emergent hydrodynamics in integrable systems out of …...a. bressan, hyperbolic conservation laws:...
TRANSCRIPT
-
'
&
$
%
Emergent hydrodynamics in integrable systems out of equilibrium
Benjamin Doyon
Department of Mathematics, King’s College London
Bled, Slovenia, 8 June 2018
1
-
'
&
$
%
Collaborators:
Alvise Bastianello, SISSA
Denis Bernard, École Normale Supérieure de Paris
M. Joe Bhaseen, KCL
Isabelle Bouchoule, Institue d’Optique, Paris
Olalla Castro Alvaredo, City, University of London
Jean-Sébastien Caux, University of Amsterdam
Jacopo De Nardis, École Normale Supérieure de Paris
Jérôme Dubail, Université de Lorraine
Rosemary Harris, Queen Mary University London
Robert Konik, Brookhaven National Laboratory
Maximilian Schemmer, Institut d’Optique, Paris
Herbert Spohn, Technische Universität München
Gerard Watts, KCL
Students:
Jason Myers, KCL (CANES)
Takato Yoshimura, CEA Saclay (on leave from KCL)
2
-
'
&
$
%
Main papers
O. A. Castro-Alvaredo, B. Doyon, T. Yoshimura, Emergent hydrodynamics in integrable quantum systems
out of equilibrium, Phys. Rev. X 6, 041065 (2016)
B. Bertini, M. Collura, J. De Nardis, M. Fagotti, Transport in Out-of-Equilibrium XXZ Chains: Exact
Profiles of Charges and Currents, Phys. Rev. Lett. 117, 207201 (2016)
Both selected for a Viewpoint in Physics written by Jérôme Dubail
(http://physics.aps.org/articles/v9/153#c1)
Advertisements:
Conference on Non-Equilibrium Systems (CONES), King’s College London
25-27 June 2018, www.cones2018.com
Les Houches summer school on integrability in atomic physics and condensed matter
August 2018 (org.: J.-S. Caux, K. Kitanine, A. Kluemper, R. Konik)
Programme in Natal: Hydrodynamic of low-dimensional quantum systems
May 2019 (org.: BD, J. Dubail, G. Mussardo, M. Rajabpour, J. Viti)
3
-
'
&
$
%
Plan
I. Basic theory of generalised hydrodynamics (GHD)
II. Numerical simulations and comparison with cold atom experiment
III. Further developments:
– steady states, geometry, method of characteristics
– zero temperature
– correlation functions, Drude weights
– large-deviation theory (a.k.a. full counting statistics)
– diffusion and viscosity
4
-
'
&
$
%
I. Basic theory
5
-
'
&
$
%
1. The problem: inhomogeneous dynamics of many-body integrable systems
Effects of integrability in the famous “Quantum Newton Cradle” experiment:
[Kinoshita, Wenger, Weiss 2006]
“Our results are probably explainable
by the well-known fact that a homoge-
neous 1D Bose gas with point-like colli-
sional interactions is integrable”
“Until now, however, the time evolution
of out-of-equilibrium 1D Bose gases
has been a theoretically unsettled is-
sue, as practical factors such as har-
monic trapping and imperfectly point-
like interactions may compromise inte-
grability”
6
-
'
&
$
%
1. The problem: inhomogeneous dynamics of many-body integrable systems
The Lieb-Liniger model describes point-like interactions of Galilean invariant Bose gases:
H =
∫dx h(x) =
∫dx
(1
2m∂xΨ
†∂xΨ +c
2Ψ†Ψ†ΨΨ
).
A simplification of the experimental problem is as follows. The initial state is an equilibrium
state with a inhomogeneous potential coupled to the density n(x) = Ψ†(x)Ψ(x):
〈A〉 = Tr (ρiniA)Trρini
, ρini = exp
[−β(H +
∫dxVini(x)n(x)
)].
Then the evolution occurs with the Hamiltonian in a different inhomogeneous potential
〈A(t)〉 = 〈eiHevotAe−iHevot〉, Hevo = H +∫
dxVevo(x)n(x).
7
-
'
&
$
%
1. The problem: inhomogeneous dynamics of many-body integrable systems
How can we compute the evolution of such a gas?
What are the general principles?
Can we reproduce the effects seen in the quantum Newton cradle experiment?
8
-
'
&
$
%
2. From Gibbs ensembles to conventional hydrodynamics
For nice descriptions of standard concepts in hydrodynamics, see e.g.
A. Bressan, Hyperbolic conservation laws: An illustrated tutorial. In Modelling and Optimisation of Flows
on Networks, Cetraro, Italy 2009, Lecture Notes in Mathematics 2062, Springer, 2013
H. Spohn, Large Scale Dynamics of Interacting Particles, Springer Verlag, Heidelberg, 1991
9
-
'
&
$
%
2. From Gibbs ensembles to conventional hydrodynamics
Hydrodynamics is the natural framework to describe inhomogeneous many-body phenomena.
The main idea behind hydrodynamics is “local thermodynamic equilibrium”.
It says that locally and on very short time scales (in fluid cells), the many-body system
“equilibrates” or relaxes. This means (naively) that locally we observe Gibbs states with
inverse temperature β(x, t) and chemical potential µ(x, t). Since things can be moving, then
in general these will be boosted by the local fluid velocity, with boost parameter ν(x, t):
ρGE(x, t) = e−β(x,t)(H−µ(x,t)N−ν(x,t)P )
and the hydrodynamic approximation is
〈O(x, t)〉ini ≈Tr (ρGE(x, t)O(0, 0))
TrρGE(x, t).
For instance (“local density approximation”):
ρini = e−β(H+
∫dxVini(x)n(x)) ⇒ ρGE(x, 0) = e−β(H+Vini(x)N).
10
-
'
&
$
%
2. From Gibbs ensembles to conventional hydrodynamics
11
-
'
&
$
%
2. From Gibbs ensembles to conventional hydrodynamics
Hydrodynamics follows from local thermodynamic equilibrium + current conservation.
