Matrix gauge fields and Noether's theorem
Citation for published version (APA):Graaf, de, J. (2014). Matrix gauge fields and Noether's theorem. (CASA-report; Vol. 1414). Technische Universiteit Eindhoven.
Document status and date: Published: 01/01/2014
Document Version:
Publisherβs PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
β’ A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
β’ The final author version and the galley proof are versions of the publication after peer review.
β’ The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. β’ Users may download and print one copy of any publication from the public portal for the purpose of private study or research. β’ You may not further distribute the material or use it for any profit-making activity or commercial gain
β’ You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the βTaverneβ license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
EINDHOVEN UNIVERSITY OF TECHNOLOGY Department of Mathematics and Computer Science
CASA-Report 14-14 May 2014
Matrix Gauge fields and Noetherβs theorem by
J. de Graaf
Centre for Analysis, Scientific computing and Applications Department of Mathematics and Computer Science
Eindhoven University of Technology P.O. Box 513
5600 MB Eindhoven, The Netherlands ISSN: 0926-4507
Matrix Gauge Fields
and
Noetherβs Theorem
J. de GRAAF
Eindhoven University of Technology, Mathematics, Casa Reports 14-14, May 2014
Preface and Summary
These notes are about systems of 1st and 2nd order (non-)linear partial differential equa-tions which are formed from a Lagrangian density LΟ : RN β C ,
Symbolically : x7β LΟ(x) = L(Ο(x) ; βΟ(x) ; x) ,
by means of the usual Euler-Lagrange variational rituals. The non subscripted L will denote the βproto-Lagrangianβ, which is a function of a finite number of variables:
L : CrΓcΓ CN rΓc
Γ RN
β C .
In this L one has to substitute matrix-valued functions Ο : RN β CrΓc
and βΟ : RN β CN rΓc for obtaining the Lagrangian density LΟ. In our considerations the role and the
special properties of the proto-Lagrangian L are crucial.
These notes have been triggered by physicistβs considerations: (1) on obtaining the βclassi-calβ, that is the βpre-quantizedβ, wave equations for matter fields from variational principles, (2) on conservation laws and (3) on βgauge field extensionsβ. For the humble mathematical anthropologist the rituals in physics textbooks have not much changed during the last four decades. Neither have they become much clearer. Compare e.g. [DM] and [W].
The underlying notes give special attention to the following
β’ In expressions (=βequationsβ) for Lagrange densities often both Ο and its hermitean transposed Οβ appear. Are they meant as independent variables or not? Mostly, from the context the suggestion arises that βvariationβ of Ο and βvariationβ of Οβ lead
to the same Euler-Lagrange equations. Why? Our remedy is doubling the matrix entries in the proto-Lagrangian and thereby making the Lagrangian density explicitly dependent on both Ο , Οβ and their derivatives: So for LΟ(x) we take expressions like
LΟ(x) = L (Ο(x) ; Ο(x)β ; βΟ(x) ; βΟ(x)β ; x). A suitable condition is then that the
Lagrangian functional
L[Ο] = Z
RN
LΟ(x) dx
only takes real values (Thm 2.4).
β’ For βfree gauge fieldsβ the situation is somewhat different. Now the dependent varia-bles, named AΒ΅, 1 β€ Β΅ β€ N , take their values in some fixed Lie-algebra g β CcΓc.
Although g mostly contains complex matrices it is a real vector space in interesting cases. (Note that u(1) = iR is a real vector space!). Therefore it needs a separate treatment.
β’ The traditional conservation laws for quantities like energy, momentum, moment of momentum, . . . , turn out to be based on External Infinitesimal Symmetries of the proto-Lagrangian. This means the existence of a couple of linear mappings
K : CrΓc β CrΓc, L : CN rΓc β CN rΓc, together with an affine mapping
x 7β βsa + esAx , such that for all matrices P β CrΓc, Q β CN rΓc and x β RN ,
L(esKP ; esLQ ; βsa + esAx) = L(P , Q ; x) + O(s2) .
Of course the presented conservation laws are just special cases of Noetherβs Theorem. β’ For the construction of gauge theories one needs, in physicistβs terminology, a βglo-bal symmetry of the Lagrangianβ. To achieve this, an Internal Symmetry of the proto-Lagrangian L is required here: For some fixed Lie-group G β CcΓc, the
proto-Lagrangian satisfies
L(PU ; QU ; x) = L(P ; Q ; x) , for all P β CrΓc, Q β CN rΓc , U β G , x β RN. Roughly speaking, a gauge theory for a Lagrangian based system of PDEβs is some kind of symmetry preserving extension of the original Lagrangian density with new (dependent) βfieldβ-variables x 7β A(x) = [A1(x), . . . , AN(x)] on RN added, such that
the original βquantitiesβ Ο become subjected to the βgauge fieldsβ A and viceversa. Since about a century, Weyl 1918, it is well known that, given the existence of some βglobal symmetry groupβ G of L, an extension of type
LΟ,A(x) = L(Ο ; βΟ + Ο Β·A ; x) + G(A ; βA ; x) ,
is often possible. This extension has to exhibit what physicists call, a βLocal Symme-tryβ : The Lagrangian density remains unaltered if in LΟ,Athe quantities Ο and A are,
which is the group of smooth maps RN β G. The added βgauge fieldsβ A have to take their values in the Lie Algebra g of the symmetry group G.
Summarizing, βlocally symmetricβ means, symbolically,
L ΟU ; β(ΟU )+(ΟU )Β·(A C U ) ; x)+G(A C U ; β(A C U ) ; x =
= L Ο ; βΟ + Ο Β·A ; x + G A ; βA ; x , for all U β Gloc.
β’ The considerations in the underlying notes not only include the standard hyperbolic evolution equations of pre-quantized fields. Wide classes of parabolic/elliptic systems turn out to have gauge extensions as well. Note the subtle extra condition (5.14) in Thm 5.5 which is, besides internal symmetry of the proto-Lagrangian, necessary for gauge extensions. Its necessity lies in the fact that one has to reconcile the complex vector space, in which the Ο take their values, with the real vector space g, the Lie-Algebra. In the standard preludes to quantum field the requirement (5.14) is never discussed, but manifestly met with.
β’ These notes do not contain functional analysis or differential geometry. The reader will find only bare elementary considerations on matrix-valued functions: The co-lumns of the x 7β Ο(x) β CrΓc might describe the βpre-quantized wave functionsβ of
individual elementary particles, whereas the βcomponentsβ of x 7β A(x) β gN, with
g β CcΓc, might represent the pre-quantized gauge fields. For an elementary and
very readable account on the differential geometrical aspects, see the contributions 3-4 in [JP].
CONTENTS
1. Foretaste: Some gauge-type calculations p.3 2. Stationary points of complex-valued functionals p.6 3. Free Gauge Fields p.13
4. Noether Fluxes p.19
5. Static/Dynamic Gauge Extensions of Lagrangians p.26 A. Addendum on Free Gauge Fields p.34 B. Electromagnetism p.35
1
Foretaste: Some gauge-type calculations
For functions Ξ¨ : RN β CrΓc we consider, by way of example, the PDE
ΞΒ΅ β¡Ψ + Ξ¨AΒ΅ + MΞ¨ = f, (1.1)
with prescribed matrix valued coefficients
ΞΒ΅: RN β CrΓr, AΒ΅: RN β CcΓc, 1 β€ Β΅ β€ N, M : RN β CrΓr,
and prescribed right hand side f : RN β CrΓc. All considered functions are supposed to
be sufficiently smooth. The summation convention for upper and lower indices applies. In physics each column of Ξ¨ may represent a βclassical-particle waveβ. The AΒ΅ may then
represent βgauge fieldsβ. Theorem 1.1
Let U , V : RN β CcΓc and suppose them invertible with Uβ1, Vβ1
: RN β CcΓc.
The function ΛΞ¨ = Ξ¨U : RN β CrΓk, with Ξ¨ any solution of (1.1) is a solution of
ΞΒ΅ β¡Ψ + ΛΛ Ξ¨ ΛAΒ΅ + M ΛΞ¨ = Λf , (1.2)
if and only if we take the new coefficients ΛAΒ΅ = Uβ1AΒ΅U β Uβ1(βΒ΅U ) and Λf = f U .
In addition we have AΛΛΒ΅= (U V)β1AΒ΅(U V) β (U V)β1(βΒ΅(U V)) = Vβ1AΛΒ΅V β Vβ1(βΒ΅V).
Proof: Multiply (1.1) from the right by U and rearrange. In the next Theorem a βtransformation propertyβ for matrix valued functions is derived. Theorem 1.2
Let AΒ΅: RN β CcΓc and ΛAΒ΅ = Uβ1AΒ΅U β Uβ1(βΒ΅U ). Define
F¡ν = βΒ΅AΞ½ β βΞ½AΒ΅β AΒ΅AΞ½β AΞ½AΒ΅. (1.3)
Then
Λ
F¡ν = βΒ΅AΛΞ½β βΞ½AΛΒ΅β AΛΒ΅AΛΞ½ β ΛAΞ½AΛΒ΅ = Uβ1F¡νU . (1.4)
Proof: First note that from βΒ΅(Uβ1U ) = βΒ΅I = 0 it follows that βΒ΅(Uβ1) = βUβ1(βΒ΅U )Uβ1.
Calculate βΒ΅AΛΞ½ = βΒ΅ Uβ1AΞ½U β Uβ1(βΞ½U ) = = Uβ1(βΒ΅AΞ½)U β Uβ1(βΒ΅U )Uβ1AΞ½U + Uβ1AΞ½(βΒ΅U ) + Uβ1(βΒ΅U )Uβ1(βΞ½U ) β Uβ1(βΒ΅βΞ½U ). and Λ AΒ΅AΛΞ½ =Uβ1AΒ΅U β Uβ1(βΒ΅U ) Uβ1AΞ½U β Uβ1(βΞ½U ) = = Uβ1 AΒ΅AΞ½Uβ Uβ1AΒ΅U Uβ1(βΞ½U )β Uβ1(βΒ΅U ) Uβ1AΞ½U+ Uβ1(βΒ΅U ) Uβ1(βΞ½U ).
Interchange the indices for two more terms and add according to (1.4). All rubbish terms
cancel out.
Condition 1.3
K : RN β CrΓr, is such that
i: KΞΒ΅ = (KΞΒ΅)β , ii: β
Β΅(KΞΒ΅) = 0, iii: KM + Mβ Kβ = 0.
Here, the dagger β denotes βHermitean transpositionβ.
Note that in the important special case that ΞΒ΅ = (ΞΒ΅)β , ΞΒ΅ is constant and M = βMβ ,
the condition is satisfied by K = I, the identity matrix. In the case of the Dirac equation one could take K = Ξ0. Cf. [M], Messiah II pp. 890-899. 1
Theorem 1.4
Let K : RN β CrΓr satisfy Condition 1.3.
Fix some J β CcΓc.
Let AΒ΅: RN β CcΓc satisfy Aβ Β΅J + J AΒ΅= 0, 1 β€ Β΅ β€ N .
Let U : RN β CcΓc satisfy Uβ
(x)J U (x) = J , x β RN.
a. For any solution Ξ¨ of (1.1) with f = 0, there is the conservation law
N
X
Β΅=1
βΒ΅Tr Jβ1[Ξ¨β KΞ¡Ψ] = 0 . (1.5)
b. This conservation law is a gauge invariant local conservation law. That means Tr Jβ1[ ΛΞ¨β KΞ¡Ψ] = Tr JΛ β1[Ξ¨β KΞ¡Ψ] , 1 β€ Β΅ β€ N . Proof
a. Take f = 0 in (1.1)and multiply from the left with Ξ¨β K:
Ξ¨β KΞΒ΅ β¡Ψ + Ξ¨β KΞ¡ΨAΒ΅+ Ξ¨β KM Ξ¨ = 0. (1.6)
The Hermitean transpose reads β¡Ψ
β
(KΞΒ΅)β Ξ¨ + Aβ ¡Ψβ (KΞΒ΅)β Ξ¨ + Ξ¨β Mβ Kβ Ξ¨ = 0. (1.7) Multiply (1.6) from the right with Jβ1 and (1.7) from the left with Jβ1. Add those two identities and take the trace. Use Condition 1.3 and the properties Tr(AB) = Tr(BA), Tr(A + B) = Tr(A) + Tr(B) and βΒ΅Tr(A) = Tr(βΒ΅A). The sum of the 1st terms of (1.6),
(1.7) result in TrJβ1Ξ¨β
(KΞΒ΅)β¡Ψ + (β¡Ψ)β (KΞΒ΅)β Ξ¨ =
= βΒ΅TrJβ1Ξ¨β (KΞΒ΅)Ξ¨ β TrJβ1Ξ¨β βΒ΅(KΞΒ΅)Ξ¨ = βΒ΅TrJβ1Ξ¨β (KΞΒ΅)Ξ¨ .
