You are on page 1of 8

Chapter 4

Vectors in Astrometry, an
Introduction

4.1 Vectors in Astrometry

In astrometry, vectors are extensively used to describe the geometrical relationships among
celestial bodies, for example between the observer and the observed object. Practical cal-
culations using computer software are today mainly carried out with the help of vector and
matrix algebra, rather than the trigonometry formulae typically found in older textbooks.
It turns out that this often provides a better insight into the problem, and hence reduces
the risk of errors in the derived algorithms, in addition to being advantageous in terms of
computational speed and accuracy.
This chapter provides a brief introduction to the use of vectors and matrices in astrome-
try. It broadly uses the notational conventions from C.A. Murray’s Vectorial Astrometry
(1983), which seem to provide a particularly clear and consistent framework for theoretical
work as well as practical calculations. By way of illustration some useful transformations
are explained in detail, while references to the general literature are provided for other
applications. Only vectors in three-dimensional Euclidean space are considered.

4.1.1 What are vectors?

Vectors and matrices Classically, a vector is defined as a physical entity having both
magnitude (length) and direction, as opposed to a scalar that only has magnitude. Vectors
can be visualized as arrows that exist in space quite independently of any coordinate
system. The usual vector operations – addition, subtraction, multiplication by a scalar,
scalar (dot) product, and vector (cross) product – have simple geometrical interpretations
that are independent of the coordinate system. For example, the sum of two vectors a and
b can be constructed by means of the usual parallelogram, and their scalar product can be
obtained as ab cos ✓, where a and b are the lengths of the vectors and ✓ the angle between
them. Much of the theoretical development in astrometry results in vector expressions
that can be interpreted in this general way without reference to any particular coordinate
system.

1
2 CHAPTER 4. VECTORS IN ASTROMETRY, AN INTRODUCTION

When the vector expressions are being used for actual calculations, it is however necessary
to agree on a numerical representation of the vectors and to map the vector operations
into a corresponding set of numerical operations. Normally this is done by using Cartesian
coordinates, arranged as column (3 ⇥ 1) matrices; thus:
2 3 2 3
ax bx
a = a y , b = by 5
4 5 4 (4.1)
az bz

where ax is the coordinate of a along the adopted x axis, etc. All the above-mentioned
vector operations have simple and well-known equivalents in Cartesian coordinates, e.g.,
2 3 2 3
a x + bx a y bz a z by
a + b = 4 a y + b y 5 , a0 b = a x b x + a y by + a z b z , a ⇥ b = 4 a z b x a x b z 5 (4.2)
a z + bz a x by a y bx

Following Murray (1983) the prime (0 ) is here used to denote both the scalar product of
vectors and the transpose of a matrix. Most of the time there is no need to worry about
the distinction between the vector as a physical entity and its coordinate representation
by means of a matrix. The a0 b in Eq. (4.2) can therefore be interpreted either as the
scalar product of vectors a and b, or as row matrix a0 (the transpose of column matrix a)
multiplied by column matrix b.

p
Unit vectors The length of vector a is the non-negative scalar quantity |a| = a0 a =
(a2x + a2y + a2z )1/2 . Sometimes the corresponding variable in italic (a) is used to denote the
length of a vector.
Unit vectors (of length 1) are ubiquitous in astrometry as representing directions, for
example from the observer towards an object, the tangent direction of a light ray, or the
direction of a coordinate axis. Constructing the unit vector from a given non-zero vector
a is a common operation for which the special symbol h i is sometimes used,
1
hai = a|a| (4.3)

Some important formulae For arbitrary vectors or column matrices a, b, c we have

b 0 a = a0 b (4.4)
b⇥a= (a ⇥ b) (4.5)
0 0 0 0 0 0
(a ⇥ b) c = (b ⇥ c) a = (c ⇥ a) b = a (b ⇥ c) = b (c ⇥ a) = c (a ⇥ b) (4.6)
0 0
(a ⇥ b) ⇥ c = bc a ab c (4.7)

