Académique Documents
Professionnel Documents
Culture Documents
CURVILINEAR COORDINATES
1. Introduction
All the notations used, but not defined, here are defined in the next
sections.
It is well known in differential geometry that a flat Riemannian space
can be isometrically immersed, at least locally, in the Euclidean space with
the same dimension. For instance, if R3 is a simply-connected domain
endowed with a flat Riemannian metric (gij ) of class C 2 in , then there
exists an immersion : R3 of class C 3 in such that
gij = i j in .
The assumption that the metric (gij ) is flat means that its Riemannian
curvature tensor vanishes in , i.e., that
Rtkij := gt` i `jk j `ik + tjk `it tik `jt = 0 in ,
where
1
`jk := g `t (j gtk + k gtj t gjk )
2
denote the Christoffel symbols associated with the metric (gij ).
The above immersion is recovered from the metric (gij ) by first solving
the system
i gj = kij gk in ,
with unknowns gj C 2 (; R3 ), then by solving the system
i = gi in .
The mapping C 3 (; R3 ) found in this fashion is the sought immersion.
Note that the first system above has solutions because the Riemannian cur-
vature tensor vanishes in and that the second system above has solutions
because the Christoffel symbols satisfy `jk = `kj .
Our main objective here is to establish an infinitesimal version of this
result. More specifically, let there be given a simply-connected domain
R3 and let (eij ) : S3 be a symmetric matrix field of class C 2 in (this
regularity assumption, chosen here for simplicity, will be in fact substantially
weakened). Then we show in Theorem 4 that if the matrix field (eij ) satisfies
the Saint Venant equations in curvilinear coordinates, viz.,
lin
Rtkij := ekikjt + etjkik ekjkit etikjk = 0 in ,
then there exists a vector field v : R3 of class C 3 such that
1
eij = (i v gj + gi j v) in .
2
The notation ekikjt , as well as the notation ekjki below, respectively denote
second and first covariant derivatives of the matrix field (eij ) (see Section 3
for their explicit expressions).
The recovery of the vector field v from the matrix field (eij ) is obtained
by first solving the system
aijkk = ekjki ekikj in ,
where the unknown is an antisymmetric matrix field (aij ) C 2 (; A3 ), then
by solving the system
i v = (eij + aij )gj in .
The vector field v C 3 (; R3 ) found in this fashion has the desired prop-
erties. Note that the first system above has solutions because the Saint
Venant equations are satisfied and that the second system above has solu-
tions because the matrix fields (eij ) and (aij ) are respectively symmetric
and antisymmetric.
As shown in Theorem 7, this last result may be viewed as an infinitesimal
version of the recovery of an immersion from its associated metric (gij )
(see the begining of the introduction) because the left-hand side of the Saint
Venant equations is nothing but the linearized part with respect to v of the
Riemann curvature tensor associated with the immersion ( + v).
The Saint Venant equations derived here in curvilinear coordinates gener-
alize the classical Saint Venant equations in Cartesian coordinates (see, e.g.,
SAINT VENANT COMPATIBILITY EQUATIONS IN CURVILINEAR COORDINATES 3
[2]). In this respect, note that these equations have been likewise extended
to Saint Venant equations on a surface; see [5].
The results of this paper have been announced in [4]
an element of H 2 ().
and they belong to the space H 1 () (cf. Lemma 1). Since the matrix fields
gi gj := gi (gj )T
form a basis in the space C 2 (; M3 ) and since
k (gi gj ) = ik` g` gj jk` gi g` ,
the above definition of the covariant derivatives Tijkk shows that
Tijkk gi gj = k (Tij gi gj ). (1)
Note that such equations are to be understood in the distributional sense,
the functions Tij being only in L2 ().
Finally, for all third-order tensor fields with covariant components Tijk
H 1 (), we define the covariant derivatives
Tijkk` := ` Tijk t`i Ttjk t`j Titk t`k Tijt ,
which, in view of Lemma 1, are well defined as distributions in H 2 ().
If Tij L2 (), the second-order covariant derivatives Tijkk` are defined
by the relations
Tijkk` := ` Tijkk t`i Ttjkk t`j Titkk t`k Tijkt .