The sum of the 2nd terms of (1.6), (1.7) is TrΞ¨β
KΞ¡Ψ AΒ΅Jβ1+ Jβ1Aβ Β΅ = 0.
1In the non-covariant form, i.e. the original form, of Diracβs equation one has Ξ0= I, ΞΞΊ= Ξ³0Ξ³ΞΊ, 1 β€
ΞΊ β€ 3 , where the Ξ³Β΅, 0 β€ Β΅ β€ 3 are Dirac-Clifford matrices, which make the Dirac equation covariant
The sum of the 3rd terms of (1.6), (1.7) TrJβ1
Ξ¨β (KM + Mβ Kβ )Ξ¨ = 0. Thus, we find (1.5)
b. By putting hats on Ξ¨ and AΒ΅ our considerations can be rephrased for PDE (1.2).
Remind that from Uβ J U = J it follows that Jβ1Uβ = Uβ1Jβ1. Finally
Tr Jβ1Uβ [Ξ¨β KΞ¡Ψ]U = Tr Uβ1Jβ1[Ξ¨β KΞ¡Ψ]U = Tr Jβ1[Ξ¨β KΞ¡Ψ].
2
Stationary points of complex-valued functionals
In this section we pay some attention to the Euler Lagrange field equations in the com-plex field case. Most physics textbooks start, in a rather verbose way, with 18th century variational rituals. However most of them become suddenly very vague, or fall completely silent, when state functions involving complex variables come into play! In order to get some feeling for such Lagrangians, we first mention a finite dimensional toy result.
Theorem 2.1 Let
f : CnΓ Cn
3 (z; w) 7β f (z, w) β C
be an analytic function of 2n complex variables with the special property f (z, z?) β R , for all z β Cn. Here z = x + iy, z? = xβ iy.
a. Consider the function
RnΓ Rn 3 (x; y) 7β g(x, y) = f (z, z?) = f (x + iy, x β iy) β R .
The relations between the (real) partial derivatives of g at (x, y) and the (complex) partial derivatives of f at (z, z?) are βg βx(x, y) = βf βz(z, z ? ) + βf βw(z, z ? ) βf βz(z, z ? ) = 1 2 βg βx(x, y) β i βg βy(x, y) βg βy(x, y) = i βf βz(z, z ?) β iβf βw(z, z ?) βf βw(z, z ?) = 1 2 βg βx(x, y) + i βg βy(x, y) (2.1) βf βw(z, z ?) = βf βz(z, z ?)
b. For g to have a stationary point at (a ; b) β RnΓ Rn each one of the following three
conditions is necessary and sufficient β’ βg βx(a, b) = βg βy(a, b) = 0 , β’ βf βz(a + ib, a β ib) = 0 , β’ βf βw(a + ib, a β ib) = β βf βz?(a + ib, a β ib) β = 0 . (2.2)
c. If the special property f (x + iy, x β iy) β R is relaxed to Ο(f (x + iy, x β iy)) β R for some non-constant analytic Ο : C β C, then the βstationary point resultβ b. still holds. Proof: Straightforward calculation In Theorem 2.4 an β-dimensional generalisation of this result is presented.
A special bookkeeping
In the sequel, for the above variable z, usually a matrix Z β CrΓc will be taken. In order to
explain our bookkeeping and also for some special properties, we now consider an analytic function of 2 matrix variables
F : CrΓc
Γ CcΓr
β C : (Z ; W) 7β F (Z , W) . (2.3) Because of Hartogβs Theorem, see [H] Thm 2.2.8, it is enough to assume analyticity with respect to each entry of each matrix separately.
The (complex!) partial derivatives of F are gathered in matrices,
(Z; W) 7β F(1)(Z, W) β CcΓr, (Z; W) 7β F(2)(Z, W) β CrΓc, with F(1) ij = βF βZ ij = βF βZji , F(2) k` = βF βW k` = βF βW`k . (2.4) In our notation the C-linearization of F at (Z, W), for Ξ΅ β C , |Ξ΅| small, reads
F (Z + Ξ΅H, W + Ξ΅K) = F (Z, W) + Ξ΅Tr[F(1)]H} + Ξ΅Tr[F(2)]K +O(|Ξ΅|2
). (2.5) Notation: Sometimes, in order to avoid excessive use of brackets, it is convenient to write TrF(1) : H} instead of Tr[F(1)]H}.
Also, without warning, in proofs sometimes Einsteinβs summation convention for repeated upper and lower indices will be used.
Next split Z in real and imaginary parts Z = X + iY and introduce the function f
F : RrΓc
Γ RrΓc β C : (X; Y) 7β fF (X , Y) = F (Z, Zβ ) =F (X + iY , X>β iY>). (2.6) The R -linearization of fF at (X , Y) for Ξ΅ β R , |Ξ΅| small, can now be written
f F (X + Ξ΅A , Y + Ξ΅B) = fF (X , Y) + Ξ΅Trβ fF βXA + Ξ΅Tr β fF βY B +O(Ξ΅ 2 ), (2.7) with Tr β fF βXA = Tr[F (1)]A + Tr[F(2)]A> = Tr [F(1)] + [F(2)]>A , Tr β fF βYB = Tr i[F (1)]B + Tr β i[F(2)]B> = Tr i [F(1)] β [F(2)]>B , (2.8) where the matrices X, Y, A, B are all real. The (complex) derivatives F(1),F(2) are taken at (Z, Zβ ). In the usual (somewhat confusing) notation, this corresponds to
β fF βX = βF βX = βF βZ + βF βZβ > , β fF βY = βF βY = i βF βZ β i βF βZβ > , (2.9) and, similarly sloppy,
βF βZ = 1 2 βF βX β i βF βY , βF βZβ > = 1 2 βF βX + i βF βY. (2.10) If it happens that Z 7βF (Z, Zβ ) is R -valued, the results of Theorem (2.1) can be rephrased. Theorem 2.2
Let, as in (2.3),
F : CrΓc
Γ CcΓr 3 (Z; W) 7β F (Z, W) β C .
be analytic. Suppose F (Z, Zβ ) β R , for all Z β CrΓc. Write Z = X + iY. Denote
f F : RrΓc Γ RrΓc β R : (X; Y) 7β fF (X , Y) = F (Z, Zβ ) =F (X + iY , X>β iY>) , β’ We have F(1)(Z, Zβ ) = [F(2)(Z, Zβ )]β . (2.11) Further, for the function fF to have a stationary point at (A ; B) β RrΓcΓ RrΓc each one
of the following three conditions is necessary and sufficient β’ β fF βX(A, B) = β fF βY(A, B) = 0 β’ F(1)(A + iB, A>β iB> ) = ββF βZ(A + iB, A >β iB> ) β = 0 β’ F(2)(A + iB, A>β iB>) = ββF βZβ (A + iB, A >β iB> ) β = 0. (2.12)
Proof: Is mostly a reformulation of the preceding theorem. It follows directly from
(2.9)-(2.10).
In order to build the concept of Lagrangian density we need an analytic function, named proto-Lagrangian, L : CrΓc Γ CcΓr Γ CN rΓc Γ CcΓN r Γ RN β C, (P; Q>; R ; S>; x) 7β L (P; Q>; R ; S> ; x) , (2.13) where P β CrΓc , R = col R 1, . . . , RN , RΒ΅β CrΓc , 1 β€ Β΅ β€ N , Q>β CcΓr , S>= row S>1 , . . . , S>N , S>Β΅ β CcΓr , 1 β€ Β΅ β€ N . Instead of (2.13) it will be convenient sometimes to denote the proto Lagrangian by
L (P; Q>
; . . . , RΒ΅, . . . ; . . . , S>Β΅, . . . ; x). It will be required that L (O; O>; O ; O>; x) = 0.
The (complex) partial derivatives of L , cf. (2.4)-(2.5), with respect to its 2N + 2 matrix arguments are denoted, respectively,
L(o), L(o?), L(1), . . . ,L(N ), L(1?), . . . ,L(N ?).
The (real) partial derivatives ofL , with respect to the vector variable x is denoted L(β). For any given matrix-valued function Ξ¨ : RN β CrΓc, we define a Lagrangian density
LΟ : RN β C, by substitution of Ξ¨, its 1st derivatives β¡Ψ = Ξ¨, Β΅, 1 β€ Β΅ β€ N , and the
hermitean transposed of all those, in L :
x 7β LΟ(x) =L (Ξ¨(x); Ξ¨β (x); βΞ¨(x) ; βΞ¨β (x) ; x ), (2.14) where βΞ¨(x) = colβ1Ξ¨(x) , . . . , βNΞ¨(x) β CN rΓc, βΞ¨β (x) = rowβ1Ξ¨β (x) , . . . , βNΞ¨β (x) β CcΓN r. Also the matrix-valued functions
x 7β [LΟ(Β΅)](x) = [L(Β΅)](Ξ¨(x); Ξ¨β (x); βΞ¨(x) ; βΞ¨β (x) ; x ) β CcΓr, similarly x 7β [LΟ(Β΅?)] β CrΓc, and x 7βLΟ(β) β RN, will be used.
On a suitable space of functions Ξ¨ : RN β CrΓc, it often makes sense to define the
Lagrangian functional Ξ¨ 7β L(Ξ¨ , Ξ¨β ) = Z RN L (Ξ¨(x); Ξ¨β (x); βΞ¨(x) ; βΞ¨β (x) ; x ) dx β C. (2.15)
Remark 2.3 The Lagrangian functional L remains the same if we replace L by L (Ξ¨; Ξ¨β
; βΞ¨ ; βΞ¨β ; x) + βΒ΅wΒ΅(Ξ¨, Ξ¨β , x),
with wΒ΅ a vectorfield which vanishes sufficiently rapidly at infinity. Therefore the functional Ξ¨ 7β L(Ξ¨ , Ξ¨β ) is R -valued if
L (Ξ¨; Ξ¨β
; βΞ¨ ; βΞ¨β ; x) β L (Ξ¨; Ξ¨β ; βΞ¨ ; βΞ¨β ; x) = βΒ΅WΒ΅(Ξ¨, Ξ¨β , x) ,
i.e. the divergence of a vector field.
Note that L may be R -valued while LΟ is not !!
If we split Ψ into real and imaginary parts: Ψ = ΨRe + iΨIm and Ψ,¡ = ΨRe ,¡+ iΨIm ,¡ ,
the R -directional derivatives with respect to Ξ¨Re and Ξ¨Im of the Lagrangian functional
L are explained by DΞ¨ReL , A = d dΞ΅L(Ξ¨ + Ξ΅A , Ξ¨ β + Ξ΅A>) Ξ΅=0 = = d dΞ΅ Z RN L (Ξ¨(x) + Ξ΅A(x); Ξ¨β
(x) + Ξ΅A>(x); β Ξ¨(x) + Ξ΅A(x) ; β Ξ¨β (x) + Ξ΅A>(x) ; x) dx
Ξ΅=0
, with A : RN β RrΓc, and Ξ΅ β R , |Ξ΅| small.
