ON THE DISTRIBUTION OF VALUES OF ANGLES

DETERMINED BY COPLANAR POINTS

J. H. CONWAY, H. T. CROFT, P. ERDijS AND M. J. T. GUY

1.
Given a configuration +Zof yt distinct points in the plane, no 3 colhnear, then in all
N = $n(n - l)(n - 2) angles are determined by triples of them. Define the functions
f(% 4 @I, gh 6, u>
to be the number of angles in %?strictly greater than M, strictly less than CI,respectively,
where 0 < c1< n. Further, attach superscripts + to these functions (and the functions
below) to denote the corresponding functions defined using inequalities in the wide
sense. Next, put
fr(n, a) = minf(n, CI,U); gI(vt, a) = ruin g(n, CI,@)e>;
Y? v
f2(n, u) = maxf(n, u, %); g&z, a) = maxg(n, CI,59);
w 0
(and similarly with superscripts). Finally,
Pi(rX) = limfi(n, a)/iV; G,(a) = lim gi(lt, c/-)/N; (i = 1, 2).
n’m n-rm
We shall show below (Theorem 2) that these limits do indeed exist. There are some
trivial logical relations between some of the functions Fi, G,, Iii+, Gif, thus:
CT&) = l-P,+@); F,+(cx) = I-GI(ol); F,(a) = l-G,+(@); G2+(=) = l-F,(a).
Further, we remark that the number of angles actually equal to any particular
fixed cw(0< CI< z) is o(n”) as n + co (this is a weak consequence of Theorem 1
below, or else of a result of Croft [2], giving the exact upper bound); hence it follows
that also
F 1 + = F,; G1+ = G,.

We are thus reduced to consideration of the 2 functions F,, G,: from now on we
usually drop embellishments and write F, G: crudely, F(E), G(E) denote ‘( the ultimate
proportion of angles necessarily > ~1,< 01” respectively.

2.
Exactly similar functions can be defined for conhgurations of points in 3
dimensions; we denote the functions corresponding to F, G by 5, 6; in general
similar remarks to the 2-dimensional case are in order. We content ourselves with
proving the following theorem, which obviates the need for the superscript + in 3
dimensions, as in 2:

Received 30 November, 1977.

[J. LONDON MATH. Sot. (2), 19 (1979), 137-1431
138 J.H. CONWAY,H.T.CROFT,P.ERD& AND M.J.T. GUY

THEOREM 1. In 3 dimensions, for a configuration of n points, the number of angles

exactly equal to any particular CI (0 < a < 7~) is o(n3), as n --f c;o.

Proof. Suppose not. Then by a combinatorial result of ErdBs [3; Theorem 11,
given any integer I, as large as we please, if n be sufficiently large (n > n,,(Z)), we may
extract a subset (Ai, Bj, C,; i, j, k = 1,2, . . . . I} of 31points, from our configuration of
IZ points, with the property that
A; ~j C, = ~1 (for each choice of i, j, k).
Taking first i = 1, j = 1, we see that the points C all lie on a cone vertex B1, axis
A1 B,. Keeping j = 1, but letting i = 2, 3, . . . . 2 in turn, we obtain similar results;
the upshot is that
either the points C lie on 2 straight lines (at most) both through B,,
or the points A lie on one straight line through B1. (“1
{The result holds true also in the special caseCI= +c.> For I > 5, this result contradicts
the non-collinearity condition. {Even if collinear points were admitted, use of (*),
and the corresponding statements for j = 2, 3, . . . . together with the conditions o! > 0,
and the distinctness of the points, is soon seen to lead to a contradiction.}

3.
THEOREM2. The limits F,(a) = lim fi(n, a), G,(a) = lim g,(n, a) exist.
n-tm n-c.3

Remark. The proof is of a quite general nature, and the idea can be applied to
prove the existence of a limit in other problems.

