Ophthal. Physiol. Opt.

2006 26: 497–501

Curvature of ellipsoids and other surfaces

W. F. Harris
Optometric Science Research Group, Department of Optometry, and Department of Mathematics
and Statistics, University of Johannesburg, PO Box 524, Auckland Park, 2006 South Africa

Abstract
From differential geometry one obtains an expression for the curvature in any direction at a point on a
surface. The general theory is outlined. The theory is then specialised for surfaces that are
represented parametrically as height over a transverse plane. The general ellipsoid is treated in detail
as a special case. A quadratic equation gives the principal directions at the point and, hence, the
principal curvatures associated with them. Equations are obtained for ellipsoids in general that are
generalisations of Bennett’s equations for sagittal and tangential curvature of ellipsoids of revolution.
Equations are also presented for the locations of umbilic points on the ellipsoid.

Keywords: curvature, ellipsoid, principal curvature, radius of curvature, sagittal curvature, tangen-
tial curvature, umbilic point

satisfactory: it is not obvious why the sagittal radius of

Introduction
curvature should be given, as Bennett asserts (see his
The anterior surface of the cornea is often represented as Equation 22), by the distance PG in his Figure 13.
ellipsoidal in form (e.g. Kiely et al., 1982; Douthwaite However, even if one accepts the derivation, it is not
et al., 1996; Douthwaite and Pardhan, 1998; Douthwa- obvious how to generalise Bennett’s approach for
ite and Evardson, 2000). In such studies the expressions ellipsoids that are not ellipsoids of revolution.
used for radius of curvature are those derived by The reader unfamiliar with differential geometry
Bennett (1988). However Bennett’s equations apply might consult any of several good texts on the subject
only to ellipsoids of revolution. As many if not most (e.g. Struik, 1957; Lipschutz, 1969; Stoker, 1969; do
corneae are better represented by general ellipsoids, that Carmo, 1976). We begin by summarising the general
is ellipsoids that are not necessarily ellipsoids of revo- approach to calculating curvature at points in a surface.
lution, there would seem to be a need for equations for We specialise the analysis for the common situation in
curvature (or radius of curvature) that hold for ellip- optometry in which the surface is conveniently repre-
soids in general. Accordingly the primary purpose of sented as height above a transverse plane. We then treat
this paper is to derive such equations. A secondary the general ellipsoid as an example. An expression is
purpose is to draw attention to and illustrate the fact obtained for the curvature in any direction through a
that differential geometry provides a general method for point on the ellipsoid. An equation allows the deter-
obtaining expressions for curvature of surfaces and, mination of the principal meridians at the point and,
hence, that it is not necessary to analyse each new hence, the principal curvatures. Bennett’s equations are
surface from scratch. obtained as the special case for ellipsoids of revolution.
Although Bennett’s (1988) equations (Equations 55 Points occur on an ellipsoid at which the curvature is
and 56 below) turn out to be quite correct for ellipsoids independent of direction through the point; they are
of revolution his derivation seems to me not entirely called ÔumbilicsÕ. (In optometric terms an umbilic would
be stigmatic or spherical in optical character.) The
positions of the umbilics on the ellipsoid are derived.
Received: 4 May 2005
Revised form: 29 August 2005
Accepted: 30 September 2005 General surfaces

Correspondence and reprint requests to: W.F. Harris.

A point in space can be represented by means of its
Tel.: +27 11 489 2785; Fax: +27 11 489 2091 position vector relative to the origin of an orthogonal
E-mail address: wfh@na.rau.ac.za coordinate system:

ª 2006 The College of Optometrists 497

498 Ophthal. Physiol. Opt. 2006 26: No. 5

x ¼ ðx1 ; x2 ; x3 Þ: ð1Þ Edu2 þ 2Fdudv þ Gdv2 : ð9Þ

Suppose each of the coordinates is a function of two We define the cross product
parameters u and v. Then one has
@x @x
x ¼ ðx1 ðu; vÞ; x2 ðu; vÞ; x3 ðu; vÞÞ ð2Þ n¼  ð10Þ
@u @v
or of the two tangent vectors at P. It is a vector normal to
the surface at P. Its length is
x ¼ xðu; vÞ: ð3Þ
 
