MÖBIUS PHOTOGRAMMETRY
MATTEO GALLET, GEORG NAWRATIL, AND JOSEF SCHICHO
Abstract. Motivated by results on the mobility of mechanical devices called
pentapods, this paper deals with a mathematically freestanding problem, which
we call Möbius Photogrammetry. Unlike traditional photogrammetry, which
tries to recover a set of points in three–dimensional space from a finite set
of central projection, we consider the problem of reconstructing a vector of
points in R3 starting from its orthogonal parallel projections. Moreover, we
assume that we have partial information about these projections, namely that
we know them only up to Möbius transformations. The goal in this case is
to understand to what extent we can reconstruct the starting set of points,
and to prove that the result can be achieved if we allow some uncertainties
in the answer. Eventually, the techniques developed in the paper allow us to
show that for a pentapod with mobility at least two, either some anchor points
are collinear, or platform and base are similar, or they are planar and affine
equivalent.
1. Introduction
In this paper we consider the following problem: given a vector of 5 points
⃗A = (A1, . . . , A5) in R3, suppose we have partial information about its orthogonal
projections along all directions in R3; in particular, we suppose to know each of
them only up to Möbius transformations of the plane. Then we ask if and to what
extent we can extract information on ⃗A starting from this partial knowledge about
its orthogonal projections.
In order to deal with this question, in Section 2 we start formalizing it in the
following way:
· the set of directions of R3 is identified with the unit sphere S2;
· a set of 5 points in the plane, considered up to Möbius transformations,
gives a point in the moduli space M5 of five points in P1
C.
So all the information about orthogonal projections of a vector ⃗A of points can
be encoded in one function f ⃗A : S2 −→M5, which we call a Möbius camera. In
Section 3 we give a formal definition of the map f ⃗A and we explain how it can
be thought as a map between projective varieties. We explore some properties of
the Möbius camera, in particular we relate the degree of the image of f ⃗A (which is
always a curve if the points are not all aligned) with the geometric configuration of
the points {Ai}. The main result of this section is Theorem 3.11:
1991 Mathematics Subject Classification. 53A17 (Kinematics), 14L35 (Classical groups),
14P99 (Real algebraic geometry).
Key words and phrases. Bond Theory, n-pods, self-motion.
1
arXiv:1408.6716v2  [math.AG]  16 Jan 2015
MÖBIUS PHOTOGRAMMETRY
2
Theorem. Let ⃗A and ⃗B be two 5-tuples of points in R3 such that no 4 of them
are collinear. Assume that f ⃗A(S2) and f ⃗B(S2) are equal as curves in M5. If ⃗A is
coplanar, then ⃗B is also coplanar and affine equivalent to ⃗A. If ⃗A is not coplanar,
then ⃗B is similar to ⃗A.
Eventually, Section 4 presents an application of the theory developed so far to
pentapods with mobility greater than or equal to 2. In fact, and this is where the
authors took one of the motivations for this paper, Theorem 3.19 of [2] gives a
necessary condition for mobility 2 of n-pods in the form of a disjunction of several
statements, one of which being: “there are infinitely many pairs (L, R) of elements
of S2 such that the points πL(p1), . . . , πL(pn) and πR(P1), . . . , πR(Pn) differ by an
inversion or a similarity”. We focus on this case, and using Theorem 3.11 we show
that base and platform points are either similar or planar and affine equivalent.
2. Setting up the Möbius photogrammetry problem
We are going to consider the following problem: given a vector ⃗A = (A1, . . . , A5)
of 5 points in R3, we want to define a map f ⃗A, which we will call Möbius cam-
era, associating to each direction ε ∈S2 the orthogonal projection of ⃗A along ε,
considered up to Möbius transformations. Moreover, starting from the collection
of all orthogonal projections of ⃗A, we want to understand to what extent we can
reconstruct ⃗A.
In order to set up this photogrammetric problem in a formal way, first of all we
have to make clear what do we mean by “consider up to Möbius transformations”.
Recall that a Möbius transformation is a map g of the complex projective line P1
C
to itself of the form
g :







(1 : z)
7→

1 : a z+b
c z+d

if z ̸= −d/c
(1 : −d/c)
7→
(0 : 1)
(0 : 1)
7→
(1 : a/c)
where ad −bc ̸= 0
with the convention that g(0 : 1) = (0 : 1) if c = 0. If we are given two n-tuples
(m1, . . . , mn) and (n1, . . . , nn) of points in the plane R2, we say that they are Möbius
equivalent if, once we identify R2 with C ,→P1
C, there is a Möbius transformation g
sending mi to ni for every i ∈{1, . . . , n}.
The rest of this and the next section are aimed to define the concept of Möbius
camera. We will clarify what are the domain and the codomain of this map, and
what is its explicit formulation.
We start discussing the domain of our desired map: as described before, it should
be S2, thought as the set of directions in R3. On the other hand, we would like
it to be an algebraic variety. We are going to see now that not only S2 can be
considered as a complex projective curve, but it also naturally carries the structure
of a real variety. This property will be crucial in the proofs of Subsection 3.3.
Definition 2.1. A real structure on a complex variety is a pair (X, α), where X
is a complex variety and α is an anti-holomorphic involution (see [11], Chapter 1,
Proposition 1.3).
MÖBIUS PHOTOGRAMMETRY
3
Remark 2.2. An example of a real structure is given by the complex projective
space Pn
C together with componentwise complex conjugation. One can prove
that there exist exactly two real structures on P1
C (up to isomorphism), and
they are given by the following two involutions:
(s, t) 7→(s, t)
and
(s : t) 7→(−t : s)
In particular, the fixed points of the first involution are precisely the closed
points of P1
R, while the second one does not have any fixed point. Moreover
there is a natural bijection between P1
C and S2, and under this bijection the
second involution corresponds to the antipodal map. Hence we can think of S2
as a real algebraic variety, whose anti-holomorphic involution is given by the
antipodal map.
The following result provides another identification of S2 with an algebraic curve
which is simply a different projective embedding of P1
C, but which enables us to
perform the computations needed to define the Möbius camera in a simpler way.
Lemma 2.3. There is a bijection γ : S2 −→C =

