Flips in Tilings
Thomas Fernique CNRS & Univ. Paris 13
1 Examples
2 Questions
3 Results
4 Context-sensitive flips
Examples Questions Results Context-sensitive flips
1 Examples
2 Questions
3 Results
4 Context-sensitive flips
Dimers
Consider the square or triangular grid.
Examples Questions Results Context-sensitive flips
Dimers
Adimer (domino or lozenge) is the union of two neighboor cells.
Dimers
Tilings by dimers.
Examples Questions Results Context-sensitive flips
Dimers
Aflipis a local exchange of tiles.
Dimers
Aflipis a local exchange of tiles.
Examples Questions Results Context-sensitive flips
Rectangles
Two dominoes two rectangles.
Rectangles
Tiling by two rectangles.
Examples Questions Results Context-sensitive flips
Rectangles
Flip: exchange rectangles which tile alcm(a,c)×lcm(b,d) box.
Rectangles
Flip: exchange rectangles which tile alcm(a,c)×lcm(b,d) box.
Examples Questions Results Context-sensitive flips
Rectangles
Flip bis: swap alcm(a,c)×(b+d) or (a+c)×lcm(b,d) box.
Rectangles
Flip bis: swap alcm(a,c)×(b+d) or (a+c)×lcm(b,d) box.
Examples Questions Results Context-sensitive flips
Parallelograms
n pairwise non-colinear vectors n2
parallelograms.
Parallelograms
Tiling by parallelograms (no more underlying lattice).
Examples Questions Results Context-sensitive flips
Parallelograms
Flip: exchange parallelograms which tile a hexagon.
Parallelograms
Flip: exchange parallelograms which tile a hexagon.
Examples Questions Results Context-sensitive flips
Kagome
Kagome or trihexagonal lattice.
Kagome
Tile: hexagon with two neighboor triangles (three up to rotation).
Examples Questions Results Context-sensitive flips
Kagome
Tile: hexagon with two neighboor triangles (three up to rotation).
Kagome
Flip: exchange of two triangles between two neighboor hexagons.
Examples Questions Results Context-sensitive flips
Kagome
Flip: exchange of two triangles between two neighboor hexagons.
Kagome
Different types of flips (six up to isometry).
Examples Questions Results Context-sensitive flips
Kagome
Different types of flips (six up to isometry).
T -tetraminoes
Two types of flips.
Examples Questions Results Context-sensitive flips
T -tetraminoes
Two types of flips.
3-d
Flips/trits on 3d dominoes and flips on parallelotopes.
Examples Questions Results Context-sensitive flips
3-d
Flips/trits on 3d dominoes and flips on parallelotopes.
1 Examples
2 Questions
3 Results
4 Context-sensitive flips
Examples Questions Results Context-sensitive flips
Tiling graph
Graph(region, tiles, flip):
vertices: tilings of the region;
edges: connect tilings which differ by a flip.
Properties? (connexity, diameter, eccentricity, treewidth. . . )
Limit shape
Does a tiling chosen u.a.r. satisfy a.s. some macroscopic property?
Examples Questions Results Context-sensitive flips
Mixing time
Ergodic random walkP on a graph stationnary distribution π.
Convergence measured by: d(t) := max
x∈V
1 2
X
y∈V
|Pt(x,y)−π(y)|.
Mixing time(half-life):
τmix:= min{t |d(t)≤1/4}.
The distance toπ is divided by (at least) 2 eachτmix steps. Rapid mixing: τmix logarithmic in the size of the graph.
Examples Questions Results Context-sensitive flips
Mixing time
Ergodic random walkP on a graph stationnary distribution π.
Convergence measured by:
d(t) := max
x∈V
1 2
X
y∈V
|Pt(x,y)−π(y)|.
τmix:= min{t |d(t)≤1/4}.
The distance toπ is divided by (at least) 2 eachτmix steps. Rapid mixing: τmix logarithmic in the size of the graph.
Examples Questions Results Context-sensitive flips
Mixing time
Ergodic random walkP on a graph stationnary distribution π.
Convergence measured by:
d(t) := max
x∈V
1 2
X
y∈V
|Pt(x,y)−π(y)|.
Mixing time(half-life):
τmix:= min{t |d(t)≤1/4}.
The distance toπ is divided by (at least) 2 eachτmix steps.
Rapid mixing: τmix logarithmic in the size of the graph.
Mixing time
Ergodic random walkP on a graph stationnary distribution π.
Convergence measured by:
d(t) := max
x∈V
1 2
X
y∈V
|Pt(x,y)−π(y)|.
Mixing time(half-life):
τmix:= min{t |d(t)≤1/4}.
