Dirac operator on the Riemann sphere

A. A. Abrikosov, jr.∗
ITEP, B. Cheremushkinskaya 25,
117 259, Moscow, Russia
arXiv:hep-th/0212134v1 11 Dec 2002

Abstract
We solve for spectrum, obtain explicitly and study group properties of eigenfunc-
tions of Dirac operator on the Riemann sphere S 2 . The eigenvalues λ are nonzero
integers. The eigenfunctions are two-component spinors that belong to representa-
tions of SU (2)-group with half-integer angular momenta l = |λ| − 21 . They form
on the sphere a complete orthonormal functional set alternative to conventional
spherical spinors. The difference and relationship between the spherical spinors in
question and the standard ones are explained.

PACS: 02.20.Qs, 02.30.Gp, 02.30.Nw, 02.30.Tb.

keywords: Dirac operator, spherical functions, spherical spinors, orthogonal poly-
nomials, monopole harmonics.
arXiv: hep-th/0212134

Introduction
The question of spectrum and eigenstates of Dirac operator on the Riemann sphere S 2
is a well posed mathematical problem that may be solved exactly. The resulting family
of orthogonal two-component spinors plays the role of orthogonal spherical functions in
problems with half-integer momenta. Therefore the solutions are of fundamental value.
The eigenspinors were first found by Newman and Penrose [1] and the relation with
SU(2) matrix elements was deduced later in [2]. These eigenspinors also reappear from
time to time under the name of “monopole harmonics”, see, for instance [3]. A general
construction of Dirac operator eigenfunctions on N-dimensional spheres was given in
paper [4] and the two-dimensional case was specially addressed in [5] (see also [6]).
One must mention several physical situations where the solutions may find application.
The first involves the so-called spectral boundary conditions. Those have been introduced
by Atiah, Patodi and Singer [7] and are widely used now in studies of spectral properties of
operators in the limited space [8]. Being compatible with chiral symmetry these boundary
conditions are perfectly suited for investigations of chirality breaking in quantum field
theory.
∗
E-mail: [email protected]

1
 The spectral boundary conditions are defined in terms of eigenvalues of Dirac operator
on the boundary. Thus our problem bears a direct relation to the (spherical) spectral bag
that allows to hold quarks while preserving the chiral symmetry.
Another application refers to physics of electrons in substances called fullerenes. Their
molecules consist of carbon atoms located in vertices of polyhedra or on the surface of
clusters with a spheroidal crystallographic structure. The most abundant and the most
spherical is the C60 molecule (known also as Buckminsterfullerene or buckyball) that looks
like a standard soccer ball. The interest to electronic structure of fullerenes was heated
by their unconventional superconducting properties with transition temperatures higher
than those of the classic superconductors (reaching up to 60 K!). In the continuum limit
the conducting electrons of fullerene molecules obey the Dirac equation [9] and our results
are pertinent for the continuous description of electron states in fullerenes.
Our present goal was to implement an explicit classical approach to eigenstates of the
Dirac operator with the intention to make them a tool that could be later employed in
physical problems. A short presentation of the topic was given in [10]. In principle one
could extract the required information from the already cited papers but this would take
some effort.
It was shown in [4] that for Dirac operator on the N-dimensional sphere one may
separate the dependence on principal polar angle and solve the obtained second order
differential equation. The result was written in terms of Jacobi polynomials. This led to
a recurrent formula that related spinor spherical functions on S N to those on S N −1 . The
subsequent reduction (quite conceivable for S 2 though) was left to the reader.
The two-dimensional case was considered as an example in [5] (Chapter 9) and may
be also found in [6]. However there the problem was elegantly solved with the help
of complex coordinates on the Riemann sphere whereas the transformation to standard
spherical coordinates was left aside.
With practical applications in mind we shall try to make the paper self-contained. It
has the following structure. Section 1 introduces the notation and expounds on general
properties of Dirac operator on the Riemann sphere. The spectrum and eigenfunctions
themselves are investigated in Section 2. We start from the eigenvalues (Sect. 2.1), analyse
SU(2) properties of solutions (Sect. 2.2), then write them out explicitly (Sect. 2.3) and
check their properties with respect to time-reversal (Sect. 2.4). Section 3 explains the link
between the newly found spherical spinors and the conventional ones [11]. After remind-
ing the spinor transformation rules (Sect. 3.2) we convert our eigenspinors to Cartesian
coordinates and derive the relationship in question (Sect 3.3). The conclusion is followed
by two technical Appendices dwelling on properties of Jacobi polynomials (Appendix A)
and spherical functions (Appendix B).

1 Generalities
We shall start with introducing spherical coordinates and writing down the Dirac operator
for free massless fermions on the Riemann sphere S 2 . Then we shall show that Dirac
operator has has no zero eigenvalues on the sphere.
The sphere of unit radius S 2 may be parameterized by two spherical angles q 1 = θ,
2
q = φ that are related to Cartesian coordinates x, y, z as follows:

x = sin θ cos φ; y = sin θ sin φ; z = cos θ. (1)

2
The metric tensor and the natural diagonal zweibein on the sphere are well known:

gαβ = diag(1, sin2 θ); eaα = diag(1, sin θ); gαβ = δab eaα ebβ . (2)

From here on we shall denote zweibein and coordinate indices by Latin and Greek letters
respectively. Multiplication by the zweibein converts the two types of indices into each
other:
Aa = eαa Aα ; Aa = eaα Aα ; eaα eαb = δba . (3)
In curvilinear coordinates the ordinary Euclidean partial derivatives must be replaced
by covariant ones. For quantities with Latin indices those include the spin connection
Rαa b 1 :
Dα Aa = ∂α Aa + Rαa b Ab and Dα Aa = ∂α Aa − Ab Rαb a . (4)
In our case the nonzero components of spin connection are:

Rφ12 = −Rφ21 = − cos θ. (5)

Spinors in two dimensions have two components and the role of Dirac matrices belongs
to Pauli matrices: γ a → (σx , σy ) where
! ! !
0 1 0 −i 1 0
σx = ; σy = ; σz = . (6)
1 0 i 0 0 −1

