WorksPhysics and Astronomy
Odd viscosity suppresses intermittency in direct turbulent cascades
Abstract
Intermittency refers to the broken self-similarity of turbulent flows caused by anomalous spatio-temporal fluctuations. In this Letter, we ask how intermittency is affected by a non-dissipative viscosity, known as odd viscosity (also Hall or gyro-viscosity), which appears in parity-breaking fluids such as magnetized polyatomic gases, electron fluids under magnetic field and spinning colloids or grains. Using a combination of Navier-Stokes simulations and theory, we show that intermittency is suppressed by odd viscosity at small scales. This effect is caused by parity-breaking waves, induced by odd viscosity, that break the multiple scale invariances of the Navier-Stokes equations. Building on this insight, we construct a two-channel helical shell model that reproduces the basic phenomenology of turbulent odd-viscous fluids including the suppression of anomalous scaling. Our findings illustrate how a fully developed direct cascade that is entirely self-similar can emerge below a tunable length scale, paving the way for designing turbulent flows with adjustable levels of intermittency.
In 3D turbulent fluids, energy is transferred from large to small length scales through a process known as a turbulent cascade: large eddies tend to split into smaller and smaller eddies [1, 2, 3, 4, 5, 6]. The hierarchical structure of the cascade suggests that it may exhibit a self-similarity relating larger and smaller eddies [7, 8]. It turns out that this self-similarity is only approximate [5]: smaller eddies exhibit an increasing propensity for extreme velocities, characterized by non-gaussian, fat-tailed distributions [9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29] (see Figure 1a). This phenomenon is known as intermittency [5, 11].
Figure 1. Intermittency in odd turbulence. (a) Probability distribution of -direction vorticity in DNS (renormalized by the standard deviation), for . Gaussian distribution is shown in dashed line for comparison. Inset: illustration of a chiral active fluid. Odd viscosity can be caused by self-spinning of particles in direction. (b) The growth of kurtosis in DNS is suppressed by odd viscosity at small scales, for same as in (a). Errorbar is shown as shaded area. (inset) Rescaled kurtosis ( being the kurtosis for ) vs rescaled lengthscale , with . (c) Kurtosis predicted from the modified Parisi-Frisch theory, for . (inset) Rescaled kurtosis vs rescaled lengthscale . Note that the scale separation in (c) is considerably larger than what is achievable in DNS in (b). See Sec. II D for results with same scale separation as that in DNS (SI [60]).
The precise conditions that lead a turbulent cascade to produce intermittency are still unknown [1, 30]. In the direct cascade of the 3D Navier-Stokes equation, intermittency (i.e. non-gaussianity) increases along the forward cascade, a property known as “anomalous scaling”. Reductions of this tendency have been reported in magneto-hydrodynamics [31], nonlinear viscosity fluids [32], fluids on fractal Fourier sets [33–36] or subject to rotation [37–42]. It has been conjectured that self-similarity may be restored in a direct cascade by a mathematical “surgery” of Navier-Stokes equations that reduces the backward energy transfer through certain triadic interactions classified by the helicity of the modes involved [43]. However, we are not aware of any example of a fully developed direct cascade with no intermittency described by the 3D Navier-Stokes equations.
In this Letter, we show that the increase of intermittency, that typically occurs along a forward turbulent cascade, can be suppressed in the inertial range by parity-breaking waves occurring in fluids that display a non-dissipative viscosity, known as odd viscosity, such as magnetized polyatomic gases, graphene under magnetic field and spinning colloids or grains [44–59]. By controlling odd viscosity, we can tune the length scale below which the growth of intermittency terminates, ultimately creating a fully developed forward turbulent cascade that is completely self-similar, i.e. anomalous scaling is suppressed.
Turbulence with odd viscosity
While the viscosity tensor is typically assumed to be symmetric (), an antisymmetric part of the viscosity tensor (), known as odd, Hall or gyro viscosity, naturally arises in fluids that break both parity and time-reversal symmetry [56]. Odd viscosity with rotational symmetry in (e.g., generated by self-spinning particles, see Figure 1(a)) modifies the incompressible Navier-Stokes equation to (see Supplementary Information (SI) [60])
where is the velocity, the convective derivative, the helicity-free external forcing, the dissipative viscosity, the unit vector along , and the odd viscosity. The odd viscosity term, that can be seen as a scale-dependent Coriolis force, does not directly dissipate or transfer energy, but affects energy transfer by inducing “odd waves” with frequency in the fluid (SI) [60]. Odd waves are parity-breaking, i.e. the frequency changes sign when the definite helicity is flipped [57]. As shown in Ref. [61], odd waves progressively arrest the forward energy flux starting from the characteristic scale ( is the energy injection rate) where is comparable with the eddy turnover rate. While a qualitative picture of the energy spectrum of odd turbulence has emerged [61], the effect of odd viscosity on intermittency is yet to be elucidated.
Intermittency in odd turbulence—We first analyze the vorticity distribution obtained from direct numerical simulations (DNS), which captures the intermittency around the dissipation scale. Details of the DNS method are provided in SI [60]. Remarkably, odd viscosity leads to more Gaussian-like distributions (compare the different curves in Figure 1a). Intermittency can be captured by the kurtosis of the structure functions in which are known as the longitudinal velocity increment. Due to the anisotropy induced by odd viscosity, the structure function is evaluated for perpendicular to . At the injection scale, the flow has a Gaussian kurtosis imposed by the forcing. In ordinary turbulence (), the kurtosis increases as energy cascades towards smaller scales (Figure 1b). For finite , the growth of the kurtosis is suppressed at small scales where odd viscosity is dominant. The rescaled data (inset) shows that the transition between the intermittent and non-intermittent regimes is located around , where odd viscosity becomes relevant [61].