DΞ¨ImL , B = d dΞ΅L(Ξ¨ + Ξ΅ iB , Ξ¨ β β Ξ΅ iB>) Ξ΅=0 = = d dΞ΅ Z RN L (Ξ¨(x) + Ξ΅ iB(x); Ξ¨β (x) β Ξ΅ iB>(x); β Ξ¨(x) + Ξ΅ iB(x) ; β Ξ¨β (x) β Ξ΅ iB>(x) ; x) dx Ξ΅=0 , with B : RN β RrΓc , and Ξ΅ β R , |Ξ΅| small.
When calculating the C-directional derivatives DΞ¨L , DΞ¨β L , the variables Ξ¨ , Ξ¨ β
are con-sidered to be independent. These derivatives are supposed to be elements in the (complex) linear dual of L2(RN; CrΓc). They are explained by
DΞ¨L , H = d dΞ΅L(Ξ¨ + Ξ΅H , Ξ¨ β ) Ξ΅=0 = = d dΞ΅ Z RN L (Ξ¨(x) + Ξ΅H(x); Ξ¨β (x); β Ξ¨(x) + Ξ΅H(x) ; βΞ¨β ; x ) dx Ξ΅=0, with H : RN β CrΓc, and Ξ΅ β C , |Ξ΅| small.
DΞ¨β L , K = d dΞ΅L(Ξ¨ , Ξ¨ β + Ξ΅K) Ξ΅=0 = = d dΞ΅ Z RN L (Ξ¨(x); Ξ¨β (x) + Ξ΅K(x); βΞ¨(x) ; β(Ξ¨β (x) + Ξ΅K(x)) ; x ) dx Ξ΅=0 , with K : RN β CcΓr, and Ξ΅ β C , |Ξ΅| small.
For H , K , A , B vanishing sufficiently rapidly at β a partial integration leads to the standard Euler-Lagrange expressions for the functional derivatives of L.
Theorem 2.4
Assume that L is R -valued. (Cf. Remark 2.3). If Ξ¨ satisfies any one of the following three Lagrangian systems
DΞ¨L = [L (o) Ο ] β N X Β΅=1 β βxΒ΅[L (Β΅) Ο ] = 0 , DΞ¨β L = [LΟ(o?)] β N X Β΅=1 β βxΒ΅[L (Β΅?) Ο ] = 0 , ο£±     ο£²     ο£³ DΞ¨ReL = βL βΞ¨Re β N X Β΅=1 β βxΒ΅ βL βΞ¨Re ,Β΅ = 0 , DΞ¨ImL = βL βΞ¨Im β N X Β΅=1 β βxΒ΅ βL βΞ¨Im ,Β΅ = 0 . , (2.16) with L = L (Ξ¨(x); Ξ¨β (x); βΞ¨(x) ; βΞ¨β (x) ; x ) , then it also satisfies the other two. Proof: With the notation (2.8)-(2.10) we obtain
βL βΞ¨Re =L(o)+ [L(o?)]> , βL βΞ¨Im = iL(o)β i[L(o?)]> , (2.17) and, the other way round,
L(o?)> = 1 2 βL βΞ¨Re + i βL βΞ¨Im , L(o) = 1 2 βL βΞ¨Re β i βL βΞ¨Im , (2.18) and similar expressions with (o) , (o?) replaced by (Β΅) , (Β΅?) and Ξ¨ , Ξ¨Re , Ξ¨Im replaced
by Ξ¨,Β΅, Ξ¨Re ,Β΅, Ξ¨Im ,Β΅. Then DΞ¨L = 12 DΞ¨ReL β iDΞ¨ImL DΞ¨β L > = 12 DΞ¨ReL + iDΞ¨ImL DΞ¨ReL = DΞ¨L +DΞ¨β L > DΞ¨ImL > = iDΞ¨L β iDΞ¨β L > . If we take into account that the entries of the matrix valued functions DΞ¨ReL and DΞ¨ImL
are R -valued, we find
DΞ¨β L
β
= DΨL , (2.19)
from which the theorem easily follows. Examples 2.5 (Matter Fields)
a) Let ΞΒ΅ and M be constant complex matrices with ΞΒ΅β = ΞΒ΅ and M = βMβ . Then the
Lagrangian density
LΟ = i TrΞ¨β ΞΒ΅β¡Ψ + Ξ¨β M Ξ¨ , (2.20)
for Ξ¨ : RN β CrΓc, satisfies the condition of Theorem (2.4) and leads to (1.1) with A = 0.
b) Let ΞΒ΅, 1 β€ Β΅ β€ N : RN β CrΓr. Let AΒ΅, 1 β€ Β΅ β€ N : RN β CcΓc.
Let M : RN β CrΓr.
Suppose both the existence of K : RN β CrΓr, having inverse Kβ1
(x), for all x β RN,
(KΞΒ΅)β = KΞΒ΅, , 1 β€ Β΅ β€ N, Aβ
Β΅(x)J + J AΒ΅(x) = 0, 1 β€ Β΅ β€ N, x β RN,
and KM + Mβ Kβ β βΒ΅ KΞΒ΅ = 0.
Then the Lagrangian density
LΟ = i TrΞ¨β K(ΞΒ΅β¡Ψ)Jβ1+ Ξ¨β K(Ξ¡ΨAΒ΅)Jβ1+ Ξ¨β KM Ξ¨Jβ1 , (2.21)
for Ξ¨ : RN β CrΓc satisfies L β L = β
Β΅w and hence the condition of Theorem (2.4).
It leads to the βmatter-field equationβ
ΞΒ΅β¡Ψ + Ξ¡ΨAΒ΅+ M Ξ¨ = 0 (2.22)
Indeed. Taking suitable combinations we find respectively TrΞ¨β KΞΒ΅(β ¡Ψ)Jβ1+ Jβ1(β¡Ψ)β (KΞΒ΅)β Ξ¨) = TrJβ1β Β΅[Ξ¨β KΞ¡Ψ)] + TrJβ1[Ξ¨β βΒ΅(KΞΒ΅)Ξ¨)] , TrΞ¨β K(Ξ¡ΨA Β΅)Jβ1+ Jβ1Aβ ¡Ψ β (KΞΒ΅)β Ξ¨ = TrAΒ΅Jβ1+ Jβ1Aβ ¡Ψ β (KΞΒ΅)Ξ¨ = 0, TrΞ¨β KM Ξ¨Jβ1+ Jβ1Ξ¨β Mβ Kβ Ξ¨ = TrJβ1Ξ¨β KM Ξ¨ + Jβ1Ξ¨β Mβ Kβ Ξ¨ = = TrJβ1Ξ¨β (KM + Mβ Kβ )Ξ¨ . Ultimately we find LΟβLΟ = βΒ΅TrJβ1[Ξ¨β KΞ¡Ψ)] = βΒ΅Tr[Ξ¨β KΞ¡Ψ)]Jβ1 . (2.23)
The Euler-Lagrange equations are
K ΞΒ΅β¡Ψ + Ξ¡ΨAΒ΅+ M Ξ¨Jβ1 = 0, (2.24)
from which K and Jβ1 can be cancelled. c) The Lagrangian density
LΟ = Tr[β¡Ψ]β Ξ¡ν[βΞ½Ξ¨] + Ξ¨β RΞ¨ , (2.25)
with Ξ¡ν, R : RN β CrΓr and [Ξ¡ν]β = ΞΞ½Β΅, Rβ
= R, is R -valued. It leads to the 2nd order equation X Β΅,Ξ½ β βx¡Π¡ν β βxΞ½Ξ¨ β R Ξ¨ = 0 . (2.26)
d. The Lagrangian density for functions Ξ¨ = col[ Ο1 Ο2 ] : R N +1 β C2, LΟ = Tr h Ξ¨β ( iβtΞ¨ + βΞ¨ + V Ξ¨) i , with x 7β V (x) β C2Γ2, Vβ = V , (2.27) leads to a R -valued Lagrangian functional L. Indeed
LΟβLΟ = iβtTr h Ξ¨β Ξ¨i+βx1Tr h Ξ¨β (βx1)Ξ¨β(βx1Ξ¨) β Ξ¨i+. . .+β xNTr h Ξ¨β (βxN)Ξ¨β(βxNΞ¨) β Ξ¨i.
3
Free Gauge Fields
The βfield variablesβ to be considered in this section are smooth functions A : RN β CcΓc Γ Β· Β· Β· Γ CcΓc | {z } N times : x 7β A(x) = col[A1(x), . . . , AΒ΅(x), . . . , AN(x)] , (3.1)
with AΒ΅(x) β g, with g β CcΓc some fixed real Lie algebra. 2 This means that g is a
R -linear subspace in CcΓc which is not necessarily a C-linear subspace. On g we impose the usual βcommutatorβ-Lie product
{AΒ΅, AΞ½} = AΒ΅AΞ½ β AΞ½AΒ΅ .
Important examples are matrix Lie Algebras of type gJ = { X β CrΓr
X
β
J + J X = 0 } , with fixed invertible J β CrΓr. Note that gJ is always a R -linear subspace in CrΓr, but not necessarily C-linear.
However: {Jβ1 = Jβ } β {X β gJ β Xβ β gJ}.
Next, byPg: CcΓc β g, we denote the real orthogonal projection with respect to the real
inner product X, Y 7β Re Tr[Xβ Y ]. Remarks 3.1
Consider CcΓcas a real vector space with standard real inner product X, Y 7β Re Tr[Xβ Y ]. ByPg : CcΓcβ g, we denote the real orthogonal projection with respect this inner product.
β’ The Hermitean conjugation map X 7β Xβ
is R -linear symmetric and orthogonal. β’ If β X β g : Xβ β g, in short gβ
= g, it follows that βX β CcΓc :P
g(Xβ ) = (PgX)β .
β’ For fixed K, L β CcΓc the mapping X 7β KXβ
L is R -linear. Its R -adjoint is Y 7β LYβ K.
β’ For any fixed invertble J β CcΓc the mapping
QJ : CcΓcβ CcΓc : X 7βQJX =
1
2(X β J
β1Xβ J ) ,
(3.2) is a R -linear mapping which reduces to the identity map when restricted to gJ.
β’ QJ is a R -linear projection on gJ iff J = Jβ .
β’ QJ is a R -linear orthogonal projection on gJ if J = Jβ1 = Jβ .
In this special case QJ =Pg, with g = gJ. 2In physics textbooks one often denotes iA
Β΅, instead of AΒ΅, cf. [DM]. For resemblance with
Electro-magnetism, I suppose. Because of u(1) = iR ? To this author the factor i is not convenient in all other cases.
β’ If we modify the standard real inner product on CcΓc to X, Y 7β Re Tr[Xβ J2Y ], the
projection QJ is orthogonal iff J = Jβ .
Proof
β’ Re Tr[(Xβ )β Y ] = Re Tr[XY ] = Re Tr[Xβ (Yβ )]. Also Re Tr[(Xβ )β (Yβ )] = Re Tr[(X)β (Y )].
β’ Since g is supposed to be an invariant subspace for X 7β Xβ and the latter is symmetric,
also gβ₯ is invariant.
β’ Re Tr[(KXβ L)β Y ] = Re Tr[KXβ LYβ ] = Re Tr[Xβ (LYβ K)] .
β’ For X β g holds (I βQJ)X = 0 , iff X β g .
β’ Q2 J = QJ iff J = Jβ . β’β’ 1 2Re Tr[(X β J β1Xβ J )β J2Y ] = 1 2Re Tr[X β J2Y ] β 1 2Re Tr[X β J2(Jβ1Yβ Jβ 2Jβ1)] .
The 2nd term equals β12Re Tr[Xβ J2(Jβ1Yβ J ] , for all X, Y , iff J = Jβ . Associated with A, cf. (3.1), we introduce covariant-type partial derivatives
βA
Β΅, 1 β€ Β΅ β€ N of functions U βC β
(RN: CcΓc) by
βAΒ΅U = βΒ΅U β {AΒ΅, U } = βΒ΅U β adAΒ΅U . (3.3)
One has the Leibniz-type rules βA Β΅(U V ) = (βAΒ΅U )V + U (βAΒ΅V ) , TrU (βA Β΅V ) = βΒ΅TrUV β Tr(βAΒ΅U )V . (3.4) Note that if U βCβ(RN: g) then also βA
Β΅U βC β
(RN: g).
