J. Geom.

88 (2008) 134 – 139

0047–2468/08/010134 – 6
© Birkhäuser Verlag, Basel, 2008
DOI 10.1007/s00022-007-1961-4

The Erdős-Mordell inequality is equivalent to non-positive curvature

Victor Pambuccian

Abstract. We prove that, in plane absolute geometry, the Erdős-Mordell inequality is equivalent to the statement
that the sum of the angles of a triangle is less than or equal to two right angles.
Mathematics Subject Classification (2000): 51M04, 51M09, 51F05, 03C35.
Key words: Erdős-Mordell inequality, hyperbolic geometry, Toponogov’s theorem.

1. Introduction

The Erdős-Mordell inequality has been the subject of many papers, all — with two
exceptions, [10] and [15], in which different inequalities are proved for spherical trian-
gles — presenting alternate proofs or generalizations of the inequality in the Euclidean
setting. Proofs are presented in [1], [3], [5]-[7], [9], and a generalization to the convex
n-gon in [8]. The aim of this note is to investigate what axiomatic framework is needed for
its validity. The theorem can be stated as

THEOREM 1. For any triangle ABC and any point P in its interior or on its boundary, the
sum S1 (ABC) of the distances from P to the sides and the sum S2 (ABC) of the distances
from P to the vertices satisfy the inequality
S2 (ABC) ≥ 2S1 (ABC). (1)
Equality holds only in case ABC is equilateral and P is its centroid.
We are also interested in the following stronger version of the original Erdős-Mordell
theorem:

THEOREM 2. (Barrow [9]). Inequality (1) holds with S1 (ABC), the sum of the lengths of
the internal bisectors of angles  AP B,  BP C, and  CP A, instead of S1 (ABC). Equality
holds only in case ABC is equilateral and P is its centroid.
To even state Theorem 1 in axiomatic geometry one needs metric concepts that make the
notion of distance from a point to a line meaningful, as well as the notion of betweenness
to make sense both of the fact that P lies inside or on the boundary of ABC and of the
inequality sign in (1).

134
Vol. 88, 2008 The Erdős-Mordell inequality is equivalent 135

To make sense of segment addition we need free mobility as well, i. e. the ability to lay
off a segment on a given ray. The axiom system for such a geometry is the one consisting
of the plane axioms in groups I, II, and III of Hilbert’s Grundlagen der Geometrie, being
equivalent with the axioms A1-A9 in [14], the models of which have been called H-planes
in [4].
All models of H-planes have been described algebraically by W. Pejas [12]. The descrip-
tion reads:
Let K be a field of characteristic  = 2, and k an element of K. By the affine-metric plane
A(K, k) (cf. [4, p.215]) over the field K with the metric constant k we mean the projective
plane P(K) over the field K from which the line [0, 0, 1], as well as all the points on it
have been removed (and we write A(K) for the remaining point-set), for whose points of
the form (x, y, 1) we shall write (x, y) (which is incident with a line [u, v, w] if and only if
xu+yv +w = 0), together with a notion of orthogonality, the lines [u, v, w] and [u , v  , w ]
being orthogonal if and only if uu + vv  + kww  = 0. If K is an ordered field, then one
can order A(K) in the usual way.
The algebraic characterization of the H -planes consists in specifying a point-set E of an
affine-metric plane A(K, k), which is the universe of the H -plane. Since E will always lie
in A(K), the H -plane will inherit the order relation B from A(K). The congruence of two
segments ab and cd will be given by the usual Euclidean formula (a1 − b1 )2 + (a2 − b2 )2 =
(c1 − d1 )2 + (c2 − d2 )2 if E ⊂ A(K, 0), and by

F (a, b)2 F (c, d)2

= ,
Q(a)Q(b) Q(c)Q(d)

if E ⊂ A(K, k) with k  = 0, where F (x, y) = k(x1 y1 + x2 y2 ) + 1, Q(x) = F (x, x), and

x = (x1 , x2 ), y = (y1 , y2 ).
Let now K be an ordered Pythagorean field (i. e. the sum of any two squares of elements
of K is the square of an element of K), R the ring of finite elements, i. e. R = {x ∈ K :
(∃n ∈ N) |x| < n} and P the ideal of infinitely small elements of K, i. e. P = {0} ∪ {x ∈
K : x −1  ∈ R}. All H -planes are isomorphic to a plane of one of the following three types:

