Self-Stimulated Avalanche Dynamics

• Self-stimulated avalanche is a cascade phenomenon where an initial seed triggers an internal feedback loop, resulting in a disproportionately large event.
• It encompasses diverse systems—from gravitational superfluorescence and semiconductor ionization to neural bursting and optical bistability—each with unique amplification dynamics.
• Studies show that while external triggers initiate the event, factors like structural disorder, occupation feedback, and coherent excitation determine the avalanche scale.

Searching arXiv for recent and relevant papers on “self-stimulated avalanche” and closely related avalanche mechanisms. “Self-stimulated avalanche” is not a uniformly standardized term across the contemporary literature, but a consistent family resemblance is visible across several domains. In the strongest and most literal usage, it denotes a cascade in which an initially seeded transition is amplified by feedback generated by the system’s own evolving state or emitted field, as in gravitational superfluorescence from superradiant axion clouds (Lyu et al., 9 Jun 2026). In broader and often implicit usage, it refers to avalanches, cascades, or switching events whose immediate trigger is insufficient to explain their scale unless one also accounts for internally developed susceptibility, occupation-dependent scattering, collective coherence, or self-organized structural disorder (Ramos et al., 2008, Burdine et al., 2 May 2026, Selvakumaran et al., 25 Jun 2026). Taken together, these works suggest that the defining feature is not the absence of a trigger, but the presence of an endogenous amplification loop that converts a seed into a disproportionately large response.

1. Terminological scope and conceptual boundary

The phrase itself is unevenly distributed. Several relevant papers explicitly state that they do not use the expression “self-stimulated avalanche,” even when they analyze mechanisms that are close to it conceptually. The sandpile work on avalanche prediction is framed in the language of self-organized criticality (SOC), large and very large avalanches, foreshocks, aftershocks, and internal structural changes rather than “self-stimulated avalanche” (Ramos et al., 2008). The NEGF study of semiconductor avalanche photodiodes speaks instead of impact ionization, carrier multiplication, and self-consistent impact-ionization self-energy (Burdine et al., 2 May 2026). The oscillator paper uses “self-organized avalanches” and attributes them to amplification of “noise” provided by local rearrangements (Gilpin, 2019). By contrast, the gravitational-wave paper explicitly describes a “self-stimulated avalanche” in a superradiant boson cloud (Lyu et al., 9 Jun 2026).

A useful boundary condition is that self-stimulation is not equivalent to complete autonomy. In several systems, an external drive or seed remains essential. The sandpile is driven by grains added one by one (Ramos et al., 2008); the optical bistability problem uses a coherently driven Kerr cavity plus an incident single photon (Selvakumaran et al., 25 Jun 2026); the QED cascade begins with a laser pulse and target electrons (Serebryakov et al., 21 Aug 2025); the LWIR self-guiding scenario requires the propagating pulse and, in practice, aerosol seed sites (Woodbury et al., 2020). What changes is the causal interpretation: the large event is no longer treated as a direct and proportionate response to the latest perturbation, but as the outcome of internally accumulated gain, coherence, or structural readiness.

A second boundary condition is that self-stimulation is not synonymous with criticality. Some systems do sit near critical or quasi-critical boundaries, but others are more naturally described in terms of bistable switching, metastability, self-consistent many-body transport, or cooperative radiative feedback. The adaptive-neuron coherent-bursting model explicitly argues that power-law avalanches “need not be signatures of criticality” (Chan et al., 2024), while the SOqC analysis attributes avalanches to stochastic oscillations around a weakly damped focus rather than exact SOC (Kinouchi et al., 2018).

2. Endogenous preparation in slowly driven and self-organized media

A central route to self-stimulated avalanche behavior is slow endogenous preparation of a susceptible state. In the quasi-two-dimensional steel-bead sandpile, avalanches span about three decades and display a power-law distribution with exponent $-1.6$. Large avalanches are essentially uncorrelated in direct event-event terms, with exponential waiting times of $26 \pm 2$ steps for large events and $132 \pm 9$ steps for very large events, yet they are preceded by measurable structural evolution (Ramos et al., 2008). The key structural disorder variable is the Voronoi shape factor

$\zeta=\frac{C^2}{4\pi S},$

whose spatial average rises continuously for about 50 steps before a large avalanche. The authors conclude that “not any small avalanche can cascade into a large event” and that “Large avalanches require certain conditions in the structure of the system that have to be developed” (Ramos et al., 2008). This is an experimentally precise statement of self-preparation: the added grain remains the immediate trigger, but the avalanche scale depends on the pile’s own internally evolved disorder.

