J. Geod. Sci.

2021; 11:111–121

Research Article Open Access

G. Panou and R. Korakitis

Analytical and numerical methods of converting

Cartesian to ellipsoidal coordinates
DOI: https://doi.org/10.1515/jogs-2020-0126 Usually, three variants of ellipsoidal coordinates,
Received March 8, 2021; accepted September 27, 2021 namely (s1 , s2 , s3 ), (ρ, µ, ν) and (u, β, λ) are used in the
Abstract: In this work, two analytical and two numerical literature. These variants are presented e.g. by Tabanov
methods of converting Cartesian to ellipsoidal coordinates (1999), Dassios (2012) and Panou (2014). However, the co-
of a point in space are presented. After slightly modifying ordinates (u, β, λ) are the only ones with invariant inter-
a well-known exact analytical method, a new exact ana- vals of variation as the system transforms to the oblate
lytical method is developed. Also, two well-known numer- spheroidal system (Dassios 2012). In other words, using
ical methods, which were developed for points exactly on the ellipsoidal coordinates (u, β, λ) we avoid the degen-
the surface of a triaxial ellipsoid, are generalized for points eracy of the coordinates s3 and ν, as the triaxial ellipsoid
in space. The four methods are validated with numerical approaches an oblate spheroid. This is of special impor-
experiments using an extensive set of points for the case tance for geodetic applications related to the Earth, so we
of the Earth. Then, a theoretical and a numerical compar- will study this variant only in the present work.
ative assessment of the four methods is made. Further- The Cartesian coordinates (x, y, z) of a point in space
more, the new exact analytical method is applied for an can be obtained from the ellipsoidal coordinates (ellip-
almost oblate spheroid and for the case of the Moon and soidal parameter u ∈ [0, +∞) , ellipsoidal latitude β ∈
 π π
the results are compared. We conclude that, the general- − 2 , + 2 and ellipsoidal longitude λ ∈ (−π, +π ]), in-
ized Panou and Korakitis’ numerical method, starting with troduced by Tabanov (1999) and also presented in Panou
approximate values from the new exact analytical method, (2014), by
  12 1
is the best choice in terms of accuracy of the resulting el- x = u2 + E2x B 2 cosλ (1)
lipsoidal coordinates.   21
y = u2 + E2y cosβsinλ (2)
Keywords: exact method, Laplace’s equation, orthogonal
net, triaxial ellipsoid 1
z = usinβL 2 (3)

where
1 Introduction B = cos2 β +
E2e
sin2 β (4)
E2x
Ellipsoidal coordinates, as a triply orthogonal coordi- and
nate system, have the advantage of decomposing the E2e
L =1− cos2 λ (5)
Laplace partial differential equation into separable func- E2x
tions. Thus, these coordinates are particularly valuable in while
potential theory. Also, the coordinate surface of a triax-   12   21   12
ial ellipsoid, that fits the physical boundary of a problem, E x = a2x − b2 , E y = a2y − b2 , E e = a2x − a2y ,
helps to express a boundary condition in a simple way. (6)
For instance, one may use ellipsoidal coordinates in the are the linear eccentricities (obviously E2e = E2x − E2y ) of the
gravity field which has a triaxial ellipsoid as a level sur- triaxial ellipsoid (u = b):
face. Other examples of the use of these coordinates can
be found in Dassios (2012). x2 y2 z2
+ + = 1, 0 < b < a y < a x (7)
a2x a2y b2
G. Panou: Department of Surveying Engineering, National where a x , a y and b are its three semi-axes.
Technical University of Athens, 15780 Athens, Greece, E-mail:
The geometrical interpretation of these ellipsoidal co-
geopanou@survey.ntua.gr
R. Korakitis: Department of Surveying Engineering, National Techni-
ordinates is given by Panou (2014). Specifically, the ellip-
cal University of Athens, 15780 Athens soidal parameter u is the polar semi-axis of the confocal el-

Open Access. © 2021 G. Panou and R. Korakitis, published by De Gruyter. This work is licensed under the Creative Commons
Attribution alone 4.0 License.
112 | G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates

lipsoid which passes through the point (x, y, z). The ellip- the conversion, the conventions with regard to the proper
soidal latitude β represents the inclination of the asymp- quadrant for the β and λ need to be applied from the signs
totes of the confocal principal hyperbola on the plane x = of x, y, and z. Furthermore, in order to ensure the precision
0. The ellipsoidal longitude λ represents the inclination of the numerical computations, we use functions which
of the asymptotes of the confocal principal hyperbola on are continuous and without singularities within their do-
the plane z = 0. In the case of an oblate spheroid, where main of use and beyond their endpoints if possible, e.g.
a x = a y ≡ a, i.e. E x = E y ≡ E and E e = 0, Eqs. (1)- using absolute values and variants of functions for differ-
(3) reduce to well-known expressions (see Heiskanen and ent domains of space.
Moritz 1967). From Eqs. (2) and (3), we obtain
In the case where u = b, i.e. for points located on the
sin2 λ
surface of a triaxial ellipsoid, Eqs. (1)-(3) reduce to well- tan2 β = M (8)
L
known expressions introduced by Jacobi (1839). These
curvilinear ellipsoidal coordinates (β, λ) constitute an or- and from Eqs. (1) and (2)
thogonal net of curves on the surface of a triaxial ellip-
B
soid, therefore they eliminate cumbersome mathematical tan2 λ = P (9)
cos2 β
expressions for problems like the study of geodesics (e.g.
Panou and Korakitis 2019) and for simple determinations, where
z2 u2 + E2y
such as the surface area. Thus, the problem of converting M= · (10)
y2 u2
the Cartesian coordinates (x, y, z) of a point on the sur-
face of a triaxial ellipsoid to the corresponding ellipsoidal and
y2 u2 + E2x
coordinates (β, λ) is of special importance. For the solu- P= · (11)
x2 u2 + E2y
tion of this problem, an exact analytical and a numerical
method were developed by Panou and Korakitis (2019). In Inverting Eqs. (8)-(11) results to
that work, an extensive test set was used, in order to eval- L
uate the performance of the two methods in double and cot2 β = N (12)
sin2 λ
quad precision. These numerical validations using points
on the surface of the triaxial ellipsoid gave results bet- and
cos2 β
ter than 1 mm for double and 1 nm for quad precision. cot2 λ = Q (13)
B
We should also mention another numerical method which where
was developed by Bektas (2015) and was used in the works y2 u2
N = M −1 = 2
· 2 (14)
of Florinsky (2018) and Pędzich (2019). z u + E2y
In the present work, we attempt to solve the prob- and
lem of converting the Cartesian coordinates (x, y, z) of a x2 u2 + E2y
Q = P−1 = · (15)
point in space to the corresponding ellipsoidal coordinates y2 u2 + E2x
(u, β, λ), using two analytical and two numerical meth- Hence, from Eqs. (8) and (9), we have
ods. In Section 2.1., we present a modification of the ex-
"  1 #
act analytical method described by Panou (2014) with a −1 M 2
change of variable in the relevant cubic equation. In Sec- β = tan sinλ (16)
L
tion 2.2., a new exact analytical method is developed. In
1
" #
Section 3., the numerical methods, developed by Bektas −1 (PB) 2
(2015) and Panou and Korakitis (2019), are generalized for λ = tan (17)
cosβ
points in space (not only located on the surface). Thus,
the coordinate u is needed in the development of the rele- and from Eqs. (12) and (13)
vant algorithms. Finally, both a theoretical and a numeri- " 1
−1 ( NL ) 2
#
π
cal comparative assessment of the four methods is made, β = − tan (18)
2 sinλ
mainly in terms of the accuracy of the resulting ellipsoidal
coordinates. "  1 #
π Q 2
In the remainder of this section, we present the re- λ = − tan−1 cosβ (19)
2 B
quired mathematical formulas for the development of the
methods. Also, for simplicity and without loss of general-
ity, we always assume x ≥ 0, y ≥ 0 and z ≥ 0. Hence, after
 G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates | 113

and
2 Analytical methods ! 21
−1 −E2y − t3
λ = tan , if | − E2y − t3 | ≤ |E2x + t2 | (34)
2.1 Case of a triaxial ellipsoid only E2x + t3
or
The ellipsoidal coordinates (u, β, λ) can be obtained from ! 12
the Cartesian coordinates (x, y, z) of a point in space by π E2x + t3
λ = −tan−1 , if |−E2y −t3 | > |E2x +t3 | (35)
solving the following cubic equation in t 2 −E2y − t3
In the case of an oblate spheroid, where E x = E y ≡ E
t3 + c2 t2 + c1 t + c0 = 0 (20)
and therefore t3 = −E2 , the ellipsoidal longitude cannot
where be computed either from Eq. (34) or Eq. (35). Therefore, we
c2 = E2x + E2y − x2 − y2 − z2 (21) cannot avoid the degeneracy of the variable t3 for an oblate
  spheroid, thus Eq. (34) or Eq. (35) produce inaccurate re-
c1 = E2x E2y − E2y x2 − E2x y2 − E2x + E2y z2 (22) sults in the case of an almost oblate spheroid. In addition,
the operations for the computation of ellipsoidal latitude
c0 = −E2x E2y z2 (23)
and longitude lead to a loss of precision for points near
Comparing Eq. (20) with the corresponding equation the principal planes (x = 0 or y = 0 or z = 0). All these rea-
presented by Panou (2014), the two cubic equations are sons constitute the motivation for the development of the
related with the transformation s = t − b2 . Here, the coef- following analytical method which uses different mathe-
ficients are expressed in terms of the linear eccentricities matical formulas for such computations. However, for the
and thus keep the magnitude to a moderate size. Also, Eq. development of all other methods of converting Cartesian
(20) has three real roots, which are distributed as follows to ellipsoidal coordinates which will be presented, the el-
lipsoidal parameter u is considered known, as computed
− E2x ≤ t3 ≤ − E2y ≤ t2 ≤ 0 ≤ t1 < +∞ (24) from Eq. (31).

and can be represented using trigonometric functions as

ω c
1
t1 = 2p 2 cos − 2 (25) 2.2 Case of a triaxial ellipsoid and an oblate
3 3
spheroid
 
1 ω − 2π c
t2 = 2p cos
2 − 2 (26) The ellipsoidal parameter u now being known, eliminat-
3 3
  ing λ between Eqs. (12) and (13) we obtain a quadratic
1 ω − 4π c
t3 = 2p 2 cos − 2 (27) equation for cot2 β, whose solution is
3 3
( "  12 #) 12
where E2e

π −1 1 2
c22 − 3c1 β = − tan W + W +4 2N (36)
p= (28) 2 2 Ex
9
where
9c1 c2 − 27c0 − 2c32 E2y x2 u2 E2e
q= (29) W=N+ · · − (37)
54
E2x z2 u2 + E2x E2x
and   Also, eliminating β between Eqs. (8) and (9) we obtain a
q 3
ω = cos−1 3 , if q < p 2 else ω = 0 (30) quadratic equation for tan2 λ, whose solution is
p 2
  ! 12  21
Then, the ellipsoidal coordinates (u, β, λ) are given 
1 E 2 
y
by λ = tan−1 U + U2 + 4 2 P  (38)
1 2 Ex 
u = (t1 ) 2 (31)
  12 where
β = tan −1 −t2
, if | − t2 | ≤ |E2y + t2 | (32) E2e z2 u2 + E2x E2y
E2y + t2 U =P+ · · − 2 (39)
E2x x2 u2 Ex
or Finally, eliminating β between Eqs. (8) and (13) we ob-
! 21 tain a quadratic equation for cot2 λ, whose solution is
π E2y + t2
β = − tan−1 , if | − t2 | > |E2y + t2 | (33) ( "
2
 12 #) 21
2 −t2

