Holographic Deep Thermalization

• Holographic deep thermalization is a phenomenon where boundary observables equilibrate nearly instantaneously, governed by deep bulk structures such as black-brane horizons, quasinormal modes, and trapped surfaces.
• Methodologies include pump–probe setups in finite-density strange metals and shockwave-induced trapped-surface formation in confining models, emphasizing the control of slow vector modes and intermediate bulk regions.
• Practical insights reveal that thermalization times are dictated by the probe’s depth and pump spectral content, providing actionable criteria for analyzing equilibration in various holographic scenarios.

“Holographic deep thermalization” denotes a class of holographic equilibration phenomena in which the decisive dynamics are controlled by structures deep in the bulk geometry: black-brane horizons, low-lying quasinormal modes, extremal surfaces, or trapped surfaces. In the literature represented here, the expression is used most explicitly for pump–probe optical conductivity in finite-density strange metals, where the striking result is that the conductivity can assume its thermal value essentially immediately after the pump ends if the pump does not efficiently excite the slow vector mode (Bagrov et al., 2017). The same expression is also used for short-time trapped-surface formation in confining backgrounds, where thermalization occurs in an intermediate bulk window rather than strictly near the UV boundary (Ageev et al., 2014). Across related works on shell collapse, Lifshitz and hyperscaling-violating quenches, and finite-coupling corrections, the common theme is that equilibration is controlled by how boundary observables couple to the slowest bulk sector (Balasubramanian et al., 2011, Baron et al., 2012, Keranen et al., 2011, Fonda et al., 2014, Stricker, 2014).

1. Conceptual scope

In the pump–probe strange-metal setting, “deep thermalization” refers to the effectively instantaneous thermalization of the optical conductivity after energy injection, despite the system having been driven far from equilibrium (Bagrov et al., 2017). The mechanism is not the absence of dynamics, but the suppression of overlap with the low-lying quasinormal mode that carries slow momentum relaxation. In that sense, the adjective “deep” points to the bulk origin of the result: the observable is controlled by whether the pump excites a specific sector of the black-brane perturbation spectrum.

In confining heavy-ion-inspired models, the same expression is used differently. There, “deep” refers to trapped-surface formation in an intermediate radial region of the bulk, $z \in [1.3~\mathrm{fm},1.8~\mathrm{fm}]$, together with a short hydrodynamization time $\tau_{\rm th}\sim 0.25~\mathrm{fm}$ and an entropy scaling $s\propto E^{1/3}$ in that window (Ageev et al., 2014). This suggests that the expression is not a single standardized label, but a family resemblance: bulk depth, horizon formation, and slow-mode control recur across otherwise distinct holographic thermalization problems.

A second recurring distinction is between observables. Local one-point functions, retarded two-point functions, heavy-operator correlators, Wilson loops, and entanglement entropy need not thermalize on the same timescale. Several works emphasize that larger or more nonlocal probes extend deeper into the bulk and therefore thermalize later, while specific observables tied to near-horizon or retarded dynamics can thermalize effectively instantaneously (Balasubramanian et al., 2011, Ebrahim et al., 2010).

2. Pump–probe strange metals and the holographic model

The most explicit formulation of holographic deep thermalization studies a spatially homogeneous, finite-density $2+1$D strange metal with mildly broken translational symmetry, driven by a time-dependent electric field pulse and then probed through the optical conductivity $\sigma(\omega,t)$ (Bagrov et al., 2017). The pump is an in-plane electric field along $x$ with Gaussian envelope and central frequency $\omega_P$,

$$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$

with a smoothed shutoff at $t_{\rm end}=t_0+3\Delta t$. Representative parameters are $\mu=1$, $\tau_{\rm th}\sim 0.25~\mathrm{fm}$0, $\tau_{\rm th}\sim 0.25~\mathrm{fm}$1, $\tau_{\rm th}\sim 0.25~\mathrm{fm}$2, $\tau_{\rm th}\sim 0.25~\mathrm{fm}$3, and amplitudes chosen so that $\tau_{\rm th}\sim 0.25~\mathrm{fm}$4 is raised to $\tau_{\rm th}\sim 0.25~\mathrm{fm}$5.

