JOURNAL OF MATHEMATICAL PHYSICS VOLUME 4, NUMBER 4 APRIL 1963

Lie Algebraic Solution of · Linear Differential Equations

JAMES WEI ..u�n Enw,o�.Ro NoR�t,o�.:-.o

Socony Mobil Oil Company, Inc., Reseo.rch Departmmt, Paulsboro, NeuJ Jersey
(Received 13 November 1962)

The solution U(t) to the linear differential equation dU/dt h(t)U can be represented by a finite
product of exponential operators; In many interesting cases the representation is global U(t) =
-

�1(t}HJ exp (g.(l)H2] exp(g.(t)H.] where g,(t) are scalar functions and H, are constant opera­
tors. The number, n, of terms in this expansion is equal to the dimension of the Lie algebra generated
• • .

by H(t). Each term in this product has time-independent eigenvectors. Some applications of this
solution to physical problems are given.

I. INTRODUCTION This is an exact solution.

ONSIDER
C order a linear differential equation of the The scalar functions g1(t) are the solutions to a
set of nonlinear differential equations
first

dU(t)jdt = H(t)U(t), U(O) =I, (1) g1(0) = 0, (6)

where Hand U are tim�ependent linear operators
in a Banach space or a fini�ensional space. where 'lllt are nonlinear functions of the g's. Thus
The operator H(t) is a given function of time. The we have reduced the linear differential equations of
initial value of U is the identity operator, and one (1) to the nonlinear differential equations of (6),
seeks the wlution U as a function of time. remini?;cent of the reduction of the linear Master
The operator H often appears, in physics as · he equation to the nonlinear Boltzmann equation. For
t

Hamiltonian in an equation of motion, or a.s a the important case where Lis a solvable Lie algebra,
transition probability matrix in a Master equation. the '11; • form a triangular matrix, and the system (6)
It usually has a finite basis, and can be written as can be solved by quadrature.
Approximate and formal solutions of Eq. (1) in
a finite sum
.. terms of a single exponential have been given by
H(t) =L:a1(t)H;, (2) Birkhoff/ Feynman,2 Kubo,1 and Fer.• Magnus, u
showed rigorously that the solution U(t) can be
i -l

where a1(t) are a set of linearly independent complex expressed locally as

valued functions of time, and H 1 are constant
operators. The set of operators {H1 • • • H } may ..
(7)
be enlarged by repeated commutation to a Lie
algebra L. The number of linearly independent where the f1(t) are a set of scalar functions of time
operators H 1 in L (the dimension of L) is often a satisfying a set of nonlinear differential equations.
finite number n 2:: m; this is always the case if H(t) Magnus also showed that if two sets of operators
is a finite matrix operator. have the same abstract Lie algebra and expansion
It will be shown that if Lis finite dimensional the functions at(t), they will have solutions in the same
I

system of Eq. (1) with a time-dependent operator form with the same functions f1(t); other properties
can be uncoupled into a set of independent systems of H(t) become irrelevant to the problem. It will
U(t) = U1 (t)Uz(t) . U,.(t),
. ·
(3) be shown that this often results in the reduction of
many separate problems to one problem. If the
where each component U1(t) is an operator satisfying
dimension n of the Lie algebra is smaller than the
1t U;(t) = g,(t)H1U1 (t), U1(0) = I,. (4) dimension of the space on which H(t) operates (this

where Q;(t) is a complex valued function of time. The 1 G. Birkhoff, J. Math. Phys. 16, 104 (1937).
1 R. Feynman, Pbys. Rev. 84, 108 (1951).
solutions to (4) are U1(t) exp [g1(t)H1J, and U(t)
= 1 R . Kubo, J. Chern. Phys. 20, 770 (195 2).
becomes a product of exponentials ' F. Fer, Bull. Classe Sci. Acad. Roy. Belg. 44, 5, 818
(1958).
..
' W. MO¥n�s, Commun. Pure Appl Math. 7, 649 (1954).
U(t) = II exp [g1(t)H;}. (5) • J. Ma.nam and W. M
lij5II us, Research rept. no . :BR-37'