π 1 E
λ = − tan−1 V + V 2 + 4 2x Q (40)
2 2 Ey
114 | G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates

where quantities M, N, P and Q, Eqs. (16) and (18) are rewritten

E2 E2 as
V = Q − 2e M − 2x (41) "  1 #
Ey Ey M 2
−1
β i+1 = tan sinλ i (46)
Hence, starting from the given Cartesian coordinates and Li
the semi-axes of the triaxial ellipsoid, we compute the el- " 1
#
π −1 ( NL ) 2
lipsoidal parameter u from Eq. (31) and the quantities M, i
β i+1 = − tan (47)
2 sinλ i
N, P and Q from Eqs. (10), (14), (11) and (15), respectively.
Then, the algorithm proceeds as follows: and Eqs. (17) and (19) as
If z ≤ y ≤ x, we first compute λ from Eq. (38) and then β " 1
#
from Eq. (16). The case y = 0 (i.e. x-axis) is excluded since −1 (PB i+1 ) 2
λ i+1 = tan (48)
β = λ = 0. Also, if z ≤ y > x, we first compute λ from Eq. cosβ i+1
(40) and then β from Eq. (16). "  12 #
If z > y ≥ x, we first compute β from Eq. (36) and then λ π Q
λ i+1 = − tan−1 cosβ i+1 (49)
from Eq. (19). The case y = 0 (i.e. z-axis) is excluded since 2 B i+1
β = λ = 2π . Also, if z > y < x, we compute β from Eq. (36) where
and λ from Eq. (38). E2e
B i+1 = cos2 β i+1 + sin2 β i+1 (50)
E2x
and
2.2.1 Case of an oblate spheroid E2e
Li = 1 − cos2 λ i (51)
E2x
In the case of an oblate spheroid, where E x = E y ≡ E, the where i = 0, 1, 2, . . . and an approximate initial value of
ellipsoidal coordinates are computed by the well-known ellipsoidal longitude λ0 is required. Then, the algorithm
expressions (e.g. Heiskanen and Moritz 1967) proceeds as follows:
1 If z ≤ y ≤ x, we first compute β from Eq. (46) and then λ
 
1  1  2
from Eq. (48). The case y = 0 (i.e. x-axis) is excluded since

2 2 2 2
u= R + R + 4E z (42)
2 β = λ = 0. Also, if z ≤ y > x, we first compute β from Eq.
where (46) and then λ from Eq. (49).
2 2
R=x +y +z −E 2 2
(43) If z > y ≥ x, we first compute β from Eq. (47) and then λ
from Eq. (49). The case y = 0 (i.e. z-axis) is excluded since
while β is given by β = λ = 2π . Also, if z > y < x, we first compute β from
"  2 1 # "  1 #Eq. (47) and then λ from Eq. (48). Thus, the xz-plane is ex-
u + E2 2
 2
−1 z π −1 u x + y2 2
β = tan · = −tan · cluded and the method fails to give a solution.
u x2 + y2 2 z u2 + E2
In all cases, the process is repeated until the values
(44)
dβ = β i+1 − β i and dλ = |λ i+1 − λ i | become negligibly
and λ by
small. We note that, the iterative method described in Bek-
 
−1 y
  π
−1 x tas (2015) uses only Eqs. (46) and (48) and approximate
λ = tan = − tan (45)
x 2 y values for both coordinates β0 and λ0 .
It is worth mentioning that Eqs. (31) and (42), although
different in form, are equivalent in the case of an oblate
spheroid, where E x = E y ≡ E and E e = 0. Furthermore,
3.2 Generalized Panou Korakitis’ method
Eqs. (36), (38) and (40) after the necessary simplifications,
Considering the Cartesian coordinates (x, y, z) of a point
reduce to Eqs. (44) and (45).
in space as measurements, the application of the method
of least squares (e.g. Ghilani and Wolf 2006) requires writ-
ing of Eqs. (1)-(3) in the form
3 Numerical methods   12 1
u2 + E2x B 2 cosλ = x + υ x (52)
3.1 Generalized Bektas’ method   12
u2 + E2y cosβsinλ = y + υ y (53)
Knowing the Cartesian coordinates (x, y, z) of a point
1
in space and the ellipsoidal parameter u, hence also the usinβL 2 = z + υ z (54)
 G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates | 115

which, in linear approximation, can be represented as

" # 4 Numerical tests and comparisons
dβ
J = δl + υ (55)
dλ 4.1 Test set
where the 3 × 2 Jacobian matrix is
 ∂x ∂x  In order to validate the four methods of conversion pre-
∂β ∂λ sented above and to evaluate their performance, we used
∂y ∂y
J= (56) an extensive test set of points. This is a set of 7023 points
 
∂β ∂λ 
∂z ∂z in space, distributed into thirteen groups, as described in
∂β ∂λ
Table 1. For the ellipsoidal parameter u, three representa-
with elements
tive values were chosen: u = 2b (inside ellipsoid), u = b (on
∂x  2 E2 sin 2β
1
( ) the ellipsoid) and u = 2b (outside ellipsoid). Also, for sim-

