Bouncing Singularities: Dynamics & Applications

• Bouncing singularities are singular configurations that serve as turning points or matching rules rather than terminal breakdowns, appearing in holography, cosmology, BKL transitions, and singular ODEs.
• In holography and black-hole perturbation theory, they manifest as complex-time singularities from bouncing geodesics, offering insights into thermal correlator behavior.
• In cosmology and BKL analysis, they underpin bounce transitions and elastic impact laws, providing frameworks for resolving or traversing big-bang and Kasner singularities.

In recent literature, “bouncing singularities” is not a single universally fixed term but a family of related constructions. In holography and black-hole perturbation theory, it denotes singularities of analytically continued thermal or retarded correlators produced by null geodesics that cross horizons, reach a singular region, and return; in cosmology, it denotes either the replacement of a big-bang singularity by a classical bounce or the imposition of effective scattering laws across a singular hypersurface; in BKL analysis, it refers to Kasner transitions near spacelike singularities; and in singular ODEs it denotes elastic impact solutions at a repulsive singularity (Dodelson et al., 12 Nov 2025, Grozdanov et al., 16 Mar 2026, Arnaudo et al., 15 May 2026, 1803.01961, LeFloch, 2021, Li, 2024, Rojas et al., 2020).

1. Terminological scope and principal meanings

The phrase appears in at least five technically distinct senses in the cited literature. In AdS/CFT, a “bouncing singularity” is a complex-time singularity of a boundary correlator associated with a bulk null geodesic that “bounces” off a black-hole singularity (Dodelson et al., 12 Nov 2025). In asymptotically flat Schwarzschild perturbation theory, the same term is used for singularities of the analytically continued retarded Green’s function caused by a null curve that leaves a source point, hits the black-hole singularity, and re-emerges (Arnaudo et al., 15 May 2026). In singularity-scattering approaches to cosmology, a “bouncing singularity” is a spacelike curvature singularity across which past asymptotic data are mapped to future asymptotic data by a singularity scattering map (LeFloch, 2021, LeFloch, 2021). In BKL theory, bounces are instability-driven Kasner transitions near a spacelike singularity (Li, 2024). In singular ODEs, the term refers to periodic solutions that hit a weak repulsive singularity and continue by elastic reflection (Rojas et al., 2020).

A useful synthesis is that all these usages involve a singular structure that is not treated as a terminal endpoint of the analysis. Instead, it becomes either a source of non-analyticity in observables, a locus across which asymptotic data are matched, or a turning point in an effective dynamical description. This suggests a common editorial characterization: a “bouncing singularity” is a singular configuration that participates in a rule of continuation rather than only marking breakdown. That characterization is interpretive; the precise meaning remains model-dependent.

2. Holographic black holes: bouncing geodesics and complex-time singularities

In the classical AdS/CFT limit of large $N$ and infinite ’t Hooft coupling, bulk null geodesics control sharp singularities of boundary correlators. For the AdS$_5$-Schwarzschild black brane,

$$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$

a radial null geodesic satisfies

$$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$

and crossing the horizon shifts the imaginary part of the coordinate time by $\pm\beta/4$ under the standard $i\epsilon$ prescription (Dodelson et al., 12 Nov 2025). A geodesic that starts on one boundary, crosses the horizon, hits the spacelike singularity, and returns to the other boundary crosses the horizon twice and gives

$\Delta t = \frac{(1+i)\,\beta}{2},$

while more general wrapped configurations give

$\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$

These times lie on distinguished horizontal lines in the complex plane,

$$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$

the “real sections” of the bulk causal structure (Dodelson et al., 12 Nov 2025).

For a heavy operator $\mathcal O$ of dimension $_5$0, the WKB or geodesic approximation gives

$_5$1

When a geodesic becomes null, $_5$2, producing a non-analytic contribution in $_5$3. In this sense, complex-time singularities or branch points of $_5$4 are direct boundary avatars of null bulk geodesics, including those that bounce from the singularity (Dodelson et al., 12 Nov 2025).

The same structure can be derived without the large-mass limit from asymptotic quasinormal modes via the thermal product formula

$_5$5

together with an asymptotic QNM line

$_5$6

Fourier transformation then yields singularities at

$_5$7

with

$_5$8

The outermost points $_5$9 and $$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$0 are the one-bounce singularities, and their imaginary separation matches the geodesic computation (Dodelson et al., 12 Nov 2025).