Proof. We need consider only F. Let lim supf, = A. Then, given an E > 0,
there exists an n, such that for some n > no,

Thus for this n (now fixed), there is a configuration Gk; of RZpoints with

Take M (‘ copies ” of this configuration <‘ on top of one another “, where M is very
large; the distinctness (and non-collinearity) of points is preserved by having the copies
of a typical point P-say P,, .. ., P-all distinct but very close to the original P;
so sufficiently close that, if A, B, C are any 3 points of the original (e, with ABC > ~1,
then Ai Bj C, > CX,where Ai, Bj, Ck are any copies of A, B, C, respectively. This gives
us an admissible configuration VMn of Mn points. The number of angles greater than d
in C,, is at least
(A-&)(Mn(Mtl-l)(Mn-2)-+Mn(Mn-l)(M+M)}, 0)
the second term corresponding to angles such as Ai Bj Bk having at least 2 of their
vertices copies of the same original. Indeed, for any v with Ma $ v < (M+ 1) la, we
have a %, with as many such angles as in (I), by merely taking the extra v - Mn points
arbitrarily. Then, on dividing by +v(v - l)(v - 2), we seethat the proportion of angles
 VALUES OF ANGLES DETERMINED BY COPLANAR POINTS 139

greater than a in such a ‘+ZVis at least

A--E-c/~-a,(l),
for some (absolute) constant c, where oM(l) can be made arbitrarily small merely
by making A4 large. Hence, if first n, and then A4 are chosen (suitably and) sufficiently
large, both depending on c, we obtain the result that
f(v, a) > A-2E
for all sufficiently large v. Since E is arbitrary, the lower limit is equal to the upper
limit A, as desired.
4.
We now pass to the functions F and G themselves. Easily we have:
(i) F(a) is a decreasing, and G(a) an increasing, function in (0, x), both in the
wide sense.
(ii) F is continuous on the left, G on the right. We prove merely the first, the
second being similar. For a given CQ,and a given B > 0, there exists an n, and a
configuration w8 whose proportion of angles greater than or equal to a,, is at most
F(a,) + E(where we are using the results of Theorems 1 and 2). Now, for some sufficiently
small 6 (depending on E, II and %,,), all these angles are greater than a0 -6. Now, by a
“ copying ” argument, as used in the proof of Theorem 2, we obtain some configura-
tion wMn, with arbitrarily large M, whose proportion of angIes greater than a0 - 26,
say, is at most F(Q) SE+O~(~), where oM(l) can be made small by sufficiently large
choice of M. (We need a,, - 26 rather than or,,4 because of the small displacements
caused by the (‘ copying ” operation.) Thus
F(ao-26) < F(rx,)+~+o~(l);
so, since .s is arbitrary, lim F(ol,-, -26) < F(aO). Since F is decreasing, we have
equality here, and F is c&&uous on the left.
(iii) F(a) = 3 (0 < a < 3~).
For, on the one hand, at least one angle of the 3 angles of each triangle
determined by any triplet of any V is at least $c; and on the other hand a V,, given by
taking II points all on a small circular arc of large radius gives fi(rz, CI,%‘,J = $N,
for each n (and any fixed a).
(iv) G(a) = 3 (a > 4~).
For, on the one hand, 2 angles of the 3 of any triangle of any V are necessarily
acute; and on the other hand, a %?”as in (iii) gives gl{n, a, $?,,)= +N, for each n (and
any fixed a).
There are no other obvious values of F or G; we turn to estimating them, from
above and below, for certain dt of interest.

5. F(a), a = $c. How many angles are necessarily obtuse?

Upper Bound.

THEOREM 3. F&c) < 4/27.

