Tδhoku Math. Journ.

23(1971), 21-36.

THE DERIVATIONS OF THE LIE ALGEBRAS

TERUKIYO SATO

(Received May 23, 1970)

Some results have been given by several authors as to the derivations of Lie
algebras. Namely, E. Schenkman and N. Jacobson [1] proved that every non-zero
nilpotent Lie algebra has an outer derivation. The author [4] sharpened this
theorem and showed that every nilpotent Lie algebra over a field of characteristic
0 possesses an outer derivation in the radical of its derivation algebra. On the
other hand G. Leger [2] has given a necessary and sufficient condition for a Lie
algebra to have an outer derivation, and in [3] he has shown that if a Lie algebra
has no outer derivations and its center is not zero then it is not solvable and its
radical is nilpotent and is not quasi-cyclic. Moreover, S. Togo [6] proved that
such a Lie algebra coincides with its derived algebra and he proceeded the studies
concerning to the Lie algebra which has outer derivations. The purpose of this
paper is to add some results to them.
In § 2 we shall introduce the notion of the free nilpotent Lie algebra, and
interpret a nilpotent Lie algebra as a quotient algebra of the free nilpotent Lie
algebra by a suitable ideal. We shall also investigate the relationship between the
derivations of the former and of the latter. In § 3, we shall study some applications
of the results by Leger [2]. As it is well known, every semi-simple Lie algebra
has no outer derivations. It is also known that there exists a solvable Lie algebra
with null center which has no outer derivations. However, it seems to the author
that it is unknown whether there exists a Lie algebra with non-zero center which
has no outer derivations. In § 4 we shall give an example of such a Lie algebra
of dimension 41 and with one dimensional center.

1. Preliminaries and notations. Throughout this paper, we suppose that

the Lie algebras have the coefficient field of characteristic 0. For a subset M of
a vector space, we denote by {M} the subspace generated by the elements of Λf.
When M is a subset of a Lie algebra L and k is a natural number, we denote
by Mk the subspace generated by the elements of the form

[ml9 [ra2, [ [%.!, mk] ] (ra1? ra2, , mk z M).

Furthermore we shall employ the following notations :

22 T. SATO

Z(L) : the center of a Lie algebra L

S)(Z/) : the Lie algebra of all derivations of L
3(L) : the ideal of ®(L) consisting of all the inner derivations of L
S»(L) : t h e radical of ®(L)
@(L) : a maximal semi-simple subalgebra of 3)(L).

2. The nilpotent Lie algebras generated by m elements. Let U be an

m-dimensional vector space, and let el9 , em be its basis. We consider the
tensor products U®U, U ®U ®U ,-- - ,U ®U ® - - - ®U (w-times) of U, and
make a space of direct sum of them :

We define a distributive and non-associative product X in V as follows :

*t x (eJι ® ®<?jk) = et ®eh ® ®eh for £ ^ n — 1

^i x (eh ® — ® ejn) = 0.

Next, when the product a x b has been defined already for an element azV and
for every element b of V, we define the product (e ® a) X b by

(tf®<z) χ j = e ® ( α x i ) - f l x ( e ® έ ) ,

where £ denotes an arbitrary element of U. Then the next equality holds.

(1) (axb) X c = ax (b X c) — bx{ax c) for a,b and c € V

To prove this it is enough to show (1) in the case where a, b and c are
homogeneous. W e will prove this by mathematical induction. So we assume that
a = e®a1 for ez U and our assertion holds for aλ. Then,

((e ®a1)xb)xc=[e® {aλ xb))xc— [ax X (e ® b)) x c

= e ® ((#! x b) x c) — (a x b) x {e ® c)

— ax x ((e ® b) x c) + (e ® b) x {ax x c)

= e®(a1x (b X c)) —e®(b x [ax x c)) — ax x (b x (e®c))

+ b x [ax x (e ® c)) — ax x (e ® {b x c)) + ax X (b X (e ® c))

+ e ® (b x (at x c)) — b x (e ® (ax x c))

= [e®ax) x (b x c) — b x ((e ® aλ) x c).

