The Curvature and Geodesics of the Torus

http://www.rdrop.com/~half/math/torus/index.xhtml

Mark L. Irons (half@proaxis.com)

The torus is a standard example in introductory discussions of the curvature of surfaces.

However, calculation of some measures of its curvature are hard to find in the literature.
This paper offers full calculation of the torus’s shape operator, Riemann tensor, and
related tensorial objects. In addition, we examine the torus’s geodesics by comparing a
solution of the geodesic equation with results obtained from the Clairaut parameter-
ization. Families of geodesics are classified. Open questions are considered. The
connection form and parallel transport on the torus are investigated in an appendix.

1. The Line Element and Metric

Our model of a torus has major radius c and minor radius a. We only consider the ring torus, for
which c>a.

We use a u,v coördinate system for which planes of constant u pass through the torus’s axis.

 x = (c + a cos v ) cos u 
 
We parameterize the surface x by x(u, v ) =  y = (c + a cos v ) sin u .
 z = a sin v 
 

Mark L. Irons 17 November 2005 1

THE CURVATURE AND GEODESICS OF THE TORUS

We begin by calculating the coefficients E, F, and G of the first fundamental form.

x u = (−(c + a cos v ) sin u,(c + a cos v ) cos u,0 )

x v = (−a cos u sin v, − a sin u sin v,a cos v )
E = xu $ xu
= (−(c + a cos v ) sin u ) + ((c + a cos v ) cos u ) + 0
2 2

= (c + a cos v )
2

F = xu $ xv
= (−(c + a cos v ) sin u )(−a sin v cos u ) + ((c + a cos v ) cos u )(−a sin v sin u ) + (0 )(a cos v )
= a sin v cos u sin u(c + a cos v ) − a sin v cos u sin u(c + a cos v )
=0
G = xv $ xv
= (−a sin v cos u ) + (−a sin v sin u ) + (a cos v )
2 2 2

= a 2 sin 2 v cos 2 u + a 2 sin 2 v sin 2 u + a 2 cos 2 v

= a 2 sin 2 v + a 2 cos 2 v
= a2

This gives us the line element ds2 = (c + a cos v) 2du2 + a2 dv2 and metric:
1
(c + a cos v ) 2 0 (c+a cos v ) 2
0
g ij = , g ij = 1 .
0 a2 0 a2

For later computations we’ll need the partial derivatives of the metric:

0 0 −2a sin v(c + a cos v ) 0

g ij,u = , and g ij,v =
0 0 0 0

2. The Shape Operator

The normal to the surface is N = (cos u cos v, sin u cos v, sin v ). Taking the partial derivatives of N
with respect to u and v gives the shape operator in those directions:

− S(x u ) = N u = (− sin u cos v, cos u cos v,0 )

− S(x v ) = N v = (− cos u sin v, − sin u sin v, cos v )

Mark L. Irons 17 November 2005 2

 THE CURVATURE AND GEODESICS OF THE TORUS

Comparing these to xu and xv, the partial derivatives of the parameterization x, we find that they
are multiples:

S(x u ) = − c +cos v
a cos v x u
S(x v ) = − 1
a xv

The Gaussian curvature K is the determinant of S, and the mean curvature H is the trace of S.

− c+acoscosv v 0
K= = cos v
0 − 1a a(c+a cos v )

H = 12 (− c+acoscosv v − 1a ) = 1
2 − a(c+a
a cos v
cos v )
− c+a cos v
a(c+a cos v )
= c+2a cos v
2a(c+a cos v )

3. The Curvature Tensor

The Christoffel symbols of the second kind

 uuu = 12 [g uu (g uu,u + g uu,u − g uu,u ) + g uv (g vu,u + g vu,u − g uu,v )]

= 1 [g uu (0 + 0 − 0 ) + 0(g vu,u + g vu,u − g uu,v )]

2
=0

 uuv = 12 [g uu (g uv,u + g uu,v − g uv,u ) + g uv (g vv,u + g vu,v − g uv,v )]

= 1 [g uu (0 + g uu,v − 0 ) + 0(g vv,u + g vu,v − g uv,v )]

