Professional Documents
Culture Documents
Coordinate Transformations
2.1 Introduction
Partial dierential equations in the physical domain X n can be solved on a
structured numerical grid obtained by mapping a reference grid in the logical
region n into X n with a coordinate transformation x() : n X n . The
structured grid concept also gives an alternative way to obtain a numerical
solution to a partial dierential equation, by solving the transformed equation
with respect to the new independent variables i on the reference grid in the
logical domain n . Some notions and relations concerning the coordinate
transformations yielding structured grids are discussed in this chapter. These
notions and relations are used to represent some conservation-law equations
in the new logical coordinates in a convenient form. The relations presented
will be used in Chap. 3 to formulate various grid properties.
Conservation-law equations in curvilinear coordinates are typically deduced from the equations in Cartesian coordinates through the classical formulas of tensor calculus, by procedures which include the substitution of tensor derivatives for ordinary derivatives. The formulation and evaluation of
the tensor derivatives is rather dicult, and they retain some elements of
mystery. However, these derivatives are based on specic transformations of
tensors, modeling in the equations some dependent variables, e.g. the components of a uid velocity vector, which after the transformation have a clear
interpretation in terms of the contravariant components of the vector. With
this concept, the conservation-law equations are readily written out in this
chapter without application of the tensor derivatives, but utilizing instend
only some specic transformations of the dependent variables, ordinary derivatives, and one basic identity of coordinate transformations derived from the
formula for dierentiation of the Jacobian.
For generality, the transformations of the coordinates are mainly considered for arbitrary n-dimensional domains, though in practical applications the
dimension n equals 1, 2, 3, or 4 for time-dependent transformations of threedimensional domains. We also apply chiey a standard vector notation for
V.D. Liseikin, Grid Generation Methods, Scientic Computation,
c Springer Science+Business Media B.V. 2010
DOI 10.1007/978-90-481-2912-6 2,
32
2 Coordinate Transformations
= ( 1 , . . . , n ), x = (x1 , . . . , xn ),
33
1
j =
, i, j = 1, . . . , n,
xj
and consequently
i
det
xj
=
1
,
J
i, j = 1, . . . , n.
(2.2)
xi
3j
= (1)i+j J 3i , i, j = 1, 2.
j
(2.3)
,
xj
J i+1 i+2
i+2 i+1
(2.4)
j+1 j+2
xi
j+1 j+2
=J
, i, j = 1, 2, 3,
j
xi+1 xi+2
xi+2 xi+1
where for each superscript or subscript index, say l, l + 3 is equivalent to l.
With this condition the sequence of indices (l, l + 1, l + 2) is the result of a
cyclic permutation of (1, 2, 3) and vice versa; the indices of a cyclic sequence
(i, j, k) satisfy the relation j = i + 1, k = i + 2.
2.2.2 Tangential Vectors
The value of the function x() = [x1 (), . . . , xn ()] in the Cartesian basis
(e1 , . . . , en ), i.e.
x() = x1 ()e1 + + xn ()en ,
is a position vector for every n . This vector-valued function x() generates the nodes, edges, faces, etc. of the cells of the coordinate grid in the
domain X n . Each edge of the cell corresponds to a coordinate line i for some
i and is dened by the vector
i x = x( + hei ) x(),
where h is the step size of the uniform grid in the i direction in the logical
domain n . We have
34
2 Coordinate Transformations
i x = hxi + t,
where
x i =
x1
xn
,
,
i
i
35
Fig. 2.2. Disposition of the base tangential and normal vectors in two dimensions
i
i
,
,
,
x1
xn
i = j,
and thus the vector i is a normal to the coordinate surface i = 0i . Therefore the vectors i , i = 1, . . . , n, are also called the normal base vectors.
Since
xi i = 1
for each xed i = 1, . . . , n, the vectors xi and i intersect each other at
an angle which is less than /2. Now, taking into account the orthogonality
of the vector i to the surface i = 0i , we nd that these two vectors xi
and i are directed to the same side of the coordinate surface (curve in two
dimensions). An illustration of this fact in two dimensions is given in Fig. 2.2.
The length of any normal base vector i is linked to the distance di
between the corresponding opposite boundary segments (joined by the vector
xi ) of the n-dimensional parallelepiped formed by the base tangential vectors,
namely,
di = 1/| i |, | i | = i i .
To prove this relation we recall that the vector i is a normal to all of the
vectors xj , j = i, and therefore to the boundary segments formed by these
36
2 Coordinate Transformations
P (x) > 0, i, j = 1, . . . , n,
ji = 1 if i = j.
i, j = 1, . . . , n,
(2.5)
where aij are the elements of the matrix (aij ) which is the inverse of the
tensor (aij ), aij = ai aj , i, j = 1, . . . , n. It is assumed in (2.5) and later that
a summation is carried out over repeated indices unless otherwise noted.
