PLAQUETTE MODELS FOR GLASSES AND CONSTRAINT - - PowerPoint PPT Presentation

plaquette models for glasses and constraint satisfaction
SMART_READER_LITE
LIVE PREVIEW

PLAQUETTE MODELS FOR GLASSES AND CONSTRAINT - - PowerPoint PPT Presentation

PLAQUETTE MODELS FOR GLASSES AND CONSTRAINT SATISFACTION PROBLEMS Giacomo Gradenigo With: Silvio Franz (LPTMS,Orsay) Stefano Spigler (LPTMS,Orsay).


slide-1
SLIDE 1

PLAQUETTE ¡MODELS ¡FOR ¡GLASSES ¡ ¡ AND ¡ ¡ CONSTRAINT ¡SATISFACTION ¡PROBLEMS ¡

Giacomo ¡Gradenigo ¡ ¡ ¡ With: ¡Silvio ¡Franz ¡(LPTMS,Orsay) ¡ Stefano ¡Spigler ¡(LPTMS,Orsay). ¡

LIPhy, ¡27-­‑01-­‑2015 ¡

slide-2
SLIDE 2

WHAT ¡IS ¡A ¡GLASS ¡? ¡

KineVcally ¡Constrained ¡Models ¡ Thermodynamic ¡theories ¡ ¡ (Replicas ¡and ¡all ¡of ¡that…) ¡

…Just a very viscous fluid !!! …a new phase of matter !!!

log(η) Tg/T

slide-3
SLIDE 3

Ÿ Set ¡of ¡microscropic ¡ dynamic ¡rules ¡ Ÿ Glassy ¡behaviour ¡at ¡low ¡ temperatures ¡but ¡no ¡glass-­‑ transiVon ¡ Ÿ Exactly ¡solved ¡in ¡finite ¡ space ¡dimensions ¡

Thermodynamic ¡theories ¡ ¡ (Replicas ¡and ¡all ¡of ¡that…) ¡

Ÿ Well ¡defined ¡Hamiltonian ¡ and ¡parVVon ¡funcVon ¡ Ÿ Ideal ¡glass ¡transiVon, ¡ landscape ¡scenario, ¡predicVons ¡

  • n ¡the ¡dynamics ¡

Ÿ Exactly ¡solved ¡only ¡in ¡ approximate ¡theories ¡ ¡ (mean-­‑field, ¡random ¡graphs) ¡

KineVcally ¡ Constrained ¡Models ¡

¡

Review: ¡J.P. ¡Garrahan, ¡P. ¡Sollich, ¡C. ¡ Toninelli, ¡arXiv:1009.6113 ¡ Recent ¡result ¡on ¡jamming ¡(hard ¡spheres ¡in ¡d=∞): ¡ ¡

  • P. ¡Charbonneau, ¡ ¡J.Kurchan, ¡G.Parisi, ¡P.Urbani, ¡

F.Zamponi, ¡Nature ¡Comm. ¡(2014): ¡arXiv.1404.6809 ¡

slide-4
SLIDE 4

KineVcally ¡ Constrained ¡Models ¡

¡

Thermodynamic ¡theories ¡ ¡ (Replicas ¡and ¡all ¡of ¡that…) ¡

Triangular Plaquette Model Ÿ Set ¡of ¡microscropic ¡ dynamic ¡rules ¡ Ÿ Glassy ¡ behaviour ¡at ¡low ¡ temperatures ¡but ¡ no ¡glass-­‑transiVon ¡ Ÿ Exactly ¡solved ¡in ¡finite ¡ space ¡dimensions ¡ Ÿ Well ¡defined ¡Hamiltonian ¡ and ¡parVVon ¡funcVon ¡ Ÿ Ideal ¡glass ¡transiVon, ¡ landscape ¡scenario, ¡predicVons ¡

