The Parameter Space of Apollonian Circle Packings
Summer Haag
University of Colorado Boulder
Aix Marseille Univerité Geometry Seminar
Introduction
We want to study Apollonian circle packings
what does it mean for two circle packings to be close to each other?
what type of packings can we generate?
there is a map from \(\PP^1(\C)\) to packings
we will find an alternative map that is equivalent
Apollonian Circle Packings
Apollonian Circle Packings
Apollonian Circle Packings
Apollonian Circle Packings
Apollonian Circle Packings
Apollonian Circle Packings
\([-1,\,2,\,2,\,3]\)
Types of Packings
There are bounded packings, the strip packing, half-plane packings, and full-plane packings
From Jan Holly
Where do the packings live?
Looking at packings over \(\hat{\C}=\C\cup\{\infty\}\)
This means we have circles and straight lines
Two parallel straight lines are tangent at infinity and have curvature \(0\)
We consider packings equivalent under rotations, shifts, reflections, and scaling
How do we tell these packings apart?
Starting with 3 circles fixes a packing
choose four starting circles and call that a quadruple
from here we can generate a packing
Theorem of Elizabeth and Descartes
The two red circles are called Soddy Circles
Theorem of Elizabeth and Descartes
Curvatures \(a\), \(b\), \(c\), \(d\) of four mutually tangent circles
(a Descartes Quadruple ) satisfy
\[(a + b + c + d)^2 = 2(a^2 + b^2 + c^2 + d^2).\]
Theorem of Elizabeth and Descartes
Given \(a,b,c\), there exist \(d,d'\) such that
\[d+d'=2(a+b+c)\]
So if \(a,b,c,d\in\Z\) then curvatures \(\in\Z\)
Descartes Quadruples
Two different Desacrtes Quadruples can generate the same packing
for example \((3,-2,6,7)\) and \( (3,34,6,7) \)
So we want to know which quadruples generate the same packing and how to move around the packing
Apollonian Group
Acting on a quadruple of circles
\(S_4\) (4 swaps)
\(\mathcal{A}=\la S_1, S_2, S_3, S_4 : S_i^2=1\ra < O_Q(\Z)\)
for example
\[S_1=\begin{bmatrix}
-1 & 2 & 2 & 2 \\
0 & 1 & 0 & 0 \\
0 & 0 & 1 & 0 \\
0 & 0 & 0 & 1
\end{bmatrix}\]
Apollonian Group example
Let's see the group act on \((-2,3,6,7)\)
\((-2,3,6,7)\)
\(S_1(-2,3,6,7)=(34,3,6,7)\)
\(S_2(-2,3,6,7)=(-2,19,6,7)\)
\(S_3(-2,3,6,7)=(-2,3,10,7)\)
\(S_4(-2,3,6,7)=(-2,3,6,7)\)
Apollonian Group
Acting on a quadruple of circles
\(S_4\) (4 swaps)
\(\mathcal{A}=\la S_1, S_2, S_3, S_4 : S_i^2=1\ra < O_Q(\Z)\)
\(\mathcal{A}\) is a thin group so
it has infinite index in \(O_Q(\Z)\)
Zariski dense in \(O_Q(\Z)\)
Space of ACPs
We want to build a space who's points correspond to ACPs with a choice of circle
Let's start by fixing a circle in a fixed packing and looking at those quadruples
Why fix a circle?
For integral packings, if we fix an integer what packings contain it?
