The Kepler-Poinsot Polyhedra

A polyhedron is regular if the faces are a single kind of regular polygon and
the vertices are all the same. The 5 Platonic Solids are the convex regular
polyhedrons. If we remove the constraint of convexity it turns out that there are
only four more solids that can be added to the list; these are known as the
Kepler-Poinsot Polyhedra.

It was Johann Kepler who, in 1619, first realized that 12 pentagrams can be
joined in pairs along their edges in two different ways that result in regular
solids. If five pentagrams meet at each vertex, the resulting solid has come to
be known as the small stellated dodecahedron.

Small Stellated Dodecahedron

If three pentagrams meet at each vertex, the resulting solid is now named the
great stellated dodecahedron (The perhaps surprising reason for these names
will be made evident shortly).

Great Stellated Dodecahedron

Two centuries later, in 1809, Louis Poinsot discovered two more non-convex
regular solids: the great dodecahedron and the great icosahedron. The twelve
faces of the great dodecahedron are pentagons (as with the ordinary
dodecahedron), but which intersect each other. Likewise, the faces of the great
icosahedron are the 20 triangles of the ordinary icosahedron, but intersecting
each other.

The great dodecahedron is a most pleasing and intriguing solid, giving the
illusion of a pentagonal star embossed on each pentagon; each star shares each
of its arms with another, so that one star disappears as soon as you bring your
attention to another!

Great Dodecahedron

It has been proven that these four solids, together with the 5 platonic solids, are
the only regular solids possible. Another wonderful truth is that the Kepler
solids are the duals of the Poinsot solids!

Great Icosahedron

It was not until 1811 that the French mathematician Augustin Cauchy showed
that the Kepler-Poinsot solids are stellated forms of the dodecahedron or the
icosahedron. It was this insight that led to the names for these solids that we use
today.

The Platonic Solids

A platonic solid is a polyhedron all of whose faces are congruent regular polygons, and
where the same number of faces meet at every vertex. The best know example is a cube
(or hexahedron ) whose faces are six congruent squares.

Manipulating the shapes on this page.

If you have a Java compatible browse you can click on any of the figures of this page and
bring up a new window in which you can rotate the polyhedron and so get a better feeling
for its shape. The applets are special cases of a simple 3D geometry viewer that comes
with instructions.

There are only five!

The Greeks recognized that there are only five platonic solids. But why is this so? The
key observation is that the interior angles of the polygons meeting at a vertex of a
polyhedron add to less than 360 degrees. To see this note that if such polygons met in a
plane, the interior angles of all the polygons meeting at a vertex would add to exactly 360
degrees. Now cut an angle out of paper, and fold another piece of paper to that angle
along a line. The first piece will fit into the second piece when it is perpendicular to the
fold. Think of the fold as a line coming out of our polyhedron. The faces of the
polyhedron meet at the fold at angles less than 90 degrees. How can this be possible? Try
wiggling your first piece of paper within the second. To be able to incline it with respect
to the fold you have to decrease the angle of the first piece, or increase the angle of the
second.

Next we'll consider all possibilities for the number of faces meeting at a vertex of a
regular polyhedron. For each possibility we actually construct such a polyhedron, a
picture of which you can see close by on this page. Here are the possibilities:

 Triangles. The interior angle of an equilateral triangle is 60 degrees. Thus on a

regular polyhedron, only 3, 4, or 5 triangles can meet a vertex. If there were more
than 6 their angles would add up to at least 360 degrees which they can't.
Consider the possibilities:
o 3 triangles meet at each vertex. This gives rise to a Tetrahedron.
o 4 triangles meet at each vertex. This gives rise to an Octahedron.
o 5 triangles meet at each vertex. This gives rise to an Icosahedron
 Squares. Since the interior angle of a square is 90 degrees, at most three squares
can meet at a vertex. This is indeed possible and it gives rise to a hexahedron or
cube.
 Pentagons. As in the case of cubes, the only possibility is that three pentagons
meet at a vertex. This gives rise to a Dodecahedron.
  Hexagons or regular polygons with more than six sides cannot form the faces of a
regular polyhedron since their interior angles are at least 120 degrees.

But now things get a little more subtle. We have looked at all possibilities of congruent
regular polygons meeting at a vertex of a polyhedron, but how do we know that there isn't
another regular polyhedron for some of these cases? For example, why is the cube the
only polyhedron for which three squares meet at each vertex? The rest of this page gives
the answer to this question, but the going will be much harder!