Modified Parisi-Frisch formalism—To rationalize this suppression of intermittency, we revisit the Parisi-Frisch framework [62] relating intermittency to the existence of multiple scale invariances in Navier-Stokes equations.
In the absence of odd viscosity, the unforced dissipation-free Navier-Stokes equation is invariant under the scale transformation , , for all . Kolmogorov’s K41 theory [7, 8] effectively assumes that the exponent is selected, leading to a self-similar solution. In reality, turbulent flows exhibit a statistical distribution of scaling exponents , as allowed by the scaling symmetry. This intertwining of different exponents , referred to as multifractality, breaks the self-similarity of the flow statistics at different scales, heralding the emergence of intermittency [5] [63].
Intermittency is suppressed by odd viscosity because it violates the multiple scaling symmetry of Navier-Stokes equations (SI [60]). We conjecture that intermittent solutions with arbitrary can still appear at large scales where has a small effect, and that they reduce
to the same wave-affected solution at small scales, governed by a single exponent (SI [60]).
We model this crossover by adapting the approach of Ref. [65] and obtain
in which is the characteristic lengthscale, which is determined by equating the eddy turnover time and the odd wave period [61]. Eq. (2) qualitatively reproduces the arrest of intermittency observed in DNS, see Figure 1c ( is adopted from Ref. [63]). For small odd viscosity, is non-monotonic, in agreement with DNS with extended inertial range (SI [60]). This is due to the variation of , such that the crossover between two regimes takes place at different lengthscale for different . Similar non-monotonicity was observed in Ref. [65].
Nonlinear energy transfer—In the Navier-Stokes equa-
tion, nonlinear energy transfer occurs through triads of modes with wavenumbers satisfying [30] [66, 67, 68, 69]. As helicity is conserved, it is convenient to consider modes with definite helicity . When odd waves are present, these modes get out of phase. As a consequence, the contribution to the non-linear energy transfer of a given triad is multiplied by in which [61]. When , the fast oscillation effectively cancels the time-averaged energy transfer during . Therefore, in odd turbulence energy can only transfer through quasi-resonant triads (for which ) [70]. We estimate the average energy transfer from the fraction of quasi-resonant triads and the energy transfer rate in ordinary turbulence, leading to .
For quasi local triads () [71] we have and , hence for , corresponding to a regular turbulence regime for . For (), odd waves arrest the energy transfer with and . Interestingly, odd waves have a strong impact on the forward transfer of heterochiral triads but only a weak impact on the inverse transfer of homochiral triads [43, 72], see Figure 2(a), because for homochiral triads with exact locality we always have and [73]. This asymmetry enhances the flux loop formed by the two channels (Figure 3b) and triggers the aforementioned breakdown of the multiple scale invariances of Navier-Stokes equations, which ultimately suppresses intermittency (SI [60]). Importantly, such a mechanism requires the energy to be symmetrically distributed on modes with both definite helicities (+ and −), which ensures that the energy flux is not dominated by homochiral channels only.
Figure 2. Effect of odd waves on triadic interactions and helical shell model. (a) Asymmetric effects of odd viscosity on homochiral and heterochiral triads. Odd viscosity strongly affects heterochiral triads by reducing the fraction of resonant triads, . Odd viscosity has no effect on local homochiral triads for which . (b) Two-channel helical shell model. Each shell contains two velocity variables representing modes with and chiralities, respectively. Energy transfers forward in heterochiral channel and backward in homochiral channel, through triads formed by nearest neighbors.
Figure 3. Odd turbulence in DNS and the helical shell model. (a). Energy spectrum in DNS, for , , , , . The spectrum in the inertial range changes from a K41 regime () to a wave-affected regime (). (inset) Rescaled spectrum ( being the spectrum for ) vs rescaled wave number , with . (b) Energy flux and flux loop formed by homochiral and heterochiral components in DNS. . (c) Kurtosis in DNS, same as Figure 1(d) with x-axis changed from to . (inset) Rescaled kurtosis ( being the kurtosis for ) vs rescaled lengthscale . (d) Energy spectrum in the shell model, for , , , , , . (inset) Rescaled spectrum vs rescaled wave number . (e) Energy fluxes in the shell model, for . (f) Kurtosis truncated at the dissipation scale, with [29] [64], for values same as that in (d). (inset) Rescaled kurtosis vs rescaled wavenumber . In the shell model simulations, we have used shells, , , an injection rate , , and . A constant injection forcing is used. Note that the scale separation in the shell model (bottom row) is considerably larger than what is achievable in DNS (top row). See Sec. II D for results with same scale separation as that in DNS (SI [60]).
Helical shell model
We now construct a two-channel helical shell model that captures most odd-turbulence phenomenology (Figure 2b) [74]. Shell models have proven to be a particularly useful tool to study the inertial-range intermittency in turbulence, as they can probe scales far beyond what is accessible in DNS [19] (Although they behave differently from DNS at the dissipation scale [65]). Shells are equispaced on a logarithmic scale and each shell has two complex dynamical variables, and , representing velocity fluctuations of modes with + and − definite helicities, respectively. We impose conservation of energy and helicity [76], resulting in
where is the forcing, and
are the nonlinear transfer of heterochiral and homochiral channels, illustrated in purple and green in Figure 2. When , the channels reduce to the Sabra shell models for the forward and inverse cascade regimes, respectively [64, 77–80] ( and are determined from conservation laws, SI [60]). To reproduce the effect of odd viscosity, we let
where ensures that the model is dominated by the forward cascade when . This functional form is chosen to introduce a transition around . Here, mimics the fraction of resonant triads: for small and for large . Finally, we choose so that the suppression effect is weaker for the homochiral channel. In the SI, we generalize the model to include “generalized odd viscosity”, for which [60].