Next, as in section 1, for given AΒ΅, AΞ½ βCβ(RN: g) , 1 β€ Β΅, Ξ½ β€ N , define
F¡ν = βΒ΅AΞ½ β βΞ½AΒ΅β {AΒ΅, AΞ½} β Cβ(RN: g) , (3.5)
to which Theorem 1.2 applies.
For the construction of a R -valued Lagrangian density GAfor the Gauge field(s) A we again
employ a proto Lagrangian G , which is now an analytic function of N(N β 1) complex-matrix variables and just smooth in N real variables:
G : CcΓc Γ Β· Β· Β· Γ CcΓc | {z } 1 2N (N β1) times Γ CcΓc Γ Β· Β· Β· Γ CcΓc | {z } 1 2N (N β1) times Γ RN β C . (3.6) The 1st set of entries to this function is labeled by the ordered pairs (¡ν) , 1 β€ Β΅ < Ξ½ β€ N . The 2nd set of entries is labelled by the ordered triple (ΞΈΟ?) , 1 β€ ΞΈ < Ο β€ N . We denote
{ . . . , P¡ν, . . . ; . . . , QΞΈΟ?, . . . ; x} 7β G ( . . . P¡ν, . . . ; . . . QΞΈΟ?, . . . ; x) β C ,
with 1 β€ Β΅ < Ξ½ β€ N and 1 β€ ΞΈ < Ο β€ N . The 3 bunches of variables get their corresponding partial derivatives denoted by, respectively, cf. (2.4),
G(¡ν)(. . . , P
Let the Lie algebra g be fixed. On G we put the condition, take QΞΈΟ? = P β ΞΈΟ,
β {P¡ν}1β€Β΅<Ξ½β€N β g βx β RN : G (. . . , P¡ν, . . . ; . . . , PΞΈΟβ , . . . ; x) β R . (3.7)
The Lagrangian density we want to consider is found by replacing P¡ν β F¡ν, QΞΈΟ? β F β ΞΈΟ,
x 7β GA(x) =G ( . . . , F¡ν(x), . . . ; . . . , F β
ΞΈΟ(x), . . . ; x ) β R . (3.8)
Note that if g = gJ, for some fixed J β CcΓc, we have F β
ΞΈΟ= βJ FΞΈΟJβ1, ΞΈ < Ο.
As in the previous section, a corresponding useful notation is
x 7β GA(¡ν)(x) =G(¡ν)( . . . , F¡ν(x), . . . ; . . . , FΞΈΟβ (x), . . . ; x ) β CcΓc. (3.9)
The Lagrangian density GA depends on the field variables x 7β AΒ΅(x) , 1 β€ Β΅ β€ N , and
their derivatives. All being functions in a vectorspace over R . In the important special case g = gJ the hermitean conjugate notation of the field variables AΒ΅need not even occur.
Finally, note that, because of (2.11) and (3.8), we have G(ΞΈΟ?) A (x) = (G (ΞΈΟ) A ) β (x) , 1 β€ ΞΈ < Ο β€ N . (3.10) Notation 3.2 In order to visually simplify the formulae to come, it is useful to extend the set of functions GA(¡ν), cf.(3.9), to βfullβ labels 1 β€ Β΅, Ξ½ β€ N in the following way,
Λ G(¡ν) A = ο£±  ο£²  ο£³ G(¡ν) A if 0 β€ Β΅ < Ξ½ β€ N , as before, 0 if Β΅ = Ξ½ , βGA(Ξ½Β΅) if 0 β€ Ξ½ < Β΅ β€ N . (3.11) Theorem 3.3
Fix a matrix Lie algebra g β CcΓc. Consider the Lagrangian density G
A of (3.8).
A. The Euler-Lagrange equations for the free gauge fields AΒ΅, 1 β€ Β΅ β€ N , with values in
the Lie algebra g β CcΓc, read
N X Β΅=1 Pg βA Β΅ [PgGΛ (¡κ?) A ] β β = 0 , 1 β€ ΞΊ β€ N , (3.12) with βA Β΅ as in (3.3).
B. In the special case gβ = g the Euler-Lagrange equations simplify to
N X Β΅=1 βA Β΅PgGΛ (¡κ) A = 0 , 1 β€ ΞΊ β€ N . (3.13) C. If we take g = gJ, with J = Jβ = Jβ1, the latter becomes
N X Β΅=1 βAΒ΅QJ[ ΛGA(¡κ)] = 0 , 1 β€ ΞΊ β€ N , (3.14) where QJZ = 12Z β 12J Zβ J , Z β CcΓc.
Proof
A. In order to calculate the (directional) derivatives of the Lagrangian functional G = R
GA dx with respect to the free gauge fields AΞΊ, 1 β€ ΞΊ β€ N , we first expand a perturbation
of x 7β F¡ν(x) by substitution of the gauge fields x 7β AΒ΅(x) + Ρδ¡κH(x) , Ξ΅ β R ,
F¡ν;Ξ΅,ΞΊ = h βΒ΅(AΞ½+ ΡδνκH) β βΞ½(AΒ΅+ Ρδ¡κH) β {AΒ΅+ Ρδ¡κH , AΞ½+ ΡδνκH} i = = h βΒ΅AΞ½ β βΞ½AΒ΅β {AΒ΅, AΞ½} i + Ξ΅ δνκ h βΒ΅H β {AΒ΅, H} i β Ξ΅ δ¡κ h βΞ½H β {AΞ½, H} i = = F¡ν+ ΡδνκβAΒ΅H β Ρδ¡κβAΞ½H .
Consider the expansion G (. . . , F¡ν;Ξ΅,ΞΊ, . . . ; . . . , F β ΞΈΟ;Ξ΅,ΞΊ, . . . ; x) β G (. . . , F¡ν, . . . ; . . . , F β ΞΈΟ, . . . ; x) = = Ξ΅ X 1β€Β΅<Ξ½β€N Trh[GA(¡ν)][δνκβAΒ΅H β δ¡κβAΞ½H ] + + Ξ΅ X 1β€ΞΈ<Οβ€N Trh[GA(ΞΈΟ?)][Ξ΄ΟΞΊβΞΈAH β δθκβAΟHΞΊ]β i +O(Ξ΅2) = = Ξ΅ 2 N X Β΅, Ξ½=1 Trh[ ΛGA(¡ν)][δνκβAΒ΅H β δ¡κβAΞ½H ] + +Ξ΅ 2 N X ΞΈ, Ο=1 Tr h [ ΛGA(ΞΈΟ?)][Ξ΄ΟΞΊβAΞΈH β δθκβAΟH ] β i +O(Ξ΅2) = = Ξ΅ 2 N X Β΅=1 Trh[ ΛGA(¡κ)][βAΒ΅Hiβ Ξ΅ 2 N X Ξ½=1 Trh[ ΛGA(ΞΊΞ½)][βAΞ½H ]i+ +Ξ΅ 2 N X ΞΈ=1 Trh[ ΛGA(ΞΈΞΊ?)][βAΞΈHi β Ξ΅ 2 N X Ο=1 Trh[ ΛGA(ΞΊΟ?)][βAΟH ]β i +O(Ξ΅2) = = Ξ΅ N X Β΅=1 Tr h [ ΛGA(¡κ)][βAΒ΅Hi + Ξ΅ N X Β΅=1 Tr h [ ΛGA(¡κ?)][βAΒ΅H ]β i +O(Ξ΅2) = = 2Ξ΅Re N X Β΅=1 Trh[ ΛGA(¡κ?)]β [βAΒ΅H]i + O(Ξ΅2) = 2Ξ΅Re N X Β΅=1 Trh[PgGΛ (¡κ?) A ] β [βAΒ΅H]i + O(Ξ΅2) = = β2Ξ΅Re N X Β΅=1 TrhβA Β΅ [PgGΛ (¡κ?) A ] β Hi + N X Β΅=1 βΒ΅(. . .) +O(Ξ΅2) =
= β2Ξ΅Re N X Β΅=1 Trh Pg βA Β΅ [PgGΛ (¡κ?) A ] β β β Hi + N X Β΅=1 βΒ΅(. . .) +O(Ξ΅2) . (3.15)
In this derivation we used, respectively, the antisymmetry Β΅ β Ξ½ of [ ΛGA(¡ν)] and [δνκβAΒ΅H β
δ¡κβAΞ½H ], the Leibniz rule(3.4), the fact that Re Tr
h
. . .β Hi expresses the real inner product on CcΓc and P
g the real orthogonal projection on g.
Also properties like Tr[AB] = Tr[BA] , Tr[A{B , C}] = Tr[{A , B}C] play a crucial role. The result now follows by the usual variational practices.
B. If gβ = g the real linear mappings {.}β and Pg commute, which greatly simplifies the
result of A.
C. Use Remarks 3.1.
Example 3.4
A. For convenience we restrict to Lie-algebras with property gβ = g. We will consider general Lagrangians which are (real) quadratic in F¡ν. Here, in our summation expressions,
we write Β΅ < Ξ½ instead of 1 β€ Β΅ < Ξ½ β€ N . Start from the proto Lagrangian G = X Β΅<Ξ½ , ΞΈ<Ο h(¡ν)(ΞΈΟ)Tr[P¡νQΞΈΟ?] with h(¡ν)(ΞΈΟ) = h(ΞΈΟ)(¡ν) β C . (3.16) Note X Β΅<Ξ½,ΞΈ<Ο h(¡ν)(ΞΈΟ)Tr[P¡νP β ΞΈΟ] β R .
For the derivatives of G we find, G(¡ν)(. . . , P ¡ν, . . . ; . . . , QΞΈΟ?, . . . ) = X Ξ±<Ξ² h(¡ν)(Ξ±Ξ²)QΞ±Ξ²? G(ΞΈΟ?)(. . . , P ¡ν, . . . ; . . . , QΞΈΟ?, . . . ) = X Ξ±<Ξ² h(Ξ±Ξ²)(ΞΈΟ)PΞ±Ξ² If we take QΞΈΟ? = P β
ΞΈΟ, one easily checks (3.8),
G(¡ν)β (. . . , P ¡ν, . . . ; . . . , P β ΞΈΟ, . . . ) = X Ξ±<Ξ² h(¡ν)(Ξ±Ξ²)PΞ±Ξ² = X Ξ±<Ξ² h(Ξ±Ξ²)(¡ν)PΞ±Ξ² =G(¡ν?).
The Lagrangian density
GA= X Β΅<Ξ½, ΞΈ<Ο h(¡ν)(ΞΈΟ)Tr[F¡νF β ΞΈΟ] , (3.17)
can now be put in (3.13) to find the Euler-Lagrange equations. Note however, that Pg
So, let us restrict to gβ = g Γ nd h(¡ν)(ΞΈΟ) β R . Anti-symmetrize h(¡ν)(ΞΈΟ) to full labels: Λ h(¡ν)(ΞΈΟ) = ο£±    ο£²    ο£³ h(¡ν)(ΞΈΟ) if Β΅ < Ξ½ , ΞΈ < Ο or Β΅ > Ξ½ , ΞΈ > Ο 0 if Β΅ = Ξ½ and/or ΞΈ = Ο βh(Ξ½Β΅)(ΞΈΟ) if Β΅ > Ξ½ , ΞΈ < Ο βh(¡ν)(ΟΞΈ) if Β΅ < Ξ½ , ΞΈ > Ο
In this special case
Λ G(¡ν) A = 1 2 N X Ξ±,Ξ²=1 Λ h(¡ν)(Ξ±Ξ²)FΞ±Ξ²β ,
and, since FΞ±Ξ²β β g, the E-L-equations (3.13) become 1 2 N X Ξ±,Ξ²=1 N X Β΅=1 Λ h(¡κ)(Ξ±Ξ²) βΒ΅F β Ξ±Ξ²β {AΒ΅, F β Ξ±Ξ²} = 0 , 1 β€ ΞΊ β€ N . (3.18)
B. For gauge fields on Minkowski space, with coordinates x0, x1, x2, x3 and
metric [g¡ν] = diag(1, β1, β1, β1), one usually takes, cf. [DM],
h(¡ν)(Ξ±Ξ²) = g¡αgΞ½Ξ² = (β1)1+δ¡0δ¡α(β1)1+δν0δνβ = (β1)δ¡0+δν0δ¡αδνβ.