Covariant derivatives of 2-spinors are also expressed in terms of the spin connection:
i
∇α ψ = ∂α ψ + Rαab σab ψ. (7)
4
Here σab are the rotation generators for spin- 12 fields:

i
σ1 2 = −σ2 1 = − [γ1 , γ2 ] = σz . (8)
2
A standard calculation gives for covariant derivatives:
iσz
∇θ = ∂θ and ∇φ = ∂φ − cos θ. (9)
2
Now we may define the Dirac operator. In spherical coordinates it is given by the
convolution of covariant derivatives in spinor representation with the zweibein and σ-
matrices:
− i∇ˆ = −i eα a σa ∇α . (10)
With the help of definitions (2, 6, 9) it is straightforward to obtain:
!
ˆ = −iσx ∂θ + cot θ
− i∇ −
iσy
∂φ . (11)
2 sin θ

Finally let us show that Dirac operator on the sphere has no zero eigenvalues. This is a
general property of manifolds with positive curvature known as Lichnerowicz theorem [12].
1
Since there is actually no difference between upper and lower Latin indices the second of eqs. (4) is
nothing but the skew symmetry of R̂.

3
 ˆ 2 . Obviously if −i∇
Consider the square of Dirac operator (−i∇) ˆ had a zero eigenvalue
ˆ 2
then (−i∇) must have it either. Let us split the product of σ-matrices in (∇) ˆ 2 into
symmetric and antisymmetric parts σ α σ β = 12 {σ α , σ β } + 21 [σ α , σ β ]. Implementing the
relation between commutators of covariant derivatives and curvature we obtain:

ˆ 2 + ∇2 = − i σ αβ [∇α , ∇β ] = 1 Rαβ .
(−i∇) (12)
F
2 4 αβ
αβ
Here Rαβ is the trace of Riemann curvature tensor and ∇2F is the covariant Laplace
operator in fundamental representation of the SU(2)-group:
1 √
∇2F = g αβ (∇α ∇β − Γγαβ ∇γ ) = √ ∇α g αβ g ∇β , (13)
g

where ∇α, β, γ are covariant derivatives in fundamental representation (7), Γγαβ is the
Christoffel symbol and g is the determinant of metric, g = det kgαβ k = sin2 θ.
αβ
The result (12) is called the Lichnerowicz formula. In the case of sphere Rαβ = 2 so
that
ˆ 2 + ∇2 = 1 .
(−i∇) (14)
2
(certainly this may be proved directly). The positive difference is brought about by the
curvature of Riemann sphere. The covariant Laplacian is a non-positive operator ∇2 ≤ 0.
Hence, by virtue of (14), (−i∇)ˆ 2 must be strictly positive and Dirac operator −i∇ˆ has
no zero eigenvalues.
In conclusion I would like to present the explicit formulae for the square of Dirac
operator and the fundamental Laplace operator on the unit sphere:

ˆ 2 = − 1 ∂θ sin θ ∂θ − ∂φ2 cos θ 1 1

(−i∇) 2 + iσz 2 ∂φ + + ; (15a)
sin θ sin θ sin θ 4 sin2 θ
1 ∂φ2 cos θ 1 1
−∇2F = − ∂θ sin θ ∂θ − 2 + iσz 2 ∂φ − + . (15b)
sin θ sin θ sin θ 4 sin2 θ
Obviously the difference between the two agrees with (14).

2 Dirac operator on the sphere

Now we shall study the eigenvalue problem for Dirac operator on the sphere S 2 . First
we shall derive the differential equation for spinor components, find eigenvalues and the
general form of solutions. Then we shall study properties of the eigenfunctions that arise
from the SU(2)-invariance of the problem and classify the solutions. Finally we shall
present them explicitly and discuss their behaviour under the complex conjugation and
time-reversal.

2.1 The eigenvalue problem

Eigenfunctions of the Dirac operator (11) are two-component spinors that satisfy the
eigenvalue equation: ! !
ˆ αλ (θ, φ) αλ (θ, φ)
−i∇ =λ . (16)
βλ (θ, φ) βλ (θ, φ)

4
 This system of first order partial differential equations in α and β allows separation
of variables. The first thing is to isolate the φ-dependence by expanding the spinors into
Fourier series:
! !
αλ (θ, φ) X exp i m φ αλm (θ) 1 3
= √ ; m = ± , ± ... (17)
βλ (θ, φ) m 2π βλm (θ) 2 2

where m are half-integers since we work with the spin- 21 field. This converts Eqn. (16)
into
!
cot θ m
− i ∂θ + + βλm (θ) = λ αλm (θ); (18a)
2 sin θ
!
cot θ m
−i ∂θ + − αλm (θ) = λ βλm (θ). (18b)
2 sin θ
By analogy with the ordinary quantum mechanics the number m may be called the
projection of angular momentum onto the polar axis.
Separate equations for spinor components α and β may be obtained by taking the
square of Dirac operator (15a):
m2
" # ! !
1 m cos θ 1 1 αλm 2 αλm
− ∂θ sin θ ∂θ + 2 − σz 2 + + =λ . (19)
sin θ sin θ sin θ 4 4 sin2 θ βλm βλm

Mark that because of the σz present in the third term the equations for upper and lower
components are different. As may be seen from eqn. (14) the difference arises already at
the level of Laplace operator in fundamental representation (15b).
The further simplification comes from the change of variables x = cos θ, x ∈ [−1, 1],
that converts (19) into the generalized hypergeometric equation:
1
m2 − σz m x +
" # ! !
d d αλm (x) 1 αλm (x)
 
(1 − x2 ) − 4
=− λ − 2
. (20)
dx dx 1 − x2 βλm (x) 4 βλm (x)

Take the notice that the replacement x → −x (or m → −m) is equivalent to trading α
for β. Thus the upper and lower spinor components are conjugate with respect to mirror
reflection.
Equation (20) is singular at the poles of the sphere x = ±1. After the redefinition of
the unknowns
 
