Convective Lyapunov Exponent

• Convective Lyapunov exponent is a velocity-dependent measure that quantifies the asymptotic growth or decay of infinitesimal perturbations in a moving frame, distinguishing it from fixed-frame Lyapunov exponents.
• It generalizes to a two-parameter spectrum Λ(v,n) which captures the expansion or contraction of higher-dimensional tangent-space volumes in spatially extended chaotic systems.
• Both direct moving-window and chronotopic approaches reveal numerical challenges and a critical integrated density (n_c) beyond which the spectrum exhibits multiple branches and discontinuous behavior.

Searching arXiv for the cited paper and closely related chronotopic work to ground the article. arXivSearch[{"query":"id:(Jiotsa et al., 2012)","max_results":5},{"query":"chronotopic Lyapunov spectrum Lepri Politi Torcini arXiv","max_results":10}]} The convective Lyapunov exponent is a velocity-dependent growth rate that quantifies the asymptotic amplification or decay of infinitesimal perturbations in a moving frame, rather than in a fixed frame. In spatially extended chaotic systems, this distinction is essential because perturbations not only grow or decay but also propagate in space. The standard object, usually denoted $\Lambda(v)$, measures growth along the world line $i=vt$; the central result of "Convective Lyapunov Spectra" is that this notion can be generalized to an entire two-parameter family $\Lambda(v,n)$, where $v$ is the frame velocity and $n$ is an integrated density labelling the position in the spectrum (Jiotsa et al., 2012). In this formulation, convective instability is no longer a one-exponent phenomenon but part of a full moving-frame analogue of the Lyapunov spectrum.

1. Definition and motivation

In an $N$-dimensional dynamical system, linear stability is described by the usual Lyapunov exponents $\{\lambda_l\}$, ordered from largest to smallest. In a large one-dimensional extended system of length $L$, the spectrum becomes a smooth function of the integrated density

$n=\frac{l}{L},$

so one writes $\lambda(n)$ in the thermodynamic limit. This fixed-frame description, however, does not separate local amplification from transport of disturbances through the medium.

The convective Lyapunov exponent addresses this limitation by measuring the growth of a perturbation in a frame moving at velocity $i=vt$0. Starting from a localized infinitesimal perturbation $i=vt$1, initially supported in a finite region, the standard definition is

$i=vt$2

The notation $i=vt$3 emphasizes that this is the maximal convective exponent; earlier literature usually denotes it simply by $i=vt$4 (Jiotsa et al., 2012).

This quantity measures how unstable the system looks to an observer moving at speed $i=vt$5. If $i=vt$6, perturbations amplify exponentially in that moving frame; if $i=vt$7, the frame moves at the threshold of growth; if $i=vt$8, perturbations decay there. Ordinary Lyapunov exponents are measured in a fixed frame and are insensitive to the transport of disturbances, whereas $i=vt$9 explicitly separates local growth from propagation.

2. Standard convective Lyapunov exponent and instability regimes

The standard convective exponent has a direct physical interpretation in terms of propagation of infinitesimal disturbances. In systems with left-right symmetry, $\Lambda(v,n)$0 is symmetric in $\Lambda(v,n)$1, maximal at $\Lambda(v,n)$2, and there equals the largest ordinary Lyapunov exponent. The largest velocity $\Lambda(v,n)$3 such that

$\Lambda(v,n)$4

gives the maximal propagation velocity of infinitesimal perturbations (Jiotsa et al., 2012).

This framework distinguishes two regimes. In an absolute instability, $\Lambda(v,n)$5, so perturbations grow even in the stationary frame. In a convective instability, $\Lambda(v,n)$6, but $\Lambda(v,n)$7 is positive for some nonzero $\Lambda(v,n)$8; perturbations are amplified only while being advected. The distinction is not visible in the ordinary Lyapunov spectrum alone, because fixed-frame exponents do not encode the velocity dependence of perturbation growth.

The paper argues that $\Lambda(v,n)$9 is nevertheless incomplete as a descriptor of stability in extended systems. It captures maximal amplitude growth in a moving frame, but not the behavior of higher-dimensional tangent-space volumes. This motivates the generalization from a single velocity-dependent exponent to a full convective Lyapunov spectrum.

