HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I

J.C.

ALVAREZ PAIVA
Abstract. Hilberts fourth problems asks to construct and study the
geometries in which the straight line segment is the shortest connection
between two points. In this paper the reader shall nd an elementary
introduction to the problem and its solutions in dimension two by Buse-
mann, Pogorelov, and Ambartzumian. The relationship between inte-
gral geometry and inverse problems in variational calculus is emphasized.
Contents
1. Introduction 1
2. Basic denitions and interpretations of the problem 2
3. Minkowski planes 6
4. Hilbert geometries 9
5. The Crofton formula 10
6. Busemanns construction of projective metrics 13
7. Ambartzumians construction 14
8. Variational interpretation of Hilberts fourth problem 15
9. Analytic solution of Hilberts fourth problem 17
10. A glimpse ahead 20
References 21
1. Introduction
Hilberts list of problems, read at the International Congress of Mathe-
maticians in 1900, is perhaps one of the most inuential documents in the
history of mathematics. The twenty-three problems in this list have been
the subject of numerous investigations for the last hundred years and con-
tinue to yield much beautiful mathematics. Even when one of Hilberts
problems has been solved in its original formulation, its variations and the
developments arising from its solution continue to pique the curiosity of
mathematicians. One example is Hilberts third problem on the decompo-
sition of polyhedra (see chapter seven in [1]). This problem was solved by
M. Dehn ([23]) just two years after Hilberts address, but the concepts he
introduced have evolved in dierent directions. For instance, the theory of
Key words and phrases. Integral geometry, Crofton formula, Hilberts fourth problem,
Minkowski spaces, cosine transform.
1
2 J.C.

ALVAREZ PAIVA
valuations, a central part of modern convex geometry, is a direct descendent
of Dehns solution.
In posing his problems, Hilbert did not shy away from vague statements
which would be subject to interpretation. Problem six on the mathematical
treatment of the axioms of physics is perhaps the prime example of this.
Hilberts fourth problem, the subject of this note, is another.
In its original formulation, Hilberts fourth problem asks to construct
and study the geometries in which the straight line segment is the shortest
connection between two points. The original wording (see [25]) makes one
think the problem is part of Hilberts project to study the foundations of
geometry. However, the dierent modern approaches make it clear that the
problem is at the basis of integral geometry, inverse problems in the calculus
of variations, and Finsler geometry.
In this paper the reader will nd various approaches to the solution of
Hilberts fourth problem in two dimensions. We will start with the examples
given by Minkowski and Hilbert of metrics where the shortest connection
between two points is a line segment, and proceed to the constructions of
Busemann and Ambartzumian which together give the most elegant and
natural solution to the problem. Later, we will cover Pogorelovs approach
using Hamels equations and the cosine transform. A second paper shall
treat the subject from the symplectic viewpoint.
An eort has been made to make the paper partly accessible to undergrad-
uate students and mathematically minded readers from other disciplines.
The material is presented as a collection of exercises some of which point at
extensions or digressions of the ideas which led to the solution of Hilberts
fourth problem in two dimensions. From the authors viewpoint, Hilberts
fourth problem is simply an axis around which revolve two fruitful lines of
research: the relation between integral geometry and variational calculus
and the relation between metric and symplectic geometry. The advanced
reader is asked to keep these two themes in mind as he/she reads the paper.
For more information on Hilberts problems the reader is referred to the
two volumes Mathematical developments arising from Hilbert problems,
Proceedings of Symposia in Pure Math. Vol. XXVII Part 1, F. Browder
(Ed.), AMS Rhode Island, 1974, and to the recently published The Honors
Class: Hilberts problems and their solvers by Benjamin Yandell, A K
Peters, Ltd, 2001.
2. Basic definitions and interpretations of the problem
Most interpretations of Hilberts fourth problem revolve around the notion
of distance or metric. To nd all the possible denitions of distance on the
plane for which the shortest path between two given points is the line segment
that joins them is a clear and attractive formulation. Of course, our usual
notion of distance in two dimensions, borrowed from Euclidean geometry,
satises this property and we may wonder if there is any other that does. In
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 3
Fig. 1. Charting the northern hemisphere.
fact, all the three classical geometries, Euclidean, elliptic, and hyperbolic,
dene distances on the plane or portions of it for which the shortest path
between two points is the line segment that joins them. Hyperbolic geometry
and the generalization that led Hilbert to pose the fourth problem will be
treated in the next section, but the elliptic case the geometry of the
sphere is easy to explain to the proverbial rst man on the street:
Suppose youre a very small being living in the northern hemisphere of
a perfectly round planet. At the last planetary convention, your nations
geographers proposed the incredibly impractical idea of charting an innite
map of the surface of the northern hemisphere by drawing above the north
pole a virtual plane parallel to the equator. For every point in the surface
of the planet, they would compute the line from the center of the planet to
the point and then compute where this line cut the plane. That would give
them a point on the plane for every point on the surface. If we dene the
distance between two points in the map as the distance of the corresponding
points on the surface of the planet, we will get a new sort of distance very
dierent from the Euclidean, but for which the shortest curve between two
points is still the straight line that joins them.
In order to tackle Hilberts fourth problem we need to understand pre-
cisely the notion of distance. We have seen that the distance between two
points in the plane can be dened in dierent ways, but it would be hard
for us to accept that the distance between two points be a negative number,
or that two distinct points be at zero distance from each other. Mathemati-
cians abstracted such common sense properties of distances and came up
with the following three axioms:
(1) The distance between any two points is greater than or equal to zero,
and it is equal to zero if and only if the points coincide.
(2) The distance between a point x and a point y is the same as the
distance between y and x.
(3) If x, y, and z are three points, then the distance between x and z
is not greater than the distance between x and y plus the distance
between y and z.
4 J.C.