2
= 1 g uu g uu,v
2
−2a sin v(c + a cos v )
= 1
2 (c + a cos v ) 2

=− a sin v
(c + a cos v )

 uvu = 12 [g uu (g uu,v + g uv,u − g vu,u ) + g uv (g vu,v + g vv,u − g vu,v )]

= 1 [g uu (g uu,v + 0 − 0 ) + 0(g vu,v + g vv,u − g vu,v )] = 1 g uu g uu,v =  uuv

2 2
=− a sin v
(c + a cos v )

 uvv = 12 [g uu (g uv,v + g uv,v − g vv,u ) + g uv (g vv,v + g vv,v − g vv,v )]

= 1 [g uu (0 + 0 − 0 ) + 0(g vv,v + g vv,v − g vv,v )]
2
=0

Mark L. Irons 17 November 2005 3

THE CURVATURE AND GEODESICS OF THE TORUS

 vuu = 12 [0(g uu,u + g uu,u − g uu,u ) + g vv (0 + 0 − g uu,v )]

= − 1 g vv g uu,v
2
= − 1 12 (−2a sin v(c + a cos v ))
2 a
= 1
a sin v(c + a cos v )

 vuv = 12 [g vu (g uv,u + g uu,v − g uv,u ) + g vv (g vv,u + g vu,v − g uv,v )]

= 1 [0(g uv,u + g uu,v − g uv,u ) + g vv (0 + 0 − 0 )]

2
=0

 vvu = 12 [g vu (g uu,v + g uv,u − g vu,u ) + g vv (g vu,v + g vv,u − g vu,v )]

= 1 [0(g uu,v + g uv,u − g vu,u ) + g vv (0 + 0 − 0 )]

2
=0

 vvv = 12 [g vu (g uv,v + g uv,v − g vv,u ) + g vv (g vv,v + g vv,v − g vv,v )]

= 1 [0(g uv,v + g uv,v − g vv,u ) + g vv (0 + 0 − 0 )]

2
=0

Partial derivatives of the nonzero Christoffel symbols:

 uuv,v =  uvu,v = − (a sin v )(−1 )(c + a cos v ) −2 (−a sin v ) + (c + a cos v ) −1 (a cos v )
= −(a sin v ) 2 (c + a cos v ) −2 − (c + a cos v ) −1 (a cos v )
(a sin v ) 2
=− − a cos v
(c + a cos v ) 2 (c + a cos v )

 vuu,v = 1a [sin v(−a sin v ) + (c + a cos v ) cos v ] = 1a [c cos v + a cos 2 v − a sin 2 v ]

The Riemann tensor

Throughout this section we use the identity R ijkl = −Rijlk .

R uuuu =  uuu,u −  uuu,u −  uuu  uuu −  uvu  vuu +  uuu  uuu +  uvu  vuu
= 0 − 0 − 0 −  uvu % 0 + 0 + 0 = 0

R uuuv = −R uuvu =  uuv,u −  uuu,v −  uuv  uuu −  uvv  vuu +  uuu  uuv +  uvu  vuv
= 0−0−0−0+0+0 = 0

R uuvv =  uuv,v −  uuv,v −  uuv  uuv −  uvv  vuv +  uuv  uuv +  uvv  vuv

=  uuv,v −  uuv,v − ( uuv ) 2 − 0 + ( uuv ) 2 + 0 = 0

Mark L. Irons 17 November 2005 4

 THE CURVATURE AND GEODESICS OF THE TORUS

R uvuu =  uvu,u −  uvu,u −  uuu  uvu −  uvu  vvu +  uuu  uvu +  uvu  vvu
=0−0−0−0+0+0 =0

R uvuv = −R uvvu =  uvv,u −  uvu,v −  uuv  uvu −  uvv  vvu +  uuu  uvv +  uvu  vvv

= 0 −  uvu,v − ( uvu ) − 0 + 0 − 0
2

2 2
= − − c +a asin v
cos v − c +a cos v a sin v
a cos v − c + a cos v
2 2
= c +a asin v
cos v + c +a cos v a sin v
a cos v − c + a cos v
= c +a cos v
a cos v
R uvvv =  uvv,v −  uvv,v −  uuv  uvv −  uvv  vvv +  uuv  uvv +  uvv  vvv
= 0−0−0−0+0+0 = 0

R vuuu =  vuu,u −  vuu,u −  vuu  uuu −  vvu  vuu +  vuu  uuu +  vvu  vuu
=0−0−0−0+0+0 =0