(1 − x) 2 |m− 2 | (1 + x) 2 |m+ 2 | ξλm (x) 
! 1 1 1 1
αλm (x)
=  , (21)
(1 − x) 2 |m+ 2 | (1 + x) 2 |m− 2 | ηλm (x)
1 1 1 1
βλm (x)

we arrive to the separate equations of hypergeometric type in ξλm and ηλm :

d2
( " # ) !
m d 1 ξλm

2
(1 − x ) 2 + σz − (2|m| + 2) x − m (m + 1) + λ2 − = 0.
dx |m| dx 4 ηλm
(22)
In order that these equations had on the interval x ∈ [−1, 1] square integrable solutions
λ must fulfill the condition (see (A.1)–(A.3))
2
1

2
λ = n + |m| + , (23)
2

5
with non-negative integer n ≥ 0. Thus λ are nonzero integers

λ = ±1, ±2 . . . , (24)

and indeed the Dirac operator has no zero-modes. Equations (22) are Jacobi-type equa-
tions and their solutions are Jacobi polynomials of the n-th order [13, 14]:
(|m− 1 |, |m+ 21 |)
!  
ξλm (x) Cαmn Pn 2 (x)
= (|m+ 21 |, |m− 21 |)
. (25)
ηλm (x) Cβmn Pn (x)

A brief review of Jacobi polynomials may be found in Appendix A. Properties of solutions

(25) will be discussed in Section 2.3.
The constants Cαmn , Cβmn may be fixed by substituting the solutions into equation (16)
and normalizing the obtained functions. Let us first find the relation between Cα and Cβ .
The Dirac operator is sensitive to the sign of m. Applying it to functions (21) at m > 0
gives for the Fourier components (the same can be done for m < 0):
   1 h  i 
! 1+x 4 1 d
ˆ αλm m  − m+ ηλm − (1 − x) dx ηλm 
− i∇ = i(1 − x2 ) 2  1−x
 1 h
2
(26)
βλm 
1−x 4
m+ 1

d
ξλm + (1 + x) dx ξλm
i 
1+x 2

(Note that coefficients Cα, β were absorbed in functions ξ, η.) Substituting this into the
eigenvalue equation (16) written in terms of ξ and η and using identities (A.7) of the
Appendix we find the condition on Cαmn , Cβmn which, after generalization to negative m,
takes the form: ! !
1 Cαmn Cβmn
 
i n + |m| + = λ sgn m . (27)
2 −Cβmn Cαmn
Provided that λ is given by (23) this equation has nonzero solutions and the ratio of
constants Cα /Cβ is defined by the sign of the product mλ:

Cβmn = i Cαmn sgn(m λ). (28)

The absolute values of the constants can be found from the normalization conditions,
Z 2π Z π h i
dφ |αλ (θ, φ)|2 + |βλ (θ, φ)|2 sin θ dθ = 1. (29)
0 0

This immediately refers us to the norms of Jacobi polynomials with the result:
q
n! (n + 2m)!
|Cαmn | = |Cβmn | = m+ 12
. (30)
2 Γ(n + m + 21 )

The last condition determines the eigenfunctions up to a complex phase that will be
fixed in Section 2.4. Presently we shall not write the solutions explicitly leaving this till
Section 2.3.
The last remark due here concerns an apparent contradiction between the integer
spectrum (24) and formula (14). The latter implies that spectra of the two operators
(15) are similar but to the shift 21 . Hence according to formula (24) the lowest eigenvalue
of the (minus) covariant Laplace operator −∇2F is 21 and not zero as one could naively

6
expect by analogy with scalar fields. The confusion is readily resolved if one recalls
that the aforementioned integer spectrum was obtained for spinors. Unlike scalars those
satisfy antiperiodic boundary conditions α, β(0) = −α, β(2π) and are expanded in half-
integer Fourier harmonics. Therefore although formally −∇2 and −∇ ˆ 2 differ only by the
numerical constant in fact both spectra and eigenfunctions for scalars and spinors are
different.

2.2 The SU (2) algebra

Let us show that the Dirac operator on the sphere S 2 is invariant under transformations
of the SU(2) group. The Weyl —Cartan set of generators looks as follows:
∂
L̂z = −i ; (31a)
∂φ
!
iφ ∂ ∂ σz
L̂+ = e + i cot θ + ; (31b)
∂θ ∂φ 2 sin θ
!
∂ ∂ σz
L̂− = −e−iφ − i cot θ − . (31c)
∂θ ∂φ 2 sin θ
These operators satisfy the standard commutation relations of the SU(2) algebra:
h i
L̂z , L̂+ = L̂+ ; (32a)
h i
L̂z , L̂− = −L̂− ; (32b)
h i
L̂+ , L̂− = 2L̂z . (32c)

In our case representations of SU(2)-group are characterized by half-integer highest

weight l (see (37). It will be shown in Section 3.3 that it corresponds to the total angular
momentum of the fermion state. The Casimir operator L̂2 takes the value
1
hl | L̂2 | li = hl | L̂2z + (L̂+ L̂− + L̂− L̂+ ) | li = l(l + 1). (33)
2
The basis vectors of the representation are classified by half-integer projection of angular
momentum onto the polar axis that varies in the range m = lz = −l, . . . l. The number
of basis vector in the representation is 2l + 1. The raising and lowering operators L̂+ and
L̂− change the projection by unity L̂± | l, mi ∝| l, m ± 1i.
A direct check proves that generators (31) commute with spherical Dirac operator
(11): h i h i h i
ˆ L̂z = −i∇,
−i∇, ˆ L̂+ = −i∇, ˆ L̂− = 0. (34)
This means that the Dirac operator is SU(2) invariant and its eigenfunctions may be clas-
sified according to their SU(2) transformation properties. Moreover, the action of raising
and lowering generators L̂+ and L̂− transforms its eigenfunctions into the eigenfunctions
with the same λ but greater or smaller m respectively.
The square of Dirac operator (15a) and Casimir operator (33) may be diagonalized
simultaneously
−∇ ˆ 2 = L̂2 + 1 , (35)
4
7
and their eigenvalues are interrelated:
1
hλ, n|L̂2 |λ, ni = l(l + 1) = λ2 − = (n + |m|)(n + |m| + 1). (36)
4
We conclude that the proper values of angular momentum are indeed half-integers related
to the eigenvalues of Dirac operator
1 1 3 5
l = |λ| − = n + |m| = , , . . . (37)
2 2 2 2
The final remark concerns reducibility of our representation. It is well known that
spinor representations in even dimensions are reducible and one may separate the left and
right spinor components. One may see that formally this is the case. Operators (31) are
diagonal, so the upper and lower spinor components behave with respect to the SU(2)-
group like two singlets with equal momenta l. However diagonalizing the Dirac operator
requires joining them into the pair of doublets with λ = ±(l + 12 ).