The distance toπ is divided by (at least) 2 eachτmix steps.
Rapid mixing: τmix logarithmic in the size of the graph.
Examples Questions Results Context-sensitive flips
Motivation
Flips on tilings: model local rearrangements in materials.
Limit shape: macroscopic properties of materials.
Mixing time: practicability of the model.
1 Examples
2 Questions
3 Results
4 Context-sensitive flips
Examples Questions Results Context-sensitive flips
Connexity
For a simply connected region, the following graphs are connected:
Dimers (Thurston 1990)
Parallelograms (Kenyon 1994)
Two rectangles(R´emila 2004)
Kagome (Bodini, 2006)
T-tetraminoes(Korn-Pak 2007)
Additional conditions are necessary:
Parallelotopes(Desoutter-Destainville 2005)
3-d. dominoes (Milet-Saldanha 2017)
Examples Questions Results Context-sensitive flips
Height function
Height function: real map defined on the vertices of the tiling.
each tiling determined by its height function; the height is monotone w.r.t. flips.
Yields an intuitive lift inR3, used to prove connexity: Dimers (Thurston)
Rectangles (R´emila)
Kagome (Bodini)
T-tetraminoes(Korn-Pak)
Examples Questions Results Context-sensitive flips
Height function
Height function: real map defined on the vertices of the tiling.
Useful when
each tiling determined by its height function;
the height is monotone w.r.t. flips.
Yields an intuitive lift inR3, used to prove connexity: Dimers (Thurston)
Rectangles (R´emila)
Kagome (Bodini)
T-tetraminoes(Korn-Pak)
Height function
Height function: real map defined on the vertices of the tiling.
Useful when
each tiling determined by its height function;
the height is monotone w.r.t. flips.
Yields an intuitive lift inR3, used to prove connexity:
Dimers (Thurston)
Rectangles (R´emila)
Kagome (Bodini)
T-tetraminoes(Korn-Pak)
Examples Questions Results Context-sensitive flips
Diameter
Easy for dimers via height function.
Logarithmic in the number of tilings in any non-degenerate case.
Not necessarily a good indicator for mixing time. . .
Straightness
Straighttiling graph: any two tilings can be connected by a sequence of flips, with no flip being performed back and forth.
Holds for dimers (the height function orientates flips).
Holds for parallelogram iffn≤4 (Bodini-F.-Rao-R´emila 2011). Does it imply rapid mixing?
Examples Questions Results Context-sensitive flips
Limit shape
Well-studied for dimers(Kenyon, Okounkov, Propp. . . ). Other cases?
Limit shape
Well-studied for dimers(Kenyon, Okounkov, Propp. . . ). Other cases?
Examples Questions Results Context-sensitive flips
Limit shape
Well-studied for dimers(Kenyon, Okounkov, Propp. . . ). Other cases?
Examples Questions Results Context-sensitive flips
Mixing time
Rapid for dimers(Luby-Randall-Sinclair 1995, Randall-Tetali 1999, Wilson 2004, Caputo-Martinelli-Toninelli 2011, Laslier-Toninelli 2013).
Connexity still to be characterized in this case.
Conjectured to be rapid for 1×1 and 2×2 squares (Ugolnikova 2016). Claimed rapid for T-tetraminoes(Kayibi-Pirzada 2015).
Conjectured rapid for parallelograms whenn= 4 (Destainville 2001). Conjectured slow for parallelograms whenn= 4 (F. 2016).
Examples Questions Results Context-sensitive flips
Mixing time
Rapid for dimers(Luby-Randall-Sinclair 1995, Randall-Tetali 1999, Wilson 2004, Caputo-Martinelli-Toninelli 2011, Laslier-Toninelli 2013).
Rapid for Kagome without “fish”(Ugolnikova 2016). Connexity still to be characterized in this case.
Conjectured to be rapid for 1×1 and 2×2 squares (Ugolnikova 2016). Claimed rapid for T-tetraminoes(Kayibi-Pirzada 2015).
Conjectured rapid for parallelograms whenn= 4 (Destainville 2001). Conjectured slow for parallelograms whenn= 4 (F. 2016).
Examples Questions Results Context-sensitive flips
Mixing time
Rapid for dimers(Luby-Randall-Sinclair 1995, Randall-Tetali 1999, Wilson 2004, Caputo-Martinelli-Toninelli 2011, Laslier-Toninelli 2013).
Rapid for Kagome without “fish”(Ugolnikova 2016). Connexity still to be characterized in this case.
Conjectured to be rapid for 1×1 and 2×2 squares (Ugolnikova 2016).