The components of the vector b in the natural basis of the tangential
vectors xi , i = 1, . . . , n, are called contravariant. Let them be denoted by
i
b , i = 1, . . . , n. Thus
37
b = b x 1 + + b x n .
Assuming in (2.5) ai = xi , i = 1, . . . , n, we obtain
i
k
x
i
mj
k x
b =a
, i, j, k, m = 1, . . . , n,
b
j m
(2.6)
i j
,
xk xk
k = 1, . . . , n.
b = bj
i
,
xj
i, j = 1, . . . , n,
(2.7)
i = 1, . . . , n.
(2.8)
b = b i ,
Thus, in this case (2.5) has the form
b = (b i )xi ,
i = 1, . . . , n.
(2.9)
For example, the normal base vector i is expanded through the base
tangential vectors xj , j = 1, . . . , n, by the following formula:
i =
i k
x k,
xj xj
i, j, k, = 1, . . . , n.
(2.10)
xj
= b x i ,
i
i = 1, . . . , n,
(2.11)
and consequently
b = bi i = (b xi ) i ,
i = 1, . . . , n.
(2.12)
xj xj
k ,
i k
i, j, k = 1, . . . , n.
(2.13)
38
2 Coordinate Transformations
i, j = 1, . . . , n,
whose elements gij are the dot products of the pairs of the basic tangential
vectors xi ,
gij = xi xj =
xk xk
,
i j
i, j, k = 1, . . . , n,
(2.14)
i = 1, . . . , n.
Also,
gij = |xi ||xj |cos =
(2.15)
where is the angle between the tangent vectors xi and xj . In these expressions for gii and gij the subscripts ii and jj are xed, i.e. here the summation
over the repeated indices is not carried out.
The matrix (gij ) is called the metric tensor because it denes distance
measurements with respect to the coordinates 1 , . . . , n :
ds = gij d i d j , i, j = 1, . . . , n.
Thus the length s of the curve in X n prescribed by the parametrization
x[(t)] : [a, b] X n
is computed by the formula
s=
gij
d i d j
dt.
dt dt
39
and hence
J 2 = g.
The covariant metric tensor is a symmetric matrix, i.e. gij = gji . If a
coordinate system at a point is orthogonal then the tensor (gij ) has a simple
diagonal form at this point. Note that these advantageous properties are in
general not possessed by the Jacobi matrix (xi / i ) from which the covariant
metric tensor (gij ) is dened.
2.2.5.2 Contravariant Metric Tensor
The contravariant metric tensor of the domain X n in the coordinates 1 , . . . , n
is the matrix
(g ij ), i, j = 1, . . . , n,
inverse to (gij ), i.e.
gij g jk = ik ,
i, j, k = 1, . . . , n.
Therefore
det(g ij ) =
(2.16)
1
.
g
i j
,
xk xk
i, j, k = 1, . . . , n.
(2.17)
(2.18)
Geometric Interpretation
Now we discuss the geometric meaning of a xed diagonal element g ii , say
g 11 , of the matrix (g ij ). Let us consider a three-dimensional coordinate transformation x() : 3 X 3 . Its tangential vectors x1 , x2 , x3 represent
geometrically the edges of the parallelepiped formed by these vectors. For the
distance d1 between the opposite faces of the parallelepiped which are dened
by the vectors x2 and x3 , we have
d1 = x1 n1 ,
where n1 is the unit normal to the plane spanned by the vectors x2 and x3 .
It is clear, that
1 xj = 0, j = 2, 3,
and hence the unit normal n1 is parallel to the normal base vector 1 . Thus
we obtain
40
2 Coordinate Transformations
n1 = 1 /|1 | = 1 / g 11 .
Therefore
d1 = 1 1 / g 11 = 1/ g 11 ,
and consequently
g 11 = 1/(d1 )2 .
Analogous relations are valid for g 22 and g 33 , i.e. in three dimensions the
diagonal element g ii for a xed i means the inverse square of the distance di
between those faces of the parallelepiped which are connected by the vector
xi . In two-dimensional space the element g ii (where i is xed) is the inverse
square of the distance between the edges of the parallelogram dened by the
tangential vectors x1 and x2 .
The same interpretation of g ii is valid for general multidimensional coordinate transformations:
g ii = 1/(di )2 ,
i = 1, . . . , n,
(2.19)
where the index i is xed, and di is the distance between those faces of the
n-dimensional parallelepiped which are linked by the tangential vector xi .
2.2.5.3 Relations Between Covariant and Contravariant Elements
Now, in analogy with (2.3) and (2.4), we write out very convenient formulas
for natural relations between the contravariant elements g ij and the covariant
ones gij in two and three dimensions.
