You are on page 1of 10

J Geod (2005) 79: 93102

DOI 10.1007/s00190-005-0446-1

O R I G I N A L A RT I C L E

J. L. Awange E. W. Grafarend Y. Fukuda


S. Takemoto

The application of commutative algebra to geodesy:


two examples

Received: 16 June 2003 / Accepted: 31 January 2005 / Published online: 14 May 2005
Springer-Verlag 2005

Abstract Algebra, in particular commutative algebra, is ature on commutative algebra in textbooks such as Cox et al.
applied here to provide a general unified solution to nonlinear (1998), Becker and Weispfenning (1998), McCoy and Janusz
systems of equations encountered in geodesy. Starting with (2001), Lauritzen (2003) and Irving (2004). A well-known
the Abelian group, the polynomial ring is defined and application of algebra to geodesy could perhaps be the use
used to form generators of ideals. By applying Buchber- of Legendre polynomials in spherical harmonic expansion
ger or polynomial resultant algorithms, these generators are studies. A more recent application of algebra to geodesy is
reduced to simple structures often comprising a univariate evidenced in Lannes and Durand (2003) who propose a new
polynomial in one of the unknowns. The advantage of the approach to differential GPS based on algebraic graph theory.
proposed unified approach is that it provides exact solutions The present contribution proposes the application of com-
to geodetic nonlinear systems of equations without the tra- mutative algebra to provide a unified solution to nonlinear
ditional requirements of linearization, iterations or approx- systems of equations encountered in geodetic practice.
imate starting values. The commutative algebraic approach In geodesy, nonlinear systems of equations are often
therefore alleviates the need for isolated exact solutions to encountered in several applications since they relate the obser-
various geodetic nonlinear systems of equations. The proce- vations (measurements) to the unknown parameters to be
dure is applied to GPS meteorology to compute refraction determined. When the number of observations and the num-
angles, and Helmerts one-to-one mapping of topographical ber of unknowns are equal, the unknown parameters may
points onto the reference ellipsoid. be obtained by solving explicitly (in a closed form) nonlin-
ear systems of equations relating observations to unknown
Keywords Commutative algebra Polynomial ring parameters.
Groebner basis Polynomial resultant Ideal GPS Because of the difficulty in obtaining reliable closed form
meteorology Helmerts projection procedures, approximate numerical methods were adopted
in practice. Such procedures depend on some approximate
1 Introduction starting values, linearization and iterations. In some cases,
the solutions of numerical methods are unstable and the iter-
Algebra was widely applied in fields such as robotics (for ations may fail to converge. For example, Cox et al. (1998,
kinematic modelling of robots), engineering (for offset sur- pp. 2832) illustrate that systems of equations with exact
face construction in solid modelling), computer science (for solutions become vulnerable to small errors introduced dur-
automated theorem proving), and computer-aided design ing root-finding. When extending the partial solution to the
(CAD). An interested reader can find more elaborate liter- complete solutions of the system, errors may accumulate and
thus become too large. If the partial solution was derived by
J. L. Awange (B) iterative procedures, then the errors incurred during the root-
Maseno University, School of Environment and Earth Sciences, finding may blow up during the extension of the partial
P.O.Box 333, Maseno, Kenya. solution to the complete solution (back substitution).
E-mail: jawange@yahoo.co.uk
In some applications, symbolic rather than numerical
Y. Fukuda S. Takemoto solutions are desired. In such cases, explicit procedures are
Department of Geophysics, Kyoto University,
Kitashirakawa Oiwake-cho, Sakyo-ku, Kyoto City, usually employed. The resulting symbolic expressions often
Kyoto 606-8502, Japan consist of univariate polynomials relating unknown parame-
E. W. Grafarend
ters (unknown variables) to known variables (observations).
Department of Geodesy and GeoInformatics By inserting known values into these univariate polynomials,
Geschwister-Scholl str.24D, 70174 Stuttgart, Germany numerical solutions are readily computed for the unknown
94 J. L. Awange et al.

Fig. 1 Tetrahedron for closed form 3D-resection

variables. Advantages of explicit solutions have been listed 2 A unified commutative algebraic solution
by Merrit (1949). In geodesy for example, the minimum dis-
tance mapping (MDM) problem (Grafarend and Lohse 1991) In geodesy, we normally work with a set of finite polynomi-
relates a point on the Earths topographical surface uniquely als P over the field of real numbers R in a finite number n of
(one-to-one) to a point on the international reference ellip- different variables expressed as
soid. The solution of such an optimization problem requires
that the equations be solved explicitly. P = R[x1 , . . . , xn ]. (1)
The drawback experienced with explicit solutions was Consider a 3D resection problem where horizontal and
partly because the methods required extensive computations vertical directions were measured from an unknown station
for the results to be obtained, and partly because the resulting P1 to three known stations {P1 , P2 , P3 } in Fig. 1.
symbolic expressions were too large and required computers The measured directions are then converted to the dimen-
with large storage capacity. Until recently, computers could sionless space angles and related to the unknown distances
hardly handle large computations due to lack of fast central {x1 , x2 , x3 } through nonlinear systems of polynomial equa-
processing units (CPUs), shortage of random access mem- tions (e.g., Grafarend et al. 1989, Eq. 1.30, p. 128)
ory (RAM) and limited hard disk storage capacity. The other 2
setback experienced by explicit procedures was that some x1 + 2a12 x1 x2 + x22 + aoo = 0

