Non-Abelian Berry's phase, accidental degeneracy, and angular momentum

Jan Segert8 )
Physics Department, Princeton University, Princeton, New Jersey 08544

(Received 23 December 1986; accepted for publication 13 May 1987)

The non-Abelian Berry's phase effect for a family of operators Ho + k·V is considered, where
H is rotationally invariant, V is a vector operator, and k varies over the unit vectors in R3. The
p:rameter space is the two-sphere. The time evolution in the a~iabatic li~it is gi~en by a
connnection on a fiber bundle over the two-sphere. All connectIOns consistent with the
rotational symmetry are classified, and the time evolution is explicitly calculated for a
nondegenerate Hamiltonian, as well as for a Hamiltonian with a double degeneracy. In the
nondegenerate case, the connection is uniquely determined by the symmetry. In the doubly
degenerate case, the connection is in some instances .not dete~ined by the sy~metry. T~e case
of approximate degeneracy is also discussed. A possible expenmental test of this effect usmg
optical pumping is described elsewhere.

I. INTRODUCTION bundle Fn' The Berry connection must be invariant under

this action. We shall topologically classify the bundles Fn in
The time evolution of a quantum system governed by a
the case of no degeneracy, and in the case ofa double degen-

Downloaded from [Link]

slowly changing Hamiltonian can be analyzed geometrical-
eracy. This degeneracy is an accidental, where we use the
ly. This was first noted in the work ofBe~ry, 1 ~nd [Link]
term to mean that the symmetry group of the Hamiltonian
by Simon. 2 The case of degenerate Hamdtomans, otherwise
does not act irreducibly on the degenerate subspace. 5 We
known as non-Abelian Berry's phase, was considered by
will detennine the action of the rotation group on the bun-
Wilczek and Zee. 3 The main results of these papers is that
dles Fn. We then classify all the invariant connections. The
the time evolution is given by the parallel transport in a cer-
classification depends only on the angular momentum quan-
. tain connection on a fiber bundle over the parameter space.
tum numbers of the state or pair of degenerate states under
We shall call this the Berry connection. In this paper, we
consideration.
shall analyze a certain class of such systems. The systems
In the nondegenerate case, we find that there is only one
described here can be experimentally realized in a number of
rotationally invariant connection, which must then be the
ways. In another paper, 4 we propose an optical pumping
Berry connection. In the degenerate case, the results are
experiment using Pb 208 atoms. The analysis of such experi-
more interesting. If the two degenerate states have consecu-
ments requires the results of the work presented below.
tive J·k quantum numbers, e.g., m 1 - m 2 = ± 1, then there
We shall consider a family of Hamiltonians parame-
is more than One invariant connection. In fact, there is an
trized by the unit vectors k in R3. To each unit vector k we
uncountable infinity of invariant connections. If the quan-
associateaHamiltonianHk = Ho + k·V, whereHoisa rota-
tum numbers of the two degenerate states do not satisfy this
tionally invariant operator, and V is a vector operator. This
condition, then the connection is unique, and we can imme-
generalizes the example of Berry of an atom in ~ magnetic
diately identify the Berry connection as the unique invariant
field. l For example, the Hamiltonian of an atom 10 constant
connection. If the two degenerate states satisfy the above
colinear electric and magnetic fields is of this form. 4 The
condition, we have to do some additional work to identify
results of this paper are, in fact, valid for a more general class the Berry connection.
of Hamiltonians than those of the form Hk = Ho + k·V.
The work of Berry and Simon can be divided into two
This class, which will be fully characterized, includes such
statements: (1) The evolution of a system under adiabatic
potentially interesting examples as the quadratic Zeeman
change of parameters is determined by some connection on a
Hamiltonian for an atom in a strong field.
certain bundle over parameter space. (2) The Schrodinger
The rotation group acts transitively on the parameter
equation uniquely determines this connection. Statement
space, since any unit vector can be rotated ~nto any o~he~ unit ( 1) follows almost immediately from the definition of a con-
vector. Thus all the operators in the famlly are umtanly e-
nection. 4 •6 When the connection is already uniquely deter-
quivalent and have the same spectrum. Consider first the
mined by the symmetry group, as is often the case, then state-
nondegenerate case. We choose an eigenvalue En' Following ment (2) contains no information. But ifthere is more than
Berry and Simon, we form a line bundle Fn over parameter
one connection compatible with the symmetry group, state-
space, where the fiber over a point k of S 2 is the space of
ment (2) has nontrivial content, and is subject to experimen-
scalar multiples of eigenvectors of Hk with eigenvalue En. If
tal confirmation. We comment here on some recent related
the spectrum has a double degeneracy at En' we must consid- papers. The work of Chiao and Wu 7 interprets the polariza-
er a two-dimensional complex vector bundle over S 2, which
tion rotation oflight in a bent optical fiber as a manifestation
we will also denote by Fn. The rotation group acts on the
of Abelian Berry's phase. The Berry connection for this sys-
tem is uniquely determined by the symmetry. This is a conse-
.) Address after September 1987: Department of Physics, California Insti- quence of the results of the present paper. The polarization
tute of Technology, Pasadena, California 91125. rotation can alternately be interpreted as a purely classical

2102 J. Math. Phys. 28 (9), September 1987 0022-2488/87/092102-13$02.50 © 1987 American Institute of PhysiCS 2102
effect,6.8 and the result is again uniquely determined by the
rotational invariance. 6 The results of the classical and quan- 1= (~ ~), Ux = (~ ~),
(2.2)
tum derivations are identical, and have been confirmed by
the experiments of Tomita and Chiao. 9 This experiment can-
not test statement (2), since the symmetry alone is enough to
uy =(0i ~ i), Uz = (~ ~ J.
determine the result. Moody, Shapere, and Wilczek JO have The obvious Lie algebra isomorphism is Lx ...... (i/2)ux '
proposed a spin-resonance experiment to look for manifesta- Ly ...... (i12)uy , L z ...... (i/2)uz ' We shall call SU(2) the rota-
tions of Abelian Berry's phase. The family of Hamiltonians tion group. This lets us treat half-integral spins as well as
for this system is ofthe form Hk = Ho + k·V, with each Hk integral spins. We denote by (iJx,iJy,iJz ) the standard basis
nondegenerate. Thus by the results of the present paper, the of a representation of the Lie algebra of SU (2), where the
rotational symmetry again uniquely determines the result, generators ( J x , J y , Jz) are self-adjoint. The corresponding
and this experiment also cannot test statement (2). In the unitary representations of SU (2) will be denoted by U. We
same paper, Moody, Shapere, and Wilczek discuss the ef- shall denote by R the vector representation of SU(2), the
fects of non-Abelian Berry's phase on the energy levels of a real three-dimensional representation generated by the Lie
diatomic molecule. This is formally similar to the systems algebra of SO ( 3 ), using the isomorphism (i/2) u x ...... Lx,
studied in the present paper. We consider a system governed (i/2)uy ...... L y , (i/2)uz ...... L z . This is, in fact, the adjoint
by a Hamiltonian of the above form, Ho + k·V. When this representation ofSO(3), and is unitarily equivalent, but not
Hamiltonian has accidental degeneracies of a certain type, identical, to the usual spin-1 representation of SU(2). We
the Berry connection cannot be determined just from the list for convenience some standard results,
rotational invariance. Then statement (2) is a nontrivial pre- [Jx,Jy] =iJz ' [Jy,Jz] =iJx , [Jz,Jx ] =iJy , (2.3)

Downloaded from [Link]

diction, which is experimentally testable. 4 We shall find the
J21j,m) =j(j+ 1)lj,m), Jzlj,m) =mlj,m), (2.4)
Berry connection explicitly using Simon's prescription. We
also study the case of approximate degeneracy. This is im- (j,ml Jz Ij,m - 1) = H(j + m) (j - + 1)] 1/2, (2.5)
m
portant for the analysis of experimental tests. (j,ml J y Ij,m - 1) = (i/2)[ (j + m) (j - m + 1)] 1/2.
In Sec. II we fix conventions and collect standard angu-
(2.6)
lar momentum results. In Sec. III we review the construc-
A vector operator V is a triplet of operators ( Vx , Vy , Vz ) sat-
tion of vector bundles using projection operators. In Sec. IV
isfying5 •11
we study vector bundles over S 2. In Sec. V we determine the
topological structure of eigenvector bundles over S 2. In Sec. [Jx,vy] =iVz , [Jy,Vz ] =iVx , [Jz'Vx ] = iVy .
VI, we determine the action of the rotation group on the (2.7)
eigenvector bundles, and present the results ofthe invariant We let i, y, zdenote the standard orthonormal basis of
connection classification. In Sec. VII we calculate parallel lR 3 • We shall use the standard conventions for spherical co-
transport in the Berry connection. Section VIII concerns the ordinates in lR 3 ,
case of approximate degeneracy. Section IX consists of con-
z=rcos(O), x=rsin(O)cos(cp), y=rsin(O)sin(cp).
cluding remarks. The Appendix contains the technical de-
tails of the invariant connection classification. (2.8)
On the unit sphere S2, the south pole corresponds to the
II. CONVENTIONS
vector - z, the north pole to the vector z. By the equator of
S 2 we shall mean the map eq: S I -+ S 2 by cp ...... (1T12,cp ). We
We now specify some of our conventions. The group of shall alternately represent a point on S 2 by k or (O,cp).
rotations in lR 3 is SO(3), the group of orthogonal real matri-
ces of dimension 3, with positive determinant. A rotation can III. MATHEMATICAL PREREQUISITES ON VECTOR
be specified by giving the axis of the rotation, a unit vector k, BUNDLES
and the angle of rotation in the positive sense a. We define
We discuss a characterization of complex vector bun-
the matrices
dles over a compact connected metric space M, for example a
compact finite-dimensional smooth manifold. This charac-
L,~G ~} L'~( ~ ~),
0 0
0 0 terization is appropriate for the bundles of eigenstates over
parameter space. Consider the product space M X K, where
o -1 0
K is a complex separable Hilbert space. The Hilbert space
(2.1 )
K may be finite or infinite dimensional.
L, ~( ~ 1 0
0 ~). Denote by 9 (K) the space of compact positive self-
adjoint projection operators on K. An operator P is in
9 (K) if and only if all the following conditions are satis-
which act on lR 3 • The SO ( 3) matrix corresponding to the fied: PP = P, P t = P, the range of P is finite dimensional,
rotation is exp(ak·L). Recall that the group SU(2) is a dou- and (xIPlx»Oforalllx)EK. We give 9(K) the opera-
ble cover of SO (3). The Lie algebra of SU (2) is isomorphic tor norm topology. In this topology, 9 (K) consists of dis-
to the Lie algebra of SO ( 3 ). We shall take as our basis of the joint components 9 n (K). Here 9 n (K) consists of all
Lie algebra of SU(2) the matrices (i/2)ux ,(i/2)uy , projections whose range has dimension n. This follows easily
(i/2)uz ), with the usual convention for Pauli matrices, from the spectral theorem for normal operators,12 and the