For n = 2,
g3i 3j
,
g
gij = (1)i+j gg 3i 3j ,
g ij = (1)i+j
(2.20)
i, j = 1, 2,
where the indices i, j on the right-hand side of the relations (2.20) are xed,
i.e. summation over the repeated indices is not carried out here. For n = 3 we
have
1
(gi+1 j+1 gi+2 j+2 gi+1 j+2 gi+2 j+1 ),
g
gij = g(g i+1 j+1 g i+2 j+2 g i+1 j+2 g i+2 j+1 ),
g ij =
(2.21)
i, j = 1, 2, 3,
with the convention that any index, say l, is identied with l 3, so, for
instance, g45 = g12 .
We also note that, in accordance with the expressions (2.14), (2.17) for gij
and g ij , respectively, the relations (2.10) and (2.13) between the basic vectors
xi and j can be written in the form
xi = gik k ,
i = g ik xk ,
i, k = 1, . . . , n.
(2.22)
41
(2.23)
i, j, m = 1, . . . , n.
e1 e2 e3
a b = det a1 a2 a3 ,
(2.24)
b1 b2 b3
where (e1 , e2 , e3 ) is the Cartesian vector basis of the Euclidean space R3 .
Thus
a b = (a2 b3 a3 b2 , a3 b1 a1 b3 , a1 b2 a2 b1 ),
or, with the previously mentioned convention in three dimensions of the identication of any index j with j 3,
a b = (ai+1 bi+2 ai+2 bi+1 )ei ,
i = 1, 2, 3.
(2.25)
We will now state some facts connected with the cross product operation.
2.2.6.1 Geometric Meaning
We can readily see that a b = 0 if the vectors a and b are parallel. Also,
from (2.25) we nd that a (a b) = 0 and b (a b) = 0, i.e. the vector
a b is orthogonal to each of the vectors a and b. Thus, if these vectors are
not parallel then
a b = |a b|n,
(2.26)
where = 1 or = 1 and n is a unit normal vector to the plane determined
by the vectors a and b.
Now we show that the length of the vector a b equals the area of the
parallelogram formed by the vectors a and b, i.e.
|a b| = |a||b| sin ,
(2.27)
where is the angle between the two vectors a and b. To prove (2.27) we rst
note that
42
2 Coordinate Transformations
3
ai ai
i=1
3
3
bj bj
j=1
3
2
a k bk
k=1
k=1
3
(al bm am bl )2 ,
k=1
(2.28)
what proves (2.27). Thus we obtain the result that if the vectors a and b are
not parallel then the vector a b is orthogonal to the parallelogram formed
by these vectors and its length equals the area of the parallelogram. Therefore
the three vectors a, b and a b are independent in this case and represent a
base vector system in the three-dimensional space R3 . Moreover, the vectors
a, b and a b form a right-handed triad since a b = 0, and consequently
the Jacobian of the matrix determined by a, b, and a b is positive; it equals
a b a b = (a b)2 .
2.2.6.2 Relation to Volumes
Let c = (c1 , c2 , c3 ) be one more vector. The volume V of the parallelepiped
whose edges are the vectors a, b and c equals the area of the parallelogram
formed by the vectors a and b multiplied by the modulas of the dot product
of the vector c and the unit normal n to the parallelogram. Thus
V = |a b||n c|
and from (2.26) we obtain
V = |(a b) c|.
Taking into account (2.25), we obtain
(a b) c = c1 (a2 b3 a3 b2 ) + c2 (a3 b1 a1 b3 ) + c3 (a1 b2 a2 b1 ).
(2.29)
43
The right-hand side of this equation is the Jacobian of the matrix whose rows
are formed by the vectors a, b, and c, i.e.
1
a a2 a3
(2.30)
(a b) c = det b1 b2 b3 .
c1 c2 c3
From this equation we readily obtain
(a b) c = a (b c) = (c a) b.
Thus the volume of the parallelepiped determined by the vectors a, b, and c
equals the Jacobian of the matrix formed by the components of these vectors.
In particular, we obtain from (2.1) that the Jacobian of a three-dimensional
coordinate transformation x() is expressed as follows:
J = x1 (x2 x3 ).
(2.31)
1
(x l xm ).
(2.32)
J
Thus the elements of the three-dimensional contravariant metric tensor (g ij )
are computed through the tangential vectors xi by the formula
i =
g ij =
1
(x i+1 xi+2 ) (xj+1 xj+2 ),
g
i, j = 1, 2, 3.
i = 1, 2, 3,
(2.33)
44
2 Coordinate Transformations
i, j = 1, 2, 3.
(2.34)
Thus the volume of the parallelepiped formed by the base normal vectors
1 , 2 , and 3 is the modulus of the inverse of the Jacobian J of the
transformation x().
