Zero-temperature dynamics of the spherical model with non-reciprocal interactions

This paper analytically solves the zero-temperature dynamics of the spherical model with non-reciprocal interactions, revealing that antisymmetric couplings induce a transition from slow power-law relaxation to exponential decay and eventually to damped oscillatory behavior, thereby breaking time-translation invariance and preventing equilibrium.

The Gist

Imagine a giant, bustling dance floor filled with thousands of dancers. Each dancer is connected to every other dancer by an invisible spring. This is our "Spherical Model." In physics, we use this model to understand how complex systems—like brains, ecosystems, or even the spread of a virus—behave when they are out of balance.

Usually, in these dance floors, if Dancer A pulls on Dancer B, Dancer B pulls back with the exact same force. This is called reciprocity (or symmetry). It's like a fair handshake.

But in this paper, the authors ask: What happens if the handshake isn't fair? What if Dancer A pulls hard on B, but B barely pulls back? This is called non-reciprocal interaction. The authors wanted to see how this "unfairness" changes the dance, specifically when the music stops (zero temperature) and the dancers are just trying to settle down.

Here is the story of their discovery, broken down into simple concepts:

1. The Old Way: The Slow, Sticky Dance (Symmetric)

In the past, scientists studied the "fair handshake" version. They found that when the music stopped, the dancers didn't just stop immediately. Instead, they got stuck in a slow, sluggish state.

• The "Aging" Effect: If you waited a long time before checking on the dancers, they would still be moving very slowly. The longer you waited, the slower they seemed to relax. It's like trying to untangle a knot of wet spaghetti; the more time passes, the harder it seems to get it to stop moving.
• The Pattern: Their movement followed a "power law," which is a fancy way of saying they slowed down gradually, like a car coasting to a stop without brakes.

2. The New Discovery: The Unfair Handshake (Non-Reciprocal)

The authors introduced the "unfair handshake" (where the pull isn't equal). They found that this completely changes the rules of the dance floor.

The Big Surprise: No More Slow Motion
When the interactions are unfair, the dancers don't get stuck in that slow, aging state. Instead, they relax exponentially.

• The Analogy: Imagine a ball rolling down a hill. In the old "fair" model, the ball rolls down a muddy, sticky slope, slowing down more and more the further it goes. In the new "unfair" model, the ball hits a patch of ice. It slides down quickly and stops much faster than you'd expect. The "unfairness" actually helps the system settle down faster!

3. The Two Types of Unfairness

The authors discovered that the type of unfairness matters. They used a dial called $\eta$ (eta) to control how unfair the interactions are.

Case A: Mostly Fair, but a Little Unfair ($\eta$ is positive)

If the interactions are mostly fair but have a tiny bit of unfairness:

• The dancers still start with a brief pause (a "plateau"), just like in the old model.
• But then, instead of getting stuck, they quickly slide into a fast, exponential decay. They stop moving much faster than before.
• Key Takeaway: Even a tiny bit of non-reciprocity destroys the "aging" effect. The system forgets its history and settles down quickly.

Case B: Totally Unfair and Backwards ($\eta$ is negative)

If the interactions are strongly "anti-symmetric" (Dancer A pulls B, but B pushes A away):

• The Oscillation: The dancers don't just stop; they start swinging back and forth like a pendulum.
• The Transition: At first, they might seem to pause, but then they enter a rhythmic phase where they oscillate.
• The Damping: However, this swinging isn't forever. The amplitude (how wide they swing) shrinks rapidly, like a swing in the wind that eventually stops.
• The Period: The speed of this swinging depends on how "unfair" the interaction is. The more extreme the unfairness, the faster they swing.

4. Why Does This Matter?

You might ask, "Why do we care about a theoretical dance floor?"

• Real-World Systems: Many real-world systems are non-reciprocal.
  • Neural Networks: In your brain, one neuron might fire and trigger another, but the second one doesn't necessarily trigger the first back with the same strength.
  • Ecosystems: A predator eats a prey, but the prey doesn't "eat" the predator back.
  • Social Media: One person posts a viral video, influencing millions, but those millions don't influence the original poster in the same way.

