Asymptotics and Universal Edge Fluctuations with Applications to Lozenge Tiling Models
ERIK DUSE
Doctoral Thesis Stockholm, Sweden 2015
ISRN KTH/MAT/A–15/12-SE ISBN 978-91-7595-732-6
100 44 Stockholm SWEDEN Akademisk avhandling som med tillstånd av Kungl Tekniska högskolan framlägges till offentlig granskning för avläggande av teknologie doktorsexamen i matematik veckodag den 4 december 2015 kl 13.00 i D3, Kungl Tekniska högskolan, Lindstedts- vägen 5, Stockholm.
Erik Duse, 2015c
Tryck: Universitetsservice US AB
Abstract
This thesis concerns uniformly random discrete interlacing particle sys- tems and their connections to certain random lozenge tiling models. In par- ticular it contains the first derivation of a relatively unknown universal scaling limit, which we call the Cusp-Airy process, of certain lozenge tiling models at a cusp point. In addition it contains a characterization of the geometry of the macroscopic behavior of uniformly random discrete interlaced parti- cle systems that, although not complete, shows many new and interesting features.
Sammanfattning
Denna avhandling behandlar likformigt slumpmässiga system av samman- flätade partiklar (eng: interlacing particles) och deras koppling till vissa rom- biska tesselerings modeller. Den innehåller i synnerhet den första härledning- en av en relativt okänd universell skalningsgräns, som vi kallar Cusp-Airy processen, för vissa rombiska tesselerings modeller vid en spetspunkt. Den innehåller dessutom en karakterisering av geometrin för det makroskopiska beteendet hos likformigt slumpmässiga system av sammanflätade partiklar, som trots att den inte är fullständig, visar på nya och intressanta egenskaper.
Contents v
Acknowledgements vii
Part I: Introduction and Summary
1 Introduction 2
1.1 Random Lozenge Tilings . . . . 2 1.2 Cut-Corner Hexagon Dimer Model and Related Models . . . . 4 1.3 Discrete Interlacing Systems . . . . 10 1.4 Discrete Orthogonal Polynomial Ensembles and Equilibrium Measures 38 1.5 Relation To Continuous Interlacing Model . . . . 49
2 Summary of Results 53
2.1 Paper A . . . . 53 2.2 Paper B . . . . 53 2.3 Paper C . . . . 53
References 55
v
First and foremost I would like to express my gratitude to my scientific advisor Kurt Johansson for all his support and encouragement during these past years. In particular I thank him for his generosity in sharing his mathematical knowledge and his friendship. I would also like to thank Anthony Metcalfe for a fruitful cooperation. In addition I would like to thank Maurice Duits and Sunil Chhita for many useful and interesting discussions regarding my research. I am also grateful to Professor Richard Kenyon at Brown University for letting me use his figure of a simulation of a random tiling of the cut-corner hexagon, figure 1 in the introduction.
My graduate studies were funded by the grant KAW 2010.0063 from the Knut and Alice Wallenberg Foundation. For this I am very grateful.
I would like to thank all the friends I have gained at the department of mathe- matics at KTH during my years as a graduate student. In particular, I would like to thank Martin Strömqvist, Joel Andersson, Andreas Minne, Gaultier Lambert, Mauriusz Hynek, Christopher Svedberg, Oscar Forsman, Katharina Heinrich, Se- bastian Öberg, Samuel Holmin, André Laestedius, Gustav Sændén Ståhl and Antti Haimi for all interesting discussions over the years.
Last but not least, I would like to thank my family for all support during these years. In particular I would like to express my gratitude and love to my fiancée Frida, my parents Alfred and Maria and my grandmother Margareta. This thesis would not have been made possible without you.
Finally I would like to dedicate this thesis to my late grandfather Erik Öhlin, who served as a great inspiration to me during my childhood.
vii
1
1.1 Random Lozenge Tilings
Figure 1.1: A simulation of a random lozenge tiling of the cut-corner hexagon model.
The Cut-Corner Hexagon Dimer Model
This thesis was born out of an attempt to understand a special feature of the cut- corner hexagon dimer model. More precisely, consider a regular hexagon with unit side length, and remove from one corner a rhombi of side length 1{2. We call this polygon the cut-corner hexagon. See figure 1.2 below.
Figure 1.2: The cut-corner regular hexagon.
