--- title: Weakly Nonholonomic Systems url: https://www.emergentmind.com/topics/weakly-nonholonomic-systems type: topic --- # Weakly Nonholonomic Systems Searching arXiv for recent and foundational papers relevant to weakly nonholonomic systems. {"query":"weakly nonholonomic systems arXiv Tatarinov stochastic nonholonomic constraints strong friction weakly degenerate Chaplygin", "max_results": 10} {"query":"weakly nonholonomic systems Tatarinov stochastic nonholonomic constraints strong friction weakly degenerate Chaplygin", "max_results": 10} Weakly nonholonomic systems are mechanical or control systems in which nonholonomic structure appears in a singular, approximate, or low-order sense rather than as a fixed ideal distribution. In the cited literature, the phrase is used in more than one technical meaning: constraints may depend on a small parameter and become integrable at the singular limit; large viscous friction may realize an exponentially attracting slow manifold that is nearly nonholonomic but permits small slip; or driftless control-affine systems may require Lie brackets up to second order to recover full accessibility. Closely related developments treat stochastic deformations of constraints, weakly degenerate Chaplygin reduction, and weak variational formulations for impacts, which together place weak nonholonomy within geometric mechanics, singular perturbation theory, and nonlinear control [2508.10803] [2408.00095] [1811.09120]. ## 1. Terminological scope and principal meanings A common source of confusion is that “weakly nonholonomic” does not denote a single definition across the literature. The term labels distinct, though related, mathematical regimes. | Sense | Defining feature | Source | |---|---|---| | Tatarinov-type weak nonholonomy | At $\varepsilon=0$, the constraints become completely integrable | [2508.10803] | | Strong-friction weak nonholonomy | For small $\varepsilon$ and large $\mu=1/\varepsilon$, the system is nearly nonholonomic with $h_\varepsilon=O(\varepsilon)$ slip | [2408.00095] | | Second-degree weak nonholonomy | Lie brackets up to order $2$ span the tangent space | [1811.09120] | | Weakly degenerate Chaplygin systems | $g$ is degenerate on $TQ$ but $g|_{\Delta_Q}$ is positive-definite and the reduced Lagrangian is non-degenerate | [1109.6056] | In the Tatarinov framework, one studies linear velocity constraints $$ f_s(x,\dot x,\varepsilon)=\sum_{i=1}^{n+m} a_{si}(x,\varepsilon)\,\dot x_i=0,\qquad s=1,\dots,m, $$ with $\operatorname{rank}(a_{si})=m$, and declares the system weakly nonholonomic when the constraints become completely integrable at $\varepsilon=0$. In the strong-friction framework, the terminology refers to the regime in which viscous friction with large coefficients realizes a slow manifold $D_\varepsilon$ close to the ideal nonholonomic distribution $D$. In the control-theoretic framework, “weakly nonholonomic of degree 2” refers not to approximate integrability but to controllability generated by vector fields together with first- and second-order Lie brackets. This multiplicity of meanings implies that the phrase is best interpreted relative to the analytic mechanism under study: singular integrable limit, singular frictional realization, or bracket-generating accessibility. A plausible implication is that the unifying theme is not a single definition but a family of “near-nonholonomic” regimes in which ideal nonholonomic motion is recovered only after reduction, averaging, or bracket closure. ## 2. Small-parameter weak nonholonomy and the transgression effect In the sense developed from Tatarinov’s theory, a mechanical