\section{Joint recovery and alignment of two rigid configurations} \label{sec:approximation} Fix a compact family of collision-free configurations containing a neighborhood of the exact trajectory. All bodies in this family have a common positive clearance from the outer wall. Constants in this section are uniform on this family. Angles are represented by real continuous lifts, rather than reduced modulo $2\pi$. \subsection{Retained Cartesian spaces and a joint lift} For $h=1/M$, let $S_h\subset H^2_0(\Omega)$ be the clamped bicubic Hermite space on the uniform Cartesian grid. Its vertex data are $\psi,\psi_x,\psi_y,\psi_{xy}$. On an enriched coarse cell $K$, add the four center degrees of freedom of its subdivision into four squares, keeping the value and gradient zero on $\partial K$. Writing this local space as $B_h(K)$, set \[ X_{h,\mathcal A}=\curl\left(S_h+\sum_{K\in\mathcal A}B_h(K)\right), \qquad \curl\psi=(\psi_y,-\psi_x). \] Every configuration contains $\curl S_h$ and is a conforming subspace of $U$. This is the concrete retained-coarse construction of Parts~I--II, based on Cartesian Bogner--Fox--Schmit interpolation \citep{bfs2020,partii2026}. Its coarse interpolant satisfies \begin{equation}\label{eq:joint-hermite} \|I_h\psi-\psi\|_{H^j}\le Ch^{4-j}\|\psi\|_{H^4}, \qquad j=1,2. \end{equation} Choose a fixed $\chi\in C_c^5(\Omega)$ equal to one on a neighborhood of every body in the admissible configuration family. The plateau includes a fixed-width neighborhood of their union. With $x_*=(1/2,1/2)$, $r=x-x_*$, and $\zeta=(W,\eta)$, put \begin{equation}\label{eq:joint-lifting-definition} \begin{aligned} \beta_q(\zeta)&=W-\eta J(c-x_*),\\ p_{q,\zeta}(x)&=\beta_{q,1}(\zeta)r_2 -\beta_{q,2}(\zeta)r_1-\tfrac12\eta|r|^2,\\ L_q\zeta&=\curl(\chi p_{q,\zeta}),\qquad L_{h,q}\zeta=\curl I_h(\chi p_{q,\zeta}). \end{aligned} \end{equation} This fixes the additive normalization before multiplication by the cutoff. Direct differentiation gives $\curl p_{q,\zeta} =\mathcal R_q\zeta$. \begin{lemma}[A common joint rigid lift]\label{lem:joint-lift} For sufficiently small $h+\bar\eps$, the coarse lift $L_{h,q}\zeta$ equals $\mathcal R_q\zeta$ on every admissible body and throughout every regularization support centered in that body or on its boundary. Consequently $(L_{h,q}\zeta,\zeta)\in\mathbb K_h(q)$, independently of the enrichments and both moment partitions. Uniformly in $q$, \[ \|L_{h,q}\zeta\|_U\le C|\zeta|,\qquad \|(L_{h,q}-L_q)\zeta\|_U\le Ch^2|\zeta|, \quad \|(L_{h,q}-L_q)\zeta\|_0\le Ch^3|\zeta|. \] For a Lipschitz path $q(t)$ and $\zeta\in H^1(0,T;\R^3)$, \begin{equation}\label{eq:joint-lift-time} \|\partial_t[(L_{h,q(t)}-L_{q(t)})\zeta(t)]\|_{L^2_tL^2_x} \le Ch^3(1+\|q'\|_{L^\infty})\|\zeta\|_{H^1_t}. \end{equation} \end{lemma} \begin{proof} Every cell meeting a kernel support lies within $\bar\eps+\sqrt2h$ of the admissible body union. On such cells $\chi p_{q,\zeta}=p_{q,\zeta}$, a quadratic reproduced exactly by Hermite interpolation. Unit mass and zero first moment of the kernel preserve its affine curl. The same argument covers interior cells. The estimates follow from \eqref{eq:joint-hermite}; the coefficients of $p_{q(t),\zeta(t)}$ have $H^1$ time derivatives controlled by the right-hand side of \eqref{eq:joint-lift-time}. Only the fixed interpolant $I_h$ is commuted with differentiation. \end{proof} \subsection{Volume-preserving configuration alignment} The comparison follows the cutoff solenoidal-flow construction of \citet[Section 4.1]{legendre2008}, together with the transformed rigid velocities in their Section 4.2. We give its realization and the continuous-time bounds needed for the present Cartesian recovery. For two sufficiently close configurations $q=(c,\theta)$ and $\widehat q=(\widehat c,\widehat\theta)$, set $d_q=|\widehat c-c|+|\widehat\theta-\theta|$ and $O=Q(\widehat\theta-\theta)$. A comparison of velocities on the two different bodies must account for their geometry. Merely shifting a trace estimate would introduce $\sqrt{d_q}$ into the error equation and would not yield the Gronwall estimate used below. \begin{lemma}[Constructive solenoidal alignment]\label{lem:alignment} There is a smooth volume-preserving diffeomorphism $\Psi=\Psi(q,\widehat q):\Omega\to\Omega$, equal to the identity near the outer wall and satisfying \begin{equation}\label{eq:alignment-rigid-region} \Psi(x)=\widehat c+O(x-c) \quad\text{on a fixed-width neighborhood of }\overline{B(q)}. \end{equation} For every fixed finite $k$, uniformly for small $d_q$, \begin{equation}\label{eq:alignment-estimates} \|\Psi-\mathrm{id}\|_{W^{k,\infty}} +\|\Psi^{-1}-\mathrm{id}\|_{W^{k,\infty}}\le C_kd_q. \end{equation} For Lipschitz paths $q(t),\widehat q(t)$ with bounded speeds, \begin{equation}\label{eq:alignment-time} \|\partial_t\Psi\|_{W^{k,\infty}} \le C_k(d_q+|\widehat q'-q'|) \quad\text{a.e.} \end{equation} The Piola map \begin{equation}\label{eq:piola-map} (\mathcal T_\Psi v)(\Psi(x))=D\Psi(x)v(x) \end{equation} preserves $U$ and maps the exact joint space at $q$ to that at $\widehat q$, with rigid velocity transformed by $(V,\omega)\mapsto(OV,\omega)$. \end{lemma} \begin{proof} Let $\Delta c=\widehat c-c$, $\Delta\theta=\widehat\theta-\theta$, $c_s=c+s\Delta c$, and $\theta_s=\theta+s\Delta\theta$, for $0\le s\le1$. Extend the rigid $s$-velocity $\Delta c+\Delta\theta J(x-c_s)$ by the compactly supported streamfunction formula \eqref{eq:joint-lifting-definition}, but use a separate auxiliary cutoff $\chi_A\in C_c^\infty(\Omega)$ with the same plateau, and call the resulting field $H_s$. The $C^5$ cutoff used in the numerical lift is not required to be smooth to all orders. Define $\Psi$ as the time-one map of $\partial_s\Psi_s=H_s\circ\Psi_s$, $\Psi_0=\mathrm{id}$. Smallness of $d_q$ keeps all intermediate rigid neighborhoods in the plateau. Uniqueness of this ordinary differential equation then gives \eqref{eq:alignment-rigid-region}. The field $H_s$ is divergence free and zero near the outer wall. The Jacobian equation gives $\det D\Psi_s=1$. All spatial derivatives of $H_s$ are bounded by $C_kd_q$. Differentiating its coefficients along the physical time paths gives $\|\partial_tH_s\|_{W^{k,\infty}} \le C_k(d_q+|\widehat q'-q'|)$. The differentiated flow equations, Gronwall's inequality on $0\le s\le1$, and the backward flow yield \eqref{eq:alignment-estimates}--\eqref{eq:alignment-time}. These estimates use the geometric clearance, not a regularization width. Finally, change of variables in the distributional divergence identity proves that \eqref{eq:piola-map} preserves solenoidality. Near the body $D\Psi=O$, and $OJ=JO$ in two dimensions, so \[ O\mathcal R_q(V,\omega)(x) =OV+\omega J(\Psi(x)-\widehat c). \] This proves the claimed rigid trace and interior transformation. \end{proof} \subsection{Recovery of a whole rigid interior} Let $q(t)$ now be any Lipschitz admissible path. Consider $\widetilde X(t)=(\widetilde u(t),\widetilde\xi(t))\in\mathbb K(q(t))$ with $\widetilde\xi\in H^1_t$ and bounded speed of $q$. Set $w=\widetilde u-L_q\widetilde\xi$. It vanishes on the entire body, not just its boundary. Suppose $w\in H^1_tL^2_x$, its piecewise $W^{3,\infty}$ norm is uniformly bounded, and the piecewise $W^{3,\infty}$ norm of its Eulerian derivative is in $L^2_t$. \begin{lemma}[Joint common recovery]\label{lem:joint-recovery} Put $\delta=4(\bar\eps+\sqrt2h)$. For small $\delta$ there is a trajectory $Z_h(t)=(z_h(t),\widetilde\xi(t))$ in every current joint discrete