This paper provides a "benchmark" or a rulebook for understanding these systems. It tells us that when interactions are asymmetric (which is very common in nature), we shouldn't expect the slow, "aging" behavior seen in simpler, symmetric models. Instead, we should expect systems to either settle down quickly or start oscillating in a rhythmic, decaying pattern.

Summary in a Nutshell

• Old Model (Fair Handshake): Slow, sticky, "aging" behavior. The system gets stuck in the past.
• New Model (Unfair Handshake):
  • Slightly Unfair: The system wakes up and settles down quickly (exponential decay). No more getting stuck.
  • Very Unfair: The system starts swinging back and forth (oscillations) before finally stopping.

The authors solved the math to prove exactly how fast this happens and when the swinging starts, giving us a clear map for understanding complex, one-way interactions in the real world.

Technical Summary

1. Problem Statement

The paper addresses the zero-temperature ($T=0$) nonequilibrium dynamics of the spherical model with non-reciprocal (asymmetric) interactions.

• Context: The spherical model is a simplified framework for studying glassy dynamics and complex systems (e.g., neural networks, ecosystems). In the standard case with symmetric couplings ($\eta=1$), the system exhibits aging: two-time correlation and response functions depend on both the waiting time ($t_w$) and the elapsed time ($\tau$), displaying slow, power-law relaxation and a breakdown of time-translation invariance (TTI).
• Gap: Previous studies (e.g., Crisanti and Sompolinsky, 1988) analyzed finite-temperature dynamics with asymmetric couplings, finding that asymmetry suppresses the spin-glass phase and restores TTI. However, the zero-temperature dynamics remained poorly understood. Specifically, it was unclear if aging persists at $T=0$ when interactions are non-reciprocal, or if the system exhibits new dynamical regimes (e.g., oscillations) driven by the antisymmetric part of the interaction matrix.
• Goal: To analytically solve the $T=0$ dynamics for the spherical model where interaction matrix elements $J_{ij}$ are drawn from the real elliptic ensemble, characterized by a correlation parameter $\eta \in [-1, 1]$.

2. Methodology

The authors employ a combination of random matrix theory (RMT) and dynamical mean-field theory to solve the model analytically in the thermodynamic limit ($N \to \infty$).

• Model Definition:

  • The system evolves via Langevin equations at $T=0$ with a global spherical constraint $\sum S_i^2 = N$.
  • The interaction matrix $J$ follows the real elliptic ensemble:
    $$\langle J_{ij}^2 \rangle = \frac{1}{N}, \quad \langle J_{ij}J_{ji} \rangle = \frac{\eta}{N}$$
    where $\eta=1$ is symmetric, $\eta=-1$ is purely antisymmetric, and $\eta=0$ implies independent entries.

• Analytical Approach:

  1. Biorthogonal Decomposition: Since $J$ is non-Hermitian for $\eta \neq 1$, the authors decompose the dynamics using the left ($\langle L_\alpha|$) and right ($|R_\alpha\rangle$) eigenvectors of $J$, which form a biorthogonal set.
  2. Finite-Size Solution: They derive exact expressions for the two-time correlation $C_N(t, t')$ and response $G_N(t, t')$ functions for finite $N$ in terms of the eigenvalues $\lambda_\alpha$ and eigenvector overlaps.
  3. Thermodynamic Limit: They assume self-averaging and utilize known results for the spectral density $\rho(x,y)$ and the two-point eigenvector correlator $D(z_1, z_2)$ of the real elliptic ensemble.
  4. Key Functions: The dynamics are governed by a central two-time function $W(t, t')$, which is expressed as an integral over the complex eigenvalue spectrum weighted by the eigenvector correlations. This leads to closed-form expressions involving Modified Bessel functions ($I_n$) for $\eta \ge 0$ and Bessel functions ($J_n$) for $\eta < 0$.

