pdq − H dt
VIOREL PETREHUS¸
We prove the existence and uniqueness of a Cartan form for Lagrangians onJkY. We restricted to fiber manifoldsπX,Y :Y →X whereX is unidimensional.
AMS 2000 Subject Classification: Primary 57R55, Secondary 53Z05.
Key words: fiber space, manifold of jets, Cartan form.
In this paper we follow the lines of [7], [6], [4] from Lagrangians onJ1Y orJ2Y to Lagrangians on JkY. The first part is standard and more general computations can be found in [6], [4], [8]. Our result is that the form θLag
associated with a lagrangian Lag onJkY is well defined on the space J2k−1Y, and is contained in Theorem 3. Many other forms θ with some properties of θLagcan be defined, see [5].This form is important in the study of conservation laws for Lag or in the study of conservative numerical methods for the Euler equations associated with Lag.
The basic space of our computations isJkY, the space of k jets of sections of the fiber spaceπXY :Y →X. In the sequelX is the real field and Y is an N+ 1 dimensional manifold. The fibre ofπXY :Y →X is diffeomorphic to a given manifoldQ. All manifolds are finite dimensional. Ak jet over x∈X is a class of equivalence of sections ofY defined in a neighborhood ofxsuch that two sections are equivalent iff their Taylor developments agree up to orderk.
The jet corresponding to sections will be denoted by jks(x). The definition is independent of the local coordinates onX or Q. If x is a local coordinate onXand
y1, y2, . . . , yN
are local coordinates on Q, then onJkY we use the coordinates
x, yA, yiA
A=1,...,N, i=1,...,k, where for the jet of a section s:X → Y, s(x) =
x, s1(x), s2(x), . . . , sN(x)
we have yA = sA(x), yAi = ddxiisA(x).
In what follows y0A is identical with yA. The correspondence x → jks(x) is a section of JkY and is denoted by jks. By truncation of a k development to a l development (l < k) we get a projection πJlY,JkY : JkY → JlY. The projective limit of these spaces is denoted byJ∞Y and the limit of their jets
MATH. REPORTS9(59),4 (2007), 357–368
corresponding to section s at x ∈X is denoted j∞s(x). For more about jet spaces and the change of coordinates see [10],
A projetable diffeomorphism η of Y is a diffeomorphism ηY : Y → Y such that there exists a diffeomorphismηX of X such that the diagram
(1)
Y −→ηY Y
πX,Y
πX,Y
X −→ηX X
is commutative; η can be extended to a diffeomorphism ηJkY of JkY such that the diagram
jkY ηJkY−→ jkY
πJk−1Y,JkY πJk−1Y,JkY
... ...
j1Y ηJ−→1Y j1Y
πY,J1Y
πY,J1Y
Y −→ηY Y
πX,Y πX,Y
X −→ηX X
is commutative. The extension is given by
(2) ηJkY
jks(x)
=jk
ηY ◦s◦η−X1
(ηX(x)).
Using local coordinates we define the total derivativeDx = ∂x∂ +y1A ∂
∂yA+ yA2 ∂
∂yA1 +· · ·+yAk∂y∂A
k−1 as a function from JkY to the tangent spaceT Jk−1Y, which sends a pointγ ∈JkY to a vector tangent toJk−1Y atπJk−1Y,JkY (γ).
More conveniently, we considerDx as a “vector field” on J∞Y by extending the preceding definition toDx= ∂x∂ +∞
k=0
N
A=1yAk+1∂y∂A k .
