Thermodynamic Uncertainty Relation

• TUR is a fundamental inequality linking current fluctuations to entropy production, defining precision–dissipation trade-offs in nonequilibrium systems.
• It has been generalized from classical Markovian and Langevin dynamics to underdamped, non-Markovian, and quantum settings, widening its applicability.
• TUR bounds can guide experimental designs to diagnose performance limits in nanoscale engines, sensors, and feedback-controlled thermodynamic systems.

The thermodynamic uncertainty relation (TUR) is a fundamental information-theoretic inequality governing precision–dissipation trade-offs in nonequilibrium thermodynamics. TURs lower-bound the relative fluctuations of empirical currents or accumulated observables in terms of entropy production or generalized irreversibility metrics, providing strict constraints on the attainable precision for finite thermodynamic cost. While originally formulated for classical Markovian jump processes and overdamped Langevin diffusions, TURs have been generalized to arbitrary initial conditions, finite-time regimes, underdamped dynamics with odd-parity variables, non-Markovian systems, quantum dynamics, systems with feedback control and memory, and to settings involving non-trivial transition structure (unidirectional steps, feedback, etc.). TURs are now a cornerstone in the quantitative theory of out-of-equilibrium thermodynamics, with direct implications for nanoscale engine design, stochastic inference, and understanding universal limits of nonequilibrium steady states.

1. TURs in Classical Markovian Processes and Langevin Dynamics

For a continuous-time Markov process or overdamped Langevin dynamics, consider an accumulated current observable $J$ over time interval $T$, with mean $\langle J \rangle$ and variance $\mathrm{Var}[J]$. The total entropy production over the same interval is $$\Sigma = \int_0^T dt \, \sum_{x \neq y, \nu} P(x;t) r^\nu(x,y) \ln \frac{P(x;t) r^\nu(x,y)}{P(y;t) r^\nu(y,x)}$$ for a rate-matrix $\mathcal{R}$ and arbitrary initial state.

The original TUR asserts

$$\frac{\mathrm{Var}[J]}{\langle J \rangle^2} \, \Sigma \ge 2,$$

in units $k_B = 1$ (Lee et al., 2021). This tightly bounds the inverse precision achievable for a fixed dissipation. For underdamped Langevin dynamics (where dynamics depend on both position and velocity, introducing parity under time reversal), the situation is complicated by the presence of velocity-dependent forces and dynamical activity. Unified bounds have the form

$$\frac{\mathrm{Var}[J]}{\Omega_T^2 (\Sigma + I)} \ge 2,$$

where $\Omega_T$ is the sensitivity of the current to scaling parameters (observation time, reversible force amplitude, length scale, driving speed), and $I$ is a Fisher-information term dependent on the initial state (negligible at long times or steady state). This "universal" TUR reduces to the overdamped form for vanishing mass or friction (Lee et al., 2021).

In velocity-dependent underdamped systems, further generalization introduces dynamical activity $\Upsilon$, leading to

$$\frac{\mathrm{Var}[J]}{\langle J \rangle^2} \ge \frac{2}{\Sigma + 4\Upsilon},$$

restoring the bound in situations where velocity-dependent forces would otherwise violate the classical TUR (Lee et al., 2019).

2. TURs for Arbitrary Initial States, Finite Time, and Discrete-Time Dynamics

Finite-time TURs valid for arbitrary initial distributions have been developed based on the Cramér–Rao/Fisher information framework (Liu et al., 2019). In continuous-time Markov processes: $\frac{\mathrm{Var}[J]}{(T j(T))^2} \Sigma \ge 2,$ where $j(T)$ is the instantaneous current at the final observation time. For discrete-time Markov chains, the bound becomes

$$\frac{\mathrm{Var}[J]}{(n j(t_{n-1}))^2} \frac{\tilde{\sigma}}{a} \ge 2,$$

with $\tilde{\sigma}$ the total entropy production plus a sum of KL divergences and $a$ the minimal staying probability. These bounds exponentially tighten earlier "exponentiated EP" TURs in the large-dissipation regime, and enable tight uncertainty relations outside the steady-state or asymptotic limits, including feedback-controlled and transient scenarios (Liu et al., 2019).

3. TURs in Quantum Thermodynamics and Quantum Transport

Quantum generalizations of the TUR bound the mean and variance of quantum observables in terms of quantum entropy production (relative entropy between evolved system and reference states) (Salazar, 2024). For forward process state $\rho$, reference $\sigma$, and Hermitian observable $\hat\theta$, the inequality is

$$\Sigma \ge F\left( \Delta\langle\hat\theta\rangle, \mathrm{Var}_\rho(\hat\theta), \mathrm{Var}_\sigma(\hat\theta) \right),$$

where $F$ is an explicitly constructed function of the differences and variances. In the incoherent (diagonal) limit this reproduces classical TURs, while in the small-perturbation regime it reduces to the quantum Cramér–Rao inequality, setting the ultimate quantum-limited precision for multiparameter estimation.

Quantum transport under steady-state conditions, e.g. electron current in atomic-scale junctions, is constrained by

$$\beta V \frac{\langle\!\langle I^2 \rangle\!\rangle}{\langle I \rangle} \ge 2,$$

with $\beta$ inverse temperature, $V$ the bias, and $\langle\!\langle I^2 \rangle\!\rangle$ the noise power (Friedman et al., 2020). For quantum-coherent, energy-independent transmission, the TUR is saturated and can be used to diagnose deviations from noninteracting transport (Agarwalla et al., 2018).