Before continuing, let us collect some data. Let

 m be the number of polygons meeting at a vertex,
 n the number of vertices of each polygon,
 f the number of faces of the polyhedron,
 e the number of edges of the polyhedron, and
 v the number of vertices of the polyhedron.

The values of these numbers for each of the polyhedra are listed in this table:

n m f e v

Tetrahedron 3 3 4 6 4

Octahedron 3 4 8 12 6

Icosahedron 3 5 20 30 12

Hexahedron 4 3 6 12 8

Dodecahedron 5 3 12 30 20

Table: Combinatorics of Regular Polyhedra

Our aim now is to show that for any pair of number n and m the values of the other
parameters, f, e, and v are determined uniquely.

First we note that since two faces meet in one edge, we must have

e = nf/2

Next, since every vertex is shared by m faces, we must have

v = nf/m
It is apparent from the Table that for all five regular polyhedra

f=2+e-v (E)

We'll see below that this equations actually holds for all convex polyhedra. Given m and
n the above three equations determine f, e, and v uniquely, and so there are only five
possible regular polyhedra.

The result (E) is known as

Euler's Polyhedron Theorem

To see why it is true we proceed in several steps. First we remove one face from the
polyhedron. Let
F=f-1
be the new number of faces. We need to show
F=1+e-v (*)

Now think of the remaining faces of the polyhedron as made of rubber and stretched out
on a table. This will certainly change the shape of the polygons and the angles involved,
but it will not alter the number of vertices, edges, and faces. Now we draw diagonals in
the stretched faces out of the Polygons. Every diagonal increase the number e of edges by
one, and also the number F of faces, so that our equation (*) remains valid. We continue
this process until all polygons have been changed into triangles.

In the final stage we remove triangles until we are left with only one triangle for which
(*) is obviously true. How do we do that? If the removed triangle has exactly one edge on
the boundary then F and e are both decreased by 1 and (*) remains true. If it has two
edges on the boundary then F is reduced by 1, e is reduced by 2, and v is reduced by 1, so
that (*) remains true.

There is one final subtlety. Can we really dismantle the triangles as described? The
answer is yes. But as an exercise you may wish to modify the dismantling procedure to
remove all doubts in your mind. A similar dismantling procedure could be designed for a
tessellation of a polyhedron by polyhedra, but in that case it is not always possible. For
an illustration you may want to visit my page that describes Rudin's example of an
unshellable triangulation.

If you'd like to play with polyhedron with many more faces, here is a crude rendering of a
sphere, which is of course not a platonic solid!
Platonic Solids in All Dimensions

John Baez

November 12, 2006

What's the pattern in this sequence?

infinity, five, six, three, three, three, three, three, ...

In 2 dimensions, the most symmetrical polygons of all are the 'regular

polygons'. All the edges of a regular polygon are the same length, and
all the angles are equal. If you only count the convex ones, it's easy to
list all the regular polygons: the equilateral triangle, the square, the
regular pentagon, and so on. In short, there is an infinity of regular
polygons: one with n sides for each n > 3. (The cases n = 0,1, and 2
are bit degenerate.)

In 3 dimensions, the most symmetrical polyhedra of all are the 'regular

polyhedra', also known as the 'Platonic solids'. All the faces of a
Platonic solid are regular polygons of the same size, and all the
vertices look identical. We also demands that our Platonic solids be
convex. There are only five Platonic solids:

 The tetrahedron, with 4 triangular faces:

 The cube, with 6 square faces:

 The octahedron, with 8 triangular faces:

 The dodecahedron, with 12 pentagonal faces.

  The icosahedron, with 20 triangular faces:

The story goes on... but in higher dimensions one usually uses the term
'regular polytopes' instead of 'Platonic solids'. All the faces of a regular
polytope must be lower-dimensional regular polytopes of the same size
and shape, and all the vertices, edges, etc. have to look identical.
Maximal symmetry, that's the name of the game! (Also, I'll only be
talking about convex polytopes.)

In 4 dimensions, there are exactly six regular polytopes.

How can visualize these? Well, a Platonic solid looks a lot like a sphere
in ordinary 3-dimensional space, with its surface chopped up into
polygons. So, a 4d regular polytope looks a lot like a sphere in 4-
dimensional space with its surface chopped up into polyhedra! A
sphere in 4-dimensional space is called a '3-sphere', since people living
on its surface would experience it as a 3-dimensional universe with the
curious feature that if you hop aboard a rocket and shoot off straight in
any direction, you eventually wind up back where you started. (This is
just like what happens when you start walking in a straight line in any
direction on an ordinary sphere.)