ALVAREZ PAIVA
Any measurement that assigns to any two points a number and that sat-
ises the three properties above is deemed to be a possible way of measuring
distances. It is sometimes convenient to admit that two points are at innite
distance from each other. The denition that you would nd in a textbook
runs as follows:
Denition 2.1. A metric space is a pair (X, d), where X is a set and the
distance function, or metric,
d : X X R
satises following properties:
Positivity: d(x, y) 0, and d(x, y) = 0 if and only if x = y.
Symmetry: d(x, y) = d(y, x).
Triangle inequality: d(x, z) d(x, y) + d(y, z).
A very simple example of a metric space with a surprising connection to
Hilberts fourth problem is constructed as follows:
Take S to be the collection of all subsets of 1, . . . , n. Some of the
elements of S are the empty set, 1, 2, 3, and 1, . . . , n. If x and y are
in S dene d(x, y) as the number of elements in x which are not in y plus
the number of elements of y that are not in x. For example,
d(1, 2, 2, 5, 6) = 3.
Exercise 2.1. Show that (S, d) is a metric space. Moreover, S satises
the following stronger version of the triangle inequality: if x
1
, . . . , x
k
are
elements of S and b
1
, . . . , b
k
are integer numbers that add up to 1, then
k

i,j=1
d(x
i
, x
j
)b
i
b
j
0.
For k = 3 the above property is just a fancy way to rewrite the triangle
inequality. When a metric space satises this property for any positive
integer k, it is called hypermetric. Remarkably, all the solutions to Hilberts
fourth problem in dimension two are hypermetric. This was rst remarked
by R. Alexander in [2].
Exercise 2.2. Dene the distance between two points (x, y, z) and (x

, y

, z

)
in three-dimensional space as the biggest of the three numbers [xx

[, [yy

[,
and [z z

[. Show that this distance is not hypermetric.

Now that we understand the freedom we have in the choice of a distance
function, it remains to makes sense of the term the shortest path that
appears in our metric interpretation of Hilberts fourth problem. For that
we must understand how to compute the length of a path in a metric space.
If all we know is to measure distances between pairs of points, a natural
way to estimate the length of a path is to mark a number of points on it,
measure the distances between consecutive points, and add them up.
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 5
Fig. 2. Estimating the length of a path.
Of course, this is just an estimate. By taking the points more and more
densely along the path, our intuition tells us that, in the limit, we will nd its
length. For many reasons, it is sometimes convenient not to evoke directly
this passage to the limit and it is preferred to dene the length of a path as
follows:
Denition 2.2. Let (X, d) be a metric space and let : [a, b] X be a
continuous curve. The length of is dened as the supremum of

n1

i=0
d((t
i
), (t
i+1
)) : a = t
0
< t
1
< < t
n
= b is any partition of [a, b].
For almost every example one comes up with, the intuitive and the formal
denition coincide.
Weve seen that if we can measure distances, we can measure lengths of
curves. However, if we can measure lengths, we may redene the distance
between two points as the inmum of the lengths of all curves joining them.
Sometimes the redened distance does not agree with the original. For
example, think of a plane where you have taken out a circular obstacle (Fig.
3). The Euclidean distance between two points is the inmum of the lengths
of the curves joining them only if the line segment that joins them does not
pass through the interior of the obstacle.
Metric spaces where the distance between two points is the inmum of
the lengths of all curves joining them are called length spaces. From now on
we will be solely concerned with length spaces.
Denition 2.3. A continuous curve in a metric space is called a segment
if the distance between its endpoints equals the length of the curve. A
continuous curve is called a geodesic if it can partitioned into segments.
Denition 2.4. A metric on the plane is said to be projective if straight
lines are geodesics. A metric on the sphere is said to be projective if great
circles are geodesics.
Hilberts fourth problem can now be given a precise statement: Construct
and study all projective metrics on the plane. This statement is usually
extended to include not only the plane, but also the sphere, the projective
plane, and convex domains in R
2
.
6 J.C.

ALVAREZ PAIVA
Fig. 3. Plane with an obstacle
Exercise 2.3. Extend the denition of projective metrics to the projective
plane and try your hand at a denition of locally projective metrics on gen-
eral surfaces. Hint: in this last case, your denition should be independent
of coordinate changes.
We end this section with a very simple characterization of projective met-
rics on the plane.
Exercise 2.4. Show that a metric d on the plane is projective if and only
if for any triple of collinear points x, y, and z, with y between x and z, we
have that
d(x, z) = d(x, y) + d(y, z).
3. Minkowski planes
In Hilberts text, he mentions that Minkowski had already found many
examples of projective metrics on the plane and that these metrics had
the additional property that translating a line segment does not change
its length. Hilbert was speaking about what we would now call nite-
dimensional normed spaces, and which many people still call Minkowski
spaces (see [32]).
Let us start, as Hilbert did, by remarking the importance of the invariance
of distances under translations. Let us say that a metric, or distance, on
the plane is invariant under translations if the distance between any two
points x and y equals the distance between the points x + v and y + v for
any vector v.
Exercise 3.1. Show that a metric on the plane that is invariant under
translations is automatically projective.
In view of this exercise, constructing and studying distances that are
invariant under translations is a logic rst step in the solution of Hilberts
fourth problem.
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 7
x
y
x + v
y + v
Fig. 4. Translations dont change distances.
Exercise 3.2. Assume d is a translation-invariant metric on R
2
and dene
the norm of a vector x, denoted by |x|, as its distance from the origin.
Show that the function
| | : R
2
R
satises the following properties:
Positivity: |x| 0 and |x| = 0 if and only if x is the origin.
Homogeneity: if is any real number |x| = [[|x|.
Convexity: |x + y| |x| +|y|.
Exercise 3.3. A function on the plane satisfying the properties in exer-
cise 3.2 is called a norm. Show that if | | is a norm on the plane, then the
function
d : R
2
R
2
[0, )
dened by d(x, y) = |x y| is a metric that is invariant under translations.
Denition 3.1. A pair (R
2
, d) where d is a metric that is invariant under
translations shall be called a Minkowski plane.
In order to construct and understand Minkowski planes, we look at their
unit discs. Since these are all the same up to translation, we can concentrate
on the disc with center at the origin.
Denition 3.2. A subset K of the plane is said to be convex if every two
points in K are joined by a line segment entirely contained in K. A convex
set is said to be a convex body if it is compact and its interior is not empty.
Exercise 3.4. The unit disc of a Minkowski plane is the set of vectors whose
distance from the origin does not exceed one. Show that the unit disc of a
Minkowski plane is a convex body symmetric around the origin.
Moreover, any convex body that is symmetric about the origin is the unit
disc of some Minkowski plane: simply dene the norm of a nonzero vector
8 J.C.