space, independent of both moment partitions and the enrichment history, for which \begin{equation}\label{eq:joint-recovery-estimate} \sup_t\|z_h-\widetilde u\|_U +\|\partial_t(z_h-\widetilde u)\|_{L^2_tL^2_x} \le C\sqrt\delta, \qquad \sup_t\|z_h-\widetilde u\|_0\le C\delta^{3/2}. \end{equation} The constant depends on the displayed regularity bounds, $\|\widetilde\xi\|_{H^1}$, and $\|q'\|_{L^\infty}$, but on no acceleration of $q$. \end{lemma} \begin{proof} Let $\curl\psi=w$, with outer clamping. Since $w=0$ in the connected body, $\psi$ is a scalar $c_\psi(t)$ there. With signed distance $d(t,x)$ and a smooth cutoff $\chi_0$ equal to one on $[-1,1]$ and zero outside $[-2,2]$, set \[ \psi_\delta=\psi-\chi_0(d/\delta)(\psi-c_\psi) \quad\text{near the interface},\qquad z_h=\curl I_h\psi_\delta+L_{h,q}\widetilde\xi. \] The streamfunction is unchanged away from the interface tube and is constant throughout the body and the $\delta$ collar. Every coarse cell meeting the body or a regularization support lies in this constant region, possibly together with its original constant interior. Thus the centered coarse velocity is zero on the entire body and all its kernel supports. Both sets of moments therefore vanish exactly, and Lemma~\ref{lem:joint-lift} completes membership. For completeness, the width-dependent estimates are the moving collar estimates of Part~II \citep{partii2026}, and require only a Lipschitz configuration path. Writing $r=\psi-c_\psi$, one has $r=O(d^2)$, $\nabla r=O(d)$, and $r_t=O(d)$. The last assertion follows from the moving trace identity $c_\psi'=\psi_t|_{\Gamma(t)}$, since $\nabla\psi|_{\Gamma(t)}=0$. The Eulerian derivative is \[ (\psi_\delta-\psi)_t =-\chi_0(d/\delta)r_t -\delta^{-1}\chi_0'(d/\delta)d_t r. \] The support has area $O(\delta)$; these quadratic and linear vanishing orders give the streamfunction bounds \[ \|\psi_\delta-\psi\|_{H^2}=O(\sqrt\delta),\quad \|\psi_\delta-\psi\|_{H^1}=O(\delta^{3/2}),\quad \|(\psi_\delta-\psi)_t\|_{L^2_tH^1_x}=O(\sqrt\delta). \] The corresponding $H^4$ norms of $\psi_\delta$ and $\psi_{\delta,t}$ are $O(1+\delta^{-3/2})$ in $L^\infty_t$ and $O(1+\delta^{-5/2})$ in $L^2_t$. Here $d_t=-(c'+\theta'J(x-c))\cdot\nabla d$ has the necessary uniform spatial bounds using only $q'$, not $q''$. The fixed Hermite errors in \eqref{eq:joint-hermite} are thus $h^2\delta^{-3/2}$ in the velocity $H^1$ norm and $h^3\delta^{-5/2}$ in the time-derivative velocity $L^2$ norm. Both are at most $C\sqrt\delta$. The lifting errors from Lemma~\ref{lem:joint-lift} are smaller. This proves \eqref{eq:joint-recovery-estimate}. The moving-cutoff derivative has been included explicitly; no commutation of a time-dependent recovery map with differentiation is asserted. \end{proof} \begin{lemma}[Alignment errors at the same physical points] \label{lem:alignment-physical-norms} Suppose $u\in U$ is piecewise $W^{2,\infty}$ across $\Gamma(q)$. For small $d_q$, \begin{equation}\label{eq:alignment-physical-norms} \|\mathcal T_\Psi u-u\|_0\le Cd_q\|u\|_{H^1},\qquad \|\mathcal T_\Psi u-u\|_U \le C\sqrt{d_q}\, (\|u\|_{W^{2,\infty}(B(q))}+\|u\|_{W^{2,\infty}(F(q))}). \end{equation} \end{lemma} \begin{proof} Integrating $u$ along the alignment flow, whose velocity is $O(d_q)$, proves the $L^2$ bound; the factor $D\Psi-I$ is also $O(d_q)$. For the gradient bound, remove the $Cd_q$ interface tube. Off this tube, an alignment trajectory remains in one smooth phase, and its gradient difference is $O(d_q)$ by the chain rule. Inside the tube, both endpoint gradients are bounded and its area is $O(d_q)$, giving the stated square-root bound. The map is the identity near the outer wall. This estimate allows a jump of the first derivatives across the fluid--rigid interface. \end{proof}