As shown in Figure 3e, results from the numerical simulation of this shell model reproduce the energy flux observed in Navier-Stokes simulation for as well as other qualitative features and scaling relations (Figure 3a, Figure 3d). In the SI [60], we show that a flux loop state [30] consistent with DNS results [61] is observed in the inverse cascade regime where forcing is injected at small lengthscale .
Suppression of anomalous scaling
Our shell model results (Figure 3f) mimic the DNS kurtosis (Figure 3c) and its non-monotonic dependence predicted by the modified Parisi-Frisch argument. We also verify that intermittency is suppressed in higher-order structure functions up to , in both DNS and shell model (SI [60]). For generalized odd viscosity with , we find a wave-affected scaling regime of the structure functions for , where the wave-affected exponents are extracted (see Figure 4(a) for an example). Figure 4(b) shows that the extracted in the wave-affected regime are consistent with the self-similar exponents (solid lines), in contrast with the anomalous exponents usually observed in forward cascades (black dashed line). The we measure are close to the theoretical approximation (SI [60]).
Figure 4. Scaling exponents in the helical shell model. (a) Rescaled structure function () in shell model with generalized odd viscosity for , with , and . (b) Exponents in the wave-affected regime for . Exponents for are fitted by the shaded area in (a), see SI for details [60]. Solid lines are monofractal fit . Error bars are estimated from shifting the fitting range by shells. Exponents for are shown for comparison, which is consistent with She-Lévêque (dashed line) [63]. (inset) for and for . Shell model parameters are same as that in Figure 3. Inertial range in the simulation is extended (SI [60]).
To sum up, we have demonstrated the existence of a fully developed forward turbulent energy cascade that is entirely self-similar below a tunable length scale, paving the way for designing turbulent flows with adjustable levels of intermittency. In future work it would be interesting to generalize the mechanism studied here to more complex flows where other sources of waves may exist [81].
Acknowledgments
We thank Alexandros Alexakis, Luca Biferale and Tali Khain for discussions. We are grateful for the support of the Netherlands Organisation for Scientific Research (NWO) for the use of supercomputer facilities (Snellius) under Grant No. 2021.035, 2023.026. This publication is part of the project “Shaping turbulence with smart particles” with project number OCENW.GROOT.2019.031 of the research programme Open Competitie ENW XL which is (partly) financed by the Dutch Research Council (NWO). S.C. and M.F. acknowledge a Kadanoff–Rice fellowship funded by the National Science Foundation under award no. DMR-2011854. M.F. acknowledges partial support from the National Science Foundation under grant DMR-2118415 and the Simons Foundation. V.V. acknowledges partial support from the Army Research Office under grant W911NF-22-2-0109 and W911NF-23-1-0212. M.F. and V.V acknowledge partial support from the France Chicago center through a FACCTS grant. This research was partly supported from the National Science Foundation through the Center for Living Systems (grant no. 2317138), the National Institute for Theory and Mathematics in Biology and the Chan Zuckerberg Foundation. This work was completed in part with resources provided by the University of Chicago’s Research Computing Center.
Data availability
The code used for simulating the helical shell model, processing the data, generating the figures, as well as the data generated during this study are available on Zenodo [82] under the 2-clause BSD licence.
Supplementary Information:
Odd viscosity suppresses intermittency in direct turbulent cascades
Odd viscosity
In this section we introduce odd viscosity and emphasize its non-dissipative nature.
In fluids, the stress tensor is related to the strain rate via a viscosity tensor : . The viscosity tensor is usually assumed to be symmetric, i.e., , which applies to passive fluids. However, in active and parity-breaking fluids, e.g., fluids formed by self-spinning particles, asymmetric viscosity tensor () arises because the system is out of equilibrium [1]. We decompose the viscosity tensor into a symmetric part and an antisymmetric part . The incompressible Navier-Stokes equation reads:
where is the density. In the last equality, the second term is the dissipation term that arises from .
In this work we consider odd viscosity generated by the self-spinning of particles in . This corresponds to a cylindrically symmetric viscosity tensor which has eight independent odd viscosities [2]. Only two independent odd viscosity terms and remain in the incompressible Navier-Stokes equation [3]:
One example of the corresponding odd viscosity tensor is
In Eq. (6) Einstein summation is not used. Note that Eq. (6) is not the only possible viscosity tensor that gives Eq. (5). In our DNS we assume and . In this case the force due to odd viscosity reduces to , where the gradient term can be absorbed into the pressure. Equation (5) then reduces to
which is Eq. (1) of the main text.
In Ref. [2], it is proved that the asymmetric viscosity tensor does not generate or dissipate energy. For an intuitive understanding, we consider the following reduced dynamical equation of the odd viscosity term:
We perform Fourier transform , which leads to
The antisymmetric, non-Hermitian dynamical matrix has imaginary eigenvalues , which is distinct from that of positive viscosity (negative eigenvalues) and negative viscosity (positive eigenvalues). Equation (9) has oscillatory solutions , i.e., energy is neither generated nor dissipated by odd viscosity. Note that in Eq. (8) the incompressibility is not included. Imposing the incompressibility changes the eigenvalues to [2, 3].