j•1 Y
New York University, New ork, 1961.
575
576 J. WEI AND E. NORMAN

is certainly the case if the operand spaceis a Banach abstract Lie algebra if they have corresponding ba.seS •
·

space), the difficiJ.J.ty of the problem is reduced. which give riSe to the same structural constants.
These properties of Magnus ' form of the solution An example of a Lie algebra is the set { Q, P, I}
of Eq. (I) are retained by the form given by Eq. (5). with the commutation rules
One of the advantages of (5) is that it is globally
[Q, P) =ihl, [Q,/]=0, [P, I] =0.
valid in a number of interestin g cases. Another
advantage of (5) is its relative ease in computation A sub8et S of L is called a subalgebra if it is closed
and added physical insight. As Eq. (7) is the ex­ under the operations of addition, multiplication by a
ponential of a sum, U(t) must be reevaluated for scalar and commutation. A suba.lgebra S is called an
each value of t by a power series or an eigenfunction ideal if the commutator [X, Y] of X E S and Y E L
expansion; in Eq. (5), the eigenvalues and eigen­ is in s.
functions of each constant operator H1 can be The set of those elements of L which are the result
evaluated once for all times, and may possess some of commutation of two Lie elements forms the
physical interpretation. The method of Magnus as derived algebra L'. L' is an ideal of L. The derived
outlined by Wichmann7 requires a finite-dimensional algebra of L' is denoted by L". By induction, one
representation of the Lie algebra-not necessarily a forms the derived series
trivial problem unless the algebra is semisimple. No L :) L' :) L" :) · · · .
such representation will be required in the method
of this paper. A Lie algebra Lis said to be solvable if L<A> = {0}
A brief description of the Lie algebraic terms rele­ for some h; this clearly means the rest of the terms
vant to this paper is given in this section.•·e in the derived series are also zero.
The solution U(t) lies in a subset of all linear The union of two solvable ideals is again a solvable
ideal; the union of all solvable ideals is called the
·

operators GL; this subset is called the associative

algebra R generated by H(t). The algebra R consists radical. An algebra is said to be semisimple if its
of all operators in H(t) plus all products radical is { 0}. An algebra is said to be simple if it
has no ideal other than Land { 0}, and if L' ¢ { 0}.
01· =I to m
' The lower centr� series is constructed by re­
R, = I 2 ... labeling L2 = L', an'd'·qefining L•+l to be all the
#J, ' ' 3 , ,
- elements that result from commutation of elements
1
and all linear combinations of such products. from L with elements from L", i.e. L"+ = [L, L"). L :)
Every associative algebra gives rise to a Lie or L2 :) L8 :) • • • A Lie algebra L is said to be nil­
•

commutator algebra L. We shall consider the Lie potent if LA = {0} for some h.
algebra generated by H(t). The Lieproduct of ele­ Examples of these terms will be given in Sec. III.
ments H1 and H1 is [H1, H.] or H;Hk - H$1.­
II, THE UNCOUPLING THEOREM
The Lie algebra L consists of all operators in H(t)
plus all Lieproducts The results of this section are contained in a
paper by the present authors.10 It is, however,
[H"' [H.. ,[H... · · · [H.,,_, H.,,] · · · ])) 011 = 1 to m, worthwhile to present here the proofs of Eqs. (5)
and all linear combinations of such products. and (6) so that the calculations in Sec. III of this
Every element of L generated by H(t) is also an paper will be meaningful.
element of R generated by H(t). The reverse is not The main theorem is based upon two lemmas.
true; H1H2 - H2H1 is a Lie element, but H1H2, in (i) (Baker-Hausdorff) If X, Y E L, then
general, is not. The Lie algebra L is thus a subset e.zYe-x E L, and exYe-x = Y + [X, Y] +
of R, and is defined by the commutation structural [X, [X, Y]]/2! + [X,. [X, [X, Y]]]/3! + · · · ·
constants Let us define the linear operator adX, X E L by
"
the equation
[H,., H.] = L 'Y;1H,. (adX) Y = [X, Y], Y E L.
Powers of the operator adX are defined in the usual
Two Lie algebras of operators have the same
way and thus
'B. H. Wichm:J.on, J. Math. Phys. 2, 876 {1961). 2
aN. Jacobson, Lie .Aluebras (Interscience Publishers, Inc., (adX) Y = [X, [X, Y]], etc.
New York, 1962). ----

• N. Bourbaki, Elements de Mathematique (Hermann & 10J. Wei and E. Norma.n, "On Global �presentations of
Cie, Paris, France, 1961), Vol. 26, Chap. on "Algebres de the Solutions of Linear Differential Equations as a. Product
Lie." of Exponentials" (to be published in Proc. Am. Math. Soc.).
 LINEAR D IFFERENTIAL EQUATIONS 577
..