Type I. E = {(a, b) | a, b ∈ M} ⊂ A(K, 0), where M is an R-module  = (0);

Type II. E = {(a, b) | a, b ∈ M} ⊂ A(K, k) with k  = 0, where M is an R-module  = (0)

included in {a ∈ K | ka 2 ∈ P }, which satisfies the condition

a ∈ M ⇒ ka 2 + 1 ∈ K 2 ;

Type III. E = {x | Q(x) > 0, Q(x)  ∈ J } ⊂ A(K, k) with k < 0, where J ⊆ P is a prime
136 Victor Pambuccian J. Geom.

ideal of R which satisfies the condition

ka 2 + 1 > 0, ka 2 + 1  ∈ J ⇒ ka 2 + 1 ∈ K 2 ,

with K satisfying

{a ∈ K | ka 2 ∈ R \ P } = ∅.

In planes of type I there exist rectangles, so their metric is Euclidean, and we may think of
them as ‘finite’ neighborhoods of the origin inside a Cartesian plane. Those of type II can
be thought of as infinitesimally small neighborhoods of the origin in a non-Archimedean
ordered affine-metric plane. There is no rectangle in them, and their metric may be of
hyperbolic type (should k < 0) or of elliptic type (should k > 0) — in the latter the sum
of the angles of a triangle can exceed two right angles only by an infinitesimal amount.
Planes of type III are generalizations of the Klein inner-disc model of hyperbolic geometry.
A certain infinitesimal collar around the boundary may be deleted from the inside of a
disc, and the metric constant k, although negative, may not be normalizable to −1, as the
coordinate field is only Pythagorean and not necessarily Euclidean. In case K is a Euclidean
field (every positive element has a square root) and J = (0), we can normalize the metric
constant k to −1 and we have Klein’s inner-disc model of plane hyperbolic geometry with
K as coordinate field.
The main result of this paper is that

THEOREM 3. The full text of Theorem 1 is true only in planes of type I, i. e. when k = 0.
The inequality (1) holds in all planes with k ≤ 0, and does not hold in planes with k > 0.

2. Facts and lemmas

Let H and E stand for the real hyperbolic plane and the real Euclidean plane. Let A stand
for plane absolute geometry, the theory axiomatized by axioms A1-A9 in [14]. Then:

FACT 1 (holds in A) If ABC and AB  C are triangles such that AB ≡ AB  and  CAB <
 CAB  , then BC < B  C.

FACT 2 (holds in A) If ABC is a triangle, with m( BAC) ≥ 90◦ , and B  is a point on the
side AB, C  is a point on the side AC, such that {B  , C  }  = {B, C}, then B  C  < BC.

FACT 3 (Toponogov’s theorem, in the form that compares geodesic hinges, cf. [2, Theorem
7.8, B]). If ABC is a triangle in H and A B  C  is a triangle in E such that AB has the same
length as A B  , AC has the same length as A C  ,  BAC has the same measure as  B  A C  ,
then the length of B  C  is less than that of BC.
Vol. 88, 2008 The Erdős-Mordell inequality is equivalent 137

LEMMA 1. If ABC is a triangle in H and A B  C  is a triangle in E such that AB has the

same length as A B  , AC has the same length as A C  ,  BAC has the same measure as
 B  A C  , then the length of the altitude AD in triangle ABC is less than the length of the
angle bisector A D  of  B  A C  in triangle A B  C  .