ALVAREZ PAIVA
(x,y)
(x,y)
(x,y)
(x,y)
d((x,y),(x,y)) := max {|xx|,|yy|}.
d((x,y),(x,y)) := |xx| + |yy| .
Fig. 5. Examples of Minkowski geometries.
x as the only positive number |x| for which x/|x| belongs to the boundary
of the unit disc.
As a result of this simple remark, constructing Minkowski planes is very
easy indeed: just draw a closed convex curve that is symmetric about the
origin (e.g., a square, a rectangle, an ellipse, a rhombus) and we have a norm
and hence a projective metric.
Another simple way of constructing and studying norms that is central to
Pogorelovs approach to Hilberts fourth problem is given in the following
exercises:
Exercise 3.5. Given a piecewise continuous function f : S
1
R, its cosine
transform is dened as the function
L(v
1
, v
2
) :=

2
0
[v
1
cos() + v
2
sin()[f() d
(1) Show that if f nonnegative and takes positive values in at least three
distinct points, then L is a norm on the plane.
(2) Show that if f is odd (i.e., f( +) = f()), then its cosine trans-
form is identically zero. Deduce from this that in order to study the
range of the cosine transform we need only consider the case where
f is even (i.e. f( + ) = f()).
(3) Assume that f is piecewise continuous and even. Show that in polar
coordinates, (r, ), the cosine transform of f is given by the following
formula:
L(r, ) = 2r

+/2
/2
cos( )f()d.
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 9
(4) With f and L as in the previous item, show that
f( + /2) :=
1
4r

L(r, ) +

2
L

2
(r, )

.
Exercise 3.6. Show that if f is a smooth positive function on the circle and
L is its cosine transform in cartesian coordinates (v
1
, v
2
), then the matrix
formed by the second partial derivatives of L
2
is positive denite. Geomet-
rically, this means that the unit circle of the Minkowski plane (R
2
, L) is
smooth and has positive curvature everywhere.
A norm L : R
2
[0, ) that is smooth outside the origin and such that
the matrix formed by the second partial derivatives of L
2
is positive denite
is said to satisfy the Legendre condition. The previous exercise shows that
the cosine transform of a smooth positive function on the circle is a such a
norm, while item (4) on exercise 3.5 shows that every norm satisfying the
Legendre condition is the cosine transform of a smooth positive function on
the circle. Summarizing, we have the following result:
Theorem 3.1 (Blaschke, [17]). The cosine transform establishes a bijection
between the set of smooth, even, positive functions on the circle and norms
on the plane that satisfy the Legendre condition.
4. Hilbert geometries
The reader has probably heard or read about the history of Euclids par-
allel axiom, which states that given a line and a point not belonging to it,
there is a unique line passing through the point and not intersecting the
original line. For a long time geometers sought to show that this axiom
perhaps the less self-evident of Euclids axioms was a logical consequence
of the others. In time it was shown that this axiom is independent of the
others and that geometries in which it does not hold are not only possible,
but natural. In fact, imagine a geometry in which the points range over the
interior of the unit disc and the lines are the straight line segments joining
pairs of points in the unit circle. Given a line and a point not belonging
to it, there are innitely many lines that pass through the point, but never
intersect the original line (Fig. 6).
Such a geometry really exists and is none other than the hyperbolic geom-
etry discovered in the nineteenth century independently by Gauss, Bolyai,
and Lobachevsky. The way in which we shall present it here is due to Arthur
Cayley and Felix Klein.
Let D denote the open unit disc and let C denote its boundary. If x and
y are two points in D, denote by a and b the points of intersection of C
with the straight line passing through x and y with the provision that x lie
between a and y and that y lie between x and b. Thus if we interchange x
and y, we are forced to interchange a and b.
10 J.C.

ALVAREZ PAIVA
Fig. 6. Innitely many parallel lines passing through the same point.
The hyperbolic distance between x and y is dened by the equation
d(x, y) :=
1
2
ln

