Professional Documents
Culture Documents
All You Need To Know About The Dirac Equation
All You Need To Know About The Dirac Equation
All You Need To Know About The Dirac Equation
net/publication/233128414
CITATIONS READS
8 251
1 author:
P. Weinberger
New York University
449 PUBLICATIONS 8,210 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by P. Weinberger on 17 May 2014.
Fo
rP
ee
Author:
Note: The following files were submitted by the author for peer review, but cannot be converted
to PDF. You must view these files (e.g. movies) online.
Dirac-3.tex
ly
http://mc.manuscriptcentral.com/pm-pml
Page 1 of 14 Philosophical Magazine & Philosophical Magazine Letters
1
2
3
4
5
6
7
8
9
10 All you need to know about the Dirac equation
11
12
13 P. Weinberger
14 Center for Computational Nanoscience
15 Seilerstätte 10/22, A-1010 Vienna, Austria
16
Fo
21 A very brief introduction is given to all that is needed to appreciate the formal structure of the Dirac
22 equation and why — without destroying this structure — it cannot be reduced to a Paul-Schrödinger type
23 equation.
24
peer-00513915, version 1 - 1 Sep 2010
ee
25
26 1 Introduction
27
28 Since about 30 years I run around in Europe and in the USA and tell everybody in the field that if relativistic
rR
29 effects are important for a certain physical phenomenon, one has to use the Dirac equation rather than the
30 Pauli-Schrödinger equation. Nearly as long as this I am confronted with the canonical question, but how big are
31 the relativistic effects? Usually whenever this question was posed in the past it was accompanied by a pitiful
32 smile which clearly was intended to indicate that I obviously was addressing something completely irrelevant.
33 The perhaps less sarcastic reply consisted frequently in a friendly statement that I should not worry since all
ev
34 that is needed is to throw in some spin-orbit interaction. My standard reply, namely the counterquestion of
35 why should I neglect completely Einstein and his special theory of relativity, usually was brushed aside with
36 the comment that we are in solid state physics and not astrophysics. This kind of attitude was (is) perhaps
iew
37 best expressed by Richard Feynman in his Six Not-So-Easy Pieces [1], quoted below:
38 Newton’s Second Law, which we have expressed by the equation
39
40 F = d(mv)/dt ,
41
was stated with the tacit assumption that m is a constant, but we now know that this is not true, and that the
42
43 mass of a body increases with velocity. In Einstein’s corrected formula m has the value
On
44 m0
m= p ,
45 1 − v2 /c2
46
47 where the "rest mass" m0 represents the mass of a body that is not moving and c is the speed of light, which is
ly
http://mc.manuscriptcentral.com/pm-pml
Philosophical Magazine & Philosophical Magazine Letters Page 2 of 14
1
2
3
4
5
6
7 destroying the inherent algebraic structure that follows from the general postulates of quantum mechanics and
8 Einstein’s (special) theory of relativity. The arguments given are based on two completely different stand-points
9 of view, namely (a) application of group theory and (b) making use of the condition of relativistic covariance.
10 The basics of these arguments are not new, however, I hope it perhaps helps to repeat them yet another time.
11
12
13
2 Minkowski-space
14 Suppose the set of space-time vectors is given by
15
16 M = {xμ } , (1)
Fo
17
18 xμ ≡ (x0 , x1 , x2 , x3 ) = (x0 , xk ) = (ct, r) , (2)
19
20 μ = 0, 1, 2, 3 , k = 1, 2, 3 ,
rP
21 where x0 = ct is the time component and r = (x1 , x2 , x3 ) the space component of an arbitrary space-time vector
22 xμ . For any arbitrary pair of elements x, y ∈ M the scalar product in M is defined as follows
23
24 3
X 3
X
peer-00513915, version 1 - 1 Sep 2010
xμ y μ = x0 y 0 − xk y k
ee
25 (x, y) ≡ , (3)
26 μ=0 k=1
27
28 and in particular therefore the norm as
rR
29
k x k = (x, x) = x0 x0 − (r, r) . (4)
30
31 The metric in M is said to be pseudo-euclidean, since the metric tensor gμν is of the following form
32
33
⎛ ⎞
ev
1 0 0 0 µ ¶
34 ⎜ 0 −1 0 0 ⎟ 1 0
35 ⎜
gμν = ⎝ ⎟ = = g μν . (5)