x2 + y2 + z2 = 0
	
⊆P2
C such
that the following diagram commutes:
S2
antipodal map
/
∼
=
γ

S2
∼
=
γ

C
componentwise
conjugation
/ C
Proof. Let ε ∈S2, then pick ε′, ε′′ in the orthogonal space ⟨ε⟩⊥such that
· ε′, ε′′ form an orthonormal basis of ⟨ε⟩⊥;
· ε, ε′, ε′′ form a right basis of R3, namely det ( ε ε′ ε′′ ) > 0.
If ε′ = (λ′, µ′, ν′) and ε′′ = (λ′′, µ′′, ν′′), then we consider the vector
ε′ + i ε′′ = (λ′ + i λ′′, µ′ + i µ′′, ν′ + i ν′′) ∈C3
By a direct computation one can check that the point in P2
C given by ε′ + i ε′′ lies
on C. We notice that a different choice of ε′ and ε′′ leads to the same point in P2
C.
Then the map γ : ε 7→ε′ + i ε′′ is well-defined and satisfies the requirements of the
thesis.
□
Remark 2.4. Recalling Remark 2.2, we have that S2 is in bijection with P1
C, and
via the previous map γ we obtain an isomorphism of real algebraic curves be-
tween P1
C and C given by homogeneous polynomials of degree 2. We get the
following triangle of bijections and involutions:

S2, antipodal map

i
)
o
/

x2 + y2 + z2 = 0
	
, componentwise conj.

3
s

P1
C, (s : t) 7→(−t : s)

The identification of S2 with the conic C ⊆P2
C becomes very useful when we
want to deal with orthogonal projections.
MÖBIUS PHOTOGRAMMETRY
4
Definition 2.5. Given a unit vector ε ∈S2, we say that a linear map πε : R3 −→
R2 is an orthogonal projection along ε if ker πε = ⟨ε⟩and πε is an isometry on ⟨ε⟩⊥.
Moreover we ask that the preimages of the standard bases of R2 lying on ⟨ε⟩⊥form,
together with ε, a positively oriented bases. Note that in this way πε is well-defined
only up to direct Euclidean isometry of the image, namely rotations around the
origin.
Remark 2.6. From the definition of γ given in the proof of Lemma 2.3 we see that the
vectors ε′ and ε′′ we form starting from ε ∈S2 satisfy ( ε ε′ ε′′ ) ∈SO(3, R). One
can check that ( ε′ ε′′ )t gives the matrix of πε, the orthogonal projection along ε.
Identifying R2 with the complex plane C one finds that if γ(ε) = (x : y : z) and
A = (p, q, r) is a point in R3, then πε(A) is given, as a point in C, by p x+q y+r z.
If now we change the representative of γ(ε), this modifies the image under the
orthogonal projection by possibly a rotation and a dilation.
Hence taking into account Remark 2.6 we can realize the orthogonal projection πε
as the dot product
⟨(x, y, z), ·⟩: R3 −→C
where (x : y : z) is any representative of γ(ε) with Hermitian norm equal to
√
2.
Thus we can view any orthogonal projection of n points ⃗A = (A1, . . . , An) as an
n-tuple of points in C. Then πε( ⃗A) can be seen as one single point in Cn. We can
use the embedding C ,→P1
C sending z to (z : 1) to identify πε( ⃗A) with a point
in
P1
C

n. We will extensively use this fact in the definition of the Möbius camera,
and for proving some of its properties.
In order to understand what should be the codomain of our photographic map we
start with the following known result (see, for example, [5], Chapter 2, Sections 2.1
and 2.2):
Proposition 2.7.

M¨obius transformations
	 ∼= PGL(2, C) ∼= Aut(P1
C).
Due to Proposition 2.7 we can use the natural action of PGL(2, C) on
P1
C

n to
express that two orthogonal projections are Möbius equivalent. This leads us to the
following definition, which we denote as temporary because unfortunately, despite
the fact that it looks all-embracing and clean, it will not be very useful for us.
Temporary Definition. Let ⃗A be a finite set of distinct points in R3 and ε ∈
S2. The Möbius picture of
⃗A along ε is the equivalence class, under the action
of PGL(2, C), of any orthogonal parallel projection of ⃗A along the direction ε,
considered as an n-tuple of points in P1
C.
Remark 2.8. We notice that the concept of Möbius picture is well-defined. In fact,
although the choice of different orthogonal projections along the same direction
determines different points of
P1
C

n, they all differ by a Möbius transforma-
tion (given by a rotation, as mentioned in the end of Definition 2.5), so their
equivalence class is the same.
MÖBIUS PHOTOGRAMMETRY
5
Remark 2.9. Since the group of Möbius transformations is 3-transitive, then all
configurations of n points in R2 are Möbius equivalent for 1 ≤n ≤3, so Möbius
photogrammetry can be reasonably approached only if n ≥4.
The problem with the Temporary Definition is that it involves an object, the
quotient of
P1
C

n by the action of PGL(2, C), which does not have good geometric
properties, namely it does not have a natural structure of an algebraic variety. In
Subsection 3.1 we will see that we can obtain a much better object at a fair price,
and we will focus on the case which is most interesting for us, namely n = 5. On
the other hand, this operation has a cost: we will not be able to define Möbius
pictures for any arbitrary configuration of points (see Footnote 3), but we need to
put some restrictions. However, we will see that the concept of Möbius camera is
meaningful for any configuration of points (see Remark 3.5).
3. The Möbius camera
3.1. A projective embedding of the moduli space of 5 points in P1
C. Geo-
metric Invariant Theory tells us that, in order to obtain our desired set of equiv-
alence classes under the action of PGL(2, C), which we denote by M51, we cannot
consider the equivalence classes of all 5-tuples, but we have to restrict to an open
subset of
P1
C