In terms of ad)( the Baker-Hausdorff formula .cari

be rewritten as
� a,(t)H, = � g,U)[ g exp (g;H;) J
(ii) Let H1, H2, H,. be a basis for L. Then
(8)
X H, [ .tr
,-,-1
exp (-g1H1) ]
� g,(t >[ ti exp { g1adH1} ] H,.
•
• • ,

[ tr exp (g;H;) ]H,[ U exp (-g1H;) J

(14)
t•l 1 •r
Application of lemma (ii) to the tel'IDB on the right
r = 1 ' .. · n ' '
(9) of Eq. (14) yields
.
• ft ft

where each. ti• �t• (gt, · • · , g,) is an analytic

==
L ak(t)H�e = L E g,(t)��:;H�;. (15)
function of its arguments.
Since the operators H are linearly independent, we
The proof of lemma (i) is accomplished by direct
�:

have a linear relation between the a�e(t) and the

computation. Equation (9) of lemma (ii) follows by
repeated application of lemma (i), and the analyticity (J,(t). 'fhe elements t�:• of the transform matrix �
of the ��:; is easily established. are analytic functions of the g, 's;
Theorem. Suppose the linear operator H(t) can be
expressed in the form a � g
!
a a tu . .. �;.. fit
"'

H(t) = 2:, a1(t)H; m finite, (10) a2 •

02
-
• •
' g(O)= 0. (16)
where the a1(t) are scalar functions of time, and the
•

• •

-
H1 are. time-independent operators. Let tfe Lie �ol ... �.... g..
algebra L generated by H(t} .�e of finite dimens�on n.
Then there exists a neighborhoo-d of t= 0, in which Since the tk; are analytic functions of g, we have
the solution of the equation. that the determinant A of � is an analytic function
of g. We also know that at t= 0 � = I, and hence
dUjdt = H(t)U, U(O) = I ·.
(11) A(O) ;;C 0. These two facts show that there must
may be expressed in the form exist a neighborhood No of f = 0 in which A ;;C 0,
i.e. in which � is invertable. We can thus write (16)
U(t) = exp [gt(t)Hd in the form
X exp [g2(t)H2} · · · exp [g.(t)H,.J, (12) dgjdt = f(a, g) = F'a, g(O) = 0. (17)
where H1, ··· is a basis for L, and the g,(t)
, H,. Since �- � is analytic in No, we are assured of a
are scalar functions of time. Moreover, the g,(t) neighborhood of t = 0 in which the solution of (17)
depend only on the Lie algebra L and the a,(t). exists and is unique. This completes the proof of
Proof: First note that we might just as well write the theorem.
A(t)= L�-� a,(t)H, instead of A(t) = 2:,;-_, a,(t)H,, The authors have shown11 that the uncoupling
by simply setting a,(t) =
0 for i > m. Note also theorem is global for all solvable Lie algebras, and
that at time t = 0, U(O) I is in the form (12) =
for the real "split 3-dimensional" simple Lie algebra.
with all g,(t) = 0.
Now let U be of the form (12). Since ill. SOME PHYSICAL EXAMPLES

�= � g (t{ g exp (g1H1)J

,
(a) The method of uncoupled exponentials will
be illustrated in this section by examples from low­

x,[fr exp (g1H1) ] ,

dimensional Lie algebras. Many physical systems
(13) give rise to low-dimensional Lie algebras.
X
·-· There is only one one-dimensional Lie algebra,
.. which is of course Abelian. For any Abelian Lie
x r x+r
AU= L a,(t)H,· U. algebra, e e = e , and the solution of our problem
is trivial.
We obtain, upon substitution of (13) into (12) and
post-multiplication by the inverse operator u-', u Reference 10, Theorem 2.
578 J. WEI AND E . NORMAN

