Ellipses and Ellipsoids Required Reading

This required reading document is reproduced from the original NAIF document available at https://naif.jpl.nasa.gov/pub/naif/misc/toolkit_docs_N0067/C/req/ellipses.html

Note

These required readings documents were translated from documentation for N67 CSPICE. These pages may not be updated as frequently as the CSPICE version, and so may be out of date. Please consult the changelog for more information.

Important

NOTE any functions postfixed by "_" mentioned below are Fortan-SPICE functions unavailable in SpiceyPy as the NAIF does not officially support these with "_c" function wrappers within the CSPICE API. If these functions are necessary for your work please contact the NAIF to request that they be added to the CSPICE API

Abstract

SPICE contains a substantial set of subroutines that solve common mathematical problems involving ellipses and triaxial ellipsoids. This required reading file documents those routines, gives examples of their use, and presents some of the mathematical background required to understand the routines.

Introduction

The 'ellipse' is a structured data type used in SPICE to represent ellipses in three-dimensional space. SPICE ellipses exist to simplify calling sequences of routines that output or accept as input data that defines ellipses.

Ellipses turn up frequently in the sort of science analysis problems SPICE is designed to help solve. The shapes of extended bodies--planets, satellites, and the Sun--are frequently modeled by triaxial ellipsoids. The IAU has defined such models for the Sun, all of the planets, and most of their satellites, in the IAU/IAG/COSPAR working group report IAU_IAG_COSPAR. Many geometry problems involving triaxial ellipsoids give rise to ellipses as 'mathematical byproducts'. Ellipses are also used in modeling orbits and planetary rings.

References

  1. 'Report of the IAU/IAG/COSPAR Working Group on Cartographic Coordinates and Rotational Elements of the Planets and Satellites: 2009', December 4, 2010.

  2. 'Calculus, Vol. II'. Tom Apostol. John Wiley and Sons,

    • See Chapter 5, 'Eigenvalues of Operators Acting on Euclidean Spaces'.

  3. Planes required reading (planes).

Ellipse Data Type Description

The following representation of an ellipse is used throughout SPICE, and in particular by the ellipse access routines: An ellipse is the set of points
ellipse = CENTER    +    cos(theta) * V1    +    sin(theta) * V2

where CENTER, V1, and V2 are 3-vectors, and theta is in the range

(-pi, pi].

The set of points "ellipse" is an ellipse (see Appendix A: Mathematical notes). The ellipse defined by this parametric representation is non-degenerate if and only if V1 and V2 are linearly independent. We call CENTER the 'center' of the ellipse, and we refer to V1 and V2 as 'generating vectors'. Note that an ellipse centered at the coordinate origin (0, 0, 0,) is completely specified by its generating vectors. Further mention of the center or generating vectors for a particular ellipse, means vectors that play the role of CENTER or V1 and V2 in defining that ellipse.

This representation of ellipses has the particularly convenient property that it allows easy computation of the image of an ellipse under a linear transformation. If M is a matrix representing a linear transformation, and E is the ellipse

CENTER    +    cos(theta) * V1    +    sin(theta) * V2,

then the image of E under the transformation represented by M is

M*CENTER    +    cos(theta) * M*V1    +    sin(theta) * M*V2.

If we accept that the first set of points is an ellipse, then we can see that the image of an ellipse under a linear transformation is always another (possibly degenerate) ellipse. Since many geometric computations involving ellipses and ellipsoids may be greatly simplified by judicious application of linear transformations to ellipses, it is useful to have a representation for ellipses that allows ready computation of their images under such mappings.

The internal design of the ellipse data type is not part of its specification. The design is an implementation choice based on the programming language and so the design may change. Users should not write code based on the current implementation; such code might fail when used with a future version of CSPICE.

NAIF implemented the SPICE ellipse data type in C as a structure with the fields

SpiceDouble      center    [3];
SpiceDouble      semiMajor [3];
SpiceDouble      semiMinor [3];

In SpiceyPy, this structure is defined by spiceypy.utils.support_types.Ellipse, although users are encouraged not to directly interact with this object and to instead use the spice routines described below for creating and using them.

The fields are set and accessed by a small set of access routines provided for that purpose. Do not access the fields in any other way. The elements of SPICE ellipses are set using cgv2el() (center and generating vectors to ellipse) and accessed using el2cgv() (ellipse to center and generating vectors).