Consider tiling the cut-corner hexagon by the rhombi forming the faces of a cube under the projection in the p1, 1, 1q direction with side lengths 1{p2nq. The rhombi, or lozenges, are shown in figure 1.3.
Y B R
Figure 1.3: Three types of lozenges with sides of length 1.
Now, consider all possible such tessellations. Put the uniform probability distri- bution on the set of all tessellations. One may now ask many interesting questions about this model. For example, one may ask what a “typical” tessellation look like when n is very large, if there is any such thing. Indeed, a simulation of a typical tessellation is shown in figure 1 in [KO07] and reprinted in this introduction as figure 1.1 by permission of Professor Richard Kenyon. From, this figure one may immediately observe some special features. To start with one observes a brick-like pattern of only one type of lozenge in each of the corners. These frozen regions are interrupted by a curve. Inside the region of this curve one observes all types of lozenges. One also observes that the curve touches the sides of the polygon tan- gentially, see figure 1.1. By letting n become large, this curve will look more and
more like a smooth curve. In particular, this smooth curve will look very similar to a cardioid curve, again see figure 1.1. Moreover, this curve has a singularity in the form of a cusp. The starting point of this thesis was to attempt to understand the local microscopic behavior of the stochastic process of the lozenges in the vicinity of the cusp as n Ñ 8. Before discussing how this question was resolved, we will in the next sections first describe more general lozenge tiling models and their relation to interlaced particle systems.
1.2 Cut-Corner Hexagon Dimer Model and Related Models
Decomposition of Polygons into Interlacing Systems
Consider first a tessellation of a regular hexagon. An example of such is shown in figure 1.4. In figure 1.5 we also depict the “frozen regions” and the “frozen boundary” separating the frozen region from from the “liquid region”. Also, the asymptotic shape of the frozen boundary is shown.
1
1 1 1
1
1
Y B R
Figure 1.4: Left: A regular hexagon with sides of length 1.
Middle: The three different types of lozenges with sides of length 1{n.
Right: An example tiling when n “ 4.
B R Y B R
Y
1
1 1 1
1
1
Figure 1.5: Left: The frozen boundary of the example tiling of figure 1.4.
Right: The asymptotic shape of the frozen boundary of a “typical random tiling”
as n Ñ 8.
We will now restrict our attention to the configuration of yellow tiles. We see that we may encode the configuration of yellow tiles as interlacing systems of particles. In in figure 1.6 this is done in two different ways. In the figure to the left in figure 1.6, we encode the configuration of yellow tiles as two interlacing systems. One interlacing system between row 1 and row 4, and one interlacing system between row 7 and row 4, being glued together along their common row, row 4. Moreover, since we pick tessellations of the regular hexagon uniformly at random, the configuration of tiles at row 4 is a random configuration. On the other hand, in the figure to the right in figure 1.6 we encode the configuration of yellow tiles as one interlacing system by adding virtual particles on the side of the polygon.
Now, the configuration of yellow tiles at row 8, the “top line”, is deterministic as opposed to the configuration of tiles on row 4.
We may now note that the configuration of yellow tiles entirely fixes the config- urations of the other tiles, the red and the blue tiles. This can be seen as follows.
In between every row of yellow tiles we may we have a row of red and blue tiles.
Firstly, we see that the position of the yellow tile on the first row fixes the positions of the tiles on first row of red and blue tiles. Secondly, we see that the position of yellow tiles on the first and the second row fixes the position of red and blue tiles on the the second row of blue and red tiles. The process may be continued until we reach the “top row” of the first interlacing system, row 4, and the position of the tiles on row 3 and row 4 have determined the position of the red and blue tiles on the fourth row of the rows of red and blue tiles. We have now fixed the position of all the tiles in the lower interlacing system. We may now repeat the process in the upper interlacing system, using that the position of the yellow tile on row 7 fixes position of red and blue tiles on row 8 and so on. This completes the tessellation of the regular hexagon, given the position of the yellow tiles. In fact we notice that the choice of considering yellow tiles was arbitrary, we may in fact equally well have
chosen to consider the configurations of red or the blue tiles instead.
row 1 row 2 row 3 row 4 row 5 row 6 row 7 row 8
Figure 1.6: Left: Equivalent interlaced particle configuration of the example tiling of figure 1.4.
Right: Equivalent interlaced particle configuration with added deterministic lozenges/particles. The unfilled circles represent the deterministic particles.