@x @x
Equation (3) may be a parametric equation of a surface n ¼   
 ð11Þ
(for example a sphere with u and v as longitude and @u @v
latitude respectively). The curvature of the surface can and
be determined from the first and second derivatives of x
with respect to u and v and then via what are called the ^n ¼ n=n ð12Þ
first and second fundamental forms of the surface. It is is a unit normal vector at P. One can express the Ôsecond
to these concepts that we now turn. fundamental formÕ of the surface at P as
Let us fix v at v0. Then Equation (3) becomes
@2x @2x @2x
x ¼ xðu; v0 Þ ð4Þ  ^
n du2
þ 2  ^
n dudv þ  ^ndv2 ð13Þ
@u2 @u@v @v2
the equation of a curve in the surface (a line of latitude or
in the example of the sphere). The curve is the
u-parameter curve with v ¼ v0; u varies along the curve. Ldu2 þ 2Mdudv þ N dv2 ð14Þ
Similarly x ¼ x(u0, v) defines the v-parameter curve in
where L, M and N are the coefficients in Equation (13)
the surface with u fixed at u ¼ u0 (a line of longitude on
(the Ôsecond fundamental coefficientsÕ).
the sphere). The u-parameter curve and the v-parameter
It turns out (see, Struik, 1957, p. 80, for example) that
curve intersect in point P on the surface; the point has
the curvature of the surface at point P in the direction
position vector x(u0, v0). The derivative
PQ is the ratio
@x @x
¼ ðu0 ; v0 Þ ð5Þ Ldu2 þ 2Mdudv þ N dv2
@u @u j¼ ð15Þ
Edu2 þ 2Fdudv þ Gdv2
is a vector tangent to the u-parameter curve at P.
Similarly ¶x/¶v is a vector tangent to the v-parameter of the two fundamental forms at P. Angle h is defined
curve at P. The three second derivates at point P are such that
¶2x/¶u2, ¶2x/¶u¶v and ¶2x/¶v2. tan h ¼ dv=du: ð16Þ
Consider a second and distinct point Q that is on the
surface and near to P; it has position vector This is interpreted to mean that )p/2 £ h £ p/2 for
x(u0 + du, v0 + dv). du and dv are two differentials du „ 0 and h ¼ p/2 for du ¼ 0. Then Equation (15)
which approach 0 but which are not both exactly zero. can be written as

dx ¼ xðu0 þ du; v0 þ dvÞ  xðu0 ; v0 Þ ð6Þ L cos2 h þ 2M sin h sin h þ N sin2 h

j¼ : ð17Þ
E cos2 h þ 2F sin h sin h þ G sin2 h
is a differential vector from P to Q which can be written
h represents a direction in the surface through point P.
@x @x
dx ¼ du þ dv: ð7Þ (We shall give a geometrical interpretation in the special
@u @v case of the ellipsoid treated below.)
The dot product At a point on a surface the curvature usually
depends on direction h. The extrema of curvature j
@x @x 2 @x @x @x @x 2
dx  dx ¼  du þ 2  dudv þ  dv over h are the two principal curvatures at the point
@u @u @u @v @v @v and the corresponding angles h define the two
ð8Þ corresponding principal directions. The principal
is called the Ôfirst fundamental formÕ of the surface at the directions are orthogonal and given by the solutions
point P. (du2 is interpreted as (du)2.) Abbreviating the to the equation
coefficients in Equation (8) as E, F and G (the Ôfirst
ðEM  FLÞ cos2 h þ ðEN  GLÞ sin h cos h
fundamental coefficientsÕ) one writes the first funda-
mental form as þ ðFN  GMÞ sin2 h ¼ 0: ð18Þ

ª 2006 The College of Optometrists

 Curvature of ellipsoids and other surfaces: W. F. Harris 499

For h „ 0 this can be written as a quadratic equation

in tan h (for example, Struik, 1957, p. 80). An umbilic is
a point in the surface at which the curvature is
independent of h.