2.3 The spinor spherical functions

Here we shall show how the formerly obtained functions (25) can be grouped into SU(2)
multiplets. The multiplets can be constructed in the standard manner, i. e by successive
application of the lowering operator L̂− to the highest weight vector, namely to the
function with the maximum projection m = l of angular momentum. Let us pass right
to the result.
It is quite natural to use instead of n the angular momentum l = n + |m|. Let us
introduce the integers l± = l ± 21 and m± = m ± 21 . Having in mind that according to
formula (27) solutions for positive and negative λ are different we may write (here again
x = cos θ):
q
l+ 1
(m− +|m− |)
(l + m)!(l − m)!
Υ±
lm (x, φ) = ±i (−1)
2
1
2|m|+ 2 Γ(l+ )
(|m− |, |m+ |)
 q 
eimφ  ∓i ρ(|m− |, |m+ |) (x) Pl−|m| (x) 
× √ q
+ |, |m− |) (|m+ |, |m− |)
; (38)
2π sgn m ±i ρ(|m (x) Pl−|m| (x)

Here the ±-superscripts stand for the sign of λ and ρ(α, β) = (1 − x)α (1 + x)β is the
weight function for Jacobi polynomials P (α, β) (x) defined in the Appendix A. The overall
multiplicative constants in these expressions were chosen in order to fix the right signs of
matrix elements (45) and ensure the correct behavior of spherical functions under complex
conjugation (47).
The relation between Υ+ and Υ− is very simple:

Υ± ∓
lm (x, φ) = ±iσz Υlm (x, φ). (39)

The structure of this representation allows to prove directly the orthogonality of spher-
ical functions:
Z 2π Z π
hΥεl11m1 |Υεl22m2 i = dφ (Υεl11m1 )† Υεl22m2 sin θ dθ = δ ε1 ε2 δl1 l2 δm1 m2 . (40)
0 0

8
Integration over dφ generates the Kronecker δm1 m2 -symbol. After that the integrand be-
± ∓ (|m± |, |m∓ |) (|m± |, |m∓ |)
comes a sum of two expressions of the type ρ(|m |, |m |) Pl1 −|m| Pl2 −|m| that adverts
us to the orthogonality of Jacobi polynomials (A.5). If the only distinction between the
Υ-functions is ǫ1 6= ǫ2 one should turn back to eqn. (39). Due to the presence of σz for
ǫ1 = −ǫ2 the equal contributions of upper and lower spinor components are subtracted
from each other and cancel.
Two more economic representations can be found with the help of identities (A.9) of
the Appendix. The first is:
v
+ −
il (−1)l ut (l + m)!
u
Υ±
lm (x, φ) = ± l + +
2 Γ(l ) (l − m)!
 √
m− m+

eimφ  ∓i(1 − x)− 2 (1 + x)− 2 dl−m − +
(1 − x)l (1 + x)l 
× √ √ m+ m−
dxl−m
l−m + −
. (41)
2π ±i(1 − x)− 2 (1 + x)− 2 d
dxl−m
(1 − x)l (1 + x)l
The second one differs from it mainly by the sign of m.
v
+
il (−1)l−m u
t (l − m)!
u
Υ±
lm (x, φ) = l + +
2 Γ(l ) (l + m)!
 √
m− m+

eimφ  ∓ ∓i(1 − x) 2 (1 + x) 2 dxdl+m l+ l−
l+m (1 − x) (1 + x)
× √ √ + −
 . (42)
2π ± ±i(1 − x) m2 (1 + x) m2 dx
dl+m
l+m (1 − x)
l−
(1 + x) l+

The important feature of these functions is that they are not zero only for values of m
ranging from −l to l. Hence each multiplet contains 2l + 1 terms. Indeed once l = n + |m|
is a half-integer then l+ and l− in their turn must be integers. As a result the products
± ∓
(1 − x)l (1 + x)l are polynomials of order 2l and their derivatives of orders higher than
2l are zero.
These expressions are perfectly suited for acting on them by L̂+ and L̂− operators.
First let us note that for any m
 h i− 1   h i1 
(m− , m+ ) 2 (m− , m+ ) 2
imφ 
ρ ρ
L̂− e  h + − i− 1  = L̂+ eimφ  h + − i 1  = 0. (43)
  
ρ(m ,m ) 2
ρ(m ,m ) 2

If we take Υ in the form (41) when acting by L̂− and in the form (42) when acting by L̂+
we shall readily find that
q
L̂− Υl, m = (l + m)(l − m + 1) Υl, m−1 ; (44a)
q
L̂+ Υl, m = (l + m + 1)(l − m) Υl, m+1 . (44b)
Thus we retain the standard expressions for the matrix elements of momentum operators,

q
hl, m − 1 |L̂− | l, mi = hl, m |L̂+ | l, m − 1i = (l + m)(l − m + 1); (45a)
hl, m |L̂z | l, mi = m. (45b)
Deducing these relations directly from formulae (38) would require engaging but to (43)
the Jacobi polynomial differentiation formula (A.6).