R vuuv = −R vuvu =  vuv,u −  vuu,v −  vuv  uuu −  vvv  vuu +  vuu  uuv +  vvu  vuv
= 0 −  vuu,v − 0 − 0 +  vuu  uuv + 0

=−1 1 a sin v
a (sin v(−a sin v ) + cos v(c + a cos v )) + a sin v(c + a cos v ) − c + a cos v

=−1
a [−a sin v + cos v(c + a cos v ) + a sin v ]
2 2

=−1
a cos v(c + a cos v )
R vuvv =  vuv,v −  vuv,v −  vuv  uuv −  vvv  vuv +  vuv  uuv +  vvv  vuv
=0−0−0−0+0+0 =0

R vvuu =  vvu,u −  vvu,u −  vuu  uvu −  vvu  vvu +  vuu  uvu +  vvu  vvu
= 0 − 0 −  vuu  uvu − 0 +  vuu  uvu + 0 = 0

R vvuv = −R vvvu =  vvv,u −  vvu,v −  vuv  uvu −  vvv  vvu +  vuu  uvv +  vvu  vvv
= 0−0−0−0+0+0 = 0

R vvvv =  vvv,v −  vvv,v −  vuv  uvv −  vvv  vvv +  vuv  uvv +  vvv  vvv
= 0−0−0−0+0+0 = 0

The Ricci tensor

R ij = R mimj

R uu = R mumu = 1a cos v(c + a cos v )

R vv = R mvmv = c +a cos v
a cos v
1
a cos v(c + a cos v ) 0
R ij = a cos v
0 (c+a cos v )

Mark L. Irons 17 November 2005 5

THE CURVATURE AND GEODESICS OF THE TORUS

The Ricci scalar, a.k.a. the curvature scalar

R = g ij R ij = g uu R uu + g vv R vv

= 1 1 cos v(c + a cos v ) + 1 a cos v

(c + a cos v ) 2 a a2 c + a cos v

= cos v + cos v
a(c + a cos v ) a(c + a cos v )

R= 2 cos v
a(c + a cos v )

R is twice the Gaussian curvature, as expected.

4. The Geodesic Equation

Let’s look at the geodesic equation ẍ a +  abc x. b x. c = 0. Plugging in Christoffel symbols and
components of the Riemann tensor yields two equations.

(i ) ü + 2 uuv u. v. = ü − c+a
2a sin v . .
cos v uv = 0
(ii ) v̈ +  vuu u. 2 = v̈ + 1a sin v(c + a cos v )u. 2 = 0

To solve (i), let w = c + a cos v. Divide (i) by u. and integrate:

. .
w = −a (sin v )v
. .
− 2w = 2a (sin v )v
2 w.
¶ 1. ü = ¶ − w
u
.
ln u = −2 ln w + ln k
= ln w −2 + ln k
= ln(kw −2 )
.
u = kw −2

= k2
w
= k
(c + a cos v ) 2

Mark L. Irons 17 November 2005 6

 THE CURVATURE AND GEODESICS OF THE TORUS

.
To solve (ii) multiply by v and integrate, using the same w = c + a cos v substitution:

. . 2.
v̈v + 1
a sin v(c + a cos v )(u ) v = 0
.
v̈v + 1a sin v(c + a cos v ) k2 .
v=0
(c + a cos v ) 4
. 2 .
v̈v + ka 1 sin(v )v = 0
(c + a cos v ) 3
(sin v )v. = − 1
aw
.

. 2 .
v̈v + ka 13 − 1 a w=0
w
. 2 .
v̈v = k 2 13 w
a w
. 2 .
¶ v̈v = k 2 ¶ 13 w
a w
1 v. 2 = − k2 1
2 + 2l
2 2a (c + a cos v )
2

.
v2 = − k2
2 +l
a 2 (c + a cos v )

Which yields
. k
u=
(c + a cos v ) 2
.
v=! − k2
2 +l
a 2 (c + a cos v )

This is the general solution to the geodesic equation. To find actual geodesics, we must find a
. .
unit speed parameterization of the curve defined by u, v.