2103 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2103
fact that the group of unitary operators, with the operator us 1T. Let (fJ,¢) denote the polar coordinates on D 1T' where fJ
norm topology, is connected. is the radial coordinate, and ¢ the angular coordinate. Let
Consider a continuous map I: M - fJI (JY'). The range DA be the S2 minus the south pole. Let UA : DA -D1T by
of I must be contained in fJI n (JY') for some n. So for any (fJ,¢ )t--+( fJ,¢) , where in the first instance (fJ,¢) are the
two points s,vEM, the dimensions of the projections spherical coordinates on S2, and in the second instance the
I(S),J(V)EfJI (JY') coincide. Denote by r the range map r: polar coordinates on D 1T' Similarly, let DB be the sphere S 2
fJI (JY') -JY'. The range map assigns to a projection opera- minus the north pole. We now define the map UB: DB -D1T
tor P the subspace of JY' which is not annihilated by P, the set by (fJ,¢ )t--+( 1T - fJ,¢). The choice of coordinates is a little
of vectors xEJiY' satisfying Px = x. The composition r °I unfortunate, but the map is clearly nonsingular. The inter-
maps Minto subspaces of JY' of constant dimension. The sectionDA nDB is the sphere minus both poles. The image of
graph reM xJY' of the map r °lis a vectorbundleAfover this region under U A is the disc with the origin removed,
M. The projection map 1T: Af -M is the restriction of the D1T - O. The image under UB is also D1T - O. We see that
projection 1T: M xJY' -M. We shall now demonstrate the UBO(U A ) -I: (D 1T - 0) - (D 1T - 0), by (fJ,¢) _ (1T - fJ,¢).
local product structure. Each pointsEM has a neighborhood We examine vector bundles on S 2. First note that each of
Ns which is contractible to the point s. The restriction f Ns the coordinate patches defined above is contractible. Recall
- fJI (JY') is then homotopic to the constant map Cs : that any vector bundle over a contractible space is trivial. All
Ns -/(s). The local product structure follows. Let g be an- the topological information is contained in the transition
other map from M to fJI (JY'). This map specifies another function. The transition function is a map g: DA nDB -G,
vector bundleA g over M. The two vector bundles are topolo- where G is the structure group of the bundle, which is U(n)
gically isomorphic if and only if the maps l,g: M - fJI (JY') in our case. The toplogical structure of the bundle is deter-
mined by the homotopy class of the transition function g.

Downloaded from [Link]

are homotopic.
There is an analog for real vector bundles which may be SinceDA nDB is homotopic to the circleS I around theequa-
more familiar. We shall construct the tangent bundle of the tor, we need only consider homotopy classes of maps
two-sphere S2. Consider the usual embedding of S2 as the S 1-- U(n). This is simply the fundamental group 1T1(U(n»),
unit sphere in R3. A unit vector k defines a point on S 2. The which is known to be Z, the integers. It is easy to verify that
tangent space at the point k consists of all vectors orthogonal 1T 1(U(n») = Z, corresponding to the winding number. Con-

to k. This is nothing but the range of the two-dimensioinal sider the determinant map det: U (n ) -+ U ( 1 ). One can show
projectionPk which acts on a vector a by Pk a = a - (k·a)k. that two maps g, g': S I _ U (n) are in the same homotopy
In fact, any finite-dimensional complex vector bundle class if and only if the composition maps detog, detog':
over M can be constructed this way if the dimension of JY' is S I -+ U ( 1) are in the same homotopy class, i.e., have the
sufficiently large. In particular, if the dimension of JY' is same winding number. One can then conclude that
infinite, all finite-dimensional vector bundles can be con- 1T2( fJI n (JY'») = Z if the dimension of JY' is sufficiently large.
structed. This follows from a theorem of Serre and Swan. 13 The theory of characteristic classes l4 ,I6,I7 provides an-
In other words, fJI n (JY') is the universal classifying space other approach to the classification of vector bundles. The
for U (n), also known as the infinite Grassmanian. Any bun- integer invariant, determined above from the winding num-
dle over M is the pullback of the universal bundle over the ber, can also be expressed as - C I , where C I is the first
classifying space. 14 In fact, there exists a universal connec- Chern number of the bundle. This is valid for complex bun-
tion on the universal bundle. This follows from the work of dles over S2 of any dimension. The Chern-Weil th eory l7,18
Narasimhan and Ramanan. 15 Any Hermitian connection on gives an expression for the first Chern number as the integral
a complex vector bundle over M can be obtained as the pull- over S 2 of a differential form C I of degree 2. This differential
back of the universal connection by the appropriate map of form is explicitly constructed from the curvature of a con-
M into the classifying space. IfM is the parameter space for a nection on the bundle. Chern showed that although the form
family of Hamiltonians, then the spectral projections l2 of a c i depends on the choice of connection, the integral C) of c i
given eigenvalue give a map of M into the classifying space. over S 2 does not, and is always an integer. In the case of a line
The pullback of the universal bundle by this map defines the bundle, the two-form CI is equal to the curvature two-form
eigenstate bundle, which will be studied in detail in Sec. V. :!Il. For bundles of higher dimension, the curvature form:!ll
We can also use this map to pull back the universal connec- is matrix valued, and the Chern class C I is the trace of :!Il. The
tion. The resulting connection is simply the Berry connec- theory of Chern classes is not necessary for what follows, but
tion. We shall not discuss these results in any detail, but we it facilitates understanding of the rather abstract classifica-
mention that the classification of rotationally invariant con- tion of invariant connections.
nections, given in the Appendix, can alternately be proved
using these results. The classification then reduces to a ques- V. STRUCTURE OF THE EIGENSTATE BUNDLES
tion in representation theory, which is easily answered. We In this section we shall study the family of Hamiltonians
choose instead to present the direct proof in the Appendix, Hk = Ho + k·V, where k is a unit vector in R 3 , V is a vector
because it uses only elementary techniques. operator, and Ho is a rotationally invariant operator. The
parameter space is the space of unit vectors in R 3 , or the two-
IV. VECTOR BUNDLES ON THE TWO-SPHERE dimensional sphere S2. By definition, [J,Ho ] = 0, so
We now study the manifold S 2. Define an atlas on S 2 in [J,Hk ] = [J,k'V] = ikXV. Obviously Hk is invariant un-
the following way. Denote by D 1T the open disc in R 2of radi- der rotation about the k axis, [k'J,Hk ] = O. In fact, this is

2104 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2104
the only property that we require. All the results of this pa- IB(o.<p» = exp(i(1T - 0)( - sin(cp)Jx + cos(cp) J y )
per actually hold for a more general class of Hamiltonians
Xexp( -i1TJx )lj,m). (5.2)
H k • Consider a single Hamiltonian Hi.' satisfying
[ Jz ,Hi] = O. We obtain a family of Hamiltonians by letting One can easily check that these are indeed local sections of
the rotation group act on Hi' Let g be an element of the the bundle by verifying p(O,<p) IA(o,<p) ) = IA(o,<p) ), and, like-
rotation group, and U(g) the corresponding unitary opera- wise, for IB(o,<p»' A global section does not generally exist,
tor on the Hilbert space. The Hamiltonian Hz is mapped to as we will see. We now find the transition function relative to
the Hamiltonian Hk = U(g)Hi ut(g), where the unit vec- this trivialization. The transition function gBA satisfies
tor k is defined by k = R (g)i. One easily checks that gBA (O,cp)IA(o,<p» = IB(o,<p»' This can be expressed as
[Jok,Hk ] = O. We thus obtain a family Hk of Hamiltonians gBA (O,cp) = IB({;J,<p) ) (A (O,<p) I· The projection p(O,<P) acts as
parametrized by the sphere S2 of unit vectors in R3. It is the identity operator, so
easily verified that Hk = Ho + koV is a family of this type. gBA (O,cp) = p(O,<P) gBA (O,cp)
Since all the Hamiltonians in the family are unitarily related,
they have the same spectrum. The representation U of = IA(o.<p) ) (A (O.<p) IB(o,<p) ) (A (O,<p) I
SU(2) on the Hilbert space is generally reducible. We shall = (A (O,<p) IB(o,<p) )p(O,<P) . (5.3 )
be interested only in the discrete spectrum of the Hamilto-
nian. If the Hamiltonian also has a continuous spectrum, we We need to compute the matrix element (A (O,<p) IB(o,<p»'
shall restrict attention to the subspace of Hilibert space (A(o,<p) 1= (j,mlexp(iO( - sin (cp)Jx + cos(cp)Jy ) , (5.4)
spanned by the eigenvectors of the Hamiltonian. We will IB(o,<p» = exp(i(1T - 0)( - sin (cp)Jx + cos(cp)Jy )
thus assume from now on that the Hamiltonian has only a
X exp( - i1TJx ) Ij,m) ,