of the methods, especially those from algebraic fields, were 2
formulated based on theoretical concepts that were hard to x2 + 2b23 x2 x3 + x32 + boo = 0 (2)

realize or comprehend without the help of computers. For a
long time, these setbacks hampered progress of the explicit x32 + 2c31 x3 x1 + x12 + coo = 0,
procedures. Equation (2) has some common features from the point of
The advances made in computer technology in recent view of modern commutative algebra. All three equations
years, however, have led to improvements in explicit compu- in Eq. (2) are polynomials in three variables or unknowns
tational procedures, which hitherto were difficult to achieve. {x1 , x2 , x3 }, which here stand for unknown distances. The
Apart from the improvements in the existing computational coefficients of these variables are constants from the field of
procedures, new computational techniques are continuously reals R.
being added with the aim of optimizing computational speed Using the basic properties of the usual addition (+) and
and efficiency. The commutative algebraic method discussed multiplication (*), we can formulate the following axioms:
herein lies in this category. With the proposed unified com-
mutative algebraic approach to geodetic nonlinear systems of a, b, c R
equations, this contribution hopes to alleviate the laborious In modern algebra, any set with the axioms {G1+, G2+,
task of having to develop exact solution for each individual G3+} or {G1, G2, G3} is said to be a group. In addi-
nonlinear system of geodetic equations that is encountered tion, if the group operation is commutative, then the group is
in practice. called an Abelian group. Examples of groups include:
The proposed procedure is applied in this study to solve
the problems of bending (refraction) angles in GPS meteo- (a) The group of integers Z under addition.
rology and Helmerts projection of topographical points to (b) The group of non-zero rational numbers Q under multi-
the international reference ellipsoid. We organize the pre- plication.
sentation as follows: in Sect. 2, we present the unified com- (c) The set of rotations about the origin in the Euclidean
mutative algebraic solution together with Theorem 2.1 that plane under the operation of composite functions.
allows us to convert geodetic nonlinear systems of equations The elements of the first two examples are numbers, and
into algebraic (polynomials). Section 3 applies the proposed being group with respect to the selected operations is self-
procedure to solve nonlinear problems encountered in GPS evident. In the third example, the element of group is not a
meteorology and topographic mapping. number, but a rotational operation, and the group operation
The application of commutative algebra to geodesy: two examples 95

Axiom Addition Multiplication


(a, b) =: a + b (a, b) =: a b
1 Associativity G1+ : (a + b) + c G1 : (a b) c
= a + (b + c) = a (b c)
(additive associativity) (multiplicative associativity)
2 Identity G2+ : a + 0 = a G2 : a 1 = a
(additive identity, (multiplicative identity,
neutral element) neutral element)
3 Inverse G3+ : a + (a) = 0 G3 : a a1 = 1
(additive inverse) (multiplicative inverse)
4 Commutativity G4+ : a + b = b + a G4 : a b = b a
(additive commutativity, (multiplicative commutativity,
Abelian axiom) Abelian axiom)

Axiom Comments
1 dist. D1 +; a (b + c) = a b + a c Left and Right
(a + b) c = a c + b c additive distributivity w.r.t
internal multiplication

is the composition of two rotations. This example shows that < f1 , . . . ., fs > is an ideal, and if a subset I R [x1 , . . ., xn ]
the elements of a group may be of any nature, provided that is an ideal, it must satisfy the following conditions Cox et al.
the group operation is properly defined. (1997, p. 29)
Furthermore, we know that in the case of the first two
0 I,
examples, apart from the group operations, there are other
If f, g I, then f +g I (i.e. {I } is an additive subgroup
operations too. These operations are connected with each
of the additive group of the field R)
other in the so-called distributive property, which can be pos-
If f I and c R [x1 , . . ., xn ], then cf I (i.e. I is
tulated as follows:
closed under multiplication ring element).
The algebraic structure that holds these properties is called
a commutative ring. The set of real numbers is a simple In geodesy, we deal with finitely generated ideals of P
example of a ring with respect to usual addition and mul- such that
tiplication 
operations. If two polynomials
 be given such that
f1 (x) = ai x i and f2 (x) = bj x j , then two binary I = g1 P + g2 P + + gn P
i j =: {g1 , . . . , gn } P , (4)
addition and multiplication operations can be defined 
on these polynomials such that: i.e., every polynomial of the form ci pi , where ci P is
i
 element of ideal I . We now search for solutions {x1 , . . . , xn }
(a) Addition: f1 (x) + f2 (x) = ck x k ,
k where every element of I gets 0. This is equivalent to express-
ck = ak + bk , ck R ing the generators g1 , . . . , gn of the ideal I equal to 0. Hence
 the resection problem in Sect. 2.2 becomes
(b) Multiplication: f1 (x).f2 (x) = ck x k , ck
k {x12 + 2a12 x1 x2 + x22 + aoo , x22 + 2b23 x2 x3 + x32 + boo , x32
= ai bj , ck R. +2c31 x3 x1 + x12 + coo } I. (5)
i+j =k
An easier structure of an ideal is now sought, namely
With these operations, polynomials form a commutative ring, an easier set of generators. In order to obtain this set, one
which is often called a polynomial ring. Designated P , a employs the Buchberger (2001) algorithm that computes the
polynomial ring is represented by n unknown variables xi Groebner basis. This algorithm strongly depends on the cho-
over R and is expressed as P := R [x1 , . . ., xn ] . Polynomial sen polynomial order and returns a minimal basis with respect
rings behave much the same as a ring of numbers and can to this chosen polynomial order. In particular, the lexico-
be manipulated as seen above. Their derivatives follow Leib- graphic ordering is preferable since it returns a univariate
nizs rule and are defined via linearity. For polynomial rings, polynomial as one of its bases. Let us consider an example
ideals {I } are defined. An ideal is constructed by a family of Buchberger (2001)
of generators consisting of the set of linear combinations of Consider the polynomials in a set F (x, y) = {f1 , f2 } to
these generators with polynomial coefficients. Let f1 , . . . , fs have as their elements
and c1 , . . . , cs be polynomials in R [x1 , . . . , xn ], then
f1 = xy 2y
< f1 , . . . , fs > (6)
 s  f2 = 2y 2 x 2 ,

