The algebra and geometry ofSU(3) matrices

PRAMANA © Printed in India Vol. 49, No. 4,

--journal of October 1997

physics pp. 371-383

The algebra and geometry of SU(3) matrices

K S MALLESH and N MUKUNDA*

Department of Studies in Physics, University of Mysore, Mysore 570 006, India

*Centre for Theoretical Studies and Department of Physics, Indian Institute of Science, Bangalore 560012, India

*Also at Jawaharlal Nehru Centre for Advanced Scientific Research, Jakkur, Bangalore 560 064, India

MS received 13 March 1997

Abstract. We give an elementary treatment of the defining representation and Lie algebra of the three-dimensional unitary unimodular group SU(3). The geometrical properties of the Lie algebra, which is an eight dimensional real linear vector space, are developed in an SU(3) covariant manner.

T h e f and d symbols of SU(3) lead to two ways of 'multiplying' two vectors to produce a third, and several useful geometric and algebraic identifies are derived. The axis-angle parametrizafion of SU(3) is developed as a generalization of that for SU(2), and the specifically new features are brought out. Application to the dynamics of three-level systems is outlined.

Keywords. SU(3) matrices; octet algebra; octet geometry; SU(3) axis-angle parameters.

PACS Nos 02.20; 03.65

1. Introduction

The unitary unimodular group SU(2) in two complex dimensions is the simplest nontrivial example of a nonabelian compact Lie group. Its many uses in physics-spin of the electron, proton, neutron ... isotopic spin of nucleons, description of two-level atoms and two-level quantum systems in general are very well known. At the same time its adjoint representation coincides with the three-dimensional real proper rotation group SO(3), with its associated concepts of three dimensional vectors in R3 and their algebra.

The Pauli matrices oj, j = 1,2, 3 mediate in a natural way between the defining two- dimensional and the adjoint three-dimensional representations. Being the generators of the defining representation, the expression of a finite SU(2) element as the exponential of a generator in closed form is also well known. Thus one has the familiar collection of results:

[aj, as] = 2i~jkt at,

{aj, as} = 26jk; (1.1a)

a . a b . a = a . b + ia_^b.a_; (1.1b)

a(6, 0) = exp(i0& • _a) = cos 0 + i&. a_ sin 0 c SU(2),

I & l = I , 0 < 0 < 2 7 r . (1.1c)

371

K S Mallesh and N Mukunda

Here a, b are (real) three-dimensional vectors, ejkt is the Levi-Civita symbol (structure constants of SU(2)), and (&, 0) are axis-angle coordinates for the general element a(&, 0) E SU(2). The two-to-one homomorphism SU(2) ~ SO(3) determines an element R(a) E SO(3) for each a E SU(2):

R(a)j k : 1Tr(aj acrk at),

R(a')R(a) = R(a'a), a',a E SU(2). (1.2)

The next group after SU(2) in the classical unitary family is the eight-dimensional group SU(3) of unitary unimodular matrices in three complex dimensions. Many representation theoretic complications expected for compact Lie groups in general, but not yet seen with SU(2), do show up with SU(3), making it quite nontrivial in comparison. Its use in elementary particle physics [1, 2] often exploits the canonical subgroup chain SU(3) D U(2) D SU(2) D U(I), while its use in the nuclear physics context involves the chain SU(3) D SO(3) D SO(2). The description of three-level systems [3-5] in general quantum mechanics (atoms, for instance) also involves SU(3).