5 (for an introduction to this topic, see [1], in particular Chapter 6,
or [8], in particular Chapters 0 and 1). Therefore we will be forced to impose some
conditions on the vector ⃗A of points in R3, in order to ensure that it is possible
to define its Möbius picture along a given direction. After accepting this limita-
tion one can construct an embedding of the quotient in projective space defined by
invariants of the 5-tuple2. It is possible to embed M5 as a quintic surface in P5
C:
in [4] it is explained a possible way to determine this surface and the quotient
map
P1
C

5 99K M5. We briefly describe the procedure:
1. Consider a convex pentagon P in the plane, and construct all plane undi-
rected multigraphs without loops whose set of vertices coincides with the
set of vertices of P, and which satisfy the following conditions:
· edges are given by segments;
· any two edges do not intersect;
· the valency of every vertex is 2;
There are exactly 6 of these graphs, showed in Figure 1.
2. Associate to each graph G = (E, V ) a homogeneous polynomial in the
coordinates {(ai : bi)} of
P1
C

5 according to the following rules:
i. for every edge e ∈E, e = (i, j) with i < j, define
ϕe = ai bj −aj bi
1In the literature this object is usually denoted by M0,5, since it is the moduli space of genus 0
smooth curves with 5 marked points, but here we will always omit the index 0.
2For n = 4, there are two invariants, defining an open embedding M4 ,→P1
C; the quotient of
the two projective invariants is an absolute invariant, the cross ratio. This is in accordance with
the fact that the projective equivalence of two 4-tuples of points in P1
C is completely determined
by their cross ratio.
MÖBIUS PHOTOGRAMMETRY
6
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
Figure 1. The only six planar undirected multigraphs without
loops with vertices on a regular pentagon, valency 2 and non-
intersecting edges.
ii. set
ϕG =
Y
e∈E
ϕe
For example the polynomial associated to the first graph in Figure 1 is
ϕ0 = (a1 b2 −a2 b1)(a2 b3 −a3 b2)(a3 b4 −a4 b3)(a4 b5 −a5 b5)(a1 b5 −a5 b1)
3. These polynomials determine a rational map ϕ :
P1
C

5 99K P5
C.
4. Consider the open set
U =

(m1, . . . , m5) : no three of the mi coincide
	
⊆
P1
C

5
5. We have that image
ϕ|U


= M5. It turns out that if we take coordinates
t, x1, . . . , x5 in P5
C, then the equations for M5 are:
xi−2 xi+2 = t xi + t2
∀i ∈{1, . . . , 5}
where the indices are taken modulo 5.
Remark 3.1. It is known that M5 is a Del Pezzo surface of degree 5: exactly 10
lines lie on such a surface and they correspond to equivalence classes of 5-tuples
(m1, . . . , m5) for which at least two points coincide. We denote by Lij the line
in M5 corresponding to classes for which mi = mj.
Remark 3.2. The variety M5 has a canonical real structure, inherited from the real
structure of the projective line. An equivalence class is real (namely, it is a
fixed point for the anti-holomorphic involution on M5) if and only if it can
be represented by a 5-tuple of real points in P1
C. Equivalently, the points are
colinear or cocircular; in fact, we recall these two facts about automorphisms
of P1
C:
· Möbius transformations map lines and circles to lines and circles;
MÖBIUS PHOTOGRAMMETRY
7
· the action of Möbius transformations is transitive on lines and circles.
Since a 5-tuple is given by real points in P1
C if and only if the corresponding points
in R2 lie on a line, the claim follows from the previous two considerations.
3.2. Definition of the Möbius camera. From this construction we infer the
condition we have to impose on the vector ⃗A of points in R3 so that we can speak
of a Möbius picture along an arbitrary direction3: no 3 points among the {Ai}
should be aligned. In this way for every ε ∈S2 we will have that πε( ⃗A) lies on U,
hence its equivalence class is a well-defined element of M5.
Definition 3.3. Let ⃗A be a vector of 5 points in R3 and ε ∈S2. Suppose that no
three points of ⃗A lie on a line parallel to ε. The Möbius picture of ⃗A along ε is the
point in M5 given by the equivalence class (under the action of PGL(2, C)) of any
orthogonal parallel projection of ⃗A along the direction ε, considered as an n-tuple
of points in P1
C.
Our next task is to define the notion of Möbius camera, namely a function which,
once fixed a vector of points, takes a direction ε ∈S2 and associates to it the Möbius
picture of the points along that direction.
Definition 3.4. Let ⃗A be a vector of 5 points in R3.
The Möbius camera, or
photographic map, for ⃗A is the morphism of varieties given by:
f ⃗A :
C
−→
M5 ⊆P5
C
c
7→
ϕ
 πε(A1) : 1


, . . . ,
πε(A5) : 1

 
= ϕ
 ⟨c, A1⟩: 1


, . . . ,
⟨c, A5⟩: 1

 
where C is the curve {x2 + y2 + z2 = 0} ⊆P2
C, and c ↔ε under the bijection γ
established in Lemma 2.3; moreover the map ϕ :
P1
C