|y a|
|x a|
|x b|
|y b|

.
Exercise 4.1. Show that the function d satises the three following prop-
erties:
(1) Positivity: d(x, y) 0, and d(x, y) = 0 if and only if x = y.
(2) Symmetry: d(x, y) = d(y, x).
(3) If x, y, and z are three aligned points in D such that y lies between
x and z. Show that d(x, z) = d(x, y) + d(y, z).
The proof of the triangle inequality for d is somewhat more elaborate, but
it only makes use of elementary projective geometry (see chapter 5 in [6]).
Part (3) of exercise 4.1 shows that this metric is projective.
In [26], Hilbert shows that it is possible to replace the open unit disc D
in the construction above by the interior of any convex body in the plane
without changing the fact that d is a projective metric.
Hilbert geometries, as these metric spaces are called, are a natural gen-
eralization of hyperbolic geometry and furnish us with a second large class
of examples of projective metrics. Recently, much progress has been made
in the understanding of the Hilbert geometries (see, for example, [30] and
[27]).
5. The Crofton formula
The key to the solution of many problems in geometry is simply a change
of viewpoint. So far we have concentrated on points and the distances
between them. Hilberts fourth problem asks for those metrics for which the
geodesics are straight lines, so perhaps the best strategy is to replace the
point for the straight line as the fundamental object of our investigations.
We shall take as departing point the following naive question: what would
Euclidean geometry look like if we take the straight line instead of the point
as our basic geometric entity?
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 11
l
z = 1
Fig. 7. The cylinder as the space of oriented lines.
The space of oriented lines. We may identify the space of oriented
straight lines on the plane with the right circular cylinder x
2
+ y
2
= 1 in
three-dimensional space by means of the following construction:
An oriented line lying on the plane z = 0 in three-dimensional space is
uniquely determined by the unit vector q dening its direction, and by the
vector v representing the point on the line that is closest to the origin. The
map q + q v identies the space of oriented lines on the plane z = 0
with the cylinder x
2
+ y
2
= 1.
Another, more geometric, correspondence may be given if we consider not
the plane z = 0, but the plane z = 1. In fact, any line on z = 1 is the
intersection of this plane with some plane through the origin. Using the
orientation on the line, we may determine which side of the plane is up
and which is down: if youre walking head up along the line in the sense
of its orientation, then the origin of R
3
should be to your left. In order to
identify with a point on the cylinder, simply trace an orthogonal ray up
from at the origin and mark where it intersects the cylinder (Fig. 7).
Exercise 5.1. Discover the simple relationship between both identications
of the space of oriented lines with the right circular cylinder. In the rest of
the paper, we use the rst of these identications.
The aim of the following exercises is to get an intuitive feeling for the
correspondence between points on the cylinder x
2
+ y
2
= 1, and henceforth
denoted by (, and oriented lines on the plane. This intuition is crucial for
the understanding of the rest of the paper. Please take ten minutes to do
them on your own and then come back to the paper.
Exercise 5.2. Draw the portion of the cylinder that corresponds to all
oriented lines passing through
(1) a point,
(2) a line segment,
12 J.C.

ALVAREZ PAIVA
(3) the unit disc centered at the origin,
(4) two sides of a given triangle.
Exercise 5.3. The cylinder ( admits the parameterization
(p, ) (cos(), sin(), p),
where 0 < 2, and p is any real number. Draw the region in the
(p, )-plane that corresponds to all oriented lines passing through
(1) a point,
(2) a line segment,
(3) the unit disc centered at the origin,
(4) two sides of a given triangle.
Euclidean transformations preserve areas. A rigid motion, or Eu-
clidean transformation, on the plane is a transformation from the plane
onto itself that preserves distances. It is not too hard to see that these
transformations are composed of translations and rotations, and that they
send oriented straight lines to oriented straight lines. We have then that
Euclidean transformations of the plane induce a class of transformations on
the space of oriented lines (. Let us now understand this class of transfor-
mations and discover what geometric properties they may have.
Exercise 5.4. Parameterize each point in the cylinder ( by its height p and
the angle, , that its projection makes with the positive x-axis (i.e., a point
in ( has coordinates (cos(), sin(), p)).
(1) Show that rotating the oriented line represented by (p, ) counter-
clockwise by an angle around the origin results in the oriented line
represented by (p, + ).
(2) Show that translating the oriented line represented by (p, ) by a
vector (x, y) results in the oriented line represented by (p+xcos()+
y sin(), ).
Exercise 5.5. Show that the area of the portion of the right circular cylinder
x
2
+ y
2
= 1 given by all points with coordinates (cos(), sin(), p), where
(p, ) ranges over some region is just

R
dpd.
Conclude that the area of a region in the space of oriented lines is invariant
under the action of Euclidean transformations.
Dening distance in terms of areas. So far our project of studying Eu-
clidean geometry by taking the line instead of the point as basic geometric
entity has led us to consider the space of oriented lines, to study the action
of Euclidean transformations on this space, and to the remarkable fact that
these transformations preserve areas. However, in order to gain full under-
standing, we must try to reconstruct the Euclidean distance on the plane
by using the geometry on the space of lines.
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 13
Exercise 5.6. Use exercises 5.3 and 5.5 to show that the area in ( of the
region representing the set of all oriented lines passing through a line segment
equals four times its length.
Exercise 5.7. Use exercise 5.6 to show that the length of a polygonal curve
in the plane is given by the following integral over the space of oriented
lines (
1
4

C
#( ) dA,
where dA is the element of area on the cylinder.
By approximating rectiable curves by polygonal ones, the previous exer-
cises furnishes us with a proof of the Crofton formula in Euclidean geometry.
Theorem 5.1 (Crofton formula). If R
2
is a rectiable curve and dA is
the standard area element on the cylinder (, then
length() =
1
4

C
#( ) dA.
Exercise 5.8. Use the Crofton formula to give an easy proof of the follow-
ing, apparently obvious, fact: if a convex body K R
2
contains a second
convex body L, then the perimeter of K is greater than or equal to the
perimeter of L with equality holding if and only if K = L.
6. Busemanns construction of projective metrics
The simple remark that underlies the solution of Hilberts fourth problem
is that the Crofton formula implies that straight lines are geodesics. In fact,
if is a rectiable curve on the plane joining the points x and y, then any
oriented line intersecting the line segment that joins them must intersect .
Applying Croftons formula, we see that the length of is greater than or
equal to that of the line segment joining x and y. The independence of this
argument on the special nature of the area element on ( suggests replacing
dA by other ways of measuring areas on the space of oriented lines.
Any continuous area element on the space of oriented lines, (, has the
form fdA, where f is some positive, continuous function on (. Let us agree
to call the area, measured with fdA, of a region in ( the f-area of .
If f is a positive, continuous function of (, dene a metric on the plane by
setting the distance between any two points as one fourth times the f-area
of the set of all oriented lines that intersect the segment that joins them.
Exercise 6.1. Show that the construction above does indeed dene a dis-
tance function on the plane and that this distance is necessarily projective.
This is Busemanns construction of projective metrics.
The same arguments in your solution of exercise 5.7 show that Croftons
formula holds for the metrics dened by Busemanns construction. In fact,
these metrics were dened so that Croftons formula be a tautology.
14 J.C.

