Metriplectic Lotka--Volterra systems with noise:interactions invisible at stationarity, visible in time
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Mathematical Physics. The day in mathematical physics: rigorous models of physical systems.
Kai: I'm Kai.
Mira: And I'm Mira, here with Lev.
Kai: We're looking at the paper: "Metriplectic Lotka--Volterra systems with noise".
Mira: Here's the gist of the paper. The paper shows that Lotka–Volterra systems with noise admit a metriplectic structure if and only if Volterra’s dissipativity condition holds, and this structure reveals interactions invisible at stationarity but visible in time The stationary distribution determines only the equilibrium and the matrix C/k, while the derivative at zero lag of the covariance of log-abundances equals A(C/k)−1 Metriplectic Structure and Conditions Dissipative LV systems admit a metriplectic structure with energy VC + s and entropy s if and only if Volterra’s dissipativity condition CA + A⊤C ≤ 0 holds This structure is constructed explicitly with a log-canonical Poisson bivector and the symmetric part of AC−1 as the metric The necessity of Volterra’s condition holds for an arbitrary bivector and an arbitrary metric Noise Selection and Stationary Law The fluctuation–dissipation relation of stochastic GENERIC selects a family of compatible noises whose covariance is fixed by A, C and a temperature k With this noise the energy VC + s is conserved exactly The stationary law is an explicit Gibbs measure, a product of independent Gamma distributions, for every interaction matrix satisfying Volterra’s condition, predator–prey interactions included Stationary Blindness and Dynamical Identifiability Given a product of Gamma distributions, the interaction matrices that produce it as the stationary law form an open cone of dimension n 2 Every such system has exactly the same stationary law as a system of uncoupled stochastic logistic equations driven by the same noise The derivative at zero lag of the covariance R(τ) of the log-abundances satisfies the exact identity R˙(0+) = A Cov(x, log x) This allows for recovering the interaction matrix from one-time moments and short-lag correlations Time Asymmetry and Entropy Production The stationary entropy production rate of the compatible system is given in closed form, does not depend on k, and vanishes if and only if the Poisson part of the structure vanishes The whole irreversibility of the stationary stochastic process is carried by the Poisson part The symmetric part of R˙(0+) is fixed by the noise, and its antisymmetric part is exactly the time asymmetry of the cross-covariances of the log-abundances Rigidity and Bounds on Correlations For an arbitrary constant noise covariance Q, any stationary solution satisfies the exact Lyapunov-type identity AK + K⊤A⊤ + Q = 0 The stationary populations are independent if and only if the noise is compatible The distance of Q to the compatible family controls the size of the correlations quantitatively, as shown by Proposition 7.4 Ecological Implications The results show that stationary abundance statistics, such as Gamma-distributed and uncorrelated fluctuations, may conceal interspecific interactions Short-lag correlations of log-abundances identify the interaction matrix exactly, without the linearization on which multivariate autoregressive methods rely The stationary correlations between species reflect the interplay between noise and dissipation, rather than providing a direct measure of interaction strength Numerical Illustration Simulations confirm these results, showing that the off-diagonal entries of the predator–prey system, which encode the interaction, are recovered to within 0.6% The time series distinguish the two communities, and applying Corollary 6.4 recovers the interaction matrix with errors of a few percent The stationary entropy production rate grows without bound as self-limitation decreases, in agreement with Theorem 6.5 Concluding Remarks The Volterra function determines the stationary law, the metric determines the noise, and the Poisson bivector is invisible at stationarity The reversible part of the interaction is exactly the time asymmetry of these correlations The exact identity R˙(0+) = A Cov(x, log x), valid for any noise, recovers the interaction matrix from onetime moments and short-lag correlations of the log-abundances The stationary entropy production rate is carried entirely by the Poisson part The exact identity AK + K⊤A⊤ + Q = 0 shows that the stationary populations are independent if and only if the noise is compatible The lower bound of Proposition 7.4 is nearly sharp over two decades of kappa, in agreement with our two-species example The scope of these conclusions should be stated clearly References
1: G.
Kai: Biroli, G.
Mira: Bunin, C.
Kai: Cammarota, Marginally stable equilibria in critical ecosystems, New J.