5 99K M5 is the quotient
map defined in Subsection 3.1. We denote by ef ⃗A the precomposition of f ⃗A by the
parametrization of C described in Remark 2.4.
Remark 3.5. The Möbius picture of ⃗A cannot be defined for those c ∈C such that
there exist three points in ⃗A lying on a line parallel to the direction defined
by c. Since the points c ∈C for which the Möbius picture is defined form an
open subset of C, then the map f ⃗A is a priori a rational one. However, since
C is a smooth curve, then f ⃗A extends to a regular map, namely it is defined
also on the points which do not admit a Möbius picture. In algebraic terms,
the polynomials defining the function have a common factor vanishing at those
points, which can be canceled.
Remark 3.6. The map ϕ is given by homogeneous polynomials of degree 5 in the
coordinates {(ai : bi)} of
P1
C

5. Hence ef ⃗A is given by homogeneous polynomials
of degree 10 in the coordinates (s : t) of P1
C.
3 This is the condition on configurations mentioned at the end of Section 2.
MÖBIUS PHOTOGRAMMETRY
8
3.3. Properties of the Möbius camera. The following lemmata describe the
behavior of the image of a Möbius camera depending on the geometry of the vec-
tor ⃗A.
Lemma 3.7. Let ⃗A = (A1, . . . , A5) be a 5-tuple of coplanar points which are not
collinear. Then the photographic map f ⃗A : C −→M5 is 2 : 1 to a rational curve of
degree 5, 4, 3, or 2 in M5.
Proof. Suppose that the Ai are coplanar: then, after a suitable change of coordi-
nates, we can assume that Ai = (pi, qi, 0) for every i, and in this case the photo-
graphic map f ⃗A factors through the restriction to C of the projection τx,y : P2
C −→
P1
C sending (x : y : z) 7→(x : y), which is a 2 : 1 map. Hence we get
C
f ⃗
A
/
τx,y

M5
P1
C
g ⃗
A
>
If we show that g ⃗A is birational, then f ⃗A is 2 : 1.
The map g ⃗A is given by 6
components, each of which is the product of five linear polynomials in x and y.
Each of these polynomials is of the form Gij = x(pi −pj) + y(qi −qj). Hence the
components of g ⃗A have the following structure:
(1)
g ⃗A


0
=
G12
G23
G34
G45
G15
g ⃗A


1
=
G12
G25
G15
G34
G34
g ⃗A


2
=
G12
G23
G13
G45
G45
g ⃗A


3
=
G23
G34
G24
G15
G15
g ⃗A


4
=
G34
G45
G35
G12
G12
g ⃗A


5
=
G14
G45
G15
G23
G23
We notice that if the lines −−−→
AiAj and −−−→
AhAk are parallel, then Gij and Ghk only differ
by a scalar multiple. Since the 5 points are not collinear, only four configurations
are allowed (after possibly relabeling the points), as shown in Figure 2.
Case (a): The components of g ⃗A do not have factors in common, so
deg
g ⃗A(P1
C)


· deg (g ⃗A) = 5
Hence either g ⃗A is a birational map to a curve of degree 5, or it is a 5 : 1 map
to a line. If the image of g ⃗A were a line, then because of Remark 3.1 that
line would coincide with one of the 10 lines of M5. This would mean that
whatever direction we use, the projections of two points always coincide in
any Möbius picture, and this is not possible. Hence this possibility must
be ruled out, obtaining that g ⃗A is birational.
Case (b): Here G12, G23 and G13 are equal up to scalar multiplication, so all
the components have one factor in common, which can be removed. Hence
deg
g ⃗A(P1
C)


· deg (g ⃗A) = 4
this leading to three possibilities: deg (g ⃗A) = 1, 2 or 4. The case when
deg (g ⃗A) is 4 can be discarded as in Case (a), so in order to prove the thesis
MÖBIUS PHOTOGRAMMETRY
9
A1
A2
A3
A4
A5
(a)
A1
A2
A3
A4
A5
(b)
A1
A2
A3
A4
A5
(c)
A1
A2
A3
A4
A5
(d)
Figure 2. Possible configurations of 5 points in the plane: (a)
no 3 points are aligned, (b) exactly 3 points are aligned, (c) 3 + 3
points are collinear, (d) exactly 4 points are collinear.
we only have to consider the situation deg (g ⃗A) = 2. In this case the image
of g ⃗A would be a conic, but from the general theory of Del Pezzo surfaces
we have the following:
Claim. g ⃗A(P1
C) cannot be a conic.
Proof. It is well known that the surface M5 contains 5 families of conics,
and every irreducible conic belongs exactly to one of them. These families
arise in the following way: fix an index i ∈{1, . . . , 5} and consider the
map M5 −→M4 ∼= P1
C sending the equivalence class of (m1, . . . , m5) to
the equivalence class of (m1, . . . , mi−1, mi+1, . . . , m5), namely we remove
the i-th point; the fibers of this map give one family of conics. From this
description, recalling the definition of the lines Lij (see Remark 3.1), we see
that the i-th family of conics intersects only 4 lines, namely Lij for j ̸= i
(recall that Lij = Lji).
On the other hand, by inspecting our current
situation, we see that the image g ⃗A(P1
C) has to intersect the lines L14, L15,
L24, L25, L34 and L35 (in general it will also intersect the line L45, but this
does not happen if −−−→
A1A3 and −−−→
A4A5 are parallel). Thus g ⃗A(P1
C) cannot be
one of the conics in M5.
Hence g ⃗A can only be birational to a curve of degree 4.
Case (c): Here G12, G23 and G13 are equal up to scalar multiplication and
the same for G14, G45 and G15. One can check that all components have
two factors in common. Thus, considering what we did in Case (a), the
only possible situation is the one in which g ⃗A is birational to a curve of
degree 3.
Case (d): In this case G12, G23, G13, G24, G34 and G14 are equal up to
scalar multiplication. One deduces that all components have three factors
MÖBIUS PHOTOGRAMMETRY
10
in common and so analogously as in Case (c) we have that g ⃗A is birational
to a curve of degree 2.
□
Lemma 3.8. Let ⃗A = (A1, . . . , A5) be a 5-tuple of points. If the {Ai} are not
coplanar, then the photographic map f ⃗A : C −→M5 is birational to a rational
curve of degree 10 or 8 in M5.
Proof. We argue as in the proof of Lemma 3.7: if we write Hij for the linear
polynomial x(pi −pj) + y(qi −qj) + z(ri −rj), then the components of f ⃗A have the
same structure as described by Equation (1), where we replace Gij by Hij. Since
the {Ai} are not coplanar, we can have only three possibilities (after a possible
relabeling of the points), showed in Figure 3.
A1
A2
A3
A4
A5
(a)
A1
A2
A3
A4
A5
(b)
A1
A2
A3
A4
A5
(c)
Figure 3. Possible configurations of 5 non coplanar points in the
space: (a) no 4 points are coplanar, (b) 4 coplanar points, no 3 of
them aligned, (c) 3 aligned points.
Case (a/b): In this situation the components
f ⃗A