4.1.2 Coordinate systems and triads

In Eqs. (4.1)–(4.3) the coordinate system is not explicitly defined but implied by the use of
Cartesian coordinates ax , etc. This is fine as long as we are consistently working in a single
coordinate system. When more than one system is involved, for example in coordinate
transformations, it helps to clarify the relation between the vectors and their di↵erent
4.1. VECTORS IN ASTROMETRY 3

coordinate representations if an explicit notation is introduced for the systems. This can
be done by means of coordinate triads.
A coordinate triad is a row matrix of three mutually orthogonal unit vectors, for example
Z = [x y z], representing the axes of a Cartesian coordinate system. From the orthogo-
nality and unit length of the vectors, it follows that Z0 Z = I, the 3 ⇥ 3 identity matrix.1
Coordinate triads are normally right-handed, thus x ⇥ y = z or det(Z) = (x ⇥ y)0 z = +1.
In the system defined by the coordinate triad Z = [x y z] the Cartesian coordinates of
the arbitrary vector a are given by the column matrix
2 03 2 0 3 2 3
x xa ax
Z a = y a = y a = ay 5
0 4 0 5 4 0 5 4 (4.8)
z0 z0 a az

The vector can be written in terms of its coordinates with respect to Z as


3 2
ax
a = xax + yay + zaz = Z 4ay 5 (4.9)
az

Introducing now a second coordinate system, represented by the triad K = [i j k], it is


seen that the coordinates of a in the new system are given by the column matrix
2 3 2 3
ai ax
4 a j 5 = K0 a = K0 Z 4 a y 5 (4.10)
ak az

Transforming the coordinates of a (or any other vector) from the Z to the K system is
therefore accomplished through pre-multiplication with the 3 ⇥ 3 matrix K0 Z. Conversely,
transforming the coordinates from K to Z is done through pre-multiplication with Z0 K.
Note that the latter matrix is the transpose of K0 Z, and also its inverse, since they are
orthogonal matrices.

4.1.3 Spherical coordinates

Celestial positions are often specified by means of spherical coordinates, for example (↵, )
in the equatorial system or (l, b) in the galactic system. Transformation of spherical coordi-
nates from one system to another is conveniently done by means of a matrix multiplication
as in Eq. (4.10). This involves some additional trigonometric operations, to be discussed
below, for the conversions between spherical and Cartesian coordinates.

From spherical coordinates to vector In the generic coordinate system Z = [x y z]


let ( , ✓) be the spherical coordinates representing the direction of the non-zero vector a.
The angles are defined in the usual astronomical sense, with the longitude-like angle
1
It is a useful exercise to write out this equation in full. When transposing the matrix Z it is necessary
to transpose its elements (x, etc) as well, as they are not scalar quantities; cf. Eq. (4.8).
4 CHAPTER 4. VECTORS IN ASTROMETRY, AN INTRODUCTION

(for example ↵ or l) and ✓ the latitude-like angle (for example or b). The Cartesian
coordinates of a in Z are 2 3 2 3
ax a cos ✓ cos
Z0 a = 4ay 5 = 4 a cos ✓ sin 5 (4.11)
az a sin ✓
where a = |a| is the length of the vector. If a is the unit vector towards ( , ✓), then a = 1
in (4.11).

From vector to spherical coordinates The inversion of (4.11) requires some care in
order to avoid potential numerical difficulties and ambiguities. The following formulae are
recommended because they work in all reasonable circumstances and give good numerical
accuracy even for positions close to the poles:2
⇣ q ⌘
= atan2 (ay , ax ) , ✓ = atan2 az , a2x + a2y (4.12)

Here atan2(y, x) is the four-quadrant inverse tangent available in the mathematical li-
braries of many programming languages, including C++, Fortran, Java, MATLAB, Perl
and Python. Equation (4.12) returns in the interval [ ⇡, ⇡], which is usually fine for
subsequent calculations, and there is normally no need to ensure that the angle falls in
the ‘standard’ range [0, 2⇡) by adding 2⇡ for negative values.