It is then easily seen, in view of relation (1), that these second-order covari-
ant derivatives satisfy
Tijkk` gi gj = (`k t`k t )(Tij gi gj ). (2)
Indeed, relation (1) implies that
`k (Tij gi gj ) = ` (Tijkk gi gj )
= (` Tijkk )gi gj (i`t Tijkk )gt gj (j`t Tijkk )gi gt
= (` Tijkk t`i Ttjkk t`j Titkk )gi gj
= (Tijkk` + t`k Tijkt )gi gj
= (Tijkk` )gi gj + t`k t (Tij gi gj ).
Note that relation (2) implies in particular that the second-order covariant
derivatives satisfy:
Tijkk` = Tijk`k .
Since {gi (x) gj (x)} form a basis in M3 for all x , the matrix field P
can be expanded over this basis as
P = pij gi gj in ,
and since pij = tr(PT (gi gj )), we clearly have pij L2 ().
Hence the matrix fields Hk can be re-written as
Hk = hijk gi gj = k (pij gi gj ).
But the definition of the covariant derivative shows that
k (pij gi gj ) = pijkk gi gj .
Combining the last two equations, we finally find that
hijk = pijkk .
That the solution pij is unique up to additive constants is clear.
i.e., (eij ) and (aij ) are respectively the symmetric and the antisymmetric
parts of the tensor (uikj ). The derivatives i u L2 (; R3 ) of the vector
field u are then given by
i u = (i u gj )gj = (eij + aij )gj ,
which shows that these derivatives are completely determined by the sym-
metric tensor (eij ) and the antisymmetric tensor (aij ). In fact, they are
determined only by the tensor (eij ), as we now show. Since i gj = j gi , we
first have
2k aij = ki u gj + i u k gj kj u gi j u k gi
= i (2ekj j u gk ) j (2eki i u gk ) + i u k gj j u k gi
= 2(i ekj j eki + i u j gk j u i gk ),
all equalities being valid in the distributional sense. Combining this last
equality with the relations
i u j gk = `jk (i u g` ) = `jk (ei` + ai` ),
j u i gk = `ik (j u g` ) = `ik (ej` + aj` ),
we next deduce that
k aij `jk ai` + `ik aj` = (i ekj `ik ej` ) (j eki `jk ei` )
= ekjki ekikj .
But the first term is equal to the covariant derivative aijkk , since aj` = a`j .
Hence the previous equality becomes
aijkk = ekjki ekikj . (5)
Then the relations (see Section 3)
aijkk` = aijk`k in H 2 (), (6)
imply that
ekjki` ekikj` = e`jkik e`ikjk in H 2 ().
This means that the Saint Venant equations (4) are satisfied in the distri-
butional sense.
Remarks. (1) Equation (5) shows that the antisymmetric matrix field (aij )
is uniquely determined by the linear strains ij (u), up to an antisymmetric
matrix field that is constant in each connected component of .
(2) Equation (6) shows that the Saint Venant equations in curvilinear
coordinates simply express that uikjk` = uikj`k . To see this, we note that
uikj = eij + aij .
These relations combined with relations (5) show that
uikjk` = eijkk` + ekjki` ekikj` .
10 P.G. CIARLET, C. MARDARE AND M. SHEN
Hence the Saint Venant equations hold if and only if uikjk` = uikj`k . These
relations are also equivalent to the relations
i ukk` = i uk`k ,
where the second-order covariant derivatives of the vector fields i u
L2 (; R3 ) are defined by replacing Ti with i u in the definition of the second-
order covariant derivatives of a first-order tensor field with covariant com-
ponents Ti . More specifically, if v i := i u L2 (; R3 ), then
v i kk := k v i rki v r in H 1 (; R3 ),
v i kk` := (v i kk )k` = ` v i kk r`i v r kk r`k v i kr in H 2 (; R3 ).
Remark. Theorem 3 of Ciarlet & C. Mardare [3] shows that, if the open
set is connected, any vector field v H 1 (; R3 ) that satisfies
1
eij = (i v gj + gi j v) in L2 (; S3 )
2
is necessarily of the form
v(x) = v(x) + (a + b (x)) for almost all x ,
where a and b are vectors in R3 .