Downloaded from [Link]

discrete spectrum. (5.5)
Assume that the Hamiltonian H k is nondegenerate. We
now construct a one-dimensional complex vector bundle
(A({;J,<p) IB(o,<p» = (j,mlexp(hT( - sin(cp)Jx + cos(cp)Jy )

over S 2 corresponding to each eigenvalue En of the Hamilto- xexp( - i1TJx ) Ij,m) , (5.6)
nian. Let Ik,n) be a normalized eigenvector of H k , Hk Ik,n) (A (O,<p) IB(o,<p» = (j,mlexp( - 2icpJz ) Ij,m) .
= En Ik,n), (k,nlk,n) = 1. This is unique only up to a
We then find
phase. Construct the one-dimensional spectral projections
Pk •n = Ik,n)(k,nl. This is determined uniquely, the phase gBA (O,cp) = exp( - i2mcp)P(o,<p) . (5.7)
ambiguity cancels. Consider the map In: S 2 -+ .9' 1 (,Jr"') by
We need to compute the winding number ofthe map ofthe
k -+ Pk.n' Each In defines a one-dimensional complex vector
equator into U ( 1), gBA : S 1 -+ U ( 1). The winding number is
bundle Fn over S2. Since [Jok,Hk ] = 0, Ik,n) is an eigen-
clearly - 2m.
vector of Jok. We denote by m the eigenvalue,
We now consider a Hamiltonian for which the eigenval-
Joklk,n) = mlk,n). The span of the vectors Ik,n) for all k
ue En is doubly degenerate. Now the bundle Fn is two di-
defines a subspace J7'"n of J7"'. The rotation group acts irredu-
mensional, given by a map In :S 2 -+ .9' 2 (J7"'). It is clear that
cibly on J7"'n. The dimension of an irreducible representation
the bundle Fn is just the direct sum of two one-dimensional
of SU(n) determines the spin quantum number j, by
bundles with the appropriate quantum numbers. The sub-
dim J7"'n = 2j + 1. Each eigenvalue En of the Hamiltonian
space J7"'n spanned by the eigenvectors need not be an irredu-
carries the angular momentum quantum numbers (j,m),
cible subspace under the action of SU (2). If it is irreducible,
determined in the above manner. Usually, the operator Vok
we shall say the two vectors are in the same j multiplet. We
can be considered a small perturbation to the rotationally
again omit the n index.
invariant H 0' and is obvious how to define (j,m). In general,
Since [Jok,Hk ] = 0, we can choose normalized eigen-
however, since [ J 2 ,Hk ] #0, we need the above procedure to
vectors Ik,l), Ik,2) satisfying Hk Ik,i) = En Ik,i), Joklk,i)
determine j.
= milk,i), and (k,1Ik,2) = O. Define Pk,1 = Ik,l)(k,ll,
We will now determine the topological structure of such
Pk,2 = Ik,2) (k,21, p(O,<P) = Pk = Pk,1 + Pk,2' The rest is
a line bundle. We choose one of the eigenvalues En, and omit
analogous to the one-dimensional case. We can find the tran-
the n subscript in the following. Recall that every bundle is
sition function for the bundle exactly as above. The transi-
trivial when restricted to DA or DB' We will give a trivializa-
tion function gBA is in this case a map into U(2), given by
tion and find the transition function. We first fix a product
structure for the restriction to D A • Make a specific choice of gBA (O,cp) = exp( - i2m I CP)P(o.<p),1
phase for li,n), which will remain fixed from now on. We
will follow the usual conventions and call this vector Ij,m). + exp( - i2m 2CP )p(O,<P),2 . (5.8)
Here (O,cp ) are the spherical coordinates on S 2, and We could also note that the bundle is the direct sum of two
PUJ.<P) = Pk , where k = cos(O)i + sin(O)cos(cp)x line bundles, and obtain the U(2) matrix gBA by imbedding
+ sin (0) sin (cp ) Y. We define a local section of Fn over D A as U ( 1) xU (1) diagonally in U (2). It is easy to see that the
follows: result is the same. The topological classification requires de-
IA(o.<p) ) = exp( - iO ( - sin(cp )Jx + cos(cp )Jy »Ij,m) . termining the winding number of the map detog BA :
S 1-+ U( 1). This is easily seen to be equal to - 2(m l + m 2 ).
(5.1 ) This is also clear from the fact that the Chern class of a sum
Similarly, we define a section IB(o.<p» over DB' of bundles is the sum of Chern classes.

2105 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2105
VI. CLASSIFICATION OF ROTATIONALLY INVARIANT r = an is a smooth simple closed curve on S2. Then the
CONNECTIONS holonomy for the path r is simply the integral of yt over n.
We now discuss the action of the rotation group on the This follows from the definition of curvature as the holon-
bundles defined above, and the classification of the connec- omy of a small path. We recover Berry's result that the ho-
tions which are invariant under this action. In the one-di- lonomy for r is equal to m times the area of n. The curvature
mensional case, most of the results can be deduced from the is not quite sufficient to determine the connection. There are
rotational in variance of the curvature and the Chern- connections with the same curvature which are not rotation-
Gauss-Bonnet theorem. In the two-dimensional case, we ap- ally invariant. There is, however, a unique rotationally in-
peal to a generalization of a theorem of Wang. 19 The details variant connection. One could show this explicitly by noting
are discussed in the Appendix. that in this case two connections with the same curvature are
We discuss now the action of the rotation group on the related by a principal bundle automorphism, or gauge trans-
eigenvector bundles. We first consider the one-dimensional formation. A rotationally invariant connection remains in-
complex vector bundles corresponding to a nondegenerate variant under gauge transformation if and only if the gauge
eigenstate with quantum numbers (),m). The fiber over a transformation commutes with the action of the rotation
point kES 2 is a one-dimensional subspace of a finite-dimen- group. One may verify that there are no nontrivial gauge
sional complex vector space, the irreducible subspace of the transformations commuting with the rotation group action.
Hilbert space under the action of the rotation group. Recall We will instead rely on the classification results of the Ap-
that such an irreducible subspace has dimension (2) + 1). pendix to prove the uniqueness of the invariant connection.
The entire bundle space is the (2) + 1)-dimensional com- The two-dimensional case is more complicated. The
plex vector space, the fiber is a one-dimensional complex curvature yt is now a matrix-valued two-form. It is much
more complicated to determine all such two-forms compati-

Downloaded from [Link]