Surfaces over a plane

In many optometric applications it is convenient to
represent the surface in question in terms of the height
of the surface over a reference plane. Let
Figure 1. An ellipsoid whose principal diameters have lengths 2a,
y ¼ ðy1 ; y2 Þ ð19Þ 2b1 and 2b2 and are along the right-handed orthogonal set of axes x,
y1 and y2. Axis x is perpendicular to the plane of the paper in panel a
represent a point in the reference plane with rectangular
and points towards the reader. In panel b axis y1 is perpendicular to
coordinates y1 and y2. We shall call y the transverse the plane of the paper and points away from the reader. Only that
position of the point. The height x of the surface over half of the ellipsoid with x ‡ 0 is shown in panel b. y1 and y2 are the
the point is represented by the function coordinates of a point in the plane of the paper with transverse
position vector y. h represents a direction through the point.
x ¼ xðy1 ; y2 Þ: ð20Þ Corresponding to the point with position vector y is a point P on
Now x, y1 and y2 can be regarded as the coordinates of the ellipsoid with position vector x whose coordinates are x, y1 and
y2. One can think of x as the height of the ellipsoid over the point with
the three-dimensional vector
position vector y. The two small circles in panels a and b represent
x ¼ ðx; y1 ; y2 Þ: ð21Þ umbilic points on the ellipsoid. Their positions are given by Equation
(54). One might be looking at a cornea from the front in panel a and
Equation (20) is Equation (1) but with symbols changed. from the right-hand side in panel b.
Substituting from Equation (19) we have
x ¼ ðxðy1 ; y2 Þ; y1 ; y2 Þ: ð22Þ a special case of Equation (20). Substituting into
Equation (21) we obtain a special case of Equation
Equation (22) is a special case of Equation (2), the (22), namely
second and third functions on the right of the latter sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
having reduced to the parameters y1 and y2 themselves. y12 y22
Thus to calculate the curvature of the surface one x ¼ a 1  2  2 ; y1 ; y2 : ð25Þ
b1 b2
applies the general procedure described above to this
special case, as we now proceed to do for the ellipsoid in Equation (27) is a parametric representation of the
particular. half ellipsoid. Calculation of curvature is now routine.
We begin by calculating the first and second deriva-
Ellipsoids tives of x with respect to the two parameters y1 and
y2:
Figure 1 shows half an ellipsoid. Its principal diameters  2 
have lengths 2a, 2b1 and 2b2. In Figure 1a and b one is @x a y1
¼  2 ; 1; 0 ; ð26Þ
looking along the principal diameters with lengths 2a @y1 b1 x
and 2b1 respectively. The axes of a right-handed
orthogonal coordinate system lie along the three prin-  2 
@x a y2
cipal diameters. The ellipsoid can be represented by the ¼  2 ; 0; 1 ; ð27Þ
@y2 b2 x
equation
   
x2 y12 y22 @2x a2 a2 y12
þ þ ¼ 1: ð23Þ ¼  1 þ ; 0; 0 ; ð28Þ
a2 b21 b22 @y12 b21 x b21 x2
Solving for x we obtain  4 
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi @2x a y1 y2
¼  2 2 3 ; 0; 0 ; ð29Þ
y2 y2 @y1 @y2 b1 b2 x
x ¼ a 1  12  22 ð24Þ
b1 b2    
@2x a2 a2 y22
where we have taken only the positive square root. This ¼  1 þ ; 0; 0 : ð30Þ
@y22 b22 x b22 x2
implies the half ellipsoid shown explicitly in Figure 1b. x
is defined only for y12 =b21 þ y22 =b22  1. Equation (24) is Then, from Equation (10) the normal vector is,

