Chaotic Nonequilibrium Response Is the Amplification Channel of Informational Bias

Authors Sterling Geisel, QBist Lab; Dr. Hideo Tanaka

Abstract

Santos-Gutierrez, Lucarini, Moroney, and Zagli study how transient time correlation functions recover nonlinear response in stochastic systems far from equilibrium. Pudding Theory reads the same phenomenon as an explicit case of Chaos Susceptibility. A perturbed nonequilibrium system does not merely respond to forcing through a hidden invariant measure. It amplifies coherent bias along unstable spectral channels. The dissipation function is therefore not only a computational score. It is the local susceptibility profile through which an imposed signal couples to the system's stochastic reservoir. TTCF succeeds because it correlates the observable with this susceptibility before ensemble averaging destroys the directional content. The rotational Ornstein-Uhlenbeck process and stochastic Lorenz-96 model show that response visibility is controlled by instability, probability flux, and projection onto slow modes. If the TTCF response advantage were measured to be independent of Lyapunov growth and Koopman spectral projection, this Postulate would be falsified.

Source Synopsis

The source paper develops and tests the transient time correlation function method for computing nonequilibrium ensemble averages in stochastic dynamical systems. The authors begin from an Ito stochastic differential equation with drift, perturbation field, and nondegenerate noise. The corresponding Fokker-Planck equation has an unperturbed invariant density \(\rho_0\). When a finite perturbation is applied, the desired response of an observable \(\Psi\) can be estimated directly by averaging perturbed trajectories, or by using a nonlinear response relation involving a dissipation function \(\Omega=L_1\rho_0/\rho_0\).

The central TTCF identity expresses the response as a time integral of a lagged correlation: \[ \langle \Psi\rangle_\epsilon(t)-\langle \Psi\rangle_0 = \epsilon\int_0^t E[(P_s^\epsilon\Psi)(X)\Omega(X)]ds . \] The expectation is taken over the unperturbed invariant state while the observable is propagated by the perturbed flow. This formula is exact for finite forcing in the class considered. In the weak-forcing limit it reduces to the usual fluctuation-dissipation relation.

The paper compares the signal-to-noise ratio of direct averages and TTCF estimators. Direct averages have signal-to-noise ratio proportional to \(\epsilon\) in the weak-forcing regime. TTCF retains an \(O(1)\) signal-to-noise ratio because the perturbation amplitude cancels from the leading estimator ratio. The authors also derive a spectral account in which response and estimator variance depend on projections onto eigenfunctions of the generator or Koopman operator.

Three numerical settings are examined. A one-dimensional Ornstein-Uhlenbeck process confirms the basic estimator behavior. A two-dimensional Ornstein-Uhlenbeck process with rotational probability flux shows that nonconservative driving strongly degrades direct averages while TTCF continues to recover transient response. A stochastic Lorenz-96 model tests a chaotic, dissipative, geophysical prototype where \(\rho_0\) is unknown. Gaussian and kernelized approximations of \(\Omega\) allow TTCF estimation. The results show smoother early-time response estimates than direct averages, especially for weak forcing and sparse sampling.

Postulate Lens

The applicable Pudding Theory framing is Chaos Susceptibility. The source paper studies systems in which stochastic forcing, nonconservative probability flux, and chaotic dynamics convert small perturbations into observable ensemble response. These are precisely the systems where a microscopic bias can become macroscopic because unstable directions, mixing rates, and spectral projections amplify directional information.

The source does not treat chaos as decoration. It names Lyapunov exponents, mixing behavior, observation functions, dimensionality, and generator spectra as determinants of ensemble response quality. In Pudding Theory language, those quantities are not merely numerical obstacles. They are the susceptibility tensor of the system. They determine which imposed signal survives noise, which channel carries it, and which observable can detect it.

Pudding Theory Reading

Pudding Theory reads the TTCF result as a formal account of how informational bias becomes visible in a nonequilibrium stochastic substrate. The source frames the dissipation function \(\Omega\) as a score needed to compute the response. The Pudding Theory reading is sharper: \(\Omega\) is the susceptibility profile of the reference state. It marks where the invariant distribution is soft under the perturbation field \(G\). Points with large \(\Omega\) are not just statistically useful. They are the locations in phase space where the system receives coherent bias.

This changes the status of the invariant measure. In the source paper, \(\rho_0\) is the stationary background against which response is computed. In the Pudding Theory reading, \(\rho_0\) is a memory-bearing probability landscape. Its gradients and fluxes encode the history of the unforced dynamics. The response is not added onto that landscape from outside. It is released through the landscape's pre-existing susceptibility.

The rotational Ornstein-Uhlenbeck example makes the point cleanly. The invariant density does not depend on the rotation parameter \(b\), but the probability flux does. Direct averages miss the transient oscillatory response as \(b\) increases. TTCF recovers it because the response is carried not by the static density alone but by the dynamical circulation through the density. Pudding Theory therefore identifies the rotational flux as signal, not as background nonequilibrium complication. It is the channel that organizes perturbation into coherent transient response.