vector space, and the base space is S 2. The structure group is
U ( 1). The action of SU (2) on the bundle space is the action ble with rotational invariance. Even if we could determine
of the corresponding irreducible representation. The action the curvature form, the holonomy is not easy to find. In the
induced on the base space is the usual action of the rotation U ( 1 ) case we had a formula relating the holonomy for r to
group on S 2. The discussion for the two-dimensional vector the integral of yt over n. There is no analogous formula for
bundles corresponding to a doubly degenerate eigenstate is the U (2) case. The holonomy may be different for two paths
similar. Let (); < m i ), i = 1, 2 be the quantum numbers of r, r' which enclose, respectively, regions n, n' of the same
the two degenerate states. The bundle is just the direct sum area, but different shape. We must then appeal to the meth-
of two one-dimensional bundles. The fiber is now a two- ods of the Appendix.
dimensional complex vector space, the base space is again S 2, We now present the results of the invariant connection
and the entire bundle space is a (2)1 + 2)2 + 2)-dimensional classification. In the one-dimensional case, we find there is a
vector space, the sum of the bundle spaces of the two corre- unique invariant connection. In the two-dimensional case,
sponding one-dimensional bundles. The structure group is the results depend on the quantum numbers of the two states
U ( 1) xU ( 1). The action of the rotation group on the vector under consideration. Let ()i ,m i ) be the angular momentum
bundle is clear. It is straightforward to deduce the action of quantum numbers of the two degenerate states. Since each
the rotation group on the corresponding principal bundles. one-dimensional bundle has a unique invariant connection,
We remark that the action of the rotation group is transitive the direct sum of the bundles must have at least one invariant
on the U ( 1) principle bundle corresponding to the non de- connection,corresponding to the sum of the connections on
generate eigenstate, but is not transitive on the U ( 1) xU ( 1) the line bundles. We shall say that a connection splits if it is a
principal bundle corresponding to the doubly degenerate ei- sum of connections on each of the line bundles. The holon-
genstate. omy ofa connection which splits is U( 1) XU( 1) CU(2). It
We consider the one-dimensional case. Choose an eigen- is clear that there is only one invariant connection which
valueEn of the Hamiltonian, and denote by (j,m) the corre- splits, since the invariant connections on the line bundles are
sponding angular momentum quantum numbers. Construct unique. The results of the classification are the following. If
the bundle Fn as above. Consider a connection ~ on Fn, and m I - m 2 =1= ± 1, then there is a unique invariant connection.
let yt denote its curvature. The curvature is a two-form tak- This connection must then split. If m I - m 2 = ± 1, then
ing values in the Lie algebra of U ( 1), which we identify with there is a two-parameter family of invariant connections,
R. The curvature is then an ordinary two-form on S2. The indexed by (a,/1)ER2. The connection corresponding to
Chern-Gauss-Bonnet theorem states that the integral of a = {3 = 0 splits. No other connection splits. There is now a
21Tyt over S 2 is equal to C 1, the first Chern number of the nontrivial subgroup of the gauge transformations which
bundle Fn . We have determined above that C 1 = 2m. A rota- commutes with the action of the rotation group. We find that
tionally invariant connection has a rotationally invariant two invariant connections (a,{3) and (a',{3') are gauge
curvature form. The only rotationally invariant two-forms equivalent if and only if a 2 + {3 2 = a,2 + {3 '2. The connec-
on S2 are constant multiples of the area two-form 1], tion (a,{3) is thus equivalent to a connection of the form
1] = sin (e) de A d¢ in spherical coordinates. The integral of (~a2 + {3 2,0). Up to gauge equivalence, the set of invariant
1] over S 2 is simply the area of S 2, which is 41T. Thus we find connections can then be parametrized by R + , the non-nega-
yt = m1]. tive real numbers.
The curvature is, in this case, sufficient to determine This classification has important consequences. In the
holonomy. Consider a region n c S 2, such that the boundary one-dimensional case, we do not need to do any work to find

2106 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2106
the Berry connection. The Berry connection is just the (7.1 )
unique invariant connection. In the two-dimensional case, Now
we can immediately determine the Berry connection if
m I - m 2 # ± 1. The Berry connection is again the unique Un tUn_I = U(h n -1)U(h n _ l ) = U(h n _ 1 -Ian -lh n _ l ) .
invariant connection. When m 1 - m 2 = ± 1, we have to ac- (7.2)
tually do some work to find the Berry connection. This is
Thus Un tUn _ I is always close to the identity operator. We
unfortunate, but not tragic. We shall do this in the following
write U r as
section.
U r = UNANA N _ I .. 'A n " 'A 2 AI' (7.3)
VII. THE BERRY CONNECTION where An = Pk" Un tUn _ I Pk". Each An is a partial isometry
In this section we determine parallel transport in the from the range of Pk" onto itself. We can write Un tUn _ I
Berry connection. We use Simon's prescription, 2generalized = exp(iG n oJ) for some vector G n . Since Un tUn _ I is close
by Wilczek and Zee3 to the degenerate case. We do the two- to the identity operator, we can take G n to have small norm,
dimensional case for arbitrary (ij ,mj ). Recall that the split and approximate Un tUn _ I = 1 + iG noJ + o( IG n 12). Since
connection is unique for any bundle. We shall find that if we are interested in the limit N ..... 00, we need only consider
il #i2' then the Berry connection splits for any m l , m 2. For the first-order term.
m l - m 2 # ± 1, this follows from the classification, but if It is convenient to assume that k; = ko = z. We wish to
m I - m 2 = ± 1, this is a nontrivial result. Using this result evaluate expressions of the form
and the classification, we see that the only possible case
An = P z (1 + iGn°J)Pz . (7.4)
where the Berry connection might not split is il = i2'

Downloaded from [Link]

m l - m 2 = ± 1. We will, in fact, find that if the two eigen- By definition, we have
vectors are in the same i multiplet, and m l - m 2 = ± 1, P z = lil,ml)(il,mll + 1i2,m 2) (i2,m 21, (7.5)
then the Berry connection never splits. Note that il = i2 is a
P z JzPz = mllil,ml)(il,mll + m 2Ii2,m 2)(i2,m 21, (7.6)
weaker statement than saying that the states are in the same
i multiplet. P z JxPz = Ijl,ml)(il,mll J x li2,m2)(i2,m21
Choose an eigenvalue En' and define Pk as before. Sup- + li2,m 2) (i2,m 21J x lil,ml)(il,mll . (7.7)
pose we have an eigenstate of Hk which we wish to transport
to an eigenstate of H k " where k' = k + Ea, and a is in the Recalll that <il,mll J x li2,m 2) vanishes if lil,m l ) and
tangent space at k, a·k = 0, and E is small. We thus want a li2,m 2 )are not in the same i multiplet. If lil,m l ) and
partial isometry Uk',k from the range of P k to the range of 1i2,m 2 )are in the same i multiplet, the matrix element van-
P k" Uk',kPk = P k , P k' Uk',k = Uk'.k' When k = k', this ishes if m I - m 2 # ± 1. The same statement holds for ma-
should simply be the identity on the range of P k' Uk,k = P k • trix elements of Jy • If these matrix elements vanish, the Ber-
It is easy to check that ifk' = k + Ea, Uk'.k = Pk,Pk has the ry connection splits, and is determined by the classification.
desired properties to first order in E, Uk',k t Uk',k We can now see that the only instance in which the Berry
= Pk + 0 (E2) , According to Simon, 2and Wilczek and Zee, 3 connection might not split is when lil,m l ) and li2,m 2) are
this is the correct prescription for the Berry connection. To in the same imultiplet, and m l - m 2 = ± 1. We shall from
parallel transport along an arbitrary path ron S 2, starting at now on use the shorthand il = j2 to mean that the states are
k j and ending at kf , choose N + 1 points (kj in the same i multiplet, and define i = i I = i2' We need only
= ko,kl, ... ,kn , ••• ,kN _ l,k N = kf ) equally spaced along r, in consider the case m I - m 2 = 1, and define
the end taking the limit N ----+ 00. The partial isometry m = !(m l + m 2). We then have m l = m +!, m 2 = m -!.
U r =PkNPkN_t "'Pkn "'Pk,Pk" maps the range ofkj to the We will use the shorthand Im+) = li,m + ~),
range ofkf . This determines parallel transport along r in the 1m _) = i,m - p. In the following equations, all matrices
1

Berry connection. act on the two-dimensional complex space spanned by 1m +)

Since the action of the rotation group SU (2) on S 2 is and Im-):
transitive, there exists for any two unit vectors k, k' an P z = Im+)(m+1 + Im_)(m_1 = (~ ~) = 1, (7.8)
aESU (2) such that R (a) : k ----+ k'. Let Gk denote the isotropy
group of the vector k, the set of hESU (2) satisfying R (h) : P z Jz P z = m(lm+)(m+1 + Im_)(m_l)
k ----+ k. The elment a above is not unique, since h 'ahESU (2)
also maps k ----+ k' if hEGk , h 'EGk ·• Consider the path r, and + !(lm+)(m+I-lm_)(m_l)
take again N + 1 equally spaced along the path. We can
choose (a l ,a2, ... ,a n,... ,a n ), where a n ESU(2), and R(a n ):
= m(ol 0) +21(1° - 0)1 =ml+-u
1
1
2 z,
k n _ I ..... k n • We can choose these so all the an are close to the (7.9)
identity e of SU(2). It is convenient to define h n P z JxPz = H (j + m + !)(i - m + P] 1/2
= anan _ I .. 'a 2a l , so that R (h n ): ko ..... k n • Recall that U de-
notes the representation of SU(2) on the Hilbert space of X(lm+)(m_1 + Im_)(m+l)
states &". Define Un = U(h n ), then P kn = UnPk" Un t.
It is straightforward to write a formula for Ur ' = ! [(j + m + !) (j - m + p ]1/2 (~ ~)
U r = UNPk" UNtUN_IPk" UN_I tUN _ 2 = [(j + m + P (j - m + P] 1/2 !ux , (7.10)

2107 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2107
Pz Jy Pz = H(j + m +!) (j - m +!) ]1/2 We need to exponentiate these matrices. For any vector B,
we have
x ( - i/m+) (m_/ + i/m_) (m+ /)
exp(iBou) = cos ( /B/)1 + sine /B/) (iBoul/B/) . (7.24 )
=! [(j + m +!) (j - m +!) ]1/2(~ - ~) Now P z GoJ P z = Bol + Bou, where
(7.11) Bo = !tmGz , (7.25 )

