Parallel Postulates Revised

Parallel postulates and continuity axioms: a mechanized
study in intuitionistic logic using Coq

Pierre Boutry, Charly Gries, Julien Narboux, Pascal Schreck
To cite this version:

Pierre Boutry, Charly Gries, Julien Narboux, Pascal Schreck. Parallel postulates and continuity
axioms: a mechanized study in intuitionistic logic using Coq. Journal of Automated Reasoning,
Springer Verlag, 2017, pp.68. <10.1007/s10817-017-9422-8>. <hal-01178236v2>
HAL Id: hal-01178236

https://hal.inria.fr/hal-01178236v2
Submitted on 8 Mar 2017
HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Noname manuscript No.
(will be inserted by the editor)
Parallel postulates and continuity axioms: a mechanized

study in intuitionistic logic using Coq.
Pierre Boutry · Charly Gries · Julien

Narboux · Pascal Schreck
the date of receipt and acceptance should be inserted later
Abstract In this paper we focus on the formalization of the proofs of equiva-

lence between different versions of Euclid’s 5th postulate. Our study is performed
in the context of Tarski’s neutral geometry, or equivalently in Hilbert’s geome-
try defined by the first three groups of axioms, and uses an intuitionistic logic,
assuming excluded-middle only for point equality. Our formalization provides a
clarification of the conditions under which different versions of the postulates are
equivalent. Following Beeson, we study which versions of the postulate are equiv-
alent, constructively or not. We distinguish four groups of parallel postulates. In
each group, the proof of their equivalence is mechanized using intuitionistic logic
without continuity assumptions. For the equivalence between the groups additional
assumptions are required. The equivalence between the 34 postulates is formal-
ized in Archimedean planar neutral geometry. We also formalize a theorem due to
Szmiliew. This theorem states that, assuming Archimedes’ axiom, any statement
which hold in the Euclidean plane and does not hold in the Hyperbolic plane is
equivalent to Euclid’s 5th postulate. To obtain all these results, we have devel-
oped a large library in planar neutral geometry, including the formalization of the
concept of sum of angles and the proof of the Saccheri-Legendre theorem, which
states that assuming Archimedes’ axiom, the sum of the angles in a triangle is at
most two right angles.
Keywords Euclid, parallel postulate, formalization, neutral geometry, Coq,

classification, foundations of geometry, decidability of intersection, Aristotle’s
axiom, Archimedes’ axiom, Saccheri-Legendre theorem, sum of angles
1 Introduction
In this paper we focus on the formalization of results about Euclid’s 5th postulate:
ICube, UMR 7357 CNRS, University of Strasbourg

Pôle API, Bd Sébastien Brant, BP 10413, 67412 Illkirch, France
E-mail: {boutry, narboux, schreck}@unistra.fr, [email protected]
2 Pierre Boutry et al.
“If two lines are drawn which intersect a third in such a way that the sum of
the inner angles on one side is less than two right angles, then the two lines
inevitably must intersect each other on that side if extended far enough.”
This postulate is of historical importance because for centuries, many mathe-
maticians believed that this statement was rather a theorem which could be de-
rived from the first four Euclid’s postulates. History is rich with incorrect proofs
of Euclid’s 5th postulate. In 1763, Klügel provided, in his dissertation written un-
der the guidance of Kästner, a survey of about 30 attempts to “prove” Euclid’s
parallel postulate” [Klu63]. Legendre published a geometry textbook Eléments de
géométrie in 1774. Each edition of this popular book contained an (incorrect) proof
of Euclid’s postulate. Even in 1833, one year after the publication by Bolyai of
an appendix about non-euclidean geometry, Legendre was still convinced of the
validity of his proofs of Euclid’s 5th postulate:
“Il n’en est pas moins certain que le théorème sur la somme des trois angles
du triangle doit être regardé comme l’une de ces vérités fondamentales qu’il
est impossible de contester, et qui sont un exemple toujours subsistant de la
certitude mathématique qu’on recherche sans cesse et qu’on n’obtient que
bien difficilement dans les autres branches des connaissances humaines.”1
– Adrien Marie Legendre [Leg33]
These proofs are incorrect for different reasons. Some proofs rely on an as-
sumption which is more or less explicit but that the author takes for granted.
Some other proofs are incorrect because they rely on a circular argument.
Proving the equivalence of different versions of the parallel postulate requires
extreme rigor, as Richard J. Trudeau has written:
“Pursuing the project faithfully will require that we take the extreme mea-
sure of shutting out the entreaties of our intuitions and imaginations - a
forced separation of mental powers that will quite understandably be con-
fusing and difficult to maintain [...].”
– Richard J. Trudeau [Tru86]
To help us in this task, we have a perfect tool which possesses no intuition:

a computer. In this paper we provide formal proofs, verified using the Coq proof
assistant, that 34 different versions of Euclid’s 5th postulate are equivalent in
the theory defined by a subset of the axioms of Tarski’s geometry, namely the
2-dimensional neutral geometry2 using Archimedes’ axiom. We also provide more
precise results showing the equivalence in intuitionist logic of four groups of axioms
without any continuity assumption.
1 “It is no less certain that the theorem on the sum of the three angles of the triangle must
be regarded as one of those fundamental truths which is impossible to dispute and which
are an enduring example of mathematical certitude, which one continually pursues and which
one obtains only with difficulty in the other branches of human knowledge.” The English
translation is borrowed from [LP13].
2 Neutral geometry designates the set of theorems which are valid in both Euclidean and
hyperbolic geometry. Therefore, for any given line and any given point, there exists at least
a line parallel to this line and passing through this point. This renders elliptic geometry
inconsistent with neutral geometry.
Parallel postulates and continuity axioms: a mechanized study. 3
Our formal proofs rely on the systematic development of geometry based on

Tarski’s system of geometry [SST83] that Schwabhäuser, Szmielew and Tarski pro-
duced. Those results have been formalized previously [Nar07, BN12, BN17, BBN16a]
using the Coq proof assistant, and completed by some new results in neutral geom-
etry for this study. The equivalence between 26 versions of Euclid’s 5th postulate
can be found in [Mar98]. Greenberg also proves (or leaves as exercises) the equiv-
alence between several versions of the parallel postulate [Gre93]. However, these
proofs are not checked mechanically and sometimes only sketched. Moreover, since
we restrict ourselves to intuitionist logic and we use continuity axioms only when
necessary, we could not reuse directly all these proofs in our context, because some
proofs in these books use the law of excluded middle or a continuity axiom (see
Section 3).
Following the classical approach to prove that Euclid’s 5th postulate is not
a theorem of neutral geometry, Timothy Makarios has provided a formal proof
of the independence of Tarski’s Euclidean axiom [Mak12]. He used the Isabelle
proof assistant to construct the Klein-Beltrami model, where the postulate is not
verified. A close result is due to Filip Marić and Danijela Petrović who formal-
ized the complex plane using the Isabelle/HOL proof assistant [MP14]. Recently,
Michael Beeson has also studied the equivalence of different versions of the parallel
postulate in the context of a constructive geometry [Bee16].
The paper is structured as follows. In Section 2, we describe the axiom system
that we use and its formalization in Coq. In Section 3, we give an overview of the
results that we formalized, based on four example postulates, each representing a
group of postulate which are equivalent. Then, in the following Sections 4, 5, 6
and 7 (one for each group of postulates), we give the precise statements and an
overview of the proofs. The list of all the studied postulates is given in appendix A,
and the summary of the main definitions and notations is given in appendix B.
2 The context
In this section, we will first present the axiomatic system that we used as a basis
for our proofs.
It is crucial to be precise about the context and the definitions. This paper
is about equivalence properties, but an equivalence is relative to a theory and
a logic. We prove the equivalences within the higher order intuitionistic logic of
Coq and using Tarski’s axiom system for neutral geometry. Using the language
of logic, the assertion saying that P and Q are equivalent formally means3 that :
T |= P ⇔ Q holds for a given theory T . It could be the case that T |= P ⇔ Q
holds because both T |= P and T |= Q hold. For every version P of the parallel
postulates presented in this paper it is also true that T 6|= P , but we do not
prove the independence results. Defining accurately the theory T is of primary
3 The meaning of equivalence does not seem to be clear for everyone. For instance, in the
French version of the wikipedia page about the parallel postulate (april 2015) one can read that
(our own translation) : “These propositions are roughly equivalent to the axiom of parallels. By
equivalent, we mean that using some adapted vocabulary, these axioms are true in Euclidean
geometry but not true in elliptic nor spherical geometry”. Thanks to a theorem of Szmielew,
it is true that this definition of equivalence is equivalent to the logical one when you have the
continuity axiom, but it is far from obvious.
interest because the equivalence results depend on the precise definition of T . For
example, Millman and Parker [MP81, p. 226] have shown that the Pythagorean
theorem is equivalent to the parallel postulate in the context of an axiom system
in the style of Birkhoff axioms based on a ruler and protractor [Bir32]. Yet there
is a Cartesian plane over a (non-archimedean) Pythagorean field4 which does not
verify the parallel postulate: see example 18.4.3 [Har00, p. 161]).
2.1 A set of axioms for neutral geometry
Proofs are given within Tarski’s system of neutral geometry. We adopted the ax-
ioms given in [SST83] excluding the axiom corresponding to Euclid’s 5th postulate.
For an explanation of the axioms and their history see [TG99]. Table 1 lists the
axioms for neutral geometry. The consistency of this axiom system has been me-
chanically proven by Makarios [Mak12]. As we already mentioned, Makarios has
also proven formally the independence of the parallel postulate in this axiom sys-
tem using the Klein-Beltrami model [BS60]. One should notice that he could have
employed the Poincaré disk model, which also satisfies the axioms of Tab. 1. These
two properties make Tarski’s system of geometry suitable for proofs of equivalence
of statements of Euclid’s parallel postulate.
Let us recall that Tarski’s axiom system is based on a single primitive type
depicting points and two predicates, namely congruence and betweenness. AB≡CD
states that the segments AB and CD have the same length. A B C means that
A, B and C are collinear and B is between A and C (and B may be equal to A or
C ).
A1 Symmetry AB ≡ BA
A2 Pseudo-Transitivity AB ≡ CD ∧ AB ≡ EF ⇒ CD ≡ EF
A3 Cong Identity AB ≡ CC ⇒ A = B
A4 Segment construction ∃E, A B E ∧ BE ≡ CD
A5 Five-segment AB ≡ A0 B 0 ∧ BC ≡ B 0 C 0 ∧
AD ≡ A0 D0 ∧ BD ≡ B 0 D0 ∧
A B C ∧ A0 B 0 C 0 ∧ A 6= B ⇒ CD ≡ C 0 D0
A6 Between Identity A B A⇒A=B
A7 Inner Pasch A P C ∧ B Q C ⇒ ∃X, P X B ∧ Q X A
A8 Lower Dimension ∃ABC, ¬A B C ∧ ¬B C A ∧ ¬C A B
A9 Upper Dimension AP ≡ AQ ∧ BP ≡ BQ ∧ CP ≡ CQ ∧ P 6= Q
⇒ A B C ∨ B C A ∨ C A B.
Table 1: Tarski’s axiom system for neutral geometry.
The symmetry axiom (A1 on Tab.1) for equidistance together with the transi-
tivity axiom A2 for equidistance imply that the equidistance relation is an equiv-
alence relation between pair of points.
The identity axiom for equidistance A3 ensures that only degenerated segments
can be congruent to a degenerated segment.
4 A Pythagorean field is a field in which every sum of two squares is a square.

C D
C
D
B E
E A
B
A
Fig. 1: Axiom of segment construction A4.
The axiom of segment construction A4 allows to extend a segment by a given

length (Fig. 15 ).
D D0 D0
A B C A0 B0 C 0
A B C A0 B 0 C0
Fig. 2: Five-segment axiom A5.
The five-segment axiom A5 corresponds to the well-known Side-Angle-Side

postulate but is expressed with the betweenness and congruence relations only.
The lengths of AB , AD and BD and the fact that A B C fix the angle ∠CBD
(Fig. 2).
The identity axiom for betweenness A6 expresses that the only possibility to
have B between A and A is to have A and B equal. It also insinuates that the
relation of betweenness is non-strict, unlike Hilbert’s one. As Beeson suggests
in [Bee15], this choice was probably made to have a reduced number of axioms by
allowing degenerated cases of the Pasch’s axiom.
The inner form of Pasch’s axiom A7 is the axiom Pasch introduced in [Pas76]
to repair the defects of Euclid. It intuitively says that if a line meets one side of a
triangle, then it must meet one of the other sides of the triangle. There are three
forms of this axiom. Thanks to Gupta’s thesis [Gup65], one knows that the inner
form and the outer form of this axiom are equivalent and that both of them allow
5 We will provide figures both in the Euclidean model and a non-Euclidean model, namely
the Poincaré disk model. The figure on the left hand side will illustrate the validity of the
axiom or lemma in Euclidean geometry. The figure on the right hand side will either depict
the validity of the statement in the Poincaré disk model or exhibit a counter-example. We
exhibit a counter-example for statements which are equivalent to the parallel postulate.
C
C
P Q
P X
X Q
A A
Fig. 3: Pasch’s axiom A7.
us to prove the weak form. The inner form enunciates Pasch’s axiom without any
case distinction. Indeed, it indicates that the line BP must meet the triangle ACQ
on the side AQ, as Q is between B and C (Fig. 3).
The lower 2-dimensional axiom A8 asserts that the existence of three non-
collinear points.
P
P C C
A
B Q
A B
0
C
Q
Fig. 4: Upper dimensional axiom A9.
The upper 2-dimensional axiom A9 means that all the points are coplanar.
Since A, B and C are equidistant to P and Q, which are different, they belong to
the hyperplane consisting of all the points equidistant to P and Q. Because the
upper 2-dimensional axiom specifies that A, B and C are collinear, this hyperplane
is of dimension one and it fixes the dimension of the space to two. It forbids the
existence of the point C 0 (Fig. 4).
2.1.1 Formalization in Coq
Contrary to the formalization of Hilbert’s axiom system [DDS01, BN12, BBN16b],

which leaves room for interpretation of natural language, the formalization in
Coq of Tarski’s axiom system is straightforward, as the axioms are stated very
precisely. We defined the axiom system using three type classes (Tab. 2). The first
class regroups the axioms for neutral geometry in any dimension greater than or
equal to two (A1-A8).
With the second class, we also assume that we can reason by cases on the point
equality (point equality decidability). This axiom does not appear in [SST83],
although reasoning by cases on point equality is done as soon as the second chap-
ter (the first chapter being dedicated to the axioms), because it is a tautology
in classical logic, while the logic of Coq is intuitionist. We say that a predicate
is decidable when it verifies the excluded middle property. We have shown pre-
viously [BNSB14a], by modifying and reordering the proofs of [SST83] that we
had formalized that decidability of point equality implies decidability of the main
predicates defined in [SST83] besides the intersection of two lines.
Finally, the third class corresponds to the axioms of planar neutral geometry
(A1-A9) with excluded middle for point equality.
We have proven formally that Hilbert’s axioms Group I, II and III are bi-
interpretable with Tarski’s axiom A1-A9 [BN12, BBN16b]. Hence, the result pre-
sented in this paper are also valid in this axiom system, the models of which are
called Hilbert-planes by Hartshorne [Har00].
3 Four categories of parallel postulates
In this paper, we classify different statements of the parallel postulate into four
categories. Throughout this section, we will focus on one postulate for each of
these four categories. These four main postulates are equivalent in Archimedean
neutral geometry using classical logic. However, in an intuitionistic logic, and in a
non-Archimedean context, they are not equivalent.
3.1 Independent parallel postulates
Let us consider four versions of the parallel postulate.

1. The first postulate that we present was chosen by Tarski in [Tar51] and re-
tained in [SST83]. Therefore, we will refer to it as Tarski’s parallel postulate.
It expresses that given a point D between the points B and C and a point T
further away from A than D on the half line AD, one can build a line which goes
through T and intersects the sides BA and BC of the angle ∠ABC respectively
further away from B than A and C (Fig. 13).
2. The second postulate that we will study in this paper was adopted by Hilbert
in [Hil60] and is known as Playfair’s postulate. It states that there is a unique
parallel to a given line going through some point (Fig. 14).
3. The third postulate, which we will designate as the triangle postulate, corre-
sponds to the implicit assumption made by Legendre in the quote by him from
the introduction. It asserts that the sum of the interior angles of a triangle is
equal to two right angles (Fig. 15).
4. Finally, the fourth postulate is due to Bachmann [Bac64]. Following Pambuc-
cian [Pam09], we will refer to it as Bachmann’s Lotschnittaxiom. It formulates
that given the lines l, m, r and s, if l and r are perpendicular, r and s are
perpendicular and s and m are perpendicular, then l and m must meet.
In classical logic, these four postulates are equivalent in Archimedean neutral
geometry. By Archimedean planar neutral geometry we mean neutral geometry in
Class Tarski_neutral_dimensionless :=
{
Tpoint : Type;
Bet : Tpoint -> Tpoint -> Tpoint -> Prop;
Cong : Tpoint -> Tpoint -> Tpoint -> Tpoint -> Prop;
cong_pseudo_reflexivity : forall A B, Cong A B B A;
cong_inner_transitivity : forall A B C D E F,
Cong A B C D -> Cong A B E F -> Cong C D E F;
cong_identity : forall A B C, Cong A B C C -> A = B;
segment_construction : forall A B C D,
exists E, Bet A B E /\ Cong B E C D;
five_segment : forall A A’ B B’ C C’ D D’,
Cong A B A’ B’ ->
Cong B C B’ C’ ->
Cong A D A’ D’ ->
Cong B D B’ D’ ->
Bet A B C -> Bet A’ B’ C’ -> A <> B -> Cong C D C’ D’;
between_identity : forall A B, Bet A B A -> A = B;
inner_pasch : forall A B C P Q,
Bet A P C -> Bet B Q C ->
exists X, Bet P X B /\ Bet Q X A;
lower_dim : exists A, exists B, exists C,
~ (Bet A B C \/ Bet B C A \/ Bet C A B)
}.
Class Tarski_neutral_dimensionless_with_decidable_point_equality
‘(Tn : Tarski_neutral_dimensionless) :=
{
point_equality_decidability : forall A B : Tpoint, A = B \/ ~ A = B
}.
Class Tarski_2D
‘(TnEQD : Tarski_neutral_dimensionless_with_decidable_point_equality) :=
{
upper_dim : forall A B C P Q,
P <> Q -> Cong A P A Q -> Cong B P B Q -> Cong C P C Q ->
(Bet A B C \/ Bet B C A \/ Bet C A B)
}.
Table 2: Formalization of the axiom system in Coq .
which Archimedes’ axiom6 holds. Archimedes’ axiom is a corollary of the continuity

axiom of Tarski (A11, which we do not present here) which can be expressed in the
following way. Given two segments AB and CD such that A is different from B ,
there exist some positive integer n and n + 1 points A1 , · · · , An+1 on line CD, such
that Aj is between Aj−1 and Aj +1 for 2 < j < n, Aj Aj +1 and AB are congruent
for 1 < j < n, A1 = C and D is between An and An+1 .
Nevertheless, by weakening the theory, this equivalence ceases to hold. We
presented these postulates, which fall into four distinct categories, in decreasing
order of strength.
6 In constrast to Euclid, we treat the words “postulate” and “axiom” as synonyms. However,
we will restrict the use of the word “postulate” to statements of the parallel postulate. For
the reader interested in the difference between these two words in terms of meaning we refer
to [Pam06].
3.1.1 Tarski’s parallel postulate is strictly stronger than Playfair’s postulate
By dropping the law of excluded middle7 , Tarski’s parallel postulate becomes

strictly stronger than Playfair’s postulate. Indeed, a particular instance of the law
of excluded middle, namely the decidability of intersection of lines, is required to
prove that Tarski’s parallel postulate follows from Playfair’s postulate. We will
present this proof in Section 5 since we did not prove a direct implication, and
therefore first need to introduce some other postulates. Now, to prove that the
decidability of intersection of lines is indeed needed for the proof, it suffices to show
that Tarski’s parallel postulate implies the decidability of intersection of lines and
that Playfair’s postulate does not. We will prove the first of these facts in Section 5.
The second of these facts can be proven in the same fashion as the independence
of the parallel postulate from the other axioms of Tarski’s system of geometry
was proved in [BBN15]. This proof consists in remarking that Tarski’s parallel
postulate allows us to construct arbitrary far away points while the other axioms
do not. This remark together with Herbrand’s theorem give an independence proof.
Here the same remark can be made about the decidability of the intersection of
lines and the other axioms of Tarski’s system of geometry plus Playfair’s postulate.
However, Playfair’s postulate can be proved in neutral geometry with decid-
able point equality assuming Tarski’s parallel postulate. This proof is actually
in [SST83]: it corresponds to Satz 12.11 which we formalized. In this proof, one
does not need to reason by cases on the possibility for two lines to intersect.
3.1.2 Playfair’s postulate is strictly stronger than the triangle postulate
Just as Tarski’s parallel postulate becomes strictly stronger than Playfair’s postu-
late by dropping the law of excluded middle, Playfair’s postulate becomes strictly
stronger than the triangle postulate when dropping Archimedes’ axiom. Indeed,
Max Dehn, a student of Hilbert, has shown that Playfair’s postulate is indepen-
dent from the axioms of planar neutral geometry extended with the triangle pos-
tulate [Deh00]: he gave a non-Archimedean model in which the triangle postulate
holds and Playfair’s postulate does not. One could then think that Archimedes’ ax-
iom is the missing link between these postulates. Actually, Greenberg has showed
that, in order to prove that the triangle postulate implies Playfair’s postulate, a
corollary of Archimedes’ axiom is sufficient [Gre10], which we will refer to as Green-
berg’s axiom8 (Fig. 18). In fact, Greenberg proves that this axiom is a corollary
of Archimedes’ axiom by proving that it follows from Aristotle’s axiom9 (Fig. 17),
itself following from Archimedes’ axiom. Greenberg defines Aristotle’s and Green-
berg’s axioms in the following way.
“Given any acute angle, any side of that angle, and any challenge segment
P Q, there exists a point Y on the given side of the angle such that if X is
the foot of the perpendicular from Y to the other side of the angle, then
Y X > P Q.”
7 As mentioned in Section 2.1.1, we assume the decidability of the point equality, which is
a tautology in classical logic.
8 One should remark that this axiom is not named after Greenberg in [Gre10].
9 For the sake of conciseness, we adopted the same name as Greenberg for this axiom which
is also known under the name of Aristotle’s angle unboundedness axiom.