system with $n+m$ generalized coordinates $x=(x_1,\dots,x_{n+m})$ and Lagrangian $L=L(x,\dot x,\varepsilon)$ is weakly nonholonomic when, at $\varepsilon=0$, its $m$ linear constraints become completely integrable. Precisely, there exist $m$ independent functions $\varphi_\mu(x)$ and an invertible matrix $k_{rs}(x)$ such that $$ \sum_{s=1}^m k_{rs}(x)\,a_{s i}(x,0)\,\dot x_i =\frac{d}{dt}\,\varphi_r(x), \qquad r=1,\dots,m. $$ Hence the singular limit integrates to $$ \varphi_\mu(x)=R_\mu=\mathrm{const},\qquad \mu=1,\dots,m, $$ and the original system reduces to an $n$-degree-of-freedom holonomic system depending on the parameters $(R_1,\dots,R_m)$. After introducing adapted coordinates $$ q_i=x_i,\quad i=1,\dots,n,\qquad q_{n+\mu}=\varphi_\mu(x),\quad \mu=1,\dots,m, $$ the constraints take the form $$ \dot q_{n+\mu}=0\quad(\varepsilon=0), \qquad \dot q_{n+\mu} =\varepsilon\sum_{\lambda=1}^n c_{\mu\lambda}(q,\varepsilon)\,\dot q_\lambda \quad(\varepsilon\ne0), $$ or equivalently $$ f_\mu(q,\dot q,\varepsilon) = \dot q_{n+\mu} -\varepsilon\sum_{\lambda=1}^n c_{\mu\lambda}(q,0)\,\dot q_\lambda +O(\varepsilon^2)=0. $$ The reduced $\varepsilon=0$ dynamics is assumed to be a completely integrable Hamiltonian system with Hamiltonian $H_0(q,p;R)$ and action-angle variables $(I,\varphi)$ satisfying $$ H_0=H_0(I;R),\qquad \dot I=0,\qquad \dot\varphi=\omega(I;R). $$ For $\varepsilon\ne0$, the formerly frozen coordinates $R_\mu=q_{n+\mu}$ become slow variables, and to first order one obtains $$ \widehat L = L_0(q,\dot q;R)+\varepsilon L_1(q,\dot q;R)+O(\varepsilon^2), \qquad \dot R_\mu = \varepsilon\,G_\mu(q,\dot q;R)+O(\varepsilon^2). $$ In the Hamiltonian form, $$ H(q,p,R,P;\varepsilon) = H_0(q,p;R)+\varepsilon H_1(q,p;R)+O(\varepsilon^2), $$ and averaging over the fast torus yields the leading-order normalized equations $$ \dot I_k=O(\varepsilon^2), \qquad \dot R_\mu=\varepsilon\,\overline{G_\mu}(I;R)+O(\varepsilon^2). $$ The paper identifies this slow $O(\varepsilon)$ drift of the former integration constants as the transgression effect [2508.10803]. The worked example with coordinates $(x,y,z)$, quadratic potential $V(x,y)=\frac{c}{2}(x^2+y^2)$, and weak constraint $$ \dot z=\varepsilon\bigl(y\,\dot x-x\,\dot y\bigr) $$ shows the mechanism explicitly. At $\varepsilon=0$, $z=C_0$ is constant and $(x,y)$ execute decoupled harmonic oscillations. For $\varepsilon\ne0$, the in-plane motion remains $$ \ddot x+\omega^2x=0,\qquad \ddot y+\omega^2y=0,\qquad \omega^2=\frac{c}{m}, $$ while $$ \dot z = \varepsilon\,\omega\,(C_2C_3-C_1C_4), \qquad z(t)=z(0)+\varepsilon\,\omega\,(C_2C_3-C_1C_4)\,t. $$ Thus the holonomic approximation is not stationary on times $t\sim1/\varepsilon$; instead, the system remains close to an integrable family with slowly drifting parameters. The stated validity regime is “small” nonintegrability $\varepsilon\ll1$ and time intervals up to $O(1/\varepsilon)$. ## 3. Strong friction, slow manifolds, and weak realization of constraints A different usage of weak nonholonomy arises when ideal nonholonomic constraints are realized as the singular limit of strong viscous friction. The geometric setup consists of a configuration manifold $Q$, a Riemannian metric $G$, a nonholonomic distribution $D\subset TQ$ defined by linearly independent one-forms $\{a^l\}$, the orthogonal complement $D^\perp$, and a Rayleigh dissipation function $$ R=\frac12\,a^*\,\bar\mu\,a. $$ With large friction scaled by $\mu=1/\varepsilon\gg1$, the forced affine-connection equations are $$ \nabla_{v(t)}v(t) = -\,\nabla V + F^{\sharp_G}(v)/\varepsilon, \qquad \dot q(t)=v(t), $$ where $F=-F_R$. In