9
2.4 Complex conjugation and time-reversal
Finally I would like to comment the choice of coefficients in definition (38). We have al-
ready mentioned that insofar as the momentum operators (31) are diagonal the upper and
lower spinor components of Υl, m belong to different representations of the SU(2)-group.
The eigenvalue equation (16) and normalization condition (29) link the components to
each other and fix their ratio (28) and absolute values. In the mean time the overall
complex phase remains free. It may be determined from the behaviour of eigenfunctions
under the complex conjugation. The latter in its own turn is closely related to properties
with respect to time-reversal symmetry in 2 + 1-dimensions. Action of the time-reversal
(T ) transformation onto spinors is described by the formulae [15]:

T : ψ → −iσy ψ ∗ . (46a)

The −iσy matrix guarantees that the time reversed spinors belong to the same funda-
mental representation of SU(2) group and not to the complex-conjugated one. This
definition agrees with the time-reversal properties of ordinary spherical functions Yl, m
(see Appendix B)
T : Yl, m → Yl,∗−m = (−1)l−m Yl, −m . (46b)
Definitions (46) guarantee the equivalence of integer-momentum representations of SU(2)
written both in terms of spinors and spherical functions.
The same line was pursued by our definitions of Υ-functions. The idea was that the
T -transformation must convert Υ’s into themselves: T : Υl, m → (−1)l−m Υl, −m . This
relates components of the complex conjugated spinor Υ∗l, m to those of Υl, −m as follows:
! !
−βl,∗ m αl, −m
T : Υl, m = −iσy Υ∗l, m = = (−1) l−m
. (47)
αl,∗ m βl, −m

One may show that for the functions introduced in the previous section this is indeed
the case. The simplest way to is to take Υl, m and Υl, −m in different representations, say
(41) and (42). A direct comparison of the coefficients proves (47) (mark that (−m)± =
−(m∓ )).
The time-reversal transformation performed twice changes the sign of the eigenfunc-
tions:
T 2 : Υl, m = (−1)2l Υl, m = −Υl, m . (48)
This also complies with what should be expected from general considerations for repre-
sentations with half-integer momentum.

3 Relation between Υ and Ω spherical spinors

3.1 Cartesian coordinates
Now we shall study the relation of the newly found functions Υ to the standard spherical
spinors. In order to do this we are going to convert Υ-functions to Cartesian coordi-
nates. This requires passing from the two-dimensional Riemann sphere S 2 to the three-
dimensional Euclidean space. In order to access the new dimension we must first add

10
to the two polar angles q 1 = θ and q 2 = φ the radial coordinate q 3 = r. The relations
between the 3-dimensional spherical and Euclidean coordinates are standard:
x1 = x = r sin θ cos φ; x2 = y = r sin θ sin φ; x3 = z = r cos θ. (49)
The old metric tensor and zweibein on the sphere (2) must be substituted by the new
spherical metric and dreibein
gαβ = diag(r 2 , r 2 sin2 θ, 1); eaα = diag(r, r sin θ, 1). (50)
The place of the third Dirac matrix is taken by γ 3 = σz .

3.2 Spinor transformation rules

Now we would like to transform our formulae to the Cartesian coordinates. Note that the
transformation includes a rotation of the dreibein. Accordingly the spinors will change.
Let us remind the corresponding transformation rules.
Local rotations of the dreibein associated with coordinate changes generate the spin
connection. The spinor and dreibein connections depend on each other so that γ a -matrices
with dreibein indices stay constant.
In order to find the explicit form of spinor transformation let us consider an auxiliary
local bilinear Aα (q) = ψ † (q)σ α ψ(q) with ψ(q) being an arbitrary spinor. Let q α be the old
and xµ the new coordinates (there was no reason yet to claim that xm were Euclidean).
Being a simple contravariant vector A obeys the following transformation law:
∂xµ α ∂xµ α
Aµ (x) = A (q) and Aa
= ea
µ (x) e (q) ψ †(q)σ b ψ(q), (51)
∂q α ∂q α b
where x and q refer to the same spatial point. Take the note of the two different dreibeins
that appear in this expression: the first one eaα (q) refers to the original frame and ebµ (x)
µ
to the new one. The matrix Uba = eaµ (x) ∂x eα (q) is orthogonal Ũ U = U Ũ = 1 (where
∂q α b
Ũ denotes the transposed matrix Ũba = Uab ). It describes the local rotation of objects
with dreibein indices that arise from the coordinate transformation. Let us introduce into
equation (51) an auxiliary unitary 2 × 2 matrix V (q) (V V † = 1):
† α ∂xµ α
a
A = Uba ψσ ψ= eaµ (x) α
eb (q) ψ †(q) V V † σ b V V † ψ(q). (52)
∂q
We shall show shortly that it is always possible to choose V so that it will compensate
the rotation U of the dreibein putting the Pauli matrices back into the conventional form:
∂xµ α
eaµ (x) e (q) V † (q) σ b V (q) = σ a . (53)
∂q α b
Provided that this condition holds the σ-matrices will stay the same in the new coordinate
frame. The associated transformation rule for spinors must be:
ψ(x) = V † ψ(q), ψ † (x) = ψ † (q) V. (54)
Let us calculate the V -matrix. Similarly to any other rotation the matrix Uba may be
written in terms of the adjoint O(3) rotation generators L̂ as:
h ib
L̂a = −i ǫabc . (55)
c

11
Let unit vector nc (q) point along the axis and θ(q) be the angle of local rotation. Then
h ia h ia
Uba (q) = exp iL̂c nc (q)θ(q) = na nb + (δba − na nb ) cos θ + i nc L̂c sin θ. (56)
b b

From the commutation relations of the Pauli matrices it is straightforward to see that
ia  i i
h 
exp iL̂c nc θ exp σ c nc θ σ b exp − σ c nc θ = σ a . (57)
b 2 2
Therefore the matrices V and V † compensating the rotation (56) are
i i
V (q) = exp − σ c nc (q)θ(q) and V † (q) = exp σ c nc (q)θ(q). (58)
2 2
Thus the transformation of spinors is unambiguously determined by the rotation angle θ
and direction of the axis n which may be found either from geometrical considerations or
algebraically by solving equation (53).