Ellipse and ellipsoid routines

Constructing ellipses

Let 'center', 'v1', and 'v2' be a center vector and two generating vectors for an ellipse.

Let 'center', 'v1', and 'v2' be declared with assigned values:

import numpy as np

center = np.array([ -1.0, 1.0, -1.0 ])
v1     = np.array([  1.0, 1.0, 1.0  ])
v2     = np.array([  1.0, -1.0, 1.0 ])

After 'center', 'v1', and 'v2' have been assigned values, you can construct a SPICE ellipse using cgv2el():

import spiceypy

ellipse = spiceypy.cgv2el( center, v1, v2 )

This call produces the SPICE ellipse 'ellips', which represents the same mathematical ellipse as do 'center', 'v1', and 'v2'. The generating vectors need not be linearly independent. If they are not, the resulting ellipse will be degenerate. Specifically, if the generating vectors are both zero, the ellipse will be the single point represented by 'center', and if just one of the semi-axis vectors (call it V) is non-zero, the ellipse will be the line segment extending from

CENTER - V

to

CENTER + V

Access to ellipse data elements

Let 'ellips' be a SPICE ellipse. To produce the center and two generating vectors for 'ellips', we can make the call
center,  v1,  v2  = spiceypy.el2cgv( ellips )

On output, 'v1' will be a semi-major axis vector for the ellipse represented by 'ellips', and 'v2' will be a semi-minor axis vector. Semi-axis vectors are never unique; if X is a semi-axis vector; then so is -X. 'v1' is a vector of maximum norm extending from the ellipse's center to the ellipse itself; 'v2' is an analogous vector of minimum norm. 'v1' and V2 are orthogonal vectors.

cgv2el() and el2cgv() are not inverses

Because the routine el2cgv() always returns semi-axes as generating vectors, if 'v1' and 'v2' are not semi-axes on input to cgv2el(), the sequence of calls
ellips = spiceypy.cgv2el( center, v1, v2 )
center,  v1,  v2  = spiceypy.el2cgv( ellips )

will certainly modify 'v1' and 'v2'. Even if 'v1' and 'v2' are semi-axes to start out with, because of the non-uniqueness of semi-axes, one or both of these vectors could be negated on output from el2cgv(). There is a sense in which cgv2el() and el2cgv() are inverses, though: the above sequence of calls returns a center and generating vectors that define the same ellipse as the input center and generating vectors.

Triaxial ellipsoid routines

The routines used to perform geometric calculations involving ellipsoids:
edlimb()

Ellipsoid limb

inedpl()

Intersection of ellipsoid and plane

nearpt()

Nearest point on ellipsoid to point

npedln()

Nearest point on ellipsoid to line

sincpt()

Surface intercept

surfnm()

Surface normal on ellipsoid

surfpt()

Surface intercept point on ellipsoid

Ellipse routines

The CSPICE routines used to perform geometric calculations involving ellipses:
inelpl()

Intersection of ellipse and plane

npelpt()

Nearest point on ellipse to point

pjelpl()

Projection of ellipse onto plane

saelgv()

Semi-axes of ellipse from generating vectors

Examples

Finding the 'limb angle' of an instrument boresight

If we want to find the angle of a ray above the limb of an ellipsoid, where the angle is measured in a plane containing the ray and a 'down' vector, we can follow the procedure given below. We assume the ray does not intersect the ellipsoid. Name the result 'angle'.

We assume that all vectors are given in body-fixed coordinates.

  • 'observ' is the body-center to observer vector.

  • 'raydir' is the boresight ray's direction vector in body-fixed coordinates.

  • 'limb' is an ellipse, the result of the limb calculation.