Title and authors: Kai: So we're looking at this paper titled "Metriplectic Lotka--Volterra systems with noise:interactions invisible at stationarity, visible in time". It deals with how you can structure a system of predator and prey dynamics when there's some randomness involved.
Mira: Right, it’s about showing that these systems admit a metriplectic structure if and only if Volterra's dissipativity condition holds, which is CA plus A transpose C less than or equal to zero. It builds an explicit structure using a log-canonical Poisson bivector and the symmetric part of AC minus one as the metric.
Lev: From an engineering standpoint, that condition CA plus A transpose C less than or equal to zero is key because it dictates whether this energy VC plus s and entropy s can be conserved exactly. If that inequality isn't met, you lose that precise control over the dynamics.
Kai: And then they add noise through the fluctuation dissipation relation of stochastic GENERIC, which fixes the covariance of that multiplicative environmental noise using the symmetric part of AC minus one. This leads to a stationary law that’s an explicit Gibbs measure, specifically a product of independent Gamma distributions for every interaction matrix.
Mira: That's interesting because most Gibbs measures known for these systems are only available when you have symmetric interactions, but this holds even for nonsymmetric ones like predator and prey pairs. The authors show this is possible with multiplicative environmental noise, which is what you get from the GENERIC framework.
Lev: For us on hardware, having that stationary law be a product of Gamma distributions means the statistical description is simple enough to handle computationally, but we still have to deal with how that noise covariance Q relates to the dissipation matrix A and C.
Kai: The paper then shows that given a set of interaction matrices satisfying Volterra's condition, there’s an open cone of dimension n two where those matrices can produce that specific stationary law. This is a big statement about the space of possible ecological interactions.
Mira: It means every system in that cone has the exact same stationary law as a simpler system with no interactions at all, driven by the same noise. That’s because every such system has exactly the same stationary law as a system of uncoupled stochastic logistic equations driven by the same noise, which reproduces those Gamma statistics we see in microbial communities.
Lev: If that's true, then for us building simulators or models, it suggests we don't need to model every single interaction term explicitly if we only care about the final abundance distribution. But what about identifying which interactions are actually there?
Title and authors: Kai: That’s where they get into dynamical identifiability. They show that for any stationary stochastic Lotka--Volterra system with a constant noise covariance, the derivative at zero lag of the covariance R tau of the log abundances satisfies an exact identity called R dot zero plus equals A Cov x log x.
Mira: That identity is powerful because it means you can recover the interaction matrix from just one time moment and short lag correlations of those log abundances, which is a direct nonlinear counterpart to methods used in multivariate autoregressive models.
Lev: If we can recover the matrix that way, that’s a huge advantage for experimental work. It bypasses some of the linearization hurdles we usually face when trying to map time series data back onto a full interaction structure.
Kai: Now let's talk about what they suggest improving this idea. The paper points out that the stationary entropy production rate of the compatible system is given in closed form, and it doesn't depend on k, temperature. It also vanishes if and only if the Poisson part of the structure goes to zero.
Mira: And that tells us that all the irreversibility in this stationary process comes entirely from the Poisson part of the structure. The symmetric part of R dot zero plus is fixed by our noise, but its antisymmetric part is what captures the time asymmetry of cross covariances between those log abundances.
Lev: So, it seems like we can use that antisymmetric part to directly measure how interactions are influencing the time evolution, even when you’re looking at a steady state where things look static. That connects the stationary analysis back to real-time dynamics.
Kai: They also have this rigidity result where for an arbitrary constant noise covariance Q, any stationary solution must satisfy an exact Lyapunov type identity: A K plus K transpose A transpose plus Q equals zero. This is important because it tells us when the populations are truly independent, which happens if and only if the noise is compatible with the structure.
Mira: And they quantify how much freedom you have in estimating interactions by looking at the distance of that noise covariance Q to the compatible family. The lower bound from proposition seven point four shows how small those correlations can be, depending on kappa over two decades of kappa, which they say is nearly sharp for their two species example.
Lev: Quantifying the lower bound is useful because it sets a concrete limit on what you can observe in your data before you have to assume interactions are present when the noise structure might just be mismatched.
Title and authors: Kai: So we’ve seen that stationary abundance statistics, like Gamma-distributed fluctuations, can hide interspecific interactions. But short lag correlations of log abundances allow us to identify the interaction matrix exactly without relying on linear approximations.