i of f ⃗A do not have any
common factor, hence either f ⃗A is a birational map with image a degree 10
curve, or f ⃗A is 2 : 1 to a curve of degree 5. We prove that in the second
case the points should be coplanar, so this cannot happen. If we suppose
that the map f ⃗A is 2 : 1, we have the following:
Claim. It is possible to define a regular map r ⃗A : C −→C which respects
the real structure on C and such that
r2
⃗A = id
and
f ⃗A
r ⃗A(ε)


= f ⃗A(ε)
Proof. Suppose in fact that we are given a finite map f : C −→D where C
is a smooth curve and f is generically 2 : 1. If eD is the normalization of D,
we can lift f to a finite map ef : C −→eD which is also generically 2 : 1.
Then we define set-theoretically an involution r : C −→C in the following
way: pick a point P ∈C; in particular, P is a prime divisor of C, so we
map it to f ∗(f∗(P))−P 4, which is also a prime divisor of C, namely a point
(we passed to the normalization in order to have good functorial properties
of divisors; generically this map swaps the two elements in a fiber of f).
In order to prove that this map is regular, since the map f is generically
2 : 1 we can suppose that locally it is given by the canonical injection
4Here f∗: Div(C) −→Div(D) and f∗: Div(D) −→Div(C) denote respectively the pushfor-
ward and the pullback induced by f between the groups of divisors of the curves C and D. For
definitions and properties of these notions see, for example, [3] Appendix A.
MÖBIUS PHOTOGRAMMETRY
11
R −→R[x]⧸(x2 + bx + c) = S, where Spec(R) is an open set in eD and
Spec(S) is an open set in C. Hence r is locally given by the homomorphism
S −→S sending x 7→−b −x, which exchanges the two roots of x2 + bx + c.
In this way we see that r is regular. Moreover, if C is a real variety and f
is a real map, then also r is a real map.
If we think of C as the unit sphere S2, because of its properties r ⃗A has to
be a rotation of S2 of 180◦along an axis, which also proves that r ⃗A has
two fixed points (the intersections of S2 with the axis of rotation). Recall
the definition of the lines Lij in M5 (see Remark 3.1). Then we get that
f −1
⃗A (Lij) =
 Ai −Aj
∥Ai −Aj∥, Aj −Ai
∥Ai −Aj∥