0 0 −1 0 ⎠ 0 −13
36 0 0 0 −1
iew
37
38 The set M is sometimes also called Minkowski space.
39 In M a vector a is called contravariant (usually denoted by e.g. aμ ) if it ”transforms like a space-time
40 vector” xμ and covariant (usually denoted by, e.g., aμ ) if it transforms like ∂/∂xμ . The transformation of a
41 contravariant vector by means of the metric tensor gμν yields a covariant vector:
42
43 3
X
On
48 3
X
49 aμ = g μν aν ≡ g μν aν . (7)
50 v=0
51
52 It should be noted that in either case a0 = a0 . The implicit summation over repeated indices as indicated in
53 the last two equations is usually called the Einstein sum convention. The product of the metric tensor with
54 itself ½
X3
55 1, ν=μ
56 gμρ g ρν ≡ gμρ g ρν = δμ ν , δμ ν = , (8)
0 , ν 6= μ
ρ=0
57
58 is a unit matrix. A vector aμ is called a space-like vector if its norm aμ aμ < 0 and oppositely a time-like
59 vector if the norm is positive.
60
http://mc.manuscriptcentral.com/pm-pml
Page 3 of 14 Philosophical Magazine & Philosophical Magazine Letters
1
2
3
4
5
6
7 Defined in M the gradient can be written as a covariant vector ∂μ ,
8 ∂ ∂ ∂ ∂ ∂ ∂
9 ∂μ ≡ μ
= ( 0, 1, 2, 3) = ( , ∇) , (9)
∂x ∂x ∂x ∂x ∂x ∂ct
10
11 or, as a contravariant vector ∂ μ ,
12 ∂
∂μ = ( , −∇) , (10)
13 ∂ct
14 whereby
15 1 ∂2 1 ∂2
16 ∂μ ∂ μ ≡ ¤ = − ∇ · ∇ = −∆ (11)
c2 ∂t2 c2 ∂t2
Fo
21 Aμ = (φ, A) , (12)
22
23 such that the electric and magnetic field, E and H, respectively, are given by
24
peer-00513915, version 1 - 1 Sep 2010
∂A
ee
37 0
38 −Ez −Hy Hx 0
39
40 The gradient and the electromagnetic field vectors finally can be combined to yield the following four component
41 vector Dμ
42 ∂
Dμ = δμ + ieAμ = ( 0 + ieφ, −∇ + ieA) . (17)
43 ∂x
On
44
45 3 Poincaré and Lorentz transformations
46
47 Poincaré transformations are inhomogeneous (linear transformations that preserve the quadratic form xμ xμ ),
ly
http://mc.manuscriptcentral.com/pm-pml
Philosophical Magazine & Philosophical Magazine Letters Page 4 of 14
1
2
3
4
5
6
7 From the condition that the norm has to be left invariant and that the transformations are real, the properties
8 of the matrices Ωμν can be deduced, namely
9 Ω∗μυ = Ωμυ , (20)
10
11 Ωμυ Ωμλ = Ωνμ Ωλμ = δνλ , (21)
12 det | Ωμv | = ±1 . (22)
13
14 The set of operators (Ω/a) forms a group, the so-called Poincaré group,
15 P = {(Ω | a) | (Ω | a)(Ω0 | a0 ) = (ΩΩ0 | Ωa0 + a)} , (23)
16
Fo
17 in which the identity element ( | 0) has the following representation for the pure space-time point operation
18 µ ¶
1 0
19 D( ) = . (24)
0 13
20
rP
21 Similarly, the pure time-inversion operator (T | 0) and pure space inversion operator (J | 0) are defined by the
22 representations of their corresponding space-time point operations
23 µ ¶ µ ¶
24 −1 0 1 0
peer-00513915, version 1 - 1 Sep 2010
25
26 The set of operator (Ω | a) for which Ω00 > 0, i.e., which preserve the direction of time, forms a subgroup
27 P ⊂ P of index two:
28 00
P = {(Ω | a) | Ω ≥ 0} , (26)
rR
29
30 since the complement (P − P ) is defined by
31 (P − P ) = { (Ω | a) | (Ω | a) = (Ω | a)(T | 0)} . (27)
32
33 P is called orthochronous Poincaré group, which in turn has a subgroup of index two, namely the so-called
ev
34 proper orthochronous Poincaré group, P + , which is the set of time conserving transformations for which
35 det | Ωμv | = 1:
00
36 P + = {(Ω | a) | Ω ≥ 0, det | Ωμv | = 1} . (28)
iew
37
In terms of left cosets, the Poincaré group P ⊃ P ⊃ P + can therefore be written as
38
39 P = {P , (T | 0)P } , (29)
40
41 P = {P + , (J | 0)P + } . (30)
42 These three Poincaré groups contain as corresponding subgroups all those operations for which the translational
43 part is zero, i.e., a = 0:
On
48 group.