w= ci fi , c1 , . . . , cs R [x1 , . . . ., xn ] . (3) where {f1 , f2 } I are the generators of the ideal I . We
i=1
now seek a simplified set of generators of this ideal using
96 J. L. Awange et al.

the Buchberger algorithm. This simplified set of generators to make the observational equations algebraic or polyno-
is called the Groebner basis, and is computed by reducing the mial, and implement a least squares solution, the objective
set formed by Eq. (6) into another set G of F as function which is of order l = 4 reduces by one to order
l = 3 upon differentiating once. The normal equations are of
G := {2x 2 + x 3 , 2y + xy, x 2 + 2y 2 }. (7) order l = 3 as expected. 

In Mathematica software, using the lexicographic order x
comes before y, the Groebner basis can simply be computed The significance of the Theorem 2.1 is that all nonlinear
by entering the command (see e.g., Awange and Grafarend observation equations of interest are successfully converted
2003b) to algebraic or polynomial equations. Finally, the com-
mutative algebraic solution of nonlinear systems of equations
GroebnerBasis[F, {x, y}]. (8) in geodesy is performed in the following steps:
The set G in Eq. (7) contains one set of univariate polyno- Step 1: Using Theorem 2.1, convert any geodetic nonlinear
mials 2x 2 + x 3 that can easily be solved using the roots system of equations into algebraic (polynomials).
command in Matlab (e.g., Hanselman and Littlefield 1997, Step 2: Prepare the set of generators of ideal {I }, i.e., equate
p. 146) for solutions {x = 0, x = 0, x = 2}, and substituted the polynomials to 0.
in any of the remaining elements of the set to solve for y. The Step 3: With the lexicographic ordering specified, employ
solutions of G, i.e., the roots {x = 0, x = 0, x = 2}), sat- the Buchberger (2001) algorithm to compute the
isfy those of F . As an alternative to computing the Groebner Groebner basis of the generators in step 2. Alter-
basis, the polynomial resultant approach (e.g., Awange and natively, directly solve the systems of equations
Grafarend 2003c) could be used to directly solve Eq. (6). formed by the generators using polynomial resul-
In order to be able to solve nonlinear systems of equations tants (e.g., Awange and Grafarend 2003c).
in geodesy, they have to be expressed as elements of poly- Step 4: From the results of step 3, there exist a univariate
nomial rings defined above using Theorem 2.1 (i.e., Awange polynomial whose roots can be obtained using, e.g.,
and Grafarend 2003a). Matlabs roots command.
Step 5: Substitute the roots of step 4 in the other elements of
Theorem 2.1. Given algebraic (polynomial) observational Groebner basis to obtain solutions to the remaining
equations (n observations, where n is the dimension of the variables. The substitution steps could be avoided
observation space Y) of order l in m variables (unknowns) by computing a reduced Groebner basis (Cox et al.
(m is the dimension of the parameter space X), the applica- 1998;Awange 2002) or solving directly for the poly-
tion of a least squares solution to the algebraic observation nomial resultants of each variable (see Awange and
equations gives (2l 1) as the order of the set of nonlinear Grafarend 2003c).
algebraic normal equations. There exist m normal equations
of the polynomial order (2l 1) to be solved.

Proof. Given nonlinear algebraic equations fi k 3 Two applications


{1 , . . . , m } expressed as
The commutative algebraic algorithm is applied to GPS-
f1 k{1 , . . . , m } meteorology to compute refraction (bending) angles and to
f2 k{1 , . . . , m } Helmerts projection of topographical points onto the refer-