At finite ’t Hooft coupling, the paper proposes that stringy corrections introduce zeroes of $$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$1 in addition to poles. In the toy deformation

$$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$2

the large-$$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$3 asymptotics shift to

$$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$4

and each classical singularity splits into a pair

$$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$5

The strict lattice on the real sections is therefore destroyed. The former divergence becomes a finite-height “bump,” interpreted heuristically as a classical bouncing geodesic replaced by a finite-size worldsheet (Dodelson et al., 12 Nov 2025).

The infinite-temperature SYK model provides a microscopic realization of this picture. Using the moment expansion

$$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$6

with moments computed up to $$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$7, and diagonal Padé continuation, the outermost singularities for $$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$8 lie very close to $$ds^2 = -f(r)\,dt^2 + \frac{dr^2}{f(r)} + r^2 d\Omega_3^2,\qquad f(r) = r^2 - \frac{\mu}{r^2},$$9 but are slightly displaced; for example,

$$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$0

The singular set is not additive, so the strict lattice structure is absent, and as $$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$1 increases the singularities move toward the imaginary axis, disappearing in the strict large-$$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$2 limit (Dodelson et al., 12 Nov 2025). The local behavior near a branch point is

$$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$3

for $$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$4, while for $$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$5,

$$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$6

This ties the branch structure to the nonlinear operator dynamics rather than to Padé artifacts (Dodelson et al., 12 Nov 2025).

3. Retarded propagators, Hadamard theory, and Schwarzschild convergence boundaries

A more general formulation replaces WKB intuition by the Hadamard theory of hyperbolic equations. For a scalar Green’s function $$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$7 solving

$$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$8

Hadamard’s theorem implies that the retarded Green’s function diverges whenever $$\frac{dt}{dr} = \pm\frac{1}{f(r)},\qquad t(r) = \int_r^\infty \frac{dr'}{f(r')},$$9 and $\pm\beta/4$0 are connected by a null geodesic (Grozdanov et al., 16 Mar 2026). In holography, after taking the boundary limit, the same statement applies to the retarded boundary correlator: singularities occur whenever the corresponding boundary points are connected by a null geodesic, including null limits of bouncing spacelike or timelike bulk geodesics (Grozdanov et al., 16 Mar 2026).

For static planar black-brane metrics

$\pm\beta/4$1

the radial geodesic equation can be written as

$\pm\beta/4$2

If near $\pm\beta/4$3,

$\pm\beta/4$4

then the effective potential diverges and there exist bouncing geodesics and corresponding bouncing singularities (Grozdanov et al., 16 Mar 2026). This provides a sufficient criterion. It is not necessary: the paper gives an explicit counterexample in the self-dual linear axion model, where the metric

$\pm\beta/4$5

has scalar curvature

$\pm\beta/4$6

so there is a curvature singularity, but $\pm\beta/4$7 is finite at $\pm\beta/4$8, and there are no bouncing geodesics (Grozdanov et al., 16 Mar 2026). A similar absence occurs in BTZ, where $\pm\beta/4$9 is also finite at the origin (Grozdanov et al., 16 Mar 2026).

Schwarzschild provides an asymptotically flat analogue. In Kruskal coordinates

$i\epsilon$0

the singularity is at

$i\epsilon$1

with $i\epsilon$2 the tortoise coordinate

$i\epsilon$3

The analytically continued retarded Green’s function has an infinite tower of bouncing singularities at

$i\epsilon$4

and

$i\epsilon$5

with $i\epsilon$6 (Arnaudo et al., 15 May 2026). These are the future- and past-bounce loci. They explain the convergence boundary of the QNM expansion, which is

$i\epsilon$7

The first term is the direct null travel time; the second is the time for a null ray that “scatters” from the effective potential at

$i\epsilon$8

The paper shows that this previously mysterious real-time boundary is the real-axis manifestation of the nearest bouncing singularity in the complex $i\epsilon$9-plane (Arnaudo et al., 15 May 2026).

The same singular set governs an annular region of convergence for the Matsubara mode sum near the horizon. The inner radius is set by the past-bounce singularity, while the outer radius is set by the ingoing lightcone singularity (Arnaudo et al., 15 May 2026). This places bouncing singularities at the center of both QNM and Matsubara analytic structure.