Hence
Λ
h(¡κ)(Ξ±Ξ²) = sgn(ΞΊ β Β΅) sgn(Ξ² β Ξ±) (β1)δ¡0+δκ0δ¡αδκβ.
In this special case the Lagrangian density (3.17) reads GA=
X
0β€Β΅<Ξ½β€3
(β1)δ¡0+δν0TrF
¡νF¡νβ . (3.19)
The corresponding Euler-Lagrange equations are
3 X Β΅=0 (β1)δ¡0+δκ0βA Β΅F β ¡κ= 0 , 0 β€ ΞΊ β€ 3 . (3.20)
For dim g = 1 the term adAΒ΅F
β
¡κ vanishes. This simplification, viz. βAΒ΅ = βΒ΅ , leads
to standard electromagnetism in Minkowski space. Indeed, if we put Aβ 0 = βΞ¦ and col[Aβ 1, Aβ 2, Aβ 3] = A, then (3.20) turns into Maxwellβs equations βin potential formβ
ο£±  ο£²  ο£³ β βtdivA + βΞ¦ = 0 β2 βt2A β βA + grad β βtΞ¦ + divA = 0 (3.21)
If the pair A, B satisfies (3.21), then the pair E = ββ A
βt β gradΞ¦ , B = rotA , satisfies the classical Maxwell equations.
Finally, imposing the βLorenz-Gaugeβ β
βtΞ¦ + divA = 0, we find the usual wave equations β2
tΞ¦ β βΞ¦ = 0 , βt2Aβ βA = 0 . For more details see Appendix B.
4
Noether Fluxes
βInfinitesimal symmetriesβ of the Lagrangian densityL lead to local conservation laws for the solutions of the Euler Lagrange equations. So we are told by Emmy Noetherβs famous theorem. First we have a short look at the needed concepts as formulated within our special (simple) context.
Definition 4.1 A Conservation Law or Noether Flux is a vectorfield on RN, with
com-ponents VΟΒ΅, 1 β€ Β΅ β€ N , which arise from a set of functions of Proto-Lagrangian type, VΒ΅, 1 β€ Β΅ β€ N , cf. (2.13), such that for all solutions Ξ¨ of the Euler Lagrangian system,
cf. Th 2.4, we have N X Β΅=1 β βxΒ΅V Β΅ Ο(x) = 0 , where V Β΅ Ο(x) = V Β΅(Ξ¨(x), Ξ¨β (x), Ξ¨,Β΅(x), Ξ¨β ,Β΅(x), x) . (4.1)
A conservation law can be named βtrivialβ for several reasons: It may happen that for all solutions Ξ¨ the fluxes VΟΒ΅ = 0. Another reason for triviality occurs if for all functions Ξ¨, whether they are solutions or not, the identity (4.1) is satisfied. For example if the components VΟΒ΅ arise from the curl of an arbitrary vector field depending on Ξ¨.
Two types of symmetries will be considered here: βInternal symmetriesβ and βExternal symmetriesβ. They can be formulated in terms of the proto-Lagrangian only.
External symmetries regard transformations of the spatial variables x. We restrict to affine transforms.
Definition 4.2 (Internal symmetries) A set of linear mappings K , LΞ»
Β΅ : CrΓc β CrΓc, 1 β€ Ξ», Β΅ β€ N , is said to generate an
internal (local) symmetry of the proto-Lagrangian L if for all P, QΒ΅ β CrΓc, all x β RN,
and s β R , |s| small, one has L (esKP; (esKP)β ; . . . esLλ¡Q Ξ». . . ; . . . (esL Ξ» Β΅Q Ξ»)β . . . ; x) = =L (P; Pβ ; . . . QΒ΅. . . ; . . . Qβ Β΅. . . ; x) +O(s 2) , (4.2)
In many cases the K , LΞ»
Β΅ are realized by left and/or right multiplication with some fixed
Many times there is a special type of internal symmetry which is related to a linear mapping A : RN β RN in the βoutside worldβ,
L (P; Pβ
; . . . (esA)λ¡QΞ». . . ; . . . ((esA)λ¡QΞ»)β . . . ; x) =
=L (P; Pβ ; . . . QΒ΅. . . ; . . . Qβ Β΅. . . ; x) +O(s
2) , (4.3)
Definition 4.3 (External symmetries)
The affine mapping x 7β βsa + esAx on RN, where a β RN and A : RN β RN, a linear mapping, is said to generate an external (local) symmetry of the proto-Lagrangian L if for all P, QΒ΅β CrΓc, all x β RN, and s β R , |s| small, one has
L (P; Pβ
; . . . QΒ΅. . . ; . . . (QΒ΅)β . . . ; βsa + esAx) =
=L (P; Pβ ; . . . QΒ΅. . . ; . . . Qβ Β΅. . . ; x) +O(s
2) . (4.4)
Remarks 4.4
β’ The order constant in O(s2) may depend on all independent variables of L .
β’ If in (4.2)-(4.4) exponents like esK are replaced by I + sK we get equivalent
conditi-ons. However in many practical applications the terms O(s2) are identically zero if exponentials are used.
β’ Local symmetry (4.4) implies L(β)(P; Pβ
; . . . QΒ΅. . . ; . . . Qβ Β΅. . . ; x) Β· (Ax β a) = 0 .
We now first consider two types of conservation laws in connection with affine transforma-tions in space.
For any vector a β RN we define the Translation operator Ta by
TaΞ¨(x) = Ξ¨(x β a).
For any matrix A β RN ΓN we define the dilation operator R A by
RAΨ(x) = Ψ(eAx).
Theorem 4.5
Suppose that, for some K : CrΓc β CrΓc
and some a β RN, the proto-Lagrangian L has internal local symmetry (4.2) with Lλ¡ = δλ
Β΅K and external local symmetry (4.4) with
A = O. Then for any solution Ξ¨ of the Euler-Lagrange system one has the conservation law N X Β΅=1 β βxΒ΅ n Trh[LΟ(Β΅)] Β· (KΞ¨ β aΞ»βλΨ) + [L (Β΅?) Ο ] Β· (KΞ¨ β a Ξ»β λΨ)β i + aΒ΅LΟ o = 0 . (4.5)
Proof: By βΌ= we mean equality up to a term O(s2). We study L esK TsaΞ¨, TsaΞ¨β esK β , βΒ΅[esKTsaΞ¨], βΒ΅[TsaΞ¨β esK β ], x β sa . With our conditions it can be written
L (esKΞ¨(xβsa); (esKΞ¨(xβsa))β
; . . . βΒ΅esKΞ¨(xβsa) . . . ; . . . βΒ΅(esKΞ¨(xβsa))β . . . ; xβsa) βΌ=
βΌ
=L (Ξ¨(x β sa); Ξ¨(x β sa)β ; . . . Ξ¨,Β΅(x β sa) . . . ; . . . Ξ¨,Β΅(x β sa)β . . . ; x β sa) =
=LΟ(x β sa) = (TsaLΟ)(x) . (4.6)
Differentiate the first line of this at s = 0 and useL(β)Β·a = 0 , Tr[L(o) Ο ](KΞ¨ β a Ξ»β λΨ) + [L (o?) Ο ](Ξ¨ β Kβ β aΞ»β λΨβ )+ +[LΟ(Β΅)](Kβ¡Ψ β aΞ»βΞ»β¡Ψ) + [L (Β΅?) Ο ](β¡Ψβ Kβ β aΞ»βΞ»β¡Ψβ ) . (4.7)
If Ξ¨ is a solution we use (2.16) and replace [LΟ(o)] by βxβΒ΅[L
(Β΅)
Ο ], etc. Now (4.7) can be
written as a divergence, which constitutes the left hand side of (4.5), apart from the last term inside { }. Together with the derivative aΞ»β
Ξ»LΟ = βΒ΅(aΒ΅LΟ) at s = 0 of the final
line of (4.6) we arrive at the wanted conserved current (4.5). Example 4.6 Let ΞΒ΅and M be constant complex matrices with ΞΒ΅β = ΞΒ΅and M = βMβ . Then the Lagrangian density
LΟ = Tr iΞ¨β ΞΒ΅β¡Ψ + Ξ¨β M Ξ¨ , (4.8)
for Ξ¨ : RN β CrΓc
satisfies the condition of Theorem 4.1 for K = O and all a β RN. The conservation law reads
β βxΒ΅Tr βa λΨβ ΞΒ΅βλΨ + a¡Ψβ ΞΞ»βλΨ + a¡Ψβ M Ξ¨ = β βxΒ΅Tr βa λΨβ ΞΒ΅βλΨ = 0. (4.9) This can be checked directly for solutions of the PDE: ΞΒ΅β¡Ψ + M Ξ¨ = 0. Observe that
in this special case LΟ = 0 for solutions.
Also the Lagrangian of Example (2.5b), with constant matrices K, M, ΞΒ΅, A
Β΅ leads to
conservation laws of this type. Theorem 4.7
Suppose that, for some K : CrΓc β CrΓc
and some A β RN ΓN with TrA = 0, the proto-Lagrangian L has internal local symmetry (4.2) with LΞ»
¡ = K + [A]λ¡I and external local
symmetry (4.4) with a = 0. Then for any solution Ξ¨ of the Euler-Lagrange system one has the conservation law
N X Β΅=1 β βxΒ΅ n Trh[LΟ(Β΅)](KΞ¨(x) + AΞ±Ξ²xΞ²Ξ¨,Ξ±(x)) + + [LΟ(Β΅?)](KΞ¨(x) + AΞ±Ξ²xΞ²Ξ¨,Ξ±(x))β i β A¡βxΞ²LΟ o = 0 . (4.10)
Proof: We study L esKR sAΞ¨ ; RsAΞ¨β esK β ; . . . βΒ΅[esKRsAΞ¨] . . . ; . . . βΒ΅[RsAΞ¨β esK β ] . . . ; esAx . With our conditions it can be written,
L (Ξ¨(esA
x); Ξ¨(esAx)β ; . . . β¡Ψ(esAx) . . . ; . . . β¡Ψ(esAx)β . . . ; esAx) βΌ=
βΌ =L (Ξ¨(esAx); Ξ¨(esAx)β ; . . . (esA)λ¡Ψ,Ξ»(esAx) . . . ; . . . (esA)λ¡Ψ,Ξ»(esAx)β . . . ; esAx) βΌ= βΌ =L (Ξ¨(esAx); Ξ¨(esAx)β ; . . . Ξ¨,Β΅(esAx) . . . ; . . . Ξ¨,Β΅(esAx)β . . . ; esAx) βΌ= βΌ =L (Ξ¨(esAx); Ξ¨(esAx)β ; . . . Ξ¨,Β΅(esAx) . . . ; . . . Ξ¨,Β΅(esAx)β . . . ; esAx) = =LΟ(esAx) = (RsALΟ)(x). (4.11)
Differentiate the first line of this at s = 0 and use L(β)Β·Ax = 0 :
Tr[L(o) Ο ](KΞ¨(x)+A Ξ± Ξ²x Ξ²Ξ¨ ,Ξ±(x))+[L (Β΅) Ο ]βΒ΅(KΞ¨(x)+AΞ±Ξ²x Ξ²Ξ¨ ,Ξ±(x)) + + [LΟ(o?)](KΞ¨(x) + AΞ±Ξ²xΞ²Ξ¨,Ξ±(x))β + [L (Β΅?) Ο ]βΒ΅(KΞ¨(x) + AΞ±Ξ²x Ξ²Ξ¨ ,Ξ±(x))β . (4.12)
If Ξ¨ is a solution we use (2.16) and replace [LΟ(o)] by βxβΒ΅[L
(Β΅)
Ο ], etc. Now (4.12) can
be written as a divergence, which constitutes the left hand side of (4.10), apart from the last term between { }. Together with the derivative at s = 0 of the final line in (4.11): A¡ββΒ΅LΟ = βΒ΅(A
Β΅ Ξ²x
Ξ²L
Ο), use TrA = 0, we arrive at the conserved current (4.10).