49 The subset of operators of the Poincaré group that corresponds to pure space translations only also forms a
50 subgroup, the so-called Euclidean group:
51
P ⊃ E = {(Ω | a) | ∀a0 = 0} . (33)
52
53 The corresponding subgroup of the Lorentz group is the familiar Rotation-Inversion group in R3 . It should
54 be appreciated that the above very brief characterization in terms of the left coset representatives (T | 0) and
55 (J | 0) most likely is the most compact way of viewing the general structure of these groups. Specific aspects
56 of group theory, namely in particular representation theory, will also be used in the following sections in order
57 to pin-point the relation between Paul spin matrices and Dirac matrices. For historical reasons it has to be
58 mentioned that of course Dirac in his 1928 paper [2] carefully "checked" the Lorentz invariance of his newly
59 found equation in an extensive separate section.
60
http://mc.manuscriptcentral.com/pm-pml
Page 5 of 14 Philosophical Magazine & Philosophical Magazine Letters
1
2
3
4
5
6
7 4 Dirac equation
8
For a single particle of charge e and mass m the relativistic Hamilton function is given by1 , c = 1,
9
p
10 H = eφ + (p − eA)2 + m2 , (34)
11
12 where φ is the scalar potential, A the vector potential and p the momentum. Assuming now in accordance with
13 the postulates of quantum mechanics that the probability density ρ = ψ ∗ ψ is positive definite then it follows
14 immediately that the corresponding Hamilton operator, Ĥ has to be Hermitian, since:
15 Z Z ³ ´
16 ∂ρ 3 i
d x=− ψ ∗ Ĥ ψ − (Ĥ ψ)∗ ψ d3 x = 0 . (35)
Fo
17 ∂t
|{z} ~
18 linear
19
For the sake of simplicity in the following discussion only the Hamilton function in the absence of a field
20
rP
shall be considered: p
21
H = p2 + m2 . (36)
22
23 Since the lhs of (35) is linear in ∂/∂t ≡ ∂/∂x0 , see in particular Eq. (9), this implies that also Ĥ on the rhs
24 of (35) has to be linear with respect to ∂/∂xk , k = 1, 3, i.e., with respect to components of the momentum
peer-00513915, version 1 - 1 Sep 2010
ee
25 operator p̂. This condition is usually called the condition of relativistic covariance.
26 If one replaces according to the correspondence principle E → i∂/∂t and p → −i∇, i.e.,
27 ³p ´
28 i∂
Ĥ ψ = ψ= p̂2 + m2 ψ , (37)
rR
29 ∂t
|{z} | {z }
30 linear has to b e linear
31
32 one immediately can see that the condition of linearity cannot be fulfilled in a straightforward manner, since
33 the square root is not a linear operator. As is perhaps less known the Dirac problem [2, 3], but also the problem
ev
34 of Pauli’s spin theory [4], can be viewed in terms of a special polynomial algebra [13].
35
36 4.1 Polynomial algebras
iew
37
38 Let P2 (x) be a second order polynomial of the following form
39 X X
P2 (x) = a21 xi xj + a22 x2j , i, j = 1, 2, . . . , m , (38)
40
i6=j j
41
42 where the aij are elements of a symmetric matrix. Consider further that the linear form
43
On
m
X
44
45 L(x) = αj xj (39)
j=1
46
47 satisfies the condition
ly
http://mc.manuscriptcentral.com/pm-pml
Philosophical Magazine & Philosophical Magazine Letters Page 6 of 14
1
2
3
4
5
6
7 the so-called Grassmann algebra2 and
8 a21 = 0 , a22 = −1 → [αi , αj ]+ = 2δij , (44)
9
10 the so-called Clifford algebra. Comparing now Eq. (37) with Eq. (36) one can see that exactly the case of
11 the Clifford algebra is needed in tackling the problem of the linearization of the square root:
12 v
13 uX
u m 2 Xm
14 t( pj ) = αj pj . (45)
15 j=1 j=1
| {z } | {z }
16
Fo
P2 (p) L(p)
17
18 In the following first the case for m = 2 and 3 (Pauli spin theory) is discussed by considering the smallest groups
19 with Clifford algebraic structure and only then in a similar way the Dirac problem (m = 4) is addressed.
