Simple Partitions of Space

Author(s): Gerald L. Alexanderson and John E. Wetzel

Source: Mathematics Magazine, Vol. 51, No. 4 (Sep., 1978), pp. 220-225
Published by: Taylor & Francis, Ltd. on behalf of the Mathematical Association of America
Stable URL: https://www.jstor.org/stable/2689466
Accessed: 01-05-2020 05:55 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Taylor & Francis, Ltd., Mathematical Association of America are collaborating with
JSTOR to digitize, preserve and extend access to Mathematics Magazine

This content downloaded from 128.32.10.230 on Fri, 01 May 2020 05:55:58 UTC
All use subject to https://about.jstor.org/terms
 Simple Partitions of Space

As a general plane sweeps through space

it counts the various types of regions
produced by a given collection of planes.

This article is respectfully dedicated to George Polya

on the occasion of his 90th day, December 13,1977

GERALD L. ALEXANDERSON
University of Santa Clara
Santa Clara, CA 95053

JOHN E. WETZEL
University of Illinois, Urbana-Champaign
Urbana, IL 61801

Consider an arrangement of n planes in general position in Euclidean 3-space, where by "general

position" we mean that at most three of the planes pass through each point, no more than two of th
planes pass through each line, there are no parallel planes, and no lines of intersection are parallel
planes of the arrangement. Into how many cells is space divided?
This problem is studied from the heuristic point of view by G. P6lya in [4] and again in his film
"Let Us Teach Guessing." The solution,

C=() ( 1) (2) 3+ )
was found in 1826 by J. Steiner [6]. We shall give a simple proof of a more general result using an
ingenious technique developed by A. Brousseau [1] but introduced earlier for a somewhat different
purpose by H. Hadwiger [3].
An arrangement of n planes in space is simple provided no more than three planes pass through
any point of intersection and no more than two planes pass through any line of intersection. A
completely arbitrary arrangement of planes in space can have various kinds of degeneracies: there
could be "multiple" points (i.e., points through which more than three planes pass); there could be
"multiple" lines (lines through which more than two planes pass); there could be parallel planes; or
there could be planes parallel to the line of intersection of other planes. The first two kinds of
degeneracies are ruled out in a simple arrangement, but the second two are permitted.
The planes of a simple arrangement intersect to form intersections of various dimensions. Let C be
the number of three-dimensional cells formed by the planes, F the number of faces (two-dimensional
regions) formed by the lines of intersection on the planes, E the number of edges (segments and rays)
formed on the lines of intersection by the points in which the planes meet, and V the number of
points (vertices) determined by the planes. Our first objective is to find formulas for C, F, and E in
terms of appropriate data.

The Plane

In studying a problem in 3-space, one can frequently gain insight from a thoughtful examinati
of its analogue in, or specialization to, the plane. So we first consider the problem of finding the
number R of regions and the number S of segments and rays formed in the plane by a simple
arrangement of lines.

220 MATHEMATICS MAGAZINE

This content downloaded from 128.32.10.230 on Fri, 01 May 2020 05:55:58 UTC
All use subject to https://about.jstor.org/terms
 An arrangement of lines in the plane is simple provided no more than two lines pass through any
point. A completely arbitrary arrangement of lines in the plane can have multiple points (i.e., points
through which more than two lines pass) or parallel lines. The first kind of degeneracy is ruled out in
a simple arrangement, but parallels are permitted.

THEOREM 1. If the n lines in a simple arrangement meet to form p points of intersection, then they
partition the plane into,

R=l+n+p regions
S=n+2p segments and rays, and
P=p points

Proof. Consider an auxiliary line, chosen so as not to be parallel to any of the n lines of the
arrangement. With Brousseau, we call this line a sweep-line. If initially this sweep-line is positioned
well outside the bounded regions, it is intersected by each of the n lines and cut by the points of
intersection into 1 + n segments and rays, each of which lies in (and so counts) a region in the plane.
As the sweep-line is moved parallel to its initial position across the plane, it enters a new region of the
arrangement precisely when it passes through a point of intersection, and at each such point it picks
up exactly one new region because the arrangement is simple. Since there are p points, there must be
equally many new regions. So R = 1 + n +p.
One can apply this same technique to count the segments and rays on the partitioning lines. The
sweep-line initially (that is, far removed from the bounded regions) accounts for n rays, one on each
of the partitioning lines. As the sweep-line passes across the plane, it picks up two new segments or
rays at each of the p points of intersection. Therefore, S = n + 2p.