Any functionf :JkY →R can be considered as a function also denoted abusively f, defined on any JnY (n > k) by the composition JnY πJkY,JnY−→
JkY −→f R and consequently defined on J∞Y. Analogously, any form ω ∈ ΛJkY gives a form on JnY (n > k) by the formula πJ∗nY,JkYω, so it gives a form onJ∞Y. We defineDx,p =Dx◦Dx◦ · · · ◦Dx (p times), which gives for f :JkY → R, a function Dx,pf = DxDx(. . . Dx(f)) defined on Jk+pY. For
p= 0 we define Dx,0= identity. For a section sof Y we have d
dx
f
x, sA(x),dsA(x)
dx , . . . ,dksA(x) dxk
=
= (Dxf)
x, sA(x),dsA(x)
dx , . . . ,dksA(x)
dxk ,dk+1sA(x) dxk+1
= (Dxf)
jk+1s(x) . For the one parameter group of diffeomorphisms ηYε : Y → Y which covers the one parameter groupηXε :X →X, its infinitesimal generator is of the form
(3) V =Vx(x) ∂
∂x+VA(x, y) ∂
∂yA
(by convention the sum is taken over all values of A from 1 to N). The extended groupηεJkY :JkY →JkY has a generator denoted by
(4) jk(V) =Vx(x) ∂
∂x + N A=1
VA(x, y) ∂
∂yA + N A=1
k j=1
VjA ∂
∂yAj . The components ofjk(V) are given (see [8]) by
VjA=Dx,jVvA+VxyAj+1, where
(5) VvA =VA−y1AVx.
The field Vv =
AVvA ∂∂yA =
A
VA−y1AVx ∂
∂yA is called the vertical part ofV. The field jk(V) can be split as
(6) jk(V) =jk(V)v+jk(V)h, where the vertical part is
(7) jk(V)v =
A
VvA ∂
∂yA + N A=1
k j=1
Dx,jVvA ∂
∂yAj = N A=1
k j=0
Dx,jVvA ∂
∂yjA and the horizontal one is
(8) jk(V)h =Vx ∂
∂x+ N A=1
k j=0
VxyAj+1
∂
∂yAj =VxDx. One sees that
jk(V)h
jks(x) =jks∗(x)
Vx ∂
∂x
.
The field jk(V) is tangent to JkY, butjk(V)v and jk(V)h have extra terms involvingykA+1 which cancel each other; jk(V)v and jk(V)h are always defined on J∞Y by extending the sumation over j to ∞. To avoid such
technical difficulties as (8) not being tangent to JkY, we shall consider all fields as functions or forms defined onJ∞Y. Functions and forms are supposed depending onJkY for some finitek depending on that form or function. For a vector field V = Vx ∂∂x+
A∞
j=0VjA, we assume that Vx is defined on JmY and VjAdepends on Jj+mY for a fixedm which depends on V;Dx is an example of such a vector field. In this case, the basic operations of calculus with differential forms: outer product, exterior derivative, Lie derivative, etc.
make sense (see [1]).
The differential forms θA0 = θA = dyA −y1Adx, θjA = dyjA−yAj+1dx, for j ≥ 1 are defined using local coordinates. For any section s we have jks∗
θAj
= 0, andθAj (Dx) = 0. For more on jet spaces see [1], [6], [10].
By a Lagrangian on JkY we understand a differential form of degree 1 which with any jetγ ∈JkY associates a value in π∗X,JkY
Λ1X
. Using local coordinates, a Lagrangian is a differential form
(9) Lag =L
x, yA, yA1, . . . , yAk dx.
Consider now aC∞-family of sections sε: [a(ε), b(ε)]→Y defined forε in a neighborhood of 0, and letζ = ddε|ε=0sε. Then if all fits into a coordinate chart, integration by parts gives (s0 equals s):
d dε
ε=0
b(ε)
a(e)
L
jksε(x) dx= (10a)
= d dε
ε=0
b(ε)
a(e)
L
x, sεA(x),dsεA(x)
dx , . . . ,dksεA(x) dxk
dx=
= N A=1
k j=0
b
a (−1)j d dxj
∂L
∂yjA
jks(x) ·ζA(x)dx+
(10b)
+L
jks(b) b(0)−L
jks(a) a(0)+
(10c)
+ N A=1
k j=1
j−1
m=0
(−1)m d dxm
∂L
∂yjA
jks(x) ·dj−m−1ζA dxj−m−1
b
a
=
= N A=1
k j=0
b
a (−1)jDx.j ∂L
∂yAj
j2ks(x) ·ζA(x)dx+
(10d)
+L
jks(b) b(0)−L
jks(a) a(0)+
(10e)
+ N A=1
k j=1
j−1
m=0
(−1)mDx,m ∂L
∂yAj
j2k−1s(x) ·dj−m−1ζA dxj−m−1
b
a.
The variation splits into an integral over [a, b] and a boundary part.