20
rP
21
22 4.2 The Pauli groups
23 4.2.1 The Pauli group for m = 2
24
peer-00513915, version 1 - 1 Sep 2010
For m = 2 the smallest set of elements αi that shows group closure [10, 11, 12] is given by
ee
25
26 (m=2)
27 GP = {±I, ±α1 , ±α2 , ±α1 α2 } . (46)
28
This group is of order 8 and has 5 classes (Ci ), as easily can be found out by using Eq. (44)
rR
29
30 (m=2)
Table 1: class structure of GP
31
C1 : {I}
32
C2 : {−I}
33
ev
34 C3 : {±α1 }
35 C4 : {±α2 }
36 C5 : {±α1 α2 }
iew
37 (m=2)
38 There are therefore 5 irreducible representations (Γi , i = 1, 5) of dimensions ni such that
39 5
X
40 n2i = 8 . (47)
41 i=1
42
(m=2)
43 This implies that 4 irreducible representations (Γi , i = 1, . . . , 4) have to be one-dimensional and one two-
On
44 dimensional. Since one-dimensional representations are commutative, i.e., do not satisfy the conditions of a
45 Clifford algebra, only the two-dimensional representation (Γ5
(m=2)
) is of help. The matrices for this irreducible
46 representation are listed below:
47 µ ¶ µ ¶
ly
48 (m=2) 1 0 (m=2) 0 1
49 Γ5 (±I) = ± , Γ5 (±α1 ) = ± ,
0 1 1 0
50 µ ¶ µ ¶
51 (m=2) 0 −i (m=2) i 0
Γ5 (±α2 ) = ± , Γ5 (±α1 α2 ) = ± . (48)
52 i 0 0 −i
53 (m=2)
54 Using this set of matrices it is easy to show that it indeed forms a representation of GP and that these
55 matrices are Clifford algebraic. For the case of m = 2 the problem of the linearization of the square root is
56 therefore solved: q µ ¶ µ ¶ µ ¶
2 2 1 0 0 1 0 −i
57 p̂1 + p̂2 = p̂1 + p̂2 . (49)
58 0 1 1 0 i 0
59 2 this is exactly the algebra of creation and annihilation operators for fermions.
60
http://mc.manuscriptcentral.com/pm-pml
Page 7 of 14 Philosophical Magazine & Philosophical Magazine Letters
1
2
3
4
5
6
7 4.2.2 The Pauli group for m = 3
8 For m = 3 the smallest set of elements σi forming a group is given by
9
(m=3)
10 GP = {±I, ±α1 , ±α2 , ±α3 , ±α1 α2 , ±α1 α3 , ±α2 α3 , ±α1 α2 α3 } . (50)
11
12 The order of this group is 16. It has 10 classes,
13 Table 2: class structure of GP
(m=3)
14
C1 : {I} C6 : {±α1 α2 }
15
C2 : {−I} C7 : {±α1 α3 }
16
Fo
C3 : {±α1 } C8 : {±α2 α3 }
17
C4 : {±α2 } C9 : {α1 α2 α3 }
18
19 C5 : {±α3 } C10 : {−α1 α2 α3 }
20 and therefore 10 irreducible representations,
rP
21
10
X
22
23 n2i = 16 , (51)
i=1
24
peer-00513915, version 1 - 1 Sep 2010
ee
25 of which 8 (Γi
(m=3)
, i = 1, 8) are one-dimensional and two (Γi
(m=3)
, i = 9, 10) are two-dimensional. Again only
26 the two-dimensional irreducible representations are Clifford algebraic.
27 For α1 and α2 one can use the same matrix representatives as in the m = 2 case,
28
rR
34 (m=3)
The second two-dimensional irreducible representation (Γ10 ) is by the way the complex conjugate represen-
35 (m=3)
36 tation of Γ9 . It is rather easy to proof that these two irreducible representations are indeed non equivalent.
For the m = 3 case the problem of the linearization of the square root reduces therefore to the following
iew
37
38 matrix equation:
39 q
(m=3) (m=3) (m=3) (m=3)
40 p̂21 + p̂22 + p̂23 Γ9 (I) = p̂1 Γ9 (α1 ) + p̂2 Γ9 (α2 ) + p̂3 Γ9 (α3 ) . (54)
41
The matrices µ ¶ µ ¶
42 (m=3) 0 1 (m=3) 0 −i
43 Γ9 (α1 ) ≡ σ1 = , Γ9 (α2 ) ≡ σ2 = ,
On
1 0 i 0
44 µ ¶
45 (m=3) 1 0
Γ9 (α3 ) ≡ σ3 = , (55)
46 0 −1
47 are nothing but the famous Pauli spin matrices, usually - as indicated in the last equation - denoted simply
ly
48 (m=2)
by σ1 , σ2 and σ3 . For m = 2, 3 the corresponding groups, GP and GP
(m=3)
, are called Pauli groups (as
49 indicated by the index P ).
50
51
52 4.3 The Dirac group
53 For m = 4 the following subset of the Clifford algebra forms the smallest group
54
(m=4)
55 GD = {±I, ±αi (i ≤ 4), ±αi αj (i < j), ±αi αj αk (i < j < k) ,
56
57 ±α1 α2 α3 α4 ≡ ±α5 } , (56)
58 where ”traditionally” the elements αi are usually denoted in the literature also by γμ . The order of this group
59 is 32. It has 17 classes,
60
http://mc.manuscriptcentral.com/pm-pml
Philosophical Magazine & Philosophical Magazine Letters Page 8 of 14
1
2
3
4
5
6 (m=4)
7 Table 3: class structure of GD
8 C1 : {I}
9 C2 : {−I}
10 C3−6 : {±αi | i ≤ 4}
11 C7−12 : {±αi αj | i < j ≤ 4}
12 C13−16 : {±αi αj αk | i < j < k ≤ 4}
13 C17 : {±α1 α2 α3 α4 }
14
15 and therefore 17 irreducible representations.