45
i, j, k = 1, . . . , n,
(2.35)
j, k, m, p = 1, . . . , n.
(2.36)
i, j, k, l = 1, . . . , n,
(2.37)
and consequently
p
2 xp
m x
= kj
,
j
k
m
2 xl i
,
k j xl
or in vector form,
i
= xk j i .
kj
(2.38)
i
2 xl xl
,
k j m
j, k, l, m = 1, . . . , n,
(2.39)
and consequently
xk j = [kj, m] m .
(2.40)
i, j, k, m = 1, . . . , n.
(2.41)
Multiplying (2.39) by g im and summing over m we nd that the space Christoffel symbols of the rst and second kind are connected by the following relation:
i
kj
= g im [kj, m],
i, j, k, m = 1, . . . , n.
(2.42)
j, k, l, m = 1, . . . , n.
(2.43)
46
2 Coordinate Transformations
The space Christoel symbols of the rst kind [kj, m] can be expressed through
the rst derivatives of the covariant elements gij of the metric tensor (gij ) by
the following readily veried formula:
gkm
gkj
1 gjm
+
m , i, j, k, m = 1, . . . , n.
[kj, m] =
(2.44)
2 k
j
Thus, taking into account (2.42), we see that the space Christoel symbols of
i
can be written in terms of metric elements and their rst
the second kind kj
derivatives. In particular, in the case of an orthogonal coordinate system i ,
we obtain from (2.42), (2.44)
1 ii gii
gii
gkj
i
kj = g
+
.
g
k
j
i
Here the index i is xed, i.e. the summation over i is not carried out.
2.3.2 Dierentiation of the Jacobian
Of critical importance in obtaining compact conservation-law equations with
coecients derived from the metric elements in new curvilinear coordinates
1 , . . . , n is the formula for dierentiation of the Jacobian
i
x
x
J
2 xi m
J
Jdivx k ,
k
k m xi
xi k
i, k, m = 1, . . . , n.
(2.45)
In accordance with (2.37), this identity can also be expressed through the
i
space Christoel symbols of the second kind kj
by
J
i
= Jik
,
k
i, k = 1, . . . , n,
aim im
det(aij ) =
G ,
k
k
i, j, k, m = 1, . . . , n,
(2.46)
where Gim is the cofactor of the element aim . For the Jacobi matrix (xi / j )
of the coordinate transformation x() we have
m
, i, j = 1, . . . , n.
xi
Therefore, applying (2.46) to the Jacobi matrix, we obtain (2.45).
Gim = J
47
j
J i 0, i, j = 1, . . . , n,
j
x
(2.47)
,
xk xm
l j xk xm xp
i, j, k, l, m, p = 1, . . . , n.
(2.48)
Now, using this relation and the formula (2.45) for dierentiation of the Jacobian in the identity
2 j xk
j
J j
+
J
,
J
=
j
xi
j xi
xi xk j
we obtain
2 xp m l j xk
j
2 xk p j
J
=J p j k i J l m
j
i
x
x x
xi xk xp j
2 xk p j
2 xp l m
J
= 0,
p j xk xi
l m xp xi
i, j, k, l, m, p = 1, . . . , n,
=J
x3i
j
= (1)i+j 3j ,
i
x
i, j = 1, 2,
x3i
j
i+1
J
=
(1)
= 0,
j
xi
1 2
2 1
i, j = 1, 2.
48
2 Coordinate Transformations
account (2.25), we readily obtain the following formula for the dierentiation
of the cross product of two three-dimensional vector-valued functions a and b:
(a b) = i a b + a i b,
i
i = 1, 2, 3.
j
j
i
(x
x
)
=
x
x
+
xj xk i ,
k
k
i
i=1
i=1
i=1
(2.49)
xj xk i =
i=1
3
xk xi j .
i=1
(xj xk ) = 0,
i
i=1
since
xi xj k = xj k xi
and (2.32) implies (2.47) for n = 3.
The identity (2.47) can help one to obtain conservative or compact forms
of some dierential expressions and equations in the curvilinear coordinates
1 , . . . , n . For example, for the rst derivative of a function f (x) with respect
to xi we obtain, using (2.47),
f
1
j
=
f
, j = 1, . . . , n.
(2.50)
J
xi
J j
xi
For the Laplacian
2 f =
f
,
xj xj
j = 1, . . . , n
(2.51)
(2.52)
49
2 f = P
(2.53)
1
mj f
Jg
=P
J j
m
(2.54)
Ai
,
xi
i = 1, . . . , n,
(2.55)
1
j
(JA ) = F,
j
J
j = 1, . . . , n,
(2.56)
A = Ai
j
,
xi
i, j = 1, . . . , n.