So, we can visualize the regular polytopes in 4 dimensions by taking a

3-sphere and drawing it chopped up into polyhedra. A 3-sphere is hard
to draw until you realize it looks just like ordinary 3d space except that
it 'wraps around'... very far away from here. But if we ignore that, and
just draw a nearby portion of the 3-sphere chopped up into polyhedra,
with everything outside this portion being one big polyhedron, we'll do
okay. And this is what we get:

 The 'hypertetrahedron' - mathematicians call it the '4-simplex' -

with 5 tetrahedral faces:

Some people call this a 'pentatope' or 'pentachoron'.

 The 'hypercube' - science fiction writers call it the 'tesseract' -

with 8 cubical faces:
 You'll notice the edges are bulging out on these pictures: that's
because they're drawn in a 3-sphere! We can also draw the
pictures in a 'flat' style, which may be more familiar, especially
for the hypercube:

This shows the 'walls' of the 8 cubical faces, as well as their

edges. Do you see the 8 cubical faces? You may only see 7, but
that's because you're ignoring the cube on the outside of the
whole picture.... remember, we're in a 3-sphere here.

 The 'hyperoctahedron' - mathematicians call it the '4-

dimensional cross-polytope' or '16-cell', with 16 tetrahedral
faces:
 A few people call this an 'orthoplex', or a 'hexadecachoron'.

 The 'hyperdodecahedron' - mathematicians call it the '120-cell' -

with 120 dodecahedral faces. This one is one of my favorites, so
let's see it made of platinum struts attached by gold spheres:
 The 'hypericosahedron' - mathematicians call it the '600-cell' -
with 600 tetrahedral faces:
 Zounds!

 Last but not least, the '24-cell', with 24 octahedral faces. This is a
denizen of the 4th dimension with no analog in lower
dimensions:

You might things would keep getting more complicated in higher

dimensions. But it doesn't! 4-dimensional space is the peak of
complexity as far as regular polytopes go. From then on, it gets pretty
boring. This is one of many examples of how 4-dimensional geometry
and topology are more complicated, in certain ways, than geometry
and topology in higher dimensions. And the spacetime we live in just
happens to be 4-dimensional. Hmm.

In 5 or more dimensions, there are only three regular polytopes:

 There is a kind of hypertetrahedron, called the 'n-simplex',

having (n+1) faces, all of which are (n-1)-simplices.
 There is a kind of hypercube, called the 'n-cube', having 2n
faces, all of which are (n-1)-cubes.
 And there is a kind of hyperoctahedron, called the 'n-dimensional
cross-polytope', having 2n faces, all of which are (n-1)-simplices.

How can we understand the proliferation of regular polytopes in 4

dimensions? And how can we visualize them?

For starters, try Tony Smith's webpage. If you cross your eyes while
gazing at this stereoscopic pair of images, you will see a 24-cell
rotating in the 4th dimension, with the 4th dimension depicted using
color. If you gaze long enough, you may become enlightened.
Foldout Models of 4-Dimensional Platonic Solids

Here's another way to visualize the 4-dimensional regular polytopes.

Ever make a cube out of paper? You draw six squares on the paper in a
cross-shaped pattern, cut the whole thing out, and then fold it up... it's
called a 'foldout model' of a cube.

When you do this, you're taking advantage of the fact that the interior
angles of 3 squares don't quite add up to 360 degrees: they only add
up to 270 degrees. So if you try to tile the plane with squares in such a
way that only 3 meet at each vertex, the pattern naturally 'curls up'
into the 3rd dimension - and becomes a cube!

The same idea applies to all the other Platonic solids. And we can
understand the 4d regular polytopes in the same way!

For example: suppose you take a cube and push in the middle of each
face, making a dent shaped like an inverted pyramid. Keep pushing in
until the tips of all these pyramids meet at the cube's center.

Now you have a cube with 6 pyramid-shaped dents that meet at a

point in the center. Isn't it tempting to take 6 regular octahedra and fit
their corners into these dents? If they fit perfectly, maybe we could tile
3-dimensional space with regular octahedra, 6 meeting at each vertex!

Alas, they don't fit perfectly: there's a little 'wiggle room'. You can
either take my word for this, or check it yourself....

But we can snatch victory from the jaws of defeat. We can't tile 3d
space with octahedra this way, but if we let the pattern 'curl up' into
the 4th dimension, we get a 4d regular polytope! This is the 24-cell. It
has 24 octahedral faces, 6 meeting at each vertex.