We will now be more precise on how we encode the positions of the yellow tiles as an interlaced particle system. Let yiprq denote the position of the i:th particle on the r:th row and let βipnq:“ yipnqdenote the position of the particles on the top line, indicated by unfilled circles inside the tiles. Then the particles on row r ` 1 will interlace with the particles on row r according to
ypr`1q1 ą y1prqą y2pr`1qą y2prq... ą yrprqą yr`1pr`1q,
for every r “ 1, ..., n ´ 1.
We may now try to repeat the idea of encoding lozenge tessellations of the regular cut-corner hexagon by interlacing systems. This can be done in the three different ways depicted in figure (1.7).
Figure 1.7: The picture to the left depicts a decomposition of the cut-corner hexagon into two interlacing regions of yellow tiles. The picture in the middle depicts a decomposition of the cut-corner hexagon into two interlacing regions of red tiles.
The picture to the right depicts a decomposition of the cut-corner hexagon into two interlacing regions of blue tiles
In particular, the methods developed in this thesis will apply to those polygons that can be decomposed into two interlacing regions for at least one of the three different types of tiles. For a more general decomposition of such a polygon see figure 1.10 in the next section.
It will be convenient when considering interlacing system to make a coordinate transformation according to figure 1.8. For more details see section 1.4 in [DM15a]
and section 2.1 in [Pet14] .
Figure 1.8: Coordinate transformation of lozenge tiles.
After the coordinate transformation, figure 1.4 becomes figure 1.9. Furthermore the interlacing condition between row r ` 1 and row r has changed into
ypr`1q1 ě y1prqą y2pr`1qě y2prq... ě yrprqą yr`1pr`1q.
Figure 1.9: An example tiling and its equivalent interlaced particle configuration after the coordinate transformation. The unfilled circles represent the deterministic lozenges/particles.
Discrete Orthogonal Polynomial Ensembles
As was described in the previous section we are interested in those tiling models that can be decomposed into two regions, such that after possibly adding virtual particles, these regions become interlacing regions of the type describe in section 1.2, glued together along a common line as depicted in figure 1.10.
Interlacing direction T1
T2
Figure 1.10: Decomposition of a polygon into two interlacing regions T1 and T2 glued together along the thick black line. The blue dots indicate the positions of virtual particles/tiles and the black dots indicate the positions of ordinary parti- cles/tiles.
Recall that the number of interlacing configurations with a given top line con-
figuration py1, y2, ...., yNq P ZN is given by Weyl’s dimension formula in [Wey39] for the irreducible characters of the unitary group U pnq,
N7py1, ..., yNq “ ś
1ďiăjďN|yi´ yj| ś
1ďiăjďN|i ´ j| :“ 1 2CN
ź
1ďi,jďN i‰j
|yi´ yj|. (1.2.1)
Let py1, y2, ...., yNq be the positions of the particles/tiles on the intersecting thick black line as in figure 1.10. Let V be the index set for the virtual or frozen particles and let F be the index set for the free particles, so that |V| ` |F | “ N and yi
is a virtual particle if i P V and free otherwise. Assume that |F | “ n, and let g : t1, ...., nu Ñ F be a set bijection such that xi :“ ygpiq, and x1ă x2ă ... ă xn. Furthermore, the virtual particles are densely packed, which implies that they vill form wedge shaped frozen regions. However, the fact that the two interlacing regions T1 and T2 need not be symmetrical implies that we need not have frozen regions on both sides of the intersecting black line. Let VL Ď V be the index set of those virtual particles such that they form a frozen region to the left, and let VRĎ V be the index set of those virtual particles such that they form a frozen region to the right. Then by (1.2.1), the number of interlacing configuration with a given fixed configuration of free particles at positions px1, ..., xnq is given by
N7px1, ..., xnq “ 1 2CN
ź
1ďiăjďN
|yi´ yj|
“ 1
2CN
ź
i,jPF i‰j
|yi´ yj|2 ź
jPViPFL
|yi´ yj| ź
jPViPFR
|yi´ yj| ź
i,jPVL
i‰j
|yi´ yj| ź
i,jPVR
i‰j
|yi´ yj|
“ 1
2CN
ź
1ďk,lďn k‰l
|xk´ xl|2
n
ź
k“1
ź
jPVL
|xk´ yj|
n
ź
k“1
ź
jPVR
|xk´ yj| ź
i,jPVL
i‰j
|yi´ yj| ź
i,jPVR
i‰j
|yi´ yj|.