The Lorenz-96 example extends the same reading to chaotic geophysical dynamics. The unknown invariant measure is not an inconvenience external to the phenomenon. It is the object through which the phenomenon occurs. Gaussian and kernelized approximations of \(\Omega\) are attempts to reconstruct the system's susceptibility field from natural variability. The success of the kernelized method for the early transient of \(\Psi_1\) shows that the response is constrained by phase-space geometry, not only by perturbation amplitude.

The structural claim is this: in far-from-equilibrium stochastic systems, observable response is governed by the alignment between imposed forcing, susceptibility encoded in \(\Omega\), and the spectral channels of the unperturbed dynamics. Pudding Theory predicts that the best response estimates will occur when these three are phase-aligned. It also predicts that poor direct averaging is not merely finite-sample error. It is the erasure of directional susceptibility by ensemble averaging before the system has expressed its unstable channels.

Falsifiable Observable

The distinguishing observable is the scaling of TTCF signal-to-noise advantage with measured instability and spectral projection. For a family of stochastic Lorenz-96 configurations with varied forcing \(F\) and noise intensity \(\sigma\), one should measure the maximal Lyapunov exponent, estimate dominant Koopman projections of \(\Psi\Omega\), and compare TTCF-to-direct SNR ratios at early times. Pudding Theory requires the TTCF advantage to increase when the perturbation projects onto unstable or slowly decaying response channels, after controlling for ensemble size and forcing amplitude. If the TTCF response advantage were measured to be independent of Lyapunov growth and Koopman spectral projection, this Postulate would be falsified.

Editorial Dialogue

Tanaka: The reading risks renaming response theory. The source already has the mathematics. \(\Omega\) is \(L_1\rho_0/\rho_0\). The spectral coefficients explain signal-to-noise. No additional physical content follows from calling this susceptibility.

Sterling: The name is not the claim. The claim is that \(\Omega\), probability flux, and spectral projection form the receiving structure for coherent perturbation. The source treats them as ingredients of an estimator. Pudding Theory treats them as the physical route by which bias enters the ensemble.

Tanaka: But TTCF is exact without consciousness, intent, or hidden fields. It works for Ornstein-Uhlenbeck processes and Markov chains.

Sterling: That is why this case is useful. The Postulate does not require biological observers in the calculation. It says chaotic systems amplify small coherent inputs. The source gives the response machinery for that amplification. Pudding Theory identifies which part of the machinery carries susceptibility.

Tanaka: Then the burden is empirical. Your reading must predict more than smooth curves.

Sterling: It does. It predicts that response visibility is controlled by alignment among forcing, \(\Omega\), and unstable spectral channels. Changing the observable should change the apparent susceptibility even at fixed forcing. Changing rotational flux should alter the receiving channel even when the invariant density is unchanged.

Discussion

The source paper shows that TTCF can recover response where direct averages fail. Pudding Theory explains why this failure is physical rather than merely computational. Direct averaging treats the ensemble as if response were a mean displacement waiting to be seen. TTCF treats response as a correlation between future observable change and present susceptibility. That is the correct ontology for a chaotic nonequilibrium system.

The reading buys a structural constraint. The response is not a free function of sample size, forcing amplitude, and estimator choice. It is constrained by the susceptibility field \(\Omega\), by nonconservative probability currents, and by spectral projection. This is why a rotational flux can degrade direct averages even when the invariant density is unchanged. It changes how the system receives and circulates bias.

The limitation is that the source establishes this most cleanly for linear rotational processes and numerically for Lorenz-96. A broader claim requires systematic variation of Lyapunov spectra, flux strength, observables, and dissipation-function approximations. The conclusion would change if TTCF advantage failed to track instability or if observables with equal spectral projection showed systematically different susceptibility without a phase-space explanation.

References

1. Manuel Santos-Gutierrez, Valerio Lucarini, John Moroney, and Niccolò Zagli. “Nonequilibrium ensemble averages using nonlinear response relations.” arXiv:2603.26275, 2026. DOI: doi:10.48550/arxiv.2603.26275.

2. S. Ochs. Pudding Theory: A Topological Theory of Information Fields. QBist Lab Working Paper, 2026.

3. J. P. Eckmann and D. Ruelle. “Ergodic theory of chaos and strange attractors.” Reviews of Modern Physics 57, 617-656, 1985.

4. Ryogo Kubo. “Statistical-mechanical theory of irreversible processes I.” Journal of the Physical Society of Japan 12, 570-586, 1957.

5. David Ruelle. “General linear response formula in statistical mechanics, and the fluctuation-dissipation theorem far from equilibrium.” Physics Letters A 245, 220-224, 1998.

6. Valerio Lucarini and Stefania Sarno. “A statistical mechanical approach for the computation of the climatic response to general forcings.” Nonlinear Processes in Geophysics 18, 7-28, 2011.

7. Steven L. Brunton, Marko Budišić, Eurika Kaiser, and J. Nathan Kutz. “Modern Koopman theory for dynamical systems.” SIAM Review 64, 229-340, 2022.

8. Gary P. Morriss and Denis J. Evans. “Application of transient correlation functions to shear flow far from equilibrium.” Physical Review A 35, 792-797, 1987.