Formal definition of the pack Intermediate The structure theory in the first part of Serre's Lie Algebras and Lie Groups runs from the definitions of nilpotent and solvable Lie algebras through the two foundational nilpotency/solvability theorems — Engel's (a Lie algebra acting by nilpotent operators is nilpotent) and Lie's (a solvable Lie algebra over an algebraically closed field of characteristic zero acts by upper-triangular matrices) — into the Killing form and Cartan's criterion that detects solvability and semisimplicity, and finally into the formal side: the Campbell-Baker-Hausdorff formula expressing log ( e X e Y ) as a Lie series, and the free Lie algebra with its Hall basis.
This pack collects nine such exercises — two easy, four medium, three hard — each with a hint and a full solution. It is meant to be read alongside its prerequisite units rather than as a standalone development. The exercises are loosely grouped: the central/derived series and basic nilpotency (easy), Engel/Lie theorems and the Killing form (medium), and Cartan's criterion, BCH, and free Lie algebras (hard).
The conventions throughout are Serre's: the lower central series is g 1 = g , g k + 1 = [ g , g k ] , nilpotent meaning g k = 0 for some k ; the derived series is g ( 0 ) = g , g ( k + 1 ) = [ g ( k ) , g ( k ) ] , solvable meaning g ( k ) = 0 ; the Killing form is κ ( X , Y ) = tr ( ad X ∘ ad Y ) . The base field is algebraically closed of characteristic zero unless stated.
Key theorem with full solution Intermediate Before the pack proper, we work one exercise in full as an exemplar of the format. The remaining eight follow the same structure (problem, hint, full answer in <details> blocks).
Lead exercise. Prove Engel's theorem: if every element of a Lie algebra g ⊆ gl ( V ) acts as a nilpotent operator on a nonzero finite-dimensional V , then there is a nonzero v ∈ V killed by all of g (hence V has a flag with g strictly upper-triangular, and g is a nilpotent Lie algebra).
Solution. Induct on dim g . The base case dim g = 0 or 1 is immediate (X nilpotent has a kernel vector). For the inductive step, let h ⊊ g be a maximal proper subalgebra. The key claim is that h is an ideal of codimension one.
For X ∈ h , the operator ad X is nilpotent on gl ( V ) (since X is a nilpotent operator, left- and right-multiplication by X are nilpotent and commute, so their difference ad X is nilpotent), hence acts nilpotently on the quotient g / h . By the inductive hypothesis applied to the image of h acting on g / h , there is a nonzero Y ˉ ∈ g / h killed by ad h ; lifting, [ h , Y ] ⊆ h for some Y ∈ / h . Then h + F Y is a subalgebra strictly containing h , so by maximality it equals g , and h is an ideal of codimension one.
By induction, W = { v ∈ V : h v = 0 } is nonzero. Since h is an ideal, W is Y -stable: for w ∈ W and H ∈ h , H ( Y w ) = Y ( H w ) + [ H , Y ] w = 0 . The nilpotent operator Y restricted to W has a kernel vector v = 0 , killed by both h and Y , hence by all of g = h + F Y . Iterating produces a full flag, so g is strictly upper-triangular and therefore nilpotent. □
Exercises Intermediate
Exercise 1 (easy, symbolic). Heisenberg algebra is nilpotent but not abelian.
The Heisenberg Lie algebra h 3 has basis X , Y , Z with [ X , Y ] = Z , Z central. Compute the lower central series and confirm h 3 is nilpotent of class 2 .
Hint
h 3 2 = [ h 3 , h 3 ] is spanned by the brackets of basis elements.
Answer
The only nonzero bracket among basis elements is [ X , Y ] = Z , so
h 3 2 = [ h 3 , h 3 ] = F Z , h 3 3 = [ h 3 , F Z ] = 0 ,
since Z is central. The lower central series terminates after two steps: h 3 ⊋ F Z ⊋ 0 . So h 3 is nilpotent of class 2 . It is not abelian because h 3 2 = F Z = 0 . (Nilpotent of class 1 would mean abelian.)
Exercise 2 (easy, symbolic). Upper-triangular matrices: solvable, not nilpotent.
Let b ⊂ gl 2 be the upper-triangular 2 × 2 matrices. Show b is solvable but not nilpotent.
Hint
Compute [ b , b ] (the strictly upper-triangular part), then the derived and lower central series.
Answer
b has basis H 1 = E 11 , H 2 = E 22 , E = E 12 . Brackets: [ H 1 , E ] = E , [ H 2 , E ] = − E , [ H 1 , H 2 ] = 0 . So [ b , b ] = F E = n (strictly upper-triangular).