Mira: The whole point is that the reversible part of the interaction is exactly captured by this time asymmetry in those correlations, which was previously invisible at stationarity. The paper "Metriplectic Lotka--Volterra systems with noise:interactions invisible at stationarity, visible in time" suggests we need to look beyond just the steady-state distribution.
Lev: For real hardware applications, this means that if you measure the short lag correlations correctly, you can get the interaction matrix back with a few percent error. That’s much better than relying on fitting complex nonlinear models to noisy data.
Kai: To wrap things up, the paper shows that the Volterra function determines the stationary law, while the metric determines which noise structure is compatible with it. The Poisson bivector itself is invisible at stationarity.
Mira: And specifically, that reversible part of the interaction is exactly reflected in those correlations between species. This entire framework hinges on that exact identity R dot zero plus equals A Cov x log x, which works for any noise you choose.
Lev: The final identity AK plus K transpose A transpose plus Q equals zero confirms that populations are independent if and only if the noise is compatible with the structure. That’s a clean condition we can check in simulations or experiments to see if we are seeing true interactions or just noise mismatch.
Kai: So, looking at the full scope of this work on "Metriplectic Lotka--Volterra systems with noise:interactions invisible at stationarity, visible in time", it’s about how structure and observation can be separated in these complex ecological models.
Mira: It gives us tools to look at ecological data and say that the stationary distribution alone isn't enough to tell the full story of what's happening in those systems. The way this paper uses the metriplectic framework lets us see the interactions that are otherwise hidden at stationarity but become visible in time through these specific correlation measures.
Lev: For me, it’s about establishing a rigorous way to connect the microscopic dynamics, like those described by A and C, to macroscopic observable statistics like those Gamma distributions. It shows how the noise selection process is intrinsically linked to the energy conservation of that metriplectic structure.
Kai: So we’ve seen that even in systems where you don't explicitly model interactions, short lag correlations give us the interaction matrix exactly. It moves ecological inference from a purely statistical fit to a dynamical recovery problem.
Title and authors: Mira: It suggests that the stationary entropy production rate is carried entirely by the Poisson part of the structure, which is what drives all the irreversible behavior we observe in these processes. We need to focus our analysis there if we want to understand how energy flows in these ecosystems.
Lev: The paper points out that some limitations are that this entire framework relies on Volterra's dissipativity condition holding for the specific structure they construct, and the noise selection process is tied to a very specific fluctuation dissipation relation of GENERIC. If your real-world system doesn't fit those assumptions, you might not get this exact conservation result.
Kai: So, looking at "Metriplectic Lotka--Volterra systems with noise:interactions invisible at stationarity, visible in time", the main points are that we need to look at time dynamics to see interactions that are otherwise hidden in the stationary distribution. And we have a concrete way to recover those interaction matrices from short lag correlations of log abundances.
Mira: It's a big step because it connects the mathematical structure of the system, like the metriplectic one, directly to observable ecological statistics like Gamma distributions and time-dependent cross covariances. The whole idea is that these two things are deeply linked through this structure.
Lev: I think what’s most exciting for hardware research is how this links to error correction because the noise compatibility test gives us a concrete way to check if the noise we're adding aligns with the fundamental dissipation matrix A and C of our physical system.
Kai: Exactly, it lets us use these correlation measures not just as descriptive statistics but as tools to validate whether our model assumptions about the underlying physics are actually holding in the presence of environmental noise.
Mira: And for those of us working on theory, this provides a new lens: we stop thinking about interactions only in terms of equilibrium distributions and start seeing them as something that manifests dynamically through time evolution.
Lev: So, if you're interested in applying this to modeling complex systems, the paper shows that the dynamics themselves carry the information about interactions that is otherwise static in the stationary state.
Kai: That’s what it boils down to. We've got a lot of work left on building up these experimental setups and testing these dynamical identifiability results on real physical systems.
Mira: Definitely. The next step is seeing how robust these results are when we move away from the ideal Volterra condition or when the noise doesn't perfectly follow that specific fluctuation dissipation relation.
Lev: Agreed. We need to see where the bounds of this structure truly lie when you introduce more complex, non-symmetric interactions that aren't just simple predator prey pairs.
The paper's summary: Kai: So, this paper shows that Lotka--Volterra systems with noise have a specific structure called a metriplectic structure if Volterra’s dissipativity condition is met.