  • n ¡the ¡dynamics ¡

Ÿ Exactly ¡solved ¡only ¡in ¡ approximate ¡theories ¡ ¡ (mean-­‑field, ¡random ¡graphs) ¡ The two scenarios are complementary rather than alternative

slide-5
SLIDE 5

TRIANGULAR ¡PLAQUETTE ¡MODEL ¡

H = − X

hijki

σiσjσk

WHY ¡INTERSTING? ¡

2D: ¡Exactly ¡solvable ¡

GLASSY ¡DYNAMICS ¡

No ¡phase ¡transiVons ¡

τrel ∼ exp

  • A/T 2

σi σj σk

εp = −σi(p)σj(p)σk(p) Z = X

{ε}

e−βΣpεp

σi = ±1

  • M. ¡Newman ¡and ¡C. ¡Moore, ¡Phys. ¡Rev. ¡E ¡60, ¡5068 ¡ ¡(1999) ¡

J.P. ¡Garrahan ¡and ¡M. ¡Newman, ¡Phys. ¡Rev. ¡E ¡62, ¡7670 ¡(2000) ¡

slide-6
SLIDE 6

WHY ¡RELAXATION ¡IS ¡SLOW? ¡

(ACTIVATED ¡TRANSPORT ¡OF ¡DEFECTS) ¡ ✚ ¡ ✚ ¡ ✚ ¡ ✚ ¡ ✚ ¡ ✚ ¡ ✚ ¡ ✚ ¡ (1) ¡ (3) ¡ (4) ¡ (2) ¡

εp = −σi(p)σj(p)σk(p) ∆ε = 2 ∆ε = 2 ∆ε = 1 ∆ε = 3

✚ ¡ Spin ¡to ¡be ¡flipped ¡

ε = 1

slide-7
SLIDE 7

✚ ¡ ✚ ¡ ✚ ¡ ✚ ¡ ✚ ¡ ✚ ¡ (1) ¡ (4) ¡ (2) ¡ Effective Kinetic Constraint: creation

  • f a defect is favoured only when there are

already two closeby

τrel ∼ eβ∆E(β) cdef(β) ∼ e−β = ⇒ ξ(β) ∼ eβ ∆E(β) ∼ log[ξ(β)] τrel = eβ log ξ = [ξ(β)]β = exp(β2)

∆ε = 2 ∆ε = 1 ∆ε = 3 εp = −σi(p)σj(p)σk(p)

✚ ¡ Spin ¡to ¡be ¡flipped ¡

ε = 1

slide-8
SLIDE 8

TRIANGULAR ¡PLAQUETTE ¡MODEL ¡

H = − X

hijki

σiσjσk

A ¡RECENT ¡RESULT ¡

Two ¡‘’replicas’’ ¡are ¡coupled ¡

HTOT = H[(a)] + H[(b)] − ✏

N

X

i=1

(a)

i

(b)

i

Arbitrary ¡small ¡coupling ¡leads ¡to ¡ correlaVon ¡between ¡the ¡two ¡replicas ¡

0.5 1 1.5 2

!

0.5 1 1.5 2

T

high

  • verlap

low

  • verlap

critical endpoint self-dual line

J.P. ¡Garrahan, ¡Phys. ¡Rev. ¡E ¡89, ¡030301(R) ¡(2014) ¡

σi σj σk

slide-9
SLIDE 9

TRIANGULAR ¡PLAQUETTE ¡MODEL ¡

H = − X

hijki

σiσjσk

OUR ¡PURPOSE ¡

H = H0 + H1 = −J0 X

hijki

σiσjσk − J1 X

r

σi(r)σj(r)σk(r) H1 H0 ⌧ 1

Show ¡that ¡with ¡arbitrarily ¡small ¡perturbaVons ¡the ¡ system ¡has ¡a ¡finite-­‑temperature ¡glass ¡transiVon. ¡ Marginal ¡nature ¡of ¡the ¡model. ¡

σi σj σk

slide-10
SLIDE 10

THE ¡FORMATION ¡OF ¡A ¡GLASS ¡PHASE ¡ (configuraVonal) ¡ENTROPY ¡CRISIS ¡

Energy ¡landscape ¡with ¡‘’many’’ ¡local ¡minima ¡

f = U − T(s + Σ) N = exp [NΣ(T)]