Above we have shown that odd viscosity is non-dissipative. As a linear viscosity term, it does not directly affect the nonlinear triadic interaction. Hence, the energy balance equation follows that of the ordinary turbulence [4, 5, 6]:
in which
This term describes the non-linear energy transfer between scales, while corresponds to energy injection by the forcing term . The term represents standard viscous dissipation. In Eq. (11), the sum runs on momenta and such that , and is the projector on incompressible flows [5, 6, 7, 8, 9].
Simulation Details
Navier-Stokes
We perform direct numerical simulations (DNS) of the Navier-Stokes equation incorporating odd viscosity (Eq. (1) in the main text) within a cubic domain of size , employing periodic boundary conditions. Our approach utilizes a pseudo-spectral method coupled with Adams-Bashforth time-stepping and a 2/3-dealiasing rule [10]. The dissipative and odd viscous terms are integrated exactly using integrating factors. The forcing acts within a wavenumber band , with random phases delta-correlated in space and time, ensuring a consistent average energy injection rate . This forcing possesses a zero mean component and covariance . The time-step is selected to resolve the fastest odd wave with frequency , where denotes the highest resolved wavenumber in the domain. We observe that stable integration requires a time step .
Since the forcing is helicity-free, and odd viscosity conserves helicity, the net total helicity of the flow remains identically zero.
The parameters used for the simulations in the main text are as in Ref. [3]. In Section II C we show additional simulations with increased resolution and hyperdissipation to further extend the inertial range. All simulation parameters are provided in Table I.
| I | 3 | 0.34 | |||||||||
| II | 1.0 | (*) | 2 | 18 | 0.35 | 0.6 | 200 |
Table I. Parameters that are used for DNS in the main text (I) and in Section II C (II). Listed are the average energy injection rate , normal shear viscosity , odd viscosity , injection wavenumber , odd viscosity wavenumber , grid resolution , Kolmogorov length , total simulation time , time-step , Kolmogorov time and the integral scale Reynolds number without odd viscosity .
(*) hyperdissipation (quadratic Laplacian)
Shell Model
In the two-channel helical shell model, shells are logarithmically equispaced with . The dynamic equation reads (Eq. (3) of the main text)
where is the forcing.
are the nonlinear transfer of heterochiral and homochiral channels.
are the wave-affected transfer coefficients. and are determined from the conservation laws of the energy and the helicity [11], which leads to
The energy balance equation reads [6, 12]
is the total energy flux
where
are the heterochiral and homochiral components of the flux, respectively. In the main text we use a velocity-based structure function . The definition reduces the 3-shell oscillation that is observed in shell models [13]. A flux-based structure function is also used in § 10 [13, 14]. In the main text we mainly focus on the kurtosis . We also explore higher order ‘kurtosis’ , which shows that intermittency is suppressed for structure functions up to in both DNS and the shell model (Figure S1).
Figure S1. in DNS (a) and for the shell model (b).
We conduct numerical simulations of the two-channel helical shell models with using the 4th order exponential-Runge-Kutta scheme. See Table II for parameters.
| I | ||||||||||
| II |
Table II. Parameters that are used for the helical shell model in the main text (I) and in Section II D (II). Listed are the energy injection rate , normal shear viscosity , odd viscosity , homochiral parameters and , wave number unit , odd viscosity wavenumber , total shell number , total simulation time , time-step . Note that parameters (II) of the shell model mimic parameters (I) of DNS.
Dissipation Scale and Extended Inertial Range
As discussed in Ref. [3] and the main text, odd viscosity affects the dissipation scale with . This change of the dissipation scale is due to the change of the energy spectrum caused by odd viscosity: odd viscosity arrest direct cascade and changes the spectrum from to , which enhances the dissipation rate at given simultaneously. Therefore, when odd viscosity is present the dissipation dominates at larger lengthscales . In Figure S2,Figure S2) we plot the rescaled energy spectrum in both DNS and the shell model ( is the spectrum for ). We find that increases with for because of the arrested cascade. For , decreases with due to the dissipation. This effectively leads to a wavelength selection phenomenon, which is also known as pattern formation [3].
Figure S2. Wavelength selection in odd turbulence. (a) Rescaled energy spectrum ( is the spectrum for ) vs wavenumber in DNS for . (b) Rescaled energy spectrum vs wavenumber in the shell model for . (c) Rescaled energy spectrum estimated from the bootstrap method for , and . In (a-c), stars indicate condensation scale .
Here we provide a simple bootstrap method which reproduces the pattern formation in odd turbulence. Similar to the shell model, we divide the system into logarithmically distributed shells . The velocity fluctuations in each shell is with shell energy . Assuming , the wave-affected energy transfer rate from shell to is (see § 6). In the steady state, the energy conservation at shell requires , i.e., the energy injected in the shell equals to the energy ejected from the shell plus dissipation. Because this equation holds for all shells in the steady state, an estimated energy spectrum can be calculated from an iteration of the energy conservation of all shells, assuming . The results are shown in Figure S2. We indeed find pattern formation at , and the dependence of on is consistent with our expectation.
Because decreases for increasing , the inertial range also shrinks, and the statistics of the kurtosis is affected more by the dissipation when odd viscosity is present. To avoid this effect, we conduct simulations with extended inertial range for both DNS and the shell model. This is achieved by reducing the viscosity to for large . This approach ensures that the inertial range remains approximately same for different values, see Figure S3,Figure S3). The suppression of intermittency is again observed in both DNS and the shell model (Figure S3,Figure S3,Figure S3,Figure S3)). Interestingly, in DNS we observe a weak non-monotonicity of the kurtosis, in alignment with the shell model result and the prediction of the modified Parisi-Frisch theory.