The purpose of this paper is to present a generalization of relations of the forms (1.1, 1.2) from SU(2 ) to SU(3), bringing out the algebraic and geometric features of the eight- dimensional octet or adjoint representation of SU(3). In doing so we describe the minimum and unavoidable new features in both algebraic and geometric aspects that one must accept in the SU(2) ~ SU(3) transition. Among real eight component vectors in

~8, apart from the Euclidean inner product, two kinds of bilinear products of vectors leading again to vectors - one antisymmetric and the other symmetric - play important roles, and are essential in developing a formula generalizing equation (1. lc). One of our results will indeed be an axis-angle description of SU(3) elements, namely a closed-form expansion of the exponential of a general matrix in the Lie algebra SU(3) of SU(3) yielding a general finite SU(3) matrix. This will be seen to be considerably more complicated than the SU(2) result (1.1 c). In general our aim is to develop useful identities which help in getting closed form expressions, and to build up geometric pictures in some situations.

The contents of this paper are arranged as follows. Section 2 recalls the definition of the group SU(3) and the generators - the A-matrices - in the defining representation.

From their commutation and anticommutation relations the structure constants frst and symmetric invariant tensor drst can be read off. Their independent nonzero components are listed. Section 3 discusses the eight-dimensional adjoint or octet representation of SU(3). Based on the available invariant t e n s o r s

frst,

drst, two kinds of vector products among octet vectors - elements of R8 _ are defined: an antisymmetric wedge product and a symmetric star product. Both are SU(3) covariant. Apart from the geometric expression of the trilinear Jacobi identity using wedge products, several other identities involving these products and the Euclidean scalar product on R8 are developed. In § 4 we take up a detailed analysis of the algebraic properties of a single generator matrix in the defining representation of SU(3). The geometric tools of §3 are used to get convenient forms for products, inverses, powers, determinants and the minimal equation for a general three dimensional generator matrix. A convenient way of characterizing the eigenvalue spec- trum of a (suitably normalized) generator matrix, and the notion of its 'rest frame' or specific diagonal form, are developed. At all stages the SU(3) covariance of the 3 7 2 Pramana - J. Phys., Vol. 49, No. 4, October 1997

The algebra and geometry of

SU(3)

matrices

procedures is kept in view. Section 5 introduces the concept of axis and angle parameters for SU(3). This is a way of describing general one-parameter subgroups in the group. The essential differences compared to SU(2) are emphasized. We also obtain a closed form expression for a general element of SU(3) expressed as the exponential of a generator matrix; for this the method of going to the 'rest frame' is exploited. Section 6 briefly describes the features of Hamiltonian dynamics for three level quantum systems, based on a generalization of the Bloch spin equation familiar from two level systems. Section 7 contains some concluding remarks.

2. The defining representation of SU(3) and the A-matrices The defining representation of the group SU(3) is given by [6, 7]

SU(3) = {A = 3 x 3 complex matrix

[A~A

= 1, det A = 1}. (2.1) This is an eight-parameter compact Lie group. The U(2) and SO(3) subgroups are identified (up to conjugation) by [8]

U(2)= {A(u)= (~

(det u ) _ 1 ) u E U(2)} C SU(3);

0 (2.2a) (2.2b)

SO(3) = {A = 3 x 3 real matrix

IArA

= 1, detA = 1} C SU(3).

The generalization of the Pauli matrices trj, in a form adapted to the U(2) subgroup, leads to the eight hermitian traceless generators A~, r = 1 , 2 , . . . , 8 defined as follows [6,7]:

(i a i) (i-ii) (i°i)

A 1 = 0 , A 2 -~ 0 , A3 = --1 ,

0 0 0

(!0!) (!0/i (!0!)

) t 4 -~- 0 , )~5 "~- 0 , A 6 ~-- 0 ,

0 0 1

A7 = 0 i , A8 = ~ 1 .

i 0

These are trace-orthonormal in the sense

(2.3)

(2.4) Tr(A~As) =

26rs,

r,s = 1 , 2 , . . . , 8.

