--- title: Code-Mixing Guided Preference-Learning Framework url: https://www.emergentmind.com/topics/code-mixing-guided-preference-learning-framework type: topic --- # Code-Mixing Guided Preference-Learning Framework Searching arXiv for the cited papers and closely related work on variance cascades. Variance cascade refers, in the cited literature, to the transfer of a variance-like quadratic quantity across an ordered hierarchy of scales or stages. In turbulence this quantity is typically temperature variance, thermal variance, scalar variance, or surface potential energy, and the cascade may be direct, toward smaller scales, or inverse, toward larger scales. The same vocabulary is extended in other fields to volatility across resolutions in financial time series and to variance propagation across sequential slots or iterations in learning algorithms. This suggests a unifying usage: redistribution of fluctuations through a cascade structure, with fluxes, spectra, or recursive variance terms providing the diagnostic language [2406.07235, 2207.12606, 2403.05401, 2010.12270]. ## 1. Variance as a cascaded invariant in geophysical turbulence In the \(\alpha\)-turbulence family, the active scalar \(\theta(x,t)\) is advected by a two-dimensional incompressible velocity \(v=\nabla^\perp\psi\), with constitutive law \[ \psi = |\Delta|^{-\alpha/2}\theta, \qquad \widehat v(k)= i\,k^\perp\,k^{-\alpha}\,\widehat\theta(k). \] For \(\alpha=1\) this is the Surface Quasi-Geostrophic model, while \(\alpha=2\) gives the 2D Navier-Stokes vorticity equation. In the inviscid, unforced limit these models conserve the generalized energy \(E=\tfrac12\langle \psi\theta\rangle\) and the generalized enstrophy, or in SQG language the surface potential energy (variance), \(P=\tfrac12\langle \theta^2\rangle\). When forcing injects variance at a scale \(\ell_f\), there is a direct cascade of \(P\) toward \(\ell\to0\) and, for \(\alpha>0\), an inverse cascade of \(E\) toward \(\ell\to\infty\) [2406.07235]. A rigorous SQG-oriented treatment formulates the cascade through localized fluxes and multi-scale averaging rather than through a spectral closure. In that framework, the governing equation is \((\partial_t+u\cdot\nabla+\Lambda^\alpha)\theta=0\), with \(u=(-\partial_2\Psi,\partial_1\Psi)\) determined by \(\Lambda\Psi=-\theta\). The key object is the localized variance flux into balls of radius \(R\), and a sufficient condition for a direct variance cascade is expressed by the inequality \(\sigma_0<\beta R_0\), where \(\sigma_0\) is a cascade-cutoff length built from macro-scale averages of \(\theta^2\), its extension \((\theta^*)^2\), and the dissipation density \(|\nabla\theta^*|^2 z^{1-\alpha}\). Under this condition, for every \(R\) with \((1/\beta)\sigma_0\le R\le R_0\), the ensemble-averaged localized flux obeys \[ \frac{1}{4K_1}\,\varepsilon_0^* \le \langle F\rangle_R \le 4K_2\,\varepsilon_0^*, \] which identifies an inertial range carrying a direct temperature variance cascade [1309.1962]. The geophysical significance of this formulation is that the cascaded quantity is not merely a passive scalar norm. In SQG, the velocity is slaved to \(\theta\) by a non-local law, and the dissipation is fractional. The cascade therefore depends on non-local inertial transport and on the Caffarelli-Silvestre extension that restores a local positive dissipation density in the extended variable \(z\) [1309.1962]. ## 2. Non-equilibrium corrections in the direct SQG cascade A recent development studies not only the mean cascade but the temporal fluctuations of the flux of surface potential energy. The shell-integrated variance spectrum \(P(k,t)\) satisfies the exact spectral balance \[ \partial_t P(k,t)=F(k,t)-\partial_k\Pi(k,t)-D(k,t), \qquad D(k,t)=2\kappa k^2 P(k,t). \] In the inertial range \(k_f\ll k\ll k_\kappa\), forcing and dissipation are neglected, so \(\partial_t P+\partial_k\Pi\approx0\). The central claim is that the instantaneous imbalance in the energy budget originates a subleading correction to the spectrum of the turbulent cascade [2406.07235]. The analysis assumes a multiple-scale expansion \[ P=P_0+\varepsilon P_1,\qquad \Pi=\Pi_0+\varepsilon \Pi_1, \] with slow temporal drift \(\partial_t\to \varepsilon\,\partial_t\), and closes the flux through the Kovasznay-type