3. Generalization to the convective Lyapunov spectrum

The generalized object is the convective Lyapunov spectrum

$v$0

where $v$1 is the velocity of the comoving frame and $v$2 is the integrated density associated with the $v$3-th Lyapunov exponent in a moving window of size $v$4. At $v$5, the generalized spectrum reduces to the standard convective Lyapunov exponent: $v$6

The role of $v$7 is the same as in the standard Lyapunov spectrum $v$8: $v$9 corresponds to the top exponent, while larger $n$0 probes deeper directions in tangent space. The interpretation of $n$1 is therefore a moving-frame analogue of the full Lyapunov spectrum. While $n$2 measures growth of a single perturbation amplitude along a moving frame, $n$3 measures the expansion or contraction of higher-dimensional tangent-space volumes in that frame (Jiotsa et al., 2012).

The model used for illustration is a standard coupled-map lattice,

$n$4

where $n$5, $n$6 is discrete time, and $n$7 is a diffusive coupling. The linearized dynamics of infinitesimal perturbations is built from the local multipliers

$n$8

For both $n$9 and $N$0, sign has the same dynamical meaning. Positive values indicate exponential growth in the moving frame, zero indicates a marginal propagation or growth threshold, and negative values indicate decay. For $N$1, positivity means that an entire $N$2-fraction of tangent-space directions experiences net volume expansion in that frame.

4. Direct moving-window construction and the chronotopic approach

A direct definition of $N$3 can be constructed in a moving window of fixed size $N$4. Since perturbations would otherwise collapse onto the most unstable direction, one iterates $N$5 linearly independent perturbations inside the window and repeatedly orthogonalizes them, in analogy with ordinary Lyapunov computations. The paper introduces two elementary moves in tangent space: $N$6, a static window, and $N$7, a right shift by one site. The tangent evolution is

$N$8

with $N$9, where $\{\lambda_l\}$0 for a $\{\lambda_l\}$1-move and $\{\lambda_l\}$2 for a $\{\lambda_l\}$3-move. To realize a given average velocity $\{\lambda_l\}$4, one chooses a sequence of $\{\lambda_l\}$5 and $\{\lambda_l\}$6 moves whose fraction of $\{\lambda_l\}$7s equals $\{\lambda_l\}$8; for rational $\{\lambda_l\}$9, one can choose a periodic binary sequence with density $L$0 (Jiotsa et al., 2012).

This direct method gives an intuitive moving-frame definition, but the paper identifies significant drawbacks. The perturbations are artificially confined by boundary conditions at the window edges; numerical vectors become strongly localized near one edge of the window; and double precision is often insufficient because many components become exponentially small. The method is therefore numerically delicate and can easily produce artifacts.

The more powerful and reliable method is the chronotopic approach. Instead of a localized perturbation, one considers an exponentially tilted perturbation

$L$1

Then $L$2 evolves according to

$L$3

For each $L$4, this defines a standard Lyapunov spectrum in the tilted dynamics,

$L$5

called the temporal Lyapunov spectrum. Here $L$6 is a spatial growth or decay rate: positive $L$7 corresponds to profiles that decay in one direction, while negative $L$8 corresponds to profiles that grow in that direction.

5. Legendre-transform relation and the temporal Lyapunov spectrum

The central theoretical relation is that the convective spectrum is obtained from the temporal spectrum by a Legendre-transform-like construction: $L$9 and conversely

$n=\frac{l}{L},$0

This generalizes to all $n=\frac{l}{L},$1 the previously established formula for $n=\frac{l}{L},$2 (Jiotsa et al., 2012).

For fixed $n=\frac{l}{L},$3, $n=\frac{l}{L},$4 is viewed as a function of the spatial tilting parameter $n=\frac{l}{L},$5. Its derivative selects the velocity through

$n=\frac{l}{L},$6

and the convective spectrum is then obtained by subtracting the kinematic term $n=\frac{l}{L},$7. The paper explicitly compares this structure with a Legendre transform in thermodynamics or large-deviation theory.

Operationally, the convective spectrum is encoded in the $n=\frac{l}{L},$8-dependence of the temporal spectrum. For each fixed $n=\frac{l}{L},$9, one computes $\lambda(n)$0 as a function of $\lambda(n)$1, determines $\lambda(n)$2 from $\lambda(n)$3, and evaluates $\lambda(n)$4. The chronotopic approach is presented as the recommended procedure for computing the entire convective spectrum.

