Distributional adjoints of natural operators 245
3. Differential operators and their distributional adjoints Let v : M
→ TM be a smooth vector field. If α : M → ∧
r
T
∗
M is a smooth ordinary current and β : M
→ ∧
m −r
T
∗
M is a test current, then
hv.α, βi = −hα, v.βi , where v. denotes the Lie derivative operator. This formula can be taken as the definition of v.α
for an arbitrary current α , and it turns out that the mapping α 7→ v.α is a continuous
linear operator in DM,
∧
r
T
∗
M . A similar construction can be carried out for scalar
generalized half-densities [4]. In the domain of a coordinate chart, in particular, on has the partial derivatives of a scalar current or generalized half-density.
Consider now E-valued currents the following argument is quite similar for E- valued generalized half-densities. Given a frame of E , the components of any such
current are scalar currents, whose partial derivatives with respect to the base coor- dinates are well-defined. If D is a differential operator acting on smooth currents,
which can be locally expressed in terms of partial derivatives, then D has a unique continuous extension to the whole space of currents, and this extension maintains the
same coordinate expression. More specifically, one considers a polynomial deriva- tion operator of order k
∈ N, which means that its coordinate expression is of the type Dα
i
= P
|p|≤k
c
i p j
∂
p
α
j
; here, p = { p
1
, .., p
m
} is a multi-index of order | p| :=
P
a
p
a
, and c
i p j
: M
→ C are smooth functions. Because the operator D can be expressed in terms of Lie derivatives, it yields a
distributional adjoint operator D
′
, fulfilling D
′
φ, θ = hφ, Dθi ; this acts in the dual
space, is still a polynomial operator of degree k, and extends to the whole, appropriate space of generalized sections. In practice, its coordinate expression is usually found
by assuming φ to be C
k
, expressing hφ, Dθi as an integral and applying integration by
parts boundary terms disappear. In the case of scalar-valued currents one has some important instances. Let v :
M → TM be a smooth vector field, and ω : M → ∧ T
∗
M a smooth form; indicating
by |θ| the degree of the current θ, by simple exterior algebra calculations and taking
into account ∂M = ∅ one obtains
hv|α, βi = −1
|α|+1
hα, v|βi , |β| = m − |α| + 1 .
hω ∧ α, βi = −1
|α|·|ω|
hα, ω ∧ βi , |β| = m − |α| − |ω| .
hdα, βi = −1
|α|+1
hα, dβi , |β| = m − |α| − 1 .
hdv|α, βi = −hα, v|dβi , |β| = m − |α| .
hv|dα, βi = −hα, dv|βi , |β| = m − |α| .
hv.α, βi = −hα, v.βi , |β| = m − |α| .
h∗α, βi = −1
|α|m+1
hα, ∗βi , |β| = |α| .
hδα, βi = −1
|α|
hα, δβi , |β| = m − |α| + 1 .
hdδα, βi = hα, δdβi , |β| = m − |α| .
hdδ + δdα, βi = hα, dδ + δdβi , |β| = m − |α| .
246 D. Canarutto
In the four last formulas, ∗ stands for the Hodge isomorphism relatively to a given non-
degenerate pseudometric on M. The operators δ :
= ∗d∗ and dδ + δd are, up to sign, the usual codifferential and Laplacian of scalar-valued currents; in general, the latter is
different from the connection-induced Laplacian defined on E-valued currents §6.
4. Operators induced by the Fr¨olicher-Nijenhuis bracket Let N be a classical manifold. Let us recall that the Fr¨olicher-Nijenhuis bracket [7,
9] of two differentiable tangent-valued forms F : N
→ ∧
p
T
∗
N ⊗
N
TN, G : N →
∧
r
T
∗
N ⊗
N
TN, is a tangent-valued form [F, G] : N
→ ∧
p +r
T
∗
N ⊗
N
TN .
This operation can be carried on fibred manifolds; in particular consider the case when N
≡ E, the total space of our vector bundle. A smooth current α : M → ∧
r
T
∗
M ⊗
M
E can also be seen as a tangent-valued r -form on E because of the natural inclusions
T
∗
M ⊂ T
∗
E and E ⊂ E ⊗
M
E ≡ VE ⊂ TE. Then we can consider the F-N bracket
[ω, α] , whenever ω is a smooth tangent-valued form on E. In practice, one is interested mainly in basic forms ω : M
→ ∧
p
T
∗
M ⊗
E
TE which are projectable over TM-valued
forms. Taking fibred coordinates x
a
, y
i
: E
X
→ R
m
×R
n
, the projectability condition for ω becomes ∂
j
ω
b a
1
... a
p
= 0 , and one obtains the local expression [ω, α]
= ω
b a
1
... a
p
∂
b
α
i a
p +1
... a
p +r
+ −1
pr
r ∂
a
r
ω
b a
r +1
... a
r + p
α
i a
1
... a
r −1
b
+ + −1
pr +1
∂
j
ω
i a
r +1
... a
r + p
α
j a
1
... a
r
d x
a
1
∧ . . . ∧ d x
a
p +r
⊗ ∂ y
i
. In order that [ω, α] be still a current, then, one has to assume that ω is a linear morphism
over M, which in coordinates reads ω
i a
1
... a
p
= y
j
ω
i j a
1
... a
p
with ω
i j a
1
... a
p
: M → R .
Now it is clear that the mapping α 7→ [ω, α] extends naturally to a continuous operator
acting on the distributional space of E-valued currents. The coordinate expression of its distributional adjoint is readily written. We see that, in general, its geometrical in-
terpretation involves viewing α as a T
∗
M ⊗
M
E-valued r −1 current, leading to a rather
involved situation. However, the really interesting cases for field theory applications are simpler: essentially, when ω is either a vertical-valued form or a connection.
Consider the first instance, namely ω : M
→ ∧
p
T
∗
M ⊗
M
VE, so that ω
b a
1
... a
p
= 0 . We obtain the distributional adjoint formula
h[ω, α], βi = −1
pr +1
Z
M
α
j a
1
... a
r
ω
i j a
r +1
... a
r + p
β
i a
r + p+1
... a
m
d x
a
1
∧ . . . ∧ d x
a
m
⊗ d y
j
= −1
pr +1
α, [ω
∗
, β ]
,
where β : M
→ ∧
m −p−r
T
∗
M ⊗
M
E
∗
and ω
∗
: E
∗
→ ∧
p
T
∗
M ⊗
M
E
∗
is the transpose
morphism over M.
Distributional adjoints of natural operators 247
A similar result holds for a linear connection on the bundle E  M; this can be defined [9] as a section γ : E