Next we deal with internal symmetries only. They play a crucial role in Gauge theories. A simple case first.
Theorem 4.8
Suppose that, for some linear K : CrΓc β CrΓc the proto-Lagrangian L satisfies (4.2) with
Lλ¡= δλ¡K. Then for any solution Ξ¨ of the Euler-Lagrange system one has the conservation law N X Β΅=1 β βxΒ΅Tr[L (Β΅) Ο ]KΞ¨ + [L (Β΅?) Ο ](KΞ¨) β = 0 , (4.13) Proof: Calculate the derivative
β βsL e
sKΞ¨, (esKΞ¨)β
, βΒ΅[esKΞ¨], βΒ΅[esKΞ¨]β , x, at s = 0 .
With the notation of (2.5) one finds Tr[L(o) Ο ][KΞ¨] + [L (o?) Ο ][KΞ¨] β + [LΟ(Β΅)][KΞ¨,Β΅] + [L (Β΅?) Ο ][KΞ¨,Β΅]β = 0.
If Ξ¨ happens to be a solution of the Lagrangian system, then with (2.16) this becomes Tr[ β βxΒ΅L (Β΅) Ο ][KΞ¨] + [ β βxΒ΅L (Β΅?) Ο ][KΞ¨] β + [LΟ(Β΅)][KΞ¨], Β΅+ [L (Β΅?) Ο ][KΞ¨] β , Β΅ = 0,
which leads to the wanted βconserved currentβ, since K is supposedly constant. In gauge applications K is often realized by a right multiplication by some A β CcΓc. In
such cases KΨ in (4.13) should be replaced by ΨA.
All previous considerations can be applied to matrix gauge fields as well if we replace Ξ¨ by A = col[. . . , AΒ΅, . . .]. Some subtleties occur however because the range of the functions
AΒ΅ is not the whole of CcΓc but some real linear subspace g of it. See Appendix A for more
details.
This section is concluded with conservation laws for non-commutative free gauge fields which come from the special Lagrangian density (3.8).
Theorem 4.9
Consider the proto-Lagrangian G of (3.6) with property (3.7) and Lagrange density as denoted in (3.8). For convenience restrict to g = gβ only.
a. Suppose GA(β)Β·a = 0, for some a β RN then we have the conservation law N X Β΅=1 β βxΒ΅ XN ΞΊ=1 Re TrhPgGΛ (¡κ) A : (a Β·β)AΞΊ i β aΒ΅G A = 0 . (4.14) b. If for some S = [S¡λ] β RN ΓN, with TrS = 0, the assumptions
G(β) A Β·Sx = 0 and Re N X Β΅, Ξ½=1 TrhGΛA(¡ν): N X Ξ±=1 S¡αβΞ±AΞ½ i = 0 , (4.15) hold, then we have the conservation law
N X Β΅=1 β βxΒ΅ XN ΞΊ=1 2Re TrhPgGΛ (¡κ) A (Sx Β· β)AΞΊ i β (Sx Β· eΒ΅)GA = 0 . (4.16) Proof a. Start from d dsG ( . . . , F¡ν(x β sa), . . . ; . . . , F β ΞΈΟ(x β sa), . . . ; x β sa ) s=0 = d dsGA(x β sa) s=0. Calculate the left hand side with the chain rule and use the assumptions
βX Β΅<Ξ½ TrhGA(¡ν) : (a Β· β)F¡ν i β X Β΅<Ξ½ TrhGA(¡ν?) : (a Β· β)F¡νβ i β a Β·GAβ =
= β2Re X Β΅<Ξ½ TrhGA(¡ν) : (a Β· β)F¡ν i . (4.17) With
(aΒ· β)F¡ν = βΒ΅(a Β· βAΞ½) β βΞ½(a Β· βAΒ΅) β {AΒ΅, a Β· βAΞ½} + {AΞ½, a Β· βAΒ΅} ,
and the antisymmetries Β΅ β Ξ½, the expression (4.17) becomes, (mind the hatΛ), βRe N X Β΅,Ξ½=1 Trh ΛGA(¡ν) : βΒ΅(a Β· βAΞ½) β {AΒ΅, a Β· βAΞ½} i = βRe N X Β΅,Ξ½=1 β βxΒ΅Trh ΛG (¡ν) A : (aΒ·βAΞ½) i + Re N X Β΅,Ξ½=1 TrhβΒ΅GΛ (¡ν) A : (aΒ·βAΞ½) + ΛG (¡ν) A : {AΒ΅, aΒ·βAΞ½} i . The 2nd term is equal to
Re N X Ξ½=1 N X Β΅=1 Tr h βAΒ΅PgGΛA(¡ν): (aΒ· βAΞ½) i = 0 , because of the E-L-equations (3.13).
The right hand side of the 1st formula of this proof equals ββΒ΅(aΒ΅LA). Hence (4.14).
b. Start from d dsG ( . . . , F¡ν(e sS x), . . . ; . . . , FΞΈΟβ (esSx), . . . ; esSx ) s=0 = d dsGA(e sS x) s=0.
Calculate the left hand side with the chain rule and use GA(β)Β·Sx = 0, 2Re X Β΅<Ξ½ TrhGA(¡ν) : (Sx Β· β)F¡ν i = = Re N X Β΅, Ξ½=1 Trh ΛGA(¡ν) : βΒ΅ (Sx Β· β)AΞ½ β {AΒ΅ , (Sx Β· β)AΞ½} β S¡αβΞ±AΞ½ i .
Because of the assumption the very final contribution vanishes. Then we proceed as in
part a.
Note The orthogonality condition (4.15) is inspired by combining Thm 4.7 with Appendix A. Indeed, another way to obtain the preceding Theorem is to rewrite Thms 4.5, 4.7 in terms of A with the aid of the table in Appendix A.
Theorem 4.10
Consider the proto-Lagrangian G of (3.6) with property (3.7) and Lagrange density as denoted in (3.8). For convenience consider g = gβ only. Suppose G satisfies
G ( . . . , esBP ¡νeβsB, . . . ; . . . , eβsB β PΞΈΟβ esBβ , . . . ; x) =G ( . . . , P¡ν, . . . ; . . . , P β ΞΈΟ, . . . ; x) , (4.18)
for all P¡ν β g β CcΓc, 1 β€ Β΅ < Ξ½ β€ N , some fixed B β g and (small) s β R .
Then, for any solution x7β . . . AΒ΅(x) . . . of the Lagrangian system of Theorem 3.3 one has
the conservation law
N X Β΅=1 β βxΒ΅Re XN Ξ½=1 Tr h [ ΛGA(¡ν)] : {B , AΞ½} i = 0 . (4.19) Proof In (4.18) replace P¡ν β F¡ν and QΞΈΟ β F
β
ΞΈΟ and put the derivative to s equal to 0
at s = 0, X 1β€Β΅<Ξ½β€N Trh[GA(¡ν)] : (BF¡νβ F¡νB)] i + X 1β€ΞΈ<Οβ€N Trh[GA(ΞΈΟ?)] : (βBβ FΞΈΟβ + FΞΈΟβ Bβ )]i = 0 . (4.20) Due to the anti-symmetry in Β΅ β Ξ½ of
BF¡ν β F¡νB = βΒ΅{B , AΞ½} β βΞ½{B , AΒ΅} β {B , {AΒ΅, AΞ½}} ,
applying convention (3.11), together with GA(¡ν?) = [GA(¡ν)]β , the 1st term of (4.20) equals the Re -part of N X Β΅=1 N X Ξ½=1 Trh[ ΛGA(¡ν)] : (BF¡νβ F¡νB) i = = N X Β΅=1 N X Ξ½=1 β βxΒ΅Tr h [ ΛGA(¡ν)]{B , AΞ½} i β N X Ξ½=1 N X Β΅=1 β βxΞ½Tr h [ ΛGA(¡ν)]{B , AΒ΅} i + β N X Ξ½=1 N X Β΅=1 Tr h [βΒ΅GΛA(¡ν)]{B , AΞ½} i + N X Β΅=1 N X Ξ½=1 Tr h [βΞ½GΛA(¡ν)]{B , AΒ΅} i + β N X Β΅=1 N X Ξ½=1 Trh[ ΛGA(¡ν)]{B , {AΒ΅, AΞ½}} i . (4.21) On the 2nd line we apply the E-L-equations (3.13) together with βΞ½GΛ
(¡ν)
A = ββΞ½GΛ (Ξ½Β΅)
A .
This together with the 3rd line leads to β N X Ξ½=1 N X Β΅=1 Tr h {AΒ΅, ΛGA(¡ν)}{B , AΞ½} i + N X Β΅=1 N X Ξ½=1 Tr h {AΞ½, ΛGA(¡ν)}{B , AΒ΅} i + β N X Β΅=1 N X Ξ½=1 Trh[ ΛGA(¡ν)]{B , {AΒ΅, AΞ½}} i .
These 3 terms add up to 0 because for each pair Β΅, Ξ½ separately we can apply the identity βTrh{M , G} : {B , N}i + Trh{N , G} : {B , M}i = TrhG : {B , {M , N}}i, (4.22)
for matrices G, B, M, N β CrΓr.
(Of course the two terms on the 3rd line of (4.21) are equal. But then, using that equality, the latter trick no longer works for each index pair Β΅, Ξ½ separately!)
Thus we found out that (4.20) corresponds to (4.19).
5
Static/Dynamic Gauge Extensions of Lagrangians
A basic ingredient for this section is a (fixed) Lie-group G β CcΓcof invertible cΓc-matrices.
Its Lie-algebra g is a R -linear subspace of CcΓc. Important examples are (subgroups of) GJ, for some fixed invertible matrix J β CcΓc. The relevant definitions are as in section 3,
GJ = U β CcΓc
Uβ J U = J , gJ = A β CcΓc
Aβ J + J A = 0 . (5.1) In the discussion to follow suitable subspaces of
the group Gloc =Cβ(RN: G) and the R -linear space Cβ(RN: g)
will be used. It will be tacitly assumed that the behaviour at β of the considered subspaces is such that our formulae make sense. TheCβ-smoothness condition can often be relaxed. Neither of those assumptions will bother us.
The group action from the right of Cβ(RN: G) on Cβ
(RN: CrΓc) is naturally defined by
Cβ
(RN: CrΓc) ΓCβ(RN: G) β Cβ(RN: CrΓc) : (Ξ¨U )(x) = Ξ¨(x)U (x). For each 1 β€ Β΅ β€ N , a group action from the right of Cβ(RN : G) on Cβ
(RN : g) is
defined by Cβ
(RN: g)ΓCβ(RN: G) β Cβ(RN: g) : (AΒ΅C U)(x) = Uβ1(x)AΒ΅(x)U (x)βUβ1(x)(βΒ΅U )(x) .
In the proof of Thm 1.2 it has been shown that this action (βgauge transformβ)is indeed a (inhomogeneous) group action. This means
[AΒ΅C U] C V = AΒ΅C (UV) . (5.2)
As before, for given AΒ΅, AΞ½ βCβ(RN: g) , 1 β€ Β΅, Ξ½ β€ N , define
F¡ν = βΒ΅AΞ½ β βΞ½AΒ΅β {AΒ΅, AΞ½} β Cβ(RN: g) . (5.3)
Then
Uβ1F¡νU = βΒ΅(AΞ½C U) β βΞ½(AΒ΅C U) β {(AΒ΅C U) , (AΞ½C U)} . (5.4)
Theorem 5.1
Fix a matrix Lie-Group G β CcΓc. Suppose a proto-Lagrangian L , cf. (2.13), to be
G-invariant, i.e. 3
β U β G β P β CrΓc
β R β CN rΓc
β x β RN :
3Property (5.5) is named Global Gauge Invariance by physicists. The conclusion of Theorem 5.1 is
named, in physicistsβ vernacular, the property of Local Gauge Invariance. In mathematiciansβ jargon however, the usage of βglobalβ, as opposed to βlocalβ, usually refers to a more involved (more difficult) notion.
