A Multiscale Camassa–Holm Equation Darryl D. Holm, Maneesh Kumar Singh Abstract A system of equations for Multiscale Geodesic Flow (MGF) is introduced whose solutions illustrate the paradigm of whorls within whorls within whorls, introduced by L. F. Richardson in 1922 to describe the cascade of energy in fluid turbulence. Numerical simulations are given for MGF on S 1 S^{1} , where the MGF equation comprises a multiscale generalisation of the Camassa-Holm equation (CHE) whose emergent singular solutions generalise the peakon solutions of the CHE. 1 Introduction Based on the soliton-bearing Camassa–Holm (CH) equation on S^{1} derived in [ Camassa and Holm, 1993 ] a new class of models for the mean motion of ideal incompressible fluids in three dimensions (3D) was proposed and analysed in [ Holm et al., 1998b ] . In the 3D CH equation the Lagrangian–averaged amplitude of the fluctuations introduces a spatial scale, \alpha , and below this scale the Lagrangian transport velocity is filtered by nonlinear dispersion. This \alpha -filtering enhances the stability and regularity of the new fluid models without compromising either their large scale behaviour, or their conservation laws. These models also describe geodesic motion on the volume-preserving diffeomorphism group for a metric containing the H^{1} norm of the fluid velocity. The 3D CH equation came to be called the Lagrangian–Averaged Euler- \alpha (LAE- \alpha ) model in [ Marsden et al., 2000 ] where the vector field solutions that generate the geodesic Lagrangian path of the 3D CH flow were proved to be smooth and converge to solutions of the Euler fluid equation in the limit \alpha\to 0 . The viscous 3D CH equation was proposed as a closure approximation for the Reynolds-averaged equations of the divergence-free Navier-Stokes flow in [ Chen et al., 1998 ] . This closure approximation was evaluated on turbulent channel and pipe flows with steady mean, and its analytical solutions for the mean velocity and the Reynolds shear stress were found to be consistent with experiments in most of the flow region. In [ Foias et al., 2002 ] , the viscous 3D CH equation was shown to possess global strong solutions which converge to solutions of the Navier–Stokes fluid equation in the limit \alpha\to 0 . Thereafter, the viscous 3D CH equation was called the Lagrangian-Averaged Navier–Stokes alpha (LANS- \alpha ) model and was tested as a Large Eddy Simulation (LES) model in various physical applications, including magnetohydrodynamics (LAMHD- \alpha ) in astrophysics [ Pietarila Graham et al., 2006 ] , as well as rotation and stratification in oceanography [ Hecht et al., 2008 ] . The LANS- \alpha model also satisfies the Kármán–Howarth turbulence theorem, [ Holm, 2002 ] . The present paper derives N -dimensional equations for multiscale geodesic models of the famous Richardson cascade of energy and circulation [ Richardson, 1922 ] , as illustrated by computational simulations of their solutions on S^{1} . Figure 1 : The MGF equation ( 2.3 ) is solved by midpoint scheme in time and Galerkin FEM in space. Fig. (a)-(g) show velocity profiles u_{1} (blue), u_{2} (red), u_{3} (green). An initial 3-component peakon-antipeakon pair starts on either side of the periodic domain of length 100 at time t=0 . A train forms of rightward moving (3,2,1) peakons and leftward moving (3,2,1) antipeakons at t=9 . Head-on collisions for t=10 (3-3), t=16 (2-3) and t=19 (1-3). Fig. (h) shows the flow of total constant energy from 1-2-3 with sudden energy exchanges at the collision times. 