The commutators and anticommutators among the A's, lead to the completely anti- symmetric structure constants

frst

of SU(3) and to the completely symmetric d-symbols (for which there are no SU(2) analogues):

[At, As] ⁼

2ifrs, At,

{Ar, As} = 4 tSrs ~-

2d,,, At;

^(2.5a)

Pramana - J. Phys., Vol. 49, No. 4, October 1997 373

K S Mallesh and N Mukunda

f123 : 1; f456 : f678 : - - 2 '

f147 : f 2 4 6 : f 2 5 7 : f 3 4 5 : f 5 1 6 : f 6 3 7 : 1/2; (2.5b) d l l s = d22s = d33s =- -dsss = l/v/3;

d146 : d157 : -d247 : 8256 : d344 : d355 : -d366 : -d377 : 1/2;

d44s = d55s = d66s = d77s = -1/2"v'~. (2.5c)

(Here only the independent nonvanishing components of frst and drst are given). The product of two A's involves three kinds of terms:

)tr)ts ~-- 2 ~rs "~ (drst -~- i f , st)A,. (2.6)

The U(2) subgroup generators are A1, A2, A3 (for SU(2)) and As, while those of SO(3) are A2, A5 and A 7.

3. The adjoint representation of SU(3) and the geometry of octet vectors

The adjoint representation of SU(3)[7] arises upon conjugation of the A's by general A E SU(3) and expressing the result in terms of the A's. It is a faithful representation, not of SU(3), but of the quotient SU(3)/Z3, where Z3 is the centre of SU(3):

Z3 = {a = e/~. llw = 0, 27r/3,aTr/3} C SU(3). (3.1) Thus we have a three-to-one homomorphism SU(3) ~ SU(3)/Z3. Each A E SU(3) is mapped onto an eight dimensional real orthogonal matrix D(A) = (D(A)rs) E SO(8), whose matrix elements are easy to calculate:

A E SU(3) --* aAra -1 = D(a)srAS, O(A)s ~ = ½Tr(AsAA~AI),

D(A')D(A) = D(A'A). (3.2)

Thus these matrices D(A) form a very small part of the full twenty-eight dimensional group SO(8). In comparison, the adjoint representation of SU(2) is the same as SO(3).

Let us denote general real eight component vectors in Rs _ octet vectors - by ~ , ~,

% . . . . Among them we have the usual Euclidean inner product

~ " fl : Otr/~r. (3.3)

We now define two 'vector products', one an antisymmetric wedge and the other a symmetric star, both of which lead to octet vectors again:

oz,__fl E 7"~8: (o~^fl_) r =frstasflt,

=

(0~:~ ~) r : v/3drstOLs/~t,

a * / 3 = f l * a . (3.4)

374 Pramana - J. Phys., Vol. 49, No. 4, October 1997

The algebra and geometry of

SU(3)

matrices

The basic SU(3) covariant properties of these products follow from the fact that fr, t and

drst are

invariant tensors:

A E SU(3): D(A)_~A

D(A)/3

= D(A)(0:A_fl),

D(A)0: , D(A)/3 = D(A)(0: •/3).

(3.5)

To express the components of ~A/3 and ~ , /3 in convenient forms in terms of those of 0: and /3, it is useful to assemble the components an, 0:5, 0:6, 0:7 of _~ into a two- component complex column vector ~b(0:):

7"~8: ~(0:) = (0:4 --i0:5

~ (3.6)

0:

c

-- _ 0:6

i0:7 J"

(This is related to the fact that the A's are adapted to the U(2) subgroup (2.29) of SU(3)).

Then we have these expressions for the components of 0:A/3:

( 0:__. ^/3)j = £jkl 0:k/31 -t-

l l m ~b(/3)tcrj ~b(o~);

¢g

(-~ ^~)s : ~ Im ~b(/3)t~b(0:);

¢(0: ^--fl) = 2 /31 at- i/32 -/33 a t- v~fl8 Similarly for the components of 0: */3 we have:

(_~ */3)j : 0:8/3j +/380:j + ?Re¢(_~)t~j ¢(_~);

(~-- * ~)8 :

aj/3j -

c~8/38 - ~Re ~b(/3)t~b(0:);