Space

Now we try the analogous argument with a sweep-plane in space.

THEOREM 2. If the n planes in a simple arrangement in space meet to form p points of intersection an
1 lines of intersection, then they partition space into

C=1+n+l+p cells,
F=n+21+3p faces,

E= 1+ 3p edges, and
V=p points.

Proof We imitate the earlier argument and introduce a sweep-plane, which we pick n
any of the partitioning planes or to any of their lines of intersection and located initially far removed
from the bounded cells. In its initial position, this plane counts certain cells, faces, and edges. Each
region on the sweep-plane corresponds to one cell (which it cuts in two), each segment or ray on the
sweep-plane corresponds to a face of the arrangement, and each point on the sweep-plane corre-
sponds to a ray on the arrangement. Therefore, by Theorem 1, the sweep-plane initially counts
1 + n + l cells, n + 21 faces, and l edges. To justify this application of Theorem 1, one has to observe
that the lines produced in the sweep-plane by the given planes form a simple arrangement, but this is
easy to see.
Now, as the sweep-plane passes through space moving parallel to its initial position, it picks up one
new cell, 3 new faces (one on each plane passing through the point), and 3 new edges (one on each
line of intersection through the point) at each point of intersection, because the arrangement is simple.
Since there are p such points, the sums increment by exactly p cells, 3p faces, and 3p edges. This
completes the argument.

The generalization of this result to higher dimensions is strongly suggested by the patterns. The
analogue in Euclidean d-dimensional space Ed of a point on a line, a line in the plane, and a plane in

VOL. 51, NO. 4, SEPTEMBER 1978 221

This content downloaded from 128.32.10.230 on Fri, 01 May 2020 05:55:58 UTC
All use subject to https://about.jstor.org/terms
space is called a "hyperplane." It has dimension d- 1, and it has the important property that it splits
Ed into two parts, called halfspaces. The objects in Ed analogous to

points in E.
points and lines in E2
points, lines, and planes in E3,

are called "k-flats," where k indicates the dimension; so in Ed, a 0-flat is a point, a 1-flat is a line, a
2-flat is a plane,..., a (d- 1)-flat is a hyperplane.
An arrangement of n hyperplanes in Ed is simple if there are no multiple k-flats for any k
(precisely, we demand for each k that no more than d- k of the hyperplanes be "concurrent" in any
k-flat). Parallelisms of various sorts are all permitted; we do not demand that each d- k hyperplane
actually intersect to form a k-flat. Then these n hyperplanes form a partition of Ed that has, for each
with 0 < r < d, exactly
d

Crd(n) = (d- r)
k=d-r

r-dimensional cells, where fd = 1, fd- I = n, and for each k with 0 < k <
formed by the arrangement of hyperplanes. A proof using some sophisticated algebraic machinery
may be found in Zaslavsky [7], but a convincing intuitive argument can be based on the sweep-hyper-
plane method.

Steiner Data

The n lines (or planes) in a simple arrangement fall naturally into parallel families (the equivalence
classes of the parallelism relation, if we agree to say that a line (plane) is parallel to itself). If there are
s such parallel families having n1, n2, ... ., lines (planes) respectively (remember, we allow ni = 1) we
call the s-tuple <nl,n2,... ,n,> the Steiner data for the simple arrangement.
Now suppose a simple arrangement of n lines in the plane has Steiner data <ni,n2,.. ., n>. It is easy
to see that these

a, =al(nl, n2,**n.)= I 1<i ,ni = n

lines meet to determine

c2 =a2(n1, n2, ... ,n)= E1<.i<j<.,ninj

points of intersection.

THEOREM 3. A simple arrangement of lines in the plane with Steiner data <n1,n2,. . . ,n,> forms

R = aO+ al + a2 regions,

S= a, +2a2 segments and edges, and

P= G2 points,

where aO= 1, a,, and a2 are the elementary symmetric functions on the Steiner data.