We shall express the variation in terms of the differential form below defined locally onJ2k−1Y as
θLag=Ldx+ N A=1
k j=1
j−1
m=0
(−1)mDx,m ∂L
∂yjA·θj−m−A 1 = (11)
=Ldx+ N A=1
k−1
b=0 k−b−1
m=0
(−1)mDx,m ∂L
∂ymA+b+1 ·θAb .
Let us look at this variation in a different way, as in [6], [4]. Let M the space of C∞-applications ϕ : [0,1] → Y and ϕX : [0,1] → X with ϕX injective such thatϕX =πX,Y ◦ϕ. Thens=ϕ◦ϕ−X1 is a section ofY defined on [a, b] =ϕX([0,1])⊂X and we define the action S by
S(ϕ) = b
a L
jks(x) dx= b
a L jk
ϕ◦ϕ−X1
(x) dx= (12)
= b
a jk
ϕ◦ϕ−X1∗ Lag.
LetηYε be a one-parameter group of diffeomorphisms of Y which covers the one parameter groupηXε ofX, and letV be its infinitesimal generator. If ϕεdef= ηεY ◦ϕand ϕεX def= ηXε ◦ϕX, then we have ϕεX =πX,Y ◦ϕε. In this case, sε =ϕε◦ϕε−X 1is a section of Y defined on [a(ε), b(ε)] =ϕεX([a, b]) and there is defined the action
S(ϕε) = b(ε)
a(ε)
L
jksε(x) dx= b(ε)
a(ε)
L jk
ϕε◦ϕε−X 1
(x) dx=
= b(ε)
a(ε)
jk
ϕε◦ϕε−X 1∗ Lag.
The definition of ηεJkY as ηJεkY
jk
ϕ◦ϕ−X1
(x) =jk
ηεY ◦ϕ◦ϕ−X1◦ηε−X 1
(ηεX(x)) =
=jk
ϕε◦ϕε−X 1
(ηεX(x)) implies
jk
ϕ◦ϕ−X1∗
◦ηJε∗kY =ηXε∗◦jk
ϕε◦ϕε−X 1∗ .
Using this, we get (L is the Lie derivative)
∂
∂ε
ε=0S(ϕε) = ∂
∂ε
ε=0
b(ε)
a(ε)
jk
ϕε◦ϕε−X1∗ Lag =
= ∂
∂ε
ε=0
b
a ηε∗X jk
ϕε◦ϕε−X 1∗ Lag =
= ∂
∂ε
ε=0
b
a jk
ϕ◦ϕ−X1∗
ηε∗JkYLag =
= b
a jk
ϕ◦ϕ−X1∗
Ljk(V)Lag.
Now, we split the Lie derivative asLjk(V)=Ljk(V)h+Ljk(V)v and obtain
[a,b]
jk
ϕ◦ϕ−X1∗
Ljk(V)hLag =
[a,b]
jk
ϕ◦ϕ−X1∗
LVxDxLag = (13a)
=
[a,b]
jk
ϕ◦ϕ−X1∗
Ljk(ϕ◦ϕ−1X)∗(Vx ∂∂x)Lag = (13b)
=
[a,b]
LVx ∂
∂xjk
ϕ◦ϕ−X1∗ Lag =
=
[a,b]
diVx ∂
∂x +iVx ∂
∂xd jk
ϕ◦ϕ−X1∗ Lag = (13c)
=
[a,b]diVx ∂
∂xL jk
ϕ◦ϕ−X1
(x) dx=
= L jk
ϕ◦ϕ−X1
(x) ·Vxb
a=L jks(b)
b(0)−L
jks(a)
a(0) = (13d)
= j2k−1
ϕ◦ϕ−1X ∗
ij2k−1(V)hθLagb
a= (13e)
=
∂[a,b]
j2k−1
ϕ◦ϕ−X1∗
ijk(V)hθLag.
So, the first two terms of the boundary part of the variation (10e) come fromj2k−1(V)h. Necessarily, the third term of (10e) is equal tob
ajk
ϕ◦ϕ−X1∗
·
Ljk(V)vLag. To connect the third term in (10e) withθLag, we look at the equa- tions
ζ(x) = d dε
ε=0sε= d dε
ε=0ϕε◦ϕε−X 1(x) = d dε
ε=0
ηεY ◦ϕ◦ϕ−X1◦ηε−X 1 (x) =
= d dε
ε=0
ηYε ◦s◦ηε−1X
(x) =V(s(x))−ds
Vx(x) ∂
∂x
=
=
VA(s(x))−Vx(x)dsA(x) dx
∂
∂yA =Vv(s(x)). Hence
dj−m−1ζA(x)
dxj−m−1 = dj−m−1
dxj−m−1VvA(s(x)) =Dx,j−m−1VvA
jj−m−1s(x)
=
=j2k−1(V)vAj−m−1.