(2.57)
50
2 Coordinate Transformations
i = 1, . . . , n,
(2.58)
where is the gas density, and ui is the ith component of the ow velocity
vector u in the Cartesian coordinates x1 , . . . , xn . With the substitution Ai =
ui , (2.58) is transformed to the following divergent form with respect to the
new dependent variables and ui in the coordinates 1 , . . . , n :
(Jui ) = 0,
i
i = 1, . . . , n.
(2.59)
i
,
xj
i, j = 1, . . . , n.
(2.60)
(ui ) = S, i = 1, . . . , n,
(2.61)
i i +
x
x
xi
where and denote the density and diusion coecient of the uid, respectively. Taking
Ai = ui i ,
x
we obtain, in accordance with the relation (2.57),
j
A = uj
kj
g .
k
Therefore, using (2.56), the convectiondiusion equation (2.61) in the curvilinear coordinates 1 , . . . , n is expressed by the divergent form
j
kj
J u g
= JS, j, k = 1, . . . , n.
(2.62)
j
k
51
f
= 0,
xj xj
j = 1, . . . , n,
(2.63)
f
,
xi
i = 1, . . . , n.
A = g ij
f
,
i
i = 1, . . . , n.
(2.64)
i, j = 1, . . . , n.
(2.65)
A = Akm
i j
,
xk xm
km
i, j, k, m = 1, . . . , n.
(2.66)
52
2 Coordinate Transformations
i
km x
k
Aij = A
xj
,
m
i, j, k, m = 1, . . . , n.
(2.67)
Therefore we can obtain a system of equations for the new dependent variables
ij
A by replacing the dependent quantities Aij in (2.65) with their expressions
(2.67). As a result we obtain
i
j
Aij
km x x
=
A
xj
xj
k m
j
km
2 i
i
x
A xi
km x
km x
=
+
A
+
A
m k
k m
k xj m
= F i,
i, j, k, m = 1, . . . , n.
The use of the formula (2.45) for dierentiation of the Jacobian in the summation in the equation above yields
km
2 i
Aij
A xi
1 km xi J
km x
=
+A
+ A
= F i,
j
m
k
k
m
x
J
k m
i, j, k, l, m = 1, . . . , n.
i, j, k, l = 1, . . . , n,
(2.68)
where
i
, i, j = 1, . . . , n,
xj
is the ith contravariant component of the vector F = (F 1 , . . . , F n ) in the
basis x1 , . . . , xn . The quantities ( 2 xl / k j )( i /xl ) in (2.68) are the
i
. Thus the system (2.68) has,
space Christoel symbols of the second kind kj
i
using the notation jk
, the form
i
F = Fj
1
ij
kj
i
i
(JA ) + kj
A =F ,
J j
i, j, k = 1, . . . , n.
(2.69)
We see that all coecients of (2.69) are derived from the metric tensor (gij ).
Equations of the form (2.69), in contrast to (2.65), do not have a conservative form. The conservative form of (2.65) in new dependent variables is
obtained, in analogy with (2.56), from the system
1
j
(JAi ) = F i ,
J j
j
i, j = 1, . . . , n,
(2.70)
where Ai is the jth component of the vector Ai = (Ai1 , . . . , Ain ) in the basis
xj , j = 1, . . . , n, i.e.
Ai = Aik
j
,
xk
i, j, k = 1, . . . , n.
53
(2.71)
In fact, (2.70) is the result of the application of (2.56) to the ith line of (2.65).
Therefore in the relations (2.65), (2.70), (2.71) we can assume an arbitrary
range for the index i, i.e. the matrix Aij in (2.65) can be a nonsquare matrix
with i = 1, . . . , m, j = 1, . . . , n.
Though the system (2.70) is conservative and more compact than (2.69),
it has its drawbacks. In particular, mathematical simulations of uid ows are
generally formulated in the form (2.65) with the tensor Aij represented as
Aij = B ij + ui uj ,
i, j = 1, . . . , n,
j
= ui uj ,
xk
i, j, k = 1, . . . , n,
i j
(u
u
)
+
= F i ,
xj
xi
xj xj
i, j = 1, . . . , n,
(2.72)
where ui is the ith Cartesian component of the vector of the uid velocity u,
is the density, p is the pressure and, is the viscosity. The tensor form of
(2.65) is given by
Aij = ui uj + ji p
ui
,
xj
i, j = 1, . . . , n.
A = ui uj + g ij p
ul i j
,
xk xl xk
(2.73)
xl
,
j
j, l = 1, . . . , n.
(2.74)
54
2 Coordinate Transformations
Therefore
m
l
ul
p x
=
u
xk
m
p xk
=
2 xl m
up xl m
+ up p m
.
m
p
k
x
xk
u
, i, j, m, p = 1, . . . , n,
pm
x x x
m
since (2.37) applies. Thus (2.73) has the form
i
u
ij
i j
ij
mj
i
p
A = u u + g p g
+ pm u ,
m
i, j, m, p = 1, . . . , n, (2.75)
g
+
u
J
u
pm
J j
m
k
i
i
k j
kj
mj u
k
p
+ kj u u + g p g
+ pm u
= F ,
m
i, j, k, m, p = 1, . . . , n.