We want to know which Descartes quadruples can have this integer
So if we are in a packing and we fix a circle let's consider the curvatures tangent to the mother circle
\(n\)-quadruple
quadruples are ordered so we consider the group generated by
\(\mathcal{A}\) and permutations
an \(n\)-quadruple is a Descartes quadruple \((n,a,b,c)\)
two \(n\)-quadruples are equivalent if they are in the same orbit of
\(\la (23),(24),S_4\ra\)
Equivalent \(n\)-quadruples: \( (3,6,7,-2)\), \( (3,10,42,7) \), \( (3,10,7,42)\), \( (3,42,10,7)\), \( (3,7,42,10)\),
\( (3,6,7,34)\), \( (3,15,10,-2)\)
Descartes quadruples to BQF quadruples
Theorem: (Sarnack, Graham-Lagarias-Mallows-Wilks-Yan) We have the bijection
\[
\begin{Bmatrix}
\text{curvatures tangent to a fixed}\\
\text{circle with curvature }n
\end{Bmatrix}
\;\longrightarrow\;
\begin{Bmatrix}
Q_n(x,y)-n:\;\gcd(x,y)=1
\end{Bmatrix}
\]
This is true regardless of integrality
Descartes quadruples to BQFs
Explicitly we have the map:
\[
\phi:\begin{Bmatrix}
n\text{-quadruples up to}\\
\text{scaling by }\R^+
\end{Bmatrix}
\longrightarrow
\begin{Bmatrix}
\text{BQF quadruples up to}\\
\text{scaling by }\R^+
\end{Bmatrix}
\]
\[\phi\bigl((n,a,b,c)\bigr)=[n,\;n+a,\;n+a+b-c,\;n+b]\]
\[\phi^{-1}([n,A,B,C])=[n,\;A-n,\;C-n,\;A+C-B-n]\]
Theorem: (Rickards)
\(\phi\bigl(\la(23),(24),S_4\ra\cdot(n,a,b,c)\bigr)
=\PGL_2(\Z)\cdot\phi\bigl((n,a,b,c)\bigr)\)
Essentially, we can decide if two \(n\)-quadruples are equivalent if their BQFs are
This is nice because in the integral case we can explicitly say which packings contain a fixed integer
Binary Quadratic forms
Def: A binary quadratic form (BQF) is
\[\begin{align}
Q(x,y) &= Ax^2 + Bxy + Cy^2 \\
&= \begin{pmatrix}x&y\end{pmatrix}
\begin{pmatrix}A&B/2\\B/2&C\end{pmatrix}
\begin{pmatrix}x\\y\end{pmatrix} \\
&= [A,B,C]
\end{align}\]
with \(x,y\in\Z\) coprime and discriminant \(\Delta=B^2-4AC\)
Def: \(\PGL_2(\Z)\) acts on BQFs by
\[\begin{align}
MQ(x,y) &= Q(ax+by,\,cx+dy) \\
&= \begin{pmatrix}x&y\end{pmatrix}
M^T\begin{pmatrix}A&B/2\\B/2&C\end{pmatrix}M
\begin{pmatrix}x\\y\end{pmatrix}
\end{align}\]
where \(M=\begin{pmatrix}a&b\\c&d\end{pmatrix}\)
BQF actions
Def: Two BQFs \(Q,Q'\) are equivalent if there is some
\(M\in\PGL_2(\Z)\) such that \(MQ(x,y)=Q'(x,y)\)
Def: A BQF quadruple is of the form \([n,A,B,C]\)
where \(\Delta=-4n^2\)
BQFs
We know more about equivalences of BQFs than of \(n\)-quadruples
So let's use things we know about BQFs to learn about ACPs and quadruples
BQF quadruples to \(\PP^1(\C)\)
We have a bijection! (Rickards)
\[
\rho:\begin{Bmatrix}
\text{BQF quadruples up to}\\
\text{scaling by }\R^+
\end{Bmatrix}
\longrightarrow \PP^1(\C)
\]
\[
\rho([n,A,B,C])=\begin{cases}
\dfrac{B}{2A}+\dfrac{n}{A}\,i & \text{if }A\neq 0\\[6pt]
\infty & \text{if }A=0
\end{cases}
\]
\[
\rho^{-1}(x+iy)=\begin{cases}
[0,0,0,1] & x+iy=\infty\\
[y,1,-2x,x^2+y^2] & \text{else}
\end{cases}
\]
This means we can send an \(n\)-quadruple to a point in \(\PP^1(\C)\) which is the map we want!
\(\PGL_2(\Z)\) with \(\rho\)
Def: \(M\in\PGL_2(\Z)\) acts on \(z\in\PP^1(\C)\) by
\[
M\cdot z:=\begin{cases}
Mz & \text{if }\det(M)=1\\
M\bar z & \text{if }\det(M)=-1
\end{cases}
\]
Theorem: (Rickards) Let \(Q\) be a BQF quadruple and \(M\in\PGL_2(\Z)\).
Then \(\rho(MQ)=M^{-1}\rho(Q)\).
What do we have now?