The paper also marks an important limitation. For $\lambda(n)$5, the Legendre relation is physically justified by the asymptotically exponential shape of a spreading localized perturbation. For $\lambda(n)$6, the interpretation is more formal, because there is no equally direct profile-based argument for subleading tangent directions. The consistency between the chronotopic construction and the direct moving-window method is used to support the validity of the generalized definition.

6. Critical integrated density, multiple branches, and numerical evidence

The principal new phenomenon concerns the shape of the temporal Lyapunov spectrum as one moves down in the spectrum. For the largest exponent, $\lambda(n)$7 is monotone in $\lambda(n)$8, producing a single well-behaved branch for $\lambda(n)$9. For larger $i=vt$00, however, $i=vt$01 can become non-monotone and change concavity as a function of $i=vt$02. The paper identifies a critical integrated density $i=vt$03 such that for $i=vt$04, the temporal spectrum has the usual concavity structure and $i=vt$05 behaves smoothly, whereas for $i=vt$06, a change of concavity occurs and the convective spectrum develops multiple branches (Jiotsa et al., 2012).

This phenomenon arises because the Legendre construction assumes a one-to-one relation between $i=vt$07 and $i=vt$08 via $i=vt$09. If $i=vt$10 changes concavity, the same slope $i=vt$11 can be realized at several different $i=vt$12-values, yielding several candidate values of $i=vt$13. The paper further reports that for $i=vt$14, the branch that is physically selected switches abruptly when $i=vt$15 crosses zero, so $i=vt$16 becomes discontinuous at $i=vt$17. The abstract describes $i=vt$18 as a dynamical invariant, because it is determined by a structural property of the Lyapunov spectrum itself.

An explicit analytic illustration is provided by coupled Bernoulli maps $i=vt$19, for which the temporal spectrum is

$i=vt$20

By expanding for small $i=vt$21, the critical density is obtained as

$i=vt$22

provided that $i=vt$23. For the parameters $i=vt$24 and $i=vt$25, the spectra for $i=vt$26 and very small $i=vt$27 are almost identical for $i=vt$28, but differ by a finite amount for $i=vt$29, directly demonstrating the discontinuity in velocity dependence.

The paper also analyzes branch structure. For a fixed velocity and $i=vt$30, the Legendre transform gives three branches: one lower branch associated with positive $i=vt$31, and two upper branches associated with negative $i=vt$32. The authors argue that the upper branches are likely not physically relevant in the moving-window interpretation, because negative $i=vt$33 corresponds to profiles larger on the right side of the window; such a profile would be impossible unless one assumes a source of perturbation or noise entering from that side. By contrast, the lower branch is consistent with the expected decay away from the perturbation source and is the one matched by direct simulations. This indicates that the multiple branches are mathematically generated by the Legendre construction, while physical considerations are needed for branch selection.

Numerical comparison is carried out for a chain of coupled logistic maps with $i=vt$34 and $i=vt$35. For $i=vt$36, the standard convective exponent is maximal at $i=vt$37 and crosses zero at

$i=vt$38

which is the maximal propagation speed of infinitesimal perturbations. For $i=vt$39, the convective spectrum at $i=vt$40 is close to zero, indicating that roughly half the ordinary Lyapunov exponents are positive. The direct moving-window method initially agrees poorly with the chronotopic one because Lyapunov vectors in the moving window become exponentially localized near the left border, so many components fall below double-precision resolution. To improve stability, the paper introduces a weighted scalar product

$i=vt$41

with adjustable $i=vt$42, and finds $i=vt$43 often near-optimal. This significantly improves agreement, although large velocities and large windows remain challenging.

The significance of the convective Lyapunov spectrum lies in the extension from amplitude growth to volume growth in moving frames, the chronotopic computation through the temporal spectrum, and the discovery of a critical integrated density beyond which new qualitative behavior appears. The paper also leaves several issues open: the precise physical meaning of $i=vt$44, the conditions under which the upper branches might become relevant, whether a cleaner direct definition can be formulated without artificial window boundary conditions, and whether covariant Lyapunov vectors can help define convective spectra more naturally in open systems. These open questions indicate that the theory establishes a general framework while leaving branch selection and physical interpretation only partially resolved (Jiotsa et al., 2012).