Find the limb of the ellipsoid as seen from the point 'observ'. Here 'a', 'b', and 'c' are the lengths of the semi-axes of the ellipsoid.

limb = spiceypy.edlimb( a, b, c, observ )

The ray direction vector is 'raydir', so the ray is the set of points

OBSERV + t * RAYDIR

where t is any non-negative real number. The 'down' vector is just - 'observ'. The vectors OBSERV and RAYDIR are spanning vectors for the plane we're interested in. We can use psv2pl() to represent this plane by a SPICELIB plane.

plane = spiceypy.psv2pl( observ, observ, raydir )

Find the intersection of the plane defined by 'observ' and 'raydir' with the limb.

nxpts, xpt1, xpt2 = spiceypy.inelpl( limb, plane )

We always expect two intersection points, if 'down' is valid. If 'nxpts' has value less-than two, the user must respond to the error condition. Form the vectors from 'observ' to the intersection points. Find the angular separation between the boresight ray and each vector from 'observ' to the intersection points.

vec1 = spiceypy.vsub( xpt1, observ )
vec2 = spiceypy.vsub( xpt2, observ )

sep1 = spiceypy.vsep( vec1, raydir )
sep2 = spiceypy.vsep( vec2, raydir )

The angular separation we're after is the minimum of the two separations we've computed.

angle = min(sep1, sep2)

Use of ellipses with planes

The nature of geometry problems involving planes often includes use of the SPICE ellipse data type. The example C code listed in the headers of the routines inelpl_c and pjelpl_c show examples of problems solved using both the ellipse and plane data type that can be converted into the equivalent python by readers.

Summary of routines


The following table summarizes the SPICE ellipse and ellipsoid routines.
cgv2el()

Center and generating vectors to ellipse

edlimb()

Ellipsoid limb

edterm()

Ellipsoid terminator

el2cgv()

Ellipse to center and generating vectors

inedpl()

Intersection of ellipsoid and plane

inelpl()

Intersection of ellipse and plane

nearpt()

Nearest point on ellipsoid to point

npedln()

Nearest point on ellipsoid to line

npelpt()

Nearest point on ellipse to point

pjelpl()

Projection of ellipse onto plane

saelgv()

Semi-axes of ellipse from generating vectors

sincpt()

Surface intercept

surfnm()

Surface normal on ellipsoid

surfpt()

Surface intercept point on ellipsoid

surfpv()

Surface point and velocity

Appendix A: Mathematical notes

Defining an ellipse parametrically

Our aim is to show that the set of points
CENTER    +    cos(theta) * V1    +    sin(theta) * V2

where CENTER, V1, and V2 are specified vectors in three-dimensional space, and where theta is a real number in the interval (-pi, pi], is in fact an ellipse as we've claimed. Since the vector CENTER simply translates the set, we may assume without loss of generality that it is the zero vector. So we'll re-write our expression for the alleged ellipse as

cos(theta) * V1    +    sin(theta) * V2

where theta is a real number in the interval (-pi, pi]. We'll give the name S to the above set of vectors. Without loss of generality, we can assume that V1 and V2 lie in the x-y plane. Therefore, we can treat V1 and V2 as two-dimensional vectors. If V1 and V2 are linearly dependent, S is a line segment or a point, so there is nothing to prove. We'll assume from now on that V1 and V2 are linearly independent.

Every point in S has coordinates ( cos(theta), sin(theta) ) relative to the basis

{V1, V2}.

Define the change-of-basis matrix C by setting the first and second columns of C equal to V1 and V2, respectively. If (x,y) are the coordinates of a point P on S relative to the standard basis

{ (1,0), (0,1) },

then the coordinates of P relative to the basis

{V1, V2}

are

           +- -+
      -1   | x |
     C     |   |
           | y |
           +- -+

         +-          -+
         | cos(theta) |
=        |            |
         | sin(theta) |
         +-          -+

Taking inner products, we find

     +-    -+      -1 T     -1   +- -+
     | x  y |   ( C  )     C     | x |
     +-    -+                    |   |
                                 | y |
                                 +- -+


     +-                      -+  +-          -+
=    | cos(theta)  sin(theta) |  | cos(theta) |
     +-                      -+  |            |
                                 | sin(theta) |
                                 +-          -+

=    1

The matrix

   -1  T   -1
( C   )   C

is symmetric; let's say that it has entries

+-          -+
|   a   b/2  |
|            |.
|  b/2   c   |
+-          -+

We know that a and c are positive because they are squares of norms of the columns of

 -1
C

which is a non-singular matrix. Then the equation above reduces to

   2                2
a x   +  b xy  + c y   =  1,     a, c  >  0.