 THE DERIVATIONS OF THE LIE ALGEBRAS 23

Hence our assertion holds for a.

Now we denote by W the left ideal generated by the set {v X v v € V}
and make the quotient space N = V/W. For any elements a, b € V it holds that

(a + b) X (α + &) = α x α + h H α x H ί ' X α = O mod W

Hence we get

axb = —b X a mod W.

PROPOSITION 1. N is a nilpotent Lie algebra.

PROOF. It is obvious that we can define the product uniquely in N. We

denote by [ , ] this product. Let x,y,z be any elements of N9 and a, b, c be
representatives of x, y, z respectively. Then the relation [x, x] = 0 follows from
the fact that a x azW. Moreover,

ax (b xc) -\- b x (c x a) +c x (ax b)

= ax (b x c) — b x (ax c) +c X (ax b)

= (axb) xc + cx(axb) Ξ O mod W.

Hence we get the Jacobi's identity :

[x, [y, z\] + [y, [z, x]] + [z, [x,y]] = 0.

Therefore N is a Lie algebra, and it is nilpotent because of Nn+1 = 0.

We call the Lie algebra N the free nilpotent Lie algebra of length n
generated by U. Assume that a and b are any monomials in V. W is spanned
by the elements of the following types :

a x a, a x b + b x a, C(a x a), C(a xb + b X a),

where C means the product of repeated multiplications by elements of U. Therefore

W is homogeneous and also is N.
From now on we regard the elements of U as imbedded in N.

PROPOSITION 2. Any linear mapping from U to N may be extended to

a derivation of N uniquely.
24 T. SATO

PROOF. Let D be a linear mapping from U to N, and eί9 , em be a basis

of £7. At first we extend D to linear mapping from V to iV as follows.

D(etι X (*<f x ( X (<Vi X eik) )

*
- ΣΣ K> K>
Then it holds that

(2) D{axb) = [Da,b] + [a,Db] for α,^F.

This can be samely proved as in ( 1 ) by induction on the degree of a. For

any a, bu b2, , bk z V,

D[bλ x (b2 x — x (bk x (a x a)) — )

= Σ V>» IK [ - , [Ob,, [ - -, [6 4 , [α, «]] •]

+ [δi, [6,, [ , [bk9 [Da, a] + [a, Da]]. . . ] = 0.

Hence D maps W into 0. Therefore D defines a linear endomorphism of the Lie

algebra N= V/W, which is a derivation of N by ( 2 ) .

PROPOSITION 3. Every linear mapping from U into N may be extended

to an endomorphism of the Lie algebra N. In particular if fu ,fm are
2 n
linearly independent elements of U and gi, ,gmzU + -{- U , then the
linear mapping which sends et to ft + g% (i = 1, , m) defines an automorphism
ofN.

PROOF. We may prove the first half samely as the preceeding proposition.
So we shall prove the latter half. We denote by σx or σ a linear mapping from
U into N which maps et to f or f + gt respectively, and we take the inverse
1
transformation σf of σx in U. We denote by σλ again the extention of σλ to an
endomorphism of N. Because of

= σ ifσ Γ 1 ^ , Wϊ^hy [ , [<rΓleh-ι> <*\leji\ ' * " L

sends iV onto N. Next, we assign to f the element σ 2 /i = f% + ^i, then the

 THE DERIVATIONS OF THE LIE ALGEBRAS 25
extension σ2 to an endomorphism of N must transform N onto N. In fact, at
first it is obvious that σ2Un = Un. So we suppose that

+ Uk+1 + + Un) = Uk

Then as it follows that

, ΓΛ, [ ,Γ/wΛJ ]
= fΛ> ΓΛ, [ , [/ιM, A J ] mod £/* + C/*+1 + + £/»,

σ2 transforms Uk~ι + £/fc + + Un onto itself. Hence by induction we can see

that σ2 is a surjection, and σ = σ2 ° σ*i is also, whence cr is an automorphism of N.