From microscopic conservation laws (for operators) to hydrodynamic equations (for Gibbs
averages)
∂th(x, t) + ∂xjh(x, t) = 0
∂tp(x, t) + ∂xjp(x, t) = 0
∂tn(x, t) + ∂xp(x, t) = 0
Take averages,hydro approximation⇒
∂th(x, t) + ∂xjh(x, t) = 0
∂tp(x, t) + ∂xjp(x, t) = 0
∂tn(x, t) + ∂xp(x, t) = 0
where
h(x, t) =Tr(ρGE(x, t) h)
TrρGE(x, t), jh(x, t) =
Tr(ρGE(x, t) jh)
TrρGE(x, t)
etc.
12
-
'
&
$
%
2. From Gibbs ensembles to conventional hydrodynamics
Since there are only three parameters µ, ν, β to determine a boosted Gibbs state, there must
be two relations:
jh = F (h, p, n), jp = G(h, p, n).
These are the equations of state of the gas (which are highly model-dependent), and
combined with the above give the hydrodynamic equations for h, p, n.
• Note that h, p, n fix the potentials β, µ, ν. Thus hydrodynamics fixes the localspace-time dependent state.
• This is valid in the Euler limit: the limit of infinitesimal small variations in space and time.Beyond this scale, there are higher derivative corrections, such as viscosity terms.
• One usually write p = n v, giving the usual Euler equation
∂tv + v∂xv = −1
n∂xP
where the pressure is P = jp − nv2.
13
-
'
&
$
%
3. From Gibbs ensembles to Generalized Gibbs ensembles (GGEs)
An old idea that got a lot of new interest recently starting with M. Rigol, V. Dunjko, V. Yurovsky, M.
Olshanii, Phys. Rev. Lett. 97, 050405 (2007).
Modern reviews:
A. Polkovnikov, K. Sengupta, A. Silva, M. Vengalattore, Rev. Mod. Phys. 83, 863 (2011)
J. Eisert, M. Friesdorf, C. Gogolin, Nat. Phys. 11, 124 (2015)
C. Gogolin, J. Eisert, Rep. Prog. Phys. 79, 056001 (2016)
F. Essler, M. Fagotti, J. Stat. Mech. 2016, 064002 (2016)
L. Vidmar, M. Rigol, J. Stat. Mech. 2016, 064007 (2016)
E. Ilievski, M. Medenjak, T. Prosen, L. Zadnik, J. Stat. Mech. 2016, 064008 (2016)
14
-
'
&
$
%
3. From Gibbs ensembles to Generalized Gibbs ensembles (GGEs)
But the Lieb-Liniger model is integrable. Integrable models possess an infinite number of
local conserved quantities
Qi =
∫dx qi(x), ∂tqi + ∂xji = 0, i = 0, 1, 2, 3, . . . unboundedly.
This gives rise to generalised Gibbs ensembles:
ρGGE = exp
[−∞∑
i=0
βiQi
].
15
-
'
&
$
%
Combining: inhomogeneous dynamics of integrable systems
∂tqi + ∂xji = 0, qi, ji =Tr [ρGGE(x, t) qi, ji]
Tr ρGGE(x, t), i = 0, 1, 2, . . .
This is generalized hydrodynamics (GHD).
• The hydrodynamic principle is the emergence of local entropy maximization with respect toall available conserved charges, valid when variation lengths are large enough:
〈O(x, t)〉ini ≈Tr [ρGGE(x, t)O]
Tr ρGGE(x, t)
• There are equations of states: ji = Fi({qj}), and a bijective relation qi ↔ βi.
• Equations of conservations give dynamical equations determining the space-timedependent (generalized) Gibbs ensembles.
16
-
'
&
$
%
Combining: inhomogeneous dynamics of integrable systems
17
-
'
&
$
%
4. GHD in the quasi-particle language
for (generalized) thermodynamic Bethe ansatz:
C.N. Yang, C.P. Yang, J. Math. Phys. 10, 1115 (1969)
A. Zamolodchikov, Nucl. Phys. B 342, 695 (1990)
J. Mossel, J.-S. Caux, J. Phys. A 45, 255001 (2012)
For an understanding the relation GGE = (G)TBA in the XXZ chain:
E. Ilievski, E. Quinn, J.-S. Caux, Phys. Rev. B 95, 115128 (2017)
For GHD in the quasi-particle language:
O. A. Castro-Alvaredo, BD, T. Yoshimura, Phys. Rev. X 6, 041065 (2016)
B. Bertini, M. Collura, J. De Nardis, M. Fagotti, Phys. Rev. Lett. 117, 207201 (2016)
and the terms due to external force fields, temperature fields etc. were found in
BD, T. Yoshimura, SciPost Phys. 2, 014 (2017)
18
-
'
&
$
%
4. GHD in the quasi-particle language
Gibbs ensembles and generalized Gibbs ensembles can be described, in Bethe ansatz
integrable models, by using the (generalized) thermodynamic Bethe ansatz.
This is based on the fact that there emerge stable quasi-particles. The set of their momenta
and other quantum numbers is preserved under scattering, thus giving good quantum numbers
used to describe GGEs.
These quantum numbers are gathered into a spectral parameter θ characterizing the
quasi-particle. So, instead of the conserved densities qi, we use the conserved quasi-particle
densities ρp(θ):
{qi : i} ⇒ {ρp(θ) : θ}
19
-
'
&
$
%
4. GHD in the quasi-particle language
The quantity of charge Qi carried by the quasi-particle θ is a function hi(θ). So we have
qi =
∫dθ hi(θ)ρp(θ).
The GGE current averages ji are also given by linear functional on that space. Since ρp fully
determines the GGE, one can always write, for some veff(θ) = veff[ρp](θ),
ji =
∫dθ hi(θ)v
eff(θ)ρp(θ).