ALVAREZ PAIVA
Exercise 6.2. Let f be a positive, continuous function on the space of ori-
ented lines that depends only on the direction of the oriented line. Show that
Busemanns construction yields a translation invariant metric, and hence a
norm, on the plane.
A metric d on the plane is said to be periodic if there exists two linearly
independent vectors v and w such that translation by any vector of the form
mv + nv, m, n any integers, preserves distances. In general, periodic met-
rics are not invariant under all translations, however we have the following
exercise:
Exercise 6.3. Suppose a metric arising from Busemanns construction is
periodic. Show that it is invariant under translations and, hence, comes
from a norm on the plane.
7. Ambartzumians construction
Busemanns construction shows that to any continuous way of measur-
ing areas in the space of oriented lines we can associate a continuous way
of measuring distances in the plane such that geodesics are straight lines.
The natural question now is whether all projective metrics can be obtained
from Busemanns construction. In [13], Ambartzumian gives a beautiful and
simple construction that answers this question armatively.
We start by remarking a nearly obvious property of Busemanns construc-
tion.
Exercise 7.1. Let fdA be a continuous element of area in the space of
oriented lines and let d be the projective metric on the plane obtained from
fdA using Busemanns construction. Show that the f-area of the set of all
oriented lines passing through both the xy and yz sides of a triangle xyz is
equal to two times the quantity d(x, y) + d(y, z) d(x, z).
This simple remark is the basis of Ambartzumians construction: if d
is a continuous distance function on the plane that satises the projective
condition d(a, b)+d(b, c) = d(a, c) whenever b lies in the line segment joining
a and c, we dene the area of the set of all lines passing through both
the xy and yz sides of a triangle xyz as two times the quantity d(x, y) +
d(y, z) d(x, z). The projective condition guarantees that this set function
is additive, and the family of sets on which it is dened is suciently large so
that the set function admits a unique extension to a continuous measure on
the set of oriented lines provided we specify that this measure be invariant
under the map that changes the orientation of the lines. While the details
in the proofs of these statements are beyond the scope of this article, the
exercises below should give the reader a good sense of why they are true.
We start by describing the set of all oriented lines passing through two
sides of a triangle in the plane as a subset of the cylinder. Of course, the
conscientious reader has already done this exercise in section 5.
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 15
Fig. 8. A short triangle on the sphere.
Exercise 7.2. Dene a short triangle on the unit sphere in three dimen-
sional space as the interior of a spherical triangle in which the length of
each side is less than (see Fig. 8). In particular, a short triangle is disjoint
from its image under the antipodal map. Show that a set in the cylinder
x
2
+y
2
= 1 corresponds to the interior of the set of all oriented lines passing
through two sides of a triangle in the plane if and only if it is the image of a
short triangle and its antipode under the radial projection from the sphere
to the cylinder. Use this description to see that any open set in the cylinder
that is invariant under the antipodal map can be approximated by a disjoint
union of such sets.
Exercise 7.3. Show that the measure of the set of all lines intersecting the
segment xy equals four times the distance between x and y. Use this to
explain why the projective condition on d is necessary for the additivity of
the measure dened in Ambartzumians construction.
8. Variational interpretation of Hilberts fourth problem
In the next two sections, we shall tackle Hilberts fourth problem in a
way that is apparently very dierent from the approach of Busemann and
Ambartzumian. Our starting point is to consider Hilberts fourth problem
as an inverse problem in variational calculus: construct and study all vari-
ational problems on the plane for which the extremals are straight lines. In
order to give a more precise statement, we must review some basic notions.
Let us start by specifying what we mean by a variational problem on
the plane. If L : R
2
R
2
R is a continuous function, then given a
dierentiable parameterized curve : [a, b] R
2
we may compute the
integral
S() :=

b
a
L((t), (t))dt.
Given two points x and y on the plane, we consider the following problem:
nd the inmum of S(), where ranges over all dierentiable parameterized
curves joining x and y. We are also interested in determining if this inmum
is attained and, if so, in the curve or curves at which it is attained.
16 J.C.

ALVAREZ PAIVA
The integrand to keep in mind is the function
L
0
(x
1
, x
2
, v
1
, v
2
) :=

v
2
1
+ v
2
1
,
which yields the variational problem of nding the shortest curve between
two points in the Euclidean plane. This suggests taking an arbitrary function
L(x
1
, x
2
, v
1
, v
2
) and dening the length of a parameterized curve : [a, b]
R
2
as the integral
S() :=