“Given any segment P Q, line l through Q perpendicular to P Q, and ray r

of l with vertex Q, if θ is any acute angle, then there exists a point R on r
such that P R Q <b θ10 .”
3.1.3 The triangle postulate is strictly stronger than Bachmann’s Lotschnittaxiom
Similarly, when dropping Archimedes’ axiom, the triangle postulate becomes strictly
stronger than Bachmann’s Lotschnittaxiom. Bachmann demonstrated that this
postulate, that he named Lotschnittaxiom, was strictly weaker than the triangle
postulate [Bac73]. Pambuccian proved that Aristotle’s axiom is sufficient to prove
that the triangle postulate is implied by Bachmann’s Lotschnittaxiom [Pam94].
Pambuccian’s proof uses Pejas’ classification of Hilbert planes [Pej61] and, up to
our knowledge, there is no synthetic proof of the fact that this corollary is suffi-
cient, therefore we did not formalize this proof.
We can summarize the results from the previous subsections using Fig. 5.
Decidability of
Archimedes’ Aristotle’s Greenberg’s
intersection
axiom axiom axiom
of lines
Bachmann’s Tarski’s
Triangle Playfair’s
Lotschnit- parallel
postulate postulate
taxiom postulate
Fig. 5: Graphical summary of the independence results from Sections 3.1.1-3.1.3.
These independence results confirm that the theory in which the statements
are proven needs to be precisely defined. Moreover, they illustrate the fact that
some postulates can cease to be equivalent if the logic is changed. Therefore,
the notion of equivalence is not only relative to the theory but also to the logic.
Since, in this paper, we classify parallel postulates according to the theory and
the logic in which they are equivalent, we will now introduce a few notations for
the different kinds of equivalence that will be considered. Let us denote by N the
axioms of planar neutral geometry (A1-A9) with decidability of point equality, by
A Archimedes’ axiom and by G Greenberg’s axiom. We adopt the symbols |=LJ
and |=LK to differentiate the intuitionistic and the classical setting. We say that
G A
two properties P and Q are respectively NLJ -equivalent, NLJ -equivalent, NLJ -
equivalent or NLK -equivalent if N |=LJ P ⇔ Q, N ; G |=LJ P ⇔ Q, N ; A |=LJ
P ⇔ Q or N |=LK P ⇔ Q. The rest of this paper will be organized according to
10 We use <
b for the strict comparison between angles.
Fig. 5. In order to determine in which category a version of the parallel postulate

belongs we will formalize its NLJ -equivalence with one of this four postulates. For
the sake of avoiding references to the NLJ -equivalence of some postulates, we will
start by studying the postulates NLJ -equivalent to Playfair’s postulate. Then we
will proceed in decreasing order of strength, thus considering the postulates NLJ -
equivalent to Tarski’s parallel postulate, then those NLJ -equivalent to the triangle
postulate and finally those NLJ -equivalent to Bachmann’s Lotschnittaxiom.
3.2 Formal definitions of acute angles, parallelism and the sum of non-oriented
angles
In order to formalize these four postulates and these four axioms, we first need
to define acute angles, parallelism and the sum of non-oriented angles. Indeed,
Tarski’s parallel postulate is the only postulate expressed without any definition11 .
Thus, this section will be dedicated to defining these concepts. In this paper, the
exact Coq syntax of the axioms, definitions and main theorems is listed without
any pretty printing, to give the reader the opportunity to check what is the exact
statement we proved. For the auxiliary lemmas and all the proofs, we will use
classical mathematical notations. The proofs given in this paper serve only as a
documentation; the correctness of the results is assured by the mechanical proof
checker. Recall that for each postulate, we will provide the figure representing the
statement in the Euclidean plane and a counter-example in Poincaré disk model.
Having a counter-example in non-Euclidean geometry is interesting, as Szmielew
proved (assuming full continuity) that every statement which is false in hyperbolic
geometry and correct in Euclidean geometry is equivalent to the fifth parallel
postulate [Szm59] (we formalize this theorem in Section 11).
3.2.1 Formal definition of parallelism
In this section we define parallelism, which will be one of the most important
definitions for this work. In [SST83] one can find two definitions of parallelism.
The common way of defining it is to consider two lines as parallel if they belong to
the same plane but do not meet. This implies that we will also define collinearity
and coplanarity. The other definition of parallelism includes the previous one and
add the possibility for the lines to be equal. Therefore we will talk about strict
parallelism in the first case and about parallelism in the second.
Definition Col A B C := Bet A B C \/ Bet B C A \/ Bet C A B.
This predicate corresponds to Definition 4.10 of [SST83]. Among the first defi-
nitions which are introduced, there is the predicates expressing collinearity. It can
be defined using only the betweeness predicate. Col A B C expresses that A, B and
C are collinear if and only if one of the three points is between the other two.
11 That is the reason why Tarski chose this postulate, as he wanted to avoid definitions in
his axiom system.

Definition Coplanar A B C D :=
exists X, (Col A B X /\ Col C D X) \/
(Col A C X /\ Col B D X) \/
(Col A D X /\ Col B C X).
X2
X2 C
C
D D
X3 X3
X1
A B A X1
B
Fig. 6: Definition of Coplanar.
We did not define coplanarity in the same way as in [SST83]; we chose to

express coplanarity as a 4-ary predicate to avoid the definition of a predicate with
an arbitrary number of terms. Restricting ourselves to characterize coplanarity of
four points, we could use Satz 9.33 in [SST83] as a definition of coplanarity. This
definition states that 4 points are coplanar if two out of these four points form a
line which intersect the line formed by the remaining two points (either X1 , X2
or X3 on Fig. 6). Since we are in a 2-dimensional space in this paper, 4 points
are always coplanar. Yet we keep this definition, because we plan to extend our
formalization to higher dimensions in the future, as a large part of our library
is available in arbitrary dimension. In fact, in [SST83] the proofs are performed
in a n-dimensional space for a fixed positive integer n, given by the statement of
variants of the dimensional axioms.
Definition Par_strict A B C D :=
A <> B /\ C <> D /\ Coplanar A B C D /\ ~ exists X, Col X A B /\ Col X C D.
This predicate corresponds to Definition 12.2 of [SST83]. Note that one could
have chosen other definitions. For instance, one could have defined two lines (when
we consider lines, it is implied that the two points defining them are distinct) to be
parallel when they are at constant distance. According to Papadopoulos [Pap12],
this definition was introduced by Posidonius, an early commenter of Euclid’s El-
ements. As we will see in Section 6, an implicit change in a definition can have
severe consequences in the validity of a proof.
Definition Par A B C D :=
Par_strict A B C D \/ (A <> B /\ C <> D /\ Col A C D /\ Col B C D).
This predicate corresponds to Definition 12.3 of [SST83]. This definition asserts

that two lines are parallel if they are strictly parallel or if they are equal, since
with the previous definition, one line is not parallel to itself.
3.2.2 Formal definition of the sum of non-oriented angles
This section will be devoted to the definition of the sum of non-oriented angles.
It is based on the notions of congruence of angles and sides of line, which will be
presented in this section. It should be pointed out that there is no definition for
the sum of non-oriented angles in [SST83].
Definition CongA A B C D E F :=
A <> B /\ C <> B /\ D <> E /\ F <> E /\
exists A’, exists C’, exists D’, exists F’,
Bet B A A’ /\ Cong A A’ E D /\
Bet B C C’ /\ Cong C C’ E F /\
Bet E D D’ /\ Cong D D’ B A /\
Bet E F F’ /\ Cong F F’ B C /\
Cong A’ C’ D’ F’.
D0
C0
D D0
D
C C0
C
E
F E
B
F
B
A F0
A
0
A0
F
A0
Fig. 7: Definition of CongA.
This predicate corresponds to Definition 11.2 of [SST83]. Two angles are said to
be congruent if one can prolong the sides of the angles to obtain congruent triangles
(Fig. 7). A B C =b D E F means that angles ∠ABC and ∠DEF are congruent. It
should be noticed that even though the definition does not explicitly states that
BA0 ≡ EF 0 or BC 0 ≡ ED0 , these congruences are provable thanks to Satz 2.11
of [SST83]. This proposition corresponds to a degenerate case of the five-segment
axiom A5 (Fig. 2). This is technically important in Tarski’s system of geometry,
as it allows us to have fewer axioms in the system. However, the non-degenerate
case of this axiom is independent of the other axioms of our theory to which one
would add the degenerate case of this axiom [Hil60]. It is therefore questionable
to assume such axioms when using intuitionistic logic. Nevertheless, as proved
in [Bee15] and [BNSB14a], assuming the decidability of point equality suffices to
recover all the propositions of [SST83] in an intuitionistic setting.
Definition TS A B P Q :=
A <> B /\ ~ Col P A B /\ ~ Col Q A B /\ exists T, Col T A B /\ Bet P T Q.
This predicate corresponds to Definition 9.1 of [SST83]. The name of this

predicate corresponds the abbreviation for two sides. It describes a special case of
P P
B T
A
A B T Q
Fig. 8: Definition of TS.
the coplanarity (Fig. 6), namely when the intersection point is between the two
points defining one of the lines (Fig. 8). In this case one says that these first two
PQ
points are on opposite sides of the other line. A B indicates that P and Q are
PQ
on opposite sides of line AB . This definition being a special case of coplanarity,
it has the advantage of being valid in spaces of dimension higher than two. This
notion is absent in Euclid’s Elements [EHD02], in which the relative position of
the points on the figure is not justified, but inferred from the figure.
Definition OS A B P Q := exists R, TS A B P R /\ TS A B Q R.
Q
Q P
B
A
A B
R
Fig. 9: Definition of OS.
The last predicate needed to be able to define the sum of non-oriented angles
captures the property for two points to be on the same side of a line. This predicate
corresponds to Definition 9.7 of [SST83]. The name of this predicate corresponds
the abbreviation for one side. Two points are said to be on the same side of a line
if there exists a third point with which both of the points are on opposite side of
PQ
this line (Fig. 9). A B indicates that P and Q are on the same side of line AB .
PQ
Definition SumA A B C D E F G H I :=
exists J, CongA C B J D E F /\ ~ OS B C A J /\ CongA A B J G H I.
E D
F
D
E
A F
A
C B C B
J J
G
G H
H I
I
Fig. 10: Definition of SumA.
As we want to study the impact of Archimedes’ axiom, we cannot define the

sum of angles through the use of a measure for the angles. Indeed, Archimedes’
axiom would be needed to define a measure function [Rot14]. Another approach
could be to define a function that given two angles would return an angle rep-
resenting their sum. Again this approach would necessitate an extra axiom: the
axiom of choice. As a matter of fact, the axiom of choice would be used to select
a representative within the equivalence class of the angles congruent to the sum
of our given angles. Thus the sum of angles has to be defined geometrically.
To obtain the sum of two angles ∠ABC and ∠DEF , one constructs a point
J such that ∠ABC and ∠CBJ are adjacent and C B J = b D E F (Fig. 10). Then
the sum of the angles ∠ABC and ∠DEF is ∠GHI if A B J = b G H I . One thing
which could be surprising is the fact that we specified that the angles ∠ABC and
∠CBJ are adjacent by the fact that A and J are not on the same side of line BC .
This choice allows us to do without a disjunction of cases (either A and J are on
opposite side of line BC or J belongs to line BC ). Actually one cannot simply
state that these points are on opposite sides of line BC , as this would imply that
the sum of angles is not defined when one of the angles is straight or null.
Definition TriSumA A B C D E F :=
exists G H I, SumA A B C B C A G H I /\ SumA G H I C A B D E F.
The triangle postulate expresses a property about the sum of the interior angles
of a triangle, so we decided to define a predicate stating that the sum of the interior
angles of a triangle is congruent to a specific angle. Namely, S (4ABC ) = b DEF
means that the sum of the interior angles of triangle ABC is congruent to angle
∠DEF . The fact that we did not define the sum of angles as a function but as
a predicate motivated this choice. Indeed, it avoids carrying the witness of the
partial sum of the first two angles. Of course, to be able to talk about the sum of
the angles of a triangle, it has to be commutative and associative. We will later see
that it is only the case under certain conditions that are fulfilled when considering
the interior angles of a triangle.
3.2.3 Formal definition of acute angles
In order to be able to formalize a predicate specifying that an angle is acute, we

need to define the concepts of angle comparison and right triangles. Defining these
concepts was straightforward, as both of them were already present in [SST83].

For the sake of completeness we will now present them.
To express a predicate specifying that an angle is acute, we do not need a
definition for perpendicularity, but only for right triangles. The definition of a
right triangle is more general than the definition of a right angle since it includes
the case of a degenerated triangle. In [SST83], right triangles are defined through
midpoints. Following, we first present Tarski’s definition of midpoint and right
triangle.
Definition Midpoint M A B := Bet A M B /\ Cong A M M B.
This predicate corresponds to Definition 7.1 of [SST83]. It states that M is the

midpoint of A and B . It is the case when M is between A and B and equidistant
from them. It is interesting to notice that the existence of the midpoint appears
quite late in [SST83]. This is because its proof, which does not involve the continu-
ity axiom and is due to Gupta [Gup65], cannot be done earlier in the development.
Definition Per A B C := exists C’, Midpoint B C C’ /\ Cong A C A C’.
C0 B C A
C0 C
Fig. 11: Definition of Per.
This predicate corresponds to Definition 8.1 of [SST83]. In the case where B

is different from A and C , A B C (Fig. 11) means that A, B and C form a right
triangle with the right angle at vertex B . But A B C is also true when B is equal
to A and/or C . Therefore, we will need either to specify that these points are
different or a new definition to avoid this case.
The notion of angle comparison is defined by means of a predicate stating that
a point belongs to the interior of an angle, itself formulated using a predicate
asserting that a point belongs to a ray.
Definition Out P A B := A <> P /\ B <> P /\ (Bet P A B \/ Bet P B A).
This predicate corresponds to Definition 6.1 of [SST83]. P A B indicates that

P belongs to line AB but does not belong to the segment AB . This implies that A
and B belong to the same ray with initial point P and that neither of these points
coincide with P . This predicate is symmetric in its last two points, but we usually
choose to prioritize the first of these points to define the ray. Thus, most of the
time, P A B will express the fact that B belongs to the ray P A.
Definition InAngle P A B C :=
A <> B /\ C <> B /\ P <> B /\ exists X, Bet A X C /\ (X = B \/ Out B X P).
A
P X
B B P X
C C
Fig. 12: Definition of InAngle.
This predicate corresponds to Definition 11.23 of [SST83]. P ∈ b A B C states

that P belongs to the interior of angle ∠ABC (Fig. 12). A point P is said to
belong to the angle ABC if this angle is well defined, meaning that B is distinct
from both A and C , and if there exists a point X on the segment AC such that
either P belongs to the ray BX or B and X are equal. This last case occurs when
angle ∠ABC is straight and one consider that any point belongs to a straight
angle, except its vertex. An alternative to this definition would have been the
one Greenberg uses in [Gre93], namely that P belongs to the interior of the angle
ABC if P and A are one the same side of line BC and if P and C are on the same
side of line BA. The definition from [SST83] is more general, because according
to Greenberg’s definition, a point on one of the sides of an angle is not inside it.
Assuming that the point we consider is not on a side of the angle, the one we
adopted trivially implies the one Greenberg uses, and the converse can be proved
by applying Pasch’s axiom. However, we chose to adopt the version from [SST83]
since it directly provides the point X . Let us here emphasize again the importance
of definitions. The reader could be tempted to define P ∈ b A B C as the existence
of a segment with endpoints on the sides of a given angle which passes through
P . Yet, it is not always the case that this segment exists. Indeed, this property
corresponds to Tarski’s parallel postulate.
Definition LeA A B C D E F := exists P, InAngle P D E F /\ CongA A B C D E P.
This predicate corresponds to Definition 11.27 of [SST83]. An angle ∠ABC is

said to be smaller than or equal to another angle ∠DEF if there exists a point P
in the interior of this second angle such that angle ∠DEP is congruent to the first
angle. The witness point P , which is needed for proving different properties about
this order relation, is omitted by this predicate.
Definition LtA A B C D E F := LeA A B C D E F /\ ~ CongA A B C D E F.
This predicate corresponds to Definition 11.38 of [SST83]. It is more straight-

forward to first define the non-strict comparison between angles. However, in order
to obtain a predicate characterizing acute angles, we need to define the strict com-
parison between angles. This is done by simply excluding the case where the angles
are congruent.
Definition Acute A B C :=
exists A’, exists B’, exists C’, Per A’ B’ C’ /\ LtA A B C A’ B’ C’.
Finally, we can define a predicate characterizing acute angles. This predicate

corresponds to Definition 11.39 of [SST83]. An angle ∠ABC is said to be acute
if there exists a right triangle A0 B 0 C 0 with the right angle at vertex B 0 such that
angle ∠ABC is strictly smaller than angle ∠A0 B 0 C 0 . One can recall that A0 B 0 C 0
means that angle ∠A0 B 0 C 0 is right only in the case where B 0 is distinct from both
A0 and C 0 . This is the case thanks to the definition of the angle comparison. This
enforces that the angle to which we compare the angle ABC is indeed right.
3.3 Formalization of the four particular versions of the parallel postulate and of
the continuity axioms
In this section, we formalize Tarski’s parallel postulate, Playfair’s postulate, the

triangle postulate and Bachmann’s Lotschnittaxiom, as well as the decidability of
intersection of lines, Archimedes’, Aristotle’s and Greenberg’s axioms. Now that
we defined acute angles, parallelism and the sum of non-oriented angles, we will be
able to define these postulates and axioms easily, except for Archimedes’, which
requires a few extra definitions.
3.3.1 Tarski’s parallel postulate
Postulate 1 (Tarski’s parallel postulate).
Definition tarski_s_parallel_postulate := forall A B C D T,

Bet A D T -> Bet B D C -> A <> D ->
exists X, exists Y, Bet A B X /\ Bet A C Y /\ Bet X T Y.
This postulate (Fig. 13) is the official version of the parallel postulate found
in [SST83]. The statement, due to Lorentz [Gup65], is a modification of an implicit
assumption made by Legendre while attempting to prove that Euclid’s parallel
postulate was a consequence of Euclid’s other axioms, namely Legendre’s parallel
postulate which will be introduced in Section 7.3. This version is particularly
interesting, as it has the advantages of being easily expressed only in term of
betweenness, and being valid in spaces of dimension higher than two.
A
A
C
B
B D
D C
X
X T Y
Fig. 13: Tarski’s parallel postulate (Postulate 1).
3.3.2 Playfair’s postulate
Postulate 2 (Playfair’s postulate).
Definition playfair_s_postulate := forall A1 A2 B1 B2 C1 C2 P,

Par A1 A2 B1 B2 -> Col P B1 B2 ->
Par A1 A2 C1 C2 -> Col P C1 C2 ->
Col C1 B1 B2 /\ Col C2 B1 B2.
C2
C1 B B1
1
P P
C2 C1 B2
B2
A1
A1 A2
A2
Fig. 14: Playfair’s postulate (Postulate 2).
Playfair’s postulate (Fig. 14) is one of the best-known versions of the par-
allel postulate for the modern reader. This postulate corresponds to Satz 12.13
in [SST83]. Note that it does not state the existence of the parallel line but only
its uniqueness, because the existence can be proved from the axioms of Tarski’s
neutral geometry (Satz 12.10 of [SST83]). Proclus, another early commenter of
Euclid’s Elements, already recognized that an incorrect proof of Euclid’s postulate
by Ptolemy was making this implicit assumption.
3.3.3 Triangle postulate
Postulate 3 (Triangle postulate).
Definition triangle_postulate := forall A B C D E F,