Now we investigate the structure of the derivation algebra SD(iV) of N. By-

Proposition 2 every endomorphism of U may be extended to a derivation of JV.
Now we denote by E the extension of the identity transformation of C7, and by
© the collection of all extensions of linear endomorphisms of U whose traces are
0. © is a simple Lie algebra of type AOT-i Let 9Ϊ be the ideal consisting of
all the extensions of linear transformations which map U to [N,N]. Then any
element of 9Ϊ is a derivation which transforms N into [N9N], whence 9ί is a
nilpotent ideal of Ί)(N). E is commutative with the elements of © and
®(jV) = @ + [E] + SR. Therefore {E}+31 is the radical of ®(iV), and © is a
maximal semi-simple Lie algebra.

PROPOSITION 4. Let N be a nilpotent Lie algebra of length n generated

by m linearly independent elements. Then there exists an ideal A of N such
that N is isomorphίc to N/A.

PROOF. Let fl9 ,fm be linearly independent, and N= {fl9 ,/ m ) 0 [N, N].
Then /i, ,/ m generate the Lie algebra N. This may be proved similarly as in
the proof of Proposition 3. Hence the set fl9 ,/ m is a minimal system of
generators of the Lie algebra N. We extend the linear transformation r from U
to N which sends et to f, to the one from V to N as follows :

ett x {eu x ( X {eik_x X ej) . . . )->[fϊ, [f2, [ [A. 1? fk] ].

Then we may prove the fact τ(a x b) = [ra, τb\ samely as in the proof of
Proposition 1. Therefore r maps W to zero and r induces a homomorphism from
N onto N.

From now on, we identify the element of U the corresponding one of N.

 26 T. SATO

PROPOSITION 5. Let a nilpotent Lie algebra N be isomorphic to N/A,

and ®! be the subalgebra of all the derivations of N which leave A
invariant. We denote by ®0 the collection of all the derivations which map
N into A. Then Ί)o is an ideal of % and ®i/®0 * 5 isomorphic to the
derivation algebra ©(JNΓ) of N.

PROOF. We may easily see that ® 0 is an ideal of ®i and we can define

naturally an isomorphism from ®i/2) 0 into ®(JV) Conversely let D be an
arbitrary derivation of N. Now we take a basis /,•••,/& on N, and let
fu * *# >/Λ; be mapped to / Ί , ,/& respectively by the natural homomorphism />
of iV onto N. Let e 1 ? ,em be the minimal system of generators of N. If we
k ^
set Όe% = Σ °ίjιfj> then by Proposition 2 there exists a derivation D oί N such
J> = 1

that Det = Σajifr -^ o r a n y ^i, , ^ ^ t/ it holds that

.7=1

= Σ ] [P«i, [pΛ2, [ , [D o paJ9 [ , [pat-l9 ρaL] ]

= D o p[α 1? [α2, [ ,K ,[ [>,_!, Λ J ].

Hence we get poD= Dop, especially DA a A. Thus 3)i/S)0 is isomorphic to

Let Φ(JV) be the derivation algebra of a nilpotent Lie algebra N, and © be

a maximal semi-simple subalgebra of 2)(iV). iV is a completely reducible ©-module
and [iV, N] is an ©-invariant submodule, therefore N is decomposed into a direct
sum of [iV, N] and an ©-invariant subspace U. Then ί7 generates the whole Lie
algebra N. @ is isomorphic to a subalgebra of %l(U). When we take the free
nilpotent Lie algebra N generated by U, © may be imbedded in the maximal
semi-simple subalgebra @ of Φ(iV).
When the nilpotent Lie algebra N has a subspace U such that

N=U + [N,N], Un [N,N] = {0},

and N is represented as direct sum of subspaces U\ it was called quasi-cyclic by

Leger [ 3 ], and he showed the following property :
 THE DERIVATIONS OF THE LIE ALGEBRAS 27

Let N be isomorphic to N/A. Then N is quasi-cyclic if and only if A

is a homogeneous ideal. If N is quasi-cyclic, then there exists an outer
derivation of N which is commutative with the elements of the maximal
semi-simple subalgebra © of 3)(JV).