But first we check the solution. For convenience in the checks, we compute ü and v̈ :

ü = 2ka sin v v.
(c + a cos v ) 3

Mark L. Irons 17 November 2005 7

THE CURVATURE AND GEODESICS OF THE TORUS

1
. 2
v = l − k 2 (c + a cos v )
−2 2

a
2 − 12 2
v̈ = 1 l − k 2 (c + a cos v )
−2
− k 2 (−2 )(c + a cos v )
−3
(−a sin v )v.
2 a a
2 − 12 2
1
2
v̈ = 1 l − k 2 (c + a cos v ) l − k 2 (c + a cos v ) − k 2 (−2 )(c + a cos v )
−2 −2 2
−3
(−a sin v )
2 a a a
2
v̈ = 1 − k 2 (−2 )(c + a cos v ) (−a sin v )
−3
2 a
v̈ = − k 2 sin v
a(c + a cos v )
3

i. Check of (i):
2ka sin v v. − 2a sin v k .
2v =
(c + a cos v ) 3
(c + a cos v ) (c + a cos v )
2ka sin v v. − 2ka sin v v. = 0
(c + a cos v ) 3 (c + a cos v ) 3
„

ii. Check of (ii)

2
k 2 sin v sin v(c + a cos v ) k
− 3 + =
a(c + a cos v ) a (c + a cos v ) 2
k 2 sin v k 2 sin v(c + a cos v )
− + =
a(c + a cos v ) a(c + a cos v )
3 4

− k 2 sin v
3 +
k 2 sin v
3 =0 „
a(c + a cos v ) a(c + a cos v )

The unclear geometric role of k and l

One problem with this solution to the geodesic equation is that we have two constants of
integration, k and l, yet given a point on a surface a geodesic’s path is determined by only one
extra parameter, its direction. It’s unclear from this solution precisely how k and l encode this
information. This makes the solution to the geodesic equation useless for determining the paths
of geodesics.

No unit speed parameterization

For a curve  to be a geodesic, it must be a unit speed curve (æa.(t ) æ = 1). Unfortunately, that
pesky constant of integration l makes a general solution to this problem difficult.

Mark L. Irons 17 November 2005 8

 THE CURVATURE AND GEODESICS OF THE TORUS

æ. (t ) æ = u. 2 + v. 2

= k2 k2
4 − 2 +l
(c + a cos v(t )) a 2 (c + a cos v(t ))

2
If l were 4ak 4 , we could complete the square under the radical and integrate. However, we can’t
make that assumption. The meridian geodesics defined by k = 0, l = 1 are a counterexample.

Since we’re not making much headway here, let’s see whether the Clairaut parameterization
helps.

5. The Clairaut Parameterization

Unfortunately, the parameterization we initially chose is the reverse of what is normally used for
a Clairaut parameterization, so in this section the roles of E and G are reversed.

x = (c + a cos v ) cos u
Recall our parameterization of the torus: y = (c + a cos v ) sin u , and the first fundamental form
z = a sin v
E = (c + a cos v ) , F = 0, G = a .
2 2

From O’Neill §7.5.5, a geodesic  can be parameterized as (v ) = x(u(v ), v ) where

du = ! h G
dv E E − h2

=! ah
(c + a cos v ) (c + a cos v ) 2 − h 2

If we could integrate this, we’d have a nice formula for u in terms of v:

u(v) = u(v0 ) ! ah ¶v0

v dv
(c+acosv) (c+acosv) 2 −h2

Alas, this integral likely has no closed form solution. But the formula for du
dv is nice: it depends
on only one parameter, h, the geodesic’s slant. Following O’Neill §7.5.3, if  = x(a 1 , a 2 ) is a
unit-speed geodesic and $ the angle from x v to  , there is a constant h such that
∏

h = E(a 1 )a 2 ∏

= E (a 1 ) sin $
= (c + a cos a 1 ) sin $

Hence h measures of the angle between the geodesic and the x v curves.