TriSumA A B C D E F -> Bet D E F.
E B F
D E
A C
A C
Fig. 15: Triangle postulate (Postulate 3).
This postulate (Fig. 15) corresponds to Proposition I.32 in [EHD02] and Satz 12.23
in [SST83]. We formalized it slightly differently, as it precisely formulates that the
sum is equal to a straight angle instead of two right angles. Nevertheless, we have
proved that the sum of an angle with itself is equal to a straight angle if and only
if the angle is right. This postulate results of a failed attempt at proving Euclid’s
parallel postulate due to Legendre. This statement was implicitly used in one of
Legendre’s proofs. Interestingly, the sum of the angles of a triangle allows to set
apart hyperbolic, Euclidean and elliptic geometry. This sum is respectively lower,
equal or higher than two right angles in hyperbolic, Euclidean and elliptic case.
3.3.4 Bachmann’s Lotschnittaxiom
Postulate 4 (Bachmann’s Lotschnittaxiom).
Definition bachmann_s_lotschnittaxiom := forall P Q R P1 R1,

P <> Q -> Q <> R -> Per P Q R -> Per Q P P1 -> Per Q R R1 ->
exists S, Col P P1 S /\ Col R R1 S.
This postulate (Fig.16) expresses that, given that the lines P Q and QR are
perpendicular, the lines P Q and P P1 are perpendicular and the lines QR and RR1
are perpendicular, we know that the lines P P1 and RR1 must intersect. Here,
the perpendicularity hypotheses are expresses using the Per predicates, hence we
had to add non-degeneracy hypotheses to exclude the cases where the points P
and Q as well as the points Q and R are equal. However, since the property is
trivially true in the cases where P = P1 or R = R1 we did not exclude these cases.
According to Hartshorne [Har00], it “characterizes geometries in which the angle
sum of a triangle differs at most infinitesimally” from two right angles.
P1
P P1 S P
R1
R1
Q R
Q R
Fig. 16: Bachmann’s Lotschnittaxiom (Postulate 4).
3.3.5 Decidability of intersection of lines
Axiom 1 (Decidability of intersection of lines).
Definition decidability_of_intersection := forall A B C D,

(exists I, Col I A B /\ Col I C D) \/
~ (exists I, Col I A B /\ Col I C D).
This axiom corresponds to a simple decidability property. However, it holds a

special place in this study. Indeed, we previously studied the impact of working
in an intuitionistic setting in Tarski’s system of geometry [BNSB14a]. During this
work, we were trying to either prove that Axiom 1 could be derived from the
axioms of Tarski’s system of geometry with decidable point equality or find an
argument justifying its independence. Once we discovered its close relationship
with the parallel postulates, we started to investigate which versions of the parallel
postulates were implying it. Thus, Axiom 1 can be considered as the starting point
of the classification of the parallel postulates that we present in this paper.
3.3.6 Archimedes’ axiom
Archimedes’ axiom can be expressed almost directly using the betweeness and
congruence predicates. Following Duprat’s work [Dup10], we defined it inductively
without introducing the natural numbers. To state Archimedes’ axiom, we first
formalized the fact that “there exists some positive integer n and n + 1 points
A1 , · · · , An+1 on line AB , such that Aj is between Aj−1 and Aj +1 for 2 < j < n,
Aj Aj +1 and AB are congruent for 1 < j < n, A1 = A and An+1 = C ” as the Grad
predicate. This predicate and its variants, which will be presented in Section 7.3,
represent the only inductive definitions of our library. Actually, we will not specify
that “D is between A1 and An+1 ” in our definition but we will use the definition
for non-strict comparison between segments from [SST83].
Definition Le A B C D := exists E, Bet C E D /\ Cong A B C E.
This predicate corresponds to Definition 5.4 of [SST83]. A segment AB is said

to be less than or equal to another one CD if one can construct a point E such that
this point is between C and D and the segments AB and CE are congruent. For
convenience, this witness, which is needed for proving different properties about
this order relation, is omitted by this predicate.
Using this predicate, it suffices to assert that CD ≤ A1 An+1 , which allow us
to set A1 = A and to have A1 , · · · , An+1 on line AB .
Inductive Grad : Tpoint -> Tpoint -> Tpoint -> Prop :=

| grad_init : forall A B, Grad A B B
| grad_stab : forall A B C C’,
Grad A B C ->
Bet A C C’ -> Cong A B C C’ ->
Grad A B C’.
Definition Reach A B C D := exists B’, Grad A B B’ /\ Le C D A B’.
Grad A B C expresses that C is on the graduation based on the segment AB .

Then, this definition allows us to define Archimedes’ axiom in a straightforward
manner.
Axiom 2 (Archimedes’ axiom).
Definition archimedes_axiom := forall A B C D, A <> B -> Reach A B C D.
3.3.7 Aristotle’s axiom
Before defining Aristotle’s axiom, we need to introduce the notion of strict com-
parison between segments.
Definition Lt A B C D := Le A B C D /\ ~ Cong A B C D.
This predicate corresponds to Definition 5.14 of [SST83]. The reason for the
non-strict comparison to appear before the strict one is simple. Unlike Hilbert,
Tarski uses a non-strict betweenness relation. In order to obtain a strict comparison
of segments, it suffices to exclude the case where they are congruent.
We are now ready to state Aristotle’s axiom.
Axiom 3 (Aristotle’s axiom).
Definition aristotle_s_axiom := forall P Q A B C,

~ Col A B C -> Acute A B C ->
exists X, exists Y, Out B A X /\ Out B C Y /\ Per B X Y /\ Lt P Q X Y.
This axiom is very close to the statement from Greenberg [Gre10] that we gave
in Subsection 3.1.2. Here the acute angle is the angle ∠ABC (Fig. 17). Compared
to Greenberg’s statement, we had to add the condition that this angle is non-null12 .
Moreover, since the triangle BXY can be proved non-degenerate from the other
assumptions, we can establish that X is the foot of the perpendicular from Y to
the other side of the angle by specifying that BXY is a right triangle with the right
angle at vertex X . The other subtle difference is the fact that our version states the
12 Non-degeneracy conditions are often omitted in textbook proofs.
P
P
Q
Q
Y Y
C
C
A X
B
B A X
Fig. 17: Aristotle’s axiom (Axiom 3).
existence of both points X and Y . This is due to the fact that one cannot define
a function for the orthogonal projection in our current axiom system. In order to
obtain such a function, one would either need a stronger axiom system where one
would introduce function symbols in the axioms which are not already quantifier-
free, or one would require an extra axiom. For example, one could have used the
constructive definite description axiom provided by the standard library:
Axiom constructive_definite_description :
forall (A : Type) (P : A->Prop), (exists! x, P x) -> { x : A | P x }.
It allows us to convert a relation which has been proved to be functional to

a proper Coq function. As the use of the axiom turns the intuitionistic logic of
Coq into an almost classical logic [Bel93], we decided to avoid adding this axiom.
3.3.8 Greenberg’s axiom
Axiom 4 (Greenberg’s axiom).
Definition greenberg_s_axiom := forall P Q R A B C,

~ Col A B C ->
Acute A B C -> Q <> R -> Per P Q R ->
exists S, LtA P S Q A B C /\ Out Q S R.
As for Aristotle’s axiom, this axiom does not differ much from Greenberg’s
statement seen in Subsection 3.1.2. Again the acute angle is the angle ∠ABC
(Fig. 18). The ray r is given through point R. In order to make sure this ray is well
defined we had to add the condition that points Q and R are different. Finally the
point S asserted to exist corresponds to the point R from the statement given by
Greenberg.
Both of these axioms are consequences of Archimede’s axiom, but not con-
versely [Gre88, Gre10]. Indeed Aristotle’s axiom is a weaker axiom than Archimede’s
axiom and Greenberg’s axiom is a consequence of Aristotle’s axiom.
Q P
P
Q
R
C
R
B
C
B S
S A A
Fig. 18: Greenberg’s axiom (Axiom 4).
4 Postulates equivalent to Playfair’s postulate
In this section, we present the postulates which are NLJ -equivalent to Playfair’s
postulate in planar neutral geometry. Some of these properties are expressed using
definitions present in [SST83]. Thus we also give these definitions in this section.
Then we will discuss the formalization of the equivalence proofs.
4.1 Statements of postulates equivalent to Playfair’s postulate
Here, we introduce eight postulates which are NLJ -equivalent to Postulate 2 (Play-
fair’s postulate). They correspond to properties about various subjects, namely
parallelism, perpendicularity, angles and distance. This variety of subjects repre-
sents a specificity of the parallel postulate. We will see in the next section how
this variety affected the way we proved the equivalence of all of these statements.
Postulate 5 (Postulate of transitivity of parallelism).
Definition postulate_of_transitivity_of_parallelism := forall A1 A2 B1 B2 C1 C2,

Par A1 A2 B1 B2 -> Par B1 B2 C1 C2 ->
Par A1 A2 C1 C2.
C1
B1
C2
C2 C1
B2
B1
A1 B2 A1
A2
A2
Fig. 19: Postulate of transitivity of parallelism (Postulate 5).

The first of these postulates (Fig. 19) is the postulate of transitivity of par-
allelism. It states that, given two lines A1 A2 and C1 C2 parallel to the same line
B1 B2 , these lines are also parallel. This postulate, which corresponds to Propo-
sition I.30 in [EHD02] and Satz 12.15 in [SST83], would have been inconsistent
with the other axioms if we would have taken Euclid’s definition of the parallelism
(wikipedia’s translation), which matches what we identify as strict parallelism:
“Parallel straight lines are straight lines which, being in the same plane and
being produced indefinitely in either direction, do not meet one another in
either direction.”
Indeed, it is possible for lines A1 A2 and C1 C2 to be equal. One should notice here
that again definitions are essential.
Postulate 6 (Midpoint converse postulate).
Definition midpoint_converse_postulate := forall A B C P Q,

~ Col A B C ->
Midpoint P B C -> Par A B Q P -> Col A C Q ->
Midpoint Q A C.
C C
Q
Q
P
P
A
A B
B
Fig. 20: Midpoint converse postulate (Postulate 6).
This postulate (Fig. 20) is a part of the converse of the midpoint theorem and
corresponds to a special case of the intercept theorem. We will therefore refer to it
as midpoint converse postulate. This postulate expresses that, in a non-degenerate
triangle ABC , the intersection point Q of side AC with the parallel to a side AB
which passes through the midpoint P of side BC is the midpoint of side AC .
One should notice that the midpoint theorem is valid in planar neutral geometry,
whereas its converse is equivalent to the parallel postulate. Indeed, it follows easily
from the Satz 13.1 of [SST83]. It is interesting to remark that the second part of
the converse of the midpoint theorem, namely that, in any triangle, the midline
(the segment P Q on Fig. 20) is congruent to half of the basis (the segment AB
on Fig. 20), is equivalent to another statement of the parallel postulate which is
strictly weaker than the triangle postulate in the theory of metric planes [AP16,
Bac73].
Postulate 7 (Alternate interior angles postulate).
Definition alternate_interior_angles_postulate := forall A B C D,

TS A C B D -> Par A B C D ->
CongA B A C D C A.
C
C
D D
B A
B
Fig. 21: Alternate interior angles postulate (Postulate 7).
This postulate (Fig. 21) is commonly known as alternate interior angles theo-
rem. It asserts that a line falling on parallel lines makes the alternate angles equal
to one another. One can remark that this postulate, like others, was a proposition
in [EHD02] (a part of Proposition I.29) as well as in [SST83] (Satz 12.21). However,
Satz 12.21 of [SST83] is an equivalence and enunciates more than the alternate
interior angles theorem. One side of the equivalence corresponds to the alternate
interior angles theorem, while the other corresponds to its converse, which is valid
in neutral planar geometry, just as for the previous postulate.
Postulate 8 (Consecutive interior angles postulate).
Definition consecutive_interior_angles_postulate := forall A B C D P Q R,

OS B C A D -> Par A B C D -> SumA A B C B C D P Q R ->
Bet P Q R.
C P
Q
C P R
R
D D
Q
B A
Fig. 22: Consecutive interior angles postulate (Postulate 8).

This postulate (Fig. 22) is commonly known as consecutive interior angles

theorem. It states that a line falling on parallel lines makes the sum of interior
angles on the same side equal to two right angles. It was proved together with the
previous postulate in [EHD02] (as a part of Proposition I.29) but not in [SST83],
since the notion of supplementary angles is never introduced in this book. Similarly
to the triangle postulate, we formalized this postulate slightly differently, as it
precisely formulates that the sum is equal to a straight angle.
With a view to defining the next postulate, we need to define perpendicularity,
something which we postponed. We adopted the definition given in [SST83], which
used the following intermediate definition.
Definition Perp_at X A B C D :=
A <> B /\ C <> D /\ Col X A B /\ Col X C D /\
(forall U V, Col U A B -> Col V C D -> Per U X V).
B
B
X
C
X C D D
Fig. 23: Definition of Perp at.
We recall that we already defined a predicate for right triangles, but this def-
inition included the case where the sides of the right angle could be degenerate.
Therefore, in order to define perpendicularity using this predicate, one must know
the intersection point of the perpendicular lines and exclude the case of the degen-
erate right triangle. AB ⊥ CD means that lines AB and CD meet at a right angle
X
in X (Fig. 23). The part of the definition that specifies that any point on the first
line together with any point on the second line and the intersection point form a
right angle is essential to the possibility of choosing any couple of different points
to represent the lines.
Definition Perp A B C D := exists X, Perp_at X A B C D.
This predicate and the previous one correspond to Definition 8.11 of [SST83].
Most of the time, we just want to consider the perpendicularity of two lines AB
and CD without specifying the point in which they meet. In such cases, we will
use AB ⊥ CD.
Postulate 9 (Perpendicular transversal postulate).
Definition perpendicular_transversal_postulate := forall A B C D P Q,

Par A B C D -> Perp A B P Q ->
Perp C D P Q.
C Q Q
C
D
D
A A
B
B
P P
Fig. 24: Perpendicular transversal postulate (Postulate 9).
This postulate (Fig. 24) is commonly known as perpendicular transversal the-

orem. It expresses that given two parallel lines, any line perpendicular to the first
line is perpendicular to the second line. Just as for the previous postulates, the
converse of the perpendicular transversal postulate is valid in neutral planar ge-
ometry. It corresponds to Satz 12.9 in [SST83] and the perpendicular transversal
postulate corresponds to a special case of Satz 12.22 in [SST83].
Postulate 10 (Postulate of parallelism of perpendicular transversals).
Definition postulate_of_parallelism_of_perpendicular_transversals :=
forall A1 A2 B1 B2 C1 C2 D1 D2,
Par A1 A2 B1 B2 -> Perp A1 A2 C1 C2 -> Perp B1 B2 D1 D2 ->
Par C1 C2 D1 D2.
B1 C2 C2
B1
D2
D2
B2
D1 B2
A1 A1
D1 A2
A2
C1 C1
Fig. 25: Postulate of parallelism of perpendicular transversals (Postulate 10).
This postulate (Fig. 25), which will be designated as postulate of parallelism of

perpendicular transversals, is less known than the previous ones. This is probably
due to the fact that it can be easily deduced from the perpendicular transversal
postulate and its converse. This could explain why it does not appear as a proposi-
tion in the most well-known axiomatic developments of Euclidean geometry, which
are those of Euclid [EHD02], Hilbert [Hil60] and Tarski [SST83]. Nevertheless, it
is listed amongst the statements equivalent to the parallel postulate in [Gre93]
and [Mar98]. It asserts that two lines, each perpendicular to one of a pair of par-
allel lines, are parallel. It is easy to take this property for granted and assume it
implicitly since it corresponds to Satz 12.9 in [SST83], which is valid in neutral
planar geometry, when the two lines known to be parallel are equal.
Postulate 11 (Universal Posidonius’ postulate).
Definition universal_posidonius_postulate := forall A1 A2 A3 A4 B1 B2 B3 B4,

Par A1 A2 B1 B2 ->
Col A1 A2 A3 -> Col B1 B2 B3 -> Perp A1 A2 A3 B3 ->
Cong A3 B3 A4 B4.
B3 B1
B2B3 B1 B4 B4
B2
A3 A1 A2 A4
A3 A1 A2 A4
Fig. 26: Universal Posidonius’ postulate (Postulate 11).
This postulate (Fig. 26) is a property of parallel lines in Euclidean geometry

which was taken as definition of parallelism by Posidonius. We will refer to it as
universal Posidonius’ postulate because another postulate (Postulate 22), known
as Posidonius’ postulate, can be expressed in a similar way with the exception
that the points A1 , A2 , B1 and B2 are quantified existentially and not universally
and that the hypothesis of parallelism is replaced by a non-degeneracy condition.
It states that, if two lines A1 A2 and B1 B2 are parallel, then they are everywhere
equidistant. This can be formalized by specifying that any two points B3 and B4
on B1 B2 form with the feet of the orthogonal projection of these points onto the
line A1 A2 , respectively A3 and A4 , congruent segments. However, as we will later
see, everywhere equidistant lines only exist in Euclidean geometry. This statement
being equivalent to the parallel postulate motivates the fact that we list all of
the definitions we chose, since, as already mentioned, definitions are critical when
studying statements of the parallel postulate.
The last postulate we will analyze in this section is a special case of Playfair’s
postulate where one of the parallel lines shares a common perpendicular with its
parallel. Thus, to state this postulate, we first present a refinement of this property
which was defined in [SST83].
Definition Perp2 A B C D P :=
exists X, exists Y, Col P X Y /\ Perp X Y A B /\ Perp X Y C D.
A X B B
X
A
P P
C
C D D
Y
Y
Fig. 27: Definition of Perp2.
|=
This predicate corresponds to Definition 13.9 of [SST83]. AB CD not only
P
means that the lines AB and CD have a common perpendicular XY but also that
|=
XY passes through the point P . One should remark that AB CD implies, in
P
neutral planar geometry, that the lines AB and CD are parallel. However, not
any pair of parallel lines share a common perpendicular. In fact, in hyperbolic
geometry, ultraparallel lines only share a unique common perpendicular, and lim-
iting parallels do not share any common perpendicular [BS60]. Therefore, even
in the case of ultraparallel lines, there may be no common perpendicular passing
through a given point, since it suffices that this point lies outside their unique
common perpendicular.
Postulate 12 (Alternative Playfair’s postulate).
Definition alternative_playfair_s_postulate := forall A1 A2 B1 B2 C1 C2 P,

Perp2 A1 A2 B1 B2 P -> Col P B1 B2 ->
Par A1 A2 C1 C2 -> Col P C1 C2 ->
Col C1 B1 B2 /\ Col C2 B1 B2.
Because of the similarity of Postulate 12 (Fig. 28) with Postulate 2 (Playfair’s

postulate) we decided to name it alternative Playfair’s postulate. It asserts that
any line parallel to a given line passing through a given point is equal to the
line that passes through the given point and shares a common perpendicular with
the given line that passes through the given point. One should mention that this
postulate does not have the same importance as the other ones, because its role is
just to simplify the proofs.
4.2 Formalizing the equivalence proof
In this section, we focus on the formalization of the proof that the postulates of
the previous subsection (Postulates 5-12) are indeed NLJ -equivalent to Postulate 2
C2
C1 B1
B1
B2 P
C2 C1
P
B2
A1
A1 A2
A2
Fig. 28: Alternative Playfair’s postulate (Postulate 12).
(Playfair’s postulate). To make sure it holds, it suffices to prove it within the

context of the Tarski_2D type class from Tab. 2. Thus we need a definition for an
n-ary equivalence relation. We use the following definition using lists:
Definition all_equiv (l : list Prop) :=

forall x y, In x l -> In y l -> (x<->y).
We chose to define this n-ary equivalence relation as a predicate on list of propo-

sitions. This list of propositions contains the equivalent predicates. This predicates
expresses that any two propositions in this list are equivalent. It allows us to reduce
the proof of the equivalence or the implication between two properties by checking
the membership of these properties to a list. The Coq statement corresponding to
the equivalence of any two of Postulates 2, 5-12 is the following.
Theorem equivalent_postulates_without_decidability_of_intersection_of_lines :
all_equiv
(alternate_interior_angles_postulate::
alternative_playfair_s_postulate::
consecutive_interior_angles_postulate::
midpoint_converse_postulate::
perpendicular_transversal_postulate::
playfair_s_postulate::
universal_posidonius_postulate::
postulate_of_parallelism_of_perpendicular_transversals::
postulate_of_transitivity_of_parallelism::
nil).
In order to lower the number of equivalences to be proven to complete the

proof of the previous theorem, we introduced an alternative predicate for n-ary
equivalence relation and proved its equivalence with all_equiv.
Definition all_equiv’_aux (l: list Prop) : Prop.

induction l; [exact True|].
exact ((a -> a0) /\ IHl).
Defined.
Definition all_equiv’ (l: list Prop) : Prop.

exact ((last l a -> a) /\ all_equiv’_aux (a::l)).