y
= − u2 + E2x · cosλ (57)
∂β E2x 1
2B 2 plicity and without loss of generality, β and λ were chosen
∂y   21 in [0◦ , 90◦ ].
= − u2 + E2y sinβsinλ (58) Models of three different bodies were considered: a
∂β
∂z 1 triaxial ellipsoid like the Earth, with a x = 6378172 m,
= ucosβL 2 (59) a y = 6378102 m and b = 6356752 m (Panou et al. 2020),
∂β
∂x   12 1 an almost oblate spheroid a x = 6378137.001 m, a y =
= − u2 + E2x B 2 sinλ (60) 6378137 m and b = 6356752 m and a triaxial ellipsoid
∂λ
∂y  2  12 like the Moon, with a x = 1738056 m, a y = 1737843 m
= u + E2y cosβcosλ (61) and b = 1735485 m (Iz 2009). The corresponding Carte-
∂λ
∂z E2 sin (2λ) sian coordinates for each body and for any point were com-
= u 2e sinβ 1 (62) puted using Eqs. (1)-(3), keeping 27 significant digits in the
∂λ Ex 2L 2
which are computed from the known ellipsoidal parameter results.
u and approximate ellipsoidal coordinates β0 and λ0 . The All algorithms were implemented in C++ and com-
3 × 1 vector piled using the GNU Compiler Collection version 9.2 at op-
timization level 3, running under the Debian Linux op-
 
x − x0
δl =  y − y0 
 
(63) erating system. We employed the open-source “libquad-
z − z0 math”, the GCC Quad-Precision Math Library, which pro-
vides a precision of 33 digits (referred to as “quad”). In
is computed from the Cartesian coordinates (x0 , y0 , z0 ) of
contrast, the long double data type (referred to simply as
the approximate ellipsoidal coordinates β0 and λ0 and υ
“double”) maps to the IEEE-754 80-bit extended precision
is a 3 × 1 vector of residuals. The updated values of the
type on this platform and provides a precision of 18 digits.
approximate ellipsoidal coordinates are
" # " # " # The codes were executed on a personal computer running
β β0 dβ a 64-bit Linux Debian operating system. The main charac-
= + (64)
λ λ0 dλ teristics of the hardware were: AMD Ryzen 7 4800H CPU
(with 8 cores and clocked at 2.9 GHz) and 16 GB of RAM.
where the solution vector is given by
" # In addition, we note that all figures, presented in the
dβ following sections, were produced using the MATLAB soft-
= N−1 JT δl (65)
dλ ware package.
Furthermore, we added a test set of 300 points in space
while
with β = λ = 45◦ and values of u from 100 km to
" #
T n11 n12
N=J J= (66) 30000 km, with a step size 100 km (Group 14), in order to
n21 n22
validate the performance of the methods with respect to a
and hence wider range of values of the ellipsoidal parameter u.
" #
−1 1 n22 −n12
N = (67)
n11 n22 − n12 n21 −n21 n11

We note that the determinant of the matrix N equals 5 Results

zero only at the umbilical points, where β = 2π and λ = 0,
by definition. The process is repeated until the corrections All four methods were tested, in the case of the Earth, us-
|dβ| and |dλ| become negligibly small. ing double and quad precision and the Cartesian coordi-
116 | G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates

Table 1. Description of the points in the test set (total 7023 points)

Group Case 𝑢𝑢 𝛽𝛽 𝜆𝜆 Number

𝑏𝑏
1 G 1st octant , 𝑏𝑏, 2𝑏𝑏 5° − 85° every 5° 5° − 85° every 5° 867
2
𝑏𝑏
2 A1 𝑥𝑥𝑥𝑥 − plane , 𝑏𝑏, 2𝑏𝑏 0° 5° − 90° every 5° 54
2
𝑏𝑏
3 A2.1 , 𝑏𝑏, 2𝑏𝑏 0° − 90° every 5° 0° 57
𝑥𝑥𝑥𝑥 − plane 2
𝑏𝑏
4 A2.2 , 𝑏𝑏, 2𝑏𝑏 90° 5° − 90° every 5° 54
2
𝑏𝑏
5 A3 𝑦𝑦𝑦𝑦 − plane , 𝑏𝑏, 2𝑏𝑏 5° − 85° every 5° 90° 51
2
𝑏𝑏 1.0° − 0.0 … 1° up to 14
6 B1 near 𝑥𝑥𝑥𝑥 − plane , 𝑏𝑏, 2𝑏𝑏 5° − 90° every 5° 810
2 decimals

𝑏𝑏 1.0° − 0.0 … 1° up to 14
7 B2.1 , 𝑏𝑏, 2𝑏𝑏 0° − 90° every 5° 855
2 decimals
near 𝑥𝑥𝑥𝑥 − plane
𝑏𝑏 89.0° − 89.9 … 9° up to
8 B2.2 , 𝑏𝑏, 2𝑏𝑏 5° − 90° every 5° 810
2 14 decimals

𝑏𝑏 89.0° − 89.9 … 9° up to
9 B3 near 𝑦𝑦𝑦𝑦 − plane , 𝑏𝑏, 2𝑏𝑏 5° − 85° every 5° 765
2 14 decimals

𝑏𝑏 1.0° − 0.0 … 1° up to 14 1.0° − 0.0 … 1° up to 14

10 C1 near 𝑥𝑥 − axis , 𝑏𝑏, 2𝑏𝑏 675
2 decimals decimals

𝑏𝑏 1.0° − 0.0 … 1° up to 14 89.0° − 89.9 … 9° up to

11 C2 near 𝑦𝑦 − axis , 𝑏𝑏, 2𝑏𝑏 675
2 decimals 14 decimals

𝑏𝑏 89.0° − 89.9 … 9° up to 89.0° − 89.9 … 9° up to

12 C3 near 𝑧𝑧 − axis , 𝑏𝑏, 2𝑏𝑏 675
2 14 decimals 14 decimals

𝑏𝑏 89.0° − 89.9 … 9° up to 1.0° − 0.0 … 1° up to 14

13 D near umbilic , 𝑏𝑏, 2𝑏𝑏 675
2 14 decimals decimals

nates x, y and z as input data. From the resulting u′ , β′ and analytical method of Section 2.1. and we recorded the
λ′ at any point, we computed the differences δu = u − u′ , mean and the maximum value of iterations i, which were
δβ = β − β′ and δλ = λ − λ′ , and recorded the maximum needed. As criterion of convergence of the dβ and dλ, the
absolute values, i.e. max |δu| , max |δβ| and max |δλ| for values 10−19 rad and 10−34 rad, for double and quad pre-
every Group. Furthermore, because Eqs. (1)-(3) are numer- cision, respectively, were used.
ically stable, the results at any point were checked
 by
 All results are presented in Tables 2 to 5, where lengths
comparing the resulting Cartesian coordinates x′ , y′ , z′ δu and δr are expressed in meters and angular quantities
with the original Cartesian coordinates (x, y, z) by the δβ and δλ in arcseconds. Whenever the value of a quantity
simple Euclidean distance formula appears as “0”, this implies that it is less than the corre-
sponding machine precision. Since the values of max |δu|
 2 2 2  21
are common in all methods, they appear in Table 2 only.
 