b
a
L((t), (t))dt.
Two basic properties of length that we would like our denition to satisfy
are
(1) The length of any dierentiable curve is a nonnegative number and
it is zero if and only if the curve is constant (i.e., it is a point).
(2) The length of a curve does not depend on the way we traverse it.
In other words, length should be invariant under changes of param-
eterizations.
Exercise 8.1. Let L : R
2
R
2
R be a continuous function and use it
to dene a length for dierentiable parameterized curves in the manner
described above.
Show that property (1) holds if and only if L(x
1
, x
2
, v
1
, v
2
) is non-
negative and it is zero if and only if v
1
and v
2
are both equal to
zero.
Show that property (2) holds if and only if L is homogeneous of order
one in the last two coordinates: L(x
1
, x
2
, v
1
, v
2
) = [[L(x
1
, x
2
, v
1
, v
2
).
The third requirement we shall make of our denition of length is more
subtle: using lengths, we may dene the distance between two points as
the inmum of the lengths of all dierentiable curves joining them. We
may then use this denition of distance to redene the lengths of curves as
was done in section 2. The two ways of measuring lengths may not agree.
Indeed, Busemann and Mayer proved in [22] that they agree if and only if
the function L is convex in the last two variables:
L(x
1
, x
2
, v
1
+ w
1
, v
2
+ w
2
) L(x
1
, x
2
, v
1
, v
2
) + L(x
1
, x
2
, w
1
, w
2
).
We may summarize all three requirements on the function L by saying
that for every point (x
1
, x
2
) on the plane the function (v
1
, v
2
) L(x
1
, x
2
, v
1
, v
2
)
is a norm.
Exercise 8.2. Let L : R
2
R
2
R be a continuous function such that for
every point (x
1
, x
2
) on the plane the function (v
1
, v
2
) L(x
1
, x
2
, v
1
, v
2
) is a
norm. Dene the length of a dierentiable parameterized curve : [a, b]
R
2
as the integral
S() :=

b
a
L((t), (t))dt,
and dene d(x, y) as the inmum of the lengths of all dierentiable param-
eterized curves joining the points x and y. Show that d is a distance.
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 17
Denition 8.1. A function L : R
2
R
2
R is called a continuous Finsler
metric if it is continuous and if for every point (x
1
, x
2
) on the plane the
function (v
1
, v
2
) L(x
1
, x
2
, v
1
, v
2
) is a norm.
Exercise 8.2 shows that a continuous Finsler metric denes a metric on
the plane. By an abuse of notation we shall also call this metric (i.e., the
distance function) a continuous Finsler metric. A possible interpretation
of Hilberts fourth problem is to construct and study all continuous Finsler
metrics on the plane for which straight lines are geodesics. However, in order
to apply the standard techniques of the calculus of variations in the study
of the integrand L, we shall require the following regularity conditions:
(1) The function L is smooth in R
2
(R
2
(0, 0)).
(2) The matrix of second partial derivatives of L
2
with respect to the
last two variables is positive denite at every point (x
1
, x
2
, v
1
, v
2
)
with (v
1
, v
2
) ,= (0, 0).
In the terminology of section three, we require that at each point (x
1
, x
2
)
the norm L(x
1
, x
2
, ) satisfy the Legendre condition. This condition is quite
standard in the calculus of variation and, among other things, it guarantees
that all geodesics of the Finsler metric are smooth curves and that there is
only one geodesic passing through each point in each direction.
Denition 8.2. A continuous Finsler metric L on the plane is said to be
a (smooth) Finsler metric if it is smooth in R
2
(R
2
(0, 0)) and at each
point (x
1
, x
2
) the norm L(x
1
, x
2
, ) satises the Legendre condition.
From now until the end of this paper, Hilberts fourth problem will be
interpreted as: to construct and study all Finsler metrics on the plane whose
geodesics are straight lines. In [28], Pogorelov showed that the restriction
to smooth Finsler metrics is not fundamental: any continuous projective
Finsler metric on the plane can be uniformly approximated in each compact
set by smooth projective Finsler metrics.
The reader interested in knowing more about Finsler metrics is referred to
the books of Bao, Chern, and Shen ([15]),

Alvarez and Duran ([8]), and to the
survey article [7]. The last two works may be downloaded from the Finsler
Geometry Newsletter, http://www.math.poly.edu/research/finsler, along
with many other papers on the subject.
9. Analytic solution of Hilberts fourth problem
The rst step in the analytic solution of Hilberts fourth problem is the
following simple result.
Theorem 9.1 (Hamel, [24]). Let L : R
2
(R
2
(0, 0)) R be a smooth
Finsler metric. Straight lines are geodesics for the metric dened by L if
and only if L satises the following partial dierential equation:

2
L
x
1
v
2
=

2
L
x
2
v
1
.
18 J.C.

ALVAREZ PAIVA
In order to prove this theorem we recall (or state) that if L is a smooth
Finsler metric on the plane and (t) is one of its geodesics, then the Euler-
Lagrange equations
d
dt
L
v
1
((t), (t))
L
x
1
((t), (t)) = 0
d
dt
L
v
2
((t), (t))
L
x
2
((t), (t)) = 0
hold. These equations also hold for more general integrands than Finsler
metrics (see, for example, Arnolds book [14] for a complete account of
Lagrangian mechanics and the Euler-Lagrange equations), but that need
not concern us at this point.
Exercise 9.1. Prove Hamels theorem using the Euler-Lagrange equations
above and the following hints:
(1) Use Eulers formula for homogeneous functions to replace the rst
partial derivatives of L with respect to x
1
and x
2
in the Euler-
Lagrange equations by expressions in the second partial derivatives
of L.
(2) Use that lines t (x
1
+ tv
1
, x
2
+ tv
2
, v
1
, v
2
) are geodesics.
From exercise 3.5 we know that any norm L(v
1
, v
2
) satisfying the Legendre
condition can be written in a unique way as
L(v
1
, v
2
) =

2
0
[v
1
cos() + v
2
sin()[f()d,
where f is a smooth, positive, and even function on the circle. Since for every
xed (x
1
, x
2
) a smooth Finsler metric L(x
1
, x
2
, v
1
, v
2
) is a norm satisfying
the Legendre condition, there is a unique function f(x
1
, x
2
, ) that is positive
and even in its last variable, and that satises
L(x
1
, x
2
, v
1
, v
2
) =