4. Cosmological uses: replacing, traversing, or softening the big-bang singularity

In cosmology, the term usually refers to a different problem: replacing or traversing a big-bang singularity by a bounce. In standard FRW cosmology,

$\Delta t = \frac{(1+i)\,\beta}{2},$0

a classical non-singular bounce is a smooth transition from $\Delta t = \frac{(1+i)\,\beta}{2},$1 to $\Delta t = \frac{(1+i)\,\beta}{2},$2 with

$\Delta t = \frac{(1+i)\,\beta}{2},$3

and a finite minimum scale factor $\Delta t = \frac{(1+i)\,\beta}{2},$4. In pure Einstein gravity with flat spatial slices and ordinary matter, this requires violation of the null energy condition and/or modified gravity (1803.01961).

“Bouncing Cosmology made simple” develops a geometric “wedge diagram” for this case. In the standard big-bang picture all wedges meet at a vertex where $\Delta t = \frac{(1+i)\,\beta}{2},$5, while in a classical non-singular bounce the wedge extends through a contracting phase, the scale factor reaches a finite minimum, and there is no vertex. With a contracting phase characterized by $\Delta t = \frac{(1+i)\,\beta}{2},$6, the horizon size grows very rapidly relative to the patch size, and the model addresses the horizon, flatness, anisotropy, and low-entropy problems while remaining geodesically complete into the past (1803.01961). The same paper emphasizes that the bounce itself must be realized by an NEC-violating sector or scalar-tensor modification that remains classically smooth and stable at sub-Planckian densities (1803.01961).

Several EFT or modified-gravity realizations are represented in the cited literature. In the non-local gravity model

$\Delta t = \frac{(1+i)\,\beta}{2},$7

an exact bouncing solution is

$\Delta t = \frac{(1+i)\,\beta}{2},$8

with

$\Delta t = \frac{(1+i)\,\beta}{2},$9

The universe contracts for $\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$0, expands for $\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$1, and reaches the minimum $\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$2 at $\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$3. No additional matter is needed; the bounce is generated by the non-local gravitational sector itself (Koshelev et al., 2012). The paper explicitly leaves perturbative stability of this exact background as an open problem (Koshelev et al., 2012).

In fourth-order gravity, order reduction is used to construct covariant effective actions $\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$4 whose reduced Friedmann equations reproduce the loop-quantum-cosmology form

$\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$5

The resulting cosmologies are perturbatively close to GR away from Planckian curvature and replace the big-bang singularity by a bounce at $\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$6, where curvature invariants remain finite (Miranda et al., 2022).

A different use of “bouncing singularity” appears in Gauss–Bonnet models with a Type IV finite-time singularity at or near the bounce. In the ghost-free Gauss–Bonnet setup of (Odintsov et al., 2022), the Hubble rate is

$\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$7

or, in the localized version,

$\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$8

For $\Delta t_n = \frac{(\pm1 + i)\,n\beta}{2}.$9, the singularity is Type IV: $$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$0, $$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$1, and $$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$2 remain finite, but higher derivatives diverge (Odintsov et al., 2022). When the Type IV term globally affects the spacetime, the scalar power spectrum is strongly red and $$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$3 is too large compared with Planck data; when the singularity acts only locally, perturbations are generated deep in the contracting phase, and for

$$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$4

one obtains $$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$5 and $$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$6 (Odintsov et al., 2022). By contrast, in Einstein-scalar-Gauss-Bonnet cosmology, flat and open FRW bounces are excluded under the paper’s analyticity and stability assumptions, while closed-universe bounces are linearly unstable in the scalar sector (Sberna et al., 2017).

The literature also contains a classical GR example of repeated, nonsingular bounces: the “Simple Harmonic Universe,” a closed FRW model with $$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$7, $$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$8, and matter satisfying

$$\operatorname{Im} t = \frac{m\beta}{2},\qquad m\in\mathbb{Z},$$9

For $\mathcal O$0,

$\mathcal O$1

with

$\mathcal O$2

If $\mathcal O$3, the model cycles through an infinite set of nonsingular bounces. Moderate bounces with $\mathcal O$4 are classically stable at linear order, while extreme bounces with $\mathcal O$5 develop perturbative and quantum instabilities after many cycles (Graham et al., 2011).

5. Singularity scattering maps and BKL bounces

A more radical cosmological interpretation does not remove the singularity at all. Instead, it formulates junction conditions across it. In the singularity-scattering framework, one works in Gaussian time $\mathcal O$6 near a spacelike singular hypersurface $\mathcal O$7 and defines singularity initial data

$\mathcal O$8

with

$\mathcal O$9

subject to the asymptotic constraints

$_5$00

The corresponding asymptotic profile is

$_5$01

A past-to-future singularity scattering map is then a map

$_5$02

which is diffeomorphism-covariant and ultra-local (LeFloch, 2021, LeFloch, 2021).