The bulk dual is a charged Reissner–Nordström black brane in AdS$\tau_{\rm th}\sim 0.25~\mathrm{fm}$6 with linear axions, namely the Einstein–Maxwell–axion model

$\tau_{\rm th}\sim 0.25~\mathrm{fm}$7

The static background is

$\tau_{\rm th}\sim 0.25~\mathrm{fm}$8

$\tau_{\rm th}\sim 0.25~\mathrm{fm}$9

$s\propto E^{1/3}$0

Here $s\propto E^{1/3}$1 controls momentum relaxation and $s\propto E^{1/3}$2 acts as an elastic mean free path. The equilibrium thermodynamics are

$s\propto E^{1/3}$3

$s\propto E^{1/3}$4

$s\propto E^{1/3}$5

At weak momentum relaxation, the equilibrium optical conductivity displays a Drude-like peak with relaxation time

$s\propto E^{1/3}$6

The conductivity is defined in equilibrium by the Kubo formula

$s\propto E^{1/3}$7

After the pump, the relevant response is the differential conductivity

$s\propto E^{1/3}$8

and a useful time-dependent frequency-space object is

$s\propto E^{1/3}$9

The computation proceeds by solving the fully time-dependent pumped bulk geometry, then evaluating linearized Maxwell–Einstein–axion perturbations on that background with ingoing horizon conditions and standard holographic renormalization (Bagrov et al., 2017).

3. Quasinormal modes, instantaneous response, and slow ringdown

The central result is a sharp dichotomy. At zero density, the dual geometry reduces to a Vaidya spacetime, and causality implies that one-point functions and retarded two-point functions probed after the quench coincide with their thermal values. In this regime, $2+1$0 assumes its post-pump equilibrium value immediately after the pumping has ended (Bagrov et al., 2017).

At finite density with mild translation breaking, the same effectively instantaneous behavior persists for oscillating pump pulses without significant DC content. By contrast, pulses with a sizeable DC component excite the slow vector quasinormal mode and produce a clean exponential ringdown. The DC conductivity obeys

$2+1$1

The lowest vector quasinormal mode has

$2+1$2

and in the regime studied $2+1$3, so the relaxation is purely relaxational rather than oscillatory. Because the metric component associated with momentum enters the conductivity effectively squared, the conductivity relaxes with

$2+1$4

hence $2+1$5 (Bagrov et al., 2017).

The dependence on pump shape is equally explicit. A significant DC component overlaps strongly with the lowest vector mode and generates net momentum. For oscillatory pumps with large central frequency $2+1$6 and smooth envelope, the vector perturbation induced during pumping satisfies

$2+1$7

and after the pulse it is suppressed more strongly than any power of $2+1$8. For a Gaussian of width $2+1$9,

$\sigma(\omega,t)$0

Accordingly, the residual deviation amplitude follows

$\sigma(\omega,t)$1

The practical criterion is therefore

$\sigma(\omega,t)$2

together with a smooth envelope. In this regime the low-lying mode is excited only exponentially weakly, and $\sigma(\omega,t)$3 is indistinguishable, within numerical resolution, from the thermal $\sigma(\omega,t)$4 at $\sigma(\omega,t)$5 immediately after the pump (Bagrov et al., 2017).

A common misconception is that “deep thermalization” means universal instantaneous equilibration. The conductivity result is more specific: it depends on the pump’s spectral weight at the slowest vector quasinormal mode. The same work states explicitly that extensions to other observables “likely follow the same slow-mode logic but must be checked case by case” (Bagrov et al., 2017).

4. Deep bulk formation in confining and collision geometries

A different use of the expression arises in confining holographic models of heavy-ion collisions, where thermalization is identified with trapped-surface formation after colliding domain shock waves (Ageev et al., 2014). The bulk metric is taken in Einstein frame as

$\sigma(\omega,t)$6

with a confining warp factor chosen to reproduce the Cornell potential and QCD $\sigma(\omega,t)$7-function behavior. In the explicit trapped-surface analysis, the paper uses

$\sigma(\omega,t)$8

with $\sigma(\omega,t)$9, $x$0, and $x$1.

The key physical assumption is that trapped-surface formation occurs during a short time window, which restricts the relevant radial range to an intermediate domain. In that window, $x$2, the confining geometry is well approximated by

$x$3

Shock-wave collisions with source strength

$x$4

form a trapped surface between $x$5 and $x$6 satisfying

$x$7

and the entropy density is

$x$8

In the intermediate window this yields the asymptotic scaling

$x$9

in contrast to the conformal AdS result $\omega_P$0 (Ageev et al., 2014).

The corresponding thermalization time is estimated from the trapped interval,

$\omega_P$1

and for $\omega_P$2, $\omega_P$3 one finds

$\omega_P$4

Here “deep thermalization” therefore refers to short-time dynamics governed by an intermediate, relatively deep region of a confining geometry rather than to instantaneous equilibration of a retarded transport coefficient (Ageev et al., 2014).

