Dynamical Partition Function

• The dynamical partition function is a generating function that extends the Gibbs ensemble to entire system trajectories, revealing dynamic phase transitions through nonanalytic behavior.
• It uses an intensive dynamical field analogous to temperature to bias trajectory distributions and generate cumulants that characterize large-deviation statistics.
• Its applications span nonequilibrium phenomena, quantum quench dynamics, and integrable combinatorial models, unifying diverse trajectory-based analyses across physics.

A dynamical partition function is a generating function that encodes the statistical properties of ensembles of system histories, as opposed to static configurations, and serves as the cornerstone of modern nonequilibrium statistical mechanics and integrable combinatorial models. By formal analogy with the ordinary (Gibbs) partition function, which sums over states weighted by Boltzmann factors, the dynamical partition function sums over entire trajectories, histories, or combinatorial objects, typically with an exponential bias imposed by a conjugate intensive field. Nonanalyticities and zeros in this function signal dynamic phase transitions (DPTs) and singular behavior in time-dependent or random systems. The dynamical partition function unifies trajectory-centric ensemble methods in physics, exactly solvable models of statistical mechanics with dynamical boundaries, and the framework of integrable dynamical systems in combinatorics.

1. Formal Definitions: Dynamical Partition Functions in Trajectory Space

For systems evolving over time, the dynamical partition function $Z(s)$ generalizes the spatial Gibbs ensemble to the trajectory (history) space. Given an ensemble of trajectories $\mathcal{T} = \{\tau_1, ..., \tau_K\}$, where $K$ counts the total number of events in a fixed observation window, the partition function is defined as

$$Z(s) = \sum_{\mathcal{T}} P[\mathcal{T}]\, e^{-s K[\mathcal{T}]}$$

or, equivalently, in terms of the activity distribution $p(K)$: $Z(s) = \sum_{K=0}^{K_{\text{max}}} p(K)\, e^{-s K}$ This construction extends to path integrals over microscopic phase-space histories $X(t)$: $Z(s) = \int \mathcal{D}[X]\, P[X]\, e^{-s K[X]}$ When temporal correlations are present, internal trajectory interactions $U_0(\mathcal{T})$ arise, resulting in

$$Z(s) = \sum_{\mathcal{T}} e^{- [ s K(\mathcal{T}) + U_0(\mathcal{T}) ] / k_D }$$

where $k_D$ is the dynamical analogue of Boltzmann’s constant and $U_0$ embodies temporal correlations as pairwise "interactions" along the time slices (Ye et al., 2022).

2. Intensive Dynamical Fields and Statistical Mechanics Analogy

The parameter $s$ acts as the intensive dynamical field conjugate to the time-extensive observable $K$ (e.g., activity, number of events), fully analogous to $\beta=1/k_B T$ in equilibrium statistical mechanics. The biased trajectory probability distribution reads

$$p(\mathcal{T}) = \frac{1}{Z(s)}\, \exp \left\{ - \frac{ s K(\mathcal{T}) + U_0(\mathcal{T}) }{ k_D } \right\}$$

In the large-$K$ (long-time) limit: $$s = - \lim_{K \to \infty} \frac{ k_D\, \ln p(\mathcal{T}) - U_0(\mathcal{T}) }{ K } + \text{const}$$ Thus, $s$ is not merely a computational artifact but an intrinsic, physically meaningful control parameter specifying the "equilibrium dynamical state" of the trajectory ensemble (Ye et al., 2022).

3. Cumulant Generating Functions and Large-Deviation Structure

The scaled cumulant generating function (SCGF) or dynamical free-energy-like function is defined as

$\psi(s) = \lim_{t \to \infty} \frac{1}{t}\ln Z(s)$

$\psi(s)$ is the dynamical analog of free energy density or pressure in equilibrium contexts. Its derivatives generate cumulants of $K$ (mean, variance, etc.). Nonanalyticities in $\psi(s)$ denote true phase transitions at the trajectory ensemble level—i.e., dynamic phase transitions—where macroscopic dynamical behavior changes abruptly (Ye et al., 2022, Takahashi et al., 2013).