2 Multiscale geodesic flows (MGF) Geodesic flows on the manifold of smooth invertible maps G acting on \mathbb{R}^{N} may be lifted as curves g_{t}\in G parametrised by time t as x_{t}={g_{t}}x_{0}\in\mathbb{R}^{N} . The flows g_{t}\in G may be derived via Hamilton’s principle 0={\delta}S={\delta}\int\ell(u)\,dt for a kinetic energy Lagrangian \ell(u)=\tfrac{1}{2}\|u\|^{2}=\tfrac{1}{2}\int_{\mathbb{R}}u\,Q^{op}u\,dx with vector-field velocity u=\dot{g}_{t}{g_{t}}^{-1}\in\mathfrak{g}\simeq T_{e}G and positive definite, symmetric differential operator Q^{op} 0={\delta}S={\delta}\int_{0}^{T}\ell(u)+\left_{\mathfrak{g}^{*}\times\mathfrak{g}}\,dt\,. Here, the brackets \left<\cdot\,,\,\cdot\right>_{\mathfrak{g}^{*}\times\mathfrak{g}} denote L^{2} pairing between the Lie algebra of vector fields \mathfrak{g} and its dual \mathfrak{g}^{*} . The dual Lie algebra \mathfrak{g}^{*} is a Poisson manifold, whose Lie–Poisson bracket is discussed below. Variations of this Hamilton principle yield an evolution equation expressed by the Lie chain rule , \frac{d}{dt}{g_{t}}_{*}m(x,t)=-\,{g_{t}}_{*}({\cal L}_{u}m)\,,\quad\hbox{with}\quad u=\dot{g}_{t}g_{t}^{-1}\,, in which {g_{t}}_{*} denotes push-foward by the map g_{t}\in G [ Holm et al., 1998a , Crisan et al., 2022 ] Thus, the coordinate-free Lie derivative {\cal L}_{u} is defined as (minus) the tangent to the push-forward {g_{t}}_{*} along the Eulerian vector field u=\dot{g}_{t}g_{t}^{-1}\in T_{e}G which generates Lagrangian flow g_{t} and satisfies the equation 1 1 1 The Lie-derivative and differential-form coordinate-free notation used here is appropriate, since solutions of the CH equation follow geodesic paths on the manifold of smooth invertible maps (diffeomorphisms) with respect to the kinetic energy metric given by H^{1}(\mathbb{R}^{N}) for any integer N . Thus, for example, m(x,t)={g_{t}}_{*}m(x_{0},0) with g_{0}=Id is the push-forward of the initial momentum density m_{0} by the smooth invertible flow map g_{t} depending on time t for any number of dimensions, N . \partial_{t}m=-\,{\cal L}_{u}m=-\,(\partial_{x}m+m\partial_{x})u\,, with m(x,t)={\delta}\ell/{\delta}u=Q^{op}u for geodesic flow. Inverting the operator Q^{op} solves for u=\dot{g}_{t}g_{t}^{-1} from which g_{t} may be reconstructed for Lagrangian fluid trajectories. After Legendre transforming to the conserved energy Hamiltonian h(m)=\left_{\mathfrak{g}^{*}\times\mathfrak{g}}-\ell(u)\,, the Lie-Poisson Hamiltonian formulation of geodesic flow on the Poisson manifold m\in P=\mathfrak{g}^{*} , with Lie-Poisson bracket \{f,h\} written in terms of Lie algebra bracket [u,v]=:{\rm ad}_{u}v , \displaystyle\{f,h\}=-\left_{\mathfrak{g}^{*}\times\mathfrak{g}}=-\left<{\rm ad}^{*}_{{\delta}h/{\delta}m}m\,,\,{\delta}f/{\delta}m\right>_{\mathfrak{g}^{*}\times\mathfrak{g}}=-\left<{\cal L}_{{\delta}h/{\delta}m}m\,,\,{\delta}f/{\delta}m\right>_{\mathfrak{g}^{*}\times\mathfrak{g}}\,, which will be extended here from GF to MGF by introducing a multiscale Hamiltonian. 