Then the third term of the boundary part of the variation (10e) is N
A=1
k j=1
j−1
m=0
(−1)mDx,m ∂L
∂yjA
j2k−1s(x) ·j2k−1(V)vAj−m−1b
a
= N A=1
k j=1
j−1
m=0
(−1)mDx,m ∂L
∂yjA
j2k−1s(x) ·θj−m−A 1
j2k−1(V)v b
a
= j2k−1s∗ij2k−1(V)vθLagb (14) a
=
∂[a,b]
j2k−1(s)∗ij2k−1(V)vθLag. (15)
Now, (13e) and (15) yield
∂[a,b]j2k−1
ϕ◦ϕ−X1∗
ij2k−1(V)θLag for the boundary part of the variation. These results may be summarized as
Lemma 1. Let Lag = L
x, yA, yA1, . . . , yAk
dx be a Lagrangian defined on JkY andηεY :Y →Y a one parameter group of projetable diffeomrphisms of Y with infinitesimal generator V. Let ηJεkY be the extension of ηεY to JkY and jk(V) its infinitesimal generator. Then for ϕ ∈ M, ϕX([0,1]) = [a, b], such that the image ofϕis contained into a coordinate chart, the variation of the action functional(12) is given by
∂
∂ε
ε=0S(ϕε) = N A=1
k j=0
b
a (−1)jDx,j ∂L
∂yjA j2k
ϕ◦ϕ−X1
(x) ·VvA(s(x)) dx+
+
∂[a,b]
j2k−1
ϕ◦ϕ−X1∗
ij2k−1(V)θLag.
The exterior derivative of θLag is given by dθLag=
N A=1
k j=1
j−1
m=0
(−1)md
Dx,m ∂L
∂yAj
∧θj−m−1A −
−N
A=1
k j=1
j−1
m=0
(−1)mDx,m ∂L
∂yAj dyj−mA ∧dx+ dL∧dx.
Now, we are going to compute j2k−1
ϕ◦ϕ−X1∗
iWdθLag for a vertical vector fieldW =N
A=1
2k−1
j=0 WjA∂y∂A
j onJ2k−1Y. We have iWdθLag=
N A=1
k j=1
j−1
m=0
(−1)md
Dx,m ∂L
∂yAj
(W)·θj−m−A 1−
−N
A=1
k j=1
j−1
m=0
(−1)mWj−m−A 1·d
Dx,m ∂L
∂yjA
−
− N A=1
k j=1
j−1
m=0
(−1)mDx,m ∂L
∂yjA ·Wj−mA ·dx+ N A=1
k j=0
∂L
∂yAj WjA·dx.
Taking into account thatj2k−1
ϕ◦ϕ−X1∗
θAj = 0, we get j2k−1
ϕ◦ϕ−X1∗
iWdθLag
(16a)
=−N
A=1
k j=1
j−1
m=0
(−1)mWj−m−A 1· d dxm+1
∂L
∂yjA
ϕ◦ϕ−X1 (x)
dx (16b)
−N
A=1
k j=1
j−1
m=0
(−1)mWj−mA · d dxm
∂L
∂yjA
ϕ◦ϕ−X1 (x)
·dx (16c)
+ N A=1
k j=0
∂L
∂yjA
ϕ◦ϕ−X1 (x)
·WjA·dx (16d)
= N A=1
∂L
∂y0A
ϕ◦ϕ−X1 (x)
·W0A+ (16e)
+ N A=1
k j=1
(−1)j d dxj
Dx,j ∂L
∂yjA
ϕ◦ϕ−X1 (x)
·W0Adx
= N A=1
k
j=0
(−1)j d dxj
Dx,j ∂L
∂yjA
ϕ◦ϕ−X1 (x)
·W0Adx.