We now make GGEs space-time dependent. This means we promote ρp(θ) 7→ ρp(x, t; θ).We use completeness:
∂tqi + ∂xji = 0 ⇒∂
∂tρp(x, t; θ) +
∂
∂x
[veff(x, t; θ)ρp(x, t; θ)
]= 0
These are the GHD equations in the quasi-particle language.
20
-
'
&
$
%
4. GHD in the quasi-particle language
In order to get veff – the GGE equation of state – we need the model characteristics.
• spectral space: θ ∈ S , typically S = R× I where R represents the continuum ofvelocities and I a space of emerging particle types (including TBA strings, etc.).
• momentum function that defines relation between spectral space and real space, forinstance p(θ) = mθ or p(θ) = m sinh(θ).
• energy function that controls time evolution, for instance E(θ) = mθ2/2 orE(θ) = m cosh θ.
• differential scattering phase or TBA kernel that controls interactions, for instance ϕ(θ− θ′)=−id logS(θ − θ′)/dθ where S(θ − θ′) is the two-particle scattering amplitude.
21
-
'
&
$
%
4. GHD in the quasi-particle language
Using these model characteristics we can express the effective velocity. We find:
veff(θ) =E′(θ)
p′(θ)+
∫dα
ϕ(θ, α) ρp(α)
p′(θ)(veff(α)− veff(θ))
This is the “effectivization” of the group velocity E′(θ) / p′(θ).
This can be seen as the GGE equations of state (relating conserved currents to densities).
veff is in fact the velocity of excitations above finite-density states [Bonnes, Essler, Läuchli 2015].
The GGE equations of state can be proven using crossing symmetry in relativistic QFT, and in
the Lieb-Liniger model by taking the non-relativistic limit. It has been checked numerically in
the XXZ chain. There are now more complete proofs using form factor expansions [De Nardis,
Bernard, BD in progress; Yoshimura et al. in progress]
22
-
'
&
$
%
4. GHD in the quasi-particle language
We can also define the occupation function (or filling fraction)
n(θ) =ρp(θ)
ρs(θ)=
2πρp(θ)
p′(θ) +∫
dαϕ(θ, α)ρp(α)
These new hydrodynamic variables are the normal modes of GHD:
∂tn(x, t; θ) + veff(x, t; θ)∂xn(x, t; θ) = 0.
That is, the occupation function is the fluid coordinate that diagonalizes GHD. It is convectively
transported by the fluid, with propagation velocities veff(x, t; θ).
23
-
'
&
$
%
4. GHD in the quasi-particle language
All of this generalizes to the presence of force fields, etc.
It is the energy function that controls the time evolution. Assume that it is explicitly space
dependent E(θ) = E(x; θ). For instance in the repulsive Lieb-Liniger model in a force field,
Hevo = H +
∫dxVevo(x)n(x) ⇒ E(x; θ) = mθ2/2 + Vevo(x).
or evolved with inhomogeneous energy density,
Hevo =
∫dxJevo(x)h(x) ⇒ E(x; θ) = Jevo(x)mθ2/2.
Then the two following equivalent equations hold (here suppressing x, t; θ dependence):
∂tρp + ∂x[veffρp
]+ ∂θ
[aeffρp
]= 0
∂tn+ veff∂xn+ a
eff∂θn = 0
where the aeff is the “effectivization” of the acceleration−∂xE /∂θp.The local density approximation of thermal states e−βHevo are stationary solutions.
24
-
'
&
$
%
5. Molecular dynamics and the effective velocity
BD, T. Yoshimura, J.-S. Caux, Phys. Rev. Lett. 120, 045301 (2018)
25
-
'
&
$
%
5. Molecular dynamics and the effective velocity
We can simulate GHD using classical dynamics of particles in one dimension.
Consider a simple gas of hard rods, a gas of segments of lengths a lying on the line, and
propagating freely except for elastic collisions.
Upon collision, velocities are exchanged. A “quasi-particle” is a tracer of a given velocity. A
tracer jumps by a distance a at every collision.
By elastic collision, the number of tracers in a given velocity interval [v, v + dv] is preserved
by the dynamics, giving infinitely many conservation laws.
26
-
'
&
$
%
5. Molecular dynamics and the effective velocity
Accumulating the jumps, an effective velocity develops at large scales:
It was shown rigorously that [Boldrighini, Dobrushin, Sukhov 1982]
veffcl (v) = v − a∫
dw ρp(w)(veffcl (w)− veffcl (v))
This is GHD if we identify v = θ and ϕ(θ, α) = −a !!!!
27
-
'
&
$
%
5. Molecular dynamics and the effective velocity
We can generalize this to velocity-dependent jump lengths d(v, w) in any direction: the
“flea gas”.
The quasi-particles act like classical solitons of an integrable field theory, and indeed GHD
equations were found in soliton gases.
[Zakharov 1971; El 2003; El, Kamchatnov 2005; El, Kamchatnov, Pavlov, Zykov 2011]
28
-
'
&
$
%
5. Molecular dynamics and the effective velocity
Simple counting of the additional distance due to the jumps give
veffcl (v) = v +
∫dw d(v, w) ρcl(w) (v
effcl (v)− veffcl (w))
One can check numerically that it works (here Lieb-Liniger model):
−2 −1 0 1 2
−1
0
1
v
veff
29
-
'
&
$
%
II. Numerics and experiments
30
-
'
&
$
%
6. An example: quantum Newton cradle
[J.-S. Caux, BD, J. Dubail, R. Konik, T. Yoshimura, 2017]
We are now ready to write the full dynamics for the original problem.
We represent the Bragg pulse used in the original experiment. We calculate ρharmop (x, θ)
associated to e−β(H+VharmoN), and then set the initial condition to
ρp(x, 0; θ) =1
2
[ρharmop (x, θ + θBragg) + ρ
harmop (x, θ − θBragg)
].
as the effect of the Bragg pulse followed by fast local entropy maximization.