We now study a class of paths for which we can compute Bx =!t [(j + m +!) (j - m + p ]1/2Gx , (7.26)
parallel transport explicitly. These are the paths which are By =!t [(j + m + P (j - m + p ]1/2Gy , (7.27)
the orbit of a point on S 2under rotation about an axis. Let G Bz = !tGz . (7.28)
be a unit vector. Consider the rotation about the G axis by
angle t. Let a denote the corresponding element of SU (2), We find
a = exp( - itG-u). For simplicity, we assume that the ini- C = exp(iBo)Pz. _ (cos ( /B/)1
tial point of the path is k; = z. The final point of the path is
+ sine /B/) (IBoul/B/) )Pz. + (7.29)
given by kf = R(a)z. The action on vectors in the Hilbert
space is by exp ( - it GoJ). We recall the rule (7.3) for paral- Using the identities
lel transport. Consider t N = tiN. Parallel transport is given p z.- uzpz.+ = 0, p z.- uxpz.+ = !(ux - iUy ) ,
by the operator U y , (7.30)
pz. _ uyPZ, + = (i/2) (ux - iuy ) ,
Uy=exp(-itGoJ)T, (7.12) we find

Downloaded from [Link]

T= lim (A(tN)t. (7.13) C = exp(iBo) [sine /B/ )I/B/] (Bxux - iByuy ) . (7.31)
N-oo
It is now easy to compute the transition probability W. Re-
We write A for An' since all are the same, and let t N denote
call Wis equal to the operator norm ofCtC,
the dependence on N,
A(tN) =Pz +itNPz(GoJ)Pz +O(tN)2. (7.14)
W = [(B; + B; )1/B/2] sin2( /B/) . (7.32)
We express G in spherical coordinates,
We recall an elementary formula,
Gx = sin(O)cos(tP), Gy = sin(O)sin(tP), Gz = cos(O).
lim (1 + xln)n = exp(x) . (7.15 ) Then
Bx =!t sin(O)cos(tP) [(j + m +!) (j - m + p ]1/2,
In our case this gives
(7.33)
T= Pz exp(itPz (GoJ)Pz)Pz , (7.16)
By =!t sin(O)sin(tP) [(j + m +!) (j - m +!) ]112,
U y = exp( - it GoJ)Pz exp (itP z GoJ [Link] . (7.17) (7.34)
We are interested in the probability of a transition
B z = ~tcos(O), (7.35)
between eigenstates of Jok. The projection pz. + projects 2 2
onto eigenstates ofk;oJ with eigenvalue m + !. The operator /B/ =!t [(j + m + P (j - m + !)sin (O) + cos (O) ]1/2,
exp( - it GoJ)Pz. _ exp(it G-J) projects onto eigenstates of (7.36)
k["J with eigenvalue m -~. We define B; +B;
C'=exp(-itGoJ)Pz._ exp(itGoJ)UyPz.+' (7.18) /B/2
We consider C'/m+). First note that p z.+ /m+) = /m+), _ (j+m+!)(j-m+psin 2(O)
then Uy gives the result of parallel transporting to - (j + m + p(j - m + psin 2(O) + cos 2(O) .
kf = R(a)z, and the remaining part projects onto the eigen- (7.37)
state ofk["J with eigenvalue m - !. The transition probabil-
ity between eigenstates is thus given by The transition probability W is given by
W= (m+/C'tC'/m+). Equivalently, W is the operator W=A sin2qOt) =!A (1 - cos(Ot)), (7.38)
norm ofC'tC'.1t is convenient to define
where A and 0 depend on 0,
(7.19 ) 2
A= (j+m+!)(j-m+psin (O) ,(7.39)
which has the property ctC= CitC'. Note also 2
(j + m +!) (j - m + psin (O) + cos 2(O)
P z CPz = C. This property lets us restrict attention to the
two-dimensional range of Pz' the subspace PzPt"CPt". This 0= [(j + m +!) (j - m + !)sin2(O) + cos 2(O) ]1/2.
has a basis /m+), /m_). (7.40)
We represent operators as matrices in the obvious way,
We see that the transition probability is an oscillating
Pz=l, p z.+ =!(l+uz )' p z.- =!(1-uz )' (7.20)
function of t, which is the angle of rotation about the axis G.
PzJzPz=!(ml+uz ), (7.21) We check some basic properties, 0 should be real. Note that
Pz JxPz = H(j + m + p(j - m + !)]I/2ux , (7.22) - (j - p<m<(j - !), so
P z JyPz = H(j+ m + p(j- m + !)]1/2Uy . (7.23) 0 2 = ( j + p2 _ m 2 _ l)sin2(O) + 1;~0. (7.41)

2108 J. Math. Phys .• Vol. 28. No.9, September 1987 Jan Segert 2108
Next, we see that as 0 --+ 0, A --+ 0, so W --+ O. W is invariant smallest splitting between states, excluding the approxi-
under 0 --+ ( 1T - 0), and also under t --+ - t. mately degenerate states,

8 = mine minE k-' minE n+) . (8.6)

k<m_ n>m+
VIII. APPROXIMATE DEGENERACY
We now consider the case of approximate degeneracy. Let € = IEm+ - Em_I. We suppose that UJ<t,8 for all 7. The
This is important in the analysis of experimental data, and is adiabatic theorem then determines up to a phase the time
further discussed elsewhere. 4 Consider a nondegenerate ei- evolution of all states except 1m + ) and 1m _ ). All eigenstates
genvector bfthe Hamiltonian H k • We denote the time vari- ofkj·J, except possibly Im+) and Im_), evolve into eigen-
able by 7. The Hamiltonian is time dependent, H ( 7) the time states ofkf·J. We can then deduce from the unitarity of the
dependence is specified by the motion of unit vector k ( 7) on time evolution that the states Im+) and Im_) can evolve
the parameter space S2. We shall denote by kj the initial only into linear combinations of states with kf·J eigenvalues
point of the path, and by kf the final point. We denote by m+ and m_. If €<t,UJ, then the states are effectively exactly
U( 7) the unitary time evolution operator. This satisfies the degenerate, and time evolution is governed by Berry's con-
Schrodinger equation nection. If €~UJ, then the states are effectively nondegener-
ate, and the adiabatic theorem tells us that there is no mixing
(8.1 ) between states of different J·k eigenvalues. We are, however,
interested in the intermediate case.
Denote the time derivative ofk by k'. Here k' (7) is a vector We shall again consider paths generated by rotation
tangent to S2 at the point k(7). Let UJ be the length ofk', about a unit vector G. Let 7 be the time variable, and UJ the
UJ( 7) = Ik'( 7) I. rate of rotation. The meaning of UJ is slightly different than in

Downloaded from [Link]

We wish to apply the adiabatic theorem. 2o We first note the previous section, but this should cause no difficulties.
that since all the j multiplets are mutually orthogonal, the The angle of rotation is t = UJ7 as before. We can formally
adiabatic parameter change we are considering cannot cause give an exact solution for the time evolution operator U( 7).
mixing between different j multiplets. We need only consid- The Schrodinger equation states
er one j multiplet at a time. We shall from now on assume
dU
that all states are in the same j multiplet. This reduces the i =HU. (8.7)
problem to finite dimensions. The adiabatic theorem in this d7
setting is quite simple. Suppose that k(O) = kj = Z. Consid- When H is time independent, this has the familiar solution
er a normalized eigenstate 1m) of Hz, with Hz 1m) U(7) = exp( - i7H). When His time dependent, the situa-
= Em 1m). Consider the quantity tion is more complicated. Suppose the time dependence of H
is given by H(7) = exp( - i7A)H(0)exp(i7A). Then it is
8m = min
-j<m'<j
IEm - Em' I, (8.2) easy to check that the following is the solution for the time
evolution operator U:
where 8 m is the splitting between 1m) and the state in the
same j multiplet closest to it in energy. The adiabatic U( 7) = exp( - i7A )exp( - i7(H - A») . (8.8)
theorem in this case states that We are specifically interested in the case A = UJG·J. Since we
kf·J U( 7) 1m) = maU( 7) 1m) , (8.3 ) need only consider the finite-dimensional restriction to a giv-
en j multiplet, we can, in principle, find an exact expression
where a is a complex number with unit modulus, if the fol- for U( 7). In practice, this is a little difficult. The simplest
lowing condition is satisfied: interesting case is j = 1, where we can have a degenerate
pair without the whole multiplet being degenerate. We must
sup UJ(X) <t,8 m . (8.4 ) then exponentiate three-dimensional matrices, which is
O<.x<T
In fact, we can simplify even further. Let k' be close to k. most easily done by first diagonalizing. The secular equation
Then an eigenstate ofk·J with eigenvalue m is orthogonal to is cubic, and explicit formulas exist for the roots. This gets
all eigenstates ofk'·J except, possibly, those with eigenvalues very messy very quickly. We can greatly simplify by using
m, m ± 1. We can then define the adiabatic theorem.
We study the time evolution operator in the case of ap-
8m = min(E'; ,E,;;) , (8.5)
proximate degeneracy. The adiabatic theorem tells us that
whereE,; =IEm -Em_II,E,;; =IEm -Em+ll,andthe U( 7) maps a nondegenerate eigenstate ofkj·J with eigenval-
adiabatic theorem as stated in (8.3) still holds. ue m' into an eigenstate of kf·J with the same eigenvalue.
We are interested in approximate degeneracy between Now the operator exp( - i7UJG'J) also maps a nondegener-
two states 1m!), 1m 2 ) in the same j multiplet. By approxi- ate eigenstate ofkj·J with eigenvalue m' into an eigenstate of
mate degeneracy, we mean that the energy difference kf·J with the same eigenvalue. So we see that
between the states is the same order of magnitude as UJ. From exp( - i7(H - UJG·J») must map a nondegenerate eigen-
the adiabatic theorem above, we see that no mixing between state of kj·J into itself, up to a phase. This tells us that the
eigenstates of k·J is possible unless m! - m 2 = ± 1. This operator exp( - i7(H - UJGoJ») is diagonal in the basis of
was already apparent for the case of exact degeneracy from eigenstates ofkj.J, except possibly on the pair of degenerate
the classification of invariant connections. We define as be- states. In other words, an eigenstate ofkj·J with eigenvalue
forem+ = m! = m +!, m_ = m 2 = m -!. Define8as the not equal to m + or m _ is mapped onto itself times a phase.