. ence ellipsoid.
(9)
.
.
fn k{1 , . . . , m }. 3.1 GPS-meteorology: solution of bending angles
and the order considered as l, we write the objective function In space-borne GPS meteorology, high-flying GPS satellites
to be minimized as transmit their signals to be received by low-Earth orbiting
f  = f1 + + fn | fi k{1 , . . . , m }
2 2 2
(10) (LEO) satellites. As the GPS signals traverse the atmosphere,
they remotely sense profiles of, e.g., temperature, pressure
and obtain the partial derivatives (first derivatives) of Eq. (10) and water vapour. Ground-based GPS meteorology comprises
with respect to the unknown variables {1 , . . . , m }. The order GPS satellites and ground-based GPS receivers. It is mainly
of Eq. (10), which is l 2 , then reduces to (2l 1) upon differ- involved with the determination of water vapour in the atmo-
entiating the objective function with respect to the variables sphere using the zenith delay.
1 , . . . , m , thus resulting in m normal equations of the poly- Essential for remote sensing using space-borne GPS-
nomial order (2l 1). meteorology is the ability to accurately measure the bend-
Using the example of GPS code pseudoranging or distance ing angle of the signal as it passes through the atmosphere.
equations, the order of the polynomials in the algebraic obser- Radio occultation with GPS takes place when a LEO satellite
vational equations is l = 2. If we take the pseudoranges views a transmitting GPS satellite, setting or rising behind the
squared (distances squared), a necessary procedure in order Earths limb. These signals are bent and retarded causing a
The application of commutative algebra to geodesy: two examples 97

(a2 y + a4 x a)2 = (a1 cos L a3 cos G )2 . (14)


Equation (14) results in an expression having only the trigo-
nometric product 2a1 a3 cos L cos G on the right-hand side
once the squared trigonometric values cos2 L and cos2 G
were replaced by variables {y, x}, and applying the trigo-
nometric Pythagorean theorem of a unit circle. The resulting
expression is arranged so as to have only 2a1 a3 cos L cos G
on the right-hand side. Squaring both sides of the re-arranged
Fig. 2 Geometry of GPS occultation
expression and replacing the resulting squared trigonomet-
ric values cos2 L and cos2 G completes the conversion of
Eq. (11) into algebraic form
delay in the arrival at the LEO satellite. Figure 2 shows the
occultation geometry. As the signal is bent, the total bending
angle {}, an impact parameter {a}, and a tangent radius {r} d1 x 4 + d2 x 3 + d3 x 3 y + d4 x 2 + d5 x 2 y 2 + d6 x 2 y + d7 x
define the ray passing through the atmosphere. +d8 xy 3 + d9 xy 2 + d10 xy + d00 = 0 (15)
The refraction angle is related to atmospheric a5 x + a6 y = 0,
parameters of temperature, pressure and water vapour via the
refractive index. The LEO receiver measures, at the required where d00 = d11 y 4 + d12 y 3 + d13 y 2 + d14 y + d14 . The
sampling rate, the dual-frequency carrier-phases, the C/A and coefficients d1 , ..., d15 are listed in Appendix B. The alge-
P-code group delays, and the signal strength. The data is then braic geometry of Eq. (15) is an intersection of a fourth-order
processed to remove errors arising from short-term oscillator polynomial (i.e., quartic) and a line (Fig. 3), indicating that
instabilities in the satellites and the receivers using at least there exist four solutions to the nonlinear system of equations
one ground station and at least one satellite that is not being (Eq. 11).
occulted. Once the observations are corrected for possible Next, the commutative algebraic algorithm is used to
sources of errors, the resulting Doppler shift is used to deter- compute the Groebner basis of Eq. (15). Note that the polyno-
mine the refraction angle (Fig. 2). For more details, we mials of Eq. (15) are generators of ideal I in Eq. (4). Applying
refer to Steiner (1998) and Wickert (2002). the reduced Groebner basis (see, e.g., Awange 2002) leads to
The system of nonlinear trigonometric equations for refrac- univariate polynomials
tion angles comprise two equations given as
h4 x 4 + h3 x 3 + h2 x 2 + h1 x + h0 = 0
vL cos(L L ) vG cos(G + G ) (16)
dLi g4 y 4 + g3 y 3 + g2 y 2 + g1 y + g0 = 0
= +vL cos(L L )vG cos(G + G )rG sin G
dt with the coefficients as
= rL sin L , (11)
where vL , vG are the projected LEO and GPS satellite veloc- h4 = (a64 d1 + a54 d11 a5 a63 d3 + a52 a62 d5 a53 a6 d8 )
ities in the occultation plane, rL , rG the radius of tangent h3 = (a53 a6 d12 + a64 d2 a5 a63 d6 + a52 a62 d9 )
points at LEO and GPS, respectively, and dLi /dt is the Dopp- h2 = (a5 a63 d10 + a52 a62 d13 + a64 d4 )
ler shift (see Fig. 2).
Equation (11) is normally solved using Newtons method h1 = (a5 a63 d14 + a64 d7 )
(e.g., Steiner 1998; Wickert 2002). In what follows, we ap- h0 = a64 d1 5
ply the solution of commutative algebra to solve Eq. (11) in
closed form. First, we apply Theorem 2.1 to convert Eq. (11) and
into algebraic (polynomial) form. Let us denote g4 = (a64 d1 + a54 d11 a5 a63 d3 + a52 a62 d5 a53 a6 d8 )
 g3 = (a54 d12 a5 a63 d2 + a52 a62 d6 a53 a6 d9 )
x = sin G , y = sin L , a1 = vL cos L , a2 = vL sin L
a3 = vG cos G , a4 = vG sin G , a5 = rG , a6 = rL , g2 = (a53 a6 d10 + a54 d13 + a52 a62 d4 )
(12) g1 = (a54 d14 a53 a6 d7 )
where the signs of the velocities change depending on the g0 = a54 d15
directions of motion of the satellites. Applying the trigono- Four solutions are obtained from Eq. (16) for both x and y
metric addition formulae to Eq. (12) leads to using, e.g., the roots command in Matlab software as x =