Now, consider the set of all possible lozenge tessellations of the the original polygon.
One easily sees that each such tessellation is in a bijective correspondence with two interlacing configurations on T1and T2with the same configuration of free particles xpnq“ px1, x2, ..., xnq on their common top line. In particular, xiP nΣ X Z for each i “ 1, ..., n, where Σ is a finite union of intervals defined by the tiling model. Let Cn denote the set of all configurations of free particles. Consider the set of all lozenge tessellations of the polygon with uniform distribution. Then this induces a probability distribution on Cn, given by
ppnqrpx1, ...., xnqs “ Prparticles at positions x1, x2, ..., xns “ N7px1, ..., xnq ÿ
px1,...,xnqPCn
N7px1, ..., xnq .
Let
wnpxq :“ ź
jPVL
|x ´ yj| ź
jPVR
|x ´ yj|.
Then
ppnqrpx1, ...., xnqs “ 1 Zn
ź
1ďiăjďn
pxi´ xjq2
n
ź
i“1
wnpxiq, (1.2.2)
where Zn“ř
px1,...,xnqPCn
ś
1ďiăjďnpxi´ xjq2śn
i“1wnpxiq. Associated with a par- ticular weight function wnpxq is a class of discrete orthogonal polynomials tpn,kpxquk
defined according to
ÿ
xiPnΣXZ
pn,kpxqpn,lpxqwnpxq “ δkl. (1.2.3)
Such particle processes are called discrete orthogonal polynomial ensembles, DOPE, and have been studied e.g. in [BKMM07]. In particular if one consider the random empirical measure µn “ 1nřn
i“1δxi{n, then µn á µλV in probability, where the measure µλV P Mλ1pΣq is the unique solution of the constrained variational problem
min
νPMλ1pΣq
tIVrνsu “ min
νPMλ1pΣq
"ˆ
ΣˆΣ
log |x ´ y|´1dνpxqdνpyq ` ˆ
Σ
V pxqdνpxq
* ,
(1.2.4) where V pxq “ limnÑ8´n´1logpwnpxqq. Here, Mλ1pΣq is the set of positive Borel measures ν, such that supppνq Ă Σ, }ν} “ 1 and ν ď λ, where λ is the Lebesgue measure. The constrained variational problem is discussed in more detail in section 1.4.
Again we emphasize that in these models we have a random empirical top line measure. We will not discuss the issues arising from this fact further in introduc- tion, but refer the interested reader to [DJM15]. For more on discrete orthogonal polynomial ensembles see [BKMM07] and [Fer].
1.3 Discrete Interlacing Systems
Discrete Interlacing Systems and Determinantal Point Processes We saw in the previous section that the study of certain random lozenge tiling models with uniform probability could be reduced to the study of certain discrete interlacing models. In this section we therefore give a more careful introduction to these systems. A discrete Gelfand-Tsetlin pattern of depth n is an n-tuple, denoted pyp1q, yp2q, . . . , ypnqq P Z ˆ Z2ˆ ¨ ¨ ¨ ˆ Zn, which satisfies the interlacing constraint
y1pr`1q ě yprq1 ą ypr`1q2 ě yprq2 ą ¨ ¨ ¨ ě yrprq ą yr`1pr`1q,
denoted ypr`1qą yprq, for all r P t1, . . . , n ´ 1u. For each n ě 1, fix xpnqP Zn with xpnq1 ą xpnq2 ą ¨ ¨ ¨ ą xpnqn and apnqn “ xpnqn and bpnqn “ xpnq1 , and consider the following
probability measure on the set of patterns of depth n:
νnrpyp1q, . . . , ypnqqs :“ 1 Zn ¨
"
1 ; when xpnq“ ypnqą ypn´1qą ¨ ¨ ¨ ą yp1q, 0 ; otherwise,
where Zn ą 0 is a normalisation constant. This can equivalently be considered as a measure on configurations of interlaced particles in Z ˆ t1, . . . , nu by placing a particle at position pu, rq P Zˆt1, . . . , nu whenever u is an element of yprq. Thus, νn
is the uniform probability measure on the set of all such interlaced configurations with the particles on the top row in the deterministic positions defined by xpnq. Also note that due to the interlacing constraint, the interlacing particle system is contained inside the polygon nPn, where
Pn“ tpχ, ηq P R2: 0 ď η ď 1, χ ` η ´ 1 ě apnqn , χ ď bpnqn u (1.3.1) We now construct a related probability space, the determinantal structure of which is more convenient to examine. Consider all tuples, pzp1q, . . . , zpn´1qq P Znˆ
¨ ¨ ¨ ˆ Zn with
zpr`1q1 ě zprq1 ą zpr`1q2 ě z2prq ą ¨ ¨ ¨ ą zpr`1qn ě znprq,
for all r, also denoted zpr`1q ą zprq. Fix zp0q :“ pxn` n ´ 1, . . . , xn` 1, xnq and define the following probability measure on the set of all such pn ´ 1q-tuples:
νn1rpzp1q, . . . , zpn´1qqs :“ 1 Zn1 ¨
"
1 ; when xpnqą zpn´1qą ¨ ¨ ¨ ą zp1qą zp0q, 0 ; otherwise,
(1.3.2) where Zn1 ą 0 is a normalisation constant.