the limit $\varepsilon\to0$, the fast dynamics drive $F_R^\sharp(v)\to0$, hence $v\in D$. For $0<\varepsilon\ll1$, there exists an exponentially attracting slow manifold $D_\varepsilon\cong D$, represented by a nonlinear section $$ H_\varepsilon:D\to TQ,\qquad P\circ H_\varepsilon=\mathrm{id}_{D}, $$ with decomposition $$ w=v^D+h_\varepsilon(v^D), \qquad v^D=P(w), \qquad h_\varepsilon(v^D)\in D^\perp. $$ The paper proposes a novel invariance condition based on covariant derivatives and proves that it is equivalent to the classical invariance condition based on time derivatives. The covariant formulation is $$ \nabla_v[P^\perp(v)] = \nabla_v[h_\varepsilon(P(v))]. $$ From this one obtains a coordinate-free generating equation for $h_\varepsilon$ and a recursive expansion $$ h_\varepsilon(v^D)=\sum_{k=1}^\infty \varepsilon^k h^{(k)}(v^D), $$ with $$ h^{(0)}=0, \qquad h^{(1)}(v^D) = Q\circ\bigl([\nabla_{v^D}P^\perp](v^D)-P^\perp(dV^\sharp)\bigr), $$ and an explicit formula for $h^{(2)}$. The associated slip-velocity approximation is $$ v=v^D+\varepsilon h^{(1)}(v^D)+\varepsilon^2 h^{(2)}(v^D)+O(\varepsilon^3), $$ while the zeroth-order reduced dynamics reproduces the standard nonholonomic affine-connection equation $$ \nabla_{v^D}v^D=-P(dV^\sharp),\qquad v^D\in D. $$ The first-order correction introduces viscous drift terms [2408.00095]. The vertical rolling disk illustrates the construction concretely. With $Q=SE(2)\times S^1$, coordinates $q=(\theta,x,y,\phi)$, flat metric $G=\operatorname{diag}(I,m,m,J)$, and constraints $$ a^1=dx-R\cos\theta\,d\phi,\qquad a^2=dy-R\sin\theta\,d\phi, $$ the distribution $D$ is spanned by $$ e_1=\frac{\partial}{\partial\theta}, \qquad e_2 = R\cos\theta\frac{\partial}{\partial x} + R\sin\theta\frac{\partial}{\partial y} + \frac{\partial}{\partial\phi}. $$ The first-order slip term is $$ h^{(1)}(v^D) = \frac{mR}{\mu} \,(0,\;\sin\theta\,v_\theta v_\phi,\;-\cos\theta\,v_\theta v_\phi,\;0)^T, $$ and the $\varepsilon^0$ dynamics reproduces rolling without slip. The paper states that, in the small-$\varepsilon$ large-$\mu$ regime, the system behaves nearly nonholonomically with $h_\varepsilon=O(\varepsilon)$ and explicitly calls this weakly nonholonomic. Its stated validity condition is that $\mu\gtrsim100\ldots1000$ depending on geometry so that the $O(\varepsilon^2)$ terms remain small over the time-scale of interest. ## 4. Second-degree weak nonholonomy in nonlinear control In nonlinear control, weak nonholonomy refers to a bracket-generation property of a driftless control-affine system $$ \dot x=\sum_{i=1}^m u_i\,f_i(x), \qquad x\in D,\quad u\in\mathbb R^m, $$ with $m0. $$ For small $\varepsilon>0$, $$ u_k^\varepsilon(t,x) = \sum_{i_1\in S_1} a_{i_1}(x)\,\delta_{k,i_1} +\varepsilon^{-1/2}\sum_{(j_1,j_2)\in S_2} a_{j_1j_2}(x)\,\phi^{(k,\varepsilon)}_{j_1j_2}(t) +\varepsilon^{-2/3}\sum_{(\ell_1,\ell_2,\ell_3)\in S_3} a_{\ell_1\ell_2\ell_3}(x)\,\phi^{(k,\varepsilon)}_{\ell_1\ell_2\ell_3}(t). $$ The constant terms steer along $f_i$, the $O(\varepsilon^{-1/2})$ oscillations generate first brackets, and the $O(\varepsilon^{-2/3})$ bi-periodic products generate second brackets. A non-resonance condition on the chosen frequencies ensures that no spurious bracket directions appear. The main theorem states that, under the second-degree rank condition and mild smoothness and boundedness assumptions, there exists $\bar\varepsilon>0$ such that for all $0<\varepsilon\le\bar\varepsilon$, the sampled solution remains in $\operatorname{int}D$ for all $t\ge0$ and converges to $$ Z_0=\{x\in D:\nabla P(x)=0\}. $$ If $P$ is a navigation function with a single non-degenerate