We can understand the latter half of this assertion as follows. The above-
mentioned derivation E of N is an /-multiple of the unit transformation on U\
therefore E belongs to Ί)u as A is homogeneous. Hence E induces a derivation
of N, and we denote this again by E. E and the el am ants of © are commutative
with each other in the space U of generators of N, and therefore on the whole
space N. On the other hand, E is an outer derivation as it is not nilpotent.
Now we consider in the sequel that the maximal semi-simple subalgebra ©
of Ί)(N) is imbedded in the maximal semi-simple subalgebra © of

PROPOSITION 6. If A is an ^-irreducible ideal then it is contained in

the center of N.

PROOF. AS [U,A] is an ©-invariant subspace contained in A, it is either

{0} or A itself. If [U,A] = A, then A = [17, A] = [U, [U,A\] = •• = [C/,[ϊ/,[ ,
[U, A] ] = {0}, namely A is {0}. On the other hand if [17, A] = {0}, A is
contained in the center of JV.

Now for any element b e N, we represent b as follows.

b = bx + b2 + + bn, btsUι.

We denote by p% the mapping from b to b%.

LEMMA 1. Let B be an ^-invariant subspace of N. Then ρt\B is an ©-

homomorphism. Especially for the case that B is irreducible, pιlB is either
an isomorphism or null mapping.

PROOF. The operation of © makes U invariant, and hence Uι also. Thus

PiiB is an ©-homomorphism. The rest is obvious.

For an ©-invariant subspace B, let

pι(B) = = p^{B) =0 and pt(B) Φ 0.

Then we denote the number i by ct(B).

 28 T. SATO

LEMMA 2. There exists an ^-isomorphism σ from pa(B)(B) into B such

that ρa{B) o σ is the identity mapping.

PROOF. If we set i = ct(B)9 by the assumption

S c i / i + [/i+1 + . . . + [ / n and B <£_ Uι+ι + + Un.

Then we may decompose B into the direct sum of ©-spaces :

B= (BO ( [ / ι + 1 + . .+[/n))ΘC.

Then it is easily proved that pt maps C isomorphically onto pt(B). Hence we

can find an isomorphism σ as in the assertion.

PROPOSITION 7. Let A be an (^-invariant ideal of N, and be decomposed

into the direct sum of ©-irreducible subspaces as follows:

(3) Ά = B1® -®BP

Among such decompositions, there exists the one such that for an arbitrary q
(l^q^p) the ideal generated by ρa{B) (Bq) does not contain any Pj(Bi) but
for the case either j = a{Bι) or p5{Bi) = 0.

PROOF. We set Cj = Aθ{Uj-] \-Un), then it is ©-invariant and A

= Ci D C2 D O Cw. We can find ©-invariant subspaces C{ such that

A = Cι 0 0 Cn_! + Cn, Ci = Gi 0 Ci + i.

Furthermore we decompose C/, , C n -i and Cn into the direct sum of irreducible

subspaces, and we assume that the decomposition (3) is such a one. Now let us
suppose that a nonzero pJ(Bι) is contained in the ideal generated by pa^B){Ba) and
j>a{Bι). We set (adUy-°c{BQ)Bq = B. Then ps(B%) is an irreducible subspace of

Then by Lemma 2, there exists an ©-isomorphism σ from Pj(Bι) into B which

satisfies Pj°σ= the identity. We set

Bl= ί&i-σop,^); bizBi}.

Then it is easily proved that B\ is an ©-irreducible space and pj{B'i)=0. We

 THE DERIVATIONS OF THE LIE ALGEBRAS 29
may replace the subspace Bt in (3 ) by B\ because &i^Cl(fii), (r°pj(bt) € B C Cj
and j>ct(Bi). Repeating this procedure we get the decomposition which we
desired.

LEMMA 3. Let B be an © -invariant irreducible subspace of N. If pj{B)

^pic(B) Φ 0, then j — k must be a multiple of m.