a1 cos L + a2 y + a3 cos G + a4 x = a roots[ h4 h3 h2 h1 h0 ] and y = roots[ g4 g3 g2 g1 g0 ]. An
(13) alternative solution of Eq. (15) would be by the use of the
a5 x + a6 y = 0.
Sylvester resultant approach, which leads to identical results
In order to eliminate the trigonometric terms cos L and as those of reduced Groebner basis. We refer to Awange et al.
cos G appearing in Eq. (13), they are taken to the right-hand (2004) for the use of the Sylvester resultant approach to solve
side, and the resulting expression squared is Eq. (15).
98 J. L. Awange et al.

5 Graphic solution of occultation bending angle


x 10
4

quartic polynomial f(X,Y)=0


3.5 line aX+bY=0

2.5

2
f(X,Y)

1.5

0.5

0.5
0 10 20 30 40 50 60
X,Y

Fig. 3 Algebraic curve for the solution of system of nonlinear GPS atmospheric sounding refraction (bending) angle equations

Using Eq. (16), bending angles for CHAMP occultation (so-called projection lines) that depart from P and intersect
number 133 of 3rd May 2002 that occurred from UTC time Ea,a,b
2
either not at all or at two points. There is one projec-
13:48:36 to 13:49:51.98 were computed. The results were tion line that is at minimum distance relating P to p. Fig-
compared to those of classical Newtons approach (Steiner ure 5 is an illustration of such a minimum distance mapping
1998). The computation was carried out for heights between (MDM).
5 km and 40 km. Above 40 km, the bending angles are highly Such an optimization problem is formulated by means of
influenced by the presence residual ionospheric errors and the Lagrangean (x1 , x2 , x3 , x4 ) as
are normally augmented by model values. Below 5 km, the 1 1
presence of water vapour significantly influences the results. (x1 , x2 , x3 , x4 ) := X x2 + x4
2
2
Bending angles between the range of 540 km from satellite
data are more accurate and normally directly assimilated into b2 (x12 + x22 ) + ax32 a 2 b2
1
numerical weather prediction models (Healey et al. 2003). = (X x1 )2 + (Y x2 )2 + (Z x3 )2
Figure 4 shows the deviation of the results between the 2

algebraic and Newtons iterative approaches. The nonlinear- +x4 b2 (x12 + x22 ) + ax32 a 2 b2 .
ity error, i.e., the error incurred in using iterative Newtons In the first case, we represent the Euclidean distance be-
approach in computing bending angles increases with decreas- tween the points P and p in terms of Cartesian coordinates
ing atmospheric height. For this occultation, from 40 km to of P (X, Y, Z) and p(x1 , x2 , x3 ). The Cartesian coordinates
15 km, the deviation is within 2 104 (degrees) but in- (x1 , x2 , x3 ) of the projection point P are unknown. The con-
creases to 7 104 (degrees) for the region below 15 km straint that the point p is an element of the ellipsoid-of-
with the maximum absolute deviation of 0.00069 . A detailed revolution
analysis of nonlinear GPS meteorology bending angles at
different diurnal solar times was presented in Awange et al. Ea,a,b
2
:= {x R3 |b2 (x12 + x22 )
(2004). + a 2 x23 a 2 b2 = 0, R+ a > b R+ },
is substituted into the Lagrangean by means of the Lagrange
multiplier x4 , which is also unknown. (x1 , x2 , x3 , x4 ) =
3.2 Helmerts projection of topographical points onto the
arg{(x1 , x2 , x3 , x4 ) = min} is the argument of the mini-
reference ellipsoid
mum of the constrained Lagrangean (x1 , x2 , x3 , x4 ). The re-
sult of the minimization procedure is presented by Lemma 3.1.
In order to relate a point P on the Earths topographical
surface to a point p on the international reference ellip- Lemma 3.1 (Constrained MDM). The functional (x1 , x2 ,
soid Ea,a,b
2
, one works with a bundle of half-straight lines x3 , x4 ) is minimal, if the conditions in Eqs. (17) and (19) hold.
The application of commutative algebra to geodesy: two examples 99

Deviation of bending anlges due to nonlinearity assumption: Same impact parameter p


45

40

35

30
Height(km)

25

20

15

10

5
6 4 2 0 2 4 6 8
Bending angle(degree) x 10
4

Fig. 4 Deviation of computed bending angles due to nonlinearity


((x , x , x , x )) = 0 i = 1, 2, 3, 4. (17)
xi 1 2 3 4
On taking partial derivatives with respect to xi , we have
2
(i) (x = (X x1 ) + b x1 x4 = 0
1)

(ii) = (Y x ) + b2 x x = 0
2 2 4
(x2 )
(18)
2
(iii) (x = (Z x ) + a x x = 0
3)
3 3 4