31
-
'
&
$
%
6. An example: quantum Newton cradle
32
-
'
&
$
%
6. An example: quantum Newton cradle
33
-
'
&
$
%
6. An example: quantum Newton cradle
Remarks:
• No thermalization, no steady states: Euler equations are reversible, and possess infinitelymany conserved quantities (including the entropy),∫
dxdθ ρpf(n) any function f(n), e.g. entropy n log n+ (1− n) log(1− n)
• Both “horizontal” and “vertical” integrations are measurable:∫
dθ ρp(x, t; θ) is the
particle density at x, t, and∫
dx ρp(x, t; θ) is the self-similar particle density seen
on ray x/t = vgroup(θ) at large times after taking away the longitudinal trap.
• Eventually Euler equations are broken by viscosity / dissipation and stationarity isreached (coarse graining of fine structure, has been related to chaotic trajectories in
hard-rod gas [Cao, Bulchandani, Moore 2017]). Integrability breaking effects of potential still not
perceptible (no thermalization).
34
-
'
&
$
%
7. Comparison with cold atom experiment: 1d expansion from a harmonic potential
[Bouchoule, BD, Dubail, Schemmer, in progress]
35
-
'
&
$
%
III. Further developments
36
-
'
&
$
%
8. Exact solutions
Consider the pratitioning protocol for generating non-
equilibrium steady states.
[Riemann 19th century; Spohn, Lebowitz 1977; Ruelle 2000; Tasaki
2000; Araki, Ho 2000; Aschbacher, Pillet 2003; Bernard, BD 2012]
The problem is invariant under x, t 7→ λx, λt: functions of the ray ξ = x/t:
(veff − ξ)∂ξn(ξ; θ) = 0.
In fluid dynamic terms, solution is a family (parametrised by θ) of contact singularities:
[Castro Alvaredo, BD, Yohimura & Bertini, Collura, De Nardis, Fagotti]
n(ξ; θ) = nL(θ)Θ(θ − θ?(ξ)) + nR(θ)Θ(θ? − θ(ξ)), veff(ξ; θ?(ξ)) = ξ.
Interpretations: Particles arrive either from the left
reservoir or the right reservoir depending on their
dressed velocity at the ray ξ.
37
-
'
&
$
%
8. Exact solutions
There is a geometry behind GHD. Because of jumps, there is effectively additional space
available for propagation of quasi-particles, that depends on the local state.
[BD, Spohn, Yoshimura 2017].
xAAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN2N2IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3H1FpHst7M0nQj+hQ8pAzaqzUeOqXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB5jmM/A==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN2N2IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3H1FpHst7M0nQj+hQ8pAzaqzUeOqXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB5jmM/A==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN2N2IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3H1FpHst7M0nQj+hQ8pAzaqzUeOqXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB5jmM/A==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN2N2IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3H1FpHst7M0nQj+hQ8pAzaqzUeOqXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB5jmM/A==
qAAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN3N0IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3n1BpHst7M0nQj+hQ8pAzaqzUeOyXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB252M9Q==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN3N0IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3n1BpHst7M0nQj+hQ8pAzaqzUeOyXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB252M9Q==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN3N0IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3n1BpHst7M0nQj+hQ8pAzaqzUeOyXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB252M9Q==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN3N0IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3n1BpHst7M0nQj+hQ8pAzaqzUeOyXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB252M9Q==
New coordinate q is related to real space x via density of states:
dq =2π ρs(x, t; θ)
p′(θ)dx
GHD equation becomes simple wave equation:
∂tn(q, t; θ) + vgroup(θ)∂qn(q, t; θ) = 0.
38
-
'
&
$
%
8. Exact solutions
Solution is immediate:
n(x, t; θ) = n(u(x, t; θ), 0; θ)
where the characteristics u(x, t; θ) solves∫ u
dy2π ρs(y, 0; θ)
p′(θ)=
∫ xdy
2π ρs(y, t; θ)
p′(θ)− vgroup(θ)t.
xAAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN2N2IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3H1FpHst7M0nQj+hQ8pAzaqzUeOqXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB5jmM/A==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN2N2IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3H1FpHst7M0nQj+hQ8pAzaqzUeOqXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB5jmM/A==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN2N2IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3H1FpHst7M0nQj+hQ8pAzaqzUeOqXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB5jmM/A==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN2N2IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3H1FpHst7M0nQj+hQ8pAzaqzUeOqXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB5jmM/A==
qAAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN3N0IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3n1BpHst7M0nQj+hQ8pAzaqzUeOyXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB252M9Q==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN3N0IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3n1BpHst7M0nQj+hQ8pAzaqzUeOyXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB252M9Q==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN3N0IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3n1BpHst7M0nQj+hQ8pAzaqzUeOyXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB252M9Q==AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0GPRi8cW7Ae0oWy2k3btZhN3N0IJ/QVePCji1Z/kzX/jts1BWx8MPN6bYWZekAiujet+O4W19Y3NreJ2aWd3b/+gfHjU0nGqGDZZLGLVCahGwSU2DTcCO4lCGgUC28H4dua3n1BpHst7M0nQj+hQ8pAzaqzUeOyXK27VnYOsEi8nFchR75e/eoOYpRFKwwTVuuu5ifEzqgxnAqelXqoxoWxMh9i1VNIItZ/ND52SM6sMSBgrW9KQufp7IqOR1pMosJ0RNSO97M3E/7xuasJrP+MySQ1KtlgUpoKYmMy+JgOukBkxsYQyxe2thI2ooszYbEo2BG/55VXSuqh6btVrXFZqN3kcRTiBUzgHD66gBndQhyYwQHiGV3hzHpwX5935WLQWnHzmGP7A+fwB252M9Q==
x
t
u(x, t; ✓)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
x, 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
ve↵(✓)AAAB/HicbVDLSgNBEJz1GeNrNUcvg0GIl7Argh6DXjxGMA9IYpid9CZDZh/M9AaWJf6KFw+KePVDvPk3TpI9aGJBQ1HVTXeXF0uh0XG+rbX1jc2t7cJOcXdv/+DQPjpu6ihRHBo8kpFqe0yDFCE0UKCEdqyABZ6Elje+nfmtCSgtovAB0xh6ARuGwhecoZH6dmnymHVVQMH3p5UujgDZed8uO1VnDrpK3JyUSY563/7qDiKeBBAil0zrjuvE2MuYQsElTIvdREPM+JgNoWNoyALQvWx+/JSeGWVA/UiZCpHO1d8TGQu0TgPPdAYMR3rZm4n/eZ0E/eteJsI4QQj5YpGfSIoRnSVBB0IBR5kawrgS5lbKR0wxjiavognBXX55lTQvqq5Tde8vy7WbPI4COSGnpEJcckVq5I7USYNwkpJn8krerCfrxXq3Phata1Y+UyJ/YH3+AEnSlIQ=AAAB/HicbVDLSgNBEJz1GeNrNUcvg0GIl7Argh6DXjxGMA9IYpid9CZDZh/M9AaWJf6KFw+KePVDvPk3TpI9aGJBQ1HVTXeXF0uh0XG+rbX1jc2t7cJOcXdv/+DQPjpu6ihRHBo8kpFqe0yDFCE0UKCEdqyABZ6Elje+nfmtCSgtovAB0xh6ARuGwhecoZH6dmnymHVVQMH3p5UujgDZed8uO1VnDrpK3JyUSY563/7qDiKeBBAil0zrjuvE2MuYQsElTIvdREPM+JgNoWNoyALQvWx+/JSeGWVA/UiZCpHO1d8TGQu0TgPPdAYMR3rZm4n/eZ0E/eteJsI4QQj5YpGfSIoRnSVBB0IBR5kawrgS5lbKR0wxjiavognBXX55lTQvqq5Tde8vy7WbPI4COSGnpEJcckVq5I7USYNwkpJn8krerCfrxXq3Phata1Y+UyJ/YH3+AEnSlIQ=AAAB/HicbVDLSgNBEJz1GeNrNUcvg0GIl7Argh6DXjxGMA9IYpid9CZDZh/M9AaWJf6KFw+KePVDvPk3TpI9aGJBQ1HVTXeXF0uh0XG+rbX1jc2t7cJOcXdv/+DQPjpu6ihRHBo8kpFqe0yDFCE0UKCEdqyABZ6Elje+nfmtCSgtovAB0xh6ARuGwhecoZH6dmnymHVVQMH3p5UujgDZed8uO1VnDrpK3JyUSY563/7qDiKeBBAil0zrjuvE2MuYQsElTIvdREPM+JgNoWNoyALQvWx+/JSeGWVA/UiZCpHO1d8TGQu0TgPPdAYMR3rZm4n/eZ0E/eteJsI4QQj5YpGfSIoRnSVBB0IBR5kawrgS5lbKR0wxjiavognBXX55lTQvqq5Tde8vy7WbPI4COSGnpEJcckVq5I7USYNwkpJn8krerCfrxXq3Phata1Y+UyJ/YH3+AEnSlIQ=AAAB/HicbVDLSgNBEJz1GeNrNUcvg0GIl7Argh6DXjxGMA9IYpid9CZDZh/M9AaWJf6KFw+KePVDvPk3TpI9aGJBQ1HVTXeXF0uh0XG+rbX1jc2t7cJOcXdv/+DQPjpu6ihRHBo8kpFqe0yDFCE0UKCEdqyABZ6Elje+nfmtCSgtovAB0xh6ARuGwhecoZH6dmnymHVVQMH3p5UujgDZed8uO1VnDrpK3JyUSY563/7qDiKeBBAil0zrjuvE2MuYQsElTIvdREPM+JgNoWNoyALQvWx+/JSeGWVA/UiZCpHO1d8TGQu0TgPPdAYMR3rZm4n/eZ0E/eteJsI4QQj5YpGfSIoRnSVBB0IBR5kawrgS5lbKR0wxjiavognBXX55lTQvqq5Tde8vy7WbPI4COSGnpEJcckVq5I7USYNwkpJn8krerCfrxXq3Phata1Y+UyJ/YH3+AEnSlIQ=
39
-
'
&
$
%
9. Zero-entropy GHD
[BD, J. Dubail, R. Konik, T. Yoshimura, Phys. Rev. Lett. 119, 195301 (2017)]
GHD simplifies drastically in the zero-entropy subspace
Let the occupation function n(θ) be formed of k disjoint, filled Fermi seas. Such occupation
functions have zero entropy. It turns out that the GHD equations simplify to a
finite-component hydrodynamics for the Fermi seas’ endpoints:
∂tθ±j + v
eff{θ}(θ
±j )∂xθ
±j = 0.
We refer to this as “2kHD”, for k = 1, 2, 3, . . .
40
-
'
&
$
%
9. Zero-entropy GHD
Consider 2HD. A single Fermi sea: this is what we get in a zero temperature initial state.
This is a 2-component, Galilean invariant hydrodynamics. Hence the only possibility is
conventional hydrodynamics! That is,
2HD = conventional hydrodynamics (CHD)
∂tn + ∂x(vn) = 0, ∂tv + v∂xv = −1
mn∂xP, P = P(v, n).
Indeed a single Fermi sea is determined by fixing the density of particles and the boost
velocity. So the local density matrices have the form e−β(H−µN−νP )|β→∞.
41
-
'
&
$
%
9. Zero-entropy GHD
Comparison with ABACUS quantum evolution:
42
-
'
&
$
%
9. Zero-entropy GHD
In CHD, one usually encounters shocks that are sustained over time. At shocks, entropy is
continuously produced (the production of entropy at shocks is described by higher-derivative
terms neglected in Euler hydrodynamics).
However, in GHD, shocks are not sustained. Gradient catastrophes arise in 2HD, but
immediately “dissolve,” thanks to the availability of a large fluid space, into 4HD.