On the other hand, if ε ∈f −1
⃗A (Lij), then also r ⃗A(ε) ∈f −1
⃗A (Lij), so there
are only two options:
i. either r ⃗A(ε) = −ε, meaning that ε lies on a great circle of S2 (the one
orthogonal to the axis determined by r ⃗A) since r ⃗A coincides with the
antipodal map only on this great circle;
ii. or r ⃗A(ε) = ε, meaning that ε is one of the two fixed points of r ⃗A.
If possibility i. happens for every Lij, this means that the direction of all
lines −−−→
AiAj lie on a great circle of S2, this implying that the points {Ai} are
coplanar. If this were not the case, since in our configuration no three Ai
are collinear in this case we have that possibility ii. can happen only for
one line Lij. Let us suppose that this line is L12: this would imply that
the points A2, A3, A4 and A5 are coplanar (in Case (a) here we would have
already reached a contradiction) and the line −−−→
A1A2 is orthogonal to the
plane on which the other points lie. On the other hand, the fact that all
lines but L12 fall on possibility i. implies that also A1, A2, A3 and A4 are
coplanar. Hence all points are coplanar. But this is in contradiction with
our assumption that the points {Ai} are not coplanar.
We have shown that in this case the only situation which is left possible is
that the map f ⃗A is birational to a degree 10 curve.
Case (c): Here we have that H12, H23 and H13 are equal up to a scalar
factor, so the components of f ⃗A have one factor in common, which can be
removed. Thus four situations are possible: either f ⃗A is birational to a
curve of degree 8, or it is 2 : 1 to a curve of degree 4, or it is 4 : 1 to a
conic, or it is 8 : 1 to a line. Arguments similar to the ones of Case (a) in
the proof of Lemma 3.7 rule out the last two situations. In order to prove
that the 2 : 1 situation is not possible, we proceed as in Case (a): here the
image f ⃗A(C) does not meet all the lines Lij, but from the configuration of
the points Ai it is ensured that the curve intersects L14, L24, L34, L15, L25,
L35 and L45, which is enough to prove that the points are coplanar.
□
MÖBIUS PHOTOGRAMMETRY
12
Remark 3.9. We notice that if all the 5 points of ⃗A are aligned, then all the 6
components of the photographic map are proportional, hence f ⃗A is a constant
map.
Remark 3.10. Lichtblau stated the following conjecture (see Conjecture 2 of [6]; this
was later proved by him in [7], Proposition 3 and Theorem 4): there can only
exist an infinite number of cylinders of revolution passing through five distinct
points in R3 if the points are located on two parallel lines. Under the assumption
that infinitely many circular cylinders are real we can use Lemma 3.8 to give
an alternative proof to the one of Lichtblau. In fact, if a 5-tuple has such a
property, then the image of its photographic map will have infinitely many real
points (see Remark 3.2). Since S2 does not have real points, it follows that the
photographic map cannot be birational, hence the points have to be coplanar.
From this it is well known that the points actually have to lie on two parallel
lines.
Therefore only the question if this condition is also necessary for the
existence of infinitely many circular cylinders over C passing through five real
distinct points remains open.
Now we state and prove the main result of this section.
Theorem 3.11. Let ⃗A and ⃗B be two 5-tuples of points in R3 such that no 4 of them
are collinear. Assume that f ⃗A(C) and f ⃗B(C) are equal as curves in M5. If ⃗A is
coplanar, then ⃗B is also coplanar and affine equivalent to ⃗A. If ⃗A is not coplanar,
then ⃗B is similar to ⃗A.
Proof. Suppose that ⃗A is not coplanar. Then by Lemma 3.8 we know that f ⃗A is
birational to a curve of degree 10 or 8. From Lemma 3.7 we have that ⃗B is also not
coplanar, since otherwise we would have a curve of different degree as the image
of f ⃗B. Thus f ⃗B is birational, and by composing ef ⃗A and ef −1
⃗B
we get an isomorphism
ρ : P1
C
∼
=
−→P1
C which respects the real structure, since both ef ⃗A and ef ⃗B do so. Thus
ρ is a rotation of S2. If we apply the rotation ρ to ⃗A we obtain a vector of points ⃗A′
so that the diagram
D
P1
C
e
f ⃗
A′
?
id
/ P1
C
e
f ⃗
B
_
commutes, namely ef ⃗A and ef ⃗B coincide as maps. The goal now is to show that the
direction −−−→
A′
iA′
j and −−−→
BiBj coincide for every i and j, this proving that ⃗A′ and ⃗B are
similar, from which we derive the thesis. Let us consider the situation when D has
degree 10. Recall the definition of the lines Lij in M5 (see Remark 3.1). Analogously
as in the proof of Lemma 3.8 we have that
f −1
⃗
A′ (Lij) =
(
A′
i −A′
j


A′
i −A′
j


,
A′
j −A′
i


A′
i −A′
j


)
and similarly for f −1
⃗B (Lij). Since the two maps ef ⃗
A′ and ef ⃗B coincide our claim is
proved. In the situation when D has degree 8 the argument is the same, but in this
MÖBIUS PHOTOGRAMMETRY
13
case D does not intersect all the lines Lij; however, knowing that f −1
⃗
A′ (D ∩Lij) and
f −1
⃗B (D∩Lij) are equal for ij ∈{14, 24, 34, 15, 25, 35, 45} (see Case (c) of Lemma 3.8)
gives already enough information for proving that ⃗A′ and ⃗B are similar.
Suppose that ⃗A is coplanar, then from Lemma 3.7 the map f ⃗A is 2 : 1 to a
curve of degree 5, 4 or 3 (we avoid the conic case, since no 4 points are collinear
by hypothesis; the reason for this is clarified in Remark 3.13).
Hence the only
possibility is that also ⃗B is coplanar, because otherwise from Lemma 3.8 we would
get a curve of degree 10 or 8 as the image of f ⃗B. As in the proof of Lemma 3.7, we
know that both f ⃗A and f ⃗B factor through a 2 : 1 map to P1
C followed by a birational
map. By a change of coordinates we can suppose that this 2 : 1 map is given by
sending (x : y : z) 7→(x : y). The picture of the situation is:
D
C
2:1
/
f ⃗
A
7
P1
C
∼
=
?
P1
C
∼
=
_
C
2:1
o
f ⃗
B
g
Thus we get an isomorphism P1
C
∼
=
−→P1
C which makes the previous diagram com-
mute. If M is the invertible 2 × 2 matrix representing it, and we denote by A′ the
vector of points obtained by applying the affinity associated to M to ⃗A, then the
following diagram commutes:
D
C
/
f ⃗
A′
=
P1
C
O
C
o
f ⃗
B
a
(x:y:z) 
/ (x:y)
(x:y:z)

o
In this way we reached the point where f ⃗
A′ and f ⃗B are equal as maps, thus we can
proceed as in the non planar case, proving that A′ and B are similar, so ⃗A and ⃗B
are affine equivalent.
□
Remark 3.12. We can describe an algorithm which takes as an input the image
of the photographic map of a vector of points ⃗A satisfying the conditions of
Theorem 3.11 and gives back a vector of points ⃗C which is similar to ⃗A. In
Algorithm 1 we describe the procedure in the case of non planar points, when
the degree of D is 10. This is the easiest situation, because we have information
about all the directions of the lines passing through the points of ⃗A.
We notice that we can always perform Steps 5, 6 and 7, namely, the involved
lines always intersect. This is ensured by the fact that we start from an existing
configuration of points.
When the curve D has degree 8, 5, 4 or 3 the algorithm is almost the same,
we just have to take into account that D will not intersect all the lines Lij: the
ones which are disjoint from the image of f ⃗A reveal which points in ⃗C will be
collinear, and the others can be used to identify the whole configuration.
MÖBIUS PHOTOGRAMMETRY
14
Algorithm 1 Non planar point reconstruction
Input: D ⊆M5, a degree 10 curve such that f ⃗A(S2) = D.
Output: ⃗C such that it is similar to ⃗A.
1: Parametrize D via ϕ respecting the real structure of D.
2: Compute