3.3 Cartesian realization of Υ-functions

Now we can apply the theory and find out what do Υ-spinors look like in the ordinary
Cartesian coordinates. First we shall derive the general formulae transforming spinors
from the spherical to the orthogonal frame and then apply them to the spinors in question.
Taking for input the relation between spherical and orthogonal coordinates (49), the
explicit spherical dreibein (50) and the trivial Cartesian dreibein eaµ = δµa we obtain the
rotation matrix (eaµ = δµa ).
 
cos θ cos φ − sin φ sin θ cos φ
∂xµ
Uba = eaµ (x) α eαb (q) =  cos θ sin φ cos φ sin θ sin φ  . (59)
 
∂q
− sin θ 0 cos θ

This may be decomposed into the product of two successive turns:

h ia
Uba (q) = exp −i φL̂3 exp −iθL̂2 . (60)
b

Geometrically these two rotations align the local axes of spherical frame with those of
Cartesian one. Obviously θ and φ are nothing but the corresponding Euler angles. Spinors
must also suffer the same two successive rotations with V -matrices being:
 
iφ iφ
iσy iσz e 2 cos θ2 e− 2 sin 2θ 
V = exp θ exp φ =  iφ iφ ; (61a)
2 2 −e 2 sin θ2 e− 2 cos 2θ
 
iφ iφ
iσz iσy e− 2 cos θ2 −e− 2 sin θ2 
V † = exp − φ exp − θ =  iφ iφ . (61b)
2 2 e 2 sin θ2 e 2 cos θ2

Now we shall perform the V -rotation on spinor spherical functions Υ (the trick is to
use representation (41) when calculating the upper line and apply (42) to the lower one):
 q q 
l+m
1 Yl− m− ± l−m+1 Yl+ m− 1
V † Υ±
lm
2l
= √  q l−m q 2l+2
l+m+1
 = √ (Ωl, l− , m ∓ Ωl, l+ , m ) . (62)
2 2l
Yl− m+ ∓ 2l+2
Yl+ m+ 2

12
 Here Yl m are spherical functions (see Appendix B) and Ωj, l, m are the ordinary spher-
ical spinors that realize an alternative representation of spin- 12 states on the sphere [11].
The quantum numbers characteristic of a spinor Ωj, l, m are: the half-integer total angular
momentum j > 0, integer orbital momentum l = j ± 12 and projection of the total mo-
mentum onto the polar axis m (that is again a half-integer). They are the two-component
functions defined by the formulae
 q   q 
j+m j−m+1
Yl, m− 1 − Yl, m− 1
Ωl+ 1 , l, m =  q 2j 2  and Ωl− 1 , l, m =  q 2j+2 2 . (63)
j−m j+m+1
2
2j
Yl, m+ 1 2
2j+2
Yl, m+ 1
2 2

Similarly to Υ the Ω-spinors make an orthonormal basis in the space of spinor functions
on S 2 : Z π Z 2π
sin θ dθ dφ Ωj l m (cos θ, φ) Ωj ′ l′ m′ (cos θ, φ) = δj j ′ δl l′ δm m′ . (64)
0 0
We conclude that in Cartesian coordinates the functions Υ may be expressed in terms
of the ordinary spherical spinors Ω (see (63)) but do not coincide with those. Representa-
tion (62) helps finding the standard quantum numbers of Υ-spinors. Inasmuch as they are
superpositions of states with equal values of j and m Υ-functions must the same values
of total angular momentum Jˆ and its z-projection Jˆz :

Ĵ2 Υlm = (L̂C + ŜC )2 Υlm = l(l + 1) Υlm and Jˆz Υlm = m Υlm ; (65)

where L̂C and ŜC are the Cartesian operators of vectors of orbital momentum and spin
respectively. In the mean time being composed of states with different orbital momenta
l+ and l− our spinors do not diagonalize L̂2C . Still it is interesting to calculate its average:
2
1h+ + 1
i 
hΥlm |L̂2C |Υlm >= l (l + 1) + l− (l− + 1) = l + = λ2 . (66)
2 2

You see that the average value of L̂2C for spherical spinors Υ is equal to the squared
eigenvalue of spherical Dirac operator. Clearly this is a mere coincidence since in this
representation −i∇ ˆ is diagonal whereas Lˆ2C is not.
Now it is the time to compare the two types of spinors. Either of them forms a complete
orthogonal functional system and belongs to the same representation of the SU(2) group.
Moreover Υ’s are linear combinations of Ω’s and vice versa. The question arises what is
the origin of distinction between them.
The answer is very simple. Both Ω and Υ diagonalize one more operator but to Jˆ and
Jˆz . These are the square of orbital momentum Lˆ2C for Ω and spherical Dirac operator for
Υ. The different choice of supplementary operators leads to different representations.
The ordinary spinors Ω were obtained from solving in the flat space the three-dimen-
sional Laplace equation on spinors
1 2 1
 
− ∆Ψ = L̂C + ∂r r 2 ∂r Ψ = k 2 Ψ. (67)
r2 r2
The Laplace operator is diagonal with respect to spin so that the upper and lower compo-
nents of Ψ behave as independent scalars. Separating the radial and angular dependencies
one arrives at solutions with integer values of orbital momentum lC = 0, 1, . . . Then those
are united into doublets with definite total momentum j and its projection jz = m.

13
 Now the Υ-spinors are found by diagonalizing Dirac operator on the Riemann sphere
that is an essentially curved manifold. One may clearly mark the difference from flat Eu-
clidean space from the curvature term that manifestly enters the square of Dirac operator
(14). Besides the nontrivial spin connection makes even the covariant two-dimensional
Laplace operator (15) on the sphere differ from the angular part of (67).
To summarize one may say that as a result Ω-spinors are better suited for separation
of variables in spherically symmetric problems of non-relativistic quantum mechanics. On
the other hand Υ-functions may be useful when one is specially interested in properties
of fermions localized on the sphere.
In conclusion we present for reference the Cartesian realization of Dirac operator (11)
ˆ
−i∇C = −i V † ∇V ˆ with V -matrices given by (59)

ˆ C = −iσx eiφσz (cos θ ∂θ − sin θ) + iσz (sin θ ∂θ + cos θ) − iσy eiφσz ∂φ

− i∇
sin θ
iσz φ iσz φ iσy
= −ie− 2 σx ∂θ eiσy θ e 2 − eiφσz ∂φ . (68)
sin θ
ˆ C )2 is left to the reader as an exercise.
Calculation of (−i∇

Conclusion
We have found the eigenfunctions of Dirac operator on the Riemann sphere S 2 . This
family of orthonormal two-component spinors may be used for basis in problems involving
fermions in spherically symmetric fields. The eigensponors are expressed in terms of
Jacobi polynomials. The solutions depend on spherical angles and are characterized by
the following quantum numbers:
• eigenvalues of Dirac operator λ which are nonzero either positive or negative integers
(λ = 0 is excluded by Lichnerovicz theorem);