1 2 2
(iv) (x =
) 2
[b (x1 + x22 )] + a 2 x32 a 2 b2 = 0
4

2
(x , x , x , x ) > 0 i, j {1, 2, 3}. (19)
xi xj 1 2 3 4
 
2 Fig. 5 Minimum distance mapping (MDM) of a point P on the
H3 := (x ) Earths topographic surface to a point p on the international reference
xi xj ellipsoid E2a,a,b

1 + b2 x4 0
= 0 1 + b2 x4 0 R33 (20)
0 0 1 + a 2 x4 Equation (17) provides the necessary condition to consti-
tute an extremum: the normal equations are bilinear. Products
eigenvalues
of the unknowns, for instance {x1 x4 , x2 x4 , x3 x4 }, and squares
|H3 I3 | = 0 (21) of the unknowns, for instance {x12 , x22 , x32 } appear. Finally, the
matrix of second derivatives H3 in Eq. (19) is positive defi-

1 = 2 := 1 + b2 x4 = xX = Y
x2
nite, which constitutes the sufficient condition to obtain a

1
(22) minimum. Fortunately, the matrix of second derivatives H3
is diagonal. Using Eqs. (18i)(18iii) together with Eq. (21)
3 := 1 + a 2 x4 = xZ
3 leads to Eq. (22), which are the eigenvalues of the Hesse
100 J. L. Awange et al.

matrix H3 . These values are 1 = 2 = X\x1 , 3 = Z\x3 , the values (x1 , x2 , x3 ) = (x, y, z) are finally produced. Once
and must be positive. the Cartesian coordinates on the ellipsoid {x1 , x2 , x3 } are
Now, we apply the commutative algebra algorithm to computed using Eq. (25), they are converted to Gauss ellip-
compute the nonlinear system of normal equations (Eqs. soidal coordinates {L, B, H } by means of the following closed
(18i)(18iii)). For this case, we are fortunate that the non- form solution ({X, Y, Z} T2 {x1 , x2 , x3 } Ea,a,b 2
to
linear equations are already in polynomial form, and thus {L, B, H })
do not require application of Theorem 2.1. The unknown in 
these case are the Cartesian coordinates of the point on the H := (X x1 )2 + (Y x2 )2 + (Z x3 )2 (28)
ellipsoid {x1 , x2 , x3 } and the Lagrangian multiplier x4 . Let us
Y x2 Y y
write the ideal of polynomials formed by Eqs. (18i)(18iii) tan L = = (29)
in lexicographic order x1 > x2 > x3 > x4 (read: x1 before X x1 Xx
x2 before x3 before x4 ) as Z x3
tan B = 
x1 + b2 x1 x4 X, (X x1 )2 + (Y x2 )2

x2 + b x2 x4 Y,
2
Z x3
I deal I :=< >. (23)
2 2 x3 +2 a2 x3 x42 2 Z, 2 2
2
=  (30)
b x1 + b x2 a x3 a b (X x)2 + (Y y)2
Here, we replicate the example of Grafarend and Lohse
Expressing the generators of the ideal in Eq. (23) as
(1991, p. 108). Given the geometric parameters of the ellip-
f1 := x1 + b2 x1 x4 X, soid of revolution: semi-major axis a = 6378137.000 m and
first numerical eccentricity-squared e2 = 0.00669437999013,
f2 := x2 + b2 x2 x4 Y, from which the semi-minor axis b can be computed. The input
(24) data are Cartesian coordinates of eight points on the surface
f3 := x3 + a 2 x3 x4 Z, of the Earth presented in Table 1.
Using the input data from Table 1, the coefficients of
f4 := b2 x12 + b2 x22 a 2 x32 a 2 b2 , the univariate polynomial (Eq. 26) are computed as given in
Table 2.
i.e., f1 = f2 = f3 = f4 = 0, the Groebner basis of the From these coefficients, polynomial roots
are computed
generators characteristic for the MDM problem is computed using Matlabs roots command as x4 = roots c4 c3 c2 c1 c0 .
using the Buchberger algorithm as The obtained roots are then substituted in Eq. (27) to give
GroebnerBasis[{f1 , f2 , f3 , f4 }, {x1 , x2 , x3 , x4 }]. (25) the values of {x3 , x2 , x1 }. The computed results presented in
Table 3 are identical to those obtained by Grafarend and Loh-
The Groebner basis computation (Eq. 25) leads to 14 ele- se (1991, Table 4, p.108). Once the ellipsoidal Cartesian coor-
ments presented in Appendix A interpreted as follows: dinates are derived, the Jacobi ellipsoidal coordinates (ellip-
The first expression is a univariate polynomial of order soidal longitude L, ellipsoidal latitude B and height H ) can
four (quartic) in the Lagrange multiplier, i.e., be computed using Eqs. (28), (29) and (30).
4
c4 x4 + c3 x43 + c2 x42 + c1 x4 + co = 0

4 Concluding remark
c4 = a 6 b6

The present study has presented a unified commutative alge-

c3 = (2a 6 b4 + 2a 4 b6 ) braic approach to nonlinear systems of equations in geodesy,

followed by two examples. By converting geodetic nonlin-


c = (a 6 b2 +4a 4 b4 + a 2 b6 a 4 b2 X 2 a 4 b2 Y 2 a 2 b4 Z 2 )
2 ear systems of equations into algebraic (polynomials) via