Derived series. b ( 1 ) = F E , b ( 2 ) = [ F E , F E ] = 0 . So b is solvable (of derived length 2 ).
Lower central series. b 2 = F E , b 3 = [ b , F E ] = F E (since [ H 1 , E ] = E = 0 ), and the series stabilises at F E = 0 forever. So b is not nilpotent. This is the standard example separating the two conditions: solvable does not imply nilpotent.
Exercise 3 (medium, proof). Lie's theorem produces a common eigenvector.
State and prove that a solvable Lie subalgebra g ⊆ gl ( V ) over C , V = 0 finite-dimensional, has a common eigenvector.
Hint
Induct on dim g . Pick a codimension-one ideal h (exists by solvability), find a common eigenvector for h with weight λ , then show the weight space V λ is g -stable using the invariance lemma (λ ([ g , h ]) = 0 ).
Answer
Induct on dim g . Since g is solvable, [ g , g ] ⊊ g , so g has a subspace h of codimension one containing [ g , g ] ; such an h is automatically an ideal. By the inductive hypothesis, h has a common eigenvector: there is 0 = v 0 and a linear functional λ : h → C with H v 0 = λ ( H ) v 0 for all H ∈ h .
Let V λ = { v : H v = λ ( H ) v ∀ H ∈ h } = 0 . The invariance lemma states λ ([ X , H ]) = 0 for X ∈ g , H ∈ h ; it is proved by considering, for fixed X and v ∈ V λ , the chain v , X v , X 2 v , … spanning a subspace W on which each H ∈ h is upper-triangular with diagonal λ ( H ) , so tr W ( H ) = ( dim W ) λ ( H ) ; applying this to H = [ X , H ′ ] (a commutator, hence trace zero on the X -stable W ) forces ( dim W ) λ ([ X , H ′ ]) = 0 , and char 0 gives λ ([ X , H ′ ]) = 0 .
Consequently V λ is g -stable: for X ∈ g , v ∈ V λ , H ∈ h , H ( X v ) = X ( H v ) + [ H , X ] v = λ ( H ) X v + λ ([ H , X ]) v = λ ( H ) X v . Pick X 0 ∈ g ∖ h so g = h + C X 0 ; the operator X 0 on V λ has an eigenvector v 1 (algebraically closed field), which is then a common eigenvector for all of g . □
Exercise 4 (medium, symbolic). Killing form of sl 2 .
Compute the Killing form κ ( X , Y ) = tr ( ad X ad Y ) on sl 2 in the basis H , E , F ([ H , E ] = 2 E , [ H , F ] = − 2 F , [ E , F ] = H ), and verify it is nondegenerate.
Hint
Write the 3 × 3 matrices of ad H , ad E , ad F in the basis ( H , E , F ) , then take traces of products.
Answer
In the ordered basis ( H , E , F ) , the adjoint matrices are
\mathrm{ad}\,H = \begin{psmallmatrix}0&0&0\\0&2&0\\0&0&-2\end{psmallmatrix},\quad
\mathrm{ad}\,E = \begin{psmallmatrix}0&0&1\\-2&0&0\\0&0&0\end{psmallmatrix},\quad
\mathrm{ad}\,F = \begin{psmallmatrix}0&-1&0\\0&0&0\\2&0&0\end{psmallmatrix}.
Taking traces of products: κ ( H , H ) = tr ( ad H ) 2 = 0 + 4 + 4 = 8 ; κ ( E , F ) = tr ( ad E ad F ) = 4 (the ( 1 , 1 ) and other diagonal contributions sum to 4 ); κ ( H , E ) = κ ( H , F ) = κ ( E , E ) = κ ( F , F ) = 0 .
So in the basis ( H , E , F ) the Gram matrix is
\kappa = \begin{psmallmatrix}8&0&0\\0&0&4\\0&4&0\end{psmallmatrix},
\qquad \det\kappa = 8\cdot(-16) = -128\neq 0.
The determinant is nonzero, so κ is nondegenerate. By Cartan's semisimplicity criterion, the nondegeneracy of κ confirms sl 2 is semisimple. (The general normalisation is κ = 4 ⟨ ⋅ , ⋅ ⟩ relative to the trace form ⟨ X , Y ⟩ = tr ( X Y ) on the defining representation.)
Exercise 5 (medium, proof). Killing form vanishes identically on a nilpotent Lie algebra.
Show that if g is nilpotent then its Killing form is identically zero.