A related but more explicitly dynamical formulation appears in the continuum SOC model of wave-like avalanche propagation. There the transport law is generalized to

$$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$

leading to a threshold-controlled continuum equation with hyperbolic and nonlocal structure (Oh et al., 2013). The avalanche is represented as a traveling unstable packet whose front survives only if the medium is metastable: $g_c > g_0 > \beta g_c.$ In this setting, triggering is local, but propagation is sustained by the retarded transport law itself. The front drives neighboring regions above threshold while the tail relaxes, so the packet is conditionally self-sustained once launched (Oh et al., 2013).

Crystal plasticity adds a slower feedback cycle. In slowly compressed Ni microcrystals, decreasing the nominal strain rate from $10^{-4}\,\mathrm{s}^{-1}$ to $10^{-6}\,\mathrm{s}^{-1}$ increases the slow-relaxation fraction, produces quasi-periodic avalanche bursts with periods reaching about 8 hours, and shifts the avalanche size exponent from $\tau \approx 1.5$ toward $\tau \approx 2.0$ (Papanikolaou et al., 2013). The reduced susceptibility dynamics,

$26 \pm 2$0

captures the mechanism: slow relaxation and loading rebuild the system toward criticality, while a large avalanche abruptly resets it. The resulting “self-organized avalanche oscillator” is a delayed feedback loop in which past avalanches and inter-avalanche relaxation jointly condition future ones (Papanikolaou et al., 2013).

The jammed-oscillator model gives a structurally different but conceptually similar picture. A modified Kuramoto ensemble with global attraction and short-range repulsion exhibits self-organized avalanches at intermediate $26 \pm 2$1. These avalanches originate in the amplification of “noise” provided by local rearrangements; analytically, the relevant stability exchange occurs at

$26 \pm 2$2

in the many-oscillator limit (Gilpin, 2019). The triggering event is a local rearrangement in a jammed synchronized cluster, but the avalanche is enabled by the cluster’s own accumulated repulsive stress.

3. Self-excitation, adaptive feedback, and neuronal avalanche regimes

Neural and neural-inspired systems provide several distinct realizations of internal stimulation. In a developing excitatory network with homeostatic neurite growth, the mean offspring number per spike is

$26 \pm 2$3

so low spontaneous activity $26 \pm 2$4 drives the network near the critical branching point $26 \pm 2$5 (Kossio et al., 2018). Avalanches are then exactly Hawkes clusters or branching-process trees. Their size distribution is Borel,

$26 \pm 2$6

with critical asymptotic $26 \pm 2$7, and durations have critical tail $26 \pm 2$8 (Kossio et al., 2018). Here the self-stimulation is explicit in the Hawkes sense: each spike raises the future intensity of further spikes.

A different route appears in adaptive Izhikevich networks under stationary white-noise input. When recurrent excitation is sufficiently strong and balanced by adaptation, the network enters a coherent-bursting state in which avalanche durations and sizes follow power laws over a suitable range of time bin $26 \pm 2$9, with representative exponents such as $132 \pm 9$0, $132 \pm 9$1, and $132 \pm 9$2 for $132 \pm 9$3 (Chan et al., 2024). The authors show that the avalanche statistics are reproduced by an inhomogeneous Poisson process with oscillatory rate $132 \pm 9$4, and conclude that the observed features arise from stochasticity plus coherent bursting rather than necessarily from criticality (Chan et al., 2024). This suggests a self-generated avalanche regime in which structured bursts emerge from internal excitation-adaptation dynamics under constant noisy background drive.

Self-organized quasi-criticality sharpens the distinction between exact criticality and internally generated avalanche segmentation. In adaptive gain and synaptic-depression models, the active fixed point is a stable focus with eigenvalues of modulus

$132 \pm 9$5

so demographic noise excites stochastic oscillations around a nearly indifferent focus (Kinouchi et al., 2018). Because low-activity segments of these oscillations can be knocked into the absorbing state, the time series is partitioned into avalanches and occasional dragon-king events. The avalanches are therefore internally generated by adaptive dynamics plus stochastic fluctuations, not by external drive through a fixed critical point (Kinouchi et al., 2018).

A fully deterministic version of internally generated avalanche behavior appears in nonlocally coupled FitzHugh–Nagumo arrays. Near the canard threshold $132 \pm 9$6, with $132 \pm 9$7 and $132 \pm 9$8, diffusive coupling induces an effective parameter

$132 \pm 9$9

which can push individual units into canard-mediated subthreshold trapping near the unstable equilibrium (Contreras et al., 2023). Intermittent escape then produces avalanche-like pseudo-synchronous bursts without external driving. The authors report approximate scale-free avalanche-size distributions with finite-size cutoff and perturbation-based signatures of critical slowing down near $\zeta=\frac{C^2}{4\pi S},$0 (Contreras et al., 2023).