The normal triad Associated with the coordinate system Z and spherical coordinates
( , ✓) are the three orthogonal unit vectors
2 3 2 3 2 3
sin sin ✓ cos cos ✓ cos
p = Z 4 cos 5 , q = Z 4 sin ✓ sin 5 , r = Z 4 cos ✓ sin 5 (4.13)
0 cos ✓ sin ✓
which form the so-called ‘normal triad’ [p q r] at point ( , ✓) with respect to Z. The third
vector r is just the unit vector towards the point, as seen from (4.11). Given r, the first
two vectors can be computed using vector algebra,
p = hz ⇥ ri, q=r⇥p (4.14)

The significance of p and q becomes apparent when r is di↵erentiated with respect to the
spherical coordinates,
2 3
cos ✓ sin
@r @r
= Z 4 cos ✓ cos 5 = p cos ✓, =q (4.15)
@ @✓
0
resulting in the total di↵erential
dr = p d cos ✓ + q d✓ (4.16)
p is therefore the unit tangent vector in the direction of increasing longitudinal angle ,
and q the unit tangent vector in the direction of increasing latitudinal angle ✓. (Note that
d cos ✓ is the ‘true angle’ representation of the longitude di↵erential.)
2
The first formula in Eq. (4.12) fails if both ax and ay are exactly zero, in which case is undefined.
If a = 0, both formulae fail and the spherical coordinates are completely undefined. In a computer
implementation of (4.12) one has to decide whether these conditions should result in an error condition,
or the angles being set to some conventional values. Both solutions have their pros and cons.
4.1. VECTORS IN ASTROMETRY 5

The proper motion vector The normal triad is particularly useful in dealing with
proper motions. The proper motion of a star is usually expressed as the time derivatives
of the spherical coordinates (as viewed from the Solar System Barycentre), µ = d / dt,
µ✓ = d✓/ dt. For the longitudinal component, the cos ✓ factor is usually included; in
the Hipparcos and Tycho Catalogues (ESA 1997) this is indicated by an asterisk, µ ⇤ =
( d / dt) cos ✓. Then µ ⇤ and µ✓ can be regarded as the projections of the proper motion
vector ( dr/ dt) on p and q, respectively.
When and ✓ are functions of time, for example because of the object’s proper motion,
it means that the normal triad computed from (4.13) is also changing with time. It
would however be highly inconvenient to express the proper motion components in such
a continually changing coordinate system. The practical solution is to use a fixed normal
triad, usually the one computed for the values of and ✓ corresponding to the barycentric
position at the reference epoch of the catalogue. Let u(t) be the barycentric direction
towards the star, which is then in general di↵erent from r. The proper motion vector is
given by
du
µ⌘ = p µ ⇤ + q µ✓ (4.17)
dt
and conversely
µ ⇤ = p0 µ, µ✓ = q 0 µ (4.18)

It is important to remember that p and q depend not only on the position of the object
(i.e., on r) but also on the chosen coordinate system. Thus, when transforming from one
coordinate system to another, the proper motion vector in Eqs. (4.17)–(4.18) remains the
same but its components along p and q will be di↵erent in the two systems.