4. Zeros of the Dynamical Partition Function and Dynamic Phase Transitions

$Z(s)$ is a finite polynomial in the fugacity-like variable $\gamma = e^{-s / k_D}$: $Z(s) = \sum_{K=0}^{K_{\text{max}}} p(K)\, \gamma^K$

Factoring $Z(\gamma)$ reveals its Lee–Yang or Fisher zeros: $$Z(\gamma) = p(0)\,\prod_{i=1}^{K_{\text{max}}} (1 - \gamma / \gamma_i)$$ As the system size or observation time grows, zeros may accumulate and pinch the positive real $\gamma$ (or $s$) axis at critical points $(\gamma_c,s_c)$. The corresponding nonanalyticity in $\psi(s)$ marks a dynamic phase transition (DPT), in perfect analog with equilibrium Yang-Lee theory (Ye et al., 2022, Takahashi et al., 2013). In random systems and spin glasses, two-dimensional distributions of zeros in the complex parameter plane signify dynamical singularities (“chaotic” phases distinct from equilibrium replica symmetry breaking) (Takahashi et al., 2013).

5. Dynamical Partition Functions in Integrable and Combinatorial Systems

Dynamical partition functions also arise in integrable lattice models with dynamical symmetries and boundary conditions. In elliptic SOS models with reflecting ends, the partition function (encoding domain-wall boundary constraints and dynamical reflection algebra) is represented as a single Izergin determinant. Here, the dynamical parameter (height $\theta$) enters via R- and K-matrix shifts and is essential for the determinant structure and enumeration of the model configuration space (Filali, 2010).

In enumerative combinatorics, multiplicative dynamical partition functions emerge from integrable evolution equations such as the discrete two-dimensional Toda molecule. These partition functions, $Z_\lambda(a,b;n)$, generate statistics of reverse plane partitions for arbitrary Young diagrams $\lambda$, with fields $a^{(s,t)}_n, b^{(s,t)}_n$ evolving under discrete flows. The partition function factorizes completely over cells of $\lambda$, reflecting the underlying integrability and the combinatorial encoding of nonintersecting path families via the Gessel–Viennot method. This framework recovers (as special cases) MacMahon's triple product and Gansner's multi-trace enumerators (Kamioka, 2017).

6. Physical Interpretation and Notable Applications

The dynamical partition function provides a direct analog of thermodynamic potentials, with the field $s$ playing the role of temperature or pressure, and the activity $K$ playing the role of energy or volume. For example, in high-pressure ice, there exists a critical $s_c$ where the dynamical entropy and average activity undergo sharp jumps, analogous to latent heat and first-order thermal transitions. The zeros of $Z(s)$ and associated singularities in $\psi(s)$ provide both theoretical diagnostics and practical computational schemes for DPTs in rare-event physics, glassy dynamics, and quantum quenches (Ye et al., 2022, Takahashi et al., 2013).

In quantum many-body systems, the analytic continuation of thermodynamic partition functions to imaginary parameters (e.g., $\beta = i t$) yields dynamical partition functions that control Loschmidt echo statistics and quantum quench dynamics. The structure and location of partition function zeros underpin dynamical singularities and the so-called "dynamical chaotic" phase, revealing phenomena inaccessible in equilibrium analysis (Takahashi et al., 2013). In classical integrable models and combinatorial settings, dynamical partition functions facilitate explicit determinant or multiplicative formulas, enabling direct connection between integrability, algebraic structure, and exact enumeration (Filali, 2010, Kamioka, 2017).

7. Broader Implications and Unified Theoretical Framework

The development and analysis of dynamical partition functions furnish a genuinely trajectory-based thermodynamic language, unifying rare-event physics, space-time phase transitions, quantum dynamics, and problems in dynamic combinatorics. The systematic inclusion of dynamical fields, path integral methods, and analysis of partition function zeros extends the classic theory of phase transitions to nonequilibrium and dynamical domains, supporting a unified conceptual picture across statistical physics, quantum theory, and integrable combinatorics (Ye et al., 2022, Kamioka, 2017, Filali, 2010, Takahashi et al., 2013).