Mark L. Irons 17 November 2005 9

THE CURVATURE AND GEODESICS OF THE TORUS

The possible values of h gives us an idea of the different kinds of geodesics that exist on the
torus. The term under the radical must be real, hence (c + a cos v ) 2 m h 2 d h [ c + a. This allows
us to classify the possible geodesics into several families. (We’ll only consider positive values of
h; negative values yield mirror image geodesics.) Note that technically we’re considering
pregeodesics here: to make them true geodesics, we’d need to find unit-speed parameterizations.

h=0d du
dv = 0. These are the meridians:

An intuitive way to see that meridians are geodesics is to realize that the torus has a mirror
symmetry through meridians. Anything that would push the geodesic off a meridian in one
direction is balanced on the opposite side, so a geodesic that starts on a meridian cannot leave it.
A similar argument can be made for both the inner and outer equators, which means they
must be geodesics as well.

0 < h < c − a. These geodesics cross both the inner and outer equators. We call these geodesics
unbounded, because they can pass through all points on the surface.

A consequence of the Clairaut relation h = (c + a cos a 1 ) sin $ is that these geodesics cross the
inner and outer equators at different angles. Note how the slant of the illustrated geodesic varies
with v. A second consequence is that as h increases, geodesics will approach tangency to the
inner equator faster than to outer equator.

Mark L. Irons 17 November 2005 10

 THE CURVATURE AND GEODESICS OF THE TORUS

h = c − a. As h approaches this value from below, the angle a geodesic makes with the inner
equator approaches zero. Hence when h = c − a, one geodesic is the inner equator.

What of the geodesics with h = c − a which pass through other points on the torus? They’re
similar to the unbounded geodesics, but are asymptotic to the inner equator. (Our diagrams don’t
have enough resolution to show that these geodesics circle the inner equator endlessly without
touching it.)

These “asymptotic” geodesics are an edge case of the next family of geodesics. This geodesic is
unique barring rotation about the z axis and reflection through the xy plane.

c − a < h < c + a. Another consequence of the Clairaut relation is that a geodesic  cannot leave
the region E m h 2 . For unbounded geodesics this restriction has no impact, but it does when
c − a < h < c + a. If $ = 2 for some v 0 ,  is tangent to the v 0 parallel. In that case,

(c + a cos v ) 2 m (c + a cos v 0 ) 2
c + a cos v m c + a cos v 0
cos v m cos v 0

i.e.,  is confined to the outer part of the torus between the v 0 and −v 0 parallels (the geodesic’s
barrier curves). We call these geodesics bounded. Here is one of the simplest bounded

Mark L. Irons 17 November 2005 11

THE CURVATURE AND GEODESICS OF THE TORUS

geodesics, which touches each barrier curve once:

As v 0 d , the region between the barrier curves grows to encompass the entire torus.

h = c + a. As h approaches c + a, the barrier curves approach the outer equator. Hence the one
geodesic for which h = c + a is the outer equator.

Summary

We can characterize all geodesics in terms of the absolute value of their slant h:

h Geodesics
0 Meridians
0 < h <c−a Alternately cross both equators ("unbounded" geodesics)
c−a The inner equator, and geodesics asymptotic to it
c−a < h <c+a Cross outer equator but not inner equator (“bounded” geodesics)
c+a The outer equator

For h > c + a, there are no real solutions to the geodesic equation.

Mark L. Irons 17 November 2005 12

 THE CURVATURE AND GEODESICS OF THE TORUS

6. The Clairaut Parameterization and the Geodesic Equation

The mystery of k and l solved

Returning to the question of the meaning of the constants of integration k and l which came out
of the geodesic equation, we find that the formulation for du
dv that comes from the Clairaut
parameterization offers an answer.

la 2 (c + a cos v )
2
.
v=! −k 2 +
a 2 (c + a cos v ) a 2 (c + a cos v )
2 2

−k 2 + la 2 (c + a cos v )
2
.
.
u= k v=!
a 2 (c + a cos v )
2
(c + a cos v ) 2
la 2 (c + a cos v ) − k 2
2
.
v=! − k2
2 +l
.
v=!
a 2 (c + a cos v ) a(c + a cos v )
. a(c + a cos v )
(from the geodesic equation) u. = ! k
v (c + a cos v ) 2
la 2 (c + a cos v ) − k 2
2

.
u. = ! ak
v (c + a cos v ) la 2 (c + a cos v ) 2 − k 2

.
u
.
v is identical to the Clairaut parameterization-derived formula for du
dv when k = h and l = 1
a2 .