There are two distinct two-dimensional Lie These operators also arise in the problem of t4e·
6
algebras, one of which is Abelian. The non-Abelian servicing of machines in queueing theory .1
algebra was discussed by Sa�k12 as the 11Quantum This algebra is solvable since the derived series is
Mechanical Shift Operator." This algebra is spanned
•

i'\'

L= {X, Yl, L' = {Y}, L"= {0}.

by the operators X and Y with the commutation
rules The solution of the equation

[X, Y] = Y. (18) dUIdt = H(t) U = (a(t)X + b(t) Y) U U(O) = I

The Landau-Teller transition probabilities of a can be written as
system of simple harmonic oscillators18 is a realiza­ (f(t)X exp g(t) Y].
U(t) = exp ·

tion of the above algebra in an infinite-dimensional e

Equation (14) gives aX + bY = /X + g( •"'.r)Y,
space. Defining the ra.ising and lowering operators as
and Eq. (8) gives (e"41x)Y = e1Y. One obtains then
..

[:J � e� [:J
R= L (j + 1)(-E;.; + E;+t,;), =
;-o

..
or
s= L j(-E;.; + E;-1.1) ,
/= a
we obtain the commutator (/ = be-',
"' which can easily be solved by quadrature. Thus we
[R, S] = R +s = L {-(2j + l)E;,; have reduced many different problems to the same
quadrature problem.
+ (j + 1)(E;.t+t + E;+!.;)}, (b) The simplest non-Abelian three-dimensional
where E;t is the infinite matrix in which all ele­ Lie algebra is spanned by the operators { Q, P, I}
ments are zero except in the (j, k) position where with the commutation rules
the entry is unity. This algebra is abstractly identi­ [ Q , P] = ihi, [Q, I] = 0, [P, I] = 0. (19)
cal to the algebra of (18) where R is isomorphic to X,
This algebra is solv.able, and it is well known that
and R +Sis isomorphic toY.
A second realization is provided by the kinetics eO+P= eoePeH•In.
of the deuterium exchange reaction!• It is aga.in a The operators for the scattering of x rays by
Master equation problem with only nearest-neighbor crystal Iattices16 form a 4-dimensional Lie algebra
transition probabilities, but now the operand space which is an easy extension of the algebra of (19).
is finite dimensional. The operators are given by It is spanned by the operators {Q3 + P, Q, P, I}
•- with the additional commutation rules 1
1
R'= L (s- J)(-E;.; + E;+I.;),
i-0 [P2 + Q2, QJ= - 2if�P [P2 + Q2, P] = 2ihQ, (ZO)
,

•
[P2 + Q2, I) = 0 .
S' = L j(-E;.; + E;-t ,;),
These two algebras are solvable since
which possess the commutator = {P + Q2, Q, P, I}, L' = {Q, P, I},
L
•
R', S]' = R' - S' = 2: (2j- s)E;,; L" = {I}, L"' = {0},
f-0 The algebra of (19), (20) can be expressed more
•- 1 •-1
conveniently in terms of the bases
L (j + 1)E;.i+I + Lo (s - J)Ef+L;,
;-o f- {w = P2 + Q\ X = Q - iP' y = Q + iP' II j .
where s + 1 is the dimension of the operand space, the commutation rules become
and E;• is an (s + 1) X (s + 1) matrix. This algebra
[W, X]= 2hX, [W, Y] = -2hY, [X, Y] = -2hl.
is again isomorphic to the algebra of (18) where - S'
is isomorphic to X, and R' - S' is isomorphic to Y. If an equation of motion is written as

u R. A. Sack, Phil. 1vlag. 3, 4\H (1958). 16 W. Feller, An Introduction to Probability 1'heory and
uE. W. Montroll and K. E. Shuler, Advan. Chern. Phys. Applications (John Wiley & Sons, Inc., New York, 1957),
P·
ys. 3,
1, 371 (1959 ). 416.
1• J. R. Anderson and C. Kemball, Proc. Roy. Soc. (Lon­
ath. Ph
u G. H. Weiss and A. A. Maradudin, J. M
don) A226, 472 (1954). 771 (1962).
 LIN EAR DIFFERENTIAL EQUATIO N S 579