(2.76)
The application of (2.70) to (2.72) yields the following system of stationary
equations:
1
j
ui kj
i j
p kg
J u u +
= F i ,
J j
xi
j
uj = uk k , i, j, k = 1, . . . , n.
(2.77)
x
Now, as an example of the utilization of the Christoel symbols of the
i
, we write out the expression for the transformed elements of
second kind kj
the tensor
i
u
uj
+
ij =
, i, j = 1, . . . , n,
(2.78)
xj
xi
in the coordinates 1 , . . . , n , obtained in accordance with the rule (2.66). This
tensor is very common and is important in applications simulating deformation in the theory of elasticity and deformation rate in uid mechanics. Using
the notations described above, the tensor ij can be expressed in the coordinates 1 , . . . , n through the metric elements and the Christoel symbols of
the second kind. For the component ij we have
i j
ij = mk m k
x x
i
j
jl u
il u
jl i
il j
p
= g
+g
+ (g pl + g pl )u ,
l
l
i, j, l, p = 1, . . . , n.
55
(2.79)
This formula is obtained rather easily. For this purpose one can use the relation
(2.74) for the inverse transition from the contravariant components ui to the
Cartesian components uj of the vector u = (u1 , . . . , un ) and the formula
(2.37). By substituting (2.74) in (2.78), carrying out dierentiation by the
chain rule, and using the expression (2.37), we obtain (2.79).
n , t [0, 1],
(2.80)
where the variable t represents the time and Xtn is an n-dimensional domain
whose boundary points change smoothly with respect to t. It is assumed that
x(t, ) is suciently smooth with respect to i and t and, in addition, that
it is invertible for all t [0, 1]. Therefore there is also the time-dependent
inverse transformation
(2.81)
(t, x) : Xtn n
for every t [0, 1]. The introduction of these time-dependent coordinate transformations enables one to compute an unsteady solution on a xed uniform
grid in n by the numerical solution of the transformed equations.
2.5.1 Reformulation of Time-Dependent Transformations
Many physical problems are modeled in the form of nonstationary conservation-law equations which include the time derivative. The formulas of
Sects. 2.3 and 2.4 can be used directly, by transforming the equations at every
value of time t. However, such utilization of the formulas does not inuence
the temporal derivative, which is transformed simply to the form
i
+
,
t
t i
i = 1, . . . , n,
so that does not maintain the property of divergency and its coecients are
not derived from the elements of the metric tensor.
56
2 Coordinate Transformations
Instead, the formulas of Sects. 2.3 and 2.4 can be more successfully applied
to time-dependent conservation-law equations if the set of the functions x(t, )
is expanded to an (n + 1)-dimensional coordinate transformation in which the
temporal parameter t is considered in the same manner as the spatial variables.
To carry out this process we expand the n-dimensional computational and
physical domains in (2.80) to (n + 1)-dimensional ones, assuming
X n+1 =
t Xtn .
n+1 = I n ,
t
(2.83)
satises
0 (x0 ) = x0 ,
i = 1, . . . , n,
wi = wj
i
xj i
=
,
j
x
xj
57
i, j = 1, . . . , n.
Therefore
w = w i x i ,
i.e.
wi =
i = 1, . . . , n,
xi
xi
= wj j ,
(2.84)
i, j = 1, . . . , n.
i, j = 1, . . . , n.
(2.85)
i, k, m = 0, 1, . . . , n,
(2.86)
diering from (2.45) only by the range of the indices. As a result, we obtain
from (2.86) for i = 0,
k m
x
x
1
J= m
, k, m = 0, 1, . . . , n,
= divx
k
J
j x
J= m w
j
J
xk
=
2 k
m
wm
i x
+
w
,
m
j m xk
j, k, m = 1, . . . , n.
58
2 Coordinate Transformations
Now, taking advantage of the formula for dierentiation of the Jacobian (2.45),
in the last sum of this equation we have
1
wm
J
1
J=
+ wj j ,
J
m
J
j, m = 1, . . . , n,
and consequently
1
1
J=
(Jwj ),
J
J j
j = 1, . . . , n.
(2.87)
j
J i = 0, i, j = 0, 1, . . . , n.
j
x
(2.88)
j
(J) + j J
= 0,
j = 1, . . . , n,
(2.89)
J j (Jwj ) = 0,
j = 1, . . . , n,
(2.90)
which corresponds to (2.87). For i > 0 the identity (2.88) coincides with (2.47),
i.e.
j
J i = 0, i, j = 1, . . . , n.
j
x
As a result of (2.89), we obtain, in analogy with (2.50),
f
1
1
j
k
=
f
=
(Jf
)
(Jw
f
)
,
J
t
J j
t
J
k
j = 0, 1, . . . , n, k = 1, . . . , n.