Figure S3. Simulation results with extended inertial range. The value of is reduced when odd viscosity is present, such that the dissipation scale is unchanged. (a) Energy flux in DNS for . (b) Probability distribution of -direction vorticity in DNS (renormalized by the standard deviation), for same values as in (a). Gaussian distribution is shown in dashed line for comparison. (c) Kurtosis in DNS, for same as in (a). (d) Energy flux in the shell model for . (e) Probability distribution of in the shell model (renormalized by the standard deviation), for same values as in (d). Gaussian distribution is shown in dashed line for comparison. (f) Kurtosis in the shell model, for same as in (d).
Results for smaller scale separation
In the main text, for both the modified Paris-Frisch theory and the shell model we use much larger scale separation than what is accessible in DNS. Such a large scale separation allows us to separate the results into clearly-defined regimes: for , there is a direct cascade of ordinary turbulence; for , there is a
direct cascade of odd-wave-affected turbulence. Such a distinction between the two regimes requires a sufficiently large separation between the injection scale , the cross-over scale and the dissipation scale , and the scale separation in DNS is limited by computational expense.
In this section we provide results for the modified Parisi-Frisch theory and the shell model, using similar scale separation as that in DNS. The results exhibit similar qualitative features as that observed in DNS, validating the two theoretical approaches.
Specifically, for the modified Parisi-Frisch theory we ensure the same odd viscosity scale as that in DNS. In the modified Parisi-Frisch theory, we define , where . This corresponds to . As shown in Figure S4, the modified Parisi-Frisch theory reproduces the suppression of kurtosis by odd viscosity. Note that for close to 1 we observe an unphysical discontinuity. This is because the original Parisi-Frisch theory is designed only for the asymptotic behavior of the intermittency at small scales, and is ill-defined at large scales.

Figure S4. Comparison between DNS, modified Parisi-Frisch and shell model results, for same scale separation as that in DNS. (a) and (c) Energy spectrum and kurtosis in DNS for , , , , , same as that in Figure 3 of the main text. (d) Kurtosis in the modified Parisi-Frisch theory for , , , , . (b) and (d) Energy spectrum and rescaled kurtosis in the shell model for , , , , . In (e) kurtosis is rescaled with , which ensures that at the forcing scale.
For the shell model we use exactly the same , and values as that used in DNS. The energy spectrum of the shell model reproduces the arrest of direct cascade, see Figure S4. The suppression of kurtosis by odd viscosity is also observed in Figure S4 for scale above the dissipation scale. For scale below the dissipation scale, we find anomalous increase of kurtosis, which is a known issue for most shell models [15].
Intermittency suppressed by breakdown of multiple scale invariances
As mentioned in the main text, intermittency is suppressed in odd-wave-affected turbulence, because the multiple scale invariances of the Navier-Stokes equation are broken by odd viscosity. In this section we detail how the multiple scale invariances are broken by the parity-breaking waves produced by odd viscosity.
We start with the inertial-range Navier-Stokes equation
In the absence of , the inertial Navier-Stokes equation is invariant under global scale transformation [16]
for arbitrary exponent . We define this property as the multiple global scale invariances, which implies that singularities (flow with local self-similar structure) with arbitrary can develop in turbulent flow. In Parisi-Frisch multifractal formalism, the turbulent flow contains singularities with a range of exponents that are distributed in different locations in the 3D space, which further leads to intermittency [16, 17].
For non-zero , because of the odd term, the global scale invariance only holds for . All other global scale invariances are broken by the linear odd term. This conclusion can be generalized to any wave-turbulence system described in Ref. [18], where a wave-generating linear term breaks multiple global scale invariances. However, the breaking of multiple global scale invariances does not necessarily suppress intermittency. As an example, we consider a flow described by Eq. (20) and artificially block the energy transfer between all modes with . This enforces a 2D flow with on which odd viscosity has no effect at all, and 2D singularities with arbitrary are still allowed by scale symmetry (although in reality 2D turbulence is non-intermittent for a different reason [19, 20], see discussion at the end of the section). In other words, while the global scale invariance of Eq. (20) only holds for , its 2D part has partial scale invariances for all .
This example shows that what matters is not the global scale invariance of the complete Navier-Stokes equation, but is the partial scale invariance of the energy-transferring part of the equation. The non-energy-transferring part of the Navier-Stokes equation does not contribute to the turbulent cascade, hence it does not affect the intermittency.
To identify the energy-transferring part of Eq. (20), we perform helical decomposition,
where with and frequency
refers to the definite helicity of each helical mode. The explicit dependence of the frequency on the helicity is a result of the broken parity symmetry in odd-viscous fluids. We rewrite Eq. (20) (in the inertial range) as
Here, a triad with wave vectors and definite helicities is a basic unit for energy transfer.
is a coefficient that has the dimension of wave number. is an important frequency of the triad . As we discussed in the main text, energy transfer is effectively blocked when , because the fast oscillation of cancels the average energy transfer. Therefore, we may divide all triads into quasi-resonant triads and quasi-non-resonant triads . The energy transfer is approximately unaffected by odd viscosity in quasi-resonant triads, and is blocked in quasi-non-resonant triads.
The energy-transferring part of Eq. (20) is then a decimated Navier-Stokes equation in which only quasi-resonant triads are preserved:
While Eq. (26) has a similar form as that of the original Navier-Stokes equation, a different nonlinear dependence on emerges from the number of resonant triads, because .