εij, −εij
	
= ϕ−1(Lij) for all i, j.
3: Set C1 = (0, 0, 0).
4: Pick C2 arbitrary on the line

C1 + t ε12 : t ∈R
	
.
5: Construct C3 as the intersection of the lines

C1 + t ε13
	
and

C2 + t ε23
	
.
6: Construct C4 using ε24 and ε34 as in Step 5.
7: Construct C5 using ε35 and ε45 as in Step 5.
8: Return ⃗C = (C1, . . . , C5).
Remark 3.13. We notice that we have to avoid the case when 4 points are collinear
(namely when the degree of the image of the photographic map is 2), because in
that case it is not possible to reconstruct the direction −−−→
A1A4. In fact, f ⃗A(C) ∩
L14 = ∅since projecting in that direction would give a configuration where four
points coincide, which is not allowed in M5. In this case one can show that the
images of two photographic maps f ⃗A(C) and f ⃗B(C) are equal if and only if the
cross ratios of the two 4-tuples of collinear points are equal. On the other hand,
also when we only have three aligned points the image of the photographic map
does not intersect the line L13, but in this case we can reconstruct the whole
configuration regardless the knowledge of −−−→
A1A3, since we can use −−−→
A1A4 and
−−−→
A1A5 to determine A1 starting from A4 and A5, and do the same for A2 and
A3 — this procedure cannot be applied to the previous configuration. The two
situations are described in Figure 4.
A1
A2
A3
A4
A5
(a)
A1
A2
A3
A4
A5
(b)
Figure 4. In the case of (a) four collinear points, the reconstruc-
tion algorithm does not work, since it is not possible to recover the
direction of the line on which the four points lie. Instead, if we
only allow (b) three collinear points, then the algorithm succeeds
since we can reconstruct the aligned points using the other ones.
Eventually, it is possible to extend the consequences of Theorem 3.11 to tuples
of n points when n > 5. In order to do this, starting from such an n-tuple ⃗A one
can define a photographic map f ⃗A : C −→Mn, where Mn is the moduli space of n
MÖBIUS PHOTOGRAMMETRY
15
points in P1
C, in the same way as we did in this paper. Then for every sub-tuple
of 5 elements of ⃗A, say (A1, . . . , A5), one has a commutative diagram:
(2)
C
f ⃗
A
/
f(A1,...,A5)
 