The two approaches are complementary. The formulas for u. ,v. derived from the geodesic
equation can be used to compute geodesics that are singular in the formula derived from the
Clairaut parameterization. In particular, the Clairaut parameterization-derived formula can’t be
used to compute the inner and outer equator geodesics, but the formulas derived from the
geodesic equation can.

7. A Gallery of Geodesics
The majority of geodesics on the torus are not æsthetically pleasing. They are aperiodic and
cover either the entire surface (if the geodesic is unbounded) or the outer region of the surface
bounded by the barrier curves (if bounded). The rare exceptions are the geodesics which return
to their starting point after just a few circuits around the z axis.

Define the period of a geodesic as the number of circuits it makes around the z axis before
returning to its starting point. Most geodesics never return to their starting point, eventually
covering either the entire torus surface or the region between barrier curves. However, there are
geodesics that are pleasing; these are the unbounded geodesics of period 1, and the bounded
geodesics of period 1 or 2.

Mark L. Irons 17 November 2005 13

THE CURVATURE AND GEODESICS OF THE TORUS

The unbounded geodesics with period 1 cross each equator n times (n m 1).

n=1 n=5

The interesting bounded geodesics fall into two groups. Those of period 1 do not self-intersect.
(For bounded geodesics, n denotes how many times the geodesic touches each barrier curve.)

n=1

The geodesics of period 2 intersect themselves n times (n odd, of course).

n=1 n=3

For bounded geodesics the allowed values of n depend on the ratio c/a. Unbounded geodesics
are not affected by c/a.

8. Open Questions
The influence of c/a on bounded geodesics

The kinds of bounded geodesics one can find on a particular torus are determined not only by h,
but also by the ratio c/a. For instance, given a torus with ac = 31 , there is no period 1 bounded
geodesic which touches each barrier curve more than once. Yet for a torus with ac = 81 , there is a
period 1 bounded geodesic which touches each barrier curve twice, and another which touches
each barrier curve three times.

Mark L. Irons 17 November 2005 14

 THE CURVATURE AND GEODESICS OF THE TORUS

This raises a question: what is the range of c/a of the toruses that contain bounded period 1
geodesics which touch each barrier curve exactly n times, as a function of n? How about for
different periods? There is no analytic apparatus I know of with which we can approach the
problem. Calculation appears to be the only way to go.

Note that this restriction appears to apply only to bounded geodesics. There is no corresponding
restriction for unbounded geodesics; by choosing an appropriate h, one can find a period 1
geodesic which crosses both equators as often as one pleases.

Questions about critical values of h

Another open question concerns the values of h which yield crowd-pleasing geodesics with
periods 1 and 2. As c/a changes, so does the value of h which yields a particular pleasing
geodesic (say, a period 1 geodesic which crosses both equators three times). Is there a simple
relation between these two quantities?

A similar question exists for values of h for a given c/a. Define hp as the value of h which yields
a period 1 geodesic which crosses both equators p times. As p increases, at what rate does hp
converge to 0? Is this governed by a simple rule? What about period 2 bounded geodesics?

9. Numerically Calculating Geodesic Paths

Numerically calculating points on these geodesics is a little tricky; for bounded geodesics, du
dv is
undefined where the geodesic touches a barrier curve. After initial experiments with a
spreadsheet, a short Perl script was used to generate points. A second script used successive
approximation to find h values which yielded geodesics with integral periodicity. The images in
this paper are as accurate as I could make them, but there was a slight tradeoff between precise
numeric solutions and illustrative power. Take the images with a grain of salt.

On a puzzling note, the second script worked extremely well for unbounded geodesics, but was
less successful for bounded geodesics. The reason for the discrepancy is not clear.

10. Lessons Learned

The search for the geodesics of the torus led to the creation of Irons’ First Law of Examples: if
an example seems obvious but you can’t find it in the literature, it’s more complex than you
expect.

11. References
Barrett O’Neill, Elementary Differential Geometry, 2nd Ed., Harcourt, 1997;
http://www.math.ucla.edu/~bon/.

Thanks are due to professors Tevian Dray, for his encouragement,

and Juha Pohjanpelto, for a fruitful suggestion.