Theorem 2.1, generators of ideal are formed. These genera-


c = (2a 4 b2 + 2a 2 b4 2a 2 b2 X 2 2a 2 b2 Y 2 2a 2 b2 Z 2 )
1 tors, which are multivariate systems of polynomial equations

co = (a 2 b2 b2 X 2 b2 Y 2 a 2 Z 2 ).
(26) Table 1 Cartesian coordinates of eight topographic test points

Point X(m) Y (m) Z(m)


With the admissible values x4 substituted in linear Eqs. (4),
1 3980192.960 0 4967325.285
(8) and (12) of the computed Groebner basis, i.e., 2 0 0 6356852.314
3 0 0 6357252.314
(1 + a 2 x4 )x3 Z 4 4423689.486 529842.355 4555616.169
5 4157619.145 664852.698 4775310.888

(1 + b2 x4 )x2 Y (27) 6 2125699.324 6012793.226 91773.648


7 5069470.828 3878707.846 55331.828
8 213750.930 5641092.098 2977743.624
(1 + b2 x4 )x1 X,
The application of commutative algebra to geodesy: two examples 101

Table 2 Computed polynomial coefficients

Point c0 c1 c2 c3 c4
1 7.7479e+22 1.3339e+41 1.3515e+55 4.3824e+68 4.4420e+81
2 5.1720e+22 1.3374e+41 1.3529e+55 4.3824e+68 4.4420e+81
3 2.5861e+23 1.3372e+41 1.3529e+55 4.3824e+68 4.4420e+81
4 2.5311e+24 1.3310e+41 1.3507e+55 4.3824e+68 4.4420e+81
5 1.8076e+23 1.3334e+41 1.3513e+55 4.3824e+68 4.4420e+81
6 5.1549e+21 1.3285e+41 1.3493e+55 4.3824e+68 4.4420e+81
7 2.6815e+24 1.3263e+41 1.3488e+55 4.3824e+68 4.4420e+81
8 4.5942e+24 1.3267e+41 1.3493e+55 4.3824e+68 4.4420e+81

Table 3 Computed ellipsoidal cartesian coordinates and the Lagrange factor

Point x1 (m) x2 (m) x3 (m) x4 (m2 )


1 3980099.549 0.000 4967207.921 5.808116e019
2 0.000 0.000 6356752.314 3.867016e019
3 0.000 0.000 6356752.314 1.933512e018
4 4420299.446 529436.317 4552101.519 1.897940e017
5 4157391.441 664816.285 4775047.592 1.355437e018
6 2125695.991 6012783.798 91773.503 3.880221e020
7 5065341.132 3875548.170 55286.450 2.017617e017
8 213453.298 5633237.315 2973569.442 3.450687e017


relating unknown variables to the known constants can be (2b2 Z + b4 x4 Z a 2 Z)x3 + a 4 b6 x43
reduced into simpler structures. When lexicographic ordering

is chosen, these simpler structures often comprise a univariate +(2a 4 b4 + a 2 b6 )x42 + (a 4 b2 + 2a 2 b4

polynomial that is solvable using, e.g., Matlab, Mathematica 3.
a 2 b2 X 2 a 2 b2 Y 2 b4 Z 2 )x4
and Maple computer softwares.

Acknowledgements The first author wishes to acknowledge the sup- +a 2 b2 b2 X 2 b2 Y 2 2b2 Z 2
port of the Japan Society of Promotion of Science (JSPS) for the finan-
cial support that allowed this study at Kyoto University, Japan. The
author is further grateful for the support and the good working atmo- 4. (1 + a 2 x4 )x3 Z
sphere provided by his hosts and co-authors Professors S. Takemoto
and Y. Fukuda of the Department of Geophysics, Graduate School of (a 4 2a 2 b2 + b4 )x32 + (2a 2 b2 Z 2b4 Z)x3 a 4 b6 x42
Science, Kyoto University, Japan. The data used for the GPS mete- 5.
orology example were provided by GeoForschungsZentrum Postdam
(GFZ), Germany. The authors express their utmost appreciation. Spe-
2a 4 b4 x4 a 4 b2 + a 2 b2 X 2 + a 2 b2 Y 2 + b4 Z 2 )
cial thanks to Dr. A. Steiner for the materials and discussions on the
GPS meteorology example. (2b2 a 2 + b4 x4 )x32 a 2 Zx3


+a 4 b6 x43 + (2a 4 b4 + 2a 2 b6 )x42

Appendix A 6.
+(a 4 b2 + 4a 2 b4 a 2 b2 X 2 a 2 b2 Y 2 b4 Z 2 )x
4
Computed Groebner basis for the MDM problem
+2a 2 b2 2b2 X 2 2b2 Y 2 2b2 Z 2
6 6 4
a b x4 + (2a 6 b4 + 2a 4 b6 )x43 + (a 6 b2 + 4a 4 b4 + a 2 b6 2
(X + Y 2 )x2 + a 2 b4 Y x42 + Y (a 2 b2 + b2 x32 b2 Zx3 )x4

a b X a b Y a b Z )x4
4 2 2 4 2 2 2 4 2 2 7.
+Y x32 Y 3 Y Zx3 Y X 2
1.
+(2a 4 b2 + 2a 2 b4 2a 2 b2 X 2 2a 2 b2 Y 2 2a 2 b2 Z 2 )x
4
8. (1 + b2 x4 )x2 Y
+(a b b X b Y a Z ).
2 2 2 2 2 2 2 2
9. (a 2 x3 b2 x3 + b2 Z)x2 a 2 x3 Y
4
(a Z 2a b Z + b Z)x3 a b x4
2 2 4 6 6 3