1 ( V~/33 -/38 v/3(/31 - i/32) ) ~b(~) + (~ ~ / 3 ) . (3.8)

~b(0: * _fl) = ~ \ V~(/31 + i/3z) -V~/33 -/38 -

Now we consider some cubic relations, identities involving triple vector products, with wedges and stars in various combinations. The first of these is just a statement of the Jacobi identity for the structure constants

frst

and involves two wedge products:

0:^ (/3 ^_7) + ~^ (_7 ^c~) + _7^ (~ ^/3) = 0. (3.9)

Other relations arise by calculating the triple product _~-_A/3-__A3'. __A in two ways and comparing the results. We have the equality

2 2i [2 1

-~~-~'~-x=3-v~-~*/3"Y+-J--~^/3"~+ Lg-~-~-7+~(~*/3)*'r

i ]

- (0:--A/3) A-7 + ~ ((~^/3) * "Y + (--~ */3) ~_7) "_A 2 /3*'Y + ~c~ "_/3^-7 + [~/3 -7_~ + ~-0: (/3' "7)

= 3V~0:" " *

i ]

- _~^(/3^_-r) + ~ (_~^(~, -Y)+ ~ , (/3A7_)) .A. (3.10)

Pramana - J. Phys., Vol. 49, No. 4, October 1997 375

K S Mallesh and N Mukunda

The a-independent terms lead to the obvious results

_~ •/3^_7 = s^/3._% (3.11a)

_ ~ . / 3 , 7 = a _ , / 3 . 7 , (3.11b)

similar to properties of the triple scalar product of vectors in 7~ 3. The a-dependent terms, upon separation of real and imaginary parts and some rearrangement, lead to further relations at the vector level:

_~, ( / 3 , 7 ) - 7 , (/3, ~ ) = 2(~ •/3 '7 - 7 •/3~) + 3~^(~^_7) , (3.12a) s^(/3 • "7) + 7_^(/3 * A) = ~ * (7^/3) + 7_ * (_~^/3). (3.12b) These are respectively antisymmetric and symmetric in the pair ~ , _7.

The relations (3.9,11,12) are the basic SU(3) covariant cubic vector relations involving three independent octet vectors.

4. Algebraic relations for SU(3) generator matrices

A general traceless hermitian three dimensional matrix is of the form ~ . A, _~ E k s. For a pair of such matrices we have from eq. (2.6) the product (and square) rules

2 1

s_. ,\_/3_. A = g s • ~ + ~ * ¢~. A + ic~^~. A,

2 s2 ---~3 s * A__, (4.1)

A ) 2 = _ + s .

which generalize eq. (1.1b).

Now we develop the properties of a single matrix a • A in some detail. The determinant is easily worked out in terms of the star product:

det a - A = ~ - ~ s . s • c~. 2 (4.2)

If so is a ninth 'scalar', from the SU(3) covariance property A(so + _~. A)A -1 : so + s ' . A,

~' : o ( a ) s , (4.3)

and invariance of the determinant we see that we must necessarily have

det (s0 + ~ " A) = s 3 + cso~ z + ~ - ~ s . s • s , (4.4) with no term quadratic in So, and with some constant c. Let us now diagonalize s . A using a suitable SU(3) transformation. We shall refer to this as 'putting s . A into its rest frame', and will refine this notion in the sequel. Then

So + s - A = diag(s0 + s 3 + s s / v / 3 , s0 - s 3 q - S s / V / 3 , S o - 2 s s / v ~ ) ,

2 2

det(ao + s . A) = C~o 3 - C~o(S3 2 + s8 2) + - - ~ c~8(s3 - s~/3). (4.5) V~

376 Pramana - J. Phys., Vol. 49, No. 4, October 1997

The algebra and geometry of SU(3) matrices This fixes c = - 1 , so we have the general relation

det(ao + s . A) = S3o - s 0 a 2 + ~ - ~ s . s * a 2 (4.6)

valid in any 'frame'.