43
-
'
&
$
%
9. Zero-entropy GHD
Remarks:
• CHD has been used in the past in order to describe the Lieb-Liniger model ininhomogeneous situations. It was partially successful. It has been observed that both at
collision of clouds and in propagation of density waves, shocks appear, beyond which CHD
is unable to reproduce tDMRG numerics. Here we have a full explanation of why CHD
works at zero temperature based on hydrodynamic principle only (local entropy
maximization), and exact hydrodynamic equations for describing what happens after the
gradient catastrophes.
• In real systems, exact gradient catastrophes do not ever occur: higher derivative terms(beyond Euler hydrodynamics) become important before. However, at very large scales
(Euler scales), gradients will look very sharp. GHD captures well what happens at such
scales.
44
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
But we would like to go beyond averages of single observables. It turns out that Euler
hydrodynamics gives access to scaled, fluid-cell averaged dynamical correlation functions
〈O1(x1, t1) · · · ON (xN , tN )〉Eul
= limλ→∞
λN−1 〈Ō1(λx1, λt1) · · · ŌN (λxN , λtN )〉cλ.
The parameter λ characterize the fluid cell scale, and
〈O〉λ =Tr(e−
∫R dx
∑i βi(λ
−1x)qi(x)O)
Tr(e−
∫R dx
∑i βi(λ
−1x)qi(x))
The observables Ō(x, t) are appropriate fluid-cell averages
Ō(x, t) =∫
Nλ(x,t)
dyds
|Nλ|O(y, s).
45
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
46
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
Linear fluctuating hydrodynamics and hydrodynamic projection operators
[BD, Spohn 2017]
The C operator.
A natural inner product is given by the bilinear form
〈O1|O2〉 =∫
dx 〈O1(x)O2(0)〉c =∫
dx 〈O1(x)O2(0)〉Eul
(after moding out the null space). A Hilbert space is obtained by completion.
The C operator of linear fluctuating hydrodynamics is defined by using conserved densities
Cij = 〈qi|qj〉
The qi span a sub-Hilbert space, that of pseudolocal charges. The C operator is simple to
calculate within TBA: we just have to differentiate with respect to βi the GGE average qj :
−∂qj∂βi
= − ∂∂βi
Tr(e−
∑k βkQkqj(0)
)
Tr(e−
∑k βkQk
) = 〈Qiqj(0)〉c = Cij
47
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
For this, we use the pseudo-energy of TBA, which says how the state depends on βi:
n(θ) =1
1 + e�(θ), �(θ) =
∑
i
βihi(θ)−∫
dα
2πϕ(θ, α) log(1 + e−�(α))
After some calculations this gives
Cij =
∫dθ ρp(θ) (1− n(θ))hdri (θ)hdrj (θ)
where the “dressing” operation is defined by solving
hdr(θ) = h(θ) +
∫dα
2πϕ(θ, α)n(α)hdr(α)
48
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
We can re-write this as a an integral operator on the spectral space
C = (1− nT )−1ρp(1− n)(1− Tn)−1
in the sense that
Cij =
∫dθ hi(θ)(Chj)(θ) =
∫dθdαC(θ, α)hi(θ)hj(α)
The integral operator T is
Th(θ) =
∫dθ
2πϕ(θ, α)h(α)
We see that hdr = (1− Tn)−1h.
49
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
The B operator.
The B operator is defined using conserved currents
Bij = 〈ji|qj〉.
From general principles, it is symmetric, BT = B. Again it can be obtained by differentiating:
Bij = −∂jj∂βi
.
Using our known expression of the GGE average jj we get
Bij =
∫dθ ρp(θ) (1− n(θ)) veff(θ)hdri (θ)hdrj (θ)
That is
B = (1− nT )−1veffρp(1− n)(1− Tn)−1
50
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
The Drude weight (Drude matrix).
The Drude weight is the matrix
Dij = limt→∞
∫dx 〈ji(x, t)jj(0, 0)〉c = 〈ji(t)|jj(0)〉
We realise that limt→∞〈ji(t)| projects onto the subspace of conserved quantities:
Dij =m∑
i′,j′=1
〈ji|qi′〉(C−1)i′j′〈qj′ |jj〉 that is D = BC−1B
This can be evaluated easily in the integral operator language
D = (1− nT )−1(veff)2ρp(1− n)(1− Tn)−1
giving
Dij =
∫dθ ρp(θ) (1− n(θ)) veff(θ)2hdri (θ)hdrj (θ)
51
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
The A operator.
In order to obtain space-time dependent Euler scale correlations, we may use the evolution
operator A. This is defined as the derivative of the equations of state
Aij =∂ji∂qj
.
The quantity 〈ji(x, t)qj(0, 0)〉Eul equals
−δji(x, t)δβj(0)
= −∑
k
δqk(x, t)
δβj(0)
δji(x, t)
δqk(x, t)=∑
k
〈qk(x, t)qj(0, 0)〉EulAik
Taking t = 0 and integrating we find the (known) relation
B = AC
and so we obtain
A = (1− nT )−1veff(1− nT )
52
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
The two-point functions of conserved densities.
Therefore we have
∂t〈qi(x, t)qj(0, 0)〉Eul + ∂x∑
k
Aik〈qk(x, t)qj(0, 0)〉Eul = 0
In Fourier space,
Sij(k, t) =
∫dx eikx〈qi(x, t)qj(0, 0)〉Eul = lim
k→0, t→∞kt fixed
∫dx eikx〈qi(x, t)qj(0, 0)〉c
Hence
∂tS = ikAS, S(k, 0) = C
The solution is
S(k, t) = eiktAC.