Similarly, consider a simple arrangement of n planes in space having Steiner dat

director of the arrangement is a set of s planes, one chosen from each parallel family.
arrangement non-degenerate if it has a director whose planes are in general position
erate arrangement, the a, = E ni planes intersect to form a2 =1 Ii<j<s ninj lines o
and a3= E 1<i<j<k<s ninjnk points of intersection.

222 MATHEMATICS MAGAZINE

This content downloaded from 128.32.10.230 on Fri, 01 May 2020 05:55:58 UTC
All use subject to https://about.jstor.org/terms
 THEOREM 4. A non-degenerate simple arrangement of planes in space with Steiner data <n1, n2,...,n,>
forms

C=?o+?l +?2+?3 cells,

F=al +2a2+303 faces,

E= 02 + 3?3 edges, and

V= 03 vertices,

where aO= 1, a1, a2, and 03 are the elementary symmetric functions on the Steiner data.

The formulas for R and C in these corollaries were first given by Steiner, who proved them
induction. It is clear from the pattern what happens in higher dimensions. A non-degenerate sim
arrangement of n hyperplanes in Ed with Steiner data <nl,n2,...,n,> forms exactly
d

Crd(n) = k ( d-r)gk
k=d-r

r-dimensional cells, where ak is the kth elementary symmetric func

fk = ?k for each k.

Special Cases

Suppose that the n planes of a non-degenerate simple arrangement fall into s parallel familie
of which have the same number m of planes, so that ni= m for each i. Then

i= m()= M n, 2 = M2 () 3m3( ),
and we have

C= s + m S+ m 2 + m 3(35 E=m 2 S +3m 3( S

F= m( s+2m2( )+3m3(M)3 V= m3( ).

The extended face-planes of a cube, for example, form a simple arrangeme

fall into s=3 parallel families, each of which contains m =2 planes. So, as
visually,

C= 1 +2*3+22*3+23. 1=27, E=22.3+3.23. 1=36,

F=2*3+2*22*3+3323 1=54, V=23 1=8.

If m = 1, the n = s planes are in general position, and we have

0=o+ 1 2+n)(n E=2)+3()

F=( n )+2(2)+3(3) V=(3)

In the plane the analogous formulas are

R = 0 +I "+ 2) S =( I )+2 2 P = 2"

and in Ed we have, for each r with 0 <r Ad,

d

k=dr(d r )(k
k=d-r

This content downloaded from 128.32.10.230 on Fri, 01 May 2020 05:55:58 UTC
All use subject to https://about.jstor.org/terms
r-dimensional cells formed by an arrangement of n hyperplanes in general position. This formula was
first given in 1943 by R. C. Buck [2]. The case r = d was given earlier by L. Schlafli [5].
It is interesting that by means of the easy identity
(a(b( a a-c
(bJ(cJ c (c)a-b!'
Buck's formula can be written
r

Cr n) = (d-r )E( k )
k=

which involves a sum part way across a row in Pascal's triangle. For example, the number of
5-dimensional cells formed by an arrangement of n = 10 hyperplanes in general position in E7 is the
product of () =45 and the sum of the first six entries in the ninth row of Pascal's triangle, viz.,
9855.

Bounded Cells

The question of how many of the cells are bounded was also considered in 1826 by Steiner. The
situation is considerably more transparent in the plane than in space, so we begin there.
Unless the n lines of an arrangement are all parallel, each line is cut by at least one other line to
form 2n rays; and since each of these 2n rays is the clockwise boundary of an unbounded region,
there must be 2n such regions. One can also count the unbounded regions in the following way with a
sweep-line. There are unbounded regions that correspond to each segment and ray on the sweep-line
in initial position, and after the sweep-line has moved through all the bounded regions, there are new
unbounded regions that correspond to each segment (but not to the rays) on the sweep-line.
Consequently, there must be (n +1) + (n -1) = 2n unbounded regions. We shall use the analogous
sweep-plane argument in space.
Subtracting off the unbounded rays and regions, we have formulas for the number R' of bounded
regions and the number S' of segments formed by the arrangement.