Conjectured rapid for parallelograms whenn= 4 (Destainville 2001). Conjectured slow for parallelograms whenn= 4 (F. 2016).
Examples Questions Results Context-sensitive flips
Mixing time
Rapid for dimers(Luby-Randall-Sinclair 1995, Randall-Tetali 1999, Wilson 2004, Caputo-Martinelli-Toninelli 2011, Laslier-Toninelli 2013).
Rapid for Kagome without “fish”(Ugolnikova 2016). Connexity still to be characterized in this case.
Conjectured to be rapid for 1×1 and 2×2 squares (Ugolnikova 2016). Claimed rapid for T-tetraminoes(Kayibi-Pirzada 2015).
Conjectured rapid for parallelograms whenn= 4 (Destainville 2001). Conjectured slow for parallelograms whenn= 4 (F. 2016).
Examples Questions Results Context-sensitive flips
Mixing time
Rapid for dimers(Luby-Randall-Sinclair 1995, Randall-Tetali 1999, Wilson 2004, Caputo-Martinelli-Toninelli 2011, Laslier-Toninelli 2013).
Rapid for Kagome without “fish”(Ugolnikova 2016). Connexity still to be characterized in this case.
Conjectured to be rapid for 1×1 and 2×2 squares (Ugolnikova 2016). Claimed rapid for T-tetraminoes(Kayibi-Pirzada 2015).
Conjectured rapid for parallelograms whenn= 4 (Destainville 2001).
Examples Questions Results Context-sensitive flips
Mixing time
Rapid for dimers(Luby-Randall-Sinclair 1995, Randall-Tetali 1999, Wilson 2004, Caputo-Martinelli-Toninelli 2011, Laslier-Toninelli 2013).
Rapid for Kagome without “fish”(Ugolnikova 2016). Connexity still to be characterized in this case.
Conjectured to be rapid for 1×1 and 2×2 squares (Ugolnikova 2016). Claimed rapid for T-tetraminoes(Kayibi-Pirzada 2015).
Conjectured rapid for parallelograms whenn= 4 (Destainville 2001). Conjectured slow for parallelograms whenn= 4 (F. 2016).
1 Examples
2 Questions
3 Results
4 Context-sensitive flips
Examples Questions Results Context-sensitive flips
Principle
Consider the Markov chain
choose a vertex of the tiling uniformly at random,
perform a flip around it with probability P = exp(−∆E)/T, where
T is a temperatureparameter,
theenergy E of a tiling is a sum of local interactions.
This is the Metropolis-Hastings algorithm for the Gibbs distribution P(tiling) = 1
Z(T)exp
−E(tiling) T
.
It aims to model the cooling of ordered materials (quasicrystals).
Principle
Consider the Markov chain
choose a vertex of the tiling uniformly at random,
perform a flip around it with probability P = exp(−∆E)/T, where
T is a temperatureparameter,
theenergy E of a tiling is a sum of local interactions.
This is the Metropolis-Hastings algorithm for the Gibbs distribution P(tiling) = 1
Z(T)exp
−E(tiling) T
.
It aims to model the cooling of ordered materials (quasicrystals).
Examples Questions Results Context-sensitive flips
Example 1: dimer tilings
Ergodic atT >0, Θ(n2) mixing at T =∞,O(n2.5) atT = 0.
Example 1: dimer tilings
Ergodic atT >0, Θ(n2) mixing at T =∞,Θ(n2) atT = 0.
Examples Questions Results Context-sensitive flips
Example 1: dimer tilings
Ergodic atT >0, Θ(n2) mixing at T =∞,Θ(n2) atT = 0.
Example 1: dimer tilings
Ergodic atT >0, Θ(n2) mixing at T =∞,Θ(n2) atT = 0.
Examples Questions Results Context-sensitive flips
Example 1: dimer tilings
Ergodic atT >0, Θ(n2) mixing at T =∞,Θ(n2) atT = 0.
Example 1: dimer tilings
Ergodic atT >0, Θ(n2) mixing at T =∞,Θ(n2) atT = 0.
Examples Questions Results Context-sensitive flips
Example 2: Beenker tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Example 2: Beenker tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Examples Questions Results Context-sensitive flips
Example 2: Beenker tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Example 2: Beenker tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Examples Questions Results Context-sensitive flips
Example 2: Beenker tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Example 3: Penrose tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Examples Questions Results Context-sensitive flips
Example 3: Penrose tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Example 3: Penrose tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Examples Questions Results Context-sensitive flips
Example 3: Penrose tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Example 3: Penrose tilings
Ergodic atT >0,unknown mixing atT =∞,Θ(n2)at T = 0.
Flips in Tilings
Thomas Fernique CNRS & Univ. Paris 13