For an \(n\)-quadruple, \((n,a,b,c) \), equivalence class we have a BQF quadruple, \([n,A,B,C]\), equivalence class
the orbit of \(\la (23),(24),S_4\ra\) on \((n,a,b,c) \) is equivalent to the orbit of \(\PGL_2(\Z)\) on \([n,A,B,C]\)
For a BQF quadruple, \([n,A,B,C]\), equivalence class we have a point \(z_0\) in \(\PP^1(\C)\)
the orbit of \(\PGL_2(\Z)\) on \([n,A,B,C]\) is equivalent to the orbit of \(\PGL_2(\Z)\) on \(z_0\)
So we have a way of going from \(n\)-quadruples to points in \(\PP^1(\C)\) by composing these maps
Descartes quadruples and \(\PP^1(\C)\)
Last bijection!
\[
\rho\circ\phi:\begin{Bmatrix}
n\text{-quadruples up to}\\
\text{scaling by }\R^+
\end{Bmatrix}
\longrightarrow \PP^1(\C)
\]
\[
\rho\circ\phi\bigl((n,a,b,c)\bigr)
=\frac{n+a+b-c}{2(n+a)}+\frac{n}{n+a}\,i
\]
\[
\phi^{-1}\circ\rho^{-1}(x+iy)
=\bigl(y,\;1-y,\;x^2+y^2-y,\;1+x+y^2+2x-y\bigr)
\]
How can we separate this out?
We now have \(\PGL_2(\Z)\)-equivalent points mapping to equivalent \(n\)-quadruples
Let's try to draw the sets that seperate out the types of packings, i.e. bounded, strip, half-plane, and full-plane
James Rickards
Parameter Space
\((0,0,2,2)\) — James Rickards
Theorem: (Holly, Kocik, Rickards) The space of circle packings is a circle packing! Say \(\phi^{-1}\circ\rho^{-1}(z)=q\).
Then \(q\) generates a
bounded packing iff \(z\) is in the interior of a circle
half-plane packing iff \(z\) is on the boundary but not a tangent point
strip packing iff \(z\) is a tangency point of two circles
full-plane packing iff \(z\) is not in or on any circle
What does this look like?
Let's make something that moves around the parameter space
and gives you the packing
Parameter Space Shader
Move cursor over the left panel (parameter space) to see the corresponding packing on the right
New Theorem
Let's take a different perspective on this space
The map from the parameter space to a circle packing is explicit in a surprising way: it's an inversion!
Geometrically, fix a packing in \( \hat\C\) and invert in different points to get different packings and all packings arise from this!
Let's prove this...
Möbius transformations and Circles
We know
\[
\begin{pmatrix}\alpha&\beta\\\gamma&\delta\end{pmatrix}\hat{\R}
\;\longrightarrow\;
\begin{cases}
\text{curvature} = 2\,\Im(\bar\gamma\delta)
= i(\gamma\bar\delta - \bar\gamma\delta)\\[4pt]
\text{curvature-center} = i(\alpha\bar\delta - \bar\gamma\beta)
\end{cases}
\]
Take the strip packing \((0,0,2,2)\) — we can represent these 4 circles
by Möbius transformations
James Rickards
Parameter Space and circle inversion
Take a point in the parameter space \(x+iy=z_0\)
Invert our quadruple \((0,0,2,2)\) through this point using the transformation
\(\begin{pmatrix}0&1\\1&-z_0\end{pmatrix}\)
Computing this new quadruple gives
\[2\bigl(y,\;1-y,\;x^2+y^2-y,\;1+x+y^2+2x-y\bigr)\]
Parameter Space and Inversion Shader
Move cursor over the left panel — middle shows resulting packing, right shows inversion of the strip
New Map
This complicated bijection
\[
\PP^1(\C)
\;\longrightarrow\;
\begin{Bmatrix}
\text{BQF quadruples up to}\\
\text{scaling by }\R^+
\end{Bmatrix}
\;\longrightarrow\;
\begin{Bmatrix}
n\text{-quadruples up to}\\
\text{scaling by }\R^+
\end{Bmatrix}
\]
is actually just inversion \(\displaystyle z\mapsto\frac{1}{z-z_0}\)
applied to the strip packing!
Thank you!