2109 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2109
Recall that Pi. is the two-dimensional projection onto the limit 8 ..... 0, the above result can be obtained from time-de-
approximately degenerate subspace, the subspace spanned pendent perturbation theory. II
by Im+) and Im-). It follows that
[Pz ' exp( - i'T(H - wG·J»)] = o. (8.9) IX. CONCLUSIONS

Restricted to the degenerate subspace, the time evolution We have studied the family of Hamiltonians of the form
operator then takes the form Ho + k·V. If a Hamiltonian in this family is nondegenerate,
we have seen that the adiabatic behavior is determined
U( 7) = exp( - i7WG·J)Pi. uniquely by the rotational symmetry. Experiments on such
X exp( - i7Pz (H - wG-J)Pi.)Pz . (8.10) systems can only yield limited information. If the Hamilto-
nian has a double degeneracy, or approximate degeneracy,
We now proceed exactly as in the previous section. It is con-
between two states m I and m 2 in the same j multiplet, then
venient to use t = 7W. We must now consider the operator
the rotational symmetry suffices to determine the adiabatic
(8.11 ) behavior if and only if m l - m 2 # ± 1. If m l - m 2 = ± 1,
Here we define /1 = (E+ - E_ )/w, E= (E+ + E_ )w, we need to invoke the Schrodinger equation, in the guise of
then the coefficients are Simon's prescription for the Berry connection. Experiments
A
(8.12 ) on such systems can then yield more information. An exam-
Bo = ~t(mGz - E) ,
ple of such an experiment is discussed in the paper.4
Bx = ~tGx [( j + m + !) (j - m + ~)] 1/2, (8.13 )
By = ~tGy [(j + m + !) (j - m + !)] 1/2, (8.14 ) APPENDIX: INVARIANT CONNECTIONS
Bz = !t(Gz -/1) . (8.15 ) We discuss the classification of rotationally invariant

Downloaded from [Link]

We note that when /1 = 0, all the coefficients except Bo re- connections on the eigenstate bundles over S 2. The results
duce to the values in the previous section. The Bo just gives are summarized in the main text. Here we present the techni-
the overall phase, which was neglected in the previous sec- cal details of the classification.
tion. Alternately, we could redefine our energy scale so Consider a principal fiber bundle P, with structure
E = O. The value of Bo does not affect the transition proba- group H and base space B. Recall that there is a free right
bility W. Expressing G in polar coordinates as before, we action of H on P, and B is the quotient by this action. Sup-
find pose, furthermore, that there is a Lie group G of bundle auto-
IBI = !t [(j + m + P (j - m + !) morphisms of P. This is a left action of G on P, which com-
2 mutes with the action of the structure group. The G action
xsin (8) + (cos(8) _/1)2] 1/2, (8.16 ) on P thus induces a G action on B. We will be interested in
B~ +B; classifying connections on P which are invariant under the
IBI2 action of G. We shall always assume that the induced G ac-
tion on B is transitive. The following can be generalized to
nontransitive G action on B, but the discussion becomes
(j + m + !) (j - m + psin2(8) + (cos(8) - /1f . more cumbersome. We shall assume that P, G, H, and Bare
( 8.17) all compact.
We then find a similar expression for the transition probabil- We shall formulate a generalization of a theorem of
ity W, Wang. 19 Wang's theorem gives a classification of invariant
connections for a special case. Consider a point yEiJ, and
W = A sin2(~Ot) = !A (1 - cos(Ot)) , (8.18) denote by Gy the isotropy group of y, Gy is the subgroup of G
where A and 0 depend now on /1 as well as 8, which leaves y fixed. Wang gives a classification in the case
2 when Gy acts transitively on the fiber Py over y. Recall that
A= (j+m+!)(j-m+!)sin (8) ,
G is assumed to act transitively on the base B. We shall ob-
(j + m +!) (j - m + psin2(8) + (cos(8) - /1)2
tain a result for the case when Gy does not necessarily act
( 8.19) transitively on the fiber over y.
2
0= [(j + m + P (j - m + !)sin (8) Let hEll, and denote by R h : p ..... p the right action of the
+ (cos( 8) - /1)2] 1/2. (8.20) structure group. Note that Rh -, gives a left action of H on P.
This action is not by bundle automorphisms, since it clearly
The possible transition probability is maximum at does not commute with the action of the structure group,
/1 = cos(8). The maximum has a value of unity, so a transi- unless the structure group is Abelian. This left action of H
tion probability of unity is possible. 0 has its minimum at does, however, commute with the left action of G on P. We
this value of /1. Now as /1 moves away from this value, A thus have a left action of G X H on P. It is easy to check that if
decreases, and 0 increases. At /1 = 0, the case of exact de- the G action on B is transitive, then the G XH action on P is
generacy, we recover our previous result. As a function of /1, also transitive. The action of G XH is not by bundle auto-
A is Lorentzian centered around /1 = cos(8), with width morphisms, so we cannot use Wang's theorem.
(j + m + !) (j - m + !)sin(8). Thus the effect is harder to We recall the construction of a connection on a princi-
observe at small 8, since /1 must be close to the maximum. pal fiber bundle. 17 Consider a point yEP, and the tangent
We now see why at /1 = 0, A decreases as 8 decreases. In the space Ty = TyP. Consider the orbit of y under the right ac-

2110 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2110
tion of the structure group H. Recall that the vertical space TJN s' denoted by i': Ty -- Ty . This corresponds to the deriv-
Vy C Ty is defined as the subspace tangent to the orbit of y ative of a map iEly .
under the action of the structure group. Here Vy has the We now consider an example. Let S = G X H. G acts on
same dimension as H. A connection @ is a one-form on P Sby left translation on the G factor, g': (g,h) -- (g'g,h). Here
taking values in the vertical subspace of the tangent space, H acts on S by right translation by the inverse on the H
which is invariant under the structure group action. At a factor, h ': (g,h) -- (g,hh '-I). Let FAG: A --G and FAH : A
point yEP, Wy is a linear map @y: Ty -- Ty , satisfying two --H be homomorphisms. A acts on S by a: (g,h)
conditions: (1) the range of Wy is Vy' and (2) the restriction --(g(FAGa-I),(FAHa)h ).
of@y to Vy istheidentitymapofVy,wylv = Id v . Thestruc- Consider the point s = (e,e)EG XH = S. We wish to
ture group acts by R h : P--P, where hEll. Suppose find the isotropy groupIs. AnelementofG XA XHactsons
Rh (y) = x. Then by invariance under the action of the by
structure group, we mean wxR;' = R ;'@Y' Here R ;, denotes (g,a,h): (e,e)~(g(FAGa-I), (FAHa)h -I). (AI)
the derivative of the map R h' One may construct a connec-
tion by choosing smoothly for each yEP an Wy satisfying ( I ) This element (g,a,h)Eis if and only if g = FAGa and
and (2) above. This is a priori not invariant under H. One h = FAHa. The isotropy group Iy of the corresponding point
obtains an H invariant W by averaging over the action of H, yEP is the set of (g,h) such that there exists an aEA satisfying
using the Haar measure on H. the above two conditions.
If @ is further invariant under the G action on P, we say Consider the tangent space Ts = TsS, This is equal to
that @ is a G-invariant connection. G-invariant connections the direct sum of the tangent spaces Te G Ell TeH. We will
exist. We use a similar argument. Construct at each yEP an from now on denote the tangent space at the identity of a Lie
@y satisfying (1) and (2). We now average over the Haar group Gby 2: G = Te G. Thus every ZETsS can be written as