Defined.
Lemma all_equiv_equiv : forall l, all_equiv l <-> all_equiv’ l.
This definition corresponds to the usual technique to prove equivalences that

minimize the number of implications to be proved. Indeed, for a list of length n, n
implications would suffice. This is much better than the 2n2 implications required
from all_equiv. In Coq, it is convenient to have the two definitions, one for proving
that a list of statements are equivalent and the other to use these equivalences.
In practice the all_equiv’ definition is also useful to improve the compilation
time of our proofs. Indeed, to prove the n-ary equivalence statements, we put in
the context all the implications proved previously manually and we let the tau-
tology checker of Coq (tauto) complete the proof. This technique is convenient,
but does not scale well when one uses all_equiv and the number of statements
is large. Fig. 29 provides a graphical summary of the implications we formalized
to prove Theorem 1. On this figure, a circle with a number n in its center repre-
sents Postulate n, an arrow between two circles represents an implication, and a
double-headed arrow represents an equivalence. One can observe that most of the
implications (eight out of fourteen) that we proved involve Postulate 2 (Playfair’s
postulate). Indeed, since these postulates correspond to properties about diverse
subjects, we found that it was more straightforward, when proving the implica-
tion between properties about different subjects, to use parallelism as one of the
two subjects. Postulate 2 and Postulate 5 (Postulate of transitivity of parallelism)
are the only two postulates about parallelism. Moreover, we have proved that the
equality of lines is decidable in planar neutral geometry assuming decidability of
point equality, while we could only prove the decidability of parallelism assum-
ing Postulate 5. Therefore, Postulate 2 can be proved by contradiction, whereas
Postulate 5 cannot unless we find a proof of the decidability of parallelism valid
in planar neutral geometry. Indeed, unless the conclusion is known to be decid-
able, one cannot use a proof by contradiction to derive it, because the proof by
contradiction is not valid in an intuitionistic setting. We should point out that,
in the definition of parallelism, the fact that the lines do not meet can be proved
by proof of negation 13 , while the rest of this definition can be proved by con-
tradiction. However, proving the parallelism in such a way is more tedious than
proving the equality of lines by contradiction. This explains why Postulate 2 has
such a central role in the formalization of Theorem 1. Thus, we only proved impli-
cations between properties about the same subject, such as the alternate interior
angles postulate and the consecutive interior angles postulate, besides these eight
implications.
With a view to keeping a good balance between mathematical aspects, for-
malization aspects, and explanations, we decided to focus on only one implication
which illustrates the impact of working in an intuitionistic setting rather than a
classical one. The reader who is interested in the proofs of the implications can
13 We use the expression ’proof of negation’ to describe a proof of ¬A assuming A and
obtaining a contradiction. For the reader who is not familiar with intuitionistic logic, we recall
that this is simply the definition of negation and this proof rule has nothing to do with the
proof by contradiction (to prove A it suffices to show that ¬A leads to a contradiction), which
is not valid in our intuitionistic setting.
10.
11. 9.
12. 2.
7. 5.
8. 6.
Fig. 29: Overview of the proofs of Sec. 4.
find some of them in the literature14 . In [SST83], there is the proof of the impli-
cation from Postulate 2 to Postulate 5 (Satz 12.15). In [Bee16], the implication
from Postulate 2 to Postulate 7 (Lemma 6.6) as well as the implication from Pos-
tulate 2 to Postulate 5 (Lemma 6.8) are proved. Finally, in [Mar98], the proofs of
the equivalence between Postulate 7 and Postulate 8 (Theorem 21.4) and of the
implication from Postulate 9 to Postulate 10 (Theorem 23.7) are provided.
Putting together the implications from Fig. 29, we can prove the following
theorem.
Theorem 1 Postulates 2, 5-12 are NLJ -equivalent.
Let us now focus on the proof of the implication from Postulate 6 (Midpoint
converse postulate) to Postulate 2 (Playfair’s postulate). In order to present the
proof that we formalized, we will collect the lemmas that will be used throughout
this proof. We believe it is important to list these lemmas, since we saw that
it often happens that a statement is valid in neutral planar geometry and that
its converse is equivalent to the parallel postulate. By detailing these lemmas
and only deriving new facts from the application of these lemmas in our proofs,
we make sure we do not implicitly apply a statement equivalent to the parallel
postulate, unless we have proved it to follow from the statement from which we
are proving a consequence. However, we chose not to include trivial lemmas which
state permutation properties of the predicates (e.g. AB k CD ⇒ CD k AB ). We
also decided to omit lemmas allowing to weaken a statement (e.g. AB ks CD ⇒
AB k CD). Besides, one problem one encounters with Tarski’s system of geometry
is the fact that there is no primitive type line. Therefore, when considering a line,
one represents it by two different points. This implies that we need a lemma such
as C 6= D0 ⇒ AB k CD ⇒ Col C D D0 ⇒ AB k CD0 . This kind of lemma are easily
proven in neutral geometry. Moreover, the proofs of collinearity can be automated
by a reflexive tactic that we developed in [BNSB14b]. Therefore we will simply
use them implicitly, as one would do in a pen-and-paper proof.
14 Up to our knowledge the following proofs are the only ones that resemble the ones we
formalized.
Lemma 1 (6.2115 ) Two points are equal if they are at the intersection of two different
lines.
Lemma 2 (7.8) The symmetric of a point with respect to another point is con-
structible.
Lemma 3 (7.17) There is only one midpoint to a given segment.
Lemma 4 16 A line P Q which enters a triangle ABC on side AB and does not pass
through C must exit the triangle either on side AC or on side BC.
Proposition 1 Postulate 6 implies Postulate 2.
Proof
Given that A1 A2 kB1 B2 , A1 A2 k X

C1 C2 , Col P B1 B2 and Col P C1 C2 ,
we wish to prove that B1 , B2 , C1 B2
C3 C1
P
and C2 are collinear. We can first
C2
eliminate the cases where line A1 A2 B3 B1
is equal to B1 B2 and/or C1 C2 . In-
deed, if all three lines are equal, A1 A2
we are trivially done, and if two
lines are equal and strictly paral- Fig. 30: Postulate 6 implies Postulate 2.
lel to the third one, then we may
also conclude, as we can also prove that this last case is impossible because the
lines meet in P . So we may now assume A1 A2 ks B1 B2 and A1 A2 ks C1 C2 . We can
then construct the symmetric point X of A1 with respect to P using Lemma 2. Now
we will prove that there exists a point B3 on line B1 B2 which is strictly between
A2 and X . We know that P is either different from B1 or from B2 , as otherwise
it would contradict A1 A2 ks B1 B2 . Let us prove the existence of the point B3 by
using Lemma 4 in the triangle A1 A2 X with either line P B1 or P B2 , depending
on whether P is distinct from B1 or B2 . We prove the hypotheses of this lemma
in the same way in both cases, so let us only consider the case where P and B1
are distinct. The hypotheses ¬ Col A2 P B1 and ¬ Col A1 X B1 can be proven by
proof of negation. Indeed, assuming Col A2 P B1 would contradict A1 A2 ks B1 B2 ,
and assuming Col A1 X B1 would contradict P 6= B1 , as these two points would
be on lines P A1 and P B1 and Lemma 1 would imply that they are equal. Fi-
nally, B3 cannot be between A1 and A2 , as assuming A1 B3 A2 would contradict
A1 A2 ks B1 B2 . Hence, Lemma 4 lets us derive the existence of the point B3 on
line B1 B2 which is strictly between A2 and X . In the same way, we can prove
there exists a point C3 on line C1 C2 which is strictly between A2 and X . Now,
it suffices to prove that B3 and C3 are equal, as it implies that B1 , B2 , C1 and
C2 are collinear. From Postulate 6 and Lemma 3, we know that they both are the
midpoint of the segment A2 X and are therefore equal. This completes the proof.
u
t
15 The numbers given in parentheses are the numbers of the propositions (e.g. Satz) as given
in [SST83].
16 This lemma is present in [BN12] as it corresponds to Hilbert’s version of Pasch’s axiom.
The proof of Proposition 1, while being simple, illustrates the impact of working
in an intuitionistic setting rather than a classical one. Indeed, in this proof we assert
the existence of points at the intersection between two lines, namely the points
B3 and C3 . Since we do not assume Axiom 1 (decidability of intersection of lines)
these points can be proved to exist from the axioms. However, it often happens
that, in a proof, the existence of a point at the intersection between two lines is
derived by contradiction, rendering it only valid in a classical setting. Thus, with a
view to prove Theorem 1, we had to be very careful not to employ such arguments.
5 Postulates equivalent to Tarski’s parallel postulate
This section follows the same outline to that used in the previous section. First, we
present the postulates which are NLJ -equivalent to Postulate 1 (Tarski’s parallel
postulate), together with the necessary definitions. Second, we will discuss the
formalization of the equivalence proofs.
5.1 Statements of postulates equivalent to Tarski’s parallel postulate
We introduce here eight new postulates. All are NLJ -equivalent to Tarski’s parallel
postulate. Three pairs among these eight postulates could appear to be quite
similar. Two of these pairs even express a seemingly analogous property, or so
it would seem. We will examine the slight differences which, while considering
a pair of these postulates, render unclear whether one is stronger, equivalent or
weaker than the other one. These postulates will correspond to properties about
parallelism, intersection, perpendicularity, triangles or angles. As in the previous
section, the subjects of these postulates are widely different.
Postulate 13 (Proclus’ postulate).
Definition proclus_postulate := forall A B C D P Q,

Par A B C D -> Col A B P -> ~ Col A B Q ->
exists Y, Col P Q Y /\ Col C D Y.
C D
D C
Y
A
P P
A B
B Q
Fig. 31: Proclus’ postulate (Postulate 13).

The first of these postulates (Fig. 31) is known as Proclus’ postulate. It asserts
that if a line intersects one of two parallel lines, then it intersects the other. One
can remark that this statement is the contrapositive of Postulate 5 (the postulate
of transitivity of parallelism). It is constructively stronger than its contrapositive,
which follows from the fact that in intuitionistic logic, an implication is not equiv-
alent to its contrapositive. In fact, only one of the implications remains valid when
dropping the law of excluded middle, namely (P ⇒ Q) ⇒ (¬Q ⇒ ¬P ).
Postulate 14 (Alternative Proclus’ postulate).
Definition alternative_proclus_postulate := forall A B C D P Q,

Perp2 A B C D P -> Col A B P -> ~ Col A B Q ->
exists Y, Col P Q Y /\ Col C D Y.
C D
D C
Y
A B
P P
B Q
Fig. 32: Alternative Proclus’ postulate (Postulate 14).
This postulate (Fig. 32) is a special case of Postulate 13. Compared to it,
Postulate 14 presents the same modifications as the one we applied to Postulate 2
(Playfair’s postulate) to obtain Postulate 12 (Alternative Playfair’s postulate).
Therefore we decided to name it alternative Proclus’ postulate. We recall that
there may be more than one parallel to a given line passing by a given point.
Thus, considering a particular one can be more convenient. We would like to
stress that this postulate, unlike the next postulates which will resemble another
previously defined postulate, really is just defined as a mean to ease some proofs
of implication.
Postulate 15 (Triangle circumscription principle).
Definition triangle_circumscription_principle := forall A B C,

~ Col A B C -> exists CC, Cong A CC B CC /\ Cong A CC C CC.
This postulate (Fig. 33) is referred to as triangle circumscription principle

in [Bee16]. It states that for any three non-collinear points there exists a point
equidistant from them. This version was originally used by Szmielew as an ax-
iom, but later Schwabhäuser chose Postulate 1 (Tarski’s parallel postulate) over
it [Bee16]. This postulate was the triggering factor for this study. Indeed, in [BNSB14a],
we used this version of the parallel postulate since we could derive Axiom 1 (the de-
cidability of intersection) of lines from it. Thus, we wanted to investigate whether
or not the same could be done with Tarski’s parallel postulate.
A B
A B
CC
C
C
Fig. 33: Triangle circumscription principle (Postulate 15).
Postulate 16 (Inverse projection postulate).
Definition inverse_projection_postulate := forall A B C P Q,

Acute A B C ->
Out B A P -> P <> Q -> Per B P Q ->
exists Y, Out B C Y /\ Col P Q Y.
Q
C C
Q
P
P A
A
B B
Fig. 34: Inverse projection postulate (Postulate 16).
This postulate (Fig. 34) expresses that, for any given acute angle, any point
together with its orthogonal projection on one side of the angle form a line which
intersects the other side. It will be designated as inverse projection postulate. It
is interesting to notice that although this postulate belongs to the strongest class
of postulates that we will consider, a modification of its statement (Postulate 31)
would render it much weaker to the point that it would belong to the weakest class
of postulates we will consider. It could seem like it trivially implies Postulate 1
(Tarski’s parallel postulate). Indeed, one could think it suffices to construct the
orthogonal projection of the considered point on the bisector of the angle (which
makes an acute angle with both sides of the angles) and, with the inverse projection
postulate, to assert the existence of a point on the other side of the angle which is
collinear with these two points. However, betweenness properties required in the
statement of Postulate 1 would not be satisfied and one would not be able to prove
the implication in such fashion.
The next postulates that we will present were introduced by Beeson in [Bee16].
In this paper, Beeson uses strict betweenness, similarly to Hilbert. Since we assume
the axioms of Tarski’s system of geometry, in which the betweenness is non-strict,
we need to define the strict betweenness.
Definition BetS A B C : Prop := Bet A B C /\ A <> B /\ B <> C.
In [Bee15], Beeson mentions that the strict and non-strict betweenness “are
interdefinable (even constructively)”. We adopted his definition of the strict be-
tweenness in terms of the non-strict betweenness. One can remark that, since we
assumed the decidability of point equality, in case we would have had to define the
non-strict betweenness in terms of the strict betweenness, we could have adopted a
simpler version of Beeson’s definition. Actually, he applies Gödel-Gentzen transla-
tion to the formula that we would have chosen to obtain a constructively valid def-
inition. We could have chosen to define A B C as A B C ∨A = B ∨B = C , while
he defines it ¬ (¬A B C ∧ A 6= B ∧ B 6= C ). Nevertheless, under the assumption
of the decidability of point equality, these two definitions are equivalent.
Postulate 17 (Euclid 5).
Definition euclid_5 := forall P Q R S T U,

BetS P T Q -> BetS R T S -> BetS Q U R -> ~ Col P Q S ->
Cong P T Q T -> Cong R T S T ->
exists I, BetS S Q I /\ BetS P U I.
R R
P P
U
U I T
T
Q
Q S
S
Fig. 35: Euclid 5 (Postulate 17).
This postulate (Fig. 35) is the first of two postulates introduced in [Bee16] by
Beeson. It is a formulation of Euclid’s parallel postulate in Tarski’s language. He
denotes it as Euclid 5. He writes that Euclid 5 is
“If a straight line falling on two straight lines make the interior angles on
the same side less than two right angles, the two straight lines, if produced
indefinitely, meet on that side on which are the angles less than the two
right angles.”
He reads “make the interior angles on the same side less than two right angles”
into line P U being in the interior of the angle ∠QP R while lines P R and QS make
consecutive interior angles17 with P Q equal to two right angles. Seeing that, in
neutral planar geometry, making consecutive interior angles equal to two right an-
gles is the same as making alternate interior angles equal, he uses this equivalent
statement. In his definition, given that the two straight lines that make alternate
interior angles equal are P R and QS , he formulates it as the quadrilateral P RQS
having its diagonals meeting in their midpoint. Yet, it is not obvious that a quadri-
lateral having its diagonals meeting in their midpoint means that their opposite
sides make alternate interior angles equal. This property follows from Satz 7.13
of [SST83], which is provable in neutral planar geometry and uses the definition
of angle congruence.
Postulate 18 (Strong parallel postulate).
Definition strong_parallel_postulate := forall P Q R S T U,

BetS P T Q -> BetS R T S -> ~ Col P R U ->
Cong P T Q T -> Cong R T S T ->
exists I, Col S Q I /\ Col P U I.
R R
U U
P P
I T
T
Q
Q S
S
Fig. 36: Strong parallel postulate (Postulate 18).
This postulate (Fig. 36), also introduced and named as strong parallel postulate
by Beeson in [Bee16], results of the modification of Euclid 5. Both its hypotheses
and its conclusion are weaker compared to it. The point U defined in the previous
postulate is not supposed to lie inside one of the considered alternate interior
angles, but to lie outside line P R. That is, the interior angles on the same side are
no longer required to make less than two right angles, but prevented to sum exactly
to two right angles. Moreover, the strict betweenness predicates in the conclusion
are replaced by collinearity predicates. That is, the two straight lines making
interior angles which do not sum to two right angles are asserted to meet without
any indication on the side of this intersection. Finally, contrary to Postulate 17,
the lines P R and SQ can be equal. This hypothesis was crucial for Postulate 17
as it avoids the case where P = U in which the postulate is false. Because both
the hypotheses and the conclusion are weaker compared to Euclid 5, it is not
evident whether these modifications render the strong parallel postulate stronger
than Euclid 5, equivalent, or weaker.
17 We previously referred to interior angles on the same side of a straight line as consecutive
interior angles.
To have a more faithful version of Euclid’s parallel postulate we introduced a

variant of Postulate 17 (Euclid 5). For this sake, we stated this variant in terms
of sum of angles. We first introduce a variant of Postulate 18 (Strong parallel
postulate) stated in terms of sum of angles.
Postulate 19 (Alternative strong parallel postulate).
Definition alternative_strong_parallel_postulate := forall A B C D P Q R,

OS B C A D -> SumA A B C B C D P Q R -> ~ Bet P Q R ->
exists Y, Col B A Y /\ Col C D Y.
R R
D Q
C Q
P D
C P
Y
B B
A
A
Fig. 37: Alternative strong parallel postulate (Postulate 19).
This postulate (Fig. 37) greatly resembles the previous one. Therefore we de-
cided to name it alternative strong parallel postulate. In this version we make
explicit the concept of sum of angles. In the same fashion as for the triangle pos-
tulate, the fact that the interior angles on the same side do not sum to exactly
two right angles is formulated as this sum not being equal to a straight angle. Fur-
thermore, one can notice that, compared to the previous postulate, the lines AB
and CD, which correspond to the lines P R and QS , are not equal. This is a due
to the hypothesis stating that A and D are on the same side of line BC . Nonethe-
less, since in the axiom system we adopted, the degenerate case of this statement
is trivial and the line equality is decidable, this difference does not impact the
possibility for these two postulates to be equivalent.
To define a variant of Euclid 5 making an explicit use of the concept of sum of
angles, we need to be able to characterize the property for two angles to make less
than two right angles. Incidentally, a property very similar to this one is essential
when considering the sum of angles. According to Rothe [Rot14], if this property
is not satisfied, the considered angles cannot be added, because “the sum would
be an over-obtuse angle”. In fact, the sum of angles is neither an order-preserving
function nor an associative function when some of the considered sums correspond
to over-obtuse angles. For example, 160◦ = (20◦ +170◦ )+30◦ 6= 20◦ +(170◦ +30◦ ) =
180◦ .
Definition SAMS A B C D E F :=
A <> B /\ (Out E D F \/ ~ Bet A B C) /\
exists J, CongA C B J D E F /\ ~ OS B C A J /\ ~ TS A B C J.
The name of this predicate is the abbreviation for sum at most straight. Two
angles ∠ABC and ∠DEF do not make an over-obtuse angle if there exists a point J
such that C B J =b D E F , the angles ∠ABC and ∠CBJ are adjacent and the angle
∠ABJ is not an over-obtuse angle. As for the definition of sum of angles (SumA), we
specified that angles ∠ABC and ∠CBJ are adjacent by the fact that A and J are
not on the same side of line BC , to do without the disjunction of cases between the
cases where at least one of the angles is degenerate and the case where both angles
are non-degenerate and A and J are on opposite sides of line BC . Interestingly, by
formalizing straightforwardly “do not make an over-obtuse angle”, one also avoids
such a disjunction of cases. This predicate almost corresponds to property for two
angles to make less than two right angles. It just does not exclude the case where
the two angles make exactly two right angles. Analogously to the predicates for
order relations on segments and angles, it is straightforward to exclude this case.
Postulate 20 (Euclid’s parallel postulate).
Definition euclid_s_parallel_postulate := forall A B C D P Q R,

OS B C A D -> Isi A B C B C D -> SumA A B C B C D P Q R -> ~ Bet P Q R ->
exists Y, Out B A Y /\ Out C D Y.
R P R P
Q
C D Q
C
D
Y
B
B
A A
Fig. 38: Euclid’s parallel postulate (Postulate 20).
This variant of Postulate 17 (Euclid 5) being intended as a more faithful version

of Euclid’s parallel postulate, we will refer to it as Euclid’s parallel postulate.
One can notice that, compared to Postulate 17 (Euclid 5), the strict betweenness
predicates in the conclusion are replaced by Out predicates (we recall that Out B A
Y expresses that Y belongs to the ray BA). This weakening of the conclusion is due
to the fact that, in this version, we state the hypothesis that the considered lines
“make the interior angles on the same side less than two right angles” without
referring to an angle in which one of these lines lies, namely P U being in the
interior of the angle ∠QP R in the definition of Postulate 17. Since lying in an
angle was expressed in terms of betweenness, it allowed us to be more precise
regarding the position of the intersection of the considered lines. This statement
is really close to the three previous postulates. However, once more, since both its
hypotheses and its conclusion are either stronger or weaker than the ones of these
three postulates, it is not obvious that they are equivalent.
10. 18.
11. 9. 15. 13.
12. 2. 1. 14.
7. 5. 17. 16.
8. 6. 20. 19.
This section is dedicated to the formalization of the proof that the postulates of
the previous subsection (Postulates 13-20) are indeed NLJ -equivalent to Postu-
late 1 (Tarski’s parallel postulate) as well as the formalization of the proof that
the postulates of Subsections 4.1 and 5.1 are indeed NLK -equivalent to Postulate 2
(Playfair’s postulate) and Postulate 1 (Tarski’s parallel postulate). As in the previ-
ous proof of equivalence, this equivalence is proved in the context of the Tarski_2D
type class from Tab. 2. The Coq statements corresponding to the NLJ -equivalence
of any two of Postulates 1, 13-20 is the following.
Theorem equivalent_postulates_without_decidability_of_intersection_of_lines_bis :
all_equiv
(alternative_strong_parallel_postulate::
alternative_proclus_postulate::
euclid_5::
euclid_s_parallel_postulate::
inverse_projection_postulate::
proclus_postulate::
strong_parallel_postulate::
tarski_s_parallel_postulate::
triangle_circumscription_principle::
nil).
A graphical summary of the implications that we formalized to prove Theo-

rem 2 is displayed on Fig. 39. The circles around Postulates 2, 5-12 and around Pos-
tulates 1, 13-20 mean that the postulates inside these circles are NLJ -equivalent.
One could think that Postulate 1 does not imply Postulate 15 (the triangle cir-
cumscription principle) in an intuitionisctic logic. Indeed, in order to prove this
implication, we proved that Postulate 1 implies Postulate 2, which implies Postu-
late 9 (the perpendicular transversal postulate), itself implying Postulate 15.
In fact, even if Postulate 9 (NLJ -equivalent to Postulate 2) does not imply