Next let's do the same trick starting with a regular tetrahedron. Push in
each triangular face, getting a dent in the shape of somewhat squat
triangular pyramid. Keep pushing until the tips of all these dents meet
at the center of our original tetrahedron.

Now stick a regular tetrahedron in each dent. There's a lot of wiggle

room this time. So let the pattern curl up into the 4th dimension... and
get the 4-simplex, with 5 tetrahedral faces, 4 meeting at each vertex!

In fact, there's so much room in these dents that we can even stick the
corner of a cube in each one. If we do this, there's still some wiggle
room - and if we let the pattern curl up into the 4th dimension, we get
the hypercube, with 8 cubical faces, 4 meeting at each vertex!

Actually, we can even go further - we can stick the corner of a

dodecahedron in each dent. This time there's only a tiny bit of wiggle
room. If we let the pattern curl up, we get the 120-cell, with 120
dodecahedral faces, 4 meeting at each vertex!

This is fun - so let's try another Platonic solid. This time, let's start with
a regular octahedron. Push in each of the 8 triangular faces, getting
dents in the shape of triangular pyramids. Keep pushing until the dents
meet at the middle, and then stick a regular tetrahedron in each of the
8 dents! There's some wiggle room - though not as much as last time -
so again, let the pattern curl up in 4-dimensional space... and get the
4-dimensional cross-polytope, with 16 tetrahedral faces, 8 meeting at
each vertex!

Next, let's take an icosahedron and do the same trick. Push in each of
the 20 triangular faces, making dents in the shape of triangular prisms,
and keep pushing until the tips of all these dents meet at the center of
the icosahedron. Now stick a regular tetrahedron in each dent. There's
only a tiny bit of wiggle room this time! But go ahead, let the pattern
curl up into the 4th dimension.... and get the hypericosahedron, with
600 tetrahedral faces, 20 meeting at each vertex!

(Note the pattern: the less wiggle room we have, the bigger our 4d
regular polytope is.)

Finally, let's do the same procedure starting from a dodecahedron.

Here each dent looks like it wants the corner of an icosahedron put into
it - so go ahead and try!

Hmm. Wait a minute! Is there wiggle room this time, or not? And even
if there is, are we guaranteed to get a regular polytope using this trick?

I'll leave this as a little puzzle for you... If you give up, click here.

I hope you've done your best to visualize everything I just said. But if
you had trouble, don't feel too bad! Andrew Weimholt has drawn
pictures of foldout models of the 4d regular polytopes. If you look at
these, what I said should make more sense.

Platonic Solids and the Quaternions

While you're pondering that, let me tell you another way to get some
of the 4d regular polytopes. This method involves quaternions, which
are a souped-up version of the complex numbers with three square
roots of -1, called i, j, and k. A typical quaternion looks like this:

a + bi + cj + dk

where a,b,c, and d are real numbers. To multiply the quaternions, you
need to use these rules, invented by Hamilton back in 1843:

i2 = j2 = k2 = -1

ij = -ji = k

jk = -kj = i

ki = -ik = j

Let's start with the 24-cell, since this guy has no analog in other
dimensions. Since the vertices of the 24-cell lie on the unit sphere in 4
dimensions, we can think of its vertices as certain unit quaternions.
The 24-cell happens to have, not only 24 faces, but also 24 vertices!
We can take them to be precisely the unit 'Hurwitz integral
quaternions', which are quaternions of the form

a + bi + cj + dk

where a,b,c,d are either all integers or all integers plus 1/2. One can
check that the Hurwitz integral quaternions are closed under
multiplication, so the vertices of the 24-cell form a subgroup of the unit
quaternions. A regular polytope that's a symmetry group in its own
right - ponder that while you cross your eyes and gaze at it spinning
around!

Similarly, the 600-cell has 120 vertices, which we can think of as

certain unit quaternions. We can take them to be precisely the unit
'icosians'. These are quaternions of the form

a + bi + cj + dk

where a,b,c,d all live in the 'golden field' - meaning that they're of the
form x + √5 y where x and y are rational. Since the icosians are closed
under multiplication a group under multiplication, the vertices of the
120-cell also form a group!
The vertices of the 4-dimensional cross-polytope also form a subgroup
of the unit quaternions. But this one is a little less exciting. We just
take the quaternions of the form

a + bi + cj + dk

where one of the numbers a,b,c,d is 1 or -1, and the rest are zero. This
8-element subgroup is sometimes called 'the quaternion group'.

Those are all the 4-dimensional regular polytopes that are also groups.
Three out of six ain't bad! But we can get most of the rest using
duality.