2
0
[v
1
cos() + v
2
sin()[f(x
1
, x
2
, )d.
Its not hard to see that f must be smooth as a function of its three variables.
Using this integral representation of Finsler metrics it is quite easy to
solve Hamels dierential equation.
Theorem 9.2 (Pogorelov, [28]). Let us use the cosine transform to represent
a smooth Finsler metric on the plane, L(x
1
, x
2
, v
1
, v
2
), as the integral
L(x
1
, x
2
, v
1
, v
2
) =

2
0
[v
1
cos() + v
2
sin()[f(x
1
, x
2
, )d,
where f is a smooth positive function on R
2
S
1
that is even in the last
variable (i.e., f(x
1
, x
2
, + ) = f(x
1
, x
2
, )). The function L satises
Hamels dierential equation if and only if there exists a smooth function
g : R S
1
R with f(x
1
, x
2
, ) = g(x
1
cos() + x
2
sin(), ).
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 19
Exercise 9.2. This exercise outlines the proof of theorem 9.2. The only idea
in the proof is to get rid of the absolute values inside the integral by changing
to polar coordinates in the velocities: (x
1
, x
2
, v
1
, v
2
) (x
1
, x
2
, r, ).
(1) Show that in the variables (x
1
, x
2
, r, ) Hamels equations take the
form
sin()

2
L
x
1
r
+
cos()
r

2
L
x
1

= cos()

2
L
x
2
r

sin()
r

2
L
x
2

.
(2) Show that if f(x
1
, x
2
, ) is a smooth function that is even in the last
variable, then
L(x
1
, x
2
, r, ) = 2r

+/2
/2
cos( )f(x
1
, x
2
, )d
satises Hamels equation if and only if the integral

+/2
/2

sin()
f
x
1
+ cos()
f
x
2

d
is zero for any value of .
(3) Show that the last condition on the above item is met if and only if
sin()
f
x
1
+ cos()
f
x
2
= 0
and deduce Pogorelovs theorem from this fact.
To obtain a Finsler metric L on the plane whose geodesics are straight
lines one just takes a function g : RS
1
R that is smooth, positive, and
even in the second variable, and plugs it into the formula
L(x
1
, x
2
, v
1
, v
2
) =

2
0
[v
1
cos() + v
2
sin()[g(x
1
cos() + x
2
sin(), )d.
Exercise 9.3. Show that if g(p, ) := 1+p
2
, then the corresponding Finsler
metric is
L(x
1
, x
2
, v
1
, v
2
) =
1
3

v
2
1
+ v
2
2
[(3 + x
2
1
+ x
2
2
)(v
2
1
+ v
2
2
) + (x
1
v
1
+ x
2
v
2
)
2
].
The reader may well ask about the relationship between this analytic ap-
proach and the geometric approach of Busemann and Ambartzumian. The
answer is that they are the two sides of the same coin. If g(p, ) is a smooth,
positive function that is even in its last variable, then the Finsler metric
associated to g by Pogorelovs construction is the same as the metric associ-
ated to the area element gdA in the space of oriented lines, (, by Busemanns
construction. For the details and a thorough study of the relationship be-
tween Crofton formulas and cosine transforms the reader is referred to the
paper of

Alvarez and Fernandes, [11].
20 J.C.

ALVAREZ PAIVA
10. A glimpse ahead
It would be a pity if the reader of this article should conclude that Hilberts
fourth problem is completely solved and that there is nothing that he or she
may contribute to the subject. Recent work by

Alvarez ([3, 4, 5]), Alvarez
and Fernandes ([10, 11]), Bryant ([18]), and Schneider ([29]) shows that the
subject still has much to yield. In this last section I have collected some
variations on Hilberts fourth problem that may tempt those with a taste
for simply stated geometric problems. I have followed the Russian tradition
of formulating problems in their simplest nontrivial formulation and have
included a short comment after each problem.
Problem 1 (Busemann, [20]). A nonsymmetric distance on a set X is a
function d : X X R that satises
Positivity: d(x, y) 0 and d(x, y) = 0 if and only if x = y.
Triangle inequality: d(x, z) d(x, y) + d(y, z.
Construct and study all nonsymmetric distances on the plane for which
straight lines are geodesics.
Comment. Bryant constructed in [18] a class of nonsymmetric (Finsler)
metrics on the two dimensional sphere for which the geodesics are great
circles. His examples have constant ag curvature.
Problem 2 (

Alvarez, [7]). Construct and study all metrics on the com-

plex, quaternionic, and Cayley planes for which the geodesics agree with
the geodesics for the standard homogeneous metrics in these spaces.
Comment. So far not a single nontrivial example of such a metric is known.
However,

Alvarez and Duran (see [9]) constructed a class of Finsler metrics
on projective and quaternionic projective spaces for which projective lines
are totally geodesic and all geodesics are geometric circles.
Problem 3 (Bryant, [19]). Is there a Riemannian metric on the real pro-
jective space such that projective planes are minimal surfaces and which is
not isometric to the standard?
Comment. If one allows the metric to be dened on an open subset of real
projective space then there are Riemannian metrics not isometric to the
standard for which projective planes are minimal. This has been studied by
Bekkar and Bryant (see [16] and [19]). If we allow the metric to be Finsler,
then

Alvarez and Fernandes show in [11] that there is a correspondence
between these metrics and volume forms on the Grassmannian of oriented
2-planes in R
4
.
Problem 4. Do there exist non-Riemannian Finsler metrics on the complex
projective plane for which all complex curves are minimal?
Comment. It can be shown that the examples of

Alvarez and Duran (see
[9]) of metrics in the complex projective plane whose geodesics are circles
HILBERTS FOURTH PROBLEM IN TWO DIMENSIONS I 21
satisfy the additional property that complex projective lines are minimal. It
is not known whether conics or other complex curves are minimal as well.
References
[1] M. Aigner and G.M. Ziegler, Proofs from the Book Second Corrected Printing,
Springer-Verlag, Berlin Heidelberg New York, 1999.
[2] R. Alexander, Zonoid theory and Hilberts fourth problem, Geom. Dedicata 28 (1988),
no. 2, 199211.
[3] J.C.