minimum at $x^*$, then $x(t)\to x^*$. The proof uses a one-step Volterra expansion yielding $$ x(\varepsilon)=x^0-\varepsilon\,\gamma\,\nabla P(x^0)+O(\varepsilon^{4/3}), $$ followed by a discrete-time LaSalle argument. Here weak nonholonomy is therefore an accessibility property generated by iterated brackets, not a small-slip or small-parameter perturbation of a mechanical constraint. ## 5. Stochastic deformations of nonholonomic constraints A nearby line of research studies stochastic perturbations of nonholonomic constraints through a stochastic extension of the Lagrange–d’Alembert framework. Two Stratonovich formulations are introduced. In the affine stochastic constraint, one replaces $$ \omega(q)\cdot \dot q=0 $$ by $$ \omega(q)\cdot \dot q=N(t), \qquad dN=F(q,\dot q,N)\,dt+E(q,\dot q,N)\circ dW_t. $$ The corresponding stochastic Lagrange–d’Alembert equations are $$ d\Bigl(\frac{\partial L}{\partial \dot q}\Bigr)-\frac{\partial L}{\partial q} = \omega(q)^T\lambda, \qquad \omega(q)\cdot \dot q=N(t), \qquad dN=F\,dt+E\circ dW_t. $$ In the ideal stochastic constraint, one imposes $$ \hat\omega(q,N)\cdot \dot q=0, \qquad dN=F(q,\dot q,N)\,dt+E(q,\dot q,N)\circ dW_t, $$ and obtains $$ d\Bigl(\frac{\partial L}{\partial \dot q}\Bigr)-\frac{\partial L}{\partial q} = \hat\omega(q,N)^T\lambda, \qquad \hat\omega(q,N)\cdot \dot q=0, \qquad dN=F\,dt+E\circ dW_t. $$ In the ideal case, the mechanical energy $$ E(q,\dot q)=\left\langle\frac{\partial L}{\partial \dot q},\dot q\right\rangle-L $$ is exactly preserved along Stratonovich trajectories, provided $F,E$ do not do work. In both cases, elimination of the multipliers yields Stratonovich SDEs on $(q,p,N)$, and the constraint force picks up extra terms from $dN$ or from $\partial_N\hat\omega$, thereby coupling noise into the momentum evolution [1707.03929]. The geometric interpretation is given in terms of stochastic connections and curvature. Deterministic nonholonomic constraints define an Ehresmann connection $A$ with horizontal distribution $H=\ker\omega$. In the affine stochastic case, the constraint distribution becomes the affine shift $$ H_N=\{v:\omega(v)=N\}, $$ so $A$ is replaced by $A+N$ and holonomy along loops acquires a stochastic line integral of $N(t)$. In the ideal stochastic case, the perturbed connection $\omega\mapsto\hat\omega(q,N)$ yields stochastic curvature $$ \hat\Omega=d\hat\omega+[\hat\omega,\hat\omega]. $$ On Lie groups with left-invariant constraints, these deformations give stochastic Euler–Poincaré–Suslov equations on $\mathfrak g^*$. The paper develops these constructions for the stochastic Suslov problem on $SO(3)$ and $SE(3)$. For example, the affine stochastic Suslov constraint is $$ a\cdot\Omega=N(t), $$ while the ideal version is $$ N(t)\cdot\Omega=0. $$ The analysis then tracks the fate of classical integrals. For the Kharlamova integral $I_K=I\Omega\cdot x$, the affine case is nonconservative in general, whereas in the ideal case one has $$ d(I\Omega\cdot x)=(a\,N\cdot x)\,dt $$ and conservation follows if $N\cdot x\equiv0$. For the Clebsch–Tisserand integral $I_C=I|\Omega|^2-A\,T\cdot T$, the affine case gives $$ dI_C=I_3\,\Omega_3\,N\,dt, $$ while the ideal case gives $$ dI_C=(I\Omega\cdot N)\,dt. $$ The qualitative discussion distinguishes the two noise models sharply: in the affine case the energy generally drifts and the system may slide off the ideal constraint manifold, whereas in the ideal case the energy is exactly conserved by construction. This suggests that stochastic deformation provides a controlled way to study how weak violations of ideal constraints alter holonomy, momentum