Postulate 15 (equivalent to Postulate 1) in an intuitionisctic logic, we know from
Proposition 3 that the decidability of intersection of lines (Axiom 1) follows from
Postulate 18, which itself follows from Postulate 1. Furthermore, Proposition 2
demonstrates that, in an intuitionisctic logic, assuming Axiom 1 is enough to

prove that Postulate 9 implies Postulate 15.
The implications displayed on Fig. 39 allow us to prove the following theorem.
Theorem 2 Postulates 1, 13-20 are NLJ -equivalent and Postulates 1, 2, 5-20 are
NLK -equivalent.
In an earlier version of this work, we were proving directly that Postulate 17

(Euclid 5) implies Postulate 18 (the strong parallel postulate). The idea behind
this proof was to add an extra hypothesis in Postulate 18, namely that the points
P , Q, R and U are coplanar. The motivation behind this idea was double. First,
this extra hypothesis is necessary in spaces of dimension higher than two. Second,
we could then reason by distinction of cases on the 27 possibilities for these points
to be coplanar. This distinction of cases was allowing us to know to which side of
lines P R and P S the point U belongs. So we were left with four cases corresponding
to the four parts of the plane to which all the considered points belong. We could
then use Pasch’s axiom in all of these cases to construct a point permitting to
apply Postulate 17 and complete this proof.
We already mentioned that Postulate 1 is valid in spaces of dimension higher
than two. This is due to the fact that all the points in its statement are coplanar.
Therefore the extra hypothesis that we added to Postulate 18 was not altering the
possibility to prove that Postulate 1 follows from it. Because Postulate 1 was the
only postulate that was proved to directly follow from Postulate 18, we could then
prove that this modified version was equivalent to the postulates of Section 5.
This proof was really tedious, even though we could slightly simplify it when we
proved that, in the context of planar neutral geometry with decidable point equal-
ity, the upper 2-dimensional axiom was equivalent to the fact any four points are
coplanar. In doing so, we were in fact proving the “two-sides” principle from [Bee16]
without relying on Axiom 1. This principle asserts that two points A and B not on
a line l are either on the same side of l or on opposite sides of l. We would like to
stress that this demonstrates a profound difference between the axiom system we
adopted and Beeson’s modification of Tarski’s axioms [Bee15] to which the results
of [Bee16] apply. Indeed, according to Theorem 10.3 from [Bee16], the “two-sides”
principle is not provable. Therefore this proof could not be done in his system,
which is why he proves that Postulate 17 implies Postulate 18 by showing that
“Euclid 5 suffices to define coordinates, addition, multiplication, and square roots
geometrically”.
In the current version of this work, this proof is not present anymore. Indeed,
when we started to consider Postulate 19 (the alternative strong parallel postulate)
and Postulate 20 (Euclid’s parallel postulate) we realized that it was straightfor-
ward to prove not only the implication from Postulate 17 to Postulate 20 but,
more surprisingly, also the one from Postulate 20 to Postulate 19. This is a result
of the fact we started to consider these postulates after the development of a small
library for the sum of angles which proved very useful for this proof. Moreover, the
proof that Postulate 19 implies Postulate 18 was also not as cumbersome as the
proof of the implication from Postulate 17 to Postulate 18. As a matter of fact, we
had already proved that Postulate 13 (Proclus’ postulate). implies Postulate 18
and we found a proof of the implication from Postulate 19 to Postulate 13 which
was quite direct to formalize. The major idea behind this proof was to use two
intermediary steps, namely Postulate 16 (the inverse projection postulate) and

Postulate 14 (Alternative Proclus’ postulate). The purpose of using these postu-
lates as an intermediary steps was that they feature hypotheses which could be
translated into each other with ease. This highlights a central issue when working
with parallel postulates: the variety of the properties which are used to state the
different postulates. When proving the equivalence between parallel postulates,
one should be careful to which implication one will prove, since the difficulty to
translate one property into another is far from being constant.
Similarly to the previous section we will only focus on a few proofs, namely
Proposition 2 and Proposition 3. We chose to focus on these proofs because they
are the only implications that involve Axiom 1. Unlike the previous section, up to
our knowledge, the only synthetic and intuitionistic proofs of Section 5 which can
be found in the literature are the implication from Postulate 1 to Postulate 17.
This implication is proved in [Bee15] (Theorem 8.3). In order to present the proof
of Proposition 2, we will collect four lemmas that will be used throughout this
proof.
Lemma 5 (8.22) Midpoints are constructible.
Lemma 6 Given two distinct points, their perpendicular bisector is constructible.
Lemma 7 (12.9) Two lines perpendicular to the same line are parallel.
Lemma 8 (12.17) If A and B are distinct and if the segments AC and BD have the
same midpoint, then the lines AB and CD are parallel18 .
The following proposition is a classic, but we still give the proof, because we
are in a intuitionistic setting and we want to emphasize the use of the decidability
of intersection.
Proposition 2 Axiom 1 and Postulate 9 imply Postulate 15.
Proof
Given a non-degenerated tri- B2
angle ABC we wish to prove the C2
C
B
existence of point CC equidistant
to A, B and C (Fig. 40). Lemma 6 CC
lets us construct the perpendicu-
lar bisector C1 C2 of the segment C1
AB and the perpendicular bisec- B1
tor B1 B2 of the segment AC , since A
they are non-degenerate segment
Fig. 40: Axiom 1 and Postulate 9 imply Pos-
as ABC is a non-degenerate tri-
tulate 15.
angle. We now prove that it is im-
possible for lines B1 B2 and C1 C2 to not intersect to prove the existence of this
intersection19 . Assuming they do not intersect, then lines B1 B2 and C1 C2 are par-
allel by definition. Using the perpendicular transversal postulate we can deduce
18 It almost corresponds to the fact that the opposite sides of a non-degenerated quadrilateral
which has its diagonals intersecting in their midpoint are parallel. To fully correspond to this
fact one would need to add the hypothesis that A and D are distinct.
19 Note that we use here the decidability of intersection of lines.
that lines AC and C1 C2 are perpendicular. Finally Lemma 7 establishes that lines
AB and AC are parallel as they are both perpendicular to line C1 C2 . This implies
that A, B and C are collinear, which is false by hypothesis. Since it is impossi-
ble for lines B1 B2 and C1 C2 to not intersect, Axiom 1 lets us assert that Cc is
their intersection point, which is equidistant from A and B since it belongs to
its perpendicular bisector and equidistant from A and C since it belongs to its
perpendicular bisector. u t
Proposition 3 Postulate 18 implies Axiom 1.
U
Proof Given 4 points that we name R
P
P , Q, S and U (rather than A, B ,
C and D, to work with the same
T I
name as those in the definition of
the strong parallel postulate) we
Q
wish to prove that either there ex- S
ists a point I such that Col I S Q
and Col I P U or that there does
not exist such a point (Fig. 41). Fig. 41: Postulate 18 implies Axiom 1.
We first eliminate the case where
P lies on QS in which there exists such a point I , namely it is P . So we may
assume that ¬ Col P Q S and we then eliminate the case of P and U being equal,
as again there exists such a point I , namely Q (we could have also taken S to be
this point). So we may assume P and U to be different. Now we construct the
midpoint T of the segment P Q, using Lemma 5, and the symmetric point R of S
with respect to T , using Lemma 2. Finally we will distinguish two cases. Either
¬ Col P R U and the strong parallel postulate asserts there exists such a point I ,
provided that P T Q and R T S , which we easily prove as P and Q are different
and ¬ Col P Q S . The other case is when Col P R U . In this case we can prove that
lines QS and P U are strictly parallel, using Lemma 8 and the fact ¬ Col P Q S ,
and by definition of two lines being strictly parallel we know that there does not
exist such a point I . u
t
6 Postulates equivalent to the triangle postulate
The same structure that was used in the previous two sections will be used through-
out this one. First, we present the postulates which are NLJ -equivalent to Postu-
late 3 (the triangle postulate), together with the necessary definitions. Second, we
will discuss the formalization of the equivalence proofs.
6.1 Statements of postulates equivalent to the triangle postulate
This section will study ten new postulates. All are NLJ -equivalent to the triangle
postulate. One, namely Postulate 21, is very similar to Postulate 3 (the triangle
postulate) but one could wrongly think it is strictly weaker than it. Furthermore,
three pairs of postulates will present the same kind of similarity. Despite these
resemblance, the subjects of these postulates are again mostly heterogeneous. In
fact, these postulates will affirm properties about triangles, equidistant lines, cir-
cles and quadrilaterals.
Postulate 21 (Postulate of existence of a triangle whose angles sum to two
rights) and Postulate 22 (Posidonius’ postulate).
Definition postulate_of_existence_of_a_triangle_whose_angles_sum_to_two_rights :=
exists A B C D E F, ~ Col A B C /\ TriSumA A B C D E F /\ Bet D E F.
Definition posidonius_postulate :=
exists A1 A2 B1 B2,
~ Col A1 A2 B1 /\ B1 <> B2 /\
forall A3 A4 B3 B4,
Cong A3 B3 A4 B4.
These two postulates correspond to trivial consequences of Postulate 3 and

Postulate 11 (the universal Posidonius’ postulate). Indeed their definitions are
nearly the same as those of these last two postulates. Postulate 21 expresses that
there exists a triangle whose angles sum to two rights and Postulate 22 expresses
that there exists two lines which are everywhere equidistant. They mainly differ in
the type of quantifiers used for some of the considered points in these postulates:
they replace some of the universal quantifiers by existential ones. Postulate 3
and Postulate 11 are also more general and can be instantiated to cases which
are provable in planar neutral geometry. So Postulate 21 and Postulate 22 add
non-degeneracy conditions. For example, in the case of the former one adds that
the triangle whose angles sum to two rights is non-flat. The latter is particularly
interesting because it represents one of the only two postulates20 which is not NLJ -
equivalent to its universally quantified version. In this section, we will consider
three more pairs of postulates differing in the type of quantifiers, and all of them
are NLJ -equivalent. Furthermore, Playfair’s postulate is proved to be equivalent
to the existence of a point and line for which there is a unique parallel line passing
through the point in [Ami33]. However, we did not formalize this proof since it
requires the space to be of dimension higher than two. Indeed, it relies on the
existence, for any given plane, of a point not incident to it. The same theorem has
also been proven synthetically by Piesyk [Pie61], but his proof needs decidability
of intersection of lines.
Postulate 23 (Postulate of existence of similar but non-congruent triangles).
Definition postulate_of_existence_of_similar_triangles :=
exists A B C D E F,
~ Col A B C /\ ~ Cong A B D E /\
CongA A B C D E F /\ CongA B C A E F D /\ CongA C A B F D E.
The postulate of existence of similar but non-congruent triangles (Fig. 42) is

a simplication suggested by Saccheri to a postulate introduced by Wallis [Mar98]:
20 Here we restrict ourselves to the postulates that we formalized.
B
B
E
E
D F C A
D F C A
Fig. 42: Postulate of existence of similar but non-congruent triangles (Postu-

late 23).
“To every figure there exists a similar figure of arbitrary magnitude”. It asserts
that there exists two similar but non-congruent triangles. Wallis assumed this pos-
tulate in order to prove Euclid’s parallel postulate [Bon55] but could have instead
assumed Postulate 23. This postulate was also assumed by Laplace [Caj98]. More-
over, Gauss produced a proof of Euclid’s parallel postulate under the assumption of
the existence of a right triangle whose area is greater than any given area [Lew20].
“Zwar bin ich auf manches gekommen, was den meisten schon für einen
Beweis geltend würde, aber was in meinen Augen sogut wie Nichts beweiset,
z. B. wenn man beweisen könnte dass ein geradlinigtes Dreieck möglich sei,
dessen Inhalt grösser wäre als eine jede gegebne Fläche, so bin ich im Stande
die ganze Geometrie völlig streng zu beweisen.”21
– Carl Friedrich Gauss [GB99]
It is unclear if the right triangle is required to be similar to another given
right triangle. If so22 , and it is probable considering this sentence was part of an
informal letter from Gauss to Bolyai, Gauss’ assumption would be a special case
of Wallis’ postulate. One should point out that, even though the formalization of
this postulate is straightforward, the triangles need to be non-flat, as the non-
degeneracy conditions are often omitted in geometry texts.
Postulate 24 (Thales’ postulate) and Postulate 25 (Thales’ converse postu-
late) and Postulate 26 (Existential Thales’ postulate).
Definition thales_postulate := forall A B C M,

~ Col A B C -> Midpoint M A B -> Cong M A M C -> Per A C B.
Definition thales_converse_postulate := forall A B C M,

~ Col A B C -> Midpoint M A B -> Per A C B -> Cong M A M C.
Definition existential_thales_postulate :=
exists A B C M, ~ Col A B C /\ Midpoint M A B /\ Cong M A M C /\ Per A C B.
21 “It is true that I have come upon much which by most people would be held to constitute
a proof; but in my eyes it proves as good as nothing. For example, if we could show that
a rectilinear triangle whose area would be greater than any given area is possible, then I
would be ready to prove the whole of [Euclidean] geometry absolutely rigorously.” The English
translation is borrowed from [Kli90].
22 Otherwise, it would provide yet another illustration of the gravity of definitions.
A
A A
C C
C
M M M
B B
Fig. 43: Thales’ postulate (Postulate 24), Thales’ converse postulate (Postulate 25)
and existential Thales’ postulate (Postulate 26).
Here we discuss simultaneously Postulate 24, Postulate 25 and Postulate 26,

because the second one is the converse of the first one. Moreover, the third one
corresponds to the result of replacing, in the first or second one, the universal
quantifiers by existential ones, and the implication between the hypotheses and the
conclusion by a conjunction. Postulate 24 states that, if the circumcenter of a tri-
angle is the midpoint of a side of a triangle, then the triangle is right. Postulate 25
states that, in a right triangle, the midpoint of the hypotenuse is the circumcen-
ter. Finally, Postulate 26 states that there is a right triangle whose circumcenter
is the midpoint of the hypotenuse. Fig. 43 displays the figure representing the
statement of the postulates in the Euclidean plane on the left. A counter-example
in Poincaré’s disk model for Postulate 24 can be found in the center of Fig. 43 and
one for Postulate 25 can be on the right of Fig. 43. There is no counter-example for
Postulate 26, for the reason that it does not state a property that some geometric
objects verify in a given configuration, but rather the existence of some geometric
objects verifying a given property. Martin qualifies Postulate 24, which is a special
case of the inscribed angle theorem (part of Proposition III.31 in [EHD02]), as
“certainly one of the oldest theorems in mathematics”. The proofs of Postulate 24
and Postulate 25, as theorems of Euclidean geometry, have already been studied
in Coq assuming Tarski’s system of geometry [BM15]. Nevertheless, Braun et al.
proved that they both follow from Postulate 6 (the midpoint converse postulate),
which is strictly stronger than both of them. Finally, formalizing these postulates
is elementary.
Postulate 27 (Postulate of right Saccheri quadrilaterals) and Postulate 28 (Pos-
tulate of existence of a right Saccheri quadrilateral).
Definition Saccheri A B C D :=
Per B A D /\ Per A D C /\ Cong A B C D /\ OS A D B C.
Definition postulate_of_right_saccheri_quadrilaterals:= forall A B C D,

Saccheri A B C D -> Per A B C.
Definition postulate_of_existence_of_a_right_saccheri_quadrialteral :=
exists A B C D, Saccheri A B C D /\ Per A B C.
We now focus on a postulate due to Saccheri, who made “the most elaborate
attempt to prove the ‘parallel postulate’” according to Coxeter [Cox98] and was
B C B C
A D
A D
Fig. 44: Definition of Saccheri.
“perhaps before its time” [Har00]. In his attempt to prove Euclid’s parallel postu-
late, he considered a specific kind of quadrilaterals which have since been named
after him. These quadrilaterals arise when one studies points that are equidistant
to a line. Indeed, S A B C D is a quadrilateral such that the angles at A and D
are right and AB ≡ CD (Fig. 44). Still, one needs to add the fact that B and C
are on the same side of line AD23 . Saccheri’s investigation of such quadrilaterals
was influenced by Clavius’ work about Postulate 11 [Har00]. He considered three
cases for these quadrilaterals, when the remaining angles are either acute, right or
obtuse, known as Saccheri’s three hypotheses. He was meaning to prove Euclid’s
parallel postulate by eliminating the hypotheses of the acute and obtuse angle. As
we will see in the next section, in Archimedean neutral geometry, one can eliminate
the hypothesis of the obtuse angle. Nonetheless, one cannot eliminate the hypoth-
esis of the acute angle, which corresponds to hyperbolic geometry. Postulate 27
expresses that the hypothesis of the right angle holds and Postulate 28 expresses
that there exists a right Saccheri quadrilateral.
Postulate 29 (Postulate of right Lambert quadrilaterals) and Postulate
30 (Postulate of existence of a right Lambert quadrilateral).
Definition Lambert A B C D :=
A <> B /\ B <> C /\ C <> D /\ A <> D /\ Per B A D /\ Per A D C /\ Per A B C.
Definition postulate_of_right_lambert_quadrilaterals := forall A B C D,

Lambert A B C D -> Per B C D.
Definition postulate_of_existence_of_a_right_lambert_quadrialteral :=
exists A B C D, Lambert A B C D /\ Per B C D.
The last postulates that we analyze in this section are closely related to to
Saccheri quadrilaterals. Indeed, they regard quadrilaterals that were also studied
by Saccheri, though they are named after Lambert [Gre10]. L A B C D has right
angles at A, B and D (Fig. 45). The reason why Saccheri studied them is be-
cause by taking N such that A N D and M such that B M C in a Saccheri
quadrilateral S A B C D, one obtains two Lambert quadrilaterals L N M B A and
L N M C D. Lambert proceeded in the same way as Saccheri in his attempt at
proving Euclid’s parallel postulate, namely, disproving the obtuse case and trying
23 Quadrilaterals are usually implicitly assumed to be non-crossed.
C
B C
A D
A D
Fig. 45: Definition of Lambert.
to derive a contradiction from the acute case [Gre93]. Postulate 29 and Postu-
late 30 state, respectively, that all Lambert quadrilaterals are rectangles and that
there exists a rectangle. One could think that this postulate is close to Postulate 4
(Bachmann’s Lotschnittaxiom), but Postulate 4 asserts the existence of an inter-
section point, while Postulate 29 states the perpendicularity of two lines known to
intersect.
In this section, we address the formalization of the proof that the postulates of the
previous subsection (Postulates 21-30) are indeed NLJ -equivalent to Postulate 3
G
(the triangle postulate), as well as the NLJ -equivalence between Postulate 3 and
Postulate 2 (Playfair’s Postulate). Exactly like in the previous proofs of equiva-
lence, the first equivalence is proved in the context of the Tarski_2D type class
from Tab. 2. The Coq statement corresponding to the NLJ -equivalence of any two
of Postulates 3, 21-30 is the following.
Theorem equivalent_postulates_without_any_continuity :
all_equiv
(existential_thales_postulate::
posidonius_postulate::
postulate_of_existence_of_a_right_lambert_quadrilateral::
postulate_of_existence_of_a_right_saccheri_quadrilateral::
postulate_of_existence_of_a_triangle_whose_angles_sum_to_two_rights::
postulate_of_existence_of_similar_triangles::
postulate_of_right_lambert_quadrilaterals::
postulate_of_right_saccheri_quadrilaterals::
thales_postulate::
thales_converse_postulate::
triangle_postulate::
nil).
One can remark, from the graphical summary of the implications we proved
(Fig. 46), that Postulate 27 (the postulate of right Saccheri quadrilaterals) plays
a very central role. There is a simple explanation for it: most of these proofs
correspond to the formalization of the proofs of some of Saccheri’s propositions
given in [Mar98]. In this book, Martin establishes equivalences between each of
28.
30. 27.
29. 21.
26. 3.
25. 22.
24. 23.
Saccheri’s three hypotheses and whether certain angles are acute, right or obtuse.
Most of these implications follow easily from these propositions. In order to for-
malize Martin’s proofs, we often proceeded by disjunction of cases on Saccheri’s
three hypotheses. One should point out that, because case distinctions cannot be
performed in existence proofs in Beeson’s modification of Tarski’s axioms [Bee15],
some of the proofs we mechanized would not be valid in his axiomatic system.
We can now consider the visual representation of all the implications that we
formalized to prove Theorem 3 (Fig. 47). Comparing with Fig. 39 and Fig. 46, one
can see two extra implications displayed, namely from Postulate 3 to Postulate 12
(the alternative Playfair’s postulate) and from Postulate 7 (the alternate inte-
rior angles postulate) to Postulate 3. Indeed, these implications are not necessary
to prove that any two postulates that belong to the same circle are equivalent.
Nonetheless, in order to prove the following theorem, they are necessary.
G
Theorem 3 Postulates 3, 21-30 are NLJ -equivalent and Postulates 1-3, 5-30 are NLJ -
equivalent.
For the sake of completeness, we list the propositions given in [Mar98] that cor-
respond to the implications on Fig. 46. Theorems 22.3 and 22.10 allow us to prove
that Postulate 27 implies Postulate 29 and is implied by Postulate 30. The impli-
cations from Postulate 28 to Postulate 27 and from Postulate 27 to Postulate 3 are
respectively proved in Theorem 22.10 and Corollary 22.13. From Theorem 22.17
we could deduce that Postulate 27 implies Postulate 24 and is implied by Pos-
tulate 26. The fact that Postulate 24 implies Postulate 25 and that Postulate 21
implies Postulate 27 are showed in Theorem 23.7. The implications from Postu-
late 3 to Postulate 21, from Postulate 27 to Postulate 28, from Postulate 29 to
Postulate 30 and from Postulate 25 to Postulate 26 are trivial. Indeed, in each
of these implications, one needs to prove that a postulate implies another where
some of the universal quantifiers are replaced by existential ones. Thus one only
needs to assert the existence of a non-degenerated triangle, a Saccheri quadrilat-
eral, a Lambert quadrilateral and a non-degenerate right triangle. The proof that
Postulate 27 is equivalent to Postulate 22 is done in Theorem 23.6 for one side of
the equivalence (but using the notion of default for a triangle, which we avoided
in this section) and in Theorem 23.7 for the other side. Finally, the proof that
Postulate 27 is equivalent to Postulate 23 is left as exercise.
28. 10. 18.