• half-integer total angular momentum l = |λ| − 21 ;

• projection of angular momentum onto the polar axis m = −l, . . . l which is obviously
a half-integer.
The set of solutions corresponding to highest weight l realizes a 2l + 1 dimensional rep-
resentation of SU(2) group with matrix elements being given by the usual formulae for
half-integer spin. The normalization coefficients are chosen so that the properties of our
solutions with respect to complex conjugation and time reversal are in accord with those
of spherical functions.
The eigenspinors of spherical Dirac operator Υ differ from the ordinary spherical
spinors Ω. These two types of spinors diagonalize the different sets of operators. There-
fore Ω-spinors are listed by another set of quantum numbers. It includes: half-integer
total angular momentum j; integer orbital momentum l = j ± 21 ; half-integer projection
m of angular momentum onto the polar axis.
The second source of difference is the curvature of S 2 . The functions Υ diagonalize
the Dirac operator on the Riemann sphere which is a genuinely curved manifold. There-
fore the spin connection is inevitably present in the equation. On the other hand the
conventional spherical spinors are found by separating the angular variables in the flat

14
three-dimensional space R3 . Obviously the resulting equations contain no spin connection
unless one inserts it by hand.
As a consequence the Υ spherical spinors must be better suited for problems formulated
on the Riemann sphere (see Introduction for examples). We hope that the explicit and
detailed presentation of properties of Υ-spinors will help to use them for practical needs.
In conclusion I would like to thank Maximillian Kreuzer for the kind hospitality and
discussions during the preparation of the pilot version of the paper. I’m particularly
grateful to Joseph C. Varilly for introducing me to the history of the subject and rendering
me excerpts from his own work.
The work was done with partial support from RFBR grants 00–02–17808 and 00–15–
96786.

A Properties of Jacobi polynomials

Jacobi polynomials Pn(α, β) (x) (α, β > −1, n being the order) are classical orthogonal
polynomials that satisfy the equation of hypergeometric type [13, 14]

1 d
σ(x) y ′′ + τ (x) y ′ + λn y = [σ(x) ρ(x) y ′] + λn y = 0, (A.1)
ρ(x) dx

with the coefficient functions:

ρ(α, β) (x) = (1 − x)α (1 + x)β ; σ(x) = 1 − x2 ; τ (x) = β − α − (α + β + 2)x. (A.2)

The constant λn is equal to

λn = n(n + α + β + 1). (A.3)
The explicit form of Jacobi polynomials is given by the differential and integral Rodrigues’
formulas:
(−1)n 1 dn n
Pn(α, β) (x) = [σ (x) ρ(x)]; (A.4a)
2n n! ρ(x) dxn
(−1)n 1 n! σ n (z) ρ(z)
I
Pn(α, β) (x) = dz. (A.4b)
2n n! ρ(x) 2πi (z − x)n+1

The contour of complex integration in the second equation must encircle the point x.
Jacobi polynomials with given α and β are orthogonal on the interval [−1, 1] with the
weight ρ(α, β) (x)
1 2α+β+1 Γ(n + α + 1) Γ(n + β + 1)
Z
Pm(α, β) (x)Pn(α, β) (x) ρ(α, β) (x) dx = δmn . (A.5)
−1 n! (2n + α + β + 1) Γ(n + α + β + 1)

Jacobi polynomials form on this interval a complete set and any square integrable function
can be expanded in terms of Pn(α, β) .
Adjacent polynomials are related to each other by the differentiation formula:
d (α, β) 1 (α+1, β+1)
Pn (x) = (n + α + β + 1) Pn−1 (x). (A.6)
dx 2

15
 In the paper we met Jacobi polynomials with α = β ± 1 and used their special
properties. The first two identities help to relate the upper and lower components of
spinor spherical functions (25). Let us prove that for k ≥ 1
!
k d n + k (k−1, k)
− Pn(k, k−1) (x) = P (x); (A.7a)
1 − x dx 1−x n
!
k d n + k (k, k−1)
+ Pn(k−1, k) (x) = P (x). (A.7b)
1 − x dx 1+x n
 n
1
The proof is based on the integral Rodrigues’ formula (A.4b). Let C = 2πi − 12 be the
numerical factor common to the both sides of the equations. Then after differentiation
we get:
C(n + 1) (1 − z 2 )n+k dz
!
k d n + k (k, k−1)
I
− Pn(k, k−1) = P − ; (A.8a)
1 − x dx 1+x n (1 − x2 )k (z − x)n+2
C(n + 1) I (1 − z 2 )n+k dz
!
k d n + k (k−1, k)
+ Pn(k−1, k) = P + . (A.8b)
1 + x dx 1−x n (1 − x2 )k (z − x)n+2
Integrating the rightmost terms by parts opens the way to the desired result (A.7).
The second pair of relations was used in deducing representations (41, 42). They are
valid only for integer α and β and for convenience we shall use the notation of Section 2.3.
(We remind that both l± = l ± 21 ≥ 0 and m± = m ± 12 are integers.)
d l−m + −
l l
dl+m l− l+ m− (l + m)! dxl−m (1 − x) (1 + x)
(1 − x) (1 + x) = (−1) ; (A.9a)
dxl+m (l − m)! (1 − x)m+ (1 + x)m−
d l−m − +
l l
dl+m l+ l− m+ (l + m)! dxl−m (1 − x) (1 + x)
(1 − x) (1 + x) = (−1) . (A.9b)
dxl+m (l − m)! (1 − x)m− (1 + x)m+
For example let us prove (A.9a) for m > 0. Note that relation (A.9b) extends its validity
to negative m and v. v. The l + m-th derivative of the product is
l+m
dl+m X l + m dk l+m−k
!
− + l− d l+
A = l+m (1 − x)l (1 + x)l = (1 − x) (1 + x) . (A.10)
dx k=0 k dxk dxl+m−k
− +
Insofar as l+ , l− are integers the derivatives of (1 − x)l and (1 + x)l of orders higher
than l− and l+ vanish. Thus only the terms with m− ≤ k ≤ l− survive. Introducing a
new summation index j = k − m− we obtain after expanding the derivatives
l−m −
X (−1)j+m (l + m)! l− !(1 − x)l−m−j l+ !(1 + x)j
A= . (A.11)
j=0 (m− + j)!(l+ − j)! (l − m − j)! j!
Observing that
dj + dl−m−j −
(−1)j l+ ! l−m−j dxj
(1 − x)l l− ! dxl−m−j
(1 + x)l
(1 − x) = and (1 + x)j = ,
(l+ − j)! (1 − x)m+ −
(m + j)! (1 + x)m−
(A.12)
we immediately find
− l−m
(−1)m X l − m dj l−m−j
!
(l + m)! l+ d −
A= m+ m− j
(1 − x) l−m−j
(1 + x)l . (A.13)
(l − m)! (1 − x) (1 + x) j=0 j dx dx
This is nothing but the RHS of equation (A.9a). The proof of (A.9b) may be carried out
by simply reversing the sign x → −x.