(16f)
Now, Lemma 1 may be reformulated as
Proposition 2. Under the hypotheses of Lemma 1, the variation of the functional S is
∂
∂ε|ε=0S(ϕε) =
[a,b]
j2k−1
ϕ◦ϕ−X1∗
ij2k−1(V)dθLag+ (17)
+
∂[a,b]
j2k−1
ϕ◦ϕ−X1∗
ij2k−1(V)θLag.
Proof. Because
j2k−1
ϕ◦ϕ−1X ∗
ij2k−1(V)hdθLag=j2k−1
ϕ◦ϕ−1X ∗
ij2k−1(ϕ◦ϕ−1X )∗(Vx ∂∂x)dθLag
=iVx ∂
∂xdj2k−1
ϕ◦ϕ−X1∗
θLag=iVx ∂
∂x0 = 0, by (16) we have
j2k−1
ϕ◦ϕ−X1∗
ij2k−1(V)dθLag=j2k−1
ϕ◦ϕ−X1∗
ij2k−1(V)vdθLag
= N A=1
k j=0
(−1)jDx,j ∂L
∂yAj j2k
ϕ◦ϕ−X1
(x) ·VjvA
ϕ◦ϕ−X1 dx.
The result now follows from Lemma 1.
Now, we shall prove that the definition (11) of θLag is independent of any coordinate chart on the space J2k−1Y of jets and formula (17) of the variation holds for anyϕ∈ M.
Theorem3. Let the fibration πX,Y :Y →X, withX a one-dimensional manifold. Let Lag be a Lagrangian defined on JkY, which is given in a co- ordinate chart by Lag = L
x, yA, yA1, . . . , ykA
dx and let ηYε : Y → Y be a one-parameter group (1) of projetable diffeomrphisms of Y with infinitesimal generator V given by (3). Let ηεJkY be the extension (2) of ηεY to JkY and jk(V) its infinitesimal generator given by (4), (6), (7), (8). Let on J2k−1Y the formθLag given locally by (11). Then
a) θLag is well defined on J2k−1Y;
b)for ϕ∈ M, ϕX([0,1]) = [a, b], the variation of the action functional (12)is given by
∂
∂ε|ε=0S(ϕε) =
[a,b]
j2k−1
ϕ◦ϕ−X1∗
ij2k−1(V)dθLag+ +
∂[a,b]j2k−1
ϕ◦ϕ−X1∗
ij2k−1(V)θLag; c)we have the formula
jk
ϕ◦ϕ−1X ∗ Lag =
j2k−1
ϕ◦ϕ−1X ∗θLag;
d)in a standard coordinate chart for J2k−1Y, θLag only depends at each point j2k−1(s)(x) on dx and dyjA, for j≤k−1;
e) if W is a vector field on J2k−1Y, tangent to the fibers of πY,J2k−1Y, then j2k
ϕ◦ϕ−X1∗
(iWdθLag) = 0;
f)θLag is uniquely defined by b), c), d), e);
g) the Euler-Lagrange equations
(18)
0≤j≤k
(−1)j
d dxj
∂L
∂yAj
ϕ◦ϕ−X1
= 0, A= 1,2, . . . , N are equivalent to
(19) j2k−1
ϕ◦ϕ−X1∗
(iW dθLag) = 0 for any vector fieldW on J2k−1Y.
Proof. If the image of ϕis contained into a coordinate chart, then b) is Proposition 2, c) follows from (11), d) is evident from the definition ofθLag, e) follows from (16) taking into account thatW0A = 0. Assuming again that the image of ϕ is contained into a coordinate chart, let θ be any form with properties b), c), d), e) above. It follows from c) anc d) that θ = Ldx + N
A=1
k−1
j=0FAjθAj. Let ∆ = θ−θLag. Then ∆ = N
A=1
k−1
j=0GjAθjA. From b) we get
[a,b]j2k−1(s)∗ij2k−1(V)d∆ = 0 for any projetable vector field V on Y and any sections= ϕ◦ϕ−X1. From e) we have iWd∆ = 0 for any vertical vector fieldπY,J2k−1Y, whence
(20)
[a,b]
j2k−1(s)∗iWd∆ = 0 for any sections and any vector field
W =Wx(x) ∂
∂x +WA(x, y) ∂
∂yA + N A=1
2k−1 j=1
WjA ∂
∂yjA,