ConfiguraVonal ¡ Entropy ¡

Σ(T) > 0 Σ(T) = 0

Entropic ¡drive ¡vs ¡Energy ¡Barriers: ¡ system ¡remains ¡ergodic ¡ System ¡becomes ¡a ¡glass ¡

IDEAL ¡GLASS ¡TRANSITION ¡

s # valleys = Σ

slide-11
SLIDE 11

THE ¡FORMATION ¡OF ¡A ¡GLASS ¡PHASE ¡

ExtrapolaVon ¡of ¡high-­‑temperature ¡thermodynamic ¡potenVals ¡ ConfiguraVonal ¡entropy ¡crisis ¡at ¡a ¡finite ¡temperature: ¡glass ¡formaVon ¡

Σ(T) = Sliq(T) − Ssol(T)

slide-12
SLIDE 12

RANDOM-­‑DILUTED ¡ ¡ TRIANGULAR ¡PLAQUETTE ¡MODEL ¡

In ¡the ¡new ¡triplets ¡spins ¡are ¡choosen ¡randomly ¡across ¡the ¡system ¡

H = H0 + H1 = −

Ns≤N

X

s=1

σi(s)σj(s)σk(s) −

M

X

r=1

σi(r)σj(r)σk(r)

A ¡diluVon ¡of ¡the ¡plaquenes ¡in ¡the ¡2D ¡laoce ¡is ¡allowed ¡ Ns < N

slide-13
SLIDE 13

In ¡the ¡new ¡triplets ¡spins ¡are ¡choosen ¡randomly ¡across ¡the ¡system ¡

RANDOM-­‑DILUTED ¡ ¡ TRIANGULAR ¡PLAQUETTE ¡MODEL ¡

H = −

N

X

s=1

σi(s)σj(s)σk(s) −

M

X

r=1

σi(r)σj(r)σk(r) Zh.T. = 2N [cosh(β)]M

PROVED ¡! ¡ Entropy ¡crisis ¡at ¡ finite ¡temperature: ¡ Glass ¡transiBon? ¡

Sliq(β → ∞) = ✓ 1 − M N ◆ Sliq > Σ

Zh.T. = 2N [cosh(β)]M 2 41 tanh(β) X

p=1

hεpi + tanh2(β) X

p6=q

hεpεqi + . . . 3 5

εp = −σi(p)σj(p)σk(p)

slide-14
SLIDE 14

Zh.T. = 2N [cosh(β)]M

PROVEN ¡

RANDOM-­‑DILUTED ¡ ¡ TRIANGULAR ¡PLAQUETTE ¡MODEL ¡

  • 1.1
  • 1.05
  • 1
  • 0.95
  • 0.9
  • 0.85
  • 0.8
  • 0.75
  • 0.7

0.2 0.4 0.6 0.8 1 1.2

E/Nspin T

cooling heating high temperature expansion

0.1

rel 1/T2

Entropy ¡crisis ¡at ¡ finite ¡temperature: ¡ Glass ¡transiBon? ¡

Sliq(β → ∞) = ✓ 1 − M N ◆ Sliq > Σ

exp(A/T 2) vs exp(A/(T − TK)) = ? Tm

L = 64 L = 256

slide-15
SLIDE 15

HOW ¡TO ¡HANDLE ¡THE ¡PROBLEM? ¡ TOOLS ¡AND ¡IDEAS ¡FROM ¡ ¡ CONSTRAINT ¡SATISFACTION ¡PROBLEMS! ¡ ¡

slide-16
SLIDE 16

DILUTED ¡P-­‑SPIN ¡

Spin ¡connecVvity ¡= ¡Poissonian ¡ Each ¡plaquene ¡is ¡anached ¡to ¡three ¡spins ¡

H = − X

ijk

Jijk σiσjσk

p(`) = e−3↵ (3↵)` `!