30. 27. 11. 9. 15. 13.
29. 21.
12. 2. 1. 14.
26. 3.
7. 5. 17. 16.
25. 22.
24. 23. 8. 6. 20. 19.
Fig. 47: Overview of the proofs.
Lastly, we will detail one proof and compare another one to the pen-and-paper
proof from which it is inspired24 . Both of these proofs illustrate one of the main
difference between a theoretical proof and the actual Coq proof, namely dealing
with non-degeneracy conditions and betweenness properties. This difference repre-
sents one of the main difficulties that one encounters while formalizing a proof in
synthetic geometry. These proofs allow us to study the impact of using the tactics
developed in [BNSB14b] as well as their limitations. The proof we have chosen to
study is the fact that Postulate 7 (the alternate interior angles postulate) implies
Postulate 3 (the triangle postulate). The pen-and-paper proof is short:
Let ABC be a triangle, construct the parallel to AC through B (Fig. 48).
Then, the two pairs of alternate interior angles displayed on the figure are
congruent, and hence the sum of the three angles is the straight angle.
Now, we will compare this argument with the formal proof as formalized in
Coq. In order to present the rigorous proof of Proposition 4, we will collect five
lemmas that will be used throughout this proof.
Lemma 9 (8.18) Dropped perpendiculars25 are constructible.

AC AB BC
Lemma 10 (9.8) If P Q and P Q then P Q.
AC AB BC
YZ XZ XY
Lemma 11 If A X and A Y then A Z.
YZ XZ XY
26
Lemma 12 A given angle can be laid off upon a given side of a given ray.
Lemma 13 (12.2127 ) If two lines share a common transversal which makes a pair of
alternate angles equal to one another, then the two lines are parallel.
Now, we give in natural language the proof at the level of details needed for
the formalization.
24 These proofs have already been presented in French [GBN16].
25 Usually in geometry, we give two different constructions for the perpendicular to a given
line in a given point, whether the given point lies on the given line or not. If it does, we “erect”
a perpendicular at this given point, and if it does not, we “drop” a perpendicular from this
given point to this given line.
26 This corresponds to the fourth axiom of Group IV from [Hil60].
27 This lemma represents only the part that is valid in neutral planar geometry.
Proposition 4 Alternate interior angles postulate (Postulate 7) implies the triangle

postulate (Postulate 3).
Proof
Given a triangle ABC and know-

B2 B B1
ing that S (4ABC ) = b D E F , we wish
to prove D E F (Fig. 48). We
first eliminate the case where B
lies on AC , in which D E F holds
trivially. Lemma 12 lets us con-
struct point B1 such that B C A = b C B B1 A C
B1 A
and C B . From Lemma 13 and
B1 A
¬ Col A B C we have that AC ks Fig. 48: Postulate 7 implies Postulate 3.
BB1 . Then Lemma 2 lets us con-
B1 B2
struct point B2 the symmetric of B1 with respect to B . We know that B A
B1 B2
because, by construction, the segment B1 B2 intersect the line AB in B and neither
B1 nor B2 belongs to the line AB , since otherwise it would contradict the fact that
B1 C
AC ks BB1 . From Lemma 11 we obtain that A B . Lemma 10 lets us derive from
B1 C
B1 B2 B1 C CB2
B A and A B that B A. By construction of B2 , ∠B1 BB2 is a straight
B1 B2 B1 C CB2
CB2
angle, hence it suffices to show that B1 B B2 =b D E F . By Postulate 7, B A and
CB2
AC k BB1 imply that A B B2 = b C A B . By construction, B C A =
b C B B1 , so we are
done. u t
The Coq proof for Proposition 4 is the following.
Lemma alternate_interior__triangle :
alternate_interior_angles_postulate ->
triangle_postulate.
Proof.
intros AIA A B C D E F HTrisuma.
elim(Col_dec A B C); [intro; apply (col_trisuma__bet A B C); auto|intro HNCol].
destruct(ex_conga_ts B C A C B A) as [B1 [HConga HTS]]; Col.
assert (HPar : Par A C B B1)
by (apply par_left_comm, par_symmetry, l12_21_b; Side; CongA).
apply (par_not_col_strict _ _ _ _ B) in HPar; Col.
assert (HNCol1 : ~ Col C B B1) by (apply (par_not_col A C); Col).
assert (HNCol2 : ~ Col A B B1) by (apply (par_not_col A C); Col).
destruct (symmetric_point_construction B1 B) as [B2 [HBet HCong]]; assert_diffs.
assert (HTS1 : TS B A B1 B2)
by (repeat split; Col; [intro; apply HNCol2; ColR|exists B; Col]).
assert (HTS2 : TS B A C B2)
by (apply (l9_8_2 _ _ B1); auto; apply os_ts1324__os; Side).
apply (bet_conga_bet B1 B B2); auto.
destruct HTrisuma as [D1 [E1 [F1 []]]].
apply (suma2__conga D1 E1 F1 C A B); auto.
assert (CongA A B B2 C A B).
{
apply conga_left_comm, AIA; Side.
apply par_symmetry, (par_col_par _ _ _ B1); Col; Par.
}
apply (conga3_suma__suma B1 B A A B B2 B1 B B2); try (apply conga_refl); auto;
[exists B2; repeat (split; CongA); apply l9_9; auto|].

apply (suma2__conga A B C B C A); auto.
apply (conga3_suma__suma A B C C B B1 A B B1); CongA.
exists B1; repeat (split; CongA); apply l9_9; Side.
Qed.
Thanks to the tactics developed in [BNSB14b] the Coq proof is fairly close to
the proof we just gave. The first main difference is that we need to deduce two
non-degeneracy conditions, namely HNCol1 and HNCol2. The second main difference
is not visible here. In fact, the proof that we just gave is different from usual proof
that the sum of the interior angles of a triangle is equal to two right angles, such
as the one given by Amiot [Ami70]28 . In Amiot’s proof, the fact that the angles
∠CAB and ∠ABB2 are alternate interior angles, HTS2 in the Coq proof, is stated
without a proof. This lack of justification for the relative position of the points
on the figure is a critic that the modern commentators of Euclid’s Elements often
make about Euclid’s proofs. However, Avigad et al. [ADM09] claim that these gaps
can be filled by some automatic procedure, justifying in some sense the gaps in
Euclid’s original proofs. This is where we reach the limits of our tactics: they only
handle incidence problems, permutation properties and compute the consequences
of the negation of the non-degeneracy conditions, but do not provide this kind of
justification. Therefore, it would be very useful to have an implementation of the
procedure proposed in [ADM09].
Our proof that Axiom 4 and Postulate 3 imply Postulate 12 is inspired from the
one Greenberg gives in [Gre10]. Nevertheless, we needed to make two modifications
to his proof. The first one is due to the fact that we used a different definition
for a point belonging to an angle, as we explained in Subsection 3.2.3. The other
modification that we made is due to the use of a proof assistant: because of it we
cannot skip the justification for the relative position of the points on the figure.
7 Postulates equivalent to Bachmann’s Lotschnittaxiom and the role of

Archimedes’ axiom
This section is devoted to the role of Archimedes’ axiom: we study the implica-
tions of assuming this property. We will first provide a proof of the independence of
Archimedes’ axiom from the axioms of Pythagorean planes29 . Then we will intro-
duce three postulates which we will prove NLJ -equivalent to Postulate 4 (Bach-
A
mann’s Lotschnittaxiom). In order to prove these postulates NLJ -equivalent to
the other postulates we presented in this paper we will introduce a new postulate,
which was implicitly assumed by Legendre in one of his attempts to prove Euclid’s
parallel postulate. Thus, we will refer to it as Legendre’s parallel postulate. Hav-
ing defined this postulate, we will formalize the proofs of Legendre’s Theorems.
Finally, we will present the formalization of Szmielew’s theorem, which opens the
28 The comment in French Wikipedia about Amiot’s proof seems to say that the proof is
valid only in Euclidean geometry because it use the construction of THE parallel to line AC
trough B. To be precise, the proof does not rely on the uniqueness of this line, only on its
existence, so this first step of the proof is valid also in hyperbolic geometry (but not in elliptic
geometry). The Wikipedia comment fails to notice that the proof relies on Postulate 7.
29 Here we use Greenberg’s denomination for models of Hilbert’s Axioms Group I, II, III and
IV [Gre10].
A4. P2.
A1.
A3. P1.
Fig. 49: Overview of the proof of the equivalence between Axiom 1, Axiom 3 and
Axiom 4.
path towards a mechanized procedure deciding the equivalence to Euclid’s parallel

postulate.
7.1 A proof of the independence of Archimedes’ axiom from the axioms of

Pythagorean planes
In this section, we will first establish the NLJ -equivalence between Axiom 1 (Decid-
ability of intersection of lines), Axiom 3 (Aristotle’s axiom) and Axiom 4 (Green-
berg’s axiom) under the assumption that Postulate 2 (Playfair’s postulate) holds.
From this equivalence, and using a syntactic proof of the independence of Axiom 1,
we obtain a proof for the independence of Archimedes’ axiom from the axioms of
Pythagorean planes, which is not based on counter-models. We do not prove in
Coq this independence property, because it relies on a proof based on Herbrand’s
theorem, that we have not formalized.
To demonstrate the equivalence between Axiom 1, Axiom 3 and Axiom 4 we
will show the implications that are represented on Fig. 49. With a view to sim-
plifying this overview, we use the equivalences proved in the previous section to
replace any postulate NLJ -equivalent to Postulate 1 by Postulate 1 and similarly
for Postulate 2. We already showed that Postulate 1 (Tarski’s parallel postulate)
is implied by the conjunction of Postulate 2 and Axiom 1 (Proposition 2) and
that Postulate 1 implies Axiom 1 (Proposition 3). In [Gre10], Greenberg proves
that Axiom 4 follows from Axiom 3, itself following from Postulate 1. Therefore,
it remains to show that Postulate 1 can be derived from the conjunction Axiom 4
and Postulate 2. To the best of our knowledge, this proof is new and will therefore
be detailed30 .
Let us first collect two lemmas from planar neutral geometry needed for it.
31 CP AP
Lemma 14 (Crossbar) If B A and B C then P ∈
b A B C.
CP AP
Lemma 15 Given two intersecting lines AB and CD and a point P not on line AB,
PQ
there exists a point Q on line CD such that A B.
PQ
30 We already presented this proof in French [GBN16].
31 This lemma is present in [Gre93] and [Har00] (Proposition 7.3) under the name of Crossbar
Theorem. Note that in [Har00], the statement look different but actually is the same, because
Hartshorne’s definition of a point being inside an angle is based on the two-side predicate,
whereas the definition we use (borrowed from [SST83]) states that the ray BP intersects the
segment AC.
Proposition 5 Assuming Axiom 4, Postulate 7 implies Postulate 13.
Proof
Given two parallel lines AB and Q

CD, P a point on line AB and Q a P
A A1
point not on line AB , we wish to B
prove that lines CD and P Q inter-
sect (Fig. 50). We first eliminate C Q1
the case of Col C D P , in which P C1
is the point of intersection. Then C0 Y
C2
D
we drop a perpendicular from P
to line CD, meeting line CD at Fig. 50: Assuming Axiom 4, Postulate 7 im-
the foot C0 , using Lemma 9. Then plies Postulate 13.
we can eliminate the case where
C0 lies on P Q, in which C0 is at the intersection between lines CD and P Q. From
C0 Q1
Lemma 15 we know that there exist a point Q1 on line P Q such that A B,
C0 Q1
as well as the points A1 and C1 respectively on lines AB and CD such that
Q1 A1 Q1 C1
P C0 and P C0 . We now have that Q1 ∈ b C0 P A1 thanks to Lemma 14.
Q1 A1 Q1 C1
Yet we know that ∠C0 P A1 is right by an application of Postulate 7, and thus
the angle ∠A1 P Q1 is acute. Using Axiom 4, we can construct C2 such that
C0 C2 C1 and P C2 C0 < b A1 P Q1 . By another application of Postulate 7, we know
C Q
that A1 P C2 = b P C2 C0 and thus C2 ∈ b A1 P Q1 . Then we can show that P 2 1 C0
C2 Q1
C0 Q1
and P C2 imply Q1 ∈
b C0 P C2 using Lemma 14. By definition it means that
C0 Q1
there exists a point Y such that C0 Y C2 and P Y Q1 . Therefore point Y is on
both lines CD and P Q. u t
Let us recall that Postulate 2 and Postulate 7 (the alternate interior angles
postulate) are NLJ -equivalent and that Postulate 1 and Postulate 13 (Proclus’
postulate) are also NLJ -equivalent. Proposition 5 lets us prove our claim.
Theorem 4 Axiom 1, Axiom 3 and Axiom 4 are NLJ -equivalent under the assumption
that Postulate 2 holds.
This theorem is quite peculiar because it asserts the equivalence between conti-
nuity axioms and a decidability property. Theorem 4 prove this equivalence under
the strong assumption that Postulate 2 holds.
Finally, since Axiom 1 is independent from the axioms of planar neutral ge-
ometry to which Postulate 2 is added, we get that both Axiom 3 and Axiom 4 are
also independent from these axioms. From the following proposition32 , we obtain
that Archimedes’ axiom is also independent from these axioms.
Proposition 6 Axiom 3 (Aristotle’s axiom) is implied by Axiom 2 (Archimedes’ ax-

iom).
32 Proposition 6 corresponds to Theorem 22.24 in [Mar98] which we mechanized. Its proof,
which depends on the Legendre’s first theorem (Theorem 6), will be discussed in the next
section.
7.2 Postulates equivalent to Bachmann’s Lotschnittaxiom
Unlike in the previous three sections, we will mostly just present the postulates
which are NLJ -equivalent to Bachmann’s Lotschnittaxiom (Postulate 4), together
with the necessary definitions. Indeed, the most interesting proof, in terms of
A
formalization, was the proof that these postulates are NLJ -equivalent with the
previously defined postulates. Therefore, we will focus mainly on the proof of
A
NLJ -equivalence which will be detailed in the next subsection.
Postulate 31 (Weak inverse projection postulate) and Postulate 32 (Weak
Tarski’s parallel postulate).
Definition weak_inverse_projection_postulate := forall A B C D E F P Q,

Acute A B C -> Per D E F -> SumA A B C A B C D E F ->
Out B A P -> P <> Q -> Per B P Q ->
exists Y, Out B C Y /\ Col P Q Y.
Definition weak_tarski_s_parallel_postulate := forall A B C T,

Per A B C -> InAngle T A B C ->
exists X Y, Out B A X /\ Out B C Y /\ Bet X T Y.
As suggested by the names of these postulates, they correspond to statements

similar to postulates that we studied in Section 5. More precisely, Postulate 31 and
Postulate 32 are respectively weaker version of Postulate 16 (the inverse projection
postulate) and Postulate 1 (Tarski’s parallel postulate).
Postulate 31 expresses that for any angle, that, together with itself, make a
right angle, any point together with its orthogonal projection on one side of the
angle form a line which intersects the other side. Compared to Postulate 16, this
postulate only adds the hypothesis that the considered angle together with itself
make a right angle.
Postulate 32 enounces that, for every right angle and every point T in the
interior of the angle, there is a point on each side of the angle such that T is between
these two points. The differences in comparison with Postulate 1 are of two kinds.
The first one is analogous to the difference between Postulate 31 and Postulate 16.
Precisely, the statement is restricted to a certain kind of angle, namely the right
angles. The second one resembles the difference between Postulate 17 (Euclid
5) and Postulate 18 (the strong parallel postulate). Indeed, Postulate 32 is less
precise than Postulate 1 regarding the position of the points in the hypotheses,
which results in a weaker conclusion.
Before introducing the last postulate we need to introduce a definition asserting
that some points belong to the perpendicular bisector of a segment.
Definition ReflectL P’ P A B :=
(exists X, Midpoint X P P’ /\ Col A B X) /\ (Perp A B P P’ \/ P = P’).
Definition Perp_bisect P Q A B := ReflectL A B P Q /\ A <> B.
P 0 •P
This predicate corresponds to Definition 10.3 of [SST83]. A B means that
P 0 •P
0
P is the image of P by the reflection with respect to the line AB . It is interesting
to see that for ReflectL P’ P A B to be true when A and B are equal, one must
A •B
have that P 0 and P are also equal. To define P Q, which expresses that the
A •B
31.
32. 4.
33.
Fig. 51: Overview of the proofs of Sec. 7.2.
line P Q is the perpendicular bisector of the segment AB , we can therefore just

exclude the case where A and B are equal and state that B is the image of A by
the reflection with respect to the line P Q.
Postulate 33 (Weak triangle circumscription principle).
Definition weak_triangle_circumscription_principle := forall A B C A1 A2 B1 B2,