Proof. We denote by e(AB) the Euclidean length of a segment AB in E, by h(AB) the

hyperbolic length of a segment AB in H, and by m( XY Z) the measure, in degrees, of
 XY Z, for angles lying in either of H and E. Let E  be the foot of the altitude from A in
triangle A B  C  .
If both D lies on the segment BC and E  lies between B  and C  , then we can show that
h(AD) < e(A E  ), and since e(A E  ) ≤ e(A D  ), that h(AD) < e(A D  ). Suppose this
were not the case, and h(AD) ≥ e(A E  ). By Fact 3, h(BC) > e(B  C  ), and thus one
of h(BD) > e(B  E  ) or h(DC) > e(E  C  ) must hold. W. l. o. g. we may assume that
h(BD) > e(B  E  ). Let M be the point between B and D for which h(DM) = e(B  E  ),
and N be the point on the segment DA for which h(DN ) = e(A E  ). By Fact 3 applied to
triangles DMN and E  B  A , we get that h(MN ) > e(A B  ), and, since e(A B  ) = h(AB),
h(MN ) > h(AB) must hold, contradicting Fact 2.
If D lies on the segment BC, but E  does not lie between B  and C  , say, such that B  lies
on the segment E  C  , then the angle bisector A D  of  B  A C  is longer than A B  (by Fact
2 if B  is different from E  ), thus e(A D  ) > e(A B  ) = h(AB) ≥ h(AD).
If D lies outside of BC, say, such that B lies between C and D, then E  must be such that
B  lies between C  and E  .
Suppose this were not the case. Then E  lies on the segment B  C  or C  lies on the segment
B  E  . In either case we have m( A B  C  ) ≤ 90◦ . Let M  be such that e(B  M  ) = e(B  C  )
and m( A B  M  ) = m( ABC), and let N  be the point on ray B  M  for which e(B  N  ) =
h(BC). Then

m( A B  C  ) ≤ 90◦ < m( ABC) = m( A B  M  ), (2)

and, by Fact 3,

e(B  M  ) = e(B  C  ) < h(BC) = e(B  N  ). (3)

Using at first (2) with Fact 1, and then (2) and (3) with Fact 2, we get e(A C  ) < e(A M  ) <
e(A N  ). However, by Fact 3, applied to triangles BCA and B  N  A , we have e(A N  ) <
h(AC) = e(A C  ), a contradiction. Thus, m( A B  C  ) must be ≥ 90◦ , so that e(A D  ) ≥
e(A B  ) = h(AB) > h(AD).

LEMMA 2. Inequality (1) holds with strict inequality sign in H.

138 Victor Pambuccian J. Geom.

Proof. Let, in H, ABC be a triangle and P be a point inside it or on its boundary. W. l. o. g.

we may assume that P is different from the vertices of triangle ABC. Let P  , A , B  , C 
be four points in E such that e(P  A ) = h(P A), e(P  B  ) = h(P B), e(P  C  ) = h(P C),
m( A P  B  ) = m( AP B), m( B  P  C  ) = m( BP C), m( C  P  A ) = m( CP A).
Notice that S2 (ABC) = S2 (A B  C  ) and that, by Lemma 1, S1 (ABC) < S1 (A B  C  ).
Since Barrow’s Theorem 2 holds for triangle A B  C  , we have S2 (ABC) = S2 (A B  C  ) ≥
2S1 (A B  C  ) > 2S1 (ABC).

LEMMA 3. Inequality (1) does not hold in H-planes with metric constant k > 0.

Proof. According to [11], in any H-plane there is an equilateral triangle. Let, in an H-plane
with k > 0, ABC be an equilateral triangle and P be the point of intersection of its
perpendicular bisectors. Let P  be the reflection of P in the side BC. Inequality (1) amounts
in this case to the inequality BP ≥ P P  , which does not hold, given that BP ≡ BP  ,
m( BP P  ) = m( BP  P ) = 60◦ , thus, since the sum of the angles of a triangle in a plane
with k > 0 is > 180◦ (see [12, Satz 5]), m( P BP  ) > 60◦ , so P P  > BP .