We can find a new orthogonal basis such that this equation transforms to

    2           2
d1 u    +   d2 v

with respect to this new basis. Let's give the name SYM to the matrix

+-          -+
|   a   b/2  |
|            |;
|  b/2   c   |
+-          -+

since SYM is symmetric, there exists an orthogonal matrix M that diagonalizes SYM. That is, we can find an orthogonal matrix M such that

                 +-      -+
 T               | d1   0 |
M  SYM  M    =   |        |.
                 | 0   d2 |
                 +-      -+

The existence of such a matrix M will not be proved here; see reference [2]. The columns of M are the elements of the basis we're looking for: if we define the variables (u,v) by the transformation

+- -+        +- -+
| u |      T | x |
|   |  =  M  |   |,
| v |        | y |
+- -+        +- -+

then our equation in x and y transforms to the equation

    2           2
d1 u    +   d2 v

since

     2                 2
    a x   +  b xy  +  c y

     +-    -+              +- -+
=    | x  y |      SYM     | x |
     +-    -+              |   |
                           | y |
                           +- -+

     +-    -+   T          +- -+
=    | u  v |  M   SYM  M  | u |
     +-    -+              |   |
                           | v |
                           +- -+

     +-    -+  +-      -+  +- -+
=    | u  v |  | d1   0 |  | u |
     +-    -+  |        |  |   |
               | 0   d2 |  | v |
               +-      -+  +- -+


         2            2
=    d1 u    +    d2 v

This last equation is that of an ellipse, as long as d1 and d2 are positive. To verify that they are, note that d1 and d2 are the eigenvalues of the matrix SYM, and SYM is the product

   -1  T   -1
( C   )   C,

which is of the form

 T
M   M,

so SYM is positive semi-definite (its eigenvalues are non-negative). Furthermore, since the product

   -1  T   -1
( C   )   C

is non-singular if C is non-singular, and since the columns of C are V1 and V2, SYM exists and is non-singular precisely when V1 and V2 are linearly independent, a condition that we have assumed. So the eigenvalues of SYM can't be zero. They're not negative either. We conclude they're positive.

Solving intersection problems

There is one problem solving technique used in SPICE ellipse and ellipsoid routines that is so useful that it deserves special mention: using a 'distortion map' to solve intersection problems.

The distortion map (as it is referred to in CSPICE routines) is simply a linear transformation that maps an ellipsoid to the unit sphere. The distortion map defined by an ellipsoid whose semi-axes are A, B, and C is represented by the matrix

+-                -+
|  1/A   0    0    |
|   0   1/B   0    |.
|   0    0    1/C  |
+-                -+

The distortion map is (as is clear from examining the matrix) one-to-one and onto, and in particular is invertible, so it preserves set operations such as intersection. That is, if M is a distortion map and X, Y are two sets, then

M( X intersect Y ) = M(X) intersect M(Y).

The same is true of the inverse of the distortion map. The utility of these facts is that frequently it's easier to find the intersection of the images under the distortion map of two sets than it is to find the intersection of the original two sets. Having found the intersection of the 'distorted' sets, we apply the inverse distortion map to arrive at the intersection of the original sets. Some examples:

  • To find the intersection of a ray and an ellipsoid, apply the distortion map to both. Because the distortion map is linear, the ray maps to another ray, and the ellipsoid maps to the unit sphere. The resulting intersection problem is easy to solve. Having found the points of intersection of the new ray and the unit sphere, if any, we apply the inverse distortion map to these points, and we're done.

  • To find the intersection of a plane and an ellipsoid, apply the distortion map to both. The linearity of the distortion map ensures that the original plane maps to a second plane (whose formula is easily calculated). The ellipsoid maps to the unit sphere. The intersection of a plane and a unit sphere is easily found. The inverse distortion map is then applied to the circle of intersection (when the intersection is non-trivial), and the ellipse of intersection of the original plane and ellipsoid results. This procedure is used in the CSPICE routine inedpl_c.

  • To find the image under gnomonic projection onto a plane (camera projection) of an ellipsoid, given a focal point, we must find the intersection of the plane and the cone generated by ellipsoid and the focal point. Applying the distortion map to the ellipsoid, plane, and focal point, the problem is transformed into that of finding the intersection of the transformed plane with the cone generated by a unit sphere and the transformed focal point. This 'transformed' problem is much easier to solve. The resulting intersection ellipse is then mapped back to the original intersection ellipse by the inverse distortion mapping.