Вы находитесь на странице: 1из 2

Construction of Fermi-Walker transported frames

Sergei Winitzki
April 2008

Initially motivated by arxiv:0804.2502 Now we assume that 2 6= 0, continue the construction and
We work in an (1 + n)-dimensional Lorentzian manifold (sig- consider the derivative
nature + ...) and assume that a timelike curve ( ) is
given, where is the proper time along the curve. c := u e2 .

Besides c e2 we also have c u because


0.1 Frenet-Serret construction
Recall that in classical differential geometry one talks about g(u, c) = g(u u, e2 ) = 1 g(e1 , e2 ) = 0.
normals, binormals, etc. So we perform a similar construction
However, c may have a component parallel to e1 :
for the curve . First we define the normalized tangent vector
u := , g(u, u) = 1, and its covariant derivative along the
g(e1 , c) = g(u e1 , e2 ) = 2 .
curve,
a := u u. Hence
The usual interpretation of a is the proper 4-acceleration of the c = u e2 =: 2 e1 + 3 e3 ,
worldline . If nonzero, the vector a is spacelike but not nec- where e is a normalized spacelike vector orthogonal to u, e , e
3 1 2
essarily normalized. Hence there exists a normalized spacelike and is an appropriate constant. The vector e is well-defined
3 3
vector e1 such that g(e1 , e1 ) = 1 and as long as 6= 0. If = 0 then we may define e as paral-
3 3 3

a = 1 e1 , lel propagation along of an arbitrary initial vector that is


orthogonal to u, e1 , e2 at some point on .
where 1 is a number (the absolute value of the proper accel- We can continue this construction until we exhaust the di-
eration; the minus sign is a purely cosmetic convenience). If mensionality of spacetime and define an orthonormal frame
1 = 0 then the curve is a geodesic and the construction ends {u, e 1 , ..., e n and the corresponding constants 1 , ..., n . To
}
here; any orthonormal frame is parallelly transported along . illustrate the construction, let us consider what happens at the
So we assume that 1 6= 0. k-th step. At the previous step we have defined orthonormal
The vector e1 is automatically orthogonal to u since vectors {u, e 1 ..., ek } and constants 1 , ..., k , some of which
,
may be zero, and we have established the relationships (called
1 1 the Frenet-Serret equations)
g(e1 , u) = g(u u, u) = u g(u, u) = 0.
1 21
u ej = j+1 ej+1 j ej1 , 1 j k 1.
We will eventually construct an orthonormal frame out of
{u, e1 , ...}, so let us define also e0 := u. (The minus sign in the definition of 1 plays its cosmetic role
We now continue the same construction and compute the here at j = 1.) Now we consider the vector e . This vector
u k
covariant derivative of e1 along the curve, is orthogonal to e and also to u because
k

b := u e1 .
g(u ek , u) = g(ek , u u) = 0.
The vector b is orthogonal to e1 but not necessarily to u; let us
compute its projection onto u using the property g(u, e1 ) = 0: Similarly, for k 6= j 1 we have

g(u, u e1 ) = g(u u, e1 ) = g(1 e1 , e1 ) = 1 . g(u ek , ej ) = g(ek , u ej ) = 0.

Hence there exists 2 and a normalized spacelike vector e2 such However, the vector u ek may have a nonzero projection onto
that e2 e1 , e2 u and ek1 :

b = u e1 = 1 u + 2 e2 . g(u ek , ek1 ) = g(ek , u ek1 )


= g(ek , k ek k1 ek2 ) = k .
The vector e2 is well-defined as long as 2 6= 0. Let us briefly
check what happens if 2 = 0. In that case, we may choose the Hence we obtain the relationship
vector e2 arbitrarily at any point. as long as it is orthogonal to
u and e1 . Moreover, we may choose e2 parallelly transported u ek = k+1 ek+1 k ek1 ,
along ; it is easy to see that the parallel transport will conserve
the properties e2 e1 , e2 u. Let us therefore perform this which defines the vector ek+1 and the constant k+1 . If
choice if 2 = 0. the constant k+1 vanishes, the vector ek+1 is chosen as a
parallel transport of an arbitrary initial vector orthogonal to Let us derive the equation for the FW transport directly in
{u, e1 , ..., ek }. terms of the vector v and its derivative:
At the last step of the construction, the constant n+1 must
3 3
vanish since no new spacelike vector en+1 can be found in X X
the spacetime. Therefore, the construction ends by defining u v = u v0 u + vj e j = v0 a + vj u ej
j=1 j=1
an orthonormal frame {e0 , ..., en } and an array of constants
{1 , ..., n } of which some may vanish. (By construction, if X3

some j = 0 then j+1 = ... = n = 0 as well.) The frame = ag(u, v) + u g(ej , v)g(a, ej )
{e0 , ..., en } is in some sense naturally adapted to the curve j=1