•
dUjdt= {�(t)W + a,(t)X + aa(t)Y + a,(t)l}U, a l 0 g h
2 -2A
and the solution is ge g '
•

.
+•A
U = exp [gt(t)W] exp [g2(t)X]·
c tO 0 e - f
or
exp [g8(t) Y] exp [g,(t)I],
· ·

h = a - g e-2Ac, g = be2A - g"-e-2�c, j = e-2Ac. (22)

we have from Eqs. (8) and (14):
� 0 This algebra is not �lvable, and the scalar func­
1 0 0 Yt
tions h, g, and f canno t be found by quadrature. The
lla 0 e'Hr•• 0 0
•

!72
'
solution of the system (22) is reduced to quadrature
aa 0 0 e-ur,l 0 Oa after solving the Riccati equation

a, ' .
0 0 -2hg 2 1 g, u - 'l.l + p(t)u + q(t) = 0,
or p(t)= -c/c, q(t)= be - a2 - a + (cjc)a,
g�= �.
satisfied by A(t).
=
{J3 e'ltr•·aa, g, = a, + 2hg�2*" aa . The general continuous-time Markov process in
These equations are easily solved by quadrature. three-dimensi ?nal space co�sists of six independent
operators whtch form a Lie algebra {E:u - E11;
All of the algebras discussed in examples (a) and_
- Eu; Et� - E22; !5az - En; Est - Eas;
(b) are sol�able; hence the given solutions are En
global. Eu - Ess}. This algebra lS not solvable. A general
(c) A celebrated three-dimensional Lie algebra is reduction. princi�le and method will be given in the
the 11split 3-dimensional simple algebra" charac- next section which can be used to decompose any
terized by Lie algebra into the solvable radical and a number
· . of simple subalgebras. We shall be content to state
= 2F. (21) the results of the decomposition here.
-
[E, F]= H, [E, H] 2E, [F', H] =
These rules of commutation are satisfied in a finite- The Lie algebra L is decomposed to the sum
dimensional operand space by the three components L= R + S,
of angular momentum for a rotator1 7 where where R is the solvable radical spanned by
E = (m + imN)/ih,
.. F= (m.. - im.)/ih1
-1 0 0 -1 -1 -1
H = -2m./h.
A= 0 -1 0 I B= 2 2 2 '

In an infinite-dimensional operand space, this is

1 1 0 -1 -1 -1
realized by the simple harmonic oscillator18 where
E= Q2/2ih, F = P'/2ih, H = (QP + PQ)/2ih. 1 1 1

Given. this Lie algebra and the dimension of the G= 0 0 0 I

representation, it is known that these are the only -1 -1 -1

ir.reducible representations in the sense of iso-
morphism. and S is the simple subalgebra spanned by
For a simple harmonic oscillator where the re­
0 -t 0 -1 -� 0
storing force is time-dependent, the wave equation
may be written as E= 0 0 0 I F= 2 1 0 I

dUjdt= (a(t)H + b(t)E + c(t)F)U, IO ! 0 -1 -t 0

and the operator U is expanded as
-1 -1 0
U(t)= exp [h(t)H] exp [g(t)E] exp fj(t)F];
· ·
H= 0 1 0 .

then Eqs. (8) and (14) give 1 0 0

17 P. A. M. Dirac, Quantum Mechanics (Oxford University
Presst.N 81 as
a subalgebra, has the commutation rule of
.. ew York, 1957).
18
J:l.efeer nce 17. Eq. (21). The rest of the multiplication table is
580 J. WEI AND E. NORMAN