Hint
On a nilpotent g , every ad X is a nilpotent operator. A product of two operators, one of which is nilpotent and they generate a nilpotent algebra, has trace zero.
Answer
If g is nilpotent, then by definition the lower central series terminates, which means every ad X (X ∈ g ) is a nilpotent operator on g : for X ∈ g , ( ad X ) k ( g ) ⊆ g k + 1 = 0 for k large.
For X , Y ∈ g , the operators ad X and ad Y both lie in the nilpotent Lie algebra ad g ⊆ gl ( g ) , which (being nilpotent and acting by nilpotent operators, via Engel) can be simultaneously strictly upper-triangularised in a suitable basis of g . In that basis ad X and ad Y are strictly upper-triangular, so their product ad X ad Y is strictly upper-triangular, with all diagonal entries zero. Hence
κ ( X , Y ) = tr ( ad X ad Y ) = 0
for all X , Y . The Killing form is identically zero. (The converse is false: κ ≡ 0 characterises solvability of ad g via Cartan's criterion, not nilpotency — see Exercise 6.)
Exercise 6 (medium, proof). Cartan's solvability criterion, statement and one direction.
State Cartan's criterion for solvability, and prove the easy direction: if tr ( X Y ) = 0 for all X ∈ [ g , g ] and Y ∈ g (as operators on a faithful representation V ), then... which direction is immediate, and why is the converse the deep one?
Hint
Cartan: g ⊆ gl ( V ) is solvable iff tr ( X Y ) = 0 for all X ∈ [ g , g ] , Y ∈ g . The "solvable ⇒ trace condition" direction follows from Lie's theorem; the converse needs the Jordan decomposition and a replete eigenvalue argument.
Answer
Statement. A Lie subalgebra g ⊆ gl ( V ) over C is solvable if and only if tr ( X Y ) = 0 for all X ∈ [ g , g ] and all Y ∈ g .
Easy direction (solvable ⇒ trace condition). If g is solvable, Lie's theorem (Exercise 3) puts every element of g into upper-triangular form in a common basis of V . Then every X ∈ [ g , g ] , being a sum of commutators of upper-triangular matrices, is strictly upper-triangular (commutators of triangular matrices have zero diagonal). The product of a strictly upper-triangular X with an upper-triangular Y is strictly upper-triangular, so tr ( X Y ) = 0 . Done.
Why the converse is deep. The hard direction (trace condition ⇒ solvable) cannot use Lie's theorem (we do not yet know g is solvable). Instead one shows [ g , g ] consists of nilpotent operators by a Jordan-decomposition argument: for X ∈ [ g , g ] with semisimple part s having eigenvalues λ i , one constructs a "replacement" derivation that conjugates eigenvalues, and the trace hypothesis forces ∑ ∣ λ i ∣ 2 = 0 , hence X is nilpotent. Then Engel's theorem makes [ g , g ] nilpotent, so g is solvable. This eigenvalue-replacement step is the substantive content. The semisimplicity criterion (κ nondegenerate ⇔ semisimple) is a corollary applied to ad g .
Exercise 7 (hard, symbolic). Low-order terms of the Campbell-Baker-Hausdorff series.
Compute log ( e X e Y ) up to and including terms of total degree three in X , Y , expressing the answer in Lie brackets.
Hint
Expand e X e Y = 1 + ( X + Y ) + 2 1 ( X 2 + 2 X Y + Y 2 ) + ⋯ and feed into log ( 1 + u ) = u − 2 1 u 2 + 3 1 u 3 − ⋯ . The non-commuting monomials collect into brackets [ X , Y ] , [ X , [ X , Y ]] , [ Y , [ Y , X ]] .
Answer
Write e X e Y = 1 + u with u = ( X + Y ) + 2 1 ( X 2 + 2 X Y + Y 2 ) + 6 1 ( X 3 + 3 X 2 Y + 3 X Y 2 + Y 3 ) + ⋯ . Then log ( e X e Y ) = u − 2 1 u 2 + 3 1 u 3 − ⋯ .
Collecting by degree:
Degree 1 : X + Y .
Degree 2 : from u , 2 1 ( X 2 + 2 X Y + Y 2 ) ; from − 2 1 u 2 , − 2 1 ( X + Y ) 2 = − 2 1 ( X 2 + X Y + Y X + Y 2 ) . Sum: 2 1 ( X Y − Y X ) = 2 1 [ X , Y ] .
Degree 3 : a careful collection of the cubic monomials from u , − 2 1 u 2 , + 3 1 u 3 gives 12 1 ( [ X , [ X , Y ]] + [ Y , [ Y , X ]] ) .