At the same time, the measurement protocol matters. In a disordered absorbing-state model of the visual pathway, avalanches defined between silent bins can exhibit restricted power-law statistics outside the true critical region, even when stimulus-to-absorption avalanches do not (Girardi-Schappo et al., 2018). This cautions against equating internally segmented bursts with critical self-organization.

4. Microscopic occupation feedback in electronic and photonic devices

In semiconductor avalanche photodiodes, the microscopic content of self-stimulation is occupation-dependent impact ionization. The NEGF formulation writes the device Hamiltonian as

$\zeta=\frac{C^2}{4\pi S},$1

with retarded propagator

$\zeta=\frac{C^2}{4\pi S},$2

lesser Green’s function

$\zeta=\frac{C^2}{4\pi S},$3

and spectral function

$\zeta=\frac{C^2}{4\pi S},$4

The impact-ionization self-energies $\zeta=\frac{C^2}{4\pi S},$5 are functionals of band-resolved occupied and unoccupied Green’s functions, so the generation rate depends on the current nonequilibrium carrier population itself (Burdine et al., 2 May 2026). The paper repeatedly emphasizes that avalanche onset requires not just a strong electric field but both “available states” and “sufficient nonequilibrium occupation.” It also states that the results establish a “transport baseline for self-consistent calculations of the impact-ionization self-energy and carrier multiplication” rather than a complete validated multiplication model (Burdine et al., 2 May 2026).

Optical and superconducting detectors show a more circuit-oriented internal trigger. In the series-2-SNAP architecture, a photon-induced hotspot in one nanowire branch diverts current into the neighboring branch; if

$\zeta=\frac{C^2}{4\pi S},$6

the second branch also switches normal, producing a local avalanche event (Cheng et al., 2016). In the reported device, the unfired nanowire pairs provide an intrinsic current-limiting inductance with $\zeta=\frac{C^2}{4\pi S},$7, the reset decay time is reduced to $\zeta=\frac{C^2}{4\pi S},$8 ns from $\zeta=\frac{C^2}{4\pi S},$9 ns, and timing jitter improves to $$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$0 ps FWHM from $$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$1 ps (Cheng et al., 2016). Here self-stimulation is purely electrical: the geometry and inductance force one switching event to provoke another.

Optical stimulation by avalanche-generated photons is measured directly in SiPMs. At $$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$2 V over-voltage, the number of photons produced per avalanche at the source is $$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$3 for the Hamamatsu VUV4 and $$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$4 for the FBK VUV-HD3, with no significant temperature dependence within uncertainties (Raymond et al., 2024). The paper interprets these photons mainly in terms of internal and external optical cross-talk rather than same-cell self-retriggering, but it establishes that an avalanche produces secondary photons in sufficient number to stimulate additional avalanches in neighboring devices or neighboring microcells (Raymond et al., 2024).

A useful counterexample is the self-differencing APD under strong pulse illumination. There, strong synchronized optical pulses can drive nearly one avalanche per gate so uniformly that the self-differencing circuit subtracts the avalanche response away (Gao et al., 2022). This is externally stimulated, periodic avalanche creation followed by architectural self-cancellation, not a self-amplified avalanche mechanism.

5. Pump-powered, bosonic, and field-mediated avalanche amplification

Driven photonic systems often realize self-stimulation as switch-assisted access to an energy reservoir. In a coherently driven Kerr cavity, a single additional photon can trigger a transition from the low-photon branch to the high-photon branch of the bistability loop. The cavity Hamiltonian is

$$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$5

and the transmitted flux is

$$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$6

The extra photon does not itself supply the large output; rather, it catalyzes a branch switch after which the coherent pump feeds the high state (Selvakumaran et al., 25 Jun 2026). The paper reports optimal single-photon-stimulated switching probability of about $$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$7, a bright-to-dark ratio $$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$8 exceeding order $$\Gamma + \tau\frac{\partial \Gamma}{\partial t} - \tau\chi\frac{\partial^2 \Gamma}{\partial x^2} = -Q \frac{\partial T}{\partial x},$$9, and amplification factor $g_c > g_0 > \beta g_c.$0 ranging from tens to thousands in favorable regions (Selvakumaran et al., 25 Jun 2026). The avalanche is therefore pump-powered and metastable rather than free-running.