Turning to matrix inverses, and assuming a0 is not an eigenvalue of - a _ . _A, again SU(3) covariance dictates the general structure

(so + ~_" _A) -1 = (ClS~ + c2~ 2 + c3so~_" _A + c4_~ * s . A)/det(so + a-_A), (4.7) for some constants c t , . . . , c4. Using eq. (4.6) and transposing terms we have

( C l S 2 + C 2 S 2 + C3S 0 S'__~+ C 4 S * ~___')~)(S0+ S " /~) : det(so + ~'A). (4.8) Comparing powers of s0 gives cl = 1, Cz = - 1 / 3 , c3 = - 1 , ca = 1/V'3 and also the relation

~ • (a_ • a__) = a : a - i v / 3 ( a , ~_) A~ (4.9) which we shall understand in another way in a moment. So for matrix inverses we have the SU(3) covariant result

( S o + a . _ A ) - l : ( s 2 - ~ 1 2__a - S o a _ - _ _ A + ~ 3 ~ _ 3 s , s . A ) / d e t ( s o + a . A ) . . _ _ (4.10) From the determinant relation (4.6) we see that the minimal (cubic) equation for s . A is ( a .__A) 3 = s E a • A + ~--V~s - S * S. 2 (4.11)

If we substitute (4.1) here for (s-_A) 2 and compare coefficients we get the two relations

~ A ( S * S) ~--- 0, (4.12a)

s * (o~ • s ) = _sZs,_ (4.12b)

which explain the earlier result (4.9). Incidentally the first result above is obtainable from eq. (3.12b) by setting ~ = fl = 7- We also obtain the following useful property of octet vectors, which generalizes the result in three dimensions that a ^ b vanishes only if b is parallel to a:

S,__fl C ~-~8, S ^ ~ -= 0 ~:~ fl_ = CI s + C2 s * O G Cl,2 constants. (4.13) That there are no more terms here follows from the relationship of octet vectors to traceless hermitian matrices in three dimensions.

Let us now trace the consequences of the minimal equation (4.11) in more detail. For brevity, denote s • s by _~' for the moment. Then eqs (4.1), (4.11) read

(_~.A)2 -~--~S2 2 +___~_3S,. A , _ _ _

Pramana - J. Phys., Vol. 49, No. 4, October 1997 3 7 7

K S M a l l e s h a n d N M u k u n d a

(oe. _A) 3 = a 2 a • A + - ~ - ~ o e . a . 2 (4.14)

Now we substitute the first relation here into the second and keep simplifying till we have only linear terms in a . A and a ' - A:

• . a = 3 2 2

- - _ +

~ (a__2)2 2 a 2

• = - a ' . A. (4.15)

: v 3

This implies as consequences for any a~ upon substituting _a' = a * a:

(a * a) 2 = (a2)2, (4.16a)

( ~ • a ) • ( a • ~ ) = 2 a . a * a a - a 2 a , a. (4.16b) This last relation is to be contrasted with the result of taking the star product of (4.12b) with a, which is

a • ( a , (a_q_ * a_a)) = a 2 a * a. (4.17)

We are now in a position to deal in more detail with the eigenvalue spectrum of a single generator _a. _A, and in the process refine the idea of the 'rest frame' form of a • A. Let us hereafter assume _a is a unit vector, &2 = 1. Then the eigenvalues of & - _A are #1,/*2,/z3 obeying

/*1 -I-/*2 -4"- /*3 = 0,

/*2 +/*~ +/*2 = Tr(&. _A) 2 = 2. (4.18)

We can easily see that they can be ordered according to/.1 _>/*2 >_ #3 and the ranges can be fixed as follows:

1 / 3

/*1,3 = - - 2 / . 2 4 - V l - ~ / . 2 2

2 1 1 2

- ~ _ < #3 < - ~ < #2 <_ ~ < #1 _< ~ . (4.19) A convenient parametrization of all three eigenvalues is by an angle ~ in the range [7r/6, 7r/2] as follows:

# l = - ~ s i n ~ p , #2=-~sin(~p+2~r/3), #3 = sin(qo+47r/3). (4.20) The three angles occurring here are in strictly non-overlapping regions. We now define the 'rest frame' of d • _A to be that unique diagonal form in which we have

378 Pramana - J. Phys., Vol. 49, No. 4, October 1997

The algebra and geometry of

SU(3)

matrices

de. __A = diag ~ 3 (sin ~, sin(~ + 27r/3), sin(~ + 47r/3)),

&3 = cos(~o + 47r/3), de8 = - sin(~ + 47r/3),

del = de2 = de4 = de5 = de6 = de7 = 0 ,

7r/6

<_ ~y _< 7r/2. (4.21) In this 'frame' for &, we find that de * de has similar nonzero components and is by eq. (4.16a) also a unit vector, but is not in its own 'rest frame':

2 (sin(2~ - 7r/2), sin(2~p + 57r/6), sin(2qo + 7r/6)), de • de. _A = diag

(de * &)3 = 2&3&8 = - sin(2~ + 27r/3), (a * de)8 = &~ - de2 = cos(2qo + 2rr/3),

(de * &)r = 0 for r = 1,2, 4, 5, 6, 7. (4.22)

The angle ~ can be inferred from the value of the invariant { = &- & * de = - sin 3qo:

since - 1 _< { _< 1 and 7r/2 _< 3qo _< 37r/2, { fixes the value of ~o uniquely.

5. Axis-angle parameters and finite elements o f SU(3)

In this section we derive the SU(3) analogue of eq. (1.1c) for SU(2). It is easily seen, for instance by going to the diagonal form, that every A E SU(3) can be obtained by exponentiating a suitable traceless antihermitian matrix. We now compute in closed form the matrix

A(&, 0) = exp(i0de. __A) (5.1)

set up in analogy to eq. (1.1c) for SU(2). The unit octet vector & is the axis, and 0 is the angle, for the element A(&, 0). The range for 0 is discussed below. The sole SU(3) scalar we can form from & is the angle ~ given by

= &. & . & = - s i n 3 ~ ,

7r/6 < qo < 7r/2. (5.2)

Upon expanding the exponential in eq. (5.1) and using (4.1, 4.12b), we see that the only terms that arise are multiples of the unit matrix, of & • _A and & • &- _A_. We therefore write

a(&, O) =

-~3c(O, qa) + a(O, qo)& . A__ + b(O,

~p)& • &. __A, (5.3) and proceed to determine the three SU(3) scalar coefficients. For this we go to the rest frame (4.21) of &. __A - then & * de. __A is also diagonal (cf. eq. (4.22)) and so is A(de, 0).

Equation (5.3) reduces to

exp(~33'isin~) /

['a(O,~)) x/~

exp ( ~ • i sin(~ + 27r/3)) ,

zx(~o)/b(O,~o) = 5--

\c(O, qa)

exp(~3, i sin(~ + 47r/3) )

Pramana - J. Phys., Vol. 49, No. 4, October 1997 379

K S Mallesh and N Mukunda

[ sin

A(~o) = [ sin(qo + 27r/3)

\ sin(~o + 47r/3) The determinant of A(~o) is

det A(qo) = --~-cos &o,

sin(2qo- r r / 2 ) i ) sin(2qo + 5rr/6) .

sin(2~o + rr/6)

(5.4)