(m=4)
16 As easily can be checked in analogy to Eq. (47) 16 of these irreducible representations (Γi , i = 1, . . . , 16)
Fo
(m=4)
17 are one-dimensional and one is four-dimensional (Γ17 ). Again only the matrices of the four-dimensional
18 irreducible representation satisfy the conditions of the Clifford algebra.
19 The following matrices
20 µ ¶
rP
21 (m=4) 0 σi
Γ17 (αi ) ≡ αi ≡ γi = , i = 1, 2, 3 , (57)
22 σi 0
23 µ ¶
24 I2 0
peer-00513915, version 1 - 1 Sep 2010
(m=4)
Γ17 (α4 ) ≡ β ≡ γ4 = , (58)
ee
25 0 −I 2
26
where the σi are the Pauli spin matrices and I 2 is a two-dimensional unit matrix, are then irreducible rep-
27 (m=4)
28 resentatives of the elements αi ∈ GD . These particular representatives, usually as indicated above simply
rR
(m=4)
29 by αi and β, are the famous Dirac matrices, GD is called the Dirac group.
30
31 4.4 Relations between the Dirac group & the Pauli groups
32
33 4.4.1 The subgroup structure
ev
34 (m=4)
The Dirac group GD contains the Pauli groups as subgroups,
35
36 (m=2) (m=3) (m=4)
GP ⊂ GP ⊂ GD , (59)
iew
37
38 (m=2) (m=3) (m=4)
39 whereby GP is a normal subgroup in GP and GD . This implies that in a coset decomposition of
(m=4) (m=2)
40 GD in terms of GP ,
41
(m=4) (m=2) (m=2) (m=2) (m=2)
42 GD = {IGP , α3 GP , α4 GP , α3 α4 GP } , (60)
43
On
http://mc.manuscriptcentral.com/pm-pml
Page 9 of 14 Philosophical Magazine & Philosophical Magazine Letters
1
2
3
4
5
6
7 4.4.2 Subduced representations
8 The set of matrices,
9 (m=4)
Γ17 (GP
(m=2)
) ≡ {Γ17
(m=4) (m=2)
(α), ∀α ∈ GP } ,
10
11 and
(m=4) (m=3) (m=4) (m=3)
12 Γ17 (GP ) ≡ {Γ17 (α), ∀α ∈ GP } ,
13 (m=2) (m=3)
of course also forms a representation for GP
and GP ,
respectively, which, however, is reducible. Such
14
representations are called subduced representations. Reducing these two representations (for example by
15
16 means of the orthogonality relation for characters), one finds the following decompositions into irreducible
Fo
17 representations:
(m=4) (m=2) (m=2) (m=2)
18 Γ17 (GP ) = 2Γ5 (GP ) , (64)
19 and
20 (m=4) (m=3) (m=3) (m=3) (m=3) (m=3)
Γ17 (GP ) = Γ9 (GP )+Γ (GP ) ´
rP
21 ³10 ∗ . (65)
(m=3) (m=3) (m=3) (m=3)
22 = Γ9 (GP ) + Γ9 (GP )
23
24
peer-00513915, version 1 - 1 Sep 2010
(m=4) (m=4)
Since the irreducible representation Γ17 of GD always subduces only the group of p the Pauli spin
ee
25
26 matrices (and their complex conjugates), there is no way to linearize properly the square root p2 + m2 for a
27 four-component momentum in terms of 2×2 matrices only! In other words: there is no other ”truly” relativistic
28 description but the one using the Dirac matrices:.
rR
29 ⎛ ⎞1/2
30
31 ⎜ ⎟
I 4 ⎝p̂21 + p̂22 + p̂23 + p̂24 ⎠ = α1 p̂1 + α2 p̂2 + α3 p̂3 + β p̂4 (66)
32 | {z } |{z}
p̂2 m2
33
ev
37
(m=4)
38 irreducible representation, namely Γ17 .