The bosonic avalanche laser makes the feedback loop explicit. Its master equation contains irreversible three-mode jump operators

$g_c > g_0 > \beta g_c.$1

and the semiclassical transport currents are

$g_c > g_0 > \beta g_c.$2

so ladder transport is stimulated both by target occupation and by cavity occupation through a factor $g_c > g_0 > \beta g_c.$3 (Garbe et al., 5 Sep 2025). The authors call this “double stimulation.” The system exhibits an intermediate self-pulsing phase, excitability, and coherence resonance, and is proposed as a number-resolved avalanche detector for microwave photons (Garbe et al., 5 Sep 2025). This is one of the cleanest many-body examples in which “self-stimulated avalanche” is not merely interpretive but built into the transition rates themselves.

Strong-field QED provides a relativistic analogue. A single multipetawatt laser pulse reflected from a self-organized parabolic plasma mirror can initiate an avalanche-type QED cascade with threshold laser power about $g_c > g_0 > \beta g_c.$4 (Serebryakov et al., 21 Aug 2025). At $g_c > g_0 > \beta g_c.$5, the simulations show exponential positron growth, more than 15 positron generations, similar growth rates across many successive generations, $g_c > g_0 > \beta g_c.$6 for essentially all generations, and nearly generation-independent GeV-scale spectra (Serebryakov et al., 21 Aug 2025). The defining criterion is that newly created charges are re-accelerated by the field and re-enter the same photon-emission and pair-creation loop.

Long-wave infrared self-guiding mediated by avalanche ionization is less autonomous but still self-consistent. The same LWIR pulse self-focuses, raises the local intensity, initiates avalanche sites—most plausibly around aerosols—and is then refracted by the effective negative index of the resulting plasma ensemble (Woodbury et al., 2020). The paper argues that this produces moderate-intensity self-guiding only when aerosol-triggered isolated avalanche sites are included, so the process is pulse-driven and self-consistently propagation-coupled, but environmentally assisted (Woodbury et al., 2020).

The most literal field-mediated self-stimulated avalanche in the supplied material is gravitational. In a superradiant axion cloud, a weak external GW seed creates a tiny lower-level amplitude; the resulting coherence emits a quadrupolar GW field whose near-zone retarded part feeds back on the same transition. The lower-state population obeys

$g_c > g_0 > \beta g_c.$7

with solution

$g_c > g_0 > \beta g_c.$8

and the observable strain has a $g_c > g_0 > \beta g_c.$9 envelope (Lyu et al., 9 Jun 2026). This is the paper’s precise meaning of a self-stimulated avalanche and the reason it is described as gravitational superfluorescence.

6. Diagnostics, distinctions, and recurrent controversies

Across these systems, several diagnostic motifs recur. One is the distinction between direct event-event predictability and state-dependent susceptibility. The sandpile study found essentially uncorrelated large-avalanche timing but robust precursor evolution in the internal disorder variable $10^{-4}\,\mathrm{s}^{-1}$0 (Ramos et al., 2008). Another is the distinction between a seed and the gain medium that amplifies it: the single photon in a Kerr cavity, the initial nonequilibrium carriers in an APD, the seed GW in a boson cloud, or the first generated $10^{-4}\,\mathrm{s}^{-1}$1 pairs in a QED cascade are all small compared with the eventual emitted or transferred energy (Selvakumaran et al., 25 Jun 2026, Burdine et al., 2 May 2026, Lyu et al., 9 Jun 2026, Serebryakov et al., 21 Aug 2025).

A common misconception is that self-stimulation implies absence of external input. The literature does not support that stronger claim. Many of the clearest examples are externally seeded but internally amplified. Another misconception is that avalanche power laws by themselves establish criticality. The adaptive-neuron coherent-bursting model, the silent-bin analysis in absorbing-state systems, and the SOqC work all show that power-law-like avalanches can emerge from coherent bursting, stochastic oscillations, or segmentation rules without exact SOC (Chan et al., 2024, Girardi-Schappo et al., 2018, Kinouchi et al., 2018).

The open theoretical issue is therefore less whether self-stimulated avalanches exist than how to classify them. Some are structurally prepared releases in slowly driven media; some are occupation-dependent scattering cascades; some are metastable switching events that unlock pump energy; some are cooperative radiative transitions. What unifies them is a delayed or instantaneous endogenous gain mechanism by which the evolving state of the system increases the probability, rate, or scale of its own subsequent release. A plausible overarching interpretation is that “self-stimulated avalanche” denotes a class of seeded but internally amplified transitions whose microscopic gain may be geometric, statistical, kinetic, or radiative, and whose macroscopic signatures range from scale-free bursts to logistic population transfer and delayed coherent pulses.