Exponential DAMs

• Exponential DAMs are models defined by exponential interaction kernels that produce exponential cluster-size and avalanche distributions.
• They encompass diverse frameworks including time series forecasting (ADA), stochastic cellular automata (RDA), and associative memory networks (SEDAM).
• These models demonstrate analytic tractability, regime transitions, and practical applications in hydrology, statistical physics, and neural computation.

Exponential Discrete Avalanche Models (DAMs) describe a class of systems in which the interaction rules or memory kernels exhibit exponential dependence, resulting in exponential cluster-size or event-size distributions. The term encompasses models ranging from time series forecasting methods for natural reservoirs, to stochastic cellular automata producing exponential statistics in avalanche distributions, and to neural associative memory networks where the energy function contains exponential nonlinearities. Recent work in arXiv literature demonstrates the analytic tractability, diverse dynamical regimes, and high capacity of such models across fields including hydrology, statistical physics, and neural computation (Souza, 2023, Bialecki et al., 2010, Cafiso et al., 16 Jan 2026).

1. Mathematical Structures of Exponential DAMs

Exponential DAMs are characterized by system-level rules or energy functions containing explicit exponential terms. Three major representative classes are documented:

• Exponential Damped Additive Model (ADA) for Time Series:

ADA, or ETS(A,Ad,A) in the notation of Hyndman et al., defines reservoir dynamics with an observation equation,

$$y_t = \ell_{t-1} + \phi b_{t-1} + s_{t-m} + \varepsilon_t,$$

and state updates:

$$\begin{aligned} \ell_t &= \ell_{t-1} + \phi b_{t-1} + \alpha \varepsilon_t,\ b_t &= \phi b_{t-1} + \beta \varepsilon_t,\ s_t &= s_{t-m} + \gamma \varepsilon_t, \end{aligned}$$

where $(\ell_t, b_t, s_t)$ denote latent level, trend, and seasonality, and $\phi \in (0,1)$ is the exponential damping parameter (Souza, 2023).

• Exponential Discrete Avalanche Model (Random Domino Automaton, RDA):

In RDA, avalanches are triggered by a constant probability $\mu_i = \beta$ for all cluster sizes $i$, yielding exponential cluster-length statistics:

$P(i) = A e^{-B i},\quad B = -\ln \rho,$

with normalization and characteristic scale determined by the control parameters $\nu$ (loading rate) and $\beta$ (trigger rate) (Bialecki et al., 2010).

• Exponential Dense Associative Memory (DAM) Networks:

The stochastic exponential DAM (SEDAM) model defines the network energy as

$$E[\mathbf{S}] = -\sum_{\mu=1}^K \exp\left(\boldsymbol{\xi}_\mu^T \mathbf{S}\right),$$

where $\mathbf{S}$ is the binary neuron state vector and patterns $\boldsymbol{\xi}_\mu$ are stored via exponential summation, encoding all higher-order interactions. Pattern embedding follows Hebbian scaling for $p$-body terms (Cafiso et al., 16 Jan 2026).

2. Regime Classification and Dynamics

Exponential DAMs demonstrate regime transitions controlled by tuning model-specific parameters (e.g., noise intensity, loading/trigger probabilities, or smoothing/damping in time series):

• Avalanche Statistics in Cellular Automata:

In RDA, a constant trigger probability imposes an exponential regime for both cluster-size and event-size distributions. The precise exponential profile follows from a nonlinear set of steady-state equations for cluster counts, with the mean-field percolation solution yielding $P(i) \approx \frac{1}{\lambda} e^{-i/\lambda}$.

• Temporal Complexity in SEDAM:
  • Subcritical: White-noise recall with $H \approx 0.5$.
  • Critical: Complex intermittency—long-range temporal correlations $(H \approx 1.1)$, non-Gaussian event-driven diffusion $(\delta \approx 0.85)$, and power-law inter-event times $(2<\mu<3)$.
  • Supercritical: Loss of memory, return to white-noise-like statistics.

The critical window in noise narrows with increasing $K$, but does not converge to a single point, consistent with the concept of extended criticality (Cafiso et al., 16 Jan 2026).