PROOF. Let us denote by EQ the derivation of N such that ep^>8pqep (p

= 1, , m). The collection of the derivations ΣQ'Λ'QEQ (where X) g λ α = 0) makes
up a Cartan subalgebra of ©. For a monomial a in U\ we denote by mp the
m
number of ep contained in expression of a. Then we have ]>Z mp = i, and

whence a is a weight vector of the representation of ©.

Now we denote the highest weights in both representation spaces pj{B) and
pic(B) by X)λ g m g and ^2xamq respectively. Then they must coincide, whence

771

ΣX(m g - mQ) = XX(raβ -mQ-mx+ mλ) = 0.

Q= 2

This holds identically for any λ g (g^2), and therefore

m2-m2 = =mm- mm'= mλ - m[.

This implies j — k = 0 mod m, because X)mg = j and X)mg = k.

PROPOSITION 8. Let N be a nϊlpotent Lie algebra generated by m

elements, and the maximal semi-simple subalgebra © of its derivation algebra
be of dimension m2 — 1. Further we assume that N = N/A and B is an ©-
irreducible subspace of A. If P3{B) =pfc(J5) Φ 0, then j — k must be a multiple
of m, and N is either quasi-cyclic or its length is greater than m + 1.

PROOF. The first half is obvious in virtue of Lemma 3. So we assume

that N is not quasi-cyclic. Then there exists ©-irreducible subspace B of A such
that ρ5(B)ΦQ and ρk(B)Φθ for j>k. Of course ; ^ 2 and it must be k^j
^m + 2 by the first half. Thus the latter half is also proved.

Proposition 7 and a modification of Proposition 8 will be available to construct

 30 T. SATO
a nilpotent Lie algebra with a certain property in § 4.

3. The Lie algebra with outer derivations. In this section we shall be

concerned with investigation of the Lie algebra which has outer derivations.
First of all we cite the following lemma from Togo [ 5 ] :

LEMMA 4. Let the Lie algebra L be decomposed into the direct sum of
ideals in the form L = L x © L 2 . Then the derivation algebra ®(L) of L may
be expressed in the following form :

3)(L) = %{Lλ) + ®(L 2 ) + X(Ll9 L2) + 2)(L 2 , Lλ).

Here, ®(LX) is the collection of trivial extensions of all the elements in %[Li)>
and 3)(L,, Lό) is the collection of all the linear transformations which send
Li into the center Z{L0) of Lj9 and send [Lu Lt] and Lj to {0}. Moreover it
holds

[^(LO + ®(L2), 3)(Lt, L,]\ c S)(Lt, Lό) (i, j = 1 or 2).

In particular provided that Z[LX) c [Lu L ], when we denote by ©(LJ the

collection of all the linear transformations of L which send Lx into Z[Lλ)
and L2 into {0}, K(LX) is an abelian ideal of ®(L) and [Ί)(LU L2), ®(L2, Lt)]
is contained in

PROPOSITION 9. Let the Lie algebra L be decomposed into the direct

sum of two ideals as follows.

L = LX@L2 and Z(LX) c [Lu L J .

We designate by ©(L^) a maximal semi-simple subalgebra of derivation algebra

Ί){L%) of Li9 and by at(Lt) the radical of Φ(LJ. Then ©(LJ + ®(L2) is a
maximal semi-simple subalgebra of 3)(L) and

ί9 L2) + ®(L 2 , Lx)

is the radical of ®(L).

PROOF. By the above-mentioned lemma,

1?
L2), ®(L 2 , Lx)1 c g(LJ c

J + 3)(L2), ©(L,, L,)] c S)(Lt, L,)

 THE DERIVATIONS OF THE LIE ALGEBRAS 31

[<D(Lt, L,), Φ(L t , L,)] = {0} for i Φj .

Hence 9ΐ is an ideal of ®(L). If we set 9Ϊ<2) = [3t, 31], 3ΐ w = [3ΐ<2), St<IJ], ,

we have the following :

c %ϊ{Lι)m + 9Ϊ(L2)<2> + 6(L,) + Φ(L,, L2) + Φ(L2, L,)

c !«(£,)<" + 3ΐ(L2)C3> + e(L,) + $(L l t L2)

9i<» c

As 8t(L,) and 3Ϊ(L2) are solvable, for sufficiently large k, ^ ( L , ) " ' = 9ΐ(L2)<« = 0 .
Hence it follows that

9t<*> c ©(LJ + ®(L» L2) + Φ(L2, A).