As discussed in the main text, we consider quasi-local triads which dominate the energy cascade. Letting be the average velocity fluctuations at wave number , Eq. (26) is (statistically) simplified to
where is the fraction of quasi-resonant triads. For homochiral triads which transfer energy inversely , (because ), leading to . Substituting in Eq. (27), we have the following scale invariances of the homochiral channel:
for arbitrary . This is identical to Eq. (21).
For heterochiral triads which carries energy forward , . For given , is uniformly distributed in . Thus, for and for . In the wave-affected regime, substituting in Eq. (27), we have the following scale invariances of the heterochiral channel:
for arbitrary . Note that the scale transformations applied on are different in Eqs. (28),(29). Because energy is transferred through homochiral and heterochiral channels simultaneously, a singularity must satisfies both Eqs. (28),(29). In that case, the only scaling exponent consistent with both channels is , and the multiple partial scale invariances are broken. In this case, the flow is statistical self-similar, i.e., only holds for a single . Because the energy transfer is dominated by the forward flux of heterochiral triads, we have . In this case, and for consistency we have . This conclusion is used in the modified Parisi-Frisch theory in the main text (Eq. (2)).
The inconsistent scale invariances of homochiral and heterochiral triads are due to the asymmetric effects of odd viscosity on them, i.e., , which originates from the broken-parity symmetry of odd fluids. Indeed, the mechanism discussed above is only valid when the wave frequency explicitly depends on the definite helicity, which requires broken parity. This mechanism can be generalized to any generalized odd waves with frequency , as we discuss in § 7, or any parity-breaking waves. Their parity-preserving counterparts, e.g., waves with frequency that does not depend on helicity, do not break the multiple partial scale invariances of the Navier-Stokes equation. This is supported by our shell model results, see § 10. Intermittency has also been experimentally reported in parity-preserving wave turbulence, e.g., gravity capillary wave turbulence [21, 22].
Another consequence of the asymmetric effect of odd viscosity on homochiral and heterochiral channels is the enhanced flux loop, as shown in the main text. Because the homochiral and heterochiral channels transfer energy in different directions, a weak inverse flux in the homochiral channel exists in the direct cascade of ordinary turbulence [23]. The inverse flux of the homochiral channel, together with the forward flux of the heterochiral channel, forms a flux loop, which is amplified by odd viscosity for . More strikingly, when forcing at small scale , a flux loop develops at large scale in the absence of a net energy flux [3]. This is also observed in the shell model (Figure S5).
Figure S5. Flux loop for small-scale forcing. (a) Energy flux in DNS for . (b) Energy flux in the shell model for . Forcing is applied on .
Importantly, the breakdown of multiple scale invariances requires zero or small net helicity injection, which ensures that energy is equally distributed on waves with both definite helicities. When large helicity is injected, the flow can be dominated by waves with the same definite helicity as the injected one. In this case, the nonlinear transfer is dominated by the homochiral channel only, and the asymmetry between the two channels has small effect. Indeed, intermittency has been reported in rotating turbulence (parity is broken by the rotation direction) when large helicity is injected [24].
Another turbulent system that is known to be non-intermittent is 2D turbulence. Although this non-intermittency in 2D has not been fully understood, it must be governed by a different mechanism from what is discussed above. The reason is that the suppression of intermittency in 3D odd turbulence is due to the breakdown of scale invariances. 2D turbulence, on the other hand, has exactly the same scale invariances as that in 3D ordinary turbulence. Hence, intermittency is allowed in 2D from a perspective of scale invariance. It is conjectured that the absence of intermittency in 2D turbulence is related to the energy being transferred from fast degrees of freedom to slower degrees of freedom [25].
Generalized Odd Viscosity
In the main text we discuss the effect of odd viscosity on turbulent fluids, which produces linear waves with frequencies
In this section we generalize the discussion to generalized odd viscosity with wave frequencies
where is a generalized odd viscosity. can be any real number. The case for odd viscosity discussed in the main text corresponds to . When , one can show that Eq. (31) is equivalent to the Navier-Stokes equation for rotating turbulence with rotating frequency [26]. Importantly, all generalized odd waves are parity-breaking waves, since the frequency explicitly depends on the chirality.
We can repeat the analysis of the resonance condition for homochiral and heterochiral triads, as we do in the main text and § 6 for the case. For heterochiral triads with locality we find the fraction of resonant triads for and for . This modified the energy transfer rate to , which supports a steady-state, wave-affected solution . The scaling of the wave-affected solution is consistent with that obtained from the weak-turbulence theories [18, 26, 27]. The ’generalized odd’ scale is found by equating the wave-affected solution and the Kolmogorov solution. The dissipation scale can be estimated from , which leads to .
For homochiral triads with locality we again find , i.e., all local triads are resonant. Here, note that homochiral triads with exact locality do not transfer energy, because when . Therefore, homochiral triads can only transfer energy through weakly non-local triads, which are also weakly affected by the resonance condition when . For , on the other hand, we find that the resonance condition is satisfied by all homochiral triads regardless of the locality. A similar situation also happens in MHD turbulence, in which the frequencies of the Alfvén waves coincide with that of generalized odd waves with [27].
For arbitrary value, the effects of generalized odd viscosity on homochiral and heterochiral channels are asymmetric. Hence, the multiple partial scale invariances discussed in § 6 are broken for any in the wave-affected regime.