Consider the relationship between the spaces. First note that, whenever xpnq“ ypnqą ypn´1qą ¨ ¨ ¨ ą yp1q for some pyp1q, yp2q, . . . , ypnqq P Z ˆ Z2ˆ ¨ ¨ ¨ ˆ Zn, then
x1ě y1prqą ¨ ¨ ¨ ą yprqr ą xn` n ´ r ´ 1,
for all r ď n. Whenever x ą zpn´1qą ¨ ¨ ¨ ą zp1qą zp0q for some pzp1q, . . . , zpn´1qq P Znˆ Znˆ ¨ ¨ ¨ ˆ Zn,
x1ě zprq1 ą ¨ ¨ ¨ ą zrprqą zr`1prq ą zr`2prq ą ¨ ¨ ¨ ą zprqn (1.3.3)
=
xn`n´r´1
=
xn`n´r´2 . . . =
xn
for all r ď n ´ 1. We refer to zprq1 , . . . , zprqr as the free particles of zprq, and zprqr`1, . . . , znprq as the deterministic particles. Note the natural bijection between tpyp1q, . . . , ypnqq P Z ˆ Z2 ˆ ¨ ¨ ¨ ˆ Zn : x “ ypnq ą ypn´1q ą ¨ ¨ ¨ ą yp1qu and tpzp1q, . . . , zpn´1qq P Zn ˆ Zn ˆ ¨ ¨ ¨ ˆ Zn : x ą zpn´1q ą ¨ ¨ ¨ ą zp1q ą zp0qu: re- move ypnq from each n-tuple pyp1q, . . . , ypnqq and map the remaining components, yprq “ py1prq, . . . , yrprqq for each r ď n ´ 1, individually as,
yprqÞÑ py1prq, . . . , yrprq, xn` n ´ r ´ 1, xn` n ´ r ´ 2, . . . , xnq.
The measure νn1 is induced by the measure νn under this bijective map. The prob- abilistic structure of particles in the first space (measure νn) is therefore identical to the probabilistic structure of the free particles in the second space (measure νn1).
From now on we restrict to the second space.
A more convenient expression for νn1 can be obtained from the work of Warren, [War07]:
det
” 1zpr`1q
j ězprqi
ın i,j“1
“
"
1 ; when zpr`1qą zprq, 0 ; otherwise,
for all r. Equation (1.3.2) thus gives,
νn1rpzp1q, . . . , zpn´1qqs “ 1 Zn1
n´1ź
r“0
det
”
φr,r`1pziprq, zjpr`1qq ın
i,j“1, (1.3.4) where zpnq:“ x, and
φr,r`1pu, vq :“ 1věu,
for all r and u, v P Z. We may define coupling matrices Ar,r`1“ Ar,r`1pzprq, zpr`1qq, connecting configurations zprq on row r with configurations zpr`1q on row r ` 1 by letting
pAr,r`1pzprq, zpr`1qqqij :“ φr,r`1pziprq, zjpr`1qq.