2.1 Lie–Poisson Hamiltonian MGF in 1D The Hamiltonian for a 1D MGF system in \mathbb{R} is \displaystyle H(\{m\})=\frac{1}{2}\sum_{k=1}^{n}\int_{\mathbb{R}}m_{\alpha_{k}}K_{\alpha_{k}}*m_{\alpha_{k}}\,dx\,, (2.1) with Eulerian fluid velocity u_{\alpha_{k}}(x,t) in the sum over finer and finer discrete scales, \alpha_{k}\in\mathbb{R} where k=1,2,\dots,n, \alpha_{k+1}/\alpha_{k}=2^{-k} and m_{\alpha_{k}}(x,t) is defined as \displaystyle m_{\alpha_{k}}(x,t) \displaystyle={Q^{op}}_{\alpha_{k}}u_{\alpha_{k}}:=(1-\alpha_{k}^{2}\partial_{x}^{2})u_{\alpha_{k}}(x,t)\,, \displaystyle u_{\alpha_{k}} \displaystyle=K_{\alpha_{k}}*m_{\alpha_{k}}=:\int_{\mathbb{R}}K_{\alpha_{k}}(|x-y|)m_{\alpha_{k}}(y,t)\,dy\,, where K_{\alpha_{k}}(|x-y|) is the Green function for the Helmholtz operator (1-\alpha_{k}^{2}\partial_{x}^{2}) on the real line, for boundary conditions \lim_{|x|\to\infty}u_{\alpha_{k}}(x,t)=0 . The Lie–Poisson form of the dynamics for the starting Hamiltonian with \delta h/\delta m_{\alpha_{k}}=u_{\alpha_{k}} and n=3 is [ Holm, 2025 ] \displaystyle\partial_{t}\begin{pmatrix}m_{\alpha_{1}}\\ \\ m_{\alpha_{2}}\\ \\ m_{\alpha_{3}}\end{pmatrix}=-\begin{bmatrix}\mathrm{ad}^{*}_{\Box}m_{\alpha_{1}}&\mathrm{ad}^{*}_{\Box}m_{\alpha_{2}}&\mathrm{ad}^{*}_{\Box}m_{\alpha_{3}}\\ \\ \mathrm{ad}^{*}_{\Box}m_{\alpha_{2}}&\mathrm{ad}^{*}_{\Box}m_{\alpha_{2}}&\mathrm{ad}^{*}_{\Box}m_{\alpha_{3}}\\ \\ \mathrm{ad}^{*}_{\Box}m_{\alpha_{3}}&\mathrm{ad}^{*}_{\Box}m_{\alpha_{3}}&\mathrm{ad}^{*}_{\Box}m_{\alpha_{3}}\end{bmatrix}\begin{pmatrix}u_{\alpha_{1}}\\ \\ u_{\alpha_{2}}\\ \\ u_{\alpha_{3}}\end{pmatrix}=-\begin{pmatrix}\mathrm{ad}^{*}_{u_{\alpha_{1}}}m_{\alpha_{1}}+\mathrm{ad}^{*}_{u_{\alpha_{2}}}m_{\alpha_{2}}+\mathrm{ad}^{*}_{u_{\alpha_{3}}}m_{\alpha_{3}}\\ \\ \mathrm{ad}^{*}_{u_{\alpha_{1}}+u_{\alpha_{2}}}m_{\alpha_{2}}+\mathrm{ad}^{*}_{u_{\alpha_{3}}}m_{\alpha_{3}}\\ \\ \mathrm{ad}^{*}_{u_{\alpha_{1}}+u_{\alpha_{2}}+u_{\alpha_{3}}}m_{\alpha_{3}}\end{pmatrix}, in which the higher (smaller) scales are seen to apply forces to the lower (larger) scales, while the sum of velocities of the lower (larger) scales transports the momenta of the higher (smaller) scales. The matrix operator in square brackets above defines the following semidirect-product Lie-Poisson bracket [ Holm et al., 1998a ] \displaystyle\begin{split}\frac{df}{dt}=\big\{f,h\big\}&=\sum_{k=1}^{n}\left=-\sum_{k=1}^{n}\left\\ &=-\sum_{k=1}^{n}\left<{\rm ad}^{*}_{\frac{\delta h}{\delta m_{\alpha_{k}}}}m_{\alpha_{k}}\,,\,\frac{\delta f}{\delta m_{\alpha_{k}}}\right>,\end{split} (2.2) where \left<\cdot\,,\,\cdot\right> denotes L^{2} pairing of functions and [\cdot,\cdot] denotes the Lie algebra bracket of vector fields. That is, \bigg[\frac{\delta f}{\delta m},\frac{\delta h}{\delta m}\bigg]={\rm ad}_{\frac{\delta h}{\delta m}}\frac{\delta f}{\delta m}=\frac{\delta f}{\delta m}\partial_{x}\frac{\delta h}{\delta m}-\frac{\delta h}{\delta m}\partial_{x}\frac{\delta f}{\delta m}. on the dual of the following nested semidirect product Lie algebra \mathfrak{s}=\mathfrak{g}_{\alpha_{1}}\ \circledS\big(\mathfrak{g}_{\alpha_{2}}\ \circledS\ \mathfrak{g}_{\alpha_{3}}\ \big) , with dual coordinates given by m_{\alpha_{k}}\in\mathfrak{g}_{\alpha_{k}}^{*} dual to u_{\alpha_{k}}\in\mathfrak{g}_{\alpha_{k}} correspondingly for k=1,2,3 , and symbol \circledS denotes the semidirect product operation [ Holm et al., 1998a ] . Physically, the Lie-Poisson bracket \{f,h\} in ( 2.2 ) refers to three types of nested fluid flow. Its decreasing filter widths \alpha_{k} for \alpha_{k+1}/\alpha_{k}=2^{-k} mimic L. F. Richardson’s famous “whorls within whorls within whorls” reference in characterising fluid dynamics [ Richardson, 1922 ] . Big whorls have little whorls, that feed on their velocity, and little whorls have lesser whorls, and so on to viscosity. – L. F. Richardson (1922) In a mathematical sense, then, the Lie–Poisson dynamics \frac{df}{dt}=\{f,h\} in ( 2.2 ) mimics ‘Richardson’s triple’. Namely, Richardson’s triple is modelled as the composition of three non-commutative