16
B Spherical functions
Here we cite for reference the basics of spherical functions and Legendre polynomials.
The associated Legendre polynomials Plm (x) with l, m being nonnegative integers are
solutions to the second order ordinary differential equation [13, 14]
d d m2
(1 − x2 ) y − y + l(l + 1) y = 0. (B.1)
dx dx 1 − x2
The two convenient representations of functions Plm (x) are:
(−1)l 2 m dl+m
Plm (x) = (1 − x ) 2 (1 − x2 )l (B.2a)
2l l! dxl+m
(−1)l−m (l + m)! 2 −m dl−m
Plm (x) = l
(1 − x ) 2
l−m
(1 − x2 )l . (B.2b)
2 l! (l − m)! dx
They may be also expressed in terms of the Jacobi and Legendre polynomials,
(l + m)! 2 m (m, m) 2 m dm
Plm (x) = (1−x ) 2 P
l−m (x), and P l
m
(x) = (1−x ) 2 Pl (x). (B.3)
2m l! dxm
From the orthogonality of Jacobi polynomials (A.5) it follows that associated Legendre
polynomials with the same value of m are orthogonal on the interval [−1, 1] with the unit
weight:
Z 1
(l + m)! 2
dx Pkm (x) Plm(x) = δkl . (B.4)
−1 (l − m)! 2l + 1
Spherical functions Yl m (x, φ) are listed by two integers, namely, by the value of the
angular momentum l ≥ 0 and its projection onto the polar axis −l ≤ m ≤ l. The functions
Yl m (x, φ) are normalized eigenfunctions of the scalar (spin-zero) Laplace operator on the
sphere:
∂2
!
1 ∂ ∂ 1
− sin θ − Yl m (cos θ, φ) = l(l + 1) Yl m (cos θ, φ). (B.5)
sin θ ∂θ ∂θ sin2 θ ∂φ2
They may be expressed in terms of the associated Legendre polynomials:
v
eimφ u
t 2l + 1 (l − |m|)! P |m| (x).
u
m+|m|
Yl m (x, φ) = (−1) 2 il (B.6)
2π 2 (l + |m|)! l
With the help of identities (B.2) the spherical functions may be written as
v
l l imφ l−m
(−1) i e u 2l + 1 (l + m)! m d
u
Yl m (x, φ) = t (1 − x2 )− 2 l−m (1 − x2 )l . (B.7a)
2l l! 2π 2 (l − m)! dx
v
l−m l imφ l+m
(−1) i e u 2l + 1 (l − m)! m d
u
Yl m (x, φ) = l
t (1 − x2 ) 2 l+m (1 − x2 )l . (B.7b)
2 l! 2π 2 (l + m)! dx
The functions Ylm (cos θ, φ) form on the sphere the orthonormal functional basis:
Z π Z 2π
sin θ dθ dφ Yl m (cos θ, φ) Yl′ m′ (cos θ, φ) = δl l′ δm m′ . (B.8)
0 0

Inasmuch as the angular part of 3-dimensional Laplace operator (B.5) coincides with
the square of angular momentum L̂2C in coordinate representation spherical functions Yl m
are the basis vectors of 2l + 1 dimensional representation of rotation group O(3).

17
References
[1] E. T. Newman and R. Penrose, J. Math. Phys. 7 (1966) 863.

[2] J. N. Goldberg, A. J. Macfarlane, E. T. Newman, F. Rohrlich and E. C. G. Sudarshan,

J. Math. Phys. 8 (1967) 2155.

[3] T. Dray, J. Math. Phys. 26 (1985) 1030.

[4] R. Camporesi, A. Higuchi, J. Geom. Phys. 20 (1996) 1.

[5] J. M. Gracia-Bondı́a, J. C. Várilly, H. Figueroa, Elements of Noncommutative Ge-

ometry, Birkhäuser Advanced Texts, Birkhäuser, Boston, (2001).

[6] J. C. Várilly, arXiv: physics/9709045.

[7] M. F. Atiah, V. K. Patodi, I. M. Singer, Math. Proc. Camb. Phil. Soc. 77 (1975) 43.

[8] T. Eguchi, P. B. Gilkey, A. J. Hanson, Phys. Rep. C 66 (1980) 213.

[9] J. Gonzalez, F. Guinea, F. A. H. Vozmediano, Nucl. Phys. B 406 (1993) 771.

[10] A. A. Abrikosov, Int. Journ. of Mod. Phys. A 17 (2002) 885.

[11] A. I. Akhiezer, V. B. Berestetsky, Quantum Electrodynamics, Consultants Bureau,

New York, (1957).

[12] N. Berline, E. Getzler, M. Vergne, Heat Kernels and Dirac Operators, Springer,
Berlin (1992).

[13] A. F. Nikiforov, V. B. Uvarov, Special functions of mathemaical physics, Birkhäuser,

Boston, (1988).

[14] H. Bateman, A. Erdélyi Higher transcedental functions, R. E. Krieger, Malabar, Fla,

(1981).

[15] L. D. Landau, E. M. Lifshitz, Quantum Mechanics, Pergamon Press, Oxford,

New York, (1977).

18