~ Col A B C -> Per A C B ->
Perp_bisect A1 A2 B C -> Perp_bisect B1 B2 A C ->
exists I, Col A1 A2 I /\ Col B1 B2 I.
As for the previous postulates, Postulate 33 present important similarities with

another postulate, namely Postulate 15 (Triangle circumscription principle), al-
though the differences are more substantial. It asserts that the perpendicular bisec-
tors of the legs of a right triangle intersect. Postulate 33 is not only the restriction
of Postulate 15 to the case of right triangles. In fact, Postulate 15 enouces that
for any three non-collinear points there exists a point equidistant from them. As
our axioms allow to prove that all the points are coplanar, being equidistant from
two points is equivalent to belonging to the perpendicular bisector of the segment
defined by these two points. However we chose to formalize this postulate using
the notion of perpendicular bisector to have a definition which is more faithful to
its statement in [Mar98].
Fig. 51 provides a graphical summary of the implications we mechanized to
prove Theorem 5. Most of these proofs correspond to proofs found in the literature.
In [Mar98], there is the proof of the implication from Postulate 33 to Postulate 4
(Theorem 23.7) and in [Bac64], the equivalence between Postulate 4, Postulate 31
and Postulate 32 is proved.
The implications displayed on Fig. 51 allow us to prove the following theorem.
Theorem 5 Postulates 4, 31-33 are NLJ -equivalent.
Following is the Coq statement corresponding to Theorem 5.
Theorem equivalent_postulates_without_any_continuity_bis :
all_equiv
(bachmann_s_lotschnittaxiom::
weak_inverse_projection_postulate::
weak_tarski_s_parallel_postulate::
weak_triangle_circumscription_principle::
nil).
7.3 Legendre’s Theorems
This section is dedicated to one of Legendre’s flawed proofs of Euclid’s parallel

postulate. In this proof he implicitly assumed Legendre’s parallel postulate, which
we will introduce. Following [Rot14], we will split this proof into four Legendre’s
Theorems. There are two statements commonly known as Legendre’s theorems, but
Rothe mentions four such theorems in [Rot14]. What he refers as Legendre’s third
theorem and Legendre’s fourth theorem correspond to the remaining parts of this
flawed proof. Throughout this section, we will refer to this proof which we mech-
anized, while emphasizing on the formalization details.
Let us first present Legendre’s parallel postulate.
Postulate 34 (Legendre’s parallel postulate).
Definition legendre_s_parallel_postulate :=
exists A B C,
~ Col A B C /\ Acute A B C /\
forall T,
InAngle T A B C ->
exists X Y, Out B A X /\ Out B C Y /\ Bet X T Y.
This posulate formulates that there exists an acute angle such that, for every
point T in the interior of the angle, there is a point on each side of the angle such
that T is between these two points. Postulate 34 is pretty similar to Postulate 1
(Tarski’s parallel postulate). In fact, Postulate 34 mainly differs from Postulate 1
in two aspects. The first difference comes from the way in which the points A,
B and C defining the considered angle are quantified (Fig. 13). In this version of
the parallel postulate they are existentially quantified33 . The second difference is
about the relative position of the points, which is more precise in Postulate 1. Here,
the same situation as in Postulate 32 (Weak Tarski’s parallel postulate) occurs:
the point through which goes the line asserted to exist is not required to be further
away from B than the segment AC , which results in a weaker conclusion, namely
that the line intersects the sides of the angles without any precision with regards
to the position of the intersections relatively to A and C .
Let us then consider Legendre’s first theorem, which can be stated in the follow-
ing way.
Theorem 6 (Legendre’s first theorem) In Archimedean neutral geometry, the an-

gles of every triangle make less than or equal to two right angles.
Theorem 6 is now known as Saccheri-Legendre theorem, as Saccheri proved

this statement almost a century before Legendre. In order to formalize the proof
of this theorem as exposed in [Mar98], we introduced a variant of the predicate
33 One could also notice that they are also specified to form a non-degenerated acute angle.
The fact it is acute plays a minor role, contrary to the non-degeneracy condition, because if one
can find such an obtuse or right angle, every acute angle inside it fulfills the same properties.
B0 B1 B2 B3 Bn−2 Bn−1 Bn
A0 A1 A2 A3 An−2 An−1 An
Fig. 52: Considered points in Theorem 22.18 of [Mar98].
Grad. Indeed, the theorem that is central for this proof, namely Theorem 22.18,
constructs pairs of points that, given two segments, correspond to the endpoints
of segments constructed by extending the given segments by their own lengths
the same number of times (the Ai and Bi for 1 ≤ i ≤ n on Fig. 5234 ). As our
formalization of Archimedes’ axiom does not use the concept of natural number,
we had to express that the segments are extended the same number of times, using
the following inductive predicate. Grad2 A B C D E F intuitively means that there
exists n such that AC ≡ nAB and DF ≡ nDE .
Inductive Grad2 : Tpoint -> Tpoint -> Tpoint -> Tpoint -> Tpoint -> Tpoint ->
Prop :=
| grad2_init : forall A B D E, Grad2 A B B D E E
| grad2_stab : forall A B C C’ D E F F’,
Grad2 A B C D E F ->
Bet A C C’ -> Cong A B C C’ ->
Bet D F F’ -> Cong D E F F’ ->
Grad2 A B C’ D E F’.
As often in induction proofs, the difficulty lied in finding the appropriate induc-
tive hypothesis. Moreover, the same difficulties as the ones presented in Section 6
arose. Having mechanized Theorem 22.18 of [Mar98], we could demonstrate The-
orem 6. The proposition that we showed is the following.
Theorem legendre_s_first_theorem :
archimedes_axiom ->
forall A B C D E F,
~ Col A B C ->
SumA C A B A B C D E F ->
Isi D E F B C A.
We should remark that the hypotheses of this theorem are not minimal. Indeed,
Greenberg provides a proof only relying on Axiom 4 (Greenberg’s axiom) [Gre93]
that we did not formalized yet.
The next theorem has already been proven in Section 6. It asserts that Postu-
late 21 (the postulate of existence of a triangle whose angles sum to two rights)
implies Postulate 3 (the triangle postulate) and can be stated in the following way.
34 The B are not known to be collinear, but the fact B B
i i i+1 ≡ B0 B1 and the quantity
nB0 B1 appear in this proof.
Theorem 7 (Legendre’s second theorem) In planar neutral geometry, if the angles

of one triangle sum to two right angles, then the angles of all triangles sum to two right
angles.
Since Theorem 7 is a corollary of Theorem 3, we will just give the Coq statement
corresponding to it.
Theorem legendre_s_second_theorem :
postulate_of_existence_of_a_triangle_whose_angles_sum_to_two_rights ->
triangle_postulate.
Legendre’s next theorem expresses that, assuming Axiom 2 (Archimedes’ ax-

iom), Postulate 20 (Euclid’s parallel postulate) is a consequence of Postulate 3. It
can be formulated in the following manner.
Theorem 8 (Legendre’s third theorem) In Archimedean neutral geometry, if the

angles of every triangle sum to two right angles, then Euclid’s parallel postulate holds.
As for Theorem 6, the hypotheses of Theorem 8 can be weakened: from The-

orem 4, we know that Axiom 4 (Greenberg’s axiom) suffices to derive the impli-
cation from Postulate 3 to Postulate 20. Hence, in order to obtain this theorem,
we chose to formalize the proof that Axiom 2 implies Axiom 3 (Aristotle’s axiom)
(Proposition 6). Let us recall Greenberg’s definition of Axiom 3.
“Given any acute angle, any side of that angle, and any challenge segment
P Q, there exists a point Y on the given side of the angle such that if X is
the foot of the perpendicular from Y to the other side of the angle, then
Y X > P Q.”
P
Y
P
Q
Y QC
C Y0
Y0 A X
B X0
B X0 A X
Fig. 53: Axiom 2 implies Axiom 3.
Let Y0 be a point on the given side of the angle and X0 be the foot of the
perpendicular from Y0 to the other side of the angle. Then, by letting n0 be a
positive integer such that n0 Y0 X0 > P Q and B be the vertex of the angle, one
could assume that Axiom 2 would let us conclude by constructing Y such that
n0 Y0 B ≡ Y B and dropping a perpendicular from Y on the other side of the angle.
Nevertheless, following [Mar98], it is much simpler to choose a positive n such
2n Y0 X0 > P Q and prove that a point Y such that n2n Y0 B ≡ Y B would suffice35 .
Therefore, we defined the following exponential variant of the predicate Grad.
Inductive GradExp : Tpoint -> Tpoint -> Tpoint -> Prop :=

| gradexp_init : forall A B, GradExp A B B
| gradexp_stab : forall A B C C’,
GradExp A B C ->
Bet A C C’ -> Cong A C C C’ ->
GradExp A B C’.
GradExp A B C intuitively asserts that there exists n such that 2n AB ≡ AC . Yet

the positive integer n is not explicit in our definition, it is hidden in the number
of times the constructor gradexp_stab is used. We then proved the equivalence
between being reachable using Grad or GradExp to complete the proof of Theorem 8.
We now provide its Coq statement.
Theorem legendre_s_third_theorem :
archimedes_postulate ->
triangle_postulate ->
euclid_s_parallel_postulate.
Finally, the last theorem will complete Legendre’s flawed proof. It states that,
assuming Axiom 2, Postulate 21 (the postulate of existence of a triangle whose
angles sum to two rights) is a consequence of Postulate 34 (Legendre’s parallel
postulate). It can be phrased as follows.
Theorem 9 (Legendre’s fourth theorem) In Archimedean neutral geometry, if Leg-

endre’s postulate holds, then there exists a triangle for which the angles sum to two right
angles.
Theorem legendre_s_fourth_theorem :
legendre_s_postulate ->
postulate_of_existence_of_a_triangle_whose_angles_sum_to_two_rights.
To demonstrate Theorem 9, we mechanized the proof given in [Mar98]. In this

proof, a concept that we have not encountered so far plays a major role: the defect
of a triangle. For a given triangle, its defect together with the sum of the angles
of this triangle make two right angles. Having the concept of sum of angles, it is
straightforward to define this concept in Coq.
Definition Defect A B C D E F := exists G H I J K L,

TriSumA A B C G H I /\ Bet J K L /\ SumA G H I D E F J K L.
Here, D(4ABC ) = b D E F expresses that ∠DEF is the defect of the triangle

ABC . This proof eliminates the hypothesis of acute angle by reproducing a con-
struction. Given the acute angle ∠ABC asserted to exist by Postulate 34 and a
pair of point Ak and Ck respectively on the sides BA and BC , this construction
creates the points Ak+1 and Ck+1 , respectively on the sides BA and BC , such
35 One should also notice that this proof relies on Theorem 6, although it may not be obvious.
that the defect of the triangle Ak+1 BCk+1 is at least the double of the defect of
the triangle Ak BCk . To conclude its proof, Legendre uses Archimedes’ axiom to
deduce that repeating this construction will lead to a triangle An BCn which has a
defect greater than two right angles. This last step need to be detailed in Coq, as it
makes the implicit assumption that Axiom 2 implies the Archimedean property for
angles. More precisely, the implicit assumption is that every non-degenerate angle
can be doubled enough times to obtain an obtuse angle. Therefore, we defined the
following variant of the predicate Grad.
Inductive GradAExp : Tpoint -> Tpoint -> Tpoint -> Tpoint -> Tpoint -> Tpoint ->
Prop :=
| gradaexp_init : forall A B C D E F, CongA A B C D E F -> GradAExp A B C D E F
| gradaexp_stab : forall A B C D E F G H I,
GradAExp A B C D E F ->
Isi D E F D E F -> SumA D E F D E F G H I ->
GradAExp A B C G H I.
Hartshorne (Lemma 35.1 in [Har00]) provides an explicit proof that we for-

malized. We could then prove Theorem 9. Again, the difficulty lied in finding the
appropriate induction hypotheses and justifying the position of the points on the
figure. By mechanizing the proof that Postulate 34 can be derived from Postu-
late 4, which is part of Theorem 23.7 of [Mar98], we obtain the following theorem
(Fig. 54 provides a summary of the implications needed for its proof).
Theorem 10 In Archimedean neutral geometry, Postulates 1-34 are equivalent.
31. 34. 28. 10. 18.

30. 27. 11. 9. 15. 13.
29. 21.
12. 2. 1. 14.
32. 4.
26. 3.
7. 5. 17. 16.
25. 22.
33. 24. 23. 8. 6. 20. 19.
Fig. 54: Overview of the proofs.
7.4 Towards a mechanized procedure deciding the equivalence to Euclid’s parallel

postulate
Another interesting consequence of Archimedes’ axiom is a very useful result con-

cerning the parallel postulate, that was established by Szmielew in [Szm59]. It
states that “Euclid’s axiom can be replaced in the axiom system of En by any
sentence whatsoever which is valid in En but not in Hn ”. Here En denotes Tarski’s
system of geometry, where A8 and A9 are replaced by the lower n-dimensional
axiom and the upper n-dimensional axiom, and Hn corresponds to En where Pos-
tulate 1 (Tarski’s parallel postulate) A10 is replaced by its negation. Hence this
result allows us to prove the equivalence of some statements with Tarski’s parallel
postulate by checking if it holds in Euclidean geometry and does not in hyper-
bolic geometry. Moreover, because both of these theories are decidable, this gives
a procedure deciding if a statement is equivalent to the parallel postulate. In this
section, we will formalize this result and we will then discuss the possibility for
the mechanization of such a procedure.
In Section 7.3, we formalized the fact that Postulate 1 is equivalent to Postu-
late 2 when assuming Archimedes’ axiom. With a view to proving this result, we
chose to use Postulate 2 in place of Postulate 1. We would like to stress that, while
Szmielew obtained her result through metamathematical properties, we present
here a synthetic proof, so the choice of the parallel postulate matters. This choice
was motivated by the fact that the negation of Postulate 2 was easier to work
with. However, her proof is valid in spaces of dimension higher than two, whereas
our proof is only valid in planar geometry. Another difference is that her proof
assumes the axiom of continuity, while our proof only assumes Archimedes’ axiom
(Axiom 2), which is strictly weaker than the axiom of continuity. Indeed, Cantor’s
axiom is independent from the axiom of planar Archimedean geometry [MB95],
and the axiom of continuity is equivalent to the conjunction of both of these ax-
ioms [Rot14]. We defined the negation of Postulate 2 in the following way.
Definition hyperbolic_plane_postulate := forall A1 A2 P,

~ Col A1 A2 P -> exists B1 B2 C1 C2,
Par A1 A2 B1 B2 /\ Col P B1 B2 /\
Par A1 A2 C1 C2 /\ Col P C1 C2 /\
~ Col C1 B1 B2.
We can point out that this definition (we will now refer to it as hyperbolic
plane postulate) is in fact the negation of a modification of Postulate 2. This
modification expresses the existence of a line and a point not on this line such that
there is a unique line parallel to this line passing by this point. This modification
was showed to be equivalent with Postulate 2 when assuming Axiom 4 for one
of the implications. Because of this we could not classify this modified version of
Postulate 2. This explains why we did not present it in Section 4. We now collect
four lemmas which will be used in the lemma at the core of this proof.
Lemma 16 Given a right triangle ABD with the right angle at vertex A, a point C
is construtible such that ABCD is a Saccheri quadrilateral.
Lemma 17 In a Saccheri quadrilateral ABCD, the sides AD and BC are parallel.
Lemma 18 In a Saccheri quadrilateral ABCD, the sides AB and CD are strictly

parallel.
Lemma 19 (12.6) Two points on a line strictly parallel to another one are on the
same side of this last line.
Proposition 7 The hyperbolic plane postulate holds under the hypothesis of the acute
angle.
Proof
Given three non-collinear points

B1 P C1
A1 , A2 and P we wish to prove
that there exists two distinct lines
B1 B2 and C1 C2 both parallel to
line A1 A2 and passing through P
(Fig. 55). Lemma 9 lets us drop
the perpendicular from P to line X A1 Q Y A2
A1 A2 in Q. A tedious distinction
of cases would then allow us to
Fig. 55: Postulate 7 implies Postulate 3.
prove that there exists a point X
on line A1 A2 distinct from A1 , A2 and Q. Lemma 2 lets us construct Y , the sym-
metric point of X with respect to Q. Using Lemma 16, we construct B1 and C1
such that QP B1 X and S Q P C1 Y . We will now prove that both lines B1 P and
C1 P are both parallel to line A1 A2 and pass by P , and that they are distinct.
The facts that these lines are parallel to line A1 A2 and pass by P are respectively
due to Lemma 17 and trivial. We proceed by proof of negation to prove that
these lines are distinct. So let us assume that they are equal. We can prove that
B1 C1 XY B1 X C1 Y
P Q: indeed, we have P Q by definition, as well as P Q and P Q
B1 C1 XY B1 X C1 Y
from Lemma 19, hence Lemma 10 applied twice lets us show our claim. We shall
now derive a contradiction by proving that B1 P C1 < b B1 P C1 . Since angles ∠XQP
and ∠P QY do not make an over-obtuse angle, making exactly two right angles, it
suffices to show that B1 P Q < b P Q X , Q P C1 <b P Q X , B1 P Q +b Q P C1 =
b B1 P C1
and P Q X + b P QX = b B1 P C1 . We have B1 P Q < b P Q X and Q P C1 < b P Q X from
the hypothesis of the acute angle and B1 P Q + b Q P C1 =
b B1 P C1 by definition.
Thus if we can show that P Q X + P Q X =
b b B1 P C1 we will be done. We trivially
have X Q P = b P Q X and P Q Y = b P Q X . Since we have X Q P + b P QY =b X Q Y by
definition, it suffices to prove X Q Y =
b B1 P C1 , which holds since straight angles
are congruent. u t
Let us state Szmielew’s theorem. We will not detail its proof as it is a tautology
G
knowing the Legendre’s first theorem, the previous lemma and the NLJ -equivalence
between Playfair’s postulate and the hypothesis of right angle. Following are the
informal statement as well as its formulation in Coq. Note that it is the only
theorem we state in this paper that is expressed using second-order logic. In gen-
eral, second-order logic is rarely used in our formalization of geometry: it is used
only in intermediate definitions and statements needed for the proof of Pappus’
theorem [BN17].
Theorem 11 (Szmielew’s theorem) Assuming Archimedes’ axiom, any proposition

implied by Playfair’s postulate and such that its negation is implied by the hyperbolic
plane postulate is equivalent to Playfair’s postulate.
Theorem szmielew_s_theorem :
(forall P : Prop,
(playfair_s_postulate -> P) ->

(hyperbolic_plane_postulate -> ~ P)->
(P <-> playfair_s_postulate)).
A
Even if Theorem 11 only allows us to prove NLJ -equivalence, it is a very power-
ful tool. Indeed, for any statement presenting only universal quantifiers, it suffices
to show it is a consequence of any of our 34 versions of the parallel postulate and
A
to provide a counterexample in hyperbolic geometry to prove its NLJ -equivalence
with Playfair’s postulate. Moreover, both En and Hn are decidable [Szm59]. There-
fore Theorem 11 render possible a mechanized procedure deciding if a statement
is equivalent to Playfair’s postulate, using the decidability of both theories. Ac-
tually, the quantifier elimination algorithm for real closed fields which has been
formalized by Cohen and Mahboubi in [CM12] can be connected to Tarski’s sys-
tem of geometry, thanks to our formalization of the arithmetization of Euclidean
geometry [BBN16a]. Following [Szm61] we could formalize the arithmetization of
hyperbolic geometry and extend the same quantifier elimination algorithm to this
system. However, we are not sure whether this method would work in practice,
because the quantifier elimination algorithm for real closed field by Cohen and
Mahboubi has not been designed to be efficient, but to provide a theoretical re-
sult. Furthermore, the Gröbner basis method has already been integrated into Coq
by Grégoire, Pottier and Théry [GPT11]. Our work on the arithmetization of Eu-
clidean geometry lets us use this method in our axiomatic system. Thus proving
that a statement presenting only universal quantifiers is a consequence of Play-
fair’s postulate could, in some case, be done by computation, although this would
require a significant formalization work.
8 Conclusion
We have described the formalization within the Coq proof assistant of the proof
that 34 versions of the parallel postulate are equivalent. The originality of our
proofs relies on the fact that first, the equivalence between these different versions
are proved in Tarski’s neutral geometry without using the continuity axiom nor
line-circle continuity, and second, we work in an intuitionistic logic. Assuming
decidability of point equality, we clarified the role of the decidability of intersection.
We obtained the formal proof that assuming decidability of point equality, some
versions of the parallel postulate imply decidability of the intersection of lines.
The formal proofs of equivalences consist of about 7k lines of code and rely on 44k
lines of code corresponding to a library for planar neutral geometry. The proofs
make heavy use of the tactics developed previously [BNSB14b]. The use of a proof
assistant was crucial to check these proofs. Indeed, it is extremely easy to make
a mistake in a pen-and-paper proof in this context. We have to be careful not to
use any of the many statements which are equivalent to the parallel postulate, as
well as not use classical reasoning.
This work can have several extensions. First, we can take advantage of our
large library in neutral geometry, and of the proof of the equivalence between the
different versions of the postulate, to obtain a library in hyperbolic geometry.
Second, we could weaken our axiom systems, to study the more constructive
version as defined by Beeson in [Bee16] where he assumes stability of equality,
congruence and betweenness instead of excluded-middle for equality as we do, or

to use even more constructive assumptions as in [Bee17].
Third, it could be extended to some other versions of the parallel postulates,
as the one recently proposed by Pambuccian [Pam17].
Finally, it would be also interesting to complete this study by the formalization
of the different independence properties that we mentioned.
Acknowledgements
We thank Victor Pambuccian for pointing us to the relevant literature. We are