E F H A B c f nition 1. A Lie algebra L is said to be the

Dei
.
semidirect sum of the subalgebra L1 and ideal L,,
A 0 0 0 0 -B -c
L = L, E9 L2, if L, (\ L2 = 0, and if every element
B c 0 -B B 0 0 x E L can be written uniquely as a sum x = x1 + x,
with x1 E L, and x2 E £2. If L1 and L2 are both
c 0 -B c c 0 0.
ideals of L then the sum is said to be dired.
In this example and in example (c), the solution is The definition can be extended in an obvious way
again global. to any number of summands.
. Another complicated algebra is that of the proper Structure theorem. A finite-dimensional semisimple
1
orthochronous inhomogeneous Lorentz group. g The Lie algebra L may be decomposed into the di rect
Lie algebra is 11-dimensional with the following sum L L 1 ffi £2 E9 · · · EB L,, where the L, are
=

elements and commutation rules: ideals which are simple algebras.

II, H, P1, P2 , Pa, J1 1 J,, Js, d1, IJ2, Sa}; Levi's theorem. If L is a finite-dimensional Lie
algebra with radical R, then there exists a semi­
[H, P.] = [H, J,] = 0; [H , s.] = -iP,;
simple subalgebra S of L such that L is the semi­
[P1, P;) = 0; [P1, J;] = iEm,P ; �: (23) direct sum L = S ffi R.
In the equation dU/dt = H(OU, where H(t)
[P,, tJ1l = -io��I;
generates L, the decomposition L = S EB R gives
[J;, Jd = iEmJAj [J;, d;) = iE;;��;j rise to the corresponding decomposition, H = H s +
[/J;1 IJ;) = -iE;;�;J�;, HR of H, where HR E R and Hs E S. Let
where
E12a = E211 = E312 = I,
dUjdt = HU = (Hs + HR)U. (24)
Es21 = Eu2 = E2n = - 1j Define UR and Us by
all other E;;Jt are zero. �
d s
= H8U8,
d� R
=
1
(US HRUs)UR·
It will be shown that this algebra can be decomposed
into a. five-dimensional radical and two three­ Since R is an ideal in L_, we see [by lemma (i)] that
dimensional simple subalgebras. U&1 HR U8 is in R. It is easy to verify that U . Us UB
··
-
IV. REDUCTION PRINCIPLE AND METHOD
satisfies (24). The fact that R is solvable makes it
easy to find UR once Us has been found. To reduce
If, in the equation dU jdt = H(t)U, the Lie the task of finding Us we make use of the structure
algebra L generated by H(t) is solvable; then Eq. (17) theorem. Write s = sl EB s2 EB ... EB S,. Then
can be solved by quadrature. The task for high­ H = H1 + · · + H, where H, E S,.,Making use
·

dimensional algebras is more tedious-but no more -of the fact that [U , H;] = 0 fori � i, it is easy to
,

difficult-than for low-dimensional algebras. On the see that Us = U1 U2 · · · U,, where U, satisfies the
other hand, if L is not solvable, then the difficulties equation dU,jdt = H,U,.
in solving (17) mount rapidly as the dimension of L We shall now describe a method by which one
inc reases. In this section, we present a method for can carry out in practice, the reduction embodied
reducing a Lie algebra. to its component parts. If in the structure theorem and Levi's theorem:
the algebra is solvable, no reduction is needed; if Definiticn 2. A subalgebra X of a Lie algebra L
it is simple, no reduction is possible. However, as is called a Carlan subalgebra if (1) X is nilpotent
indicated in the preceding section, there do exist and (2) the set of all x E L such that- [x, h] E X
physically interesting algebras which are neither for every h E X (the normalizer of X) is simply
solvable nor simple. X itself.
First we shall present the relevant theoretical The construction of a Cartan subalgebra is the
results, and show how they are used in the reduction first step in carrying out the reduction. Frequently
of our problem. Proofs of these results may be found a. Cartan subalgebra can be found by inspection.
1
in the literature,20'2 but these proofs are of the Failing this, one may be constructed by first finding
nature of existence proofs and thus are not very a. regular element.
useful in pract.ire. Definition 3. An element h E L is called regular
u J. S. Lomont and H. E. Moses, J. Math. Phys. 3, 405 if the number of independent elements x E L so
(1962). 0 for some k, is minimal.
Reference 8.
20
that (adh)�x =

n Reference 9. If h is a regular element, then the set X defined

 LINEAR DIFFERENTIAL EQUATIONS &81
.
TAB
= L= E
= = I=. =======
by X = {x E L I (adh)1x = 0 for some k} is a . ===== ===