4.1.4 Rotations

Rotating the arbitrary vector a by the angle ✏ about the unit vector e results in the new
vector
b = a cos ✏ + ee0 a(1 cos ✏) + (e ⇥ a) sin ✏ (4.19)

In particular, application of the above rotation (e, ✏) to the coordinate triad Z = [x y z]


results in the new triad

Z̃ ⌘ [x̃ ỹ z̃] = Z cos ✏ + ee0 Z (1 cos ✏) + (e ⇥ Z) sin ✏ (4.20)

Now let r be some vector fixed in space. According to (4.10), the coordinates [rx̃ , rỹ , rz̃ ]0
of the vector in the rotated system are obtained through pre-multiplication of the original
coordinates [rx , ry , rz ]0 by the 3 ⇥ 3 matrix Z̃0 Z. Using Eq. (4.20) we find

Z̃0 Z = Z0 Z cos ✏ + Z0 ee0 Z (1 cos ✏) + (e ⇥ Z)0 Z sin ✏


2 2 3 2 3
ex ex ey ex ez 0 ez ey
= I cos ✏ + 4ey ex e2y ey ez 5 (1 cos ✏) + 4 ez 0 ex 5 sin ✏ (4.21)
ez ex ez ey e2z ey ex 0

It may be noted that the components of e are the same in the two systems.
6 CHAPTER 4. VECTORS IN ASTROMETRY, AN INTRODUCTION

Equation (4.21) is much simplified when e coincides with one of the axes of Z. Putting e
in turn equal to x, y, and z gives the elementary rotation matrices
2 3 2 3 2 3
1 0 0 c✓ 0 s✓ c s 0
4
Rx ( ) = 0 c 5 4
s , Ry (✓) = 0 1 0 , Rz ( ) = 5 4 s c 05 (4.22)
0 s c s✓ 0 c✓ 0 0 1

(using, for brevity, s and c for the sine and cosine). Since any rotation matrix can be
decomposed as the product of three (or more) elementary rotation matrices, it is often
convenient to express the relation between two coordinate systems by means of succes-
sive elementary rotations, rather than using (e, ✏). The minimum set of three successive
rotation angles is sometimes referred to as the Euler angles, but there are many possible
conventions in use depending on the order in which the axes are taken (Wertz 1978). An
alternative way of representing general rotations is by means of unit quaternions (Wertz
1978), which are usually the preferred choice for example in spacecraft attitude control
and computer games.
The rotation matrices (4.22) transform the coordinates of a fixed vector when the coordi-
nate system is rotated. What happens if instead a vector is rotated in a fixed coordinate
system? The answer is found by resolving (4.19) into the coordinates of a fixed triad, say
Z = [x y z]:

Z0 b = Z0 a cos ✏ + Z0 ee0 a(1 cos ✏) + Z0 (e ⇥ a) sin ✏


= I cos ✏ + Z0 ee0 Z(1 cos ✏) + Z0 (e ⇥ Z) sin ✏ Z0 a (4.23)

since e ⇥ a = e ⇥ (ZZ0 a) = (e ⇥ Z)Z0 a. This shows that the coordinates of b are obtained
through pre-multiplication by a matrix which is the transpose of (4.21). Since the matrix
is orthogonal, its transpose equals the inverse, which is also obtained by reversing the sign
of the rotation angle. For example, if a is rotated an angle about the x axis, the relevant
transformation matrix is R0x ( ) = Rx 1 ( ) = Rx ( ).
In some literature, for example Seidelmann (1992, p. 552), elementary rotation matrices
are defined in the opposite sense of (4.22), but since they are post-multiplied to the vector
coordinates expressed as a row matrix, the end result is the same.3 The reader should be
aware of the several potential ambiguities involved in using rotation matrices, in particular
whether they apply to column or row vectors, and to a rotation of the coordinate system
or of the vector itself.

4.1.5 Example: Conversion between equatorial and galactic coordinates

For transformation between the equatorial and galactic systems, the unit vector represent-
ing the barycentric direction to the star at the adopted reference epoch may be written
2 3 2 3
cos cos ↵ cos b cos l
r = E 4 cos sin ↵ 5 = G 4 cos b sin l 5 (4.24)
sin sin b
3
The conventions used here seem to be the ones most commonly used in astronomical literature (for
example Barbieri 2007, Eichhorn 1974, Kovalevsky & Seidelmann 2004, Ta↵ 1981; also implicitly in Sei-
delmann 1992, pp. 103, 182, etc).
4.1. VECTORS IN ASTROMETRY 7