39
40 4.4.3 Fundamental theorem of Dirac matrices
41
42 One can summarize the properties of these three groups very compactly in the below short table:
43
On
44
Table 4: summary of group properties for m ≤ 4
45
46 m Group- # of # of one- # of two- # of four-
47 order classes dim. irreps dim. irreps dim. irreps
2m+1 m2 + 1 2m
ly
48
49 2 8 5 4 1 0
50 3 16 10 8 2 0
51 4 32 17 16 0 1
52
53
The so-called fundamental theorem of Dirac matrices, namely that a necessary and sufficient condition
54
for a set of 4 matrices γi0 to be Dirac matrices, i.e., to be irreducible and Clifford algebraic, is that they have to
55
be obtained via a similarity transformation W from the matrices in (57,58):
56
57 γi0 = W −1 γi W , i = 1, 4 , (67)
58
59 is in the context of the Dirac group nothing but Schur’s lemma for irreducible representations.
60
http://mc.manuscriptcentral.com/pm-pml
Philosophical Magazine & Philosophical Magazine Letters Page 10 of 14
1
2
3
4
5
6
7 5 Dirac’s original derivation
8
Of course Dirac in his famous paper [2] did not use group theory, nor did he realize that his matrices were
9
Clifford algebraic. He found "his" matrices by trial and error, knowing very well, however, that if quantum
10
mechanics and Einstein’s special theory of relativity were to be compatible at all then only on the condition
11
that the postulates of quantum mechanics had to be fulfilled rigorously.
12
In an appendix of his first paper he introduced also the so-called elimination method in order to arrive
13
14 at expressions that at his time were very much en vogue, namely the Darwin term, the mass-velocity term
15 and the spin-orbit term, the last of which causing so much confusion in the following decades. For matters of
16 completeness his derivation of these terms is reformulated in the following section [6].
Fo
17
18
19
6 The Pauli-Schrödinger equation
20 Considering for matters of simplicity a Dirac-type Hamiltonian for a non-magnetic system, in atomic units
rP
21 (~ = m = 1),
22 H = cα · p + (β − I4 ) c2 + V I4 , (68)
23
24 where c is the speed of light. In making use of the bi-spinor property of the wavefunction3 ,
peer-00513915, version 1 - 1 Sep 2010
ee
25 µ ¶
26 |φi
|ψi = ,
27 |χi
28
rR
34 (70)
35 ¡ ¢
36 cσ · p |φi + V − 2c2 |χi = |χi .
iew
37
38 Clearly, the spinor |χi can now be expressed in terms of |φi:
39
|χi = (1/2c) B −1 σ · p |φi , (71)
40
41 ¡ ¢
B = 1 + 1/2c2 ( − V ) , (72)
42
43 leading thus to only one equation for |φi:
On
48
49 For a central field, V (r) = V (|r|), the operator D in Eq. (74) has the same constants of motion [7, 8, 9] as the
50
corresponding Dirac Hamiltonian, namely the angular momentum operators J2 , Jz , and K = β (1 + σ · L):
51
52 3 because of the block-diagonal form of the Dirac matrix β
53
54
55
56
57
58
59
60
10
http://mc.manuscriptcentral.com/pm-pml
Page 11 of 14 Philosophical Magazine & Philosophical Magazine Letters
1
2
3
4
5
6
7
8
J2 |φi = j(j + 1) |φi , J2 |χi = j(j + 1) |χi ,
9
10 Jz |φi = μ |φi , Jz |χi = μ |χi ,
11 (1 + σ · L) |φi = κ |φi , (1 + σ · L) |χi = −κ |χi ,
12 1 3 5
13 j = , , , ... ,
2 2 2
14
−j ≤ μ ≤ j ,
15 ½
16 − − 1 ; j = + 1/2
Fo
κ= .
17 ; j = − 1/2
18
19 Their simultaneous eigenfunctions are the so-called spin spherical harmonics [7],
20 X
rP
1
21 |κμi ≡ |Qi = c( ; μ − s, s) | , μ − si Φ(s) ,
22 2
s=±1/2
23 h r̂ | , μ − si = Y ,μ−s (r̂) ,
24 µ ¶ µ ¶
peer-00513915, version 1 - 1 Sep 2010
1 1 1 0
ee
25 Φ( ) = , Φ(− ) = ,
26 2 0 2 1
µ ¶
27 1 (−1) +μ−1/2 1/2 j
28 c( ; μ − s, s) = ,
2 (2j + 1)1/2 m s μ
rR
29
30 where the Y m (r̂) are (complex) spherical harmonics, Φ(± 12 ) the so-called spin eigenfunctions [4] and the c( 12 ; μ−
31 s, s) the famous Clebsch-Gordan coefficients [5], which in turn are related to the Wigner 3j-coefficients [5]. It
32 is important to note that | , μ − si Φ(s) is a tensorial product of functions belonging to different spaces.
