Geodesics∗

(COMS 4770/5770 Notes)

Yan-Bin Jia

Dec 12, 2024

Geodesics are the curves in a surface that make turns just to stay on the surface and never move
sideways. A bug living in the surface and following such a curve would perceive it to be straight.
A geodesic is a generalization of the notion of a “straight line” from a plane to a surface, on which
it represents in some sense the shortest path between two points. We will begin with a definition
of geodesics, then present various method for finding geodesics on surfaces, and later reveal their
relationships to shortest paths.
The term geodesic comes from the science of geodesy, which is concerned with measurements of
the earth’s surface [2, p. 163]. F. W. Bessel (1784–1846) was involved with determining the shape
of the earth as an ellipsoid of rotation. C. G. Jacobi (1804–1851) studied the “shortest curves” on
an ellipsoid of rotation which he referred to as “geodesic curves”. The term “shortest curves” had
earlier been used by Johannes Bernoulli (1667–1748) and Carl-Friedrich Gauss (1777–1855).
Geodesics on a surface are directly linked to shortest paths. This connection to the general
notion of distance has found extensive applications in science, engineering, and art, from classical
mechanics and relativity theory to neurology (e.g., signal propagation) to statistics (e.g., geodesic
analysis) and to architecture and structure design (e.g., domes). In computer vision, geodesics
have been primarily used in image segmentation, shape analysis, and surface parametrization [4]. In
robotics, geodesics have been considered in trajectory planning or coverage tasks such as automobile
painting [1].

1 Definition
A curve γ(t) on a surface S is called a geodesic if at every point γ(t) the acceleration γ̈(t) is either
zero or parallel to its unit normal n̂.

Example 1. Strictly following the definition, a straight line γ(t) = at + b is a geodesic since γ̈(t) = 0.
However, a reparametrization of the line as δ(u) = a tan u + b over u ∈ (− π2 , π2 ) is not a geodesic since
δ̈(u) = (a/ cos2 u)′ = −(2 sin u/ cos3 u)a 6= 0 unless u = 0. The fact is that every line can be parametrized
so that it is a geodesic.

Here is an interesting interpretation of geodesics from mechanics. Suppose a particle is moving

on the surface under a force perpendicular to the surface to maintain contact. Its trajectory would
The material is adapted from the book Elementary Differential Geometry by Andrew Pressley, Springer-Verlag,
∗

2001.

1
be a geodesic, because Newton’s second law states that the particle’s acceleration γ̈ is parallel to
the force, hence always perpendicular to the surface.
Proposition 1 A geodesic γ(t) on a surface S has constant speed.
Proof We have
d d
kγ̇k2 = (γ̇ · γ̇) = 2γ̈ · γ̇.
dt dt
Since γ is geodesic, γ̈ is perpendicular to the tangent plane which contains γ̇. Hence γ̈ · γ̇ = 0.
Subsequently, dkγ̇k2 /dt = 0. Therefore, the speed kγ̇k is constant.
Proposition 1 suggests that the unit-speed parametrization of a geodesic is still a geodesic, since
the acceleration is just scaled by a non-zero constant factor. So we can always consider unit-speed
geodesics only if needed.
Proposition 2 A curve on a surface is a geodesic if and only if its geodesic curvature is zero
everywhere.
Proof Let γ be a unit-speed geodesic in a patch σ of the surface, and N the unit normal of σ.
The geodesic curvature is
κg = γ̈ · (N × γ̇).
If γ̈ is zero, then the above implies that κg = 0. Otherwise, by definition γ̈ is parallel to N . It is
thus perpendicular to N × γ̇, thereby κg = 0.
Conversely, suppose κg = 0. Then γ̈ ⊥ N × γ̇ if γ̈ 6= 0. Meanwhile, from γ̇ × γ̇ = 1 we obtain
that γ̈ ⊥ γ̇. Therefore, γ̈ is parallel to γ̇ × (N × γ̇); in other words, it is parallel to N .
By Proposition 2, we can claim that all straight lines are geodesics. Other simple geodesics
include the rulings of any ruled surface, such as the generators of a (generalized) cylinder or cone.
Suppose a curve on a surface is its intersection with a plane that happens to be perpendicular to
the tangent plane at every point on the curve. Since theq curve lies in a normal plane, its curvature κ
equals the normal curvature κn everywhere. From κ = κ2n + κ2g , we infer that κg = 0 everywhere.

Example 2. A great circle on a sphere is its intersection with a plane Π passing through the center o of
the sphere. Every point p on the great circle defines a vector p − o which is perpendicular to the tangent
plane at p. Thus, the great circle is a geodesic.

Example 3. The intersection of a generalized cylinder with a plane Π perpendicular to the rulings of the
cylinder is a geodesic. Clearly, the unit normal at such an intersection point is perpendicular to the rulings,
and hence contained in Π. So, Π is perpendicular to the tangent plane at the point.