53
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
That is,
Sij(k, t) =
∫dθ eiktv
eff (θ)ρp(θ) (1− n(θ))hdri (θ)hdrj (θ)
Back to real space,
〈qi(x, t)qj(0, 0)〉Eul =∫
dθ δ(x− veff(θ)t) ρp(θ) (1− n(θ))hdri (θ)hdrj (θ)
54
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
More general two-point functions. [BD 2017]
Hydrodynamic projection principles allow us to generalize to generic fields:
〈O′(x, t)O(0, 0)〉Eul =∑
i,j,k,l
〈O′|qi〉C−1ij 〈qj(x, t)qk(0, 0)〉C−1kl 〈ql|O〉
55
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
We obtain the necessary overlaps by differentiating GGE averages:
〈qi|O〉 = −∂〈O〉∂βi
=:
∫dθ ρp(θ) (1− n(θ))V O(θ)hdri (θ)
This gives
〈O′(x, t)O(0, 0)〉Eul =∫
dθ δ(x− veff(θ)t) ρp(θ) (1− n(θ))V O′(θ)V O(θ)
For instance,
V qi = hdri , Vji = veff hdri
56
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
In the Lieb-Liniger model (with m = 1/2) we may take Pozsgay’s formulae [Pozsgay 2011] for
the Kth power of the particle density,
OK =1
(K!)2(Ψ†)K(Ψ)K
We find in a state defined by the occupation function n(p) (with p = θ)
V OK (p) =K∑
s=1
∫
RK−1
(K∏
r=1r 6=s
dpr2π
n(pr)hdrr (pr)
)hdrs (p) g
drs (p)
1dr(p)
where
hr(p) = pr−1, r = 1, 2, 3, . . .
and
gs(ps) =∏
j≥l
pj − pl(pj − pl)2 + (c/2)2
(defined as a function of ps with pr 6=s fixed parameters).
57
-
'
&
$
%
10. Correlation functions and linear fluctuating hydrodynamics
All results also apply to classical integrable field theories. We look at sinh-Gordon model
[Bastianello, BD, Watts, Yoshimura 2017].
We evaluate numerically by Monte Carlo simulation the following two-point functions and
compare with our formulae (g is coupling):
1
t
∫ t
0
ds s 〈Tµµ (ζs, s)T νν (0, 0)〉 and1
t
∫ t
0
ds s 〈cosh(2gΦ(ζs, s)) cosh(2gΦ(0, 0))〉
Numerics
Analytic
Free
(a)
-1 0 10
0.05
0.1
0.15
ζ
〈Tμ μTν ν〉
Numerics
Analytic
(b)
-1 0 1
0
100
200
300
ζ
〈Ch(2)Ch(2)〉
Figure 6: The predictions (black curves) for the time average of two-point functions of(a) the trace of the stress-energy tensor, (46), and (b) vertex operators, (45) with (48),are compared to the numerical simulation (red curves). In the case (a) the free result ofAppendix B (dotted curve) is plotted for comparison. In order to improve the convergence,short times are avoided: the time average is taken in the interval [t0, t], rather than [0, t]as in eq. (46), with t0 = 5 and t = 20. The same parameters as those of Figure 5 areused.
NumericsAnalytic
(a)
0 4 8 12 16 200
0.2
0.4
Time
〈Tμ μTν ν〉
NumericsAnalytic
(b)
0 4 8 12 16 200
500
1000
Time
〈Ch(2)Ch(2)〉
Figure 7: The predictions (black curves) for the space integral of two-point functions of(a) the trace of the stress-energy tensor, (47), and (b) vertex operators, (28) with (48),are compared to the numerical simulation (red curves) as functions of time. At largetimes, the oscillating ripples are due to the Metropolis noise and are observed to reduceon increasing the number of samples. The same parameters m, �, g as those of Figure 5are used.
V k(✓) is given by
V k+1(✓) =g2
4⇡Vk+1
kX
l=0
V lV l+1 (2l + 1)n(✓)
pl(✓)dl(✓)
⇢p(✓)(48)
23
58
-
'
&
$
%
11. Large deviation theory
[Myers, Bhaseen, Harris, BD in progress]
The statistics of the charge Qi being transported from the left to the right:
Prob(transfer of amount Q of charge Qi in time t ) ∼ e−tIi(Q)
where Ii(Q) is the large deviation function. Legendre transform is scaled cumulant
generating function for the integrated current at 0:
Fi(λ) =∞∑
n=1
cnn!λn = lim
t→∞t−1 log 〈 exp
[λ
∫ t
0
ji(0, s) ds
]〉
[free fermions: Levitov-Lesovik; quantum impurities: Saleur et al.; CFT: Bernard, BD]
59
-
'
&
$
%
11. Large deviation theory
Main idea: current integral gives λ-dependent homogeneous, stationary states (GGE)
〈exp[λ∫∞−∞ ji(0, s)ds
]O〉
〈exp[λ∫∞−∞ ji(0, s)ds
]〉
= 〈O〉GGEλ
We used
dλ〈O〉GGEλ =
∫ ∞
−∞ds 〈ji(0, s)O〉cGGEλ
and our formulae for one- and two-point functions to get
∂nλ∂λ
= sign(veff)nλ(1− nλ)hdri,λ
Get an expression for Fi(λ), which can be evaluated numerically, and expressions for the
cumulants cn.
60
-
'
&
$
%
11. Large deviation theory
Numerical checks in the hard rod gas:
Then by defining
n±(1 � a⇢±)�1f±(✓) (3.49)
we can show that this hydro equation with the initial conditions 3.46 is solved by:
n(⇠; ✓) = n�(✓)⇥(✓ � ✓?(⇠)) + n+(✓)⇥(�✓ + ✓?(⇠)) (3.50)
where,
veff (⇠; ✓?) = ⇠ (3.51)
The key thing to notice is that it seems there is ambiguity coming from the ambiguity inherent in⇥(0). However consider the physical situation and ensure we are consistent in the choice of conventionof the ⇥ function. If ⇥(0) = 0 then somehow, in the steady state, the particle density between thetwo connections must drop to zero. If ⇥(0) = 1 then we somehow have an extra dense point in thesteady state of the system. However with the choice ⇥(0) = 12 we get an average of the left and rightparticle densities which seems to physically be the most reasonable.
3.7 Results
Figure 1: Plot of moments scaled in time. These are for 1000000 samplings. There are 45000 particlesin this simulation.