(5.5) so A(~o) is nonsingular for rr/6 < qo < rr/2. In this range we have

f-sin2qo sin(2qo+ rr/3) sin(2qo-rr/3) '~

A ( q o ) _ l 2 / -c°s~° sin(qo+rr/6) -sin(~-Tr/6)], (5.6) 3 cos 3qo \ ½ cos 3qo ½ cos 3~o ½ cos 3qo ]

and the solution for the coefficients a(O, ~o), b(O, ~o), c(O, ~o) is

{ (exp ~ qo) exp (20.isin(qo+27r/3))--~

a(O, qo) = - ~/3" i sin • sin 2qo +

x sin(2qo+ rr/3)+ exp ( - ~ . i sin(qo+ 4rr/3))sin(2qo-7r/3)}/

V~ cos 3qo,

{ (20"isin(q°+27r/3))

b(0,~o)= -exp(--~3.isin~o ).cos~o+exp ~

( 2 0 . isin(qo + 4rr/3))sin(qo - r r / 6 ) } / x sin(T + 7r/6) - exp ~

x/3 cos 3~o, c(O,~p) = exp ~ . / s i n q o + exp ~ . isin(~ + 27r/3)

+ exp ( ~ 3 • i sin(qo + 4rr/3)) } / 2 v / 3 . (5.7) With this the computation of A(&, 0) in closed form in the generic case is complete.

The two limiting cases ~o = rr/6 and ~o = rr/2 correspond respectively to & • & = -&

and &, & = &. In these cases we find after some algebra:

= 7r/6: a(&, O) = -~c(O, rr/6) + a(O, rr/6)&- A, qo

u

a(O, rr/6) = {e i°/v5 - e-2iO/vS} /v/3, c(O, 7r/6) = {2e i°/v5 + e-2ia/v"5}/2v/3;

2 c(O, 7r/2) + a(O, rr/2)& • A, qo : rr/2: A(&, 0) : - ~

380 Pramana - J. Phys., Vol. 49, No. 4, October 1997

The algebra and geometry of

SU(3)

matrices a( O, ~-/2)

⁼ - { e -io/V'~ -

e2iOl~} lvr3,

c(O, ,r/2) = {2e -iolv5 + e2ie/'/3}/2v~. (5.8)

Now we consider the question of the range of the angle variable 0. Here the situation is more intricate than in the SU(2) case. With Pauli matrices, &. _a always has eigenvalues -4-1 independent of&, so in eq. (1.1c) we have a fixed range 0 < 0 < 2rr. Equivalently, all U(1) subgroups of SU(2) are conjugate to one another. With SU(3), however, there are infinitely many inequivalent U(1) subgroups available. This means that while every A E SU(3) lies on some one-parameter subgroup and can be written as A = A(&, 0) for some & and 0, the range of 0 depends on &, more specifically on the SU(3) invariant angle T associated with &. The determining factor is the nature of the eigenvalues # 1 , / ~ ,

#3 of & • _A. The case #2 = 0, #1 = - # 3 = 1 leads to the range 0 < 0 < 27r, which occurs for qo = rr/3. For #2 # 0, we need to distinguish between the cases of rational and irrational ratios #1/#2. For rational #1/#2, when #3/#2 is also rational, the elements A(&, 0) lie on a cyclic U(1) subgroup in SU(3) and 0 can be taken to be in some finite interval from zero to a maximum determined by #2. For irrational #1/#2, when #3/#2 is also irrational, the character of the one-parameter subgroup is very different - it is the real line g , not a cyclic U(1), so 0 E ( - o o , oo). This then is an essentially new feature with axis angle parameters for SU(3) as compared to SU(2).

6. Hamiltonian dynamics o f three-level systems

Consider a three-level quantum system whose state is represented by a density matrix p.

The properties

p t = p > O ,

T r p = l (6.1)

allow us to expand p in terms of the A's, bringing in a scalar c and a unit octet vector h:

p = ½(1 +

ch.

_A),

c < - ~ - c o s e c @ + 7r/3), (6.2)

where the angle qa E [7r/6, 7r/2] is determined by eq. (5.2): h. h * h = - sin 3qv. The pure state case corresponds to qa = 7r/2 and c = v/3: then h * h = h and the eigenvalues of p are (1,0,0).