′ ′ ′
δr = x−x + y−y + z−z (68)
From the values of max |δu| presented in Table 2, we
For every Group, we recorded the maximum value, i.e. conclude that the estimation of u is better than 10−12 m for
maxδr. We remark that only the quantity δr can be com- double and 10−19 m for quad precision (as noted earlier,
puted when starting with knowledge of the Cartesian co- this holds for all methods). On the other hand, comparing
ordinates. the results of max |δβ| , max |δλ| and maxδr, we conclude
Additionally, in the numerical methods we used ap- that only quad precision provides results suitable for most
proximate values of the coordinates β and λ from the practical applications.
 G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates | 117

Table 2. Performance of the exact analytical method of Section 2.1., in the case of the Earth, using double and quad precision

double quad
Group
max|𝛿𝛿𝑢𝑢| (m) max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m) max|𝛿𝛿𝑢𝑢| (m) max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m)
1 9.09 · 10−13 6.76 · 10−7 1.40 · 10−4 8.97 · 10−4 4.64 · 10−20 1.20 · 10−21 1.36 · 10−19 8.75 · 10−19
2 9.09 · 10−13 1.61 · 10−2 1.83 · 10−1 11.3 3.99 · 10−20 6.88 · 10−10 4.79 · 10−9 2.96 · 10−7
3 9.09 · 10 −13 40.1 699 18.2 4.99 · 10 −20
9.78 · 10 −3
1.71 · 10 −1
6.71 · 10−7
4 9.09 · 10 −13 2.12 1.24 13.9 4.90 · 10 −20
7.17 · 10 −8
1.47 · 10 −7
7.36 · 10−7
5 9.09 · 10 −13
2.54 · 10 −7
1.66 16.1 4.14 · 10 −20
8.71 · 10 −22
1.00 · 10 −7
7.04 · 10−7
6 9.09 · 10 −13
1.61 · 10 −2
2.83 · 10 −1
17.4 4.98 · 10 −20
6.06 · 10 −10
1.29 · 10 −8
7.94 · 10−7
7 9.09 · 10 −13
40.1 699 18.2 4.99 · 10 −20
1.16 · 10 −2
1.99 · 10 −1
7.72 · 10−7
8 9.09 · 10−13 2.25 4.42 17.8 4.97 · 10−20 1.14 · 10−7 1.78 · 10−7 7.20 · 10−7
9 9.09 · 10−13 3.83 · 10−7 1.66 19.8 4.93 · 10−20 1.06 · 10−21 9.97 · 10−8 7.63 · 10−7
10 9.09 · 10−13 1.44 · 10−2 3.07 · 10−1 19.0 4.97 · 10−20 6.75 · 10−10 1.28 · 10−8 7.91 · 10−7
11 9.09 · 10−13 1.47 · 10−2 2.86 · 10−1 17.6 5.00 · 10−20 6.57 · 10−10 1.10 · 10−8 6.79 · 10−7
12 9.09 · 10−13 2.42 · 10−1 4.42 20.4 5.00 · 10−20 1.32 · 10−8 2.13 · 10−7 8.63 · 10−7
13 9.09 · 10 −13
54.8 920 16.7 4.98 · 10 −20
1.16 · 10 −2
1.99 · 10 −1
7.16 · 10−7

We overemphasize that, the ellipsoidal coordinates The characteristic of the numerical method of Section
are valuable in gravity field related studies and taking into 3.2. is that the coordinates β and λ are computed indepen-
account the current estimates of the related quantities, a dently. Therefore, the worst results are usually observed at
precision better than a millimeter is sufficient. different points.
Comparing the results of max |δβ| , max |δλ| and
maxδr presented in Table 3 between the double and quad
precision, we conclude that both precisions can provide
results suitable for most practical applications (δr better
than 1 cm for double and 10−11 m for quad precision, while
angular coordinates better than 10−4 arcsec for double and
10−8 arcsec for quad precision).
It is worth noting that, in the exact analytical method
of Section 2.2., the one coordinate is computed from the
other. Therefore, the worst results are observed at the same
points.
Comparing the results of max |δβ| , max |δλ| and
maxδr presented in Table 4 between the double and quad
precision, we conclude that only quad precision can pro-
vide results suitable for most practical applications, ex-
cluding the xz-plane and the area near the umbilics.
Equations (46)-(49) may be written using as first ap-
proximate value only the ellipsoidal latitude β0 , i.e. we
first compute λ and then β. Also, these equations may be
written using approximate values for both coordinates β0
and λ0 , as suggested by Bektas (2015). However, in both
cases the results are worse than those of Table 4.
Comparing the results of max |δβ| , max |δλ| and
maxδr presented in Table 5 between the double and quad
precision, we conclude that both precisions can provide
results suitable for most practical applications (δr better
than 10−11 m for double and 10−26 m for quad precision,
Fig. 1. Values of maxδr in log10 scale for every group and method
while angular coordinates better than 10−4 arcsec for dou- in: (top) double precision (bottom) quad precision
ble and 10−8 arcsec for quad precision).
118 | G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates

Table 3. Performance of the exact analytical method of Section 2.2., in the case of the Earth, using double and quad precision

double quad
Group
max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m) max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m)
1 3.75 · 10−14 5.00 · 10−14 3.19 · 10−12 4.78 · 10−22 5.65 · 10−22 4.57 · 10−27
2 0 2.65 · 10−14 1.88 · 10−12 0 3.80 · 10−22 3.23 · 10−27
3 4.24 · 10−6 0 1.29 · 10−12 6.95 · 10−10 8.82 · 10−9 5.78 · 10−12
4 0 1.95 · 10−13 9.58 · 10−13 0 1.06 · 10−20 1.68 · 10−27
5 2.50 · 10−14 0 1.85 · 10−12 3.63 · 10−22 0 3.24 · 10−27
6 1.17 · 10−15 5.00 · 10−14 3.28 · 10−12 1.82 · 10−23 4.39 · 10−22 6.46 · 10−27
7 4.24 · 10−6 3.60 · 10−5 2.22 · 10−4 6.95 · 10−10 8.79 · 10−9 1.17 · 10−12
8 3.60 · 10−7 2.65 · 10−13 2.22 · 10−5 6.66 · 10−17 1.06 · 10−20 4.09 · 10−15
9 2.50 · 10 −14
3.60 · 10 −5 2.21 · 10−3
4.42 · 10 −22
3.79 · 10 −14
2.33 · 10−12
10 3.60 · 10 −6
3.60 · 10 −6
3.14 · 10−4
9.51 · 10 −15 1.89 · 10 −14
1.17 · 10−12
11 3.90 · 10 −16
3.60 · 10 −5
2.22 · 10−3
1.87 · 10 −23
3.79 · 10 −14
2.34 · 10−12
12 3.60 · 10 −7
3.60 · 10 −6
2.56 · 10−5
6.66 · 10 −17
6.66 · 10 −16
4.73 · 10−15
13 4.24 · 10 −6
3.60 · 10 −5
3.88 · 10−6
6.59 · 10 −10
8.79 · 10 −9
2.58 · 10−15

Table 4. Performance of the numerical method (generalized Bektas) of Section 3.1., in the case of the Earth, using double and quad preci-
sion

double quad
Group mean(𝑖𝑖) mean(𝑖𝑖)
max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m) max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m)
/max(𝑖𝑖) /max(𝑖𝑖)
1 2.50 · 10−14 5.00 · 10−14 2.03 · 10−12 8/100 4.78 · 10−22 5.65 · 10−22 4.91 · 10−27 10/150
2 0 2.50 · 10−14 1.88 · 10−12 2/2 0 3.80 · 10−22 2.28 · 10−27 2/2
3 - - - - - - - -
4 - - - - - - - -
5 2.50 · 10−14 0 1.85 · 10−12 3/3 3.63 · 10−22 0 3.24 · 10−27 3/3
6 3.90 · 10−16 2.50 · 10−14 2.03 · 10−12 2/4 1.82 · 10−23 4.39 · 10−22 4.04 · 10−27 2/4
7 2.50 · 10−14 3600 111 2/100 5.11 · 10−22 1.62 · 10−2 7.74 · 10−9 22/150
8 1.15 · 10−4 8.62 · 10−1 2.34 2/100 6.96 · 10−19 1.98 · 10−16 6.09 · 10−17 19/150
9 2.50 · 10−14 2.50 · 10−14 2.10 · 10−12 3/4 4.42 · 10−22 1.62 · 10−23 3.66 · 10−27 3/4
10 1.14 · 10−14 3.90 · 10−16 1.85 · 10−12 4/8 1.47 · 10−23 1.90 · 10−23 3.32 · 10−27 4/9
11 3.90 · 10−16 2.50 · 10−14 2.07 · 10−12 2/3 1.87 · 10−23 1.50 · 10−23 3.55 · 10−27 2/3
12 2.50 · 10−14 3.01 · 10−2 1.06 · 10−1 2/5 1.62 · 10−23 2.00 · 10−23 6.36 · 10−23 2/5
13 3600 919 33067 4/100 3262 1.99 · 10−1 32769 127/150

Comparing the results of all methods of conversion, as As a novel method, we also investigated the perfor-
they are presented in Figure 1, we conclude that only the mance of the exact analytical method of Section 2.2. for the
exact analytical method of Section 2.2. and the numerical two other bodies described in Section 4.1. and the results
method (Panou and Korakitis) of Section 3.2. are adequate are presented in Tables 6 and 7.
to provide excellent results for both precisions. The basic In the case of an almost oblate spheroid (Table 6), the
difference is that, in the numerical method (Table 5), the results of max |δu| are better than 1 · 10−12 m for double
worst results are observed in different points, as indicated and 5 · 10−20 m for quad precision. In this case, the um-
by the values of maxδr, compared with the corresponding bilics are near the z-axis, where the longitude is singular.
values in the analytical method (Table 3).
In Figure 2, we present the values of maxδr for the In the case of the Moon (Table 7), the results of
group 14 of all methods of conversion in the case of the max |δu| are better than 3 · 10−13 m for double and
Earth. We observe that there is a small increase in maxδr 7 · 10−21 m for quad precision.
with increasing u for all methods, except the exact ana- Comparing the results of the analytical method of Sec-
lytical method of Section 2.1., which shows a much bigger tion 2.2. for the three bodies, we observe that this method
deterioration of accuracy with increasing u. provides almost equivalent results for the three bodies, ex-
 G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates | 119