Using the coupling matrices we see that (1.3.4) can be written as
νn1rpzp1q, . . . , zpn´1qqs “ 1 Zn1 det
„n´1
ź
r“0
Ar,r`1pzprq, zpr`1qq
. (1.3.5)
Note, each pzp1q, . . . , zpn´1qq P Znˆ Znˆ ¨ ¨ ¨ ˆ Zn can be equivalently considered as a configuration of particles in Z ˆ t1, . . . , n ´ 1u by placing a particle at position pu, rq P Z ˆ t1, . . . , n ´ 1u whenever u is an element of zprq. The measure νn1 in equation (1.3.5) therefore defines a random point process on configurations of particles in Z ˆ t1, . . . , n ´ 1u. The Eynard-Mehta theorem, see Proposition 2.13 of Johansson, [Joh06], proves that this process is determinantal with correlation kernel,
Knppu, rq, pv, sqq “ ˜Knppu, rq, pv, sqq ´ φr,spu, vq, (1.3.6) for all r, s P t1, . . . , n ´ 1u and u, v P Z, where
K˜nppu, rq, pv, sqq :“
n
ÿ
k,l“1
φr,npu, zkpnqqpA´1qklφ0,spzlp0q, vq,
and
φr,spu, vq :“ 0 when s ď r,
φr,spu, vq :“ 1věuwhen s “ r ` 1, φr,spu, vq :“ ÿ
z1,...,zs´r´1
φr,r`1pu, z1qφr`1,r`2pz1, z2q ¨ ¨ ¨ φs´1,spzs´r´1, vq when s ą r ` 1,
A P Cnˆnwith Akl:“ φ0,npzkp0q, zpnql q for all k, l.
Note that, for all r, s P t1, . . . , nu and u, v P Z, one may in fact compute φr,spu, vq to get
φr,spu, vq “ 1sąrhv´upp1qs´rq,
for all r, s P t1, . . . , nu and u, v P Z, and where hk denotes the homogeneous sym- metric polynomials and where by convention hk :“ 0 whenever k ă 0. Using the special form of φr,spu, vq, we where in [DM15a] able to compute ˜Knppu, rq, pv, sqq, and to derive an alternative form for φr,spu, vq. We got
K˜nppu, rq, pv, sqq “ pn ´ sq!
pn ´ r ´ 1q!
n
ÿ
k“1 v
ÿ
l“v´n`s
1xkěu
śu´1
j“u`r´n`1pxk´ jq śv
j“v´n`s,j‰lpl ´ jq ź
i‰k
ˆ l ´ xi
xk´ xi
˙ ,
(1.3.7) and
φr,spu, vq “ 1věu pn ´ sq!
pn ´ r ´ 1q!
n
ÿ
k“1 v
ÿ
l“v´n`s
śu´1
j“u`r´n`1pxk´ jq śv
j“v´n`s,j‰lpl ´ jq ź
i‰k
ˆ l ´ xi
xk´ xi
˙ .
It is interesting and surprising to note that both ˜Knppu, rq, pv, sqq and φr,spu, vq have the same form. Also note that the fixed top row and the interlacing constraint implies that it is sufficient to restrict to those positions, pu, rq, pv, sq P Zˆt1, . . . , n´
1u, with u ě xpnqn ` n ´ v and v ě xpnqn ` n ´ s. We finally note that in terms of lozenge tilings as depicted in figure 1.9, the correlation kernel Kn is the correlation kernel for the positions of the yellow tiles. When we want to emphasize this fact we will sometimes write KnY. In [DJM15] we showed how to get correlation kernels of the other tiles through the particle transformations
KnRppu, rq, pv, sqq “ ´KnYppu, rq, pv, s ´ 1qq (1.3.8) KnBppu, rq, pv, sqq “ KnYppu, rq, pv ` 1, s ´ 1qq. (1.3.9) Here, KnRis a correlation kernel for the red tiles and KnBis a correlation kernel for the blue tiles.
At this point it is advantageous to pause for a moment and reflect upon the fact why we where able to compute the correlation kernel Knppu, rq, pv, sqq. We see that it depended on two critical steps:
• We were able to compute the n-fold matrix product A and get ‘nice’ formulas for the matrix elements pAqij.
• We were able to compute the inverse matrix A´1, which by Cramer’s rule reduces to computing determinants. The fact that we could compute the determinants in turn depended critically on the form of the matrix elements Aij.
For more complicated determinantal point processes, neither the first nor the second step is usually feasible. One could ask why the uniformly random discrete interlacing model has this integrable structure. We will not give a precise answer to this question in the thesis but we note that interlaced particle systems fit within the broader framework of Schur process. See for example [OR03] and [Bor11]. We also note that there is a close connection between random interlaced particles and Markov chains, see [BF15].