(2.91)
i = 1, . . . , n.
(2.92)
Using (2.88), in analogy with (2.56), this equation is transformed in the coordinates 0 , 1 , . . . , n , 0 = to
1
J
j
(JA0 )
j
59
= F,
j = 0, 1, . . . , n,
(2.93)
where by A0 we denote the jth contravariant component of the (n + 1)dimensional vector A0 = (A0 , A1 , . . . , An ) in the basis x0 / i , i = 0, 1, . . . , n,
i.e.
j
j
A0 = Ai i , i, j = 0, 1, . . . , n.
(2.94)
x
j
A0 = A0
j
j
j
+ Ai i = A A0 wj ,
t
x
i, j = 1, . . . , n,
A0 = Ak
0
= A0 ,
xk
k = 0, 1, . . . , n.
j
0
0 j
(JA ) + j [J(A A w )] = F, j = 1, . . . , n.
(2.95)
J
j = 1, . . . , n,
= j J g jk k + f wj , j, k = 1, . . . , n.
(2.96)
(2.97)
60
2 Coordinate Transformations
i = 1, 2, 3,
(2.98)
j = 1, 2, 3.
(2.99)
ConvectionDiusion Equation
The unsteady convectiondiusion conservation equation
i
() +
(u )
= S, i = 1, . . . , n,
t
xi
xi xi
has the form in the coordinates , 1 , . . . , n
j
j
kj
(J) + j J(u w ) Jg k = JS,
(2.100)
j, k = 1, . . . , n. (2.101)
(e + u2 /2) + j uj (e + u2 /2 + p/) = Fj uj ,
t
x
j = 1, 2, 3,
(2.102)
where
p=p
e = e(, p),
u2 =
ui
,
xi
i = 1, 2, 3,
3
(ui )2 ,
i=1
is transformed
J e +
1
1
m k
m k
j
j
j
gmk u u
+ j J e + gmk u u (u w ) + Jpu
2
2
m
= Jgmk f uk ,
j, m, k = 1, 2, 3,
(2.103)
(Jui ),
J i
i = 1, 2, 3.
61
(2.104)
arises in many areas such as uid dynamics, elasticity, acoustics, and magnetohydrodynamics. If the coecient c is constant than (2.104) has a divergent
form (2.92) with
A0 = ut ,
Ai = c2
u
,
xi
i = 1, . . . , n,
u
,
i
Ai = c2
u k
,
k xi
i, k = 1, . . . , n.
u
u
+ j J u wj + (c2 g mj wi wj ) i
= 0. (2.105)
J u w i i
Another representation of the linear wave equation (2.104) in the coordinates , 1 , . . . , n comes from the formula (2.64) for the Laplace operator and
the description of the temporal derivative (2.91). Taking advantage of (2.91),
we obtain
1
u
k u
utt =
J
k Jw
J
t
t
u
1
=
J u w i i
J
1
k
i u
Jw u w
.
J k
i
This equation and (2.64) allow one to derive the following form of (2.104) in
the coordinates , 1 , . . . , n :
i u
J u w
i
u
= k Jwk u wj i
+ c2 k Jg kj j ,
(2.106)
62
2 Coordinate Transformations
(2.107)
(2.108)
kj
(J) + j Jg k = JS, j, k = 1, . . . , n,
and
1
m
m k
+ j (Jpuj ) = Jgmk f uk ,
J e + gmk u u
j, m, k = 1, 2, 3,
respectively.
In the same manner, the equations can be written in the EulerLagrange
form, where some coordinates are Lagrangian while the rest are Cartesian
coordinates.
2.5.4 Equations in the Form of Vector Conservation Laws
Now we consider a formula for a vector conservation law with time-dependent
physical magnitudes Aij
ij
A = F i,
xj
i, j = 0, 1, . . . , n,
(2.109)
63
ij
A0 = Amn
i, j, m, n = 0, 1, . . . , n,
i, j, k, l = 0, 1, . . . , n,
i, j = 0, 1, . . . , n.