Modified Parisi-Frisch
We now extend the modified Parisi-Frisch formalism to generalized odd viscosity. In this case, self-similar solutions with at large scales reduce to the wave-affected solution (because ). The crossover is phenomenologically written as
which is same as Eq. (2) of the main text. Here, the crossover lengthscale is determined by equating two timescales: the eddy turnover time and the timescale of generalized odd waves , which gives . We adopt a commonly-used empirical choice of : , with and [28].
The suppression of intermittency at small lengthscales holds for arbitrary , which exhibits the same qualitative features as the results for shown in the main text. For , the generalized odd viscosity dominates at large lengthscales, e.g., in rotating turbulence, where Eq. (32) becomes invalid.
Two-channel helical shell model
We generalize here the two-channel helical shell model to include generalized odd viscosity. This is achieved by replacing Eq. (15) with
The value of controls which channel is affected more by generalized odd viscosity. According to our analysis of resonant triads in the main text, such that odd viscosity has a weaker effect on the homochiral channel. We show here the results for , which exhibits the same qualitative features as the case in the main text (Figure S6,Figure S6).

Figure S6. Shell model simulations for wave frequencies with . Inertial range is extended by reducing the normal viscosity, see Sec. II C. (a) Energy flux for parity-breaking waves () and . (b) Velocity-based kurtosis for parity-breaking waves () and (c) Flux-based kurtosis for parity-breaking waves () and same values as in (b). (d) Energy flux for parity-preserving waves () and . (e) Velocity-based kurtosis for parity-preserving waves () and values same as in (b). (f) Flux-based kurtosis for parity-preserving waves () and same values as in (b).
As shown in Figure 4(a), generalized odd viscosity modifies the scaling behavior of the structure functions for . This can be quantitatively described by a differential scaling exponent . We plot and in the shell model for different and values, vs rescaled wavenumber , see Figure S7-Figure S7, e.g. We find that changes from the usual multifractal exponents of regular turbulence for to a plateau for . The plateau corresponds to a fully wave-affected regime where . We fit the values of using the shaded area, and the results are shown in Figure 4d of the main text. We then fit using the monofractal formula . The numerical values of are close to the theoretical approximation (Figure S7). The theoretical is obtained using the resonant fraction, see below Eq. (31).

Figure S7. Differential scaling exponent of (a-c) and (e-g) in the shell model with generalized odd viscosity (inducing generalized odd waves with frequency ). Exponents are plotted vs rescaled wavenumber , with . Results are shown for , with . (a, e) and . (b, f) and . (c, g) and . (d) Coefficient obtained from the shell model. Theoretical approximation is shown for comparison.
Parity-breaking waves vs parity-preserving waves
As discussed in § 6, the multiple scale invariances of the Navier-Stokes equation are broken by the asymmetric effects of (generalized) odd viscosity on homochiral and heterochiral channels. This mechanism only applies to parity-breaking waves which break parity symmetry. For parity-preserving waves, the effects on homochiral and heterochiral channels are symmetric and intermittency may still develop.
This argument can be tested by the two-channel helical shell model with generalized odd viscosity (§ 7 B). The asymmetry between homochiral and heterochiral channels is controlled by the parameter in Eq. (33): when the two channels are affected in a symmetric way by generalized odd viscosity; when the homochiral channel is less affected. Hence, while the shell model with corresponds to parity-breaking waves with , the shell model with corresponds to parity-preserving waves with, e.g., .
We explore the intermittency for both the parity-breaking and parity-preserving models with . As expected, the flux-loop is enhanced by in the parity-breaking model (Figure S6) but is not enhanced in the parity-preserving model (Figure S6). The kurtosis is suppressed by large in the parity-breaking model (Figure S6). In the parity-preserving model, on the other hand, the kurtosis still increases with a relatively slow speed for large , suggesting that the intermittency is not completely suppressed. This difference is more obvious for the flux-based kurtosis, defined by , where is the flux-based structure function [13, 14]. We find that while the parity-breaking model is still non-intermittent in the wave-affected regime, the parity-preserving model is strongly intermittent (Figure S6,Figure S6).
A similar comparison is done for odd waves () (Figure S8, comparing a,b,c with d,e,f). Surprisingly, we find that the velocity-based kurtosis is suppressed in both the parity-breaking and parity-preserving models (Figure S8,Figure S8). When looking at the flux-based kurtosis we again observe different results: kurtosis does not develop in the parity-breaking model but develops slowly in the parity-preserving model (Figure S8,Figure S8). This suggests that the parity-breaking model is non-intermittent and the parity-preserving model is intermittent in the wave-affected regime. However, the intermittency of the parity-preserving model is not reflected on the velocity-based kurtosis. This is possibly a

Figure S8. Shell model simulations for wave frequencies . Inertial range is extended by reducing the normal viscosity, see Sec. II C. (a) Energy flux for parity-breaking waves () and . (b) Velocity-based kurtosis for parity-breaking waves () and . (c) Flux-based kurtosis for parity-breaking waves () and same values as in (b). (d) Energy flux for parity-preserving waves () and . (e) Velocity-based kurtosis for parity-preserving waves () and values same as in (b). (f) Flux-based kurtosis for parity-preserving waves () and same values as in (b).
| Shell | Theory | |
| 1.2 | 0.175 | 0.2 |
| 1.5 | 0.104 | 0.125 |
| 2 | 0.027 | 0 |
coincidence for : when , the wave-affected solution coincides with the equipartition solution (all have similar velocity fluctuations). This leads to strong fluctuations in the wave-affected regime which may mask the non-Gaussian fluctuations of . For the 3D Navier-Stokes equation the equipartition solution holds at , hence a similar coincidence may happen at instead of .