 140 J. H. CONWAY, H. T. CROm, P. ERD& AND M. J. T. GUY

Proof. We give a configuration with ‘I few ” obtuse angles. Take a triangle ABC,
with angles 2 = +X-Q, fi = eZ, and with BC = 1 unit. Assuming as we may that
12is divisible by 3, let V,, have: $n points Ai(i = 1,2, . . . . in) close to A, +n points B,
close to B, +a points Ci close to C; each set arranged equally spaced on small
circular arcs, thus:
the Ai on an arc through A of length s3, with centre B,
the Bi on an arc through B of length Q, with centre C,
the Ci on an arc through C of length es, with centre A.
All the E are small; the relationship between them will be chosen later. We consider
triangles of various “ types “: thus “ type AAB ” denotes generically triangles Ai Aj B,.
Altogether, there are $V triangles. The following types are all acute-angled, provided
the appropriate condition holds:
type ABC, provided Ed,e4, Ed 4 .slr s2,
type AAB, provided Ed 6 E3/n,
type BBC, provided (er + EJ . .ss-%c&t,
type CCA, provided Ed.a1 4 Es/n.
We observe that these conditions are all compatible: e.g. take
&I = E2= n-5, &3 = ne6, c4 = es = n-*.
Triangles of types ABB, BCC, CAA, of which there are 3N/81 f O(1) in number of
each type and triangles of types AAA, BBB, CCC, of which there are N/81 + O(1) in
number of each type are all necessarily obtuse-angled. Since there is a (l-f)-
correspondence between obtuse-angled triangles and the obtuse angles of %‘,, we
have (4/27&V + O(1) obtuse angles. Since n may be arbitrarily large, the result follows.

6.
Lower Bound.

THEOREM4. F&c) 2 6.

Proof. It is convenient to consider here f +(y1,@c,%,J, rather than f : take then

such a %‘n(for given n) that minimizes this, i.e. a configuration with the least number,
which we may unambiguously denote by f (n), of non-acute angles. Now remove a
point from %,,,leaving n - 1 points, which determine at least f (n - 1) non-acute angles,
by definition off. Remove thus in turn each one of the n points of Vn. Then the total
number of non-acute angles counted in the it deleted configurations, each of n- 1
points, is at least II .f (n- 1). But the total number that can be counted is: each of the
f(n) original angles, each counted just n- 3 times (for Pi ~j Pk cannot be counted
if and only if Pi or Pi or Pk is removed). SO (n-3)f (n) > nf (n- 1); SO, sincef (n) is
an integer,

f(n) 2 (5 J(M)).
 VALUES OF ANGLES DETERMINED BY COPLANAR POINTS 141

where {x} is the least integer not less than x. In addition to (2), we have the initial
condition f(4) = 1 (every quadrilateral has a non-acute angle).
Define an integer-valued function h(n) by

h(n) = h(4) = 1. (3)

Then, we show that

h(n) = { n2’;;3))

exactly. This follows by induction, for the truth of the hypothesis for n- 1 implies

n2(n-3)
= - &) = {““:p”) ,
18

using the facts that n2(n-3) is congruent to 0, -2 or -4 mod 18, and that for p1> 6,
2n/9(n - 3) < 4/9.
Thus h(n) w &V; but f(n) > h(n), by induction. The result follows.

Remark (a). It is a nice fluke that (3) has an exact solution. It appears that most
recurrence relations of the shape

k(n) = -!- .k(n-1) , k(b) = c

n-a 1

for given integers a, b, c do not have a solution in closed form. An interesting problem
is to characterize those which do.

Remark (b). If the ( } in (3) were dropped, and the resulting h used to determine
a lower bound, we would indeed obtain one, but a much worse one, namely
Wn) 2 itz-

Remark (c). If it were true thatf(5) = 4, then we could drastically improve the
estimate. However, this is false, though a counter-example is not very obvious; one
may be constructed thus: distort a square ABCD such that 2 becomes obtuse,
8, C, b acute, and the diagonals cut (at a point O), not at right angles, but with
A OB = &rt + 3e, say (E > 0); then construct a point E with OY& = OI?E (= &B) =
*n-s. Then the set {A, B, C, D, E} contains but 3 non-acute angles.

7. Other applications of the ‘( lower bound ” method.

TJiEoREM 5. F(+c) 2 5/171. Also I> 5/171.

Proof. For the first inequality we remark: any 6 coplanar points define some angle at
least $n; for the result is obvious if the convex hull contains all 6, andifnot, then an angle
 142 J. H. CONWAY, H. T, CROFT, I’. ERD6S AND M. I. T. GUY

of at least +c is defined at any internal point. The method of the previous section
gives similar equations (2), (3) for new functions (that we may again callf, h) except
that now the initial condition is f(6) = h(6) = 1; by calculation h(20) = 100, and
so, for n > 20, by dropping brackets, we obtain

n(p1-l)(n-2).100 5N
h(n) >
20.19.18 = 111 ’

The first result follows, The second has identical working; for it happens to have the
same initial condition, since any 6 points in 3 dimensions determine at least 1 non-acute
angle: see Croft [l], or Griinbaum [5], or Schiitte [a].