Mira: And what this means is that you get an energy VC plus s and an entropy s, and it's built using some log-canonical Poisson bivector and the symmetric part of AC minus one as the metric.
Lev: That condition CA plus A transpose C less than or equal to zero is really the gatekeeper here. If that doesn't hold, you don't get that exact energy conservation, so it's not a stable structure for us on hardware.
Kai: Right, and they also talk about noise selection through the fluctuation dissipation relation of stochastic GENERIC. This fixes the covariance of that multiplicative environmental noise using A, C and a temperature k.
Mira: And with this specific noise, you get a stationary law that is an explicit Gibbs measure—a product of independent Gamma distributions for every interaction matrix. Even for those tricky predator and prey pairs.
Lev: From my side, I see the immediate problem as how to run that on real hardware without massive overhead. A product of Gamma distributions sounds computationally nice, but we have to deal with the underlying dynamics first.
Kai: The paper then tackles what you see at stationarity versus what happens in time with dynamical identifiability. They show that the derivative at zero lag of the covariance of log abundances equals A Cov x log x.
Mira: So this means you can recover that interaction matrix from just one moment and short lag correlations of those log abundances, which is a nonlinear way to do it, unlike standard multivariate autoregressive methods.
Lev: That’s a huge win for data analysis because it bypasses the linearization problems we usually run into when trying to map time series back onto a full interaction structure.
Kai: But there's more about time asymmetry. The stationary entropy production rate is given in closed form, and it doesn't depend on k, temperature. It only vanishes if the Poisson part of the structure goes to zero.
Mira: So all the real irreversibility in this stationary process comes entirely from that Poisson part of the structure, and the symmetric part of that covariance derivative captures what they call time asymmetry.
Lev: That antisymmetric part is what reflects how interactions are actually influencing things over time, which was otherwise invisible when you just looked at the steady state.
Kai: They also have this rigidity result where for any noise covariance Q, any stationary solution must satisfy an exact Lyapunov type identity: A K plus K transpose A transpose plus Q equals zero.
Mira: That identity shows that the populations are truly independent if and only if the noise is compatible with that structure we built. And they quantify how much freedom you have in estimating interactions by looking at the distance of that noise covariance Q to the compatible family.
The paper's summary: Lev: Quantifying that lower bound is useful because it sets a concrete limit on what correlations you can actually observe before you have to assume interactions are present when the noise structure might just be mismatched.
Kai: So, essentially, they’re saying stationary abundance statistics like Gamma distributions can hide interspecific interactions, but short lag correlations of log abundances allow us to identify the interaction matrix exactly.
Mira: The whole point is that the reversible part of the interaction is exactly captured by this time asymmetry in those correlations, which was previously invisible at stationarity.
Lev: It connects the mathematical structure of the system to observable ecological statistics like Gamma distributions, and it shows how those two things are deeply linked through this metriplectic framework.
Kai: The final identity AK plus K transpose A transpose plus Q equals zero confirms that populations are independent if and only if the noise is compatible with the structure.
Mira: It suggests that we need to stop thinking about interactions only in terms of equilibrium distributions and start seeing them as something that manifests dynamically through time evolution.
Lev: If you're interested in applying this to modeling complex systems, it means the dynamics themselves carry the information about interactions that is otherwise static in the stationary state.
Kai: So we've got a concrete way to recover those interaction matrices from short lag correlations of log abundances, moving ecological inference from a purely statistical fit to a dynamical recovery problem.
Mira: It’s about using these correlation measures not just as descriptive statistics but as tools to validate whether our model assumptions about the underlying physics are actually holding in the presence of environmental noise.
Lev: I think what's most exciting for hardware research is how this links to error correction because the noise compatibility test gives us a concrete way to check if the noise we're adding aligns with our fundamental dissipation matrix A and C.
Kai: Exactly, it lets us use these correlation measures not just as descriptive statistics but as tools to validate whether our model assumptions about the underlying physics are actually holding in the presence of environmental noise.
Mira: It gives us tools to look at ecological data and say that the stationary distribution alone isn't enough to tell the full story of what's happening in those systems.
Lev: For me, it’s about establishing a rigorous way to connect the microscopic dynamics, like those described by A and C, to macroscopic observable statistics like those Gamma distributions. It shows how the noise selection process is intrinsically linked to the energy conservation of that metriplectic structure.