Shell Model with Stochastic Waves
In our two-channel helical shell model, the generalized odd waves in the Navier-Stokes equation are not explicitly included. The effect of generalized odd waves are modelled by the fractions of resonant triads . A different approach was adopted in prior shell models of rotating and MHD turbulence, which modifies the classic one-channel shell model with explicit waves [29, 30]. This approach is summarized in the following form:
where
is the nonlinear transfer of Sabra shell model with . It has been argued that to reproduce the wave-affected energy spectrum, the wave frequencies cannot be assumed constant, but has to be stochastic with [29]. The reason for this may be explained by the resonance condition: In the Navier-Stokes equation, generalized odd waves oscillate with frequencies , where is the angle between and . The 3D distribution of effectively induces a wide distribution of the wave frequencies for given , which further guarantees that the resonant condition can be satisfied by some triads. In shell models, however, this 3D structure is absent and the interaction among triads is limited to nearest neighbors. Hence, if constant frequencies is assumed, one can show that the resonant condition will never be satisfied, with .
The stochasticity of in the shell model effectively reproduces the wide distribution of wave frequencies in the Navier-Stokes equation. The difference is that in the shell model, different is distributed in different time period, while in the Navier-Stokes equation, different frequencies are distributed in different triads. This type of shell models is able to reproduce the average energy transfer of the Navier-Stokes equation, however, it would be inappropriate to study intermittency with these models. The reason is that the stochasticity naturally breaks scale invariances at any given time, hence intermittency must be suppressed. This suppression is just caused by an artifact of the stochastic shell model that is absent in the Navier-Stokes equation.
References
- [1]M. Fruchart, M. Han, C. Scheibner, and V. Vitelli, “The odd ideal gas: Hall viscosity and thermal conductivity from non-hermitian kinetic theory,” (2022), arXiv:2202.02037.arxiv.org/abs/2202.02037
- [2]T. Khain, C. Scheibner, M. Fruchart, and V. Vitelli, Journal of Fluid Mechanics 934, A23 (2022).
- [3]X. M. de Wit, M. Fruchart, T. Khain, F. Toschi, and V. Vitelli, Nature 627, 515 (2024).
- [4]P. A. Davidson, Turbulence: an introduction for scientists and engineers (Oxford university press, 2015).
- [5]A. Alexakis and L. Biferale, Physics Reports 767–769, 1–101 (2018).
- [6]M. K. Verma, Energy transfers in fluid flows: multispectral and spectral perspectives (Cambridge University Press, 2019).DOI
- [7]G. Dar, M. K. Verma, and V. Eswaran, Physica D: Nonlinear Phenomena 157, 207 (2001), 0109004 [nlin].
- [8]A. Alexakis, P. D. Mininni, and A. Pouquet, Physical Review E 72, 046301 (2005).
- [9]P. D. Mininni, A. Alexakis, and A. Pouquet, Physical Review E 72, 046302 (2005).
- [10]R. Peyret, Spectral Methods for Incompressible Viscous Flow (Springer, 2002).DOI
- [11]R. Benzi, L. Biferale, R. Kerr, and E. Trovatore, Physical Review E 53, 3541 (1996).
- [12]L. Biferale, Annual Review of Fluid Mechanics 35, 441–468 (2003).
- [13]V. S. L’vov, E. Podivilov, A. Pomyalov, I. Procaccia, and D. Vandenbroucq, Physical Review E 58, 1811 (1998).
- [14]X. M. de Wit, G. Ortali, A. Corbetta, A. A. Mailbyaev, L. Biferale, and F. Toschi, Physical Review E 109, 055106 (2024).
- [15]R. Benzi and L. Biferale, Journal of Statistical Physics 135, 977 (2009).
- [16]G. Parisi and U. Frisch, in Proceedings of the International School of Physics “E. Fermi”, edited by M. Ghil, R. Benzi, and G. Parisi (North-Holland, Amsterdam, 1985) pp. 84–87.
- [17]U. Frisch, Turbulence: The Legacy of A. N. Kolmogorov (Cambridge University Press, 1995).
- [18]V. Zakharov, V. L’vov, and G. Falkovich, Kolmogorov Spectra of Turbulence I: Wave Turbulence (Springer-Verlag, Berlin, Germany, 1992).
- [19]G. Paladin and A. Vulpiani, Physics Reports 156, 147 (1987).
- [20]G. L. Eyink and K. R. Sreenivasan, Rev. Mod. Phys. 78, 87 (2006).
- [21]E. Falcon, S. Fauve, and C. Laroche, Physical Review Letters 98, 154501 (2007).
- [22]S. Lukaschuk, S. Nazarenko, S. McLelland, and P. Denissinko, Physical Review Letters 103, 044501 (2009).
- [23]E. Deusebio, G. Boffetta, E. Lindborg, and S. Musacchio, Physical Review E 90, 023005 (2014).
- [24]P. D. Mininni and A. Pouquet, Physics of Fluids 22, 035106 (2010).
- [25]N. Vladimirova, M. Shavit, and G. Falkovich, Physical Review X 11, 021063 (2021).
- [26]S. Galtier, Physical Review E 68, 051301 (2003).
- [27]S. Galtier, S. Nazarenko, A. C. Newell, and A. Pouquet, Journal of plasma physics 63, 447 (2000).
- [28]Z.-S. She and E. Lévêque, Physical Review Letters 72, 336 (1994).
- [29]Y. Hattori, R. Rubinstein, and A. Ishizawa, Physical Review E 70, 046311 (2004).
- [30]F. Plunian and R. Stepanov, Physical Review E 82, 046311 (2010).