Asymptotic Limit Shapes
What do we mean by asymptotic limit shapes of discrete interlaced particle sys- tems? In analogy with statistical physics we may view the asymptotic limit shapes as the boundaries of different types of “phases”, i.e., microscopic behavior of the underlying systems as n Ñ 8. The phases that can appear are called liquid or solid state. Of course the limit n Ñ 8 is an idealization of a true system where n is very large. Therefore, the asymptotic limit shape only exists in an idealized sense.
For many models from statistical physics it turns out that when one consider various scaling limits of the microscopic system, i.e. one zoom in on the stochastic system at a certain scale, one gets universal scaling limits. By universal scaling limits we mean that the limit does not depend on the particular details of the original statistical model and that the same limits occur in many different models.
For a physically realistic system the limit n Ñ 8 of the original microscopic system can then be approximated by the universal scaling limit in much the same way as the the sum of n identically distributed independent random variables, rescaled by one over the square root of n, can be approximated by the normal distribution.
Which universal scaling limit one gets will depend on where one is looking in the system. Since these limits can change in a discontinuous way depending on where one is looking one expects to see different phases. In particular one expects that for n large enough the microscopic behavior of system changes in an abrupt way as one crosses the asymptotic boundary between different types of phases.
Now how is the asymptotic limit shape of uniformly discrete interlaced patterns determined? In what follows we will change notation somewhat. Change xpnq “ px1, ...., xnq Ñ βpnq “ pβ1, ..., βnq and ppu, rq, pv, sqq Ñ ppx1, y1q, px2, y2qq. Since this system is determinantal, the correlation functions are given by
ρppx1, y1q, px2, y2q, ..., pxm, ymqq :“ Prparticles at positions px1, y1q, px2, y2q, ..., pxm, ymqs
“ detrKnppxi, yiq, pxj, yjqqsmi,j“1. (1.3.10) Therefore, various scaling limits of (1.3.10) (with m fixed) are entirely determined by the correlation kernel Knppxi, yiq, pxj, yjqq. To analyze the scaling limits, in
[DJM15], we gave an integral representation (1.3.11) below of the correlation kernel Kn based on the formula (1.3.7). This reduces the problem of studying limits of the kernel (1.3.7) to the problem of determining the asymptotics of double contour integrals. By the method of steepest descent in complex analysis, this problem in turn reduces to determining the roots of certain functions fn,11 pwq, fn,21 pzq. Under certain mild assumptions the function fn,i for i “ 1, 2 converges to an asymptotic function f . Therefore the problem of determining the limit shapes of discrete interlacing systems can be put in a bijective correspondence to a certain asymptotic function f . This function will of course depend on the coordinates of the position of where one is taking this limit and the weak limit of the sequence of point masses
1 n
n
ÿ
i“1
δβpnq i
to be defined below. The double contour integral representation of Kn
given in [DJM15] is Knppx1, y1q, px2, y2qq “ 1x1ăx2
pn ´ y1q!
pn ´ y2´ 1q!
1 p2πiq2
˛
Γn
dz
˛
γn
dw śx2´1
k“x2`y2´n`1pz ´ kq śx1
k“x1`y1´npw ´ kq 1 w ´ z
n
ź
i“1
ˆ w ´ βipnq z ´ βpnqi
˙
´ 1x1ěx2
pn ´ y1q!
pn ´ y2´ 1q!
1 p2πiq2
˛
Γ1n
dz
˛
γn
dw śx2´1
k“x2`y2´n`1pz ´ kq śx1
k“x1`y1´npw ´ kq 1 w ´ z
n
ź
i“1
ˆ w ´ βpnqi z ´ βipnq
˙ ,
(1.3.11) where Γn is a counterclockwise oriented contour containing tβjpnq: βjpnqě x2u but not the set tβjpnqď x2´1u, and Γ1nis a counterclockwise oriented contour containing tβpnqj : βjpnqă x2u but not the set tβpnqj ě x2` 1u. In addition, γn contains the set tx1` y1´ n, ..., x1u and Γn or Γ1n. The integration contours are shown in figure 1.11.
an bn
Γ1n
x2
Γn
γn
Figure 1.11: Integration contours. Here, an“ βpnqn and β1pnq“ bn.
One should note that the correlation kernel of a determinantal point process is not uniquely defined. Indeed, if Knppx1, y1q, px2, y2qq is a correlation kernel of a de-