grateful to David Braun for proof reading a draft of this paper.
Availability
The full Coq development is available here: http://geocoq.github.io/GeoCoq/
References
ADM09. Jeremy Avigad, Edward Dean, and John Mumma. A formal system for euclid’s
elements. The Review of Symbolic Logic, 2:700–768, 12 2009.
Ami70. A Amiot. Eléments de géométrie [texte imprimé]: rédigés d’après le nouveau pro-
gramme de l’enseignement scientifique des lycées; suivis d’un complément à l’usage
des élèves de mathématiques spéciales/par a. amiot,... 1870.
Ami33. D. Amira. Sur l’axiome de droites parallèles. L’Enseignement Mathématique,
32:52–57, 1933.
AP16. Jesse Alama and Victor Pambuccian. From absolute to affine geometry in terms
of point-reflections, midpoints, and collinearity. Note di Matematica, 36(1):11–24,
2016.
Bac64. F Bachmann. Zur parallelenfrage. In Abhandlungen aus dem Mathematischen
Seminar der Universität Hamburg, volume 27, pages 173–192. Springer, 1964.
Bac73. F. Bachmann. Aufbau der Geometrie aus dem Spiegelungsbegriff. Grundlehren der
mathematischen Wissenschaften. Springer, 1973.
BBN15. Michael Beeson, Pierre Boutry, and Julien Narboux. Herbrand’s theorem and
non-Euclidean geometry. Bulletin of Symbolic Logic, 21(1):111–122, 2015.
BBN16a. Pierre Boutry, Gabriel Braun, and Julien Narboux. From Tarski to Descartes:
Formalization of the Arithmetization of Euclidean Geometry. In The 7th Interna-
tional Symposium on Symbolic Computation in Software (SCSS 2016), EasyChair
Proceedings in Computing, page 15, Tokyo, Japan, March 2016.
BBN16b. Gabriel Braun, Pierre Boutry, and Julien Narboux. From Hilbert to Tarski. In
Eleventh International Workshop on Automated Deduction in Geometry, Proceed-
ings of ADG 2016, page 19, Strasbourg, France, June 2016.
Bee15. Michael Beeson. A constructive version of Tarski’s geometry. Annals of Pure and
Applied Logic, 166(11):1199–1273, 2015.
Bee16. Michael Beeson. Constructive geometry and the parallel postulate. Bulletin of
Symbolic Logic, 22(1):1–104, 2016.
Bee17. Michael Beeson. Brouwer and Euclid. Indagationes Mathematicae, to appear in a
special issue devoted to Brouwer, 2017.
Bel93. John L Bell. Hilbert’s -operator in intuitionistic type theories. Mathematical
Logic Quarterly, 39(1):323–337, 1993.
Bir32. George David Birkhoff. A set of postulates for plane geometry (based on scale and
protractors). Annals of Mathematics, 33, 1932.
BM15. David Braun and Nicolas Magaud. Des preuves formelles en Coq du théorème
de Thalès pour les cercles. In David Baelde and Jade Alglave, editors, Vingt-
sixièmes Journées Francophones des Langages Applicatifs (JFLA 2015), Le Val
d’Ajol, France, January 2015.
BN12. Gabriel Braun and Julien Narboux. From Tarski to Hilbert. In Tetsuo Ida and
Jacques Fleuriot, editors, Post-proceedings of Automated Deduction in Geometry
2012, volume 7993 of LNCS, pages 89–109, Edinburgh, United Kingdom, Septem-
ber 2012. Jacques Fleuriot, Springer.
BN17. Gabriel Braun and Julien Narboux. A synthetic proof of Pappus’ theorem in
Tarski’s geometry. Journal of Automated Reasoning, 58(2):23, February 2017.
BNSB14a. Pierre Boutry, Julien Narboux, Pascal Schreck, and Gabriel Braun. A short note
about case distinctions in Tarski’s geometry. In Francisco Botana and Pedro
Quaresma, editors, Proceedings of the 10th Int. Workshop on Automated Deduc-
tion in Geometry, volume TR 2014/01 of Proceedings of ADG 2014, pages 51–65,
Coimbra, Portugal, July 2014. University of Coimbra.
BNSB14b. Pierre Boutry, Julien Narboux, Pascal Schreck, and Gabriel Braun. Using small
scale automation to improve both accessibility and readability of formal proofs in
geometry. In Francisco Botana and Pedro Quaresma, editors, Proceedings of the
10th Int. Workshop on Automated Deduction in Geometry, volume TR 2014/01 of
Proceedings of ADG 2014, pages 31–49, Coimbra, Portugal, July 2014. University
of Coimbra.
Bon55. Roberto Bonola. Non-Euclidean geometry: A critical and historical study of its
development. Courier Corporation, 1955.
BS60. Karol Borsuk and Wanda Szmielew. Foundations of geometry. North-Holland,
1960.
Caj98. Florian Cajori. A history of elementary mathematics. Macmillan, 1898.
CM12. Cyril Cohen and Assia Mahboubi. Formal proofs in real algebraic geometry: from
ordered fields to quantifier elimination. Logical Methods in Computer Science,
8(1:02):1–40, February 2012.
Cox98. Harold Scott Macdonald Coxeter. Non-euclidean geometry. Cambridge University
Press, 1998.
DDS01. Christophe Dehlinger, Jean-François Dufourd, and Pascal Schreck. Higher-Order
Intuitionistic Formalization and Proofs in Hilbert’s Elementary Geometry. In Au-
tomated Deduction in Geometry, volume 2061 of Lectures Notes in Computer Sci-
ence, pages 306–324, 2001.
Deh00. M. Dehn. Die Legendre’schen Sätze über die Winkelsumme im Dreieck. Mathe-
matische Annalen, 53(3):404–439, 1900.
Dup10. Jean Duprat. Fondements de géométrie euclidienne. https://hal.inria.fr/
hal-00661537, 2010.
EHD02. Euclid, T.L. Heath, and D. Densmore. Euclid’s Elements: all thirteen books com-
plete in one volume : the Thomas L. Heath translation. Green Lion Press, 2002.
GB99. Carl Friedrich Gauss and Farkas Bolyai. Briefwechsel zwischen Carl Friedrich
Gauss und Wolfgang Bolyai. BG Teubner, 1899.
GBN16. Charly Gries, Pierre Boutry, and Julien Narboux. Somme des angles d’un triangle
et unicité de la parallèle : une preuve d’équivalence formalisée en Coq. In Les vingt-
septièmes Journées Francophones des Langages Applicatifs (JFLA 2016), Actes
des Vingt-septièmes Journées Francophones des Langages Applicatifs (JFLA 2016),
page 15, Saint Malo, France, January 2016. Jade Algave and Julien Signoles.
GPT11. Benjamin Grégoire, Loı̈c Pottier, and Laurent Théry. Proof Certificates for Al-
gebra and their Application to Automatic Geometry Theorem Proving. In Post-
proceedings of Automated Deduction in Geometry (ADG 2008), number 6701 in
Lecture Notes in Artificial Intelligence, 2011.
Gre88. Marvin Jay Greenberg. Aristotle’s axiom in the foundations of geometry. Journal
of Geometry, 33(1):53–57, 1988.
Gre93. Marvin J. Greenberg. Euclidean and Non-Euclidean Geometries - Development
and History. Macmillan, 1993.
Gre10. Marvin J. Greenberg. Old and New Results in the Foundations of Elementary Plane
Euclidean and Non-Euclidean Geometries. The American Mathematical Monthly,
117(3):198–219, March 2010.
Gup65. Haragauri Narayan Gupta. Contributions to the axiomatic foundations of geome-
try. PhD thesis, University of California, Berkley, 1965.
Har00. Robin Hartshorne. Geometry : Euclid and beyond. Undergraduate texts in math-
ematics. Springer, 2000.
Hil60. David Hilbert. Foundations of Geometry (Grundlagen der Geometrie). Open
Court, La Salle, Illinois, 1960. Second English edition, translated from the tenth
German edition by Leo Unger. Original publication date, 1899.
Kli90. Morris Kline. Mathematical Thought From Ancient to Modern Times: Volume 3,
volume 3. OUP USA, 1990.
Klu63. G.S. Klugel. Conatuum praecipuorum theoriam parallelarum demonstrandi recen-
sio. PhD thesis, Schultz, Göttingen, 1763. German translation available:
http://www2.math.uni-wuppertal.de/~volkert/versuch.html.
Leg33. Adrien Marie Legendre. Réflexions sur différentes manières de démontrer la théorie
des parallèles ou le théorème sur la somme des trois angles du triangle. Mémoires
de l’Académie royale des sciences de l’Institut de France, XII:367–410, 1833.
Lew20. Florence P Lewis. History of the parallel postulate. The American Mathematical
Monthly, 27(1):16–23, 1920.
LP13. Reinhard Laubenbacher and David Pengelley. Mathematical expeditions: Chroni-
cles by the explorers. Springer Science & Business Media, 2013.
Mak12. Timothy James McKenzie Makarios. A Mechanical Verification of the Indepen-
dence of Tarski’s Euclidean Axiom. Master’s thesis, Victoria University of Welling-
ton, 2012. Master Thesis.
Mar98. G.E. Martin. The Foundations of Geometry and the Non-Euclidean Plane. Un-
dergraduate Texts in Mathematics. Springer, 1998.
MB95. Radu Miron and Dan Brânzei. Backgrounds of arithmetic and geometry: an in-
troduction, volume 23. World Scientific, 1995.
MP81. Richard S. Millman and George D. Parker. Geometry, A Metric Approach with
Models . Springer-Verlag, 1981.
MP14. Filip Marić and Danijela Petrović. Formalizing complex plane geometry. Annals
of Mathematics and Artificial Intelligence, 74:271–308, 2014.
Nar07. Julien Narboux. Mechanical Theorem Proving in Tarski’s geometry. In Fran-
cisco Botana Eugenio Roanes Lozano, editor, Post-proceedings of Automated De-
duction in Geometry 2006, volume 4869 of LNCS, pages 139–156, Pontevedra,
Spain, 2007. Francisco Botana, Springer.
Pam94. Victor Pambuccian. Zum stufenaufbau des parallelenaxioms. Journal of Geometry,
51(1-2):79–88, 1994.
Pam06. Victor Pambuccian. Axiomatizations of hyperbolic and absolute geometries. In
Non-Euclidean Geometries, volume 581, pages 119–153. Springer, 2006.
Pam09. Victor Pambuccian. On the equivalence of lagrange’s axiom to the lotschnittaxiom.
Journal of Geometry, 95(1-2):165–171, 2009.
Pam17. Victor Pambuccian. Another equivalent of the lotschnittaxiom. Beiträge zur Al-
gebra und Geometrie / Contributions to Algebra and Geometry, 58(1):167–170,
2017.
Pap12. Athanase Papadopoulos. Master Class in Geometry - Chapter 1 - Notes on non-
Euclidean geometry. European Mathematical Society, 2012.
Pas76. Moritz Pasch. Vorlesungen über neuere Geometrie. Springer, 1976.
Pej61. W. Pejas. Die modelle des hilbertschen axiomensystems der absoluten geometrie.
Mathematische Annalen, 143(3):212–235, 1961.
Pie61. Piesyk. The existential and universal statements on parallels. Bull. Acad. Polon.
Sci. Sér. Sci. Math. Astronom. Phys., 9:761–764, 1961.
Rot14. Franz Rothe. Several Topics from Geometry. unpublished, 2014. http://math2.
uncc.edu/~frothe/3181all.pdf.
SST83. Wolfram Schwabhäuser, Wanda Szmielew, and Alfred Tarski. Metamathematische
Methoden in der Geometrie. Springer-Verlag, Berlin, 1983.
Szm59. Wanda Szmielew. Some metamathematical problems concerning elementary hyper-
bolic geometry. Studies in Logic and the Foundations of Mathematics, 27:30–52,
1959.
Szm61. Wanda Szmielew. A new analytic approach to hyperbolic geometry. Fundamenta
Mathematicae, 50(2):129–158, 1961.
Tar51. Alfred Tarski. A decision method for elementary algebra and geometry. University
of California Press, 1951.
TG99. Alfred Tarski and Steven Givant. Tarski’s system of geometry. The bulletin of
Symbolic Logic, 5(2), June 1999.
Tru86. Richard J. Trudeau. The Non-Euclidean Revolution. Springer, 1986.
A Summary of the 34 postulates
1. (Tarski’s parallel postulate) Given a point D between the points B and C and a point T
further away from A than D on the half line AD, one can build a line which goes through
T and intersects the sides BA and BC of the angle ∠ABC respectively further away from
B than A and C.
2. (Playfair’s postulate) There is a unique parallel to a given line going through some point.
3. (Triangle postulate) The sum of the angles of any triangle is two right angles.
4. (Bachmann’s Lotschnittaxiom) Given the lines l, m, r and s, if l and r are perpendicular,
r and s are perpendicular and s and m are perpendicular, then l and m must meet.
5. (Postulate of transitivity of parallelism) If two lines are parallel to the same line then these
lines are also parallel.
6. (Midpoint converse postulate) The parallel line to one side of a triangle going through the
midpoint of another side cut the third side in its midpoint.
7. (Alternate interior angles postulate) The line falling on parallel lines makes the alternate
angles equal one to one another.
8. (Consecutive interior angles postulate) A line falling on parallel lines makes the sum of
interior angles on the same side equal to two right angles.
9. (Perpendicular transversal postulate) Given two parallel lines, any line perpendicular to
the first line is perpendicular to the second line.
10. (Postulate of parallelism of perpendicular transversals) Two lines, each perpendicular to
one of a pair of parallel lines, are parallel.
11. (Universal Posidonius’ postulate) If two lines are parallel then they are everywhere equidis-
tant.
12. (Alternative Playfair’s postulate) Any line parallel to line l passing through a point P is
equal to the line that passes through P and shares a common perpendicular with l going
through P .
13. (Proclus’ postulate) If a line intersects one of two parallel lines then it intersects the other.
14. (Alternative Proclus’ postulate) If a line intersects in P one of two parallel lines which
share a common perpendicular going through P , then it intersects the other.
15. (Triangle circumscription principle) For any three non-collinear points there exists a point
equidistant from them.
16. (Inverse projection postulate) For any given acute angle, any point together with its or-
thogonal projection on one side of the angle form a line which intersects the other side.
17. (Euclid 5) Given a non-degenerated parallelogram P RQS and a point U strictly inside the
angle ∠QP R, there exists a point I such that Q and U are respectively strictly between S
and I and strictly between P and I.
18. (Strong parallel postulate) Given a non-degenerated parallelogram P RQS and a point U
not on line P R, the lines P U and QS intersect.
19. (Alternative strong parallel postulate) If a straight line falling on two straight lines make
the sum of the interior angles on the same side different from two right angles, the two
straight lines meet if produced indefinitely.
20. (Euclid’s parallel postulate) If a straight line falling on two straight lines make the interior
angles on the same side less than two right angles, the two straight lines, if produced
indefinitely, meet on that side on which are the angles less than the two right angles.
21. (Postulate of existence of a triangle whose angles sum to two rights) There exists a triangle
whose angles sum to two rights.
22. (Posidonius’ postulate) There exists two lines which are everywhere equidistant.
23. (Postulate of existence of similar but non-congruent triangles) There exists two similar but
non-congruent triangles.
24. (Thales’ postulate) If the circumcenter of a triangle is the midpoint of a side of a triangle,
then the triangle is right.
25. (Thales’ converse postulate) In a right triangle, the midpoint of the hypotenuse is the
circumcenter.
26. (Existential Thales’ postulate) There is a right triangle whose circumcenter is the midpoint
of the hypotenuse.
27. (Postulate of right Saccheri quadrilaterals) The angles of any Saccheri quadrilateral are
right.
28. (Postulate of existence of a right Saccheri quadrilateral) There is a Saccheri quadrilateral
whose angles are right.
29. (Postulate of right Lambert quadrilaterals) The angles of any Lambert quadrilateral are
right i.e. if in a quadrilateral three angles are right, so is the fourth.
30. (Postulate of existence of a right Lambert quadrilateral) There exists a Lambert quadri-
lateral whose angles are all right.
31. (Weak inverse projection postulate) For any angle, that, together with itself, make a right
angle, any point together with its orthogonal projection on one side of the angle form a
line which intersects the other side.
32. (Weak Tarski’s parallel postulate) For every right angle and every point T in the interior
of the angle, there is a point on each side of the angle such that T is between these two
points.
33. (Weak triangle circumscription principle) The perpendicular bisectors of the legs of a right
triangle intersect.
34. (Legendre’s parallel postulate) There exists an acute angle such that, for every point T in
the interior of the angle, there is a point on each side of the angle such that T is between
these two points.
B Definitions of the Geometric Predicates
Coq Notation Explanation

Bet A B C A B C B is between A and C.
Cong A B C D AB ≡ CD The segments AB and CD are
congruent.
Col A B C Col A B C A, B and C are collinear.
Coplanar A B C D Cp A B C D A, B, C and D are coplanar.
Par strict A B C D AB ks CD The lines AB and CD are strictly
parallel.
Par A B C D AB k XY The lines AB and CD are parallel.
CongA A B C D E F ABC =b DEF The angles ∠ABC and ∠DEF are
congruent.
PQ
TS A B P Q A B P and Q are on opposite sides of
PQ
line AB.
PQ
OS A B P Q A B P and Q are on the same side of
PQ
line AB.
SumA A B C D E F G H I ABC +
b DEF =
b GH I The angles ∠ABC and ∠DEF
sum to ∠GHI.
TriSumA A B C D E F S(4ABC) =
b DEF The angle of the triangle ABC
sum to ∠DEF .
Le A B C D AB ≤ CD The segment AB is smaller or con-
gruent to the segment CD.
Lt A B C D AB < CD The segment AB is smaller than
the segment CD.
Midpoint M A B A M B M is the midpoint of the segment
AB.
Per A B C ABC The triangle ABC is a right trian-
gle with the right angle at vertex
B.
Out P A B P A B B belongs to the ray P A.
InAngle P A B C P∈
b ABC P belongs to the angle ∠ABC.
LeA A B C D E F ABC ≤
b DEF The angle ∠ABC is smaller or
congruent to the angle ∠DEF .
LtA A B C D E F ABC <
b DEF The angle ∠ABC is smaller than
the angle ∠DEF .
Acute A B C ABC <b The angle ∠ABC is acute.
Perp at X A B C D AB ⊥ CD The lines AB and CD meet at a
X
right angle in X.
Coq Notation Explanation

Perp A B C D AB ⊥ CD The lines AB and CD are perpen-
dicular.
|=
Perp2 A B C D P AB CD The lines AB and CD have a com-
P mon perpendicular which passes
through P .
BetS A B C A B C B is strictly between A and C.
SAMS A B C D E F b DEF ≤
ABC + b2 The angles ∠ABC and ∠DEF do
not make an over-obtuse angle.
Saccheri A B C D S ABC D The quadrilateral ABCD is a Sac-
cheri quadrilateral.
Lambert A B C D L ABC D The quadrilateral ABCD is a
Lambert quadrilateral.
P 0 •P
ReflectL P’ P A B A B P 0 is the image of P by the reflec-
P 0 •P tion with respect to the line AB.
A•B
Perp bisect P Q A B P Q The line P Q is the perpendicular
A•B
bisector of the segment AB.
Defect A B C D E F D(4ABC) =
b DEF The angle ∠DEF is the defect of
the triangle ABC.
Table 3: Definitions of the geometric predicates.

Parallel Postulates Revised

Uploaded by

Copyright:

Available Formats

Parallel Postulates Revised

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Parallel Postulates Revised

Uploaded by

Copyright:

Available Formats

Parallel postulates and continuity axioms: a mechanized

study in intuitionistic logic using Coq

To cite this version:

HAL Id: hal-01178236

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

Parallel postulates and continuity axioms: a mechanized

Pierre Boutry · Charly Gries · Julien

the date of receipt and acceptance should be inserted later

Abstract In this paper we focus on the formalization of the proofs of equiva-

Keywords Euclid, parallel postulate, formalization, neutral geometry, Coq,

ICube, UMR 7357 CNRS, University of Strasbourg

To help us in this task, we have a perfect tool which possesses no intuition:

Our formal proofs rely on the systematic development of geometry based on

2.1 A set of axioms for neutral geometry

Table 1: Tarski’s axiom system for neutral geometry.

4 A Pythagorean field is a field in which every sum of two squares is a square.

Fig. 1: Axiom of segment construction A4.

The axiom of segment construction A4 allows to extend a segment by a given

Fig. 2: Five-segment axiom A5.

The five-segment axiom A5 corresponds to the well-known Side-Angle-Side

Fig. 3: Pasch’s axiom A7.

Fig. 4: Upper dimensional axiom A9.

2.1.1 Formalization in Coq

Contrary to the formalization of Hilbert’s axiom system [DDS01, BN12, BBN16b],

3 Four categories of parallel postulates

3.1 Independent parallel postulates

Let us consider four versions of the parallel postulate.

Table 2: Formalization of the axiom system in Coq .

which Archimedes’ axiom6 holds. Archimedes’ axiom is a corollary of the continuity

3.1.1 Tarski’s parallel postulate is strictly stronger than Playfair’s postulate

By dropping the law of excluded middle7 , Tarski’s parallel postulate becomes

3.1.2 Playfair’s postulate is strictly stronger than the triangle postulate

is also known under the name of Aristotle’s angle unboundedness axiom.

“Given any segment P Q, line l through Q perpendicular to P Q, and ray r

3.1.3 The triangle postulate is strictly stronger than Bachmann’s Lotschnittaxiom

Fig. 5: Graphical summary of the independence results from Sections 3.1.1-3.1.3.

Fig. 5. In order to determine in which category a version of the parallel postulate

3.2.1 Formal definition of parallelism

Definition Col A B C := Bet A B C \/ Bet B C A \/ Bet C A B.

his axiom system.

Fig. 6: Definition of Coplanar.

We did not define coplanarity in the same way as in [SST83]; we chose to

This predicate corresponds to Definition 12.3 of [SST83]. This definition asserts

3.2.2 Formal definition of the sum of non-oriented angles

Fig. 7: Definition of CongA.

This predicate corresponds to Definition 9.1 of [SST83]. The name of this

Fig. 8: Definition of TS.

Fig. 9: Definition of OS.

Fig. 10: Definition of SumA.

As we want to study the impact of Archimedes’ axiom, we cannot define the

3.2.3 Formal definition of acute angles

In order to be able to formalize a predicate specifying that an angle is acute, we

concepts was straightforward, as both of them were already present in [SST83].

Definition Midpoint M A B := Bet A M B /\ Cong A M M B.

This predicate corresponds to Definition 7.1 of [SST83]. It states that M is the

Definition Per A B C := exists C’, Midpoint B C C’ /\ Cong A C A C’.

Fig. 11: Definition of Per.

This predicate corresponds to Definition 8.1 of [SST83]. In the case where B

Definition Out P A B := A <> P /\ B <> P /\ (Bet P A B \/ Bet P B A).