4. TURs from Exchange Fluctuation Theorems and Generalized Symmetries

Exchange fluctuation theorems (EFTs) provide a universal symmetry for systems exchanging energy or particles with multiple reservoirs, underpinning matrix-valued TURs. For joint probability $P(\Delta \mathbf{X})$ of exchanged quantities and affinities $(A_1,\dots,A_n)$,

$$\frac{P(\Delta\mathbf{X})}{P(-\,\Delta\mathbf{X})} = e^{\Sigma}, \quad \Sigma=\sum_i A_i \Delta X_i,$$

the tightest TUR bound is

$$\mathbf{z}^T (C - f(\langle \Sigma \rangle) \mathbf{q} \mathbf{q}^T) \mathbf{z} \ge 0 \quad \forall\, \mathbf{z},$$

where $C$ is the full covariance matrix, $\mathbf{q}$ the vector mean, and $f$ a universal function of $\langle \Sigma \rangle$. This constrains both variances and cross-correlations for arbitrary linear combinations of currents, and holds for both classical and quantum (non-Markovian, nonstationary) protocols (Timpanaro et al., 2019). Scalar TURs are recovered as the diagonal elements.

Moreover, the existence of an involution mapping each trajectory to its unique time reverse suffices to derive universal TURs without explicit reference to fluctuation theorems (Salazar, 2022), producing a hierarchy of bounds (exchange, asymmetric, response, entropy-production) solely from reversible symmetries of the process.

5. Specific System Extensions: Unidirectional Transitions, Memory, Feedback

For systems with intrinsically irreversible transitions (unidirectional steps), TUR bounds incorporate both "reversible" entropy production and "irreversible" net fluxes: $$\frac{\mathrm{Var}(J)}{\langle J \rangle^2} \ge \frac{2}{\Sigma_{\rm rev} + \Sigma_{\rm uni}},$$ where $\Sigma_{\rm rev}$ is computed from reversible cycles and $\Sigma_{\rm uni}$ from unidirectional steps (Pal et al., 2020). For optimality, edges are represented unidirectionally wherever $\Sigma_{\rm bi} > \Phi_{\rm uni}$, yielding tighter bounds.

In generalized Langevin systems (GLEs) with non-Markovian memory, TURs continue to hold but the bounding "cost function" involves both dissipation (heat) and ephemeral system-entropy changes, thus reflecting frenetic, non-dissipative costs not present in simple Markovian fluids (Terlizzi et al., 2020). Only in long-time/steady state or Markovian limits does the bound reduce to the conventional form purely in terms of entropy production.

TURs have further been extended to feedback-cooled underdamped systems, where a bound relates entropy reduction rate (cooling power) and information flow efficiency to variance of reversible currents: $|\dot S_{\rm red}| \le V \cdot \eta (1-\eta),$ with $V$ quantifying fluctuations of the (reversible) local mean velocity and $\eta$ the efficiency approaching unity at ideal cooling (but requiring diverging fluctuation cost) (Kumasaki et al., 8 Aug 2025).

6. TUR Validity, Violation Regimes, Quantum Signatures

The universality of TURs is conditioned by symmetry, statistics, and system features. Violations have been observed:

• In quantum collisional models, strong non-Markovianity or coherent drive can induce finite-time and steady-state violation of the classical TUR, though quantum versions remain intact due to additional dynamical activity and coherence terms (Maity et al., 2024).
• In thermal transport through spin-boson or correlated fermionic models, TURs can be violated in parameter regimes with negative skewness of the heat current distribution (memory-bearing baths, strong quantum coherence, or sharply resonant transmission), while harmonic or weak-coupling regimes always respect the bounds (Saryal et al., 2019, Saryal et al., 2020).
• In experimentally realized two-qubit quantum thermal machines, generalized TURs based on fluctuation theorems remain unbroken, but specialized Markovian bounds are strongly violated under transient or strong-coupling conditions (Pal et al., 2019).

Across these scenarios, universality holds for generalized TURs derived from detailed fluctuation theorems/relative entropy, but can break down for classical or Markovian specialized forms as soon as quantum coherence, memory, or correlation effects dominate.

7. Practical Measurement, Saturation, and Implications

TURs provide inferential lower bounds for entropy production given measurements of mean and fluctuations of (integrated) currents: assess $\langle J \rangle$ and $\mathrm{Var}[J]$ across many trajectories, estimate sensitivity prefactors (e.g. $\Omega_T$), and extract entropy production or related cost functions. The bounds can be experimentally saturated or approached via controlled protocols (e.g. tuning drive strength in quantum Ising circuits, Kalman-filtered feedback in cooling), and serve as diagnostic tools for the breakdown of classical cost–precision trade-offs. Matrix-valued TURs further constrain how tightly multiple currents or observables can co-fluctuate.

TURs thus consolidate the fluctuation–dissipation paradigm, unify classical, quantum, stochastic, and finance-theoretic precision–cost relations (Ziyin et al., 2022), and chart the performance frontier for molecular, quantum, and nano-scale engines, sensors, and information-processing devices.