1,p
Proof. Since Wloc () C 0 () by the Sobolev embedding theorem and
since det(gij ) > 0 in (the matrix (gij (x)) being positive definite for all
x by assumption), the definition of the inverse of a matrix shows that
1,p
(g k` ) = (gij )1 Wloc (; S3> ). Hence the Christoffel symbols
1
`jk := g `r (j grk + k grj r gjk )
2
p
belong to the space Lloc ().
Let the vectors fields gi and gj by defined by
gi := i and gi gj = ij in .
Then we deduce from relations (9) that
j grk = j gr gk + gr j gk .
This relation combined with the above definition of the Christoffel symbols
implies that
`jk = g `r (gr j gk ).
Since g` = g `k gk and since the vectors {gi (x)} form a basis in R3 for all
x , we next deduce that
p
j gk = `jk g` in Lloc (; R3 ).
Using now Schwarz lemma applied to the second derivatives of gk in the
space of distributions, we next infer from the above relation that
i (`jk g` ) = j (`ik g` )
for all i, j, k, ` in the distributional sense. Combining the last two relations
then shows that
i `jk j `ik + rjk `ir rik `jr = 0
in the distributional sense, which means that the Riemann curvature tensor
of the metric gij vanishes in .
Our objective is to show that Theorems 3 and 4 are nothing but the
infinitesimal versions of Theorems 5 and 6, respectively. To this end, we
will show that the left-hand side of the Saint Venant equations is in a specific
sense the linear part of the Riemann curvature tensor.
Theorem 7. Let be a bounded open subset in R3 and let there be given
a matrix field (gij ) C 2 (; S3> ) whose associated Riemann curvature tensor
field vanishes in .
Then, for all increment symmetric matrix fields (eij ) W 1,p (; S3 ),
p > 3, the linear part with respect to (eij ) of the covariant components of
the Riemann curvature tensor associated with the metric (gij +eij ) are given
by
lin
Rskij (eij ) = ekikjs + esjkik ekjkis esikjk , (13)
where ekikjs denote the second-order covariant derivatives of eki (cf. Section
3).
Proof. For all , define the matrix field
(gij ()) := (gij ) + (eij ) W 1,p (; S3 ).
Since W 1,p () C 0 () by the Sobolev embedding theorem, there exists a
number 0 > 0 such that, for all || < 0 , the matrix field (gij ()) is positive
definite in . As in the proof Theorem 5, this implies that g k` () W 1,p (),
where (g k` ()) = (gij ())1 is the inverse of the matrix field (gij ()). Hence
the Christoffel symbols
1
rjk () := {j grk () + k gjr () r gjk ()} and `jk () := g `r ()rjk ()
2
belong to the space Lp (). Consequently, the Riemann curvature tensor
associated with the metric (gij ()) is well defined in the sense of distributions
by its mixed components
`
Rkij () := i `jk () j `ik () + rjk ()`ir () rik ()`jr (),
or by its covariant components
`
Rskij () = gs` ()Rkij ().
The linear part with respect to (eij ) of each covariant component of the
Riemann curvature tensor associated with the metric (gij + eij ) is then
defined by the limit
lin Rskij ()
Rskij := lim ,
0
6=0
since the Riemann curvature tensor of the metric (gij ), whose covariant
components are defined by
Rskij := gs` i `jk j `ik + rjk `ir rik `jr ,
vanishes in by assumption.
SAINT VENANT COMPATIBILITY EQUATIONS IN CURVILINEAR COORDINATES15
In order to compute this linear part, we expand all the above functions
as power series in . Using the notation O(2 ) for any function f such that
(2 f ) is bounded in spaces that will be specified at each occurence, we have
Consequently,
1
kij () = kij + {i (gkj () gkj ) + j (gki () gki ) k (gij () gij )}
2
= kij + (i ekj + j eki k eij ) + O(2 ) in Lp ().
Likewise,
Letting
X`ij := e`jki + e`ikj eijk` and Xijk = g k` X`ij ,
we thus have the following relations in Lp ():
hold in the space W 1,p (), hence also in the space H 1 ().