In general, the 'dual' of a regular polytope is another polytope, also

regular, having one vertex in the center of each face of the polytope
we started with. The dual of the dual of a regular polytope is the one
we started with (only smaller). So polytopes come in mated pairs -
except for some 'self-dual' ones.

In 2 dimensions, every regular polytope is its own dual.

In 3 dimensions, the tetrahedron is self-dual. The dual of the cube is

the octahedron. And the dual of the dodecahedron is the icosahedron.

In 4 dimensions, the 4-simplex is self-dual. The 24-cell is also self-dual -

that's why it had 24 faces and also 24 vertices! The dual of the
hypercube is the 4-dimensional cross-polytope. The dual of the 120-cell
is the 600-cell.

In higher dimensions, the n-simplex is self-dual, and the dual of the n-

cube is the n-dimensional cross-polytope.

But what is so special about 4 dimensions, exactly?

Well, there are very few dimensions in which the unit sphere is also a
group. It happens only in dimensions 1, 2, and 4! In 1 dimensions the
unit sphere is just two points, which we can think of as the unit real
numbers, -1 and 1. In 2 dimensions we can think of the unit sphere as
the unit complex numbers, exp(i theta). In 4 dimensions we can think
of the unit sphere as the unit quaternions.

Only in these dimensions do we get polytopes that are also groups in a

natural way. In 2 dimensions all the regular n-gons correspond to
groups consisting of the unit complex numbers exp(2πi / n). In 4
dimensions things are more subtle and interesting. It's especially
interesting because the group of unit quaternions, also known as
SU(2), happens to be the 'double cover' of the rotation group in 3
dimensions. Roughly speaking, this means that there is a nice function
sending 2 elements of SU(2) to each rotation in 3 dimensions.

This gives a slick way to construct the 600-cell, or hypericosahedron.

Take the icosahedron in 3 dimensions. Consider its group of rotational
symmetries. This is a 60-element subgroup of the rotation group in 3
dimensions. Now look at the corresponding subgroup of SU(2) - its
'double cover', so to speak. This is a 120-element subgroup of the unit
quaternions. These are the vertices of the hypericosahedron! So in a
very real sense, the hypericosahedron is just the symmetries of the
icosahedron! This trick doesn't work in higher dimensions. This is one
thing that's very cool about 4 dimensions - it inherits the
hypericosahedron and the hyperdodecahedron from the the fact that
the icosahedron and dodecahedron happen to exist in 3 dimensions.

Similarly, the 24-cell comes from the symmetries of the tetrahedron!

Directions for Further Study

I copied the pictures of rotating Platonic solids from the Wikipedia

article on Platonic solids under the terms of the GNU Free
Documentation License. I also copied the pictures of 4d regular
polytopes from the Wikipedia articles on these polytopes, under the
terms of the relevant copyrights. These articles are a great place to get
started on understanding the Platonic solids.

For more information try Eric Weisstein's Mathworld website. He has

lots of information on Platonic solids and 4d geometry. You can rotate
the Platonic solids and 4d polytopes using your mouse!

If you have access to VRML, you can also have fun with George Hart's
Encyclopedia of Polyhedra, which has over 1000 polyhedra in it. (VRML
stands for "virtual reality modelling language", and it's available as a
plugin for most browsers.)

If you want to learn a lot about regular polytopes, read this book by the
king of geometry:

 H. S. M. Coxeter, Regular Polytopes, 3rd edition, New York, Dover

Publications, 1973.

For more on the dodecahedron, see "Tales of the dodecahedron: from

Pythagoras through Plato to Poincaré".
For more about the icosahedron, see "Some thoughts on the number
six".

For more about icosians and related marvels, see week20 of This
Week's Finds.

For more about the Platonic solids, how fool's gold fooled the Greeks
into inventing the regular dodecahedron, and highly symmetric
structures in higher dimensions, see week62, week63, week64, and
week65 of my weekly column on mathematical physics. This story
continues at a deeper level in week186 and week187.

For more on the Hurwitz integral quaternions and the mysteries of

triality in 8 dimensions, see week91.

For a deeper look at relations between different Platonic solids, and

also more stuff about the 24-cell and 600-cell, see week155.

Everything sufficiently beautiful is connected to all other beautiful

things! Follow the beauty and you will learn all the coolest stuff. The
Platonic solids are a nice place to start.

From Kepler's Mysterium Cosmographicium,

in which he modeled the orbits of the five known planets using Platonic
 solids.
The cube fits outside all those shown above.