THEIXEM 6. For small E, F(TT-E) >, 2/8l’“.

Proof. By a result of Szekeres [7], and Erdos and Szekeres [4], n points determine
an angle greater than 7~-E if log, n > 6-l. So the initial condition to be fed into (3)
for our new function h is h((2l/“}) = 1. Dropping brackets, we obtain, for large n,

n(n-l>(n-2).1 2N
f (4 > h(n) a
{21’e)(2% 1}(21’” + 2) a 81/” *
The result follows.

We may also apply the “ lower-bound ” method to G(a),

THEOREM 7. (i) G(E) > 2.s3/n3for small 8,

(ii) G(Qn) > 3.

Proof. (i) By a result of Croft [2], n points determine some angle less than or
equal to x/n (with equality only for the regular polygon); we feed in the initial
condition h({n/e}) = 1 in (3), and working as in the last theorem, obtain
f(n) > ~NE~/~c~.
(ii) 4 points determine at least 4 angles less than or equal to 37~(at least 1 in each
triangle). The initial condition for our function h is h(4) = 4. By calculation
h(12) = 220; so for larger n, dropping brackets,

h(n) <

The result follows.

8.
THEOREM 8. F is discontinuous at a = +n; in particular, F(+n + E) < 7127.

Proof. To prove the second clause of the theorem it suffices, given an arbitrary
E, to produce a cotiguration of n points with IZ arbitrarily large, the proportions of
angles greater than 97~+E in which is at most 7/27. Without essential loss of
generality, we may, as before, take n divisible by 3; now place the n points on a
 VALUES OF ANGLES DETERMINED BY COPLANAR POINTS 143
circle, with one-third of them near each of the vertices of an inscribed equilateral
triangle ABC, With obvious notation the points are Ai, Bi, Ci (1 < i < &z). We
speak of triangles of types AAB, and so on, as before. If the points in each
collection are close enough to A, B, and C (a constant multiple of E will be found
sufficient), then all angles of triangles of type ABC are not greater than +rc+.s.
However, every other triangle is obtuse-angled with acute angles less than +n+e;
and the proportion of these triangles is 9; the result follows. The first clause of the
theorem follows on recalling that F(&r) = $.

9.
Many problems of interest on the functions F and G remain; in particular:
(i) are they discontinuous for any other a, or for infinitely many values of a?
(ii) are they strictly monotonic (outside the trivial intervals noted in $4)?-or
maybe even-strongly in the opposite sense-neither possessesat any point
a non-zero derivative?
(iii} What are the true orders of magnitude of F(n-s) and G(s), for small E?

References
1. H. T. Croft, IL On 6-point configurations in 3-space “, J. London Math. Sac., 36 (1961), 289-306,
2. H. T. Croft, “ Some geometrical thoughts II “, Math. Guz., 51 (1967), 125-129.
3. P. Erdas, “ On extremal problems of graphs and generalised graphs “, Israel J. Math., 2 (1964),
183-190.
4. P. ErdBs and C. Szckercs, ” On some extremum problems in elementary geometry “, Annales
Univ. Sci. Budapest, E&v%, Sect. Math. IIZ/ZV(196&61), 53-62.
5. B. Grtinbaum, “ Strictly antipodal sets “, Israel J. Math., 1 (1963), 5-10.
6. K. Schiitte, “ Minimale Durchmesser endlicher Punktmengen mit vorgeschriebenem Mindest-
abstand “, Mat/z. Ann., 150 (1963), 91-98.
7. G. Szekeres, “ On an extremum problem in the plane “, Amer. J. Math., 63 (1941), 208-210.

J, H. Conway, H. T. Croft,
Department of Pure Mathematics, Peterhouse,
16 Mill Lane, Cambridge.
Cambridge.

P. ErdBs, M. J. T. Guy,
Hungarian Academy of Sciences, Computing Laboratory,
Budapest. Corn Exchange Street,
Cambridge.