The classification theorem states that only two classes of ultra-local spacelike scattering maps are available: isotropic bounces and non-isotropic bounces (LeFloch, 2021). In the rigidly conformal isotropic case,

$_5$03

In the rigidly conformal non-isotropic case,

$_5$04

$_5$05

with

$_5$06

(LeFloch, 2021).

Any ultra-local quiescent bounce obeys three universal laws. First, the traceless part of the extrinsic curvature scales as

$_5$07

for some dissipation parameter $_5$08. Second, matter undergoes a canonical transformation

$_5$09

Third, the metric rescales directionally according to

$_5$10

These laws interpret the bounce as a constrained scattering problem on singularity data rather than as a smooth continuation of the metric (LeFloch, 2021).

A related but distinct use of “bounce” appears in rigorous BKL theory beyond spatial homogeneity. For Gowdy-symmetric vacuum spacetimes, the metric in areal gauge is

$_5$11

and the Einstein vacuum equations reduce to the coupled PDE system

$_5$12

$_5$13

Near the singularity $_5$14, the dynamics are asymptotically velocity term dominated, yet nonlinear BKL bounces and spikes occur (Li, 2024). Along timelike curves one obtains an ODE model

$_5$15

whose heteroclinic behavior realizes a BKL bounce. In the one-bounce regime, the effective velocity map is

$_5$16

which matches the classical Kasner transition map in the relevant Gowdy sector (Li, 2024). Here “bouncing singularity” does not mean singularity removal; it means the instability mechanism by which one Kasner epoch transitions into another as the singularity is approached.

6. Regularized bounces, dynamical pathologies, and singular impacts

The cosmological literature distinguishes between geometric singularities and pathologies of perturbation theory. In the geometric sigma-model construction with four scalar fields, any regular bouncing FRW metric can be embedded into a model whose vacuum realizes that metric and whose target-space metric remains invertible throughout the bounce. The sufficient conditions are

$_5$17

which make the sigma-model target metric non-degenerate everywhere (Vasilić, 2017). Linear perturbations split into $_5$18 tensor, $_5$19 vector, and $_5$20 scalar degrees of freedom, and the apparent singularities in the scalar perturbation equations at $_5$21 are shown by a field redefinition and power-series analysis to be gauge artifacts rather than genuine breakdowns (Vasilić, 2017). In that precise EFT sense, “bouncing singularities” are avoided.

A different dynamical viewpoint is provided by singular second-order ODEs of Lazer–Solimini type,

$_5$22

with $_5$23 continuous, $_5$24-periodic, and negative (Rojas et al., 2020). Here the repulsive singularity at $_5$25 is weak enough that trajectories reach it in finite time with finite velocity, and a bouncing solution is defined as a continuous $_5$26 whose zero set is discrete, which is $_5$27 away from impacts and satisfies the elastic collision law

$_5$28

at each impact time $_5$29 (Rojas et al., 2020). The paper proves, by a Poincaré–Birkhoff argument, the existence of abundant periodic bouncing solutions. For sufficiently large $_5$30, there are at least two $_5$31-periodic solutions with exactly one impact in $_5$32, and for any prescribed $_5$33 there exists $_5$34 such that for all $_5$35 there is at least one $_5$36-periodic solution with exactly $_5$37 impacts in the period interval (Rojas et al., 2020). In this setting, the singularity is neither smoothed nor crossed by regular geometry; it is an impact surface endowed with a reflection rule.

A final point of contrast is that not every purported bounce is stable or even available. In Einstein-scalar-Gauss-Bonnet gravity, the paper (Sberna et al., 2017) proves a no-go theorem for nonsingular flat and open FRW bounces under analyticity and tensor-stability assumptions, and shows that explicit closed-universe bouncing solutions are linearly unstable in the scalar sector. This is a reminder that “bouncing singularity” is not itself a guarantee of viability; it is a descriptive category whose realization depends on the detailed field content and stability properties.

Across these literatures, the common thread is not the elimination of singular behavior but its controlled reappearance in another language: as a complex singularity of a correlator, as a matching rule on asymptotic data, as a Kasner transition, as a finite-height stringy bump, or as an elastic impact law. The principal technical divide is between frameworks where the singularity remains geometric but becomes diagnostically accessible, and frameworks where it is replaced or bypassed by an effective bounce.