl-2.  Trying the next case will reveal p(x)vl-6 = (l + (l - 2) + (l - 4))vl-4 allowing us to see the pattern p(x)vl-2(i+1) = (l + (l - 2) + ... + (l - 2i))vl-2i which can be proved by induction.  We can simplify (l + (l - 2) + ... + (l - 2i)) = (i + 1)l - (i + 1)i = (i + 1)(l - i) and conclude p(x)vl-2(i+1) = (i + 1)(l - i)vl-2i.  This shows W to be p(x) invariant thus p(L) invariant.
            As W = span{vl, vl-2, vl-4, ..., vl-2k} is p(L) invariant we get V = W = Cvl (+) Cvl-2 (+) ... (+) Cvl-2k.  We also have the formulas  p(h)vl-2i = (l-2i)vl-2i, p(y)vl-2i = vl-2(i+1), and p(x)vl-2i = i(l-i+1)vl-2(i+1).  This gives us the matrices for p(h), p(y),and p(x) under the basis {vl, vl-2, vl-4, ..., vl-2k}.  We can gain even more information, however.
            Now l has to satisfy a condition.  We can use the definition of a representation to write p(h)vl-2k = p([xy])vl-2k = (p(x)p(y) - p(x)p(y))vl-2k.  As p(y)vl-2k = 0 and by the relations above we get (l - 2k)vl-2k = p(h)vl-2k = (p(x)p(y) - p(x)p(y))vl-2k = -p(y)p(x)vl-2k = -p(y)k(l - k + 1)vl-2(k-1) = -k(l - k + 1)vl-2.  Thus (l - 2k) = -k(l - k + 1) or l(k + 1) = k(k + 1).  As the smallest value of k is zero, k + 1 is nonzero and we get l = k.  We have shown l is a nonnegative integer.  Also V now equals span{vk, vk-2, vk-4, ..., vk-2k} so S, our set of weights, equals {k, k-2, ..., -k}.
            What we have shown is that if we have some irreducible representation, then it has this structure.  We have not yet proved such a structure exists.  This part is easy.  For each k > 0 we may have an irreducible representation of dimension k + 1 with weights Sk such that p(h), p(y),and p(x) have the following matrices.
 
k
 0
 0
 ...
 0
 
 0
 0
 ...
 0
 0
 
 0
 ...
 0
 0
 0
 
 0
k-2
 0
...
 0
 
 1
 0
 ...
 0
 0
 
 0
 0
 i(k-i+1)
 0
0
p(h)= 
 0
 0
k-4
 ...
 0
p(x)=
 0
 1
 ...
 0
 0
p(y)=
 0
 0
 0
 ...
0
 
 ...
... 
 ...
 ...
 ...
 
 ...
 ...
 ...
 0
 0
 
 0
 0
 0
 0
 ...
 
 0
 0
 0
 ...
-k 
 
 0
 0
 1
 0
 
 0
 0
 0
 0
            At this point all that remains is to check that these three matrices are really a representation of sl(2, C) by matrix multiplication.  This is an easy exercise and we are done.

1/28/99

Def:  Suppose L and L' are Lie algebras.  A linear map f from L to L' is called a homomorphism if f([xy])=[f(x)f(y)].

Def:  Suppose K is a subspace of a Lie algebra L.  Then K is called a subalgebra if for any x,y in K then [xy] is also in K.  K
is called an ideal if for any x in L, y in K then [xy] is also in K.

Note:  Every ideal in L is a subalgebra of L.

Lemma:  If f : L --> L' is a homomorphism then Im(f)={f(x) : x in L} is a subalgebra of L and Ker(f)={x in L : f(x)=0} is an ideal in L.

Proof:  Say x,y are in Im(f).  Then for some x',y' in L f(x') = x and f(y') = y.  Then f([x'y'])=[f(x')f(y')]=[xy] in Im(f).  Thus Im(f) is a subalgebra of L'.  Next say y is in Ker(f) and x is in L.  Then f([xy])=[f(x)f(y)]=[f(x)0]=0 so [xy] is in Ker(f) and Ker(f) is an ideal of L. n

Lemma:  If K is an ideal of L then the coset space L/K={x+K: x in L} is a Lie algebra with [x+K, y+K] = [xy] + K.

Proof:  W must show [x+K, y+K] = [xy]+K is well defined and the properties will follow from the bracket of L.  Suppose x+K = x'+K, then x-x' is in K.  Then [xy]-[x'y]=[x-x',y] in K as K is an ideal and so [x'y]+K = [x'y]+[xy]-[x'y]+K = [xy]+K. n

Def: The Lie algebra L/K is called the quotient Lie algebra.