Table 5. Performance of the numerical method (generalized Panou and Korakitis) of Section 3.2., in the case of the Earth, using double and
quad precision

double quad
Group mean(𝑖𝑖) mean(𝑖𝑖)
max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m) max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m)
/max(𝑖𝑖) /max(𝑖𝑖)
1 2.50 · 10−14 2.50 · 10−14 1.95 · 10−12 6/30 4.78 · 10−22 5.65 · 10−22 3.64 · 10−27 17/30
2 0 1.25 · 10 −14 1.88 · 10 −12
4/30 0 3.80 · 10 −22
2.28 · 10−27 11/30
3 5.17 · 10−6 7.70 · 10−5 1.29 · 10−12 5/30 6.95 · 10−10 8.82 · 10−9 2.42 · 10−27 12/30
4 0 1.06 · 10−13 9.58 · 10−13 6/30 0 1.06 · 10−20 1.70 · 10−27 14/30
5 1.25 · 10−14 0 1.85 · 10−12 4/30 3.63 · 10−22 0 1.87 · 10−27 11/30
6 3.90 · 10−16 2.50 · 10−14 2.03 · 10−12 4/30 1.82 · 10−23 4.39 · 10−22 4.86 · 10−27 10/30
7 5.17 · 10−6 7.70 · 10−5 2.03 · 10−12 5/30 6.95 · 10−10 8.79 · 10−9 3.61 · 10−27 11/30
8 2.50 · 10−14 4.20 · 10−13 1.83 · 10−12 6/30 1.85 · 10−22 1.06 · 10−20 3.33 · 10−27 19/30
9 2.50 · 10−14 2.50 · 10−14 2.07 · 10−12 4/30 4.42 · 10−22 1.62 · 10−23 3.61 · 10−27 15/30
10 3.90 · 10−16 7.81 · 10−16 2.73 · 10−12 2/3 1.47 · 10−23 1.90 · 10−23 3.23 · 10−27 2/3
11 7.81 · 10−16 2.50 · 10−14 1.92 · 10−12 2/3 1.87 · 10−23 1.50 · 10−23 3.39 · 10−27 8/30
12 2.50 · 10−14 2.50 · 10−14 1.83 · 10−12 3/3 1.62 · 10−23 2.45 · 10−23 3.43 · 10−27 12/30
13 5.17 · 10 −6
7.70 · 10 −5
1.82 · 10 −12
2/30 6.60 · 10 −10
8.79 · 10 −9
3.25 · 10−27 21/30

Table 6. Performance of the exact analytical method of Section 2.2., in the case of an almost oblate spheroid, using double and quad preci-
sion

double quad
Group
max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m) max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m)
1 2.65 · 10−12 1.52 · 10−11 2.67 · 10−12 4.79 · 10−22 5.99 · 10−22 6.27 · 10−27
2 0 2.50 · 10−14 1.74 · 10−12 0 3.87 · 10−22 2.42 · 10−27
3 2.65 · 10−12 1.43 1.29 · 10−12 1.70 · 10−12 1.48 · 10−8 4.04 · 10−27
4 0 5.67 · 10−5 9.63 · 10−13 2.47 · 10−16 7.72 · 10−21 2.73 · 10−15
5 2.50 · 10 −14 0 −12
1.29 · 10 3.47 · 10 −22 0 2.02 · 10−27
6 1.13 · 10 −14
4.50 · 10 −14 −12
2.61 · 10 1.74 · 10 −23
4.33 · 10 −22
6.26 · 10−27
7 2.65 · 10 −12
1.43 2.22 · 10 −4
1.70 · 10 −12
1.47 · 10 −8
1.16 · 10−12
8 6.09 · 10 −9
5.67 · 10 −5 2.22 · 10 −7
1.53 · 10 −21
1.28 · 10 −20
9.11 · 10−20
9 2.50 · 10 −14
3.60 · 10 −5 2.21 · 10 −3
4.54 · 10 −22
3.79 · 10 −14
2.20 · 10−12
10 3.60 · 10−6 3.60 · 10−6 3.14 · 10−4 9.51 · 10−15 1.89 · 10−14 1.17 · 10−12
11 3.90 · 10−16 3.60 · 10−5 2.22 · 10−3 1.90 · 10−23 3.79 · 10−14 2.34 · 10−12
12 3.60 · 10−9 3.60 · 10−6 2.27 · 10−7 1.53 · 10−21 1.66 · 10−19 9.44 · 10−20
13 1.76 · 10 −4
1.43 3.88 · 10 −6
1.62 · 10 −13
1.30 · 10 −10
2.04 · 10−14

hibiting a slightly better performance in the case of the lipsoidal latitude and longitude are different. In the first
Moon. exact analytical method the formulas for such calculations
are well-known but in the second exact analytical method
the corresponding formulas are developed for the first time
in this work. The first numerical method is a generalization
6 Conclusions of a method of Bektas (2015) and the second of a method of
Panou and Korakitis (2019), in order to enable both meth-
In this work, a theoretical and a numerical comparative as-
ods to be applicable for points in space, not only on the
sessment of four methods (two analytical and two numer-
surface of a triaxial ellipsoid.
ical) of converting Cartesian (x, y, z) to ellipsoidal coor-
Theoretically, the precision of the two exact analyti-
dinates (u, β, λ) of a point in space have been presented.
cal methods is limited only when one approaches singu-
In all four methods the ellipsoidal parameter u is calcu-
lar points in the cumbersome expressions involved. On the
lated by the same formula, while the computations of el-
other hand, numerical methods with iterative approxima-
120 | G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates

Table 7. Performance of the exact analytical method of Section 2.2., in the case of the Moon, using double and quad precision

double quad
Group
max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m) max|𝛿𝛿𝛿𝛿| (") max|𝛿𝛿𝛿𝛿| (") max 𝛿𝛿𝑟𝑟 (m)
1 2.50 · 10−14 4.50 · 10−14 5.12 · 10−13 7.74 · 10−22 1.49 · 10−21 1.28 · 10−27
2 0 2.50 · 10−14 3.22 · 10−13 0 3.27 · 10−22 1.18 · 10−27
3 1.70 · 10 −5 0 4.69 · 10−13
2.98 · 10 −9
1.56 · 10 −8
5.96 · 10−12
4 0 2.60 · 10 −13
2.41 · 10−13
3.58 · 10 −13
4.03 · 10 −21
2.31 · 10−12
5 2.50 · 10 −14 0 2.57 · 10−13
3.75 · 10 −22 0 5.08 · 10−28
6 1.17 · 10 −15
2.65 · 10 −14
6.65 · 10−13
1.43 · 10 −23
7.08 · 10 −22
1.86 · 10−27
7 1.70 · 10 −5
7.14 · 10 −5
1.01 · 10 −4
2.98 · 10 −9
1.56 · 10 −8
3.18 · 10−13
8 3.60 · 10 −6
2.73 · 10 −13
6.06 · 10 −5
2.40 · 10 −15 6.87 · 10 −21
3.98 · 10−14
9 2.50 · 10 −14
3.60 · 10 −5 6.04 · 10 −4
7.06 · 10 −22
3.79 · 10 −14
6.37 · 10−13
10 3.60 · 10−6 3.60 · 10−6 8.39 · 10−5 9.51 · 10−15 1.89 · 10−14 3.19 · 10−13
11 3.90 · 10−16 3.60 · 10−5 6.06 · 10−4 1.15 · 10−23 3.79 · 10−14 6.39 · 10−13
12 3.60 · 10−6 3.60 · 10−6 6.31 · 10−5 2.40 · 10−15 2.40 · 10−15 4.20 · 10−14
13 1.70 · 10−5 7.15 · 10−5 1.06 · 10−7 3.59 · 10−9 1.56 · 10−8 1.31 · 10−17

Fig. 2. Values of maxδr in log10 scale for the group 14 and all meth- Fig. 3. Values of maxδr in log10 scale for every group of the exact
ods in: (top) double precision (bottom) quad precision analytical method of Section 2.2. for the three bodies in: (top) dou-
ble precision (bottom) quad precision

tions are more precise but the execution time is longer.

Practically, all methods are validated with numerical ex- Comparing the results of all methods of conversion,
periments using an extensive set of test points, common for a test set of 7023 points for the Earth, we conclude that
for all methods. only the exact analytical method of Section 2.2. and the
 G. Panou and R. Korakitis, Analytical and numerical methods of converting Cartesian to ellipsoidal coordinates | 121

numerical method (Panou and Korakitis) of Section 3.2. Panou G., 2014. A Study on Geodetic Boundary Value Problems in
are adequate to provide excellent results for both preci- Ellipsoidal Geometry. Ph.D. Thesis, Department of Surveying
Engineering, National Technical University of Athens, Greece.
sions. Specifically, the numerical validations of the analyt-
Panou G. and Korakitis R., 2019. Geodesic equations and their nu-
ical method produce values of δr better than 1 cm for dou-
merical solution in Cartesian coordinates on a triaxial ellip-
ble and 10−11 m for quad precision, whereas in the numeri- soid. Journal of Geodetic Science, 9, 1-12.
cal method the values of δr are better than 10−11 m for dou- Panou G., Korakitis R. and Pantazis G., 2020. Fitting a triaxial ellip-
ble and 10−26 m for quad precision. Through the analytical soid to a geoid model. Journal of Geodetic Science, 10, 69-82.
method, we also have at hand formulas for the calculation Pędzich P., 2019. A low distortion conformal projection of a tri-axial
ellipsoid and its application for mapping of extra-terrestrial
of ellipsoidal latitude and longitude useful for other cal-
objects. Planetary and Space Science, 178, 104697.
culations, such as the derivatives etc., while the numerical Tabanov M.B., 1999. Normal forms of equations of wave functions
method provides only a way of computation. Also, another in new natural ellipsoidal coordinates. In: Uraltseva N.N. (Ed.),
difference is that, in the numerical method, the worst re- Proceedings of the St. Petersburg Mathematical Society, Vol-
sults for β and λ are observed in different points compared ume V. American Mathematical Society Translations – Series 2,
193, 225-238.
with the analytical method.
We also investigated the performance of all methods
of conversion with respect to a wide range of values of
ellipsoidal parameter u (300 points for the Earth). Again,
the two aforementioned methods provide reliable results.
Also, as a novel exact analytical method, which can
provide approximate values for the numerical method,
we also analyzed its performance for two other bodies
(an almost oblate spheroid and the Moon), with almost
equivalent results. Finally, considering all above detailed
comparisons, we conclude that the generalized Panou and
Korakitis’ numerical method (Section 3.2.), starting with
approximate values from the new exact analytical method
(Section 2.2.), is the best choice in terms of accuracy of the
resulting ellipsoidal coordinates.

Acknowledgements: The authors greatly appreciate

valuable suggestions and comments by anonymous
referees which improved the readability of the article.

References
Bektas S., 2015. Geodetic computations on triaxial ellipsoid. Inter-
national Journal of Mining Science, 1, 25-34.
Dassios G., 2012. Ellipsoidal Harmonics: Theory and Applications.
Cambridge University Press, Cambridge.
Florinsky I.V., 2018. Geomorphometry on the surface of a triaxial
ellipsoid: towards the solution of the problem. International
Journal of Geographical Information Science, 32, 1558-1571.
Ghilani C. and Wolf P., 2006. Adjustment Computations: Spatial
Data Analysis. 4th ed., John Wiley & Sons, Inc.
Heiskanen W.A. and Moritz H., 1967. Physical Geodesy. W.H. Free-
man and Co., San Francisco and London.
Iz H. Bâki., 2009. New parameters of geometrically best fitting lunar
figures. Journal of Applied Geodesy, 3, 155-162.
Jacobi C.G.J., 1839. Note von der geodätischen Linie auf einem El-
lipsoid und den verschiedenen Anwendungen einer merk-
würdigen analytischen Substitution. Journal für die Reine und
Angewandte Mathematik, 1839, 309-313.