As [e(L 1 ),a)(L < ,L i )]ce(L 1 )nΦ(L M L J ) = {0}, SR^+^ceίL!) and therefore 9ϊ α + 2 )

= 0. This implies that 9i is a solvable ideal. On the other hand ©(LJ + @(L2)
is a semi-simple subalgebra and 3)(L) is expressed as ©(LJ + ©(^2) + 91 This
concludes the proof.

Let L = S + R be a Levi decomposition of the Lie algebra L. For an

element 5 in S, we denote by adRs the restriction of ad s to R, and by adRS all
of them. Then Leger [2] has proved the following.

PROPOSITION 10. The Lie algebla has no outer derivations if and only
if any derivation of R which is commutable with all the elements of adRS
is an inner derivation.

We shall now state a number of corollaries obtained from this proposition:

COROLLARY 1. Let S be a semi-simple Lie algebra and R be a solvable

Lie algebra with outer derivations. Then the direct sum of S and R has an
outer derivations.

PROOF. It is obvious as adRS = 0.

COROLLARY 2. Let the radical N of L be nilpotent, and the derivation

algebra of N be solvable. Then L has an outer derivation.

PROOF. Let L = S + N be a Levi decomposion of L. It holds that adNS = 0

and N has an outer derivation by Jacobson [1], and so our assertion holds.
 32 T. SATO

COROLLARY 3. The notation being as in Proposition 10, let adRS be

contained in a proper ideal ©x of a maximal semi-simple sub algebra @(22)
of 3) (22). Then L has an outer derivation.

PROOF. Because, when we decompose ©(22) into a direct sum of ideals

©(22) = @! + @2, any element of @2 is commutable with the elements of adRS
and is an outer derivation.

Let 3>(22) be the derivation algebra of a solvable Lie algebra 22, and ©(22)
be its maximal semi-simple subalgebra. If there exists an outer derivation which
is commutable with all the elements of ©(22), we call that 22 belongs to the class
D. It is easily seen that this definition is independent of the choice of ©(22).

PROPOSITION 11. If the radical R of a Lie algebra L belongs to the class

D, then L has an outer derivation,

PROOF. We may verify this immediately by cosidering a maximal semi-simple

subalgebra of ©(22) which contains adRS.

PROPOSITION 12. If a solvable Lie algebra R is expressed as a direct

sum of two ideals Rx and 222, and Rλ belongs to the class D, then R also
belongs to the class £).

PROOF. Let D1 be an outer derivation which is commutable with the elements

of ©(22J. We denote by Dx the trivial extension of Dx on 22 again. At first
we suppose that Z(222) c [222, 222]. Then Dx is commutable with the elements of
©(22j) + @(222), which coincides with ©(22) by Proposition 9. As Dλ is an outer
derivation, so in this case 22 belongs to the class D. When Z(222) is not contained
in [222, 222], we may decompose 22 into the direct sum of ideals such as 22 = 223
0 224, where Z(223) c [223, 223] and 224 is abelian. As the abelian Lie algebra 224
belongs to class D, 22 also belongs to the class D by the above-stated. Hence
the proposition is proved.

It is known that the following Lie algebras belong to the class D.

abelian Lie algebra
nilpotent Lie algebra of dimension less than 6
quasi-cyclic Lie algebra.
On the other hand, Leger [3] and Togo [6] proved the following:

PROPOSITION 13. If a Lie algebra L possesses the non-zero center and

has no outer derivations, then L is not solvable and its radical is nilpotent,
and moreover L = [L, L].
 THE DERIVATIONS OF THE LIE ALGEBRAS 33

By the above-stated, we can find many examples of Lie algebras with outer
derivations. But there has not been found a Lie algebra with non-zero center
which has no outer derivations. In the next section we shall give an example of
such a Lie algebra. For the purpose of this we need to construct a nilpotent Lie
algebra which does not belong to the class D.