L (PU ; Uβ
Pβ ; RU ; Uβ Rβ ; x) =L (P ; Pβ ; R ; Rβ ; x) (5.5) Then, for all xβ RN, the statically gauge extended Lagrangian density
LΟ, A(x) = L (Ξ¨ ; Ξ¨β ; . . . , β¡Ψ + Ξ¨AΒ΅, . . . ; . . . , β¡Ψβ + Aβ ¡Ψ β
, . . . ; x) , (5.6) with any Ξ¨ βCβ(RN: CrΓc) , AΒ΅βCβ(RN: g) , 1 β€ Β΅ β€ N ,
equals the statically gauge extended Lagrangian density
LΟU , AC U(x) = (5.7)
=L (Ξ¨U ; Uβ Ξ¨β ; . . . , βΒ΅(Ξ¨U )+(Ξ¨U )(AΒ΅C U) , . . . ; . . . , βΒ΅(Ξ¨U )β +(AΒ΅C U)β (Ξ¨U )β , . . . ; x) ,
with any U βCβ
(RN: G) . In (5.6),(5.7) we wrote Ξ¨ instead of Ξ¨(x), etc.
Proof Straightforward calculation. Example 5.2 Consider the proto-Lagrangian, cf. (2.13),
L (P; Q>
; R ; S>; x) = i Tr[Q>(X
Β΅
ΞΒ΅RΒ΅+ M P)]
with fixed ΞΒ΅, M β CrΓr and [ΞΒ΅]β = ΞΒ΅, Mβ = βM . Put G = U(c) β CcΓc, that is the unitary group GI, with I the identity matrix. Our proto-Lagrangian is U(c)-invariant
i Tr[Uβ Pβ (ΞΒ΅RΒ΅U + M PU)] = i Tr[Pβ (ΞΒ΅RΒ΅+ M P)] , U β U(c) ,
because Uβ = Uβ1 and the properties of Tr. Then the statically extended Lagrangian density
LΟ, A(x) = i Tr[Ξ¨β ΞΒ΅(β¡Ψ + Ξ¨AΒ΅) + M Ξ¨] , (5.8)
with any Ξ¨ βCβ(RN: CrΓc) , AΒ΅βCβ(RN: u(c)) , 1 β€ Β΅ β€ N ,
equals the statically extended Lagrangian density
LΟU , ACU(x) = i Tr[Uβ Ξ¨β ΞΒ΅(βΒ΅(Ξ¨U ) + Ξ¨U (Uβ1AΒ΅U β Uβ1βΒ΅U )) + M Ξ¨U] , (5.9)
with any U βCβ
(RN: U(c)) .
Theorem 5.3
β’ Suppose that the statically gauge extended Lagrange density LΟ, A, cf. (5.6) leads to an
R -valued Langrangian functional LΟ, A. The E-L-equations are
L(o) Ο,Aβ N X Β΅=1 β βxΒ΅[L (Β΅) Ο,A] β [AΒ΅L (Β΅) Ο,A] = 0 , Pg Ξ¨β [LΟ,A(ΞΊ)β +LΟ,A(ΞΊ?)]= 0 , Pg Ξ¨β [LΟ,A(ΞΊ)β βLΟ,A(ΞΊ?)] i = 0 , 1 β€ ΞΊ β€ N . (5.10) Here Pg : CcΓcβ CcΓc denotes the R -orthogonal projection on g.
β’ If it happens that Pg( iZ) = iPgβ₯Z , Z β CcΓc, the 2nd line in (5.10) reduces to
Ξ¨β LΟ,A(ΞΊ)β + (PgβPgβ₯)Ξ¨
β L(ΞΊ?)
Ο,A = 0 , 1 β€ ΞΊ β€ N . (5.11)
β’ In the important special case g = gJ, with J = Jβ = Jβ1, (5.11) can be written
L(ΞΊ)
Ο,AΞ¨ β J Ξ¨
β L(ΞΊ?)
Ο,A J = 0 , 1 β€ ΞΊ β€ N . (5.12)
Proof β’ The perturbed statically extended LagrangianLΟ, A reads
L (Ξ¨ + Ξ΅H ; Ξ¨β + Ξ΅?K ; . . . , βΒ΅(Ξ¨ + Ξ΅H) + (Ξ¨ + Ξ΅H)(AΒ΅+ Ρκδ¡κH) , . . . ; ; . . . , βΒ΅(Ξ¨β + Ξ΅?K) + (Aβ Β΅+ Ρκδ¡κHβ )(Ξ¨β + Ξ΅?K) , . . . ; x) The results of d dΞ΅ Ξ΅=0, d dΞ΅? Ξ΅?=0, d dΡκ
Ρκ=0, 1 †κ †N , being put to 0 are,
for all functions H , K , H , TrL(o): H + X Β΅ TrL(Β΅): βΒ΅H + X Β΅ TrL(Β΅): HAΒ΅ = 0 , TrL(o?): K + X Β΅ TrL(Β΅?): βΒ΅K + X Β΅ TrL(Β΅?): Aβ Β΅K = 0 , X Β΅ TrL(Β΅): Ψδ¡κH + X Β΅ TrL(Β΅?): δ¡κHβ Ξ¨β = 0 , 1 β€ ΞΊ β€ N .
The usual partial integration techniques applied to the first two lines lead to the E-L-equations for Ξ¨. Also use Theorem 2.4.
From the final line we arrive at (5.10) because of the trace identity TrhXZ + YZβ i = Re TrhXβ + Y β Zi β i Re TrhX β β Y i β Zi. (5.13) β’ If for X, Y β CcΓc one has P
g(X + Y) = 0 and Pgβ₯(X β Y) = 0, it follows that
X + (PgβPgβ₯)Y = 0 and also Y + (PgβPgβ₯)X = 0.
β’ In this special case (PgβPgβ₯)Y = βJ Y β
Examples 5.4
Note that in the E-L-equations (5.10) the AΒ΅ occur only βalgebraicallyβ.
The βΒ΅A are not involved!
a. For the Lagrangian densities from examples 2.5a and 5.2 the 2nd set of E-L-equations (5.12) does not depend on A. If we choose g = gJ, the 2nd line reads
Ξ¨β ΞΞΊΞ¨ = 0 , 1 β€ ΞΊ β€ N .
It means that Ξ¨ can only take values in a cone in CrΓc. If one of the ΞΞΊ = ΞΞΊβ is strictly
positive, the only solutions are Ξ¨ = 0, the trivial ones. If a nontrivial choice for Ξ¨ is possible it can be substituted in the 1st E-L-equation and we are left with an algebraic equation for the AΞΊ.
b. For the Lagrangian densities from example 2.5c, again with g = gJ , the 2nd set of
E-L-equations becomes N X Β΅=1 [β¡Ψ + Ξ¨AΒ΅]β Ξ¡κΨ β J XN Β΅=1 [Ξ¨β ΞΞΊΒ΅[β¡Ψ + Ξ¨AΒ΅] J = 0 , 1 β€ ΞΊ β€ N , which is algebraic in the AΞΊ.
Finally we want to consider the dynamically gauge extended Lagrangian density or Gauge field extended Lagrangian density of typeLΟ, A(x) +GA(x) .
Theorem 5.5
Fix a matrix Liegroup G β CcΓc with Lie algebra g β CcΓc and property gβ = g. Fix a proto Lagrangian of type (2.13)
(P; Q>; R ; S>; x) 7β L (P; Q>; R ; S>; x) ,
leading to a R -valued Lagrangian functional L. Require the special property β P β R β x : Pg
Pβ L(ΞΊ)β (P; Pβ ; R ; Rβ ; x) β L(ΞΊ?)(P; Pβ ; R ; Rβ ; x) i
= 0 . (5.14) Fix a second proto Lagrangian of type (3.6) and such that
β R¡ν β g : G (. . . , R¡ν, . . . ; . . . , R β
ΞΈΟ, . . . ; x) β R .
Consider the dynamically extended Lagrangian density
LΟ, A(x) +GA(x) = L (Ξ¨ ; Ξ¨β ; . . . , β¡Ψ + Ξ¨AΒ΅, . . . ; . . . , β¡Ψβ + Aβ ¡Ψ β , . . . ; x) + +G ( . . . , F¡ν(x), . . . ; . . . , F β ΞΈΟ(x), . . . ; x ) (5.15)
with any Ξ¨ βCβ(RN: CrΓc) , AΒ΅βCβ(RN: g) , 1 β€ Β΅ β€ N .
β’ The Euler-Lagrange equations are, with LΟ,A(o) instead of LΟ,A(o)(x), etc., [LΟ,A(o)] β N X Β΅=1 β βxΒ΅[L (Β΅) Ο,A] β [AΒ΅L (Β΅) Ο,A] = 0 , Pg Ξ¨β [LΟ,A(ΞΊ)β +LΟ,A(ΞΊ?)] β 2PN Β΅=1 βΒ΅Pg[ ΛGA(¡κ)] β {AΒ΅,Pg[ ΛGA(¡κ)]} β = 0 , 1 β€ ΞΊ β€ N . (5.16) Here Pg : CcΓcβ CcΓc denotes the R -orthogonal projection on g.
β’ In the special case g = gJ, with J = Jβ = Jβ1, the 2nd line in (5.16) can be rewritten
L(ΞΊ) Ο,AΞ¨ β J Ξ¨ β L(ΞΊ?) Ο,A J β 2 N X Β΅=1 βΒ΅Pg[ ΛG (¡κ) A ] β {AΒ΅,Pg[ ΛG (¡κ) A ]} = 0 , 1 β€ ΞΊ β€ N . (5.17) Proof β’ The perturbed gauge supplemented Lagrangian reads
L (Ξ¨+Ξ΅H ; Ξ¨β +Ξ΅?K ; . . . , βΒ΅(Ξ¨+Ξ΅H)+(Ξ¨+Ξ΅H)(AΒ΅+Ρκδ¡κH) , . . . ; ; . . . , βΒ΅(Ξ¨β +Ξ΅?K)+(Aβ Β΅+Ρκδ¡κHβ )(Ξ¨β +Ξ΅?K) , . . . ; x) + +G (. . . , F¡ν,Ρκ, . . . ; . . . , F β ΞΈΟ,Ρκ, . . . ; x) , 1 β€ ΞΊ β€ N , where F¡ν;Ξ΅,ΞΊ= F¡ν+ Ρκδνκ h βΒ΅H β {AΒ΅, H} i β Ρκδ¡κ h βΞ½H β {AΞ½, H} i , The results of d dΞ΅ Ξ΅=0, d dΞ΅? Ξ΅?=0 d dΡκ
Ρκ=0, being put to 0 are, respectively,
TrL(o): H + X Β΅ TrL(Β΅): βΒ΅H + X Β΅ TrL(Β΅): HAΒ΅ = 0 , TrL(o?): K + X Β΅ TrL(Β΅?): βΒ΅K + X Β΅ TrL(Β΅?): Aβ Β΅K = 0 , X Β΅ Tr L(Β΅): Ψδ ¡κH + X Β΅ Tr L(Β΅?): Ξ΄ ¡κHβ Ξ¨β + β 2X Β΅ Re TrhPgβΒ΅GΛ (¡κ?) A +Pg{Aβ Β΅ ,PgGΛ (¡κ?) A } β [H]i = 0 , 1 β€ ΞΊ β€ N . With (5.13) the 3rd set of equations can be rewritten
Re Trh Ξ¨β ([L(ΞΊ)]β + [L(ΞΊ?)]β Hi + iRe Trh iΞ¨β ([L(ΞΊ)]β β [L(ΞΊ?)]β Hi+
β 2X Β΅ Re TrhPgβΒ΅GΛ (¡κ) A β {AΒ΅ ,PgGΛ (¡κ) A } β β [H]i = 0 , 1 β€ ΞΊ β€ N . Because of assumption (5.14) the iRe Tr-term cancels. The assumption gβ = g enables us to interchange β and Pg.
β’ Finally (5.17) follows as in the proof of Thm (5.3). Finally we want to find the conservation law of βconserved currentsβ.