drift, and long-time transport. ## 6. Geometric, Hamilton–Jacobi, and impact formulations The broader theory of weakly nonholonomic phenomena intersects with two additional frameworks: Dirac reduction for degenerate constrained systems and weak variational formulations for impacts. In the Lagrange–Dirac setting, a regular constraint distribution $\Delta_Q\subset TQ$ and a possibly degenerate Lagrangian $L:TQ\to\mathbb R$ define an induced Dirac structure $D_{\Delta_Q}\subset TT^*Q\oplus T^*T^*Q$. A Lagrange–Dirac system is a triple $(L,\Delta_Q,X)$ such that $$ \bigl(X(z),\mathfrak D L(u)\bigr)\in D_{\Delta_Q}(z), $$ which in coordinates yields the implicit Euler–Lagrange equations $$ p=\frac{\partial L}{\partial v}, \qquad \dot q=v\in\Delta_Q, \qquad \dot p-\frac{\partial L}{\partial q}\in\Delta_Q^\circ. $$ The associated Dirac–Hamilton–Jacobi equation for a section $$ \Upsilon(q)=\mathcal X(q)\oplus\gamma(q) $$ is $$ d(\mathcal E\circ\Upsilon)\in\Delta_Q^\circ, $$ with generalized energy $\mathcal E(q,v,p)=p\cdot v-L(q,v)$. When $Q$ is connected and $\Delta_Q$ is bracket-generating, this is equivalent to $\mathcal E\circ\Upsilon=E$ [1109.6056]. Within this framework, the cited paper singles out weakly degenerate Chaplygin systems: Chaplygin systems for which $$ L(q,v)=\frac12\,g_q(v,v)-V(q) $$ has degenerate $g$ on $TQ$, but $g|_{\Delta_Q}$ is positive-definite so that the reduced Lagrangian $\bar L:T\bar Q\to\mathbb R$ is non-degenerate. Reduction yields an almost Hamiltonian system on $T^*\bar Q$, $$ i_{\bar X}\Omega^{\mathrm{nh}}=d\bar H, \qquad \Omega^{\mathrm{nh}}=\bar\Omega-\Xi, $$ where the non-closed two-form $\Xi$ encodes the curvature of the principal connection and the momentum map. The reduced Dirac–HJ problem requires $$ \bar H\circ\bar\gamma=E, \qquad d\bar\gamma+\bar\gamma^*\Xi=0. $$ This is not itself a definition of weak nonholonomy, but it supplies an exact integration framework for systems in which degeneracy and nonholonomic reduction coexist. A different extension is the weak-solution theory for collisions in nonholonomic systems. There, one replaces classical trajectories by curves $x(\cdot)\in H^1([t_1,t_2];M)$ satisfying the constraint almost everywhere and the weak integral identity $$ \int_{t_1}^{t_2} \left( \frac{\partial L}{\partial x}(x,\dot x)\cdot\psi + \frac{\partial L}{\partial \dot x}(x,\dot x)\cdot\dot\psi \right)\,dt =0 $$ for all compactly supported virtual displacements $\psi$ tangent to the constraint distribution. If a collision occurs on a submanifold $N\subset M$ with stronger instantaneous constraint $\ker B(x)\subset\ker A(x)$, then the momentum jump satisfies $$ p^+-p^-\in(\ker B)^\perp, $$ equivalently $p^+-p^-=\Lambda$ with $\Lambda\in\operatorname{Im}B^T$. Under energy conservation and reversibility, the unique rebound law is $$ v^+=(I-2P)v^-, $$ where $P$ is the $G$-orthogonal projector onto $(\ker B)^\perp$ [1206.1496]. These frameworks broaden the meaning of “weak” in nonholonomic mechanics. In one direction, weak degeneracy allows reduction to almost Hamiltonian form and exact Hamilton–Jacobi integration. In another, weak solutions permit discontinuous velocities and collision laws inside the Lagrange–d’Alembert formalism. Taken together with the small-parameter, strong-friction, second-degree, and stochastic settings, they show that weakly nonholonomic systems are best understood as a family of singular or generalized limits in which ideal nonholonomic structure is either approached, perturbed, or reconstructed rather than imposed once and for all. Source: https://www.emergentmind.com/topics/weakly-nonholonomic-systems