N
Π

2
2 Geodesic Equations
We have seen two examples of geodesics. To determine all the geodesics on a given surface, we need
to solve differential equations stated in the following theorem.

Theorem 3 A curve γ on a surface S is a geodesic if and only if for any part γ(t) = σ(u(t), v(t))
contained in a surface patch σ of S, the following two equations are satisfied:
d 1
(E u̇ + F v̇) = (Eu u̇2 + 2Fu u̇v̇ + Gu v̇ 2 ), (1)
dt 2
d 1
(F u̇ + Gv̇) = (Ev u̇2 + 2Fv u̇v̇ + Gv v̇ 2 ), (2)
dt 2
where Edu2 + 2F dudv + Gdv 2 is the first fundamental form of σ.

Proof The tangent plane is spanned by σu and σv . By definition the curve γ is a geodesic if
and only if γ̈ · σu = γ̈ · σv = 0. Since γ̇ = u̇σu + v̇σv , γ̈ · σu = 0 becomes
 
d
(u̇σu + v̇σv ) · σu = 0.
dt

We rewrite the left hand side of the above equation:

 
d d dσu
(u̇σu + v̇σv ) · σu = ((u̇σu + v̇σv ) · σu ) − (u̇σu + v̇σv ) ·
dt dt dt
d
= (E u̇ + F v̇) − (u̇σu + v̇σv ) · (u̇σuu + v̇σuv )
dt
d 
= (E u̇ + F v̇) − u̇2 (σu · σuu ) + u̇v̇(σu · σuv + σv · σuu )
dt 
+ v̇ 2 (σv · σuv ) . (3)

We have that
1 ∂ 1
σu · σuu = (σu · σu ) = Eu ,
2 ∂u 2
1
σv · σuv = Gu ,
2
σu · σuv + σv · σuu = Fu .

Substituting them into (3), we obtain

 
d d 1
(u̇σu + v̇σv ) · σu = (E u̇ + F v̇) − (Eu u̇2 + 2Fu u̇v̇ + Gu v̇ 2 ).
dt dt 2

This establishes the first differential equation (1). Similarly, equation (2) can be established from
 
d
(u̇σu + v̇σv ) · σv = 0.
dt

3
 The two equations in Theorem 3 are called the geodesic equations. They are nonlinear and
solvable analytically on rare occasions only. The following are two examples where a close-form
solution can be found.

Example 4. Let us find the geodesics on the unit sphere S 2 by solving equations (1) and (2). Consider the
patch under the parametrization
σ(θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ).
In Example 1 of the notes titled “Surface Curves and Fundamental Forms”, the first fundamental form
is found to be dθ2 + cos2 θ dφ2 , with E = 1, F = 0, and G = cos2 θ. We restrict to unit-speed curves
γ(t) = σ(θ(t), φ(t)) so that
E θ̇2 + 2F θ̇φ̇ + Gφ̇2 = θ̇2 + φ̇2 cos2 θ = 1. (4)
If γ is a geodesic, equation (2) is satisfied. Here, the equation reduces to
d
(φ̇ cos2 θ) = 0, (5)
dt
is satisfied. Hence
φ̇ cos2 θ = C, (6)
for some constant C. There are two cases.
• C =0 Then φ̇ = 0. In this case, φ is constant and γ is part of a meridian.
• C 6= 0 Substituting (6) into the unit-speed condition (4), we have
C2
θ̇2 = 1 − .
cos2 θ
Combining the above with (6), along the geodesic it holds that
 2
dφ φ̇2 1
= = cos2 θ
.
dθ θ̇ 2 2
cos θ( C 2 − 1)
Integrate the derivative dφ/dθ:
dθ
Z
φ − φ0 = ± q ,
cos2 θ
cos θ C2 −1

where φ0 is a constant. The substitution u = tan θ yields

 
du u
Z
−1
φ − φ0 = ± √ = sin √ ,
C −2 − 1 − u2 C −2 − 1
which leads to p
tan θ = ± C −2 − 1 · sin(φ − φ0 ).
Multiply both sides of the above equation with cos θ:
p
sin θ = ± C −2 − 1(cos φ0 cos θ sin φ − sin φ0 cos θ cos φ).
Since σ(θ, φ) = (xy, z), we have
p p
z = ∓(sin φ0 C −2 − 1)x ± (cos φ0 C −2 − 1)y.
Clearly, z = 0 when x = y = 0. Therefore, γ is contained in the intersection of S 2 with a plane
through the center of the sphere. Hence it is part of a great circle.
In both cases, γ is part of a great circle.

4
Example 5. We now find the geodesics of the circular cylinder

σ = (a cos φ, a sin φ, z).

First, we obtain the coefficients of the first fundamental form:

E = a2 , F = 0, G = 1.