Alvarez Paiva, Anti-self-dual symplectic forms and integral geometry, in Anal-
ysis, Geometry, Number Theory, The Mathematics of Leon Ehrenpreis., 15 25,
Contemp. Math., 251, Amer. Math. Soc., Providence, RI, 2000.
[4] J.C.

Alvarez Paiva, Contact topology, taut immersions, and Hilberts fourth problem,
in Dierential and Symplectic Topology of Knots and Curves, 121, Amer. Math.
Soc. Transl. Ser. 2, 190, Amer. Math. Soc., Providence, RI, 1999.
[5] J.C.

Alvarez Paiva, Symplectic geometry and Hilberts fourth problem, preprint (2002).
[6] J.C.

Alvarez Paiva, Interactive Course on Projective Geometry,
http://www.math.poly.edu/courses/projective geometry.
[7] J.C.

Alvarez Paiva, Some problems on Finsler geometry, preprint (2000).
[8] J.C.

Alvarez Paiva and C. Duran, An Introduction to Finsler Geometry, Notas de
la Escuela Venezolana de Matematicas, 1998.
[9] J.C.

Alvarez Paiva and C. Duran, Isometric submersions of Finsler manifolds, to
appear in Proc. of the Amer. Math. Soc.
[10] J.C.

Alvarez Paiva and E. Fernandes, Crofton formulas in projective Finsler spaces,
Electronic Research Announcements of the Amer. Math. Soc. 4 (1998), 91100.
[11] J.C.

Alvarez Paiva and E. Fernandes, Crofton formulas and Gelfand transforms,
preprint 2000.
[12] J.C.

Alvarez Paiva, I.M. Gelfand, and M. Smirnov, Crofton densities, symplectic
geometry, and Hilberts fourth problem, in Arnold-Gelfand Mathematical Seminars,
Geometry and Singularity Theory, V.I. Arnold, I.M. Gelfand, M. Smirnov, and V.S.
Retakh (eds.). Birkhauser, Boston, 1997, pp. 7792.
[13] R. Ambartzumian, A note on pseudo-metrics on the plane, Z. Wahrsch. Verw. Gebiete
37 (1976), 145 155.
[14] V.I. Arnold, Mathematical Methods of Classical Mechanics, Graduate Texts in
Mathematics, Springer-Verlag, 1989.
[15] D. Bao and S.S. Chern and Z. Shen, An Introduction to Riemann-Finsler Geometry,
Graduate Texts in Mathematics, Springer-Verlag, Berlin, 2000.
[16] M. Bekkar, Sur les metriques admettant les plans comme surfaces minimales, Proc.
Amer. Math. Soc 124 (1996), 30773083.
[17] W. Blaschke, Kreis und Kugel, Chelsea, New York, 1955.
[18] R.L. Bryant, Projectively at Finsler 2-spheres of constant curvature, Selecta Math-
ematica (New Series) 3 (1997), 161203.
[19] R.L. Bryant, On metrics in 3-space for which the planes are minimal, preprint, 1995.
[20] H. Busemann, Problem IV: Desarguesian spaces in Mathematical developments aris-
ing from Hilbert problems, Proc. Sympos. Pure Math., Vol. XXVIII, Amer. Math.
Soc., Providence, R. I., 1976.
[21] H. Busemann, Geometries in which the planes minimize area, Ann. Mat. Pura Appl.
(4) 55 (1961), 171190.
[22] H. Busemann and W. Mayer, On the foundations of variational calculus, Trans. AMS
49 (1941), 173198.
[23] M. Dehn,

Uber den Rauminhalt, Math. Ann. 55 (1902), 465478.
22 J.C.

ALVAREZ PAIVA
[24] Hamel,

Uber die Geometrien, in denen die Geraden die k urzesten sind, Math. Ann.
57 (1903), 231 264.
[25] D. Hilbert, Mathematical problems, translation in Mathematical developments aris-
ing from Hilbert problems, Proceedings of Symposia in Pure Math. Vol. XXVII Part
1, F. Browder (Ed.), AMS Rhode Island, 1974.
[26] D. Hilbert, Foundations of Geometry, Open Court Classics, Lasalle, Illinois, 1971.
[27] A. Karlsson and G.A. Noskov, The Hilbert metric and Gromov hyperbolicity, preprint
2001.
[28] A.V. Pogorelov, Hilberts Fourth Problem, Scripta Series in Mathematics, Winston
and Sons, 1979.
[29] R. Schneider, Crofton formulas in hypermetric projective Finsler spaces, Arch. Math.
77 (2001), 8597.
[30] E. Socie-Methou, Comportements asymptotiques et rigidites en geometries de Hilbert,
th`ese, Universite de Strasbourg, 2000.
[31] Z.I. Szabo, Hilberts fourth problem, I, Adv. in Math. 59 (1986), 185301.
[32] A.C. Thompson, Minkowski Geometry, Encyclopedia of Math. and Its Applica-
tions, Vol. 63, Cambridge Univ. Press, Cambridge, 1996.
J.C.

Alvarez Paiva, Polytechnic University Brooklyn, Six MetroTech Cen-
ter, Brooklyn, NY 11201 USA.
E-mail address: jalvarez@duke.poly.edu