So
log ( e X e Y ) = X + Y + 2 1 [ X , Y ] + 12 1 ( [ X , [ X , Y ]] − [ Y , [ X , Y ]] ) + O ( deg 4 ) .
Every term is a Lie polynomial — the deep content of the BCH theorem (the Friedrichs / Dynkin assertion that the series is bracket-expressible), here verified explicitly through degree three. The symmetry X ↔ Y with log ( e Y e X ) = − log ( e − X e − Y ) checks the signs.
Exercise 8 (hard, numeric). Dimension of the free Lie algebra by Witt's formula.
The free Lie algebra on r generators is graded; the dimension of its degree-n part is the necklace (Witt) number n 1 ∑ d ∣ n μ ( d ) r n / d . Compute the dimensions for r = 2 generators and n = 1 , 2 , 3 , 4 , and interpret each as a count of Hall basis elements.
Hint
μ is the Mobius function: μ ( 1 ) = 1 , μ ( 2 ) = − 1 , μ ( 3 ) = − 1 , μ ( 4 ) = 0 .
Answer
Witt's formula W ( n ) = n 1 ∑ d ∣ n μ ( d ) 2 n / d for r = 2 :
n = 1 : 1 1 ( 2 ) = 2 . The two generators x , y .
n = 2 : 2 1 ( 2 2 − 2 1 ) = 2 1 ( 4 − 2 ) = 1 . The single bracket [ x , y ] .
n = 3 : 3 1 ( 2 3 − 2 1 ) = 3 1 ( 8 − 2 ) = 2 . The brackets [ x , [ x , y ]] and [ y , [ x , y ]] .
n = 4 : 4 1 ( 2 4 − 2 2 ) = 4 1 ( 16 − 4 ) = 3 (the μ ( 4 ) = 0 term drops the 2 1 contribution). Three degree-4 Hall basis elements.
So the graded dimensions are 2 , 1 , 2 , 3 , … . Each W ( n ) counts the standard Hall (Lyndon) basis monomials of degree n — equivalently the number of aperiodic necklaces of length n over a 2 -letter alphabet. The total agrees with the Poincare-Birkhoff-Witt count: the free associative algebra of degree n has dimension 2 n , and ∏ n ( 1 − t n ) − W ( n ) = ( 1 − 2 t ) − 1 as generating functions.
Exercise 9 (hard, proof). Poincare-Birkhoff-Witt gives a vector-space basis of U ( g ) .
State the Poincare-Birkhoff-Witt theorem for the universal enveloping algebra U ( g ) , and explain why it shows g ↪ U ( g ) is injective and why the free Lie algebra embeds in the free associative algebra.
Hint
PBW: ordered monomials x i 1 ⋯ x i k with i 1 ≤ ⋯ ≤ i k in a fixed ordered basis { x i } of g form a basis of U ( g ) . The degree-one part of this basis is exactly g .
Answer
Statement. Fix an ordered basis { x i } i ∈ I of g . Then the ordered monomials x i 1 x i 2 ⋯ x i k with i 1 ≤ i 2 ≤ ⋯ ≤ i k (and the empty monomial 1 ) form a vector-space basis of the universal enveloping algebra U ( g ) = T ( g ) / ⟨ x ⊗ y − y ⊗ x − [ x , y ]⟩ .
Injectivity of g ↪ U ( g ) . The degree-one ordered monomials are exactly the basis elements x i of g , and PBW asserts they are linearly independent in U ( g ) . So the canonical map g → U ( g ) carries a basis to a linearly independent set — it is injective. (Without PBW the relations could in principle collapse g .)
Free Lie algebra embeds in the free associative algebra. For the free Lie algebra L ( X ) on a set X , its universal enveloping algebra is the free associative algebra A ( X ) = T ( span X ) (both satisfy the same universal property: a Lie/associative map from X to any associative algebra factors uniquely). PBW for L ( X ) then says L ( X ) ↪ U ( L ( X )) = A ( X ) is injective. Concretely, the Lie monomials sit inside the tensor algebra as the bracket-expressible elements (the image of the Dynkin idempotent n 1 ∑ acting on degree-n tensors), recovering the Friedrichs criterion: an element of A ( X ) is Lie iff it is primitive for the comultiplication Δ ( x ) = x ⊗ 1 + 1 ⊗ x . □
Exercise pack EP2 for Chapter 07.06. Serre Lie Algebras and Lie Groups supplement: nilpotent/solvable Lie algebras, the Engel and Lie theorems, the Killing form and Cartan's criterion, the Campbell-Baker-Hausdorff formula, and free Lie algebras.