Downloaded from [Link]

measure on G XH. Thus at least one G-invariant connection Z = (X,!), for some XEY G and YE2: H' Consider
exists. There may be more, however. Consider a point yEP, (g,a,h)Els' We wish to find the derivative of the map ats. We
consider exp~Z) = (exp(tX), exp(ty»)EG XH, and
and denote by Iy the isotropy group for the left G XH action.
Herely is the subgroup ofG X Hwhich leaves y fixed. IfW is (FAGa,a,FAHa )Els. Recall that all elements off can be writ-
a G-invariant connection, then clearly I y must leave @ y in- 'x
ten in this way. The element offs gives a map G H -- G X H
variant. Recall that the G XH action is transitive. Suppose as follows:
we have at one point yEP an @Y' satisfying (I) and (2), (FAGa,a,FAHa): (exp(tX), exp(tY»)
which is invariant under I y • Then @ is defined at all points of ~(FAGa)exp(tX)(FAGa-I), (FAHa)
P by the action of G XH. We then have the following propo-
sition. X exp(tY) (FAHa- I ») . (A2)
Proposition: Fix a point yEP. The G-invariant connec- The derivative of this maps is clearly (X,Y)~(Ad(FAGa)X,
tions on P are in one-to-one correspondence with linear Ad(FAHa) Y). Here Ad denotes the adjoint representation
maps Wy: Ty -- Vy which satisfy (1) and (2), and which are of a Lie group on its Lie algebra.
invariant under the action of I y • The null space Ns is the tangent along the A orbit of seS.
By invariance under the Iy action we mean the follow- Let WE2: A' Then exp(tW)EA. Under the A action on
ing. Let iEly, then i: y--y, the derivative map i': Ty--Ty . S = G XH, this gives
Recall that Wy: Ty -- T y . We say that @y is invariant under exp(tA): (e,e)~(FAG exp( - tA), FAH exp(tA»).
the action of Iy if i'Wy = @yi' for all iEly . This proposition (A3)
contains Wang's theorem as a special case.
~OWFAG': Y A --2: G,FAH ': 2:A --2: H' ThenullspaceNs
We shall be interested in principal H-bundles construct-
ed as follows. Let A be a compact Lie group. Consider a IS spanned by vectors of the form (- FAG' W,FAH ' W),
manifold S with a transitive left G XA XH action. Let seS, where WEY A' In other words, Ns is the range of the map
and denote by Is the isotropy group ofs under the G XA X H ( - FAG' Ell FAH '): 2: A - 2: G Ell 2: H'
action. Is is a subgroup ofG XA XH. LetPbe the quotient of We now consider the nondegenerate eigenstate bundle
S under the A action, P = S / A, and yEP the point corre- over S 2. This is a complex line bundle, so the structure group
sponding to seS. Here P admits a left G XH action, and can is U(1). The SU(2) action is transitive on the principal bun-
thus be considered as a fiber bundle with structure group H, dleP. In fact, it is easy to check thatP = SU(2) X U(I) U( I).
and a group G of bundle automorphism. The structure group The action of U ( I) on the U ( I) factor depends on the m
acts on the right, so we must reverse the H action appropri- quantum number. We now elaborate. Let G = SU(2),
ately. A point yEP corresponds to an A orbit in S. Let H = U( 1), A = U( 1). A basis for Y G = Y SU(2) is given
(a,g,h)EisCA XG XH. Then (g,h)ElyCG XH. Converse- by Lx: L y' L z ' where Lx = (i/2)ux ' Ly = (i/2)uy,
ly, supp~se (g,h)Ely ' Then there exists an aEA such that L z = (1/2)uz ' These are all anti-Hermitian, and have the
(a,g,h)Els' commutation relations [Lx,Ly] = L z • LetR = - ibe a ba-
sis for 2: A = 2: U(ll' exp(aR) = exp( - ia). Similarly, let
The tangent space Ty corresponds to the quotient
Q be a basis for Y H = 2: U(I)' We define the homomor-
Ts / N s' where Ns C Ts is tangent to the A orbit of s. Consider
phisms FAG and FAH by their derivatives at the identity,
the action of the derivative map ofT': Ts -- T s' where [Link] is
easy to see that the subspace Ns is preserved by this map, I: FAG ':2: U (I) -2:SU(2), R~2Lz' (A4)
Ns --Ns. Thus l' passes to a map of the quotient space F AH ':2: U(I) - 2 : U(I) , R~2mQ. (A5)

2111 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2111
 Let sES be the point (e,e )ESU (2) xU (1 ). The tangent Consider the points = (e,e,e)EG XB XH = S. We wish
space Ts = .!£'SU(2) EEl.!£'U(I) is spanned by Lx, L y' L z' Q. to find the isotropy group Is. Let (a,g,h) EA X G X H,
The null subpsace Ns is the range of (a,g,h): (e,e,e)~(g(FAGa-I),
(-FAG/)EElFAH': .!£'u(l)-Ts (A14)
RI---+ - 2L z + 2mQ . (A6) The isotropy group Is is specified by the following condi-
Let TEis' Then i = (FAG a, a, FAHa). Suppose that tions: F~Ba = F~a, g = FAG a, h = FAHa. Thus (g,h)Ely if
a = exp( - ~tR). Recall that the derivative of this map is and only if there exists an aEA satisfying the three conditions
above.
T I: .!£' SU(2) EEl.!£' U(l) -.!£' SU(2) EEl.!£' U(l) ,
An element T = (a,g,h )Eis maps s to itself. The deriva-
LxI---+Ad(exp(tLz ))Lx , tive thus gives a linear map Ts - Ts. The derivative T' is given
by the action of (a,g,h)Eis on (exp(tX) ,
LyI---+Ad(exp(tL z ) )Ly ,
exp(tY),exp(tZ»)EG XB XH, in the limit t-O. Since
L z I---+L z , Q---+Q . (A7) (a,g,h)Eis' F~Ba = F~Ba. We denote the common value as
We consider an infinitesimal action. Let t = E ~ 1. Denote by FABa,
TE the corresponding element of Is. The action of T; is to first T: (exp(tX) , exp(tY), exp(tZ»)~(g,b,ii) , (AI5)
order in E where
Lxl---+Lx + ELy, Lyl---+Ly - ELx ,
(A8) g= (FAGa)exp(tX)(FAG a - I
) ,

b = (FABa)exp(tY) (FABa- l ) , (AI6)

Downloaded from [Link]

Let L 'be the subspace of Ts spanned by Lx and L y . Let Fbe
the subspace of Ts spanned by L z and R. Here Ts decom-
Ii = (FAHa)exp(tZ)(FAHa-l) .
poses as Ts = L ' EEl F. The null subspace is a subspace of F, In the limit t - 0, we obtain the derivative map
Ns C F. The vertical subspace, tangent to the structure group T': (X,Y,Z)~(Ad(FAGa)X, Ad(FABa)Y, Ad(FAHa)Z).
action, is also a subspace of F, Vs C F. Also,
(AI7)
TJNs = L' EElF INs, and VJNs = FINs' Note that an ele-
ment of Is acts trivially on F, and acts by rotation on L '. In The null space N is tangent to the orbit of s under A,
particular, the action on VsINs is trivial. a: (e,e,e)I---+(FAGa- I ) , (F~Ba)(F~Ba-I), (FAHa»).
We need to consider maps Ws: TJNs - VJNs . Here w (AI8)
must act as the identity on VsINs' We need maps of
L '-F INs which commute with the mapT;. First note that Consider WE.!£' A' a = exp(tW)EA. In the limitt-O, we ob-
the composition woT; maps tain the derivative of the above map
Lxl---+liJ (Lx) + EW (Ly) , Lyl---+liJ (Ly) - EW (Lx) . (-FAG')EEl(F~B'-F~B')EEl(FAH'): .!£'A-Ts·
(A9) (AI9)
The composition T ; Ow maps The range of this map is the null subspace Ns C Ts. In other
words, Ns is generated by vectors of the form
Lxl---+liJ(Lx) , Lyl---+liJ(Ly)' (AlO)
(-FAG'W, F~B'W -F~B'W, FAH'W) , (A20)
For these two maps to be equal, we must have
where WE.!£' A •
w(Lx) =O=w(Ly )' (All)
We now wish to consider the bundle corresponding to
There is then a unique map w satisfying the necessary condi-
two degenerate eigenstates. Suppose we have two states with
tions. We conclude that there is a unique SU(2) invariant
quantum numbers m l and m 2 • We have seen above that the
connection for any value of m. This result also follows from
principal U ( 1) bundle associated to a line bundle over S 2 is
an easy application of Wang's theorem.
p = SU(2) XU(I) U( 1). Here the U( 1) action depends on
We now construct a more complicated example. Let
m. The U ( 1) xU ( I) principal bundle corresponding to the
S = G XB XH. The action ofG XA XHis as follows. Gacts
direct sum of the two line bundles can be written as
on S by left translation on the G factor, g':
p= SU(2) X U(I) U(1) XU(1), where U(l) acts on
(g,b,h) - (g'g,b,h). Here H acts on Sby right translation by
U(1) XU(1) in the obvious way. The structure group is
the inverse on the H factor, h ': (g,b,h) _ (g,b,hh ,-I). Con-
U( 1) XU( I). We wish to imbed our U( 1) X U( 1) principal
sider the group homomorphisms
bundle into a U(2) principal bundle. This amounts to con-
FAG: A -G, F~B: A -B, structing the principal bundle
(A12)
F~B: A-B, FAH : A-H,
p= (SU(2) XU(I) U(1) xU(1»)X U (l)XU(l) U(2) .
where A acts on S by
(A2l)
a: (g,b,h)l---+(g(FAGa- I ), (F~Ba)b(F~Ba-I), (FAHa)h).
We now elaborate. We shall write P as the quotient of a space
(A13) S by the action of a group A. We take
Note that the actions of A, G, and H mutually commute, so G = SU(2), B = U( 1) xU(1), H = U(2). Let
we indeed have an action of A X G XH on S. A = (U(1) XU(1) XU(1»). As before, we take Lx , Ly,L z as

2112 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2112
a basis for 2' SU(2)' We denote by Ko = (i12)1. (F~B' - F~B'): !t' U(I)XU(I)XU(I) -+!t' U(l)xU(I) ,

Kx = (i12)ux' Ky = (i12)uy • K z = (i12)uz a basis for RI1---+2mlQI + 2m 2Q2'