Theorem: (Fundamental Theorem of Lie Algebras)
1) Suppose f: L --> L is a homomorphism of Lie algebras, then L/Ker(f) is isomorphic to Im(f).
2) If I is an ideal of L so I is contained in Ker(f) then there is a unique y such that f(x)=y(p(x)).
3) Suppose I and J are ideals of L with I an ideal of J.  Then J/I is an ideal of L/I and (L/I)/(J/I) is isomorphic to L/J.
4)  Suppose I and J are ideals of L, then I+J={x+y : x in I, y in J} is an ideal of L, I intersect J is an ideal of L and (I+J)/J is isomorphic to I/(I intersect J).
 
Proof:  Exercise.

Corollary:  There is a 1-1 correspondence between the ideals of L/I and the ideals of L containing I.

Proof:  Let J be an ideal in L containing I.  Then let f map J to the ideal J/I of L/I (which we know exists and is an ideal from the third theorem).  Suppose f(J) = 0, meaning the zero ideal in L/I.  Then J is in I, but as J is an ideal containing I, J can only be I.  Thus Ker(f) ={J} and f is 1-1.  Next let K be an ideal of L/I.  Consider p-1(K) = {x in L : x + I is in K}.  This is an ideal in L containing I which gets sent back to K by f.  This shows f is surjective and we are done. n

Def:  If L is a Lie algebra we define L(1) = [L, L] = {[xy] : x,y in L}, L(2) = [L(1), L(1)], ... , L(k) = [L(k-1), L(k-1)].  This series is called the derived series of L.  We call L(k) the kth derived algebra of L.

Corollary:  If L is a Lie algebra, every term in the derived series of L is an ideal of L.

Lemma:  For any ideals I, J of L, [IJ] is also an ideal of L.

Proof:  Take z in L, x in I, y in J.  Then [xy] is in [IJ].  Here [z[xy]] = [[zx]y] + [x[zy]].  As I is an ideal, [zx] is in I, so [[zx]y] is in [IJ].  As J is an ideal, [zy] is in J, so [x[zy]] is in [IJ].  Hence [z[xy]] is in [IJ] and we are done. n

Corollary:  If L is a Lie algebra, every term in the derived series of L is an ideal of L.

Def:  If L is a Lie algebra we define L1 = [L, L] = {[xy] : x,y in L}, L2 = [L, L1], ... , Lk = [L, Lk-1].  This series is called the descending central series of L.

Note:  For all k,  L(k) is an ideal of Lk which is an ideal of L.

Def:  A Lie algebra is called solvable if L(k) = 0 for some k.  A Lie algebra is called nilpotent if Lk = 0 for some k.

Note:  As L(k) is an ideal of Lk, if Lk = 0 for some k then L(k) = 0 for the same k.  Thus if L is nilpotent L is solvable.

Note:  Consider any L generated by one element x.  Thus L = Cx.  We can call such Lie algebras cyclic.  As [xx] = 0, we know that cyclic Lie algebras are abelian, meaning [LL] = 0.  Abelian Lie algebras are nilpotent as L1 = 0.  Finally nilpotent Lie algebras are solvable.  Thus we get the following chain which also occurs in Group Theory: Cyclic ~> Abelian ~> Nilpotent ~> Solvable.

Ex:  Let L = {n by n upper triangular matrices (excluding the digonal)}.  Call the main diagonal (entries of form aii) the first diagonal, the diagonal above that the second, and so on.  If x and y and in L, x has zeros on the first p-diagonals, and y has zeros on the first q diagonals then xy has zeros on the first p + q diagonals.  We get this from the formula EijEkl = djkEil.  Now Eij is on diagonal (j - i) + 1 (> p), Ekl is on diagonal (l - i) + 1 (> q), and Eil is on diagonal (l-i) + 1 with l - i + 1 > p + q.  Similarly this is true for yx, so it is true for xy - yx.  Thus Lk had k + 1zero diagonals and L is nilpotent.  Moreover if K is a subalgebra of L then Kk < Lk so any subalgebra of L is nilpotent.

Ex:  Let K = {{n by n upper triangular matrices (including the digonal)}.  If A, B are in K then AB and BA in K may have nonzero entries on the diagonal.  However, the diagonal of AB equals the diagonal of BA so AB - BA will have zeros on the main diagonal and thus is in L from our last example.  Now [K[KK]] = [KK] so K is not nilpotent.  However as [[KK],[KK]] does have more zero diagonals by our last example.  In fact K(k) has zeros on the first 2k-1 diagonals.  Thus K is solvable.  We can also conclude that any subalgebra of K is also solvable.

Note:  Over C these example are all the solvable and nilpotent Lie algebras.  This is a consequence of Engel's theorem, Lie's theorem, and Ado's theorem.
 

Click here to return.