Ideal Quantum Glass Transitions: Many-body localization without quenched disorder TIDS 15 1-5 September 2013 Sant Feliu de Guixols Markus Müller Mauro.

Ideal Quantum Glass Transitions:Many-body localization without

quenched disorder

TIDS 15 1-5 September 2013 Sant Feliu de Guixols

Markus Müller

Mauro Schiulaz

MotivationG. Carleo, F. Becca, M. Schiro, M. Fabrizio,

Scientific Reports 2, 243 (2012).

L=8, 10, 12

Dynamics starting from inhomogeneous initial condition At large U: relaxation time grows (diverges?) with L!

→ Glass transition ?! Why – how ??

Glasses =

Systems defying thermodynamic equilibration

• Breaking of ergodicity (τrel → ∞)• Absence of full thermalization

How can this occur in general?

Routes to glassiness

Spin glasses + cousins: (Coulomb glasses?)Usual classical ingredients: Disorder + Frustration

→ Barriers grow with size L τrel ~ exp[Lθ]

Structural glasses (viscous, supercooled liquids):Steric frustration + self-generated disorder → growing time- and length-scales (Kirkpatrick, Thirumalai, Wolynes)

But: Eternal debate without conclusion: Is there an “ideal” glass transition at finite T: Can τrel → ∞, before full jamming and incompressibility are reached ???

Ideal glass transition?

Classical structural glasses in finite d: ??? How can barriers become infinite without jamming?

BUTQuantum ideal (disorder free) glasses can exist! Extra ingredient: Anderson localization! - Properties:

• τrel = ∞ (ergodicity broken) • Self-generated disorder• No d.c. transport / no diffusion in thermodynamic limit

• Classical frustration plays no role!• Glass because of quantum effects, NOT despite of them (↔ quantum spin glass, superglass)

Model: inhibited hopping in 1d

H0 : non-ergodic

T : potentially restores ergodicity

Anderson:Many-body quantum glass:

Inhibited hopping model Eigenstate at λ = 0

Aim: show that for λ ≤ λc many-body localized quantum glass!

Signatures of quantum glass, many-body localization

• No thermalization

• Persistence of spatial inhomogeneity in long-time average

• Spontaneous breaking of translational invariance

• No d.c. transport

Essential ingredients

H0 is fully localized: has an extensive set of local, conserved operators → “ integrable system”

γ

Simple arguments for localization

• Hybridization between eigenstates with different τi,l;m is suppressed, since energies El;m~ J >> λ

→ Expect: “Hopping ~λ in the lattice labeled with τi,l;m is localized”

• BUT: Caveat: El;m does not depend on site i! → Spectrum of H0 is extensively degenerate → resonant delocalization?

Summary of the argument• Degeneracies are lifted at low orders in perturbation

theory in λ

• Near-degeneracies are much more weakly coupled than their level splitting (typically)

→ Rare resonances occur locally, but don’t percolate

→ Perturbation theory [in lifted basis] converges for λ << J(same reasons as in systems with quenched disorder,

[Basko, Aleiner, Altshuler 2006, Imbrie&Spencer, unpub.])

→ Eigenstates are localized close to inhomogeneous eigenstates of H0

→ Initial inhomogeneities remain frozen in dynamics! → Eigenstate thermalization hypothesis is not obeyed!

Eigenstate perturbation theory

Perturb localized eigenstates of H0, expand in λ

That is: Choose basis of fixed barrier positions, do not fix momentum!

Lifting of degeneracies?

Resonances and hybridization3 degenerate configurations (with li = lj ±1): Degeneracy is lifted by hybridization at order O(λ)

But: Most configurations remain degenerate at first order.

Lifting degeneracies at order O(λ2)

Generic lifting mechanism: Virtual barrier hops: ΔΕ ~λ2/J

→ In general two eigenstates don’t hybridize unless they can be connected by only two barrier hops (matrix element ~ λ2/J).

Lifting degeneracies at order O(λ2)

But: some eigenstates remain exactly degenerate at order O(λ2):

Lifting degeneracies at order O(λ2)

But: some eigenstates remain exactly degenerate at order O(λ2):

In random eigenstates their density is small ~ 0.034 (ρbarrier=1/2).

→ Eigenstates with same shift ΔΕ ~λ2/J are connected by matrix elements ~ λn~30 <<< λ4. Resonances are very rare!

→ In typical, random eigenstates, perturbation theory converges!

Dynamic localizationIn typical, random eigenstates, perturbation theory

converges at small λ!

→ Eigenstates are localized close to inhomogeneous eigenstates of H0

→ Initial inhomogeneities remain frozen in dynamics! [Expand the initial state in eigenstates and check!]

BUT: highly atypical, nearly periodic eigenstates hybridize over large distances and delocalize! Nevertheless, generic initial conditions have exponentially small weight on such eigenstates, and remain localized.

Independent, direct check?

Numerical verification?

Make use of translational invariance!

Spontaneous symmetry breaking

Break translational invariance by very weak disorder W

Check eigenstate inhomogeneity

Spontaneous dynamical breaking of translational invariance = self-induced many-body localization

;

Susceptibility to disorder

?

All ρbarrierL barriers must be moved to hybridize the degenerate barrier configurations (rigid rotations)

→ exponentially large “mass”, → exponentially small splitting of the band,→ expoentially strong response to disorder

Analytics: Disorder-response dominated by mixing of L nearly degenerate momentum states

Susceptibility to disorder

?

Linear response Slope: Susceptibility

L

Analytics: Disorder-response dominated by mixing of L nearly degenerate momentum states

Susceptibility to disorder

?

Many-body problem Non-intercating particles

>>

Comparison with free particles

Susceptibility to disorder

?Glass transition?

No exponential sensitivity to W!

Very rough estimate:

Ideal quantum glass exists in a substantial range 0 < λ < λc !

For the expertsRecent conjectures (Huse& Oganesyan; Serbyn&Papic&Abanin 2013)

“Many-body localization ↔ Existence of an extensive set of local conserved quantities, as in integrable models.These conservation laws prohibit thermalization.”

? ?

In disorder-free quantum glasses, such operators seem not to exist for λ > 0 .

They seem to be inconsistent with rare delocalized states.

→ When is the above conjecture correct? → What does it imply when it does not hold?

Conclusions

• Self-generated disorder [initial conditions] can induce many-body Anderson localization in closed, disorder-free quantum systems

• Here: Manybody localization = spontaneous dynamical breaking of translational symmetry

→ Genuine, ideal dynamic quantum glass Induced by quantum effects BUT: requires coherence = absence of noise/dephasing → In reality: ergodicity breaking up to time scale controlled

by remaining dissipative processes.

Open questions

• Nature of the glass / localization transition as function of λ?

• Temperature dependence? Delocalization due to reduced disorder at low T?

• Relation with Anderson orthogonality catastrophy (if any)?Localization and non-thermalization in strongly correlated systems?

• Interplay between manybody Anderson localization and classical frustration in phase space? Interacting insulator-to-conductor transitions?