ª 2006 The College of Optometrists

500 Ophthal. Physiol. Opt. 2006 26: No. 5

 
 a2 y1  ðN  GLÞ sin h cos h ¼ 0: ð41Þ
 2 1 0
@x @x  b21 x 
 Thus, as expected from symmetry, for N ) GL „ 0, the
n¼  ¼ a y  ð31Þ
@y1 @y2   2 2 0 1 principal directions are at h ¼ 0 and p/2 and the
 b2 x 
  corresponding principal curvatures are
i j k
jðð0; y2 Þ; 0Þ ¼ L; ð42Þ
where i, j and k represent unit vectors in the positive
senses along the coordinate axes x, y1 and y2 respect-
ively. One obtains jðð0; y2 Þ; p=2Þ ¼ N =G: ð43Þ
 2  Substituting for the coefficients, making use of Equa-
a y1 a2 y2
n ¼ 1; 2 ; 2 : ð32Þ tions (24) and (33) for x and n respectively, and
b1 x b2 x
simplifying one obtains
The length of vector n is (Equation 11) a
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi jðð0; y2 Þ; 0Þ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  ; ð44Þ
a4 y 2 a4 y 2 b21 1 þ a2 =b22  1 y22 =b22
n ¼ 1 þ 4 12 þ 4 22 : ð33Þ
b1 x b2 x
a
jðð0; y2 Þ; p=2Þ ¼    3=2 : ð45Þ
^.
Equation (12) then gives the unit normal vector n b22 1 þ a =b2  1 y22 =b22
2 2
Dot products of the vectors of Equations (26) and
(27) give the coefficients of the first fundamental form In particular, at the apex of the ellipsoid, that is, at
according to Equations (8) and (9): y ¼ o where o is a null vector, the curvatures are
a4 y12 jðo; 0Þ ¼ a=b21 ; ð46Þ
E¼ þ 1; ð34Þ
b41 x2
jðo; p=2Þ ¼ a=b22 : ð47Þ
4
a y1 y2
F ¼ ; ð35Þ Equations (44) and (45) can be written as
b21 b22 x2
1
jðð0; y2 Þ; 0Þ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð48Þ
a4 y22 1=j2 ðo; 0Þ þ ðp1 =p2 Þ2 ð1  p2 Þy22
G¼ þ 1: ð36Þ
b42 x2
Dot products of the vectors of Equations (28) to (30) p1 j3 ðð0; y2 Þ; 0Þ
jðð0; y2 Þ; p=2Þ ¼ ð49Þ
with ^
n give the coefficients of the second fundamental p2 j2 ðo; 0Þ
form according to Equations (13) and (14):
where
 
a4 b22  y22
L¼ ; ð37Þ p1 ¼ b21 =a2 ð50Þ
b21 b22 nx3
and similarly for p2. The radius of curvature q at y in
4
a y1 y2 direction h is
M¼ ; ð38Þ
b21 b22 nx3 q ¼ qðy; hÞ ¼ 1=j: ð51Þ
  Equations (48) and (49) lead to
a4 b21  y12 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
N¼ : ð39Þ
b21 b22 nx3 qðð0; y2 Þ; 0Þ ¼ q2 ðo; 0Þ þ ðp1 =p2 Þ2 ð1  p2 Þy22 ; ð52Þ
For a point P on the ellipsoid with transverse position y
(Equation 19), Equation (17) then gives the curvature p2 q3 ðð0; y2 Þ; 0Þ
qðð0; y2 Þ; p=2Þ ¼ : ð53Þ
j(y, h) in the direction h (see Figure 1a). p1 q2 ðo; 0Þ
In the special case of points on the ellipsoid with
y1 ¼ 0 the fundamental coefficients simplify and one At a point where N ) GL ¼ 0 Equation (41) is
obtains the curvature satisfied for all h. Curvature there is independent of
L cos2 h þ N sin2 h direction h and the point is an umbilic. To find the
jðð0; y2 Þ; hÞ ¼ : ð40Þ umbilics we can equate the right-hand sides of Equa-
cos2 h þ G sin2 h
tions (44) and (45) and solve. We find them in the plane
Furthermore, Equation (18) reduces to y1 ¼ 0 at a point with y2 that satisfies