THEOREM 5. In any simple arrangement of n > 2 lines in the plane not all of which are parallel,
R' = 1 - n +p = co- a1 + 02,
S'= -n+2p= -a,+2a2.
Now consider a simple arrangement of n planes in space. If the planes are all paral
space, then the arrangement has constant cross-section and forms no bounded cells,
all. Otherwise, there are unbounded cells corresponding to each unbounded region on the sweep-
plane and, as one can see with a few moments thought, to each bounded region on the sweep-plane in
initial position and again after the sweep-plane has moved through the bounded cells. There are,
therefore, 2n + 2[l - n + 1] = 2 + 21 unbounded cells. Similarly, there are unbounded faces correspond-
ing to each ray on the sweep-plane and to twice the number of segments on the sweep-plane. So there
are 2n + 2[ - n + 21] unbounded faces. And the number of rays formed is twice the number of lines, 21.
Subtracting these from the total, we find formulas for the numbers C', F', and E' of bounded cells,
faces, and edges:

THEOREM 6. In any simple arrangement of n > 3 planes in space that are not all parallel to a line,
C'= -1 + n-I +p,
F'=n-21+3p,
E'= -1+3p.

For non-degenerate simple arrrangements, these formulas can also be express

ric functions on the Steiner data.

224 MATHEMATICS MAGAZINE

This content downloaded from 128.32.10.230 on Fri, 01 May 2020 05:55:58 UTC
All use subject to https://about.jstor.org/terms
 These results extend in the obvious manner to give formulas for the number of bounded
r-dimensional cells in simple arrangements of hyperplanes in Ed. In particular, an arrangement
hyperplanes in general position in Ed forms
d

rdn= 2 (1 d-r) k)
bounded r -dimensional cells. This can be written as
r

the last equality being a well-known identity easily proved by induction.

Euler's Formulas

For any simple arrangement of n planes in space, we have found the formulas

C= l +n+l+p C'=-l +n-l+p

F=n+21+3p F'= n-21+3p

E=1+3p E'=-1+3p

V=p V=p.

If one takes alternating signs and adds, one finds that

C-F+E-V=l and C'-F'+E'-V=-l.

These formulas are analogous to Euler's polyhedral formul

faces, edges, and vertices of a convex polyhedron.
In the plane, the corresponding formulas are

R-S+P=l and R'-S'+P=l.

The higher dimensional analogues are also correct, with constant + 1 when all the r-dimensional cells
are counted and (- l)d when only the bounded r-dimensional cells are counted. The key to the proof
is the familiar identity

fk)k(k1+2k._... +(_w)k k =0
for k = 1, 2,3, . .., which arises most easily from setting x I- in the binomial expansion o 1+

References

[I] Bro. U. Alfred (Brousseau), A mathematician's progress,, Math. Teacher, 59 (1966) 722-727.
[2] R. C. Buck, Partition of space, Amer. Math. Monthly, 50 (1943) 541-544.
[3] H. Hadwiger, Eulers charakteristik und kombinatorische geometrie, J. Reine Angew. Math., 194 (1955)
101-110.
[4] G. P6lya, Induction and Analogy in Mathematics, Vol. 1 of Mathematics and Plausible Reasoning,
Princeton University Press, 1954.
[5] Ludwig Schlafli, Theorie der vielfachen Kontinuitat, Denkschriften der Schweizerischen naturforschenden
Gesellschaft, 38 (1901) 1-237; in his Gesammelte Mathematische Abhandlungen, Band I, Verlag Birk-
hauser, Basel, 1950.
[6] J. Steiner, Einige Gesetze uber die Theilung der Ebene und des Raumes, J. Reine Angew. Math., 1 (1826)
349-364.
[7] T. Zaslavsky, A combinatorial analysis of topological dissections, Advances in Math., 25 (1977) 267-285.

VOL. 51, NO. 4, SEPTEMBER 1978 225

This content downloaded from 128.32.10.230 on Fri, 01 May 2020 05:55:58 UTC
All use subject to https://about.jstor.org/terms