Let H be a general (time-independent) Hamiltonian which we express in terms of an octet vector h and a scalar ho:

H = ½ (ho + h- _A). (6.3)

The equation of motion for p (with c and h regarded as functions of time),

i-~f=dP [H,p],

(6.4)

is independent of ho and at first leads to

~h + ch = ch^h.

(6.5)

Pramana - J. Phys., Vol. 49, No. 4, October 1997 381

K S Mallesh and N Mukunda

However since n and -hAh are both orthogonal to h, we get b = 0 as expected, and the equation of motion for h becomes:

n = -hAh. L (6.6)

This is the three level version of the Bloch equation for the spin vector familiar from two- level systems. One can now use some of the identities derived in earlier sections to verify that h • h * h is a constant of motion:

d h . h , h = h ' h * h + 2 n ' h * h dt

= _hAft. ( h , fi) + 2ft. h • (_hAh)

= h. h^(h • h) + 2h. (hA(h * -h) + -hA(h * h))

= 2-h. (h * h) Ah

= 0. (6.7)

Here we used eqs (3.1 la, 4.12a) and eq. (3.12b) with a = [3 = h, 3' = h to simplify terms at various stages. Thus, as expected, both the scalar c in p and the SU(3) scalar angle characteristic of h are constant in time. This is consistent with the fact that h(t) evolves essentially according to the octet representation matrix D(A (t, h)) where in the notation of eq. (5.1) the time t is the angle and the vector h the axis of 'rotation'.

7. C o n c l u d i n g r e m a r k s

We have presented a set of practical tools for carrying out calculations with finite matrices of SU(3) as well as with its Lie algebra, exploiting both algebraic and geometric aspects of the situation. The space of octet vectors bears the same relation to SU(3) as does ordinary Euclidean three-dimensional space to SU(2). The constructions we have given for working with these vectors should prove useful in dealing with three level system dynamics. The existence and interpretation of the cubic invariant a • a * a and the general solution (4.13) to _aA/3 -- 0 are noteworthy. We have also brought out the fact that in contrast to SU(2), there are infinitely many distinct kinds of one-parameter subgroups in SU(3), and this shows up in the axis-angle description in this case.

The main new feature in the SU(3) situation, absent with SU(2), is the occurrence of the d-symbols. However, for all groups SU(n), n _> 4, nothing new apart from such a d- symbol is expected since one cannot in any case go beyond the commutators and anticommutators of the generators in the defining representation. It therefore is to be expected that the methods of this paper can be systematically extended to these higher dimensional groups as well.

A c k n o w l e d g e m e n t s

One of us (KSM) wishes to thank the Jawaharlal Nehru Centre for Advanced Scientific Research for the award of a Visiting Fellowship, and the Centre for Theoretical Studies, Indian Institute of Science for providing facilities.

382 Pramana - J. Phys., Vol. 49, No. 4, October 1997

The algebra and geometry of SU(3) matrices R e f e r e n c e s

[1] D Griffiths, Introduction to elementary particles (John Wiley and Sons, New York, 1983) [2] C Quigg, Gauge theories of the strong, weak and electromagnetic interactions (The

Benjamin-Cummings Pub. Co., London, 1983) [3] J N Elgin, Phys. Lett. A80, 140 (1980) [4] P K Aravind, J. Opt. Soc. Am. B3, 1025 (1986) [5] F T Hioe, Phys. Rev. A28, 879 (1983)

[6] M Gell-Mann and Y Neeman, The eighOCold way (W A Benjamin Inc., New York, 1964) [7] J J de Swart, Rev. Mod. Phys. 35, 916 (1963)

[8] G Khanna, S Mukhopadhyay, R Simon and N Mukunda, Ann. Phys. (NY) 253, 55 (1997)

Pramana - J. Phys., Vol. 49, No. 4, October 1997 383