dimensional Ansatz \[ \Pi(k,t)=C\,k^{(7-2\alpha)/2}\,P(k,t)^{3/2}. \] At leading order one obtains \(\Pi_0(k,t)=\varepsilon(t)\), the instantaneous dissipation-injection rate, and the classical inertial-range law \[ P_0(k,t)=C_0\,\varepsilon(t)^{2/3}\,k^{(2\alpha-7)/3}. \] At first order, assuming \(P_1(k,t)=g(t)k^\beta\), exponent matching yields \[ \beta=\frac{4\alpha-11}{3}, \qquad P_1(k,t)=C_1\,\varepsilon(t)^{-2/3}\,\dot\varepsilon(t)\,k^{(4\alpha-11)/3}. \] Because \(0<\alpha<2\), the correction is steeper and subdominant at high \(k\) [2406.07235]. For \(\alpha=1\), corresponding to SQG, the equilibrium exponent is \(-5/3\) and the non-equilibrium correction exponent is \(-7/3\). The time-averaged inertial range displays the \(k^{-5/3}\) law, while the instantaneous deviation exhibits a \(k^{-7/3}\) tail. As \(\alpha\to2\), the prefactor \(C_1=2C_0^2/[3(2-\alpha)]\) diverges, so the perturbation theory breaks down and one recovers the well-known logarithmic corrections of the enstrophy cascade [2406.07235]. The same work gives an operational decomposition of instantaneous spectra into equilibrium and non-equilibrium parts. Defining \[ A(k)=\langle P(k,t)\,\varepsilon(t)^{-2/3}\rangle_t, \qquad P_0(k,t)=\varepsilon(t)^{2/3}A(k), \qquad P_1(k,t)=P(k,t)-P_0(k,t), \] one isolates the pure \(k^{(2\alpha-7)/3}\) law in \(A(k)\) and the steeper correction in \(P_1\). This disentangling procedure is explicitly proposed as generalizable to other turbulent systems [2406.07235]. ## 3. Linear stochastic realization and fractional-Gaussian scaling A different realization of variance cascade is provided by a linear stochastic PDE designed to model the spatial structure of turbulent velocity fields. In Fourier space, the field \(\widehat u(t,k)\) evolves according to \[ \partial_t \widehat u + c\,\mathrm{div}_k\!\left[\frac{k}{|k|}\widehat u\right] + c\,\frac{H+1/2}{|k|}\widehat u = -\nu(2\pi|k|)^2\widehat u +\widehat f(t,k), \] with \(\widehat f\) a real-valued, mean-zero Gaussian forcing that is \(\delta\)-correlated in time and space and supported in an annulus \(\kappa\le |k|\le k_f\). The cascade mechanism is the radial transport operator \(\mathrm{div}_k((k/|k|)\widehat u)\), which in polar coordinates becomes \(\partial_\rho \widehat u + (d-1)\widehat u/\rho\). In the inviscid, unforced limit, and ignoring the weaker \((H+1/2)/\rho\) term, the dynamics reduces to \(\partial_t\widehat u + c\,\partial_\rho\widehat u=0\), so any packet of variance originally at scale \(\rho_0\) is carried to larger wavenumbers \(\rho=\rho_0+ct\) [2403.05401]. Because the white-in-time forcing injects variance only into \(\kappa\le |k|\le k_f\), transport carries that injected variance outward to all \(\rho\). In the stationary inviscid limit, and for \(\rho\gg k_f\), the mode-by-mode covariance obeys \[ \mathbb E[|\widehat u|^2](\rho)\propto \rho^{-(2H+d)}, \] which is the classic power-law variance spectrum of a fractional Gaussian field of Hölder index \(H\). In physical space, the exact relation \[ S_2(\ell)=2\int [1-\cos(2\pi k\cdot \ell)]\,E(k)\,dk \] leads to \(S_2(\ell)\sim C_d|\ell|^{2H}\) for \(|\ell|\ll k_f^{-1}\). In particular, for \(d=3\) and \(H=1/3\), one recovers \(E(k)\sim k^{-11/3}\), the shell-integrated energy density \(4\pi k^2E(k)\sim k^{-5/3}\), and \(S_2(\ell)\sim \ell^{2/3}\) [2403.05401]. The numerical evidence is based on one-, two-, and three-dimensional finite-volume discretizations in \(k\)-space with a splitting integrator in time. Angle-averaged periodograms display the exact power-law slope \(-(2H+d)\) over an increasingly wide inertial range at smaller \(\nu\), and the structure functions transition from \(S_2\sim \ell^{2H}\) to \(\ell^2\) in the viscous subrange. In this model, the cascade of variance is analytically explicit: large-scale stochastic injection, deterministic outward transport in Fourier space, and viscous arrest at high \(k\) [2403.05401]. ## 4. Inverse and coupled variance cascades in convection and electrokinetic turbulence Variance cascades are not confined to direct small-scale transfer. In three-dimensional, horizontally extended turbulent Rayleigh-Bénard convection with imposed heat-flux boundary conditions, the total thermal variance is \(\tfrac12\langle \Theta^2\rangle\), with spectral density \(E_{\Theta\Theta}(\mathbf k,t)=\tfrac12|\widehat\Theta(\mathbf k,t)|^2\). The mode-to-mode transfer is \[ S_\Theta(\mathbf k\mid \mathbf p\mid \mathbf q,t) = \Im\!