Kai: So we’ve seen that even in systems where you don't explicitly model interactions, short lag correlations give us the interaction matrix exactly. It moves ecological inference from a purely statistical fit to a dynamical recovery problem.
The paper's improvements: Tom: So we’re looking at how this paper improves on the initial model of Lotka--Volterra systems with noise by showing how you can actually use that structure in a more useful way.
Kai: The main improvement is about adaptive noise selection. They propose using a specific method, Corollary four point four, to select a family of compatible environmental noises whose covariance is fixed by A, C and k.
Mira: And what that does is it ensures the energy VC plus s remains conserved exactly under the stochastic dynamics. That's crucial because without exact conservation, you lose control over the system's stability and behavior.
Lev: From a hardware standpoint, that’s a big deal for error correction setups because it means we can design our noise injection to match the fundamental dissipation matrix A and C of our physical system.
Kai: Then there's the Stationary Distribution Prediction part. Given any set of interaction matrices that satisfy Volterra’s condition, the AI can predict that the stationary law will be an explicit Gibbs measure, a product of independent Gamma distributions for every interaction matrix.
Mira: That’s important because it confirms that we don't need symmetric interactions to get those simple Gamma statistics; this holds even for nonsymmetric predator and prey pairs.
Lev: That makes the statistical description much more powerful for modeling real ecological systems, even if the underlying dynamics are complex and not symmetric.
Kai: The next big thing is about interaction matrix recovery from time series. They show that you can recover the interaction matrix by calculating the derivative at zero lag of the covariance of log abundances, which is A Cov x log x.
Mira: This means you can get the interaction matrix from just one moment and short lag correlations of those log abundances, providing an exact nonlinear counterpart to those multivariate autoregressive models we use.
Lev: That’s a huge win for data analysis because it bypasses the linearization problems we usually run into when trying to map time series back onto a full interaction structure.
Kai: They also focus on using the time asymmetry of lagged cross-covariances R xy tau and R yx tau to distinguish between different community structures, like cyclic predation versus symmetric competition.
Mira: That asymmetry is what reveals that Poisson part of the structure we talked about earlier, which was otherwise invisible at stationarity.
Lev: So they’re showing how to use time evolution data to find structural information that the steady state just completely hides.
Kai: Finally, they discuss noise compatibility testing. You can test if observed correlations between species are due to real interactions or just a mismatch between noise and dissipation by checking if the covariance of species i and log species j is zero for all i not equal j.
Mira: That’s equivalent to testing if the noise covariance Q is compatible with that metriplectic structure we defined at the start.
Lev: That gives us a concrete way to check if our model assumptions about the underlying physics are actually holding in the presence of environmental noise, which is exactly what we need for robust error correction.
Kai: So these improvements show that we can use dynamic data to find hidden structure and test if our physical noise setup is correct.
Mira: The whole point is that this framework connects the mathematical structure of the system directly to observable ecological statistics like Gamma distributions and time-dependent cross covariances.
Lev: It’s a big step because it moves us toward using these correlation measures not just as descriptive statistics but as tools to validate whether our model assumptions about the underlying physics are actually holding in the presence of environmental noise.
Kai: The next thing we need to figure out is how robust these results are when you move away from the ideal Volterra condition or when the noise doesn't perfectly follow that specific fluctuation dissipation relation.
Conclusion: Kai: So we’ve covered how the paper "Metriplectic Lotka--Volterra systems with noise:interactions invisible at stationarity, visible in time" shows that you need to look at time dynamics to see interactions hidden in a stationary state.
Mira: Exactly, because they connect the mathematical structure of the system directly to observable ecological statistics like Gamma distributions and time-dependent cross covariances.
Lev: For me, it’s about establishing a rigorous way to connect the microscopic dynamics, like those described by A and C, to macroscopic observable statistics like those Gamma distributions. It shows how the noise selection process is intrinsically linked to the energy conservation of that metriplectic structure.
Kai: That's what it boils down to; we move from just looking at steady-state pictures to understanding what happens as time evolves in these complex systems.
Mira: They give us tools to look at ecological data and say that the stationary distribution alone isn't enough to tell the full story of what's happening in those systems.