Glass ¡ Supercooled ¡ liquid ¡ Liquid ¡ BiparVte ¡graph ¡ FuncVon ¡nodes: ¡Plaquenes ¡ Variable ¡nodes: ¡Spins ¡ M ¡= ¡number ¡of ¡plaquenes ¡ N ¡= ¡number ¡of ¡spins ¡

α = M/N

TK Td

slide-17
SLIDE 17

DILUTED ¡P-­‑SPIN ¡

Spin ¡connecVvity ¡= ¡Poissonian ¡ Each ¡plaquene ¡is ¡anached ¡to ¡three ¡spins ¡

H = − X

ijk

Jijk σiσjσk

p(`) = e−3↵ (3↵)` `!

Glass ¡ Supercooled ¡ liquid ¡ Liquid ¡

log(N) ¡loops ¡

BiparVte ¡graph ¡

α = M/N

TK Td

slide-18
SLIDE 18

αs

UNSAT ¡

XORSAT ¡

(ni1 + nj1 + nk1)mod2 = J1 (niM + njM + nkM )mod2 = JM . . .

SAT ¡

  • M. ¡Mezard, ¡F. ¡Ricci-­‑Tersenghi, ¡R. ¡

Zecchina, ¡JSTAT ¡111, ¡505 ¡(2003) ¡

BiparVte ¡graph ¡

log(N) ¡loops ¡

ni(p) = 0, 1 Jp = 0, 1

System ¡of ¡M ¡linear ¡equaVons ¡ with ¡boolean ¡variables ¡

α = M/N

slide-19
SLIDE 19

αs αd

UNSAT ¡ Clustered ¡ Non-­‑clustered ¡

XORSAT ¡

∆E = O(N) #clusters = Σ

  • M. ¡Mezard, ¡F. ¡Ricci-­‑Tersenghi, ¡R. ¡

Zecchina, ¡JSTAT ¡111, ¡505 ¡(2003) ¡

Glass ¡ Supercooled ¡ liquid ¡ Liquid ¡

(ni1 + nj1 + nk1)mod2 = J1 (niM + njM + nkM )mod2 = JM . . .

log(N) ¡loops ¡

slide-20
SLIDE 20

XORSAT ¡and ¡LEAF-­‑REMOVAL ¡ ¡

1) ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡is ¡ ¡present ¡in ¡a ¡single ¡equaVon ¡(interacVon): ¡LEAF ¡ 2) ¡Remove ¡the ¡equaVon ¡(interacVon) ¡ ¡

nip

3) ¡Check ¡if ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡and ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡are ¡new ¡LEAVES ¡ 4) ¡Repeat ¡from ¡1) ¡ ¡for ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡or ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡or ¡ ¡search ¡for ¡

  • ther ¡leaves ¡

njp njp nkp nkp nip + njp + nkp = Jp Mc Nc

Number ¡of ¡plaquenes ¡(equaVons) ¡in ¡the ¡CORE ¡ Number ¡of ¡spins ¡(variables) ¡in ¡the ¡CORE ¡

CORE

Part ¡of ¡the ¡interacVon ¡network ¡such ¡that ¡each ¡ variable ¡parVcipates ¡to ¡at ¡least ¡two ¡interacVons ¡

slide-21
SLIDE 21

Mc Nc

Number ¡of ¡plaquenes ¡(equaVons) ¡in ¡the ¡CORE ¡ Number ¡of ¡spins ¡(variables) ¡in ¡the ¡CORE ¡

nip njp nkp nkp

Part ¡of ¡the ¡interacVon ¡network ¡such ¡that ¡each ¡ variable ¡parVcipates ¡to ¡at ¡least ¡two ¡interacVons ¡

CORE

slide-22
SLIDE 22

αs αd

UNSAT ¡ Clustered ¡ Non-­‑clustered ¡ All ¡variables ¡and ¡interacVons ¡ are ¡removed ¡ CORE ¡ CORE ¡

Mc/Nc > 1 Mc/Nc < 1 nip njp nkp nkp

slide-23
SLIDE 23

αs αd

UNSAT ¡ Clustered ¡ Non-­‑clustered ¡ All ¡variables ¡and ¡interacVons ¡ are ¡removed ¡ Exact ¡soluVon ¡for ¡ locally ¡tree-­‑like ¡graph ¡