3. Key Contributions

• Exact Analytical Solution: The paper provides the first exact analytical solution for the zero-temperature dynamics of the spherical model with arbitrary non-reciprocal interactions ($\eta \in [-1, 1]$).
• Breakdown of TTI: It demonstrates that for any $\eta < 1$, the system never reaches equilibrium at $T=0$; the two-time observables explicitly depend on both $t_w$ and $\tau$, confirming the persistence of aging-like features (dependence on waiting time) even in the presence of non-reciprocity.
• Transition from Power-Law to Exponential Decay: A major finding is that while symmetric interactions ($\eta=1$) yield slow power-law relaxation, any degree of non-reciprocity ($|\eta|<1$) induces exponential decay in the long-time limit, albeit with algebraic prefactors.
• Oscillatory Regime: The authors identify a critical transition at $\eta=0$. For $\eta < 0$ (dominant antisymmetric interactions), the system undergoes a transition to a regime of damped periodic oscillations after a characteristic time scale $\tau^*$.

4. Detailed Results

Regime I: $0 \le \eta \le 1$ (Symmetric to Weakly Asymmetric)

• Correlation Function $C(t_w+\tau, t_w)$:
  • Exhibits an initial plateau ($C \approx 1$) whose duration increases with $t_w$ and $\eta$.
  • Long-time behavior ($\tau \gg t_w$): The decay is exponential rather than power-law:
    $$C \sim \left(\frac{t_w}{\tau}\right)^{5/4} \exp[-r(\eta)\tau]$$
    where the decay rate $r(\eta)$ depends on $\eta$. As $\eta \to 1$, the decay rate vanishes, recovering the power-law behavior of the symmetric case.
• Response Function $G(t_w+\tau, t_w)$:
  • Shows a weak dependence on $t_w$ compared to the correlation function.
  • Decays exponentially for $\eta < 1$, contrasting with the power-law decay for $\eta=1$.

Regime II: $-1 \le \eta < 0$ (Dominantly Antisymmetric)

• Characteristic Time Scale $\tau^*$:
  • A transition time $\tau^* = \frac{1-|\eta|}{|\eta|} t_w$ separates two dynamical regimes.
  • For $\tau < \tau^*$: Non-oscillatory decay (similar to the $\eta > 0$ case).
  • For $\tau > \tau^*$: The system enters an oscillatory regime.
• Oscillatory Behavior:
  • The correlation and response functions exhibit periodic oscillations with period $T = \pi/\sqrt{|\eta|}$.
  • The amplitude decays as $\tau^{-5/4} e^{-(1-|\eta|)\tau}$.
  • Purely Antisymmetric Case ($\eta = -1$): The dynamics become time-translation invariant with pure power-law decay of the amplitude ($\tau^{-3/2}$) and no exponential damping.
• Phase Dependence: The phase of the oscillations in the correlation function depends on the waiting time $t_w$, whereas the response function's phase is constant.

Finite-Size Effects

• Numerical simulations for finite $N$ confirm that the analytical predictions for $N \to \infty$ are approached systematically.
• For $\eta < 0$, finite-size effects are notably weaker than in the symmetric case, showing excellent agreement with theory even for moderate $N$.

5. Significance and Implications

• Resolution of a Long-Standing Gap: The work clarifies the zero-temperature behavior of non-reciprocal systems, showing that while TTI is broken (aging persists), the nature of the relaxation changes fundamentally from slow glassy dynamics to faster exponential relaxation.
• Mechanism of Acceleration: The authors link the faster relaxation to the reduction in the number of stationary states (fixed points). In the symmetric case, the number of stationary states scales with $N$ (related to the complex energy landscape). In the non-reciprocal case, the number of physical (real) stationary states scales only as $\sqrt{N}$, leading to a sparser landscape and faster escape from metastable states.
• Oscillations in Complex Systems: The discovery of a transition to damped oscillations for $\eta < 0$ provides a theoretical benchmark for complex systems with asymmetric interactions (e.g., predator-prey models, neural networks) where non-reciprocity is intrinsic.
• Future Directions: The paper suggests that in the "weakly non-Hermitian" regime ($\eta \approx 1 - O(1/N)$), the system may retain slow dynamics, bridging the gap between the symmetric and asymmetric limits.

In summary, this paper establishes that non-reciprocity fundamentally alters the relaxation landscape of the spherical model, replacing the slow, scale-invariant aging of glassy systems with faster exponential decay and, under strong antisymmetry, inducing damped oscillatory dynamics.