Lev: The next thing we need to figure out is how robust these results are when you move away from the ideal Volterra condition or when the noise doesn't perfectly follow that specific fluctuation dissipation relation.
Kai: Right, so if your real-world system doesn't fit those specific assumptions, you might not get that exact energy conservation result.
Mira: And we have to be careful; this whole framework relies on Volterra’s dissipativity condition holding for the specific structure they construct.
Lev: That’s a fair caveat because in real-world biology, things rarely fit perfect mathematical models, so knowing the limits of that assumption is essential.
Kai: Anyway, looking at "Metriplectic Lotka--Volterra systems with noise:interactions invisible at stationarity, visible in time", the main point is that we can recover interaction matrices from short lag correlations of log abundances.
Mira: It’s a big step because it connects the mathematical structure of the system directly to observable ecological statistics like Gamma distributions and time-dependent cross covariances.
Lev: I think what's most exciting for hardware research is how this links to error correction because the noise compatibility test gives us a concrete way to check if the noise we're adding aligns with our fundamental dissipation matrix A and C.
Kai: Exactly, it lets us use these correlation measures not just as descriptive statistics but as tools to validate whether our model assumptions about the underlying physics are actually holding in the presence of environmental noise.
Mira: So they show that stationary entropy production is carried entirely by the Poisson part of the structure, which is what drives all the irreversible behavior we observe in these processes.
Lev: That means if you want to understand energy flow in these ecosystems, you need to focus your analysis there because that’s where the real irreversibility lies.
Kai: It’s a lot of work left on building up experimental setups and testing those dynamical identifiability results on real physical systems.
Mira: Definitely. The next step is seeing how robust these results are when you move away from the ideal Volterra condition or when the noise doesn't perfectly follow that specific fluctuation dissipation relation.
JOSE A. CAPIT ¡N, ALBERTO MEG ¡IAS, CRISTINA SARDON¡
math-ph, math.MP
Submitted: 2026-11-11
Updated: 2026-11-11
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 81/100
The gist: The paper shows that Lotka–Volterra systems with noise admit a metriplectic structure if and only if Volterra’s dissipativity condition holds…
Key concepts
- Metriplectic Structure
- This is a mathematical framework describing the system's dynamics using energy (VC+s) and entropy (s). It requires Volterra’s dissipativity condition to exist, providing a geometric description of how the system evolves over time.
- Volterra’s Dissipativity Condition
- This condition, CA + A⊤C ≤ 0, is necessary for the system to have a metriplectic structure. It ensures that the dissipative nature of the Lotka–Volterra dynamics is compatible with the required geometric framework for energy and entropy conservation.
- Stationary Blindness
- At equilibrium (stationarity), complex interspecies interactions become invisible in abundance statistics. The stationary distribution only reveals basic equilibrium parameters like C/k, hiding the detailed interaction matrix until time-dependent correlations are analyzed.
Terminology
Summary
The paper shows that Lotka–Volterra systems with noise admit a metriplectic structure if and only if Volterra’s dissipativity condition holds, and this structure reveals interactions invisible at stationarity but visible in time The stationary distribution determines only the equilibrium and the matrix C/k, while the derivative at zero lag of the covariance of log-abundances equals A(C/k)−1
Metriplectic Structure and Conditions
Dissipative LV systems admit a metriplectic structure with energy VC + s and entropy s if and only if Volterra’s dissipativity condition CA + A⊤C ≤ 0 holds This structure is constructed explicitly with a log-canonical Poisson bivector and the symmetric part of AC−1 as the metric The necessity of Volterra’s condition holds for an arbitrary bivector and an arbitrary metric
Noise Selection and Stationary Law