˙ n`(t) = −`1 + `0 + 2 3(↵ − t) [(` + 1)n`+1(t) − `n`(t)]

t = niteration N

CORE ¡ CORE ¡

Mc/Nc > 1 Mc/Nc < 1

0.5 1 1.5 2 f1() /

  • < d

= d > d * 0.5 1 1.5 2 f1() /

  • < d

= d > d * 0.5 1 1.5 2 f1() /

  • < d

= d > d * 0.5 1 1.5 2 2.5 3 0.8 0.85 0.9 0.95 1 d * vs.

slide-24
SLIDE 24

RD-­‑TPM ¡model ¡and ¡LEAF-­‑REMOVAL ¡ (zero ¡temperature ¡phase-­‑diagram) ¡ ¡

H = −

Ms

X

s=1

σi(s)σj(s)σk(s) −

ML

X

r=1

σi(r)σj(r)σk(r)

Plaquenes ¡in ¡the ¡2D ¡plane ¡ Extra ¡random ¡plaquenes ¡

αs = Ms N αL = ML N ˙ n`(t) = −`1 + `0 + 2 3(↵ − t) [(` + 1)n`+1(t) − `n`(t)]

Diluted ¡p-­‑spin ¡

n`(0) = e−3↵ (3↵)` `!

Our ¡model ¡

n`(0) = N −1

`0

X

r=0

ρr(αs)p`−r(αL)

pk(αL) = e−3αL (3αL)k k! ρk(αs) = ✓3 k ◆ αk

s(1 − αs)3−k

slide-25
SLIDE 25

RD-­‑TPM ¡model ¡and ¡LEAF-­‑REMOVAL ¡ (zero ¡temperature ¡phase-­‑diagram) ¡ ¡

H = −

Ms

X

s=1

σi(s)σj(s)σk(s) −

ML

X

r=1

σi(r)σj(r)σk(r)

Plaquenes ¡in ¡the ¡2D ¡plane ¡ Extra ¡random ¡plaquenes ¡

αs = Ms N αL = ML N

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

L s SAT UNSAT TPM

L 1-s

b)

L

th(N) [s=1]

1/N1/2

a)

Main ¡Panel: ¡ ¡ Points, ¡numerical ¡results ¡in ¡2D ¡ (1024x1024); ¡ Lines, ¡analyVcal ¡results ¡on ¡ random ¡regular ¡graph ¡

slide-26
SLIDE 26

RD-­‑TPM ¡model ¡and ¡LEAF-­‑REMOVAL ¡ (zero ¡temperature ¡phase-­‑diagram) ¡ ¡

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

L s SAT UNSAT TPM

L 1-s

b)

L

th(N) [s=1]

1/N1/2

a)

Main ¡Panel: ¡ ¡ Points, ¡numerical ¡results ¡in ¡2D ¡ (1024x1024); ¡ Lines, ¡analyVcal ¡results ¡on ¡ random ¡regular ¡graph ¡

Triangular ¡Plaquene ¡Model ¡is ¡the ¡only ¡2D ¡model ¡(in ¡ the ¡phase ¡diagram) ¡such ¡that ¡every ¡concentraVon ¡ ¡ ¡ makes ¡the ¡problem ¡unsaVsfiable ¡(at ¡T=0 ¡glass ¡phase) ¡

αL > 0

slide-27
SLIDE 27

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

L s SAT UNSAT TPM

  • th(N) [ =1]

LEAF-­‑REMOVAL ¡and ¡INTEGRABILITY ¡

0.001 0.002 0.003 0.004 0.005 0.006 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

L 1 - s

CORE NO CORE

x2

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

Nc/N L

0.001 0.003 0.005 0.007 0.0008 0.0016 0.0024

L(N) N1/2

ExtrapolaVon ¡of ¡numerical ¡data ¡(in ¡ 2D) ¡to ¡the ¡thermodynamic ¡limit ¡

Z = 2N[cosh(β)]Ms+ML

NO ¡CORE ¡= ¡Summability ¡of ¡Z ¡ with ¡parVal ¡traces ¡

slide-28
SLIDE 28

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

L s SAT UNSAT TPM

  • th(N) [ =1]