\left\{[\mathbf k\cdot \widehat{\mathbf u}(\mathbf q,t)]\,\widehat\Theta(\mathbf p,t)\,\widehat\Theta^*(\mathbf k,t)\right\}, \] and the cumulative planar flux among vertically homogeneous modes is denoted \(\Pi_\Theta^{\mathrm h}(k_h',t)\). A negative sign of \(\Pi_\Theta^{\mathrm h}\) on some range indicates an inverse cascade, transferring variance to larger horizontal scales [2207.12606]. In the non-rotating case, at all Rayleigh numbers up to \(\Ra\sim10^7\), one finds \(\Pi_\Theta^{\mathrm h}(k_h')<0\) for \(k_h'\lesssim k_{\rm inv}\), and positive flux above that scale, so the dynamics contains both an inverse and a direct branch. The peak of the thermal spectrum \(\bar E_{\Theta\Theta}(k_h,0)\) drifts toward the domain fundamental \(k_{\min}=2\pi/\Gamma\), and once the aggregation fills the domain the inverse cascade ceases. The negative transfer is dominated by planar triads with \(k_z=p_z=q_z=0\), while interactions involving nonzero \(k_z\) do not produce inverse transfer. Weak rotation arrests the inverse cascade at a finite scale \(\Lambda(\Ro)\), with the empirical law \[ \Lambda(\Ro)\approx a(\Ra)\,\Ro+b(\Ra), \qquad 2.5\lesssim \Ro\lesssim \Ro^*. \] A parallel inverse cascade of kinetic energy persists even after the thermal variance saturates, sustaining the large \(k_z=0\) rolls that feed the planar thermal-variance cascade [2207.12606]. Electrokinetic turbulence provides a different coupled setting in which scalar variance and kinetic energy are linked by an externally forced conservation law. Under statistical stationarity, and in the subrange where electric body force dominates and dissipation is negligible, the forcing components satisfy \(F_u(k)-MkF_\theta(k)=0\), giving the universal conservation law \[ \Pi_u(k)+Mk\,\Pi_\theta(k)=0. \] This produces a quad-cascade picture with four asymptotic regimes. For the scalar-variance spectrum \(E_\theta(k)\), the predicted exponents are \(k^{-5/3}\) when both \(\Pi_u\) and \(\Pi_\theta\) are constant, \(k^{-9/5}\) when \(\Pi_\theta\) is constant and \(\Pi_u\) is variable, \(k^{-7/3}\) when \(\Pi_u\) is constant and \(\Pi_\theta\) is variable, and \(k^{-3}\) when both are variable [2301.07547]. Microfluidic experiments support three of these regimes. At \(\mathrm{Ra}_e\simeq2.45\times10^6\), a clear \(k^{-7/3}\) range appears; between \(\mathrm{Ra}_e\approx3.3\)–\(4.3\times10^6\), the spectrum shifts toward a \(k^{-9/5}\) range; and above \(\mathrm{Ra}_e\approx7.8\times10^6\), a \(k^{-5/3}\) inertial subrange emerges. The cross-over wavenumber scales as \(k_z\propto \mathrm{Ra}_e^{0.72\pm0.05}\), in close agreement with the model prediction \(k_z\propto \mathrm{Ra}_e^{3/4}\) [2301.07547]. ## 5. Cascade constructions beyond fluid turbulence In financial modeling, a volatility cascade is formulated as a stochastic differential equation in the log-scale variable \(\lambda=\log(L/s)\), with \(x(\lambda)=|W_\psi Z[\cdot,s(\lambda)]|\) as the volatility proxy: \[ dx(\lambda) = x(\lambda)\bigl[-\gamma_M\,d\lambda+\sigma_M\,dB_M(\lambda)\bigr] +a_A(\lambda)\,d\lambda+b_A(\lambda)\,dB_A(\lambda). \] The multiplicative source governs the cascade, while the additive source is treated perturbatively through \(a_A(\lambda)=\varepsilon e^{-\lambda/2}\), \(b_A(\lambda)=O(\varepsilon^{1/2})e^{-\lambda/2}\), with \(\varepsilon\ll1\). The stationary Fokker-Planck density \[ p_{\rm st}(x) = C\,[b_A^2+\sigma_M^2x^2]^{-(1+\gamma_M/\sigma_M^2)} \exp\!\left\{\frac{2a_A}{\sigma_M b_A}\arctan\!\left(\frac{\sigma_M x}{b_A}\right)\right\} \] has a Student-\(t\) tail when \(a_A\to0\), and the \(q\)th moments scale as \(S_q(\lambda)\sim s^{\tau(q)}\) with \[ \tau(q)= -1 + (\gamma_M+\tfrac12\sigma_M^2)q - \tfrac12\sigma_M^2 q^2. \] The non-quadratic dependence of \(\tau(q)\) is identified as the hallmark of multifractality, and the paper explicitly frames the mechanism as a variance-cascade analogy to turbulence [2010.12270]. In ranking-policy evaluation under the cascade behavior model, the cascade is sequential rather than spectral. A user examines slot \(l\) only if no click occurred at earlier slots, with \(e_1\equiv1\) and \(e_l=\prod_{k