4. Example of a nilpotent Lie algebra which does not belong to the

class D.

THEOREM. There exists a Lie algebra which has no outer derivation

and has non-zero center.

To prove the theorem, we shall construct the nilpotent Lie algebra N of

dimension 38, and further the Lie algebra of dimension 41 whose radical is N.
Let xl9 # 2 , m , #33 be the basis of N9 and let N be generated by the elements
xί9 # 2 , # 3 and # 4 . The multiplication in N is given in the following table, but as
for the multiplication which does not appear in the table, let it be commutative.

= =
|_#i, X2\ — # 5 l_#l> #4] — # 7 L#2> #4j #8
= = =
L#i> # 3 j #6 L#2> # 3 j — #7 #5 L#3> #4J #5

L#2>Xτ\
= X = =
\X\9 XQ] = Xg \\ L#3> #8J #15
L#l, ΛΓ7J = #10 L#2> #8J — #12 L#4> #6J = #14
[#1> #8J = #11 L#3> #βJ — #13 [#4> #7] = #15
[ # 2 , Xβ] = # 1 0 [#3, # 7 ] = # 1 4 [#4, # 8 ] = #16

[ # ! , #9] = #17 L#2> # l l J — #20 L#4> #13j — 3 0 # 3 5

[ # ! , # 1 0 ] = #18 [#2, #12] = #21 [#4, #14] = 20#3β
[#!, # n ] = #19 [#3, #13] = 60#34 [#4, #15] = 15#37
[#1, #12] = #20 [#3, #14] = 30#35 t#4, #lβ] =
[#2, #9] = #18 [#3, #15] = 20#3β
[#2, #10] — # 1 9 L#3> #16J —

[#1, #17] = #22 t#3, #lδ] = #29 [#6, # l l ] = — #30

[XU X18] = X2Z [X3, X19] = X30 [XQ, X12] = - X31
[#1, #19] = #24 [#3, #2θ] = #31 [#7, # 9 ] = - #29
= = =
L#i, #20] #25 L#3> # 2 l ] #32 L#7>#lθJ #30
=
[ # ! , ^2lJ ~ #26 L#4> #17] #29 L#7> # l l ] ~ — # 3 1
== = = z=
L#2> #17J #23 L#4> #18J #30 L#7> #12] #32
= = = =
[#2> #18J X24 L#4> #19] #31 [#8> # 9 ] #30
[ # 2 , # 1 9 ] = X25 [#4, #20] = #32 [#8> # l θ ] = — #31
= =
[ # 2 , #2θJ #26 L#4> #2lJ — #33 [#8> # l l ] #32
34 T. SATO

2> ^ 2 l J — "^27 L ^-6? ^ J — -^28 L ^8> ^ 1 2 J — *^33

3, X\i\ -~~ X21& L^βί •^loJ - ~ X29

- X$i \?£%i 3^26J — »^37 L ^'7> «^2θJ — (•*•/"/"^37

LXl, X 3 0 J — X 3 5 I X 3 , X27J — — *^38 L*^7> *^2lJ — ^/^/'>C38

— OX34 ^ 4 ? *^25J — \- L / z -'/*^37 L^δ? ^ 2 0 ] — V ^ / *-'/**'38

=
""^ "^35 L ^-'4J *^26J ~ \ ^-/*^/^38 L ^9> *-*-'lθJ ""^ Λ^34
= =
~~ •^'36 L*^6> *^18j — "<^34 L ^ Θ J •^-'llj *->X%§
1
— ^ - -/"/*^'37 L*^6> *^Ί9J — ^ ^ S δ L^-9> *^12J — «->^36
/ 1 / C \ ^ Γ'v 'r 1 9 ^ ΓT T 1 'r*
— ^ -L/ j y ^ S S L^6> ^-20] — ^ ^ 3 6 L^IO? ^ l l J — *^36
= =
— ^34 L ^6> *^2lJ ^•^'37 L^l
=
[^3, ^24] ~~ X35 YXT> -^17] =
~"4^34 [Xι
=
«^--36 L*2-7> •^-lδJ ""• "^35