Substitute these coefficients into (1) and (2):

 
d  a2 φ̇
q  = 0,
dt 2 2 2
a φ̇ + ż
 
d  ż
q  = 0,
dt
a2 φ̇2 + ż 2

Integrate the two equations above:

a2 φ̇
q = C1 ,
a2 φ̇2 + ż 2
ż
q = C2 .
a2 φ̇2 + ż 2

Dividing the last equation by the one before, we obtain

dz
= D1 ,
dφ
which has the solution
z = D1 φ + D2 .
The geodesic between two points on a cylinder thus is a helix lying on the cylinder. Given two points
σ(φ0 , z0 ) and σ(φ1 , z1 ), the helix is described as
 
z0 − z1 φ0 z1 − φ1 z0
α(φ) = a cos φ, a sin φ, φ+ .
φ0 − φ1 φ0 − φ1

The figure on the top right shows a geodesic (cos φ, sin φ, 2φ) connecting two points with φ0 = 0 and
φ1 = 23 on a cylinder (cos φ, sin φ, z) which is plotted over the subdomain [0, 2π] × [−1, 4].

The following theorem states that a unique geodesic exists on a surface that passes through any
of its point in any given tangent direction.1

Theorem 4 Let p be a point on a surface S, and t̂ a unit tangent vector at p. There exists a
unique unit-speed geodesic γ on S which passes through p with velocity γ ′ = t̂.

A trivial example would be straight lines in a plane. At any point, in any direction, there is a
unique straight line through the point. Theorem 4 says that there are no other geodesics. On a
sphere, the great circles are the only geodesics. This is because, in any tangent direction at a point
1
For a proof of the theorem, we refer to [5, p. 178].

5
on the sphere, there is a great circle through the point which is the intersection of the sphere with
the plane determined by the tangent line in that direction and the sphere’s center.
Theorem 4 specifies four initial values needed to solve the geodesic equations (1) and (2) for
u and v as functions of the curve parameter. Suppose we need to construct a geodesic passing
through a point p on a surface patch σ(u, v) in the unit tangent direction t̂, with p = σ(u0 , v0 ).
We consider a unit-speed parametrization γ(t) = σ(u(t), v(t)) such that

u(0) = u0 ,
v(0) = v0 .

The unit tangent at p is σu (u0 , v0 )u̇(0)+ σv (u0 , v0 )v̇(0). Because σu and σv span the tangent plane
at p, the values of the two derivatives u̇(0) and v̇(0) can be determined from the equation

t̂ = σu (u0 , v0 )u̇(0) + σv (u0 , v0 )v̇(0).

One of the three scalar equations in the above equation is redundant since t̂, σu (u0 , v0 ), and
σv (u0 , v0 ) are co-planar. To eliminate the redundancy, we take the dot products of the equation
with σu (u0 , v0 ) and σv (u0 , v0 ) separately, obtaining
    
E F u̇(0) t̂ · σu (u0 , v0 )
= .
F G v̇(0) t̂ · σv (u0 , v0 )

This gives us    
u̇(0) t̂ · σu (u0 , v0 )
= F1−1 ,
v̇(0) t̂ · σv (u0 , v0 )
where the first fundamental form matrix F1 is evaluated at (u0 , v0 ). The four initial values
u(0), v(0), u̇(0), v̇(0) uniquely determine functions u(t) and v(t), and thus a solution curve to the
geodesic equations. In fact, this is the idea behind the proof of Theorem 4.

3 Preservation of Geodesics Under Isometry

Let S1 and S2 be two surfaces. A map f : S1 → S2 is called a diffeomorphism if it is smooth
and bijective and its inverse map f −1 : S2 → S1 is also smooth. A diffeomorphism f is called an
isometry if it maps curves in S1 to curves of the same length in S2 . If an isometry exists, then S1
and S2 are isometric. A sufficient and necessary condition for f to be an isometry is any surface
patch σ1 of S1 and its image patch f ◦ σ1 of S2 have the same first fundamental form. The proof,
given in [5,qp. 102], relies on the fact that the length of a curve γ(t) = σ(u(t), v(t)) on a surface
patch is
R ˙ 2 + 2F u̇v̇ + Gv̇ 2 dt.
E (u)

Example 6. Let S1 be the infinite strip in the xy-plane given by 0 < x < 2π, and S2 be the circular cylinder
x2 + y 2 = 1 with the ruling x = 1, y = 0 removed. Then S1 is covered by a single patch σ1 (u, v) = (u, v, 0),
and S2 by the patch σ2 (u, v) = (cos u, sin u, v), where 0 < u < 2π for both cases. Consider the map
f : (u, v, 0) 7→ (cos u, sin u, v) from S1 to S2 . We can verify that both patches have the same first fundamental
form du2 + dv 2 . Therefore f is an isometry, and S1 and S2 are isometric.