LEAF-­‑REMOVAL ¡and ¡INTEGRABILITY ¡

Z = 2N[cosh(β)]Ms+ML

NO ¡CORE ¡= ¡Summability ¡of ¡Z ¡ with ¡parVal ¡traces ¡

START: ¡find ¡a ¡spin ¡which ¡parVcipates ¡ to ¡a ¡single ¡interacVon ¡ Go ¡to ¡START ¡

Z = X

{σiq ,...}

eβσiq σjq σkq eβ P

r6=q σir σjr σkr

Z = cosh(β) X

{σiq ,...}

(1 + σiqσjqσkq tanh(β)) eβ P

r6=q σir σjr σkr

Z = 2 cosh(β) X

{σ}

eβ P

r6=q σir σjr σkr

0) ¡ 3) ¡ 2) ¡ 1) ¡

slide-29
SLIDE 29

GLASS ¡TRANSITION ¡AT ¡FINITE ¡TEMPERATURE ¡

UNSAT ¡PHASE: ¡Ground ¡states ¡have ¡extensive ¡energy ¡(random ¡graph) ¡ Minimum ¡allowed ¡value ¡for ¡the ¡concentraVon ¡of ¡excited ¡plaquenes ¡ But ¡for ¡the ¡concentraVon ¡of ¡defects ¡

c ∼ e−aβ

… ¡so ¡that ¡

α ∼ e−bβK cmin ∼ εmin ∼ αg

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.05 0.1 0.15 0.2 0.25 0.3

T L (s=1) glass liquid

b = a g εmin ∼ αg

DiluVon ¡of ¡‘’short ¡range ¡ plaquenes ¡is ¡fixed’’ ¡ Exact ¡result ¡on ¡the ¡random ¡graph ¡

αs = 1 αs = 0.96

and ¡therefore ¡ cmin ∼ e−aβK

slide-30
SLIDE 30

GLASS ¡TRANSITION ¡AT ¡FINITE ¡TEMPERATURE ¡

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.05 0.1 0.15 0.2 0.25 0.3

T L (s=1) glass liquid

On ¡a ¡locally ¡tree-­‑like ¡random ¡graph, ¡ ¡ by ¡means ¡of ¡the ¡cavity ¡method, ¡

  • ne ¡can ¡exactly ¡compute ¡the ¡

CONFIGURATIONAL ¡ENTROPY ¡

Σ = fliq − fmeta

UNSAT ¡PHASE: ¡Ground ¡states ¡have ¡extensive ¡energy ¡(random ¡graph) ¡ Minimum ¡allowed ¡value ¡for ¡the ¡concentraVon ¡of ¡excited ¡plaquenes ¡ But ¡for ¡the ¡concentraVon ¡of ¡defects ¡

c ∼ e−aβ

… ¡so ¡that ¡

α ∼ e−bβK cmin ∼ εmin ∼ αg b = a g εmin ∼ αg

and ¡therefore ¡ cmin ∼ e−aβK

slide-31
SLIDE 31

CONCLUSIONS ¡

We ¡have ¡introduced ¡a ¡model ¡which ¡interpolates ¡from ¡ the ¡Triangular ¡Plaquene ¡Model ¡to ¡the ¡diluted ¡p-­‑spin, ¡ The ¡Random-­‑Diluted ¡TPM. ¡ By ¡using ¡the ¡RD-­‑TPM ¡we ¡have ¡shown ¡that ¡the ¡TPM ¡is ¡arbitrarily ¡ close ¡to ¡models ¡with ¡a ¡finite ¡temperature ¡glass-­‑transiVon. ¡ TO ¡BE ¡PROVEN: ¡when ¡Nc ¡< ¡Mc ¡in ¡our ¡model ¡there ¡are ¡ ground ¡states ¡with ¡extensive ¡energy ¡