3. Proof of the main theorem

We are now ready to prove Theorem 3. Let M be an H-plane with coordinate field K and
metric constant k < 0. Notice that, expressed in terms of the relations of equidistance ≡
and betweenness B, as in [14], inequality (1) becomes a universal statement σ . Let K be the
real closure of K and let K be the Klein model of hyperbolic geometry with K as coordinate
field (i.e. the H-plane of Type III, with K as coordinate field, k as metric constant, and
in which J = {0} — although the metric constant k can be normalized to −1, we prefer
to think of it as not normalized). Thus M is a submodel of K. By Tarski’s Theorem the
first-order theory of real-closed fields is complete (cf. [13], Theorem 4.28 or [14]). Thus
the Transfer Principle (cf. the discussion of the meaning of the notion of completeness in
[13], Chapter 4, p. 89), which has become part of the algebraic geometry folklore, where
it is referred to as the Seidenberg-Tarski principle, is applicable to R and K. Therefore, by
Lemma 2, σ must hold in K since it holds in H. Being a universal sentence, σ must hold in
all submodels of K as well, thus it must hold in M. The usual Euclidean proofs of Theorem
1 are valid in all H-planes of Type I, so σ holds in all H-planes with k ≤ 0. By Lemma 3,
it does not hold in H-planes with k > 0. Since, by Lemma 2, equality cannot occur in (1),
the full statement of Theorem 1 holds only in planes with k = 0.

4. Remarks and acknowledgement

We do not know whether Barrow’s generalization, Theorem 2, or the generalization of

Theorem 1 to convex n-gons proved in [8] holds in H.
Vol. 88, 2008 The Erdős-Mordell inequality is equivalent 139

This paper was written while the author was visiting the University of Dortmund with a study
visit grant of the DAAD. I thank Professor Tudor Zamfirescu for very fruitful conversations
and the DAAD for its support. I thank the referee for suggestions which improved the
presentation.

References
[1] L. Bankoff, An elementary proof of the Erdős-Mordell theorem, Amer. Math. Monthly 65 (1958) 521.
[2] I. Chavel, Riemannian geometry: a modern introduction, Cambridge University Press, Cambridge, 1993.
[3] P. Erdős, Problem 3740, Amer. Math. Monthly 42 (1935) 396.
[4] G. Hessenberg and J. Diller, Grundlagen der Geometrie, 2. Auflage, Walter de Gruyter, Berlin, 1967.
[5] D. K. Kazarinoff, A simple proof of the Erdős-Mordell inequality for triangles, Michigan Math. J. 4 (1957)
97–98.
[6] V. Komornik, A short proof of the Erdős-Mordell theorem, Amer. Math. Monthly 104 (1997) 57–60.
[7] H. Lee, Another proof of the Erdős-Mordell theorem, Forum Geom. 1 (2001) 7–8.
[8] H.-C. Lenhard, Verallgemeinerung und Verschärfung der Erdős-Mordellschen Ungleichung für Polygone,
Arch. Math. 12 (1961) 311–314.
[9] L. J. Mordell and D. F. Barrow, Solution to 3740, Amer. Math. Monthly 44 (1937) 252–254.
[10] A. Oppenheim, Some inequalities for a spherical triangle and an internal point. Publ. Fac. Electrotech.
Univ. Belgrade, Ser. Math. Phys. 200-209 (1967) 13–16.
[11] V. Pambuccian, Zur Existenz gleichseitiger Dreiecke in H-Ebenen, J. Geom. 63 (1998) 147–153.
[12] W. Pejas, Die Modelle des Hilbertschen Axiomensystems der absoluten Geometrie, Math. Ann. 143 (1961)
212–235.
[13] A. Robinson, Introduction to model theory and to the metamathematics of algebra, North-Holland, Ams-
terdam, 1963.
[14] W. Schwabhäuser, W. Szmielew and A. Tarski, Metamathematische Methoden in der Geometrie, Springer
Verlag, Berlin, 1983.
[15] L. Si, B. W. He and G. S. Leng, Erdős-Mordell inequality on a sphere in R3 , J. Shanghai Univ. Nat. Sci.
10 (2004) 56–58.

Victor Pambuccian
Department of Integrative Studies
Arizona State University - West campus
P. O. Box 37100
Phoenix, AZ 85069-7100
U.S.A.
e-mail: pamb@math.west.asu.edu

Received 17 July 2006; revised 31 May 2007