We set C7 = [xu x2, X3, x*} - After the long and tedious calculation, we may
verify that the multiplication stated above satisfies Jacobi's identities. Using the
facts U7 = 0, [C/2, U2] = 0 and that ^:5 belongs to the center of N, we can abridge
the calculation a little.
Now we take the following linear endomorphisms of U :

50 : ^i ->x1? α: 2 -> - x2, x* -+X3, x ± - + - x±

S2 '. ^ 1 —>0, X2—>Xi, X3-+O, X^—^^Ά

Then © = [s0, su s2} forms a simple Lie algebra. We are going to enlarge s0, sϊ9 s2
to derivations of N. This is possible because N is decomposed into the direct
sum of the ©-invariant irreducible modules as follows :

N = {xu x2] Θ [xs, x,} Θ {x,} Θ {^β, xi - (1/2)^5, ^ } θ

{^9, XIQ, Xn, X12} 0 {^13> -^14» ^15? -

Φ l ^-22> Kit) ^24> «^25> ^-26? ^27J Φ

Φ {60^34, 30^35, 20.T36,15^37,

In regard to operation of ©, N has the structure indicated in the following

diagram. Here let { } be an ©-irreducible subspace, and we denote by *•
process of generating ideals, and by = identification of subspaces.
 THE DERIVATIONS OF THE LIE ALGEBRAS 35

{xQ,x7-{l/2)x5yx&}

Now we will prove that the nilpotent Lie algebra N does not belong to the
class D. Let ΰ be a derivation of N commutable with the elements of ©.
There is no direct summand of N which is ©-isomorphic to {xu x2] or {xs, x^}
but for themselves. Therefore by Schur's lemma, D must have the following
form :

i = axι

Dx2 = CCX2 +

Then D acts on N as follows :

x5 = [xu 1 + yx3, x2] + [xu ax2 + y r j = (2a + γ)x 5

δ^i, x 4 ] + \xs, βx± + δx 2 ] = (2/3 + δ) r 5

= [xl9 xA] - [x2y xs] ->[ΛTX + 7x3, Λ J + [#1, /3^4 + δx 2 ]

i] = (a + /3 + 2γ
36 T. SATO

xQ = [xlf x3] -> [ax1 -f yx3f x3] + [xu βx3 + fcrj = (Λ + β)x6

x9 = [xl9 x6] -> [axι + γ^ 3 ,

= (2a + /β).rβ + yx13

0 = fo, ^ β ]

Hence we have

γ = δ = 0,

and this implies that a = β. Then,

^13 = [x*9 xQ]->[ax3, xQ] + [x3, 2ax6]

As fe^ia] =60^34, we get

ct = β = y = δ = O, namely D = 0.

Therefore a derivation of JNΓ which is commutable with the elements of © is only

0, and this implies that N does not belong to D.
When we take a semi-direct sum ©+JV, it has a non-zero center {x5}, and
it has no outer derivations by Proposition 10. Hence the theorem is proved. We
remark that [© + N9 © + N] = © + JV, which is a demand of Proposition 13.

REFERENCES

[ 1 ] N. JACOBSON, A note on automorphisms and derivations of Lie algeras, Proc. Amer. Math.
Soc., 6(1955), 281-283.
[ 2 ] G. LEGER, A note on derivations of Lie algebras, Proc. Amer. Math. Soc, 5(1953),
511-514.
[ 3 ] G. LEGER, Derivations of Lie algebras, III, Duke Math. J., 30(1963), 637-645.
[ 4 ] T. SATO, On the derivations of nilpotent Lie algebras, Tόhoku Math. J , 17(1965), 244-249.
[ 5 ] S. TOGO, On the derivation algebras of Lie algebras, Canad. J. Math., 13(1961), 201-216.
[ 6 ] S. TOGO, Outer derivations of Lie algebras, Trans. Amer. Math. Soc, 128(1967), 264-276.

FACULTY OF ENGINEERING
IBARAKI UNIVERSITY
HITACHI, JAPAN