5. Nonlocal probes, scale dependence, and generalized scaling laws

A large part of the holographic thermalization literature studies nonlocal probes whose extremal curves or surfaces penetrate progressively deeper into the bulk as their boundary size increases. In the thin-shell AdS–Vaidya analysis of two-point functions, Wilson loops, and entanglement entropy, the universal picture is: a slight delay in the onset of thermalization, an apparent non-analyticity at the endpoint of thermalization, and top-down thermalization in which the UV thermalizes first (Balasubramanian et al., 2011). For homogeneous quenches, entanglement entropy thermalizes slowest and sets a timescale that saturates a causality bound,

$\omega_P$5

whereas two-point functions and Wilson loops can thermalize slightly faster (Balasubramanian et al., 2011).

The thin-shell study with general equations of state makes the same dependence on bulk depth more explicit. Equal radii shells with scalar-field matter $\omega_P$6 and dust $\omega_P$7 thermalize almost as fast as AdS–Vaidya, while conformal matter $\omega_P$8 exhibits markedly larger $\omega_P$9, and the delay grows with spacetime dimension (Baron et al., 2012). When the inner and outer cosmological constants differ, only the scalar-field shell collapses generically; other shells expand unless the radii are tuned extremely close. This identifies shell microphysics, not only bulk causal depth, as a control parameter for deep-probe thermalization (Baron et al., 2012).

Nonrelativistic generalizations preserve the same structural idea. In Lifshitz–Vaidya backgrounds, the thermalized region still propagates with a finite front velocity despite the boundary theory lacking a relativistic speed limit. The analytic upper bounds are

$$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$0

for two-point functions, and

$$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$1

for entanglement entropy (Keranen et al., 2011). In hyperscaling-violating Lifshitz backgrounds, the early-time HEE growth becomes

$$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$2

while the deep linear regime obeys

$$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$3

with $$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$4 determined by $$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$5, $$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$6, and the effective dimension $$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$7 (Fonda et al., 2014). These results generalize the notion of deep thermalization from relativistic CFTs to anisotropic and hyperscaling-violating fixed points without abandoning the central role of near-horizon geometry and extremal-surface depth.

6. Limits, misconceptions, and open issues

Several developments delimit the scope of the strongest versions of deep thermalization. First, finite-coupling corrections weaken the strict strong-coupling top-down picture. In $$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$8 SYM with leading $$E_x(t)=A\cos(\omega_P t)e^{-(t-t_0)^2/(\Delta t)^2},$$9 corrections, the quasinormal-mode towers bend toward the real axis as $t_{\rm end}=t_0+3\Delta t$0 decreases, meaning smaller damping rates $t_{\rm end}=t_0+3\Delta t$1 and longer-lived excitations; the spectral deviations $t_{\rm end}=t_0+3\Delta t$2 then turn up at high $t_{\rm end}=t_0+3\Delta t$3, especially for nearly on-shell probes (Stricker, 2014). This does not eliminate deep thermalization, but it weakens the UV-first pattern.

Second, immediate thermalization beyond one-point functions is possible but not generic. In AdS$t_{\rm end}=t_0+3\Delta t$4–Vaidya, the Brownian-motion correlator of a probe string endpoint thermalizes instantly after the shell, and the fluctuation–dissipation theorem holds immediately for $t_{\rm end}=t_0+3\Delta t$5 (Ebrahim et al., 2010). That result shows that some $t_{\rm end}=t_0+3\Delta t$6 observables can thermalize as fast as the geometry permits, yet it does not imply that general nonlocal observables behave similarly.

Third, chemical potential and higher-curvature corrections can pull in opposite directions. In AdS–Reissner–Nordström–Vaidya, the entanglement entropy thermalizes latest, and the dependence of thermalization time on $t_{\rm end}=t_0+3\Delta t$7 is non-monotonic for $t_{\rm end}=t_0+3\Delta t$8 but becomes monotonic for $t_{\rm end}=t_0+3\Delta t$9 (Caceres et al., 2012). In Einstein–Maxwell–Gauss–Bonnet gravity, larger Gauss–Bonnet coefficient shortens the thermalization time, while larger charge lengthens it; the thermalization velocity is non-monotonic, and a phase transition point separates acceleration and deceleration phases (1311.0718).

The pump–probe strange-metal paper frames the broader methodological question directly: do these results reflect artefacts of the large-$\mu=1$0 limit, or do they possess enough IR/UV independence to remain qualitatively robust in experiment (Bagrov et al., 2017)? The literature surveyed here supports neither an unqualified yes nor an unqualified no. What it does support is a sharper statement: holographic thermalization is governed by bulk causal structure, slow-mode content, and probe depth, and “deep thermalization” names the regime in which those bulk mechanisms become directly visible in the boundary observable.