(since it is constructed purely out of and its derivatives). = ag(u, v) ug(a, v).

This is the basic equation of the FW transport.


0.2 Fermi-Walker transport
Consider an observer who moves along the timelike, non- 0.3 Computing the Fermi-Walker frame
geodesic worldline in a 3+1-dimensional spacetime. We would
like to construct the observers reference frame that does not How can we compute the FW propagated frame { ej } with re-
have any spatial rotation; this will be appropriate if the observer spect to a timelike worldline( )? The method in Maluf and
keeps track of the spatial directions by using strong gyroscopes. Faria is to take an arbitrary spacelike frame {ej }, j = 1, 2, 3,
The timelike basis vector is naturally chosen as e0 = . The along and compute a local proper rotation ( ) of the frame
main problem is to obtain the spacelike frame {e1 , e2 , e3 }. {ej },
3
Suppose some frame {e } is given; then we can compute the X
accelerations u e and express them as linear combinations of j :=
ej 7 e j k ( )ek , j = 1, 2, 3,
the frame vectors, k=1

such that the rotated frame {


ej } is FW propagated. (Of course,
3
X
the timelike vector e0 is unchanged by the rotation.) We need
u e =: f e .
to determine the -dependent matrix j k ( ). To determine this
=0
matrix, let us write the equations for the covariant derivatives
This defines an antisymmetric matrix f . The spatial compo- of {ej } and {
ej } along :
nents of that matrix (f1 2 , f2 3 , f3 1 ) describe the purely spatial
3
rotation of the frame. Therefore, we wish to choose the space- X
like vectors {e1 , e2 , e3 } such that these components vanish. u ej = Aj u + fj k ek ,
k=1
If the spacelike vectors {ej } were so chosen, we would have
j = Bj u.
u e
u ej = Aj u, Aj fj 0 , j = 1, 2, 3.
Since the frame {ej } is given, we may compute the coefficients
The coefficients Aj satisfy fj k ( ); if these coefficients are nonzero, the frame {ej } is not
FW propagated. Let us now substitute e j:
j = e
Aj = g(u, u ej ) = g(u u, ej ) = g(a, ej ),
3
!
X
where a := u u is the proper 4-acceleration of . Therefore, we k
Bj u = u e j = u j ek
may compute the suitable vectors {ej } by solving the ordinary k=1
differential equations (along ), 3   3 3
!
X d l
X
k
X
l
= j el + j Ak u + fk el
u ej = g(a, ej ), d
l=1 k=1 l=1
3 3
! 3
with arbitrary (but orthonormal) initial vectors ej . The frame
X d l
X
k l
X
= j + j fk el + u j k Ak .
{ej } obtained in this way is called the Fermi-Walker propa- d
l=1 k=1 k=1
gated frame with respect to the given curve .
Given the FW propagated frame, one can define the notion The frame { ej } is FW propagated if the terms at el above
of Fermi-Walker transport (along ) of an arbitrary vector v vanish: !
3 3
defined at some point p of . One first decomposes the vector X d l
X
k l
j + j fk = 0.
v into components with respect to the FW propagated frame d
l=1 k=1
{u, ej } and obtains the coefficients
This is a differential equation for that can be rewritten in the
3
X matrix form as
v = v0 u + vj ej , 1 d
+ f = 0.

j=1 d
v0 = g(u, v), Practical solution of this equation is perhaps easier when one
vj = g(ej , v), j = 1, 2, 3. uses the orthogonality property 1 =
T and the quaternionic
representation of rotations.
Then one defines the FW transported vector v at all other
points of as the linear combination of {u, ej } with the same
coefficients v0 , ..., v3 .

Вам также может понравиться