Mark L. Irons 17 November 2005 15

THE CURVATURE AND GEODESICS OF THE TORUS

Appendix A: Parallel Transport on the Torus

Our surface parameterization is an orthogonal patch (F = x u $ x v = 0 ), so we can easily compute
the associated frame field E1, E2.

xu
E1 = = (− sin u, cos u, 0 )
E
x
E 2 = v = (− cos u sin v, − sin u sin v, cos v )
G

Let’s check that their dot product is zero and their cross product is the normal
N = (cos u cos v, sin u cos v, sin v ) .

E 1 $ E 2 = sin u cos u sin v − sin u cos u sin v + 0

=0 „
x y z
E1 % E2 = − sin u cos u 0
− cos u sin v − sin u sin v cos v
= (cos u cos v, sin u cos v, sin 2 u sin v + cos 2 u sin v )
= (cos u cos v, sin u cos v, sin v )
=N „
We can use partial derivatives of E and G to compute the connection form ' 12 , which encodes
pretty much everything we’d ever want to know about how vectors change when parallel
transported on the torus:

' 12 = ( E )v
G
du +
( G )u
E
dv = a du +
a sin v
0 = sin v du

The du term tells us that parallel transport along lines of constant u (longitude lines) doesn’t
affect vectors. We could have predicted this from the symmetry of the torus; along a line of
longitude, the neighborhoods to the left and right are mirror images, so there’s no preferred
direction for a vector to rotate.

Mark L. Irons 17 November 2005 16

 THE CURVATURE AND GEODESICS OF THE TORUS

The sin v term tells us that parallel transport along lines of constant v (latitude lines) causes
vectors to rotate through 2 sin v during their journey back to their starting point. Let’s look at
some specific cases, starting with parallel transport along the outer equator. Here sin v = 0 ,
which means the vectors do not rotate as they are parallel transported along the outer equator.

At the top of the torus (v = 2 ) sin v = 1 , so a vector rotates through a full 2 during its journey.
Note the angle between the blue vector and its path (red) as the vector is parallel transported.

Something else interesting is happening simultaneously: the vector’s origin is also rotating
through2. These rotations cancel, leaving the blue vector pointing in the same direction in the
embedding space. Someone living on the torus would say the vector rotates as it is parallel
transported, while someone living outside the surface would not. (At v = 2 the torus’s Gaussian
curvature is zero, so it’s not surprising that vectors parallel transported along that path don’t
appear to rotate in the embedding space.)

Parallel transport along other lines of latitude causes vectors to rotate varying amounts (2 sin v).
In the next illustration, four frames aligned with the u and v axes (at v = 0, 6 , 3 , and 2 ) are
parallel transported widdershins around the torus.

Mark L. Irons 17 November 2005 17

THE CURVATURE AND GEODESICS OF THE TORUS

The second line of latitude from the bottom is at v = 6 , so vectors parallel transported along it
will have rotated through 2 sin 6 = 2 12 =  , as indeed they have.

Putting all of this together, here’s how a whole bunch of frames rotate while being parallel
transported widdershins along lines of latitude, starting at the red longitude line.

When creating these images, I was surprised by how quickly the values of 2 sin v change near
dv = cos v , which has extremes at integer multiples of  . So the
v = 0 . Then I remembered that d sin v

rate of change of the effect of parallel transport along lines of latitude is most extreme at the
outer and inner equators.

Mark L. Irons 17 November 2005 18

 THE CURVATURE AND GEODESICS OF THE TORUS

Finally, parallel transport on the bottom half of the torus is the same except for direction of
rotation, since sin v is negative there.

Students of differential geometry may have noticed that sin v du is also the ' 12 of the sphere.
The difference is that for the sphere, the domain of v is − 2 , 2 , while for the torus it is [−,  ] .
Parallel transport on the outer half of the torus mirrors parallel transport on the sphere.

While we’re at it, let’s recompute the Gaussian curvature from E and G:

K= −1
EG
( G )u
E u
+
( E )v
G v
= −1
a(c+a cos v )
( −a sin
a
v
)v =
(sin v ) v
a(c+a cos v )
= cos v
a(c+a cos v )
„
This agrees with the value for the Gaussian curvature we computed from the shape operator, but
unlike that calculation this one doesn’t require a normal to the surface. Thus, if we lived on the
torus, we could compute our space’s Gaussian curvature directly from measurements made
within our space, without assuming the existence of an embedding space. That’s the beauty of
differential geometry.

Mark L. Irons 17 November 2005 19