where E and G are the equatorial and galactic coordinate triads. Pre-multiplication by G0
(remembering that G0 G = I) gives the required formula for transformation from equatorial
to galactic; pre-multiplication by E0 gives the inverse transformation.
The rotation matrix relevant for converting from equatorial to galactic coordinates is
therefore G0 E. Let (↵G , G ) denote the equatorial coordinates of the north galactic pole
and l⌦ the galactic longitude of the first intersection of the galactic plane with the equator.
It is then seen that a triad originally aligned with E can be brought into alignment with
G by the following successive rotations: (i) by the angle ↵G + 90 about the third axis;
(ii) by the angle 90 G about the resulting first axis; (iii) by the angle l⌦ about the
resulting third axis. Consequently

G0 E = Rz ( l⌦ )Rx (90 G )Rz (↵G + 90 ) (4.25)

If E stands for the International Celestial Reference System (ICRS), as realized for example
by the Hipparcos and Tycho Catalogues (ESA 1997), then we may adopt the following
values as defining G:

↵G = 192.85948 , G = 27.12825 , l⌦ = 32.93192 (4.26)

(Vol. 1, p. 91 ESA 1997), from which4


" #
0.054 875 560 416 215 0.873 437 090 234 885 0.483 835 015 548 713
G0 E = +0.494 109 427 875 584 0.444 829 629 960 011 +0.746 982 244 497 219 (4.27)
0.867 666 149 019 005 0.198 076 373 431 202 +0.455 983 776 175 067

In order to transform the proper motion components, the proper motion vector is written

µ = pE µ↵⇤ + qE µ = pG µl⇤ + qG µb (4.28)

where [pE qE r] and [pG qG r] are the normal triads at r with respect to E and G. When
transforming from equatorial to galactic components, use the first equality to compute the
proper motion vector in the E system, then obtain µl⇤ = p0G µ etc as in (4.18). This last
computation can be done in either system. If the G system is used, transform r and µ
into galactic coordinates and compute G0 pG and G0 qG by means of (4.14), noting that
G0 zG = [0, 0, 1]0 ; if E is used, then compute E0 pG and E0 qG from (4.14), noting that
E0 zG makes up the third column of E0 G or the third row of G0 E.

4
The matrix in Eq. (4.27) is the transpose of the AG given in the Hipparcos and Tycho Catalogues,
Vol. 1, Eq. [1.5.11]. This is consistent with the use of A0G in Eq. [1.5.13]. AG is only given to 10 decimals
in that reference, which was sufficient for the data in the Hipparcos Catalogue tabulated to a precision of
0.01 mas. The result is here given to 15 decimals, which is adequate even for nanoarcsec astrometry. Nev-
ertheless, this matrix – or any other finite-precision matrix – cannot be taken as defining the ICRS/galactic
relation; that should be done by means of a set of conventional angles as in Eq. (4.26).
Bibliography

Barbieri C., 2007, Fundamentals of astronomy, Boca Raton, FL: Taylor & Francis

Eichhorn H., 1974, Astronomy of star positions - A critical investigation of star catalogues,
the methods of their construction and their purpose, New York: Ungar

ESA, 1997, The Hipparcos and Tycho Catalogues, ESA SP-1200

Kovalevsky J., Seidelmann P.K., 2004, Fundamentals of Astrometry, Cambridge, UK:


Cambridge University Press

Murray C.A., 1983, Vectorial astrometry, Bristol: Adam Hilger

Seidelmann P.K.E., 1992, Explanatory Supplement to the Astronomical Almanac, Mill


Valley, CA: University Science Books

Ta↵ L.G., 1981, Computational spherical astronomy, New York: Wiley-Interscience

Wertz J.R.E., 1978, Spacecraft Attitude Determination and Control, Astrophysics and
Space Science Library, Vol. 73, Dordrecht: D. Reidel

You might also like