ª 2006 The College of Optometrists

 Curvature of ellipsoids and other surfaces: W. F. Harris 501

an ellipsoid of revolution the equations reduce to

y22 b21  b22 Bennett’s equations for tangential and sagittal radius
¼ : ð54Þ
b22 a2  b22 of curvature. We are provided, therefore, with a formal
proof of his equations. Equations (52) and (53) gener-
For an ellipsoid of revolution b1 ¼ b2; and Equation
alise Bennett’s equations to general ellipsoids. The
(54) shows that the apex is an umbilic in agreement with
locations of umbilic points on the ellipsoid are given
what we already know. For other ellipsoids, in the plane
by Equation (54).
y1 ¼ 0, there are no (real) umbilics if the right-hand side
of Equation (54) is negative and two umbilics in the half
ellipsoid, equidistant from the apex, if the right-hand Acknowledgement
side is positive. Figure 1 is drawn for the case
This note is based on research supported under Grant
a ‡ b1 ‡ b2. The umbilics are shown by means of small
No. 2053699 to W.F. Harris from the National
circles.
Research Foundation. I thank W.A. Douthwaite for
For the special case of an ellipsoid of revolution
communications and W.D.H. Gillan for continuing
p1 ¼ p2; Equations (52) and (53) reduce to Bennett’s
discussions.
Equations (24) and (26), that is,
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rS ¼ r02 þ ð1  pÞy 2 ð55Þ References

and Bennett, A. G. (1988) Aspherical and continuous curve contact

lenses. Part Three. Optom. Today 28, 238–242.
rT ¼ rS3 =r02 ð56Þ do Carmo, M. P. (1976) Differential Geometry of Curves and
Surfaces. Prentice-Hall, Englewood Cliffs, NJ.
for the sagittal and tangential radii of curvature Douthwaite, W. A. and Evardson, W. T. (2000) Corneal
respectively. topography by keratometry. Br. J. Ophthalmol. 84, 842–847.
Douthwaite, W. A. and Pardhan, S. (1998) Surface tilt
measured with the EyeSys Videokeratoscope: influence on
Concluding remarks corneal asymmetry. Invest. Ophthalmol. Vis. Sci. 39, 1727–
From differential geometry one obtains an expression 1737.
Douthwaite, W. A., Pardhan, S. and Burek, H. (1996) Extent
(Equation 17) for the curvature in any direction at a
and effect of surface tilt on the data display of the EyeSys
point on a surface. The principal directions, and, hence,
videokeratoscope. Br. J. Ophthalmol. 80, 986–983.
the principal curvatures, at the point are given via Kiely, P. M., Smith, G. and Carney, L. G. (1982) The mean
Equation (18). The analysis simplifies if the surface can shape of the human cornea. Optica Acta 29, 1027–1040.
be represented parametrically in terms of transverse Lipschutz, M. M. (1969) Schaum’s Outline of Theory and
position (Equation 22). For ellipsoids, in particular, the Problems of Differential Geometry. McGraw-Hill, New
curvature in any direction at any point is given by York, NY.
Equation (17) with E, F, G, L, M and N given by Stoker, J. J. (1969) Differential Geometry. Wiley, New York,
Equations (34) to (39). x and n in the latter set of NY.
equations can be calculated by means of Equations (24) Struik, D. J. (1957) Lectures on Classical Differential Geom-
and (33) respectively. Equation (40) gives the curvature etry. Addison-Wesley, Reading, MA.
in a principal plane of the ellipsoid. In the special case of

ª 2006 The College of Optometrists