10. Y x1 Xx2

(2a b + a b )x4 (a b + 2a b a b X
6 4 4 6 2 6 2 4 4 4 2 2
11. Xx1 + a 2 b4 x42 + (a 2 b2 + b2 x32 b2 Zx3 )x4
2.
a 4 b2 Y 2 a 2 b4 Z 2 )x a 4 b2 + a 2 b2 X 2 +x32 Zx3 + Y x2 X2 Y 2
4

+a 2 b2 Y 2 + 2a 2 b2 Z 2 b4 Z 2 12. (1 + b2 x4 )x1 X
102 J. L. Awange et al.

13. (a 2 x3 b2 x3 + b2 Z)x1 a 2 Xx3 Awange JL, Grafarend E (2003b) Groebner basis solution of the three-
dimensional resection problem (P4P). J Geod 77:327337
Awange JL, Grafarend E (2003c) Multipolynomial resultant solution of
14. x12 + a 2 b4 x42 + (2a 2 b2 + b2 x32 b2 Zx3 )x4 the three-dimensional resection problem (P4P). Bollettino di Geo-
+2x32 2Zx3 + x22 X2 Y 2 desia e Science Affini 62:79102
Awange JL, Fukuda Y, Wickert J, Aoyama Y (2004) Analytic solution
of GPS atmospheric sounding refraction angles. Earth Planets Space
56:573587
Appendix B Becker T, Weispfenning V (1998) Grobner bases. A computational ap-
proach to commutative algebra. Graduate text in mathematics 141,
2nd edn. Springer, Berlin Heidelberg New York
Coefficients d1 , ..., d15 for GPS-meteorology Buchberger B (2001) Grobner bases. A short introduction for system
theorists. In: Moreno-Diaz R et al. (eds): EUROCAST 2001, LNCS
d1 = b42 2178, pp. 119
Cox D, Little J, OShea D (1997) Ideals, varieties, and algorithms. An
d2 = 2b4 b5 introduction to computational algebraic geometry and commutative
d3 = 2b4 b3 algebra. Springer, Berlin Heidelberg New York
Cox D, Little J, OShea D (1998) Using algebraic geometry. Graduate
d4 = 2b6 b4 + b52 + b72 Text in mathematics 185. Springer, Berlin Heidelberg New York
d5 = 2b1 b4 + b32 b72 Grafarend E, Lohse P (1991) The minimal distance mapping of the topo-
graphic surface onto the (reference) ellipsoid of revolution. Manu-
d6 = 2b3 b5 + 2b2 b4 scripta Geodaetica 16:92110
d7 = 2b6 b5 Grafarend E, Lohse P, Schaffrin B (1989) Dreidimensional-
er Ruckwartsschnitt. Zeitschrift fur Vermessungswesen 114:61
d8 = 2b1 b3 67,127137,172175,225234,278287
Hanselman D, Littlefield B (1997) The students edition of Matlab. Pren-
d9 = 2b1 b5 + 2b2 b3 tice Hall, New Jersey
d10 = 2b3 b6 + 2b5 b2 Healey S, Jupp A, Offiler D, Eyre J (2003) The assimilation of radio
occultation measurements. In: Reigber C, Luhr H, Schwintzer P (eds)
d11 = b12 First CHAMP mission results for gravity, magnetic and atmospheric
d12 = 2b1 b2 studies, Springer, Berlin Heidelberg New York
Irving RS (2004) Integers, polynomials and rings. Springer, Berlin Hei-
d13 = 2b1 b6 + b22 + b72 delberg New York
d14 = 2b2 b6 Lannes A, Durand S (2003) Dual algebraic formulation of differential
GPS. J Geod 77:2229
d15 = b62 b72 Lauritzen N (2003) Concrete abstract algebra. From numbers to
Grobner basis. Cambridge University Press, Cambridge
with McCoy NH, Janusz GJ (2001) Introduction to abstract algebra. Harcout
Academic Press, San Diego
b1 = a12 + a22 Merrit EL (1949) Explicit three-point resection in space. Phot Eng
b2 = 2aa2 15:649665
SteinerAK (1998) High resolution sounding of key climate variables us-
b3 = 2a2 a4 ing the radio occultation technique. Dissertation, Institute for Mete-
b4 = (a32 + a42 ) orology and Geophysics, vol 3, University of Graz
Wickert, J. (2002) Das CHAMP-Radiookkultationsexperiment: algo-
b5 = 2aa4 rithmen, Prozessierungssystem und erste Ergebnisse. Dissertation.
b6 = a 2 a12 a32 Scientific Technical Report STR02/07, GFZ Potsdam
b7 = 2a1 a3

References

Awange JL (2002) Groebner basis solution of planar resection. Survey


Rev 36(285):528543
Awange JL, Grafarend E (2003a) Explicit solution of the overdeter-
mined threedimensional resection problem. J Geod 76:605616

You might also like