Note that the covariant derivatives
` `
Xjk ki := i Xjk rij Xrk
`
rik Xjr
`
+ `ir Xjk
r
,
X`jkki := i X`jk rij X`rk rik X`jr ri` Xrjk ,
g k` ki := i g k` + kir g r` + `ir g kr ,
satisfy
`
Xjk ki = g `r Xrjkki + g `r ki Xrjk in H 1 ().
16 P.G. CIARLET, C. MARDARE AND M. SHEN
Moreover, the definition of the Christoffel symbols associated with the metric
(gij ) shows that
1
g k` ki = i g k` + (g ks g r` + g kr g s` )sir
2
1
k`
= i g + {g ks (g r` i gsr ) + g kr (g s` i gsr ) + (g ks g r` + g kr g s` )(r gis s gir )}
2
1
= i g k` {g ks (i g r` )gsr + g kr (i g s` )i gsr }
2
= 0 in Lp (),
which, combined with the previous relations, implies that
`
Xjk ki = g `r Xrjkki in H 1 ().
` (), we thus obtain
Using this relation in the previous expression of Rkij
the following relations in H 1 ():
` ` `
Rkij () = (Xjk ki Xik kj ) + O(2 )
= g `r (Xrjkki Xrikkj ) + O(2 )
= g `r (erjkki ejkkri erikkj + eikkrj ) + O(2 )
and
R`kij () = g`r g rs (esjkki ejkksi esikkj + eikksj ) + O(2 )
= e`jkki ejkk`i e`ikkj + eikk`j + O(2 ).
This completes the proof of the theorem.
where wi (x) denotes the vector in R3 whose components are (k i )(x). This
implies that wi (x) = gi (x) for all x , or equivalently, that
(k i ) = [gi ]k in ,
where [gi ]k denotes the k-th component of the vector field gi .
18 P.G. CIARLET, C. MARDARE AND M. SHEN
lin
= [g` ]s [gk ]r [gi ]p [gj ]q ekikj` + e`jkik ekjki` e`ikjk
Rsrpq
= [g` ]s [gk ]r [gi ]p [gj ]q R`kij
lin
where [gi ]p denotes the p-th component of the vector field gi . Then the
definition of the functions epq shows that
1
(i v gj + gi j v) = [gi ]p [gj ]q [gk ]p [g` ]q ek` = eij
2
in L2 (). This completes the proof.
SAINT VENANT COMPATIBILITY EQUATIONS IN CURVILINEAR COORDINATES19
References
[1] Adams, R.A.: Sobolev Spaces, Academic Press, New York, 1975.
[2] Ciarlet, P.G. and Ciarlet, P. Jr.: Another approach to linearized elasticity and a new
proof of Korns inequality, Math. Models Methods Appl. Sci. 15 (2005), 259271.
[3] Ciarlet, P.G. and Mardare, C.: On rigid and infinitesimal rigid displacements in
three-dimensional elasticity, Mathematical Models and Methods in Applied Sciences
13 (2003), 1589-1598.
[4] Ciarlet, P.G., Mardare, C. and Shen, M.: Recovery of a displacement field from its
linearized strain tensor field in curvilinear coordinates, C.R. Acad. Sci. Paris, Ser. I
(submitted).
[5] Ciarlet, P.G., Gratie, L., Mardare, C., and Shen, M.: Saint Venant compatibility
equations on a surface Application to linear intrinsic shell theory, to appear.
[6] Grisvard P.: Elliptic Problems in Nonsmooth Domains, Pitman, Boston, 1985.
20 P.G. CIARLET, C. MARDARE AND M. SHEN
[7] Necas, J.: Les Methodes Directes en Theorie des Equations Elliptiques, Masson, Paris,
1967.
[8] Mardare, S.: On Pfaff systems with Lp coefficients and their applications in differen-
tial geometry, J. Math. Pures Appl. 84 (2005), 1659-1692.
[9] Mardare, S.: On isometric immersions of a Riemannian space with little regularity,
Analysis and Applications 2 (2004), 193-226.
[10] Mardare, S.: Sur quelques problemes de geometrie differentielle lies a la theorie de
lelasticite, Doctoral Disertation, Universite Paris 6, 2003.
Ming Shen, Department of Mathematics, City University of Hong Kong, 83 Tat Chee
Avenue, Kowloon, Hong Kong, E-mail address: geoffrey.shen@student.cityu.edu.hk