6
 isometric
geodesic geodesic

0 2π ruling 1
removed

Theorem 5 An isometry f : S1 → S2 maps the geodesics of S1 to the geodesics of S2 .

Proof Let γ(t) be a geodesic in S1 . Suppose a part of γ lies in the patch σ(u, v) in S1 so that
γ(t) = σ(u(t), v(t). Then, u and v satisfy the geodesic equations (1) and (2). Now, consider the
patch f ◦ σ of S2 as the image of σ under the isometry. Because both patches have the same first
fundamental form, the curve f (σ(u(t), v(t))) on S2 must also satisfy the geodesic equations. Thus
f ◦ γ is a geodesic on S2 by Theorem 3.
Example 7. We know that on the circular cylinder x2 +y 2 = 1, geodesics include cross sections where planes
parallel to the xy-plane intersect the cylinder, as well as straight lines parallel to the z-axis. To find the
missing geodesics, we make use of the isometry given in Example 5 which maps (u, v, 0) to (cos u, sin u, v).
In the isometric strip in the xy-plane given by 0 < x < 2π, any line not parallel to the y-axis has a
parametrization (u, mu + c). Its image on the cylinder is the curve

γ(u) = (cos u, sin u, mu + c).

This curve is a geodesic on the cylinder by Theorem 5. In the previous figure, a diagonal in the strip is
drawn, along with its image geodesic curve on the isometric cylinder.

4 Geodesics vs. Shortest Paths

We all know that the shortest path between two points p and q in the plane is the line segment
pq. It is almost as well known that great circles are the shortest paths on a sphere. The following
can be established.

Theorem 6 Let γ be a shortest path on a surface S connecting two points p and q. Then the part
of γ contained in any surface patch σ of S must be a geodesic.

The converse of the statement in the theorem is not necessarily true, however. If γ is a geodesic
on σ connecting p and q. Then γ need not be a shortest path between the two points. As an
example, the great circle connecting two points p and q on a sphere is split into two circular arcs
by the points. Both arcs are geodesics. Only the shorter one of the two is the shortest path joining
p and q.
In general, a shortest path joining two points on a surface may not exist. For example, consider
the surface P which is the xy-plane with its origin removed. There is no shortest path from the
point p = (−1, 0) to the point q = (1, 0). Such a path would be the line segment, except it passes

7
 p q

p 0 q

through the origin and does not lie entirely on the surface. Any short path in P from p to q would
walk in a straight line as long as possible, and then move around the origin, and continue in a
straight line. It can always be improved on by moving a little closer to the origin before circling
around it. Therefore, there is no shortest path connecting the two points.
If a surface S is a closed subset2 of R3 , and if there is some path in S joining any two points p
and q, then there always exists a shortest path joining the two points. For instance, a sphere is a
closed subset of R3 , and the short great circle arc joining two points on the sphere is their shortest
path. The surface P , the xy-plane with its origin removed, is not a closed subset of R3 because
any open ball containing the origin must contain points of P (so the set of points not in P is not
open).

5 Geodesic Coordinates
The existence of geodesics on a surface S gives a
way to construct a useful atlas for S in practice. Let
β ′v (0)
p be a point in S, and γ(v) a unit-speed geodesic
p = γ(0) σ(u, v) = β v (u)
γ on S with γ(0) = p. For any value of v, there
γ(v)
γ ′ (v) exists a unique unit-speed geodesic βv (u) such that
βv (0) = γ(v) and βv′ (0) ⊥ γ ′ (v). We define a patch
βv σ(u, v) = βv (u) and call it a geodesic patch, and u
and v geodesic coordinates.

Example 8. Consider a point p on the equator of the unit sphere S 2 . Let the equator be parametrized
with the longitude φ as γ(φ). Let βφ be the meridian parametrized by the latitude θ and passing through
the point on the equator with longitude φ. The corresponding geodesic patch is the usual one in latitude
and longitude, for which the first fundamental form is dθ2 + cos2 θdφ2 .

References
[1] P. N. Atkar, A. Greenfield, D. C. Conner, H. Choset, and A. A. Rizzi. Uniform coverage of
automotive surface patches. International Journal of Robotics Research, 24(11):883-898.

[2] J. McCleary. Geometry from a Differentiable Viewpoint. Cambridge University Press, 1994.
2
In other words, the set R3 − S is open.

8
[3] B. O’Neill. Elementary Differential Geometry. Academic Press, Inc., 1966.

[4] G. Peyré, M. Péchaud, R. Keriven, and L. D. Cohen. Geodesic methods in computer vision and
graphics. Foundations and Trends in Computer Graphics and Vision, 5(3–4):197-397, 2010.

[5] A. Pressley. Elementary Differential Geometry. Springer-Verlag London, 2001.