right group actions of the smooth invertible maps G_{\alpha_{1}}\times G_{\alpha_{2}}\times G_{\alpha_{3}} . the corresponding fluid velocities are \displaystyle u_{\alpha_{1}} \displaystyle\in\mathfrak{g}_{\alpha_{1}}\otimes u_{\alpha_{2}}\in\mathfrak{g}_{\alpha_{2}}/G_{\alpha_{1}}\otimes u_{\alpha_{3}}\in(\mathfrak{g}_{\alpha_{3}}/G_{\alpha_{2}})/G_{\alpha_{1}}\,. Here, the flow of G_{\alpha_{1}} can represent the big whorls, the flow of G_{\alpha_{2}} can represent the little whorls carried in the frame of motion of the big whorls, and the flow of G_{\alpha_{3}} can represent the lesser whorls carried along successively by the two other whorls. 2.2 Diagonalising the Hamiltonian formulation The linear transformation of variables to \mu_{\alpha_{1}}=m_{\alpha_{1}}-m_{\alpha_{2}}\,,\quad\mu_{\alpha_{2}}=m_{\alpha_{2}}-m_{\alpha_{3}}\,,\quad\mu_{\alpha_{3}}=m_{\alpha_{3}}\,, diagonalises the previous entangled Lie–Poisson matrix and leads to equivalent untangled equations, \displaystyle\partial_{t}\begin{pmatrix}\mu_{\alpha_{1}}\\ \mu_{\alpha_{2}}\\ \mu_{\alpha_{3}}\end{pmatrix}=-\begin{pmatrix}\mathrm{ad}^{*}_{u_{\alpha_{1}}}\mu_{\alpha_{1}}\\ \mathrm{ad}^{*}_{(u_{\alpha_{1}}+u_{\alpha_{2}})}\mu_{\alpha_{2}}\\ \mathrm{ad}^{*}_{(u_{\alpha_{1}}+u_{\alpha_{2}}+u_{\alpha_{3}})}\mu_{\alpha_{3}}\end{pmatrix}=-\begin{pmatrix}\big(\partial_{x}\mu_{\alpha_{1}}+\mu_{\alpha_{1}}\partial_{x}\big)u_{\alpha_{1}}\\ \big(\partial_{x}\mu_{\alpha_{2}}+\mu_{\alpha_{2}}\partial_{x}\big)(u_{\alpha_{1}}+u_{\alpha_{2}})\\ \big(\partial_{x}\mu_{\alpha_{3}}+\mu_{\alpha_{3}}\partial_{x}\big)(u_{\alpha_{1}}+u_{\alpha_{2}}+u_{\alpha_{3}})\end{pmatrix}\,, where \partial_{x} denotes partial derivative along the real line, \mathbb{R} and boundary conditions are chosen so that \lim_{|x|\to\infty}u_{\alpha_{k}}(x,t)=0 for k=1,2,3 . The untangled equations arise from a diagonal Poisson bracket \{f,h\}=\langle\mu,[df,dh]\rangle on the dual of the following direct product Lie algebra \mathfrak{s}_{diag}=\mathfrak{g}_{\alpha_{1}}\ \otimes\mathfrak{g}_{\alpha_{2}}\ \otimes\ \mathfrak{g}_{\alpha_{3}} . Untangled equations for momentum differences is typical of Lagrangian reduction by stages based on the composition of Lie group actions [ Holm et al., 1998a ] . Consequently, the sum of velocities of larger scales transports the pairwise differences of momentum densities of the smaller scales, \partial_{t}\mu_{\alpha_{k}}+{\rm ad}^{*}_{\sum_{j=1}^{k}\!u_{\alpha_{j}}}\mu_{\alpha_{k}}=0\,. (2.3) 2.3 Multiscale measure-valued peakon dynamics Remarkably, the form of the dynamics in terms of pairwise differences in singular momentum densities at smaller scales admits emergent measure-valued solutions. Namely, the singular solution for N_{i} peakons , \mu_{\alpha_{i}}(x,t)=\sum_{a=1}^{N_{i}}\,p^{(a)}_{i}(t)\,\delta\big(x-x^{(a)}_{i}(t)\big)\,, in which x\in\mathbb{R} and the parameters p^{(a)}_{i} are the momenta of the N_{i} peakons at the i -th level. The starting Hamiltonian in ( 2.1 ) may be written as H=\frac{1}{2}\sum_{k=1}^{n}\int_{\mathbb{R}}u_{\alpha_{k}}m_{\alpha_{k}}\,dx=\frac{1}{2}\sum_{k=1}^{n}\int_{\mathbb{R}}u_{\alpha_{k}}\sum_{j=k}^{n}\,\mu_{\alpha_{j}}\,dx with velocity u_{\alpha_{k}} written in terms of pairwise differences in singular momentum density, given by u_{\alpha_{k}}(x,t)=\sum_{i=k}^{n}K_{\alpha_{i}}*\mu_{\alpha_{i}}=\sum_{i=k}^{n}\sum_{b=1}^{N_{i}}\,p^{(b)}_{i}(t)\,K_{\alpha_{i}}({x}-{x}^{(b)}_{i}(t)). This expression reveals the canonical Hamiltonian, \displaystyle H(\{x\},\{p\})=\frac{1}{2}\sum_{i