Carlan subalgebra. ·
operator eigenva.lues multiplicity
There exists a straightforward method for finding
a regular element.22 ad I 0 11
We will now illustrate the way in which a Cartan 0 11
0
adH
ad PI 11
subalgebra is used to effect the decomposition of L. 0, 1, -1 5, 3, 3
0, i, -1
ad Jl
The general method is easily inferred from. the ad dt 7, 2, 2
example.
A Cartan subalgebra of the proper orthochronous
BLE
TA
inhomogeneous Lorentz group can be determined by II.
inspection to be X = {I, H, dt, Jt, Pt}. Next, we
determine all the eigenvalues of the linear operators operator eigenvalues generalized eigenvectors
adi, ad.H, .etc: The matrices corresponding to these
ad! 0 all elements of L
linear transforms are quite simple, and the eigen- adH 0 all elements of L
.· values are readily found and given in Table I. ad P, 0 all elements of L
ad 11, 0 I, H, d., J,, P., Ps, P,
lis- J,, 11, + J,
Next we define an element x of L to be a
i
generalized eigenvector of h with eigenvalue X if - t
0
lis+J,, lis- J,
I, H, 11., J., P1
(abh - X)kx = 0 for some positive integer k. The ad J,
1 lis - ida, J2- iJ,, P�- iPa
generalized eigenvectors are found and tabulated - - 1 lit + id ,, J2 + iJ,, P� +iP,
in Table II. '
Next, by a suitable linear combination, one ob­
tains a set of simultaneous eigenvectors to these TABI.E Ill.
five operators, that span L. The simultaneous
generali,zed eigenvectors and the corresponding
-

eigenvector ad I adHad P, ad 11, ad J1

eigenvalues are tabulated m'..Table III. \
0 0 0 o· - 1
'
.
At this stage, it is convenienfto P, +iPs
0 0 0 0 1
remove the ele-
P2- iPa
ments belonging to the radical from Table III. The (lit 0 0 0 . t -1
0 0 0
J,) +i(lla + J2)
.
-

algebra is (lit + Ja) + i(da - J,) -t - 1

0 0 0
method of finding the radical of a Lie
(lit + J,) - i(lla - J,) -t 1
quite straightforward, and is fully described by (lis - J4) i(da + J,)
- 0 0 0 1. I

·-· ---·---·-- ·- .. ·-======

Jacobson.23 The radical of this algebra is found to ·
···· - ·-·---- ...·..·-·
==,=-:=:==-
--
be R = II, H, Pt, P2, P3}. After deletion of the
--· -· ---

radical, Table III becomes Table IV. TABLE IV.

The contents of Table IV forms a semisimple
algebra of six dimensions. Thus we have accom­ eigenvector ad J,

plished the decomposition given by Levi's theorem,

L = S EB R. The next step is to decompose S D � (112 - J,) + i(lla + J2) t - 1
B = (112 + Ja) + i(da - J,) - � - 1
into simple subalgebras according to the struc­ C = (d2 + Ja) - i(lla - J2) l 1
l 1
-
ture theorem. By a linear combination of the ad­ A = ( lit - J,) - i(lla + Js)
joint operators, we replace Table IV by Table V.
=
By noting that [B, A] = 2(J1- id1) and [D, C)
TA LE V.
Z(J1 + id1), the decomposition is now complete: B

eigenvector ad(J 1 - ill,) ad(/,+ ill,)

where
A 2 0
R = {I, H, PI, P2, P3} 1 B -2 0
c 0 2
= D 0 -2
L1 fJ1 - id1, (d2- Ja)- i(d3 + J2),
(d2 + Js) + i(da- J2)} 1

=
L2 {JI + id1, (d2 + Ja)- i(d3- J2), ACKNOWLEDGMENT

(d2- Ja) + i(da + J2)}. The authors are indebted to Professor Wilhelm

Reference 8, p. 60. Magnus for his stimulating discussions and en­

22
21
Reference 8, p. 73. couragements.