---
title: 'MEGNO: Mean Exponential Growth for Orbits'
url: https://www.emergentmind.com/topics/mean-exponential-growth-for-nearby-orbits-megno
type: topic
---

# MEGNO: Mean Exponential Growth for Orbits

Mean Exponential Growth for Nearby Orbits (MEGNO) is a nonlinear time-series indicator introduced to probe the stability and chaoticity of dynamical systems via the variance of exponential separation rates for nearby initial conditions. Though originally formulated outside the framework of Birkhoff-type ergodic averages, subsequent developments have tightly linked MEGNO to the theory of weighted Birkhoff averages (WBA). It now occupies a central place among orbit stability discriminators, numerical Lyapunov spectrum estimators, and multifractal analysis methodologies, especially within the context of smooth and weakly chaotic dynamical regimes.

## 1. Definition and Connection to Weighted Averages

The MEGNO indicator quantifies the mean exponential divergence rate between infinitesimally close orbits over finite time \( T \). Classical MEGNO is defined for a tangent vector \( v_t \) along an orbit \( x_t \) as:
\[
Y_T(x_0,v_0) = \frac{2}{T} \int_0^T t \, R(x_t, v_t) \, dt
\]
where
\[
R(x_t, v_t) = \ln \frac{ \|Df(x_t) v_t \| }{ \|v_t\| }
\]
with \( Df \) the Jacobian of the map or flow. \( R \) is the local "stretching rate" or infinitesimal Lyapunov increment. The time-increasing weight \( t \) makes MEGNO sensitive to persistent exponential divergence.

Recent work rigorously reformulates MEGNO as a specific instance of a weighted Birkhoff average. Define a weight function \( g_{\rm MEG}(\tau) = \tau^2(1-\tau) \), with \( \tau = t/T \) on \( [0,1] \). Then the "weighted MEGNO average" is:
\[
\mathrm{WB}_T^{\rm MEG}(R) = \frac{1}{A_T} \sum_{t=0}^{T-1} g_{\rm MEG}(t/T) R(x_t, v_t)
\]
where \( A_T = \sum_{t=0}^{T-1} g_{\rm MEG}(t/T) \). This recasts MEGNO as a non-uniform average in the WBA framework, enabling direct comparison with alternative weighted averaging protocols such as \( C^\infty \) bump weights that yield superpolynomial convergence on regular dynamics [2409.08496].

## 2. Theoretical Properties and Convergence Regimes

The rate at which MEGNO and related weighted averages converge to their asymptotic values is strongly determined by the underlying orbit regularity and the smoothness of the weighting function:

- **Nonchaotic orbits (quasiperiodic/tori with Diophantine rotation vectors and smooth observables):** Super-polynomial (faster than \( O(T^{-m}) \) for all \( m \)) convergence is provable when using weights that are \( C^\infty \) and flat at endpoints. In contrast, MEGNO weights \( g_{\rm MEG} \) are only \( C^1 \) at \( \tau=0 \) and \( C^0 \) at \( \tau=1 \), so MEGNO averages convergence like \( O(T^{-2}) \), i.e., much faster than unweighted \( O(1/T) \) but inferior to \( C^\infty \) bump-weighted averages [2409.08496].

- **Chaotic orbits:** For both MEGNO and any WBA, convergence slows to \( O(1/T) \), or \( O(\ln T/T) \) in the presence of shear, matching the behavior of the standard finite-time Lyapunov estimator. No deterministic weighting overcomes this barrier for truly irregular orbits.

A summary of observed convergence rates is given below:

| Orbit Type   | Weight Type        | Convergence Rate                        |
|--------------|-------------------|-----------------------------------------|
| Regular (torus) | \( C^\infty \) bump | Super-polynomial, \( O(T^{-m}) \)     |
| Regular (torus) | MEGNO (\( C^1 \))   | \( O(T^{-2}) \)                       |
| Chaotic         | Any               | \( O(1/T) \), or \( O(\ln T/T) \) w/ shear |

## 3. Algorithmic Implementation

The MEGNO (and more general WBA-based) algorithm for Lyapunov exponent estimation entails:

```python
# Pseudocode for weighted Lyapunov exponent estimation
Input: initial point x0, tangent vector v0 (‖v0‖=1), steps T, weight function g(·) on [0,1]
Precompute: G[t] = g(t/T), S = sum(G)
Initialize: x = x0, v = v0, mu_acc = 0
for t in 0 ... T-1:
    y = Df(x)·v
    r = norm(y)
    v = y / r
    mu_acc += G[t]/S * log(r)
    x = f(x)
Output: mu ≈ mu_acc
```
For MEGNO, \( g = g_{\rm MEG}(\tau) \). For full Lyapunov spectrum one exponentiates the Gram–Schmidt or QR orthogonalization procedure, accumulating weighted logarithms of singular values at each step [2409.08496].