Mn
δ(1,...,5)
}
M5
where δ(1,...,5) associates the equivalence class of the n-tuple (m1, . . . , mn) to the
equivalence class of the 5-tuple (m1, . . . , m5) (this is a rational map).
Corollary 3.14. Theorem 3.11 holds true also if we take ⃗A and ⃗B to be two n-
tuples of points in R3 where no n −1 points are collinear, provided that n ≥5.
Proof. We prove the statement by reducing to the n = 5 case and applying The-
orem 3.11. Suppose that ⃗A is not coplanar; we want to prove that ⃗A and ⃗B are
similar. After possibly relabeling the points, we can suppose that A1, . . . , A4 are
not coplanar. By hypothesis we have that f ⃗A(C) = f ⃗B(C), so from Diagram 2 we
can infer that for every k ≥5 we have f(A1,...,A4,Ak)(C) = f(B1,...,B4,Bk)(C). Thus
by Theorem 3.11 we get that for every k ≥5, the two 5-tuples (A1, . . . , A4, Ak) and
(B1, . . . , B4, Bk) are not coplanar and similar. Now, since in this case there exists a
unique similarity sending (A1, . . . , A4) to (B1, . . . , B4), from what we said we have
that the same similarity sends Ak to Bk for all k ≥5. Hence ⃗A and ⃗B are similar.
If ⃗A is coplanar, then from the commutativity of Diagram 2 and by Theorem 3.11
we obtain that also ⃗B is coplanar. Now we can proceed as before to get the thesis,
but here in order to be able to use Theorem 3.11 we have to make sure that we
can choose A1, . . . , A4 so that for every k ≥5 there are no 4 collinear points among
A1, . . . , A4, Ak. This is ensured by the hypothesis that no n−1 among the {Ai} are
collinear, since the latter is the only case when this choice cannot be made. Hence
we can conclude as before, since an affinity is completely determined by the image
of 3 non collinear points.
□
4. A necessary condition for pentapods with mobility 2
We can finally apply the theory we developed so far to get necessary conditions
for mobility of pentapods. The geometry of this kind of mechanical manipulators
is defined by the coordinates of the 5 platform anchor points p1, . . . , p5 ∈R3 and
of the 5 base anchor points P1, . . . , P5 ∈R3 in one of their possible configurations.
All pairs of points (pi, Pi) are connected by a rigid body, called leg, so that for all
possible configurations the distance di = ∥pi −Pi∥is preserved. The dimension of
the set of possible configurations of a pentapod is called its mobility (for a formal
definition of this concept, see [2], Section 3, Definition 3.2).
In [2] we proved the following conditions for n-pods (replace 5 by n in the previous
paragraph):
Theorem 4.1. Let Π be an n-pod with mobility 2 or higher.
Then one of the
following holds:
MÖBIUS PHOTOGRAMMETRY
16
(a) there are infinitely many pairs (L, R) of elements of S2 such that the points
πL(p1), . . . , πL(pn) and πR(P1), . . . , πR(Pn) differ by an inversion or a sim-
ilarity;
(b) there exists m ≤n such that p1, . . . , pm are collinear and Pm+1 = . . . = Pn,
up to permutation of indices and interchange between base and platform;
(c) there exists m ≤n with 1 < m < n −1 such that p1, . . . , pm lie on a
line g ⊆R3 and pm+1, . . . , pn are located on a line g′ ⊆R3 parallel to g;
moreover P1, . . . , Pm lie on a line G ⊆R3 and Pm+1, . . . , Pn are located on
a line G′ ⊆R3 parallel to G, up to permutation of indices.
For n = 5 we can use our Möbius Photogrammetry technique to reformulate
condition (a) above in a more geometric fashion.
Theorem 4.2. Let Π a pentapod with mobility 2 or higher. Then one of the fol-
lowing conditions holds:
(a) the platform and the base are similar;
(b) the platform and the base are planar and affine equivalent;
(c) there exists m ≤5 such that p1, . . . , pm are collinear and Pm+1, . . . , P5
coincide, up to permutation of indices and interchange of platform and
base;
(d) the points p1, p2, p3 lie on a line g ⊆R3 and p4, p5 lie on a line g′ ⊆R3
parallel to g, and P1, P2, P3 lie on a line G ⊆R3 and P4, P5 lie on a line
G′ ⊆R3 parallel to G, up to permutation of indices.
Proof. Since Π has mobility at least 2, then by Theorem 4.1 either we are in cases (c)
or (d), or there are infinitely many pairs (L, R) of elements of S2 such that the points
πL(p1), . . . , πL(p5) and πR(P1), . . . , πR(P5) differ by an inversion or a similarity. Let
us consider then this last case. Since we can suppose that no 4 point of the base
or platform are aligned (otherwise we are in case (c) or (d)), we have in particular
that the photographic maps f ⃗P and f⃗p of base and platform points of Π are not
constant. Hence, if we re-interpret the assumption in the language we developed
in this paper, we have that the images f ⃗P (C) and f⃗p(C) have infinitely points in
common. Since both are irreducible algebraic curves, they must coincide, and we
get (a) or (b) by Theorem 3.11.
□
Remark 4.3. For quadropods the analogous statement of Theorem 4.2 does not hold.
In fact, all quadropods have mobility at least 2, but the general quadropod does
not fulfill any of the conditions (a)–(d) of the theorem. For tripods the statement
is trivially true, since conditions (b) and (c) are always fulfilled.
For n-pods with n > 5 one can prove a statement analogous to Theorem 4.2
by using Corollary 3.14.
Based on Theorem 4.2 a complete classification of pentapods with mobility 2
was given in [9] and [10].
MÖBIUS PHOTOGRAMMETRY
17
Acknowledgments
The first and third author’s research is supported by the Austrian Science Fund
(FWF): W1214-N15/DK9 and P26607 - “Algebraic Methods in Kinematics: Motion
Factorisation and Bond Theory”. The second author’s research is funded by the
Austrian Science Fund (FWF): P24927-N25 - “Stewart Gough platforms with self-
motions”.
References
[1] Igor Dolgachev. Lectures on invariant theory, volume 296 of London Mathematical Society
Lecture Note Series. Cambridge University Press, Cambridge, 2003.
[2] Matteo Gallet, Georg Nawratil, and Josef Schicho. Bond theory for pentapods and hexapods.
Journal of Geometry, pages 1–18, 2014. Accepted [doi 10.1007/s00022-014-0243-1].
[3] Robin Hartshorne. Algebraic geometry. Springer-Verlag, New York-Heidelberg, 1977. Gradu-
ate Texts in Mathematics, No. 52.
[4] Benjamin Howard, John Millson, Andrew Snowden, and Ravi Vakil. The equations for the
moduli space of n points on the line. Duke Math. J., 146(2):175–226, 2009.
[5] Gareth A. Jones and David Singerman. Complex functions. An algebraic and geometric view-
point. Cambridge University Press, Cambridge, 1987.
[6] Daniel Lichtblau. Cylinders through five points: Complex and Real Enumerative geometry.
In Automated Deduction in Geometry, volume 4869 of Lecture Notes in Computer Science,
pages 80–97. Springer, Berlin, 2007.
[7] Daniel Lichtblau. Cylinders through five points: Computational Algebra and Geometry. J.
Math. Res., 4(6):65–82, 2012.
[8] D. Mumford, J. Fogarty, and F. Kirwan. Geometric invariant theory, volume 34 of Ergebnisse
der Mathematik und ihrer Grenzgebiete (2) [Results in Mathematics and Related Areas (2)].
Springer, Berlin, third edition, 1994.
[9] Georg Nawratil and Josef Schicho. Pentapods with Mobility 2. ASME Journal of Mechanisms
and Robotics. In press [doi: 10.1115/1.4028934].
[10] Georg Nawratil and Josef Schicho. Self-motions of pentapods with linear platform. Submitted
[arXiv:1407.6126].
[11] Robert Silhol. Real algebraic surfaces, volume 1392 of Lecture Notes in Mathematics.
Springer, Berlin, 1989.
Research Institute for Symbolic Computation, Johannes Kepler University, Al-
tenberger Strasse 69, 4040 Linz, Austria.
E-mail address: mgallet@risc.jku.at
Institute of Discrete Mathematics and Geometry, Vienna University of Technol-
ogy, Wiedner Hauptstrasse 8-10/104, 1040 Vienna, Austria.
E-mail address: nawratil@geometrie.tuwien.ac.at
Research Institute for Symbolic Computation, Johannes Kepler University, Al-
tenberger Strasse 69, 4040 Linz, Austria.
E-mail address: josef.schicho@risc.jku.at