The fluctuation–dissipation relation of stochastic GENERIC selects a family of compatible noises whose covariance is fixed by A, C and a temperature k With this noise the energy VC + s is conserved exactly The stationary law is an explicit Gibbs measure, a product of independent Gamma distributions, for every interaction matrix satisfying Volterra’s condition, predator–prey interactions included
Stationary Blindness and Dynamical Identifiability
Given a product of Gamma distributions, the interaction matrices that produce it as the stationary law form an open cone of dimension n 2 Every such system has exactly the same stationary law as a system of uncoupled stochastic logistic equations driven by the same noise The derivative at zero lag of the covariance R(τ) of the log-abundances satisfies the exact identity R˙(0+) = A Cov(x, log x) This allows for recovering the interaction matrix from one-time moments and short-lag correlations
Time Asymmetry and Entropy Production
The stationary entropy production rate of the compatible system is given in closed form, does not depend on k, and vanishes if and only if the Poisson part of the structure vanishes The whole irreversibility of the stationary stochastic process is carried by the Poisson part The symmetric part of R˙(0+) is fixed by the noise, and its antisymmetric part is exactly the time asymmetry of the cross-covariances of the log-abundances
Rigidity and Bounds on Correlations
For an arbitrary constant noise covariance Q, any stationary solution satisfies the exact Lyapunov-type identity AK + K⊤A⊤ + Q = 0 The stationary populations are independent if and only if the noise is compatible The distance of Q to the compatible family controls the size of the correlations quantitatively, as shown by Proposition 7.4
Ecological Implications
The results show that stationary abundance statistics, such as Gamma-distributed and uncorrelated fluctuations, may conceal interspecific interactions Short-lag correlations of log-abundances identify the interaction matrix exactly, without the linearization on which multivariate autoregressive methods rely The stationary correlations between species reflect the interplay between noise and dissipation, rather than providing a direct measure of interaction strength
Numerical Illustration
Simulations confirm these results, showing that the off-diagonal entries of the predator–prey system, which encode the interaction, are recovered to within 0.6% The time series distinguish the two communities, and applying Corollary 6. p. 14 recovers the interaction matrix with errors of a few percent The stationary entropy production rate grows without bound as self-limitation decreases, in agreement with Theorem 6. p. 25
Concluding Remarks
The Volterra function determines the stationary law, the metric determines the noise, and the Poisson bivector is invisible at stationarity The reversible part of the interaction is exactly the time asymmetry of these correlations The exact identity R˙(0+) = A Cov(x, log x), valid for any noise, recovers the interaction matrix from onetime moments and short-lag correlations of the log-abundances The stationary entropy production rate is carried entirely by the Poisson part The exact identity AK + K⊤A⊤ + Q = 0 shows that the stationary populations are independent if and only if the noise is compatible The lower bound of Proposition 7.4 is nearly sharp over two decades of kappa, in agreement with our two-species example The scope of these conclusions should be stated clearly
References
[1] G. Biroli, G. Bunin, C. Cammarota, Marginally stable equilibria in critical ecosystems, New J. p. 1 Phys. 20 (2018) 083051
[2] A. Altieri, F. Roy, C. Cammarota, G. Biroli, Properties of equilibria and glassy phases of the random Lotka–Volterra model with demographic noise, Phys. p. 0 Rev. Lett. 126 (2021) 258301
[3] G. Bunin, Ecological communities with Lotka–Volterra dynamics, Phys. p. 0 Rev. E 95 (2017) 042414
[4] P. Duarte, R.L. Fernandes, W.M. Oliva, Dynamics of the attractor in the Lotka–Volterra equations, J. p. 0 Differential Equations 149 (1998) 143–189
[5] N.S. Goel, S.C. Maitra, E.W. Montroll, On the Volterra and other nonlinear models of interacting populations, Rev. p. 1 Mod. Phys. 43 (1971) 231–276
[6] J. Grilli, Macroecological laws describe variation and diversity in microbial communities, Nat. p. 1 Commun. 11 (2020) 4743
[7] M.Y. Gueddari, W. Hachem, Spin glass analysis of the invariant distribution of a Lotka–Volterra SDE with a large random interaction matrix, arXiv:2510. p. 115754 (2025)
[8] M. Grmela, H.C. Ottinger, Dynamics and thermodynamics of complex fluids. p. 0 I. Development of a general formalism, Phys. p. 0 Rev. E 56 (1997) 6620–6632