Theorem 5.6
Consider proto-Lagrangians L and G as in Theorem 5.5. Suppose for some B β g they both have the invariance properties
L (PesB; (PesB)β ; . . . QΞ»esB. . . ; . . . (QΞ»esB)β . . . ; x) = =L (P; Pβ ; . . . QΞ». . . ; . . . Q β Ξ». . . ; x) +O(s 2) , (5.18) G ( . . . , eβsB R¡νesB, . . . ; . . . , esB β Rβ ΞΈΟeβsBβ , . . . ; x ) = = G ( . . . , R¡ν, . . . ; . . . , R β ΞΈΟ, . . . ; x ) +O(s 2) . (5.19)
Then, the solutions to the E-L-system (5.16) satisfy the conservation law
N X Β΅=1 β βxΒ΅ n TrhLΟ,A(Β΅): Ξ¨Bi + TrhLΟ,A(Β΅?): Bβ Ξ¨β i+ N X ΞΊ=1 2Re TrhPgGΛ (¡κ) A : {AΞΊ, B} i o = 0. (5.20) Proof Add the Lagrange densitiesLΟ,A and GA and put to 0 the
d ds of the expression L (Ξ¨esB; esBβ Ξ¨β ; . . . , β¡ΨesB+ Ξ¨AΒ΅esB, . . . ; . . . , esB β β¡Ψβ + esB β Aβ ¡Ψβ , . . . ; x) + +G ( . . . , eβsBF¡νesB, . . . ; . . . , esB β FΞΈΟβ eβsBβ , . . . ; x ) One finds, TrhLΟ,A(o): Ξ¨Bi+X Β΅ TrhLΟ,A(Β΅): β¡ΨB i +X Β΅ TrhLΟ,A(Β΅): Ξ¨AΒ΅B i + + TrhLΟ,A(o?): Bβ Ξ¨β i +X Β΅ TrhLΟ,A(Β΅?): Bβ β¡Ψβ i +X Β΅ TrhLΟ,A(Β΅?): Bβ Aβ ¡Ψβ i + + X Β΅<Ξ½ TrhGA(¡ν): {F¡ν, B} i + X ΞΈ<Ο TrhGA(ΞΈΟ?): {Bβ , FΞΈΟβ }i = 0 . (5.21)
Rewrite the 3rd term and the 6th term: X Β΅ TrhLΟ,A(Β΅): Ξ¨AΒ΅B i = X ΞΊ TrhLΟ,A(ΞΊ): Ξ¨{AΞΊ, B} i +X Β΅ TrhAΒ΅L (Β΅) Ο,A: Ξ¨B i , X Β΅ TrhLΟ,A(Β΅?): (Ξ¨AΒ΅B)β i = X ΞΊ TrhLΟ,A(ΞΊ?): (Ξ¨{AΞΊ, B})β i +X Β΅ TrhAβ Β΅LΟ,A(Β΅?): (Ξ¨B)β i. These identities, together with the 1st E-L-equation of (5.16) turn the first 6 terms of (5.21) into X Β΅ βΒ΅TrhL (Β΅) Ο,A: Ξ¨B i +X Β΅ βΒ΅TrhL (Β΅?) Ο,A : B β Ξ¨β i+ +X ΞΊ TrhLΟ,A(ΞΊ): Ξ¨{AΞΊ, B} i +X ΞΊ TrhLΟ,A(ΞΊ?): (Ξ¨{AΞΊ, B})β i With Trace identity (5.13) and condition (5.14) the latter becomes
X Β΅ βΒ΅TrhL (Β΅) Ο,A: Ξ¨B i +X Β΅ βΒ΅TrhL (Β΅?) Ο,A : B β Ξ¨β i+ + 2 N X ΞΊ, Β΅=1 Re TrhPgβΒ΅GΛ (¡κ) A β {AΒ΅ ,PgGΛ (¡κ) A } : {AΞΊ, B} i . (5.22) Next, because of (anti)symmetry, B β g being constant and the definition of F¡ν, the final
2 terms of (5.21) equal to Re N X Β΅,Ξ½=1 Trh ΛGA(¡ν): {F¡ν, B} i = Re N X Β΅,Ξ½=1 Trh ΛGA(¡ν) : βΒ΅{AΞ½, B} i + β Re N X Β΅, Ξ½=1 Trh ΛGA(¡ν): βΞ½{AΒ΅, B} i β Re N X Β΅, Ξ½=1 TrhGA(¡ν): {{AΒ΅, AΞ½} , B} i = = 2Re N X Β΅,Ξ½=1 Trh ΛGA(¡ν) : βΒ΅{AΞ½, B} i β Re N X Β΅, Ξ½=1 TrhGA(¡ν): {{AΒ΅, AΞ½} , B} i . (5.23) If we add (5.22), (5.23), we arrive at (5.20), up to a term
β Re N X ΞΊ, Β΅=1 2 Trh{AΒ΅ ,PgGΛ (¡κ) A } : {AΞΊ, B} i + TrhPgG (¡κ) A : {{AΒ΅, AΞΊ} , B} i .
Split the first term in this summation. It becomes, β Re N X ΞΊ, Β΅=1 Trh{AΒ΅ ,PgGΛ (¡κ) A } : {AΞΊ, B} i β Trh{AΞΊ ,PgGΛ (¡κ) A } : {AΒ΅, B} i + + TrhPgG (¡κ) A : {{AΒ΅, AΞΊ} , B} i . Each term in this sum equals 0 because of the trace identity
Tr h {M , G} : {K , B}i β Trh{K , G} : {M , B}i + Tr h G : {{M , K} , B} i = 0. Indeed, note that for any M, G, K, B β CcΓc,
Trh MGKB β GMKB β MGBK + GMBK β KGMB + GKMB +
+ KGBM β GKBM + GMKB β GKMB β GBMK + GBKM i
= 0 .
A
Addendum on Free Gauge Fields
If we put GA(x) = G ( . . . , F¡ν(x), . . . ; . . . , F β ΞΈΟ(x), . . . ; x ) = =G (. . . , βΒ΅AΞ½ββΞ½AΒ΅β{AΒ΅, AΞ½}, . . . ; . . . , βΒ΅Aβ Ξ½ββΞ½Aβ Β΅+{A β Β΅, A β Ξ½}, . . . ; x ) == L (A(x) ; Aβ (x) ; . . . , βΒ΅A(x), . . . ; . . . , βΒ΅Aβ (x), . . . ; x) , (A.1)
with A= col[. . . , AΒ΅, . . .], which now plays the role of Ξ¨ in section 2, we get, in accordance
with our notation in section 2, L(o) A = row [ . . . β PN Β΅=1{ ΛG (¡κ) A , AΒ΅} . . . ] L(1) A = row [ 0 G (12) A G (13) A . . . G (1ΞΊ) A . . . G (1N ) A ] L(2) A = row [ βG (12) A 0 G (23) A . . . G (2ΞΊ) A . . . G (2N ) A ] L(3) A = row [ βG (13) A βG (23) A 0 . . . G (3ΞΊ) A . . . G (3N ) A ] . . . = row [ . . . ] L(ΞΊ) A = row [ βG (1ΞΊ) A βG (2ΞΊ) A βG (3ΞΊ) A . . . 0 . . . G (ΞΊN ) A ] . . . = row [ . . . ] L(N ) A = row [ βG (1N ) A βG (2N ) A βG (3N ) A . . . βG (ΞΊN ) A . . . 0 ] (A.2) With convention (3.15) the lower N rows of this table simplify to
L(Β΅)
A = row [. . . , ΛG (¡κ)
A , . . .] , 1 β€ Β΅, ΞΊ β€ N . (A.3)
Table (A.2) enables to reduce the proof of Theorem 3.2 to an application of Theorem 2.4. Because of property (3.7) it is obvious that all βcomponentsβ ofLA(Β΅?), 0 β€ Β΅ β€ N, are the hermitean transposed of the components of LA(Β΅), 0 β€ Β΅ β€ N . Only for LA(o?) this is not immediately obvious. Let us check it in an ad hoc way by calculating the ΞΊ-th component of LA(o?). In (A.1) replace {Aβ Β΅, Aβ Ξ½} by the perturbation {Aβ
Β΅ + Ρδ¡κH , Aβ Ξ½ + ΡδνκH}.
Now differentiate the result to Ξ΅. At Ξ΅ = 0 it becomes X 1β€Β΅<Ξ½β€N TrhGA(¡ν?) : {δ¡κH , Aβ Ξ½} + {A β Β΅, δνκH} i = = X ΞΊ<Ξ½β€N TrhGA(ΞΊΞ½?) : {H , Aβ Ξ½}i + X 1β€Β΅<ΞΊ TrhGA(¡κ?) : {Aβ Β΅, H} i = = X ΞΊ<Ξ½β€N Trh{Aβ Ξ½,GA(ΞΊΞ½?)} : Hi+ X 1β€Β΅<ΞΊ Trh{GA(¡κ?), Aβ Β΅} : Hi = Trh N X Β΅=1 { ΛGA(¡κ?), Aβ Β΅} : Hi.
Finally one finds hXN Β΅=1 { ΛGA(¡κ?), Aβ Β΅}iβ = β N X Β΅=1 { ΛGA(¡κ), AΒ΅} .
Remark on Thm 4.9-b: If it happens that G (. . . , esSΞ» Β΅β Ξ»AΞ½βesS ΞΈ Ξ½β ΞΈAΒ΅β{AΒ΅, AΞ½}, . . . ; . . . , esS Ξ» Β΅β Ξ»Aβ Ξ½βe sSΞΈ Ξ½β ΞΈAβ Β΅+{A β Β΅, A β Ξ½}, . . . ; x ) = =G (. . . , βΒ΅AΞ½ββΞ½AΒ΅β{AΒ΅, AΞ½}, . . . ; . . . , βΒ΅Aβ Ξ½ββΞ½Aβ Β΅+{A β Β΅, A β Ξ½}, . . . ; x ) +O(s 2) , it follows that Re X Β΅<Ξ½ TrhGA(¡ν) : S¡λβΞ»AΞ½ β SΞ½ΞΈβΞΈAΒ΅ i = 0 .
B
Electromagnetism
Some more details on Example 3.4B: GA= X 0β€Β΅<Ξ½β€3 (β1)δ¡0+δν0TrFβ ¡νF¡ν G(01) A = βF β 01 G (02) A = βF β 02 G (03) A = βF β 03 G (12) A = F β 12 G (13) A = F β 13 G (23) A = F β 23
Now (3.19) reads, for 0 β€ ΞΊ β€ 3, ΞΊ = 0 : β1G (01) A + β2G (02) A + β3G (03) A = = ββ1(β0A β 1β β1A β 0) β β2(β0A β 2β β2A β 0) β β3(β0A β 3β β3A β 0) = ββ0(β1Aβ 1+ β2Aβ 2+ β3Aβ 3) + β1β1Aβ 0+ β2β2Aβ 0+ β3β3Aβ 0 ΞΊ = 1 : ββ0G (01) A + β2G (12) A + β3G (13) A = = β0(β0A β 1β β1A β 0) + β2(β1A β 2β β2A β 1) + β3(β1A β 3 β β3A β 1) = β0β0A β 1+ β1(ββ0A β 0+ β1A β 1+ β2A β 2+ β3A β 3) β (β1β1 + β2β2+ β3β3)A β 1 ΞΊ = 2 : ββ0G (02) A β β1G (12) A + β3G (23) A = = β0(β0A β 2β β2A β 0) β β1(β1A β 2β β2A β 1) + β3(β2A β 3β β3A β 2) = β0β0A β 2+ β2(ββ0A β 0+ β1A β 1+ β2A β 2+ β3A β 3) β (β1β1 + β2β2+ β3β3)A β 2 ΞΊ = 3 : ββ0G (03) A β β1G (13) A β β2G (23) A = = β0(β0Aβ 3β β3Aβ 0) β β1(β1A3β β β3Aβ 1) β β2(β2Aβ 3β β3Aβ 2) = β0β0A β 3+ β3(ββ0A β 0+ β1A β 1+ β2A β 2+ β3A β 3) β (β1β1 + β2β2+ β3β3)A β 3