MEGNO formulas are also directly extensible to continuous flows, where time integrals are replaced with appropriately normalized numerical quadrature sums.

## 4. Empirical Performance and Comparative Behavior

Systematic empirical comparisons have established the following findings:

- For nonchaotic orbits, the standard (unweighted) Birkhoff method requires \( T \sim 10^6 \) to reach 5–6 digits, the MEGNO-weighted scheme attains \( O(T^{-2}) \) so \( 10^5 \) steps yield 4–5 digits; \( C^\infty \) bump weighting achieves machine precision (\( <10^{-14} \)) with only \( 3\times10^5 \) steps.
- In maps with shear, both weighted and unweighted estimators display decay only like \( \ln T / T \) due to transient effects in the tangent dynamics.
- For attractors with singularities in \( Df \), such as the Tinkerbell map, the accuracy for non-dominant exponents is lost for all methods.
- For uniformly hyperbolic (constant-Jacobian) maps, geometric convergence is observed, limited ultimately by numerical roundoff.

In all cases, the weighting function's differentiability determines the prefactor and power of the convergence, with MEGNO providing a universal improvement over unweighted but not attaining the exponential or superpolynomial rates achievable by smooth, vanishing-end bump weights [2409.08496].

## 5. Role in Regularity-Chaos Discrimination

MEGNO, by accelerating convergence toward the largest Lyapunov exponent and amplifying divergence in the presence of chaos, offers an efficient test for distinguishing regular from chaotic orbits. This capability is shared with other WBA-based criteria, such as the dual-segment digit-accuracy test: comparing the difference between two WBA values on successive segments of an orbit, high "digit accuracy" is indicative of regularity (e.g. reaching double-precision accuracy for super-convergent tori), while persistence of low accuracy signals chaos [2212.10783, 2001.00086]. The functional form of MEGNO leads to especially rapid separation between these two dynamical regimes, although at the cost of reduced convergence order compared to the bump-weighted average.

## 6. Limitations and Special Considerations

Several limitations and subtleties arise in MEGNO analysis:

- Only for smooth orbits (e.g., KAM tori) and smooth observables does MEGNO accelerate convergence. In chaotic, singular, or fractal-attractor regimes, the asymptotic advantage is lost.
- The choice of weight in MEGNO is suboptimal in smooth regimes compared to analytically constructed \( C^\infty \) weights, which can achieve exponential or arbitrarily high polynomial convergence depending on analyticity and Diophantine properties [2205.09496, 2505.03210].
- The MEGNO weight is not symmetric in time and is not bounded away from zero at \( t=0 \), making it \( C^1 \)-shrinking rather than infinitely smooth at endpoints, hence strictly less effective for super-convergent averaging.
- For singular Jacobians or in systems where Lyapunov exponents are not well-defined, MEGNO and other WBA-based indicators may produce erroneous or meaningless averages due to numerical divergence in the stretching rate [2409.08496].

## 7. Broader Context: Weighted Birkhoff Average Theory

The rigorous study and extension of MEGNO has illuminated the general principle: appropriate endpoint-suppressing, high-differentiability weighting of ergodic averages offers universal acceleration of convergence for smooth, regular orbits. The effectiveness hierarchy is:

- **Classical Birkhoff (uniform weight):** \( O(1/T) \)
- **MEGNO-style (polynomial weight, \( C^1 \) at best):** \( O(1/T^2) \)
- **General WBA with \( C^\infty \) bump:** \( O(T^{-M}) \) for all \( M \)
- **Analytic weight and observable:** Exponential convergence

These behaviors are explained and generalized in the deterministic and probabilistic analysis of WBAs for toral, quasiperiodic, almost periodic, and periodic dynamics [2505.03210, 2409.08496, 2205.09496]. The deep associations between MEGNO, Lyapunov spectrum computation, multifractal analysis, and entropy spectra are made explicit in the context of thermodynamic formalism and multifractal spectrum of weighted Birkhoff averages for symbolic systems [2109.03898, 2006.06774].

The MEGNO indicator thus represents a pivotal bridge between practical orbit characterization and rigorous ergodic-theoretic acceleration methodologies, with theoretical and computational roles in both chaos detection and fine Lyapunov exponent estimation.

Source: https://www.emergentmind.com/topics/mean-exponential-growth-for-nearby-orbits-megno