33
ev
Because of the constants of motion J2 , Jz , and K = β (1 + σ · L) Eq. (73) is separable with respect to the
34 radial and angular variables, i.e., an eigenfunction of D belonging to a particular eigenspace of these constants
35 of motions is then of the form
36 Rκ (r)
h r | φQ i = h r̂ | Qi , (75)
iew
37 r
38
where the radial amplitudes Rκ (r) are solutions of the following differential equation [2], [6]
39
40 ∙ µ ¶ ¸
1 d2 ( + 1)
41 − 2+ + V (r) − Rκ (r) =
2 dr r2
42
43 1 −2 dV (r) κ
On
B (r) Rκ (r)
44 4c2∙ dr r µ ¶¸
45 1 −1 d2 ( + 1)
+ 2 ( − V (r)) B (r) − 2 + Rκ (r)
46 4c dr r2
47 ∙ ¸
1 dV (r) d
+ 2 B −2 (r)
ly
48 Rκ (r) . (76)
4c dr dr
49
50 Equation (76) shows a remarkably “physical structure” , namely
51
52 1. For c = ∞ (non-relativistic limit) this equation is reduced to the well-known radial Schrödinger equation.
53
54 2. By approximating the elimination operator B in Eq. (72) by unity (B = 1) the so-called (radial) Pauli-
55 Schrödinger equation is obtained. The terms on the right-hand side of Eq. (76) are then in turn the
56 spin-orbit coupling, the mass velocity term, and the Darwin shift.
57
58 3. For B 6= 1 relativistic corrections in order higher than c−4 , enter, e.g., via the normalization of the
59 wavefunction.
60
11
http://mc.manuscriptcentral.com/pm-pml
Philosophical Magazine & Philosophical Magazine Letters Page 12 of 14
1
2
3
4
5
6
7 4. It should be noted that although all three terms on the right-hand side of Eq. (76) have a prefactor 1/4c2 ,
8 i.e., are of relativistic origin, the only one, however, which explicitly depends on a (relativistic) quantum
9 number, namely κ, is spin-orbit coupling.
10 5. It should be noted in particular that dV /dr, has the unpleasant property of being singular for r → 0.
11
12
13 7 Comparison to the "radial Dirac equation"
14
15 Clearly in the case of a central field also Eq. (69) can be separated using polar coordinates, i.e., using the
16 constants of motion J2 , Jz , and K = β (1 + σ · L),
Fo
17 µ ¶
18 gκ (r) h r̂ | κμi
h r | ψκμ i = , (77)
19 ifκ (r) h r̂ | −κμi
20
rP
21 Pκ (r) Qκ (r)
gκ (r) = , fκ (r) = , (78)
22 r cr
23 where the radial amplitudes Pκ (r) and Qκ (r) are solution of the following differential equation (in atomic
24
peer-00513915, version 1 - 1 Sep 2010
Rydberg units)
ee
25
26 dQκ (r) Qκ (r)
27 =κ − ( − V (r))Pκ (r) , (79)
dr r
28 dPκ (r) Pκ (r) − V (r)
rR
34 d2 ( + 1)
− 2+ + V (r) − Rκ (r) =
35 dr r2
36 1 d(V (r)) κ
Rκ (r)
iew
37 4c2 ∙ dr r µ
38 ¶¸
1 d2 ( + 1)
39 + 2 ( − V (r)) − 2 + Rκ (r)
4c dr r2
40 ∙ ¸
41 1 d(V (r)) d
+ 2 Rκ (r) . (81)
42 4c dr dr
43
On
44 As easily can be seen, Eq. (81) is a second order differential equation, while Eqs. (79) - (80) form a system of
45 coupled first order differential equations.
46 Remembering now that the key requirement for a proper inclusion of relativity into quantum mechanics is
47 the condition of relativistic covariance, see in particular Eqs. (9) - (10), namely the condition of linearity
ly
48 for a relativistic Hamiltonian, see Eq. (37), when using the correspondence principle (one of the postulates of
49 quantum mechanics), then one has to arrive immediately at the conclusion that the Pauli-Schrödinger equation
50 does not meet this requirement. This equation only partially satisfies the postulates of (relativistic)
51 quantum mechanics! It has, however, the big advantage that one immediately can prove that in the non-
52 relativistic limit (c → ∞) the Schrödinger equation is recovered, a fact which is less easy to be seen inspecting
53 the Dirac equation.
54 It is utterly important not to confuse the "large component" gκ (r) in Eq. (77) or for that matter h r | φQ i
55 in Eq. (75) with a corresponding solution of the Schrödinger equation: only in the limit of c → ∞ such a
56 relationship can be established.