4 The Full SCGF
Consider the first moment:
c1 =
Zd✓
2⇡E0(✓)n(✓)H(✓) (4.1)
16
Then by defining
n±(1 � a⇢±)�1f±(✓) (3.49)
we can show that this hydro equation with the initial conditions 3.46 is solved by:
n(⇠; ✓) = n�(✓)⇥(✓ � ✓?(⇠)) + n+(✓)⇥(�✓ + ✓?(⇠)) (3.50)
where,
veff (⇠; ✓?) = ⇠ (3.51)
The key thing to notice is that it seems there is ambiguity coming from the ambiguity inherent in⇥(0). However consider the physical situation and ensure we are consistent in the choice of conventionof the ⇥ function. If ⇥(0) = 0 then somehow, in the steady state, the particle density between thetwo connections must drop to zero. If ⇥(0) = 1 then we somehow have an extra dense point in thesteady state of the system. However with the choice ⇥(0) = 12 we get an average of the left and rightparticle densities which seems to physically be the most reasonable.
3.7 Results
Figure 1: Plot of moments scaled in time. These are for 1000000 samplings. There are 45000 particlesin this simulation.
4 The Full SCGF
Consider the first moment:
c1 =
Zd✓
2⇡E0(✓)n(✓)H(✓) (4.1)
16
61
-
'
&
$
%
12. Diffusion and viscosity
[De Nardis, Bernard, BD in progress]
The diffusion matrix Dij is defined via the Kubo formula (which involves subtracting the Drude
matrix Dij )
(DC)ij =
∫
Rdt
(∫
Rdx 〈ji(x, t)jj(0, 0)〉c −Dij
)
This changes the evolution of two-point functions as
∂tSij(x, t) +∑
k
(Aik∂x ∓
1
2Dik∂
2x
)Skj(x, t) = 0 (t ≷ 0)
or, in Fourier transform,
S(k, t) = exp
[−ikAt− 1
2k2D|t|
]C
It also adds a “Navier-Stokes” term to the GHD equation
∂tρp(θ) + ∂x(veff(θ)ρp(θ)) =
1
2∂x
∫dαD(θ, α)∂xρp(α)
62
-
'
&
$
%
12. Diffusion and viscosity
We find an exact expression for the diffusion operator
(DC)ij =
∫dθdαK(θ, α)hdri (θ)h
drj (α)
with
K(θ, α) = g(θ)w(θ)δ(θ − α)− g(θ)U(θ, α)g(α), g(θ) = ρp(θ)(1− n(θ))
where
U(θ, α) =ϕdr(θ, α)ϕdr(α, θ)
(2πρs(θ))(2πρs(α))|veff(θ)− veff(α)|
w(θ) =
∫dα g(α)
ϕdr(θ, α)2
(2πρs(θ))2|veff(θ)− veff(α)|
63
-
'
&
$
%
12. Diffusion and viscosity
Remarks:
• The operator DC is symmetric and positive, as it should.
• There is positive Yang-Yang entropy production:
S = −∫
dθdx ρs (n log n+ (1− n) log(1− n))
satisfies
∂tS = appropriate diagonal matrix element (DC)ii > 0
• Corresponding formulae for quantum chains, classical gases and field theories, andcorrectly specialises to the known diffusion operator in the hard rod gas [Boldrighini, Suhov 97].
64
-
'
&
$
%
12. Diffusion and viscosity
Main steps of calculation:
• Expansion into intermediate excitations of particles and holes in the GGE background:
〈ji(x, t)jj(0, 0)〉c
=
∫dθpdθhe
i(kp−kh)x−i(�p−�h)t〈GGE|ji|θpθh〉〈θpθh|jj |GGE〉
+
∫dθ1,2p dθ
1,2h e
i∑2a=1(k
ap−k
ah)x−i(�
ap−�
ah)t〈GGE|ji|θ1,2p θ1,2h 〉〈θ1,2p θ
1,2h |jj |GGE〉
+ · · ·
• Intergal over x gives δ(∑(kp − kh)). First line (one particle-hole pair) gives Drude weight,subtracted.
• Use of conservation equations:
〈GGE|ji|θ1,2,...p θ1,2,...h 〉 =∑
a
(�ap − �ah) fi(θ1,2,...p θ1,2,...h )
where fi has simple structure, with simple “kinematical poles” at θap = θ
bh.
65
-
'
&
$
%
12. Diffusion and viscosity
• Intergal over t gives δ(∑(�p − �h)).• At two particle-hole pairs: the conditions of conservation of energy and momentum impliesconservation of the set of momenta:
2∑
a=1
kap =2∑
a=1
kah,2∑
a=1
�2p =2∑
a=1
�ah ⇒ {θap} = {θah}
This specialises to the pole of fi, cancelling the factor (�ap − �ah)2.
• At higher particle-hole pairs: the conditions don’t fix the set, hence the factor (�ap − �ah)2dominates and makes the contribution vanish.
• General properties of integrable systems fix the kinematical poles.
Main message, interpreting holes as outgoing particles: two-body processes give rise to
diffusion, with amplitude given by two particle-hole pair matrix element of conserved currents.
66
-
'
&
$
%
Conclusions
We have obtained a set of hydrodynamic equations describing the dynamics at large scales
(Euler scale) in integrable systems of all types, and the diffusive (Navier-Stokes) correction.
This is a theoretical framework “solving” the famous quantum Newton cradle experiment, and
giving exact results for a wealth of problems in non-equilibrium dynamics of integrable systems.
• GHD equations in the quasi-particle language have wide applicability:– quantum chains and quantum field theory;
– classical gases such as the hard rod gas and soliton gases;
– classical integrable field theory and most likely classical integrable chains as well.
• We also found formulae for Euler-scale two-point functions in inhomogeneous,non-stationary states [BD 2017].
Still to do: superdiffusion, subdiffusion; higher-derivative corrections; integrability breaking and
emergence of conventional hydrodynamics from GHD, ...
67
-
'
&
$
%
...
68