!t' U(2) • Let R I' R 2• R3 be the basis for
!t'A =!t'U(I)XU(I)XU(I)' Let QI' Q2 be the basis for
2' B = !t' U(I)xU(I)' We specify the homomorphisms by (A27)
giving their derivatives at the identity
Consequently !t' A' is one dimensional, spanned by the vec-
tor R ' = !R I + m IR2 + m 2R 3. The isotropy group is genera-
FAG': !t' U(I)XU(I)XU(I) -+!t'SU(2) •
ted by !t'I, which is the image of !t' A' under the map
R 11---+2L z • R 2>-----+O. R 3>-----+O, (A22)
( - FAG') (JJ (F~B' - F~B') (JJ (FAH '):
FAH ': !t' U(I)XU(I)XU(I) -+!t' U(2) ,
!t'A,-+!t'G(JJ!t'A (JJ!t'H'
R 1>-----+0. R21---+K O + K z • R31---+K O - K z • (A23)
R 'I---+L z + m (Ko + K z ) + m 2(Ko - K z ) ,
l

F~B': !t' U(I) xU(I) xU(I) -+!t' U(I) xU(I) •

I---+L z + (m + m 2)Ko + (m m 2)Kz •
l l -

I---+U. (A28)
F~B': !t' U(I)XU(I)XU(I) -+!t' U(I)XU(I)
Recall T = !t' G (JJ !t' B (JJ !t' H' The isotropy group Is acts on
•

(A25) Tby Ad(tU). Suppose t = €~ 1. Denote the corresponding

map TE : T -+ T. To first order,
We now define a decomposition of the tangent space T = Ts·

Downloaded from [Link]

LetX I.X2, ... ,XnET be a set of vectors. Denote by CIf (X I,X2•
.... Xn ). the linear span of the vectors. This is a subspace of T. (A29)
Define the following subspaces of T: Then T; acts trivially on E and K c. The action on L 1 and K 1

E = CIf (L z ,QpQ2) •L ([Link]) • K'

1 = '{; = '{; ([Link] ) • is given by
KC='{;(Ko,Kz )' F=E(JJK c . (A26) Lxl---+Lx + €L y • Lyl---+Ly - €L x •
Note that T = L 1 (JJF(JJK ' . Let s = (e,e.e). We need to de- Kxl---+Kx + €(m l - m 2)Ky ' Kyl---+Ky - €(ml - m 2)Kx .
termine Is. We must first find the subgroup of A ' CA satisfy-
(A30)
ing F ~B a = F ~B a for all aEA '. The vector space !t' A. C !t' A
satisfies F ~B' W = F ~B' W. for all WE!t' A.. Here !t' A' is the The null space N = Ns is tangent to the orbit of A. Here
kernel of the map N is generated by the range of the map

( - F~G') (JJ (F~B' - F~B') (JJ (FAH '): !t' U(I)XU(I)XU(I) -+!t'SU(2) (JJ!t' U(I)XU(I) (JJ!t' U(2) •

RII---+ - 2L z + 2mlQI + 2m 2Q2'

R21---+ - QI + Ko + K z ,
R31---+ - Q2 + Ko - K z • (A31)

Note that UEN. Also NCF, so N is invariant under Is· TIN -+ V IN,suchthatliJrestrictedto V INC T INistheiden-
We now determine the relation between TIN and VI N. tity. liJ must commute with the action orIs. This amounts to
Recall N C E, consider 0 = E IN. Recall E is five dimension- classifying maps liJ: L 1 -+ 0 (JJ K 1 which commute with the
al. N is three dimensional. so the dimension of 0 is two and action of Is ,

TIN=L'(JJO(JJK ' . (A32) liJ: L 1 -+ 0 (JJ K 1 ,

Now consider V=Kc(JJK'CT. We need to find

VI N C TIN. One checks that K (JJ N = F. We thus find C
Lyl---+Oy + Jy • (A34)
VIN=O(JJK ' • TIN=L'(JJVIN. (A33) where Ox, OyEO, and J x • JyEK I. Suppose J x = aKx + /3Ky,
Recall that the action of Is is trivial on F, and since Jy = yKx + 8Ky.
OCF. the action is trivial on O. We need to classify maps liJ: We now find the action ofT; 0liJ and liJoT; :

Lxl---+Ox + a{Kx + €(m l - m 2)Ky) + /3{Ky - €(m l - m 2))Kx •

I---+Ox + {a -/3€(m l - m2))Kx + (f3 + a€(m l - m 2))Ky ,

2113 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2113
 Ly~Oy + y(Kx + E(m l - m 2)Ky ) + t5(Ky - E(m l - m2)Kx) '
~Oy + (y - t5€(m l - m 2»)Kx + (15 + YE(m l - m2»)Ky , (A35)
wo"[ ;: L' -+ 0 (f) K' ,
Lx~Ox + EOy + aKx + /3Ky + EyKx + Et5Ky ,
~Ox + EOy + (a + q)Kx + (/3 + Et5)Ky ,

Ly~Oy - EOx + yKx + t5Ky - EaKx - E/3Ky ,

~Oy - EOx + (y - w)Kx + (15 - E/3)Ky . (A36)

We demand wOI; = I; ow. This requires Ad(exp(tKz »): : f V(2) -+:f V(2) ,

Ox=Oy=O, -/3(m l -m 2)=y, a(m l -m 2)=t5, Kx~cos(t)Kx + sin(t)Ky ,

-t5(m l -m 2) = -a, y(m l -m 2)= -/3. (A37) Ky~ - sin(t)Kx + cos(t)Ky ,

We first notice that w: L t -+K t, since Ox and Oy vanish. We

see that 15 and yare determined by a and /3. Combining (A40)
equations, we find
where t is a real number. So we see that two invariant connec-
/3(m l -m 2)2=/3, a(m l -m2)2=a. (A38) tions (a,/3) and (a ,/3 / ) are gauge equivalent if and only if
l

Downloaded from [Link]

We now see that if (m l - m 2 ) =1= ± 1, then all the param- a 2 + /3 2 = a'2 + /3 '2. An invariant connection (a,/3) is then
eters are zero, and the only solution is w: L '-+ 0 (f) K t by the gauge equivalent to the connection (a 2 + /3 2) 112,0). We see
zero map. If (m l - m 2 ) = ± 1, then a and/3 are arbitrary that the gauge equivalence classes of invariant connections
real numbers. We then get solutions are parametrized by R +, the non-negative real numbers.

w: L'-+K', 'M. V. Berry, Proc. R. Soc. London Ser. A 392, 45 (1984).

2B. Simon, Phys. Rev. Lett. 51, 2167 (1983).
Lx~Kx + /3Ky , 'F. Wilczek and A. Zee, Phys. Rev. Lett. 52,2111 (1984).
41. Segert, "Non-Abelian Berry's phase effects and optical pumping of
Ly~± (-/3Kx +aKy )' (A39) atoms," submitted to Ann. Phys.
'M. Tinkham, Group Theory and Quantum Mechanics (McGraw-Hill,
The space of invariant connections is parametrized by New York, 1964).
(a,/3)ER2. The Berry connection for (j,m) corresponds to 61. Segert, Phys. Rev. A 36, 10 (1987).
a = ~ [(j + m + P (j - m + !) ] 1/2, /3 = O. The connec- 7R. Y. Chiao and Y. S. Wu, Phys. Rev. Lett. 57, 933 (1986).
tion corresponding to a = /3 = 0 splits, and no other invar- "F. D. M. Haldane, Opt. Lett. 11,730 (1986).
9A. Tomita and R. Y. Chiao, Phys. Rev. Lett. 57, 937 (1986).
iant connection splits. 101. Moody, A. Shapere, and F. Wilczek, Phys. Rev. Lett. 56, 893 (1986).
We should consider two connections as equivalent if IlL. D. Landau and E. M. Lifshitz, Quantum Mechanics (Pergamon, New
they can be brought into one another by gauge transforma- York, 1977).
tion. A gauge transformation is a principal bundle automor- 12M. Reed and B. Simon, Methods ofModern Mathematical Physics (Aca-
demic, New York, 1980), Vol. I.
phism which induces the identity map on the base space. A 13M. Karoubi, K-Theory, An Introduction (Springer, Berlin, 1974).
gauge transformation maps a G-invariant connnection into 141. W. Milnor and I. D. Stasheff, Characteristic Classes (Princeton V.P.,
another G-invariant connection if and only if the gauge Princeton, 1974).
15M. S. Narasimhan and S. Ramanan, Am. I. Math. 83, 563 (1961).
transformation commutes with the G action. Since the G lOR. Bott and L. W. Tu, Differential Forms in Algebraic Topology (Springer,
action on the base space is transitive, it suffices to determine Berlin, 1982).
the action of the gauge transformation on the fiber over one 17S. Kobayashi and K. Nomizu, Foundations ofDifferential Geometry (Wi-
point. Take the point zES 2 • The gauge transformation at z ley, New York, 1969), Vols. I and II.
'"Y. Choquet-Bruhat, C. DeWitt-Morette, and M. Dillard-Bleick, Analysis,
must commute with the action of the isotropy subgroup, the Manifolds, and Physics, revised edition (Elsevier, Amsterdam, 1982).
subgroup of G which leaves zfixed. The only such automor- 19H. C. Wang, Nagoya Math. 1.13, I (1958).
phisms are generated by 2<>r. Kato, Phys. Soc. Ipn. 5,435 (1950).

2114 J. Math. Phys., Vol. 28, No.9, September 1987 Jan Segert 2114