57
58
59
60
12
http://mc.manuscriptcentral.com/pm-pml
Page 13 of 14 Philosophical Magazine & Philosophical Magazine Letters
1
2
3
4
5
6
7 8 A last remark
8
We all know that nowadays nearly all calculations in solid state physics are performed using Density Functional
9
Theory (DFT), in particular local DFT. For a magnetic system the Kohn-Sham-Dirac Hamiltonian is given by
10
11 H(r) = cα · p + (β − I4 ) c2 + V ef f (r)I4 + βΣz Bzef f (r) , (82)
12
13 where V ef f (r) is the effective potential, Bzef f (r) the effective exchange field and
14
µ ¶
15 σz 0
16 Σz = . (83)
0 σz
Fo
17
18 Because LDFT provides Bzef f (r) only with respect to a fictitious z-axis, in order to evaluate anisotropy energies
19 for example, it is necessary to "rotate" H
20
rP
ee
25 µ ¶
26 U (R) 0
S(R) = , (85)
27 0 det[±]U (R)
28
rR
29 where U (R) is a (unimodular) 2 × 2 matrix and det[±] = det[D(3) (R)] is the determinant of D(3) (R), the latter
30 one being the representation of R in R3 . Clearly enough by such a transformation not only βΣz Bzef f (r) is
31 transformed but also cα · p.
32 On the other hand considering a two-component formulation
33
ev
37
38 energy part is properly transformed (a special case of Lorentz group invariance), while in a two-component
39 formulation it is not. Of course the Pauli-Schrödinger and the Dirac Hamiltonian do have different spectra.
40
41
42
9 Summary
43
On
The formal deficiencies of the Pauli-Schrödinger equation discussed above were in the past always my main
44
arguments for insisting to use directly the Dirac equation and not some alternative two-component descriptions.
45
Clearly enough many more things can be said about the Dirac equation, see for example Refs. [5, 9].
46
Nowadays — as it seems — things have changed for very practical reasons as it is much easier to use the Dirac
47
equation in actual calculations then fiddling around with the spin-orbit term in the Pauli-Schrödinger equation.
ly
48
The theoretical description of anisotropic magnetic properties of magnetic nanostructures, even of domain walls
49
would not have been possible without this numerical advantage!
50
51 To my great satisfaction the use of the time-dependent Dirac equation in the presence of an external electro-
52 magnetic field and of the concept of the so-called polarization operator led very recently [14] to a quantum
53 mechanically correct identification of spin currents, spin-transfer, and spin-Hall effects, which in turn hopefully
54 will lead to a completely new stage in spintronics!
55 I very much would like to acknowledge the influence of my former teachers Per-Olov Löwdin (Uppsala,
56 together with Harold MacIntosh), Simon Altmann (Oxford) and Walter Kohn (S. Barbara). While Per-Olov
57 raised my interest in "formal structures", it was Simon, who completely indoctrinated me with group theory as
58 a formal tool to view "structures". And Walter tought me to keep the right distance to my own research.
59
60
13
http://mc.manuscriptcentral.com/pm-pml
Philosophical Magazine & Philosophical Magazine Letters Page 14 of 14
1
2
3
4
5
6
7 References
8
[1] R. P. Feynman, Six-Not-So-Easy Pieces, Perseus Books, Cambridge, Massachusetts, 1997.
9
10 [2] P. A. M. Dirac, Proc. Roy. Soc. A117, 610 (1928); Proc. Roy. Soc. A126, 360 (1930).
11
12 [3] P. A. M. Dirac, The Principles of Quantum Mechanics, Oxford University Press (first edition 1930), pa-
13 perback 1981.
14
15 [4] W. Pauli, Die allgemeinen Prinzipien der Wellenmechanik, Handbuch der Physik, Springer-Verlag, Berlin,
16 1933.
Fo
21 [7] E. M. Rose, Relativistic Electron Theory, John Wiley and Sons, 1961.
22
[8] J. D. Bjorken and S. D. Drell, Relativistic Quantum Mechanics, McGraw Hill Inc., 1964
23
24
peer-00513915, version 1 - 1 Sep 2010
[9] L. D. Landau and E. M. Lifschitz, Lehrbuch der Theoretischen Physik, Band IV, Relativistische Quanten-
ee
29
30 [11] S. L. Altmann, Induced Representations in Crystals and Molecules, Academic Press, 1977.
31 [12] S. L. Altmann, Rotation, Quaternions and Double Groups, Oxford University Press 1986.
32
33 [13] I. V. V. Raghavaryulua and N. B. Menon, J. Math. Phys. 11, 3055 (1970).
ev
34
35 [14] A. Vernes, B. L. Gyorffy, and P. Weinberger, Phys. Rev. B 76, 012408 (2007).
36
iew
37
38
39
40
41
42
43
On
44
45
46
47
ly
48
49
50
51
52
53
54
55
56
57
58
59
60
14
http://mc.manuscriptcentral.com/pm-pml