[9] P. Guha, Metriplectic structure, Leibniz dynamics and dissipative systems, J. p. 0 Math. Anal. Appl. 326 (2007) 121–136
[10] A. Hening, D.H. Nguyen, Coexistence and extinction for stochastic Kolmogorov systems, Ann. p. 1 Appl. Probab. 28 (2018) 1893–1942
[11] J. Hofbauer, K. Sigmund, Evolutionary Games and Population Dynamics, Cambridge University Press, 1998
[12] C.-R. Hwang, S.-Y. Hwang-Ma, S.-J. Sheu, Accelerating Gaussian diffusions, Ann. Appl. Probab. 3 (1993) 897–913
[13] G. Ivan, M. Ivan, C. Pop, Numerical integration and synchronization for the 3-dimensional metriplectic Volterra system, Math. p. 0 Probl. Eng. 2011 (2011) 723629
[14] A.R. Ives, B. Dennis, K.L. Cottingham, S.R. Carpenter, Estimating community stability and ecological interactions from time-series data, Ecol. p. 2 Monogr. 73 (2003) 301–330
[15] D.-Q. Jiang, M. Qian, M.-P. Qian, Mathematical Theory of Nonequilibrium Steady States, Lecture Notes in Mathematics 1833, Springer, 2004
[16] A.N. Kaufman, Dissipative Hamiltonian systems: a unifying principle, Phys. p. 0 Lett. A 100 (1984) 419–422
[17] E.H. Kerner, A statistical mechanics of interacting biological species, Bull. p. 0 Math. Biophys. 19 (1957) 121–146
[18] E.H. Kerner, Further considerations on the statistical mechanics of biological associations, Bull. p. 0 Math. Biophys. 21 (1959) 217–255
[19] R. Khasminskii, Stochastic Stability of Differential Equations, 2nd ed. p. 0, with contributions by G. p. 1N. Milstein and M.B. Nevelson, Stochastic Modelling and Applied Probability 66, Springer, 2012
[20] X. Mao, S. Sabanis, E. Renshaw, Asymptotic behaviour of the stochastic Lotka–Volterra model, J. p. 0 Math. Anal. Appl. 287 (2003) 141–156
[21] P.J. Morrison, Bracket formulation for irreversible classical fields, Phys. p. 1 Lett. A 100 (1984) 423–427
[22] P.J. Morrison, A paradigm for joined Hamiltonian and dissipative systems, Physica D 18 (1986) 410–419
[23] H.C. Ottinger, ¨ Beyond Equilibrium Thermodynamics, Wiley, 2005
[24] M.A. Peletier, M. Seri, A geometric formulation of GENERIC stochastic differential equations, arXiv:2509. p. 2
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Adaptive Noise Selection: The AI can select a family of compatible environmental noises whose covariance is fixed by A, C, and k using Corollary 4.4, which ensures that the energy VC + s is conserved exactly under stochastic dynamics.
-
Stationary Distribution Prediction: Given a set of interaction matrices satisfying Volterra's condition (CA + A⊤C ≤ 0), the AI can predict the stationary law as an explicit Gibbs measure, a product of independent Gamma distributions, even for nonsymmetric interactions like predator–prey pairs.
-
Interaction Matrix Recovery from Time Series: The AI can recover the interaction matrix by calculating R˙(0+) = A Cov(x, log x) from one-time moments and short-lag correlations of log-abundances, providing an exact nonlinear counterpart to MAR models.
-
Sensitivity to Interaction Structure: The AI can distinguish between different community structures (e.g., cyclic predation versus symmetric competition) by analyzing the time asymmetry of lagged cross-covariances Rxy(tau) and Ryx(tau), which are invisible at stationarity but reveal the Poisson part of the structure.
-
Correlation Lower Bound Estimation: The AI can quantify how much freedom is left in interaction estimation by calculating dF (Q, QA), which provides a lower bound for cross-covariances between species in terms of the distance of the noise covariance Q to the compatible family.
-
Noise Compatibility Testing: The AI can test whether observed correlations between species are due to interactions or merely a mismatch between noise and dissipation by checking if Cov(xi, log xj) is zero for all i not equal j, which is equivalent to testing if the noise covariance Q is compatible with a metriplectic structure.
Sources
- Spin glass analysis of the invariant distribution of a Lotka-Volterra SDE with a large random interaction matrix
- A geometric formulation of GENERIC stochastic differential equations
Related papers
- Existence of a fluid phase for the two-dimensional one-component plasma
- The Huang-Yang conjecture for the low-density Fermi gas
- Anderson localisation in infinite and finite resonator crystals with long-range interactions
- Free energy and spectral edge of the SYK model
- Anderson localization for 1-d quasi-periodic Schr"odinger operators with degenerate weights
- Irregular SLE(4) martingales and isomonodromic deformations