3. Analytic Tractability and Key Formulas

The analytic structures of exponential DAMs facilitate tractable closed-form solutions in specific regimes:

• Closed-form Distributions in RDA:

The cluster- and avalanche-size distributions are given by geometric/exponential forms under the assumption of independent clusters. The characteristic scale $\lambda$ is explicitly determined from the control ratio $\theta = \beta/\nu$ via

$$\rho = \frac{2}{3\theta + \sqrt{9\theta^2 + 4(1-\theta)}},$$

and

$P(i) = (1-\rho)\rho^i.$

• State-space Forecasting Equations in ADA:

Multi-step forecasting utilizes

$$\phi_h = \sum_{j=1}^h \phi^j = \phi \frac{1-\phi^h}{1-\phi},$$

to propagate trend with damping over the seasonal cycle, and the model accommodates trend and seasonality simultaneously (Souza, 2023).

• Infinite-order Interactions in Exponential DAM Networks:

The expansion

$$E[\mathbf{S}] = -\sum_{p=0}^\infty \sum_{i_1 < \cdots < i_p} W^{(p)}_{i_1\cdots i_p} S_{i_1} \cdots S_{i_p}$$

encodes exponentially many memory attractors, providing a mechanism for extreme storage capacity (Cafiso et al., 16 Jan 2026).

4. Model Selection, Fitting, and Empirical Results

Model selection in exponential DAM frameworks employs likelihood-based criteria and systematic taxonomy:

• Time Series (Reservoir Level Forecasting):

Souza et al. (Souza, 2023) fit the full Pegels taxonomy of exponential smoothing models (15 forms) to daily reservoir levels, selecting ADA (ETS(A,Ad,A)) by Akaike Information Criterion (AIC), with

$$\text{AIC} = T \ln\left(\frac{\mathrm{SSE}}{T}\right) + 2c,$$

where $c$ counts estimated parameters. For both Descoberto and Santa Maria reservoirs, ADA yielded the lowest AIC and accurate multi-year forecasts, as residuals exhibited Gaussianity and negligible autocorrelation.

• Critical Intermittency in SEDAM:

Cafiso & Paradisi (Cafiso et al., 16 Jan 2026) use temporal complexity metrics (DFA exponent $H$, diffusion entropy $\delta$, IET autocorrelation $\mathcal T_c$, event PDFs) to demarcate phase regions and quantify the trade-off between memory load and emergent complexity, with detailed scaling values tabulated for varying $K$ and $p$.

5. Interpretive Insights and Implications

Exponential DAMs reveal diverse structures unified by exponential law statistics or exponential nonlinearities in their dynamics:

• Unified Exponential Regime:

Exponential distributions in cluster or avalanche sizes are not signatures of criticality—as in SOC systems—but of memoryless triggering (μ constant), as demonstrated in RDA (Bialecki et al., 2010).

• Capacity and Stability in Associative Networks:

Infinite-order, exponentially weighted interactions in DAMs stabilize stored patterns, enabling explosive memory capacity while still supporting complex dynamic regimes in the presence of noise (Cafiso et al., 16 Jan 2026). This suggests a plausible route toward artificial networks with both ultra-high memory and biologically plausible stochastic dynamics.

• Forecasting in Environmental Systems:

The ADA exemplifies the capacity of exponential DAMs to synthesize trend, seasonality, and damping for long-range, reliable forecasting, as validated on multi-decade hydrological datasets (Souza, 2023).

• Extended Criticality:

The SEDAM model’s critical regime spans a finite parameter interval rather than a sharp point, supporting the extended criticality paradigm relevant to neural and complex adaptive systems.

6. Application Domains and Cross-Disciplinary Significance

Exponential DAMs have been applied and analyzed in a broad set of contexts:

Domain	Model Type	Key Observable(s)
Time Series Forecasting	ETS(A,Ad,A) (ADA)	Level, trend, seasonality
Stochastic Automata	RDA (Exponential DAM)	Cluster-size, avalanche-size
Associative Memory	Exponential DAM/SEDAM	Event statistics, temporal complexity

The analytical tractability, extensible capacity, and experimentally validated performance of Exponential DAMs continue to drive interest in their cross-disciplinary applications. The convergence of exponential law statistics, infinite-order interactions, and emergent critical phenomena in these diverse models suggests a broad relevance for understanding the dynamics of both artificial and natural complex systems (Souza, 2023, Bialecki et al., 2010, Cafiso et al., 16 Jan 2026).