As in the case of the scalar conservation law, we represent all of the terms of
(2.110) through A00 and the spatial components:
i j
,
xk xm
2 xl i
i
= k j l,
kj
x
i
i
F = Fj j ,
x
i
xj i
=
wi =
,
t
xj
ij
A = Akm
i, j, m, n = 1, . . . , n,
i, j, k, l = 1, . . . , n,
i, j = 1, . . . , n,
i, j = 1, . . . , n.
ij
For A0 we obtain
00
A0 = A00 ,
0i
i
i
0i
+ A0m m = A A00 wi ,
t
x
i0
= A A00 wi ,
A0 = A00
i0
A0
ij
A0
0j
i0
ij
= A00 wi wj A wi A wj + A ,
i = 1, . . . , n,
i = 1, . . . , n,
i, j = 1, . . . , n.
kj = 0,
i
k, j = 0, 1, . . . , n,
i
w
i
+ wl wm lm
,
t
i
w
i
= 0j
=
+ wl jli ,
j
00 =
i
j0
i
i
jk = jk
,
i
and for F 0 ,
i, l, m = 1, . . . , n,
i, j, l = 1, . . . , n,
i, j, k = 1, . . . , n,
64
2 Coordinate Transformations
0
F 0 = F 0,
i
F 0 = F A0 wi ,
i = 1, . . . , n.
Using these expression in (2.110), we obtain a system of equations for the vector conservation law in the coordinates , 1 , . . . , n with an explicit expression
for the components of the speed of the grid movement:
0i
(JA00 ) + j [J(A A00 wj )] = JF 0 ,
i0
ij
0j
i0
J(A A00 wj ) + j J(A + A00 wi wj A wj A wi )
i
i
w
00
l w
l j i
+w
+ w w lj
+ JA
(2.111)
l
i
w
j0
0j
+ J(A + A 2A00 wj )
+ wl jli
j
lj
0j
0l
= J(F F 0 wi ),
i, j, l = 1, . . . , n.
1
(JAi0 ) + j J Aik k Ai0 wj
= F, j, k = 1, . . . , n. (2.112)
J
x
Recall that this approach is not restricted to a square form of the system
(2.109), i.e. the ranges for the indices i and j can be dierent.
As an illustration of these equations for a vector conservation law in the
curvilinear coordinates , 1 , . . . , n , we write out a joint system for the conservation of mass and momentum, which in the coordinates t, x1 , x2 , x3 has
the following form:
+
ui = 0, i = 1, 2, 3,
t
xi
ui
+
(ui uj + pji ) = f i ,
t
xj
(2.113)
i, j = 1, 2, 3,
where
p=p
ui
,
xi
ji = 0 if i = j
and
ji = 1
if i = j.
i.e.
u1
1 1
u u + p
u2 u1
u3 u1
u1
ij
(A ) =
u2
u3
65
u3
u1 u3
,
u2 u3
u3 u3 + p
u2
u1 u2
u2 u2 + p
u3 u2
i, j = 0, 1, 2, 3
(2.114)
and
F 0 = 0,
F i = f i ,
i = 1, 2, 3.
0i
ij
= ,
i0
=A
= (ui wi ),
i = 1, 2, 3,
= (u w )(u w ) + pg ij ,
i
i, j = 1, 2, 3.
(J) + j [J(uj wj )] = 0, i = 1, 2, 3,
wi
wi
+ J(2uj wj ) j + J(ul uj + pg lj )lji
+ J
= Jf ,
(2.115)
i, j, l = 1, 2, 3.
(J) = 0,
(2.116)
ui
ui
i
J
+ j (Jpg ij ) + Juj j + J(ul uj + pg lj )lji = Jf .
Note that the rst equation of the system (2.115) coincides with (2.99) if
F = 0; this was obtained as the scalar mass conservation law.
In the same manner, we can obtain an expression for the general Navier
Stokes equations of mass and momentum conservation by inserting the tensor
( ij ) described by (2.78) in the system (2.113) and the tensor ( ij ) represented
by (2.79) in the system (2.115).
66
2 Coordinate Transformations
A divergent form of (2.113) in arbitrary coordinates , 1 , . . . , n is obtained by applying (2.112). With this, we obtain the system
(J) + j [J(uj wj )] = 0,
i
i j
j
(Ju ) + j J u (u w ) + p i
= JF i ,
x
j
uj = uk k , i, j, k = 1, 2, 3,
x
(2.117)
(J) = 0,
j
(Jui ) + j Jp i = JF i ,
(2.118)
i, j = 1, 2, 3.
2.6 Comments
Many of the basic formulations of vector calculus and tensor analysis may be
found in the books by Kochin (1951), Sokolniko (1964) and Gurtin (1981).
The formulation of general metric and tensor concepts specically aimed
at grid generation was originally performed by Eiseman (1980) and Warsi
(1981).
Very important applications of the most general tensor relations to the
formulation of unsteady equations in curvilinear coordinates in a strong conservative form were presented by Vinokur (1974). A strong conservation-law
form of unsteady Euler equations also was also described by Viviand (1974).
A derivation of various forms of the NavierStokes equations in general
moving coordinates was described by Ogawa and Ishiguto (1987).
http://www.springer.com/978-90-481-2911-9